diff options
author | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-11 08:17:27 +0000 |
---|---|---|
committer | Daniel Baumann <daniel.baumann@progress-linux.org> | 2024-04-11 08:17:27 +0000 |
commit | f215e02bf85f68d3a6106c2a1f4f7f063f819064 (patch) | |
tree | 6bb5b92c046312c4e95ac2620b10ddf482d3fa8b /src/VBox/ExtPacks/VBoxDTrace/onnv/uts | |
parent | Initial commit. (diff) | |
download | virtualbox-f215e02bf85f68d3a6106c2a1f4f7f063f819064.tar.xz virtualbox-f215e02bf85f68d3a6106c2a1f4f7f063f819064.zip |
Adding upstream version 7.0.14-dfsg.upstream/7.0.14-dfsg
Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>
Diffstat (limited to '')
36 files changed, 31971 insertions, 0 deletions
diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.c new file mode 100644 index 00000000..d4ca1d30 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.c @@ -0,0 +1,1214 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2010 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/errno.h> +#include <sys/cpuvar.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/cmn_err.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/ksynch.h> +#include <sys/conf.h> +#include <sys/kmem.h> +#include <sys/kcpc.h> +#include <sys/cap_util.h> +#include <sys/cpc_pcbe.h> +#include <sys/cpc_impl.h> +#include <sys/dtrace_impl.h> + +/* + * DTrace CPU Performance Counter Provider + * --------------------------------------- + * + * The DTrace cpc provider allows DTrace consumers to access the CPU + * performance counter overflow mechanism of a CPU. The configuration + * presented in a probe specification is programmed into the performance + * counter hardware of all available CPUs on a system. Programming the + * hardware causes a counter on each CPU to begin counting events of the + * given type. When the specified number of events have occurred, an overflow + * interrupt will be generated and the probe is fired. + * + * The required configuration for the performance counter is encoded into + * the probe specification and this includes the performance counter event + * name, processor mode, overflow rate and an optional unit mask. + * + * Most processors provide several counters (PICs) which can count all or a + * subset of the events available for a given CPU. However, when overflow + * profiling is being used, not all CPUs can detect which counter generated the + * overflow interrupt. In this case we cannot reliably determine which counter + * overflowed and we therefore only allow such CPUs to configure one event at + * a time. Processors that can determine the counter which overflowed are + * allowed to program as many events at one time as possible (in theory up to + * the number of instrumentation counters supported by that platform). + * Therefore, multiple consumers can enable multiple probes at the same time + * on such platforms. Platforms which cannot determine the source of an + * overflow interrupt are only allowed to program a single event at one time. + * + * The performance counter hardware is made available to consumers on a + * first-come, first-served basis. Only a finite amount of hardware resource + * is available and, while we make every attempt to accomodate requests from + * consumers, we must deny requests when hardware resources have been exhausted. + * A consumer will fail to enable probes when resources are currently in use. + * + * The cpc provider contends for shared hardware resources along with other + * consumers of the kernel CPU performance counter subsystem (e.g. cpustat(1M)). + * Only one such consumer can use the performance counters at any one time and + * counters are made available on a first-come, first-served basis. As with + * cpustat, the cpc provider has priority over per-LWP libcpc usage (e.g. + * cputrack(1)). Invoking the cpc provider will cause all existing per-LWP + * counter contexts to be invalidated. + */ + +typedef struct dcpc_probe { + char dcpc_event_name[CPC_MAX_EVENT_LEN]; + int dcpc_flag; /* flags (USER/SYS) */ + uint32_t dcpc_ovfval; /* overflow value */ + int64_t dcpc_umask; /* umask/emask for this event */ + int dcpc_picno; /* pic this event is programmed in */ + int dcpc_enabled; /* probe is actually enabled? */ + int dcpc_disabling; /* probe is currently being disabled */ + dtrace_id_t dcpc_id; /* probeid this request is enabling */ + int dcpc_actv_req_idx; /* idx into dcpc_actv_reqs[] */ +} dcpc_probe_t; + +static dev_info_t *dcpc_devi; +static dtrace_provider_id_t dcpc_pid; +static dcpc_probe_t **dcpc_actv_reqs; +static uint32_t dcpc_enablings = 0; +static int dcpc_ovf_mask = 0; +static int dcpc_mult_ovf_cap = 0; +static int dcpc_mask_type = 0; + +/* + * When the dcpc provider is loaded, dcpc_min_overflow is set to either + * DCPC_MIN_OVF_DEFAULT or the value that dcpc-min-overflow is set to in + * the dcpc.conf file. Decrease this value to set probes with smaller + * overflow values. Remember that very small values could render a system + * unusable with frequently occurring events. + */ +#define DCPC_MIN_OVF_DEFAULT 5000 +static uint32_t dcpc_min_overflow; + +static int dcpc_aframes = 0; /* override for artificial frame setting */ +#if defined(__x86) +#define DCPC_ARTIFICIAL_FRAMES 8 +#elif defined(__sparc) +#define DCPC_ARTIFICIAL_FRAMES 2 +#endif + +/* + * Called from the platform overflow interrupt handler. 'bitmap' is a mask + * which contains the pic(s) that have overflowed. + */ +static void +dcpc_fire(uint64_t bitmap) +{ + int i; + + /* + * No counter was marked as overflowing. Shout about it and get out. + */ + if ((bitmap & dcpc_ovf_mask) == 0) { + cmn_err(CE_NOTE, "dcpc_fire: no counter overflow found\n"); + return; + } + + /* + * This is the common case of a processor that doesn't support + * multiple overflow events. Such systems are only allowed a single + * enabling and therefore we just look for the first entry in + * the active request array. + */ + if (!dcpc_mult_ovf_cap) { + for (i = 0; i < cpc_ncounters; i++) { + if (dcpc_actv_reqs[i] != NULL) { + dtrace_probe(dcpc_actv_reqs[i]->dcpc_id, + CPU->cpu_cpcprofile_pc, + CPU->cpu_cpcprofile_upc, 0, 0, 0); + return; + } + } + return; + } + + /* + * This is a processor capable of handling multiple overflow events. + * Iterate over the array of active requests and locate the counters + * that overflowed (note: it is possible for more than one counter to + * have overflowed at the same time). + */ + for (i = 0; i < cpc_ncounters; i++) { + if (dcpc_actv_reqs[i] != NULL && + (bitmap & (1ULL << dcpc_actv_reqs[i]->dcpc_picno))) { + dtrace_probe(dcpc_actv_reqs[i]->dcpc_id, + CPU->cpu_cpcprofile_pc, + CPU->cpu_cpcprofile_upc, 0, 0, 0); + } + } +} + +static void +dcpc_create_probe(dtrace_provider_id_t id, const char *probename, + char *eventname, int64_t umask, uint32_t ovfval, char flag) +{ + dcpc_probe_t *pp; + int nr_frames = DCPC_ARTIFICIAL_FRAMES + dtrace_mach_aframes(); + + if (dcpc_aframes) + nr_frames = dcpc_aframes; + + if (dtrace_probe_lookup(id, NULL, NULL, probename) != 0) + return; + + pp = kmem_zalloc(sizeof (dcpc_probe_t), KM_SLEEP); + (void) strncpy(pp->dcpc_event_name, eventname, + sizeof (pp->dcpc_event_name) - 1); + pp->dcpc_event_name[sizeof (pp->dcpc_event_name) - 1] = '\0'; + pp->dcpc_flag = flag | CPC_OVF_NOTIFY_EMT; + pp->dcpc_ovfval = ovfval; + pp->dcpc_umask = umask; + pp->dcpc_actv_req_idx = pp->dcpc_picno = pp->dcpc_disabling = -1; + + pp->dcpc_id = dtrace_probe_create(id, NULL, NULL, probename, + nr_frames, pp); +} + +/*ARGSUSED*/ +static void +dcpc_provide(void *arg, const dtrace_probedesc_t *desc) +{ + /* + * The format of a probe is: + * + * event_name-mode-{optional_umask}-overflow_rate + * e.g. + * DC_refill_from_system-user-0x1e-50000, or, + * DC_refill_from_system-all-10000 + * + */ + char *str, *end, *p; + int i, flag = 0; + char event[CPC_MAX_EVENT_LEN]; + long umask = -1, val = 0; + size_t evlen, len; + + /* + * The 'cpc' provider offers no probes by default. + */ + if (desc == NULL) + return; + + len = strlen(desc->dtpd_name); + p = str = kmem_alloc(len + 1, KM_SLEEP); + (void) strcpy(str, desc->dtpd_name); + + /* + * We have a poor man's strtok() going on here. Replace any hyphens + * in the the probe name with NULL characters in order to make it + * easy to parse the string with regular string functions. + */ + for (i = 0; i < len; i++) { + if (str[i] == '-') + str[i] = '\0'; + } + + /* + * The first part of the string must be either a platform event + * name or a generic event name. + */ + evlen = strlen(p); + (void) strncpy(event, p, CPC_MAX_EVENT_LEN - 1); + event[CPC_MAX_EVENT_LEN - 1] = '\0'; + + /* + * The next part of the name is the mode specification. Valid + * settings are "user", "kernel" or "all". + */ + p += evlen + 1; + + if (strcmp(p, "user") == 0) + flag |= CPC_COUNT_USER; + else if (strcmp(p, "kernel") == 0) + flag |= CPC_COUNT_SYSTEM; + else if (strcmp(p, "all") == 0) + flag |= CPC_COUNT_USER | CPC_COUNT_SYSTEM; + else + goto err; + + /* + * Next we either have a mask specification followed by an overflow + * rate or just an overflow rate on its own. + */ + p += strlen(p) + 1; + if (p[0] == '0' && (p[1] == 'x' || p[1] == 'X')) { + /* + * A unit mask can only be specified if: + * 1) this performance counter back end supports masks. + * 2) the specified event is platform specific. + * 3) a valid hex number is converted. + * 4) no extraneous characters follow the mask specification. + */ + if (dcpc_mask_type != 0 && strncmp(event, "PAPI", 4) != 0 && + ddi_strtol(p, &end, 16, &umask) == 0 && + end == p + strlen(p)) { + p += strlen(p) + 1; + } else { + goto err; + } + } + + /* + * This final part must be an overflow value which has to be greater + * than the minimum permissible overflow rate. + */ + if ((ddi_strtol(p, &end, 10, &val) != 0) || end != p + strlen(p) || + val < dcpc_min_overflow) + goto err; + + /* + * Validate the event and create the probe. + */ + for (i = 0; i < cpc_ncounters; i++) { + char *events, *cp, *p, *end; + int found = 0, j; + size_t llen; + + if ((events = kcpc_list_events(i)) == NULL) + goto err; + + llen = strlen(events); + p = cp = ddi_strdup(events, KM_NOSLEEP); + end = cp + llen; + + for (j = 0; j < llen; j++) { + if (cp[j] == ',') + cp[j] = '\0'; + } + + while (p < end && found == 0) { + if (strcmp(p, event) == 0) { + dcpc_create_probe(dcpc_pid, desc->dtpd_name, + event, umask, (uint32_t)val, flag); + found = 1; + } + p += strlen(p) + 1; + } + kmem_free(cp, llen + 1); + + if (found) + break; + } + +err: + kmem_free(str, len + 1); +} + +/*ARGSUSED*/ +static void +dcpc_destroy(void *arg, dtrace_id_t id, void *parg) +{ + dcpc_probe_t *pp = parg; + + ASSERT(pp->dcpc_enabled == 0); + kmem_free(pp, sizeof (dcpc_probe_t)); +} + +/*ARGSUSED*/ +static int +dcpc_usermode(void *arg, dtrace_id_t id, void *parg) +{ + return (CPU->cpu_cpcprofile_pc == 0); +} + +static void +dcpc_populate_set(cpu_t *c, dcpc_probe_t *pp, kcpc_set_t *set, int reqno) +{ + kcpc_set_t *oset; + int i; + + (void) strncpy(set->ks_req[reqno].kr_event, pp->dcpc_event_name, + CPC_MAX_EVENT_LEN); + set->ks_req[reqno].kr_config = NULL; + set->ks_req[reqno].kr_index = reqno; + set->ks_req[reqno].kr_picnum = -1; + set->ks_req[reqno].kr_flags = pp->dcpc_flag; + + /* + * If a unit mask has been specified then detect which attribute + * the platform needs. For now, it's either "umask" or "emask". + */ + if (pp->dcpc_umask >= 0) { + set->ks_req[reqno].kr_attr = + kmem_zalloc(sizeof (kcpc_attr_t), KM_SLEEP); + set->ks_req[reqno].kr_nattrs = 1; + if (dcpc_mask_type & DCPC_UMASK) + (void) strncpy(set->ks_req[reqno].kr_attr->ka_name, + "umask", 5); + else + (void) strncpy(set->ks_req[reqno].kr_attr->ka_name, + "emask", 5); + set->ks_req[reqno].kr_attr->ka_val = pp->dcpc_umask; + } else { + set->ks_req[reqno].kr_attr = NULL; + set->ks_req[reqno].kr_nattrs = 0; + } + + /* + * If this probe is enabled, obtain its current countdown value + * and use that. The CPUs cpc context might not exist yet if we + * are dealing with a CPU that is just coming online. + */ + if (pp->dcpc_enabled && (c->cpu_cpc_ctx != NULL)) { + oset = c->cpu_cpc_ctx->kc_set; + + for (i = 0; i < oset->ks_nreqs; i++) { + if (strcmp(oset->ks_req[i].kr_event, + set->ks_req[reqno].kr_event) == 0) { + set->ks_req[reqno].kr_preset = + *(oset->ks_req[i].kr_data); + } + } + } else { + set->ks_req[reqno].kr_preset = UINT64_MAX - pp->dcpc_ovfval; + } + + set->ks_nreqs++; +} + + +/* + * Create a fresh request set for the enablings represented in the + * 'dcpc_actv_reqs' array which contains the probes we want to be + * in the set. This can be called for several reasons: + * + * 1) We are on a single or multi overflow platform and we have no + * current events so we can just create the set and initialize it. + * 2) We are on a multi-overflow platform and we already have one or + * more existing events and we are adding a new enabling. Create a + * new set and copy old requests in and then add the new request. + * 3) We are on a multi-overflow platform and we have just removed an + * enabling but we still have enablings whch are valid. Create a new + * set and copy in still valid requests. + */ +static kcpc_set_t * +dcpc_create_set(cpu_t *c) +{ + int i, reqno = 0; + int active_requests = 0; + kcpc_set_t *set; + + /* + * First get a count of the number of currently active requests. + * Note that dcpc_actv_reqs[] should always reflect which requests + * we want to be in the set that is to be created. It is the + * responsibility of the caller of dcpc_create_set() to adjust that + * array accordingly beforehand. + */ + for (i = 0; i < cpc_ncounters; i++) { + if (dcpc_actv_reqs[i] != NULL) + active_requests++; + } + + set = kmem_zalloc(sizeof (kcpc_set_t), KM_SLEEP); + + set->ks_req = + kmem_zalloc(sizeof (kcpc_request_t) * active_requests, KM_SLEEP); + + set->ks_data = + kmem_zalloc(active_requests * sizeof (uint64_t), KM_SLEEP); + + /* + * Look for valid entries in the active requests array and populate + * the request set for any entries found. + */ + for (i = 0; i < cpc_ncounters; i++) { + if (dcpc_actv_reqs[i] != NULL) { + dcpc_populate_set(c, dcpc_actv_reqs[i], set, reqno); + reqno++; + } + } + + return (set); +} + +static int +dcpc_program_cpu_event(cpu_t *c) +{ + int i, j, subcode; + kcpc_ctx_t *ctx, *octx; + kcpc_set_t *set; + + set = dcpc_create_set(c); + + set->ks_ctx = ctx = kcpc_ctx_alloc(KM_SLEEP); + ctx->kc_set = set; + ctx->kc_cpuid = c->cpu_id; + + if (kcpc_assign_reqs(set, ctx) != 0) + goto err; + + if (kcpc_configure_reqs(ctx, set, &subcode) != 0) + goto err; + + for (i = 0; i < set->ks_nreqs; i++) { + for (j = 0; j < cpc_ncounters; j++) { + if (dcpc_actv_reqs[j] != NULL && + strcmp(set->ks_req[i].kr_event, + dcpc_actv_reqs[j]->dcpc_event_name) == 0) { + dcpc_actv_reqs[j]->dcpc_picno = + set->ks_req[i].kr_picnum; + } + } + } + + /* + * If we already have an active enabling then save the current cpc + * context away. + */ + octx = c->cpu_cpc_ctx; + + kcpc_cpu_program(c, ctx); + + if (octx != NULL) { + kcpc_set_t *oset = octx->kc_set; + kmem_free(oset->ks_data, oset->ks_nreqs * sizeof (uint64_t)); + kcpc_free_configs(oset); + kcpc_free_set(oset); + kcpc_ctx_free(octx); + } + + return (0); + +err: + /* + * We failed to configure this request up so free things up and + * get out. + */ + kcpc_free_configs(set); + kmem_free(set->ks_data, set->ks_nreqs * sizeof (uint64_t)); + kcpc_free_set(set); + kcpc_ctx_free(ctx); + + return (-1); +} + +static void +dcpc_disable_cpu(cpu_t *c) +{ + kcpc_ctx_t *ctx; + kcpc_set_t *set; + + /* + * Leave this CPU alone if it's already offline. + */ + if (c->cpu_flags & CPU_OFFLINE) + return; + + /* + * Grab CPUs CPC context before kcpc_cpu_stop() stops counters and + * changes it. + */ + ctx = c->cpu_cpc_ctx; + + kcpc_cpu_stop(c, B_FALSE); + + set = ctx->kc_set; + + kcpc_free_configs(set); + kmem_free(set->ks_data, set->ks_nreqs * sizeof (uint64_t)); + kcpc_free_set(set); + kcpc_ctx_free(ctx); +} + +/* + * The dcpc_*_interrupts() routines are responsible for manipulating the + * per-CPU dcpc interrupt state byte. The purpose of the state byte is to + * synchronize processing of hardware overflow interrupts wth configuration + * changes made to the CPU performance counter subsystem by the dcpc provider. + * + * The dcpc provider claims ownership of the overflow interrupt mechanism + * by transitioning the state byte from DCPC_INTR_INACTIVE (indicating the + * dcpc provider is not in use) to DCPC_INTR_FREE (the dcpc provider owns the + * overflow mechanism and interrupts may be processed). Before modifying + * a CPUs configuration state the state byte is transitioned from + * DCPC_INTR_FREE to DCPC_INTR_CONFIG ("configuration in process" state). + * The hardware overflow handler, kcpc_hw_overflow_intr(), will only process + * an interrupt when a configuration is not in process (i.e. the state is + * marked as free). During interrupt processing the state is set to + * DCPC_INTR_PROCESSING by the overflow handler. When the last dcpc based + * enabling is removed, the state byte is set to DCPC_INTR_INACTIVE to indicate + * the dcpc provider is no longer interested in overflow interrupts. + */ +static void +dcpc_block_interrupts(void) +{ + cpu_t *c = cpu_list; + uint8_t *state; + + ASSERT(cpu_core[c->cpu_id].cpuc_dcpc_intr_state != DCPC_INTR_INACTIVE); + + do { + state = &cpu_core[c->cpu_id].cpuc_dcpc_intr_state; + + while (atomic_cas_8(state, DCPC_INTR_FREE, + DCPC_INTR_CONFIG) != DCPC_INTR_FREE) + continue; + + } while ((c = c->cpu_next) != cpu_list); +} + +/* + * Set all CPUs dcpc interrupt state to DCPC_INTR_FREE to indicate that + * overflow interrupts can be processed safely. + */ +static void +dcpc_release_interrupts(void) +{ + cpu_t *c = cpu_list; + + ASSERT(cpu_core[c->cpu_id].cpuc_dcpc_intr_state != DCPC_INTR_INACTIVE); + + do { + cpu_core[c->cpu_id].cpuc_dcpc_intr_state = DCPC_INTR_FREE; + membar_producer(); + } while ((c = c->cpu_next) != cpu_list); +} + +/* + * Transition all CPUs dcpc interrupt state from DCPC_INTR_INACTIVE to + * to DCPC_INTR_FREE. This indicates that the dcpc provider is now + * responsible for handling all overflow interrupt activity. Should only be + * called before enabling the first dcpc based probe. + */ +static void +dcpc_claim_interrupts(void) +{ + cpu_t *c = cpu_list; + + ASSERT(cpu_core[c->cpu_id].cpuc_dcpc_intr_state == DCPC_INTR_INACTIVE); + + do { + cpu_core[c->cpu_id].cpuc_dcpc_intr_state = DCPC_INTR_FREE; + membar_producer(); + } while ((c = c->cpu_next) != cpu_list); +} + +/* + * Set all CPUs dcpc interrupt state to DCPC_INTR_INACTIVE to indicate that + * the dcpc provider is no longer processing overflow interrupts. Only called + * during removal of the last dcpc based enabling. + */ +static void +dcpc_surrender_interrupts(void) +{ + cpu_t *c = cpu_list; + + ASSERT(cpu_core[c->cpu_id].cpuc_dcpc_intr_state != DCPC_INTR_INACTIVE); + + do { + cpu_core[c->cpu_id].cpuc_dcpc_intr_state = DCPC_INTR_INACTIVE; + membar_producer(); + } while ((c = c->cpu_next) != cpu_list); +} + +/* + * dcpc_program_event() can be called owing to a new enabling or if a multi + * overflow platform has disabled a request but needs to program the requests + * that are still valid. + * + * Every invocation of dcpc_program_event() will create a new kcpc_ctx_t + * and a new request set which contains the new enabling and any old enablings + * which are still valid (possible with multi-overflow platforms). + */ +static int +dcpc_program_event(dcpc_probe_t *pp) +{ + cpu_t *c; + int ret = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + kpreempt_disable(); + + dcpc_block_interrupts(); + + c = cpu_list; + + do { + /* + * Skip CPUs that are currently offline. + */ + if (c->cpu_flags & CPU_OFFLINE) + continue; + + /* + * Stop counters but preserve existing DTrace CPC context + * if there is one. + * + * If we come here when the first event is programmed for a CPU, + * there should be no DTrace CPC context installed. In this + * case, kcpc_cpu_stop() will ensure that there is no other + * context on the CPU. + * + * If we add new enabling to the original one, the CPU should + * have the old DTrace CPC context which we need to keep around + * since dcpc_program_event() will add to it. + */ + if (c->cpu_cpc_ctx != NULL) + kcpc_cpu_stop(c, B_TRUE); + } while ((c = c->cpu_next) != cpu_list); + + dcpc_release_interrupts(); + + /* + * If this enabling is being removed (in the case of a multi event + * capable system with more than one active enabling), we can now + * update the active request array to reflect the enablings that need + * to be reprogrammed. + */ + if (pp->dcpc_disabling == 1) + dcpc_actv_reqs[pp->dcpc_actv_req_idx] = NULL; + + do { + /* + * Skip CPUs that are currently offline. + */ + if (c->cpu_flags & CPU_OFFLINE) + continue; + + ret = dcpc_program_cpu_event(c); + } while ((c = c->cpu_next) != cpu_list && ret == 0); + + /* + * If dcpc_program_cpu_event() fails then it is because we couldn't + * configure the requests in the set for the CPU and not because of + * an error programming the hardware. If we have a failure here then + * we assume no CPUs have been programmed in the above step as they + * are all configured identically. + */ + if (ret != 0) { + pp->dcpc_enabled = 0; + kpreempt_enable(); + return (-1); + } + + if (pp->dcpc_disabling != 1) + pp->dcpc_enabled = 1; + + kpreempt_enable(); + + return (0); +} + +/*ARGSUSED*/ +static int +dcpc_enable(void *arg, dtrace_id_t id, void *parg) +{ + dcpc_probe_t *pp = parg; + int i, found = 0; + cpu_t *c; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * Bail out if the counters are being used by a libcpc consumer. + */ + rw_enter(&kcpc_cpuctx_lock, RW_READER); + if (kcpc_cpuctx > 0) { + rw_exit(&kcpc_cpuctx_lock); + return (-1); + } + + dtrace_cpc_in_use++; + rw_exit(&kcpc_cpuctx_lock); + + /* + * Locate this enabling in the first free entry of the active + * request array. + */ + for (i = 0; i < cpc_ncounters; i++) { + if (dcpc_actv_reqs[i] == NULL) { + dcpc_actv_reqs[i] = pp; + pp->dcpc_actv_req_idx = i; + found = 1; + break; + } + } + + /* + * If we couldn't find a slot for this probe then there is no + * room at the inn. + */ + if (!found) { + dtrace_cpc_in_use--; + return (-1); + } + + ASSERT(pp->dcpc_actv_req_idx >= 0); + + /* + * DTrace is taking over CPC contexts, so stop collecting + * capacity/utilization data for all CPUs. + */ + if (dtrace_cpc_in_use == 1) + cu_disable(); + + /* + * The following must hold true if we are to (attempt to) enable + * this request: + * + * 1) No enablings currently exist. We allow all platforms to + * proceed if this is true. + * + * OR + * + * 2) If the platform is multi overflow capable and there are + * less valid enablings than there are counters. There is no + * guarantee that a platform can accommodate as many events as + * it has counters for but we will at least try to program + * up to that many requests. + * + * The 'dcpc_enablings' variable is implictly protected by locking + * provided by the DTrace framework and the cpu management framework. + */ + if (dcpc_enablings == 0 || (dcpc_mult_ovf_cap && + dcpc_enablings < cpc_ncounters)) { + /* + * Before attempting to program the first enabling we need to + * invalidate any lwp-based contexts and lay claim to the + * overflow interrupt mechanism. + */ + if (dcpc_enablings == 0) { + kcpc_invalidate_all(); + dcpc_claim_interrupts(); + } + + if (dcpc_program_event(pp) == 0) { + dcpc_enablings++; + return (0); + } + } + + /* + * If active enablings existed before we failed to enable this probe + * on a multi event capable platform then we need to restart counters + * as they will have been stopped in the attempted configuration. The + * context should now just contain the request prior to this failed + * enabling. + */ + if (dcpc_enablings > 0 && dcpc_mult_ovf_cap) { + c = cpu_list; + + ASSERT(dcpc_mult_ovf_cap == 1); + do { + /* + * Skip CPUs that are currently offline. + */ + if (c->cpu_flags & CPU_OFFLINE) + continue; + + kcpc_cpu_program(c, c->cpu_cpc_ctx); + } while ((c = c->cpu_next) != cpu_list); + } + + /* + * Give up any claim to the overflow interrupt mechanism if no + * dcpc based enablings exist. + */ + if (dcpc_enablings == 0) + dcpc_surrender_interrupts(); + + dtrace_cpc_in_use--; + dcpc_actv_reqs[pp->dcpc_actv_req_idx] = NULL; + pp->dcpc_actv_req_idx = pp->dcpc_picno = -1; + + /* + * If all probes are removed, enable capacity/utilization data + * collection for every CPU. + */ + if (dtrace_cpc_in_use == 0) + cu_enable(); + + return (-1); +} + +/* + * If only one enabling is active then remove the context and free + * everything up. If there are multiple enablings active then remove this + * one, its associated meta-data and re-program the hardware. + */ +/*ARGSUSED*/ +static void +dcpc_disable(void *arg, dtrace_id_t id, void *parg) +{ + cpu_t *c; + dcpc_probe_t *pp = parg; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + kpreempt_disable(); + + /* + * This probe didn't actually make it as far as being fully enabled + * so we needn't do anything with it. + */ + if (pp->dcpc_enabled == 0) { + /* + * If we actually allocated this request a slot in the + * request array but failed to enabled it then remove the + * entry in the array. + */ + if (pp->dcpc_actv_req_idx >= 0) { + dcpc_actv_reqs[pp->dcpc_actv_req_idx] = NULL; + pp->dcpc_actv_req_idx = pp->dcpc_picno = + pp->dcpc_disabling = -1; + } + + kpreempt_enable(); + return; + } + + /* + * If this is the only enabling then stop all the counters and + * free up the meta-data. + */ + if (dcpc_enablings == 1) { + ASSERT(dtrace_cpc_in_use == 1); + + dcpc_block_interrupts(); + + c = cpu_list; + + do { + dcpc_disable_cpu(c); + } while ((c = c->cpu_next) != cpu_list); + + dcpc_actv_reqs[pp->dcpc_actv_req_idx] = NULL; + dcpc_surrender_interrupts(); + } else { + /* + * This platform can support multiple overflow events and + * the enabling being disabled is not the last one. Remove this + * enabling and re-program the hardware with the new config. + */ + ASSERT(dcpc_mult_ovf_cap); + ASSERT(dcpc_enablings > 1); + + pp->dcpc_disabling = 1; + (void) dcpc_program_event(pp); + } + + kpreempt_enable(); + + dcpc_enablings--; + dtrace_cpc_in_use--; + pp->dcpc_enabled = 0; + pp->dcpc_actv_req_idx = pp->dcpc_picno = pp->dcpc_disabling = -1; + + /* + * If all probes are removed, enable capacity/utilization data + * collection for every CPU + */ + if (dtrace_cpc_in_use == 0) + cu_enable(); +} + +/*ARGSUSED*/ +static int +dcpc_cpu_setup(cpu_setup_t what, processorid_t cpu, void *arg) +{ + cpu_t *c; + uint8_t *state; + + ASSERT(MUTEX_HELD(&cpu_lock)); + + switch (what) { + case CPU_OFF: + /* + * Offline CPUs are not allowed to take part so remove this + * CPU if we are actively tracing. + */ + if (dtrace_cpc_in_use) { + c = cpu_get(cpu); + state = &cpu_core[c->cpu_id].cpuc_dcpc_intr_state; + + /* + * Indicate that a configuration is in process in + * order to stop overflow interrupts being processed + * on this CPU while we disable it. + */ + while (atomic_cas_8(state, DCPC_INTR_FREE, + DCPC_INTR_CONFIG) != DCPC_INTR_FREE) + continue; + + dcpc_disable_cpu(c); + + /* + * Reset this CPUs interrupt state as the configuration + * has ended. + */ + cpu_core[c->cpu_id].cpuc_dcpc_intr_state = + DCPC_INTR_FREE; + membar_producer(); + } + break; + + case CPU_ON: + case CPU_SETUP: + /* + * This CPU is being initialized or brought online so program + * it with the current request set if we are actively tracing. + */ + if (dtrace_cpc_in_use) { + c = cpu_get(cpu); + (void) dcpc_program_cpu_event(c); + } + break; + + default: + break; + } + + return (0); +} + +static dtrace_pattr_t dcpc_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_CPU }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t dcpc_pops = { + dcpc_provide, + NULL, + dcpc_enable, + dcpc_disable, + NULL, + NULL, + NULL, + NULL, + dcpc_usermode, + dcpc_destroy +}; + +/*ARGSUSED*/ +static int +dcpc_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +/*ARGSUSED*/ +static int +dcpc_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)dcpc_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +static int +dcpc_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(dcpc_pid) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + + mutex_enter(&cpu_lock); + unregister_cpu_setup_func(dcpc_cpu_setup, NULL); + mutex_exit(&cpu_lock); + + kmem_free(dcpc_actv_reqs, cpc_ncounters * sizeof (dcpc_probe_t *)); + + kcpc_unregister_dcpc(); + + return (DDI_SUCCESS); +} + +static int +dcpc_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + uint_t caps; + char *attrs; + + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (kcpc_pcbe_loaded() == -1) + return (DDI_FAILURE); + + caps = kcpc_pcbe_capabilities(); + + if (!(caps & CPC_CAP_OVERFLOW_INTERRUPT)) { + cmn_err(CE_NOTE, "!dcpc: Counter Overflow not supported"\ + " on this processor"); + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "dcpc", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("cpc", &dcpc_attr, DTRACE_PRIV_KERNEL, + NULL, &dcpc_pops, NULL, &dcpc_pid) != 0) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + mutex_enter(&cpu_lock); + register_cpu_setup_func(dcpc_cpu_setup, NULL); + mutex_exit(&cpu_lock); + + dcpc_ovf_mask = (1 << cpc_ncounters) - 1; + ASSERT(dcpc_ovf_mask != 0); + + if (caps & CPC_CAP_OVERFLOW_PRECISE) + dcpc_mult_ovf_cap = 1; + + /* + * Determine which, if any, mask attribute the back-end can use. + */ + attrs = kcpc_list_attrs(); + if (strstr(attrs, "umask") != NULL) + dcpc_mask_type |= DCPC_UMASK; + else if (strstr(attrs, "emask") != NULL) + dcpc_mask_type |= DCPC_EMASK; + + /* + * The dcpc_actv_reqs array is used to store the requests that + * we currently have programmed. The order of requests in this + * array is not necessarily the order that the event appears in + * the kcpc_request_t array. Once entered into a slot in the array + * the entry is not moved until it's removed. + */ + dcpc_actv_reqs = + kmem_zalloc(cpc_ncounters * sizeof (dcpc_probe_t *), KM_SLEEP); + + dcpc_min_overflow = ddi_prop_get_int(DDI_DEV_T_ANY, devi, + DDI_PROP_DONTPASS, "dcpc-min-overflow", DCPC_MIN_OVF_DEFAULT); + + kcpc_register_dcpc(dcpc_fire); + + ddi_report_dev(devi); + dcpc_devi = devi; + + return (DDI_SUCCESS); +} + +static struct cb_ops dcpc_cb_ops = { + dcpc_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops dcpc_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + dcpc_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + dcpc_attach, /* attach */ + dcpc_detach, /* detach */ + nodev, /* reset */ + &dcpc_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type */ + "DTrace CPC Module", /* name of module */ + &dcpc_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.conf new file mode 100644 index 00000000..3f186940 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dcpc.conf @@ -0,0 +1,33 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License (the "License"). +# You may not use this file except in compliance with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2009 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +name="dcpc" parent="pseudo" instance=0; + +# +# dcpc-min-overflow is the lower limit on the rate of overflow that can be +# set on a probe. The default value is 5000. Decrease this value to create +# probes with lower overflow values. Setting this value too low on a rapidly +# increasing event could render a system unusable. +# +#dcpc-min-overflow=5000; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.c new file mode 100644 index 00000000..201ce497 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.c @@ -0,0 +1,16292 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved. + */ + +/* + * DTrace - Dynamic Tracing for Solaris + * + * This is the implementation of the Solaris Dynamic Tracing framework + * (DTrace). The user-visible interface to DTrace is described at length in + * the "Solaris Dynamic Tracing Guide". The interfaces between the libdtrace + * library, the in-kernel DTrace framework, and the DTrace providers are + * described in the block comments in the <sys/dtrace.h> header file. The + * internal architecture of DTrace is described in the block comments in the + * <sys/dtrace_impl.h> header file. The comments contained within the DTrace + * implementation very much assume mastery of all of these sources; if one has + * an unanswered question about the implementation, one should consult them + * first. + * + * The functions here are ordered roughly as follows: + * + * - Probe context functions + * - Probe hashing functions + * - Non-probe context utility functions + * - Matching functions + * - Provider-to-Framework API functions + * - Probe management functions + * - DIF object functions + * - Format functions + * - Predicate functions + * - ECB functions + * - Buffer functions + * - Enabling functions + * - DOF functions + * - Anonymous enabling functions + * - Consumer state functions + * - Helper functions + * - Hook functions + * - Driver cookbook functions + * + * Each group of functions begins with a block comment labelled the "DTrace + * [Group] Functions", allowing one to find each block by searching forward + * on capital-f functions. + */ +#ifndef VBOX +#include <sys/errno.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#include <sys/strsubr.h> +#include <sys/sysmacros.h> +#include <sys/dtrace_impl.h> +#include <sys/atomic.h> +#include <sys/cmn_err.h> +#include <sys/mutex_impl.h> +#include <sys/rwlock_impl.h> +#include <sys/ctf_api.h> +#include <sys/panic.h> +#include <sys/priv_impl.h> +#include <sys/policy.h> +#include <sys/cred_impl.h> +#include <sys/procfs_isa.h> +#include <sys/taskq.h> +#include <sys/mkdev.h> +#include <sys/kdi.h> +#include <sys/zone.h> +#include <sys/socket.h> +#include <netinet/in.h> + +#else /* VBOX */ +# include <sys/dtrace_impl.h> +# include <VBox/sup.h> +# include <iprt/assert.h> +# include <iprt/cpuset.h> +# include <iprt/err.h> +# include <iprt/mem.h> +# include <iprt/mp.h> +# include <iprt/string.h> +# include <iprt/process.h> +# include <iprt/thread.h> +# include <iprt/timer.h> +# include <limits.h> + +# undef offsetof +# define offsetof RT_OFFSETOF + +/* + * Use asm.h to implemente some of the simple stuff in dtrace_asm.s. + */ +# include <iprt/asm.h> +# include <iprt/asm-amd64-x86.h> +# define dtrace_casptr(a_ppvDst, a_pvOld, a_pvNew) \ + VBoxDtCompareAndSwapPtr((void * volatile *)a_ppvDst, a_pvOld, a_pvNew) +DECLINLINE(void *) VBoxDtCompareAndSwapPtr(void * volatile *ppvDst, void *pvOld, void *pvNew) +{ + void *pvRet; + ASMAtomicCmpXchgExPtrVoid(ppvDst, pvNew, pvOld, &pvRet); + return pvRet; +} + +# define dtrace_cas32(a_pu32Dst, a_pu32Old, a_pu32New) \ + VBoxDtCompareAndSwapU32(a_pu32Dst, a_pu32Old, a_pu32New) +DECLINLINE(uint32_t) VBoxDtCompareAndSwapU32(uint32_t volatile *pu32Dst, uint32_t u32Old, uint32_t u32New) +{ + uint32_t u32Ret; + ASMAtomicCmpXchgExU32(pu32Dst, u32New, u32Old, &u32Ret); + return u32Ret; +} + +#define dtrace_membar_consumer() ASMReadFence() +#define dtrace_membar_producer() ASMWriteFence() +#define dtrace_interrupt_disable() ASMIntDisableFlags() +#define dtrace_interrupt_enable(a_EFL) ASMSetFlags(a_EFL) + +/* + * NULL must be set to 0 or we'll end up with a billion warnings(=errors). + */ +# undef NULL +# define NULL (0) +#endif /* VBOX */ + +/** Check if the given address is a valid kernel address. + * The value can be uintptr_t or uint64_t. */ +#ifndef VBOX +# define VBDT_IS_VALID_KRNL_ADDR(a_uAddr) ((a_uAddr) >= KERNELBASE) +#else +# define VBDT_IS_VALID_KRNL_ADDR(a_uAddr) \ + ( (sizeof(a_uAddr) == sizeof(uintptr_t) || (uintptr_t)(a_uAddr) == (a_uAddr)) \ + && RTR0MemKernelIsValidAddr((void *)(uintptr_t)(a_uAddr)) ) +#endif + + +/* + * DTrace Tunable Variables + * + * The following variables may be tuned by adding a line to /etc/system that + * includes both the name of the DTrace module ("dtrace") and the name of the + * variable. For example: + * + * set dtrace:dtrace_destructive_disallow = 1 + * + * In general, the only variables that one should be tuning this way are those + * that affect system-wide DTrace behavior, and for which the default behavior + * is undesirable. Most of these variables are tunable on a per-consumer + * basis using DTrace options, and need not be tuned on a system-wide basis. + * When tuning these variables, avoid pathological values; while some attempt + * is made to verify the integrity of these variables, they are not considered + * part of the supported interface to DTrace, and they are therefore not + * checked comprehensively. Further, these variables should not be tuned + * dynamically via "mdb -kw" or other means; they should only be tuned via + * /etc/system. + */ +int dtrace_destructive_disallow = 0; +dtrace_optval_t dtrace_nonroot_maxsize = (16 * 1024 * 1024); +size_t dtrace_difo_maxsize = (256 * 1024); +dtrace_optval_t dtrace_dof_maxsize = (256 * 1024); +size_t dtrace_global_maxsize = (16 * 1024); +size_t dtrace_actions_max = (16 * 1024); +size_t dtrace_retain_max = 1024; +dtrace_optval_t dtrace_helper_actions_max = 32; +dtrace_optval_t dtrace_helper_providers_max = 32; +dtrace_optval_t dtrace_dstate_defsize = (1 * 1024 * 1024); +size_t dtrace_strsize_default = 256; +dtrace_optval_t dtrace_cleanrate_default = 9900990; /* 101 hz */ +dtrace_optval_t dtrace_cleanrate_min = 200000; /* 5000 hz */ +dtrace_optval_t dtrace_cleanrate_max = (uint64_t)60 * NANOSEC; /* 1/minute */ +dtrace_optval_t dtrace_aggrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_statusrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_statusrate_max = (hrtime_t)10 * NANOSEC; /* 6/minute */ +dtrace_optval_t dtrace_switchrate_default = NANOSEC; /* 1 hz */ +dtrace_optval_t dtrace_nspec_default = 1; +dtrace_optval_t dtrace_specsize_default = 32 * 1024; +dtrace_optval_t dtrace_stackframes_default = 20; +dtrace_optval_t dtrace_ustackframes_default = 20; +dtrace_optval_t dtrace_jstackframes_default = 50; +dtrace_optval_t dtrace_jstackstrsize_default = 512; +int dtrace_msgdsize_max = 128; +hrtime_t dtrace_chill_max = 500 * (NANOSEC / MILLISEC); /* 500 ms */ +hrtime_t dtrace_chill_interval = NANOSEC; /* 1000 ms */ +int dtrace_devdepth_max = 32; +int dtrace_err_verbose; +hrtime_t dtrace_deadman_interval = NANOSEC; +hrtime_t dtrace_deadman_timeout = (hrtime_t)10 * NANOSEC; +hrtime_t dtrace_deadman_user = (hrtime_t)30 * NANOSEC; + +/* + * DTrace External Variables + * + * As dtrace(7D) is a kernel module, any DTrace variables are obviously + * available to DTrace consumers via the backtick (`) syntax. One of these, + * dtrace_zero, is made deliberately so: it is provided as a source of + * well-known, zero-filled memory. While this variable is not documented, + * it is used by some translators as an implementation detail. + */ +const char dtrace_zero[256] = { 0 }; /* zero-filled memory */ + +/* + * DTrace Internal Variables + */ +#ifndef VBOX +static dev_info_t *dtrace_devi; /* device info */ +#endif +static vmem_t *dtrace_arena; /* probe ID arena */ +#ifndef VBOX +static vmem_t *dtrace_minor; /* minor number arena */ +static taskq_t *dtrace_taskq; /* task queue */ +#endif +static dtrace_probe_t **dtrace_probes; /* array of all probes */ +static VBDTTYPE(uint32_t,int) dtrace_nprobes; /* number of probes */ +static dtrace_provider_t *dtrace_provider; /* provider list */ +static dtrace_meta_t *dtrace_meta_pid; /* user-land meta provider */ +static int dtrace_opens; /* number of opens */ +static int dtrace_helpers; /* number of helpers */ +#ifndef VBOX +static void *dtrace_softstate; /* softstate pointer */ +#endif +static dtrace_hash_t *dtrace_bymod; /* probes hashed by module */ +static dtrace_hash_t *dtrace_byfunc; /* probes hashed by function */ +static dtrace_hash_t *dtrace_byname; /* probes hashed by name */ +static dtrace_toxrange_t *dtrace_toxrange; /* toxic range array */ +static int dtrace_toxranges; /* number of toxic ranges */ +static int dtrace_toxranges_max; /* size of toxic range array */ +static dtrace_anon_t dtrace_anon; /* anonymous enabling */ +static kmem_cache_t *dtrace_state_cache; /* cache for dynamic state */ +static uint64_t dtrace_vtime_references; /* number of vtimestamp refs */ +#ifndef VBOX +static kthread_t *dtrace_panicked; /* panicking thread */ +#endif +static dtrace_ecb_t *dtrace_ecb_create_cache; /* cached created ECB */ +static dtrace_genid_t dtrace_probegen; /* current probe generation */ +static dtrace_helpers_t *dtrace_deferred_pid; /* deferred helper list */ +static dtrace_enabling_t *dtrace_retained; /* list of retained enablings */ +static dtrace_genid_t dtrace_retained_gen; /* current retained enab gen */ +static dtrace_dynvar_t dtrace_dynhash_sink; /* end of dynamic hash chains */ +static int dtrace_dynvar_failclean; /* dynvars failed to clean */ + +/* + * DTrace Locking + * DTrace is protected by three (relatively coarse-grained) locks: + * + * (1) dtrace_lock is required to manipulate essentially any DTrace state, + * including enabling state, probes, ECBs, consumer state, helper state, + * etc. Importantly, dtrace_lock is _not_ required when in probe context; + * probe context is lock-free -- synchronization is handled via the + * dtrace_sync() cross call mechanism. + * + * (2) dtrace_provider_lock is required when manipulating provider state, or + * when provider state must be held constant. + * + * (3) dtrace_meta_lock is required when manipulating meta provider state, or + * when meta provider state must be held constant. + * + * The lock ordering between these three locks is dtrace_meta_lock before + * dtrace_provider_lock before dtrace_lock. (In particular, there are + * several places where dtrace_provider_lock is held by the framework as it + * calls into the providers -- which then call back into the framework, + * grabbing dtrace_lock.) + * + * There are two other locks in the mix: mod_lock and cpu_lock. With respect + * to dtrace_provider_lock and dtrace_lock, cpu_lock continues its historical + * role as a coarse-grained lock; it is acquired before both of these locks. + * With respect to dtrace_meta_lock, its behavior is stranger: cpu_lock must + * be acquired _between_ dtrace_meta_lock and any other DTrace locks. + * mod_lock is similar with respect to dtrace_provider_lock in that it must be + * acquired _between_ dtrace_provider_lock and dtrace_lock. + */ +static kmutex_t dtrace_lock; /* probe state lock */ +static kmutex_t dtrace_provider_lock; /* provider state lock */ +static kmutex_t dtrace_meta_lock; /* meta-provider state lock */ + +/* + * DTrace Provider Variables + * + * These are the variables relating to DTrace as a provider (that is, the + * provider of the BEGIN, END, and ERROR probes). + */ +static dtrace_pattr_t dtrace_provider_attr = { +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_STABLE, DTRACE_STABILITY_STABLE, DTRACE_CLASS_COMMON }, +}; + +static void +dtrace_nullop(void) +{} + +static int +dtrace_enable_nullop(void) +{ + return (0); +} + +static dtrace_pops_t dtrace_provider_ops = { + (void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop, + (void (*)(void *, struct modctl *))dtrace_nullop, + (int (*)(void *, dtrace_id_t, void *))(uintptr_t)dtrace_enable_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop, + NULL, + NULL, + NULL, + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop +}; + +static dtrace_id_t dtrace_probeid_begin; /* special BEGIN probe */ +static dtrace_id_t dtrace_probeid_end; /* special END probe */ +dtrace_id_t dtrace_probeid_error; /* special ERROR probe */ + +/* + * DTrace Helper Tracing Variables + */ +uint32_t dtrace_helptrace_next = 0; +uint32_t dtrace_helptrace_nlocals; +char *dtrace_helptrace_buffer; +int dtrace_helptrace_bufsize = 512 * 1024; + +#ifdef DEBUG +int dtrace_helptrace_enabled = 1; +#else +int dtrace_helptrace_enabled = 0; +#endif + +/* + * DTrace Error Hashing + * + * On DEBUG kernels, DTrace will track the errors that has seen in a hash + * table. This is very useful for checking coverage of tests that are + * expected to induce DIF or DOF processing errors, and may be useful for + * debugging problems in the DIF code generator or in DOF generation . The + * error hash may be examined with the ::dtrace_errhash MDB dcmd. + */ +#ifdef DEBUG +static dtrace_errhash_t dtrace_errhash[DTRACE_ERRHASHSZ]; +static const char *dtrace_errlast; +static kthread_t *dtrace_errthread; +static kmutex_t dtrace_errlock; +#endif + +/* + * DTrace Macros and Constants + * + * These are various macros that are useful in various spots in the + * implementation, along with a few random constants that have no meaning + * outside of the implementation. There is no real structure to this cpp + * mishmash -- but is there ever? + */ +#define DTRACE_HASHSTR(hash, probe) \ + dtrace_hash_str(*((char **)((uintptr_t)(probe) + (hash)->dth_stroffs))) + +#define DTRACE_HASHNEXT(hash, probe) \ + (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_nextoffs) + +#define DTRACE_HASHPREV(hash, probe) \ + (dtrace_probe_t **)((uintptr_t)(probe) + (hash)->dth_prevoffs) + +#define DTRACE_HASHEQ(hash, lhs, rhs) \ + (strcmp(*((char **)((uintptr_t)(lhs) + (hash)->dth_stroffs)), \ + *((char **)((uintptr_t)(rhs) + (hash)->dth_stroffs))) == 0) + +#define DTRACE_AGGHASHSIZE_SLEW 17 + +#define DTRACE_V4MAPPED_OFFSET (sizeof (uint32_t) * 3) + +/* + * The key for a thread-local variable consists of the lower 61 bits of the + * t_did, plus the 3 bits of the highest active interrupt above LOCK_LEVEL. + * We add DIF_VARIABLE_MAX to t_did to assure that the thread key is never + * equal to a variable identifier. This is necessary (but not sufficient) to + * assure that global associative arrays never collide with thread-local + * variables. To guarantee that they cannot collide, we must also define the + * order for keying dynamic variables. That order is: + * + * [ key0 ] ... [ keyn ] [ variable-key ] [ tls-key ] + * + * Because the variable-key and the tls-key are in orthogonal spaces, there is + * no way for a global variable key signature to match a thread-local key + * signature. + */ +#ifndef VBOX +#define DTRACE_TLS_THRKEY(where) { \ + uint_t intr = 0; \ + uint_t actv = CPU->cpu_intr_actv >> (LOCK_LEVEL + 1); \ + for (; actv; actv >>= 1) \ + intr++; \ + ASSERT(intr < (1 << 3)); \ + (where) = ((curthread->t_did + DIF_VARIABLE_MAX) & \ + (((uint64_t)1 << 61) - 1)) | ((uint64_t)intr << 61); \ +} +#else +#define DTRACE_TLS_THRKEY(where) do { \ + (where) = (((uintptr_t)RTThreadNativeSelf() + DIF_VARIABLE_MAX) & (RT_BIT_64(61) - 1)) \ + | (RTThreadIsInInterrupt(NIL_RTTHREAD) ? RT_BIT_64(61) : 0); \ +} while (0) +#endif + +#define DT_BSWAP_8(x) ((x) & 0xff) +#define DT_BSWAP_16(x) ((DT_BSWAP_8(x) << 8) | DT_BSWAP_8((x) >> 8)) +#define DT_BSWAP_32(x) ((DT_BSWAP_16(x) << 16) | DT_BSWAP_16((x) >> 16)) +#define DT_BSWAP_64(x) ((DT_BSWAP_32(x) << 32) | DT_BSWAP_32((x) >> 32)) + +#define DT_MASK_LO 0x00000000FFFFFFFFULL + +#define DTRACE_STORE(type, tomax, offset, what) \ + *((type *)((uintptr_t)(tomax) + (uintptr_t)offset)) = (type)(what); + +#ifndef __i386 +#define DTRACE_ALIGNCHECK(addr, size, flags) \ + if (addr & (size - 1)) { \ + *flags |= CPU_DTRACE_BADALIGN; \ + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval = addr; \ + return (0); \ + } +#else +#define DTRACE_ALIGNCHECK(addr, size, flags) +#endif + +/* + * Test whether a range of memory starting at testaddr of size testsz falls + * within the range of memory described by addr, sz. We take care to avoid + * problems with overflow and underflow of the unsigned quantities, and + * disallow all negative sizes. Ranges of size 0 are allowed. + */ +#define DTRACE_INRANGE(testaddr, testsz, baseaddr, basesz) \ + ((testaddr) - (baseaddr) < (basesz) && \ + (testaddr) + (testsz) - (baseaddr) <= (basesz) && \ + (testaddr) + (testsz) >= (testaddr)) + +/* + * Test whether alloc_sz bytes will fit in the scratch region. We isolate + * alloc_sz on the righthand side of the comparison in order to avoid overflow + * or underflow in the comparison with it. This is simpler than the INRANGE + * check above, because we know that the dtms_scratch_ptr is valid in the + * range. Allocations of size zero are allowed. + */ +#define DTRACE_INSCRATCH(mstate, alloc_sz) \ + ((mstate)->dtms_scratch_base + (mstate)->dtms_scratch_size - \ + (mstate)->dtms_scratch_ptr >= (alloc_sz)) + +#ifndef VBOX +#define DTRACE_LOADFUNC(bits) \ +/*CSTYLED*/ \ +VBDTSTATIC uint##bits##_t \ +dtrace_load##bits(uintptr_t addr) \ +{ \ + size_t size = bits / NBBY; \ + /*CSTYLED*/ \ + uint##bits##_t rval; \ + int i; \ + processorid_t me = VBDT_GET_CPUID(); \ + volatile uint16_t *flags = (volatile uint16_t *) \ + &cpu_core[me].cpuc_dtrace_flags; \ + \ + DTRACE_ALIGNCHECK(addr, size, flags); \ + \ + for (i = 0; i < dtrace_toxranges; i++) { \ + if (addr >= dtrace_toxrange[i].dtt_limit) \ + continue; \ + \ + if (addr + size <= dtrace_toxrange[i].dtt_base) \ + continue; \ + \ + /* \ + * This address falls within a toxic region; return 0. \ + */ \ + *flags |= CPU_DTRACE_BADADDR; \ + cpu_core[me].cpuc_dtrace_illval = addr; \ + return (0); \ + } \ + \ + *flags |= CPU_DTRACE_NOFAULT; \ + /*CSTYLED*/ \ + rval = *((volatile uint##bits##_t *)addr); \ + *flags &= ~CPU_DTRACE_NOFAULT; \ + \ + return (!(*flags & CPU_DTRACE_FAULT) ? rval : 0); \ +} +#else /* VBOX */ +# define DTRACE_LOADFUNC(bits) \ +VBDTSTATIC uint##bits##_t \ +dtrace_load##bits(uintptr_t addr) \ +{ \ + size_t const size = bits / NBBY; \ + uint##bits##_t rval; \ + processorid_t me; \ + int i, rc; \ + \ + /*DTRACE_ALIGNCHECK(addr, size, flags);*/ \ + \ + for (i = 0; i < dtrace_toxranges; i++) { \ + if (addr >= dtrace_toxrange[i].dtt_limit) \ + continue; \ + \ + if (addr + size <= dtrace_toxrange[i].dtt_base) \ + continue; \ + \ + /* \ + * This address falls within a toxic region; return 0. \ + */ \ + me = VBDT_GET_CPUID(); \ + cpu_core[me].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; \ + cpu_core[me].cpuc_dtrace_illval = addr; \ + return (0); \ + } \ + \ + rc = RTR0MemKernelCopyFrom(&rval, (void const *)addr, size); \ + if (RT_SUCCESS(rc)) \ + return rval; \ + \ + /* \ + * If not supported, pray it won't fault... \ + */ \ + if (rc == VERR_NOT_SUPPORTED) \ + return *(uint##bits##_t const *)addr; \ + \ + me = VBDT_GET_CPUID(); \ + cpu_core[me].cpuc_dtrace_flags |= CPU_DTRACE_BADADDR; \ + cpu_core[me].cpuc_dtrace_illval = addr; \ + return (0); \ +} + +#endif /* VBOX */ + +#ifdef _LP64 +#define dtrace_loadptr dtrace_load64 +#else +#define dtrace_loadptr dtrace_load32 +#endif + +#define DTRACE_DYNHASH_FREE 0 +#define DTRACE_DYNHASH_SINK 1 +#define DTRACE_DYNHASH_VALID 2 + +#define DTRACE_MATCH_FAIL -1 +#define DTRACE_MATCH_NEXT 0 +#define DTRACE_MATCH_DONE 1 +#define DTRACE_ANCHORED(probe) ((probe)->dtpr_func[0] != '\0') +#define DTRACE_STATE_ALIGN 64 + +#define DTRACE_FLAGS2FLT(flags) \ + (((flags) & CPU_DTRACE_BADADDR) ? DTRACEFLT_BADADDR : \ + ((flags) & CPU_DTRACE_ILLOP) ? DTRACEFLT_ILLOP : \ + ((flags) & CPU_DTRACE_DIVZERO) ? DTRACEFLT_DIVZERO : \ + ((flags) & CPU_DTRACE_KPRIV) ? DTRACEFLT_KPRIV : \ + ((flags) & CPU_DTRACE_UPRIV) ? DTRACEFLT_UPRIV : \ + ((flags) & CPU_DTRACE_TUPOFLOW) ? DTRACEFLT_TUPOFLOW : \ + ((flags) & CPU_DTRACE_BADALIGN) ? DTRACEFLT_BADALIGN : \ + ((flags) & CPU_DTRACE_NOSCRATCH) ? DTRACEFLT_NOSCRATCH : \ + ((flags) & CPU_DTRACE_BADSTACK) ? DTRACEFLT_BADSTACK : \ + DTRACEFLT_UNKNOWN) + +#define DTRACEACT_ISSTRING(act) \ + ((act)->dta_kind == DTRACEACT_DIFEXPR && \ + (act)->dta_difo->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) + +static size_t dtrace_strlen(const char *, size_t); +static dtrace_probe_t *dtrace_probe_lookup_id(dtrace_id_t id); +static void dtrace_enabling_provide(dtrace_provider_t *); +static int dtrace_enabling_match(dtrace_enabling_t *, int *); +static void dtrace_enabling_matchall(void); +static dtrace_state_t *dtrace_anon_grab(void); +#ifndef VBOX +static uint64_t dtrace_helper(int, dtrace_mstate_t *, + dtrace_state_t *, uint64_t, uint64_t); +static dtrace_helpers_t *dtrace_helpers_create(proc_t *); +#endif +static void dtrace_buffer_drop(dtrace_buffer_t *); +static intptr_t dtrace_buffer_reserve(dtrace_buffer_t *, size_t, size_t, + dtrace_state_t *, dtrace_mstate_t *); +static int dtrace_state_option(dtrace_state_t *, dtrace_optid_t, + dtrace_optval_t); +static int dtrace_ecb_create_enable(dtrace_probe_t *, void *); +#ifndef VBOX +static void dtrace_helper_provider_destroy(dtrace_helper_provider_t *); +#endif + +/* + * DTrace Probe Context Functions + * + * These functions are called from probe context. Because probe context is + * any context in which C may be called, arbitrarily locks may be held, + * interrupts may be disabled, we may be in arbitrary dispatched state, etc. + * As a result, functions called from probe context may only call other DTrace + * support functions -- they may not interact at all with the system at large. + * (Note that the ASSERT macro is made probe-context safe by redefining it in + * terms of dtrace_assfail(), a probe-context safe function.) If arbitrary + * loads are to be performed from probe context, they _must_ be in terms of + * the safe dtrace_load*() variants. + * + * Some functions in this block are not actually called from probe context; + * for these functions, there will be a comment above the function reading + * "Note: not called from probe context." + */ +void +dtrace_panic(const char *format, ...) +{ + va_list alist; + + va_start(alist, format); + dtrace_vpanic(format, alist); + va_end(alist); +} + +#ifndef VBOX /* We have our own assertion machinery. */ +int +dtrace_assfail(const char *a, const char *f, int l) +{ + dtrace_panic("assertion failed: %s, file: %s, line: %d", a, f, l); + + /* + * We just need something here that even the most clever compiler + * cannot optimize away. + */ + return (a[(uintptr_t)f]); +} +#endif + +/* + * Atomically increment a specified error counter from probe context. + */ +static void +dtrace_error(uint32_t *counter) +{ + /* + * Most counters stored to in probe context are per-CPU counters. + * However, there are some error conditions that are sufficiently + * arcane that they don't merit per-CPU storage. If these counters + * are incremented concurrently on different CPUs, scalability will be + * adversely affected -- but we don't expect them to be white-hot in a + * correctly constructed enabling... + */ + uint32_t oval, nval; + + do { + oval = *counter; + + if ((nval = oval + 1) == 0) { + /* + * If the counter would wrap, set it to 1 -- assuring + * that the counter is never zero when we have seen + * errors. (The counter must be 32-bits because we + * aren't guaranteed a 64-bit compare&swap operation.) + * To save this code both the infamy of being fingered + * by a priggish news story and the indignity of being + * the target of a neo-puritan witch trial, we're + * carefully avoiding any colorful description of the + * likelihood of this condition -- but suffice it to + * say that it is only slightly more likely than the + * overflow of predicate cache IDs, as discussed in + * dtrace_predicate_create(). + */ + nval = 1; + } + } while (dtrace_cas32(counter, oval, nval) != oval); +} + +/* + * Use the DTRACE_LOADFUNC macro to define functions for each of loading a + * uint8_t, a uint16_t, a uint32_t and a uint64_t. + */ +DTRACE_LOADFUNC(8) +DTRACE_LOADFUNC(16) +DTRACE_LOADFUNC(32) +DTRACE_LOADFUNC(64) + +static int +dtrace_inscratch(uintptr_t dest, size_t size, dtrace_mstate_t *mstate) +{ + if (dest < mstate->dtms_scratch_base) + return (0); + + if (dest + size < dest) + return (0); + + if (dest + size > mstate->dtms_scratch_ptr) + return (0); + + return (1); +} + +static int +dtrace_canstore_statvar(uint64_t addr, size_t sz, + dtrace_statvar_t **svars, int nsvars) +{ + int i; + + for (i = 0; i < nsvars; i++) { + dtrace_statvar_t *svar = svars[i]; + + if (svar == NULL || svar->dtsv_size == 0) + continue; + + if (DTRACE_INRANGE(addr, sz, svar->dtsv_data, svar->dtsv_size)) + return (1); + } + + return (0); +} + +/* + * Check to see if the address is within a memory region to which a store may + * be issued. This includes the DTrace scratch areas, and any DTrace variable + * region. The caller of dtrace_canstore() is responsible for performing any + * alignment checks that are needed before stores are actually executed. + */ +static int +dtrace_canstore(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + /* + * First, check to see if the address is in scratch space... + */ + if (DTRACE_INRANGE(addr, sz, mstate->dtms_scratch_base, + mstate->dtms_scratch_size)) + return (1); + + /* + * Now check to see if it's a dynamic variable. This check will pick + * up both thread-local variables and any global dynamically-allocated + * variables. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)vstate->dtvs_dynvars.dtds_base, + vstate->dtvs_dynvars.dtds_size)) { + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + uintptr_t base = (uintptr_t)dstate->dtds_base + + (dstate->dtds_hashsize * sizeof (dtrace_dynhash_t)); + uintptr_t chunkoffs; + + /* + * Before we assume that we can store here, we need to make + * sure that it isn't in our metadata -- storing to our + * dynamic variable metadata would corrupt our state. For + * the range to not include any dynamic variable metadata, + * it must: + * + * (1) Start above the hash table that is at the base of + * the dynamic variable space + * + * (2) Have a starting chunk offset that is beyond the + * dtrace_dynvar_t that is at the base of every chunk + * + * (3) Not span a chunk boundary + * + */ + if (addr < base) + return (0); + + chunkoffs = (addr - base) % dstate->dtds_chunksize; + + if (chunkoffs < sizeof (dtrace_dynvar_t)) + return (0); + + if (chunkoffs + sz > dstate->dtds_chunksize) + return (0); + + return (1); + } + + /* + * Finally, check the static local and global variables. These checks + * take the longest, so we perform them last. + */ + if (dtrace_canstore_statvar(addr, sz, + vstate->dtvs_locals, vstate->dtvs_nlocals)) + return (1); + + if (dtrace_canstore_statvar(addr, sz, + vstate->dtvs_globals, vstate->dtvs_nglobals)) + return (1); + + return (0); +} + + +/* + * Convenience routine to check to see if the address is within a memory + * region in which a load may be issued given the user's privilege level; + * if not, it sets the appropriate error flags and loads 'addr' into the + * illegal value slot. + * + * DTrace subroutines (DIF_SUBR_*) should use this helper to implement + * appropriate memory access protection. + */ +static int +dtrace_canload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + volatile uintptr_t *illval = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + /* + * You can obviously read that which you can store. + */ + if (dtrace_canstore(addr, sz, mstate, vstate)) + return (1); + + /* + * We're allowed to read from our own string table. + */ + if (DTRACE_INRANGE(addr, sz, (uintptr_t)mstate->dtms_difo->dtdo_strtab, + mstate->dtms_difo->dtdo_strlen)) + return (1); + + DTRACE_CPUFLAG_SET(CPU_DTRACE_KPRIV); + *illval = addr; + return (0); +} + +/* + * Convenience routine to check to see if a given string is within a memory + * region in which a load may be issued given the user's privilege level; + * this exists so that we don't need to issue unnecessary dtrace_strlen() + * calls in the event that the user has all privileges. + */ +static int +dtrace_strcanload(uint64_t addr, size_t sz, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t strsz; + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + strsz = 1 + dtrace_strlen((char *)(uintptr_t)addr, sz); + if (dtrace_canload(addr, strsz, mstate, vstate)) + return (1); + + return (0); +} + +/* + * Convenience routine to check to see if a given variable is within a memory + * region in which a load may be issued given the user's privilege level. + */ +static int +dtrace_vcanload(void *src, dtrace_diftype_t *type, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate) +{ + size_t sz; + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + /* + * If we hold the privilege to read from kernel memory, then + * everything is readable. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (1); + + if (type->dtdt_kind == DIF_TYPE_STRING) + sz = dtrace_strlen(src, + vstate->dtvs_state->dts_options[DTRACEOPT_STRSIZE]) + 1; + else + sz = type->dtdt_size; + + return (dtrace_canload((uintptr_t)src, sz, mstate, vstate)); +} + +/* + * Compare two strings using safe loads. + */ +static int +dtrace_strncmp(char *s1, char *s2, size_t limit) +{ + uint8_t c1, c2; + volatile uint16_t *flags; + + if (s1 == s2 || limit == 0) + return (0); + + flags = (volatile uint16_t *)&cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + + do { + if (s1 == NULL) { + c1 = '\0'; + } else { + c1 = dtrace_load8((uintptr_t)s1++); + } + + if (s2 == NULL) { + c2 = '\0'; + } else { + c2 = dtrace_load8((uintptr_t)s2++); + } + + if (c1 != c2) + return (c1 - c2); + } while (--limit && c1 != '\0' && !(*flags & CPU_DTRACE_FAULT)); + + return (0); +} + +/* + * Compute strlen(s) for a string using safe memory accesses. The additional + * len parameter is used to specify a maximum length to ensure completion. + */ +static size_t +dtrace_strlen(const char *s, size_t lim) +{ + uint_t len; + + for (len = 0; len != lim; len++) { + if (dtrace_load8((uintptr_t)s++) == '\0') + break; + } + + return (len); +} + +/* + * Check if an address falls within a toxic region. + */ +static int +dtrace_istoxic(uintptr_t kaddr, size_t size) +{ + uintptr_t taddr, tsize; + int i; + + for (i = 0; i < dtrace_toxranges; i++) { + taddr = dtrace_toxrange[i].dtt_base; + tsize = dtrace_toxrange[i].dtt_limit - taddr; + + if (kaddr - taddr < tsize) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval = kaddr; + return (1); + } + + if (taddr - kaddr < size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval = taddr; + return (1); + } + } + + return (0); +} + +/* + * Copy src to dst using safe memory accesses. The src is assumed to be unsafe + * memory specified by the DIF program. The dst is assumed to be safe memory + * that we can store to directly because it is managed by DTrace. As with + * standard bcopy, overlapping copies are handled properly. + */ +static void +dtrace_bcopy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst; + const uint8_t *s2 = src; + + if (s1 <= s2) { + do { + *s1++ = dtrace_load8((uintptr_t)s2++); + } while (--len != 0); + } else { + s2 += len; + s1 += len; + + do { + *--s1 = dtrace_load8((uintptr_t)--s2); + } while (--len != 0); + } + } +} + +/* + * Copy src to dst using safe memory accesses, up to either the specified + * length, or the point that a nul byte is encountered. The src is assumed to + * be unsafe memory specified by the DIF program. The dst is assumed to be + * safe memory that we can store to directly because it is managed by DTrace. + * Unlike dtrace_bcopy(), overlapping regions are not handled. + */ +static void +dtrace_strcpy(const void *src, void *dst, size_t len) +{ + if (len != 0) { + uint8_t *s1 = dst, c; + const uint8_t *s2 = src; + + do { + *s1++ = c = dtrace_load8((uintptr_t)s2++); + } while (--len != 0 && c != '\0'); + } +} + +/* + * Copy src to dst, deriving the size and type from the specified (BYREF) + * variable type. The src is assumed to be unsafe memory specified by the DIF + * program. The dst is assumed to be DTrace variable memory that is of the + * specified type; we assume that we can store to directly. + */ +static void +dtrace_vcopy(void *src, void *dst, dtrace_diftype_t *type) +{ + ASSERT(type->dtdt_flags & DIF_TF_BYREF); + + if (type->dtdt_kind == DIF_TYPE_STRING) { + dtrace_strcpy(src, dst, type->dtdt_size); + } else { + dtrace_bcopy(src, dst, type->dtdt_size); + } +} + +/* + * Compare s1 to s2 using safe memory accesses. The s1 data is assumed to be + * unsafe memory specified by the DIF program. The s2 data is assumed to be + * safe memory that we can access directly because it is managed by DTrace. + */ +static int +dtrace_bcmp(const void *s1, const void *s2, size_t len) +{ + volatile uint16_t *flags; + + flags = (volatile uint16_t *)&cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + + if (s1 == s2) + return (0); + + if (s1 == NULL || s2 == NULL) + return (1); + + if (s1 != s2 && len != 0) { + const uint8_t *ps1 = s1; + const uint8_t *ps2 = s2; + + do { + if (dtrace_load8((uintptr_t)ps1++) != *ps2++) + return (1); + } while (--len != 0 && !(*flags & CPU_DTRACE_FAULT)); + } + return (0); +} + +/* + * Zero the specified region using a simple byte-by-byte loop. Note that this + * is for safe DTrace-managed memory only. + */ +static void +dtrace_bzero(void *dst, size_t len) +{ + uchar_t *cp; + + for (cp = dst; len != 0; len--) + *cp++ = 0; +} + +static void +dtrace_add_128(uint64_t *addend1, uint64_t *addend2, uint64_t *sum) +{ + uint64_t result[2]; + + result[0] = addend1[0] + addend2[0]; + result[1] = addend1[1] + addend2[1] + + (result[0] < addend1[0] || result[0] < addend2[0] ? 1 : 0); + + sum[0] = result[0]; + sum[1] = result[1]; +} + +/* + * Shift the 128-bit value in a by b. If b is positive, shift left. + * If b is negative, shift right. + */ +static void +dtrace_shift_128(uint64_t *a, int b) +{ + uint64_t mask; + + if (b == 0) + return; + + if (b < 0) { + b = -b; + if (b >= 64) { + a[0] = a[1] >> (b - 64); + a[1] = 0; + } else { + a[0] >>= b; + mask = 1LL << (64 - b); + mask -= 1; + a[0] |= ((a[1] & mask) << (64 - b)); + a[1] >>= b; + } + } else { + if (b >= 64) { + a[1] = a[0] << (b - 64); + a[0] = 0; + } else { + a[1] <<= b; + mask = a[0] >> (64 - b); + a[1] |= mask; + a[0] <<= b; + } + } +} + +/* + * The basic idea is to break the 2 64-bit values into 4 32-bit values, + * use native multiplication on those, and then re-combine into the + * resulting 128-bit value. + * + * (hi1 << 32 + lo1) * (hi2 << 32 + lo2) = + * hi1 * hi2 << 64 + + * hi1 * lo2 << 32 + + * hi2 * lo1 << 32 + + * lo1 * lo2 + */ +static void +dtrace_multiply_128(uint64_t factor1, uint64_t factor2, uint64_t *product) +{ + uint64_t hi1, hi2, lo1, lo2; + uint64_t tmp[2]; + + hi1 = factor1 >> 32; + hi2 = factor2 >> 32; + + lo1 = factor1 & DT_MASK_LO; + lo2 = factor2 & DT_MASK_LO; + + product[0] = lo1 * lo2; + product[1] = hi1 * hi2; + + tmp[0] = hi1 * lo2; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); + + tmp[0] = hi2 * lo1; + tmp[1] = 0; + dtrace_shift_128(tmp, 32); + dtrace_add_128(product, tmp, product); +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the user credentials of the process that enabled the + * invoking ECB match the target credentials + */ +static int +dtrace_priv_proc_common_user(dtrace_state_t *state) +{ + cred_t *cr, *s_cr = state->dts_cred.dcr_cred; + + /* + * We should always have a non-NULL state cred here, since if cred + * is null (anonymous tracing), we fast-path bypass this routine. + */ + ASSERT(s_cr != NULL); + + if ((cr = CRED()) != NULL && + s_cr->cr_uid == cr->cr_uid && + s_cr->cr_uid == cr->cr_ruid && + s_cr->cr_uid == cr->cr_suid && + s_cr->cr_gid == cr->cr_gid && + s_cr->cr_gid == cr->cr_rgid && + s_cr->cr_gid == cr->cr_sgid) + return (1); + + return (0); +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the zone of the process that enabled the invoking ECB + * matches the target credentials + */ +static int +dtrace_priv_proc_common_zone(dtrace_state_t *state) +{ + cred_t *cr, *s_cr = state->dts_cred.dcr_cred; + + /* + * We should always have a non-NULL state cred here, since if cred + * is null (anonymous tracing), we fast-path bypass this routine. + */ + ASSERT(s_cr != NULL); + + if ((cr = CRED()) != NULL && + s_cr->cr_zone == cr->cr_zone) + return (1); + + return (0); +} + +/* + * This privilege check should be used by actions and subroutines to + * verify that the process has not setuid or changed credentials. + */ +static int +dtrace_priv_proc_common_nocd(VBDTVOID) +{ +#ifndef VBOX + proc_t *proc; + + if ((proc = VBDT_GET_PROC()) != NULL && + !(proc->p_flag & SNOCD)) + return (1); + + return (0); +#else + return (1); +#endif +} + +#ifndef VBOX +static int +dtrace_priv_proc_destructive(dtrace_state_t *state) +{ + int action = state->dts_cred.dcr_action; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE) == 0) && + dtrace_priv_proc_common_zone(state) == 0) + goto bad; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER) == 0) && + dtrace_priv_proc_common_user(state) == 0) + goto bad; + + if (((action & DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG) == 0) && + dtrace_priv_proc_common_nocd() == 0) + goto bad; + + return (1); + +bad: + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} +#endif /* !VBOX */ + +static int +dtrace_priv_proc_control(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC_CONTROL) + return (1); + + if (dtrace_priv_proc_common_zone(state) && + dtrace_priv_proc_common_user(state) && + dtrace_priv_proc_common_nocd()) + return (1); + + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} + +static int +dtrace_priv_proc(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_PROC) + return (1); + + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_UPRIV; + + return (0); +} + +static int +dtrace_priv_kernel(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL) + return (1); + + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; + + return (0); +} + +static int +dtrace_priv_kernel_destructive(dtrace_state_t *state) +{ + if (state->dts_cred.dcr_action & DTRACE_CRA_KERNEL_DESTRUCTIVE) + return (1); + + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_KPRIV; + + return (0); +} + +/* + * Note: not called from probe context. This function is called + * asynchronously (and at a regular interval) from outside of probe context to + * clean the dirty dynamic variable lists on all CPUs. Dynamic variable + * cleaning is explained in detail in <sys/dtrace_impl.h>. + */ +VBDTSTATIC void +dtrace_dynvar_clean(dtrace_dstate_t *dstate) +{ + dtrace_dynvar_t *dirty; + dtrace_dstate_percpu_t *dcpu; + dtrace_dynvar_t **rinsep; + int i, j, work = 0; + + for (i = 0; i < NCPU; i++) { + dcpu = &dstate->dtds_percpu[i]; + rinsep = &dcpu->dtdsc_rinsing; + + /* + * If the dirty list is NULL, there is no dirty work to do. + */ + if (dcpu->dtdsc_dirty == NULL) + continue; + + if (dcpu->dtdsc_rinsing != NULL) { + /* + * If the rinsing list is non-NULL, then it is because + * this CPU was selected to accept another CPU's + * dirty list -- and since that time, dirty buffers + * have accumulated. This is a highly unlikely + * condition, but we choose to ignore the dirty + * buffers -- they'll be picked up a future cleanse. + */ + continue; + } + + if (dcpu->dtdsc_clean != NULL) { + /* + * If the clean list is non-NULL, then we're in a + * situation where a CPU has done deallocations (we + * have a non-NULL dirty list) but no allocations (we + * also have a non-NULL clean list). We can't simply + * move the dirty list into the clean list on this + * CPU, yet we also don't want to allow this condition + * to persist, lest a short clean list prevent a + * massive dirty list from being cleaned (which in + * turn could lead to otherwise avoidable dynamic + * drops). To deal with this, we look for some CPU + * with a NULL clean list, NULL dirty list, and NULL + * rinsing list -- and then we borrow this CPU to + * rinse our dirty list. + */ + for (j = 0; j < NCPU; j++) { + dtrace_dstate_percpu_t *rinser; + + rinser = &dstate->dtds_percpu[j]; + + if (rinser->dtdsc_rinsing != NULL) + continue; + + if (rinser->dtdsc_dirty != NULL) + continue; + + if (rinser->dtdsc_clean != NULL) + continue; + + rinsep = &rinser->dtdsc_rinsing; + break; + } + + if (j == NCPU) { + /* + * We were unable to find another CPU that + * could accept this dirty list -- we are + * therefore unable to clean it now. + */ + dtrace_dynvar_failclean++; + continue; + } + } + + work = 1; + + /* + * Atomically move the dirty list aside. + */ + do { + dirty = dcpu->dtdsc_dirty; + + /* + * Before we zap the dirty list, set the rinsing list. + * (This allows for a potential assertion in + * dtrace_dynvar(): if a free dynamic variable appears + * on a hash chain, either the dirty list or the + * rinsing list for some CPU must be non-NULL.) + */ + *rinsep = dirty; + dtrace_membar_producer(); + } while (dtrace_casptr(&dcpu->dtdsc_dirty, + dirty, NULL) != dirty); + } + + if (!work) { + /* + * We have no work to do; we can simply return. + */ + return; + } + + dtrace_sync(); + + for (i = 0; i < NCPU; i++) { + dcpu = &dstate->dtds_percpu[i]; + + if (dcpu->dtdsc_rinsing == NULL) + continue; + + /* + * We are now guaranteed that no hash chain contains a pointer + * into this dirty list; we can make it clean. + */ + ASSERT(dcpu->dtdsc_clean == NULL); + dcpu->dtdsc_clean = dcpu->dtdsc_rinsing; + dcpu->dtdsc_rinsing = NULL; + } + + /* + * Before we actually set the state to be DTRACE_DSTATE_CLEAN, make + * sure that all CPUs have seen all of the dtdsc_clean pointers. + * This prevents a race whereby a CPU incorrectly decides that + * the state should be something other than DTRACE_DSTATE_CLEAN + * after dtrace_dynvar_clean() has completed. + */ + dtrace_sync(); + + dstate->dtds_state = DTRACE_DSTATE_CLEAN; +} + +/* + * Depending on the value of the op parameter, this function looks-up, + * allocates or deallocates an arbitrarily-keyed dynamic variable. If an + * allocation is requested, this function will return a pointer to a + * dtrace_dynvar_t corresponding to the allocated variable -- or NULL if no + * variable can be allocated. If NULL is returned, the appropriate counter + * will be incremented. + */ +VBDTSTATIC dtrace_dynvar_t * +dtrace_dynvar(dtrace_dstate_t *dstate, uint_t nkeys, + dtrace_key_t *key, size_t dsize, dtrace_dynvar_op_t op, + dtrace_mstate_t *mstate, dtrace_vstate_t *vstate) +{ + uint64_t hashval = DTRACE_DYNHASH_VALID; + dtrace_dynhash_t *hash = dstate->dtds_hash; + dtrace_dynvar_t *free, *new_free, *next, *dvar, *start, *prev = NULL; + processorid_t me = VBDT_GET_CPUID(), cpu = me; + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[me]; + size_t bucket, ksize; + size_t chunksize = dstate->dtds_chunksize; + uintptr_t kdata, lock, nstate; + uint_t i; + + ASSERT(nkeys != 0); + + /* + * Hash the key. As with aggregations, we use Jenkins' "One-at-a-time" + * algorithm. For the by-value portions, we perform the algorithm in + * 16-bit chunks (as opposed to 8-bit chunks). This speeds things up a + * bit, and seems to have only a minute effect on distribution. For + * the by-reference data, we perform "One-at-a-time" iterating (safely) + * over each referenced byte. It's painful to do this, but it's much + * better than pathological hash distribution. The efficacy of the + * hashing algorithm (and a comparison with other algorithms) may be + * found by running the ::dtrace_dynstat MDB dcmd. + */ + for (i = 0; i < nkeys; i++) { + if (key[i].dttk_size == 0) { + uint64_t val = key[i].dttk_value; + + hashval += (val >> 48) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 32) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += (val >> 16) & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + hashval += val & 0xffff; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } else { + /* + * This is incredibly painful, but it beats the hell + * out of the alternative. + */ + uint64_t j, size = key[i].dttk_size; + uintptr_t base = (uintptr_t)key[i].dttk_value; + + if (!dtrace_canload(base, size, mstate, vstate)) + break; + + for (j = 0; j < size; j++) { + hashval += dtrace_load8(base + j); + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + } + } + } + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return (NULL); + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * There is a remote chance (ideally, 1 in 2^31) that our hashval + * comes out to be one of our two sentinel hash values. If this + * actually happens, we set the hashval to be a value known to be a + * non-sentinel value. + */ + if (hashval == DTRACE_DYNHASH_FREE || hashval == DTRACE_DYNHASH_SINK) + hashval = DTRACE_DYNHASH_VALID; + + /* + * Yes, it's painful to do a divide here. If the cycle count becomes + * important here, tricks can be pulled to reduce it. (However, it's + * critical that hash collisions be kept to an absolute minimum; + * they're much more painful than a divide.) It's better to have a + * solution that generates few collisions and still keeps things + * relatively simple. + */ + bucket = hashval % dstate->dtds_hashsize; + + if (op == DTRACE_DYNVAR_DEALLOC) { + volatile uintptr_t *lockp = &hash[bucket].dtdh_lock; + + for (;;) { + while ((lock = *lockp) & 1) + continue; + + if (dtrace_casptr((void *)lockp, + (void *)lock, (void *)(lock + 1)) == (void *)lock) + break; + } + + dtrace_membar_producer(); + } + +top: + prev = NULL; + lock = hash[bucket].dtdh_lock; + + dtrace_membar_consumer(); + + start = hash[bucket].dtdh_chain; + ASSERT(start != NULL && (start->dtdv_hashval == DTRACE_DYNHASH_SINK || + start->dtdv_hashval != DTRACE_DYNHASH_FREE || + op != DTRACE_DYNVAR_DEALLOC)); + + for (dvar = start; dvar != NULL; dvar = dvar->dtdv_next) { + dtrace_tuple_t *dtuple = &dvar->dtdv_tuple; + dtrace_key_t *dkey = &dtuple->dtt_key[0]; + + if (dvar->dtdv_hashval != hashval) { + if (dvar->dtdv_hashval == DTRACE_DYNHASH_SINK) { + /* + * We've reached the sink, and therefore the + * end of the hash chain; we can kick out of + * the loop knowing that we have seen a valid + * snapshot of state. + */ + ASSERT(dvar->dtdv_next == NULL); + ASSERT(dvar == &dtrace_dynhash_sink); + break; + } + + if (dvar->dtdv_hashval == DTRACE_DYNHASH_FREE) { + /* + * We've gone off the rails: somewhere along + * the line, one of the members of this hash + * chain was deleted. Note that we could also + * detect this by simply letting this loop run + * to completion, as we would eventually hit + * the end of the dirty list. However, we + * want to avoid running the length of the + * dirty list unnecessarily (it might be quite + * long), so we catch this as early as + * possible by detecting the hash marker. In + * this case, we simply set dvar to NULL and + * break; the conditional after the loop will + * send us back to top. + */ + dvar = NULL; + break; + } + + goto next; + } + + if (dtuple->dtt_nkeys != nkeys) + goto next; + + for (i = 0; i < nkeys; i++, dkey++) { + if (dkey->dttk_size != key[i].dttk_size) + goto next; /* size or type mismatch */ + + if (dkey->dttk_size != 0) { + if (dtrace_bcmp( + (void *)(uintptr_t)key[i].dttk_value, + (void *)(uintptr_t)dkey->dttk_value, + dkey->dttk_size)) + goto next; + } else { + if (dkey->dttk_value != key[i].dttk_value) + goto next; + } + } + + if (op != DTRACE_DYNVAR_DEALLOC) + return (dvar); + + ASSERT(dvar->dtdv_next == NULL || + dvar->dtdv_next->dtdv_hashval != DTRACE_DYNHASH_FREE); + + if (prev != NULL) { + ASSERT(hash[bucket].dtdh_chain != dvar); + ASSERT(start != dvar); + ASSERT(prev->dtdv_next == dvar); + prev->dtdv_next = dvar->dtdv_next; + } else { + if (dtrace_casptr(&hash[bucket].dtdh_chain, + start, dvar->dtdv_next) != start) { + /* + * We have failed to atomically swing the + * hash table head pointer, presumably because + * of a conflicting allocation on another CPU. + * We need to reread the hash chain and try + * again. + */ + goto top; + } + } + + dtrace_membar_producer(); + + /* + * Now set the hash value to indicate that it's free. + */ + ASSERT(hash[bucket].dtdh_chain != dvar); + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + /* + * Set the next pointer to point at the dirty list, and + * atomically swing the dirty pointer to the newly freed dvar. + */ + do { + next = dcpu->dtdsc_dirty; + dvar->dtdv_next = next; + } while (dtrace_casptr(&dcpu->dtdsc_dirty, next, dvar) != next); + + /* + * Finally, unlock this hash bucket. + */ + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + + return (NULL); +next: + prev = dvar; + continue; + } + + if (dvar == NULL) { + /* + * If dvar is NULL, it is because we went off the rails: + * one of the elements that we traversed in the hash chain + * was deleted while we were traversing it. In this case, + * we assert that we aren't doing a dealloc (deallocs lock + * the hash bucket to prevent themselves from racing with + * one another), and retry the hash chain traversal. + */ + ASSERT(op != DTRACE_DYNVAR_DEALLOC); + goto top; + } + + if (op != DTRACE_DYNVAR_ALLOC) { + /* + * If we are not to allocate a new variable, we want to + * return NULL now. Before we return, check that the value + * of the lock word hasn't changed. If it has, we may have + * seen an inconsistent snapshot. + */ + if (op == DTRACE_DYNVAR_NOALLOC) { + if (hash[bucket].dtdh_lock != lock) + goto top; + } else { + ASSERT(op == DTRACE_DYNVAR_DEALLOC); + ASSERT(hash[bucket].dtdh_lock == lock); + ASSERT(lock & 1); + hash[bucket].dtdh_lock++; + } + + return (NULL); + } + + /* + * We need to allocate a new dynamic variable. The size we need is the + * size of dtrace_dynvar plus the size of nkeys dtrace_key_t's plus the + * size of any auxiliary key data (rounded up to 8-byte alignment) plus + * the size of any referred-to data (dsize). We then round the final + * size up to the chunksize for allocation. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); + + /* + * This should be pretty much impossible, but could happen if, say, + * strange DIF specified the tuple. Ideally, this should be an + * assertion and not an error condition -- but that requires that the + * chunksize calculation in dtrace_difo_chunksize() be absolutely + * bullet-proof. (That is, it must not be able to be fooled by + * malicious DIF.) Given the lack of backwards branches in DIF, + * solving this would presumably not amount to solving the Halting + * Problem -- but it still seems awfully hard. + */ + if (sizeof (dtrace_dynvar_t) + sizeof (dtrace_key_t) * (nkeys - 1) + + ksize + dsize > chunksize) { + dcpu->dtdsc_drops++; + return (NULL); + } + + nstate = DTRACE_DSTATE_EMPTY; + + do { +retry: + free = dcpu->dtdsc_free; + + if (free == NULL) { + dtrace_dynvar_t *clean = dcpu->dtdsc_clean; + void *rval; + + if (clean == NULL) { + /* + * We're out of dynamic variable space on + * this CPU. Unless we have tried all CPUs, + * we'll try to allocate from a different + * CPU. + */ + switch (dstate->dtds_state) { + case DTRACE_DSTATE_CLEAN: { + void *sp = &dstate->dtds_state; + + if (++cpu >= NCPU) + cpu = 0; + + if (dcpu->dtdsc_dirty != NULL && + nstate == DTRACE_DSTATE_EMPTY) + nstate = DTRACE_DSTATE_DIRTY; + + if (dcpu->dtdsc_rinsing != NULL) + nstate = DTRACE_DSTATE_RINSING; + + dcpu = &dstate->dtds_percpu[cpu]; + + if (cpu != me) + goto retry; + + (void) dtrace_cas32(sp, + DTRACE_DSTATE_CLEAN, nstate); + + /* + * To increment the correct bean + * counter, take another lap. + */ + goto retry; + } + + case DTRACE_DSTATE_DIRTY: + dcpu->dtdsc_dirty_drops++; + break; + + case DTRACE_DSTATE_RINSING: + dcpu->dtdsc_rinsing_drops++; + break; + + case DTRACE_DSTATE_EMPTY: + dcpu->dtdsc_drops++; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_DROP); + return (NULL); + } + + /* + * The clean list appears to be non-empty. We want to + * move the clean list to the free list; we start by + * moving the clean pointer aside. + */ + if (dtrace_casptr(&dcpu->dtdsc_clean, + clean, NULL) != clean) { + /* + * We are in one of two situations: + * + * (a) The clean list was switched to the + * free list by another CPU. + * + * (b) The clean list was added to by the + * cleansing cyclic. + * + * In either of these situations, we can + * just reattempt the free list allocation. + */ + goto retry; + } + + ASSERT(clean->dtdv_hashval == DTRACE_DYNHASH_FREE); + + /* + * Now we'll move the clean list to our free list. + * It's impossible for this to fail: the only way + * the free list can be updated is through this + * code path, and only one CPU can own the clean list. + * Thus, it would only be possible for this to fail if + * this code were racing with dtrace_dynvar_clean(). + * (That is, if dtrace_dynvar_clean() updated the clean + * list, and we ended up racing to update the free + * list.) This race is prevented by the dtrace_sync() + * in dtrace_dynvar_clean() -- which flushes the + * owners of the clean lists out before resetting + * the clean lists. + */ + dcpu = &dstate->dtds_percpu[me]; + rval = dtrace_casptr(&dcpu->dtdsc_free, NULL, clean); + ASSERT(rval == NULL); + goto retry; + } + + dvar = free; + new_free = dvar->dtdv_next; + } while (dtrace_casptr(&dcpu->dtdsc_free, free, new_free) != free); + + /* + * We have now allocated a new chunk. We copy the tuple keys into the + * tuple array and copy any referenced key data into the data space + * following the tuple array. As we do this, we relocate dttk_value + * in the final tuple to point to the key data address in the chunk. + */ + kdata = (uintptr_t)&dvar->dtdv_tuple.dtt_key[nkeys]; + dvar->dtdv_data = (void *)(kdata + ksize); + dvar->dtdv_tuple.dtt_nkeys = nkeys; + + for (i = 0; i < nkeys; i++) { + dtrace_key_t *dkey = &dvar->dtdv_tuple.dtt_key[i]; + size_t kesize = key[i].dttk_size; + + if (kesize != 0) { + dtrace_bcopy( + (const void *)(uintptr_t)key[i].dttk_value, + (void *)kdata, kesize); + dkey->dttk_value = kdata; + kdata += P2ROUNDUP(kesize, sizeof (uint64_t)); + } else { + dkey->dttk_value = key[i].dttk_value; + } + + dkey->dttk_size = kesize; + } + + ASSERT(dvar->dtdv_hashval == DTRACE_DYNHASH_FREE); + dvar->dtdv_hashval = hashval; + dvar->dtdv_next = start; + + if (dtrace_casptr(&hash[bucket].dtdh_chain, start, dvar) == start) + return (dvar); + + /* + * The cas has failed. Either another CPU is adding an element to + * this hash chain, or another CPU is deleting an element from this + * hash chain. The simplest way to deal with both of these cases + * (though not necessarily the most efficient) is to free our + * allocated block and tail-call ourselves. Note that the free is + * to the dirty list and _not_ to the free list. This is to prevent + * races with allocators, above. + */ + dvar->dtdv_hashval = DTRACE_DYNHASH_FREE; + + dtrace_membar_producer(); + + do { + free = dcpu->dtdsc_dirty; + dvar->dtdv_next = free; + } while (dtrace_casptr(&dcpu->dtdsc_dirty, free, dvar) != free); + + return (dtrace_dynvar(dstate, nkeys, key, dsize, op, mstate, vstate)); +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_min(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + RT_NOREF_PV(arg); + if ((int64_t)nval < (int64_t)*oval) + *oval = nval; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_max(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + RT_NOREF_PV(arg); + if ((int64_t)nval > (int64_t)*oval) + *oval = nval; +} + +static void +dtrace_aggregate_quantize(uint64_t *quanta, uint64_t nval, uint64_t incr) +{ + int i, zero = DTRACE_QUANTIZE_ZEROBUCKET; + int64_t val = (int64_t)nval; + + if (val < 0) { + for (i = 0; i < zero; i++) { + if (val <= DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i] += incr; + return; + } + } + } else { + for (i = zero + 1; i < VBDTCAST(int)DTRACE_QUANTIZE_NBUCKETS; i++) { + if (val < DTRACE_QUANTIZE_BUCKETVAL(i)) { + quanta[i - 1] += incr; + return; + } + } + + quanta[DTRACE_QUANTIZE_NBUCKETS - 1] += incr; + return; + } + +#ifndef VBOX + ASSERT(0); +#else + AssertFatalFailed(); +#endif +} + +static void +dtrace_aggregate_lquantize(uint64_t *lquanta, uint64_t nval, uint64_t incr) +{ + uint64_t arg = *lquanta++; + int32_t base = DTRACE_LQUANTIZE_BASE(arg); + uint16_t step = DTRACE_LQUANTIZE_STEP(arg); + uint16_t levels = DTRACE_LQUANTIZE_LEVELS(arg); + int32_t val = (int32_t)nval, level; + + ASSERT(step != 0); + ASSERT(levels != 0); + + if (val < base) { + /* + * This is an underflow. + */ + lquanta[0] += incr; + return; + } + + level = (val - base) / step; + + if (level < levels) { + lquanta[level + 1] += incr; + return; + } + + /* + * This is an overflow. + */ + lquanta[levels + 1] += incr; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_avg(uint64_t *data, uint64_t nval, uint64_t arg) +{ + RT_NOREF_PV(arg); + data[0]++; + data[1] += nval; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_stddev(uint64_t *data, uint64_t nval, uint64_t arg) +{ + int64_t snval = (int64_t)nval; + uint64_t tmp[2]; + RT_NOREF_PV(arg); + + data[0]++; + data[1] += nval; + + /* + * What we want to say here is: + * + * data[2] += nval * nval; + * + * But given that nval is 64-bit, we could easily overflow, so + * we do this as 128-bit arithmetic. + */ + if (snval < 0) + snval = -snval; + + dtrace_multiply_128((uint64_t)snval, (uint64_t)snval, tmp); + dtrace_add_128(data + 2, tmp, data + 2); +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_count(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + RT_NOREF_PV(arg); RT_NOREF_PV(nval); + + *oval = *oval + 1; +} + +/*ARGSUSED*/ +static void +dtrace_aggregate_sum(uint64_t *oval, uint64_t nval, uint64_t arg) +{ + RT_NOREF_PV(arg); + *oval += nval; +} + +/* + * Aggregate given the tuple in the principal data buffer, and the aggregating + * action denoted by the specified dtrace_aggregation_t. The aggregation + * buffer is specified as the buf parameter. This routine does not return + * failure; if there is no space in the aggregation buffer, the data will be + * dropped, and a corresponding counter incremented. + */ +static void +dtrace_aggregate(dtrace_aggregation_t *agg, dtrace_buffer_t *dbuf, + intptr_t offset, dtrace_buffer_t *buf, uint64_t expr, uint64_t arg) +{ + dtrace_recdesc_t *rec = &agg->dtag_action.dta_rec; + uint32_t i, ndx, size, fsize; + uint32_t align = sizeof (uint64_t) - 1; + dtrace_aggbuffer_t *agb; + dtrace_aggkey_t *key; + uint32_t hashval = 0, limit, isstr; + caddr_t tomax, data, kdata; + dtrace_actkind_t action; + dtrace_action_t *act; + uintptr_t offs; + + if (buf == NULL) + return; + + if (!agg->dtag_hasarg) { + /* + * Currently, only quantize() and lquantize() take additional + * arguments, and they have the same semantics: an increment + * value that defaults to 1 when not present. If additional + * aggregating actions take arguments, the setting of the + * default argument value will presumably have to become more + * sophisticated... + */ + arg = 1; + } + + action = agg->dtag_action.dta_kind - DTRACEACT_AGGREGATION; + size = rec->dtrd_offset - agg->dtag_base; + fsize = size + rec->dtrd_size; + + ASSERT(dbuf->dtb_tomax != NULL); + data = dbuf->dtb_tomax + offset + agg->dtag_base; + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return; + } + + /* + * The metastructure is always at the bottom of the buffer. + */ + agb = (dtrace_aggbuffer_t *)(tomax + buf->dtb_size - + sizeof (dtrace_aggbuffer_t)); + + if (buf->dtb_offset == 0) { + /* + * We just kludge up approximately 1/8th of the size to be + * buckets. If this guess ends up being routinely + * off-the-mark, we may need to dynamically readjust this + * based on past performance. + */ + uintptr_t hashsize = (buf->dtb_size >> 3) / sizeof (uintptr_t); + + if ((uintptr_t)agb - hashsize * sizeof (dtrace_aggkey_t *) < + (uintptr_t)tomax || hashsize == 0) { + /* + * We've been given a ludicrously small buffer; + * increment our drop count and leave. + */ + dtrace_buffer_drop(buf); + return; + } + + /* + * And now, a pathetic attempt to try to get a an odd (or + * perchance, a prime) hash size for better hash distribution. + */ + if (hashsize > (DTRACE_AGGHASHSIZE_SLEW << 3)) + hashsize -= DTRACE_AGGHASHSIZE_SLEW; + + agb->dtagb_hashsize = hashsize; + agb->dtagb_hash = (dtrace_aggkey_t **)((uintptr_t)agb - + agb->dtagb_hashsize * sizeof (dtrace_aggkey_t *)); + agb->dtagb_free = (uintptr_t)agb->dtagb_hash; + + for (i = 0; i < agb->dtagb_hashsize; i++) + agb->dtagb_hash[i] = NULL; + } + + ASSERT(agg->dtag_first != NULL); + ASSERT(agg->dtag_first->dta_intuple); + + /* + * Calculate the hash value based on the key. Note that we _don't_ + * include the aggid in the hashing (but we will store it as part of + * the key). The hashing algorithm is Bob Jenkins' "One-at-a-time" + * algorithm: a simple, quick algorithm that has no known funnels, and + * gets good distribution in practice. The efficacy of the hashing + * algorithm (and a comparison with other algorithms) may be found by + * running the ::dtrace_aggstat MDB dcmd. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + hashval += data[i]; + hashval += (hashval << 10); + hashval ^= (hashval >> 6); + + if (isstr && data[i] == '\0') + break; + } + } + + hashval += (hashval << 3); + hashval ^= (hashval >> 11); + hashval += (hashval << 15); + + /* + * Yes, the divide here is expensive -- but it's generally the least + * of the performance issues given the amount of data that we iterate + * over to compute hash values, compare data, etc. + */ + ndx = hashval % agb->dtagb_hashsize; + + for (key = agb->dtagb_hash[ndx]; key != NULL; key = key->dtak_next) { + ASSERT((caddr_t)key >= tomax); + ASSERT((caddr_t)key < tomax + buf->dtb_size); + + if (hashval != key->dtak_hashval || key->dtak_size != size) + continue; + + kdata = key->dtak_data; + ASSERT(kdata >= tomax && kdata < tomax + buf->dtb_size); + + for (act = agg->dtag_first; act->dta_intuple; + act = act->dta_next) { + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + isstr = DTRACEACT_ISSTRING(act); + + for (; i < limit; i++) { + if (kdata[i] != data[i]) + goto next; + + if (isstr && data[i] == '\0') + break; + } + } + + if (action != key->dtak_action) { + /* + * We are aggregating on the same value in the same + * aggregation with two different aggregating actions. + * (This should have been picked up in the compiler, + * so we may be dealing with errant or devious DIF.) + * This is an error condition; we indicate as much, + * and return. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + + /* + * This is a hit: we need to apply the aggregator to + * the value at this key. + */ + agg->dtag_aggregate((uint64_t *)(kdata + size), expr, arg); + return; +next: + continue; + } + + /* + * We didn't find it. We need to allocate some zero-filled space, + * link it into the hash table appropriately, and apply the aggregator + * to the (zero-filled) value. + */ + offs = buf->dtb_offset; + while (offs & (align - 1)) + offs += sizeof (uint32_t); + + /* + * If we don't have enough room to both allocate a new key _and_ + * its associated data, increment the drop count and return. + */ + if ((uintptr_t)tomax + offs + fsize > + agb->dtagb_free - sizeof (dtrace_aggkey_t)) { + dtrace_buffer_drop(buf); + return; + } + + /*CONSTCOND*/ + ASSERT(!(sizeof (dtrace_aggkey_t) & (sizeof (uintptr_t) - 1))); + key = (dtrace_aggkey_t *)(agb->dtagb_free - sizeof (dtrace_aggkey_t)); + agb->dtagb_free -= sizeof (dtrace_aggkey_t); + + key->dtak_data = kdata = tomax + offs; + buf->dtb_offset = offs + fsize; + + /* + * Now copy the data across. + */ + *((dtrace_aggid_t *)kdata) = agg->dtag_id; + + for (i = sizeof (dtrace_aggid_t); i < size; i++) + kdata[i] = data[i]; + + /* + * Because strings are not zeroed out by default, we need to iterate + * looking for actions that store strings, and we need to explicitly + * pad these strings out with zeroes. + */ + for (act = agg->dtag_first; act->dta_intuple; act = act->dta_next) { + int nul; + + if (!DTRACEACT_ISSTRING(act)) + continue; + + i = act->dta_rec.dtrd_offset - agg->dtag_base; + limit = i + act->dta_rec.dtrd_size; + ASSERT(limit <= size); + + for (nul = 0; i < limit; i++) { + if (nul) { + kdata[i] = '\0'; + continue; + } + + if (data[i] != '\0') + continue; + + nul = 1; + } + } + + for (i = size; i < fsize; i++) + kdata[i] = 0; + + key->dtak_hashval = hashval; + key->dtak_size = size; + key->dtak_action = action; + key->dtak_next = agb->dtagb_hash[ndx]; + agb->dtagb_hash[ndx] = key; + + /* + * Finally, apply the aggregator. + */ + *((uint64_t *)(key->dtak_data + size)) = agg->dtag_initial; + agg->dtag_aggregate((uint64_t *)(key->dtak_data + size), expr, arg); +} + +/* + * Given consumer state, this routine finds a speculation in the INACTIVE + * state and transitions it into the ACTIVE state. If there is no speculation + * in the INACTIVE state, 0 is returned. In this case, no error counter is + * incremented -- it is up to the caller to take appropriate action. + */ +static int +dtrace_speculation(dtrace_state_t *state) +{ + int i = 0; + dtrace_speculation_state_t current; + uint32_t *stat = &state->dts_speculations_unavail, count; + + while (i < state->dts_nspeculations) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + current = spec->dtsp_state; + + if (current != DTRACESPEC_INACTIVE) { + if (current == DTRACESPEC_COMMITTINGMANY || + current == DTRACESPEC_COMMITTING || + current == DTRACESPEC_DISCARDING) + stat = &state->dts_speculations_busy; + i++; + continue; + } + + if ( (dtrace_speculation_state_t)dtrace_cas32((uint32_t *)&spec->dtsp_state, current, DTRACESPEC_ACTIVE) + == current) + return (i + 1); + } + + /* + * We couldn't find a speculation. If we found as much as a single + * busy speculation buffer, we'll attribute this failure as "busy" + * instead of "unavail". + */ + do { + count = *stat; + } while (dtrace_cas32(stat, count, count + 1) != count); + + return (0); +} + +/* + * This routine commits an active speculation. If the specified speculation + * is not in a valid state to perform a commit(), this routine will silently do + * nothing. The state of the specified speculation is transitioned according + * to the state transition diagram outlined in <sys/dtrace_impl.h> + */ +static void +dtrace_speculation_commit(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_buffer_t *src, *dest; + uintptr_t daddr, saddr, dlimit; + dtrace_speculation_state_t current, new VBDTUNASS(-1); + intptr_t offs; + + if (which == 0) + return; + + if (which > VBDTCAST(unsigned)state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + src = &spec->dtsp_buffer[cpu]; + dest = &state->dts_buffer[cpu]; + + do { + current = spec->dtsp_state; + + if (current == DTRACESPEC_COMMITTINGMANY) + break; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_COMMITTING: + /* + * This is only possible if we are (a) commit()'ing + * without having done a prior speculate() on this CPU + * and (b) racing with another commit() on a different + * CPU. There's nothing to do -- we just assert that + * our offset is 0. + */ + ASSERT(src->dtb_offset == 0); + return; + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_COMMITTING; + break; + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is active on one CPU. If our + * buffer offset is non-zero, we know that the one CPU + * must be us. Otherwise, we are committing on a + * different CPU from the speculate(), and we must + * rely on being asynchronously cleaned. + */ + if (src->dtb_offset != 0) { + new = DTRACESPEC_COMMITTING; + break; + } + RT_FALL_THRU(); + + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_COMMITTINGMANY; + break; + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", current)); +#endif + } + } while ((dtrace_speculation_state_t)dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new) != current); + + /* + * We have set the state to indicate that we are committing this + * speculation. Now reserve the necessary space in the destination + * buffer. + */ + if ((offs = dtrace_buffer_reserve(dest, src->dtb_offset, + sizeof (uint64_t), state, NULL)) < 0) { + dtrace_buffer_drop(dest); + goto out; + } + + /* + * We have the space; copy the buffer across. (Note that this is a + * highly subobtimal bcopy(); in the unlikely event that this becomes + * a serious performance issue, a high-performance DTrace-specific + * bcopy() should obviously be invented.) + */ + daddr = (uintptr_t)dest->dtb_tomax + offs; + dlimit = daddr + src->dtb_offset; + saddr = (uintptr_t)src->dtb_tomax; + + /* + * First, the aligned portion. + */ + while (dlimit - daddr >= sizeof (uint64_t)) { + *((uint64_t *)daddr) = *((uint64_t *)saddr); + + daddr += sizeof (uint64_t); + saddr += sizeof (uint64_t); + } + + /* + * Now any left-over bit... + */ + while (dlimit - daddr) + *((uint8_t *)daddr++) = *((uint8_t *)saddr++); + + /* + * Finally, commit the reserved space in the destination buffer. + */ + dest->dtb_offset = offs + src->dtb_offset; + +out: + /* + * If we're lucky enough to be the only active CPU on this speculation + * buffer, we can just set the state back to DTRACESPEC_INACTIVE. + */ + if (current == DTRACESPEC_ACTIVE || + (current == DTRACESPEC_ACTIVEONE && new == DTRACESPEC_COMMITTING)) { + uint32_t rval = dtrace_cas32((uint32_t *)&spec->dtsp_state, + DTRACESPEC_COMMITTING, DTRACESPEC_INACTIVE); + + ASSERT(rval == DTRACESPEC_COMMITTING); NOREF(rval); + } + + src->dtb_offset = 0; + src->dtb_xamot_drops += src->dtb_drops; + src->dtb_drops = 0; +} + +/* + * This routine discards an active speculation. If the specified speculation + * is not in a valid state to perform a discard(), this routine will silently + * do nothing. The state of the specified speculation is transitioned + * according to the state transition diagram outlined in <sys/dtrace_impl.h> + */ +static void +dtrace_speculation_discard(dtrace_state_t *state, processorid_t cpu, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t current, new; + dtrace_buffer_t *buf; + + if (which == 0) + return; + + if (which > VBDTCAST(unsigned)state->dts_nspeculations) { + cpu_core[cpu].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return; + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpu]; + + do { + current = spec->dtsp_state; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_COMMITTING: + case DTRACESPEC_DISCARDING: + return; + + case DTRACESPEC_ACTIVE: + case DTRACESPEC_ACTIVEMANY: + new = DTRACESPEC_DISCARDING; + break; + + case DTRACESPEC_ACTIVEONE: + if (buf->dtb_offset != 0) { + new = DTRACESPEC_INACTIVE; + } else { + new = DTRACESPEC_DISCARDING; + } + break; + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", current)); +#endif + } + } while ((dtrace_speculation_state_t)dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new) != current); + + buf->dtb_offset = 0; + buf->dtb_drops = 0; +} + +/* + * Note: not called from probe context. This function is called + * asynchronously from cross call context to clean any speculations that are + * in the COMMITTINGMANY or DISCARDING states. These speculations may not be + * transitioned back to the INACTIVE state until all CPUs have cleaned the + * speculation. + */ +static void +dtrace_speculation_clean_here(dtrace_state_t *state) +{ + dtrace_icookie_t cookie; + processorid_t cpu = VBDT_GET_CPUID(); + dtrace_buffer_t *dest = &state->dts_buffer[cpu]; + dtrace_specid_t i; + + cookie = dtrace_interrupt_disable(); + + if (dest->dtb_tomax == NULL) { + dtrace_interrupt_enable(cookie); + return; + } + + for (i = 0; i < VBDTCAST(unsigned)state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + dtrace_buffer_t *src = &spec->dtsp_buffer[cpu]; + + if (src->dtb_tomax == NULL) + continue; + + if (spec->dtsp_state == DTRACESPEC_DISCARDING) { + src->dtb_offset = 0; + continue; + } + + if (spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + if (src->dtb_offset == 0) + continue; + + dtrace_speculation_commit(state, cpu, i + 1); + } + + dtrace_interrupt_enable(cookie); +} + +#ifdef VBOX +/** */ +static DECLCALLBACK(void) dtrace_speculation_clean_here_wrapper(RTCPUID idCpu, void *pvUser1, void *pvUser2) +{ + dtrace_speculation_clean_here((dtrace_state_t *)pvUser1); + NOREF(pvUser2); NOREF(idCpu); +} +#endif + +/* + * Note: not called from probe context. This function is called + * asynchronously (and at a regular interval) to clean any speculations that + * are in the COMMITTINGMANY or DISCARDING states. If it discovers that there + * is work to be done, it cross calls all CPUs to perform that work; + * COMMITMANY and DISCARDING speculations may not be transitioned back to the + * INACTIVE state until they have been cleaned by all CPUs. + */ +static void +dtrace_speculation_clean(dtrace_state_t *state) +{ + int work = 0, rv; + dtrace_specid_t i; + + for (i = 0; i < VBDTCAST(unsigned)state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + + ASSERT(!spec->dtsp_cleaning); + + if (spec->dtsp_state != DTRACESPEC_DISCARDING && + spec->dtsp_state != DTRACESPEC_COMMITTINGMANY) + continue; + + work++; + spec->dtsp_cleaning = 1; + } + + if (!work) + return; + +#ifndef VBOX + dtrace_xcall(DTRACE_CPUALL, + (dtrace_xcall_t)dtrace_speculation_clean_here, state); +#else + RTMpOnAll(dtrace_speculation_clean_here_wrapper, state, NULL); +#endif + + /* + * We now know that all CPUs have committed or discarded their + * speculation buffers, as appropriate. We can now set the state + * to inactive. + */ + for (i = 0; i < VBDTCAST(unsigned)state->dts_nspeculations; i++) { + dtrace_speculation_t *spec = &state->dts_speculations[i]; + dtrace_speculation_state_t current, new; + + if (!spec->dtsp_cleaning) + continue; + + current = spec->dtsp_state; + ASSERT(current == DTRACESPEC_DISCARDING || + current == DTRACESPEC_COMMITTINGMANY); + + new = DTRACESPEC_INACTIVE; + + rv = dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new); + ASSERT(VBDTCAST(dtrace_speculation_state_t)rv == current); + spec->dtsp_cleaning = 0; + } +} + +/* + * Called as part of a speculate() to get the speculative buffer associated + * with a given speculation. Returns NULL if the specified speculation is not + * in an ACTIVE state. If the speculation is in the ACTIVEONE state -- and + * the active CPU is not the specified CPU -- the speculation will be + * atomically transitioned into the ACTIVEMANY state. + */ +static dtrace_buffer_t * +dtrace_speculation_buffer(dtrace_state_t *state, processorid_t cpuid, + dtrace_specid_t which) +{ + dtrace_speculation_t *spec; + dtrace_speculation_state_t current, new VBDTUNASS(-1); + dtrace_buffer_t *buf; + + if (which == 0) + return (NULL); + + if (which > VBDTCAST(unsigned)state->dts_nspeculations) { + cpu_core[cpuid].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (NULL); + } + + spec = &state->dts_speculations[which - 1]; + buf = &spec->dtsp_buffer[cpuid]; + + do { + current = spec->dtsp_state; + + switch (current) { + case DTRACESPEC_INACTIVE: + case DTRACESPEC_COMMITTINGMANY: + case DTRACESPEC_DISCARDING: + return (NULL); + + case DTRACESPEC_COMMITTING: + ASSERT(buf->dtb_offset == 0); + return (NULL); + + case DTRACESPEC_ACTIVEONE: + /* + * This speculation is currently active on one CPU. + * Check the offset in the buffer; if it's non-zero, + * that CPU must be us (and we leave the state alone). + * If it's zero, assume that we're starting on a new + * CPU -- and change the state to indicate that the + * speculation is active on more than one CPU. + */ + if (buf->dtb_offset != 0) + return (buf); + + new = DTRACESPEC_ACTIVEMANY; + break; + + case DTRACESPEC_ACTIVEMANY: + return (buf); + + case DTRACESPEC_ACTIVE: + new = DTRACESPEC_ACTIVEONE; + break; + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", current)); +#endif + } + } while ((dtrace_speculation_state_t)dtrace_cas32((uint32_t *)&spec->dtsp_state, current, new) != current); + + ASSERT(new == DTRACESPEC_ACTIVEONE || new == DTRACESPEC_ACTIVEMANY); + return (buf); +} + +/* + * Return a string. In the event that the user lacks the privilege to access + * arbitrary kernel memory, we copy the string out to scratch memory so that we + * don't fail access checking. + * + * dtrace_dif_variable() uses this routine as a helper for various + * builtin values such as 'execname' and 'probefunc.' + */ +VBDTSTATIC uintptr_t +dtrace_dif_varstr(uintptr_t addr, dtrace_state_t *state, + dtrace_mstate_t *mstate) +{ + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t ret; + size_t strsz; + + /* + * The easy case: this probe is allowed to read all of memory, so + * we can just return this as a vanilla pointer. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) != 0) + return (addr); + + /* + * This is the tougher case: we copy the string in question from + * kernel memory into scratch memory and return it that way: this + * ensures that we won't trip up when access checking tests the + * BYREF return value. + */ + strsz = dtrace_strlen((char *)addr, size) + 1; + + if (mstate->dtms_scratch_ptr + strsz > + mstate->dtms_scratch_base + mstate->dtms_scratch_size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return (NULL); + } + + dtrace_strcpy((const void *)addr, (void *)mstate->dtms_scratch_ptr, + strsz); + ret = mstate->dtms_scratch_ptr; + mstate->dtms_scratch_ptr += strsz; + return (ret); +} + +/* + * This function implements the DIF emulator's variable lookups. The emulator + * passes a reserved variable identifier and optional built-in array index. + */ +static uint64_t +dtrace_dif_variable(dtrace_mstate_t *mstate, dtrace_state_t *state, uint64_t v, + uint64_t ndx) +{ + /* + * If we're accessing one of the uncached arguments, we'll turn this + * into a reference in the args array. + */ + if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) { + ndx = v - DIF_VAR_ARG0; + v = DIF_VAR_ARGS; + } + + switch (v) { + case DIF_VAR_ARGS: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_ARGS); + if (ndx >= sizeof (mstate->dtms_arg) / + sizeof (mstate->dtms_arg[0])) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + dtrace_provider_t *pv; + uint64_t val; + + pv = mstate->dtms_probe->dtpr_provider; + if (pv->dtpv_pops.dtps_getargval != NULL) + val = pv->dtpv_pops.dtps_getargval(pv->dtpv_arg, + mstate->dtms_probe->dtpr_id, + mstate->dtms_probe->dtpr_arg, ndx, aframes); + else + val = dtrace_getarg(ndx, aframes); + + /* + * This is regrettably required to keep the compiler + * from tail-optimizing the call to dtrace_getarg(). + * The condition always evaluates to true, but the + * compiler has no way of figuring that out a priori. + * (None of this would be necessary if the compiler + * could be relied upon to _always_ tail-optimize + * the call to dtrace_getarg() -- but it can't.) + */ + if (mstate->dtms_probe != NULL) + return (val); + +#ifndef VBOX + ASSERT(0); +#else + AssertFatalFailed(); +#endif + } + + return (mstate->dtms_arg[ndx]); + + case DIF_VAR_UREGS: { +#ifndef VBOX + klwp_t *lwp; + + if (!dtrace_priv_proc(state)) + return (0); + + if ((lwp = curthread->t_lwp) == NULL) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval = NULL; + return (0); + } + + return (dtrace_getreg(lwp->lwp_regs, ndx)); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + } + + case DIF_VAR_CURTHREAD: + if (!dtrace_priv_kernel(state)) + return (0); +#ifndef VBOX + return ((uint64_t)(uintptr_t)curthread); +#else + return ((uintptr_t)RTThreadNativeSelf()); +#endif + + case DIF_VAR_TIMESTAMP: + if (!(mstate->dtms_present & DTRACE_MSTATE_TIMESTAMP)) { + mstate->dtms_timestamp = dtrace_gethrtime(); + mstate->dtms_present |= DTRACE_MSTATE_TIMESTAMP; + } + return (mstate->dtms_timestamp); + + case DIF_VAR_VTIMESTAMP: +#ifndef VBOX + ASSERT(dtrace_vtime_references != 0); + return (curthread->t_dtrace_vtime); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + + case DIF_VAR_WALLTIMESTAMP: + if (!(mstate->dtms_present & DTRACE_MSTATE_WALLTIMESTAMP)) { + mstate->dtms_walltimestamp = dtrace_gethrestime(); + mstate->dtms_present |= DTRACE_MSTATE_WALLTIMESTAMP; + } + return (mstate->dtms_walltimestamp); + + case DIF_VAR_IPL: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_IPL)) { + mstate->dtms_ipl = dtrace_getipl(); + mstate->dtms_present |= DTRACE_MSTATE_IPL; + } + return (mstate->dtms_ipl); + + case DIF_VAR_EPID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_EPID); + return (mstate->dtms_epid); + + case DIF_VAR_ID: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (mstate->dtms_probe->dtpr_id); + + case DIF_VAR_STACKDEPTH: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_STACKDEPTH)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + mstate->dtms_stackdepth = dtrace_getstackdepth(aframes); + mstate->dtms_present |= DTRACE_MSTATE_STACKDEPTH; + } + return (mstate->dtms_stackdepth); + + case DIF_VAR_USTACKDEPTH: + if (!dtrace_priv_proc(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_USTACKDEPTH)) { + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && + CPU_ON_INTR(CPU)) { + mstate->dtms_ustackdepth = 0; + } else { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + mstate->dtms_ustackdepth = + dtrace_getustackdepth(); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + mstate->dtms_present |= DTRACE_MSTATE_USTACKDEPTH; + } + return (mstate->dtms_ustackdepth); + + case DIF_VAR_CALLER: + if (!dtrace_priv_kernel(state)) + return (0); + if (!(mstate->dtms_present & DTRACE_MSTATE_CALLER)) { + int aframes = mstate->dtms_probe->dtpr_aframes + 2; + + if (!DTRACE_ANCHORED(mstate->dtms_probe)) { + /* + * If this is an unanchored probe, we are + * required to go through the slow path: + * dtrace_caller() only guarantees correct + * results for anchored probes. + */ + pc_t caller[2]; + + dtrace_getpcstack(caller, 2, aframes, + (uint32_t *)(uintptr_t)mstate->dtms_arg[0]); + mstate->dtms_caller = caller[1]; + } else if ((mstate->dtms_caller = + dtrace_caller(aframes)) == VBDTCAST(uintptr_t)-1) { + /* + * We have failed to do this the quick way; + * we must resort to the slower approach of + * calling dtrace_getpcstack(). + */ + pc_t caller; + + dtrace_getpcstack(&caller, 1, aframes, NULL); + mstate->dtms_caller = caller; + } + + mstate->dtms_present |= DTRACE_MSTATE_CALLER; + } + return (mstate->dtms_caller); + + case DIF_VAR_UCALLER: + if (!dtrace_priv_proc(state)) + return (0); + + if (!(mstate->dtms_present & DTRACE_MSTATE_UCALLER)) { + uint64_t ustack[3]; + + /* + * dtrace_getupcstack() fills in the first uint64_t + * with the current PID. The second uint64_t will + * be the program counter at user-level. The third + * uint64_t will contain the caller, which is what + * we're after. + */ + ustack[2] = NULL; + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack(ustack, 3); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + mstate->dtms_ucaller = ustack[2]; + mstate->dtms_present |= DTRACE_MSTATE_UCALLER; + } + + return (mstate->dtms_ucaller); + + case DIF_VAR_PROBEPROV: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_provider->dtpv_name, + state, mstate)); + + case DIF_VAR_PROBEMOD: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_mod, + state, mstate)); + + case DIF_VAR_PROBEFUNC: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_func, + state, mstate)); + + case DIF_VAR_PROBENAME: + ASSERT(mstate->dtms_present & DTRACE_MSTATE_PROBE); + return (dtrace_dif_varstr( + (uintptr_t)mstate->dtms_probe->dtpr_name, + state, mstate)); + + case DIF_VAR_PID: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * Note that we are assuming that an unanchored probe is + * always due to a high-level interrupt. (And we're assuming + * that there is only a single high level interrupt.) + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (pid0.pid_id); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * Further, it is always safe to dereference the p_pidp member + * of one's own proc structure. (These are truisms becuase + * threads and processes don't clean up their own state -- + * they leave that task to whomever reaps them.) + */ + return ((uint64_t)curthread->t_procp->p_pidp->pid_id); +#else + return (RTProcSelf()); +#endif + + case DIF_VAR_PPID: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (pid0.pid_id); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return ((uint64_t)curthread->t_procp->p_ppid); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); /** @todo parent pid? */ +#endif + + case DIF_VAR_TID: +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (0); + + return ((uint64_t)curthread->t_tid); +#else + return (RTThreadNativeSelf()); /** @todo proper tid? */ +#endif + + case DIF_VAR_EXECNAME: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)(uintptr_t)p0.p_user.u_comm); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return (dtrace_dif_varstr( + (uintptr_t)curthread->t_procp->p_user.u_comm, + state, mstate)); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); /** @todo execname */ +#endif + + case DIF_VAR_ZONENAME: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)(uintptr_t)p0.p_zone->zone_name); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + */ + return (dtrace_dif_varstr( + (uintptr_t)curthread->t_procp->p_zone->zone_name, + state, mstate)); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + + case DIF_VAR_UID: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)p0.p_cred->cr_uid); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return ((uint64_t)curthread->t_procp->p_cred->cr_uid); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + + case DIF_VAR_GID: + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return ((uint64_t)p0.p_cred->cr_gid); + + /* + * It is always safe to dereference one's own t_procp pointer: + * it always points to a valid, allocated proc structure. + * (This is true because threads don't clean up their own + * state -- they leave that task to whomever reaps them.) + * + * Additionally, it is safe to dereference one's own process + * credential, since this is never NULL after process birth. + */ + return ((uint64_t)curthread->t_procp->p_cred->cr_gid); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + + case DIF_VAR_ERRNO: { +#ifndef VBOX + klwp_t *lwp; +#endif + if (!dtrace_priv_proc(state)) + return (0); + +#ifndef VBOX + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate->dtms_probe) && CPU_ON_INTR(CPU)) + return (0); + + /* + * It is always safe to dereference one's own t_lwp pointer in + * the event that this pointer is non-NULL. (This is true + * because threads and lwps don't clean up their own state -- + * they leave that task to whomever reaps them.) + */ + if ((lwp = curthread->t_lwp) == NULL) + return (0); + + return ((uint64_t)lwp->lwp_errno); +#else + cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= CPU_DTRACE_ILLOP; + return (0); +#endif + } + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } +} + +/* + * Emulate the execution of DTrace ID subroutines invoked by the call opcode. + * Notice that we don't bother validating the proper number of arguments or + * their types in the tuple stack. This isn't needed because all argument + * interpretation is safe because of our load safety -- the worst that can + * happen is that a bogus program can obtain bogus results. + */ +static void +dtrace_dif_subr(uint_t subr, uint_t rd, uint64_t *regs, + dtrace_key_t *tupregs, int nargs, + dtrace_mstate_t *mstate, dtrace_state_t *state) +{ + volatile uint16_t *flags = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval; + dtrace_vstate_t *vstate = &state->dts_vstate; + +#ifndef VBOX + union { + mutex_impl_t mi; + uint64_t mx; + } m; + + union { + krwlock_t ri; + uintptr_t rw; + } r; +#endif + + switch (subr) { + case DIF_SUBR_RAND: + regs[rd] = (dtrace_gethrtime() * 2416 + 374441) % 1771875; + break; + + case DIF_SUBR_MUTEX_OWNED: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + if (MUTEX_TYPE_ADAPTIVE(&m.mi)) + regs[rd] = MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER; + else + regs[rd] = LOCK_HELD(&m.mi.m_spin.m_spinlock); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_MUTEX_OWNER: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + if (MUTEX_TYPE_ADAPTIVE(&m.mi) && + MUTEX_OWNER(&m.mi) != MUTEX_NO_OWNER) + regs[rd] = (uintptr_t)MUTEX_OWNER(&m.mi); + else + regs[rd] = 0; +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_MUTEX_TYPE_ADAPTIVE: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + regs[rd] = MUTEX_TYPE_ADAPTIVE(&m.mi); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_MUTEX_TYPE_SPIN: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (kmutex_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + m.mx = dtrace_load64(tupregs[0].dttk_value); + regs[rd] = MUTEX_TYPE_SPIN(&m.mi); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_RW_READ_HELD: { +#ifndef VBOX + uintptr_t tmp; + + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (uintptr_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_READ_HELD(&r.ri, tmp); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + } + + case DIF_SUBR_RW_WRITE_HELD: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_WRITE_HELD(&r.ri); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_RW_ISWRITER: +#ifndef VBOX + if (!dtrace_canload(tupregs[0].dttk_value, sizeof (krwlock_t), + mstate, vstate)) { + regs[rd] = NULL; + break; + } + + r.rw = dtrace_loadptr(tupregs[0].dttk_value); + regs[rd] = _RW_ISWRITER(&r.ri); +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_BCOPY: { + /* + * We need to be sure that the destination is in the scratch + * region -- no other region is allowed. + */ + uintptr_t src = tupregs[0].dttk_value; + uintptr_t dest = tupregs[1].dttk_value; + size_t size = tupregs[2].dttk_value; + + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(src, size, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + dtrace_bcopy((void *)src, (void *)dest, size); + break; + } + + case DIF_SUBR_ALLOCA: + case DIF_SUBR_COPYIN: { + uintptr_t dest = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + uint64_t size = + tupregs[subr == DIF_SUBR_ALLOCA ? 0 : 1].dttk_value; + size_t scratch_size = (dest - mstate->dtms_scratch_ptr) + size; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + + /* + * Rounding up the user allocation size could have overflowed + * a large, bogus allocation (like -1ULL) to 0. + */ + if (scratch_size < size || + !DTRACE_INSCRATCH(mstate, scratch_size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + if (subr == DIF_SUBR_COPYIN) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + + mstate->dtms_scratch_ptr += scratch_size; + regs[rd] = dest; + break; + } + + case DIF_SUBR_COPYINTO: { + uint64_t size = tupregs[1].dttk_value; + uintptr_t dest = tupregs[2].dttk_value; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!dtrace_inscratch(dest, size, mstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyin(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + break; + } + + case DIF_SUBR_COPYINSTR: { + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + + if (nargs > 1 && tupregs[1].dttk_value < size) + size = tupregs[1].dttk_value + 1; + + /* + * This action doesn't require any credential checks since + * probes will not activate in user contexts to which the + * enabling user does not have permissions. + */ + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyinstr(tupregs[0].dttk_value, dest, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + ((char *)dest)[size - 1] = '\0'; + mstate->dtms_scratch_ptr += size; + regs[rd] = dest; + break; + } + + case DIF_SUBR_MSGSIZE: + case DIF_SUBR_MSGDSIZE: { +#ifndef VBOX + uintptr_t baddr = tupregs[0].dttk_value, daddr; + uintptr_t wptr, rptr; + size_t count = 0; + int cont = 0; + + while (baddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + + if (!dtrace_canload(baddr, sizeof (mblk_t), mstate, + vstate)) { + regs[rd] = NULL; + break; + } + + wptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_wptr)); + + rptr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_rptr)); + + if (wptr < rptr) { + *flags |= CPU_DTRACE_BADADDR; + *illval = tupregs[0].dttk_value; + break; + } + + daddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_datap)); + + baddr = dtrace_loadptr(baddr + + offsetof(mblk_t, b_cont)); + + /* + * We want to prevent against denial-of-service here, + * so we're only going to search the list for + * dtrace_msgdsize_max mblks. + */ + if (cont++ > dtrace_msgdsize_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + if (subr == DIF_SUBR_MSGDSIZE) { + if (dtrace_load8(daddr + + offsetof(dblk_t, db_type)) != M_DATA) + continue; + } + + count += wptr - rptr; + } + + if (!(*flags & CPU_DTRACE_FAULT)) + regs[rd] = count; + +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + } + + case DIF_SUBR_PROGENYOF: { +#ifndef VBOX + pid_t pid = tupregs[0].dttk_value; + proc_t *p; + int rval = 0; + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + for (p = curthread->t_procp; p != NULL; p = p->p_parent) { + if (p->p_pidp->pid_id == pid) { + rval = 1; + break; + } + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + regs[rd] = rval; +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + } + + case DIF_SUBR_SPECULATION: + regs[rd] = dtrace_speculation(state); + break; + + case DIF_SUBR_COPYOUT: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyout(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_COPYOUTSTR: { + uintptr_t kaddr = tupregs[0].dttk_value; + uintptr_t uaddr = tupregs[1].dttk_value; + uint64_t size = tupregs[2].dttk_value; + + if (!dtrace_destructive_disallow && + dtrace_priv_proc_control(state) && + !dtrace_istoxic(kaddr, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_copyoutstr(kaddr, uaddr, size, flags); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + } + break; + } + + case DIF_SUBR_STRLEN: { + size_t sz; + uintptr_t addr = (uintptr_t)tupregs[0].dttk_value; + sz = dtrace_strlen((char *)addr, + state->dts_options[DTRACEOPT_STRSIZE]); + + if (!dtrace_canload(addr, sz + 1, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + regs[rd] = sz; + + break; + } + + case DIF_SUBR_STRCHR: + case DIF_SUBR_STRRCHR: { + /* + * We're going to iterate over the string looking for the + * specified character. We will iterate until we have reached + * the string length or we have found the character. If this + * is DIF_SUBR_STRRCHR, we will look for the last occurrence + * of the specified character instead of the first. + */ + uintptr_t saddr = tupregs[0].dttk_value; + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t limit = addr + state->dts_options[DTRACEOPT_STRSIZE]; + char c, target = (char)tupregs[1].dttk_value; + + for (regs[rd] = NULL; addr < limit; addr++) { + if ((c = dtrace_load8(addr)) == target) { + regs[rd] = addr; + + if (subr == DIF_SUBR_STRCHR) + break; + } + + if (c == '\0') + break; + } + + if (!dtrace_canload(saddr, addr - saddr, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + break; + } + + case DIF_SUBR_STRSTR: + case DIF_SUBR_INDEX: + case DIF_SUBR_RINDEX: { + /* + * We're going to iterate over the string looking for the + * specified string. We will iterate until we have reached + * the string length or we have found the string. (Yes, this + * is done in the most naive way possible -- but considering + * that the string we're searching for is likely to be + * relatively short, the complexity of Rabin-Karp or similar + * hardly seems merited.) + */ + char *addr = (char *)(uintptr_t)tupregs[0].dttk_value; + char *substr = (char *)(uintptr_t)tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + size_t len = dtrace_strlen(addr, size); + size_t sublen = dtrace_strlen(substr, size); + char *limit = addr + len, *orig = addr; + int notfound = subr == DIF_SUBR_STRSTR ? 0 : -1; + int inc = 1; + + regs[rd] = notfound; + + if (!dtrace_canload((uintptr_t)addr, len + 1, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!dtrace_canload((uintptr_t)substr, sublen + 1, mstate, + vstate)) { + regs[rd] = NULL; + break; + } + + /* + * strstr() and index()/rindex() have similar semantics if + * both strings are the empty string: strstr() returns a + * pointer to the (empty) string, and index() and rindex() + * both return index 0 (regardless of any position argument). + */ + if (sublen == 0 && len == 0) { + if (subr == DIF_SUBR_STRSTR) + regs[rd] = (uintptr_t)addr; + else + regs[rd] = 0; + break; + } + + if (subr != DIF_SUBR_STRSTR) { + if (subr == DIF_SUBR_RINDEX) { + limit = orig - 1; + addr += len; + inc = -1; + } + + /* + * Both index() and rindex() take an optional position + * argument that denotes the starting position. + */ + if (nargs == 3) { + int64_t pos = (int64_t)tupregs[2].dttk_value; + + /* + * If the position argument to index() is + * negative, Perl implicitly clamps it at + * zero. This semantic is a little surprising + * given the special meaning of negative + * positions to similar Perl functions like + * substr(), but it appears to reflect a + * notion that index() can start from a + * negative index and increment its way up to + * the string. Given this notion, Perl's + * rindex() is at least self-consistent in + * that it implicitly clamps positions greater + * than the string length to be the string + * length. Where Perl completely loses + * coherence, however, is when the specified + * substring is the empty string (""). In + * this case, even if the position is + * negative, rindex() returns 0 -- and even if + * the position is greater than the length, + * index() returns the string length. These + * semantics violate the notion that index() + * should never return a value less than the + * specified position and that rindex() should + * never return a value greater than the + * specified position. (One assumes that + * these semantics are artifacts of Perl's + * implementation and not the results of + * deliberate design -- it beggars belief that + * even Larry Wall could desire such oddness.) + * While in the abstract one would wish for + * consistent position semantics across + * substr(), index() and rindex() -- or at the + * very least self-consistent position + * semantics for index() and rindex() -- we + * instead opt to keep with the extant Perl + * semantics, in all their broken glory. (Do + * we have more desire to maintain Perl's + * semantics than Perl does? Probably.) + */ + if (subr == DIF_SUBR_RINDEX) { + if (pos < 0) { + if (sublen == 0) + regs[rd] = 0; + break; + } + + if (VBDTCAST(uint64_t)pos > len) + pos = len; + } else { + if (pos < 0) + pos = 0; + + if (VBDTCAST(uint64_t)pos >= len) { + if (sublen == 0) + regs[rd] = len; + break; + } + } + + addr = orig + pos; + } + } + + for (regs[rd] = notfound; addr != limit; addr += inc) { + if (dtrace_strncmp(addr, substr, sublen) == 0) { + if (subr != DIF_SUBR_STRSTR) { + /* + * As D index() and rindex() are + * modeled on Perl (and not on awk), + * we return a zero-based (and not a + * one-based) index. (For you Perl + * weenies: no, we're not going to add + * $[ -- and shouldn't you be at a con + * or something?) + */ + regs[rd] = (uintptr_t)(addr - orig); + break; + } + + ASSERT(subr == DIF_SUBR_STRSTR); + regs[rd] = (uintptr_t)addr; + break; + } + } + + break; + } + + case DIF_SUBR_STRTOK: { + uintptr_t addr = tupregs[0].dttk_value; + uintptr_t tokaddr = tupregs[1].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t limit, toklimit = tokaddr + size; + uint8_t c VBDTUNASS(0), tokmap[32]; /* 256 / 8 */ + char *dest = (char *)mstate->dtms_scratch_ptr; + VBDTTYPE(unsigned,int) i; + + /* + * Check both the token buffer and (later) the input buffer, + * since both could be non-scratch addresses. + */ + if (!dtrace_strcanload(tokaddr, size, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + if (addr == NULL) { + /* + * If the address specified is NULL, we use our saved + * strtok pointer from the mstate. Note that this + * means that the saved strtok pointer is _only_ + * valid within multiple enablings of the same probe -- + * it behaves like an implicit clause-local variable. + */ + addr = mstate->dtms_strtok; + } else { + /* + * If the user-specified address is non-NULL we must + * access check it. This is the only time we have + * a chance to do so, since this address may reside + * in the string table of this clause-- future calls + * (when we fetch addr from mstate->dtms_strtok) + * would fail this access check. + */ + if (!dtrace_strcanload(addr, size, mstate, vstate)) { + regs[rd] = NULL; + break; + } + } + + /* + * First, zero the token map, and then process the token + * string -- setting a bit in the map for every character + * found in the token string. + */ + for (i = 0; i < sizeof (tokmap); i++) + tokmap[i] = 0; + + for (; tokaddr < toklimit; tokaddr++) { + if ((c = dtrace_load8(tokaddr)) == '\0') + break; + + ASSERT((c >> 3) < sizeof (tokmap)); + tokmap[c >> 3] |= (1 << (c & 0x7)); + } + + for (limit = addr + size; addr < limit; addr++) { + /* + * We're looking for a character that is _not_ contained + * in the token string. + */ + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (!(tokmap[c >> 3] & (1 << (c & 0x7)))) + break; + } + + if (c == '\0') { + /* + * We reached the end of the string without finding + * any character that was not in the token string. + * We return NULL in this case, and we set the saved + * address to NULL as well. + */ + regs[rd] = NULL; + mstate->dtms_strtok = NULL; + break; + } + + /* + * From here on, we're copying into the destination string. + */ + for (i = 0; addr < limit && i < size - 1; addr++) { + if ((c = dtrace_load8(addr)) == '\0') + break; + + if (tokmap[c >> 3] & (1 << (c & 0x7))) + break; + + ASSERT(i < size); + dest[i++] = c; + } + + ASSERT(i < size); + dest[i] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + mstate->dtms_strtok = addr; + break; + } + + case DIF_SUBR_SUBSTR: { + uintptr_t s = tupregs[0].dttk_value; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *d = (char *)mstate->dtms_scratch_ptr; + int64_t index = (int64_t)tupregs[1].dttk_value; + int64_t remaining = (int64_t)tupregs[2].dttk_value; + size_t len = dtrace_strlen((char *)s, size); + int64_t i; + + if (!dtrace_canload(s, len + 1, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + if (nargs <= 2) + remaining = (int64_t)size; + + if (index < 0) { + index += len; + + if (index < 0 && index + remaining > 0) { + remaining += index; + index = 0; + } + } + + if (VBDTCAST(uint64_t)index >= len || index < 0) { + remaining = 0; + } else if (remaining < 0) { + remaining += len - index; + } else if (VBDTCAST(uint64_t)index + remaining > size) { + remaining = size - index; + } + + for (i = 0; i < remaining; i++) { + if ((d[i] = dtrace_load8(s + index + i)) == '\0') + break; + } + + d[i] = '\0'; + + mstate->dtms_scratch_ptr += size; + regs[rd] = (uintptr_t)d; + break; + } + + case DIF_SUBR_GETMAJOR: +#ifndef VBOX +#ifdef _LP64 + regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR64) & MAXMAJ64; +#else + regs[rd] = (tupregs[0].dttk_value >> NBITSMINOR) & MAXMAJ; +#endif +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_GETMINOR: +#ifndef VBOX +#ifdef _LP64 + regs[rd] = tupregs[0].dttk_value & MAXMIN64; +#else + regs[rd] = tupregs[0].dttk_value & MAXMIN; +#endif +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + + case DIF_SUBR_DDI_PATHNAME: { +#ifndef VBOX + /* + * This one is a galactic mess. We are going to roughly + * emulate ddi_pathname(), but it's made more complicated + * by the fact that we (a) want to include the minor name and + * (b) must proceed iteratively instead of recursively. + */ + uintptr_t dest = mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + char *start = (char *)dest, *end = start + size - 1; + uintptr_t daddr = tupregs[0].dttk_value; + int64_t minor = (int64_t)tupregs[1].dttk_value; + char *s; + int i, len, depth = 0; + + /* + * Due to all the pointer jumping we do and context we must + * rely upon, we just mandate that the user must have kernel + * read privileges to use this routine. + */ + if ((mstate->dtms_access & DTRACE_ACCESS_KERNEL) == 0) { + *flags |= CPU_DTRACE_KPRIV; + *illval = daddr; + regs[rd] = NULL; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + *end = '\0'; + + /* + * We want to have a name for the minor. In order to do this, + * we need to walk the minor list from the devinfo. We want + * to be sure that we don't infinitely walk a circular list, + * so we check for circularity by sending a scout pointer + * ahead two elements for every element that we iterate over; + * if the list is circular, these will ultimately point to the + * same element. You may recognize this little trick as the + * answer to a stupid interview question -- one that always + * seems to be asked by those who had to have it laboriously + * explained to them, and who can't even concisely describe + * the conditions under which one would be forced to resort to + * this technique. Needless to say, those conditions are + * found here -- and probably only here. Is this the only use + * of this infamous trick in shipping, production code? If it + * isn't, it probably should be... + */ + if (minor != -1) { + uintptr_t maddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_minor)); + + uintptr_t next = offsetof(struct ddi_minor_data, next); + uintptr_t name = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, name); + uintptr_t dev = offsetof(struct ddi_minor_data, + d_minor) + offsetof(struct ddi_minor, dev); + uintptr_t scout; + + if (maddr != NULL) + scout = dtrace_loadptr(maddr + next); + + while (maddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + uint64_t m; +#ifdef _LP64 + m = dtrace_load64(maddr + dev) & MAXMIN64; +#else + m = dtrace_load32(maddr + dev) & MAXMIN; +#endif + if (m != minor) { + maddr = dtrace_loadptr(maddr + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + scout = dtrace_loadptr(scout + next); + + if (scout == NULL) + continue; + + if (scout == maddr) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + + continue; + } + + /* + * We have the minor data. Now we need to + * copy the minor's name into the end of the + * pathname. + */ + s = (char *)dtrace_loadptr(maddr + name); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = ':'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + break; + } + } + + while (daddr != NULL && !(*flags & CPU_DTRACE_FAULT)) { + ddi_node_state_t devi_state; + + devi_state = dtrace_load32(daddr + + offsetof(struct dev_info, devi_node_state)); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (devi_state >= DS_INITIALIZED) { + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_addr)); + len = dtrace_strlen(s, size); + + if (*flags & CPU_DTRACE_FAULT) + break; + + if (len != 0) { + if ((end -= (len + 1)) < start) + break; + + *end = '@'; + } + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + } + + /* + * Now for the node name... + */ + s = (char *)dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_node_name)); + + daddr = dtrace_loadptr(daddr + + offsetof(struct dev_info, devi_parent)); + + /* + * If our parent is NULL (that is, if we're the root + * node), we're going to use the special path + * "devices". + */ + if (daddr == NULL) + s = "devices"; + + len = dtrace_strlen(s, size); + if (*flags & CPU_DTRACE_FAULT) + break; + + if ((end -= (len + 1)) < start) + break; + + for (i = 1; i <= len; i++) + end[i] = dtrace_load8((uintptr_t)s++); + *end = '/'; + + if (depth++ > dtrace_devdepth_max) { + *flags |= CPU_DTRACE_ILLOP; + break; + } + } + + if (end < start) + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + + if (daddr == NULL) { + regs[rd] = (uintptr_t)end; + mstate->dtms_scratch_ptr += size; + } + +#else + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + break; + } + + case DIF_SUBR_STRJOIN: { + char *d = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t s1 = tupregs[0].dttk_value; + uintptr_t s2 = tupregs[1].dttk_value; + VBDTTYPE(unsigned,int) i = 0; + + if (!dtrace_strcanload(s1, size, mstate, vstate) || + !dtrace_strcanload(s2, size, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + if ((d[i++] = dtrace_load8(s1++)) == '\0') { + i--; + break; + } + } + + for (;;) { + if (i >= size) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + if ((d[i++] = dtrace_load8(s2++)) == '\0') + break; + } + + if (i < size) { + mstate->dtms_scratch_ptr += i; + regs[rd] = (uintptr_t)d; + } + + break; + } + + case DIF_SUBR_LLTOSTR: { + int64_t i = (int64_t)tupregs[0].dttk_value; + int64_t val = i < 0 ? i * -1 : i; + uint64_t size = 22; /* enough room for 2^64 in decimal */ + char *end = (char *)mstate->dtms_scratch_ptr + size - 1; + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + for (*end-- = '\0'; val; val /= 10) + *end-- = '0' + (val % 10); + + if (i == 0) + *end-- = '0'; + + if (i < 0) + *end-- = '-'; + + regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_HTONS: + case DIF_SUBR_NTOHS: +#ifdef _BIG_ENDIAN + regs[rd] = (uint16_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_16((uint16_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONL: + case DIF_SUBR_NTOHL: +#ifdef _BIG_ENDIAN + regs[rd] = (uint32_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_32((uint32_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_HTONLL: + case DIF_SUBR_NTOHLL: +#ifdef _BIG_ENDIAN + regs[rd] = (uint64_t)tupregs[0].dttk_value; +#else + regs[rd] = DT_BSWAP_64((uint64_t)tupregs[0].dttk_value); +#endif + break; + + + case DIF_SUBR_DIRNAME: + case DIF_SUBR_BASENAME: { + char *dest = (char *)mstate->dtms_scratch_ptr; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i, j, len = VBDTCAST(int)dtrace_strlen((char *)src, size); + int lastbase = -1, firstbase = -1, lastdir = -1; + int start, end; + + if (!dtrace_canload(src, len + 1, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + /* + * The basename and dirname for a zero-length string is + * defined to be "." + */ + if (len == 0) { + len = 1; + src = (uintptr_t)"."; + } + + /* + * Start from the back of the string, moving back toward the + * front until we see a character that isn't a slash. That + * character is the last character in the basename. + */ + for (i = len - 1; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastbase = i; + + /* + * Starting from the last character in the basename, move + * towards the front until we find a slash. The character + * that we processed immediately before that is the first + * character in the basename. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) == '/') + break; + } + + if (i >= 0) + firstbase = i + 1; + + /* + * Now keep going until we find a non-slash character. That + * character is the last character in the dirname. + */ + for (; i >= 0; i--) { + if (dtrace_load8(src + i) != '/') + break; + } + + if (i >= 0) + lastdir = i; + + ASSERT(!(lastbase == -1 && firstbase != -1)); + ASSERT(!(firstbase == -1 && lastdir != -1)); + + if (lastbase == -1) { + /* + * We didn't find a non-slash character. We know that + * the length is non-zero, so the whole string must be + * slashes. In either the dirname or the basename + * case, we return '/'. + */ + ASSERT(firstbase == -1); + firstbase = lastbase = lastdir = 0; + } + + if (firstbase == -1) { + /* + * The entire string consists only of a basename + * component. If we're looking for dirname, we need + * to change our string to be just "."; if we're + * looking for a basename, we'll just set the first + * character of the basename to be 0. + */ + if (subr == DIF_SUBR_DIRNAME) { + ASSERT(lastdir == -1); + src = (uintptr_t)"."; + lastdir = 0; + } else { + firstbase = 0; + } + } + + if (subr == DIF_SUBR_DIRNAME) { + if (lastdir == -1) { + /* + * We know that we have a slash in the name -- + * or lastdir would be set to 0, above. And + * because lastdir is -1, we know that this + * slash must be the first character. (That + * is, the full string must be of the form + * "/basename".) In this case, the last + * character of the directory name is 0. + */ + lastdir = 0; + } + + start = 0; + end = lastdir; + } else { + ASSERT(subr == DIF_SUBR_BASENAME); + ASSERT(firstbase != -1 && lastbase != -1); + start = firstbase; + end = lastbase; + } + + for (i = start, j = 0; i <= end && VBDTCAST(unsigned)j < size - 1; i++, j++) + dest[j] = dtrace_load8(src + i); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_CLEANPATH: { + char *dest = (char *)mstate->dtms_scratch_ptr, c; + uint64_t size = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t src = tupregs[0].dttk_value; + int i = 0, j = 0; + + if (!dtrace_strcanload(src, size, mstate, vstate)) { + regs[rd] = NULL; + break; + } + + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + /* + * Move forward, loading each character. + */ + do { + c = dtrace_load8(src + i++); +next: + if (j + 5 >= VBDTCAST(int64_t)size) /* 5 = strlen("/..c\0") */ + break; + + if (c != '/') { + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * We have two slashes -- we can just advance + * to the next character. + */ + goto next; + } + + if (c != '.') { + /* + * This is not "." and it's not ".." -- we can + * just store the "/" and this character and + * drive on. + */ + dest[j++] = '/'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c == '/') { + /* + * This is a "/./" component. We're not going + * to store anything in the destination buffer; + * we're just going to go to the next component. + */ + goto next; + } + + if (c != '.') { + /* + * This is not ".." -- we can just store the + * "/." and this character and continue + * processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + c = dtrace_load8(src + i++); + + if (c != '/' && c != '\0') { + /* + * This is not ".." -- it's "..[mumble]". + * We'll store the "/.." and this character + * and continue processing. + */ + dest[j++] = '/'; + dest[j++] = '.'; + dest[j++] = '.'; + dest[j++] = c; + continue; + } + + /* + * This is "/../" or "/..\0". We need to back up + * our destination pointer until we find a "/". + */ + i--; + while (j != 0 && dest[--j] != '/') + continue; + + if (c == '\0') + dest[++j] = '/'; + } while (c != '\0'); + + dest[j] = '\0'; + regs[rd] = (uintptr_t)dest; + mstate->dtms_scratch_ptr += size; + break; + } + + case DIF_SUBR_INET_NTOA: + case DIF_SUBR_INET_NTOA6: + case DIF_SUBR_INET_NTOP: { +#ifndef VBOX + size_t size; + int af, argi, i; + char *base, *end; + + if (subr == DIF_SUBR_INET_NTOP) { + af = (int)tupregs[0].dttk_value; + argi = 1; + } else { + af = subr == DIF_SUBR_INET_NTOA ? AF_INET: AF_INET6; + argi = 0; + } + + if (af == AF_INET) { + ipaddr_t ip4; + uint8_t *ptr8, val; + + /* + * Safely load the IPv4 address. + */ + ip4 = dtrace_load32(tupregs[argi].dttk_value); + + /* + * Check an IPv4 string will fit in scratch. + */ + size = INET_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + + /* + * Stringify as a dotted decimal quad. + */ + *end-- = '\0'; + ptr8 = (uint8_t *)&ip4; + for (i = 3; i >= 0; i--) { + val = ptr8[i]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + (val % 10); + } + } + + if (i > 0) + *end-- = '.'; + } + ASSERT(end + 1 >= base); + + } else if (af == AF_INET6) { + struct in6_addr ip6; + int firstzero, tryzero, numzero, v6end; + uint16_t val; + const char digits[] = "0123456789abcdef"; + + /* + * Stringify using RFC 1884 convention 2 - 16 bit + * hexadecimal values with a zero-run compression. + * Lower case hexadecimal digits are used. + * eg, fe80::214:4fff:fe0b:76c8. + * The IPv4 embedded form is returned for inet_ntop, + * just the IPv4 string is returned for inet_ntoa6. + */ + + /* + * Safely load the IPv6 address. + */ + dtrace_bcopy( + (void *)(uintptr_t)tupregs[argi].dttk_value, + (void *)(uintptr_t)&ip6, sizeof (struct in6_addr)); + + /* + * Check an IPv6 string will fit in scratch. + */ + size = INET6_ADDRSTRLEN; + if (!DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + base = (char *)mstate->dtms_scratch_ptr; + end = (char *)mstate->dtms_scratch_ptr + size - 1; + *end-- = '\0'; + + /* + * Find the longest run of 16 bit zero values + * for the single allowed zero compression - "::". + */ + firstzero = -1; + tryzero = -1; + numzero = 1; + for (i = 0; i < sizeof (struct in6_addr); i++) { + if (ip6._S6_un._S6_u8[i] == 0 && + tryzero == -1 && i % 2 == 0) { + tryzero = i; + continue; + } + + if (tryzero != -1 && + (ip6._S6_un._S6_u8[i] != 0 || + i == sizeof (struct in6_addr) - 1)) { + + if (i - tryzero <= numzero) { + tryzero = -1; + continue; + } + + firstzero = tryzero; + numzero = i - i % 2 - tryzero; + tryzero = -1; + + if (ip6._S6_un._S6_u8[i] == 0 && + i == sizeof (struct in6_addr) - 1) + numzero += 2; + } + } + ASSERT(firstzero + numzero <= sizeof (struct in6_addr)); + + /* + * Check for an IPv4 embedded address. + */ + v6end = sizeof (struct in6_addr) - 2; + if (IN6_IS_ADDR_V4MAPPED(&ip6) || + IN6_IS_ADDR_V4COMPAT(&ip6)) { + for (i = sizeof (struct in6_addr) - 1; + i >= DTRACE_V4MAPPED_OFFSET; i--) { + ASSERT(end >= base); + + val = ip6._S6_un._S6_u8[i]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 10) { + *end-- = '0' + val % 10; + } + } + + if (i > DTRACE_V4MAPPED_OFFSET) + *end-- = '.'; + } + + if (subr == DIF_SUBR_INET_NTOA6) + goto inetout; + + /* + * Set v6end to skip the IPv4 address that + * we have already stringified. + */ + v6end = 10; + } + + /* + * Build the IPv6 string by working through the + * address in reverse. + */ + for (i = v6end; i >= 0; i -= 2) { + ASSERT(end >= base); + + if (i == firstzero + numzero - 2) { + *end-- = ':'; + *end-- = ':'; + i -= numzero - 2; + continue; + } + + if (i < 14 && i != firstzero - 2) + *end-- = ':'; + + val = (ip6._S6_un._S6_u8[i] << 8) + + ip6._S6_un._S6_u8[i + 1]; + + if (val == 0) { + *end-- = '0'; + } else { + for (; val; val /= 16) { + *end-- = digits[val % 16]; + } + } + } + ASSERT(end + 1 >= base); + + } else { + /* + * The user didn't use AH_INET or AH_INET6. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + regs[rd] = NULL; + break; + } + +inetout: regs[rd] = (uintptr_t)end + 1; + mstate->dtms_scratch_ptr += size; +#else /* VBOX */ + regs[rd] = 0; + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif /* VBOX */ + break; + } + + } +} + +/* + * Emulate the execution of DTrace IR instructions specified by the given + * DIF object. This function is deliberately void of assertions as all of + * the necessary checks are handled by a call to dtrace_difo_validate(). + */ +static uint64_t +dtrace_dif_emulate(dtrace_difo_t *difo, dtrace_mstate_t *mstate, + dtrace_vstate_t *vstate, dtrace_state_t *state) +{ + const dif_instr_t *text = difo->dtdo_buf; + const uint_t textlen = difo->dtdo_len; + const char *strtab = difo->dtdo_strtab; + const uint64_t *inttab = difo->dtdo_inttab; + + uint64_t rval = 0; + dtrace_statvar_t *svar; + dtrace_dstate_t *dstate = &vstate->dtvs_dynvars; + dtrace_difv_t *v; + volatile uint16_t *flags = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + volatile uintptr_t *illval = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval; + + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ + uint64_t regs[DIF_DIR_NREGS]; + uint64_t *tmp; + + uint8_t cc_n = 0, cc_z = 0, cc_v = 0, cc_c = 0; + int64_t cc_r; + uint_t pc = 0, id, opc VBDTUNASS(0); + uint8_t ttop = 0; + dif_instr_t instr; + uint_t r1, r2, rd; + + /* + * We stash the current DIF object into the machine state: we need it + * for subsequent access checking. + */ + mstate->dtms_difo = difo; + + regs[DIF_REG_R0] = 0; /* %r0 is fixed at zero */ + + while (pc < textlen && !(*flags & CPU_DTRACE_FAULT)) { + opc = pc; + + instr = text[pc++]; + r1 = DIF_INSTR_R1(instr); + r2 = DIF_INSTR_R2(instr); + rd = DIF_INSTR_RD(instr); + + switch (DIF_INSTR_OP(instr)) { + case DIF_OP_OR: + regs[rd] = regs[r1] | regs[r2]; + break; + case DIF_OP_XOR: + regs[rd] = regs[r1] ^ regs[r2]; + break; + case DIF_OP_AND: + regs[rd] = regs[r1] & regs[r2]; + break; + case DIF_OP_SLL: + regs[rd] = regs[r1] << regs[r2]; + break; + case DIF_OP_SRL: + regs[rd] = regs[r1] >> regs[r2]; + break; + case DIF_OP_SUB: + regs[rd] = regs[r1] - regs[r2]; + break; + case DIF_OP_ADD: + regs[rd] = regs[r1] + regs[r2]; + break; + case DIF_OP_MUL: + regs[rd] = regs[r1] * regs[r2]; + break; + case DIF_OP_SDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = (int64_t)regs[r1] / + (int64_t)regs[r2]; + } + break; + + case DIF_OP_UDIV: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = regs[r1] / regs[r2]; + } + break; + + case DIF_OP_SREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = (int64_t)regs[r1] % + (int64_t)regs[r2]; + } + break; + + case DIF_OP_UREM: + if (regs[r2] == 0) { + regs[rd] = 0; + *flags |= CPU_DTRACE_DIVZERO; + } else { + regs[rd] = regs[r1] % regs[r2]; + } + break; + + case DIF_OP_NOT: + regs[rd] = ~regs[r1]; + break; + case DIF_OP_MOV: + regs[rd] = regs[r1]; + break; + case DIF_OP_CMP: + cc_r = regs[r1] - regs[r2]; + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = 0; + cc_c = regs[r1] < regs[r2]; + break; + case DIF_OP_TST: + cc_n = cc_v = cc_c = 0; + cc_z = regs[r1] == 0; + break; + case DIF_OP_BA: + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BE: + if (cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BNE: + if (cc_z == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BG: + if ((cc_z | (cc_n ^ cc_v)) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGU: + if ((cc_c | cc_z) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGE: + if ((cc_n ^ cc_v) == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BGEU: + if (cc_c == 0) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BL: + if (cc_n ^ cc_v) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLU: + if (cc_c) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLE: + if (cc_z | (cc_n ^ cc_v)) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_BLEU: + if (cc_c | cc_z) + pc = DIF_INSTR_LABEL(instr); + break; + case DIF_OP_RLDSB: + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDSB: + regs[rd] = (int8_t)dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDSH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDSH: + regs[rd] = (int16_t)dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDSW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDSW: + regs[rd] = (int32_t)dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDUB: + if (!dtrace_canstore(regs[r1], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDUB: + regs[rd] = dtrace_load8(regs[r1]); + break; + case DIF_OP_RLDUH: + if (!dtrace_canstore(regs[r1], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDUH: + regs[rd] = dtrace_load16(regs[r1]); + break; + case DIF_OP_RLDUW: + if (!dtrace_canstore(regs[r1], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDUW: + regs[rd] = dtrace_load32(regs[r1]); + break; + case DIF_OP_RLDX: + if (!dtrace_canstore(regs[r1], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_KPRIV; + *illval = regs[r1]; + break; + } + RT_FALL_THRU(); + case DIF_OP_LDX: + regs[rd] = dtrace_load64(regs[r1]); + break; + case DIF_OP_ULDSB: + regs[rd] = (int8_t) + dtrace_fuword8((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSH: + regs[rd] = (int16_t) + dtrace_fuword16((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDSW: + regs[rd] = (int32_t) + dtrace_fuword32((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUB: + regs[rd] = + dtrace_fuword8((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUH: + regs[rd] = + dtrace_fuword16((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDUW: + regs[rd] = + dtrace_fuword32((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_ULDX: + regs[rd] = + dtrace_fuword64((void *)(uintptr_t)regs[r1]); + break; + case DIF_OP_RET: + rval = regs[rd]; + pc = textlen; + break; + case DIF_OP_NOP: + break; + case DIF_OP_SETX: + regs[rd] = inttab[DIF_INSTR_INTEGER(instr)]; + break; + case DIF_OP_SETS: + regs[rd] = (uint64_t)(uintptr_t) + (strtab + DIF_INSTR_STRING(instr)); + break; + case DIF_OP_SCMP: { + size_t sz = state->dts_options[DTRACEOPT_STRSIZE]; + uintptr_t s1 = regs[r1]; + uintptr_t s2 = regs[r2]; + + if (s1 != NULL && + !dtrace_strcanload(s1, sz, mstate, vstate)) + break; + if (s2 != NULL && + !dtrace_strcanload(s2, sz, mstate, vstate)) + break; + + cc_r = dtrace_strncmp((char *)s1, (char *)s2, sz); + + cc_n = cc_r < 0; + cc_z = cc_r == 0; + cc_v = cc_c = 0; + break; + } + case DIF_OP_LDGA: + regs[rd] = dtrace_dif_variable(mstate, state, + r1, regs[r2]); + break; + case DIF_OP_LDGS: + id = DIF_INSTR_VAR(instr); + + if (id >= DIF_VAR_OTHER_UBASE) { + uintptr_t a; + + id -= DIF_VAR_OTHER_UBASE; + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (!(v->dtdv_type.dtdt_flags & DIF_TF_BYREF)) { + regs[rd] = svar->dtsv_data; + break; + } + + a = (uintptr_t)svar->dtsv_data; + + if (*(uint8_t *)a == UINT8_MAX) { + /* + * If the 0th byte is set to UINT8_MAX + * then this is to be treated as a + * reference to a NULL variable. + */ + regs[rd] = NULL; + } else { + regs[rd] = a + sizeof (uint64_t); + } + + break; + } + + regs[rd] = dtrace_dif_variable(mstate, state, id, 0); + break; + + case DIF_OP_STGS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + svar = vstate->dtvs_globals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + + ASSERT(a != NULL); + ASSERT(svar->dtsv_size != 0); + + if (regs[rd] == NULL) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof (uint64_t); + } + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + svar->dtsv_data = regs[rd]; + break; + + case DIF_OP_LDTA: + /* + * There are no DTrace built-in thread-local arrays at + * present. This opcode is saved for future work. + */ + *flags |= CPU_DTRACE_ILLOP; + regs[rd] = 0; + break; + + case DIF_OP_LDLS: + id = DIF_INSTR_VAR(instr); + + if (id < DIF_VAR_OTHER_UBASE) { + /* + * For now, this has no meaning. + */ + regs[rd] = 0; + break; + } + + id -= DIF_VAR_OTHER_UBASE; + + ASSERT(VBDTCAST(int64_t)id < vstate->dtvs_nlocals); + ASSERT(vstate->dtvs_locals != NULL); + + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof (uint64_t); + ASSERT(svar->dtsv_size == NCPU * sz); + a += VBDT_GET_CPUID() * sz; + + if (*(uint8_t *)a == UINT8_MAX) { + /* + * If the 0th byte is set to UINT8_MAX + * then this is to be treated as a + * reference to a NULL variable. + */ + regs[rd] = NULL; + } else { + regs[rd] = a + sizeof (uint64_t); + } + + break; + } + + ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + regs[rd] = tmp[VBDT_GET_CPUID()]; + break; + + case DIF_OP_STLS: + id = DIF_INSTR_VAR(instr); + + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + ASSERT(VBDTCAST(int64_t)id < vstate->dtvs_nlocals); + + ASSERT(vstate->dtvs_locals != NULL); + svar = vstate->dtvs_locals[id]; + ASSERT(svar != NULL); + v = &svar->dtsv_var; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + uintptr_t a = (uintptr_t)svar->dtsv_data; + size_t sz = v->dtdv_type.dtdt_size; + + sz += sizeof (uint64_t); + ASSERT(svar->dtsv_size == NCPU * sz); + a += VBDT_GET_CPUID() * sz; + + if (regs[rd] == NULL) { + *(uint8_t *)a = UINT8_MAX; + break; + } else { + *(uint8_t *)a = 0; + a += sizeof (uint64_t); + } + + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + (void *)a, &v->dtdv_type); + break; + } + + ASSERT(svar->dtsv_size == NCPU * sizeof (uint64_t)); + tmp = (uint64_t *)(uintptr_t)svar->dtsv_data; + tmp[VBDT_GET_CPUID()] = regs[rd]; + break; + + case DIF_OP_LDTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + v = &vstate->dtvs_tlocals[id]; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + + dvar = dtrace_dynvar(dstate, 2, key, + sizeof (uint64_t), DTRACE_DYNVAR_NOALLOC, + mstate, vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + } else { + regs[rd] = *((uint64_t *)dvar->dtdv_data); + } + + break; + } + + case DIF_OP_STTS: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_value = (uint64_t)id; + key[0].dttk_size = 0; + DTRACE_TLS_THRKEY(key[1].dttk_value); + key[1].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + + dvar = dtrace_dynvar(dstate, 2, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + regs[rd] ? DTRACE_DYNVAR_ALLOC : + DTRACE_DYNVAR_DEALLOC, mstate, vstate); + + /* + * Given that we're storing to thread-local data, + * we need to flush our predicate cache. + */ + curthread->t_predcache = NULL; + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], + &v->dtdv_type, mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else { + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + } + + break; + } + + case DIF_OP_SRA: + regs[rd] = (int64_t)regs[r1] >> regs[r2]; + break; + + case DIF_OP_CALL: + dtrace_dif_subr(DIF_INSTR_SUBR(instr), rd, + regs, tupregs, ttop, mstate, state); + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + if (r1 == DIF_TYPE_STRING) { + /* + * If this is a string type and the size is 0, + * we'll use the system-wide default string + * size. Note that we are _not_ looking at + * the value of the DTRACEOPT_STRSIZE option; + * had this been set, we would expect to have + * a non-zero size value in the "pushtr". + */ + tupregs[ttop].dttk_size = + dtrace_strlen((char *)(uintptr_t)regs[rd], + regs[r2] ? regs[r2] : + dtrace_strsize_default) + 1; + } else { + tupregs[ttop].dttk_size = regs[r2]; + } + + tupregs[ttop++].dttk_value = regs[rd]; + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) { + *flags |= CPU_DTRACE_TUPOFLOW; + break; + } + + tupregs[ttop].dttk_value = regs[rd]; + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_LDTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else { + v = &vstate->dtvs_globals[id]->dtsv_var; + } + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + DTRACE_DYNVAR_NOALLOC, mstate, vstate); + + if (dvar == NULL) { + regs[rd] = 0; + break; + } + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + regs[rd] = (uint64_t)(uintptr_t)dvar->dtdv_data; + } else { + regs[rd] = *((uint64_t *)dvar->dtdv_data); + } + + break; + } + + case DIF_OP_STGAA: + case DIF_OP_STTAA: { + dtrace_dynvar_t *dvar; + dtrace_key_t *key = tupregs; + uint_t nkeys = ttop; + + id = DIF_INSTR_VAR(instr); + ASSERT(id >= DIF_VAR_OTHER_UBASE); + id -= DIF_VAR_OTHER_UBASE; + + key[nkeys].dttk_value = (uint64_t)id; + key[nkeys++].dttk_size = 0; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) { + DTRACE_TLS_THRKEY(key[nkeys].dttk_value); + key[nkeys++].dttk_size = 0; + v = &vstate->dtvs_tlocals[id]; + } else { + v = &vstate->dtvs_globals[id]->dtsv_var; + } + + dvar = dtrace_dynvar(dstate, nkeys, key, + v->dtdv_type.dtdt_size > sizeof (uint64_t) ? + v->dtdv_type.dtdt_size : sizeof (uint64_t), + regs[rd] ? DTRACE_DYNVAR_ALLOC : + DTRACE_DYNVAR_DEALLOC, mstate, vstate); + + if (dvar == NULL) + break; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) { + if (!dtrace_vcanload( + (void *)(uintptr_t)regs[rd], &v->dtdv_type, + mstate, vstate)) + break; + + dtrace_vcopy((void *)(uintptr_t)regs[rd], + dvar->dtdv_data, &v->dtdv_type); + } else { + *((uint64_t *)dvar->dtdv_data) = regs[rd]; + } + + break; + } + + case DIF_OP_ALLOCS: { + uintptr_t ptr = P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size_t size = ptr - mstate->dtms_scratch_ptr + regs[r1]; + + /* + * Rounding up the user allocation size could have + * overflowed large, bogus allocations (like -1ULL) to + * 0. + */ + if (size < regs[r1] || + !DTRACE_INSCRATCH(mstate, size)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + regs[rd] = NULL; + break; + } + + dtrace_bzero((void *) mstate->dtms_scratch_ptr, size); + mstate->dtms_scratch_ptr += size; + regs[rd] = ptr; + break; + } + + case DIF_OP_COPYS: + if (!dtrace_canstore(regs[rd], regs[r2], + mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + + if (!dtrace_canload(regs[r1], regs[r2], mstate, vstate)) + break; + + dtrace_bcopy((void *)(uintptr_t)regs[r1], + (void *)(uintptr_t)regs[rd], (size_t)regs[r2]); + break; + + case DIF_OP_STB: + if (!dtrace_canstore(regs[rd], 1, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + *((uint8_t *)(uintptr_t)regs[rd]) = (uint8_t)regs[r1]; + break; + + case DIF_OP_STH: + if (!dtrace_canstore(regs[rd], 2, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 1) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint16_t *)(uintptr_t)regs[rd]) = (uint16_t)regs[r1]; + break; + + case DIF_OP_STW: + if (!dtrace_canstore(regs[rd], 4, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 3) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint32_t *)(uintptr_t)regs[rd]) = (uint32_t)regs[r1]; + break; + + case DIF_OP_STX: + if (!dtrace_canstore(regs[rd], 8, mstate, vstate)) { + *flags |= CPU_DTRACE_BADADDR; + *illval = regs[rd]; + break; + } + if (regs[rd] & 7) { + *flags |= CPU_DTRACE_BADALIGN; + *illval = regs[rd]; + break; + } + *((uint64_t *)(uintptr_t)regs[rd]) = regs[r1]; + break; + } + } + + if (!(*flags & CPU_DTRACE_FAULT)) + return (rval); + + mstate->dtms_fltoffs = opc * sizeof (dif_instr_t); + mstate->dtms_present |= DTRACE_MSTATE_FLTOFFS; + + return (0); +} + +#ifndef VBOX /* no destructive stuff */ + +static void +dtrace_action_breakpoint(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + dtrace_provider_t *prov = probe->dtpr_provider; + char c[DTRACE_FULLNAMELEN + 80], *str; + char *msg = "dtrace: breakpoint action at probe "; + char *ecbmsg = " (ecb "; + uintptr_t mask = (0xf << (sizeof (uintptr_t) * NBBY / 4)); + uintptr_t val = (uintptr_t)ecb; + int shift = (sizeof (uintptr_t) * NBBY) - 4, i = 0; + + if (dtrace_destructive_disallow) + return; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + /* + * This is a poor man's (destitute man's?) sprintf(): we want to + * print the provider name, module name, function name and name of + * the probe, along with the hex address of the ECB with the breakpoint + * action -- all of which we must place in the character buffer by + * hand. + */ + while (*msg != '\0') + c[i++] = *msg++; + + for (str = prov->dtpv_name; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_mod; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_func; *str != '\0'; str++) + c[i++] = *str; + c[i++] = ':'; + + for (str = probe->dtpr_name; *str != '\0'; str++) + c[i++] = *str; + + while (*ecbmsg != '\0') + c[i++] = *ecbmsg++; + + while (shift >= 0) { + mask = (uintptr_t)0xf << shift; + + if (val >= ((uintptr_t)1 << shift)) + c[i++] = "0123456789abcdef"[(val & mask) >> shift]; + shift -= 4; + } + + c[i++] = ')'; + c[i] = '\0'; + + debug_enter(c); +} + +static void +dtrace_action_panic(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + /* + * It's impossible to be taking action on the NULL probe. + */ + ASSERT(probe != NULL); + + if (dtrace_destructive_disallow) + return; + + if (dtrace_panicked != NULL) + return; + + if (dtrace_casptr(&dtrace_panicked, NULL, curthread) != NULL) + return; + + /* + * We won the right to panic. (We want to be sure that only one + * thread calls panic() from dtrace_probe(), and that panic() is + * called exactly once.) + */ + dtrace_panic("dtrace: panic action at probe %s:%s:%s:%s (ecb %p)", + probe->dtpr_provider->dtpv_name, probe->dtpr_mod, + probe->dtpr_func, probe->dtpr_name, (void *)ecb); +} + +static void +dtrace_action_raise(uint64_t sig) +{ + if (dtrace_destructive_disallow) + return; + + if (sig >= NSIG) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return; + } + + /* + * raise() has a queue depth of 1 -- we ignore all subsequent + * invocations of the raise() action. + */ + if (curthread->t_dtrace_sig == 0) + curthread->t_dtrace_sig = (uint8_t)sig; + + curthread->t_sig_check = 1; + aston(curthread); +} + +static void +dtrace_action_stop(void) +{ + if (dtrace_destructive_disallow) + return; + + if (!curthread->t_dtrace_stop) { + curthread->t_dtrace_stop = 1; + curthread->t_sig_check = 1; + aston(curthread); + } +} + +static void +dtrace_action_chill(dtrace_mstate_t *mstate, hrtime_t val) +{ + hrtime_t now; + volatile uint16_t *flags; + cpu_t *cpu = CPU; + + if (dtrace_destructive_disallow) + return; + + flags = (volatile uint16_t *)&cpu_core[cpu->cpu_id].cpuc_dtrace_flags; + + now = dtrace_gethrtime(); + + if (now - cpu->cpu_dtrace_chillmark > dtrace_chill_interval) { + /* + * We need to advance the mark to the current time. + */ + cpu->cpu_dtrace_chillmark = now; + cpu->cpu_dtrace_chilled = 0; + } + + /* + * Now check to see if the requested chill time would take us over + * the maximum amount of time allowed in the chill interval. (Or + * worse, if the calculation itself induces overflow.) + */ + if (cpu->cpu_dtrace_chilled + val > dtrace_chill_max || + cpu->cpu_dtrace_chilled + val < cpu->cpu_dtrace_chilled) { + *flags |= CPU_DTRACE_ILLOP; + return; + } + + while (dtrace_gethrtime() - now < val) + continue; + + /* + * Normally, we assure that the value of the variable "timestamp" does + * not change within an ECB. The presence of chill() represents an + * exception to this rule, however. + */ + mstate->dtms_present &= ~DTRACE_MSTATE_TIMESTAMP; + cpu->cpu_dtrace_chilled += val; +} + +#endif /* !VBOX */ + +#ifndef VBOX +static void +dtrace_action_ustack(dtrace_mstate_t *mstate, dtrace_state_t *state, + uint64_t *buf, uint64_t arg) +{ + int nframes = DTRACE_USTACK_NFRAMES(arg); + int strsize = DTRACE_USTACK_STRSIZE(arg); + uint64_t *pcs = &buf[1], *fps; + char *str = (char *)&pcs[nframes]; + int size, offs = 0, i, j; + uintptr_t old = mstate->dtms_scratch_ptr, saved; +#ifndef VBOX + uint16_t *flags = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; +#else + uint16_t volatile *flags = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; +#endif + char *sym; + + /* + * Should be taking a faster path if string space has not been + * allocated. + */ + ASSERT(strsize != 0); + + /* + * We will first allocate some temporary space for the frame pointers. + */ + fps = (uint64_t *)P2ROUNDUP(mstate->dtms_scratch_ptr, 8); + size = (uintptr_t)fps - mstate->dtms_scratch_ptr + + (nframes * sizeof (uint64_t)); + + if (!DTRACE_INSCRATCH(mstate, VBDTCAST(unsigned)size)) { + /* + * Not enough room for our frame pointers -- need to indicate + * that we ran out of scratch space. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOSCRATCH); + return; + } + + mstate->dtms_scratch_ptr += size; + saved = mstate->dtms_scratch_ptr; + + /* + * Now get a stack with both program counters and frame pointers. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getufpstack(buf, fps, nframes + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + /* + * If that faulted, we're cooked. + */ + if (*flags & CPU_DTRACE_FAULT) + goto out; + + /* + * Now we want to walk up the stack, calling the USTACK helper. For + * each iteration, we restore the scratch pointer. + */ + for (i = 0; i < nframes; i++) { + mstate->dtms_scratch_ptr = saved; + + if (offs >= strsize) + break; + +#ifndef VBOX + sym = (char *)(uintptr_t)dtrace_helper( + DTRACE_HELPER_ACTION_USTACK, + mstate, state, pcs[i], fps[i]); +#else + sym = NULL; +#endif + + /* + * If we faulted while running the helper, we're going to + * clear the fault and null out the corresponding string. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + str[offs++] = '\0'; + continue; + } + + if (sym == NULL) { + str[offs++] = '\0'; + continue; + } + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + + /* + * Now copy in the string that the helper returned to us. + */ + for (j = 0; offs + j < strsize; j++) { + if ((str[offs + j] = sym[j]) == '\0') + break; + } + + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + offs += j + 1; + } + + if (offs >= strsize) { + /* + * If we didn't have room for all of the strings, we don't + * abort processing -- this needn't be a fatal error -- but we + * still want to increment a counter (dts_stkstroverflows) to + * allow this condition to be warned about. (If this is from + * a jstack() action, it is easily tuned via jstackstrsize.) + */ + dtrace_error(&state->dts_stkstroverflows); + } + + while (offs < strsize) + str[offs++] = '\0'; + +out: + mstate->dtms_scratch_ptr = old; +} +#endif /* !VBOX */ + +#ifdef VBOX +extern void dtrace_probe6(dtrace_id_t, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, uintptr_t arg5); +# define dtrace_probe_error(a1, a2, a3, a4, a5, a6) \ + dtrace_probe6(dtrace_probeid_error, (uintptr_t)a1, a2, a3, a4, a5, a6) +#endif + +/* + * If you're looking for the epicenter of DTrace, you just found it. This + * is the function called by the provider to fire a probe -- from which all + * subsequent probe-context DTrace activity emanates. + */ +void +dtrace_probe(dtrace_id_t id, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) +{ + processorid_t cpuid; + dtrace_icookie_t cookie; + dtrace_probe_t *probe; + dtrace_mstate_t mstate; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + intptr_t offs; + size_t size; + int vtime, onintr; + volatile uint16_t *flags; + hrtime_t now; + +#ifndef VBOX + /* + * Kick out immediately if this CPU is still being born (in which case + * curthread will be set to -1) or the current thread can't allow + * probes in its current context. + */ + if (((uintptr_t)curthread & 1) || (curthread->t_flag & T_DONTDTRACE)) + return; +#endif + + cookie = dtrace_interrupt_disable(); + probe = dtrace_probes[id - 1]; + cpuid = VBDT_GET_CPUID(); + onintr = CPU_ON_INTR(CPU); + + if (!onintr && probe->dtpr_predcache != DTRACE_CACHEIDNONE && + probe->dtpr_predcache == curthread->t_predcache) { + /* + * We have hit in the predicate cache; we know that + * this predicate would evaluate to be false. + */ + dtrace_interrupt_enable(cookie); + return; + } + +#ifndef VBOX + if (panic_quiesce) { + /* + * We don't trace anything if we're panicking. + */ + dtrace_interrupt_enable(cookie); + return; + } +#endif + + now = dtrace_gethrtime(); + vtime = dtrace_vtime_references != 0; + + if (vtime && curthread->t_dtrace_start) + curthread->t_dtrace_vtime += now - curthread->t_dtrace_start; + + mstate.dtms_difo = NULL; + mstate.dtms_probe = probe; + mstate.dtms_strtok = NULL; + mstate.dtms_arg[0] = arg0; + mstate.dtms_arg[1] = arg1; + mstate.dtms_arg[2] = arg2; + mstate.dtms_arg[3] = arg3; + mstate.dtms_arg[4] = arg4; + + flags = (volatile uint16_t *)&cpu_core[cpuid].cpuc_dtrace_flags; + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + dtrace_predicate_t *pred = ecb->dte_predicate; + dtrace_state_t *state = ecb->dte_state; + dtrace_buffer_t *buf = &state->dts_buffer[cpuid]; + dtrace_buffer_t *aggbuf = &state->dts_aggbuffer[cpuid]; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_provider_t *prov = probe->dtpr_provider; + int committed = 0; + caddr_t tomax; + + /* + * A little subtlety with the following (seemingly innocuous) + * declaration of the automatic 'val': by looking at the + * code, you might think that it could be declared in the + * action processing loop, below. (That is, it's only used in + * the action processing loop.) However, it must be declared + * out of that scope because in the case of DIF expression + * arguments to aggregating actions, one iteration of the + * action loop will use the last iteration's value. + */ +#ifdef lint + uint64_t val = 0; +#else + uint64_t val VBDTUNASS(0); +#endif + + mstate.dtms_present = DTRACE_MSTATE_ARGS | DTRACE_MSTATE_PROBE; + *flags &= ~CPU_DTRACE_ERROR; + + if (prov == dtrace_provider) { + /* + * If dtrace itself is the provider of this probe, + * we're only going to continue processing the ECB if + * arg0 (the dtrace_state_t) is equal to the ECB's + * creating state. (This prevents disjoint consumers + * from seeing one another's metaprobes.) + */ + if (arg0 != (uint64_t)(uintptr_t)state) + continue; + } + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) { + /* + * We're not currently active. If our provider isn't + * the dtrace pseudo provider, we're not interested. + */ + if (prov != dtrace_provider) + continue; + + /* + * Now we must further check if we are in the BEGIN + * probe. If we are, we will only continue processing + * if we're still in WARMUP -- if one BEGIN enabling + * has invoked the exit() action, we don't want to + * evaluate subsequent BEGIN enablings. + */ + if (probe->dtpr_id == dtrace_probeid_begin && + state->dts_activity != DTRACE_ACTIVITY_WARMUP) { + ASSERT(state->dts_activity == + DTRACE_ACTIVITY_DRAINING); + continue; + } + } + + if (ecb->dte_cond) { + /* + * If the dte_cond bits indicate that this + * consumer is only allowed to see user-mode firings + * of this probe, call the provider's dtps_usermode() + * entry point to check that the probe was fired + * while in a user context. Skip this ECB if that's + * not the case. + */ + if ((ecb->dte_cond & DTRACE_COND_USERMODE) && + prov->dtpv_pops.dtps_usermode(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg) == 0) + continue; + + /* + * This is more subtle than it looks. We have to be + * absolutely certain that CRED() isn't going to + * change out from under us so it's only legit to + * examine that structure if we're in constrained + * situations. Currently, the only times we'll this + * check is if a non-super-user has enabled the + * profile or syscall providers -- providers that + * allow visibility of all processes. For the + * profile case, the check above will ensure that + * we're examining a user context. + */ + if (ecb->dte_cond & DTRACE_COND_OWNER) { + cred_t *cr; + cred_t *s_cr = + ecb->dte_state->dts_cred.dcr_cred; +#ifndef VBOX + proc_t *proc; +#endif + + ASSERT(s_cr != NULL); + + if ((cr = CRED()) == NULL || + s_cr->cr_uid != cr->cr_uid || + s_cr->cr_uid != cr->cr_ruid || + s_cr->cr_uid != cr->cr_suid || + s_cr->cr_gid != cr->cr_gid || + s_cr->cr_gid != cr->cr_rgid || + s_cr->cr_gid != cr->cr_sgid || +#ifndef VBOX + (proc = VBDT_GET_PROC()) == NULL || + (proc->p_flag & SNOCD)) +#else + 0) + +#endif + continue; + } + +#ifndef VBOX + if (ecb->dte_cond & DTRACE_COND_ZONEOWNER) { + cred_t *cr; + cred_t *s_cr = + ecb->dte_state->dts_cred.dcr_cred; + + ASSERT(s_cr != NULL); + + if ((cr = CRED()) == NULL || + s_cr->cr_zone->zone_id != + cr->cr_zone->zone_id) + continue; + } +#endif + } + + if (now - state->dts_alive > dtrace_deadman_timeout) { + /* + * We seem to be dead. Unless we (a) have kernel + * destructive permissions (b) have expicitly enabled + * destructive actions and (c) destructive actions have + * not been disabled, we're going to transition into + * the KILLED state, from which no further processing + * on this state will be performed. + */ + if (!dtrace_priv_kernel_destructive(state) || + !state->dts_cred.dcr_destructive || + dtrace_destructive_disallow) { + void *activity = &state->dts_activity; + dtrace_activity_t current; + + do { + current = state->dts_activity; + } while ( (dtrace_activity_t)dtrace_cas32(activity, current, DTRACE_ACTIVITY_KILLED) + != current); + + continue; + } + } + + if ((offs = dtrace_buffer_reserve(buf, ecb->dte_needed, + ecb->dte_alignment, state, &mstate)) < 0) + continue; + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, ecb->dte_epid); + + mstate.dtms_epid = ecb->dte_epid; + mstate.dtms_present |= DTRACE_MSTATE_EPID; + + if (state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) + mstate.dtms_access = DTRACE_ACCESS_KERNEL; + else + mstate.dtms_access = 0; + + if (pred != NULL) { + dtrace_difo_t *dp = pred->dtp_difo; + int rval; + + rval = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (!(*flags & CPU_DTRACE_ERROR) && !rval) { + dtrace_cacheid_t cid = probe->dtpr_predcache; + + if (cid != DTRACE_CACHEIDNONE && !onintr) { + /* + * Update the predicate cache... + */ + ASSERT(cid == pred->dtp_cacheid); + curthread->t_predcache = cid; + } + + continue; + } + } + + for (act = ecb->dte_action; !(*flags & CPU_DTRACE_ERROR) && + act != NULL; act = act->dta_next) { + size_t valoffs; + dtrace_difo_t *dp; + dtrace_recdesc_t *rec = &act->dta_rec; + + size = rec->dtrd_size; + valoffs = offs + rec->dtrd_offset; + + if (DTRACEACT_ISAGG(act->dta_kind)) { + uint64_t v = 0xbad; + dtrace_aggregation_t *agg; + + agg = (dtrace_aggregation_t *)act; + + if ((dp = act->dta_difo) != NULL) + v = dtrace_dif_emulate(dp, + &mstate, vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + /* + * Note that we always pass the expression + * value from the previous iteration of the + * action loop. This value will only be used + * if there is an expression argument to the + * aggregating action, denoted by the + * dtag_hasarg field. + */ + dtrace_aggregate(agg, buf, + offs, aggbuf, v, val); + continue; + } + + switch (act->dta_kind) { + case DTRACEACT_STOP: +#ifndef VBOX + if (dtrace_priv_proc_destructive(state)) + dtrace_action_stop(); +#else + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + continue; + + case DTRACEACT_BREAKPOINT: +#ifndef VBOX + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_breakpoint(ecb); +#else + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + continue; + + case DTRACEACT_PANIC: +#ifndef VBOX + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_panic(ecb); +#endif + continue; + + case DTRACEACT_STACK: + if (!dtrace_priv_kernel(state)) + continue; + + dtrace_getpcstack((pc_t *)(tomax + valoffs), + VBDTCAST(int)(size / sizeof (pc_t)), probe->dtpr_aframes, + DTRACE_ANCHORED(probe) ? NULL : + (uint32_t *)arg0); + + continue; + + case DTRACEACT_JSTACK: + case DTRACEACT_USTACK: + if (!dtrace_priv_proc(state)) + continue; + + /* + * See comment in DIF_VAR_PID. + */ + if (DTRACE_ANCHORED(mstate.dtms_probe) && + CPU_ON_INTR(CPU)) { + int depth = DTRACE_USTACK_NFRAMES( + rec->dtrd_arg) + 1; + + dtrace_bzero((void *)(tomax + valoffs), + DTRACE_USTACK_STRSIZE(rec->dtrd_arg) + + depth * sizeof (uint64_t)); + + continue; + } + +#ifndef VBOX /* no helpers */ + if (DTRACE_USTACK_STRSIZE(rec->dtrd_arg) != 0 && + curproc->p_dtrace_helpers != NULL) { + /* + * This is the slow path -- we have + * allocated string space, and we're + * getting the stack of a process that + * has helpers. Call into a separate + * routine to perform this processing. + */ + dtrace_action_ustack(&mstate, state, + (uint64_t *)(tomax + valoffs), + rec->dtrd_arg); + continue; + } +#endif + + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + dtrace_getupcstack((uint64_t *) + (tomax + valoffs), + DTRACE_USTACK_NFRAMES(rec->dtrd_arg) + 1); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + continue; + + default: + break; + } + + dp = act->dta_difo; + ASSERT(dp != NULL); + + val = dtrace_dif_emulate(dp, &mstate, vstate, state); + + if (*flags & CPU_DTRACE_ERROR) + continue; + + switch (act->dta_kind) { + case DTRACEACT_SPECULATE: + ASSERT(buf == &state->dts_buffer[cpuid]); + buf = dtrace_speculation_buffer(state, + cpuid, val); + + if (buf == NULL) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + offs = dtrace_buffer_reserve(buf, + ecb->dte_needed, ecb->dte_alignment, + state, NULL); + + if (offs < 0) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + tomax = buf->dtb_tomax; + ASSERT(tomax != NULL); + + if (ecb->dte_size != 0) + DTRACE_STORE(uint32_t, tomax, offs, + ecb->dte_epid); + continue; + + case DTRACEACT_CHILL: +#ifndef VBOX + if (dtrace_priv_kernel_destructive(state)) + dtrace_action_chill(&mstate, val); +#else + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + continue; + + case DTRACEACT_RAISE: +#ifndef VBOX + if (dtrace_priv_proc_destructive(state)) + dtrace_action_raise(val); +#else + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); +#endif + continue; + + case DTRACEACT_COMMIT: + ASSERT(!committed); + + /* + * We need to commit our buffer state. + */ + if (ecb->dte_size) + buf->dtb_offset = offs + ecb->dte_size; + buf = &state->dts_buffer[cpuid]; + dtrace_speculation_commit(state, cpuid, val); + committed = 1; + continue; + + case DTRACEACT_DISCARD: + dtrace_speculation_discard(state, cpuid, val); + continue; + + case DTRACEACT_DIFEXPR: + case DTRACEACT_LIBACT: + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (!dtrace_priv_kernel(state)) + continue; + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: { +#ifndef VBOX + struct pid *pid = curthread->t_procp->p_pidp; + + if (!dtrace_priv_proc(state)) + continue; + + DTRACE_STORE(uint64_t, tomax, + valoffs, (uint64_t)pid->pid_id); + DTRACE_STORE(uint64_t, tomax, + valoffs + sizeof (uint64_t), val); +#else + DTRACE_CPUFLAG_SET(CPU_DTRACE_UPRIV); +#endif + continue; + } + + case DTRACEACT_EXIT: { + /* + * For the exit action, we are going to attempt + * to atomically set our activity to be + * draining. If this fails (either because + * another CPU has beat us to the exit action, + * or because our current activity is something + * other than ACTIVE or WARMUP), we will + * continue. This assures that the exit action + * can be successfully recorded at most once + * when we're in the ACTIVE state. If we're + * encountering the exit() action while in + * COOLDOWN, however, we want to honor the new + * status code. (We know that we're the only + * thread in COOLDOWN, so there is no race.) + */ + void *activity = &state->dts_activity; + dtrace_activity_t current = state->dts_activity; + + if (current == DTRACE_ACTIVITY_COOLDOWN) + break; + + if (current != DTRACE_ACTIVITY_WARMUP) + current = DTRACE_ACTIVITY_ACTIVE; + + if ( (dtrace_activity_t)dtrace_cas32(activity, current, DTRACE_ACTIVITY_DRAINING) + != current) { + *flags |= CPU_DTRACE_DROP; + continue; + } + + break; + } + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", act->dta_kind)); +#endif + } + + if (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF) { + uintptr_t end = valoffs + size; + + if (!dtrace_vcanload((void *)(uintptr_t)val, + &dp->dtdo_rtype, &mstate, vstate)) + continue; + + /* + * If this is a string, we're going to only + * load until we find the zero byte -- after + * which we'll store zero bytes. + */ + if (dp->dtdo_rtype.dtdt_kind == + DIF_TYPE_STRING) { + char c = '\0' + 1; + int intuple = act->dta_intuple; + size_t s; + + for (s = 0; s < size; s++) { + if (c != '\0') + c = dtrace_load8(val++); + + DTRACE_STORE(uint8_t, tomax, + valoffs++, c); + + if (c == '\0' && intuple) + break; + } + + continue; + } + + while (valoffs < end) { + DTRACE_STORE(uint8_t, tomax, valoffs++, + dtrace_load8(val++)); + } + + continue; + } + + switch (size) { + case 0: + break; + + case sizeof (uint8_t): + DTRACE_STORE(uint8_t, tomax, valoffs, val); + break; + case sizeof (uint16_t): + DTRACE_STORE(uint16_t, tomax, valoffs, val); + break; + case sizeof (uint32_t): + DTRACE_STORE(uint32_t, tomax, valoffs, val); + break; + case sizeof (uint64_t): + DTRACE_STORE(uint64_t, tomax, valoffs, val); + break; + default: + /* + * Any other size should have been returned by + * reference, not by value. + */ +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%zu\n", size)); +#endif + break; + } + } + + if (*flags & CPU_DTRACE_DROP) + continue; + + if (*flags & CPU_DTRACE_FAULT) { + int ndx; + dtrace_action_t *err; + + buf->dtb_errors++; + + if (probe->dtpr_id == dtrace_probeid_error) { + /* + * There's nothing we can do -- we had an + * error on the error probe. We bump an + * error counter to at least indicate that + * this condition happened. + */ + dtrace_error(&state->dts_dblerrors); + continue; + } + + if (vtime) { + /* + * Before recursing on dtrace_probe(), we + * need to explicitly clear out our start + * time to prevent it from being accumulated + * into t_dtrace_vtime. + */ + curthread->t_dtrace_start = 0; + } + + /* + * Iterate over the actions to figure out which action + * we were processing when we experienced the error. + * Note that act points _past_ the faulting action; if + * act is ecb->dte_action, the fault was in the + * predicate, if it's ecb->dte_action->dta_next it's + * in action #1, and so on. + */ + for (err = ecb->dte_action, ndx = 0; + err != act; err = err->dta_next, ndx++) + continue; + + dtrace_probe_error(state, ecb->dte_epid, ndx, + (mstate.dtms_present & DTRACE_MSTATE_FLTOFFS) ? + mstate.dtms_fltoffs : -1, DTRACE_FLAGS2FLT(*flags), + cpu_core[cpuid].cpuc_dtrace_illval); + + continue; + } + + if (!committed) + buf->dtb_offset = offs + ecb->dte_size; + } + + if (vtime) + curthread->t_dtrace_start = dtrace_gethrtime(); + + dtrace_interrupt_enable(cookie); +} + +/* + * DTrace Probe Hashing Functions + * + * The functions in this section (and indeed, the functions in remaining + * sections) are not _called_ from probe context. (Any exceptions to this are + * marked with a "Note:".) Rather, they are called from elsewhere in the + * DTrace framework to look-up probes in, add probes to and remove probes from + * the DTrace probe hashes. (Each probe is hashed by each element of the + * probe tuple -- allowing for fast lookups, regardless of what was + * specified.) + */ +static uint_t +dtrace_hash_str(char *p) +{ + unsigned int g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + hval &= ~g; + } + return (hval); +} + +static dtrace_hash_t * +dtrace_hash_create(uintptr_t stroffs, uintptr_t nextoffs, uintptr_t prevoffs) +{ + dtrace_hash_t *hash = kmem_zalloc(sizeof (dtrace_hash_t), KM_SLEEP); + + hash->dth_stroffs = stroffs; + hash->dth_nextoffs = nextoffs; + hash->dth_prevoffs = prevoffs; + + hash->dth_size = 1; + hash->dth_mask = hash->dth_size - 1; + + hash->dth_tab = kmem_zalloc(hash->dth_size * + sizeof (dtrace_hashbucket_t *), KM_SLEEP); + + return (hash); +} + +static void +dtrace_hash_destroy(dtrace_hash_t *hash) +{ +#ifdef DEBUG + int i; + + for (i = 0; i < hash->dth_size; i++) + ASSERT(hash->dth_tab[i] == NULL); +#endif + + kmem_free(hash->dth_tab, + hash->dth_size * sizeof (dtrace_hashbucket_t *)); + kmem_free(hash, sizeof (dtrace_hash_t)); +} + +static void +dtrace_hash_resize(dtrace_hash_t *hash) +{ + int size = hash->dth_size, i, ndx; + int new_size = hash->dth_size << 1; + int new_mask = new_size - 1; + dtrace_hashbucket_t **new_tab, *bucket, *next; + + ASSERT((new_size & new_mask) == 0); + + new_tab = kmem_zalloc(new_size * sizeof (void *), KM_SLEEP); + + for (i = 0; i < size; i++) { + for (bucket = hash->dth_tab[i]; bucket != NULL; bucket = next) { + dtrace_probe_t *probe = bucket->dthb_chain; + + ASSERT(probe != NULL); + ndx = DTRACE_HASHSTR(hash, probe) & new_mask; + + next = bucket->dthb_next; + bucket->dthb_next = new_tab[ndx]; + new_tab[ndx] = bucket; + } + } + + kmem_free(hash->dth_tab, hash->dth_size * sizeof (void *)); + hash->dth_tab = new_tab; + hash->dth_size = new_size; + hash->dth_mask = new_mask; +} + +static void +dtrace_hash_add(dtrace_hash_t *hash, dtrace_probe_t *new) +{ + int hashval = DTRACE_HASHSTR(hash, new); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + dtrace_probe_t **nextp, **prevp; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, new)) + goto add; + } + + if ((hash->dth_nbuckets >> 1) > hash->dth_size) { + dtrace_hash_resize(hash); + dtrace_hash_add(hash, new); + return; + } + + bucket = kmem_zalloc(sizeof (dtrace_hashbucket_t), KM_SLEEP); + bucket->dthb_next = hash->dth_tab[ndx]; + hash->dth_tab[ndx] = bucket; + hash->dth_nbuckets++; + +add: + nextp = DTRACE_HASHNEXT(hash, new); + ASSERT(*nextp == NULL && *(DTRACE_HASHPREV(hash, new)) == NULL); + *nextp = bucket->dthb_chain; + + if (bucket->dthb_chain != NULL) { + prevp = DTRACE_HASHPREV(hash, bucket->dthb_chain); + ASSERT(*prevp == NULL); + *prevp = new; + } + + bucket->dthb_chain = new; + bucket->dthb_len++; +} + +static dtrace_probe_t * +dtrace_hash_lookup(dtrace_hash_t *hash, dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return (bucket->dthb_chain); + } + + return (NULL); +} + +static int +dtrace_hash_collisions(dtrace_hash_t *hash, dtrace_probe_t *template) +{ + int hashval = DTRACE_HASHSTR(hash, template); + int ndx = hashval & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, template)) + return (bucket->dthb_len); + } + + return (NULL); +} + +static void +dtrace_hash_remove(dtrace_hash_t *hash, dtrace_probe_t *probe) +{ + int ndx = DTRACE_HASHSTR(hash, probe) & hash->dth_mask; + dtrace_hashbucket_t *bucket = hash->dth_tab[ndx]; + + dtrace_probe_t **prevp = DTRACE_HASHPREV(hash, probe); + dtrace_probe_t **nextp = DTRACE_HASHNEXT(hash, probe); + + /* + * Find the bucket that we're removing this probe from. + */ + for (; bucket != NULL; bucket = bucket->dthb_next) { + if (DTRACE_HASHEQ(hash, bucket->dthb_chain, probe)) + break; + } + + ASSERT(bucket != NULL); + + if (*prevp == NULL) { + if (*nextp == NULL) { + /* + * The removed probe was the only probe on this + * bucket; we need to remove the bucket. + */ + dtrace_hashbucket_t *b = hash->dth_tab[ndx]; + + ASSERT(bucket->dthb_chain == probe); + ASSERT(b != NULL); + + if (b == bucket) { + hash->dth_tab[ndx] = bucket->dthb_next; + } else { + while (b->dthb_next != bucket) + b = b->dthb_next; + b->dthb_next = bucket->dthb_next; + } + + ASSERT(hash->dth_nbuckets > 0); + hash->dth_nbuckets--; + kmem_free(bucket, sizeof (dtrace_hashbucket_t)); + return; + } + + bucket->dthb_chain = *nextp; + } else { + *(DTRACE_HASHNEXT(hash, *prevp)) = *nextp; + } + + if (*nextp != NULL) + *(DTRACE_HASHPREV(hash, *nextp)) = *prevp; +} + +/* + * DTrace Utility Functions + * + * These are random utility functions that are _not_ called from probe context. + */ +static int +dtrace_badattr(const dtrace_attribute_t *a) +{ + return (a->dtat_name > DTRACE_STABILITY_MAX || + a->dtat_data > DTRACE_STABILITY_MAX || + a->dtat_class > DTRACE_CLASS_MAX); +} + +/* + * Return a duplicate copy of a string. If the specified string is NULL, + * this function returns a zero-length string. + */ +static char * +dtrace_strdup(const char *str) +{ + char *new = kmem_zalloc((str != NULL ? strlen(str) : 0) + 1, KM_SLEEP); + + if (str != NULL) + (void) strcpy(new, str); + + return (new); +} + +#define DTRACE_ISALPHA(c) \ + (((c) >= 'a' && (c) <= 'z') || ((c) >= 'A' && (c) <= 'Z')) + +static int +dtrace_badname(const char *s) +{ + char c; + + if (s == NULL || (c = *s++) == '\0') + return (0); + + if (!DTRACE_ISALPHA(c) && c != '-' && c != '_' && c != '.') + return (1); + + while ((c = *s++) != '\0') { + if (!DTRACE_ISALPHA(c) && (c < '0' || c > '9') && + c != '-' && c != '_' && c != '.' && c != '`') + return (1); + } + + return (0); +} + +static void +dtrace_cred2priv(cred_t *cr, uint32_t *privp, uid_t *uidp, zoneid_t *zoneidp) +{ + uint32_t priv; + + if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { + /* + * For DTRACE_PRIV_ALL, the uid and zoneid don't matter. + */ + priv = DTRACE_PRIV_ALL; +#ifdef VBOX + *uidp = UINT32_MAX; + *zoneidp = 0; +#endif + } else { + *uidp = crgetuid(cr); + *zoneidp = crgetzoneid(cr); + + priv = 0; + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) + priv |= DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER; + else if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) + priv |= DTRACE_PRIV_USER; + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) + priv |= DTRACE_PRIV_PROC; + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + priv |= DTRACE_PRIV_OWNER; + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + priv |= DTRACE_PRIV_ZONEOWNER; + } + + *privp = priv; +} + +#ifdef DTRACE_ERRDEBUG +static void +dtrace_errdebug(const char *str) +{ + int hval = dtrace_hash_str((char *)str) % DTRACE_ERRHASHSZ; + int occupied = 0; + + mutex_enter(&dtrace_errlock); + dtrace_errlast = str; + dtrace_errthread = curthread; + + while (occupied++ < DTRACE_ERRHASHSZ) { + if (dtrace_errhash[hval].dter_msg == str) { + dtrace_errhash[hval].dter_count++; + goto out; + } + + if (dtrace_errhash[hval].dter_msg != NULL) { + hval = (hval + 1) % DTRACE_ERRHASHSZ; + continue; + } + + dtrace_errhash[hval].dter_msg = str; + dtrace_errhash[hval].dter_count = 1; + goto out; + } + + panic("dtrace: undersized error hash"); +out: + mutex_exit(&dtrace_errlock); +} +#endif + +/* + * DTrace Matching Functions + * + * These functions are used to match groups of probes, given some elements of + * a probe tuple, or some globbed expressions for elements of a probe tuple. + */ +static int +dtrace_match_priv(const dtrace_probe_t *prp, uint32_t priv, uid_t uid, + zoneid_t zoneid) +{ + if (priv != DTRACE_PRIV_ALL) { + uint32_t ppriv = prp->dtpr_provider->dtpv_priv.dtpp_flags; + uint32_t match = priv & ppriv; + + /* + * No PRIV_DTRACE_* privileges... + */ + if ((priv & (DTRACE_PRIV_PROC | DTRACE_PRIV_USER | + DTRACE_PRIV_KERNEL)) == 0) + return (0); + + /* + * No matching bits, but there were bits to match... + */ + if (match == 0 && ppriv != 0) + return (0); + + /* + * Need to have permissions to the process, but don't... + */ + if (((ppriv & ~match) & DTRACE_PRIV_OWNER) != 0 && + uid != prp->dtpr_provider->dtpv_priv.dtpp_uid) { + return (0); + } + + /* + * Need to be in the same zone unless we possess the + * privilege to examine all zones. + */ + if (((ppriv & ~match) & DTRACE_PRIV_ZONEOWNER) != 0 && + zoneid != prp->dtpr_provider->dtpv_priv.dtpp_zoneid) { + return (0); + } + } + + return (1); +} + +/* + * dtrace_match_probe compares a dtrace_probe_t to a pre-compiled key, which + * consists of input pattern strings and an ops-vector to evaluate them. + * This function returns >0 for match, 0 for no match, and <0 for error. + */ +static int +dtrace_match_probe(const dtrace_probe_t *prp, const dtrace_probekey_t *pkp, + uint32_t priv, uid_t uid, zoneid_t zoneid) +{ + dtrace_provider_t *pvp = prp->dtpr_provider; + int rv; + + if (pvp->dtpv_defunct) + return (0); + + if ((rv = pkp->dtpk_pmatch(pvp->dtpv_name, pkp->dtpk_prov, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_mmatch(prp->dtpr_mod, pkp->dtpk_mod, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_fmatch(prp->dtpr_func, pkp->dtpk_func, 0)) <= 0) + return (rv); + + if ((rv = pkp->dtpk_nmatch(prp->dtpr_name, pkp->dtpk_name, 0)) <= 0) + return (rv); + + if (dtrace_match_priv(prp, priv, uid, zoneid) == 0) + return (0); + + return (rv); +} + +/* + * dtrace_match_glob() is a safe kernel implementation of the gmatch(3GEN) + * interface for matching a glob pattern 'p' to an input string 's'. Unlike + * libc's version, the kernel version only applies to 8-bit ASCII strings. + * In addition, all of the recursion cases except for '*' matching have been + * unwound. For '*', we still implement recursive evaluation, but a depth + * counter is maintained and matching is aborted if we recurse too deep. + * The function returns 0 if no match, >0 if match, and <0 if recursion error. + */ +static int +dtrace_match_glob(const char *s, const char *p, int depth) +{ + const char *olds; + char s1, c; + int gs; + + if (depth > DTRACE_PROBEKEY_MAXDEPTH) + return (-1); + + if (s == NULL) + s = ""; /* treat NULL as empty string */ + +top: + olds = s; + s1 = *s++; + + if (p == NULL) + return (0); + + if ((c = *p++) == '\0') + return (s1 == '\0'); + + switch (c) { + case '[': { + int ok = 0, notflag = 0; + char lc = '\0'; + + if (s1 == '\0') + return (0); + + if (*p == '!') { + notflag = 1; + p++; + } + + if ((c = *p++) == '\0') + return (0); + + do { + if (c == '-' && lc != '\0' && *p != ']') { + if ((c = *p++) == '\0') + return (0); + if (c == '\\' && (c = *p++) == '\0') + return (0); + + if (notflag) { + if (s1 < lc || s1 > c) + ok++; + else + return (0); + } else if (lc <= s1 && s1 <= c) + ok++; + + } else if (c == '\\' && (c = *p++) == '\0') + return (0); + + lc = c; /* save left-hand 'c' for next iteration */ + + if (notflag) { + if (s1 != c) + ok++; + else + return (0); + } else if (s1 == c) + ok++; + + if ((c = *p++) == '\0') + return (0); + + } while (c != ']'); + + if (ok) + goto top; + + return (0); + } + + case '\\': + if ((c = *p++) == '\0') + return (0); + RT_FALL_THRU(); + + default: + if (c != s1) + return (0); + RT_FALL_THRU(); + + case '?': + if (s1 != '\0') + goto top; + return (0); + + case '*': + while (*p == '*') + p++; /* consecutive *'s are identical to a single one */ + + if (*p == '\0') + return (1); + + for (s = olds; *s != '\0'; s++) { + if ((gs = dtrace_match_glob(s, p, depth + 1)) != 0) + return (gs); + } + + return (0); + } +} + +/*ARGSUSED*/ +static int +dtrace_match_string(const char *s, const char *p, int depth) +{ + RT_NOREF_PV(depth); + return (s != NULL && strcmp(s, p) == 0); +} + +/*ARGSUSED*/ +static int +dtrace_match_nul(const char *s, const char *p, int depth) +{ + RT_NOREF_PV(s); RT_NOREF_PV(p); RT_NOREF_PV(depth); + return (1); /* always match the empty pattern */ +} + +/*ARGSUSED*/ +static int +dtrace_match_nonzero(const char *s, const char *p, int depth) +{ + RT_NOREF_PV(p); RT_NOREF_PV(depth); + return (s != NULL && s[0] != '\0'); +} + +static int +dtrace_match(const dtrace_probekey_t *pkp, uint32_t priv, uid_t uid, + zoneid_t zoneid, int (*matched)(dtrace_probe_t *, void *), void *arg) +{ + dtrace_probe_t template, *probe; + dtrace_hash_t *hash = NULL; + int len, rc, best = INT_MAX, nmatched = 0; + dtrace_id_t i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + /* + * If the probe ID is specified in the key, just lookup by ID and + * invoke the match callback once if a matching probe is found. + */ + if (pkp->dtpk_id != DTRACE_IDNONE) { + if ((probe = dtrace_probe_lookup_id(pkp->dtpk_id)) != NULL && + dtrace_match_probe(probe, pkp, priv, uid, zoneid) > 0) { + if ((*matched)(probe, arg) == DTRACE_MATCH_FAIL) + return (DTRACE_MATCH_FAIL); + nmatched++; + } + return (nmatched); + } + + template.dtpr_mod = (char *)pkp->dtpk_mod; + template.dtpr_func = (char *)pkp->dtpk_func; + template.dtpr_name = (char *)pkp->dtpk_name; + + /* + * We want to find the most distinct of the module name, function + * name, and name. So for each one that is not a glob pattern or + * empty string, we perform a lookup in the corresponding hash and + * use the hash table with the fewest collisions to do our search. + */ + if (pkp->dtpk_mmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_bymod, &template)) < best) { + best = len; + hash = dtrace_bymod; + } + + if (pkp->dtpk_fmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byfunc, &template)) < best) { + best = len; + hash = dtrace_byfunc; + } + + if (pkp->dtpk_nmatch == &dtrace_match_string && + (len = dtrace_hash_collisions(dtrace_byname, &template)) < best) { + best = len; + hash = dtrace_byname; + } + + /* + * If we did not select a hash table, iterate over every probe and + * invoke our callback for each one that matches our input probe key. + */ + if (hash == NULL) { + for (i = 0; i < VBDTCAST(dtrace_id_t)dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL || + dtrace_match_probe(probe, pkp, priv, uid, + zoneid) <= 0) + continue; + + nmatched++; + + if ((rc = (*matched)(probe, arg)) != + DTRACE_MATCH_NEXT) { + if (rc == DTRACE_MATCH_FAIL) + return (DTRACE_MATCH_FAIL); + break; + } + } + + return (nmatched); + } + + /* + * If we selected a hash table, iterate over each probe of the same key + * name and invoke the callback for every probe that matches the other + * attributes of our input probe key. + */ + for (probe = dtrace_hash_lookup(hash, &template); probe != NULL; + probe = *(DTRACE_HASHNEXT(hash, probe))) { + + if (dtrace_match_probe(probe, pkp, priv, uid, zoneid) <= 0) + continue; + + nmatched++; + + if ((rc = (*matched)(probe, arg)) != DTRACE_MATCH_NEXT) { + if (rc == DTRACE_MATCH_FAIL) + return (DTRACE_MATCH_FAIL); + break; + } + } + + return (nmatched); +} + +/* + * Return the function pointer dtrace_probecmp() should use to compare the + * specified pattern with a string. For NULL or empty patterns, we select + * dtrace_match_nul(). For glob pattern strings, we use dtrace_match_glob(). + * For non-empty non-glob strings, we use dtrace_match_string(). + */ +static dtrace_probekey_f * +dtrace_probekey_func(const char *p) +{ + char c; + + if (p == NULL || *p == '\0') + return (&dtrace_match_nul); + + while ((c = *p++) != '\0') { + if (c == '[' || c == '?' || c == '*' || c == '\\') + return (&dtrace_match_glob); + } + + return (&dtrace_match_string); +} + +/* + * Build a probe comparison key for use with dtrace_match_probe() from the + * given probe description. By convention, a null key only matches anchored + * probes: if each field is the empty string, reset dtpk_fmatch to + * dtrace_match_nonzero(). + */ +static void +dtrace_probekey(const dtrace_probedesc_t *pdp, dtrace_probekey_t *pkp) +{ + pkp->dtpk_prov = pdp->dtpd_provider; + pkp->dtpk_pmatch = dtrace_probekey_func(pdp->dtpd_provider); + + pkp->dtpk_mod = pdp->dtpd_mod; + pkp->dtpk_mmatch = dtrace_probekey_func(pdp->dtpd_mod); + + pkp->dtpk_func = pdp->dtpd_func; + pkp->dtpk_fmatch = dtrace_probekey_func(pdp->dtpd_func); + + pkp->dtpk_name = pdp->dtpd_name; + pkp->dtpk_nmatch = dtrace_probekey_func(pdp->dtpd_name); + + pkp->dtpk_id = pdp->dtpd_id; + + if (pkp->dtpk_id == DTRACE_IDNONE && + pkp->dtpk_pmatch == &dtrace_match_nul && + pkp->dtpk_mmatch == &dtrace_match_nul && + pkp->dtpk_fmatch == &dtrace_match_nul && + pkp->dtpk_nmatch == &dtrace_match_nul) + pkp->dtpk_fmatch = &dtrace_match_nonzero; +} + +/* + * DTrace Provider-to-Framework API Functions + * + * These functions implement much of the Provider-to-Framework API, as + * described in <sys/dtrace.h>. The parts of the API not in this section are + * the functions in the API for probe management (found below), and + * dtrace_probe() itself (found above). + */ + +/* + * Register the calling provider with the DTrace framework. This should + * generally be called by DTrace providers in their attach(9E) entry point. + */ +int +dtrace_register(const char *name, const dtrace_pattr_t *pap, uint32_t priv, + cred_t *cr, const dtrace_pops_t *pops, void *arg, dtrace_provider_id_t *idp) +{ + dtrace_provider_t *provider; + + if (name == NULL || pap == NULL || pops == NULL || idp == NULL) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "arguments", name ? name : "<NULL>"); + return (EINVAL); + } + + if (name[0] == '\0' || dtrace_badname(name)) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider name", name); + return (EINVAL); + } + + if ((pops->dtps_provide == NULL && pops->dtps_provide_module == NULL) || + pops->dtps_enable == NULL || pops->dtps_disable == NULL || + pops->dtps_destroy == NULL || + ((pops->dtps_resume == NULL) != (pops->dtps_suspend == NULL))) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider ops", name); + return (EINVAL); + } + + if (dtrace_badattr(&pap->dtpa_provider) || + dtrace_badattr(&pap->dtpa_mod) || + dtrace_badattr(&pap->dtpa_func) || + dtrace_badattr(&pap->dtpa_name) || + dtrace_badattr(&pap->dtpa_args)) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "provider attributes", name); + return (EINVAL); + } + + if (priv & ~DTRACE_PRIV_ALL) { + cmn_err(CE_WARN, "failed to register provider '%s': invalid " + "privilege attributes", name); + return (EINVAL); + } + + if ((priv & DTRACE_PRIV_KERNEL) && + (priv & (DTRACE_PRIV_USER | DTRACE_PRIV_OWNER)) && + pops->dtps_usermode == NULL) { + cmn_err(CE_WARN, "failed to register provider '%s': need " + "dtps_usermode() op for given privilege attributes", name); + return (EINVAL); + } + + provider = kmem_zalloc(sizeof (dtrace_provider_t), KM_SLEEP); + provider->dtpv_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); + (void) strcpy(provider->dtpv_name, name); + + provider->dtpv_attr = *pap; + provider->dtpv_priv.dtpp_flags = priv; + if (cr != NULL) { + provider->dtpv_priv.dtpp_uid = crgetuid(cr); + provider->dtpv_priv.dtpp_zoneid = crgetzoneid(cr); + } + provider->dtpv_pops = *pops; + + if (pops->dtps_provide == NULL) { + ASSERT(pops->dtps_provide_module != NULL); + provider->dtpv_pops.dtps_provide = + (void (*)(void *, const dtrace_probedesc_t *))dtrace_nullop; + } + + if (pops->dtps_provide_module == NULL) { + ASSERT(pops->dtps_provide != NULL); + provider->dtpv_pops.dtps_provide_module = + (void (*)(void *, struct modctl *))dtrace_nullop; + } + + if (pops->dtps_suspend == NULL) { + ASSERT(pops->dtps_resume == NULL); + provider->dtpv_pops.dtps_suspend = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + provider->dtpv_pops.dtps_resume = + (void (*)(void *, dtrace_id_t, void *))dtrace_nullop; + } + + provider->dtpv_arg = arg; + *idp = (dtrace_provider_id_t)provider; + + if (pops == &dtrace_provider_ops) { + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dtrace_anon.dta_enabling == NULL); + + /* + * We make sure that the DTrace provider is at the head of + * the provider chain. + */ + provider->dtpv_next = dtrace_provider; + dtrace_provider = provider; + return (0); + } + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + /* + * If there is at least one provider registered, we'll add this + * provider after the first provider. + */ + if (dtrace_provider != NULL) { + provider->dtpv_next = dtrace_provider->dtpv_next; + dtrace_provider->dtpv_next = provider; + } else { + dtrace_provider = provider; + } + + if (dtrace_retained != NULL) { + dtrace_enabling_provide(provider); + + /* + * Now we need to call dtrace_enabling_matchall() -- which + * will acquire cpu_lock and dtrace_lock. We therefore need + * to drop all of our locks before calling into it... + */ + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + dtrace_enabling_matchall(); + + return (0); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + return (0); +} + +/* + * Unregister the specified provider from the DTrace framework. This should + * generally be called by DTrace providers in their detach(9E) entry point. + */ +int +dtrace_unregister(dtrace_provider_id_t id) +{ + dtrace_provider_t *old = (dtrace_provider_t *)id; + dtrace_provider_t *prev = NULL; + VBDTTYPE(uint32_t,int) i, self = 0; + dtrace_probe_t *probe, *first = NULL; + + if (old->dtpv_pops.dtps_enable == + (int (*)(void *, dtrace_id_t, void *))(uintptr_t)dtrace_enable_nullop) { + /* + * If DTrace itself is the provider, we're called with locks + * already held. + */ + ASSERT(old == dtrace_provider); +#ifndef VBOX + ASSERT(dtrace_devi != NULL); +#endif + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + self = 1; + + if (dtrace_provider->dtpv_next != NULL) { + /* + * There's another provider here; return failure. + */ + return (EBUSY); + } + } else { + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + } + + /* + * If anyone has /dev/dtrace open, or if there are anonymous enabled + * probes, we refuse to let providers slither away, unless this + * provider has already been explicitly invalidated. + */ + if (!old->dtpv_defunct && + (dtrace_opens || (dtrace_anon.dta_state != NULL && + dtrace_anon.dta_state->dts_necbs > 0))) { + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + return (EBUSY); + } + + /* + * Attempt to destroy the probes associated with this provider. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != old) + continue; + + if (probe->dtpr_ecb == NULL) + continue; + + /* + * We have at least one ECB; we can't remove this provider. + */ + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + return (EBUSY); + } + + /* + * All of the probes for this provider are disabled; we can safely + * remove all of them from their hash chains and from the probe array. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != old) + continue; + + dtrace_probes[i] = NULL; + + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (first == NULL) { + first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = first; + first = probe; + } + } + + /* + * The provider's probes have been removed from the hash chains and + * from the probe array. Now issue a dtrace_sync() to be sure that + * everyone has cleared out from any probe array processing. + */ + dtrace_sync(); + + for (probe = first; probe != NULL; probe = first) { + first = probe->dtpr_nextmod; + + old->dtpv_pops.dtps_destroy(old->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); + vmem_free(dtrace_arena, (void *)(uintptr_t)(probe->dtpr_id), 1); + kmem_free(probe, sizeof (dtrace_probe_t)); + } + + if ((prev = dtrace_provider) == old) { +#ifndef VBOX + ASSERT(self || dtrace_devi == NULL); + ASSERT(old->dtpv_next == NULL || dtrace_devi == NULL); +#endif + dtrace_provider = old->dtpv_next; + } else { + while (prev != NULL && prev->dtpv_next != old) + prev = prev->dtpv_next; + + if (prev == NULL) { + panic("attempt to unregister non-existent " + "dtrace provider %p\n", (void *)id); + } + + prev->dtpv_next = old->dtpv_next; + } + + if (!self) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + } + + kmem_free(old->dtpv_name, strlen(old->dtpv_name) + 1); + kmem_free(old, sizeof (dtrace_provider_t)); + + return (0); +} + +/* + * Invalidate the specified provider. All subsequent probe lookups for the + * specified provider will fail, but its probes will not be removed. + */ +void +dtrace_invalidate(dtrace_provider_id_t id) +{ + dtrace_provider_t *pvp = (dtrace_provider_t *)id; + + ASSERT(pvp->dtpv_pops.dtps_enable != + (int (*)(void *, dtrace_id_t, void *))(uintptr_t)dtrace_enable_nullop); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + pvp->dtpv_defunct = 1; + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); +} + +/* + * Indicate whether or not DTrace has attached. + */ +int +dtrace_attached(void) +{ + /* + * dtrace_provider will be non-NULL iff the DTrace driver has + * attached. (It's non-NULL because DTrace is always itself a + * provider.) + */ + return (dtrace_provider != NULL); +} + +/* + * Remove all the unenabled probes for the given provider. This function is + * not unlike dtrace_unregister(), except that it doesn't remove the provider + * -- just as many of its associated probes as it can. + */ +int +dtrace_condense(dtrace_provider_id_t id) +{ + dtrace_provider_t *prov = (dtrace_provider_t *)id; + VBDTTYPE(uint32_t,int) i; + dtrace_probe_t *probe; + + /* + * Make sure this isn't the dtrace provider itself. + */ + ASSERT(prov->dtpv_pops.dtps_enable != + (int (*)(void *, dtrace_id_t, void *))(uintptr_t)dtrace_enable_nullop); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + /* + * Attempt to destroy the probes associated with this provider. + */ + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_provider != prov) + continue; + + if (probe->dtpr_ecb != NULL) + continue; + + dtrace_probes[i] = NULL; + + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, i + 1, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); + kmem_free(probe, sizeof (dtrace_probe_t)); + vmem_free(dtrace_arena, (void *)((uintptr_t)i + 1), 1); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + return (0); +} + +/* + * DTrace Probe Management Functions + * + * The functions in this section perform the DTrace probe management, + * including functions to create probes, look-up probes, and call into the + * providers to request that probes be provided. Some of these functions are + * in the Provider-to-Framework API; these functions can be identified by the + * fact that they are not declared "static". + */ + +/* + * Create a probe with the specified module name, function name, and name. + */ +dtrace_id_t +dtrace_probe_create(dtrace_provider_id_t prov, const char *mod, + const char *func, const char *name, int aframes, void *arg) +{ + dtrace_probe_t *probe, **probes; + dtrace_provider_t *provider = (dtrace_provider_t *)prov; + dtrace_id_t id; + + if (provider == dtrace_provider) { + ASSERT(MUTEX_HELD(&dtrace_lock)); + } else { + mutex_enter(&dtrace_lock); + } + + id = (dtrace_id_t)(uintptr_t)vmem_alloc(dtrace_arena, 1, + VM_BESTFIT | VM_SLEEP); + probe = kmem_zalloc(sizeof (dtrace_probe_t), KM_SLEEP); + + probe->dtpr_id = id; + probe->dtpr_gen = dtrace_probegen++; + probe->dtpr_mod = dtrace_strdup(mod); + probe->dtpr_func = dtrace_strdup(func); + probe->dtpr_name = dtrace_strdup(name); + probe->dtpr_arg = arg; + probe->dtpr_aframes = aframes; + probe->dtpr_provider = provider; + + dtrace_hash_add(dtrace_bymod, probe); + dtrace_hash_add(dtrace_byfunc, probe); + dtrace_hash_add(dtrace_byname, probe); + + if (id - 1 >= dtrace_nprobes) { + size_t osize = dtrace_nprobes * sizeof (dtrace_probe_t *); + size_t nsize = osize << 1; + + if (nsize == 0) { + ASSERT(osize == 0); + ASSERT(dtrace_probes == NULL); + nsize = sizeof (dtrace_probe_t *); + } + + probes = kmem_zalloc(nsize, KM_SLEEP); + + if (dtrace_probes == NULL) { + ASSERT(osize == 0); + dtrace_probes = probes; + dtrace_nprobes = 1; + } else { + dtrace_probe_t **oprobes = dtrace_probes; + + bcopy(oprobes, probes, osize); + dtrace_membar_producer(); + dtrace_probes = probes; + + dtrace_sync(); + + /* + * All CPUs are now seeing the new probes array; we can + * safely free the old array. + */ + kmem_free(oprobes, osize); + dtrace_nprobes <<= 1; + } + + ASSERT(id - 1 < dtrace_nprobes); + } + + ASSERT(dtrace_probes[id - 1] == NULL); + dtrace_probes[id - 1] = probe; + + if (provider != dtrace_provider) + mutex_exit(&dtrace_lock); + + return (id); +} + +static dtrace_probe_t * +dtrace_probe_lookup_id(dtrace_id_t id) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || id > dtrace_nprobes) + return (NULL); + + return (dtrace_probes[id - 1]); +} + +static int +dtrace_probe_lookup_match(dtrace_probe_t *probe, void *arg) +{ + *((dtrace_id_t *)arg) = probe->dtpr_id; + + return (DTRACE_MATCH_DONE); +} + +/* + * Look up a probe based on provider and one or more of module name, function + * name and probe name. + */ +dtrace_id_t +dtrace_probe_lookup(dtrace_provider_id_t prid, const char *mod, + const char *func, const char *name) +{ + dtrace_probekey_t pkey; + dtrace_id_t id; + int match; + + pkey.dtpk_prov = ((dtrace_provider_t *)prid)->dtpv_name; + pkey.dtpk_pmatch = &dtrace_match_string; + pkey.dtpk_mod = mod; + pkey.dtpk_mmatch = mod ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_func = func; + pkey.dtpk_fmatch = func ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_name = name; + pkey.dtpk_nmatch = name ? &dtrace_match_string : &dtrace_match_nul; + pkey.dtpk_id = DTRACE_IDNONE; + + mutex_enter(&dtrace_lock); + match = dtrace_match(&pkey, DTRACE_PRIV_ALL, 0, 0, + dtrace_probe_lookup_match, &id); + mutex_exit(&dtrace_lock); + + ASSERT(match == 1 || match == 0); + return (match ? id : 0); +} + +/* + * Returns the probe argument associated with the specified probe. + */ +void * +dtrace_probe_arg(dtrace_provider_id_t id, dtrace_id_t pid) +{ + dtrace_probe_t *probe; + void *rval = NULL; + + mutex_enter(&dtrace_lock); + + if ((probe = dtrace_probe_lookup_id(pid)) != NULL && + probe->dtpr_provider == (dtrace_provider_t *)id) + rval = probe->dtpr_arg; + + mutex_exit(&dtrace_lock); + + return (rval); +} + +/* + * Copy a probe into a probe description. + */ +static void +dtrace_probe_description(const dtrace_probe_t *prp, dtrace_probedesc_t *pdp) +{ + bzero(pdp, sizeof (dtrace_probedesc_t)); + pdp->dtpd_id = prp->dtpr_id; + + (void) strncpy(pdp->dtpd_provider, + prp->dtpr_provider->dtpv_name, DTRACE_PROVNAMELEN - 1); + + (void) strncpy(pdp->dtpd_mod, prp->dtpr_mod, DTRACE_MODNAMELEN - 1); + (void) strncpy(pdp->dtpd_func, prp->dtpr_func, DTRACE_FUNCNAMELEN - 1); + (void) strncpy(pdp->dtpd_name, prp->dtpr_name, DTRACE_NAMELEN - 1); +} + +/* + * Called to indicate that a probe -- or probes -- should be provided by a + * specfied provider. If the specified description is NULL, the provider will + * be told to provide all of its probes. (This is done whenever a new + * consumer comes along, or whenever a retained enabling is to be matched.) If + * the specified description is non-NULL, the provider is given the + * opportunity to dynamically provide the specified probe, allowing providers + * to support the creation of probes on-the-fly. (So-called _autocreated_ + * probes.) If the provider is NULL, the operations will be applied to all + * providers; if the provider is non-NULL the operations will only be applied + * to the specified provider. The dtrace_provider_lock must be held, and the + * dtrace_lock must _not_ be held -- the provider's dtps_provide() operation + * will need to grab the dtrace_lock when it reenters the framework through + * dtrace_probe_lookup(), dtrace_probe_create(), etc. + */ +static void +dtrace_probe_provide(dtrace_probedesc_t *desc, dtrace_provider_t *prv) +{ +#ifndef VBOX + struct modctl *ctl; +#endif + int all = 0; + + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + /* + * First, call the blanket provide operation. + */ + prv->dtpv_pops.dtps_provide(prv->dtpv_arg, desc); + +#ifndef VBOX + /* + * Now call the per-module provide operation. We will grab + * mod_lock to prevent the list from being modified. Note + * that this also prevents the mod_busy bits from changing. + * (mod_busy can only be changed with mod_lock held.) + */ + mutex_enter(&mod_lock); + + ctl = &modules; + do { + if (ctl->mod_busy || ctl->mod_mp == NULL) + continue; + + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); + + } while ((ctl = ctl->mod_next) != &modules); + + mutex_exit(&mod_lock); +#endif + } while (all && (prv = prv->dtpv_next) != NULL); +} + +#ifndef VBOX +/* + * Iterate over each probe, and call the Framework-to-Provider API function + * denoted by offs. + */ +static void +dtrace_probe_foreach(uintptr_t offs) +{ + dtrace_provider_t *prov; + void (*func)(void *, dtrace_id_t, void *); + dtrace_probe_t *probe; + dtrace_icookie_t cookie; + VBDTTYPE(uint32_t,int) i; + + /* + * We disable interrupts to walk through the probe array. This is + * safe -- the dtrace_sync() in dtrace_unregister() assures that we + * won't see stale data. + */ + cookie = dtrace_interrupt_disable(); + + for (i = 0; i < dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i]) == NULL) + continue; + + if (probe->dtpr_ecb == NULL) { + /* + * This probe isn't enabled -- don't call the function. + */ + continue; + } + + prov = probe->dtpr_provider; + func = *((void(**)(void *, dtrace_id_t, void *)) + ((uintptr_t)&prov->dtpv_pops + offs)); + + func(prov->dtpv_arg, i + 1, probe->dtpr_arg); + } + + dtrace_interrupt_enable(cookie); +} +#endif /* !VBOX */ + +static int +dtrace_probe_enable(const dtrace_probedesc_t *desc, dtrace_enabling_t *enab) +{ + dtrace_probekey_t pkey; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + dtrace_ecb_create_cache = NULL; + + if (desc == NULL) { + /* + * If we're passed a NULL description, we're being asked to + * create an ECB with a NULL probe. + */ + (void) dtrace_ecb_create_enable(NULL, enab); + return (0); + } + + dtrace_probekey(desc, &pkey); + dtrace_cred2priv(enab->dten_vstate->dtvs_state->dts_cred.dcr_cred, + &priv, &uid, &zoneid); + + return (dtrace_match(&pkey, priv, uid, zoneid, dtrace_ecb_create_enable, + enab)); +} + +/* + * DTrace Helper Provider Functions + */ +static void +dtrace_dofattr2attr(dtrace_attribute_t *attr, const dof_attr_t dofattr) +{ + attr->dtat_name = DOF_ATTR_NAME(dofattr); + attr->dtat_data = DOF_ATTR_DATA(dofattr); + attr->dtat_class = DOF_ATTR_CLASS(dofattr); +} + +static void +dtrace_dofprov2hprov(dtrace_helper_provdesc_t *hprov, + const dof_provider_t *dofprov, char *strtab) +{ + hprov->dthpv_provname = strtab + dofprov->dofpv_name; + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_provider, + dofprov->dofpv_provattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_mod, + dofprov->dofpv_modattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_func, + dofprov->dofpv_funcattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_name, + dofprov->dofpv_nameattr); + dtrace_dofattr2attr(&hprov->dthpv_pattr.dtpa_args, + dofprov->dofpv_argsattr); +} + +static void +dtrace_helper_provide_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; + dof_provider_t *provider; + dof_probe_t *probe; + uint32_t *off, *enoff; + uint8_t *arg; + char *strtab; + uint_t i, nprobes; + dtrace_helper_provdesc_t dhpv; + dtrace_helper_probedesc_t dhpb; + dtrace_meta_t *meta = dtrace_meta_pid; + dtrace_mops_t *mops = &meta->dtm_mops; + void *parg; + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_strtab * dof->dofh_secsize); + prb_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_probes * dof->dofh_secsize); + arg_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_prargs * dof->dofh_secsize); + off_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_proffs * dof->dofh_secsize); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + off = (uint32_t *)(uintptr_t)(daddr + off_sec->dofs_offset); + arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); + enoff = NULL; + + /* + * See dtrace_helper_provider_validate(). + */ + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + provider->dofpv_prenoffs != DOF_SECT_NONE) { + enoff_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_prenoffs * dof->dofh_secsize); + enoff = (uint32_t *)(uintptr_t)(daddr + enoff_sec->dofs_offset); + } + + nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; + + /* + * Create the provider. + */ + dtrace_dofprov2hprov(&dhpv, provider, strtab); + + if ((parg = mops->dtms_provide_pid(meta->dtm_arg, &dhpv, pid)) == NULL) + return; + + meta->dtm_count++; + + /* + * Create the probes. + */ + for (i = 0; i < nprobes; i++) { + probe = (dof_probe_t *)(uintptr_t)(daddr + + prb_sec->dofs_offset + i * prb_sec->dofs_entsize); + + dhpb.dthpb_mod = dhp->dofhp_mod; + dhpb.dthpb_func = strtab + probe->dofpr_func; + dhpb.dthpb_name = strtab + probe->dofpr_name; + dhpb.dthpb_base = probe->dofpr_addr; + dhpb.dthpb_offs = off + probe->dofpr_offidx; + dhpb.dthpb_noffs = probe->dofpr_noffs; + if (enoff != NULL) { + dhpb.dthpb_enoffs = enoff + probe->dofpr_enoffidx; + dhpb.dthpb_nenoffs = probe->dofpr_nenoffs; + } else { + dhpb.dthpb_enoffs = NULL; + dhpb.dthpb_nenoffs = 0; + } + dhpb.dthpb_args = arg + probe->dofpr_argidx; + dhpb.dthpb_nargc = probe->dofpr_nargc; + dhpb.dthpb_xargc = probe->dofpr_xargc; + dhpb.dthpb_ntypes = strtab + probe->dofpr_nargv; + dhpb.dthpb_xtypes = strtab + probe->dofpr_xargv; + + mops->dtms_create_probe(meta->dtm_arg, parg, &dhpb); + } +} + +static void +dtrace_helper_provide(dof_helper_t *dhp, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + VBDTTYPE(uint32_t,int) i; + + ASSERT(MUTEX_HELD(&dtrace_meta_lock)); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + dtrace_helper_provide_one(dhp, sec, pid); + } + + /* + * We may have just created probes, so we must now rematch against + * any retained enablings. Note that this call will acquire both + * cpu_lock and dtrace_lock; the fact that we are holding + * dtrace_meta_lock now is what defines the ordering with respect to + * these three locks. + */ + dtrace_enabling_matchall(); +} + +#ifndef VBOX + +static void +dtrace_helper_provider_remove_one(dof_helper_t *dhp, dof_sec_t *sec, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + dof_sec_t *str_sec; + dof_provider_t *provider; + char *strtab; + dtrace_helper_provdesc_t dhpv; + dtrace_meta_t *meta = dtrace_meta_pid; + dtrace_mops_t *mops = &meta->dtm_mops; + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = (dof_sec_t *)(uintptr_t)(daddr + dof->dofh_secoff + + provider->dofpv_strtab * dof->dofh_secsize); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + + /* + * Create the provider. + */ + dtrace_dofprov2hprov(&dhpv, provider, strtab); + + mops->dtms_remove_pid(meta->dtm_arg, &dhpv, pid); + + meta->dtm_count--; +} + +static void +dtrace_helper_provider_remove(dof_helper_t *dhp, pid_t pid) +{ + uintptr_t daddr = (uintptr_t)dhp->dofhp_dof; + dof_hdr_t *dof = (dof_hdr_t *)daddr; + VBDTTYPE(uint32_t,int) i; + + ASSERT(MUTEX_HELD(&dtrace_meta_lock)); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + dtrace_helper_provider_remove_one(dhp, sec, pid); + } +} + +#endif /* !VBOX */ + +/* + * DTrace Meta Provider-to-Framework API Functions + * + * These functions implement the Meta Provider-to-Framework API, as described + * in <sys/dtrace.h>. + */ +int +dtrace_meta_register(const char *name, const dtrace_mops_t *mops, void *arg, + dtrace_meta_provider_id_t *idp) +{ + dtrace_meta_t *meta; + dtrace_helpers_t *help, *next; + VBDTTYPE(uint32_t,int) i; + + *idp = DTRACE_METAPROVNONE; + + /* + * We strictly don't need the name, but we hold onto it for + * debuggability. All hail error queues! + */ + if (name == NULL) { + cmn_err(CE_WARN, "failed to register meta-provider: " + "invalid name"); + return (EINVAL); + } + + if (mops == NULL || + mops->dtms_create_probe == NULL || + mops->dtms_provide_pid == NULL || + mops->dtms_remove_pid == NULL) { + cmn_err(CE_WARN, "failed to register meta-register %s: " + "invalid ops", name); + return (EINVAL); + } + + meta = kmem_zalloc(sizeof (dtrace_meta_t), KM_SLEEP); + meta->dtm_mops = *mops; + meta->dtm_name = kmem_alloc(strlen(name) + 1, KM_SLEEP); + (void) strcpy(meta->dtm_name, name); + meta->dtm_arg = arg; + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (dtrace_meta_pid != NULL) { + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + cmn_err(CE_WARN, "failed to register meta-register %s: " + "user-land meta-provider exists", name); + kmem_free(meta->dtm_name, strlen(meta->dtm_name) + 1); + kmem_free(meta, sizeof (dtrace_meta_t)); + return (EINVAL); + } + + dtrace_meta_pid = meta; + *idp = (dtrace_meta_provider_id_t)meta; + + /* + * If there are providers and probes ready to go, pass them + * off to the new meta provider now. + */ + + help = dtrace_deferred_pid; + dtrace_deferred_pid = NULL; + + mutex_exit(&dtrace_lock); + + while (help != NULL) { + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, + help->dthps_pid); + } + + next = help->dthps_next; + help->dthps_next = NULL; + help->dthps_prev = NULL; + help->dthps_deferred = 0; + help = next; + } + + mutex_exit(&dtrace_meta_lock); + + return (0); +} + +int +dtrace_meta_unregister(dtrace_meta_provider_id_t id) +{ + dtrace_meta_t **pp, *old = (dtrace_meta_t *)id; + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (old == dtrace_meta_pid) { + pp = &dtrace_meta_pid; + } else { + panic("attempt to unregister non-existent " + "dtrace meta-provider %p\n", (void *)old); +#ifdef VBOX + return EINVAL; +#endif + } + + if (old->dtm_count != 0) { + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + return (EBUSY); + } + + *pp = NULL; + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_meta_lock); + + kmem_free(old->dtm_name, strlen(old->dtm_name) + 1); + kmem_free(old, sizeof (dtrace_meta_t)); + + return (0); +} + + +/* + * DTrace DIF Object Functions + */ +static int +dtrace_difo_err(uint_t pc, const char *format, ...) +{ + if (dtrace_err_verbose) { + va_list alist; + + (void) uprintf("dtrace DIF object error: [%u]: ", pc); + va_start(alist, format); + (void) vuprintf(format, alist); + va_end(alist); + } + +#ifdef DTRACE_ERRDEBUG + dtrace_errdebug(format); +#endif + return (1); +} + +/* + * Validate a DTrace DIF object by checking the IR instructions. The following + * rules are currently enforced by dtrace_difo_validate(): + * + * 1. Each instruction must have a valid opcode + * 2. Each register, string, variable, or subroutine reference must be valid + * 3. No instruction can modify register %r0 (must be zero) + * 4. All instruction reserved bits must be set to zero + * 5. The last instruction must be a "ret" instruction + * 6. All branch targets must reference a valid instruction _after_ the branch + */ +static int +dtrace_difo_validate(dtrace_difo_t *dp, dtrace_vstate_t *vstate, uint_t nregs, + cred_t *cr) +{ +#ifndef VBOX + int err = 0, i; +#else + int err = 0; + uint_t i; +#endif + int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; + int kcheckload; + uint_t pc; + + kcheckload = cr == NULL || + (vstate->dtvs_state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) == 0; + + dp->dtdo_destructive = 0; + + for (pc = 0; pc < dp->dtdo_len && err == 0; pc++) { + dif_instr_t instr = dp->dtdo_buf[pc]; + + uint_t r1 = DIF_INSTR_R1(instr); + uint_t r2 = DIF_INSTR_R2(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t rs = DIF_INSTR_RS(instr); + uint_t label = DIF_INSTR_LABEL(instr); + uint_t v = DIF_INSTR_VAR(instr); + uint_t subr = DIF_INSTR_SUBR(instr); + uint_t type = DIF_INSTR_TYPE(instr); + uint_t op = DIF_INSTR_OP(instr); + + switch (op) { + case DIF_OP_OR: + case DIF_OP_XOR: + case DIF_OP_AND: + case DIF_OP_SLL: + case DIF_OP_SRL: + case DIF_OP_SRA: + case DIF_OP_SUB: + case DIF_OP_ADD: + case DIF_OP_MUL: + case DIF_OP_SDIV: + case DIF_OP_UDIV: + case DIF_OP_SREM: + case DIF_OP_UREM: + case DIF_OP_COPYS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_NOT: + case DIF_OP_MOV: + case DIF_OP_ALLOCS: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDSB: + case DIF_OP_LDSH: + case DIF_OP_LDSW: + case DIF_OP_LDUB: + case DIF_OP_LDUH: + case DIF_OP_LDUW: + case DIF_OP_LDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + if (kcheckload) + dp->dtdo_buf[pc] = DIF_INSTR_LOAD(op + + DIF_OP_RLDSB - DIF_OP_LDSB, r1, rd); + break; + case DIF_OP_RLDSB: + case DIF_OP_RLDSH: + case DIF_OP_RLDSW: + case DIF_OP_RLDUB: + case DIF_OP_RLDUH: + case DIF_OP_RLDUW: + case DIF_OP_RLDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_ULDSB: + case DIF_OP_ULDSH: + case DIF_OP_ULDSW: + case DIF_OP_ULDUB: + case DIF_OP_ULDUH: + case DIF_OP_ULDUW: + case DIF_OP_ULDX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STB: + case DIF_OP_STH: + case DIF_OP_STW: + case DIF_OP_STX: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to 0 address\n"); + break; + case DIF_OP_CMP: + case DIF_OP_SCMP: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_TST: + if (r1 >= nregs) + err += efunc(pc, "invalid register %u\n", r1); + if (r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_BA: + case DIF_OP_BE: + case DIF_OP_BNE: + case DIF_OP_BG: + case DIF_OP_BGU: + case DIF_OP_BGE: + case DIF_OP_BGEU: + case DIF_OP_BL: + case DIF_OP_BLU: + case DIF_OP_BLE: + case DIF_OP_BLEU: + if (label >= dp->dtdo_len) { + err += efunc(pc, "invalid branch target %u\n", + label); + } + if (label <= pc) { + err += efunc(pc, "backward branch to %u\n", + label); + } + break; + case DIF_OP_RET: + if (r1 != 0 || r2 != 0) + err += efunc(pc, "non-zero reserved bits\n"); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_NOP: + case DIF_OP_POPTS: + case DIF_OP_FLUSHTS: + if (r1 != 0 || r2 != 0 || rd != 0) + err += efunc(pc, "non-zero reserved bits\n"); + break; + case DIF_OP_SETX: + if (DIF_INSTR_INTEGER(instr) >= dp->dtdo_intlen) { + err += efunc(pc, "invalid integer ref %u\n", + DIF_INSTR_INTEGER(instr)); + } + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_SETS: + if (DIF_INSTR_STRING(instr) >= dp->dtdo_strlen) { + err += efunc(pc, "invalid string ref %u\n", + DIF_INSTR_STRING(instr)); + } + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGA: + case DIF_OP_LDTA: + if (r1 > DIF_VAR_ARRAY_MAX) + err += efunc(pc, "invalid array %u\n", r1); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_LDGS: + case DIF_OP_LDTS: + case DIF_OP_LDLS: + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: + if (v < DIF_VAR_OTHER_MIN || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + break; + case DIF_OP_STGS: + case DIF_OP_STTS: + case DIF_OP_STLS: + case DIF_OP_STGAA: + case DIF_OP_STTAA: + if (v < DIF_VAR_OTHER_UBASE || v > DIF_VAR_OTHER_MAX) + err += efunc(pc, "invalid variable %u\n", v); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + break; + case DIF_OP_CALL: + if (subr > DIF_SUBR_MAX) + err += efunc(pc, "invalid subr %u\n", subr); + if (rd >= nregs) + err += efunc(pc, "invalid register %u\n", rd); + if (rd == 0) + err += efunc(pc, "cannot write to %r0\n"); + + if (subr == DIF_SUBR_COPYOUT || + subr == DIF_SUBR_COPYOUTSTR) { + dp->dtdo_destructive = 1; + } + break; + case DIF_OP_PUSHTR: + if (type != DIF_TYPE_STRING && type != DIF_TYPE_CTF) + err += efunc(pc, "invalid ref type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + case DIF_OP_PUSHTV: + if (type != DIF_TYPE_CTF) + err += efunc(pc, "invalid val type %u\n", type); + if (r2 >= nregs) + err += efunc(pc, "invalid register %u\n", r2); + if (rs >= nregs) + err += efunc(pc, "invalid register %u\n", rs); + break; + default: + err += efunc(pc, "invalid opcode %u\n", + DIF_INSTR_OP(instr)); + } + } + + if (dp->dtdo_len != 0 && + DIF_INSTR_OP(dp->dtdo_buf[dp->dtdo_len - 1]) != DIF_OP_RET) { + err += efunc(dp->dtdo_len - 1, + "expected 'ret' as last DIF instruction\n"); + } + + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) { + /* + * If we're not returning by reference, the size must be either + * 0 or the size of one of the base types. + */ + switch (dp->dtdo_rtype.dtdt_size) { + case 0: + case sizeof (uint8_t): + case sizeof (uint16_t): + case sizeof (uint32_t): + case sizeof (uint64_t): + break; + + default: + err += efunc(dp->dtdo_len - 1, "bad return size\n"); + } + } + + for (i = 0; i < dp->dtdo_varlen && err == 0; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i], *existing = NULL; + dtrace_diftype_t *vt, *et; + uint_t id, ndx; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL && + v->dtdv_scope != DIFV_SCOPE_THREAD && + v->dtdv_scope != DIFV_SCOPE_LOCAL) { + err += efunc(i, "unrecognized variable scope %d\n", + v->dtdv_scope); + break; + } + + if (v->dtdv_kind != DIFV_KIND_ARRAY && + v->dtdv_kind != DIFV_KIND_SCALAR) { + err += efunc(i, "unrecognized variable type %d\n", + v->dtdv_kind); + break; + } + + if ((id = v->dtdv_id) > DIF_VARIABLE_MAX) { + err += efunc(i, "%d exceeds variable id limit\n", id); + break; + } + + if (id < DIF_VAR_OTHER_UBASE) + continue; + + /* + * For user-defined variables, we need to check that this + * definition is identical to any previous definition that we + * encountered. + */ + ndx = id - DIF_VAR_OTHER_UBASE; + + switch (v->dtdv_scope) { + case DIFV_SCOPE_GLOBAL: + if (VBDTCAST(int64_t)ndx < vstate->dtvs_nglobals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_globals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + + case DIFV_SCOPE_THREAD: + if (VBDTCAST(int64_t)ndx < vstate->dtvs_ntlocals) + existing = &vstate->dtvs_tlocals[ndx]; + break; + + case DIFV_SCOPE_LOCAL: + if (VBDTCAST(int64_t)ndx < vstate->dtvs_nlocals) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[ndx]) != NULL) + existing = &svar->dtsv_var; + } + + break; + } + + vt = &v->dtdv_type; + + if (vt->dtdt_flags & DIF_TF_BYREF) { + if (vt->dtdt_size == 0) { + err += efunc(i, "zero-sized variable\n"); + break; + } + + if (v->dtdv_scope == DIFV_SCOPE_GLOBAL && + vt->dtdt_size > dtrace_global_maxsize) { + err += efunc(i, "oversized by-ref global\n"); + break; + } + } + + if (existing == NULL || existing->dtdv_id == 0) + continue; + + ASSERT(existing->dtdv_id == v->dtdv_id); + ASSERT(existing->dtdv_scope == v->dtdv_scope); + + if (existing->dtdv_kind != v->dtdv_kind) + err += efunc(i, "%d changed variable kind\n", id); + + et = &existing->dtdv_type; + + if (vt->dtdt_flags != et->dtdt_flags) { + err += efunc(i, "%d changed variable type flags\n", id); + break; + } + + if (vt->dtdt_size != 0 && vt->dtdt_size != et->dtdt_size) { + err += efunc(i, "%d changed variable type size\n", id); + break; + } + } + + return (err); +} + +#ifndef VBOX +/* + * Validate a DTrace DIF object that it is to be used as a helper. Helpers + * are much more constrained than normal DIFOs. Specifically, they may + * not: + * + * 1. Make calls to subroutines other than copyin(), copyinstr() or + * miscellaneous string routines + * 2. Access DTrace variables other than the args[] array, and the + * curthread, pid, ppid, tid, execname, zonename, uid and gid variables. + * 3. Have thread-local variables. + * 4. Have dynamic variables. + */ +static int +dtrace_difo_validate_helper(dtrace_difo_t *dp) +{ + int (*efunc)(uint_t pc, const char *, ...) = dtrace_difo_err; + int err = 0; + uint_t pc; + + for (pc = 0; pc < dp->dtdo_len; pc++) { + dif_instr_t instr = dp->dtdo_buf[pc]; + + uint_t v = DIF_INSTR_VAR(instr); + uint_t subr = DIF_INSTR_SUBR(instr); + uint_t op = DIF_INSTR_OP(instr); + + switch (op) { + case DIF_OP_OR: + case DIF_OP_XOR: + case DIF_OP_AND: + case DIF_OP_SLL: + case DIF_OP_SRL: + case DIF_OP_SRA: + case DIF_OP_SUB: + case DIF_OP_ADD: + case DIF_OP_MUL: + case DIF_OP_SDIV: + case DIF_OP_UDIV: + case DIF_OP_SREM: + case DIF_OP_UREM: + case DIF_OP_COPYS: + case DIF_OP_NOT: + case DIF_OP_MOV: + case DIF_OP_RLDSB: + case DIF_OP_RLDSH: + case DIF_OP_RLDSW: + case DIF_OP_RLDUB: + case DIF_OP_RLDUH: + case DIF_OP_RLDUW: + case DIF_OP_RLDX: + case DIF_OP_ULDSB: + case DIF_OP_ULDSH: + case DIF_OP_ULDSW: + case DIF_OP_ULDUB: + case DIF_OP_ULDUH: + case DIF_OP_ULDUW: + case DIF_OP_ULDX: + case DIF_OP_STB: + case DIF_OP_STH: + case DIF_OP_STW: + case DIF_OP_STX: + case DIF_OP_ALLOCS: + case DIF_OP_CMP: + case DIF_OP_SCMP: + case DIF_OP_TST: + case DIF_OP_BA: + case DIF_OP_BE: + case DIF_OP_BNE: + case DIF_OP_BG: + case DIF_OP_BGU: + case DIF_OP_BGE: + case DIF_OP_BGEU: + case DIF_OP_BL: + case DIF_OP_BLU: + case DIF_OP_BLE: + case DIF_OP_BLEU: + case DIF_OP_RET: + case DIF_OP_NOP: + case DIF_OP_POPTS: + case DIF_OP_FLUSHTS: + case DIF_OP_SETX: + case DIF_OP_SETS: + case DIF_OP_LDGA: + case DIF_OP_LDLS: + case DIF_OP_STGS: + case DIF_OP_STLS: + case DIF_OP_PUSHTR: + case DIF_OP_PUSHTV: + break; + + case DIF_OP_LDGS: + if (v >= DIF_VAR_OTHER_UBASE) + break; + + if (v >= DIF_VAR_ARG0 && v <= DIF_VAR_ARG9) + break; + + if (v == DIF_VAR_CURTHREAD || v == DIF_VAR_PID || + v == DIF_VAR_PPID || v == DIF_VAR_TID || + v == DIF_VAR_EXECNAME || v == DIF_VAR_ZONENAME || + v == DIF_VAR_UID || v == DIF_VAR_GID) + break; + + err += efunc(pc, "illegal variable %u\n", v); + break; + + case DIF_OP_LDTA: + case DIF_OP_LDTS: + case DIF_OP_LDGAA: + case DIF_OP_LDTAA: + err += efunc(pc, "illegal dynamic variable load\n"); + break; + + case DIF_OP_STTS: + case DIF_OP_STGAA: + case DIF_OP_STTAA: + err += efunc(pc, "illegal dynamic variable store\n"); + break; + + case DIF_OP_CALL: + if (subr == DIF_SUBR_ALLOCA || + subr == DIF_SUBR_BCOPY || + subr == DIF_SUBR_COPYIN || + subr == DIF_SUBR_COPYINTO || + subr == DIF_SUBR_COPYINSTR || + subr == DIF_SUBR_INDEX || + subr == DIF_SUBR_INET_NTOA || + subr == DIF_SUBR_INET_NTOA6 || + subr == DIF_SUBR_INET_NTOP || + subr == DIF_SUBR_LLTOSTR || + subr == DIF_SUBR_RINDEX || + subr == DIF_SUBR_STRCHR || + subr == DIF_SUBR_STRJOIN || + subr == DIF_SUBR_STRRCHR || + subr == DIF_SUBR_STRSTR || + subr == DIF_SUBR_HTONS || + subr == DIF_SUBR_HTONL || + subr == DIF_SUBR_HTONLL || + subr == DIF_SUBR_NTOHS || + subr == DIF_SUBR_NTOHL || + subr == DIF_SUBR_NTOHLL) + break; + + err += efunc(pc, "invalid subr %u\n", subr); + break; + + default: + err += efunc(pc, "invalid opcode %u\n", + DIF_INSTR_OP(instr)); + } + } + + return (err); +} +#endif /* !VBOX */ + +/* + * Returns 1 if the expression in the DIF object can be cached on a per-thread + * basis; 0 if not. + */ +static int +dtrace_difo_cacheable(dtrace_difo_t *dp) +{ + VBDTTYPE(uint_t,int) i; + + if (dp == NULL) + return (0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_scope != DIFV_SCOPE_GLOBAL) + continue; + + switch (v->dtdv_id) { + case DIF_VAR_CURTHREAD: + case DIF_VAR_PID: + case DIF_VAR_TID: + case DIF_VAR_EXECNAME: + case DIF_VAR_ZONENAME: + break; + + default: + return (0); + } + } + + /* + * This DIF object may be cacheable. Now we need to look for any + * array loading instructions, any memory loading instructions, or + * any stores to thread-local variables. + */ + for (i = 0; i < dp->dtdo_len; i++) { + uint_t op = DIF_INSTR_OP(dp->dtdo_buf[i]); + + if ((op >= DIF_OP_LDSB && op <= DIF_OP_LDX) || + (op >= DIF_OP_ULDSB && op <= DIF_OP_ULDX) || + (op >= DIF_OP_RLDSB && op <= DIF_OP_RLDX) || + op == DIF_OP_LDGA || op == DIF_OP_STTS) + return (0); + } + + return (1); +} + +static void +dtrace_difo_hold(dtrace_difo_t *dp) +{ +#ifndef VBOX + VBDTTYPE(uint_t,int) i; +#endif + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + dp->dtdo_refcnt++; + ASSERT(dp->dtdo_refcnt != 0); + +#ifndef VBOX + /* + * We need to check this DIF object for references to the variable + * DIF_VAR_VTIMESTAMP. + */ + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + if (dtrace_vtime_references++ == 0) + dtrace_vtime_enable(); + } +#endif +} + +/* + * This routine calculates the dynamic variable chunksize for a given DIF + * object. The calculation is not fool-proof, and can probably be tricked by + * malicious DIF -- but it works for all compiler-generated DIF. Because this + * calculation is likely imperfect, dtrace_dynvar() is able to gracefully fail + * if a dynamic variable size exceeds the chunksize. + */ +static void +dtrace_difo_chunksize(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + uint64_t sval VBDTGCC(0); + dtrace_key_t tupregs[DIF_DTR_NREGS + 2]; /* +2 for thread and id */ + const dif_instr_t *text = dp->dtdo_buf; + uint_t pc, srd = 0; + uint_t ttop = 0; + size_t size, ksize; + uint_t id, i; + + for (pc = 0; pc < dp->dtdo_len; pc++) { + dif_instr_t instr = text[pc]; + uint_t op = DIF_INSTR_OP(instr); + uint_t rd = DIF_INSTR_RD(instr); + uint_t r1 = DIF_INSTR_R1(instr); + uint_t nkeys = 0; + uchar_t scope VBDTGCC(0); + + dtrace_key_t *key = tupregs; + + switch (op) { + case DIF_OP_SETX: + sval = dp->dtdo_inttab[DIF_INSTR_INTEGER(instr)]; + srd = rd; + continue; + + case DIF_OP_STTS: + key = &tupregs[DIF_DTR_NREGS]; + key[0].dttk_size = 0; + key[1].dttk_size = 0; + nkeys = 2; + scope = DIFV_SCOPE_THREAD; + break; + + case DIF_OP_STGAA: + case DIF_OP_STTAA: + nkeys = ttop; + + if (DIF_INSTR_OP(instr) == DIF_OP_STTAA) + key[nkeys++].dttk_size = 0; + + key[nkeys++].dttk_size = 0; + + if (op == DIF_OP_STTAA) { + scope = DIFV_SCOPE_THREAD; + } else { + scope = DIFV_SCOPE_GLOBAL; + } + + break; + + case DIF_OP_PUSHTR: + if (ttop == DIF_DTR_NREGS) + return; + + if ((srd == 0 || sval == 0) && r1 == DIF_TYPE_STRING) { + /* + * If the register for the size of the "pushtr" + * is %r0 (or the value is 0) and the type is + * a string, we'll use the system-wide default + * string size. + */ + tupregs[ttop++].dttk_size = + dtrace_strsize_default; + } else { + if (srd == 0) + return; + + tupregs[ttop++].dttk_size = sval; + } + + break; + + case DIF_OP_PUSHTV: + if (ttop == DIF_DTR_NREGS) + return; + + tupregs[ttop++].dttk_size = 0; + break; + + case DIF_OP_FLUSHTS: + ttop = 0; + break; + + case DIF_OP_POPTS: + if (ttop != 0) + ttop--; + break; + } + + sval = 0; + srd = 0; + + if (nkeys == 0) + continue; + + /* + * We have a dynamic variable allocation; calculate its size. + */ + for (ksize = 0, i = 0; i < nkeys; i++) + ksize += P2ROUNDUP(key[i].dttk_size, sizeof (uint64_t)); + + size = sizeof (dtrace_dynvar_t); + size += sizeof (dtrace_key_t) * (nkeys - 1); + size += ksize; + + /* + * Now we need to determine the size of the stored data. + */ + id = DIF_INSTR_VAR(instr); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id == id && v->dtdv_scope == scope) { + size += v->dtdv_type.dtdt_size; + break; + } + } + + if (i == dp->dtdo_varlen) + return; + + /* + * We have the size. If this is larger than the chunk size + * for our dynamic variable state, reset the chunk size. + */ + size = P2ROUNDUP(size, sizeof (uint64_t)); + + if (size > vstate->dtvs_dynvars.dtds_chunksize) + vstate->dtvs_dynvars.dtds_chunksize = size; + } +} + +static void +dtrace_difo_init(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ +#ifndef VBOX + int i, oldsvars, osz, nsz, otlocals, ntlocals; +#else + int oldsvars, osz, nsz, otlocals, ntlocals; + uint_t i; +#endif + uint_t id; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_buf != NULL && dp->dtdo_len != 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, ***svarp; + size_t dsize = 0; + uint8_t scope = v->dtdv_scope; + int *np; + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + + switch (scope) { + case DIFV_SCOPE_THREAD: + while (VBDTCAST(int64_t)id >= (otlocals = vstate->dtvs_ntlocals)) { + dtrace_difv_t *tlocals; + + if ((ntlocals = (otlocals << 1)) == 0) + ntlocals = 1; + + osz = otlocals * sizeof (dtrace_difv_t); + nsz = ntlocals * sizeof (dtrace_difv_t); + + tlocals = kmem_zalloc(nsz, KM_SLEEP); + + if (osz != 0) { + bcopy(vstate->dtvs_tlocals, + tlocals, osz); + kmem_free(vstate->dtvs_tlocals, osz); + } + + vstate->dtvs_tlocals = tlocals; + vstate->dtvs_ntlocals = ntlocals; + } + + vstate->dtvs_tlocals[id] = *v; + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = &vstate->dtvs_locals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = NCPU * (v->dtdv_type.dtdt_size + + sizeof (uint64_t)); + else + dsize = NCPU * sizeof (uint64_t); + + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = &vstate->dtvs_globals; + + if (v->dtdv_type.dtdt_flags & DIF_TF_BYREF) + dsize = v->dtdv_type.dtdt_size + + sizeof (uint64_t); + + break; + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", scope)); +#endif + } + + while (VBDTCAST(int64_t)id >= (oldsvars = *np)) { + dtrace_statvar_t **statics; + int newsvars, oldsize, newsize; + + if ((newsvars = (oldsvars << 1)) == 0) + newsvars = 1; + + oldsize = oldsvars * sizeof (dtrace_statvar_t *); + newsize = newsvars * sizeof (dtrace_statvar_t *); + + statics = kmem_zalloc(newsize, KM_SLEEP); + + if (oldsize != 0) { + bcopy(*svarp, statics, oldsize); + kmem_free(*svarp, oldsize); + } + + *svarp = statics; + *np = newsvars; + } + + if ((svar = (*svarp)[id]) == NULL) { + svar = kmem_zalloc(sizeof (dtrace_statvar_t), KM_SLEEP); + svar->dtsv_var = *v; + + if ((svar->dtsv_size = dsize) != 0) { + svar->dtsv_data = (uint64_t)(uintptr_t) + kmem_zalloc(dsize, KM_SLEEP); + } + + (*svarp)[id] = svar; + } + + svar->dtsv_refcnt++; + } + + dtrace_difo_chunksize(dp, vstate); + dtrace_difo_hold(dp); +} + +#ifndef VBOX +static dtrace_difo_t * +dtrace_difo_duplicate(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + dtrace_difo_t *new; + size_t sz; + + ASSERT(dp->dtdo_buf != NULL); + ASSERT(dp->dtdo_refcnt != 0); + + new = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); + + ASSERT(dp->dtdo_buf != NULL); + sz = dp->dtdo_len * sizeof (dif_instr_t); + new->dtdo_buf = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_buf, new->dtdo_buf, sz); + new->dtdo_len = dp->dtdo_len; + + if (dp->dtdo_strtab != NULL) { + ASSERT(dp->dtdo_strlen != 0); + new->dtdo_strtab = kmem_alloc(dp->dtdo_strlen, KM_SLEEP); + bcopy(dp->dtdo_strtab, new->dtdo_strtab, dp->dtdo_strlen); + new->dtdo_strlen = dp->dtdo_strlen; + } + + if (dp->dtdo_inttab != NULL) { + ASSERT(dp->dtdo_intlen != 0); + sz = dp->dtdo_intlen * sizeof (uint64_t); + new->dtdo_inttab = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_inttab, new->dtdo_inttab, sz); + new->dtdo_intlen = dp->dtdo_intlen; + } + + if (dp->dtdo_vartab != NULL) { + ASSERT(dp->dtdo_varlen != 0); + sz = dp->dtdo_varlen * sizeof (dtrace_difv_t); + new->dtdo_vartab = kmem_alloc(sz, KM_SLEEP); + bcopy(dp->dtdo_vartab, new->dtdo_vartab, sz); + new->dtdo_varlen = dp->dtdo_varlen; + } + + dtrace_difo_init(new, vstate); + return (new); +} +#endif + +static void +dtrace_difo_destroy(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ + VBDTTYPE(uint_t,int) i; + + ASSERT(dp->dtdo_refcnt == 0); + + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_statvar_t *svar, **svarp; + uint_t id; + uint8_t scope = v->dtdv_scope; + int *np; + + switch (scope) { + case DIFV_SCOPE_THREAD: + continue; + + case DIFV_SCOPE_LOCAL: + np = &vstate->dtvs_nlocals; + svarp = vstate->dtvs_locals; + break; + + case DIFV_SCOPE_GLOBAL: + np = &vstate->dtvs_nglobals; + svarp = vstate->dtvs_globals; + break; + + default: +#ifndef VBOX + ASSERT(0); +#else + AssertFatalMsgFailed(("%d\n", scope)); +#endif + } + + if ((id = v->dtdv_id) < DIF_VAR_OTHER_UBASE) + continue; + + id -= DIF_VAR_OTHER_UBASE; + ASSERT(VBDTCAST(int64_t)id < *np); + + svar = svarp[id]; + ASSERT(svar != NULL); + ASSERT(svar->dtsv_refcnt > 0); + + if (--svar->dtsv_refcnt > 0) + continue; + + if (svar->dtsv_size != 0) { + ASSERT(svar->dtsv_data != NULL); + kmem_free((void *)(uintptr_t)svar->dtsv_data, + svar->dtsv_size); + } + + kmem_free(svar, sizeof (dtrace_statvar_t)); + svarp[id] = NULL; + } + + kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); + kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); + kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); + kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); + + kmem_free(dp, sizeof (dtrace_difo_t)); +} + +static void +dtrace_difo_release(dtrace_difo_t *dp, dtrace_vstate_t *vstate) +{ +#ifndef VBOX + VBDTTYPE(uint_t,int) i; +#endif + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + +#ifndef VBOX + for (i = 0; i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + + if (v->dtdv_id != DIF_VAR_VTIMESTAMP) + continue; + + ASSERT(dtrace_vtime_references > 0); + if (--dtrace_vtime_references == 0) + dtrace_vtime_disable(); + } +#endif + + if (--dp->dtdo_refcnt == 0) + dtrace_difo_destroy(dp, vstate); +} + +/* + * DTrace Format Functions + */ +static uint16_t +dtrace_format_add(dtrace_state_t *state, char *str) +{ + char *fmt, **new; + uint16_t ndx, len = VBDTCAST(uint16_t)strlen(str) + 1; + + fmt = kmem_zalloc(len, KM_SLEEP); + bcopy(str, fmt, len); + + for (ndx = 0; ndx < state->dts_nformats; ndx++) { + if (state->dts_formats[ndx] == NULL) { + state->dts_formats[ndx] = fmt; + return (ndx + 1); + } + } + + if (state->dts_nformats == USHRT_MAX) { + /* + * This is only likely if a denial-of-service attack is being + * attempted. As such, it's okay to fail silently here. + */ + kmem_free(fmt, len); + return (0); + } + + /* + * For simplicity, we always resize the formats array to be exactly the + * number of formats. + */ + ndx = state->dts_nformats++; + new = kmem_alloc((ndx + 1) * sizeof (char *), KM_SLEEP); + + if (state->dts_formats != NULL) { + ASSERT(ndx != 0); + bcopy(state->dts_formats, new, ndx * sizeof (char *)); + kmem_free(state->dts_formats, ndx * sizeof (char *)); + } + + state->dts_formats = new; + state->dts_formats[ndx] = fmt; + + return (ndx + 1); +} + +static void +dtrace_format_remove(dtrace_state_t *state, uint16_t format) +{ + char *fmt; + + ASSERT(state->dts_formats != NULL); + ASSERT(format <= state->dts_nformats); + ASSERT(state->dts_formats[format - 1] != NULL); + + fmt = state->dts_formats[format - 1]; + kmem_free(fmt, strlen(fmt) + 1); + state->dts_formats[format - 1] = NULL; +} + +static void +dtrace_format_destroy(dtrace_state_t *state) +{ + int i; + + if (state->dts_nformats == 0) { + ASSERT(state->dts_formats == NULL); + return; + } + + ASSERT(state->dts_formats != NULL); + + for (i = 0; i < state->dts_nformats; i++) { + char *fmt = state->dts_formats[i]; + + if (fmt == NULL) + continue; + + kmem_free(fmt, strlen(fmt) + 1); + } + + kmem_free(state->dts_formats, state->dts_nformats * sizeof (char *)); + state->dts_nformats = 0; + state->dts_formats = NULL; +} + +/* + * DTrace Predicate Functions + */ +static dtrace_predicate_t * +dtrace_predicate_create(dtrace_difo_t *dp) +{ + dtrace_predicate_t *pred; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp->dtdo_refcnt != 0); + + pred = kmem_zalloc(sizeof (dtrace_predicate_t), KM_SLEEP); + pred->dtp_difo = dp; + pred->dtp_refcnt = 1; + + if (!dtrace_difo_cacheable(dp)) + return (pred); + + if (dtrace_predcache_id == DTRACE_CACHEIDNONE) { + /* + * This is only theoretically possible -- we have had 2^32 + * cacheable predicates on this machine. We cannot allow any + * more predicates to become cacheable: as unlikely as it is, + * there may be a thread caching a (now stale) predicate cache + * ID. (N.B.: the temptation is being successfully resisted to + * have this cmn_err() "Holy shit -- we executed this code!") + */ + return (pred); + } + + pred->dtp_cacheid = dtrace_predcache_id++; + + return (pred); +} + +static void +dtrace_predicate_hold(dtrace_predicate_t *pred) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(pred->dtp_difo != NULL && pred->dtp_difo->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + pred->dtp_refcnt++; +} + +static void +dtrace_predicate_release(dtrace_predicate_t *pred, dtrace_vstate_t *vstate) +{ +#ifdef VBOX_STRICT + dtrace_difo_t *dp = pred->dtp_difo; +#endif + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dp != NULL && dp->dtdo_refcnt != 0); + ASSERT(pred->dtp_refcnt > 0); + + if (--pred->dtp_refcnt == 0) { + dtrace_difo_release(pred->dtp_difo, vstate); + kmem_free(pred, sizeof (dtrace_predicate_t)); + } +} + +/* + * DTrace Action Description Functions + */ +static dtrace_actdesc_t * +dtrace_actdesc_create(dtrace_actkind_t kind, uint32_t ntuple, + uint64_t uarg, uint64_t arg) +{ + dtrace_actdesc_t *act; + + ASSERT(!DTRACEACT_ISPRINTFLIKE(kind) || (arg != NULL && + VBDT_IS_VALID_KRNL_ADDR(arg)) || (arg == NULL && kind == DTRACEACT_PRINTA)); + + act = kmem_zalloc(sizeof (dtrace_actdesc_t), KM_SLEEP); + act->dtad_kind = kind; + act->dtad_ntuple = ntuple; + act->dtad_uarg = uarg; + act->dtad_arg = arg; + act->dtad_refcnt = 1; + + return (act); +} + +static void +dtrace_actdesc_hold(dtrace_actdesc_t *act) +{ + ASSERT(act->dtad_refcnt >= 1); + act->dtad_refcnt++; +} + +static void +dtrace_actdesc_release(dtrace_actdesc_t *act, dtrace_vstate_t *vstate) +{ + dtrace_actkind_t kind = act->dtad_kind; + dtrace_difo_t *dp; + + ASSERT(act->dtad_refcnt >= 1); + + if (--act->dtad_refcnt != 0) + return; + + if ((dp = act->dtad_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISPRINTFLIKE(kind)) { + char *str = (char *)(uintptr_t)act->dtad_arg; + + ASSERT((str != NULL && VBDT_IS_VALID_KRNL_ADDR((uintptr_t)str)) || + (str == NULL && act->dtad_kind == DTRACEACT_PRINTA)); + + if (str != NULL) + kmem_free(str, strlen(str) + 1); + } + + kmem_free(act, sizeof (dtrace_actdesc_t)); +} + +/* + * DTrace ECB Functions + */ +static dtrace_ecb_t * +dtrace_ecb_add(dtrace_state_t *state, dtrace_probe_t *probe) +{ + dtrace_ecb_t *ecb; + dtrace_epid_t epid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + ecb = kmem_zalloc(sizeof (dtrace_ecb_t), KM_SLEEP); + ecb->dte_predicate = NULL; + ecb->dte_probe = probe; + + /* + * The default size is the size of the default action: recording + * the epid. + */ + ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); + ecb->dte_alignment = sizeof (dtrace_epid_t); + + epid = state->dts_epid++; + + if (VBDTCAST(int64_t)epid - 1 >= state->dts_necbs) { + dtrace_ecb_t **oecbs = state->dts_ecbs, **ecbs; + int necbs = state->dts_necbs << 1; + + ASSERT(epid == VBDTCAST(dtrace_epid_t)state->dts_necbs + 1); + + if (necbs == 0) { + ASSERT(oecbs == NULL); + necbs = 1; + } + + ecbs = kmem_zalloc(necbs * sizeof (*ecbs), KM_SLEEP); + + if (oecbs != NULL) + bcopy(oecbs, ecbs, state->dts_necbs * sizeof (*ecbs)); + + dtrace_membar_producer(); + state->dts_ecbs = ecbs; + + if (oecbs != NULL) { + /* + * If this state is active, we must dtrace_sync() + * before we can free the old dts_ecbs array: we're + * coming in hot, and there may be active ring + * buffer processing (which indexes into the dts_ecbs + * array) on another CPU. + */ + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + dtrace_sync(); + + kmem_free(oecbs, state->dts_necbs * sizeof (*ecbs)); + } + + dtrace_membar_producer(); + state->dts_necbs = necbs; + } + + ecb->dte_state = state; + + ASSERT(state->dts_ecbs[epid - 1] == NULL); + dtrace_membar_producer(); + state->dts_ecbs[(ecb->dte_epid = epid) - 1] = ecb; + + return (ecb); +} + +static int +dtrace_ecb_enable(dtrace_ecb_t *ecb) +{ + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + + if (probe == NULL) { + /* + * This is the NULL probe -- there's nothing to do. + */ + return (0); + } + + if (probe->dtpr_ecb == NULL) { + dtrace_provider_t *prov = probe->dtpr_provider; + + /* + * We're the first ECB on this probe. + */ + probe->dtpr_ecb = probe->dtpr_ecb_last = ecb; + + if (ecb->dte_predicate != NULL) + probe->dtpr_predcache = ecb->dte_predicate->dtp_cacheid; + + return (prov->dtpv_pops.dtps_enable(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg)); + } else { + /* + * This probe is already active. Swing the last pointer to + * point to the new ECB, and issue a dtrace_sync() to assure + * that all CPUs have seen the change. + */ + ASSERT(probe->dtpr_ecb_last != NULL); + probe->dtpr_ecb_last->dte_next = ecb; + probe->dtpr_ecb_last = ecb; + probe->dtpr_predcache = 0; + + dtrace_sync(); + return (0); + } +} + +static void +dtrace_ecb_resize(dtrace_ecb_t *ecb) +{ + uint32_t maxalign = sizeof (dtrace_epid_t); + uint32_t align = sizeof (uint8_t), offs, diff; + dtrace_action_t *act; + int wastuple = 0; + uint32_t aggbase = UINT32_MAX; + dtrace_state_t *state = ecb->dte_state; + + /* + * If we record anything, we always record the epid. (And we always + * record it first.) + */ + offs = sizeof (dtrace_epid_t); + ecb->dte_size = ecb->dte_needed = sizeof (dtrace_epid_t); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + dtrace_recdesc_t *rec = &act->dta_rec; + + if ((align = rec->dtrd_alignment) > maxalign) + maxalign = align; + + if (!wastuple && act->dta_intuple) { + /* + * This is the first record in a tuple. Align the + * offset to be at offset 4 in an 8-byte aligned + * block. + */ + diff = offs + sizeof (dtrace_aggid_t); + + if ((diff = (diff & (sizeof (uint64_t) - 1)))) + offs += sizeof (uint64_t) - diff; + + aggbase = offs - sizeof (dtrace_aggid_t); + ASSERT(!(aggbase & (sizeof (uint64_t) - 1))); + } + + /*LINTED*/ + if (rec->dtrd_size != 0 && (diff = (offs & (align - 1)))) { + /* + * The current offset is not properly aligned; align it. + */ + offs += align - diff; + } + + rec->dtrd_offset = offs; + + if (offs + rec->dtrd_size > ecb->dte_needed) { + ecb->dte_needed = offs + rec->dtrd_size; + + if (ecb->dte_needed > state->dts_needed) + state->dts_needed = ecb->dte_needed; + } + + if (DTRACEACT_ISAGG(act->dta_kind)) { + dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; + dtrace_action_t *first = agg->dtag_first, *prev; + + ASSERT(rec->dtrd_size != 0 && first != NULL); + ASSERT(wastuple); + ASSERT(aggbase != UINT32_MAX); + + agg->dtag_base = aggbase; + + while ((prev = first->dta_prev) != NULL && + DTRACEACT_ISAGG(prev->dta_kind)) { + agg = (dtrace_aggregation_t *)prev; + first = agg->dtag_first; + } + + if (prev != NULL) { + offs = prev->dta_rec.dtrd_offset + + prev->dta_rec.dtrd_size; + } else { + offs = sizeof (dtrace_epid_t); + } + wastuple = 0; + } else { + if (!act->dta_intuple) + ecb->dte_size = offs + rec->dtrd_size; + + offs += rec->dtrd_size; + } + + wastuple = act->dta_intuple; + } + + if ((act = ecb->dte_action) != NULL && + !(act->dta_kind == DTRACEACT_SPECULATE && act->dta_next == NULL) && + ecb->dte_size == sizeof (dtrace_epid_t)) { + /* + * If the size is still sizeof (dtrace_epid_t), then all + * actions store no data; set the size to 0. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = 0; + + /* + * If the needed space is still sizeof (dtrace_epid_t), then + * all actions need no additional space; set the needed + * size to 0. + */ + if (ecb->dte_needed == sizeof (dtrace_epid_t)) + ecb->dte_needed = 0; + + return; + } + + /* + * Set our alignment, and make sure that the dte_size and dte_needed + * are aligned to the size of an EPID. + */ + ecb->dte_alignment = maxalign; + ecb->dte_size = (ecb->dte_size + (sizeof (dtrace_epid_t) - 1)) & + ~(sizeof (dtrace_epid_t) - 1); + ecb->dte_needed = (ecb->dte_needed + (sizeof (dtrace_epid_t) - 1)) & + ~(sizeof (dtrace_epid_t) - 1); + ASSERT(ecb->dte_size <= ecb->dte_needed); +} + +static dtrace_action_t * +dtrace_ecb_aggregation_create(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) +{ + dtrace_aggregation_t *agg; + size_t size = sizeof (uint64_t); + int ntuple = desc->dtad_ntuple; + dtrace_action_t *act; + dtrace_recdesc_t *frec; + dtrace_aggid_t aggid; + dtrace_state_t *state = ecb->dte_state; + + agg = kmem_zalloc(sizeof (dtrace_aggregation_t), KM_SLEEP); + agg->dtag_ecb = ecb; + + ASSERT(DTRACEACT_ISAGG(desc->dtad_kind)); + + switch (desc->dtad_kind) { + case DTRACEAGG_MIN: + agg->dtag_initial = INT64_MAX; + agg->dtag_aggregate = dtrace_aggregate_min; + break; + + case DTRACEAGG_MAX: + agg->dtag_initial = (uint64_t)INT64_MIN; + agg->dtag_aggregate = dtrace_aggregate_max; + break; + + case DTRACEAGG_COUNT: + agg->dtag_aggregate = dtrace_aggregate_count; + break; + + case DTRACEAGG_QUANTIZE: + agg->dtag_aggregate = dtrace_aggregate_quantize; + size = (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) * + sizeof (uint64_t); + break; + + case DTRACEAGG_LQUANTIZE: { + uint16_t step = DTRACE_LQUANTIZE_STEP(desc->dtad_arg); + uint16_t levels = DTRACE_LQUANTIZE_LEVELS(desc->dtad_arg); + + agg->dtag_initial = desc->dtad_arg; + agg->dtag_aggregate = dtrace_aggregate_lquantize; + + if (step == 0 || levels == 0) + goto err; + + size = levels * sizeof (uint64_t) + 3 * sizeof (uint64_t); + break; + } + + case DTRACEAGG_AVG: + agg->dtag_aggregate = dtrace_aggregate_avg; + size = sizeof (uint64_t) * 2; + break; + + case DTRACEAGG_STDDEV: + agg->dtag_aggregate = dtrace_aggregate_stddev; + size = sizeof (uint64_t) * 4; + break; + + case DTRACEAGG_SUM: + agg->dtag_aggregate = dtrace_aggregate_sum; + break; + + default: + goto err; + } + + agg->dtag_action.dta_rec.dtrd_size = VBDTCAST(uint32_t)size; + + if (ntuple == 0) + goto err; + + /* + * We must make sure that we have enough actions for the n-tuple. + */ + for (act = ecb->dte_action_last; act != NULL; act = act->dta_prev) { + if (DTRACEACT_ISAGG(act->dta_kind)) + break; + + if (--ntuple == 0) { + /* + * This is the action with which our n-tuple begins. + */ + agg->dtag_first = act; + goto success; + } + } + + /* + * This n-tuple is short by ntuple elements. Return failure. + */ + ASSERT(ntuple != 0); +err: + kmem_free(agg, sizeof (dtrace_aggregation_t)); + return (NULL); + +success: + /* + * If the last action in the tuple has a size of zero, it's actually + * an expression argument for the aggregating action. + */ + ASSERT(ecb->dte_action_last != NULL); + act = ecb->dte_action_last; + + if (act->dta_kind == DTRACEACT_DIFEXPR) { + ASSERT(act->dta_difo != NULL); + + if (act->dta_difo->dtdo_rtype.dtdt_size == 0) + agg->dtag_hasarg = 1; + } + + /* + * We need to allocate an id for this aggregation. + */ + aggid = (dtrace_aggid_t)(uintptr_t)vmem_alloc(state->dts_aggid_arena, 1, + VM_BESTFIT | VM_SLEEP); + + if (VBDTCAST(int64_t)aggid - 1 >= state->dts_naggregations) { + dtrace_aggregation_t **oaggs = state->dts_aggregations; + dtrace_aggregation_t **aggs; + int naggs = state->dts_naggregations << 1; + int onaggs = state->dts_naggregations; + + ASSERT(aggid == VBDTCAST(dtrace_aggid_t)state->dts_naggregations + 1); + + if (naggs == 0) { + ASSERT(oaggs == NULL); + naggs = 1; + } + + aggs = kmem_zalloc(naggs * sizeof (*aggs), KM_SLEEP); + + if (oaggs != NULL) { + bcopy(oaggs, aggs, onaggs * sizeof (*aggs)); + kmem_free(oaggs, onaggs * sizeof (*aggs)); + } + + state->dts_aggregations = aggs; + state->dts_naggregations = naggs; + } + + ASSERT(state->dts_aggregations[aggid - 1] == NULL); + state->dts_aggregations[(agg->dtag_id = aggid) - 1] = agg; + + frec = &agg->dtag_first->dta_rec; + if (frec->dtrd_alignment < sizeof (dtrace_aggid_t)) + frec->dtrd_alignment = sizeof (dtrace_aggid_t); + + for (act = agg->dtag_first; act != NULL; act = act->dta_next) { + ASSERT(!act->dta_intuple); + act->dta_intuple = 1; + } + + return (&agg->dtag_action); +} + +static void +dtrace_ecb_aggregation_destroy(dtrace_ecb_t *ecb, dtrace_action_t *act) +{ + dtrace_aggregation_t *agg = (dtrace_aggregation_t *)act; + dtrace_state_t *state = ecb->dte_state; + dtrace_aggid_t aggid = agg->dtag_id; + + ASSERT(DTRACEACT_ISAGG(act->dta_kind)); + vmem_free(state->dts_aggid_arena, (void *)(uintptr_t)aggid, 1); + + ASSERT(state->dts_aggregations[aggid - 1] == agg); + state->dts_aggregations[aggid - 1] = NULL; + + kmem_free(agg, sizeof (dtrace_aggregation_t)); +} + +static int +dtrace_ecb_action_add(dtrace_ecb_t *ecb, dtrace_actdesc_t *desc) +{ + dtrace_action_t *action, *last; + dtrace_difo_t *dp = desc->dtad_difo; + uint32_t size = 0, align = sizeof (uint8_t), mask; + uint16_t format = 0; + dtrace_recdesc_t *rec; + dtrace_state_t *state = ecb->dte_state; + dtrace_optval_t *opt = state->dts_options, nframes VBDTUNASS(0), strsize; + uint64_t arg = desc->dtad_arg; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_action == NULL || ecb->dte_action->dta_refcnt == 1); + + if (DTRACEACT_ISAGG(desc->dtad_kind)) { + /* + * If this is an aggregating action, there must be neither + * a speculate nor a commit on the action chain. + */ + dtrace_action_t *act; + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + + if (act->dta_kind == DTRACEACT_SPECULATE) + return (EINVAL); + } + + action = dtrace_ecb_aggregation_create(ecb, desc); + + if (action == NULL) + return (EINVAL); + } else { + if (DTRACEACT_ISDESTRUCTIVE(desc->dtad_kind) || + (desc->dtad_kind == DTRACEACT_DIFEXPR && + dp != NULL && dp->dtdo_destructive)) { + state->dts_destructive = 1; + } + + switch (desc->dtad_kind) { + case DTRACEACT_PRINTF: + case DTRACEACT_PRINTA: + case DTRACEACT_SYSTEM: + case DTRACEACT_FREOPEN: + /* + * We know that our arg is a string -- turn it into a + * format. + */ + if (arg == NULL) { + ASSERT(desc->dtad_kind == DTRACEACT_PRINTA); + format = 0; + } else { + ASSERT(arg != NULL); + ASSERT(VBDT_IS_VALID_KRNL_ADDR(arg)); + format = dtrace_format_add(state, + (char *)(uintptr_t)arg); + } + + RT_FALL_THRU(); + case DTRACEACT_LIBACT: + case DTRACEACT_DIFEXPR: + if (dp == NULL) + return (EINVAL); + + if ((size = dp->dtdo_rtype.dtdt_size) != 0) + break; + + if (dp->dtdo_rtype.dtdt_kind == DIF_TYPE_STRING) { + if (!(dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + + size = opt[DTRACEOPT_STRSIZE]; + } + + break; + + case DTRACEACT_STACK: + if ((nframes = arg) == 0) { + nframes = opt[DTRACEOPT_STACKFRAMES]; + ASSERT(nframes > 0); + arg = nframes; + } + + size = VBDTCAST(uint32_t)(nframes * sizeof (pc_t)); + break; + + case DTRACEACT_JSTACK: + if ((strsize = DTRACE_USTACK_STRSIZE(arg)) == 0) + strsize = opt[DTRACEOPT_JSTACKSTRSIZE]; + + if ((nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) + nframes = opt[DTRACEOPT_JSTACKFRAMES]; + + arg = DTRACE_USTACK_ARG(nframes, strsize); + + RT_FALL_THRU(); + case DTRACEACT_USTACK: + if (desc->dtad_kind != DTRACEACT_JSTACK && + (nframes = DTRACE_USTACK_NFRAMES(arg)) == 0) { + strsize = DTRACE_USTACK_STRSIZE(arg); + nframes = opt[DTRACEOPT_USTACKFRAMES]; + ASSERT(nframes > 0); + arg = DTRACE_USTACK_ARG(nframes, strsize); + } + + /* + * Save a slot for the pid. + */ + size = VBDTCAST(uint32_t)((nframes + 1) * sizeof (uint64_t)); + size += DTRACE_USTACK_STRSIZE(arg); + size = P2ROUNDUP(size, (uint32_t)(sizeof (uintptr_t))); + + break; + + case DTRACEACT_SYM: + case DTRACEACT_MOD: + if (dp == NULL || ((size = dp->dtdo_rtype.dtdt_size) != + sizeof (uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + break; + + case DTRACEACT_USYM: + case DTRACEACT_UMOD: + case DTRACEACT_UADDR: + if (dp == NULL || + (dp->dtdo_rtype.dtdt_size != sizeof (uint64_t)) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + + /* + * We have a slot for the pid, plus a slot for the + * argument. To keep things simple (aligned with + * bitness-neutral sizing), we store each as a 64-bit + * quantity. + */ + size = 2 * sizeof (uint64_t); + break; + + case DTRACEACT_STOP: + case DTRACEACT_BREAKPOINT: + case DTRACEACT_PANIC: + break; + + case DTRACEACT_CHILL: + case DTRACEACT_DISCARD: + case DTRACEACT_RAISE: + if (dp == NULL) + return (EINVAL); + break; + + case DTRACEACT_EXIT: + if (dp == NULL || + (size = dp->dtdo_rtype.dtdt_size) != sizeof (int) || + (dp->dtdo_rtype.dtdt_flags & DIF_TF_BYREF)) + return (EINVAL); + break; + + case DTRACEACT_SPECULATE: + if (ecb->dte_size > sizeof (dtrace_epid_t)) + return (EINVAL); + + if (dp == NULL) + return (EINVAL); + + state->dts_speculates = 1; + break; + + case DTRACEACT_COMMIT: { + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + } + + if (dp == NULL) + return (EINVAL); + break; + } + + default: + return (EINVAL); + } + + if (size != 0 || desc->dtad_kind == DTRACEACT_SPECULATE) { + /* + * If this is a data-storing action or a speculate, + * we must be sure that there isn't a commit on the + * action chain. + */ + dtrace_action_t *act = ecb->dte_action; + + for (; act != NULL; act = act->dta_next) { + if (act->dta_kind == DTRACEACT_COMMIT) + return (EINVAL); + } + } + + action = kmem_zalloc(sizeof (dtrace_action_t), KM_SLEEP); + action->dta_rec.dtrd_size = size; + } + + action->dta_refcnt = 1; + rec = &action->dta_rec; + size = rec->dtrd_size; + + for (mask = sizeof (uint64_t) - 1; size != 0 && mask > 0; mask >>= 1) { + if (!(size & mask)) { + align = mask + 1; + break; + } + } + + action->dta_kind = desc->dtad_kind; + + if ((action->dta_difo = dp) != NULL) + dtrace_difo_hold(dp); + + rec->dtrd_action = action->dta_kind; + rec->dtrd_arg = arg; + rec->dtrd_uarg = desc->dtad_uarg; + rec->dtrd_alignment = (uint16_t)align; + rec->dtrd_format = format; + + if ((last = ecb->dte_action_last) != NULL) { + ASSERT(ecb->dte_action != NULL); + action->dta_prev = last; + last->dta_next = action; + } else { + ASSERT(ecb->dte_action == NULL); + ecb->dte_action = action; + } + + ecb->dte_action_last = action; + + return (0); +} + +static void +dtrace_ecb_action_remove(dtrace_ecb_t *ecb) +{ + dtrace_action_t *act = ecb->dte_action, *next; + dtrace_vstate_t *vstate = &ecb->dte_state->dts_vstate; + dtrace_difo_t *dp; + uint16_t format; + + if (act != NULL && act->dta_refcnt > 1) { + ASSERT(act->dta_next == NULL || act->dta_next->dta_refcnt == 1); + act->dta_refcnt--; + } else { + for (; act != NULL; act = next) { + next = act->dta_next; + ASSERT(next != NULL || act == ecb->dte_action_last); + ASSERT(act->dta_refcnt == 1); + + if ((format = act->dta_rec.dtrd_format) != 0) + dtrace_format_remove(ecb->dte_state, format); + + if ((dp = act->dta_difo) != NULL) + dtrace_difo_release(dp, vstate); + + if (DTRACEACT_ISAGG(act->dta_kind)) { + dtrace_ecb_aggregation_destroy(ecb, act); + } else { + kmem_free(act, sizeof (dtrace_action_t)); + } + } + } + + ecb->dte_action = NULL; + ecb->dte_action_last = NULL; + ecb->dte_size = sizeof (dtrace_epid_t); +} + +static void +dtrace_ecb_disable(dtrace_ecb_t *ecb) +{ + /* + * We disable the ECB by removing it from its probe. + */ + dtrace_ecb_t *pecb, *prev = NULL; + dtrace_probe_t *probe = ecb->dte_probe; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (probe == NULL) { + /* + * This is the NULL probe; there is nothing to disable. + */ + return; + } + + for (pecb = probe->dtpr_ecb; pecb != NULL; pecb = pecb->dte_next) { + if (pecb == ecb) + break; + prev = pecb; + } + + ASSERT(pecb != NULL); + + if (prev == NULL) { + probe->dtpr_ecb = ecb->dte_next; + } else { + prev->dte_next = ecb->dte_next; + } + + if (ecb == probe->dtpr_ecb_last) { + ASSERT(ecb->dte_next == NULL); + probe->dtpr_ecb_last = prev; + } + + /* + * The ECB has been disconnected from the probe; now sync to assure + * that all CPUs have seen the change before returning. + */ + dtrace_sync(); + + if (probe->dtpr_ecb == NULL) { + /* + * That was the last ECB on the probe; clear the predicate + * cache ID for the probe, disable it and sync one more time + * to assure that we'll never hit it again. + */ + dtrace_provider_t *prov = probe->dtpr_provider; + + ASSERT(ecb->dte_next == NULL); + ASSERT(probe->dtpr_ecb_last == NULL); + probe->dtpr_predcache = DTRACE_CACHEIDNONE; + prov->dtpv_pops.dtps_disable(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg); + dtrace_sync(); + } else { + /* + * There is at least one ECB remaining on the probe. If there + * is _exactly_ one, set the probe's predicate cache ID to be + * the predicate cache ID of the remaining ECB. + */ + ASSERT(probe->dtpr_ecb_last != NULL); + ASSERT(probe->dtpr_predcache == DTRACE_CACHEIDNONE); + + if (probe->dtpr_ecb == probe->dtpr_ecb_last) { + dtrace_predicate_t *p = probe->dtpr_ecb->dte_predicate; + + ASSERT(probe->dtpr_ecb->dte_next == NULL); + + if (p != NULL) + probe->dtpr_predcache = p->dtp_cacheid; + } + + ecb->dte_next = NULL; + } +} + +static void +dtrace_ecb_destroy(dtrace_ecb_t *ecb) +{ + dtrace_state_t *state = ecb->dte_state; + dtrace_vstate_t *vstate = &state->dts_vstate; + dtrace_predicate_t *pred; + dtrace_epid_t epid = ecb->dte_epid; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(ecb->dte_next == NULL); + ASSERT(ecb->dte_probe == NULL || ecb->dte_probe->dtpr_ecb != ecb); + + if ((pred = ecb->dte_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + dtrace_ecb_action_remove(ecb); + + ASSERT(state->dts_ecbs[epid - 1] == ecb); + state->dts_ecbs[epid - 1] = NULL; + + kmem_free(ecb, sizeof (dtrace_ecb_t)); +} + +static dtrace_ecb_t * +dtrace_ecb_create(dtrace_state_t *state, dtrace_probe_t *probe, + dtrace_enabling_t *enab) +{ + dtrace_ecb_t *ecb; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + dtrace_provider_t *prov; + dtrace_ecbdesc_t *desc = enab->dten_current; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(state != NULL); + + ecb = dtrace_ecb_add(state, probe); + ecb->dte_uarg = desc->dted_uarg; + + if ((pred = desc->dted_pred.dtpdd_predicate) != NULL) { + dtrace_predicate_hold(pred); + ecb->dte_predicate = pred; + } + + if (probe != NULL) { + /* + * If the provider shows more leg than the consumer is old + * enough to see, we need to enable the appropriate implicit + * predicate bits to prevent the ecb from activating at + * revealing times. + * + * Providers specifying DTRACE_PRIV_USER at register time + * are stating that they need the /proc-style privilege + * model to be enforced, and this is what DTRACE_COND_OWNER + * and DTRACE_COND_ZONEOWNER will then do at probe time. + */ + prov = probe->dtpr_provider; + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLPROC) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) + ecb->dte_cond |= DTRACE_COND_OWNER; + + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_ALLZONE) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_USER)) + ecb->dte_cond |= DTRACE_COND_ZONEOWNER; + + /* + * If the provider shows us kernel innards and the user + * is lacking sufficient privilege, enable the + * DTRACE_COND_USERMODE implicit predicate. + */ + if (!(state->dts_cred.dcr_visible & DTRACE_CRV_KERNEL) && + (prov->dtpv_priv.dtpp_flags & DTRACE_PRIV_KERNEL)) + ecb->dte_cond |= DTRACE_COND_USERMODE; + } + + if (dtrace_ecb_create_cache != NULL) { + /* + * If we have a cached ecb, we'll use its action list instead + * of creating our own (saving both time and space). + */ + dtrace_ecb_t *cached = dtrace_ecb_create_cache; + dtrace_action_t *act2 = cached->dte_action; + + if (act2 != NULL) { + ASSERT(act2->dta_refcnt > 0); + act2->dta_refcnt++; + ecb->dte_action = act2; + ecb->dte_action_last = cached->dte_action_last; + ecb->dte_needed = cached->dte_needed; + ecb->dte_size = cached->dte_size; + ecb->dte_alignment = cached->dte_alignment; + } + + return (ecb); + } + + for (act = desc->dted_action; act != NULL; act = act->dtad_next) { + if ((enab->dten_error = dtrace_ecb_action_add(ecb, act)) != 0) { + dtrace_ecb_destroy(ecb); + return (NULL); + } + } + + dtrace_ecb_resize(ecb); + + return (dtrace_ecb_create_cache = ecb); +} + +static int +dtrace_ecb_create_enable(dtrace_probe_t *probe, void *arg) +{ + dtrace_ecb_t *ecb; + dtrace_enabling_t *enab = arg; + dtrace_state_t *state = enab->dten_vstate->dtvs_state; + + ASSERT(state != NULL); + + if (probe != NULL && probe->dtpr_gen < enab->dten_probegen) { + /* + * This probe was created in a generation for which this + * enabling has previously created ECBs; we don't want to + * enable it again, so just kick out. + */ + return (DTRACE_MATCH_NEXT); + } + + if ((ecb = dtrace_ecb_create(state, probe, enab)) == NULL) + return (DTRACE_MATCH_DONE); + + if (dtrace_ecb_enable(ecb) < 0) + return (DTRACE_MATCH_FAIL); + + return (DTRACE_MATCH_NEXT); +} + +static dtrace_ecb_t * +dtrace_epid2ecb(dtrace_state_t *state, dtrace_epid_t id) +{ + dtrace_ecb_t *ecb; NOREF(ecb); + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || VBDTCAST(int64_t)id > state->dts_necbs) + return (NULL); + + ASSERT(state->dts_necbs > 0 && state->dts_ecbs != NULL); + ASSERT((ecb = state->dts_ecbs[id - 1]) == NULL || ecb->dte_epid == id); + + return (state->dts_ecbs[id - 1]); +} + +static dtrace_aggregation_t * +dtrace_aggid2agg(dtrace_state_t *state, dtrace_aggid_t id) +{ + dtrace_aggregation_t *agg; NOREF(agg); + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (id == 0 || VBDTCAST(int64_t)id > state->dts_naggregations) + return (NULL); + + ASSERT(state->dts_naggregations > 0 && state->dts_aggregations != NULL); + ASSERT((agg = state->dts_aggregations[id - 1]) == NULL || + agg->dtag_id == id); + + return (state->dts_aggregations[id - 1]); +} + +/* + * DTrace Buffer Functions + * + * The following functions manipulate DTrace buffers. Most of these functions + * are called in the context of establishing or processing consumer state; + * exceptions are explicitly noted. + */ + +/* + * Note: called from cross call context. This function switches the two + * buffers on a given CPU. The atomicity of this operation is assured by + * disabling interrupts while the actual switch takes place; the disabling of + * interrupts serializes the execution with any execution of dtrace_probe() on + * the same CPU. + */ +static void +dtrace_buffer_switch(dtrace_buffer_t *buf) +{ + caddr_t tomax = buf->dtb_tomax; + caddr_t xamot = buf->dtb_xamot; + dtrace_icookie_t cookie; + + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + ASSERT(!(buf->dtb_flags & DTRACEBUF_RING)); + + cookie = dtrace_interrupt_disable(); + buf->dtb_tomax = xamot; + buf->dtb_xamot = tomax; + buf->dtb_xamot_drops = buf->dtb_drops; + buf->dtb_xamot_offset = buf->dtb_offset; + buf->dtb_xamot_errors = buf->dtb_errors; + buf->dtb_xamot_flags = buf->dtb_flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + buf->dtb_errors = 0; + buf->dtb_flags &= ~(DTRACEBUF_ERROR | DTRACEBUF_DROPPED); + dtrace_interrupt_enable(cookie); +} + +#ifdef VBOX +static DECLCALLBACK(void) dtrace_buffer_switch_wrapper(RTCPUID idCpu, void *pvUser1, void *pvUser2) +{ + dtrace_buffer_switch((dtrace_buffer_t *)pvUser1); + NOREF(pvUser2); NOREF(idCpu); +} +#endif + +/* + * Note: called from cross call context. This function activates a buffer + * on a CPU. As with dtrace_buffer_switch(), the atomicity of the operation + * is guaranteed by the disabling of interrupts. + */ +static void +dtrace_buffer_activate(dtrace_state_t *state) +{ + dtrace_buffer_t *buf; + dtrace_icookie_t cookie = dtrace_interrupt_disable(); + + buf = &state->dts_buffer[VBDT_GET_CPUID()]; + + if (buf->dtb_tomax != NULL) { + /* + * We might like to assert that the buffer is marked inactive, + * but this isn't necessarily true: the buffer for the CPU + * that processes the BEGIN probe has its buffer activated + * manually. In this case, we take the (harmless) action + * re-clearing the bit INACTIVE bit. + */ + buf->dtb_flags &= ~DTRACEBUF_INACTIVE; + } + + dtrace_interrupt_enable(cookie); +} + +#ifdef VBOX +static DECLCALLBACK(void) dtrace_buffer_activate_wrapper(RTCPUID idCpu, void *pvUser1, void *pvUser2) +{ + dtrace_buffer_activate((dtrace_state_t *)pvUser1); + NOREF(pvUser2); NOREF(idCpu); +} +#endif + +static int +dtrace_buffer_alloc(dtrace_buffer_t *bufs, size_t size, int flags, + processorid_t cpu) +{ +#ifndef VBOX + cpu_t *cp; +#else + RTCPUSET CpuSet; + unsigned iCpu; +#endif + dtrace_buffer_t *buf; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (VBDTCAST(int64_t)size > dtrace_nonroot_maxsize +#ifndef VBOX + && !PRIV_POLICY_CHOICE(CRED(), PRIV_ALL, B_FALSE) +#endif + ) + return (EFBIG); + +#ifndef VBOX + cp = cpu_list; +#else + RTMpGetSet(&CpuSet); +#endif + +#ifndef VBOX + do { + if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) + continue; + + buf = &bufs[cp->cpu_id]; +#else + for (iCpu = 0; iCpu < RTCPUSET_MAX_CPUS; iCpu++) { + if ( !RTCpuSetIsMember(&CpuSet, iCpu) + || (cpu != (processorid_t)DTRACE_CPUALL && cpu != iCpu)) + continue; + + buf = &bufs[iCpu]; +#endif + + /* + * If there is already a buffer allocated for this CPU, it + * is only possible that this is a DR event. In this case, + * the buffer size must match our specified size. + */ + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + continue; + } + + ASSERT(buf->dtb_xamot == NULL); + + if ((buf->dtb_tomax = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; + + buf->dtb_size = size; + buf->dtb_flags = flags; + buf->dtb_offset = 0; + buf->dtb_drops = 0; + + if (flags & DTRACEBUF_NOSWITCH) + continue; + + if ((buf->dtb_xamot = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + goto err; +#ifndef VBOX + } while ((cp = cp->cpu_next) != cpu_list); +#else + } +#endif + + return (0); + +err: +#ifndef VBOX + cp = cpu_list; + + do { + if (cpu != DTRACE_CPUALL && cpu != cp->cpu_id) + continue; + + buf = &bufs[cp->cpu_id]; +#else + for (iCpu = 0; iCpu < RTCPUSET_MAX_CPUS; iCpu++) { + if ( !RTCpuSetIsMember(&CpuSet, iCpu) + || (cpu != (processorid_t)DTRACE_CPUALL && cpu != iCpu)) + continue; + + buf = &bufs[iCpu]; +#endif + + if (buf->dtb_xamot != NULL) { + ASSERT(buf->dtb_tomax != NULL); + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_xamot, size); + } + + if (buf->dtb_tomax != NULL) { + ASSERT(buf->dtb_size == size); + kmem_free(buf->dtb_tomax, size); + } + + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + buf->dtb_size = 0; +#ifndef VBOX + } while ((cp = cp->cpu_next) != cpu_list); +#else + } +#endif + + return (ENOMEM); +} + +/* + * Note: called from probe context. This function just increments the drop + * count on a buffer. It has been made a function to allow for the + * possibility of understanding the source of mysterious drop counts. (A + * problem for which one may be particularly disappointed that DTrace cannot + * be used to understand DTrace.) + */ +static void +dtrace_buffer_drop(dtrace_buffer_t *buf) +{ + buf->dtb_drops++; +} + +/* + * Note: called from probe context. This function is called to reserve space + * in a buffer. If mstate is non-NULL, sets the scratch base and size in the + * mstate. Returns the new offset in the buffer, or a negative value if an + * error has occurred. + */ +static intptr_t +dtrace_buffer_reserve(dtrace_buffer_t *buf, size_t needed, size_t align, + dtrace_state_t *state, dtrace_mstate_t *mstate) +{ + intptr_t offs = buf->dtb_offset, soffs; + intptr_t woffs; + caddr_t tomax; + size_t total; + + if (buf->dtb_flags & DTRACEBUF_INACTIVE) + return (-1); + + if ((tomax = buf->dtb_tomax) == NULL) { + dtrace_buffer_drop(buf); + return (-1); + } + + if (!(buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL))) { + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if (VBDTCAST(uintptr_t)(soffs = offs + needed) > buf->dtb_size) { + dtrace_buffer_drop(buf); + return (-1); + } + + if (mstate == NULL) + return (offs); + + mstate->dtms_scratch_base = (uintptr_t)tomax + soffs; + mstate->dtms_scratch_size = buf->dtb_size - soffs; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return (offs); + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (state->dts_activity != DTRACE_ACTIVITY_COOLDOWN && + (buf->dtb_flags & DTRACEBUF_FULL)) + return (-1); + goto out; + } + + total = needed + (offs & (align - 1)); + + /* + * For a ring buffer, life is quite a bit more complicated. Before + * we can store any padding, we need to adjust our wrapping offset. + * (If we've never before wrapped or we're not about to, no adjustment + * is required.) + */ + if ((buf->dtb_flags & DTRACEBUF_WRAPPED) || + offs + total > buf->dtb_size) { + woffs = buf->dtb_xamot_offset; + + if (offs + total > buf->dtb_size) { + /* + * We can't fit in the end of the buffer. First, a + * sanity check that we can fit in the buffer at all. + */ + if (total > buf->dtb_size) { + dtrace_buffer_drop(buf); + return (-1); + } + + /* + * We're going to be storing at the top of the buffer, + * so now we need to deal with the wrapped offset. We + * only reset our wrapped offset to 0 if it is + * currently greater than the current offset. If it + * is less than the current offset, it is because a + * previous allocation induced a wrap -- but the + * allocation didn't subsequently take the space due + * to an error or false predicate evaluation. In this + * case, we'll just leave the wrapped offset alone: if + * the wrapped offset hasn't been advanced far enough + * for this allocation, it will be adjusted in the + * lower loop. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + if (woffs >= offs) + woffs = 0; + } else { + woffs = 0; + } + + /* + * Now we know that we're going to be storing to the + * top of the buffer and that there is room for us + * there. We need to clear the buffer from the current + * offset to the end (there may be old gunk there). + */ + while (VBDTCAST(uintptr_t)offs < buf->dtb_size) + tomax[offs++] = 0; + + /* + * We need to set our offset to zero. And because we + * are wrapping, we need to set the bit indicating as + * much. We can also adjust our needed space back + * down to the space required by the ECB -- we know + * that the top of the buffer is aligned. + */ + offs = 0; + total = needed; + buf->dtb_flags |= DTRACEBUF_WRAPPED; + } else { + /* + * There is room for us in the buffer, so we simply + * need to check the wrapped offset. + */ + if (woffs < offs) { + /* + * The wrapped offset is less than the offset. + * This can happen if we allocated buffer space + * that induced a wrap, but then we didn't + * subsequently take the space due to an error + * or false predicate evaluation. This is + * okay; we know that _this_ allocation isn't + * going to induce a wrap. We still can't + * reset the wrapped offset to be zero, + * however: the space may have been trashed in + * the previous failed probe attempt. But at + * least the wrapped offset doesn't need to + * be adjusted at all... + */ + goto out; + } + } + + while (VBDTCAST(uintptr_t)offs + total > VBDTCAST(uintptr_t)woffs) { + dtrace_epid_t epid = *(uint32_t *)(tomax + woffs); + size_t size; + + if (epid == DTRACE_EPIDNONE) { + size = sizeof (uint32_t); + } else { + ASSERT(VBDTCAST(int64_t)epid <= state->dts_necbs); + ASSERT(state->dts_ecbs[epid - 1] != NULL); + + size = state->dts_ecbs[epid - 1]->dte_size; + } + + ASSERT(woffs + size <= buf->dtb_size); + ASSERT(size != 0); + + if (woffs + size == buf->dtb_size) { + /* + * We've reached the end of the buffer; we want + * to set the wrapped offset to 0 and break + * out. However, if the offs is 0, then we're + * in a strange edge-condition: the amount of + * space that we want to reserve plus the size + * of the record that we're overwriting is + * greater than the size of the buffer. This + * is problematic because if we reserve the + * space but subsequently don't consume it (due + * to a failed predicate or error) the wrapped + * offset will be 0 -- yet the EPID at offset 0 + * will not be committed. This situation is + * relatively easy to deal with: if we're in + * this case, the buffer is indistinguishable + * from one that hasn't wrapped; we need only + * finish the job by clearing the wrapped bit, + * explicitly setting the offset to be 0, and + * zero'ing out the old data in the buffer. + */ + if (offs == 0) { + buf->dtb_flags &= ~DTRACEBUF_WRAPPED; + buf->dtb_offset = 0; + woffs = total; + + while (VBDTCAST(uintptr_t)woffs < buf->dtb_size) + tomax[woffs++] = 0; + } + + woffs = 0; + break; + } + + woffs += size; + } + + /* + * We have a wrapped offset. It may be that the wrapped offset + * has become zero -- that's okay. + */ + buf->dtb_xamot_offset = woffs; + } + +out: + /* + * Now we can plow the buffer with any necessary padding. + */ + while (offs & (align - 1)) { + /* + * Assert that our alignment is off by a number which + * is itself sizeof (uint32_t) aligned. + */ + ASSERT(!((align - (offs & (align - 1))) & + (sizeof (uint32_t) - 1))); + DTRACE_STORE(uint32_t, tomax, offs, DTRACE_EPIDNONE); + offs += sizeof (uint32_t); + } + + if (buf->dtb_flags & DTRACEBUF_FILL) { + if (offs + needed > buf->dtb_size - state->dts_reserve) { + buf->dtb_flags |= DTRACEBUF_FULL; + return (-1); + } + } + + if (mstate == NULL) + return (offs); + + /* + * For ring buffers and fill buffers, the scratch space is always + * the inactive buffer. + */ + mstate->dtms_scratch_base = (uintptr_t)buf->dtb_xamot; + mstate->dtms_scratch_size = buf->dtb_size; + mstate->dtms_scratch_ptr = mstate->dtms_scratch_base; + + return (offs); +} + +static void +dtrace_buffer_polish(dtrace_buffer_t *buf) +{ + ASSERT(buf->dtb_flags & DTRACEBUF_RING); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (!(buf->dtb_flags & DTRACEBUF_WRAPPED)) + return; + + /* + * We need to polish the ring buffer. There are three cases: + * + * - The first (and presumably most common) is that there is no gap + * between the buffer offset and the wrapped offset. In this case, + * there is nothing in the buffer that isn't valid data; we can + * mark the buffer as polished and return. + * + * - The second (less common than the first but still more common + * than the third) is that there is a gap between the buffer offset + * and the wrapped offset, and the wrapped offset is larger than the + * buffer offset. This can happen because of an alignment issue, or + * can happen because of a call to dtrace_buffer_reserve() that + * didn't subsequently consume the buffer space. In this case, + * we need to zero the data from the buffer offset to the wrapped + * offset. + * + * - The third (and least common) is that there is a gap between the + * buffer offset and the wrapped offset, but the wrapped offset is + * _less_ than the buffer offset. This can only happen because a + * call to dtrace_buffer_reserve() induced a wrap, but the space + * was not subsequently consumed. In this case, we need to zero the + * space from the offset to the end of the buffer _and_ from the + * top of the buffer to the wrapped offset. + */ + if (buf->dtb_offset < buf->dtb_xamot_offset) { + bzero(buf->dtb_tomax + buf->dtb_offset, + buf->dtb_xamot_offset - buf->dtb_offset); + } + + if (buf->dtb_offset > buf->dtb_xamot_offset) { + bzero(buf->dtb_tomax + buf->dtb_offset, + buf->dtb_size - buf->dtb_offset); + bzero(buf->dtb_tomax, buf->dtb_xamot_offset); + } +} + +static void +dtrace_buffer_free(dtrace_buffer_t *bufs) +{ + int i; + + for (i = 0; i < NCPU; i++) { + dtrace_buffer_t *buf = &bufs[i]; + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + ASSERT(buf->dtb_size == 0); + continue; + } + + if (buf->dtb_xamot != NULL) { + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + kmem_free(buf->dtb_xamot, buf->dtb_size); + } + + kmem_free(buf->dtb_tomax, buf->dtb_size); + buf->dtb_size = 0; + buf->dtb_tomax = NULL; + buf->dtb_xamot = NULL; + } +} + +/* + * DTrace Enabling Functions + */ +static dtrace_enabling_t * +dtrace_enabling_create(dtrace_vstate_t *vstate) +{ + dtrace_enabling_t *enab; + + enab = kmem_zalloc(sizeof (dtrace_enabling_t), KM_SLEEP); + enab->dten_vstate = vstate; + + return (enab); +} + +static void +dtrace_enabling_add(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb) +{ + dtrace_ecbdesc_t **ndesc; + size_t osize, nsize; + + /* + * We can't add to enablings after we've enabled them, or after we've + * retained them. + */ + ASSERT(enab->dten_probegen == 0); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + + if (enab->dten_ndesc < enab->dten_maxdesc) { + enab->dten_desc[enab->dten_ndesc++] = ecb; + return; + } + + osize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + + if (enab->dten_maxdesc == 0) { + enab->dten_maxdesc = 1; + } else { + enab->dten_maxdesc <<= 1; + } + + ASSERT(enab->dten_ndesc < enab->dten_maxdesc); + + nsize = enab->dten_maxdesc * sizeof (dtrace_enabling_t *); + ndesc = kmem_zalloc(nsize, KM_SLEEP); + bcopy(enab->dten_desc, ndesc, osize); + kmem_free(enab->dten_desc, osize); + + enab->dten_desc = ndesc; + enab->dten_desc[enab->dten_ndesc++] = ecb; +} + +static void +dtrace_enabling_addlike(dtrace_enabling_t *enab, dtrace_ecbdesc_t *ecb, + dtrace_probedesc_t *pd) +{ + dtrace_ecbdesc_t *new; + dtrace_predicate_t *pred; + dtrace_actdesc_t *act; + + /* + * We're going to create a new ECB description that matches the + * specified ECB in every way, but has the specified probe description. + */ + new = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); + + if ((pred = ecb->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_hold(pred); + + for (act = ecb->dted_action; act != NULL; act = act->dtad_next) + dtrace_actdesc_hold(act); + + new->dted_action = ecb->dted_action; + new->dted_pred = ecb->dted_pred; + new->dted_probe = *pd; + new->dted_uarg = ecb->dted_uarg; + + dtrace_enabling_add(enab, new); +} + +#ifndef VBOX +static void +dtrace_enabling_dump(dtrace_enabling_t *enab) +{ + int i; + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_probedesc_t *desc = &enab->dten_desc[i]->dted_probe; + + cmn_err(CE_NOTE, "enabling probe %d (%s:%s:%s:%s)", i, + desc->dtpd_provider, desc->dtpd_mod, + desc->dtpd_func, desc->dtpd_name); + } +} +#endif /* !VBOX */ + +static void +dtrace_enabling_destroy(dtrace_enabling_t *enab) +{ + int i; + dtrace_ecbdesc_t *ep; + dtrace_vstate_t *vstate = enab->dten_vstate; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_actdesc_t *act, *next; + dtrace_predicate_t *pred; + + ep = enab->dten_desc[i]; + + if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) + dtrace_predicate_release(pred, vstate); + + for (act = ep->dted_action; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + kmem_free(ep, sizeof (dtrace_ecbdesc_t)); + } + + kmem_free(enab->dten_desc, + enab->dten_maxdesc * sizeof (dtrace_enabling_t *)); + + /* + * If this was a retained enabling, decrement the dts_nretained count + * and take it off of the dtrace_retained list. + */ + if (enab->dten_prev != NULL || enab->dten_next != NULL || + dtrace_retained == enab) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + ASSERT(enab->dten_vstate->dtvs_state->dts_nretained > 0); + enab->dten_vstate->dtvs_state->dts_nretained--; + dtrace_retained_gen++; + } + + if (enab->dten_prev == NULL) { + if (dtrace_retained == enab) { + dtrace_retained = enab->dten_next; + + if (dtrace_retained != NULL) + dtrace_retained->dten_prev = NULL; + } + } else { + ASSERT(enab != dtrace_retained); + ASSERT(dtrace_retained != NULL); + enab->dten_prev->dten_next = enab->dten_next; + } + + if (enab->dten_next != NULL) { + ASSERT(dtrace_retained != NULL); + enab->dten_next->dten_prev = enab->dten_prev; + } + + kmem_free(enab, sizeof (dtrace_enabling_t)); +} + +static int +dtrace_enabling_retain(dtrace_enabling_t *enab) +{ + dtrace_state_t *state; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(enab->dten_next == NULL && enab->dten_prev == NULL); + ASSERT(enab->dten_vstate != NULL); + + state = enab->dten_vstate->dtvs_state; + ASSERT(state != NULL); + + /* + * We only allow each state to retain dtrace_retain_max enablings. + */ + if (state->dts_nretained >= dtrace_retain_max) + return (ENOSPC); + + state->dts_nretained++; + dtrace_retained_gen++; + + if (dtrace_retained == NULL) { + dtrace_retained = enab; + return (0); + } + + enab->dten_next = dtrace_retained; + dtrace_retained->dten_prev = enab; + dtrace_retained = enab; + + return (0); +} + +static int +dtrace_enabling_replicate(dtrace_state_t *state, dtrace_probedesc_t *match, + dtrace_probedesc_t *create) +{ + dtrace_enabling_t *new, *enab; + int found = 0, err = ENOENT; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(strlen(match->dtpd_provider) < DTRACE_PROVNAMELEN); + ASSERT(strlen(match->dtpd_mod) < DTRACE_MODNAMELEN); + ASSERT(strlen(match->dtpd_func) < DTRACE_FUNCNAMELEN); + ASSERT(strlen(match->dtpd_name) < DTRACE_NAMELEN); + + new = dtrace_enabling_create(&state->dts_vstate); + + /* + * Iterate over all retained enablings, looking for enablings that + * match the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + int i; + + /* + * dtvs_state can only be NULL for helper enablings -- and + * helper enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * Now iterate over each probe description; we're looking for + * an exact match to the specified probe description. + */ + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + dtrace_probedesc_t *pd = &ep->dted_probe; + + if (strcmp(pd->dtpd_provider, match->dtpd_provider)) + continue; + + if (strcmp(pd->dtpd_mod, match->dtpd_mod)) + continue; + + if (strcmp(pd->dtpd_func, match->dtpd_func)) + continue; + + if (strcmp(pd->dtpd_name, match->dtpd_name)) + continue; + + /* + * We have a winning probe! Add it to our growing + * enabling. + */ + found = 1; + dtrace_enabling_addlike(new, ep, create); + } + } + + if (!found || (err = dtrace_enabling_retain(new)) != 0) { + dtrace_enabling_destroy(new); + return (err); + } + + return (0); +} + +static void +dtrace_enabling_retract(dtrace_state_t *state) +{ + dtrace_enabling_t *enab, *next; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + /* + * Iterate over all retained enablings, destroy the enablings retained + * for the specified state. + */ + for (enab = dtrace_retained; enab != NULL; enab = next) { + next = enab->dten_next; + + /* + * dtvs_state can only be NULL for helper enablings -- and + * helper enablings can't be retained. + */ + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state == state) { + ASSERT(state->dts_nretained > 0); + dtrace_enabling_destroy(enab); + } + } + + ASSERT(state->dts_nretained == 0); +} + +static int +dtrace_enabling_match(dtrace_enabling_t *enab, int *nmatched) +{ + int i = 0; + int total_matched = 0, matched = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(MUTEX_HELD(&dtrace_lock)); + + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + + enab->dten_current = ep; + enab->dten_error = 0; + + /* + * If a provider failed to enable a probe then get out and + * let the consumer know we failed. + */ + if ((matched = dtrace_probe_enable(&ep->dted_probe, enab)) < 0) + return (EBUSY); + + total_matched += matched; + + if (enab->dten_error != 0) { + /* + * If we get an error half-way through enabling the + * probes, we kick out -- perhaps with some number of + * them enabled. Leaving enabled probes enabled may + * be slightly confusing for user-level, but we expect + * that no one will attempt to actually drive on in + * the face of such errors. If this is an anonymous + * enabling (indicated with a NULL nmatched pointer), + * we cmn_err() a message. We aren't expecting to + * get such an error -- such as it can exist at all, + * it would be a result of corrupted DOF in the driver + * properties. + */ + if (nmatched == NULL) { + cmn_err(CE_WARN, "dtrace_enabling_match() " + "error on %p: %d", (void *)ep, + enab->dten_error); + } + + return (enab->dten_error); + } + } + + enab->dten_probegen = dtrace_probegen; + if (nmatched != NULL) + *nmatched = total_matched; + + return (0); +} + +static void +dtrace_enabling_matchall(void) +{ + dtrace_enabling_t *enab; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + /* + * Iterate over all retained enablings to see if any probes match + * against them. We only perform this operation on enablings for which + * we have sufficient permissions by virtue of being in the global zone + * or in the same zone as the DTrace client. Because we can be called + * after dtrace_detach() has been called, we cannot assert that there + * are retained enablings. We can safely load from dtrace_retained, + * however: the taskq_destroy() at the end of dtrace_detach() will + * block pending our completion. + */ + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { +#ifndef VBOX + cred_t *cr = enab->dten_vstate->dtvs_state->dts_cred.dcr_cred; + + if (INGLOBALZONE(curproc) || + cr != NULL && getzoneid() == crgetzoneid(cr)) +#endif + (void) dtrace_enabling_match(enab, NULL); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); +} + +/* + * If an enabling is to be enabled without having matched probes (that is, if + * dtrace_state_go() is to be called on the underlying dtrace_state_t), the + * enabling must be _primed_ by creating an ECB for every ECB description. + * This must be done to assure that we know the number of speculations, the + * number of aggregations, the minimum buffer size needed, etc. before we + * transition out of DTRACE_ACTIVITY_INACTIVE. To do this without actually + * enabling any probes, we create ECBs for every ECB decription, but with a + * NULL probe -- which is exactly what this function does. + */ +static void +dtrace_enabling_prime(dtrace_state_t *state) +{ + dtrace_enabling_t *enab; + int i; + + for (enab = dtrace_retained; enab != NULL; enab = enab->dten_next) { + ASSERT(enab->dten_vstate->dtvs_state != NULL); + + if (enab->dten_vstate->dtvs_state != state) + continue; + + /* + * We don't want to prime an enabling more than once, lest + * we allow a malicious user to induce resource exhaustion. + * (The ECBs that result from priming an enabling aren't + * leaked -- but they also aren't deallocated until the + * consumer state is destroyed.) + */ + if (enab->dten_primed) + continue; + + for (i = 0; i < enab->dten_ndesc; i++) { + enab->dten_current = enab->dten_desc[i]; + (void) dtrace_probe_enable(NULL, enab); + } + + enab->dten_primed = 1; + } +} + +/* + * Called to indicate that probes should be provided due to retained + * enablings. This is implemented in terms of dtrace_probe_provide(), but it + * must take an initial lap through the enabling calling the dtps_provide() + * entry point explicitly to allow for autocreated probes. + */ +static void +dtrace_enabling_provide(dtrace_provider_t *prv) +{ + int i, all = 0; + dtrace_probedesc_t desc; + dtrace_genid_t gen; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&dtrace_provider_lock)); + + if (prv == NULL) { + all = 1; + prv = dtrace_provider; + } + + do { + dtrace_enabling_t *enab; + void *parg = prv->dtpv_arg; + +retry: + gen = dtrace_retained_gen; + for (enab = dtrace_retained; enab != NULL; + enab = enab->dten_next) { + for (i = 0; i < enab->dten_ndesc; i++) { + desc = enab->dten_desc[i]->dted_probe; + mutex_exit(&dtrace_lock); + prv->dtpv_pops.dtps_provide(parg, &desc); + mutex_enter(&dtrace_lock); + /* + * Process the retained enablings again if + * they have changed while we weren't holding + * dtrace_lock. + */ + if (gen != dtrace_retained_gen) + goto retry; + } + } + } while (all && (prv = prv->dtpv_next) != NULL); + + mutex_exit(&dtrace_lock); + dtrace_probe_provide(NULL, all ? NULL : prv); + mutex_enter(&dtrace_lock); +} + +/* + * DTrace DOF Functions + */ +/*ARGSUSED*/ +static void +dtrace_dof_error(dof_hdr_t *dof, const char *str) +{ + RT_NOREF_PV(dof); + + if (dtrace_err_verbose) + cmn_err(CE_WARN, "failed to process DOF: %s", str); + +#ifdef DTRACE_ERRDEBUG + dtrace_errdebug(str); +#endif +} + +/* + * Create DOF out of a currently enabled state. Right now, we only create + * DOF containing the run-time options -- but this could be expanded to create + * complete DOF representing the enabled state. + */ +static dof_hdr_t * +dtrace_dof_create(dtrace_state_t *state) +{ + dof_hdr_t *dof; + dof_sec_t *sec; + dof_optdesc_t *opt; + int i, len = sizeof (dof_hdr_t) + + roundup(sizeof (dof_sec_t), sizeof (uint64_t)) + + sizeof (dof_optdesc_t) * DTRACEOPT_MAX; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + dof = kmem_zalloc(len, KM_SLEEP); + dof->dofh_ident[DOF_ID_MAG0] = DOF_MAG_MAG0; + dof->dofh_ident[DOF_ID_MAG1] = DOF_MAG_MAG1; + dof->dofh_ident[DOF_ID_MAG2] = DOF_MAG_MAG2; + dof->dofh_ident[DOF_ID_MAG3] = DOF_MAG_MAG3; + + dof->dofh_ident[DOF_ID_MODEL] = DOF_MODEL_NATIVE; + dof->dofh_ident[DOF_ID_ENCODING] = DOF_ENCODE_NATIVE; + dof->dofh_ident[DOF_ID_VERSION] = DOF_VERSION; + dof->dofh_ident[DOF_ID_DIFVERS] = DIF_VERSION; + dof->dofh_ident[DOF_ID_DIFIREG] = DIF_DIR_NREGS; + dof->dofh_ident[DOF_ID_DIFTREG] = DIF_DTR_NREGS; + + dof->dofh_flags = 0; + dof->dofh_hdrsize = sizeof (dof_hdr_t); + dof->dofh_secsize = sizeof (dof_sec_t); + dof->dofh_secnum = 1; /* only DOF_SECT_OPTDESC */ + dof->dofh_secoff = sizeof (dof_hdr_t); + dof->dofh_loadsz = len; + dof->dofh_filesz = len; + dof->dofh_pad = 0; + + /* + * Fill in the option section header... + */ + sec = (dof_sec_t *)((uintptr_t)dof + sizeof (dof_hdr_t)); + sec->dofs_type = DOF_SECT_OPTDESC; + sec->dofs_align = sizeof (uint64_t); + sec->dofs_flags = DOF_SECF_LOAD; + sec->dofs_entsize = sizeof (dof_optdesc_t); + + opt = (dof_optdesc_t *)((uintptr_t)sec + + roundup(sizeof (dof_sec_t), sizeof (uint64_t))); + + sec->dofs_offset = (uintptr_t)opt - (uintptr_t)dof; + sec->dofs_size = sizeof (dof_optdesc_t) * DTRACEOPT_MAX; + + for (i = 0; i < DTRACEOPT_MAX; i++) { + opt[i].dofo_option = i; + opt[i].dofo_strtab = DOF_SECIDX_NONE; + opt[i].dofo_value = state->dts_options[i]; + } + + return (dof); +} + +static dof_hdr_t * +dtrace_dof_copyin(uintptr_t uarg, int *errp) +{ + dof_hdr_t hdr, *dof; + + ASSERT(!MUTEX_HELD(&dtrace_lock)); + + /* + * First, we're going to copyin() the sizeof (dof_hdr_t). + */ + if (copyin((void *)uarg, &hdr, sizeof (hdr)) != 0) { + dtrace_dof_error(NULL, "failed to copyin DOF header"); + *errp = EFAULT; + return (NULL); + } + + /* + * Now we'll allocate the entire DOF and copy it in -- provided + * that the length isn't outrageous. + */ + if (hdr.dofh_loadsz >= VBDTCAST(uint64_t)dtrace_dof_maxsize) { + dtrace_dof_error(&hdr, "load size exceeds maximum"); + *errp = E2BIG; + return (NULL); + } + + if (hdr.dofh_loadsz < sizeof (hdr)) { + dtrace_dof_error(&hdr, "invalid load size"); + *errp = EINVAL; + return (NULL); + } + + dof = kmem_alloc(hdr.dofh_loadsz, KM_SLEEP); + + if (copyin((void *)uarg, dof, hdr.dofh_loadsz) != 0 || + dof->dofh_loadsz != hdr.dofh_loadsz) { + kmem_free(dof, hdr.dofh_loadsz); + *errp = EFAULT; + return (NULL); + } + + return (dof); +} + +#ifndef VBOX +static dof_hdr_t * +dtrace_dof_property(const char *name) +{ +#ifndef VBOX + uchar_t *buf; + uint64_t loadsz; + unsigned int len, i; + dof_hdr_t *dof; + + /* + * Unfortunately, array of values in .conf files are always (and + * only) interpreted to be integer arrays. We must read our DOF + * as an integer array, and then squeeze it into a byte array. + */ + if (ddi_prop_lookup_int_array(DDI_DEV_T_ANY, dtrace_devi, 0, + (char *)name, (int **)&buf, &len) != DDI_PROP_SUCCESS) + return (NULL); + + for (i = 0; i < len; i++) + buf[i] = (uchar_t)(((int *)buf)[i]); + + if (len < sizeof (dof_hdr_t)) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "truncated header"); + return (NULL); + } + + if (len < (loadsz = ((dof_hdr_t *)buf)->dofh_loadsz)) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "truncated DOF"); + return (NULL); + } + + if (loadsz >= dtrace_dof_maxsize) { + ddi_prop_free(buf); + dtrace_dof_error(NULL, "oversized DOF"); + return (NULL); + } + + dof = kmem_alloc(loadsz, KM_SLEEP); + bcopy(buf, dof, loadsz); + ddi_prop_free(buf); + + return (dof); +#else /* VBOX */ + RT_NOREF_PV(name); + return (NULL); +#endif /* VBOX */ +} +#endif /* !VBOX */ + +static void +dtrace_dof_destroy(dof_hdr_t *dof) +{ + kmem_free(dof, dof->dofh_loadsz); +} + +/* + * Return the dof_sec_t pointer corresponding to a given section index. If the + * index is not valid, dtrace_dof_error() is called and NULL is returned. If + * a type other than DOF_SECT_NONE is specified, the header is checked against + * this type and NULL is returned if the types do not match. + */ +static dof_sec_t * +dtrace_dof_sect(dof_hdr_t *dof, uint32_t type, dof_secidx_t i) +{ + dof_sec_t *sec = (dof_sec_t *)(uintptr_t) + ((uintptr_t)dof + dof->dofh_secoff + i * dof->dofh_secsize); + + if (i >= dof->dofh_secnum) { + dtrace_dof_error(dof, "referenced section index is invalid"); + return (NULL); + } + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "referenced section is not loadable"); + return (NULL); + } + + if (type != DOF_SECT_NONE && type != sec->dofs_type) { + dtrace_dof_error(dof, "referenced section is the wrong type"); + return (NULL); + } + + return (sec); +} + +static dtrace_probedesc_t * +dtrace_dof_probedesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_probedesc_t *desc) +{ + dof_probedesc_t *probe; + dof_sec_t *strtab; + uintptr_t daddr = (uintptr_t)dof; + uintptr_t str; + size_t size; + + if (sec->dofs_type != DOF_SECT_PROBEDESC) { + dtrace_dof_error(dof, "invalid probe section"); + return (NULL); + } + + if (sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in probe description"); + return (NULL); + } + + if (sec->dofs_offset + sizeof (dof_probedesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated probe description"); + return (NULL); + } + + probe = (dof_probedesc_t *)(uintptr_t)(daddr + sec->dofs_offset); + strtab = dtrace_dof_sect(dof, DOF_SECT_STRTAB, probe->dofp_strtab); + + if (strtab == NULL) + return (NULL); + + str = daddr + strtab->dofs_offset; + size = strtab->dofs_size; + + if (probe->dofp_provider >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe provider"); + return (NULL); + } + + (void) strncpy(desc->dtpd_provider, + (char *)(str + probe->dofp_provider), + MIN(DTRACE_PROVNAMELEN - 1, size - probe->dofp_provider)); + + if (probe->dofp_mod >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe module"); + return (NULL); + } + + (void) strncpy(desc->dtpd_mod, (char *)(str + probe->dofp_mod), + MIN(DTRACE_MODNAMELEN - 1, size - probe->dofp_mod)); + + if (probe->dofp_func >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe function"); + return (NULL); + } + + (void) strncpy(desc->dtpd_func, (char *)(str + probe->dofp_func), + MIN(DTRACE_FUNCNAMELEN - 1, size - probe->dofp_func)); + + if (probe->dofp_name >= strtab->dofs_size) { + dtrace_dof_error(dof, "corrupt probe name"); + return (NULL); + } + + (void) strncpy(desc->dtpd_name, (char *)(str + probe->dofp_name), + MIN(DTRACE_NAMELEN - 1, size - probe->dofp_name)); + + return (desc); +} + +static dtrace_difo_t * +dtrace_dof_difo(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_difo_t *dp; + size_t ttl = 0; + dof_difohdr_t *dofd; + uintptr_t daddr = (uintptr_t)dof; + size_t max = dtrace_difo_maxsize; + int i, l, n; + + static const struct { + int section; + int bufoffs; + int lenoffs; + int entsize; + int align; + const char *msg; + } difo[] = { + { DOF_SECT_DIF, offsetof(dtrace_difo_t, dtdo_buf), + offsetof(dtrace_difo_t, dtdo_len), sizeof (dif_instr_t), + sizeof (dif_instr_t), "multiple DIF sections" }, + + { DOF_SECT_INTTAB, offsetof(dtrace_difo_t, dtdo_inttab), + offsetof(dtrace_difo_t, dtdo_intlen), sizeof (uint64_t), + sizeof (uint64_t), "multiple integer tables" }, + + { DOF_SECT_STRTAB, offsetof(dtrace_difo_t, dtdo_strtab), + offsetof(dtrace_difo_t, dtdo_strlen), 0, + sizeof (char), "multiple string tables" }, + + { DOF_SECT_VARTAB, offsetof(dtrace_difo_t, dtdo_vartab), + offsetof(dtrace_difo_t, dtdo_varlen), sizeof (dtrace_difv_t), + sizeof (uint_t), "multiple variable tables" }, + + { DOF_SECT_NONE, 0, 0, 0, NULL } + }; + + if (sec->dofs_type != DOF_SECT_DIFOHDR) { + dtrace_dof_error(dof, "invalid DIFO header section"); + return (NULL); + } + + if (sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad alignment in DIFO header"); + return (NULL); + } + + if (sec->dofs_size < sizeof (dof_difohdr_t) || + sec->dofs_size % sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "bad size in DIFO header"); + return (NULL); + } + + dofd = (dof_difohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); + n = (sec->dofs_size - sizeof (*dofd)) / sizeof (dof_secidx_t) + 1; + + dp = kmem_zalloc(sizeof (dtrace_difo_t), KM_SLEEP); + dp->dtdo_rtype = dofd->dofd_rtype; + + for (l = 0; l < n; l++) { + dof_sec_t *subsec; + void **bufp; + uint32_t *lenp; + + if ((subsec = dtrace_dof_sect(dof, DOF_SECT_NONE, + dofd->dofd_links[l])) == NULL) + goto err; /* invalid section link */ + + if (ttl + subsec->dofs_size > max) { + dtrace_dof_error(dof, "exceeds maximum size"); + goto err; + } + + ttl += subsec->dofs_size; + + for (i = 0; difo[i].section != DOF_SECT_NONE; i++) { + if (subsec->dofs_type != VBDTCAST(uint32_t)difo[i].section) + continue; + + if (!(subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "section not loaded"); + goto err; + } + + if (subsec->dofs_align != VBDTCAST(uint32_t)difo[i].align) { + dtrace_dof_error(dof, "bad alignment"); + goto err; + } + + bufp = (void **)((uintptr_t)dp + difo[i].bufoffs); + lenp = (uint32_t *)((uintptr_t)dp + difo[i].lenoffs); + + if (*bufp != NULL) { + dtrace_dof_error(dof, difo[i].msg); + goto err; + } + + if (VBDTCAST(uint32_t)difo[i].entsize != subsec->dofs_entsize) { + dtrace_dof_error(dof, "entry size mismatch"); + goto err; + } + + if (subsec->dofs_entsize != 0 && + (subsec->dofs_size % subsec->dofs_entsize) != 0) { + dtrace_dof_error(dof, "corrupt entry size"); + goto err; + } + + *lenp = subsec->dofs_size; + *bufp = kmem_alloc(subsec->dofs_size, KM_SLEEP); + bcopy((char *)(uintptr_t)(daddr + subsec->dofs_offset), + *bufp, subsec->dofs_size); + + if (subsec->dofs_entsize != 0) + *lenp /= subsec->dofs_entsize; + + break; + } + + /* + * If we encounter a loadable DIFO sub-section that is not + * known to us, assume this is a broken program and fail. + */ + if (difo[i].section == DOF_SECT_NONE && + (subsec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "unrecognized DIFO subsection"); + goto err; + } + } + + if (dp->dtdo_buf == NULL) { + /* + * We can't have a DIF object without DIF text. + */ + dtrace_dof_error(dof, "missing DIF text"); + goto err; + } + + /* + * Before we validate the DIF object, run through the variable table + * looking for the strings -- if any of their size are under, we'll set + * their size to be the system-wide default string size. Note that + * this should _not_ happen if the "strsize" option has been set -- + * in this case, the compiler should have set the size to reflect the + * setting of the option. + */ + for (i = 0; VBDTCAST(unsigned)i < dp->dtdo_varlen; i++) { + dtrace_difv_t *v = &dp->dtdo_vartab[i]; + dtrace_diftype_t *t = &v->dtdv_type; + + if (v->dtdv_id < DIF_VAR_OTHER_UBASE) + continue; + + if (t->dtdt_kind == DIF_TYPE_STRING && t->dtdt_size == 0) + t->dtdt_size = VBDTCAST(uint32_t)dtrace_strsize_default; + } + + if (dtrace_difo_validate(dp, vstate, DIF_DIR_NREGS, cr) != 0) + goto err; + + dtrace_difo_init(dp, vstate); + return (dp); + +err: + kmem_free(dp->dtdo_buf, dp->dtdo_len * sizeof (dif_instr_t)); + kmem_free(dp->dtdo_inttab, dp->dtdo_intlen * sizeof (uint64_t)); + kmem_free(dp->dtdo_strtab, dp->dtdo_strlen); + kmem_free(dp->dtdo_vartab, dp->dtdo_varlen * sizeof (dtrace_difv_t)); + + kmem_free(dp, sizeof (dtrace_difo_t)); + return (NULL); +} + +static dtrace_predicate_t * +dtrace_dof_predicate(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_difo_t *dp; + + if ((dp = dtrace_dof_difo(dof, sec, vstate, cr)) == NULL) + return (NULL); + + return (dtrace_predicate_create(dp)); +} + +static dtrace_actdesc_t * +dtrace_dof_actdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_actdesc_t *act, *first = NULL, *last = NULL, *next; + dof_actdesc_t *desc; + dof_sec_t *difosec; + size_t offs; + uintptr_t daddr = (uintptr_t)dof; + uint64_t arg; + dtrace_actkind_t kind; + + if (sec->dofs_type != DOF_SECT_ACTDESC) { + dtrace_dof_error(dof, "invalid action section"); + return (NULL); + } + + if (sec->dofs_offset + sizeof (dof_actdesc_t) > dof->dofh_loadsz) { + dtrace_dof_error(dof, "truncated action description"); + return (NULL); + } + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in action description"); + return (NULL); + } + + if (sec->dofs_size < sec->dofs_entsize) { + dtrace_dof_error(dof, "section entry size exceeds total size"); + return (NULL); + } + + if (sec->dofs_entsize != sizeof (dof_actdesc_t)) { + dtrace_dof_error(dof, "bad entry size in action description"); + return (NULL); + } + + if (sec->dofs_size / sec->dofs_entsize > dtrace_actions_max) { + dtrace_dof_error(dof, "actions exceed dtrace_actions_max"); + return (NULL); + } + + for (offs = 0; offs < sec->dofs_size; offs += sec->dofs_entsize) { + desc = (dof_actdesc_t *)(daddr + + (uintptr_t)sec->dofs_offset + offs); + kind = (dtrace_actkind_t)desc->dofa_kind; + + if (DTRACEACT_ISPRINTFLIKE(kind) && + (kind != DTRACEACT_PRINTA || + desc->dofa_strtab != DOF_SECIDX_NONE)) { + dof_sec_t *strtab; + char *str, *fmt; + uint64_t i; + + /* + * printf()-like actions must have a format string. + */ + if ((strtab = dtrace_dof_sect(dof, + DOF_SECT_STRTAB, desc->dofa_strtab)) == NULL) + goto err; + + str = (char *)((uintptr_t)dof + + (uintptr_t)strtab->dofs_offset); + + for (i = desc->dofa_arg; i < strtab->dofs_size; i++) { + if (str[i] == '\0') + break; + } + + if (i >= strtab->dofs_size) { + dtrace_dof_error(dof, "bogus format string"); + goto err; + } + + if (i == desc->dofa_arg) { + dtrace_dof_error(dof, "empty format string"); + goto err; + } + + i -= desc->dofa_arg; + fmt = kmem_alloc(i + 1, KM_SLEEP); + bcopy(&str[desc->dofa_arg], fmt, i + 1); + arg = (uint64_t)(uintptr_t)fmt; + } else { + if (kind == DTRACEACT_PRINTA) { + ASSERT(desc->dofa_strtab == DOF_SECIDX_NONE); + arg = 0; + } else { + arg = desc->dofa_arg; + } + } + + act = dtrace_actdesc_create(kind, desc->dofa_ntuple, + desc->dofa_uarg, arg); + + if (last != NULL) { + last->dtad_next = act; + } else { + first = act; + } + + last = act; + + if (desc->dofa_difo == DOF_SECIDX_NONE) + continue; + + if ((difosec = dtrace_dof_sect(dof, + DOF_SECT_DIFOHDR, desc->dofa_difo)) == NULL) + goto err; + + act->dtad_difo = dtrace_dof_difo(dof, difosec, vstate, cr); + + if (act->dtad_difo == NULL) + goto err; + } + + ASSERT(first != NULL); + return (first); + +err: + for (act = first; act != NULL; act = next) { + next = act->dtad_next; + dtrace_actdesc_release(act, vstate); + } + + return (NULL); +} + +static dtrace_ecbdesc_t * +dtrace_dof_ecbdesc(dof_hdr_t *dof, dof_sec_t *sec, dtrace_vstate_t *vstate, + cred_t *cr) +{ + dtrace_ecbdesc_t *ep; + dof_ecbdesc_t *ecb; + dtrace_probedesc_t *desc; + dtrace_predicate_t *pred = NULL; + + if (sec->dofs_size < sizeof (dof_ecbdesc_t)) { + dtrace_dof_error(dof, "truncated ECB description"); + return (NULL); + } + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in ECB description"); + return (NULL); + } + + ecb = (dof_ecbdesc_t *)((uintptr_t)dof + (uintptr_t)sec->dofs_offset); + sec = dtrace_dof_sect(dof, DOF_SECT_PROBEDESC, ecb->dofe_probes); + + if (sec == NULL) + return (NULL); + + ep = kmem_zalloc(sizeof (dtrace_ecbdesc_t), KM_SLEEP); + ep->dted_uarg = ecb->dofe_uarg; + desc = &ep->dted_probe; + + if (dtrace_dof_probedesc(dof, sec, desc) == NULL) + goto err; + + if (ecb->dofe_pred != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, + DOF_SECT_DIFOHDR, ecb->dofe_pred)) == NULL) + goto err; + + if ((pred = dtrace_dof_predicate(dof, sec, vstate, cr)) == NULL) + goto err; + + ep->dted_pred.dtpdd_predicate = pred; + } + + if (ecb->dofe_actions != DOF_SECIDX_NONE) { + if ((sec = dtrace_dof_sect(dof, + DOF_SECT_ACTDESC, ecb->dofe_actions)) == NULL) + goto err; + + ep->dted_action = dtrace_dof_actdesc(dof, sec, vstate, cr); + + if (ep->dted_action == NULL) + goto err; + } + + return (ep); + +err: + if (pred != NULL) + dtrace_predicate_release(pred, vstate); + kmem_free(ep, sizeof (dtrace_ecbdesc_t)); + return (NULL); +} + +/* + * Apply the relocations from the specified 'sec' (a DOF_SECT_URELHDR) to the + * specified DOF. At present, this amounts to simply adding 'ubase' to the + * site of any user SETX relocations to account for load object base address. + * In the future, if we need other relocations, this function can be extended. + */ +static int +dtrace_dof_relocate(dof_hdr_t *dof, dof_sec_t *sec, uint64_t ubase) +{ + uintptr_t daddr = (uintptr_t)dof; + dof_relohdr_t *dofr = + (dof_relohdr_t *)(uintptr_t)(daddr + sec->dofs_offset); + dof_sec_t *ss, *rs, *ts; + dof_relodesc_t *r; + uint_t i, n; + + if (sec->dofs_size < sizeof (dof_relohdr_t) || + sec->dofs_align != sizeof (dof_secidx_t)) { + dtrace_dof_error(dof, "invalid relocation header"); + return (-1); + } + + ss = dtrace_dof_sect(dof, DOF_SECT_STRTAB, dofr->dofr_strtab); + rs = dtrace_dof_sect(dof, DOF_SECT_RELTAB, dofr->dofr_relsec); + ts = dtrace_dof_sect(dof, DOF_SECT_NONE, dofr->dofr_tgtsec); + + if (ss == NULL || rs == NULL || ts == NULL) + return (-1); /* dtrace_dof_error() has been called already */ + + if (rs->dofs_entsize < sizeof (dof_relodesc_t) || + rs->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "invalid relocation section"); + return (-1); + } + + r = (dof_relodesc_t *)(uintptr_t)(daddr + rs->dofs_offset); + n = rs->dofs_size / rs->dofs_entsize; + + for (i = 0; i < n; i++) { + uintptr_t taddr = daddr + ts->dofs_offset + r->dofr_offset; + + switch (r->dofr_type) { + case DOF_RELO_NONE: + break; + case DOF_RELO_SETX: + if (r->dofr_offset >= ts->dofs_size || r->dofr_offset + + sizeof (uint64_t) > ts->dofs_size) { + dtrace_dof_error(dof, "bad relocation offset"); + return (-1); + } + + if (!IS_P2ALIGNED(taddr, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned setx relo"); + return (-1); + } + + *(uint64_t *)taddr += ubase; + break; + default: + dtrace_dof_error(dof, "invalid relocation type"); + return (-1); + } + + r = (dof_relodesc_t *)((uintptr_t)r + rs->dofs_entsize); + } + + return (0); +} + +/* + * The dof_hdr_t passed to dtrace_dof_slurp() should be a partially validated + * header: it should be at the front of a memory region that is at least + * sizeof (dof_hdr_t) in size -- and then at least dof_hdr.dofh_loadsz in + * size. It need not be validated in any other way. + */ +static int +dtrace_dof_slurp(dof_hdr_t *dof, dtrace_vstate_t *vstate, cred_t *cr, + dtrace_enabling_t **enabp, uint64_t ubase, int noprobes) +{ + uint64_t len = dof->dofh_loadsz, seclen; + uintptr_t daddr = (uintptr_t)dof; + dtrace_ecbdesc_t *ep; + dtrace_enabling_t *enab; + uint_t i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dof->dofh_loadsz >= sizeof (dof_hdr_t)); + + /* + * Check the DOF header identification bytes. In addition to checking + * valid settings, we also verify that unused bits/bytes are zeroed so + * we can use them later without fear of regressing existing binaries. + */ + if (bcmp(&dof->dofh_ident[DOF_ID_MAG0], + DOF_MAG_STRING, DOF_MAG_STRLEN) != 0) { + dtrace_dof_error(dof, "DOF magic string mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_ILP32 && + dof->dofh_ident[DOF_ID_MODEL] != DOF_MODEL_LP64) { + dtrace_dof_error(dof, "DOF has invalid data model"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_ENCODING] != DOF_ENCODE_NATIVE) { + dtrace_dof_error(dof, "DOF encoding mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_2) { + dtrace_dof_error(dof, "DOF version mismatch"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFVERS] != DIF_VERSION_2) { + dtrace_dof_error(dof, "DOF uses unsupported instruction set"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFIREG] > DIF_DIR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many integer registers"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_DIFTREG] > DIF_DTR_NREGS) { + dtrace_dof_error(dof, "DOF uses too many tuple registers"); + return (-1); + } + + for (i = DOF_ID_PAD; i < DOF_ID_SIZE; i++) { + if (dof->dofh_ident[i] != 0) { + dtrace_dof_error(dof, "DOF has invalid ident byte set"); + return (-1); + } + } + + if (dof->dofh_flags & ~DOF_FL_VALID) { + dtrace_dof_error(dof, "DOF has invalid flag bits set"); + return (-1); + } + + if (dof->dofh_secsize == 0) { + dtrace_dof_error(dof, "zero section header size"); + return (-1); + } + + /* + * Check that the section headers don't exceed the amount of DOF + * data. Note that we cast the section size and number of sections + * to uint64_t's to prevent possible overflow in the multiplication. + */ + seclen = (uint64_t)dof->dofh_secnum * (uint64_t)dof->dofh_secsize; + + if (dof->dofh_secoff > len || seclen > len || + dof->dofh_secoff + seclen > len) { + dtrace_dof_error(dof, "truncated section headers"); + return (-1); + } + + if (!IS_P2ALIGNED(dof->dofh_secoff, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned section headers"); + return (-1); + } + + if (!IS_P2ALIGNED(dof->dofh_secsize, sizeof (uint64_t))) { + dtrace_dof_error(dof, "misaligned section size"); + return (-1); + } + + /* + * Take an initial pass through the section headers to be sure that + * the headers don't have stray offsets. If the 'noprobes' flag is + * set, do not permit sections relating to providers, probes, or args. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (noprobes) { + switch (sec->dofs_type) { + case DOF_SECT_PROVIDER: + case DOF_SECT_PROBES: + case DOF_SECT_PRARGS: + case DOF_SECT_PROFFS: + dtrace_dof_error(dof, "illegal sections " + "for enabling"); + return (-1); + } + } + + if (DOF_SEC_ISLOADABLE(sec->dofs_type) && + !(sec->dofs_flags & DOF_SECF_LOAD)) { + dtrace_dof_error(dof, "loadable section with load " + "flag unset"); + return (-1); + } + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; /* just ignore non-loadable sections */ + + if (sec->dofs_align & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "bad section alignment"); + return (-1); + } + + if (sec->dofs_offset & (sec->dofs_align - 1)) { + dtrace_dof_error(dof, "misaligned section"); + return (-1); + } + + if (sec->dofs_offset > len || sec->dofs_size > len || + sec->dofs_offset + sec->dofs_size > len) { + dtrace_dof_error(dof, "corrupt section header"); + return (-1); + } + + if (sec->dofs_type == DOF_SECT_STRTAB && *((char *)daddr + + sec->dofs_offset + sec->dofs_size - 1) != '\0') { + dtrace_dof_error(dof, "non-terminating string table"); + return (-1); + } + } + + /* + * Take a second pass through the sections and locate and perform any + * relocations that are present. We do this after the first pass to + * be sure that all sections have had their headers validated. + */ + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (!(sec->dofs_flags & DOF_SECF_LOAD)) + continue; /* skip sections that are not loadable */ + + switch (sec->dofs_type) { + case DOF_SECT_URELHDR: + if (dtrace_dof_relocate(dof, sec, ubase) != 0) + return (-1); + break; + } + } + + if ((enab = *enabp) == NULL) + enab = *enabp = dtrace_enabling_create(vstate); + + for (i = 0; i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(daddr + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_ECBDESC) + continue; + + if ((ep = dtrace_dof_ecbdesc(dof, sec, vstate, cr)) == NULL) { + dtrace_enabling_destroy(enab); + *enabp = NULL; + return (-1); + } + + dtrace_enabling_add(enab, ep); + } + + return (0); +} + +/* + * Process DOF for any options. This routine assumes that the DOF has been + * at least processed by dtrace_dof_slurp(). + */ +static int +dtrace_dof_options(dof_hdr_t *dof, dtrace_state_t *state) +{ + int i, rval; + uint32_t entsize; + size_t offs; + dof_optdesc_t *desc; + + for (i = 0; VBDTCAST(unsigned)i < dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)((uintptr_t)dof + + (uintptr_t)dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_OPTDESC) + continue; + + if (sec->dofs_align != sizeof (uint64_t)) { + dtrace_dof_error(dof, "bad alignment in " + "option description"); + return (EINVAL); + } + + if ((entsize = sec->dofs_entsize) == 0) { + dtrace_dof_error(dof, "zeroed option entry size"); + return (EINVAL); + } + + if (entsize < sizeof (dof_optdesc_t)) { + dtrace_dof_error(dof, "bad option entry size"); + return (EINVAL); + } + + for (offs = 0; offs < sec->dofs_size; offs += entsize) { + desc = (dof_optdesc_t *)((uintptr_t)dof + + (uintptr_t)sec->dofs_offset + offs); + + if (desc->dofo_strtab != DOF_SECIDX_NONE) { + dtrace_dof_error(dof, "non-zero option string"); + return (EINVAL); + } + + if (desc->dofo_value == VBDTCAST(uint64_t)DTRACEOPT_UNSET) { + dtrace_dof_error(dof, "unset option"); + return (EINVAL); + } + + if ((rval = dtrace_state_option(state, + desc->dofo_option, desc->dofo_value)) != 0) { + dtrace_dof_error(dof, "rejected option"); + return (rval); + } + } + } + + return (0); +} + +/* + * DTrace Consumer State Functions + */ +VBDTSTATIC int +dtrace_dstate_init(dtrace_dstate_t *dstate, size_t size) +{ + size_t hashsize, maxper, min, chunksize = dstate->dtds_chunksize; + void *base; + uintptr_t limit; + dtrace_dynvar_t *dvar, *next, *start; + VBDTTYPE(size_t,int) i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(dstate->dtds_base == NULL && dstate->dtds_percpu == NULL); + + bzero(dstate, sizeof (dtrace_dstate_t)); + + if ((dstate->dtds_chunksize = chunksize) == 0) + dstate->dtds_chunksize = DTRACE_DYNVAR_CHUNKSIZE; + + if (size < (min = dstate->dtds_chunksize + sizeof (dtrace_dynhash_t))) + size = min; + + if ((base = kmem_zalloc(size, KM_NOSLEEP)) == NULL) + return (ENOMEM); + + dstate->dtds_size = size; + dstate->dtds_base = base; + dstate->dtds_percpu = kmem_cache_alloc(dtrace_state_cache, KM_SLEEP); + bzero(dstate->dtds_percpu, NCPU * sizeof (dtrace_dstate_percpu_t)); + + hashsize = size / (dstate->dtds_chunksize + sizeof (dtrace_dynhash_t)); + + if (hashsize != 1 && (hashsize & 1)) + hashsize--; + + dstate->dtds_hashsize = hashsize; + dstate->dtds_hash = dstate->dtds_base; + + /* + * Set all of our hash buckets to point to the single sink, and (if + * it hasn't already been set), set the sink's hash value to be the + * sink sentinel value. The sink is needed for dynamic variable + * lookups to know that they have iterated over an entire, valid hash + * chain. + */ + for (i = 0; i < hashsize; i++) + dstate->dtds_hash[i].dtdh_chain = &dtrace_dynhash_sink; + + if (dtrace_dynhash_sink.dtdv_hashval != DTRACE_DYNHASH_SINK) + dtrace_dynhash_sink.dtdv_hashval = DTRACE_DYNHASH_SINK; + + /* + * Determine number of active CPUs. Divide free list evenly among + * active CPUs. + */ + start = (dtrace_dynvar_t *) + ((uintptr_t)base + hashsize * sizeof (dtrace_dynhash_t)); + limit = (uintptr_t)base + size; + + maxper = (limit - (uintptr_t)start) / NCPU; + maxper = (maxper / dstate->dtds_chunksize) * dstate->dtds_chunksize; + + for (i = 0; i < NCPU; i++) { + dstate->dtds_percpu[i].dtdsc_free = dvar = start; + + /* + * If we don't even have enough chunks to make it once through + * NCPUs, we're just going to allocate everything to the first + * CPU. And if we're on the last CPU, we're going to allocate + * whatever is left over. In either case, we set the limit to + * be the limit of the dynamic variable space. + */ + if (maxper == 0 || i == NCPU - 1) { + limit = (uintptr_t)base + size; + start = NULL; + } else { + limit = (uintptr_t)start + maxper; + start = (dtrace_dynvar_t *)limit; + } + + ASSERT(limit <= (uintptr_t)base + size); + + for (;;) { + next = (dtrace_dynvar_t *)((uintptr_t)dvar + + dstate->dtds_chunksize); + + if ((uintptr_t)next + dstate->dtds_chunksize >= limit) + break; + + dvar->dtdv_next = next; + dvar = next; + } + + if (maxper == 0) + break; + } + + return (0); +} + +VBDTSTATIC void +dtrace_dstate_fini(dtrace_dstate_t *dstate) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (dstate->dtds_base == NULL) + return; + + kmem_free(dstate->dtds_base, dstate->dtds_size); + kmem_cache_free(dtrace_state_cache, dstate->dtds_percpu); +} + +static void +dtrace_vstate_fini(dtrace_vstate_t *vstate) +{ + /* + * Logical XOR, where are you? + */ + ASSERT((vstate->dtvs_nglobals == 0) ^ (vstate->dtvs_globals != NULL)); + + if (vstate->dtvs_nglobals > 0) { + kmem_free(vstate->dtvs_globals, vstate->dtvs_nglobals * + sizeof (dtrace_statvar_t *)); + } + + if (vstate->dtvs_ntlocals > 0) { + kmem_free(vstate->dtvs_tlocals, vstate->dtvs_ntlocals * + sizeof (dtrace_difv_t)); + } + + ASSERT((vstate->dtvs_nlocals == 0) ^ (vstate->dtvs_locals != NULL)); + + if (vstate->dtvs_nlocals > 0) { + kmem_free(vstate->dtvs_locals, vstate->dtvs_nlocals * + sizeof (dtrace_statvar_t *)); + } +} + +static void +dtrace_state_clean(dtrace_state_t *state) +{ + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) + return; + + dtrace_dynvar_clean(&state->dts_vstate.dtvs_dynvars); + dtrace_speculation_clean(state); +} +#ifdef VBOX +static DECLCALLBACK(void) dtrace_state_clean_timer(PRTTIMER pTimer, void *pvUser, uint64_t iTick) +{ + dtrace_state_clean((dtrace_state_t *)pvUser); + NOREF(pTimer); NOREF(iTick); +} +#endif + +static void +dtrace_state_deadman(dtrace_state_t *state) +{ + hrtime_t now; + + dtrace_sync(); + + now = dtrace_gethrtime(); + + if (state != dtrace_anon.dta_state && + now - state->dts_laststatus >= dtrace_deadman_user) + return; + + /* + * We must be sure that dts_alive never appears to be less than the + * value upon entry to dtrace_state_deadman(), and because we lack a + * dtrace_cas64(), we cannot store to it atomically. We thus instead + * store INT64_MAX to it, followed by a memory barrier, followed by + * the new value. This assures that dts_alive never appears to be + * less than its true value, regardless of the order in which the + * stores to the underlying storage are issued. + */ + state->dts_alive = INT64_MAX; + dtrace_membar_producer(); + state->dts_alive = now; +} + +#ifdef VBOX +static DECLCALLBACK(void) dtrace_state_deadman_timer(PRTTIMER pTimer, void *pvUser, uint64_t iTick) +{ + dtrace_state_deadman((dtrace_state_t *)pvUser); + NOREF(pTimer); NOREF(iTick); +} +#endif + +VBDTSTATIC dtrace_state_t * +#ifdef VBOX +dtrace_state_create(cred_t *cr) +#else +dtrace_state_create(dev_t *devp, cred_t *cr) +#endif +{ +#ifndef VBOX + minor_t minor; + major_t major; +#endif + char c[30]; + dtrace_state_t *state; + dtrace_optval_t *opt; + int bufsize = NCPU * sizeof (dtrace_buffer_t), i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + +#ifndef VBOX + minor = (minor_t)(uintptr_t)vmem_alloc(dtrace_minor, 1, + VM_BESTFIT | VM_SLEEP); + + if (ddi_soft_state_zalloc(dtrace_softstate, minor) != DDI_SUCCESS) { + vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); + return (NULL); + } + + state = ddi_get_soft_state(dtrace_softstate, minor); +#else + state = kmem_zalloc(sizeof (*state), KM_SLEEP); + if (!state) { + return (NULL); + } +#endif + state->dts_epid = DTRACE_EPIDNONE + 1; + +#ifndef VBOX + (void) snprintf(c, sizeof (c), "dtrace_aggid_%d", minor); +#else + (void) snprintf(c, sizeof (c), "dtrace_aggid_%p", state); +#endif +#ifndef VBOX /* Avoid idProbe = UINT32_MAX as it is used as invalid value by VTG. */ + state->dts_aggid_arena = vmem_create(c, (void *)1, UINT32_MAX, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); +#else + state->dts_aggid_arena = vmem_create(c, (void *)(uintptr_t)1, _1G, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); +#endif + +#ifndef VBOX + if (devp != NULL) { + major = getemajor(*devp); + } else { + major = ddi_driver_major(dtrace_devi); + } + + state->dts_dev = makedevice(major, minor); + + if (devp != NULL) + *devp = state->dts_dev; +#endif + + /* + * We allocate NCPU buffers. On the one hand, this can be quite + * a bit of memory per instance (nearly 36K on a Starcat). On the + * other hand, it saves an additional memory reference in the probe + * path. + */ + state->dts_buffer = kmem_zalloc(bufsize, KM_SLEEP); + state->dts_aggbuffer = kmem_zalloc(bufsize, KM_SLEEP); + state->dts_cleaner = CYCLIC_NONE; + state->dts_deadman = CYCLIC_NONE; + state->dts_vstate.dtvs_state = state; + + for (i = 0; i < DTRACEOPT_MAX; i++) + state->dts_options[i] = DTRACEOPT_UNSET; + + /* + * Set the default options. + */ + opt = state->dts_options; + opt[DTRACEOPT_BUFPOLICY] = DTRACEOPT_BUFPOLICY_SWITCH; + opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_AUTO; + opt[DTRACEOPT_NSPEC] = dtrace_nspec_default; + opt[DTRACEOPT_SPECSIZE] = dtrace_specsize_default; + opt[DTRACEOPT_CPU] = (dtrace_optval_t)DTRACE_CPUALL; + opt[DTRACEOPT_STRSIZE] = dtrace_strsize_default; + opt[DTRACEOPT_STACKFRAMES] = dtrace_stackframes_default; + opt[DTRACEOPT_USTACKFRAMES] = dtrace_ustackframes_default; + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_default; + opt[DTRACEOPT_AGGRATE] = dtrace_aggrate_default; + opt[DTRACEOPT_SWITCHRATE] = dtrace_switchrate_default; + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_default; + opt[DTRACEOPT_JSTACKFRAMES] = dtrace_jstackframes_default; + opt[DTRACEOPT_JSTACKSTRSIZE] = dtrace_jstackstrsize_default; + + state->dts_activity = DTRACE_ACTIVITY_INACTIVE; + + /* + * Depending on the user credentials, we set flag bits which alter probe + * visibility or the amount of destructiveness allowed. In the case of + * actual anonymous tracing, or the possession of all privileges, all of + * the normal checks are bypassed. + */ + if (cr == NULL || PRIV_POLICY_ONLY(cr, PRIV_ALL, B_FALSE)) { + state->dts_cred.dcr_visible = DTRACE_CRV_ALL; + state->dts_cred.dcr_action = DTRACE_CRA_ALL; + } else { + /* + * Set up the credentials for this instantiation. We take a + * hold on the credential to prevent it from disappearing on + * us; this in turn prevents the zone_t referenced by this + * credential from disappearing. This means that we can + * examine the credential and the zone from probe context. + */ + crhold(cr); + state->dts_cred.dcr_cred = cr; + + /* + * CRA_PROC means "we have *some* privilege for dtrace" and + * unlocks the use of variables like pid, zonename, etc. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE) || + PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { + state->dts_cred.dcr_action |= DTRACE_CRA_PROC; + } + + /* + * dtrace_user allows use of syscall and profile providers. + * If the user also has proc_owner and/or proc_zone, we + * extend the scope to include additional visibility and + * destructive power. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_USER, B_FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) { + state->dts_cred.dcr_visible |= + DTRACE_CRV_ALLPROC; + + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + } + + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) { + state->dts_cred.dcr_visible |= + DTRACE_CRV_ALLZONE; + + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + } + + /* + * If we have all privs in whatever zone this is, + * we can do destructive things to processes which + * have altered credentials. + */ + if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), + cr->cr_zone->zone_privset)) { + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; + } + } + + /* + * Holding the dtrace_kernel privilege also implies that + * the user has the dtrace_user privilege from a visibility + * perspective. But without further privileges, some + * destructive actions are not available. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_KERNEL, B_FALSE)) { + /* + * Make all probes in all zones visible. However, + * this doesn't mean that all actions become available + * to all zones. + */ + state->dts_cred.dcr_visible |= DTRACE_CRV_KERNEL | + DTRACE_CRV_ALLPROC | DTRACE_CRV_ALLZONE; + + state->dts_cred.dcr_action |= DTRACE_CRA_KERNEL | + DTRACE_CRA_PROC; + /* + * Holding proc_owner means that destructive actions + * for *this* zone are allowed. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + + /* + * Holding proc_zone means that destructive actions + * for this user/group ID in all zones is allowed. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + + /* + * If we have all privs in whatever zone this is, + * we can do destructive things to processes which + * have altered credentials. + */ + if (priv_isequalset(priv_getset(cr, PRIV_EFFECTIVE), + cr->cr_zone->zone_privset)) { + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG; + } + } + + /* + * Holding the dtrace_proc privilege gives control over fasttrap + * and pid providers. We need to grant wider destructive + * privileges in the event that the user has proc_owner and/or + * proc_zone. + */ + if (PRIV_POLICY_ONLY(cr, PRIV_DTRACE_PROC, B_FALSE)) { + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_OWNER, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER; + + if (PRIV_POLICY_ONLY(cr, PRIV_PROC_ZONE, B_FALSE)) + state->dts_cred.dcr_action |= + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE; + } + } + + return (state); +} + +static int +dtrace_state_buffer(dtrace_state_t *state, dtrace_buffer_t *buf, int which) +{ + dtrace_optval_t *opt = state->dts_options, size; + processorid_t cpu VBDTUNASS((processorid_t)DTRACE_CPUALL); + int flags = 0, rval; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + ASSERT(which < DTRACEOPT_MAX); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_INACTIVE || + (state == dtrace_anon.dta_state && + state->dts_activity == DTRACE_ACTIVITY_ACTIVE)); + + if (opt[which] == DTRACEOPT_UNSET || opt[which] == 0) + return (0); + + if (opt[DTRACEOPT_CPU] != DTRACEOPT_UNSET) + cpu = opt[DTRACEOPT_CPU]; + + if (which == DTRACEOPT_SPECSIZE) + flags |= DTRACEBUF_NOSWITCH; + + if (which == DTRACEOPT_BUFSIZE) { + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_RING) + flags |= DTRACEBUF_RING; + + if (opt[DTRACEOPT_BUFPOLICY] == DTRACEOPT_BUFPOLICY_FILL) + flags |= DTRACEBUF_FILL; + + if (state != dtrace_anon.dta_state || + state->dts_activity != DTRACE_ACTIVITY_ACTIVE) + flags |= DTRACEBUF_INACTIVE; + } + + for (size = opt[which]; size >= VBDTCAST(dtrace_optval_t)sizeof (uint64_t); size >>= 1) { + /* + * The size must be 8-byte aligned. If the size is not 8-byte + * aligned, drop it down by the difference. + */ + if (size & (sizeof (uint64_t) - 1)) + size -= size & (sizeof (uint64_t) - 1); + + if (size < state->dts_reserve) { + /* + * Buffers always must be large enough to accommodate + * their prereserved space. We return E2BIG instead + * of ENOMEM in this case to allow for user-level + * software to differentiate the cases. + */ + return (E2BIG); + } + + rval = dtrace_buffer_alloc(buf, size, flags, cpu); + + if (rval != ENOMEM) { + opt[which] = size; + return (rval); + } + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + return (rval); + } + + return (ENOMEM); +} + +static int +dtrace_state_buffers(dtrace_state_t *state) +{ + dtrace_speculation_t *spec = state->dts_speculations; + int rval, i; + + if ((rval = dtrace_state_buffer(state, state->dts_buffer, + DTRACEOPT_BUFSIZE)) != 0) + return (rval); + + if ((rval = dtrace_state_buffer(state, state->dts_aggbuffer, + DTRACEOPT_AGGSIZE)) != 0) + return (rval); + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((rval = dtrace_state_buffer(state, + spec[i].dtsp_buffer, DTRACEOPT_SPECSIZE)) != 0) + return (rval); + } + + return (0); +} + +static void +dtrace_state_prereserve(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_probe_t *probe; + + state->dts_reserve = 0; + + if (state->dts_options[DTRACEOPT_BUFPOLICY] != DTRACEOPT_BUFPOLICY_FILL) + return; + + /* + * If our buffer policy is a "fill" buffer policy, we need to set the + * prereserved space to be the space required by the END probes. + */ + probe = dtrace_probes[dtrace_probeid_end - 1]; + ASSERT(probe != NULL); + + for (ecb = probe->dtpr_ecb; ecb != NULL; ecb = ecb->dte_next) { + if (ecb->dte_state != state) + continue; + + state->dts_reserve += VBDTCAST(uint32_t)ecb->dte_needed + ecb->dte_alignment; + } +} + +static int +dtrace_state_go(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_optval_t *opt = state->dts_options, sz, nspec; + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; +#ifndef VBOX + cyc_handler_t hdlr; + cyc_time_t when; +#endif + int rval = 0, i, bufsize = NCPU * sizeof (dtrace_buffer_t); + dtrace_icookie_t cookie; + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + rval = EBUSY; + goto out; + } + + /* + * Before we can perform any checks, we must prime all of the + * retained enablings that correspond to this state. + */ + dtrace_enabling_prime(state); + + if (state->dts_destructive && !state->dts_cred.dcr_destructive) { + rval = EACCES; + goto out; + } + + dtrace_state_prereserve(state); + + /* + * Now we want to do is try to allocate our speculations. + * We do not automatically resize the number of speculations; if + * this fails, we will fail the operation. + */ + nspec = opt[DTRACEOPT_NSPEC]; + ASSERT(nspec != DTRACEOPT_UNSET); + + if (nspec > INT_MAX) { + rval = ENOMEM; + goto out; + } + + spec = kmem_zalloc(nspec * sizeof (dtrace_speculation_t), KM_NOSLEEP); + + if (spec == NULL) { + rval = ENOMEM; + goto out; + } + + state->dts_speculations = spec; + state->dts_nspeculations = (int)nspec; + + for (i = 0; i < nspec; i++) { + if ((buf = kmem_zalloc(bufsize, KM_NOSLEEP)) == NULL) { + rval = ENOMEM; + goto err; + } + + spec[i].dtsp_buffer = buf; + } + + if (opt[DTRACEOPT_GRABANON] != DTRACEOPT_UNSET) { + if (dtrace_anon.dta_state == NULL) { + rval = ENOENT; + goto out; + } + + if (state->dts_necbs != 0) { + rval = EALREADY; + goto out; + } + + state->dts_anon = dtrace_anon_grab(); + ASSERT(state->dts_anon != NULL); + state = state->dts_anon; + + /* + * We want "grabanon" to be set in the grabbed state, so we'll + * copy that option value from the grabbing state into the + * grabbed state. + */ + state->dts_options[DTRACEOPT_GRABANON] = + opt[DTRACEOPT_GRABANON]; + + *cpu = dtrace_anon.dta_beganon; + + /* + * If the anonymous state is active (as it almost certainly + * is if the anonymous enabling ultimately matched anything), + * we don't allow any further option processing -- but we + * don't return failure. + */ + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + goto out; + } + + if (opt[DTRACEOPT_AGGSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_AGGSIZE] != 0) { + if (state->dts_aggregations == NULL) { + /* + * We're not going to create an aggregation buffer + * because we don't have any ECBs that contain + * aggregations -- set this option to 0. + */ + opt[DTRACEOPT_AGGSIZE] = 0; + } else { + /* + * If we have an aggregation buffer, we must also have + * a buffer to use as scratch. + */ + if (opt[DTRACEOPT_BUFSIZE] == DTRACEOPT_UNSET || + opt[DTRACEOPT_BUFSIZE] < VBDTCAST(dtrace_optval_t)state->dts_needed) { + opt[DTRACEOPT_BUFSIZE] = state->dts_needed; + } + } + } + + if (opt[DTRACEOPT_SPECSIZE] != DTRACEOPT_UNSET && + opt[DTRACEOPT_SPECSIZE] != 0) { + if (!state->dts_speculates) { + /* + * We're not going to create speculation buffers + * because we don't have any ECBs that actually + * speculate -- set the speculation size to 0. + */ + opt[DTRACEOPT_SPECSIZE] = 0; + } + } + + /* + * The bare minimum size for any buffer that we're actually going to + * do anything to is sizeof (uint64_t). + */ + sz = sizeof (uint64_t); + + if ((state->dts_needed != 0 && opt[DTRACEOPT_BUFSIZE] < sz) || + (state->dts_speculates && opt[DTRACEOPT_SPECSIZE] < sz) || + (state->dts_aggregations != NULL && opt[DTRACEOPT_AGGSIZE] < sz)) { + /* + * A buffer size has been explicitly set to 0 (or to a size + * that will be adjusted to 0) and we need the space -- we + * need to return failure. We return ENOSPC to differentiate + * it from failing to allocate a buffer due to failure to meet + * the reserve (for which we return E2BIG). + */ + rval = ENOSPC; + goto out; + } + + if ((rval = dtrace_state_buffers(state)) != 0) + goto err; + + if ((sz = opt[DTRACEOPT_DYNVARSIZE]) == DTRACEOPT_UNSET) + sz = dtrace_dstate_defsize; + + do { + rval = dtrace_dstate_init(&state->dts_vstate.dtvs_dynvars, sz); + + if (rval == 0) + break; + + if (opt[DTRACEOPT_BUFRESIZE] == DTRACEOPT_BUFRESIZE_MANUAL) + goto err; + } while (sz >>= 1); + + opt[DTRACEOPT_DYNVARSIZE] = sz; + + if (rval != 0) + goto err; + + if (opt[DTRACEOPT_STATUSRATE] > dtrace_statusrate_max) + opt[DTRACEOPT_STATUSRATE] = dtrace_statusrate_max; + + if (opt[DTRACEOPT_CLEANRATE] == 0) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + + if (opt[DTRACEOPT_CLEANRATE] < dtrace_cleanrate_min) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_min; + + if (opt[DTRACEOPT_CLEANRATE] > dtrace_cleanrate_max) + opt[DTRACEOPT_CLEANRATE] = dtrace_cleanrate_max; + +#ifndef VBOX + hdlr.cyh_func = (cyc_func_t)dtrace_state_clean; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = 0; + when.cyt_interval = opt[DTRACEOPT_CLEANRATE]; + + state->dts_cleaner = cyclic_add(&hdlr, &when); + + hdlr.cyh_func = (cyc_func_t)dtrace_state_deadman; + hdlr.cyh_arg = state; + hdlr.cyh_level = CY_LOW_LEVEL; + + when.cyt_when = 0; + when.cyt_interval = dtrace_deadman_interval; + + state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); + state->dts_deadman = cyclic_add(&hdlr, &when); +#else /* VBOX */ + + rval = RTTimerCreateEx(&state->dts_cleaner, opt[DTRACEOPT_CLEANRATE], + RTTIMER_FLAGS_CPU_ANY, dtrace_state_clean_timer, state); + if (RT_FAILURE(rval)) { + rval = RTErrConvertToErrno(rval); + goto err; + } + + state->dts_alive = state->dts_laststatus = dtrace_gethrtime(); + rval = RTTimerCreateEx(&state->dts_deadman, dtrace_deadman_interval, + RTTIMER_FLAGS_CPU_ANY, dtrace_state_deadman_timer, state); + if (RT_FAILURE(rval)) { + RTTimerDestroy(state->dts_cleaner); + state->dts_cleaner = CYCLIC_NONE; + state->dts_deadman = CYCLIC_NONE; + rval = RTErrConvertToErrno(rval); + goto err; + } + + rval = RTTimerStart(state->dts_cleaner, 0); + if (RT_SUCCESS(rval)) + rval = RTTimerStart(state->dts_deadman, 0); + if (RT_FAILURE(rval)) { + rval = RTErrConvertToErrno(rval); + goto err; + } +#endif /* VBOX */ + + state->dts_activity = DTRACE_ACTIVITY_WARMUP; + + /* + * Now it's time to actually fire the BEGIN probe. We need to disable + * interrupts here both to record the CPU on which we fired the BEGIN + * probe (the data from this CPU will be processed first at user + * level) and to manually activate the buffer for this CPU. + */ + cookie = dtrace_interrupt_disable(); + *cpu = VBDT_GET_CPUID(); + ASSERT(state->dts_buffer[*cpu].dtb_flags & DTRACEBUF_INACTIVE); + state->dts_buffer[*cpu].dtb_flags &= ~DTRACEBUF_INACTIVE; + + dtrace_probe(dtrace_probeid_begin, + (uint64_t)(uintptr_t)state, 0, 0, 0, 0); + dtrace_interrupt_enable(cookie); + /* + * We may have had an exit action from a BEGIN probe; only change our + * state to ACTIVE if we're still in WARMUP. + */ + ASSERT(state->dts_activity == DTRACE_ACTIVITY_WARMUP || + state->dts_activity == DTRACE_ACTIVITY_DRAINING); + + if (state->dts_activity == DTRACE_ACTIVITY_WARMUP) + state->dts_activity = DTRACE_ACTIVITY_ACTIVE; + + /* + * Regardless of whether or not now we're in ACTIVE or DRAINING, we + * want each CPU to transition its principal buffer out of the + * INACTIVE state. Doing this assures that no CPU will suddenly begin + * processing an ECB halfway down a probe's ECB chain; all CPUs will + * atomically transition from processing none of a state's ECBs to + * processing all of them. + */ +#ifndef VBOX + dtrace_xcall(DTRACE_CPUALL, + (dtrace_xcall_t)dtrace_buffer_activate, state); +#else + RTMpOnAll(dtrace_buffer_activate_wrapper, state, NULL); +#endif + goto out; + +err: + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + if ((nspec = state->dts_nspeculations) == 0) { + ASSERT(state->dts_speculations == NULL); + goto out; + } + + spec = state->dts_speculations; + ASSERT(spec != NULL); + + for (i = 0; i < state->dts_nspeculations; i++) { + if ((buf = spec[i].dtsp_buffer) == NULL) + break; + + dtrace_buffer_free(buf); + kmem_free(buf, bufsize); + } + + kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); + state->dts_nspeculations = 0; + state->dts_speculations = NULL; + +out: + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + + return (rval); +} + +static int +dtrace_state_stop(dtrace_state_t *state, processorid_t *cpu) +{ + dtrace_icookie_t cookie; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE && + state->dts_activity != DTRACE_ACTIVITY_DRAINING) + return (EINVAL); + + /* + * We'll set the activity to DTRACE_ACTIVITY_DRAINING, and issue a sync + * to be sure that every CPU has seen it. See below for the details + * on why this is done. + */ + state->dts_activity = DTRACE_ACTIVITY_DRAINING; + dtrace_sync(); + + /* + * By this point, it is impossible for any CPU to be still processing + * with DTRACE_ACTIVITY_ACTIVE. We can thus set our activity to + * DTRACE_ACTIVITY_COOLDOWN and know that we're not racing with any + * other CPU in dtrace_buffer_reserve(). This allows dtrace_probe() + * and callees to know that the activity is DTRACE_ACTIVITY_COOLDOWN + * iff we're in the END probe. + */ + state->dts_activity = DTRACE_ACTIVITY_COOLDOWN; + dtrace_sync(); + ASSERT(state->dts_activity == DTRACE_ACTIVITY_COOLDOWN); + + /* + * Finally, we can release the reserve and call the END probe. We + * disable interrupts across calling the END probe to allow us to + * return the CPU on which we actually called the END probe. This + * allows user-land to be sure that this CPU's principal buffer is + * processed last. + */ + state->dts_reserve = 0; + + cookie = dtrace_interrupt_disable(); + *cpu = VBDT_GET_CPUID(); + dtrace_probe(dtrace_probeid_end, + (uint64_t)(uintptr_t)state, 0, 0, 0, 0); + dtrace_interrupt_enable(cookie); + + state->dts_activity = DTRACE_ACTIVITY_STOPPED; + dtrace_sync(); + + return (0); +} + +static int +dtrace_state_option(dtrace_state_t *state, dtrace_optid_t option, + dtrace_optval_t val) +{ + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) + return (EBUSY); + + if (option >= DTRACEOPT_MAX) + return (EINVAL); + + if (option != DTRACEOPT_CPU && val < 0) + return (EINVAL); + + switch (option) { + case DTRACEOPT_DESTRUCTIVE: + if (dtrace_destructive_disallow) + return (EACCES); + + state->dts_cred.dcr_destructive = 1; + break; + + case DTRACEOPT_BUFSIZE: + case DTRACEOPT_DYNVARSIZE: + case DTRACEOPT_AGGSIZE: + case DTRACEOPT_SPECSIZE: + case DTRACEOPT_STRSIZE: + if (val < 0) + return (EINVAL); + + if (val >= LONG_MAX) { + /* + * If this is an otherwise negative value, set it to + * the highest multiple of 128m less than LONG_MAX. + * Technically, we're adjusting the size without + * regard to the buffer resizing policy, but in fact, + * this has no effect -- if we set the buffer size to + * ~LONG_MAX and the buffer policy is ultimately set to + * be "manual", the buffer allocation is guaranteed to + * fail, if only because the allocation requires two + * buffers. (We set the the size to the highest + * multiple of 128m because it ensures that the size + * will remain a multiple of a megabyte when + * repeatedly halved -- all the way down to 15m.) + */ + val = LONG_MAX - (1 << 27) + 1; + } + } + + state->dts_options[option] = val; + + return (0); +} + +static void +dtrace_state_destroy(dtrace_state_t *state) +{ + dtrace_ecb_t *ecb; + dtrace_vstate_t *vstate = &state->dts_vstate; +#ifndef VBOX + minor_t minor = getminor(state->dts_dev); +#endif + int i, bufsize = NCPU * sizeof (dtrace_buffer_t); + dtrace_speculation_t *spec = state->dts_speculations; + int nspec = state->dts_nspeculations; + uint32_t match; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * First, retract any retained enablings for this state. + */ + dtrace_enabling_retract(state); + ASSERT(state->dts_nretained == 0); + + if (state->dts_activity == DTRACE_ACTIVITY_ACTIVE || + state->dts_activity == DTRACE_ACTIVITY_DRAINING) { + /* + * We have managed to come into dtrace_state_destroy() on a + * hot enabling -- almost certainly because of a disorderly + * shutdown of a consumer. (That is, a consumer that is + * exiting without having called dtrace_stop().) In this case, + * we're going to set our activity to be KILLED, and then + * issue a sync to be sure that everyone is out of probe + * context before we start blowing away ECBs. + */ + state->dts_activity = DTRACE_ACTIVITY_KILLED; + dtrace_sync(); + } + + /* + * Release the credential hold we took in dtrace_state_create(). + */ + if (state->dts_cred.dcr_cred != NULL) + crfree(state->dts_cred.dcr_cred); + + /* + * Now we can safely disable and destroy any enabled probes. Because + * any DTRACE_PRIV_KERNEL probes may actually be slowing our progress + * (especially if they're all enabled), we take two passes through the + * ECBs: in the first, we disable just DTRACE_PRIV_KERNEL probes, and + * in the second we disable whatever is left over. + */ + for (match = DTRACE_PRIV_KERNEL; ; match = 0) { + for (i = 0; i < state->dts_necbs; i++) { + if ((ecb = state->dts_ecbs[i]) == NULL) + continue; + + if (match && ecb->dte_probe != NULL) { + dtrace_probe_t *probe = ecb->dte_probe; + dtrace_provider_t *prov = probe->dtpr_provider; + + if (!(prov->dtpv_priv.dtpp_flags & match)) + continue; + } + + dtrace_ecb_disable(ecb); + dtrace_ecb_destroy(ecb); + } + + if (!match) + break; + } + + /* + * Before we free the buffers, perform one more sync to assure that + * every CPU is out of probe context. + */ + dtrace_sync(); + + dtrace_buffer_free(state->dts_buffer); + dtrace_buffer_free(state->dts_aggbuffer); + + for (i = 0; i < nspec; i++) + dtrace_buffer_free(spec[i].dtsp_buffer); + + if (state->dts_cleaner != CYCLIC_NONE) + cyclic_remove(state->dts_cleaner); + + if (state->dts_deadman != CYCLIC_NONE) + cyclic_remove(state->dts_deadman); + + dtrace_dstate_fini(&vstate->dtvs_dynvars); + dtrace_vstate_fini(vstate); + kmem_free(state->dts_ecbs, state->dts_necbs * sizeof (dtrace_ecb_t *)); + + if (state->dts_aggregations != NULL) { +#ifdef DEBUG + for (i = 0; i < state->dts_naggregations; i++) + ASSERT(state->dts_aggregations[i] == NULL); +#endif + ASSERT(state->dts_naggregations > 0); + kmem_free(state->dts_aggregations, + state->dts_naggregations * sizeof (dtrace_aggregation_t *)); + } + + kmem_free(state->dts_buffer, bufsize); + kmem_free(state->dts_aggbuffer, bufsize); + + for (i = 0; i < nspec; i++) + kmem_free(spec[i].dtsp_buffer, bufsize); + + kmem_free(spec, nspec * sizeof (dtrace_speculation_t)); + + dtrace_format_destroy(state); + + vmem_destroy(state->dts_aggid_arena); +#ifndef VBOX + ddi_soft_state_free(dtrace_softstate, minor); + vmem_free(dtrace_minor, (void *)(uintptr_t)minor, 1); +#else + kmem_free(state, sizeof (*state)); +#endif +} + +/* + * DTrace Anonymous Enabling Functions + */ +static dtrace_state_t * +dtrace_anon_grab(void) +{ + dtrace_state_t *state; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if ((state = dtrace_anon.dta_state) == NULL) { + ASSERT(dtrace_anon.dta_enabling == NULL); + return (NULL); + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + ASSERT(dtrace_retained != NULL); + + dtrace_enabling_destroy(dtrace_anon.dta_enabling); + dtrace_anon.dta_enabling = NULL; + dtrace_anon.dta_state = NULL; + + return (state); +} + +#ifndef VBOX +static void +dtrace_anon_property(void) +{ + int i, rv; + dtrace_state_t *state; + dof_hdr_t *dof; + char c[32]; /* enough for "dof-data-" + digits */ + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(MUTEX_HELD(&cpu_lock)); + + for (i = 0; ; i++) { + (void) snprintf(c, sizeof (c), "dof-data-%d", i); + + dtrace_err_verbose = 1; + + if ((dof = dtrace_dof_property(c)) == NULL) { + dtrace_err_verbose = 0; + break; + } + +#ifndef VBOX + /* + * We want to create anonymous state, so we need to transition + * the kernel debugger to indicate that DTrace is active. If + * this fails (e.g. because the debugger has modified text in + * some way), we won't continue with the processing. + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + cmn_err(CE_NOTE, "kernel debugger active; anonymous " + "enabling ignored."); + dtrace_dof_destroy(dof); + break; + } +#endif + + /* + * If we haven't allocated an anonymous state, we'll do so now. + */ + if ((state = dtrace_anon.dta_state) == NULL) { + state = dtrace_state_create(NULL, NULL); + dtrace_anon.dta_state = state; + + if (state == NULL) { + /* + * This basically shouldn't happen: the only + * failure mode from dtrace_state_create() is a + * failure of ddi_soft_state_zalloc() that + * itself should never happen. Still, the + * interface allows for a failure mode, and + * we want to fail as gracefully as possible: + * we'll emit an error message and cease + * processing anonymous state in this case. + */ + cmn_err(CE_WARN, "failed to create " + "anonymous state"); + dtrace_dof_destroy(dof); + break; + } + } + + rv = dtrace_dof_slurp(dof, &state->dts_vstate, CRED(), + &dtrace_anon.dta_enabling, 0, B_TRUE); + + if (rv == 0) + rv = dtrace_dof_options(dof, state); + + dtrace_err_verbose = 0; + dtrace_dof_destroy(dof); + + if (rv != 0) { + /* + * This is malformed DOF; chuck any anonymous state + * that we created. + */ + ASSERT(dtrace_anon.dta_enabling == NULL); + dtrace_state_destroy(state); + dtrace_anon.dta_state = NULL; + break; + } + + ASSERT(dtrace_anon.dta_enabling != NULL); + } + + if (dtrace_anon.dta_enabling != NULL) { + int rval; + + /* + * dtrace_enabling_retain() can only fail because we are + * trying to retain more enablings than are allowed -- but + * we only have one anonymous enabling, and we are guaranteed + * to be allowed at least one retained enabling; we assert + * that dtrace_enabling_retain() returns success. + */ + rval = dtrace_enabling_retain(dtrace_anon.dta_enabling); + ASSERT(rval == 0); + + dtrace_enabling_dump(dtrace_anon.dta_enabling); + } +} +#endif /* !VBOX */ + +/* + * DTrace Helper Functions + */ +#ifndef VBOX /* No helper stuff */ +static void +dtrace_helper_trace(dtrace_helper_action_t *helper, + dtrace_mstate_t *mstate, dtrace_vstate_t *vstate, int where) +{ + uint32_t size, next, nnext, i; + dtrace_helptrace_t *ent; + uint16_t flags = cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + + if (!dtrace_helptrace_enabled) + return; + + ASSERT(vstate->dtvs_nlocals <= VBDTCAST(int32_t)dtrace_helptrace_nlocals); + + /* + * What would a tracing framework be without its own tracing + * framework? (Well, a hell of a lot simpler, for starters...) + */ + size = sizeof (dtrace_helptrace_t) + dtrace_helptrace_nlocals * + sizeof (uint64_t) - sizeof (uint64_t); + + /* + * Iterate until we can allocate a slot in the trace buffer. + */ + do { + next = dtrace_helptrace_next; + + if (next + size < VBDTCAST(unsigned)dtrace_helptrace_bufsize) { + nnext = next + size; + } else { + nnext = size; + } + } while (dtrace_cas32(&dtrace_helptrace_next, next, nnext) != next); + + /* + * We have our slot; fill it in. + */ + if (nnext == size) + next = 0; + + ent = (dtrace_helptrace_t *)&dtrace_helptrace_buffer[next]; + ent->dtht_helper = helper; + ent->dtht_where = where; + ent->dtht_nlocals = vstate->dtvs_nlocals; + + ent->dtht_fltoffs = (mstate->dtms_present & DTRACE_MSTATE_FLTOFFS) ? + mstate->dtms_fltoffs : -1; + ent->dtht_fault = DTRACE_FLAGS2FLT(flags); + ent->dtht_illval = cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_illval; + + for (i = 0; VBDTCAST(int32_t)i < vstate->dtvs_nlocals; i++) { + dtrace_statvar_t *svar; + + if ((svar = vstate->dtvs_locals[i]) == NULL) + continue; + + ASSERT(svar->dtsv_size >= NCPU * sizeof (uint64_t)); + ent->dtht_locals[i] = + ((uint64_t *)(uintptr_t)svar->dtsv_data)[VBDT_GET_CPUID()]; + } +} + +static uint64_t +dtrace_helper(int which, dtrace_mstate_t *mstate, + dtrace_state_t *state, uint64_t arg0, uint64_t arg1) +{ + VBDTTYPE(uint16_t volatile *, uint16_t *)flags = &cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags; + uint64_t sarg0 = mstate->dtms_arg[0]; + uint64_t sarg1 = mstate->dtms_arg[1]; + uint64_t rval VBDTUNASS(666); + dtrace_helpers_t *helpers = curproc->p_dtrace_helpers; + dtrace_helper_action_t *helper; + dtrace_vstate_t *vstate; + dtrace_difo_t *pred; + int i, trace = dtrace_helptrace_enabled; + + ASSERT(which >= 0 && which < DTRACE_NHELPER_ACTIONS); + + if (helpers == NULL) + return (0); + + if ((helper = helpers->dthps_actions[which]) == NULL) + return (0); + + vstate = &helpers->dthps_vstate; + mstate->dtms_arg[0] = arg0; + mstate->dtms_arg[1] = arg1; + + /* + * Now iterate over each helper. If its predicate evaluates to 'true', + * we'll call the corresponding actions. Note that the below calls + * to dtrace_dif_emulate() may set faults in machine state. This is + * okay: our caller (the outer dtrace_dif_emulate()) will simply plow + * the stored DIF offset with its own (which is the desired behavior). + * Also, note the calls to dtrace_dif_emulate() may allocate scratch + * from machine state; this is okay, too. + */ + for (; helper != NULL; helper = helper->dtha_next) { + if ((pred = helper->dtha_predicate) != NULL) { + if (trace) + dtrace_helper_trace(helper, mstate, vstate, 0); + + if (!dtrace_dif_emulate(pred, mstate, vstate, state)) + goto next; + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + + for (i = 0; i < helper->dtha_nactions; i++) { + if (trace) + dtrace_helper_trace(helper, + mstate, vstate, i + 1); + + rval = dtrace_dif_emulate(helper->dtha_actions[i], + mstate, vstate, state); + + if (*flags & CPU_DTRACE_FAULT) + goto err; + } + +next: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_NEXT); + } + + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_DONE); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return (rval); + +err: + if (trace) + dtrace_helper_trace(helper, mstate, vstate, + DTRACE_HELPTRACE_ERR); + + /* + * Restore the arg0 that we saved upon entry. + */ + mstate->dtms_arg[0] = sarg0; + mstate->dtms_arg[1] = sarg1; + + return (NULL); +} + +static void +dtrace_helper_action_destroy(dtrace_helper_action_t *helper, + dtrace_vstate_t *vstate) +{ + int i; + + if (helper->dtha_predicate != NULL) + dtrace_difo_release(helper->dtha_predicate, vstate); + + for (i = 0; i < helper->dtha_nactions; i++) { + ASSERT(helper->dtha_actions[i] != NULL); + dtrace_difo_release(helper->dtha_actions[i], vstate); + } + + kmem_free(helper->dtha_actions, + helper->dtha_nactions * sizeof (dtrace_difo_t *)); + kmem_free(helper, sizeof (dtrace_helper_action_t)); +} + +static int +dtrace_helper_destroygen(int gen) +{ + proc_t *p = curproc; + dtrace_helpers_t *help = p->p_dtrace_helpers; + dtrace_vstate_t *vstate; + VBDTTYPE(uint_t,int) i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if (help == NULL || gen > help->dthps_generation) + return (EINVAL); + + vstate = &help->dthps_vstate; + + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + dtrace_helper_action_t *last = NULL, *h, *next; + + for (h = help->dthps_actions[i]; h != NULL; h = next) { + next = h->dtha_next; + + if (h->dtha_generation == gen) { + if (last != NULL) { + last->dtha_next = next; + } else { + help->dthps_actions[i] = next; + } + + dtrace_helper_action_destroy(h, vstate); + } else { + last = h; + } + } + } + + /* + * Interate until we've cleared out all helper providers with the + * given generation number. + */ + for (;;) { + dtrace_helper_provider_t *prov VBDTGCC(NULL); + + /* + * Look for a helper provider with the right generation. We + * have to start back at the beginning of the list each time + * because we drop dtrace_lock. It's unlikely that we'll make + * more than two passes. + */ + for (i = 0; i < help->dthps_nprovs; i++) { + prov = help->dthps_provs[i]; + + if (prov->dthp_generation == gen) + break; + } + + /* + * If there were no matches, we're done. + */ + if (i == help->dthps_nprovs) + break; + + /* + * Move the last helper provider into this slot. + */ + help->dthps_nprovs--; + help->dthps_provs[i] = help->dthps_provs[help->dthps_nprovs]; + help->dthps_provs[help->dthps_nprovs] = NULL; + + mutex_exit(&dtrace_lock); + + /* + * If we have a meta provider, remove this helper provider. + */ + mutex_enter(&dtrace_meta_lock); + if (dtrace_meta_pid != NULL) { + ASSERT(dtrace_deferred_pid == NULL); + dtrace_helper_provider_remove(&prov->dthp_prov, + p->p_pid); + } + mutex_exit(&dtrace_meta_lock); + + dtrace_helper_provider_destroy(prov); + + mutex_enter(&dtrace_lock); + } + + return (0); +} + +static int +dtrace_helper_validate(dtrace_helper_action_t *helper) +{ + int err = 0, i; + dtrace_difo_t *dp; + + if ((dp = helper->dtha_predicate) != NULL) + err += dtrace_difo_validate_helper(dp); + + for (i = 0; i < helper->dtha_nactions; i++) + err += dtrace_difo_validate_helper(helper->dtha_actions[i]); + + return (err == 0); +} + +static int +dtrace_helper_action_add(int which, dtrace_ecbdesc_t *ep) +{ + dtrace_helpers_t *help; + dtrace_helper_action_t *helper, *last; + dtrace_actdesc_t *act; + dtrace_vstate_t *vstate; + dtrace_predicate_t *pred; + int count = 0, nactions = 0, i; + + if (which < 0 || which >= DTRACE_NHELPER_ACTIONS) + return (EINVAL); + + help = curproc->p_dtrace_helpers; + last = help->dthps_actions[which]; + vstate = &help->dthps_vstate; + + for (count = 0; last != NULL; last = last->dtha_next) { + count++; + if (last->dtha_next == NULL) + break; + } + + /* + * If we already have dtrace_helper_actions_max helper actions for this + * helper action type, we'll refuse to add a new one. + */ + if (count >= dtrace_helper_actions_max) + return (ENOSPC); + + helper = kmem_zalloc(sizeof (dtrace_helper_action_t), KM_SLEEP); + helper->dtha_generation = help->dthps_generation; + + if ((pred = ep->dted_pred.dtpdd_predicate) != NULL) { + ASSERT(pred->dtp_difo != NULL); + dtrace_difo_hold(pred->dtp_difo); + helper->dtha_predicate = pred->dtp_difo; + } + + for (act = ep->dted_action; act != NULL; act = act->dtad_next) { + if (act->dtad_kind != DTRACEACT_DIFEXPR) + goto err; + + if (act->dtad_difo == NULL) + goto err; + + nactions++; + } + + helper->dtha_actions = kmem_zalloc(sizeof (dtrace_difo_t *) * + (helper->dtha_nactions = nactions), KM_SLEEP); + + for (act = ep->dted_action, i = 0; act != NULL; act = act->dtad_next) { + dtrace_difo_hold(act->dtad_difo); + helper->dtha_actions[i++] = act->dtad_difo; + } + + if (!dtrace_helper_validate(helper)) + goto err; + + if (last == NULL) { + help->dthps_actions[which] = helper; + } else { + last->dtha_next = helper; + } + + if (vstate->dtvs_nlocals > VBDTCAST(int32_t)dtrace_helptrace_nlocals) { + dtrace_helptrace_nlocals = vstate->dtvs_nlocals; + dtrace_helptrace_next = 0; + } + + return (0); +err: + dtrace_helper_action_destroy(helper, vstate); + return (EINVAL); +} + +static void +dtrace_helper_provider_register(proc_t *p, dtrace_helpers_t *help, + dof_helper_t *dofhp) +{ + ASSERT(MUTEX_NOT_HELD(&dtrace_lock)); + + mutex_enter(&dtrace_meta_lock); + mutex_enter(&dtrace_lock); + + if (!dtrace_attached() || dtrace_meta_pid == NULL) { + /* + * If the dtrace module is loaded but not attached, or if + * there aren't isn't a meta provider registered to deal with + * these provider descriptions, we need to postpone creating + * the actual providers until later. + */ + + if (help->dthps_next == NULL && help->dthps_prev == NULL && + dtrace_deferred_pid != help) { + help->dthps_deferred = 1; + help->dthps_pid = p->p_pid; + help->dthps_next = dtrace_deferred_pid; + help->dthps_prev = NULL; + if (dtrace_deferred_pid != NULL) + dtrace_deferred_pid->dthps_prev = help; + dtrace_deferred_pid = help; + } + + mutex_exit(&dtrace_lock); + + } else if (dofhp != NULL) { + /* + * If the dtrace module is loaded and we have a particular + * helper provider description, pass that off to the + * meta provider. + */ + + mutex_exit(&dtrace_lock); + + dtrace_helper_provide(dofhp, p->p_pid); + + } else { + /* + * Otherwise, just pass all the helper provider descriptions + * off to the meta provider. + */ + + VBDTTYPE(uint_t,int) i; + mutex_exit(&dtrace_lock); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provide(&help->dthps_provs[i]->dthp_prov, + p->p_pid); + } + } + + mutex_exit(&dtrace_meta_lock); +} + +static int +dtrace_helper_provider_add(dof_helper_t *dofhp, int gen) +{ + dtrace_helpers_t *help; + dtrace_helper_provider_t *hprov, **tmp_provs; + uint_t tmp_maxprovs, i; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + help = curproc->p_dtrace_helpers; + ASSERT(help != NULL); + + /* + * If we already have dtrace_helper_providers_max helper providers, + * we're refuse to add a new one. + */ + if (help->dthps_nprovs >= dtrace_helper_providers_max) + return (ENOSPC); + + /* + * Check to make sure this isn't a duplicate. + */ + for (i = 0; i < help->dthps_nprovs; i++) { + if (dofhp->dofhp_addr == + help->dthps_provs[i]->dthp_prov.dofhp_addr) + return (EALREADY); + } + + hprov = kmem_zalloc(sizeof (dtrace_helper_provider_t), KM_SLEEP); + hprov->dthp_prov = *dofhp; + hprov->dthp_ref = 1; + hprov->dthp_generation = gen; + + /* + * Allocate a bigger table for helper providers if it's already full. + */ + if (help->dthps_maxprovs == help->dthps_nprovs) { + tmp_maxprovs = help->dthps_maxprovs; + tmp_provs = help->dthps_provs; + + if (help->dthps_maxprovs == 0) + help->dthps_maxprovs = 2; + else + help->dthps_maxprovs *= 2; + if (help->dthps_maxprovs > dtrace_helper_providers_max) + help->dthps_maxprovs = dtrace_helper_providers_max; + + ASSERT(tmp_maxprovs < help->dthps_maxprovs); + + help->dthps_provs = kmem_zalloc(help->dthps_maxprovs * + sizeof (dtrace_helper_provider_t *), KM_SLEEP); + + if (tmp_provs != NULL) { + bcopy(tmp_provs, help->dthps_provs, tmp_maxprovs * + sizeof (dtrace_helper_provider_t *)); + kmem_free(tmp_provs, tmp_maxprovs * + sizeof (dtrace_helper_provider_t *)); + } + } + + help->dthps_provs[help->dthps_nprovs] = hprov; + help->dthps_nprovs++; + + return (0); +} + +static void +dtrace_helper_provider_destroy(dtrace_helper_provider_t *hprov) +{ + mutex_enter(&dtrace_lock); + + if (--hprov->dthp_ref == 0) { + dof_hdr_t *dof; + mutex_exit(&dtrace_lock); + dof = (dof_hdr_t *)(uintptr_t)hprov->dthp_prov.dofhp_dof; + dtrace_dof_destroy(dof); + kmem_free(hprov, sizeof (dtrace_helper_provider_t)); + } else { + mutex_exit(&dtrace_lock); + } +} + +static int +dtrace_helper_provider_validate(dof_hdr_t *dof, dof_sec_t *sec) +{ + uintptr_t daddr = (uintptr_t)dof; + dof_sec_t *str_sec, *prb_sec, *arg_sec, *off_sec, *enoff_sec; + dof_provider_t *provider; + dof_probe_t *probe; + uint8_t *arg; + char *strtab, *typestr; + dof_stridx_t typeidx; + size_t typesz; + uint_t nprobes, j, k; + + ASSERT(sec->dofs_type == DOF_SECT_PROVIDER); + + if (sec->dofs_offset & (sizeof (uint_t) - 1)) { + dtrace_dof_error(dof, "misaligned section offset"); + return (-1); + } + + /* + * The section needs to be large enough to contain the DOF provider + * structure appropriate for the given version. + */ + if (sec->dofs_size < + ((dof->dofh_ident[DOF_ID_VERSION] == DOF_VERSION_1) ? + offsetof(dof_provider_t, dofpv_prenoffs) : + sizeof (dof_provider_t))) { + dtrace_dof_error(dof, "provider section too small"); + return (-1); + } + + provider = (dof_provider_t *)(uintptr_t)(daddr + sec->dofs_offset); + str_sec = dtrace_dof_sect(dof, DOF_SECT_STRTAB, provider->dofpv_strtab); + prb_sec = dtrace_dof_sect(dof, DOF_SECT_PROBES, provider->dofpv_probes); + arg_sec = dtrace_dof_sect(dof, DOF_SECT_PRARGS, provider->dofpv_prargs); + off_sec = dtrace_dof_sect(dof, DOF_SECT_PROFFS, provider->dofpv_proffs); + + if (str_sec == NULL || prb_sec == NULL || + arg_sec == NULL || off_sec == NULL) + return (-1); + + enoff_sec = NULL; + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1 && + provider->dofpv_prenoffs != DOF_SECT_NONE && + (enoff_sec = dtrace_dof_sect(dof, DOF_SECT_PRENOFFS, + provider->dofpv_prenoffs)) == NULL) + return (-1); + + strtab = (char *)(uintptr_t)(daddr + str_sec->dofs_offset); + + if (provider->dofpv_name >= str_sec->dofs_size || + strlen(strtab + provider->dofpv_name) >= DTRACE_PROVNAMELEN) { + dtrace_dof_error(dof, "invalid provider name"); + return (-1); + } + + if (prb_sec->dofs_entsize == 0 || + prb_sec->dofs_entsize > prb_sec->dofs_size) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + if (prb_sec->dofs_entsize & (sizeof (uintptr_t) - 1)) { + dtrace_dof_error(dof, "misaligned entry size"); + return (-1); + } + + if (off_sec->dofs_entsize != sizeof (uint32_t)) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + if (off_sec->dofs_offset & (sizeof (uint32_t) - 1)) { + dtrace_dof_error(dof, "misaligned section offset"); + return (-1); + } + + if (arg_sec->dofs_entsize != sizeof (uint8_t)) { + dtrace_dof_error(dof, "invalid entry size"); + return (-1); + } + + arg = (uint8_t *)(uintptr_t)(daddr + arg_sec->dofs_offset); + + nprobes = prb_sec->dofs_size / prb_sec->dofs_entsize; + + /* + * Take a pass through the probes to check for errors. + */ + for (j = 0; j < nprobes; j++) { + probe = (dof_probe_t *)(uintptr_t)(daddr + + prb_sec->dofs_offset + j * prb_sec->dofs_entsize); + + if (probe->dofpr_func >= str_sec->dofs_size) { + dtrace_dof_error(dof, "invalid function name"); + return (-1); + } + + if (strlen(strtab + probe->dofpr_func) >= DTRACE_FUNCNAMELEN) { + dtrace_dof_error(dof, "function name too long"); + return (-1); + } + + if (probe->dofpr_name >= str_sec->dofs_size || + strlen(strtab + probe->dofpr_name) >= DTRACE_NAMELEN) { + dtrace_dof_error(dof, "invalid probe name"); + return (-1); + } + + /* + * The offset count must not wrap the index, and the offsets + * must also not overflow the section's data. + */ + if (probe->dofpr_offidx + probe->dofpr_noffs < + probe->dofpr_offidx || + (probe->dofpr_offidx + probe->dofpr_noffs) * + off_sec->dofs_entsize > off_sec->dofs_size) { + dtrace_dof_error(dof, "invalid probe offset"); + return (-1); + } + + if (dof->dofh_ident[DOF_ID_VERSION] != DOF_VERSION_1) { + /* + * If there's no is-enabled offset section, make sure + * there aren't any is-enabled offsets. Otherwise + * perform the same checks as for probe offsets + * (immediately above). + */ + if (enoff_sec == NULL) { + if (probe->dofpr_enoffidx != 0 || + probe->dofpr_nenoffs != 0) { + dtrace_dof_error(dof, "is-enabled " + "offsets with null section"); + return (-1); + } + } else if (probe->dofpr_enoffidx + + probe->dofpr_nenoffs < probe->dofpr_enoffidx || + (probe->dofpr_enoffidx + probe->dofpr_nenoffs) * + enoff_sec->dofs_entsize > enoff_sec->dofs_size) { + dtrace_dof_error(dof, "invalid is-enabled " + "offset"); + return (-1); + } + + if (probe->dofpr_noffs + probe->dofpr_nenoffs == 0) { + dtrace_dof_error(dof, "zero probe and " + "is-enabled offsets"); + return (-1); + } + } else if (probe->dofpr_noffs == 0) { + dtrace_dof_error(dof, "zero probe offsets"); + return (-1); + } + + if (probe->dofpr_argidx + probe->dofpr_xargc < + probe->dofpr_argidx || + (probe->dofpr_argidx + probe->dofpr_xargc) * + arg_sec->dofs_entsize > arg_sec->dofs_size) { + dtrace_dof_error(dof, "invalid args"); + return (-1); + } + + typeidx = probe->dofpr_nargv; + typestr = strtab + probe->dofpr_nargv; + for (k = 0; k < probe->dofpr_nargc; k++) { + if (typeidx >= str_sec->dofs_size) { + dtrace_dof_error(dof, "bad " + "native argument type"); + return (-1); + } + + typesz = strlen(typestr) + 1; + if (typesz > DTRACE_ARGTYPELEN) { + dtrace_dof_error(dof, "native " + "argument type too long"); + return (-1); + } + typeidx += VBDTCAST(dof_stridx_t)typesz; + typestr += typesz; + } + + typeidx = probe->dofpr_xargv; + typestr = strtab + probe->dofpr_xargv; + for (k = 0; k < probe->dofpr_xargc; k++) { + if (arg[probe->dofpr_argidx + k] > probe->dofpr_nargc) { + dtrace_dof_error(dof, "bad " + "native argument index"); + return (-1); + } + + if (typeidx >= str_sec->dofs_size) { + dtrace_dof_error(dof, "bad " + "translated argument type"); + return (-1); + } + + typesz = strlen(typestr) + 1; + if (typesz > DTRACE_ARGTYPELEN) { + dtrace_dof_error(dof, "translated argument " + "type too long"); + return (-1); + } + + typeidx += VBDTCAST(dof_stridx_t)typesz; + typestr += typesz; + } + } + + return (0); +} + +static int +dtrace_helper_slurp(dof_hdr_t *dof, dof_helper_t *dhp) +{ + dtrace_helpers_t *help; + dtrace_vstate_t *vstate; + dtrace_enabling_t *enab = NULL; + int i, gen, rv, nhelpers = 0, nprovs = 0, destroy = 1; + uintptr_t daddr = (uintptr_t)dof; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + + if ((help = curproc->p_dtrace_helpers) == NULL) + help = dtrace_helpers_create(curproc); + + vstate = &help->dthps_vstate; + + if ((rv = dtrace_dof_slurp(dof, vstate, NULL, &enab, + dhp != NULL ? dhp->dofhp_addr : 0, B_FALSE)) != 0) { + dtrace_dof_destroy(dof); + return (rv); + } + + /* + * Look for helper providers and validate their descriptions. + */ + if (dhp != NULL) { + for (i = 0; i < VBDTCAST(int)dof->dofh_secnum; i++) { + dof_sec_t *sec = (dof_sec_t *)(uintptr_t)(daddr + + dof->dofh_secoff + i * dof->dofh_secsize); + + if (sec->dofs_type != DOF_SECT_PROVIDER) + continue; + + if (dtrace_helper_provider_validate(dof, sec) != 0) { + dtrace_enabling_destroy(enab); + dtrace_dof_destroy(dof); + return (-1); + } + + nprovs++; + } + } + + /* + * Now we need to walk through the ECB descriptions in the enabling. + */ + for (i = 0; i < enab->dten_ndesc; i++) { + dtrace_ecbdesc_t *ep = enab->dten_desc[i]; + dtrace_probedesc_t *desc = &ep->dted_probe; + + if (strcmp(desc->dtpd_provider, "dtrace") != 0) + continue; + + if (strcmp(desc->dtpd_mod, "helper") != 0) + continue; + + if (strcmp(desc->dtpd_func, "ustack") != 0) + continue; + + if ((rv = dtrace_helper_action_add(DTRACE_HELPER_ACTION_USTACK, + ep)) != 0) { + /* + * Adding this helper action failed -- we are now going + * to rip out the entire generation and return failure. + */ + (void) dtrace_helper_destroygen(help->dthps_generation); + dtrace_enabling_destroy(enab); + dtrace_dof_destroy(dof); + return (-1); + } + + nhelpers++; + } + + if (nhelpers < enab->dten_ndesc) + dtrace_dof_error(dof, "unmatched helpers"); + + gen = help->dthps_generation++; + dtrace_enabling_destroy(enab); + + if (dhp != NULL && nprovs > 0) { + dhp->dofhp_dof = (uint64_t)(uintptr_t)dof; + if (dtrace_helper_provider_add(dhp, gen) == 0) { + mutex_exit(&dtrace_lock); + dtrace_helper_provider_register(curproc, help, dhp); + mutex_enter(&dtrace_lock); + + destroy = 0; + } + } + + if (destroy) + dtrace_dof_destroy(dof); + + return (gen); +} + +static dtrace_helpers_t * +dtrace_helpers_create(proc_t *p) +{ + dtrace_helpers_t *help; + + ASSERT(MUTEX_HELD(&dtrace_lock)); + ASSERT(p->p_dtrace_helpers == NULL); + + help = kmem_zalloc(sizeof (dtrace_helpers_t), KM_SLEEP); + help->dthps_actions = kmem_zalloc(sizeof (dtrace_helper_action_t *) * + DTRACE_NHELPER_ACTIONS, KM_SLEEP); + + p->p_dtrace_helpers = help; + dtrace_helpers++; + + return (help); +} + +static void +dtrace_helpers_destroy(void) +{ + dtrace_helpers_t *help; + dtrace_vstate_t *vstate; + proc_t *p = curproc; + VBDTTYPE(uint_t, int) i; + + mutex_enter(&dtrace_lock); + + ASSERT(p->p_dtrace_helpers != NULL); + ASSERT(dtrace_helpers > 0); + + help = p->p_dtrace_helpers; + vstate = &help->dthps_vstate; + + /* + * We're now going to lose the help from this process. + */ + p->p_dtrace_helpers = NULL; + dtrace_sync(); + + /* + * Destory the helper actions. + */ + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + dtrace_helper_action_t *h, *next; + + for (h = help->dthps_actions[i]; h != NULL; h = next) { + next = h->dtha_next; + dtrace_helper_action_destroy(h, vstate); + h = next; + } + } + + mutex_exit(&dtrace_lock); + + /* + * Destroy the helper providers. + */ + if (help->dthps_maxprovs > 0) { + mutex_enter(&dtrace_meta_lock); + if (dtrace_meta_pid != NULL) { + ASSERT(dtrace_deferred_pid == NULL); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provider_remove( + &help->dthps_provs[i]->dthp_prov, p->p_pid); + } + } else { + mutex_enter(&dtrace_lock); + ASSERT(help->dthps_deferred == 0 || + help->dthps_next != NULL || + help->dthps_prev != NULL || + help == dtrace_deferred_pid); + + /* + * Remove the helper from the deferred list. + */ + if (help->dthps_next != NULL) + help->dthps_next->dthps_prev = help->dthps_prev; + if (help->dthps_prev != NULL) + help->dthps_prev->dthps_next = help->dthps_next; + if (dtrace_deferred_pid == help) { + dtrace_deferred_pid = help->dthps_next; + ASSERT(help->dthps_prev == NULL); + } + + mutex_exit(&dtrace_lock); + } + + mutex_exit(&dtrace_meta_lock); + + for (i = 0; i < help->dthps_nprovs; i++) { + dtrace_helper_provider_destroy(help->dthps_provs[i]); + } + + kmem_free(help->dthps_provs, help->dthps_maxprovs * + sizeof (dtrace_helper_provider_t *)); + } + + mutex_enter(&dtrace_lock); + + dtrace_vstate_fini(&help->dthps_vstate); + kmem_free(help->dthps_actions, + sizeof (dtrace_helper_action_t *) * DTRACE_NHELPER_ACTIONS); + kmem_free(help, sizeof (dtrace_helpers_t)); + + --dtrace_helpers; + mutex_exit(&dtrace_lock); +} + +static void +dtrace_helpers_duplicate(proc_t *from, proc_t *to) +{ + dtrace_helpers_t *help, *newhelp; + dtrace_helper_action_t *helper, *new, *last; + dtrace_difo_t *dp; + dtrace_vstate_t *vstate; + int i, j, sz, hasprovs = 0; + + mutex_enter(&dtrace_lock); + ASSERT(from->p_dtrace_helpers != NULL); + ASSERT(dtrace_helpers > 0); + + help = from->p_dtrace_helpers; + newhelp = dtrace_helpers_create(to); + ASSERT(to->p_dtrace_helpers != NULL); + + newhelp->dthps_generation = help->dthps_generation; + vstate = &newhelp->dthps_vstate; + + /* + * Duplicate the helper actions. + */ + for (i = 0; i < DTRACE_NHELPER_ACTIONS; i++) { + if ((helper = help->dthps_actions[i]) == NULL) + continue; + + for (last = NULL; helper != NULL; helper = helper->dtha_next) { + new = kmem_zalloc(sizeof (dtrace_helper_action_t), + KM_SLEEP); + new->dtha_generation = helper->dtha_generation; + + if ((dp = helper->dtha_predicate) != NULL) { + dp = dtrace_difo_duplicate(dp, vstate); + new->dtha_predicate = dp; + } + + new->dtha_nactions = helper->dtha_nactions; + sz = sizeof (dtrace_difo_t *) * new->dtha_nactions; + new->dtha_actions = kmem_alloc(sz, KM_SLEEP); + + for (j = 0; j < new->dtha_nactions; j++) { + dtrace_difo_t *dp2 = helper->dtha_actions[j]; + + ASSERT(dp2 != NULL); + dp2 = dtrace_difo_duplicate(dp2, vstate); + new->dtha_actions[j] = dp2; + } + + if (last != NULL) { + last->dtha_next = new; + } else { + newhelp->dthps_actions[i] = new; + } + + last = new; + } + } + + /* + * Duplicate the helper providers and register them with the + * DTrace framework. + */ + if (help->dthps_nprovs > 0) { + newhelp->dthps_nprovs = help->dthps_nprovs; + newhelp->dthps_maxprovs = help->dthps_nprovs; + newhelp->dthps_provs = kmem_alloc(newhelp->dthps_nprovs * + sizeof (dtrace_helper_provider_t *), KM_SLEEP); + for (i = 0; i < VBDTCAST(int)newhelp->dthps_nprovs; i++) { + newhelp->dthps_provs[i] = help->dthps_provs[i]; + newhelp->dthps_provs[i]->dthp_ref++; + } + + hasprovs = 1; + } + + mutex_exit(&dtrace_lock); + + if (hasprovs) + dtrace_helper_provider_register(to, newhelp, NULL); +} + +/* + * DTrace Hook Functions + */ +static void +dtrace_module_loaded(struct modctl *ctl) +{ + dtrace_provider_t *prv; + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + + ASSERT(ctl->mod_busy); + + /* + * We're going to call each providers per-module provide operation + * specifying only this module. + */ + for (prv = dtrace_provider; prv != NULL; prv = prv->dtpv_next) + prv->dtpv_pops.dtps_provide_module(prv->dtpv_arg, ctl); + + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + + /* + * If we have any retained enablings, we need to match against them. + * Enabling probes requires that cpu_lock be held, and we cannot hold + * cpu_lock here -- it is legal for cpu_lock to be held when loading a + * module. (In particular, this happens when loading scheduling + * classes.) So if we have any retained enablings, we need to dispatch + * our task queue to do the match for us. + */ + mutex_enter(&dtrace_lock); + + if (dtrace_retained == NULL) { + mutex_exit(&dtrace_lock); + return; + } + + (void) taskq_dispatch(dtrace_taskq, + (task_func_t *)dtrace_enabling_matchall, NULL, TQ_SLEEP); + + mutex_exit(&dtrace_lock); + + /* + * And now, for a little heuristic sleaze: in general, we want to + * match modules as soon as they load. However, we cannot guarantee + * this, because it would lead us to the lock ordering violation + * outlined above. The common case, of course, is that cpu_lock is + * _not_ held -- so we delay here for a clock tick, hoping that that's + * long enough for the task queue to do its work. If it's not, it's + * not a serious problem -- it just means that the module that we + * just loaded may not be immediately instrumentable. + */ + delay(1); +} + +static void +dtrace_module_unloaded(struct modctl *ctl) +{ + dtrace_probe_t template, *probe, *first, *next; + dtrace_provider_t *prov; + + template.dtpr_mod = ctl->mod_modname; + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + + if (dtrace_bymod == NULL) { + /* + * The DTrace module is loaded (obviously) but not attached; + * we don't have any work to do. + */ + mutex_exit(&dtrace_provider_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_lock); + return; + } + + for (probe = first = dtrace_hash_lookup(dtrace_bymod, &template); + probe != NULL; probe = probe->dtpr_nextmod) { + if (probe->dtpr_ecb != NULL) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_lock); + + /* + * This shouldn't _actually_ be possible -- we're + * unloading a module that has an enabled probe in it. + * (It's normally up to the provider to make sure that + * this can't happen.) However, because dtps_enable() + * doesn't have a failure mode, there can be an + * enable/unload race. Upshot: we don't want to + * assert, but we're not going to disable the + * probe, either. + */ + if (dtrace_err_verbose) { + cmn_err(CE_WARN, "unloaded module '%s' had " + "enabled probes", ctl->mod_modname); + } + + return; + } + } + + probe = first; + + for (first = NULL; probe != NULL; probe = next) { + ASSERT(dtrace_probes[probe->dtpr_id - 1] == probe); + + dtrace_probes[probe->dtpr_id - 1] = NULL; + + next = probe->dtpr_nextmod; + dtrace_hash_remove(dtrace_bymod, probe); + dtrace_hash_remove(dtrace_byfunc, probe); + dtrace_hash_remove(dtrace_byname, probe); + + if (first == NULL) { + first = probe; + probe->dtpr_nextmod = NULL; + } else { + probe->dtpr_nextmod = first; + first = probe; + } + } + + /* + * We've removed all of the module's probes from the hash chains and + * from the probe array. Now issue a dtrace_sync() to be sure that + * everyone has cleared out from any probe array processing. + */ + dtrace_sync(); + + for (probe = first; probe != NULL; probe = first) { + first = probe->dtpr_nextmod; + prov = probe->dtpr_provider; + prov->dtpv_pops.dtps_destroy(prov->dtpv_arg, probe->dtpr_id, + probe->dtpr_arg); + kmem_free(probe->dtpr_mod, strlen(probe->dtpr_mod) + 1); + kmem_free(probe->dtpr_func, strlen(probe->dtpr_func) + 1); + kmem_free(probe->dtpr_name, strlen(probe->dtpr_name) + 1); + vmem_free(dtrace_arena, (void *)(uintptr_t)probe->dtpr_id, 1); + kmem_free(probe, sizeof (dtrace_probe_t)); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); +} + +VBDTSTATIC void +dtrace_suspend(void) +{ + dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_suspend)); +} + +VBDTSTATIC void +dtrace_resume(void) +{ + dtrace_probe_foreach(offsetof(dtrace_pops_t, dtps_resume)); +} + +#endif /* !VBOX */ + +#ifdef VBOX +typedef enum { + CPU_INVALID, + CPU_CONFIG, + CPU_UNCONFIG +} cpu_setup_t; +#endif + +#ifndef VBOX + +static int +dtrace_cpu_setup(cpu_setup_t what, processorid_t cpu) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + mutex_enter(&dtrace_lock); + + switch (what) { + case CPU_CONFIG: { + dtrace_state_t *state; + dtrace_optval_t *opt, rs, c; + + /* + * For now, we only allocate a new buffer for anonymous state. + */ + if ((state = dtrace_anon.dta_state) == NULL) + break; + + if (state->dts_activity != DTRACE_ACTIVITY_ACTIVE) + break; + + opt = state->dts_options; + c = opt[DTRACEOPT_CPU]; + + if (c != DTRACE_CPUALL && c != DTRACEOPT_UNSET && c != cpu) + break; + + /* + * Regardless of what the actual policy is, we're going to + * temporarily set our resize policy to be manual. We're + * also going to temporarily set our CPU option to denote + * the newly configured CPU. + */ + rs = opt[DTRACEOPT_BUFRESIZE]; + opt[DTRACEOPT_BUFRESIZE] = DTRACEOPT_BUFRESIZE_MANUAL; + opt[DTRACEOPT_CPU] = (dtrace_optval_t)cpu; + + (void) dtrace_state_buffers(state); + + opt[DTRACEOPT_BUFRESIZE] = rs; + opt[DTRACEOPT_CPU] = c; + + break; + } + + case CPU_UNCONFIG: + /* + * We don't free the buffer in the CPU_UNCONFIG case. (The + * buffer will be freed when the consumer exits.) + */ + break; + + default: + break; + } + + mutex_exit(&dtrace_lock); + return (0); +} + +static void +dtrace_cpu_setup_initial(processorid_t cpu) +{ + (void) dtrace_cpu_setup(CPU_CONFIG, cpu); +} + +#endif /* !VBOX */ + +static void +dtrace_toxrange_add(uintptr_t base, uintptr_t limit) +{ + if (dtrace_toxranges >= dtrace_toxranges_max) { + int osize, nsize; + dtrace_toxrange_t *range; + + osize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); + + if (osize == 0) { + ASSERT(dtrace_toxrange == NULL); + ASSERT(dtrace_toxranges_max == 0); + dtrace_toxranges_max = 1; + } else { + dtrace_toxranges_max <<= 1; + } + + nsize = dtrace_toxranges_max * sizeof (dtrace_toxrange_t); + range = kmem_zalloc(nsize, KM_SLEEP); + + if (dtrace_toxrange != NULL) { + ASSERT(osize != 0); + bcopy(dtrace_toxrange, range, osize); + kmem_free(dtrace_toxrange, osize); + } + + dtrace_toxrange = range; + } + + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_base == NULL); + ASSERT(dtrace_toxrange[dtrace_toxranges].dtt_limit == NULL); + + dtrace_toxrange[dtrace_toxranges].dtt_base = base; + dtrace_toxrange[dtrace_toxranges].dtt_limit = limit; + dtrace_toxranges++; +} + +/* + * DTrace Driver Cookbook Functions + */ +#ifdef VBOX +int dtrace_attach(void) +#else +/*ARGSUSED*/ +static int +dtrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +#endif +{ + dtrace_provider_id_t id; + dtrace_state_t *state = NULL; + dtrace_enabling_t *enab; + +#ifdef VBOX + if ( VBoxDtMutexInit(&dtrace_lock) + || VBoxDtMutexInit(&dtrace_provider_lock) + || VBoxDtMutexInit(&dtrace_meta_lock) +# ifdef DEBUG + || VBoxDtMutexInit(&dtrace_errlock) +# endif + ) + return (DDI_FAILURE); +#endif + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + +#ifndef VBOX + if (ddi_soft_state_init(&dtrace_softstate, + sizeof (dtrace_state_t), 0) != 0) { + cmn_err(CE_NOTE, "/dev/dtrace failed to initialize soft state"); + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, DTRACEMNR_DTRACE, S_IFCHR, + DTRACEMNRN_DTRACE, DDI_PSEUDO, NULL) == DDI_FAILURE || + ddi_create_minor_node(devi, DTRACEMNR_HELPER, S_IFCHR, + DTRACEMNRN_HELPER, DDI_PSEUDO, NULL) == DDI_FAILURE) { + cmn_err(CE_NOTE, "/dev/dtrace couldn't create minor nodes"); + ddi_remove_minor_node(devi, NULL); + ddi_soft_state_fini(&dtrace_softstate); + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + dtrace_devi = devi; + + dtrace_modload = dtrace_module_loaded; + dtrace_modunload = dtrace_module_unloaded; + dtrace_cpu_init = dtrace_cpu_setup_initial; + dtrace_helpers_cleanup = dtrace_helpers_destroy; + dtrace_helpers_fork = dtrace_helpers_duplicate; + dtrace_cpustart_init = dtrace_suspend; + dtrace_cpustart_fini = dtrace_resume; + dtrace_debugger_init = dtrace_suspend; + dtrace_debugger_fini = dtrace_resume; + + register_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); +#else + /** @todo some of these hooks needs checking out! */ +#endif + + ASSERT(MUTEX_HELD(&cpu_lock)); + +#ifndef VBOX /* Reduce the area a bit just to be sure our vmem fake doesn't blow up. */ + dtrace_arena = vmem_create("dtrace", (void *)1, UINT32_MAX, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); +#else + dtrace_arena = vmem_create("dtrace", (void *)(uintptr_t)1, UINT32_MAX - 16, 1, + NULL, NULL, NULL, 0, VM_SLEEP | VMC_IDENTIFIER); +#endif +#ifndef VBOX + dtrace_minor = vmem_create("dtrace_minor", (void *)DTRACEMNRN_CLONE, + UINT32_MAX - DTRACEMNRN_CLONE, 1, NULL, NULL, NULL, 0, + VM_SLEEP | VMC_IDENTIFIER); + dtrace_taskq = taskq_create("dtrace_taskq", 1, maxclsyspri, + 1, INT_MAX, 0); +#endif + + dtrace_state_cache = kmem_cache_create("dtrace_state_cache", + sizeof (dtrace_dstate_percpu_t) * NCPU, DTRACE_STATE_ALIGN, + NULL, NULL, NULL, NULL, NULL, 0); + + ASSERT(MUTEX_HELD(&cpu_lock)); + dtrace_bymod = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_mod), + offsetof(dtrace_probe_t, dtpr_nextmod), + offsetof(dtrace_probe_t, dtpr_prevmod)); + + dtrace_byfunc = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_func), + offsetof(dtrace_probe_t, dtpr_nextfunc), + offsetof(dtrace_probe_t, dtpr_prevfunc)); + + dtrace_byname = dtrace_hash_create(offsetof(dtrace_probe_t, dtpr_name), + offsetof(dtrace_probe_t, dtpr_nextname), + offsetof(dtrace_probe_t, dtpr_prevname)); + + if (dtrace_retain_max < 1) { + cmn_err(CE_WARN, "illegal value (%lu) for dtrace_retain_max; " + "setting to 1", dtrace_retain_max); + dtrace_retain_max = 1; + } + + /* + * Now discover our toxic ranges. + */ + dtrace_toxic_ranges(dtrace_toxrange_add); + + /* + * Before we register ourselves as a provider to our own framework, + * we would like to assert that dtrace_provider is NULL -- but that's + * not true if we were loaded as a dependency of a DTrace provider. + * Once we've registered, we can assert that dtrace_provider is our + * pseudo provider. + */ + (void) dtrace_register("dtrace", &dtrace_provider_attr, + DTRACE_PRIV_NONE, 0, &dtrace_provider_ops, NULL, &id); + + ASSERT(dtrace_provider != NULL); + ASSERT((dtrace_provider_id_t)dtrace_provider == id); + + dtrace_probeid_begin = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "BEGIN", 0, NULL); + dtrace_probeid_end = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "END", 0, NULL); + dtrace_probeid_error = dtrace_probe_create((dtrace_provider_id_t) + dtrace_provider, NULL, NULL, "ERROR", 1, NULL); + +#ifndef VBOX + dtrace_anon_property(); +#endif + mutex_exit(&cpu_lock); + + /* + * If DTrace helper tracing is enabled, we need to allocate the + * trace buffer and initialize the values. + */ + if (dtrace_helptrace_enabled) { + ASSERT(dtrace_helptrace_buffer == NULL); + dtrace_helptrace_buffer = + kmem_zalloc(dtrace_helptrace_bufsize, KM_SLEEP); + dtrace_helptrace_next = 0; + } + + /* + * If there are already providers, we must ask them to provide their + * probes, and then match any anonymous enabling against them. Note + * that there should be no other retained enablings at this time: + * the only retained enablings at this time should be the anonymous + * enabling. + */ + if (dtrace_anon.dta_enabling != NULL) { + ASSERT(dtrace_retained == dtrace_anon.dta_enabling); + + dtrace_enabling_provide(NULL); + state = dtrace_anon.dta_state; + + /* + * We couldn't hold cpu_lock across the above call to + * dtrace_enabling_provide(), but we must hold it to actually + * enable the probes. We have to drop all of our locks, pick + * up cpu_lock, and regain our locks before matching the + * retained anonymous enabling. + */ + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + if ((enab = dtrace_anon.dta_enabling) != NULL) + (void) dtrace_enabling_match(enab, NULL); + + mutex_exit(&cpu_lock); + } + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); + + if (state != NULL) { + /* + * If we created any anonymous state, set it going now. + */ + (void) dtrace_state_go(state, &dtrace_anon.dta_beganon); + } + + return (DDI_SUCCESS); +} + +#ifdef VBOX +int dtrace_open(dtrace_state_t **ppState, cred_t *cred_p) +#else +/*ARGSUSED*/ +static int +dtrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +#endif +{ + dtrace_state_t *state; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; + +#ifndef VBOX + if (getminor(*devp) == DTRACEMNRN_HELPER) + return (0); + + /* + * If this wasn't an open with the "helper" minor, then it must be + * the "dtrace" minor. + */ + if (getminor(*devp) != DTRACEMNRN_DTRACE) + return (ENXIO); +#endif /* !VBOX */ + + /* + * If no DTRACE_PRIV_* bits are set in the credential, then the + * caller lacks sufficient permission to do anything with DTrace. + */ + dtrace_cred2priv(cred_p, &priv, &uid, &zoneid); + if (priv == DTRACE_PRIV_NONE) + return (EACCES); + + /* + * Ask all providers to provide all their probes. + */ + mutex_enter(&dtrace_provider_lock); + dtrace_probe_provide(NULL, NULL); + mutex_exit(&dtrace_provider_lock); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + dtrace_opens++; + dtrace_membar_producer(); + +#ifndef VBOX + /* + * If the kernel debugger is active (that is, if the kernel debugger + * modified text in some way), we won't allow the open. + */ + if (kdi_dtrace_set(KDI_DTSET_DTRACE_ACTIVATE) != 0) { + dtrace_opens--; + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + return (EBUSY); + } +#endif + +#ifndef VBOX + state = dtrace_state_create(devp, cred_p); +#else + state = dtrace_state_create(cred_p); +#endif + mutex_exit(&cpu_lock); + + if (state == NULL) { +#ifndef VBOX + if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL) + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#else + dtrace_opens--; +#endif + mutex_exit(&dtrace_lock); + return (EAGAIN); + } + + mutex_exit(&dtrace_lock); + +#ifdef VBOX + *ppState = state; +#endif + return (0); +} + +#ifdef VBOX +int dtrace_close(dtrace_state_t *state) +#else +/*ARGSUSED*/ +static int +dtrace_close(dev_t dev, int flag, int otyp, cred_t *cred_p) +#endif +{ +#ifndef VBOX + minor_t minor = getminor(dev); + dtrace_state_t *state; + + if (minor == DTRACEMNRN_HELPER) + return (0); + + state = ddi_get_soft_state(dtrace_softstate, minor); +#endif + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + + if (state->dts_anon) { + /* + * There is anonymous state. Destroy that first. + */ + ASSERT(dtrace_anon.dta_state == NULL); + dtrace_state_destroy(state->dts_anon); + } + + dtrace_state_destroy(state); + ASSERT(dtrace_opens > 0); + +#ifndef VBOX + /* + * Only relinquish control of the kernel debugger interface when there + * are no consumers and no anonymous enablings. + */ + if (--dtrace_opens == 0 && dtrace_anon.dta_enabling == NULL) + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#else + dtrace_opens--; +#endif + + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + + return (0); +} + +#ifndef VBOX +/*ARGSUSED*/ +static int +dtrace_ioctl_helper(int cmd, intptr_t arg, int *rv) +{ + int rval; + dof_helper_t help, *dhp = NULL; + + switch (cmd) { + case DTRACEHIOC_ADDDOF: + if (copyin((void *)arg, &help, sizeof (help)) != 0) { + dtrace_dof_error(NULL, "failed to copyin DOF helper"); + return (EFAULT); + } + + dhp = &help; + arg = (intptr_t)help.dofhp_dof; + RT_FALL_THRU(); + + case DTRACEHIOC_ADD: { + dof_hdr_t *dof = dtrace_dof_copyin(arg, &rval); + + if (dof == NULL) + return (rval); + + mutex_enter(&dtrace_lock); + + /* + * dtrace_helper_slurp() takes responsibility for the dof -- + * it may free it now or it may save it and free it later. + */ + if ((rval = dtrace_helper_slurp(dof, dhp)) != -1) { + *rv = rval; + rval = 0; + } else { + rval = EINVAL; + } + + mutex_exit(&dtrace_lock); + return (rval); + } + + case DTRACEHIOC_REMOVE: { + mutex_enter(&dtrace_lock); + rval = dtrace_helper_destroygen(arg); + mutex_exit(&dtrace_lock); + + return (rval); + } + + default: + break; + } + + return (ENOTTY); +} +#endif /* !VBOX */ + +#ifdef VBOX +int dtrace_ioctl(dtrace_state_t *state, int cmd, intptr_t arg, int32_t *rv) +#else +/*ARGSUSED*/ +static int +dtrace_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) +#endif +{ +#ifndef VBOX + minor_t minor = getminor(dev); + dtrace_state_t *state; +#endif + int rval; + +#ifndef VBOX + if (minor == DTRACEMNRN_HELPER) + return (dtrace_ioctl_helper(cmd, arg, rv)); + + state = ddi_get_soft_state(dtrace_softstate, minor); +#endif + + if (state->dts_anon) { + ASSERT(dtrace_anon.dta_state == NULL); + state = state->dts_anon; + } + + switch (cmd) { + case DTRACEIOC_PROVIDER: { + dtrace_providerdesc_t pvd; + dtrace_provider_t *pvp; + + if (copyin((void *)arg, &pvd, sizeof (pvd)) != 0) + return (EFAULT); + + pvd.dtvd_name[DTRACE_PROVNAMELEN - 1] = '\0'; + mutex_enter(&dtrace_provider_lock); + + for (pvp = dtrace_provider; pvp != NULL; pvp = pvp->dtpv_next) { + if (strcmp(pvp->dtpv_name, pvd.dtvd_name) == 0) + break; + } + + mutex_exit(&dtrace_provider_lock); + + if (pvp == NULL) + return (ESRCH); + + bcopy(&pvp->dtpv_priv, &pvd.dtvd_priv, sizeof (dtrace_ppriv_t)); + bcopy(&pvp->dtpv_attr, &pvd.dtvd_attr, sizeof (dtrace_pattr_t)); + if (copyout(&pvd, (void *)arg, sizeof (pvd)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_EPROBE: { + dtrace_eprobedesc_t epdesc; + dtrace_ecb_t *ecb; + dtrace_action_t *act; + void *buf; + size_t size; + uintptr_t dest; + int nrecs; + + if (copyin((void *)arg, &epdesc, sizeof (epdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((ecb = dtrace_epid2ecb(state, epdesc.dtepd_epid)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + if (ecb->dte_probe == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + epdesc.dtepd_probeid = ecb->dte_probe->dtpr_id; + epdesc.dtepd_uarg = ecb->dte_uarg; + epdesc.dtepd_size = VBDTCAST(uint32_t)ecb->dte_size; + + nrecs = epdesc.dtepd_nrecs; + epdesc.dtepd_nrecs = 0; + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + epdesc.dtepd_nrecs++; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_eprobedesc_t) + + (epdesc.dtepd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&epdesc, (void *)dest, sizeof (epdesc)); + dest += offsetof(dtrace_eprobedesc_t, dtepd_rec[0]); + + for (act = ecb->dte_action; act != NULL; act = act->dta_next) { + if (DTRACEACT_ISAGG(act->dta_kind) || act->dta_intuple) + continue; + + if (nrecs-- == 0) + break; + + bcopy(&act->dta_rec, (void *)dest, + sizeof (dtrace_recdesc_t)); + dest += sizeof (dtrace_recdesc_t); + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + + case DTRACEIOC_AGGDESC: { + dtrace_aggdesc_t aggdesc; + dtrace_action_t *act; + dtrace_aggregation_t *agg; + int nrecs; + uint32_t offs; + dtrace_recdesc_t *lrec; + void *buf; + size_t size; + uintptr_t dest; + + if (copyin((void *)arg, &aggdesc, sizeof (aggdesc)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if ((agg = dtrace_aggid2agg(state, aggdesc.dtagd_id)) == NULL) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + aggdesc.dtagd_epid = agg->dtag_ecb->dte_epid; + + nrecs = aggdesc.dtagd_nrecs; + aggdesc.dtagd_nrecs = 0; + + offs = agg->dtag_base; + lrec = &agg->dtag_action.dta_rec; + aggdesc.dtagd_size = lrec->dtrd_offset + lrec->dtrd_size - offs; + + for (act = agg->dtag_first; ; act = act->dta_next) { + ASSERT(act->dta_intuple || + DTRACEACT_ISAGG(act->dta_kind)); + + /* + * If this action has a record size of zero, it + * denotes an argument to the aggregating action. + * Because the presence of this record doesn't (or + * shouldn't) affect the way the data is interpreted, + * we don't copy it out to save user-level the + * confusion of dealing with a zero-length record. + */ + if (act->dta_rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + aggdesc.dtagd_nrecs++; + + if (act == &agg->dtag_action) + break; + } + + /* + * Now that we have the size, we need to allocate a temporary + * buffer in which to store the complete description. We need + * the temporary buffer to be able to drop dtrace_lock() + * across the copyout(), below. + */ + size = sizeof (dtrace_aggdesc_t) + + (aggdesc.dtagd_nrecs * sizeof (dtrace_recdesc_t)); + + buf = kmem_alloc(size, KM_SLEEP); + dest = (uintptr_t)buf; + + bcopy(&aggdesc, (void *)dest, sizeof (aggdesc)); + dest += offsetof(dtrace_aggdesc_t, dtagd_rec[0]); + + for (act = agg->dtag_first; ; act = act->dta_next) { + dtrace_recdesc_t rec = act->dta_rec; + + /* + * See the comment in the above loop for why we pass + * over zero-length records. + */ + if (rec.dtrd_size == 0) { + ASSERT(agg->dtag_hasarg); + continue; + } + + if (nrecs-- == 0) + break; + + rec.dtrd_offset -= offs; + bcopy(&rec, (void *)dest, sizeof (rec)); + dest += sizeof (dtrace_recdesc_t); + + if (act == &agg->dtag_action) + break; + } + + mutex_exit(&dtrace_lock); + + if (copyout(buf, (void *)arg, dest - (uintptr_t)buf) != 0) { + kmem_free(buf, size); + return (EFAULT); + } + + kmem_free(buf, size); + return (0); + } + + case DTRACEIOC_ENABLE: { + dof_hdr_t *dof; + dtrace_enabling_t *enab = NULL; + dtrace_vstate_t *vstate; + int err = 0; +#ifdef VBOX + cred_t *cr = CRED(); +#endif + + *rv = 0; + + /* + * If a NULL argument has been passed, we take this as our + * cue to reevaluate our enablings. + */ + if (arg == NULL) { + dtrace_enabling_matchall(); + + return (0); + } + + if ((dof = dtrace_dof_copyin(arg, &rval)) == NULL) + return (rval); + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_lock); + vstate = &state->dts_vstate; + + if (state->dts_activity != DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EBUSY); + } + + if (dtrace_dof_slurp(dof, vstate, cr, &enab, 0, B_TRUE) != 0) { + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (EINVAL); + } + + if ((rval = dtrace_dof_options(dof, state)) != 0) { + dtrace_enabling_destroy(enab); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + dtrace_dof_destroy(dof); + return (rval); + } + + if ((err = dtrace_enabling_match(enab, rv)) == 0) { + err = dtrace_enabling_retain(enab); + } else { + dtrace_enabling_destroy(enab); + } + + mutex_exit(&cpu_lock); + mutex_exit(&dtrace_lock); + dtrace_dof_destroy(dof); + + return (err); + } + + case DTRACEIOC_REPLICATE: { + dtrace_repldesc_t desc; + dtrace_probedesc_t *match = &desc.dtrpd_match; + dtrace_probedesc_t *create = &desc.dtrpd_create; + int err; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + match->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + match->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + match->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + match->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + create->dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + create->dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + create->dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + create->dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + mutex_enter(&dtrace_lock); + err = dtrace_enabling_replicate(state, match, create); + mutex_exit(&dtrace_lock); + + return (err); + } + + case DTRACEIOC_PROBEMATCH: + case DTRACEIOC_PROBES: { + dtrace_probe_t *probe = NULL; + dtrace_probedesc_t desc; + dtrace_probekey_t pkey; + dtrace_id_t i; + int m = 0; + uint32_t priv; + uid_t uid; + zoneid_t zoneid; +#ifdef VBOX + cred_t *cr = CRED(); +#endif + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + desc.dtpd_provider[DTRACE_PROVNAMELEN - 1] = '\0'; + desc.dtpd_mod[DTRACE_MODNAMELEN - 1] = '\0'; + desc.dtpd_func[DTRACE_FUNCNAMELEN - 1] = '\0'; + desc.dtpd_name[DTRACE_NAMELEN - 1] = '\0'; + + /* + * Before we attempt to match this probe, we want to give + * all providers the opportunity to provide it. + */ + if (desc.dtpd_id == DTRACE_IDNONE) { + mutex_enter(&dtrace_provider_lock); + dtrace_probe_provide(&desc, NULL); + mutex_exit(&dtrace_provider_lock); + desc.dtpd_id++; + } + + if (cmd == DTRACEIOC_PROBEMATCH) { + dtrace_probekey(&desc, &pkey); + pkey.dtpk_id = DTRACE_IDNONE; + } + + dtrace_cred2priv(cr, &priv, &uid, &zoneid); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_PROBEMATCH) { + for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + (m = dtrace_match_probe(probe, &pkey, + priv, uid, zoneid)) != 0) + break; + } + + if (m < 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + } else { + for (i = desc.dtpd_id; i <= dtrace_nprobes; i++) { + if ((probe = dtrace_probes[i - 1]) != NULL && + dtrace_match_priv(probe, priv, uid, zoneid)) + break; + } + } + + if (probe == NULL) { + mutex_exit(&dtrace_lock); + return (ESRCH); + } + + dtrace_probe_description(probe, &desc); + mutex_exit(&dtrace_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_PROBEARG: { + dtrace_argdesc_t desc; + dtrace_probe_t *probe; + dtrace_provider_t *prov; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + if (desc.dtargd_id == DTRACE_IDNONE) + return (EINVAL); + + if (desc.dtargd_ndx == DTRACE_ARGNONE) + return (EINVAL); + + mutex_enter(&dtrace_provider_lock); + mutex_enter(&mod_lock); + mutex_enter(&dtrace_lock); + + if (desc.dtargd_id > dtrace_nprobes) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + if ((probe = dtrace_probes[desc.dtargd_id - 1]) == NULL) { + mutex_exit(&dtrace_lock); + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + return (EINVAL); + } + + mutex_exit(&dtrace_lock); + + prov = probe->dtpr_provider; + + if (prov->dtpv_pops.dtps_getargdesc == NULL) { + /* + * There isn't any typed information for this probe. + * Set the argument number to DTRACE_ARGNONE. + */ + desc.dtargd_ndx = DTRACE_ARGNONE; + } else { + desc.dtargd_native[0] = '\0'; + desc.dtargd_xlate[0] = '\0'; + desc.dtargd_mapping = desc.dtargd_ndx; + + prov->dtpv_pops.dtps_getargdesc(prov->dtpv_arg, + probe->dtpr_id, probe->dtpr_arg, &desc); + } + + mutex_exit(&mod_lock); + mutex_exit(&dtrace_provider_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_GO: { + processorid_t cpuid; + rval = dtrace_state_go(state, &cpuid); + + if (rval != 0) + return (rval); + + if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_STOP: { + processorid_t cpuid; + + mutex_enter(&dtrace_lock); + rval = dtrace_state_stop(state, &cpuid); + mutex_exit(&dtrace_lock); + + if (rval != 0) + return (rval); + + if (copyout(&cpuid, (void *)arg, sizeof (cpuid)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_DOFGET: { + dof_hdr_t hdr, *dof; + uint64_t len; + + if (copyin((void *)arg, &hdr, sizeof (hdr)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + dof = dtrace_dof_create(state); + mutex_exit(&dtrace_lock); + + len = MIN(hdr.dofh_loadsz, dof->dofh_loadsz); + rval = copyout(dof, (void *)arg, len); + dtrace_dof_destroy(dof); + + return (rval == 0 ? 0 : EFAULT); + } + + case DTRACEIOC_AGGSNAP: + case DTRACEIOC_BUFSNAP: { + dtrace_bufdesc_t desc; + caddr_t cached; + dtrace_buffer_t *buf; + + if (copyin((void *)arg, &desc, sizeof (desc)) != 0) + return (EFAULT); + + if (/*VBox value is is unsigned: desc.dtbd_cpu < 0 ||*/ desc.dtbd_cpu >= NCPU) + return (EINVAL); + + mutex_enter(&dtrace_lock); + + if (cmd == DTRACEIOC_BUFSNAP) { + buf = &state->dts_buffer[desc.dtbd_cpu]; + } else { + buf = &state->dts_aggbuffer[desc.dtbd_cpu]; + } + + if (buf->dtb_flags & (DTRACEBUF_RING | DTRACEBUF_FILL)) { + size_t sz = buf->dtb_offset; + + if (state->dts_activity != DTRACE_ACTIVITY_STOPPED) { + mutex_exit(&dtrace_lock); + return (EBUSY); + } + + /* + * If this buffer has already been consumed, we're + * going to indicate that there's nothing left here + * to consume. + */ + if (buf->dtb_flags & DTRACEBUF_CONSUMED) { + mutex_exit(&dtrace_lock); + + desc.dtbd_size = 0; + desc.dtbd_drops = 0; + desc.dtbd_errors = 0; + desc.dtbd_oldest = 0; + sz = sizeof (desc); + + if (copyout(&desc, (void *)arg, sz) != 0) + return (EFAULT); + + return (0); + } + + /* + * If this is a ring buffer that has wrapped, we want + * to copy the whole thing out. + */ + if (buf->dtb_flags & DTRACEBUF_WRAPPED) { + dtrace_buffer_polish(buf); + sz = buf->dtb_size; + } + + if (copyout(buf->dtb_tomax, desc.dtbd_data, sz) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = sz; + desc.dtbd_drops = buf->dtb_drops; + desc.dtbd_errors = buf->dtb_errors; + desc.dtbd_oldest = buf->dtb_xamot_offset; + + mutex_exit(&dtrace_lock); + + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + buf->dtb_flags |= DTRACEBUF_CONSUMED; + + return (0); + } + + if (buf->dtb_tomax == NULL) { + ASSERT(buf->dtb_xamot == NULL); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + cached = buf->dtb_tomax; + ASSERT(!(buf->dtb_flags & DTRACEBUF_NOSWITCH)); + +#ifndef VBOX + dtrace_xcall(desc.dtbd_cpu, + (dtrace_xcall_t)dtrace_buffer_switch, buf); +#else + if (desc.dtbd_cpu == DTRACE_CPUALL) + RTMpOnAll(dtrace_buffer_switch_wrapper, buf, NULL); + else + RTMpOnSpecific(desc.dtbd_cpu, dtrace_buffer_switch_wrapper, buf, NULL); +#endif + + state->dts_errors += buf->dtb_xamot_errors; + + /* + * If the buffers did not actually switch, then the cross call + * did not take place -- presumably because the given CPU is + * not in the ready set. If this is the case, we'll return + * ENOENT. + */ + if (buf->dtb_tomax == cached) { + ASSERT(buf->dtb_xamot != cached); + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + ASSERT(cached == buf->dtb_xamot); + + /* + * We have our snapshot; now copy it out. + */ + if (copyout(buf->dtb_xamot, desc.dtbd_data, + buf->dtb_xamot_offset) != 0) { + mutex_exit(&dtrace_lock); + return (EFAULT); + } + + desc.dtbd_size = buf->dtb_xamot_offset; + desc.dtbd_drops = buf->dtb_xamot_drops; + desc.dtbd_errors = buf->dtb_xamot_errors; + desc.dtbd_oldest = 0; + + mutex_exit(&dtrace_lock); + + /* + * Finally, copy out the buffer description. + */ + if (copyout(&desc, (void *)arg, sizeof (desc)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_CONF: { + dtrace_conf_t conf; + + bzero(&conf, sizeof (conf)); + conf.dtc_difversion = DIF_VERSION; + conf.dtc_difintregs = DIF_DIR_NREGS; + conf.dtc_diftupregs = DIF_DTR_NREGS; + conf.dtc_ctfmodel = CTF_MODEL_NATIVE; + + if (copyout(&conf, (void *)arg, sizeof (conf)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_STATUS: { + dtrace_status_t stat; + dtrace_dstate_t *dstate; + int i, j; + uint64_t nerrs; + + /* + * See the comment in dtrace_state_deadman() for the reason + * for setting dts_laststatus to INT64_MAX before setting + * it to the correct value. + */ + state->dts_laststatus = INT64_MAX; + dtrace_membar_producer(); + state->dts_laststatus = dtrace_gethrtime(); + + bzero(&stat, sizeof (stat)); + + mutex_enter(&dtrace_lock); + + if (state->dts_activity == DTRACE_ACTIVITY_INACTIVE) { + mutex_exit(&dtrace_lock); + return (ENOENT); + } + + if (state->dts_activity == DTRACE_ACTIVITY_DRAINING) + stat.dtst_exiting = 1; + + nerrs = state->dts_errors; + dstate = &state->dts_vstate.dtvs_dynvars; + + for (i = 0; i < NCPU; i++) { + dtrace_dstate_percpu_t *dcpu = &dstate->dtds_percpu[i]; + + stat.dtst_dyndrops += dcpu->dtdsc_drops; + stat.dtst_dyndrops_dirty += dcpu->dtdsc_dirty_drops; + stat.dtst_dyndrops_rinsing += dcpu->dtdsc_rinsing_drops; + + if (state->dts_buffer[i].dtb_flags & DTRACEBUF_FULL) + stat.dtst_filled++; + + nerrs += state->dts_buffer[i].dtb_errors; + + for (j = 0; j < state->dts_nspeculations; j++) { + dtrace_speculation_t *spec; + dtrace_buffer_t *buf; + + spec = &state->dts_speculations[j]; + buf = &spec->dtsp_buffer[i]; + stat.dtst_specdrops += buf->dtb_xamot_drops; + } + } + + stat.dtst_specdrops_busy = state->dts_speculations_busy; + stat.dtst_specdrops_unavail = state->dts_speculations_unavail; + stat.dtst_stkstroverflows = state->dts_stkstroverflows; + stat.dtst_dblerrors = state->dts_dblerrors; + stat.dtst_killed = + (state->dts_activity == DTRACE_ACTIVITY_KILLED); + stat.dtst_errors = nerrs; + + mutex_exit(&dtrace_lock); + + if (copyout(&stat, (void *)arg, sizeof (stat)) != 0) + return (EFAULT); + + return (0); + } + + case DTRACEIOC_FORMAT: { + dtrace_fmtdesc_t fmt; + char *str; + int len; + + if (copyin((void *)arg, &fmt, sizeof (fmt)) != 0) + return (EFAULT); + + mutex_enter(&dtrace_lock); + + if (fmt.dtfd_format == 0 || + fmt.dtfd_format > state->dts_nformats) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + + /* + * Format strings are allocated contiguously and they are + * never freed; if a format index is less than the number + * of formats, we can assert that the format map is non-NULL + * and that the format for the specified index is non-NULL. + */ + ASSERT(state->dts_formats != NULL); + str = state->dts_formats[fmt.dtfd_format - 1]; + ASSERT(str != NULL); + + len = VBDTCAST(int)strlen(str) + 1; + + if (len > fmt.dtfd_length) { + fmt.dtfd_length = len; + + if (copyout(&fmt, (void *)arg, sizeof (fmt)) != 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + } else { + if (copyout(str, fmt.dtfd_string, len) != 0) { + mutex_exit(&dtrace_lock); + return (EINVAL); + } + } + + mutex_exit(&dtrace_lock); + return (0); + } + + default: + break; + } + + return (ENOTTY); +} + +#ifdef VBOX +int dtrace_detach(void) +#else +/*ARGSUSED*/ +static int +dtrace_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) +#endif +{ + dtrace_state_t *state; + +#ifndef VBOX + switch (cmd) { + case DDI_DETACH: + break; + + case DDI_SUSPEND: + return (DDI_SUCCESS); + + default: + return (DDI_FAILURE); + } +#endif + + mutex_enter(&cpu_lock); + mutex_enter(&dtrace_provider_lock); + mutex_enter(&dtrace_lock); + + ASSERT(dtrace_opens == 0); + + if (dtrace_helpers > 0) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + return (DDI_FAILURE); + } + + if (dtrace_unregister((dtrace_provider_id_t)dtrace_provider) != 0) { + mutex_exit(&dtrace_provider_lock); + mutex_exit(&dtrace_lock); + mutex_exit(&cpu_lock); + return (DDI_FAILURE); + } + + dtrace_provider = NULL; + + if ((state = dtrace_anon_grab()) != NULL) { + /* + * If there were ECBs on this state, the provider should + * have not been allowed to detach; assert that there is + * none. + */ + ASSERT(state->dts_necbs == 0); + dtrace_state_destroy(state); + +#ifndef VBOX + /* + * If we're being detached with anonymous state, we need to + * indicate to the kernel debugger that DTrace is now inactive. + */ + (void) kdi_dtrace_set(KDI_DTSET_DTRACE_DEACTIVATE); +#endif + } + + bzero(&dtrace_anon, sizeof (dtrace_anon_t)); +#ifndef VBOX /** @todo CPU hooks */ + unregister_cpu_setup_func((cpu_setup_func_t *)dtrace_cpu_setup, NULL); + dtrace_cpu_init = NULL; + dtrace_helpers_cleanup = NULL; + dtrace_helpers_fork = NULL; + dtrace_cpustart_init = NULL; + dtrace_cpustart_fini = NULL; + dtrace_debugger_init = NULL; + dtrace_debugger_fini = NULL; + dtrace_modload = NULL; + dtrace_modunload = NULL; +#endif + + mutex_exit(&cpu_lock); + + if (dtrace_helptrace_enabled) { + kmem_free(dtrace_helptrace_buffer, dtrace_helptrace_bufsize); + dtrace_helptrace_buffer = NULL; + } + + kmem_free(dtrace_probes, dtrace_nprobes * sizeof (dtrace_probe_t *)); + dtrace_probes = NULL; + dtrace_nprobes = 0; + + dtrace_hash_destroy(dtrace_bymod); + dtrace_hash_destroy(dtrace_byfunc); + dtrace_hash_destroy(dtrace_byname); + dtrace_bymod = NULL; + dtrace_byfunc = NULL; + dtrace_byname = NULL; + + kmem_cache_destroy(dtrace_state_cache); +#ifndef VBOX + vmem_destroy(dtrace_minor); +#endif + vmem_destroy(dtrace_arena); + + if (dtrace_toxrange != NULL) { + kmem_free(dtrace_toxrange, + dtrace_toxranges_max * sizeof (dtrace_toxrange_t)); + dtrace_toxrange = NULL; + dtrace_toxranges = 0; + dtrace_toxranges_max = 0; + } + +#ifndef VBOX + ddi_remove_minor_node(dtrace_devi, NULL); + dtrace_devi = NULL; + + ddi_soft_state_fini(&dtrace_softstate); +#endif + + ASSERT(dtrace_vtime_references == 0); + ASSERT(dtrace_opens == 0); + ASSERT(dtrace_retained == NULL); + + mutex_exit(&dtrace_lock); + mutex_exit(&dtrace_provider_lock); +#ifdef VBOX + VBoxDtMutexDelete(&dtrace_lock); + VBoxDtMutexDelete(&dtrace_provider_lock); + VBoxDtMutexDelete(&dtrace_meta_lock); +# ifdef DEBUG + VBoxDtMutexDelete(&dtrace_errlock); +# endif +#endif + + /* + * We don't destroy the task queue until after we have dropped our + * locks (taskq_destroy() may block on running tasks). To prevent + * attempting to do work after we have effectively detached but before + * the task queue has been destroyed, all tasks dispatched via the + * task queue must check that DTrace is still attached before + * performing any operation. + */ +#ifndef VBOX + taskq_destroy(dtrace_taskq); + dtrace_taskq = NULL; +#endif + + return (DDI_SUCCESS); +} + +#ifndef VBOX +/*ARGSUSED*/ +static int +dtrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)dtrace_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +static struct cb_ops dtrace_cb_ops = { + dtrace_open, /* open */ + dtrace_close, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + dtrace_ioctl, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops dtrace_ops = { + DEVO_REV, /* devo_rev */ + 0, /* refcnt */ + dtrace_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + dtrace_attach, /* attach */ + dtrace_detach, /* detach */ + nodev, /* reset */ + &dtrace_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Dynamic Tracing", /* name of module */ + &dtrace_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} + +#endif /* !VBOX */ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.conf new file mode 100644 index 00000000..6c4ec26b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/dtrace.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="dtrace" parent="pseudo" instance=0; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/fasttrap.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/fasttrap.c new file mode 100644 index 00000000..42263e4e --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/fasttrap.c @@ -0,0 +1,2378 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/atomic.h> +#include <sys/errno.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#include <sys/strsubr.h> +#include <sys/fasttrap.h> +#include <sys/fasttrap_impl.h> +#include <sys/fasttrap_isa.h> +#include <sys/dtrace.h> +#include <sys/dtrace_impl.h> +#include <sys/sysmacros.h> +#include <sys/proc.h> +#include <sys/priv.h> +#include <sys/policy.h> +#include <util/qsort.h> + +/* + * User-Land Trap-Based Tracing + * ---------------------------- + * + * The fasttrap provider allows DTrace consumers to instrument any user-level + * instruction to gather data; this includes probes with semantic + * signifigance like entry and return as well as simple offsets into the + * function. While the specific techniques used are very ISA specific, the + * methodology is generalizable to any architecture. + * + * + * The General Methodology + * ----------------------- + * + * With the primary goal of tracing every user-land instruction and the + * limitation that we can't trust user space so don't want to rely on much + * information there, we begin by replacing the instructions we want to trace + * with trap instructions. Each instruction we overwrite is saved into a hash + * table keyed by process ID and pc address. When we enter the kernel due to + * this trap instruction, we need the effects of the replaced instruction to + * appear to have occurred before we proceed with the user thread's + * execution. + * + * Each user level thread is represented by a ulwp_t structure which is + * always easily accessible through a register. The most basic way to produce + * the effects of the instruction we replaced is to copy that instruction out + * to a bit of scratch space reserved in the user thread's ulwp_t structure + * (a sort of kernel-private thread local storage), set the PC to that + * scratch space and single step. When we reenter the kernel after single + * stepping the instruction we must then adjust the PC to point to what would + * normally be the next instruction. Of course, special care must be taken + * for branches and jumps, but these represent such a small fraction of any + * instruction set that writing the code to emulate these in the kernel is + * not too difficult. + * + * Return probes may require several tracepoints to trace every return site, + * and, conversely, each tracepoint may activate several probes (the entry + * and offset 0 probes, for example). To solve this muliplexing problem, + * tracepoints contain lists of probes to activate and probes contain lists + * of tracepoints to enable. If a probe is activated, it adds its ID to + * existing tracepoints or creates new ones as necessary. + * + * Most probes are activated _before_ the instruction is executed, but return + * probes are activated _after_ the effects of the last instruction of the + * function are visible. Return probes must be fired _after_ we have + * single-stepped the instruction whereas all other probes are fired + * beforehand. + * + * + * Lock Ordering + * ------------- + * + * The lock ordering below -- both internally and with respect to the DTrace + * framework -- is a little tricky and bears some explanation. Each provider + * has a lock (ftp_mtx) that protects its members including reference counts + * for enabled probes (ftp_rcount), consumers actively creating probes + * (ftp_ccount) and USDT consumers (ftp_mcount); all three prevent a provider + * from being freed. A provider is looked up by taking the bucket lock for the + * provider hash table, and is returned with its lock held. The provider lock + * may be taken in functions invoked by the DTrace framework, but may not be + * held while calling functions in the DTrace framework. + * + * To ensure consistency over multiple calls to the DTrace framework, the + * creation lock (ftp_cmtx) should be held. Naturally, the creation lock may + * not be taken when holding the provider lock as that would create a cyclic + * lock ordering. In situations where one would naturally take the provider + * lock and then the creation lock, we instead up a reference count to prevent + * the provider from disappearing, drop the provider lock, and acquire the + * creation lock. + * + * Briefly: + * bucket lock before provider lock + * DTrace before provider lock + * creation lock before DTrace + * never hold the provider lock and creation lock simultaneously + */ + +static dev_info_t *fasttrap_devi; +static dtrace_meta_provider_id_t fasttrap_meta_id; + +static timeout_id_t fasttrap_timeout; +static kmutex_t fasttrap_cleanup_mtx; +static uint_t fasttrap_cleanup_work; + +/* + * Generation count on modifications to the global tracepoint lookup table. + */ +static volatile uint64_t fasttrap_mod_gen; + +/* + * When the fasttrap provider is loaded, fasttrap_max is set to either + * FASTTRAP_MAX_DEFAULT or the value for fasttrap-max-probes in the + * fasttrap.conf file. Each time a probe is created, fasttrap_total is + * incremented by the number of tracepoints that may be associated with that + * probe; fasttrap_total is capped at fasttrap_max. + */ +#define FASTTRAP_MAX_DEFAULT 250000 +static uint32_t fasttrap_max; +static uint32_t fasttrap_total; + + +#define FASTTRAP_TPOINTS_DEFAULT_SIZE 0x4000 +#define FASTTRAP_PROVIDERS_DEFAULT_SIZE 0x100 +#define FASTTRAP_PROCS_DEFAULT_SIZE 0x100 + +#define FASTTRAP_PID_NAME "pid" + +fasttrap_hash_t fasttrap_tpoints; +static fasttrap_hash_t fasttrap_provs; +static fasttrap_hash_t fasttrap_procs; + +static uint64_t fasttrap_pid_count; /* pid ref count */ +static kmutex_t fasttrap_count_mtx; /* lock on ref count */ + +#define FASTTRAP_ENABLE_FAIL 1 +#define FASTTRAP_ENABLE_PARTIAL 2 + +static int fasttrap_tracepoint_enable(proc_t *, fasttrap_probe_t *, uint_t); +static void fasttrap_tracepoint_disable(proc_t *, fasttrap_probe_t *, uint_t); + +static fasttrap_provider_t *fasttrap_provider_lookup(pid_t, const char *, + const dtrace_pattr_t *); +static void fasttrap_provider_retire(pid_t, const char *, int); +static void fasttrap_provider_free(fasttrap_provider_t *); + +static fasttrap_proc_t *fasttrap_proc_lookup(pid_t); +static void fasttrap_proc_release(fasttrap_proc_t *); + +#define FASTTRAP_PROVS_INDEX(pid, name) \ + ((fasttrap_hash_str(name) + (pid)) & fasttrap_provs.fth_mask) + +#define FASTTRAP_PROCS_INDEX(pid) ((pid) & fasttrap_procs.fth_mask) + +static int +fasttrap_highbit(ulong_t i) +{ + int h = 1; + + if (i == 0) + return (0); +#ifdef _LP64 + if (i & 0xffffffff00000000ul) { + h += 32; i >>= 32; + } +#endif + if (i & 0xffff0000) { + h += 16; i >>= 16; + } + if (i & 0xff00) { + h += 8; i >>= 8; + } + if (i & 0xf0) { + h += 4; i >>= 4; + } + if (i & 0xc) { + h += 2; i >>= 2; + } + if (i & 0x2) { + h += 1; + } + return (h); +} + +static uint_t +fasttrap_hash_str(const char *p) +{ + unsigned int g; + uint_t hval = 0; + + while (*p) { + hval = (hval << 4) + *p++; + if ((g = (hval & 0xf0000000)) != 0) + hval ^= g >> 24; + hval &= ~g; + } + return (hval); +} + +void +fasttrap_sigtrap(proc_t *p, kthread_t *t, uintptr_t pc) +{ + sigqueue_t *sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP); + + sqp->sq_info.si_signo = SIGTRAP; + sqp->sq_info.si_code = TRAP_DTRACE; + sqp->sq_info.si_addr = (caddr_t)pc; + + mutex_enter(&p->p_lock); + sigaddqa(p, t, sqp); + mutex_exit(&p->p_lock); + + if (t != NULL) + aston(t); +} + +/* + * This function ensures that no threads are actively using the memory + * associated with probes that were formerly live. + */ +static void +fasttrap_mod_barrier(uint64_t gen) +{ + int i; + + if (gen < fasttrap_mod_gen) + return; + + fasttrap_mod_gen++; + + for (i = 0; i < NCPU; i++) { + mutex_enter(&cpu_core[i].cpuc_pid_lock); + mutex_exit(&cpu_core[i].cpuc_pid_lock); + } +} + +/* + * This is the timeout's callback for cleaning up the providers and their + * probes. + */ +/*ARGSUSED*/ +static void +fasttrap_pid_cleanup_cb(void *data) +{ + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + int i, later; + + static volatile int in = 0; + ASSERT(in == 0); + in = 1; + + mutex_enter(&fasttrap_cleanup_mtx); + while (fasttrap_cleanup_work) { + fasttrap_cleanup_work = 0; + mutex_exit(&fasttrap_cleanup_mtx); + + later = 0; + + /* + * Iterate over all the providers trying to remove the marked + * ones. If a provider is marked but not retired, we just + * have to take a crack at removing it -- it's no big deal if + * we can't. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + bucket = &fasttrap_provs.fth_table[i]; + mutex_enter(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + + while ((fp = *fpp) != NULL) { + if (!fp->ftp_marked) { + fpp = &fp->ftp_next; + continue; + } + + mutex_enter(&fp->ftp_mtx); + + /* + * If this provider has consumers actively + * creating probes (ftp_ccount) or is a USDT + * provider (ftp_mcount), we can't unregister + * or even condense. + */ + if (fp->ftp_ccount != 0 || + fp->ftp_mcount != 0) { + mutex_exit(&fp->ftp_mtx); + fp->ftp_marked = 0; + continue; + } + + if (!fp->ftp_retired || fp->ftp_rcount != 0) + fp->ftp_marked = 0; + + mutex_exit(&fp->ftp_mtx); + + /* + * If we successfully unregister this + * provider we can remove it from the hash + * chain and free the memory. If our attempt + * to unregister fails and this is a retired + * provider, increment our flag to try again + * pretty soon. If we've consumed more than + * half of our total permitted number of + * probes call dtrace_condense() to try to + * clean out the unenabled probes. + */ + provid = fp->ftp_provid; + if (dtrace_unregister(provid) != 0) { + if (fasttrap_total > fasttrap_max / 2) + (void) dtrace_condense(provid); + later += fp->ftp_marked; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + mutex_exit(&bucket->ftb_mtx); + } + + mutex_enter(&fasttrap_cleanup_mtx); + } + + ASSERT(fasttrap_timeout != 0); + + /* + * If we were unable to remove a retired provider, try again after + * a second. This situation can occur in certain circumstances where + * providers cannot be unregistered even though they have no probes + * enabled because of an execution of dtrace -l or something similar. + * If the timeout has been disabled (set to 1 because we're trying + * to detach), we set fasttrap_cleanup_work to ensure that we'll + * get a chance to do that work if and when the timeout is reenabled + * (if detach fails). + */ + if (later > 0 && fasttrap_timeout != (timeout_id_t)1) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, hz); + else if (later > 0) + fasttrap_cleanup_work = 1; + else + fasttrap_timeout = 0; + + mutex_exit(&fasttrap_cleanup_mtx); + in = 0; +} + +/* + * Activates the asynchronous cleanup mechanism. + */ +static void +fasttrap_pid_cleanup(void) +{ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 1; + if (fasttrap_timeout == 0) + fasttrap_timeout = timeout(&fasttrap_pid_cleanup_cb, NULL, 1); + mutex_exit(&fasttrap_cleanup_mtx); +} + +/* + * This is called from cfork() via dtrace_fasttrap_fork(). The child + * process's address space is (roughly) a copy of the parent process's so + * we have to remove all the instrumentation we had previously enabled in the + * parent. + */ +static void +fasttrap_fork(proc_t *p, proc_t *cp) +{ + pid_t ppid = p->p_pid; + int i; + + ASSERT(curproc == p); + ASSERT(p->p_proc_flag & P_PR_LOCK); + ASSERT(p->p_dtrace_count > 0); + ASSERT(cp->p_dtrace_count == 0); + + /* + * This would be simpler and faster if we maintained per-process + * hash tables of enabled tracepoints. It could, however, potentially + * slow down execution of a tracepoint since we'd need to go + * through two levels of indirection. In the future, we should + * consider either maintaining per-process ancillary lists of + * enabled tracepoints or hanging a pointer to a per-process hash + * table of enabled tracepoints off the proc structure. + */ + + /* + * We don't have to worry about the child process disappearing + * because we're in fork(). + */ + mutex_enter(&cp->p_lock); + sprlock_proc(cp); + mutex_exit(&cp->p_lock); + + /* + * Iterate over every tracepoint looking for ones that belong to the + * parent process, and remove each from the child process. + */ + for (i = 0; i < fasttrap_tpoints.fth_nent; i++) { + fasttrap_tracepoint_t *tp; + fasttrap_bucket_t *bucket = &fasttrap_tpoints.fth_table[i]; + + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == ppid && + tp->ftt_proc->ftpc_acount != 0) { + int ret = fasttrap_tracepoint_remove(cp, tp); + ASSERT(ret == 0); + + /* + * The count of active providers can only be + * decremented (i.e. to zero) during exec, + * exit, and removal of a meta provider so it + * should be impossible to drop the count + * mid-fork. + */ + ASSERT(tp->ftt_proc->ftpc_acount != 0); + } + } + mutex_exit(&bucket->ftb_mtx); + } + + mutex_enter(&cp->p_lock); + sprunlock(cp); +} + +/* + * This is called from proc_exit() or from exec_common() if p_dtrace_probes + * is set on the proc structure to indicate that there is a pid provider + * associated with this process. + */ +static void +fasttrap_exec_exit(proc_t *p) +{ + ASSERT(p == curproc); + ASSERT(MUTEX_HELD(&p->p_lock)); + + mutex_exit(&p->p_lock); + + /* + * We clean up the pid provider for this process here; user-land + * static probes are handled by the meta-provider remove entry point. + */ + fasttrap_provider_retire(p->p_pid, FASTTRAP_PID_NAME, 0); + + mutex_enter(&p->p_lock); +} + + +/*ARGSUSED*/ +static void +fasttrap_pid_provide(void *arg, const dtrace_probedesc_t *desc) +{ + /* + * There are no "default" pid probes. + */ +} + +static int +fasttrap_tracepoint_enable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_tracepoint_t *tp, *new_tp = NULL; + fasttrap_bucket_t *bucket; + fasttrap_id_t *id; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + ASSERT(!(p->p_flag & SVFORK)); + + /* + * Before we make any modifications, make sure we've imposed a barrier + * on the generation in which this probe was last modified. + */ + fasttrap_mod_barrier(probe->ftp_gen); + + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + + /* + * If the tracepoint has already been enabled, just add our id to the + * list of interested probes. This may be our second time through + * this path in which case we'll have constructed the tracepoint we'd + * like to install. If we can't find a match, and have an allocated + * tracepoint ready to go, enable that one now. + * + * A tracepoint whose process is defunct is also considered defunct. + */ +again: + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + /* + * Note that it's safe to access the active count on the + * associated proc structure because we know that at least one + * provider (this one) will still be around throughout this + * operation. + */ + if (tp->ftt_pid != pid || tp->ftt_pc != pc || + tp->ftt_proc->ftpc_acount == 0) + continue; + + /* + * Now that we've found a matching tracepoint, it would be + * a decent idea to confirm that the tracepoint is still + * enabled and the trap instruction hasn't been overwritten. + * Since this is a little hairy, we'll punt for now. + */ + + /* + * This can't be the first interested probe. We don't have + * to worry about another thread being in the midst of + * deleting this tracepoint (which would be the only valid + * reason for a tracepoint to have no interested probes) + * since we're holding P_PR_LOCK for this process. + */ + ASSERT(tp->ftt_ids != NULL || tp->ftt_retids != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = tp->ftt_ids; + membar_producer(); + tp->ftt_ids = id; + membar_producer(); + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = tp->ftt_retids; + membar_producer(); + tp->ftt_retids = id; + membar_producer(); + break; + + default: + ASSERT(0); + } + + mutex_exit(&bucket->ftb_mtx); + + if (new_tp != NULL) { + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + } + + return (0); + } + + /* + * If we have a good tracepoint ready to go, install it now while + * we have the lock held and no one can screw with us. + */ + if (new_tp != NULL) { + int rc = 0; + + new_tp->ftt_next = bucket->ftb_data; + membar_producer(); + bucket->ftb_data = new_tp; + membar_producer(); + mutex_exit(&bucket->ftb_mtx); + + /* + * Activate the tracepoint in the ISA-specific manner. + * If this fails, we need to report the failure, but + * indicate that this tracepoint must still be disabled + * by calling fasttrap_tracepoint_disable(). + */ + if (fasttrap_tracepoint_install(p, new_tp) != 0) + rc = FASTTRAP_ENABLE_PARTIAL; + + /* + * Increment the count of the number of tracepoints active in + * the victim process. + */ + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count++; + + return (rc); + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * Initialize the tracepoint that's been preallocated with the probe. + */ + new_tp = probe->ftp_tps[index].fit_tp; + + ASSERT(new_tp->ftt_pid == pid); + ASSERT(new_tp->ftt_pc == pc); + ASSERT(new_tp->ftt_proc == probe->ftp_prov->ftp_proc); + ASSERT(new_tp->ftt_ids == NULL); + ASSERT(new_tp->ftt_retids == NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + id->fti_next = NULL; + new_tp->ftt_ids = id; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + id->fti_next = NULL; + new_tp->ftt_retids = id; + break; + + default: + ASSERT(0); + } + + /* + * If the ISA-dependent initialization goes to plan, go back to the + * beginning and try to install this freshly made tracepoint. + */ + if (fasttrap_tracepoint_init(p, new_tp, pc, id->fti_ptype) == 0) + goto again; + + new_tp->ftt_ids = NULL; + new_tp->ftt_retids = NULL; + + return (FASTTRAP_ENABLE_FAIL); +} + +static void +fasttrap_tracepoint_disable(proc_t *p, fasttrap_probe_t *probe, uint_t index) +{ + fasttrap_bucket_t *bucket; + fasttrap_provider_t *provider = probe->ftp_prov; + fasttrap_tracepoint_t **pp, *tp; + fasttrap_id_t *id, **idp; + pid_t pid; + uintptr_t pc; + + ASSERT(index < probe->ftp_ntps); + + pid = probe->ftp_pid; + pc = probe->ftp_tps[index].fit_tp->ftt_pc; + id = &probe->ftp_tps[index].fit_id; + + ASSERT(probe->ftp_tps[index].fit_tp->ftt_pid == pid); + + /* + * Find the tracepoint and make sure that our id is one of the + * ones registered with it. + */ + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + mutex_enter(&bucket->ftb_mtx); + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (tp->ftt_pid == pid && tp->ftt_pc == pc && + tp->ftt_proc == provider->ftp_proc) + break; + } + + /* + * If we somehow lost this tracepoint, we're in a world of hurt. + */ + ASSERT(tp != NULL); + + switch (id->fti_ptype) { + case DTFTP_ENTRY: + case DTFTP_OFFSETS: + case DTFTP_IS_ENABLED: + ASSERT(tp->ftt_ids != NULL); + idp = &tp->ftt_ids; + break; + + case DTFTP_RETURN: + case DTFTP_POST_OFFSETS: + ASSERT(tp->ftt_retids != NULL); + idp = &tp->ftt_retids; + break; + + default: + ASSERT(0); + } + + while ((*idp)->fti_probe != probe) { + idp = &(*idp)->fti_next; + ASSERT(*idp != NULL); + } + + id = *idp; + *idp = id->fti_next; + membar_producer(); + + ASSERT(id->fti_probe == probe); + + /* + * If there are other registered enablings of this tracepoint, we're + * all done, but if this was the last probe assocated with this + * this tracepoint, we need to remove and free it. + */ + if (tp->ftt_ids != NULL || tp->ftt_retids != NULL) { + + /* + * If the current probe's tracepoint is in use, swap it + * for an unused tracepoint. + */ + if (tp == probe->ftp_tps[index].fit_tp) { + fasttrap_probe_t *tmp_probe; + fasttrap_tracepoint_t **tmp_tp; + uint_t tmp_index; + + if (tp->ftt_ids != NULL) { + tmp_probe = tp->ftt_ids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_ids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } else { + tmp_probe = tp->ftt_retids->fti_probe; + /* LINTED - alignment */ + tmp_index = FASTTRAP_ID_INDEX(tp->ftt_retids); + tmp_tp = &tmp_probe->ftp_tps[tmp_index].fit_tp; + } + + ASSERT(*tmp_tp != NULL); + ASSERT(*tmp_tp != probe->ftp_tps[index].fit_tp); + ASSERT((*tmp_tp)->ftt_ids == NULL); + ASSERT((*tmp_tp)->ftt_retids == NULL); + + probe->ftp_tps[index].fit_tp = *tmp_tp; + *tmp_tp = tp; + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was + * changed. + */ + probe->ftp_gen = fasttrap_mod_gen; + return; + } + + mutex_exit(&bucket->ftb_mtx); + + /* + * We can't safely remove the tracepoint from the set of active + * tracepoints until we've actually removed the fasttrap instruction + * from the process's text. We can, however, operate on this + * tracepoint secure in the knowledge that no other thread is going to + * be looking at it since we hold P_PR_LOCK on the process if it's + * live or we hold the provider lock on the process if it's dead and + * gone. + */ + + /* + * We only need to remove the actual instruction if we're looking + * at an existing process + */ + if (p != NULL) { + /* + * If we fail to restore the instruction we need to kill + * this process since it's in a completely unrecoverable + * state. + */ + if (fasttrap_tracepoint_remove(p, tp) != 0) + fasttrap_sigtrap(p, NULL, pc); + + /* + * Decrement the count of the number of tracepoints active + * in the victim process. + */ + ASSERT(p->p_proc_flag & P_PR_LOCK); + p->p_dtrace_count--; + } + + /* + * Remove the probe from the hash table of active tracepoints. + */ + mutex_enter(&bucket->ftb_mtx); + pp = (fasttrap_tracepoint_t **)&bucket->ftb_data; + ASSERT(*pp != NULL); + while (*pp != tp) { + pp = &(*pp)->ftt_next; + ASSERT(*pp != NULL); + } + + *pp = tp->ftt_next; + membar_producer(); + + mutex_exit(&bucket->ftb_mtx); + + /* + * Tag the modified probe with the generation in which it was changed. + */ + probe->ftp_gen = fasttrap_mod_gen; +} + +static void +fasttrap_enable_callbacks(void) +{ + /* + * We don't have to play the rw lock game here because we're + * providing something rather than taking something away -- + * we can be sure that no threads have tried to follow this + * function pointer yet. + */ + mutex_enter(&fasttrap_count_mtx); + if (fasttrap_pid_count == 0) { + ASSERT(dtrace_pid_probe_ptr == NULL); + ASSERT(dtrace_return_probe_ptr == NULL); + dtrace_pid_probe_ptr = &fasttrap_pid_probe; + dtrace_return_probe_ptr = &fasttrap_return_probe; + } + ASSERT(dtrace_pid_probe_ptr == &fasttrap_pid_probe); + ASSERT(dtrace_return_probe_ptr == &fasttrap_return_probe); + fasttrap_pid_count++; + mutex_exit(&fasttrap_count_mtx); +} + +static void +fasttrap_disable_callbacks(void) +{ + ASSERT(MUTEX_HELD(&cpu_lock)); + + mutex_enter(&fasttrap_count_mtx); + ASSERT(fasttrap_pid_count > 0); + fasttrap_pid_count--; + if (fasttrap_pid_count == 0) { + cpu_t *cur, *cpu = CPU; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_enter(&cur->cpu_ft_lock, RW_WRITER); + } + + dtrace_pid_probe_ptr = NULL; + dtrace_return_probe_ptr = NULL; + + for (cur = cpu->cpu_next_onln; cur != cpu; + cur = cur->cpu_next_onln) { + rw_exit(&cur->cpu_ft_lock); + } + } + mutex_exit(&fasttrap_count_mtx); +} + +/*ARGSUSED*/ +static int +fasttrap_pid_enable(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + proc_t *p; + int i, rc; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(id == probe->ftp_id); + ASSERT(MUTEX_HELD(&cpu_lock)); + + /* + * Increment the count of enabled probes on this probe's provider; + * the provider can't go away while the probe still exists. We + * must increment this even if we aren't able to properly enable + * this probe. + */ + mutex_enter(&probe->ftp_prov->ftp_mtx); + probe->ftp_prov->ftp_rcount++; + mutex_exit(&probe->ftp_prov->ftp_mtx); + + /* + * If this probe's provider is retired (meaning it was valid in a + * previously exec'ed incarnation of this address space), bail out. The + * provider can't go away while we're in this code path. + */ + if (probe->ftp_prov->ftp_retired) + return (0); + + /* + * If we can't find the process, it may be that we're in the context of + * a fork in which the traced process is being born and we're copying + * USDT probes. Otherwise, the process is gone so bail. + */ + if ((p = sprlock(probe->ftp_pid)) == NULL) { + if ((curproc->p_flag & SFORKING) == 0) + return (0); + + mutex_enter(&pidlock); + p = prfind(probe->ftp_pid); + + /* + * Confirm that curproc is indeed forking the process in which + * we're trying to enable probes. + */ + ASSERT(p != NULL); + ASSERT(p->p_parent == curproc); + ASSERT(p->p_stat == SIDL); + + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + sprlock_proc(p); + } + + ASSERT(!(p->p_flag & SVFORK)); + mutex_exit(&p->p_lock); + + /* + * We have to enable the trap entry point before any user threads have + * the chance to execute the trap instruction we're about to place + * in their process's text. + */ + fasttrap_enable_callbacks(); + + /* + * Enable all the tracepoints and add this probe's id to each + * tracepoint's list of active probes. + */ + for (i = 0; i < probe->ftp_ntps; i++) { + if ((rc = fasttrap_tracepoint_enable(p, probe, i)) != 0) { + /* + * If enabling the tracepoint failed completely, + * we don't have to disable it; if the failure + * was only partial we must disable it. + */ + if (rc == FASTTRAP_ENABLE_FAIL) + i--; + else + ASSERT(rc == FASTTRAP_ENABLE_PARTIAL); + + /* + * Back up and pull out all the tracepoints we've + * created so far for this probe. + */ + while (i >= 0) { + fasttrap_tracepoint_disable(p, probe, i); + i--; + } + + mutex_enter(&p->p_lock); + sprunlock(p); + + /* + * Since we're not actually enabling this probe, + * drop our reference on the trap table entry. + */ + fasttrap_disable_callbacks(); + return (0); + } + } + + mutex_enter(&p->p_lock); + sprunlock(p); + + probe->ftp_enabled = 1; + return (0); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_disable(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + fasttrap_provider_t *provider = probe->ftp_prov; + proc_t *p; + int i, whack = 0; + + ASSERT(id == probe->ftp_id); + + /* + * We won't be able to acquire a /proc-esque lock on the process + * iff the process is dead and gone. In this case, we rely on the + * provider lock as a point of mutual exclusion to prevent other + * DTrace consumers from disabling this probe. + */ + if ((p = sprlock(probe->ftp_pid)) != NULL) { + ASSERT(!(p->p_flag & SVFORK)); + mutex_exit(&p->p_lock); + } + + mutex_enter(&provider->ftp_mtx); + + /* + * Disable all the associated tracepoints (for fully enabled probes). + */ + if (probe->ftp_enabled) { + for (i = 0; i < probe->ftp_ntps; i++) { + fasttrap_tracepoint_disable(p, probe, i); + } + } + + ASSERT(provider->ftp_rcount > 0); + provider->ftp_rcount--; + + if (p != NULL) { + /* + * Even though we may not be able to remove it entirely, we + * mark this retired provider to get a chance to remove some + * of the associated probes. + */ + if (provider->ftp_retired && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + + mutex_enter(&p->p_lock); + sprunlock(p); + } else { + /* + * If the process is dead, we're just waiting for the + * last probe to be disabled to be able to free it. + */ + if (provider->ftp_rcount == 0 && !provider->ftp_marked) + whack = provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + } + + if (whack) + fasttrap_pid_cleanup(); + + if (!probe->ftp_enabled) + return; + + probe->ftp_enabled = 0; + + ASSERT(MUTEX_HELD(&cpu_lock)); + fasttrap_disable_callbacks(); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_getargdesc(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc) +{ + fasttrap_probe_t *probe = parg; + char *str; + int i, ndx; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + if (probe->ftp_prov->ftp_retired != 0 || + desc->dtargd_ndx >= probe->ftp_nargs) { + desc->dtargd_ndx = DTRACE_ARGNONE; + return; + } + + ndx = (probe->ftp_argmap != NULL) ? + probe->ftp_argmap[desc->dtargd_ndx] : desc->dtargd_ndx; + + str = probe->ftp_ntypes; + for (i = 0; i < ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_native)); + (void) strcpy(desc->dtargd_native, str); + + if (probe->ftp_xtypes == NULL) + return; + + str = probe->ftp_xtypes; + for (i = 0; i < desc->dtargd_ndx; i++) { + str += strlen(str) + 1; + } + + ASSERT(strlen(str + 1) < sizeof (desc->dtargd_xlate)); + (void) strcpy(desc->dtargd_xlate, str); +} + +/*ARGSUSED*/ +static void +fasttrap_pid_destroy(void *arg, dtrace_id_t id, void *parg) +{ + fasttrap_probe_t *probe = parg; + int i; + size_t size; + + ASSERT(probe != NULL); + ASSERT(!probe->ftp_enabled); + ASSERT(fasttrap_total >= probe->ftp_ntps); + + atomic_add_32(&fasttrap_total, -probe->ftp_ntps); + size = offsetof(fasttrap_probe_t, ftp_tps[probe->ftp_ntps]); + + if (probe->ftp_gen + 1 >= fasttrap_mod_gen) + fasttrap_mod_barrier(probe->ftp_gen); + + for (i = 0; i < probe->ftp_ntps; i++) { + kmem_free(probe->ftp_tps[i].fit_tp, + sizeof (fasttrap_tracepoint_t)); + } + + kmem_free(probe, size); +} + + +static const dtrace_pattr_t pid_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +}; + +static dtrace_pops_t pid_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + fasttrap_pid_getarg, + NULL, + fasttrap_pid_destroy +}; + +static dtrace_pops_t usdt_pops = { + fasttrap_pid_provide, + NULL, + fasttrap_pid_enable, + fasttrap_pid_disable, + NULL, + NULL, + fasttrap_pid_getargdesc, + fasttrap_usdt_getarg, + NULL, + fasttrap_pid_destroy +}; + +static fasttrap_proc_t * +fasttrap_proc_lookup(pid_t pid) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, *new_fprc; + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mutex_enter(&bucket->ftb_mtx); + + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mutex_enter(&fprc->ftpc_mtx); + mutex_exit(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); + mutex_exit(&fprc->ftpc_mtx); + + return (fprc); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mutex_exit(&bucket->ftb_mtx); + + new_fprc = kmem_zalloc(sizeof (fasttrap_proc_t), KM_SLEEP); + new_fprc->ftpc_pid = pid; + new_fprc->ftpc_rcount = 1; + new_fprc->ftpc_acount = 1; + + mutex_enter(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a proc hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fprc = bucket->ftb_data; fprc != NULL; fprc = fprc->ftpc_next) { + if (fprc->ftpc_pid == pid && fprc->ftpc_acount != 0) { + mutex_enter(&fprc->ftpc_mtx); + mutex_exit(&bucket->ftb_mtx); + fprc->ftpc_rcount++; + atomic_add_64(&fprc->ftpc_acount, 1); + ASSERT(fprc->ftpc_acount <= fprc->ftpc_rcount); + mutex_exit(&fprc->ftpc_mtx); + + kmem_free(new_fprc, sizeof (fasttrap_proc_t)); + + return (fprc); + } + } + + new_fprc->ftpc_next = bucket->ftb_data; + bucket->ftb_data = new_fprc; + + mutex_exit(&bucket->ftb_mtx); + + return (new_fprc); +} + +static void +fasttrap_proc_release(fasttrap_proc_t *proc) +{ + fasttrap_bucket_t *bucket; + fasttrap_proc_t *fprc, **fprcp; + pid_t pid = proc->ftpc_pid; + + mutex_enter(&proc->ftpc_mtx); + + ASSERT(proc->ftpc_rcount != 0); + ASSERT(proc->ftpc_acount <= proc->ftpc_rcount); + + if (--proc->ftpc_rcount != 0) { + mutex_exit(&proc->ftpc_mtx); + return; + } + + mutex_exit(&proc->ftpc_mtx); + + /* + * There should definitely be no live providers associated with this + * process at this point. + */ + ASSERT(proc->ftpc_acount == 0); + + bucket = &fasttrap_procs.fth_table[FASTTRAP_PROCS_INDEX(pid)]; + mutex_enter(&bucket->ftb_mtx); + + fprcp = (fasttrap_proc_t **)&bucket->ftb_data; + while ((fprc = *fprcp) != NULL) { + if (fprc == proc) + break; + + fprcp = &fprc->ftpc_next; + } + + /* + * Something strange has happened if we can't find the proc. + */ + ASSERT(fprc != NULL); + + *fprcp = fprc->ftpc_next; + + mutex_exit(&bucket->ftb_mtx); + + kmem_free(fprc, sizeof (fasttrap_proc_t)); +} + +/* + * Lookup a fasttrap-managed provider based on its name and associated pid. + * If the pattr argument is non-NULL, this function instantiates the provider + * if it doesn't exist otherwise it returns NULL. The provider is returned + * with its lock held. + */ +static fasttrap_provider_t * +fasttrap_provider_lookup(pid_t pid, const char *name, + const dtrace_pattr_t *pattr) +{ + fasttrap_provider_t *fp, *new_fp = NULL; + fasttrap_bucket_t *bucket; + char provname[DTRACE_PROVNAMELEN]; + proc_t *p; + cred_t *cred; + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + ASSERT(pattr != NULL); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mutex_enter(&bucket->ftb_mtx); + + /* + * Take a lap through the list and return the match if we find it. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mutex_enter(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + return (fp); + } + } + + /* + * Drop the bucket lock so we don't try to perform a sleeping + * allocation under it. + */ + mutex_exit(&bucket->ftb_mtx); + + /* + * Make sure the process exists, isn't a child created as the result + * of a vfork(2), and isn't a zombie (but may be in fork). + */ + mutex_enter(&pidlock); + if ((p = prfind(pid)) == NULL) { + mutex_exit(&pidlock); + return (NULL); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + if (p->p_flag & (SVFORK | SEXITING)) { + mutex_exit(&p->p_lock); + return (NULL); + } + + /* + * Increment p_dtrace_probes so that the process knows to inform us + * when it exits or execs. fasttrap_provider_free() decrements this + * when we're done with this provider. + */ + p->p_dtrace_probes++; + + /* + * Grab the credentials for this process so we have + * something to pass to dtrace_register(). + */ + mutex_enter(&p->p_crlock); + crhold(p->p_cred); + cred = p->p_cred; + mutex_exit(&p->p_crlock); + mutex_exit(&p->p_lock); + + new_fp = kmem_zalloc(sizeof (fasttrap_provider_t), KM_SLEEP); + new_fp->ftp_pid = pid; + new_fp->ftp_proc = fasttrap_proc_lookup(pid); + + ASSERT(new_fp->ftp_proc != NULL); + + mutex_enter(&bucket->ftb_mtx); + + /* + * Take another lap through the list to make sure a provider hasn't + * been created for this pid while we weren't under the bucket lock. + */ + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) { + mutex_enter(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); + crfree(cred); + return (fp); + } + } + + (void) strcpy(new_fp->ftp_name, name); + + /* + * Fail and return NULL if either the provider name is too long + * or we fail to register this new provider with the DTrace + * framework. Note that this is the only place we ever construct + * the full provider name -- we keep it in pieces in the provider + * structure. + */ + if (snprintf(provname, sizeof (provname), "%s%u", name, (uint_t)pid) >= + sizeof (provname) || + dtrace_register(provname, pattr, + DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER, cred, + pattr == &pid_attr ? &pid_pops : &usdt_pops, new_fp, + &new_fp->ftp_provid) != 0) { + mutex_exit(&bucket->ftb_mtx); + fasttrap_provider_free(new_fp); + crfree(cred); + return (NULL); + } + + new_fp->ftp_next = bucket->ftb_data; + bucket->ftb_data = new_fp; + + mutex_enter(&new_fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + + crfree(cred); + return (new_fp); +} + +static void +fasttrap_provider_free(fasttrap_provider_t *provider) +{ + pid_t pid = provider->ftp_pid; + proc_t *p; + + /* + * There need to be no associated enabled probes, no consumers + * creating probes, and no meta providers referencing this provider. + */ + ASSERT(provider->ftp_rcount == 0); + ASSERT(provider->ftp_ccount == 0); + ASSERT(provider->ftp_mcount == 0); + + /* + * If this provider hasn't been retired, we need to explicitly drop the + * count of active providers on the associated process structure. + */ + if (!provider->ftp_retired) { + atomic_add_64(&provider->ftp_proc->ftpc_acount, -1); + ASSERT(provider->ftp_proc->ftpc_acount < + provider->ftp_proc->ftpc_rcount); + } + + fasttrap_proc_release(provider->ftp_proc); + + kmem_free(provider, sizeof (fasttrap_provider_t)); + + /* + * Decrement p_dtrace_probes on the process whose provider we're + * freeing. We don't have to worry about clobbering somone else's + * modifications to it because we have locked the bucket that + * corresponds to this process's hash chain in the provider hash + * table. Don't sweat it if we can't find the process. + */ + mutex_enter(&pidlock); + if ((p = prfind(pid)) == NULL) { + mutex_exit(&pidlock); + return; + } + + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + p->p_dtrace_probes--; + mutex_exit(&p->p_lock); +} + +static void +fasttrap_provider_retire(pid_t pid, const char *name, int mprov) +{ + fasttrap_provider_t *fp; + fasttrap_bucket_t *bucket; + dtrace_provider_id_t provid; + + ASSERT(strlen(name) < sizeof (fp->ftp_name)); + + bucket = &fasttrap_provs.fth_table[FASTTRAP_PROVS_INDEX(pid, name)]; + mutex_enter(&bucket->ftb_mtx); + + for (fp = bucket->ftb_data; fp != NULL; fp = fp->ftp_next) { + if (fp->ftp_pid == pid && strcmp(fp->ftp_name, name) == 0 && + !fp->ftp_retired) + break; + } + + if (fp == NULL) { + mutex_exit(&bucket->ftb_mtx); + return; + } + + mutex_enter(&fp->ftp_mtx); + ASSERT(!mprov || fp->ftp_mcount > 0); + if (mprov && --fp->ftp_mcount != 0) { + mutex_exit(&fp->ftp_mtx); + mutex_exit(&bucket->ftb_mtx); + return; + } + + /* + * Mark the provider to be removed in our post-processing step, mark it + * retired, and drop the active count on its proc. Marking it indicates + * that we should try to remove it; setting the retired flag indicates + * that we're done with this provider; dropping the active the proc + * releases our hold, and when this reaches zero (as it will during + * exit or exec) the proc and associated providers become defunct. + * + * We obviously need to take the bucket lock before the provider lock + * to perform the lookup, but we need to drop the provider lock + * before calling into the DTrace framework since we acquire the + * provider lock in callbacks invoked from the DTrace framework. The + * bucket lock therefore protects the integrity of the provider hash + * table. + */ + atomic_add_64(&fp->ftp_proc->ftpc_acount, -1); + ASSERT(fp->ftp_proc->ftpc_acount < fp->ftp_proc->ftpc_rcount); + + fp->ftp_retired = 1; + fp->ftp_marked = 1; + provid = fp->ftp_provid; + mutex_exit(&fp->ftp_mtx); + + /* + * We don't have to worry about invalidating the same provider twice + * since fasttrap_provider_lookup() will ignore provider that have + * been marked as retired. + */ + dtrace_invalidate(provid); + + mutex_exit(&bucket->ftb_mtx); + + fasttrap_pid_cleanup(); +} + +static int +fasttrap_uint32_cmp(const void *ap, const void *bp) +{ + return (*(const uint32_t *)ap - *(const uint32_t *)bp); +} + +static int +fasttrap_uint64_cmp(const void *ap, const void *bp) +{ + return (*(const uint64_t *)ap - *(const uint64_t *)bp); +} + +static int +fasttrap_add_probe(fasttrap_probe_spec_t *pdata) +{ + fasttrap_provider_t *provider; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + char *name; + int i, aframes, whack; + + /* + * There needs to be at least one desired trace point. + */ + if (pdata->ftps_noffs == 0) + return (EINVAL); + + switch (pdata->ftps_type) { + case DTFTP_ENTRY: + name = "entry"; + aframes = FASTTRAP_ENTRY_AFRAMES; + break; + case DTFTP_RETURN: + name = "return"; + aframes = FASTTRAP_RETURN_AFRAMES; + break; + case DTFTP_OFFSETS: + name = NULL; + break; + default: + return (EINVAL); + } + + if ((provider = fasttrap_provider_lookup(pdata->ftps_pid, + FASTTRAP_PID_NAME, &pid_attr)) == NULL) + return (ESRCH); + + /* + * Increment this reference count to indicate that a consumer is + * actively adding a new probe associated with this provider. This + * prevents the provider from being deleted -- we'll need to check + * for pending deletions when we drop this reference count. + */ + provider->ftp_ccount++; + mutex_exit(&provider->ftp_mtx); + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. We must drop the provider lock + * before taking this lock to avoid a three-way deadlock with the + * DTrace framework. + */ + mutex_enter(&provider->ftp_cmtx); + + if (name == NULL) { + for (i = 0; i < pdata->ftps_noffs; i++) { + char name_str[17]; + + (void) sprintf(name_str, "%llx", + (unsigned long long)pdata->ftps_offs[i]); + + if (dtrace_probe_lookup(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str) != 0) + continue; + + atomic_add_32(&fasttrap_total, 1); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -1); + goto no_mem; + } + + pp = kmem_zalloc(sizeof (fasttrap_probe_t), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = 1; + + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[0].fit_tp = tp; + pp->ftp_tps[0].fit_id.fti_probe = pp; + pp->ftp_tps[0].fit_id.fti_ptype = pdata->ftps_type; + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name_str, + FASTTRAP_OFFSET_AFRAMES, pp); + } + + } else if (dtrace_probe_lookup(provider->ftp_provid, pdata->ftps_mod, + pdata->ftps_func, name) == 0) { + atomic_add_32(&fasttrap_total, pdata->ftps_noffs); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + /* + * Make sure all tracepoint program counter values are unique. + * We later assume that each probe has exactly one tracepoint + * for a given pc. + */ + qsort(pdata->ftps_offs, pdata->ftps_noffs, + sizeof (uint64_t), fasttrap_uint64_cmp); + for (i = 1; i < pdata->ftps_noffs; i++) { + if (pdata->ftps_offs[i] > pdata->ftps_offs[i - 1]) + continue; + + atomic_add_32(&fasttrap_total, -pdata->ftps_noffs); + goto no_mem; + } + + ASSERT(pdata->ftps_noffs > 0); + pp = kmem_zalloc(offsetof(fasttrap_probe_t, + ftp_tps[pdata->ftps_noffs]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_faddr = pdata->ftps_pc; + pp->ftp_fsize = pdata->ftps_size; + pp->ftp_pid = pdata->ftps_pid; + pp->ftp_ntps = pdata->ftps_noffs; + + for (i = 0; i < pdata->ftps_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), + KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = pdata->ftps_offs[i] + pdata->ftps_pc; + tp->ftt_pid = pdata->ftps_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = pdata->ftps_type; + } + + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, + pdata->ftps_mod, pdata->ftps_func, name, aframes, pp); + } + + mutex_exit(&provider->ftp_cmtx); + + /* + * We know that the provider is still valid since we incremented the + * creation reference count. If someone tried to clean up this provider + * while we were using it (e.g. because the process called exec(2) or + * exit(2)), take note of that and try to clean it up now. + */ + mutex_enter(&provider->ftp_mtx); + provider->ftp_ccount--; + whack = provider->ftp_retired; + mutex_exit(&provider->ftp_mtx); + + if (whack) + fasttrap_pid_cleanup(); + + return (0); + +no_mem: + /* + * If we've exhausted the allowable resources, we'll try to remove + * this provider to free some up. This is to cover the case where + * the user has accidentally created many more probes than was + * intended (e.g. pid123:::). + */ + mutex_exit(&provider->ftp_cmtx); + mutex_enter(&provider->ftp_mtx); + provider->ftp_ccount--; + provider->ftp_marked = 1; + mutex_exit(&provider->ftp_mtx); + + fasttrap_pid_cleanup(); + + return (ENOMEM); +} + +/*ARGSUSED*/ +static void * +fasttrap_meta_provide(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ + fasttrap_provider_t *provider; + + /* + * A 32-bit unsigned integer (like a pid for example) can be + * expressed in 10 or fewer decimal digits. Make sure that we'll + * have enough space for the provider name. + */ + if (strlen(dhpv->dthpv_provname) + 10 >= + sizeof (provider->ftp_name)) { + cmn_err(CE_WARN, "failed to instantiate provider %s: " + "name too long to accomodate pid", dhpv->dthpv_provname); + return (NULL); + } + + /* + * Don't let folks spoof the true pid provider. + */ + if (strcmp(dhpv->dthpv_provname, FASTTRAP_PID_NAME) == 0) { + cmn_err(CE_WARN, "failed to instantiate provider %s: " + "%s is an invalid name", dhpv->dthpv_provname, + FASTTRAP_PID_NAME); + return (NULL); + } + + /* + * The highest stability class that fasttrap supports is ISA; cap + * the stability of the new provider accordingly. + */ + if (dhpv->dthpv_pattr.dtpa_provider.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_provider.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_mod.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_mod.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_func.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_func.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_name.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_name.dtat_class = DTRACE_CLASS_ISA; + if (dhpv->dthpv_pattr.dtpa_args.dtat_class > DTRACE_CLASS_ISA) + dhpv->dthpv_pattr.dtpa_args.dtat_class = DTRACE_CLASS_ISA; + + if ((provider = fasttrap_provider_lookup(pid, dhpv->dthpv_provname, + &dhpv->dthpv_pattr)) == NULL) { + cmn_err(CE_WARN, "failed to instantiate provider %s for " + "process %u", dhpv->dthpv_provname, (uint_t)pid); + return (NULL); + } + + /* + * Up the meta provider count so this provider isn't removed until + * the meta provider has been told to remove it. + */ + provider->ftp_mcount++; + + mutex_exit(&provider->ftp_mtx); + + return (provider); +} + +/*ARGSUSED*/ +static void +fasttrap_meta_create_probe(void *arg, void *parg, + dtrace_helper_probedesc_t *dhpb) +{ + fasttrap_provider_t *provider = parg; + fasttrap_probe_t *pp; + fasttrap_tracepoint_t *tp; + int i, j; + uint32_t ntps; + + /* + * Since the meta provider count is non-zero we don't have to worry + * about this provider disappearing. + */ + ASSERT(provider->ftp_mcount > 0); + + /* + * The offsets must be unique. + */ + qsort(dhpb->dthpb_offs, dhpb->dthpb_noffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_noffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_offs[i] <= + dhpb->dthpb_base + dhpb->dthpb_offs[i - 1]) + return; + } + + qsort(dhpb->dthpb_enoffs, dhpb->dthpb_nenoffs, sizeof (uint32_t), + fasttrap_uint32_cmp); + for (i = 1; i < dhpb->dthpb_nenoffs; i++) { + if (dhpb->dthpb_base + dhpb->dthpb_enoffs[i] <= + dhpb->dthpb_base + dhpb->dthpb_enoffs[i - 1]) + return; + } + + /* + * Grab the creation lock to ensure consistency between calls to + * dtrace_probe_lookup() and dtrace_probe_create() in the face of + * other threads creating probes. + */ + mutex_enter(&provider->ftp_cmtx); + + if (dtrace_probe_lookup(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name) != 0) { + mutex_exit(&provider->ftp_cmtx); + return; + } + + ntps = dhpb->dthpb_noffs + dhpb->dthpb_nenoffs; + ASSERT(ntps > 0); + + atomic_add_32(&fasttrap_total, ntps); + + if (fasttrap_total > fasttrap_max) { + atomic_add_32(&fasttrap_total, -ntps); + mutex_exit(&provider->ftp_cmtx); + return; + } + + pp = kmem_zalloc(offsetof(fasttrap_probe_t, ftp_tps[ntps]), KM_SLEEP); + + pp->ftp_prov = provider; + pp->ftp_pid = provider->ftp_pid; + pp->ftp_ntps = ntps; + pp->ftp_nargs = dhpb->dthpb_xargc; + pp->ftp_xtypes = dhpb->dthpb_xtypes; + pp->ftp_ntypes = dhpb->dthpb_ntypes; + + /* + * First create a tracepoint for each actual point of interest. + */ + for (i = 0; i < dhpb->dthpb_noffs; i++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_offs[i]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; +#ifdef __sparc + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_POST_OFFSETS; +#else + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_OFFSETS; +#endif + } + + /* + * Then create a tracepoint for each is-enabled point. + */ + for (j = 0; i < ntps; i++, j++) { + tp = kmem_zalloc(sizeof (fasttrap_tracepoint_t), KM_SLEEP); + + tp->ftt_proc = provider->ftp_proc; + tp->ftt_pc = dhpb->dthpb_base + dhpb->dthpb_enoffs[j]; + tp->ftt_pid = provider->ftp_pid; + + pp->ftp_tps[i].fit_tp = tp; + pp->ftp_tps[i].fit_id.fti_probe = pp; + pp->ftp_tps[i].fit_id.fti_ptype = DTFTP_IS_ENABLED; + } + + /* + * If the arguments are shuffled around we set the argument remapping + * table. Later, when the probe fires, we only remap the arguments + * if the table is non-NULL. + */ + for (i = 0; i < dhpb->dthpb_xargc; i++) { + if (dhpb->dthpb_args[i] != i) { + pp->ftp_argmap = dhpb->dthpb_args; + break; + } + } + + /* + * The probe is fully constructed -- register it with DTrace. + */ + pp->ftp_id = dtrace_probe_create(provider->ftp_provid, dhpb->dthpb_mod, + dhpb->dthpb_func, dhpb->dthpb_name, FASTTRAP_OFFSET_AFRAMES, pp); + + mutex_exit(&provider->ftp_cmtx); +} + +/*ARGSUSED*/ +static void +fasttrap_meta_remove(void *arg, dtrace_helper_provdesc_t *dhpv, pid_t pid) +{ + /* + * Clean up the USDT provider. There may be active consumers of the + * provider busy adding probes, no damage will actually befall the + * provider until that count has dropped to zero. This just puts + * the provider on death row. + */ + fasttrap_provider_retire(pid, dhpv->dthpv_provname, 1); +} + +static dtrace_mops_t fasttrap_mops = { + fasttrap_meta_create_probe, + fasttrap_meta_provide, + fasttrap_meta_remove +}; + +/*ARGSUSED*/ +static int +fasttrap_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +/*ARGSUSED*/ +static int +fasttrap_ioctl(dev_t dev, int cmd, intptr_t arg, int md, cred_t *cr, int *rv) +{ + if (!dtrace_attached()) + return (EAGAIN); + + if (cmd == FASTTRAPIOC_MAKEPROBE) { + fasttrap_probe_spec_t *uprobe = (void *)arg; + fasttrap_probe_spec_t *probe; + uint64_t noffs; + size_t size; + int ret; + char *c; + + if (copyin(&uprobe->ftps_noffs, &noffs, + sizeof (uprobe->ftps_noffs))) + return (EFAULT); + + /* + * Probes must have at least one tracepoint. + */ + if (noffs == 0) + return (EINVAL); + + size = sizeof (fasttrap_probe_spec_t) + + sizeof (probe->ftps_offs[0]) * (noffs - 1); + + if (size > 1024 * 1024) + return (ENOMEM); + + probe = kmem_alloc(size, KM_SLEEP); + + if (copyin(uprobe, probe, size) != 0 || + probe->ftps_noffs != noffs) { + kmem_free(probe, size); + return (EFAULT); + } + + /* + * Verify that the function and module strings contain no + * funny characters. + */ + for (c = &probe->ftps_func[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + for (c = &probe->ftps_mod[0]; *c != '\0'; c++) { + if (*c < 0x20 || 0x7f <= *c) { + ret = EINVAL; + goto err; + } + } + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = probe->ftps_pid; + + mutex_enter(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mutex_exit(&pidlock); + return (ESRCH); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD | VWRITE)) != 0) { + mutex_exit(&p->p_lock); + return (ret); + } + + mutex_exit(&p->p_lock); + } + + ret = fasttrap_add_probe(probe); +err: + kmem_free(probe, size); + + return (ret); + + } else if (cmd == FASTTRAPIOC_GETINSTR) { + fasttrap_instr_query_t instr; + fasttrap_tracepoint_t *tp; + uint_t index; + int ret; + + if (copyin((void *)arg, &instr, sizeof (instr)) != 0) + return (EFAULT); + + if (!PRIV_POLICY_CHOICE(cr, PRIV_ALL, B_FALSE)) { + proc_t *p; + pid_t pid = instr.ftiq_pid; + + mutex_enter(&pidlock); + /* + * Report an error if the process doesn't exist + * or is actively being birthed. + */ + if ((p = prfind(pid)) == NULL || p->p_stat == SIDL) { + mutex_exit(&pidlock); + return (ESRCH); + } + mutex_enter(&p->p_lock); + mutex_exit(&pidlock); + + if ((ret = priv_proc_cred_perm(cr, p, NULL, + VREAD)) != 0) { + mutex_exit(&p->p_lock); + return (ret); + } + + mutex_exit(&p->p_lock); + } + + index = FASTTRAP_TPOINTS_INDEX(instr.ftiq_pid, instr.ftiq_pc); + + mutex_enter(&fasttrap_tpoints.fth_table[index].ftb_mtx); + tp = fasttrap_tpoints.fth_table[index].ftb_data; + while (tp != NULL) { + if (instr.ftiq_pid == tp->ftt_pid && + instr.ftiq_pc == tp->ftt_pc && + tp->ftt_proc->ftpc_acount != 0) + break; + + tp = tp->ftt_next; + } + + if (tp == NULL) { + mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); + return (ENOENT); + } + + bcopy(&tp->ftt_instr, &instr.ftiq_instr, + sizeof (instr.ftiq_instr)); + mutex_exit(&fasttrap_tpoints.fth_table[index].ftb_mtx); + + if (copyout(&instr, (void *)arg, sizeof (instr)) != 0) + return (EFAULT); + + return (0); + } + + return (EINVAL); +} + +static struct cb_ops fasttrap_cb_ops = { + fasttrap_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + fasttrap_ioctl, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +/*ARGSUSED*/ +static int +fasttrap_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)fasttrap_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +static int +fasttrap_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + ulong_t nent; + + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "fasttrap", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + fasttrap_devi = devi; + + /* + * Install our hooks into fork(2), exec(2), and exit(2). + */ + dtrace_fasttrap_fork_ptr = &fasttrap_fork; + dtrace_fasttrap_exit_ptr = &fasttrap_exec_exit; + dtrace_fasttrap_exec_ptr = &fasttrap_exec_exit; + + fasttrap_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "fasttrap-max-probes", FASTTRAP_MAX_DEFAULT); + fasttrap_total = 0; + + /* + * Conjure up the tracepoints hashtable... + */ + nent = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "fasttrap-hash-size", FASTTRAP_TPOINTS_DEFAULT_SIZE); + + if (nent == 0 || nent > 0x1000000) + nent = FASTTRAP_TPOINTS_DEFAULT_SIZE; + + if ((nent & (nent - 1)) == 0) + fasttrap_tpoints.fth_nent = nent; + else + fasttrap_tpoints.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_tpoints.fth_nent > 0); + fasttrap_tpoints.fth_mask = fasttrap_tpoints.fth_nent - 1; + fasttrap_tpoints.fth_table = kmem_zalloc(fasttrap_tpoints.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + /* + * ... and the providers hash table... + */ + nent = FASTTRAP_PROVIDERS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_provs.fth_nent = nent; + else + fasttrap_provs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_provs.fth_nent > 0); + fasttrap_provs.fth_mask = fasttrap_provs.fth_nent - 1; + fasttrap_provs.fth_table = kmem_zalloc(fasttrap_provs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + /* + * ... and the procs hash table. + */ + nent = FASTTRAP_PROCS_DEFAULT_SIZE; + if ((nent & (nent - 1)) == 0) + fasttrap_procs.fth_nent = nent; + else + fasttrap_procs.fth_nent = 1 << fasttrap_highbit(nent); + ASSERT(fasttrap_procs.fth_nent > 0); + fasttrap_procs.fth_mask = fasttrap_procs.fth_nent - 1; + fasttrap_procs.fth_table = kmem_zalloc(fasttrap_procs.fth_nent * + sizeof (fasttrap_bucket_t), KM_SLEEP); + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); + + return (DDI_SUCCESS); +} + +static int +fasttrap_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + int i, fail = 0; + timeout_id_t tmp; + + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + /* + * Unregister the meta-provider to make sure no new fasttrap- + * managed providers come along while we're trying to close up + * shop. If we fail to detach, we'll need to re-register as a + * meta-provider. We can fail to unregister as a meta-provider + * if providers we manage still exist. + */ + if (fasttrap_meta_id != DTRACE_METAPROVNONE && + dtrace_meta_unregister(fasttrap_meta_id) != 0) + return (DDI_FAILURE); + + /* + * Prevent any new timeouts from running by setting fasttrap_timeout + * to a non-zero value, and wait for the current timeout to complete. + */ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_cleanup_work = 0; + + while (fasttrap_timeout != (timeout_id_t)1) { + tmp = fasttrap_timeout; + fasttrap_timeout = (timeout_id_t)1; + + if (tmp != 0) { + mutex_exit(&fasttrap_cleanup_mtx); + (void) untimeout(tmp); + mutex_enter(&fasttrap_cleanup_mtx); + } + } + + fasttrap_cleanup_work = 0; + mutex_exit(&fasttrap_cleanup_mtx); + + /* + * Iterate over all of our providers. If there's still a process + * that corresponds to that pid, fail to detach. + */ + for (i = 0; i < fasttrap_provs.fth_nent; i++) { + fasttrap_provider_t **fpp, *fp; + fasttrap_bucket_t *bucket = &fasttrap_provs.fth_table[i]; + + mutex_enter(&bucket->ftb_mtx); + fpp = (fasttrap_provider_t **)&bucket->ftb_data; + while ((fp = *fpp) != NULL) { + /* + * Acquire and release the lock as a simple way of + * waiting for any other consumer to finish with + * this provider. A thread must first acquire the + * bucket lock so there's no chance of another thread + * blocking on the provider's lock. + */ + mutex_enter(&fp->ftp_mtx); + mutex_exit(&fp->ftp_mtx); + + if (dtrace_unregister(fp->ftp_provid) != 0) { + fail = 1; + fpp = &fp->ftp_next; + } else { + *fpp = fp->ftp_next; + fasttrap_provider_free(fp); + } + } + + mutex_exit(&bucket->ftb_mtx); + } + + if (fail) { + uint_t work; + /* + * If we're failing to detach, we need to unblock timeouts + * and start a new timeout if any work has accumulated while + * we've been unsuccessfully trying to detach. + */ + mutex_enter(&fasttrap_cleanup_mtx); + fasttrap_timeout = 0; + work = fasttrap_cleanup_work; + mutex_exit(&fasttrap_cleanup_mtx); + + if (work) + fasttrap_pid_cleanup(); + + (void) dtrace_meta_register("fasttrap", &fasttrap_mops, NULL, + &fasttrap_meta_id); + + return (DDI_FAILURE); + } + +#ifdef DEBUG + mutex_enter(&fasttrap_count_mtx); + ASSERT(fasttrap_pid_count == 0); + mutex_exit(&fasttrap_count_mtx); +#endif + + kmem_free(fasttrap_tpoints.fth_table, + fasttrap_tpoints.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_tpoints.fth_nent = 0; + + kmem_free(fasttrap_provs.fth_table, + fasttrap_provs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_provs.fth_nent = 0; + + kmem_free(fasttrap_procs.fth_table, + fasttrap_procs.fth_nent * sizeof (fasttrap_bucket_t)); + fasttrap_procs.fth_nent = 0; + + /* + * We know there are no tracepoints in any process anywhere in + * the system so there is no process which has its p_dtrace_count + * greater than zero, therefore we know that no thread can actively + * be executing code in fasttrap_fork(). Similarly for p_dtrace_probes + * and fasttrap_exec() and fasttrap_exit(). + */ + ASSERT(dtrace_fasttrap_fork_ptr == &fasttrap_fork); + dtrace_fasttrap_fork_ptr = NULL; + + ASSERT(dtrace_fasttrap_exec_ptr == &fasttrap_exec_exit); + dtrace_fasttrap_exec_ptr = NULL; + + ASSERT(dtrace_fasttrap_exit_ptr == &fasttrap_exec_exit); + dtrace_fasttrap_exit_ptr = NULL; + + ddi_remove_minor_node(devi, NULL); + + return (DDI_SUCCESS); +} + +static struct dev_ops fasttrap_ops = { + DEVO_REV, /* devo_rev */ + 0, /* refcnt */ + fasttrap_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + fasttrap_attach, /* attach */ + fasttrap_detach, /* detach */ + nodev, /* reset */ + &fasttrap_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Fasttrap Tracing", /* name of module */ + &fasttrap_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.c new file mode 100644 index 00000000..69c8b725 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.c @@ -0,0 +1,343 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/types.h> +#include <sys/param.h> +#include <sys/stat.h> +#include <sys/open.h> +#include <sys/file.h> +#include <sys/conf.h> +#include <sys/modctl.h> +#include <sys/cmn_err.h> +#include <sys/bitmap.h> +#include <sys/debug.h> +#include <sys/kmem.h> +#include <sys/errno.h> +#include <sys/sysmacros.h> +#include <sys/lockstat.h> +#include <sys/atomic.h> +#include <sys/dtrace.h> + +#include <sys/ddi.h> +#include <sys/sunddi.h> + +typedef struct lockstat_probe { + const char *lsp_func; + const char *lsp_name; + int lsp_probe; + dtrace_id_t lsp_id; +} lockstat_probe_t; + +lockstat_probe_t lockstat_probes[] = +{ + { LS_MUTEX_ENTER, LSA_ACQUIRE, LS_MUTEX_ENTER_ACQUIRE }, + { LS_MUTEX_ENTER, LSA_BLOCK, LS_MUTEX_ENTER_BLOCK }, + { LS_MUTEX_ENTER, LSA_SPIN, LS_MUTEX_ENTER_SPIN }, + { LS_MUTEX_EXIT, LSA_RELEASE, LS_MUTEX_EXIT_RELEASE }, + { LS_MUTEX_DESTROY, LSA_RELEASE, LS_MUTEX_DESTROY_RELEASE }, + { LS_MUTEX_TRYENTER, LSA_ACQUIRE, LS_MUTEX_TRYENTER_ACQUIRE }, + { LS_LOCK_SET, LSS_ACQUIRE, LS_LOCK_SET_ACQUIRE }, + { LS_LOCK_SET, LSS_SPIN, LS_LOCK_SET_SPIN }, + { LS_LOCK_SET_SPL, LSS_ACQUIRE, LS_LOCK_SET_SPL_ACQUIRE }, + { LS_LOCK_SET_SPL, LSS_SPIN, LS_LOCK_SET_SPL_SPIN }, + { LS_LOCK_TRY, LSS_ACQUIRE, LS_LOCK_TRY_ACQUIRE }, + { LS_LOCK_CLEAR, LSS_RELEASE, LS_LOCK_CLEAR_RELEASE }, + { LS_LOCK_CLEAR_SPLX, LSS_RELEASE, LS_LOCK_CLEAR_SPLX_RELEASE }, + { LS_CLOCK_UNLOCK, LSS_RELEASE, LS_CLOCK_UNLOCK_RELEASE }, + { LS_RW_ENTER, LSR_ACQUIRE, LS_RW_ENTER_ACQUIRE }, + { LS_RW_ENTER, LSR_BLOCK, LS_RW_ENTER_BLOCK }, + { LS_RW_EXIT, LSR_RELEASE, LS_RW_EXIT_RELEASE }, + { LS_RW_TRYENTER, LSR_ACQUIRE, LS_RW_TRYENTER_ACQUIRE }, + { LS_RW_TRYUPGRADE, LSR_UPGRADE, LS_RW_TRYUPGRADE_UPGRADE }, + { LS_RW_DOWNGRADE, LSR_DOWNGRADE, LS_RW_DOWNGRADE_DOWNGRADE }, + { LS_THREAD_LOCK, LST_SPIN, LS_THREAD_LOCK_SPIN }, + { LS_THREAD_LOCK_HIGH, LST_SPIN, LS_THREAD_LOCK_HIGH_SPIN }, + { NULL } +}; + +static dev_info_t *lockstat_devi; /* saved in xxattach() for xxinfo() */ +static kmutex_t lockstat_test; /* for testing purposes only */ +static dtrace_provider_id_t lockstat_id; + +/*ARGSUSED*/ +static int +lockstat_enable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = id; + membar_producer(); + + lockstat_hot_patch(); + membar_producer(); + + /* + * Immediately generate a record for the lockstat_test mutex + * to verify that the mutex hot-patch code worked as expected. + */ + mutex_enter(&lockstat_test); + mutex_exit(&lockstat_test); + return (0); +} + +/*ARGSUSED*/ +static void +lockstat_disable(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + int i; + + ASSERT(lockstat_probemap[probe->lsp_probe]); + + lockstat_probemap[probe->lsp_probe] = 0; + lockstat_hot_patch(); + membar_producer(); + + /* + * See if we have any probes left enabled. + */ + for (i = 0; i < LS_NPROBES; i++) { + if (lockstat_probemap[i]) { + /* + * This probe is still enabled. We don't need to deal + * with waiting for all threads to be out of the + * lockstat critical sections; just return. + */ + return; + } + } + + /* + * The delay() here isn't as cheesy as you might think. We don't + * want to busy-loop in the kernel, so we have to give up the + * CPU between calls to lockstat_active_threads(); that much is + * obvious. But the reason it's a do..while loop rather than a + * while loop is subtle. The memory barrier above guarantees that + * no threads will enter the lockstat code from this point forward. + * However, another thread could already be executing lockstat code + * without our knowledge if the update to its t_lockstat field hasn't + * cleared its CPU's store buffer. Delaying for one clock tick + * guarantees that either (1) the thread will have *ample* time to + * complete its work, or (2) the thread will be preempted, in which + * case it will have to grab and release a dispatcher lock, which + * will flush that CPU's store buffer. Either way we're covered. + */ + do { + delay(1); + } while (lockstat_active_threads()); +} + +/*ARGSUSED*/ +static int +lockstat_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +/* ARGSUSED */ +static int +lockstat_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *) lockstat_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static void +lockstat_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i = 0; + + for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) { + lockstat_probe_t *probe = &lockstat_probes[i]; + + if (dtrace_probe_lookup(lockstat_id, "genunix", + probe->lsp_func, probe->lsp_name) != 0) + continue; + + ASSERT(!probe->lsp_id); + probe->lsp_id = dtrace_probe_create(lockstat_id, + "genunix", probe->lsp_func, probe->lsp_name, + 1, probe); + } +} + +/*ARGSUSED*/ +static void +lockstat_destroy(void *arg, dtrace_id_t id, void *parg) +{ + lockstat_probe_t *probe = parg; + + ASSERT(!lockstat_probemap[probe->lsp_probe]); + probe->lsp_id = 0; +} + +static dtrace_pattr_t lockstat_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t lockstat_pops = { + lockstat_provide, + NULL, + lockstat_enable, + lockstat_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + lockstat_destroy +}; + +static int +lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0, + DDI_PSEUDO, 0) == DDI_FAILURE || + dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL, + NULL, &lockstat_pops, NULL, &lockstat_id) != 0) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + lockstat_probe = dtrace_probe; + membar_producer(); + + ddi_report_dev(devi); + lockstat_devi = devi; + return (DDI_SUCCESS); +} + +static int +lockstat_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(lockstat_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + return (DDI_SUCCESS); +} + +/* + * Configuration data structures + */ +static struct cb_ops lockstat_cb_ops = { + lockstat_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_MP | D_NEW /* Driver compatibility flag */ +}; + +static struct dev_ops lockstat_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + lockstat_info, /* getinfo */ + nulldev, /* identify */ + nulldev, /* probe */ + lockstat_attach, /* attach */ + lockstat_detach, /* detach */ + nulldev, /* reset */ + &lockstat_cb_ops, /* cb_ops */ + NULL, /* bus_ops */ + NULL, /* power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +static struct modldrv modldrv = { + &mod_driverops, /* Type of module. This one is a driver */ + "Lock Statistics", /* name of module */ + &lockstat_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, (void *)&modldrv, NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.conf new file mode 100644 index 00000000..6d29b6a8 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/lockstat.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 1997-2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="lockstat" parent="pseudo" instance=0; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.c new file mode 100644 index 00000000..c1a2d1f1 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.c @@ -0,0 +1,577 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/errno.h> +#include <sys/stat.h> +#include <sys/modctl.h> +#include <sys/conf.h> +#include <sys/systm.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/cpuvar.h> +#include <sys/kmem.h> +#include <sys/strsubr.h> +#include <sys/dtrace.h> +#include <sys/cyclic.h> +#include <sys/atomic.h> + +static dev_info_t *profile_devi; +static dtrace_provider_id_t profile_id; + +/* + * Regardless of platform, the stack frames look like this in the case of the + * profile provider: + * + * profile_fire + * cyclic_expire + * cyclic_fire + * [ cbe ] + * [ interrupt code ] + * + * On x86, there are five frames from the generic interrupt code; further, the + * interrupted instruction appears as its own stack frame, giving us a total of + * 10. + * + * On SPARC, the picture is further complicated because the compiler + * optimizes away tail-calls -- so the following frames are optimized away: + * + * profile_fire + * cyclic_expire + * + * This gives three frames. However, on DEBUG kernels, the cyclic_expire + * frame cannot be tail-call eliminated, yielding four frames in this case. + * + * All of the above constraints lead to the mess below. Yes, the profile + * provider should ideally figure this out on-the-fly by hitting one of its own + * probes and then walking its own stack trace. This is complicated, however, + * and the static definition doesn't seem to be overly brittle. Still, we + * allow for a manual override in case we get it completely wrong. + */ +#ifdef __x86 +#define PROF_ARTIFICIAL_FRAMES 10 +#else +#ifdef __sparc +#ifdef DEBUG +#define PROF_ARTIFICIAL_FRAMES 4 +#else +#define PROF_ARTIFICIAL_FRAMES 3 +#endif +#endif +#endif + +#define PROF_NAMELEN 15 + +#define PROF_PROFILE 0 +#define PROF_TICK 1 +#define PROF_PREFIX_PROFILE "profile-" +#define PROF_PREFIX_TICK "tick-" + +typedef struct profile_probe { + char prof_name[PROF_NAMELEN]; + dtrace_id_t prof_id; + int prof_kind; + hrtime_t prof_interval; + cyclic_id_t prof_cyclic; +} profile_probe_t; + +typedef struct profile_probe_percpu { + hrtime_t profc_expected; + hrtime_t profc_interval; + profile_probe_t *profc_probe; +} profile_probe_percpu_t; + +hrtime_t profile_interval_min = NANOSEC / 5000; /* 5000 hz */ +int profile_aframes = 0; /* override */ + +static int profile_rates[] = { + 97, 199, 499, 997, 1999, + 4001, 4999, 0, 0, 0, + 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +static int profile_ticks[] = { + 1, 10, 100, 500, 1000, + 5000, 0, 0, 0, 0, + 0, 0, 0, 0, 0 +}; + +/* + * profile_max defines the upper bound on the number of profile probes that + * can exist (this is to prevent malicious or clumsy users from exhausing + * system resources by creating a slew of profile probes). At mod load time, + * this gets its value from PROFILE_MAX_DEFAULT or profile-max-probes if it's + * present in the profile.conf file. + */ +#define PROFILE_MAX_DEFAULT 1000 /* default max. number of probes */ +static uint32_t profile_max; /* maximum number of profile probes */ +static uint32_t profile_total; /* current number of profile probes */ + +static void +profile_fire(void *arg) +{ + profile_probe_percpu_t *pcpu = arg; + profile_probe_t *prof = pcpu->profc_probe; + hrtime_t late; + + late = dtrace_gethrtime() - pcpu->profc_expected; + pcpu->profc_expected += pcpu->profc_interval; + + dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, + CPU->cpu_profile_upc, late, 0, 0); +} + +static void +profile_tick(void *arg) +{ + profile_probe_t *prof = arg; + + dtrace_probe(prof->prof_id, CPU->cpu_profile_pc, + CPU->cpu_profile_upc, 0, 0, 0); +} + +static void +profile_create(hrtime_t interval, const char *name, int kind) +{ + profile_probe_t *prof; + int nr_frames = PROF_ARTIFICIAL_FRAMES + dtrace_mach_aframes(); + + if (profile_aframes) + nr_frames = profile_aframes; + + if (interval < profile_interval_min) + return; + + if (dtrace_probe_lookup(profile_id, NULL, NULL, name) != 0) + return; + + atomic_add_32(&profile_total, 1); + if (profile_total > profile_max) { + atomic_add_32(&profile_total, -1); + return; + } + + prof = kmem_zalloc(sizeof (profile_probe_t), KM_SLEEP); + (void) strcpy(prof->prof_name, name); + prof->prof_interval = interval; + prof->prof_cyclic = CYCLIC_NONE; + prof->prof_kind = kind; + prof->prof_id = dtrace_probe_create(profile_id, + NULL, NULL, name, nr_frames, prof); +} + +/*ARGSUSED*/ +static void +profile_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i, j, rate, kind; + hrtime_t val = 0, mult = 1, len; + const char *name, *suffix = NULL; + + const struct { + char *prefix; + int kind; + } types[] = { + { PROF_PREFIX_PROFILE, PROF_PROFILE }, + { PROF_PREFIX_TICK, PROF_TICK }, + { NULL, NULL } + }; + + const struct { + char *name; + hrtime_t mult; + } suffixes[] = { + { "ns", NANOSEC / NANOSEC }, + { "nsec", NANOSEC / NANOSEC }, + { "us", NANOSEC / MICROSEC }, + { "usec", NANOSEC / MICROSEC }, + { "ms", NANOSEC / MILLISEC }, + { "msec", NANOSEC / MILLISEC }, + { "s", NANOSEC / SEC }, + { "sec", NANOSEC / SEC }, + { "m", NANOSEC * (hrtime_t)60 }, + { "min", NANOSEC * (hrtime_t)60 }, + { "h", NANOSEC * (hrtime_t)(60 * 60) }, + { "hour", NANOSEC * (hrtime_t)(60 * 60) }, + { "d", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "day", NANOSEC * (hrtime_t)(24 * 60 * 60) }, + { "hz", 0 }, + { NULL } + }; + + if (desc == NULL) { + char n[PROF_NAMELEN]; + + /* + * If no description was provided, provide all of our probes. + */ + for (i = 0; i < sizeof (profile_rates) / sizeof (int); i++) { + if ((rate = profile_rates[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_PROFILE, rate); + profile_create(NANOSEC / rate, n, PROF_PROFILE); + } + + for (i = 0; i < sizeof (profile_ticks) / sizeof (int); i++) { + if ((rate = profile_ticks[i]) == 0) + continue; + + (void) snprintf(n, PROF_NAMELEN, "%s%d", + PROF_PREFIX_TICK, rate); + profile_create(NANOSEC / rate, n, PROF_TICK); + } + + return; + } + + name = desc->dtpd_name; + + for (i = 0; types[i].prefix != NULL; i++) { + len = strlen(types[i].prefix); + + if (strncmp(name, types[i].prefix, len) != 0) + continue; + break; + } + + if (types[i].prefix == NULL) + return; + + kind = types[i].kind; + j = strlen(name) - len; + + /* + * We need to start before any time suffix. + */ + for (j = strlen(name); j >= len; j--) { + if (name[j] >= '0' && name[j] <= '9') + break; + suffix = &name[j]; + } + + ASSERT(suffix != NULL); + + /* + * Now determine the numerical value present in the probe name. + */ + for (; j >= len; j--) { + if (name[j] < '0' || name[j] > '9') + return; + + val += (name[j] - '0') * mult; + mult *= (hrtime_t)10; + } + + if (val == 0) + return; + + /* + * Look-up the suffix to determine the multiplier. + */ + for (i = 0, mult = 0; suffixes[i].name != NULL; i++) { + if (strcasecmp(suffixes[i].name, suffix) == 0) { + mult = suffixes[i].mult; + break; + } + } + + if (suffixes[i].name == NULL && *suffix != '\0') + return; + + if (mult == 0) { + /* + * The default is frequency-per-second. + */ + val = NANOSEC / val; + } else { + val *= mult; + } + + profile_create(val, name, kind); +} + +/*ARGSUSED*/ +static void +profile_destroy(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic == CYCLIC_NONE); + kmem_free(prof, sizeof (profile_probe_t)); + + ASSERT(profile_total >= 1); + atomic_add_32(&profile_total, -1); +} + +/*ARGSUSED*/ +static void +profile_online(void *arg, cpu_t *cpu, cyc_handler_t *hdlr, cyc_time_t *when) +{ + profile_probe_t *prof = arg; + profile_probe_percpu_t *pcpu; + + pcpu = kmem_zalloc(sizeof (profile_probe_percpu_t), KM_SLEEP); + pcpu->profc_probe = prof; + + hdlr->cyh_func = profile_fire; + hdlr->cyh_arg = pcpu; + hdlr->cyh_level = CY_HIGH_LEVEL; + + when->cyt_interval = prof->prof_interval; + when->cyt_when = dtrace_gethrtime() + when->cyt_interval; + + pcpu->profc_expected = when->cyt_when; + pcpu->profc_interval = when->cyt_interval; +} + +/*ARGSUSED*/ +static void +profile_offline(void *arg, cpu_t *cpu, void *oarg) +{ + profile_probe_percpu_t *pcpu = oarg; + + ASSERT(pcpu->profc_probe == arg); + kmem_free(pcpu, sizeof (profile_probe_percpu_t)); +} + +/*ARGSUSED*/ +static int +profile_enable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + cyc_omni_handler_t omni; + cyc_handler_t hdlr; + cyc_time_t when; + + ASSERT(prof->prof_interval != 0); + ASSERT(MUTEX_HELD(&cpu_lock)); + + if (prof->prof_kind == PROF_TICK) { + hdlr.cyh_func = profile_tick; + hdlr.cyh_arg = prof; + hdlr.cyh_level = CY_HIGH_LEVEL; + + when.cyt_interval = prof->prof_interval; + when.cyt_when = dtrace_gethrtime() + when.cyt_interval; + } else { + ASSERT(prof->prof_kind == PROF_PROFILE); + omni.cyo_online = profile_online; + omni.cyo_offline = profile_offline; + omni.cyo_arg = prof; + } + + if (prof->prof_kind == PROF_TICK) { + prof->prof_cyclic = cyclic_add(&hdlr, &when); + } else { + prof->prof_cyclic = cyclic_add_omni(&omni); + } + return (0); +} + +/*ARGSUSED*/ +static void +profile_disable(void *arg, dtrace_id_t id, void *parg) +{ + profile_probe_t *prof = parg; + + ASSERT(prof->prof_cyclic != CYCLIC_NONE); + ASSERT(MUTEX_HELD(&cpu_lock)); + + cyclic_remove(prof->prof_cyclic); + prof->prof_cyclic = CYCLIC_NONE; +} + +/*ARGSUSED*/ +static int +profile_usermode(void *arg, dtrace_id_t id, void *parg) +{ + return (CPU->cpu_profile_pc == 0); +} + +static dtrace_pattr_t profile_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +}; + +static dtrace_pops_t profile_pops = { + profile_provide, + NULL, + profile_enable, + profile_disable, + NULL, + NULL, + NULL, + NULL, + profile_usermode, + profile_destroy +}; + +static int +profile_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (ddi_create_minor_node(devi, "profile", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("profile", &profile_attr, + DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER, NULL, + &profile_pops, NULL, &profile_id) != 0) { + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + profile_max = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS, + "profile-max-probes", PROFILE_MAX_DEFAULT); + + ddi_report_dev(devi); + profile_devi = devi; + return (DDI_SUCCESS); +} + +static int +profile_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(profile_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +profile_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)profile_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +profile_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops profile_cb_ops = { + profile_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops profile_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + profile_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + profile_attach, /* attach */ + profile_detach, /* detach */ + nodev, /* reset */ + &profile_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Profile Interrupt Tracing", /* name of module */ + &profile_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.conf new file mode 100644 index 00000000..618ff60f --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/profile.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="profile" parent="pseudo" instance=0; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/sdt_subr.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/sdt_subr.c new file mode 100644 index 00000000..24218507 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/sdt_subr.c @@ -0,0 +1,1191 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved. + */ + +#include <sys/sdt_impl.h> + +static dtrace_pattr_t vtrace_attr = { +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_UNSTABLE, DTRACE_STABILITY_UNSTABLE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t info_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fc_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fpu_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_CPU }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t fsinfo_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t stab_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t sdt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pattr_t xpv_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_PLATFORM }, +}; + +static dtrace_pattr_t iscsi_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +}; + +sdt_provider_t sdt_providers[] = { + { "vtrace", "__vtrace_", &vtrace_attr, 0 }, + { "sysinfo", "__cpu_sysinfo_", &info_attr, 0 }, + { "vminfo", "__cpu_vminfo_", &info_attr, 0 }, + { "fpuinfo", "__fpuinfo_", &fpu_attr, 0 }, + { "sched", "__sched_", &stab_attr, 0 }, + { "proc", "__proc_", &stab_attr, 0 }, + { "io", "__io_", &stab_attr, 0 }, + { "ip", "__ip_", &stab_attr, 0 }, + { "tcp", "__tcp_", &stab_attr, 0 }, + { "udp", "__udp_", &stab_attr, 0 }, + { "mib", "__mib_", &stab_attr, 0 }, + { "fsinfo", "__fsinfo_", &fsinfo_attr, 0 }, + { "iscsi", "__iscsi_", &iscsi_attr, 0 }, + { "nfsv3", "__nfsv3_", &stab_attr, 0 }, + { "nfsv4", "__nfsv4_", &stab_attr, 0 }, + { "xpv", "__xpv_", &xpv_attr, 0 }, + { "fc", "__fc_", &fc_attr, 0 }, + { "srp", "__srp_", &fc_attr, 0 }, + { "sysevent", "__sysevent_", &stab_attr, 0 }, + { "sdt", NULL, &sdt_attr, 0 }, + { NULL } +}; + +sdt_argdesc_t sdt_args[] = { + { "sched", "wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "dequeue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "dequeue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "enqueue", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "enqueue", 2, 1, "disp_t *", "cpuinfo_t *" }, + { "sched", "enqueue", 3, 2, "int" }, + { "sched", "off-cpu", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "off-cpu", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "tick", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "tick", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "change-pri", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "change-pri", 2, 1, "pri_t" }, + { "sched", "schedctl-nopreempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-nopreempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-nopreempt", 2, 1, "int" }, + { "sched", "schedctl-preempt", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "schedctl-preempt", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "schedctl-yield", 0, 0, "int" }, + { "sched", "surrender", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "surrender", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-sleep", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-sleep", 1, 0, "kthread_t *", "psinfo_t *" }, + { "sched", "cpucaps-wakeup", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "sched", "cpucaps-wakeup", 1, 0, "kthread_t *", "psinfo_t *" }, + + { "proc", "create", 0, 0, "proc_t *", "psinfo_t *" }, + { "proc", "exec", 0, 0, "string" }, + { "proc", "exec-failure", 0, 0, "int" }, + { "proc", "exit", 0, 0, "int" }, + { "proc", "fault", 0, 0, "int" }, + { "proc", "fault", 1, 1, "siginfo_t *" }, + { "proc", "lwp-create", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "lwp-create", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-clear", 0, 0, "int" }, + { "proc", "signal-clear", 1, 1, "siginfo_t *" }, + { "proc", "signal-discard", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-discard", 1, 1, "proc_t *", "psinfo_t *" }, + { "proc", "signal-discard", 2, 2, "int" }, + { "proc", "signal-handle", 0, 0, "int" }, + { "proc", "signal-handle", 1, 1, "siginfo_t *" }, + { "proc", "signal-handle", 2, 2, "void (*)(void)" }, + { "proc", "signal-send", 0, 0, "kthread_t *", "lwpsinfo_t *" }, + { "proc", "signal-send", 1, 0, "kthread_t *", "psinfo_t *" }, + { "proc", "signal-send", 2, 1, "int" }, + + { "io", "start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "done", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-start", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-start", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-start", 2, 0, "buf_t *", "fileinfo_t *" }, + { "io", "wait-done", 0, 0, "buf_t *", "bufinfo_t *" }, + { "io", "wait-done", 1, 0, "buf_t *", "devinfo_t *" }, + { "io", "wait-done", 2, 0, "buf_t *", "fileinfo_t *" }, + + { "mib", NULL, 0, 0, "int" }, + + { "fsinfo", NULL, 0, 0, "vnode_t *", "fileinfo_t *" }, + { "fsinfo", NULL, 1, 1, "int", "int" }, + + { "iscsi", "async-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "async-send", 1, 1, "iscsi_async_evt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "login-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "login-command", 1, 1, "iscsi_login_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "login-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "login-response", 1, 1, "iscsi_login_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "logout-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "logout-command", 1, 1, "iscsi_logout_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "logout-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "logout-response", 1, 1, "iscsi_logout_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-request", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-request", 1, 1, "iscsi_rtt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-send", 1, 1, "iscsi_data_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "data-receive", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "data-receive", 1, 1, "iscsi_data_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "nop-send", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "nop-send", 1, 1, "iscsi_nop_in_hdr_t *", "iscsiinfo_t *" }, + { "iscsi", "nop-receive", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "nop-receive", 1, 1, "iscsi_nop_out_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "scsi-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "scsi-command", 1, 1, "iscsi_scsi_cmd_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "scsi-command", 2, 2, "scsi_task_t *", "scsicmd_t *" }, + { "iscsi", "scsi-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "scsi-response", 1, 1, "iscsi_scsi_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "task-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "task-command", 1, 1, "iscsi_scsi_task_mgt_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "task-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "task-response", 1, 1, "iscsi_scsi_task_mgt_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "text-command", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "text-command", 1, 1, "iscsi_text_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "text-response", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "text-response", 1, 1, "iscsi_text_rsp_hdr_t *", + "iscsiinfo_t *" }, + { "iscsi", "xfer-start", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "xfer-start", 1, 0, "idm_conn_t *", "iscsiinfo_t *" }, + { "iscsi", "xfer-start", 2, 1, "uintptr_t", "xferinfo_t *" }, + { "iscsi", "xfer-start", 3, 2, "uint32_t"}, + { "iscsi", "xfer-start", 4, 3, "uintptr_t"}, + { "iscsi", "xfer-start", 5, 4, "uint32_t"}, + { "iscsi", "xfer-start", 6, 5, "uint32_t"}, + { "iscsi", "xfer-start", 7, 6, "uint32_t"}, + { "iscsi", "xfer-start", 8, 7, "int"}, + { "iscsi", "xfer-done", 0, 0, "idm_conn_t *", "conninfo_t *" }, + { "iscsi", "xfer-done", 1, 0, "idm_conn_t *", "iscsiinfo_t *" }, + { "iscsi", "xfer-done", 2, 1, "uintptr_t", "xferinfo_t *" }, + { "iscsi", "xfer-done", 3, 2, "uint32_t"}, + { "iscsi", "xfer-done", 4, 3, "uintptr_t"}, + { "iscsi", "xfer-done", 5, 4, "uint32_t"}, + { "iscsi", "xfer-done", 6, 5, "uint32_t"}, + { "iscsi", "xfer-done", 7, 6, "uint32_t"}, + { "iscsi", "xfer-done", 8, 7, "int"}, + + { "nfsv3", "op-getattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-start", 2, 3, "GETATTR3args *" }, + { "nfsv3", "op-getattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-getattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-getattr-done", 2, 3, "GETATTR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-lookup-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-start", 2, 3, "LOOKUP3args *" }, + { "nfsv3", "op-lookup-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-lookup-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-lookup-done", 2, 3, "LOOKUP3res *" }, + { "nfsv3", "op-access-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-start", 2, 3, "ACCESS3args *" }, + { "nfsv3", "op-access-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-access-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-access-done", 2, 3, "ACCESS3res *" }, + { "nfsv3", "op-commit-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-start", 2, 3, "COMMIT3args *" }, + { "nfsv3", "op-commit-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-commit-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-commit-done", 2, 3, "COMMIT3res *" }, + { "nfsv3", "op-create-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-start", 2, 3, "CREATE3args *" }, + { "nfsv3", "op-create-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-create-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-create-done", 2, 3, "CREATE3res *" }, + { "nfsv3", "op-fsinfo-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-start", 2, 3, "FSINFO3args *" }, + { "nfsv3", "op-fsinfo-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsinfo-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsinfo-done", 2, 3, "FSINFO3res *" }, + { "nfsv3", "op-fsstat-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-start", 2, 3, "FSSTAT3args *" }, + { "nfsv3", "op-fsstat-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-fsstat-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-fsstat-done", 2, 3, "FSSTAT3res *" }, + { "nfsv3", "op-link-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-start", 2, 3, "LINK3args *" }, + { "nfsv3", "op-link-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-link-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-link-done", 2, 3, "LINK3res *" }, + { "nfsv3", "op-mkdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-start", 2, 3, "MKDIR3args *" }, + { "nfsv3", "op-mkdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mkdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mkdir-done", 2, 3, "MKDIR3res *" }, + { "nfsv3", "op-mknod-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-start", 2, 3, "MKNOD3args *" }, + { "nfsv3", "op-mknod-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-mknod-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-mknod-done", 2, 3, "MKNOD3res *" }, + { "nfsv3", "op-null-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-null-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-null-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-start", 2, 3, "PATHCONF3args *" }, + { "nfsv3", "op-pathconf-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-pathconf-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-pathconf-done", 2, 3, "PATHCONF3res *" }, + { "nfsv3", "op-read-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-start", 2, 3, "READ3args *" }, + { "nfsv3", "op-read-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-read-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-read-done", 2, 3, "READ3res *" }, + { "nfsv3", "op-readdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-start", 2, 3, "READDIR3args *" }, + { "nfsv3", "op-readdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdir-done", 2, 3, "READDIR3res *" }, + { "nfsv3", "op-readdirplus-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-start", 2, 3, "READDIRPLUS3args *" }, + { "nfsv3", "op-readdirplus-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readdirplus-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readdirplus-done", 2, 3, "READDIRPLUS3res *" }, + { "nfsv3", "op-readlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-start", 2, 3, "READLINK3args *" }, + { "nfsv3", "op-readlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-readlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-readlink-done", 2, 3, "READLINK3res *" }, + { "nfsv3", "op-remove-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-start", 2, 3, "REMOVE3args *" }, + { "nfsv3", "op-remove-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-remove-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-remove-done", 2, 3, "REMOVE3res *" }, + { "nfsv3", "op-rename-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-start", 2, 3, "RENAME3args *" }, + { "nfsv3", "op-rename-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rename-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rename-done", 2, 3, "RENAME3res *" }, + { "nfsv3", "op-rmdir-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-start", 2, 3, "RMDIR3args *" }, + { "nfsv3", "op-rmdir-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-rmdir-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-rmdir-done", 2, 3, "RMDIR3res *" }, + { "nfsv3", "op-setattr-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-start", 2, 3, "SETATTR3args *" }, + { "nfsv3", "op-setattr-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-setattr-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-setattr-done", 2, 3, "SETATTR3res *" }, + { "nfsv3", "op-symlink-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-start", 2, 3, "SYMLINK3args *" }, + { "nfsv3", "op-symlink-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-symlink-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-symlink-done", 2, 3, "SYMLINK3res *" }, + { "nfsv3", "op-write-start", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-start", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-start", 2, 3, "WRITE3args *" }, + { "nfsv3", "op-write-done", 0, 0, "struct svc_req *", + "conninfo_t *" }, + { "nfsv3", "op-write-done", 1, 1, "nfsv3oparg_t *", + "nfsv3opinfo_t *" }, + { "nfsv3", "op-write-done", 2, 3, "WRITE3res *" }, + + { "nfsv4", "null-start", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "null-done", 0, 0, "struct svc_req *", "conninfo_t *" }, + { "nfsv4", "compound-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-start", 2, 1, "COMPOUND4args *" }, + { "nfsv4", "compound-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "compound-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "compound-done", 2, 1, "COMPOUND4res *" }, + { "nfsv4", "op-access-start", 0, 0, "struct compound_state *", + "conninfo_t *"}, + { "nfsv4", "op-access-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-start", 2, 1, "ACCESS4args *" }, + { "nfsv4", "op-access-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-access-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-access-done", 2, 1, "ACCESS4res *" }, + { "nfsv4", "op-close-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-start", 2, 1, "CLOSE4args *" }, + { "nfsv4", "op-close-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-close-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-close-done", 2, 1, "CLOSE4res *" }, + { "nfsv4", "op-commit-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-start", 2, 1, "COMMIT4args *" }, + { "nfsv4", "op-commit-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-commit-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-commit-done", 2, 1, "COMMIT4res *" }, + { "nfsv4", "op-create-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-start", 2, 1, "CREATE4args *" }, + { "nfsv4", "op-create-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-create-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-create-done", 2, 1, "CREATE4res *" }, + { "nfsv4", "op-delegpurge-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-start", 2, 1, "DELEGPURGE4args *" }, + { "nfsv4", "op-delegpurge-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegpurge-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegpurge-done", 2, 1, "DELEGPURGE4res *" }, + { "nfsv4", "op-delegreturn-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-start", 2, 1, "DELEGRETURN4args *" }, + { "nfsv4", "op-delegreturn-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-delegreturn-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-delegreturn-done", 2, 1, "DELEGRETURN4res *" }, + { "nfsv4", "op-getattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-start", 2, 1, "GETATTR4args *" }, + { "nfsv4", "op-getattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getattr-done", 2, 1, "GETATTR4res *" }, + { "nfsv4", "op-getfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-getfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-getfh-done", 2, 1, "GETFH4res *" }, + { "nfsv4", "op-link-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-start", 2, 1, "LINK4args *" }, + { "nfsv4", "op-link-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-link-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-link-done", 2, 1, "LINK4res *" }, + { "nfsv4", "op-lock-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-start", 2, 1, "LOCK4args *" }, + { "nfsv4", "op-lock-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lock-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lock-done", 2, 1, "LOCK4res *" }, + { "nfsv4", "op-lockt-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-start", 2, 1, "LOCKT4args *" }, + { "nfsv4", "op-lockt-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lockt-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lockt-done", 2, 1, "LOCKT4res *" }, + { "nfsv4", "op-locku-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-start", 2, 1, "LOCKU4args *" }, + { "nfsv4", "op-locku-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-locku-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-locku-done", 2, 1, "LOCKU4res *" }, + { "nfsv4", "op-lookup-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-start", 2, 1, "LOOKUP4args *" }, + { "nfsv4", "op-lookup-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookup-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookup-done", 2, 1, "LOOKUP4res *" }, + { "nfsv4", "op-lookupp-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-lookupp-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-lookupp-done", 2, 1, "LOOKUPP4res *" }, + { "nfsv4", "op-nverify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-start", 2, 1, "NVERIFY4args *" }, + { "nfsv4", "op-nverify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-nverify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-nverify-done", 2, 1, "NVERIFY4res *" }, + { "nfsv4", "op-open-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-start", 2, 1, "OPEN4args *" }, + { "nfsv4", "op-open-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-done", 2, 1, "OPEN4res *" }, + { "nfsv4", "op-open-confirm-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-start", 2, 1, "OPEN_CONFIRM4args *" }, + { "nfsv4", "op-open-confirm-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-confirm-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-confirm-done", 2, 1, "OPEN_CONFIRM4res *" }, + { "nfsv4", "op-open-downgrade-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-start", 2, 1, "OPEN_DOWNGRADE4args *" }, + { "nfsv4", "op-open-downgrade-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-open-downgrade-done", 2, 1, "OPEN_DOWNGRADE4res *" }, + { "nfsv4", "op-openattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-start", 2, 1, "OPENATTR4args *" }, + { "nfsv4", "op-openattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-openattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-openattr-done", 2, 1, "OPENATTR4res *" }, + { "nfsv4", "op-putfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-start", 2, 1, "PUTFH4args *" }, + { "nfsv4", "op-putfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putfh-done", 2, 1, "PUTFH4res *" }, + { "nfsv4", "op-putpubfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putpubfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putpubfh-done", 2, 1, "PUTPUBFH4res *" }, + { "nfsv4", "op-putrootfh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-putrootfh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-putrootfh-done", 2, 1, "PUTROOTFH4res *" }, + { "nfsv4", "op-read-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-start", 2, 1, "READ4args *" }, + { "nfsv4", "op-read-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-read-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-read-done", 2, 1, "READ4res *" }, + { "nfsv4", "op-readdir-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-start", 2, 1, "READDIR4args *" }, + { "nfsv4", "op-readdir-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readdir-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readdir-done", 2, 1, "READDIR4res *" }, + { "nfsv4", "op-readlink-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-readlink-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-readlink-done", 2, 1, "READLINK4res *" }, + { "nfsv4", "op-release-lockowner-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-start", 2, 1, + "RELEASE_LOCKOWNER4args *" }, + { "nfsv4", "op-release-lockowner-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-release-lockowner-done", 2, 1, + "RELEASE_LOCKOWNER4res *" }, + { "nfsv4", "op-remove-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-start", 2, 1, "REMOVE4args *" }, + { "nfsv4", "op-remove-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-remove-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-remove-done", 2, 1, "REMOVE4res *" }, + { "nfsv4", "op-rename-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-start", 2, 1, "RENAME4args *" }, + { "nfsv4", "op-rename-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-rename-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-rename-done", 2, 1, "RENAME4res *" }, + { "nfsv4", "op-renew-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-start", 2, 1, "RENEW4args *" }, + { "nfsv4", "op-renew-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-renew-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-renew-done", 2, 1, "RENEW4res *" }, + { "nfsv4", "op-restorefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-restorefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-restorefh-done", 2, 1, "RESTOREFH4res *" }, + { "nfsv4", "op-savefh-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-savefh-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-savefh-done", 2, 1, "SAVEFH4res *" }, + { "nfsv4", "op-secinfo-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-start", 2, 1, "SECINFO4args *" }, + { "nfsv4", "op-secinfo-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-secinfo-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-secinfo-done", 2, 1, "SECINFO4res *" }, + { "nfsv4", "op-setattr-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-start", 2, 1, "SETATTR4args *" }, + { "nfsv4", "op-setattr-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setattr-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setattr-done", 2, 1, "SETATTR4res *" }, + { "nfsv4", "op-setclientid-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-start", 2, 1, "SETCLIENTID4args *" }, + { "nfsv4", "op-setclientid-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-setclientid-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-done", 2, 1, "SETCLIENTID4res *" }, + { "nfsv4", "op-setclientid-confirm-start", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-start", 2, 1, + "SETCLIENTID_CONFIRM4args *" }, + { "nfsv4", "op-setclientid-confirm-done", 0, 0, + "struct compound_state *", "conninfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 1, 0, + "struct compound_state *", "nfsv4opinfo_t *" }, + { "nfsv4", "op-setclientid-confirm-done", 2, 1, + "SETCLIENTID_CONFIRM4res *" }, + { "nfsv4", "op-verify-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-start", 2, 1, "VERIFY4args *" }, + { "nfsv4", "op-verify-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-verify-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-verify-done", 2, 1, "VERIFY4res *" }, + { "nfsv4", "op-write-start", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-start", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-start", 2, 1, "WRITE4args *" }, + { "nfsv4", "op-write-done", 0, 0, "struct compound_state *", + "conninfo_t *" }, + { "nfsv4", "op-write-done", 1, 0, "struct compound_state *", + "nfsv4opinfo_t *" }, + { "nfsv4", "op-write-done", 2, 1, "WRITE4res *" }, + { "nfsv4", "cb-recall-start", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-start", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-start", 2, 2, "CB_RECALL4args *" }, + { "nfsv4", "cb-recall-done", 0, 0, "rfs4_client_t *", + "conninfo_t *" }, + { "nfsv4", "cb-recall-done", 1, 1, "rfs4_deleg_state_t *", + "nfsv4cbinfo_t *" }, + { "nfsv4", "cb-recall-done", 2, 2, "CB_RECALL4res *" }, + + { "ip", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "ip", "send", 1, 1, "conn_t *", "csinfo_t *" }, + { "ip", "send", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "ip", "send", 3, 3, "__dtrace_ipsr_ill_t *", "ifinfo_t *" }, + { "ip", "send", 4, 4, "ipha_t *", "ipv4info_t *" }, + { "ip", "send", 5, 5, "ip6_t *", "ipv6info_t *" }, + { "ip", "send", 6, 6, "int" }, /* used by __dtrace_ipsr_ill_t */ + { "ip", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "ip", "receive", 1, 1, "conn_t *", "csinfo_t *" }, + { "ip", "receive", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "ip", "receive", 3, 3, "__dtrace_ipsr_ill_t *", "ifinfo_t *" }, + { "ip", "receive", 4, 4, "ipha_t *", "ipv4info_t *" }, + { "ip", "receive", 5, 5, "ip6_t *", "ipv6info_t *" }, + { "ip", "receive", 6, 6, "int" }, /* used by __dtrace_ipsr_ill_t */ + + { "tcp", "connect-established", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-established", 1, 1, "ip_xmit_attr_t *", + "csinfo_t *" }, + { "tcp", "connect-established", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-established", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-established", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "connect-refused", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-refused", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "connect-refused", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-refused", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-refused", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "connect-request", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "connect-request", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "connect-request", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "connect-request", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "connect-request", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "accept-established", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "accept-established", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "accept-established", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "accept-established", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "accept-established", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "accept-refused", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "accept-refused", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "accept-refused", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "tcp", "accept-refused", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "accept-refused", 4, 4, "tcph_t *", "tcpinfo_t *" }, + { "tcp", "state-change", 0, 0, "void", "void" }, + { "tcp", "state-change", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "state-change", 2, 2, "void", "void" }, + { "tcp", "state-change", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "state-change", 4, 4, "void", "void" }, + { "tcp", "state-change", 5, 5, "int32_t", "tcplsinfo_t *" }, + { "tcp", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "send", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "send", 2, 2, "__dtrace_tcp_void_ip_t *", "ipinfo_t *" }, + { "tcp", "send", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "send", 4, 4, "__dtrace_tcp_tcph_t *", "tcpinfo_t *" }, + { "tcp", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "tcp", "receive", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "tcp", "receive", 2, 2, "__dtrace_tcp_void_ip_t *", "ipinfo_t *" }, + { "tcp", "receive", 3, 3, "tcp_t *", "tcpsinfo_t *" }, + { "tcp", "receive", 4, 4, "__dtrace_tcp_tcph_t *", "tcpinfo_t *" }, + + { "udp", "send", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "udp", "send", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "udp", "send", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "udp", "send", 3, 3, "udp_t *", "udpsinfo_t *" }, + { "udp", "send", 4, 4, "udpha_t *", "udpinfo_t *" }, + { "udp", "receive", 0, 0, "mblk_t *", "pktinfo_t *" }, + { "udp", "receive", 1, 1, "ip_xmit_attr_t *", "csinfo_t *" }, + { "udp", "receive", 2, 2, "void_ip_t *", "ipinfo_t *" }, + { "udp", "receive", 3, 3, "udp_t *", "udpsinfo_t *" }, + { "udp", "receive", 4, 4, "udpha_t *", "udpinfo_t *" }, + + { "sysevent", "post", 0, 0, "evch_bind_t *", "syseventchaninfo_t *" }, + { "sysevent", "post", 1, 1, "sysevent_impl_t *", "syseventinfo_t *" }, + + { "xpv", "add-to-physmap-end", 0, 0, "int" }, + { "xpv", "add-to-physmap-start", 0, 0, "domid_t" }, + { "xpv", "add-to-physmap-start", 1, 1, "uint_t" }, + { "xpv", "add-to-physmap-start", 2, 2, "ulong_t" }, + { "xpv", "add-to-physmap-start", 3, 3, "ulong_t" }, + { "xpv", "decrease-reservation-end", 0, 0, "int" }, + { "xpv", "decrease-reservation-start", 0, 0, "domid_t" }, + { "xpv", "decrease-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "decrease-reservation-start", 2, 2, "uint_t" }, + { "xpv", "decrease-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "dom-create-start", 0, 0, "xen_domctl_t *" }, + { "xpv", "dom-destroy-start", 0, 0, "domid_t" }, + { "xpv", "dom-pause-start", 0, 0, "domid_t" }, + { "xpv", "dom-unpause-start", 0, 0, "domid_t" }, + { "xpv", "dom-create-end", 0, 0, "int" }, + { "xpv", "dom-destroy-end", 0, 0, "int" }, + { "xpv", "dom-pause-end", 0, 0, "int" }, + { "xpv", "dom-unpause-end", 0, 0, "int" }, + { "xpv", "evtchn-op-end", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 0, 0, "int" }, + { "xpv", "evtchn-op-start", 1, 1, "void *" }, + { "xpv", "increase-reservation-end", 0, 0, "int" }, + { "xpv", "increase-reservation-start", 0, 0, "domid_t" }, + { "xpv", "increase-reservation-start", 1, 1, "ulong_t" }, + { "xpv", "increase-reservation-start", 2, 2, "uint_t" }, + { "xpv", "increase-reservation-start", 3, 3, "ulong_t *" }, + { "xpv", "mmap-end", 0, 0, "int" }, + { "xpv", "mmap-entry", 0, 0, "ulong_t" }, + { "xpv", "mmap-entry", 1, 1, "ulong_t" }, + { "xpv", "mmap-entry", 2, 2, "ulong_t" }, + { "xpv", "mmap-start", 0, 0, "domid_t" }, + { "xpv", "mmap-start", 1, 1, "int" }, + { "xpv", "mmap-start", 2, 2, "privcmd_mmap_entry_t *" }, + { "xpv", "mmapbatch-end", 0, 0, "int" }, + { "xpv", "mmapbatch-end", 1, 1, "struct seg *" }, + { "xpv", "mmapbatch-end", 2, 2, "caddr_t" }, + { "xpv", "mmapbatch-start", 0, 0, "domid_t" }, + { "xpv", "mmapbatch-start", 1, 1, "int" }, + { "xpv", "mmapbatch-start", 2, 2, "caddr_t" }, + { "xpv", "mmu-ext-op-end", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 0, 0, "int" }, + { "xpv", "mmu-ext-op-start", 1, 1, "struct mmuext_op *" }, + { "xpv", "mmu-update-start", 0, 0, "int" }, + { "xpv", "mmu-update-start", 1, 1, "int" }, + { "xpv", "mmu-update-start", 2, 2, "mmu_update_t *" }, + { "xpv", "mmu-update-end", 0, 0, "int" }, + { "xpv", "populate-physmap-end", 0, 0, "int" }, + { "xpv", "populate-physmap-start", 0, 0, "domid_t" }, + { "xpv", "populate-physmap-start", 1, 1, "ulong_t" }, + { "xpv", "populate-physmap-start", 2, 2, "ulong_t *" }, + { "xpv", "set-memory-map-end", 0, 0, "int" }, + { "xpv", "set-memory-map-start", 0, 0, "domid_t" }, + { "xpv", "set-memory-map-start", 1, 1, "int" }, + { "xpv", "set-memory-map-start", 2, 2, "struct xen_memory_map *" }, + { "xpv", "setvcpucontext-end", 0, 0, "int" }, + { "xpv", "setvcpucontext-start", 0, 0, "domid_t" }, + { "xpv", "setvcpucontext-start", 1, 1, "vcpu_guest_context_t *" }, + + { "srp", "service-up", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "service-up", 1, 0, "srpt_session_t *", "srp_portinfo_t *" }, + { "srp", "service-down", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "service-down", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-command", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "login-command", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-command", 2, 1, "srp_login_req_t *", + "srp_logininfo_t *" }, + { "srp", "login-response", 0, 0, "srpt_session_t *", "conninfo_t *" }, + { "srp", "login-response", 1, 0, "srpt_session_t *", + "srp_portinfo_t *" }, + { "srp", "login-response", 2, 1, "srp_login_rsp_t *", + "srp_logininfo_t *" }, + { "srp", "login-response", 3, 2, "srp_login_rej_t *" }, + { "srp", "logout-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "logout-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "task-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-command", 2, 1, "srp_cmd_req_t *", "srp_taskinfo_t *" }, + { "srp", "task-response", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "task-response", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "task-response", 2, 1, "srp_rsp_t *", "srp_taskinfo_t *" }, + { "srp", "task-response", 3, 2, "scsi_task_t *" }, + { "srp", "task-response", 4, 3, "int8_t" }, + { "srp", "scsi-command", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "scsi-command", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "scsi-command", 2, 1, "scsi_task_t *", "scsicmd_t *" }, + { "srp", "scsi-command", 3, 2, "srp_cmd_req_t *", "srp_taskinfo_t *" }, + { "srp", "scsi-response", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "scsi-response", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "scsi-response", 2, 1, "srp_rsp_t *", "srp_taskinfo_t *" }, + { "srp", "scsi-response", 3, 2, "scsi_task_t *" }, + { "srp", "scsi-response", 4, 3, "int8_t" }, + { "srp", "xfer-start", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "xfer-start", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "xfer-start", 2, 1, "ibt_wr_ds_t *", "xferinfo_t *" }, + { "srp", "xfer-start", 3, 2, "srpt_iu_t *", "srp_taskinfo_t *" }, + { "srp", "xfer-start", 4, 3, "ibt_send_wr_t *"}, + { "srp", "xfer-start", 5, 4, "uint32_t" }, + { "srp", "xfer-start", 6, 5, "uint32_t" }, + { "srp", "xfer-start", 7, 6, "uint32_t" }, + { "srp", "xfer-start", 8, 7, "uint32_t" }, + { "srp", "xfer-done", 0, 0, "srpt_channel_t *", "conninfo_t *" }, + { "srp", "xfer-done", 1, 0, "srpt_channel_t *", + "srp_portinfo_t *" }, + { "srp", "xfer-done", 2, 1, "ibt_wr_ds_t *", "xferinfo_t *" }, + { "srp", "xfer-done", 3, 2, "srpt_iu_t *", "srp_taskinfo_t *" }, + { "srp", "xfer-done", 4, 3, "ibt_send_wr_t *"}, + { "srp", "xfer-done", 5, 4, "uint32_t" }, + { "srp", "xfer-done", 6, 5, "uint32_t" }, + { "srp", "xfer-done", 7, 6, "uint32_t" }, + { "srp", "xfer-done", 8, 7, "uint32_t" }, + + { "fc", "link-up", 0, 0, "fct_i_local_port_t *", "conninfo_t *" }, + { "fc", "link-down", 0, 0, "fct_i_local_port_t *", "conninfo_t *" }, + { "fc", "fabric-login-start", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "fabric-login-start", 1, 0, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "fabric-login-end", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "fabric-login-end", 1, 0, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-login-start", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-start", 3, 3, "int", "int" }, + { "fc", "rport-login-end", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-login-end", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-end", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-login-end", 3, 3, "int", "int" }, + { "fc", "rport-login-end", 4, 4, "int", "int" }, + { "fc", "rport-logout-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-logout-start", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-start", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-start", 3, 3, "int", "int" }, + { "fc", "rport-logout-end", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "rport-logout-end", 1, 1, "fct_local_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-end", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "rport-logout-end", 3, 3, "int", "int" }, + { "fc", "scsi-command", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "scsi-command", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-command", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "scsi-command", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-response", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "scsi-response", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "scsi-response", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "scsi-response", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "xfer-start", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "xfer-start", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-start", 4, 4, "stmf_data_buf_t *", + "fc_xferinfo_t *" }, + { "fc", "xfer-done", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "xfer-done", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-done", 2, 2, "scsi_task_t *", + "scsicmd_t *" }, + { "fc", "xfer-done", 3, 3, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + { "fc", "xfer-done", 4, 4, "stmf_data_buf_t *", + "fc_xferinfo_t *" }, + { "fc", "rscn-receive", 0, 0, "fct_i_local_port_t *", + "conninfo_t *" }, + { "fc", "rscn-receive", 1, 1, "int", "int"}, + { "fc", "abts-receive", 0, 0, "fct_cmd_t *", + "conninfo_t *" }, + { "fc", "abts-receive", 1, 1, "fct_i_local_port_t *", + "fc_port_info_t *" }, + { "fc", "abts-receive", 2, 2, "fct_i_remote_port_t *", + "fc_port_info_t *" }, + + + { NULL } +}; + +/*ARGSUSED*/ +void +sdt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + sdt_probe_t *sdp = parg; + int i; + + desc->dtargd_native[0] = '\0'; + desc->dtargd_xlate[0] = '\0'; + + for (i = 0; sdt_args[i].sda_provider != NULL; i++) { + sdt_argdesc_t *a = &sdt_args[i]; + + if (strcmp(sdp->sdp_provider->sdtp_name, a->sda_provider) != 0) + continue; + + if (a->sda_name != NULL && + strcmp(sdp->sdp_name, a->sda_name) != 0) + continue; + + if (desc->dtargd_ndx != a->sda_ndx) + continue; + + if (a->sda_native != NULL) + (void) strcpy(desc->dtargd_native, a->sda_native); + + if (a->sda_xlate != NULL) + (void) strcpy(desc->dtargd_xlate, a->sda_xlate); + + desc->dtargd_mapping = a->sda_mapping; + return; + } + + desc->dtargd_ndx = DTRACE_ARGNONE; +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.c new file mode 100644 index 00000000..b864041c --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.c @@ -0,0 +1,374 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/dtrace.h> +#include <sys/systrace.h> +#include <sys/stat.h> +#include <sys/systm.h> +#include <sys/conf.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/atomic.h> + +#define SYSTRACE_ARTIFICIAL_FRAMES 1 + +#define SYSTRACE_SHIFT 16 +#define SYSTRACE_ISENTRY(x) ((int)(x) >> SYSTRACE_SHIFT) +#define SYSTRACE_SYSNUM(x) ((int)(x) & ((1 << SYSTRACE_SHIFT) - 1)) +#define SYSTRACE_ENTRY(id) ((1 << SYSTRACE_SHIFT) | (id)) +#define SYSTRACE_RETURN(id) (id) + +#if ((1 << SYSTRACE_SHIFT) <= NSYSCALL) +#error 1 << SYSTRACE_SHIFT must exceed number of system calls +#endif + +static dev_info_t *systrace_devi; +static dtrace_provider_id_t systrace_id; + +static void +systrace_init(struct sysent *actual, systrace_sysent_t **interposed) +{ + systrace_sysent_t *sysent = *interposed; + int i; + + if (sysent == NULL) { + *interposed = sysent = kmem_zalloc(sizeof (systrace_sysent_t) * + NSYSCALL, KM_SLEEP); + } + + for (i = 0; i < NSYSCALL; i++) { + struct sysent *a = &actual[i]; + systrace_sysent_t *s = &sysent[i]; + + if (LOADABLE_SYSCALL(a) && !LOADED_SYSCALL(a)) + continue; + + if (a->sy_callc == dtrace_systrace_syscall) + continue; + +#ifdef _SYSCALL32_IMPL + if (a->sy_callc == dtrace_systrace_syscall32) + continue; +#endif + + s->stsy_underlying = a->sy_callc; + } +} + +/*ARGSUSED*/ +static void +systrace_provide(void *arg, const dtrace_probedesc_t *desc) +{ + int i; + + if (desc != NULL) + return; + + systrace_init(sysent, &systrace_sysent); +#ifdef _SYSCALL32_IMPL + systrace_init(sysent32, &systrace_sysent32); +#endif + + for (i = 0; i < NSYSCALL; i++) { + if (systrace_sysent[i].stsy_underlying == NULL) + continue; + + if (dtrace_probe_lookup(systrace_id, NULL, + syscallnames[i], "entry") != 0) + continue; + + (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i], + "entry", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_ENTRY(i))); + (void) dtrace_probe_create(systrace_id, NULL, syscallnames[i], + "return", SYSTRACE_ARTIFICIAL_FRAMES, + (void *)((uintptr_t)SYSTRACE_RETURN(i))); + + systrace_sysent[i].stsy_entry = DTRACE_IDNONE; + systrace_sysent[i].stsy_return = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[i].stsy_entry = DTRACE_IDNONE; + systrace_sysent32[i].stsy_return = DTRACE_IDNONE; +#endif + } +} + +/*ARGSUSED*/ +static void +systrace_destroy(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + + /* + * There's nothing to do here but assert that we have actually been + * disabled. + */ + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + ASSERT(systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE); +#ifdef _SYSCALL32_IMPL + ASSERT(systrace_sysent32[sysnum].stsy_entry == DTRACE_IDNONE); +#endif + } else { + ASSERT(systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE); +#ifdef _SYSCALL32_IMPL + ASSERT(systrace_sysent32[sysnum].stsy_return == DTRACE_IDNONE); +#endif + } +} + +/*ARGSUSED*/ +static int +systrace_enable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int enabled = (systrace_sysent[sysnum].stsy_entry != DTRACE_IDNONE || + systrace_sysent[sysnum].stsy_return != DTRACE_IDNONE); + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + systrace_sysent[sysnum].stsy_entry = id; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_entry = id; +#endif + } else { + systrace_sysent[sysnum].stsy_return = id; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_return = id; +#endif + } + + if (enabled) { + ASSERT(sysent[sysnum].sy_callc == dtrace_systrace_syscall); + return (0); + } + + (void) casptr(&sysent[sysnum].sy_callc, + (void *)systrace_sysent[sysnum].stsy_underlying, + (void *)dtrace_systrace_syscall); +#ifdef _SYSCALL32_IMPL + (void) casptr(&sysent32[sysnum].sy_callc, + (void *)systrace_sysent32[sysnum].stsy_underlying, + (void *)dtrace_systrace_syscall32); +#endif + return (0); +} + +/*ARGSUSED*/ +static void +systrace_disable(void *arg, dtrace_id_t id, void *parg) +{ + int sysnum = SYSTRACE_SYSNUM((uintptr_t)parg); + int disable = (systrace_sysent[sysnum].stsy_entry == DTRACE_IDNONE || + systrace_sysent[sysnum].stsy_return == DTRACE_IDNONE); + + if (disable) { + (void) casptr(&sysent[sysnum].sy_callc, + (void *)dtrace_systrace_syscall, + (void *)systrace_sysent[sysnum].stsy_underlying); + +#ifdef _SYSCALL32_IMPL + (void) casptr(&sysent32[sysnum].sy_callc, + (void *)dtrace_systrace_syscall32, + (void *)systrace_sysent32[sysnum].stsy_underlying); +#endif + } + + if (SYSTRACE_ISENTRY((uintptr_t)parg)) { + systrace_sysent[sysnum].stsy_entry = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_entry = DTRACE_IDNONE; +#endif + } else { + systrace_sysent[sysnum].stsy_return = DTRACE_IDNONE; +#ifdef _SYSCALL32_IMPL + systrace_sysent32[sysnum].stsy_return = DTRACE_IDNONE; +#endif + } +} + +static dtrace_pattr_t systrace_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t systrace_pops = { + systrace_provide, + NULL, + systrace_enable, + systrace_disable, + NULL, + NULL, + NULL, + NULL, + NULL, + systrace_destroy +}; + +static int +systrace_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + systrace_probe = (void (*)())dtrace_probe; + membar_enter(); + + if (ddi_create_minor_node(devi, "systrace", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("syscall", &systrace_attr, DTRACE_PRIV_USER, NULL, + &systrace_pops, NULL, &systrace_id) != 0) { + systrace_probe = systrace_stub; + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + systrace_devi = devi; + + return (DDI_SUCCESS); +} + +static int +systrace_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(systrace_id) != 0) + return (DDI_FAILURE); + + ddi_remove_minor_node(devi, NULL); + systrace_probe = systrace_stub; + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +systrace_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)systrace_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +systrace_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops systrace_cb_ops = { + systrace_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops systrace_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + systrace_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + systrace_attach, /* attach */ + systrace_detach, /* detach */ + nodev, /* reset */ + &systrace_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "System Call Tracing", /* name of module */ + &systrace_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.conf new file mode 100644 index 00000000..02b0dac7 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/dtrace/systrace.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="systrace" parent="pseudo" instance=0; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/dtrace_subr.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/dtrace_subr.c new file mode 100644 index 00000000..f2a9ac1b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/os/dtrace_subr.c @@ -0,0 +1,337 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/dtrace.h> +#include <sys/cmn_err.h> +#include <sys/tnf.h> +#include <sys/atomic.h> +#include <sys/prsystm.h> +#include <sys/modctl.h> +#include <sys/aio_impl.h> + +#ifdef __sparc +#include <sys/privregs.h> +#endif + +void (*dtrace_cpu_init)(processorid_t); +void (*dtrace_modload)(struct modctl *); +void (*dtrace_modunload)(struct modctl *); +void (*dtrace_helpers_cleanup)(void); +void (*dtrace_helpers_fork)(proc_t *, proc_t *); +void (*dtrace_cpustart_init)(void); +void (*dtrace_cpustart_fini)(void); +void (*dtrace_cpc_fire)(uint64_t); + +void (*dtrace_debugger_init)(void); +void (*dtrace_debugger_fini)(void); + +dtrace_vtime_state_t dtrace_vtime_active = 0; +dtrace_cacheid_t dtrace_predcache_id = DTRACE_CACHEIDNONE + 1; + +/* + * dtrace_cpc_in_use usage statement: this global variable is used by the cpc + * hardware overflow interrupt handler and the kernel cpc framework to check + * whether or not the DTrace cpc provider is currently in use. The variable is + * set before counters are enabled with the first enabling and cleared when + * the last enabling is disabled. Its value at any given time indicates the + * number of active dcpc based enablings. The global 'kcpc_cpuctx_lock' rwlock + * is held during initial setting to protect races between kcpc_open() and the + * first enabling. The locking provided by the DTrace subsystem, the kernel + * cpc framework and the cpu management framework protect consumers from race + * conditions on enabling and disabling probes. + */ +uint32_t dtrace_cpc_in_use = 0; + +typedef struct dtrace_hrestime { + lock_t dthr_lock; /* lock for this element */ + timestruc_t dthr_hrestime; /* hrestime value */ + int64_t dthr_adj; /* hrestime_adj value */ + hrtime_t dthr_hrtime; /* hrtime value */ +} dtrace_hrestime_t; + +static dtrace_hrestime_t dtrace_hrestime[2]; + +/* + * Making available adjustable high-resolution time in DTrace is regrettably + * more complicated than one might think it should be. The problem is that + * the variables related to adjusted high-resolution time (hrestime, + * hrestime_adj and friends) are adjusted under hres_lock -- and this lock may + * be held when we enter probe context. One might think that we could address + * this by having a single snapshot copy that is stored under a different lock + * from hres_tick(), using the snapshot iff hres_lock is locked in probe + * context. Unfortunately, this too won't work: because hres_lock is grabbed + * in more than just hres_tick() context, we could enter probe context + * concurrently on two different CPUs with both locks (hres_lock and the + * snapshot lock) held. As this implies, the fundamental problem is that we + * need to have access to a snapshot of these variables that we _know_ will + * not be locked in probe context. To effect this, we have two snapshots + * protected by two different locks, and we mandate that these snapshots are + * recorded in succession by a single thread calling dtrace_hres_tick(). (We + * assure this by calling it out of the same CY_HIGH_LEVEL cyclic that calls + * hres_tick().) A single thread can't be in two places at once: one of the + * snapshot locks is guaranteed to be unheld at all times. The + * dtrace_gethrestime() algorithm is thus to check first one snapshot and then + * the other to find the unlocked snapshot. + */ +void +dtrace_hres_tick(void) +{ + int i; + ushort_t spl; + + for (i = 0; i < 2; i++) { + dtrace_hrestime_t tmp; + + spl = hr_clock_lock(); + tmp.dthr_hrestime = hrestime; + tmp.dthr_adj = hrestime_adj; + tmp.dthr_hrtime = dtrace_gethrtime(); + hr_clock_unlock(spl); + + lock_set(&dtrace_hrestime[i].dthr_lock); + dtrace_hrestime[i].dthr_hrestime = tmp.dthr_hrestime; + dtrace_hrestime[i].dthr_adj = tmp.dthr_adj; + dtrace_hrestime[i].dthr_hrtime = tmp.dthr_hrtime; + dtrace_membar_producer(); + + /* + * To allow for lock-free examination of this lock, we use + * the same trick that is used hres_lock; for more details, + * see the description of this technique in sun4u/sys/clock.h. + */ + dtrace_hrestime[i].dthr_lock++; + } +} + +hrtime_t +dtrace_gethrestime(void) +{ + dtrace_hrestime_t snap; + hrtime_t now; + int i = 0, adj, nslt; + + for (;;) { + snap.dthr_lock = dtrace_hrestime[i].dthr_lock; + dtrace_membar_consumer(); + snap.dthr_hrestime = dtrace_hrestime[i].dthr_hrestime; + snap.dthr_hrtime = dtrace_hrestime[i].dthr_hrtime; + snap.dthr_adj = dtrace_hrestime[i].dthr_adj; + dtrace_membar_consumer(); + + if ((snap.dthr_lock & ~1) == dtrace_hrestime[i].dthr_lock) + break; + + /* + * If we're here, the lock was either locked, or it + * transitioned while we were taking the snapshot. Either + * way, we're going to try the other dtrace_hrestime element; + * we know that it isn't possible for both to be locked + * simultaneously, so we will ultimately get a good snapshot. + */ + i ^= 1; + } + + /* + * We have a good snapshot. Now perform any necessary adjustments. + */ + nslt = dtrace_gethrtime() - snap.dthr_hrtime; + ASSERT(nslt >= 0); + + now = ((hrtime_t)snap.dthr_hrestime.tv_sec * (hrtime_t)NANOSEC) + + snap.dthr_hrestime.tv_nsec; + + if (snap.dthr_adj != 0) { + if (snap.dthr_adj > 0) { + adj = (nslt >> adj_shift); + if (adj > snap.dthr_adj) + adj = (int)snap.dthr_adj; + } else { + adj = -(nslt >> adj_shift); + if (adj < snap.dthr_adj) + adj = (int)snap.dthr_adj; + } + now += adj; + } + + return (now); +} + +void +dtrace_vtime_enable(void) +{ + dtrace_vtime_state_t state, nstate; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_INACTIVE: + nstate = DTRACE_VTIME_ACTIVE; + break; + + case DTRACE_VTIME_INACTIVE_TNF: + nstate = DTRACE_VTIME_ACTIVE_TNF; + break; + + case DTRACE_VTIME_ACTIVE: + case DTRACE_VTIME_ACTIVE_TNF: + panic("DTrace virtual time already enabled"); + /*NOTREACHED*/ + } + + } while (cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_disable(void) +{ + dtrace_vtime_state_t state, nstate; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_ACTIVE: + nstate = DTRACE_VTIME_INACTIVE; + break; + + case DTRACE_VTIME_ACTIVE_TNF: + nstate = DTRACE_VTIME_INACTIVE_TNF; + break; + + case DTRACE_VTIME_INACTIVE: + case DTRACE_VTIME_INACTIVE_TNF: + panic("DTrace virtual time already disabled"); + /*NOTREACHED*/ + } + + } while (cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_enable_tnf(void) +{ + dtrace_vtime_state_t state, nstate; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_ACTIVE: + nstate = DTRACE_VTIME_ACTIVE_TNF; + break; + + case DTRACE_VTIME_INACTIVE: + nstate = DTRACE_VTIME_INACTIVE_TNF; + break; + + case DTRACE_VTIME_ACTIVE_TNF: + case DTRACE_VTIME_INACTIVE_TNF: + panic("TNF already active"); + /*NOTREACHED*/ + } + + } while (cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_disable_tnf(void) +{ + dtrace_vtime_state_t state, nstate; + + do { + state = dtrace_vtime_active; + + switch (state) { + case DTRACE_VTIME_ACTIVE_TNF: + nstate = DTRACE_VTIME_ACTIVE; + break; + + case DTRACE_VTIME_INACTIVE_TNF: + nstate = DTRACE_VTIME_INACTIVE; + break; + + case DTRACE_VTIME_ACTIVE: + case DTRACE_VTIME_INACTIVE: + panic("TNF already inactive"); + /*NOTREACHED*/ + } + + } while (cas32((uint32_t *)&dtrace_vtime_active, + state, nstate) != state); +} + +void +dtrace_vtime_switch(kthread_t *next) +{ + dtrace_icookie_t cookie; + hrtime_t ts; + + if (tnf_tracing_active) { + tnf_thread_switch(next); + + if (dtrace_vtime_active == DTRACE_VTIME_INACTIVE_TNF) + return; + } + + cookie = dtrace_interrupt_disable(); + ts = dtrace_gethrtime(); + + if (curthread->t_dtrace_start != 0) { + curthread->t_dtrace_vtime += ts - curthread->t_dtrace_start; + curthread->t_dtrace_start = 0; + } + + next->t_dtrace_start = ts; + + dtrace_interrupt_enable(cookie); +} + +void (*dtrace_fasttrap_fork_ptr)(proc_t *, proc_t *); +void (*dtrace_fasttrap_exec_ptr)(proc_t *); +void (*dtrace_fasttrap_exit_ptr)(proc_t *); + +/* + * This function is called by cfork() in the event that it appears that + * there may be dtrace tracepoints active in the parent process's address + * space. This first confirms the existence of dtrace tracepoints in the + * parent process and calls into the fasttrap module to remove the + * corresponding tracepoints from the child. By knowing that there are + * existing tracepoints, and ensuring they can't be removed, we can rely + * on the fasttrap module remaining loaded. + */ +void +dtrace_fasttrap_fork(proc_t *p, proc_t *cp) +{ + ASSERT(p->p_proc_flag & P_PR_LOCK); + ASSERT(p->p_dtrace_count > 0); + ASSERT(dtrace_fasttrap_fork_ptr != NULL); + + dtrace_fasttrap_fork_ptr(p, cp); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf.h b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf.h new file mode 100644 index 00000000..c23d5977 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf.h @@ -0,0 +1,364 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2004 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _CTF_H +#define _CTF_H + +#ifndef VBOX +# pragma ident "%Z%%M% %I% %E% SMI" +#endif + +#ifndef VBOX +# include <sys/types.h> +#else +# include "VBoxDTraceTypes.h" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * CTF - Compact ANSI-C Type Format + * + * This file format can be used to compactly represent the information needed + * by a debugger to interpret the ANSI-C types used by a given program. + * Traditionally, this kind of information is generated by the compiler when + * invoked with the -g flag and is stored in "stabs" strings or in the more + * modern DWARF format. CTF provides a representation of only the information + * that is relevant to debugging a complex, optimized C program such as the + * operating system kernel in a form that is significantly more compact than + * the equivalent stabs or DWARF representation. The format is data-model + * independent, so consumers do not need different code depending on whether + * they are 32-bit or 64-bit programs. CTF assumes that a standard ELF symbol + * table is available for use in the debugger, and uses the structure and data + * of the symbol table to avoid storing redundant information. The CTF data + * may be compressed on disk or in memory, indicated by a bit in the header. + * CTF may be interpreted in a raw disk file, or it may be stored in an ELF + * section, typically named .SUNW_ctf. Data structures are aligned so that + * a raw CTF file or CTF ELF section may be manipulated using mmap(2). + * + * The CTF file or section itself has the following structure: + * + * +--------+--------+---------+----------+-------+--------+ + * | file | type | data | function | data | string | + * | header | labels | objects | info | types | table | + * +--------+--------+---------+----------+-------+--------+ + * + * The file header stores a magic number and version information, encoding + * flags, and the byte offset of each of the sections relative to the end of the + * header itself. If the CTF data has been uniquified against another set of + * CTF data, a reference to that data also appears in the the header. This + * reference is the name of the label corresponding to the types uniquified + * against. + * + * Following the header is a list of labels, used to group the types included in + * the data types section. Each label is accompanied by a type ID i. A given + * label refers to the group of types whose IDs are in the range [0, i]. + * + * Data object and function records are stored in the same order as they appear + * in the corresponding symbol table, except that symbols marked SHN_UNDEF are + * not stored and symbols that have no type data are padded out with zeroes. + * For each data object, the type ID (a small integer) is recorded. For each + * function, the type ID of the return type and argument types is recorded. + * + * The data types section is a list of variable size records that represent each + * type, in order by their ID. The types themselves form a directed graph, + * where each node may contain one or more outgoing edges to other type nodes, + * denoted by their ID. + * + * Strings are recorded as a string table ID (0 or 1) and a byte offset into the + * string table. String table 0 is the internal CTF string table. String table + * 1 is the external string table, which is the string table associated with the + * ELF symbol table for this object. CTF does not record any strings that are + * already in the symbol table, and the CTF string table does not contain any + * duplicated strings. + * + * If the CTF data has been merged with another parent CTF object, some outgoing + * edges may refer to type nodes that exist in another CTF object. The debugger + * and libctf library are responsible for connecting the appropriate objects + * together so that the full set of types can be explored and manipulated. + */ + +#define CTF_MAX_TYPE 0xffff /* max type identifier value */ +#define CTF_MAX_NAME 0x7fffffff /* max offset into a string table */ +#define CTF_MAX_VLEN 0x3ff /* max struct, union, enum members or args */ +#define CTF_MAX_INTOFF 0xff /* max offset of intrinsic value in bits */ +#define CTF_MAX_INTBITS 0xffff /* max size of an intrinsic in bits */ + +/* See ctf_type_t */ +#define CTF_MAX_SIZE 0xfffe /* max size of a type in bytes */ +#define CTF_LSIZE_SENT 0xffff /* sentinel for ctt_size */ +#define CTF_MAX_LSIZE UINT64_MAX + +typedef struct ctf_preamble { + ushort_t ctp_magic; /* magic number (CTF_MAGIC) */ + uchar_t ctp_version; /* data format version number (CTF_VERSION) */ + uchar_t ctp_flags; /* flags (see below) */ +} ctf_preamble_t; + +typedef struct ctf_header { + ctf_preamble_t cth_preamble; + uint_t cth_parlabel; /* ref to name of parent lbl uniq'd against */ + uint_t cth_parname; /* ref to basename of parent */ + uint_t cth_lbloff; /* offset of label section */ + uint_t cth_objtoff; /* offset of object section */ + uint_t cth_funcoff; /* offset of function section */ + uint_t cth_typeoff; /* offset of type section */ + uint_t cth_stroff; /* offset of string section */ + uint_t cth_strlen; /* length of string section in bytes */ +} ctf_header_t; + +#define cth_magic cth_preamble.ctp_magic +#define cth_version cth_preamble.ctp_version +#define cth_flags cth_preamble.ctp_flags + +#ifdef CTF_OLD_VERSIONS + +typedef struct ctf_header_v1 { + ctf_preamble_t cth_preamble; + uint_t cth_objtoff; + uint_t cth_funcoff; + uint_t cth_typeoff; + uint_t cth_stroff; + uint_t cth_strlen; +} ctf_header_v1_t; + +#endif /* CTF_OLD_VERSIONS */ + +#define CTF_MAGIC 0xcff1 /* magic number identifying header */ + +/* data format version number */ +#define CTF_VERSION_1 1 +#define CTF_VERSION_2 2 +#define CTF_VERSION CTF_VERSION_2 /* current version */ + +#define CTF_F_COMPRESS 0x1 /* data buffer is compressed */ + +typedef struct ctf_lblent { + uint_t ctl_label; /* ref to name of label */ + uint_t ctl_typeidx; /* last type associated with this label */ +} ctf_lblent_t; + +typedef struct ctf_stype { + uint_t ctt_name; /* reference to name in string table */ + ushort_t ctt_info; /* encoded kind, variant length (see below) */ + union { + ushort_t _size; /* size of entire type in bytes */ + ushort_t _type; /* reference to another type */ + } _u; +} ctf_stype_t; + +/* + * type sizes, measured in bytes, come in two flavors. 99% of them fit within + * (USHRT_MAX - 1), and thus can be stored in the ctt_size member of a + * ctf_stype_t. The maximum value for these sizes is CTF_MAX_SIZE. The sizes + * larger than CTF_MAX_SIZE must be stored in the ctt_lsize member of a + * ctf_type_t. Use of this member is indicated by the presence of + * CTF_LSIZE_SENT in ctt_size. + */ +typedef struct ctf_type { + uint_t ctt_name; /* reference to name in string table */ + ushort_t ctt_info; /* encoded kind, variant length (see below) */ + union { + ushort_t _size; /* always CTF_LSIZE_SENT */ + ushort_t _type; /* do not use */ + } _u; + uint_t ctt_lsizehi; /* high 32 bits of type size in bytes */ + uint_t ctt_lsizelo; /* low 32 bits of type size in bytes */ +} ctf_type_t; + +#define ctt_size _u._size /* for fundamental types that have a size */ +#define ctt_type _u._type /* for types that reference another type */ + +/* + * The following macros compose and decompose values for ctt_info and + * ctt_name, as well as other structures that contain name references. + * + * ------------------------ + * ctt_info: | kind | isroot | vlen | + * ------------------------ + * 15 11 10 9 0 + * + * kind = CTF_INFO_KIND(c.ctt_info); <-- CTF_K_* value (see below) + * vlen = CTF_INFO_VLEN(c.ctt_info); <-- length of variable data list + * + * stid = CTF_NAME_STID(c.ctt_name); <-- string table id number (0 or 1) + * offset = CTF_NAME_OFFSET(c.ctt_name); <-- string table byte offset + * + * c.ctt_info = CTF_TYPE_INFO(kind, vlen); + * c.ctt_name = CTF_TYPE_NAME(stid, offset); + */ + +#define CTF_INFO_KIND(info) (((info) & 0xf800) >> 11) +#define CTF_INFO_ISROOT(info) (((info) & 0x0400) >> 10) +#define CTF_INFO_VLEN(info) (((info) & CTF_MAX_VLEN)) + +#define CTF_NAME_STID(name) ((name) >> 31) +#define CTF_NAME_OFFSET(name) ((name) & 0x7fffffff) + +#define CTF_TYPE_INFO(kind, isroot, vlen) \ + (((kind) << 11) | (((isroot) ? 1 : 0) << 10) | ((vlen) & CTF_MAX_VLEN)) + +#define CTF_TYPE_NAME(stid, offset) \ + (((stid) << 31) | ((offset) & 0x7fffffff)) + +#define CTF_TYPE_ISPARENT(id) ((id) < 0x8000) +#define CTF_TYPE_ISCHILD(id) ((id) > 0x7fff) + +#define CTF_TYPE_TO_INDEX(id) ((id) & 0x7fff) +#define CTF_INDEX_TO_TYPE(id, child) ((child) ? ((id) | 0x8000) : (id)) +#define CTF_PARENT_SHIFT 15 + +#define CTF_STRTAB_0 0 /* symbolic define for string table id 0 */ +#define CTF_STRTAB_1 1 /* symbolic define for string table id 1 */ + +#define CTF_TYPE_LSIZE(cttp) \ + (((uint64_t)(cttp)->ctt_lsizehi) << 32 | (cttp)->ctt_lsizelo) +#define CTF_SIZE_TO_LSIZE_HI(size) ((uint32_t)((uint64_t)(size) >> 32)) +#define CTF_SIZE_TO_LSIZE_LO(size) ((uint32_t)(size)) + +#ifdef CTF_OLD_VERSIONS + +#define CTF_INFO_KIND_V1(info) (((info) & 0xf000) >> 12) +#define CTF_INFO_ISROOT_V1(info) (((info) & 0x0800) >> 11) +#define CTF_INFO_VLEN_V1(info) (((info) & 0x07ff)) + +#define CTF_TYPE_INFO_V1(kind, isroot, vlen) \ + (((kind) << 12) | (((isroot) ? 1 : 0) << 11) | ((vlen) & 0x07ff)) + +#endif /* CTF_OLD_VERSIONS */ + +/* + * Values for CTF_TYPE_KIND(). If the kind has an associated data list, + * CTF_INFO_VLEN() will extract the number of elements in the list, and + * the type of each element is shown in the comments below. + */ +#define CTF_K_UNKNOWN 0 /* unknown type (used for padding) */ +#define CTF_K_INTEGER 1 /* variant data is CTF_INT_DATA() (see below) */ +#define CTF_K_FLOAT 2 /* variant data is CTF_FP_DATA() (see below) */ +#define CTF_K_POINTER 3 /* ctt_type is referenced type */ +#define CTF_K_ARRAY 4 /* variant data is single ctf_array_t */ +#define CTF_K_FUNCTION 5 /* ctt_type is return type, variant data is */ + /* list of argument types (ushort_t's) */ +#define CTF_K_STRUCT 6 /* variant data is list of ctf_member_t's */ +#define CTF_K_UNION 7 /* variant data is list of ctf_member_t's */ +#define CTF_K_ENUM 8 /* variant data is list of ctf_enum_t's */ +#define CTF_K_FORWARD 9 /* no additional data; ctt_name is tag */ +#define CTF_K_TYPEDEF 10 /* ctt_type is referenced type */ +#define CTF_K_VOLATILE 11 /* ctt_type is base type */ +#define CTF_K_CONST 12 /* ctt_type is base type */ +#define CTF_K_RESTRICT 13 /* ctt_type is base type */ + +#define CTF_K_MAX 31 /* Maximum possible CTF_K_* value */ + +/* + * Values for ctt_type when kind is CTF_K_INTEGER. The flags, offset in bits, + * and size in bits are encoded as a single word using the following macros. + */ +#define CTF_INT_ENCODING(data) (((data) & 0xff000000) >> 24) +#define CTF_INT_OFFSET(data) (((data) & 0x00ff0000) >> 16) +#define CTF_INT_BITS(data) (((data) & 0x0000ffff)) + +#define CTF_INT_DATA(encoding, offset, bits) \ + (((encoding) << 24) | ((offset) << 16) | (bits)) + +#define CTF_INT_SIGNED 0x01 /* integer is signed (otherwise unsigned) */ +#define CTF_INT_CHAR 0x02 /* character display format */ +#define CTF_INT_BOOL 0x04 /* boolean display format */ +#define CTF_INT_VARARGS 0x08 /* varargs display format */ + +/* + * Values for ctt_type when kind is CTF_K_FLOAT. The encoding, offset in bits, + * and size in bits are encoded as a single word using the following macros. + */ +#define CTF_FP_ENCODING(data) (((data) & 0xff000000) >> 24) +#define CTF_FP_OFFSET(data) (((data) & 0x00ff0000) >> 16) +#define CTF_FP_BITS(data) (((data) & 0x0000ffff)) + +#define CTF_FP_DATA(encoding, offset, bits) \ + (((encoding) << 24) | ((offset) << 16) | (bits)) + +#define CTF_FP_SINGLE 1 /* IEEE 32-bit float encoding */ +#define CTF_FP_DOUBLE 2 /* IEEE 64-bit float encoding */ +#define CTF_FP_CPLX 3 /* Complex encoding */ +#define CTF_FP_DCPLX 4 /* Double complex encoding */ +#define CTF_FP_LDCPLX 5 /* Long double complex encoding */ +#define CTF_FP_LDOUBLE 6 /* Long double encoding */ +#define CTF_FP_INTRVL 7 /* Interval (2x32-bit) encoding */ +#define CTF_FP_DINTRVL 8 /* Double interval (2x64-bit) encoding */ +#define CTF_FP_LDINTRVL 9 /* Long double interval (2x128-bit) encoding */ +#define CTF_FP_IMAGRY 10 /* Imaginary (32-bit) encoding */ +#define CTF_FP_DIMAGRY 11 /* Long imaginary (64-bit) encoding */ +#define CTF_FP_LDIMAGRY 12 /* Long double imaginary (128-bit) encoding */ + +#define CTF_FP_MAX 12 /* Maximum possible CTF_FP_* value */ + +typedef struct ctf_array { + ushort_t cta_contents; /* reference to type of array contents */ + ushort_t cta_index; /* reference to type of array index */ + uint_t cta_nelems; /* number of elements */ +} ctf_array_t; + +/* + * Most structure members have bit offsets that can be expressed using a + * short. Some don't. ctf_member_t is used for structs which cannot + * contain any of these large offsets, whereas ctf_lmember_t is used in the + * latter case. If ctt_size for a given struct is >= 8192 bytes, all members + * will be stored as type ctf_lmember_t. + */ + +#define CTF_LSTRUCT_THRESH 8192 + +typedef struct ctf_member { + uint_t ctm_name; /* reference to name in string table */ + ushort_t ctm_type; /* reference to type of member */ + ushort_t ctm_offset; /* offset of this member in bits */ +} ctf_member_t; + +typedef struct ctf_lmember { + uint_t ctlm_name; /* reference to name in string table */ + ushort_t ctlm_type; /* reference to type of member */ + ushort_t ctlm_pad; /* padding */ + uint_t ctlm_offsethi; /* high 32 bits of member offset in bits */ + uint_t ctlm_offsetlo; /* low 32 bits of member offset in bits */ +} ctf_lmember_t; + +#define CTF_LMEM_OFFSET(ctlmp) \ + (((uint64_t)(ctlmp)->ctlm_offsethi) << 32 | (ctlmp)->ctlm_offsetlo) +#define CTF_OFFSET_TO_LMEMHI(offset) ((uint32_t)((uint64_t)(offset) >> 32)) +#define CTF_OFFSET_TO_LMEMLO(offset) ((uint32_t)(offset)) + +typedef struct ctf_enum { + uint_t cte_name; /* reference to name in string table */ + int cte_value; /* value associated with this name */ +} ctf_enum_t; + +#ifdef __cplusplus +} +#endif + +#endif /* _CTF_H */ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf_api.h b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf_api.h new file mode 100644 index 00000000..7baa37e6 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/ctf_api.h @@ -0,0 +1,251 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License, Version 1.0 only + * (the "License"). You may not use this file except in compliance + * with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2005 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +/* + * This header file defines the interfaces available from the CTF debugger + * library, libctf, and an equivalent kernel module. This API can be used by + * a debugger to operate on data in the Compact ANSI-C Type Format (CTF). + * This is NOT a public interface, although it may eventually become one in + * the fullness of time after we gain more experience with the interfaces. + * + * In the meantime, be aware that any program linked with this API in this + * release of Solaris is almost guaranteed to break in the next release. + * + * In short, do not user this header file or the CTF routines for any purpose. + */ + +#ifndef _CTF_API_H +#define _CTF_API_H + +#ifndef VBOX +# pragma ident "%Z%%M% %I% %E% SMI" +#endif + +#ifndef VBOX +# include <sys/types.h> +# include <sys/param.h> +# include <sys/elf.h> +#else +# include "VBoxDTraceTypes.h" +#endif +#include <sys/ctf.h> + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * Clients can open one or more CTF containers and obtain a pointer to an + * opaque ctf_file_t. Types are identified by an opaque ctf_id_t token. + * These opaque definitions allow libctf to evolve without breaking clients. + */ +typedef struct ctf_file ctf_file_t; +typedef long ctf_id_t; + +/* + * If the debugger needs to provide the CTF library with a set of raw buffers + * for use as the CTF data, symbol table, and string table, it can do so by + * filling in ctf_sect_t structures and passing them to ctf_bufopen(): + */ +typedef struct ctf_sect { + const char *cts_name; /* section name (if any) */ + ulong_t cts_type; /* section type (ELF SHT_... value) */ + ulong_t cts_flags; /* section flags (ELF SHF_... value) */ + const void *cts_data; /* pointer to section data */ + size_t cts_size; /* size of data in bytes */ + size_t cts_entsize; /* size of each section entry (symtab only) */ +#ifndef VBOX + off64_t cts_offset; /* file offset of this section (if any) */ +#else + RTFOFF cts_offset; /* file offset of this section (if any) */ +#endif +} ctf_sect_t; + +/* + * Encoding information for integers, floating-point values, and certain other + * intrinsics can be obtained by calling ctf_type_encoding(), below. The flags + * field will contain values appropriate for the type defined in <sys/ctf.h>. + */ +typedef struct ctf_encoding { + uint_t cte_format; /* data format (CTF_INT_* or CTF_FP_* flags) */ + uint_t cte_offset; /* offset of value in bits */ + uint_t cte_bits; /* size of storage in bits */ +} ctf_encoding_t; + +typedef struct ctf_membinfo { + ctf_id_t ctm_type; /* type of struct or union member */ + ulong_t ctm_offset; /* offset of member in bits */ +} ctf_membinfo_t; + +typedef struct ctf_arinfo { + ctf_id_t ctr_contents; /* type of array contents */ + ctf_id_t ctr_index; /* type of array index */ + uint_t ctr_nelems; /* number of elements */ +} ctf_arinfo_t; + +typedef struct ctf_funcinfo { + ctf_id_t ctc_return; /* function return type */ + uint_t ctc_argc; /* number of typed arguments to function */ + uint_t ctc_flags; /* function attributes (see below) */ +} ctf_funcinfo_t; + +typedef struct ctf_lblinfo { + ctf_id_t ctb_typeidx; /* last type associated with the label */ +} ctf_lblinfo_t; + +#define CTF_FUNC_VARARG 0x1 /* function arguments end with varargs */ + +/* + * Functions that return integer status or a ctf_id_t use the following value + * to indicate failure. ctf_errno() can be used to obtain an error code. + */ +#define CTF_ERR (-1L) + +/* + * The CTF data model is inferred to be the caller's data model or the data + * model of the given object, unless ctf_setmodel() is explicitly called. + */ +#define CTF_MODEL_ILP32 1 /* object data model is ILP32 */ +#define CTF_MODEL_LP64 2 /* object data model is LP64 */ +#ifdef _LP64 +#define CTF_MODEL_NATIVE CTF_MODEL_LP64 +#else +#define CTF_MODEL_NATIVE CTF_MODEL_ILP32 +#endif + +/* + * Dynamic CTF containers can be created using ctf_create(). The ctf_add_* + * routines can be used to add new definitions to the dynamic container. + * New types are labeled as root or non-root to determine whether they are + * visible at the top-level program scope when subsequently doing a lookup. + */ +#define CTF_ADD_NONROOT 0 /* type only visible in nested scope */ +#define CTF_ADD_ROOT 1 /* type visible at top-level scope */ + +/* + * These typedefs are used to define the signature for callback functions + * that can be used with the iteration and visit functions below: + */ +typedef int ctf_visit_f(const char *, ctf_id_t, ulong_t, int, void *); +typedef int ctf_member_f(const char *, ctf_id_t, ulong_t, void *); +typedef int ctf_enum_f(const char *, int, void *); +typedef int ctf_type_f(ctf_id_t, void *); +typedef int ctf_label_f(const char *, const ctf_lblinfo_t *, void *); + +extern ctf_file_t *ctf_bufopen(const ctf_sect_t *, const ctf_sect_t *, + const ctf_sect_t *, int *); +extern ctf_file_t *ctf_fdopen(int, int *); +extern ctf_file_t *ctf_open(const char *, int *); +extern ctf_file_t *ctf_create(int *); +extern void ctf_close(ctf_file_t *); + +extern ctf_file_t *ctf_parent_file(ctf_file_t *); +extern const char *ctf_parent_name(ctf_file_t *); + +extern int ctf_import(ctf_file_t *, ctf_file_t *); +extern int ctf_setmodel(ctf_file_t *, int); +extern int ctf_getmodel(ctf_file_t *); + +extern void ctf_setspecific(ctf_file_t *, void *); +extern void *ctf_getspecific(ctf_file_t *); + +extern int ctf_errno(ctf_file_t *); +extern const char *ctf_errmsg(int); +extern int ctf_version(int); + +extern int ctf_func_info(ctf_file_t *, ulong_t, ctf_funcinfo_t *); +extern int ctf_func_args(ctf_file_t *, ulong_t, uint_t, ctf_id_t *); + +extern ctf_id_t ctf_lookup_by_name(ctf_file_t *, const char *); +extern ctf_id_t ctf_lookup_by_symbol(ctf_file_t *, ulong_t); + +extern ctf_id_t ctf_type_resolve(ctf_file_t *, ctf_id_t); +extern ssize_t ctf_type_lname(ctf_file_t *, ctf_id_t, char *, size_t); +extern char *ctf_type_name(ctf_file_t *, ctf_id_t, char *, size_t); +extern ssize_t ctf_type_size(ctf_file_t *, ctf_id_t); +extern ssize_t ctf_type_align(ctf_file_t *, ctf_id_t); +extern int ctf_type_kind(ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_type_reference(ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_type_pointer(ctf_file_t *, ctf_id_t); +extern int ctf_type_encoding(ctf_file_t *, ctf_id_t, ctf_encoding_t *); +extern int ctf_type_visit(ctf_file_t *, ctf_id_t, ctf_visit_f *, void *); +extern int ctf_type_cmp(ctf_file_t *, ctf_id_t, ctf_file_t *, ctf_id_t); +extern int ctf_type_compat(ctf_file_t *, ctf_id_t, ctf_file_t *, ctf_id_t); + +extern int ctf_member_info(ctf_file_t *, ctf_id_t, const char *, + ctf_membinfo_t *); +extern int ctf_array_info(ctf_file_t *, ctf_id_t, ctf_arinfo_t *); + +extern const char *ctf_enum_name(ctf_file_t *, ctf_id_t, int); +extern int ctf_enum_value(ctf_file_t *, ctf_id_t, const char *, int *); + +extern const char *ctf_label_topmost(ctf_file_t *); +extern int ctf_label_info(ctf_file_t *, const char *, ctf_lblinfo_t *); + +extern int ctf_member_iter(ctf_file_t *, ctf_id_t, ctf_member_f *, void *); +extern int ctf_enum_iter(ctf_file_t *, ctf_id_t, ctf_enum_f *, void *); +extern int ctf_type_iter(ctf_file_t *, ctf_type_f *, void *); +extern int ctf_label_iter(ctf_file_t *, ctf_label_f *, void *); + +extern ctf_id_t ctf_add_array(ctf_file_t *, uint_t, const ctf_arinfo_t *); +extern ctf_id_t ctf_add_const(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_enum(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_float(ctf_file_t *, uint_t, + const char *, const ctf_encoding_t *); +extern ctf_id_t ctf_add_forward(ctf_file_t *, uint_t, const char *, uint_t); +extern ctf_id_t ctf_add_function(ctf_file_t *, uint_t, + const ctf_funcinfo_t *, const ctf_id_t *); +extern ctf_id_t ctf_add_integer(ctf_file_t *, uint_t, + const char *, const ctf_encoding_t *); +extern ctf_id_t ctf_add_pointer(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_type(ctf_file_t *, ctf_file_t *, ctf_id_t); +extern ctf_id_t ctf_add_typedef(ctf_file_t *, uint_t, const char *, ctf_id_t); +extern ctf_id_t ctf_add_restrict(ctf_file_t *, uint_t, ctf_id_t); +extern ctf_id_t ctf_add_struct(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_union(ctf_file_t *, uint_t, const char *); +extern ctf_id_t ctf_add_volatile(ctf_file_t *, uint_t, ctf_id_t); + +extern int ctf_add_enumerator(ctf_file_t *, ctf_id_t, const char *, int); +extern int ctf_add_member(ctf_file_t *, ctf_id_t, const char *, ctf_id_t); + +extern int ctf_set_array(ctf_file_t *, ctf_id_t, const ctf_arinfo_t *); + +extern int ctf_update(ctf_file_t *); +extern int ctf_discard(ctf_file_t *); +extern int ctf_write(ctf_file_t *, int); + +#ifdef _KERNEL + +struct module; +extern ctf_file_t *ctf_modopen(struct module *, int *); + +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _CTF_API_H */ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace.h b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace.h new file mode 100644 index 00000000..868a94b2 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace.h @@ -0,0 +1,2276 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_DTRACE_H +#define _SYS_DTRACE_H + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * DTrace Dynamic Tracing Software: Kernel Interfaces + * + * Note: The contents of this file are private to the implementation of the + * Solaris system and DTrace subsystem and are subject to change at any time + * without notice. Applications and drivers using these interfaces will fail + * to run on future releases. These interfaces should not be used for any + * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB). + * Please refer to the "Solaris Dynamic Tracing Guide" for more information. + */ + +#ifndef _ASM + +#ifndef VBOX +#include <sys/types.h> +#include <sys/modctl.h> +#include <sys/processor.h> +#include <sys/systm.h> +#include <sys/cyclic.h> +#include <sys/int_limits.h> +#else +# include <VBoxDTraceTypes.h> +# include <sys/ctf_api.h> +#endif + +/* + * DTrace Universal Constants and Typedefs + */ +#ifdef VBOX +#define DTRACE_CPUALL ((processorid_t)-1) /* all CPUs */ +#else +#define DTRACE_CPUALL -1 /* all CPUs */ +#endif +#define DTRACE_IDNONE 0 /* invalid probe identifier */ +#define DTRACE_EPIDNONE 0 /* invalid enabled probe identifier */ +#define DTRACE_AGGIDNONE 0 /* invalid aggregation identifier */ +#define DTRACE_AGGVARIDNONE 0 /* invalid aggregation variable ID */ +#define DTRACE_CACHEIDNONE 0 /* invalid predicate cache */ +#define DTRACE_PROVNONE 0 /* invalid provider identifier */ +#define DTRACE_METAPROVNONE 0 /* invalid meta-provider identifier */ +#define DTRACE_ARGNONE -1 /* invalid argument index */ + +#define DTRACE_PROVNAMELEN 64 +#define DTRACE_MODNAMELEN 64 +#define DTRACE_FUNCNAMELEN 128 +#define DTRACE_NAMELEN 64 +#define DTRACE_FULLNAMELEN (DTRACE_PROVNAMELEN + DTRACE_MODNAMELEN + \ + DTRACE_FUNCNAMELEN + DTRACE_NAMELEN + 4) +#define DTRACE_ARGTYPELEN 128 + +typedef uint32_t dtrace_id_t; /* probe identifier */ +typedef uint32_t dtrace_epid_t; /* enabled probe identifier */ +typedef uint32_t dtrace_aggid_t; /* aggregation identifier */ +typedef int64_t dtrace_aggvarid_t; /* aggregation variable identifier */ +typedef uint16_t dtrace_actkind_t; /* action kind */ +typedef int64_t dtrace_optval_t; /* option value */ +typedef uint32_t dtrace_cacheid_t; /* predicate cache identifier */ + +typedef enum dtrace_probespec { + DTRACE_PROBESPEC_NONE = -1, + DTRACE_PROBESPEC_PROVIDER = 0, + DTRACE_PROBESPEC_MOD, + DTRACE_PROBESPEC_FUNC, + DTRACE_PROBESPEC_NAME +} dtrace_probespec_t; + +/* + * DTrace Intermediate Format (DIF) + * + * The following definitions describe the DTrace Intermediate Format (DIF), a + * a RISC-like instruction set and program encoding used to represent + * predicates and actions that can be bound to DTrace probes. The constants + * below defining the number of available registers are suggested minimums; the + * compiler should use DTRACEIOC_CONF to dynamically obtain the number of + * registers provided by the current DTrace implementation. + */ +#define DIF_VERSION_1 1 /* DIF version 1: Solaris 10 Beta */ +#define DIF_VERSION_2 2 /* DIF version 2: Solaris 10 FCS */ +#define DIF_VERSION DIF_VERSION_2 /* latest DIF instruction set version */ +#define DIF_DIR_NREGS 8 /* number of DIF integer registers */ +#define DIF_DTR_NREGS 8 /* number of DIF tuple registers */ + +#define DIF_OP_OR 1 /* or r1, r2, rd */ +#define DIF_OP_XOR 2 /* xor r1, r2, rd */ +#define DIF_OP_AND 3 /* and r1, r2, rd */ +#define DIF_OP_SLL 4 /* sll r1, r2, rd */ +#define DIF_OP_SRL 5 /* srl r1, r2, rd */ +#define DIF_OP_SUB 6 /* sub r1, r2, rd */ +#define DIF_OP_ADD 7 /* add r1, r2, rd */ +#define DIF_OP_MUL 8 /* mul r1, r2, rd */ +#define DIF_OP_SDIV 9 /* sdiv r1, r2, rd */ +#define DIF_OP_UDIV 10 /* udiv r1, r2, rd */ +#define DIF_OP_SREM 11 /* srem r1, r2, rd */ +#define DIF_OP_UREM 12 /* urem r1, r2, rd */ +#define DIF_OP_NOT 13 /* not r1, rd */ +#define DIF_OP_MOV 14 /* mov r1, rd */ +#define DIF_OP_CMP 15 /* cmp r1, r2 */ +#define DIF_OP_TST 16 /* tst r1 */ +#define DIF_OP_BA 17 /* ba label */ +#define DIF_OP_BE 18 /* be label */ +#define DIF_OP_BNE 19 /* bne label */ +#define DIF_OP_BG 20 /* bg label */ +#define DIF_OP_BGU 21 /* bgu label */ +#define DIF_OP_BGE 22 /* bge label */ +#define DIF_OP_BGEU 23 /* bgeu label */ +#define DIF_OP_BL 24 /* bl label */ +#define DIF_OP_BLU 25 /* blu label */ +#define DIF_OP_BLE 26 /* ble label */ +#define DIF_OP_BLEU 27 /* bleu label */ +#define DIF_OP_LDSB 28 /* ldsb [r1], rd */ +#define DIF_OP_LDSH 29 /* ldsh [r1], rd */ +#define DIF_OP_LDSW 30 /* ldsw [r1], rd */ +#define DIF_OP_LDUB 31 /* ldub [r1], rd */ +#define DIF_OP_LDUH 32 /* lduh [r1], rd */ +#define DIF_OP_LDUW 33 /* lduw [r1], rd */ +#define DIF_OP_LDX 34 /* ldx [r1], rd */ +#define DIF_OP_RET 35 /* ret rd */ +#define DIF_OP_NOP 36 /* nop */ +#define DIF_OP_SETX 37 /* setx intindex, rd */ +#define DIF_OP_SETS 38 /* sets strindex, rd */ +#define DIF_OP_SCMP 39 /* scmp r1, r2 */ +#define DIF_OP_LDGA 40 /* ldga var, ri, rd */ +#define DIF_OP_LDGS 41 /* ldgs var, rd */ +#define DIF_OP_STGS 42 /* stgs var, rs */ +#define DIF_OP_LDTA 43 /* ldta var, ri, rd */ +#define DIF_OP_LDTS 44 /* ldts var, rd */ +#define DIF_OP_STTS 45 /* stts var, rs */ +#define DIF_OP_SRA 46 /* sra r1, r2, rd */ +#define DIF_OP_CALL 47 /* call subr, rd */ +#define DIF_OP_PUSHTR 48 /* pushtr type, rs, rr */ +#define DIF_OP_PUSHTV 49 /* pushtv type, rs, rv */ +#define DIF_OP_POPTS 50 /* popts */ +#define DIF_OP_FLUSHTS 51 /* flushts */ +#define DIF_OP_LDGAA 52 /* ldgaa var, rd */ +#define DIF_OP_LDTAA 53 /* ldtaa var, rd */ +#define DIF_OP_STGAA 54 /* stgaa var, rs */ +#define DIF_OP_STTAA 55 /* sttaa var, rs */ +#define DIF_OP_LDLS 56 /* ldls var, rd */ +#define DIF_OP_STLS 57 /* stls var, rs */ +#define DIF_OP_ALLOCS 58 /* allocs r1, rd */ +#define DIF_OP_COPYS 59 /* copys r1, r2, rd */ +#define DIF_OP_STB 60 /* stb r1, [rd] */ +#define DIF_OP_STH 61 /* sth r1, [rd] */ +#define DIF_OP_STW 62 /* stw r1, [rd] */ +#define DIF_OP_STX 63 /* stx r1, [rd] */ +#define DIF_OP_ULDSB 64 /* uldsb [r1], rd */ +#define DIF_OP_ULDSH 65 /* uldsh [r1], rd */ +#define DIF_OP_ULDSW 66 /* uldsw [r1], rd */ +#define DIF_OP_ULDUB 67 /* uldub [r1], rd */ +#define DIF_OP_ULDUH 68 /* ulduh [r1], rd */ +#define DIF_OP_ULDUW 69 /* ulduw [r1], rd */ +#define DIF_OP_ULDX 70 /* uldx [r1], rd */ +#define DIF_OP_RLDSB 71 /* rldsb [r1], rd */ +#define DIF_OP_RLDSH 72 /* rldsh [r1], rd */ +#define DIF_OP_RLDSW 73 /* rldsw [r1], rd */ +#define DIF_OP_RLDUB 74 /* rldub [r1], rd */ +#define DIF_OP_RLDUH 75 /* rlduh [r1], rd */ +#define DIF_OP_RLDUW 76 /* rlduw [r1], rd */ +#define DIF_OP_RLDX 77 /* rldx [r1], rd */ +#define DIF_OP_XLATE 78 /* xlate xlrindex, rd */ +#define DIF_OP_XLARG 79 /* xlarg xlrindex, rd */ + +#define DIF_INTOFF_MAX 0xffff /* highest integer table offset */ +#define DIF_STROFF_MAX 0xffff /* highest string table offset */ +#define DIF_REGISTER_MAX 0xff /* highest register number */ +#define DIF_VARIABLE_MAX 0xffff /* highest variable identifier */ +#define DIF_SUBROUTINE_MAX 0xffff /* highest subroutine code */ + +#define DIF_VAR_ARRAY_MIN 0x0000 /* lowest numbered array variable */ +#define DIF_VAR_ARRAY_UBASE 0x0080 /* lowest user-defined array */ +#define DIF_VAR_ARRAY_MAX 0x00ff /* highest numbered array variable */ + +#define DIF_VAR_OTHER_MIN 0x0100 /* lowest numbered scalar or assc */ +#define DIF_VAR_OTHER_UBASE 0x0500 /* lowest user-defined scalar or assc */ +#define DIF_VAR_OTHER_MAX 0xffff /* highest numbered scalar or assc */ + +#define DIF_VAR_ARGS 0x0000 /* arguments array */ +#define DIF_VAR_REGS 0x0001 /* registers array */ +#define DIF_VAR_UREGS 0x0002 /* user registers array */ +#define DIF_VAR_CURTHREAD 0x0100 /* thread pointer */ +#define DIF_VAR_TIMESTAMP 0x0101 /* timestamp */ +#define DIF_VAR_VTIMESTAMP 0x0102 /* virtual timestamp */ +#define DIF_VAR_IPL 0x0103 /* interrupt priority level */ +#define DIF_VAR_EPID 0x0104 /* enabled probe ID */ +#define DIF_VAR_ID 0x0105 /* probe ID */ +#define DIF_VAR_ARG0 0x0106 /* first argument */ +#define DIF_VAR_ARG1 0x0107 /* second argument */ +#define DIF_VAR_ARG2 0x0108 /* third argument */ +#define DIF_VAR_ARG3 0x0109 /* fourth argument */ +#define DIF_VAR_ARG4 0x010a /* fifth argument */ +#define DIF_VAR_ARG5 0x010b /* sixth argument */ +#define DIF_VAR_ARG6 0x010c /* seventh argument */ +#define DIF_VAR_ARG7 0x010d /* eighth argument */ +#define DIF_VAR_ARG8 0x010e /* ninth argument */ +#define DIF_VAR_ARG9 0x010f /* tenth argument */ +#define DIF_VAR_STACKDEPTH 0x0110 /* stack depth */ +#define DIF_VAR_CALLER 0x0111 /* caller */ +#define DIF_VAR_PROBEPROV 0x0112 /* probe provider */ +#define DIF_VAR_PROBEMOD 0x0113 /* probe module */ +#define DIF_VAR_PROBEFUNC 0x0114 /* probe function */ +#define DIF_VAR_PROBENAME 0x0115 /* probe name */ +#define DIF_VAR_PID 0x0116 /* process ID */ +#define DIF_VAR_TID 0x0117 /* (per-process) thread ID */ +#define DIF_VAR_EXECNAME 0x0118 /* name of executable */ +#define DIF_VAR_ZONENAME 0x0119 /* zone name associated with process */ +#define DIF_VAR_WALLTIMESTAMP 0x011a /* wall-clock timestamp */ +#define DIF_VAR_USTACKDEPTH 0x011b /* user-land stack depth */ +#define DIF_VAR_UCALLER 0x011c /* user-level caller */ +#define DIF_VAR_PPID 0x011d /* parent process ID */ +#define DIF_VAR_UID 0x011e /* process user ID */ +#define DIF_VAR_GID 0x011f /* process group ID */ +#define DIF_VAR_ERRNO 0x0120 /* thread errno */ + +#define DIF_SUBR_RAND 0 +#define DIF_SUBR_MUTEX_OWNED 1 +#define DIF_SUBR_MUTEX_OWNER 2 +#define DIF_SUBR_MUTEX_TYPE_ADAPTIVE 3 +#define DIF_SUBR_MUTEX_TYPE_SPIN 4 +#define DIF_SUBR_RW_READ_HELD 5 +#define DIF_SUBR_RW_WRITE_HELD 6 +#define DIF_SUBR_RW_ISWRITER 7 +#define DIF_SUBR_COPYIN 8 +#define DIF_SUBR_COPYINSTR 9 +#define DIF_SUBR_SPECULATION 10 +#define DIF_SUBR_PROGENYOF 11 +#define DIF_SUBR_STRLEN 12 +#define DIF_SUBR_COPYOUT 13 +#define DIF_SUBR_COPYOUTSTR 14 +#define DIF_SUBR_ALLOCA 15 +#define DIF_SUBR_BCOPY 16 +#define DIF_SUBR_COPYINTO 17 +#define DIF_SUBR_MSGDSIZE 18 +#define DIF_SUBR_MSGSIZE 19 +#define DIF_SUBR_GETMAJOR 20 +#define DIF_SUBR_GETMINOR 21 +#define DIF_SUBR_DDI_PATHNAME 22 +#define DIF_SUBR_STRJOIN 23 +#define DIF_SUBR_LLTOSTR 24 +#define DIF_SUBR_BASENAME 25 +#define DIF_SUBR_DIRNAME 26 +#define DIF_SUBR_CLEANPATH 27 +#define DIF_SUBR_STRCHR 28 +#define DIF_SUBR_STRRCHR 29 +#define DIF_SUBR_STRSTR 30 +#define DIF_SUBR_STRTOK 31 +#define DIF_SUBR_SUBSTR 32 +#define DIF_SUBR_INDEX 33 +#define DIF_SUBR_RINDEX 34 +#define DIF_SUBR_HTONS 35 +#define DIF_SUBR_HTONL 36 +#define DIF_SUBR_HTONLL 37 +#define DIF_SUBR_NTOHS 38 +#define DIF_SUBR_NTOHL 39 +#define DIF_SUBR_NTOHLL 40 +#define DIF_SUBR_INET_NTOP 41 +#define DIF_SUBR_INET_NTOA 42 +#define DIF_SUBR_INET_NTOA6 43 + +#define DIF_SUBR_MAX 43 /* max subroutine value */ + +typedef uint32_t dif_instr_t; + +#define DIF_INSTR_OP(i) (((i) >> 24) & 0xff) +#define DIF_INSTR_R1(i) (((i) >> 16) & 0xff) +#define DIF_INSTR_R2(i) (((i) >> 8) & 0xff) +#define DIF_INSTR_RD(i) ((i) & 0xff) +#define DIF_INSTR_RS(i) ((i) & 0xff) +#define DIF_INSTR_LABEL(i) ((i) & 0xffffff) +#define DIF_INSTR_VAR(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_INTEGER(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_STRING(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_SUBR(i) (((i) >> 8) & 0xffff) +#define DIF_INSTR_TYPE(i) (((i) >> 16) & 0xff) +#define DIF_INSTR_XLREF(i) (((i) >> 8) & 0xffff) + +#define DIF_INSTR_FMT(op, r1, r2, d) \ + (((op) << 24) | ((r1) << 16) | ((r2) << 8) | (d)) + +#define DIF_INSTR_NOT(r1, d) (DIF_INSTR_FMT(DIF_OP_NOT, r1, 0, d)) +#define DIF_INSTR_MOV(r1, d) (DIF_INSTR_FMT(DIF_OP_MOV, r1, 0, d)) +#define DIF_INSTR_CMP(op, r1, r2) (DIF_INSTR_FMT(op, r1, r2, 0)) +#define DIF_INSTR_TST(r1) (DIF_INSTR_FMT(DIF_OP_TST, r1, 0, 0)) +#define DIF_INSTR_BRANCH(op, label) (((op) << 24) | (label)) +#define DIF_INSTR_LOAD(op, r1, d) (DIF_INSTR_FMT(op, r1, 0, d)) +#define DIF_INSTR_STORE(op, r1, d) (DIF_INSTR_FMT(op, r1, 0, d)) +#define DIF_INSTR_SETX(i, d) ((DIF_OP_SETX << 24) | ((i) << 8) | (d)) +#define DIF_INSTR_SETS(s, d) ((DIF_OP_SETS << 24) | ((s) << 8) | (d)) +#define DIF_INSTR_RET(d) (DIF_INSTR_FMT(DIF_OP_RET, 0, 0, d)) +#define DIF_INSTR_NOP (DIF_OP_NOP << 24) +#define DIF_INSTR_LDA(op, v, r, d) (DIF_INSTR_FMT(op, v, r, d)) +#define DIF_INSTR_LDV(op, v, d) (((op) << 24) | ((v) << 8) | (d)) +#define DIF_INSTR_STV(op, v, rs) (((op) << 24) | ((v) << 8) | (rs)) +#define DIF_INSTR_CALL(s, d) ((DIF_OP_CALL << 24) | ((s) << 8) | (d)) +#define DIF_INSTR_PUSHTS(op, t, r2, rs) (DIF_INSTR_FMT(op, t, r2, rs)) +#define DIF_INSTR_POPTS (DIF_OP_POPTS << 24) +#define DIF_INSTR_FLUSHTS (DIF_OP_FLUSHTS << 24) +#define DIF_INSTR_ALLOCS(r1, d) (DIF_INSTR_FMT(DIF_OP_ALLOCS, r1, 0, d)) +#define DIF_INSTR_COPYS(r1, r2, d) (DIF_INSTR_FMT(DIF_OP_COPYS, r1, r2, d)) +#define DIF_INSTR_XLATE(op, r, d) (((op) << 24) | ((r) << 8) | (d)) + +#define DIF_REG_R0 0 /* %r0 is always set to zero */ + +/* + * A DTrace Intermediate Format Type (DIF Type) is used to represent the types + * of variables, function and associative array arguments, and the return type + * for each DIF object (shown below). It contains a description of the type, + * its size in bytes, and a module identifier. + */ +typedef struct dtrace_diftype { + uint8_t dtdt_kind; /* type kind (see below) */ + uint8_t dtdt_ckind; /* type kind in CTF */ + uint8_t dtdt_flags; /* type flags (see below) */ + uint8_t dtdt_pad; /* reserved for future use */ + uint32_t dtdt_size; /* type size in bytes (unless string) */ +} dtrace_diftype_t; + +#define DIF_TYPE_CTF 0 /* type is a CTF type */ +#define DIF_TYPE_STRING 1 /* type is a D string */ + +#define DIF_TF_BYREF 0x1 /* type is passed by reference */ + +/* + * A DTrace Intermediate Format variable record is used to describe each of the + * variables referenced by a given DIF object. It contains an integer variable + * identifier along with variable scope and properties, as shown below. The + * size of this structure must be sizeof (int) aligned. + */ +typedef struct dtrace_difv { + uint32_t dtdv_name; /* variable name index in dtdo_strtab */ + uint32_t dtdv_id; /* variable reference identifier */ + uint8_t dtdv_kind; /* variable kind (see below) */ + uint8_t dtdv_scope; /* variable scope (see below) */ + uint16_t dtdv_flags; /* variable flags (see below) */ + dtrace_diftype_t dtdv_type; /* variable type (see above) */ +} dtrace_difv_t; + +#define DIFV_KIND_ARRAY 0 /* variable is an array of quantities */ +#define DIFV_KIND_SCALAR 1 /* variable is a scalar quantity */ + +#define DIFV_SCOPE_GLOBAL 0 /* variable has global scope */ +#define DIFV_SCOPE_THREAD 1 /* variable has thread scope */ +#define DIFV_SCOPE_LOCAL 2 /* variable has local scope */ + +#define DIFV_F_REF 0x1 /* variable is referenced by DIFO */ +#define DIFV_F_MOD 0x2 /* variable is written by DIFO */ + +/* + * DTrace Actions + * + * The upper byte determines the class of the action; the low bytes determines + * the specific action within that class. The classes of actions are as + * follows: + * + * [ no class ] <= May record process- or kernel-related data + * DTRACEACT_PROC <= Only records process-related data + * DTRACEACT_PROC_DESTRUCTIVE <= Potentially destructive to processes + * DTRACEACT_KERNEL <= Only records kernel-related data + * DTRACEACT_KERNEL_DESTRUCTIVE <= Potentially destructive to the kernel + * DTRACEACT_SPECULATIVE <= Speculation-related action + * DTRACEACT_AGGREGATION <= Aggregating action + */ +#define DTRACEACT_NONE 0 /* no action */ +#define DTRACEACT_DIFEXPR 1 /* action is DIF expression */ +#define DTRACEACT_EXIT 2 /* exit() action */ +#define DTRACEACT_PRINTF 3 /* printf() action */ +#define DTRACEACT_PRINTA 4 /* printa() action */ +#define DTRACEACT_LIBACT 5 /* library-controlled action */ + +#define DTRACEACT_PROC 0x0100 +#define DTRACEACT_USTACK (DTRACEACT_PROC + 1) +#define DTRACEACT_JSTACK (DTRACEACT_PROC + 2) +#define DTRACEACT_USYM (DTRACEACT_PROC + 3) +#define DTRACEACT_UMOD (DTRACEACT_PROC + 4) +#define DTRACEACT_UADDR (DTRACEACT_PROC + 5) + +#define DTRACEACT_PROC_DESTRUCTIVE 0x0200 +#define DTRACEACT_STOP (DTRACEACT_PROC_DESTRUCTIVE + 1) +#define DTRACEACT_RAISE (DTRACEACT_PROC_DESTRUCTIVE + 2) +#define DTRACEACT_SYSTEM (DTRACEACT_PROC_DESTRUCTIVE + 3) +#define DTRACEACT_FREOPEN (DTRACEACT_PROC_DESTRUCTIVE + 4) + +#define DTRACEACT_PROC_CONTROL 0x0300 + +#define DTRACEACT_KERNEL 0x0400 +#define DTRACEACT_STACK (DTRACEACT_KERNEL + 1) +#define DTRACEACT_SYM (DTRACEACT_KERNEL + 2) +#define DTRACEACT_MOD (DTRACEACT_KERNEL + 3) + +#define DTRACEACT_KERNEL_DESTRUCTIVE 0x0500 +#define DTRACEACT_BREAKPOINT (DTRACEACT_KERNEL_DESTRUCTIVE + 1) +#define DTRACEACT_PANIC (DTRACEACT_KERNEL_DESTRUCTIVE + 2) +#define DTRACEACT_CHILL (DTRACEACT_KERNEL_DESTRUCTIVE + 3) + +#define DTRACEACT_SPECULATIVE 0x0600 +#define DTRACEACT_SPECULATE (DTRACEACT_SPECULATIVE + 1) +#define DTRACEACT_COMMIT (DTRACEACT_SPECULATIVE + 2) +#define DTRACEACT_DISCARD (DTRACEACT_SPECULATIVE + 3) + +#define DTRACEACT_CLASS(x) ((x) & 0xff00) + +#define DTRACEACT_ISDESTRUCTIVE(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_PROC_DESTRUCTIVE || \ + DTRACEACT_CLASS(x) == DTRACEACT_KERNEL_DESTRUCTIVE) + +#define DTRACEACT_ISSPECULATIVE(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_SPECULATIVE) + +#define DTRACEACT_ISPRINTFLIKE(x) \ + ((x) == DTRACEACT_PRINTF || (x) == DTRACEACT_PRINTA || \ + (x) == DTRACEACT_SYSTEM || (x) == DTRACEACT_FREOPEN) + +/* + * DTrace Aggregating Actions + * + * These are functions f(x) for which the following is true: + * + * f(f(x_0) U f(x_1) U ... U f(x_n)) = f(x_0 U x_1 U ... U x_n) + * + * where x_n is a set of arbitrary data. Aggregating actions are in their own + * DTrace action class, DTTRACEACT_AGGREGATION. The macros provided here allow + * for easier processing of the aggregation argument and data payload for a few + * aggregating actions (notably: quantize(), lquantize(), and ustack()). + */ +#define DTRACEACT_AGGREGATION 0x0700 +#define DTRACEAGG_COUNT (DTRACEACT_AGGREGATION + 1) +#define DTRACEAGG_MIN (DTRACEACT_AGGREGATION + 2) +#define DTRACEAGG_MAX (DTRACEACT_AGGREGATION + 3) +#define DTRACEAGG_AVG (DTRACEACT_AGGREGATION + 4) +#define DTRACEAGG_SUM (DTRACEACT_AGGREGATION + 5) +#define DTRACEAGG_STDDEV (DTRACEACT_AGGREGATION + 6) +#define DTRACEAGG_QUANTIZE (DTRACEACT_AGGREGATION + 7) +#define DTRACEAGG_LQUANTIZE (DTRACEACT_AGGREGATION + 8) + +#define DTRACEACT_ISAGG(x) \ + (DTRACEACT_CLASS(x) == DTRACEACT_AGGREGATION) + +#define DTRACE_QUANTIZE_NBUCKETS \ + (((sizeof (uint64_t) * NBBY) - 1) * 2 + 1) + +#define DTRACE_QUANTIZE_ZEROBUCKET (VBDTCAST(int)(sizeof (uint64_t) * NBBY) - 1) + +#define DTRACE_QUANTIZE_BUCKETVAL(buck) \ + (int64_t)((buck) < DTRACE_QUANTIZE_ZEROBUCKET ? \ + -(1LL << (DTRACE_QUANTIZE_ZEROBUCKET - 1 - (buck))) : \ + (buck) == DTRACE_QUANTIZE_ZEROBUCKET ? 0 : \ + 1LL << ((buck) - DTRACE_QUANTIZE_ZEROBUCKET - 1)) + +#define DTRACE_LQUANTIZE_STEPSHIFT 48 +#define DTRACE_LQUANTIZE_STEPMASK ((uint64_t)UINT16_MAX << 48) +#define DTRACE_LQUANTIZE_LEVELSHIFT 32 +#define DTRACE_LQUANTIZE_LEVELMASK ((uint64_t)UINT16_MAX << 32) +#define DTRACE_LQUANTIZE_BASESHIFT 0 +#define DTRACE_LQUANTIZE_BASEMASK UINT32_MAX + +#define DTRACE_LQUANTIZE_STEP(x) \ + (uint16_t)(((x) & DTRACE_LQUANTIZE_STEPMASK) >> \ + DTRACE_LQUANTIZE_STEPSHIFT) + +#define DTRACE_LQUANTIZE_LEVELS(x) \ + (uint16_t)(((x) & DTRACE_LQUANTIZE_LEVELMASK) >> \ + DTRACE_LQUANTIZE_LEVELSHIFT) + +#define DTRACE_LQUANTIZE_BASE(x) \ + (int32_t)(((x) & DTRACE_LQUANTIZE_BASEMASK) >> \ + DTRACE_LQUANTIZE_BASESHIFT) + +#define DTRACE_USTACK_NFRAMES(x) (uint32_t)((x) & UINT32_MAX) +#define DTRACE_USTACK_STRSIZE(x) (uint32_t)((x) >> 32) +#define DTRACE_USTACK_ARG(x, y) \ + ((((uint64_t)(y)) << 32) | ((x) & UINT32_MAX)) + +#ifndef _LP64 +#ifndef _LITTLE_ENDIAN +#define DTRACE_PTR(type, name) uint32_t name##pad; type *name +#else +#define DTRACE_PTR(type, name) type *name; uint32_t name##pad +#endif +#else +#define DTRACE_PTR(type, name) type *name +#endif + +/* + * DTrace Object Format (DOF) + * + * DTrace programs can be persistently encoded in the DOF format so that they + * may be embedded in other programs (for example, in an ELF file) or in the + * dtrace driver configuration file for use in anonymous tracing. The DOF + * format is versioned and extensible so that it can be revised and so that + * internal data structures can be modified or extended compatibly. All DOF + * structures use fixed-size types, so the 32-bit and 64-bit representations + * are identical and consumers can use either data model transparently. + * + * The file layout is structured as follows: + * + * +---------------+-------------------+----- ... ----+---- ... ------+ + * | dof_hdr_t | dof_sec_t[ ... ] | loadable | non-loadable | + * | (file header) | (section headers) | section data | section data | + * +---------------+-------------------+----- ... ----+---- ... ------+ + * |<------------ dof_hdr.dofh_loadsz --------------->| | + * |<------------ dof_hdr.dofh_filesz ------------------------------->| + * + * The file header stores meta-data including a magic number, data model for + * the instrumentation, data encoding, and properties of the DIF code within. + * The header describes its own size and the size of the section headers. By + * convention, an array of section headers follows the file header, and then + * the data for all loadable sections and unloadable sections. This permits + * consumer code to easily download the headers and all loadable data into the + * DTrace driver in one contiguous chunk, omitting other extraneous sections. + * + * The section headers describe the size, offset, alignment, and section type + * for each section. Sections are described using a set of #defines that tell + * the consumer what kind of data is expected. Sections can contain links to + * other sections by storing a dof_secidx_t, an index into the section header + * array, inside of the section data structures. The section header includes + * an entry size so that sections with data arrays can grow their structures. + * + * The DOF data itself can contain many snippets of DIF (i.e. >1 DIFOs), which + * are represented themselves as a collection of related DOF sections. This + * permits us to change the set of sections associated with a DIFO over time, + * and also permits us to encode DIFOs that contain different sets of sections. + * When a DOF section wants to refer to a DIFO, it stores the dof_secidx_t of a + * section of type DOF_SECT_DIFOHDR. This section's data is then an array of + * dof_secidx_t's which in turn denote the sections associated with this DIFO. + * + * This loose coupling of the file structure (header and sections) to the + * structure of the DTrace program itself (ECB descriptions, action + * descriptions, and DIFOs) permits activities such as relocation processing + * to occur in a single pass without having to understand D program structure. + * + * Finally, strings are always stored in ELF-style string tables along with a + * string table section index and string table offset. Therefore strings in + * DOF are always arbitrary-length and not bound to the current implementation. + */ + +#define DOF_ID_SIZE 16 /* total size of dofh_ident[] in bytes */ + +typedef struct dof_hdr { + uint8_t dofh_ident[DOF_ID_SIZE]; /* identification bytes (see below) */ + uint32_t dofh_flags; /* file attribute flags (if any) */ + uint32_t dofh_hdrsize; /* size of file header in bytes */ + uint32_t dofh_secsize; /* size of section header in bytes */ + uint32_t dofh_secnum; /* number of section headers */ + uint64_t dofh_secoff; /* file offset of section headers */ + uint64_t dofh_loadsz; /* file size of loadable portion */ + uint64_t dofh_filesz; /* file size of entire DOF file */ + uint64_t dofh_pad; /* reserved for future use */ +} dof_hdr_t; + +#define DOF_ID_MAG0 0 /* first byte of magic number */ +#define DOF_ID_MAG1 1 /* second byte of magic number */ +#define DOF_ID_MAG2 2 /* third byte of magic number */ +#define DOF_ID_MAG3 3 /* fourth byte of magic number */ +#define DOF_ID_MODEL 4 /* DOF data model (see below) */ +#define DOF_ID_ENCODING 5 /* DOF data encoding (see below) */ +#define DOF_ID_VERSION 6 /* DOF file format major version (see below) */ +#define DOF_ID_DIFVERS 7 /* DIF instruction set version */ +#define DOF_ID_DIFIREG 8 /* DIF integer registers used by compiler */ +#define DOF_ID_DIFTREG 9 /* DIF tuple registers used by compiler */ +#define DOF_ID_PAD 10 /* start of padding bytes (all zeroes) */ + +#define DOF_MAG_MAG0 0x7F /* DOF_ID_MAG[0-3] */ +#define DOF_MAG_MAG1 'D' +#define DOF_MAG_MAG2 'O' +#define DOF_MAG_MAG3 'F' + +#define DOF_MAG_STRING "\177DOF" +#define DOF_MAG_STRLEN 4 + +#define DOF_MODEL_NONE 0 /* DOF_ID_MODEL */ +#define DOF_MODEL_ILP32 1 +#define DOF_MODEL_LP64 2 + +#ifdef _LP64 +#define DOF_MODEL_NATIVE DOF_MODEL_LP64 +#else +#define DOF_MODEL_NATIVE DOF_MODEL_ILP32 +#endif + +#define DOF_ENCODE_NONE 0 /* DOF_ID_ENCODING */ +#define DOF_ENCODE_LSB 1 +#define DOF_ENCODE_MSB 2 + +#ifdef _BIG_ENDIAN +#define DOF_ENCODE_NATIVE DOF_ENCODE_MSB +#else +#define DOF_ENCODE_NATIVE DOF_ENCODE_LSB +#endif + +#define DOF_VERSION_1 1 /* DOF version 1: Solaris 10 FCS */ +#define DOF_VERSION_2 2 /* DOF version 2: Solaris Express 6/06 */ +#define DOF_VERSION DOF_VERSION_2 /* Latest DOF version */ + +#define DOF_FL_VALID 0 /* mask of all valid dofh_flags bits */ + +typedef uint32_t dof_secidx_t; /* section header table index type */ +typedef uint32_t dof_stridx_t; /* string table index type */ + +#define DOF_SECIDX_NONE (~1U) /* null value for section indices */ +#define DOF_STRIDX_NONE (~1U) /* null value for string indices */ + +typedef struct dof_sec { + uint32_t dofs_type; /* section type (see below) */ + uint32_t dofs_align; /* section data memory alignment */ + uint32_t dofs_flags; /* section flags (if any) */ + uint32_t dofs_entsize; /* size of section entry (if table) */ + uint64_t dofs_offset; /* offset of section data within file */ + uint64_t dofs_size; /* size of section data in bytes */ +} dof_sec_t; + +#define DOF_SECT_NONE 0 /* null section */ +#define DOF_SECT_COMMENTS 1 /* compiler comments */ +#define DOF_SECT_SOURCE 2 /* D program source code */ +#define DOF_SECT_ECBDESC 3 /* dof_ecbdesc_t */ +#define DOF_SECT_PROBEDESC 4 /* dof_probedesc_t */ +#define DOF_SECT_ACTDESC 5 /* dof_actdesc_t array */ +#define DOF_SECT_DIFOHDR 6 /* dof_difohdr_t (variable length) */ +#define DOF_SECT_DIF 7 /* uint32_t array of byte code */ +#define DOF_SECT_STRTAB 8 /* string table */ +#define DOF_SECT_VARTAB 9 /* dtrace_difv_t array */ +#define DOF_SECT_RELTAB 10 /* dof_relodesc_t array */ +#define DOF_SECT_TYPTAB 11 /* dtrace_diftype_t array */ +#define DOF_SECT_URELHDR 12 /* dof_relohdr_t (user relocations) */ +#define DOF_SECT_KRELHDR 13 /* dof_relohdr_t (kernel relocations) */ +#define DOF_SECT_OPTDESC 14 /* dof_optdesc_t array */ +#define DOF_SECT_PROVIDER 15 /* dof_provider_t */ +#define DOF_SECT_PROBES 16 /* dof_probe_t array */ +#define DOF_SECT_PRARGS 17 /* uint8_t array (probe arg mappings) */ +#define DOF_SECT_PROFFS 18 /* uint32_t array (probe arg offsets) */ +#define DOF_SECT_INTTAB 19 /* uint64_t array */ +#define DOF_SECT_UTSNAME 20 /* struct utsname */ +#define DOF_SECT_XLTAB 21 /* dof_xlref_t array */ +#define DOF_SECT_XLMEMBERS 22 /* dof_xlmember_t array */ +#define DOF_SECT_XLIMPORT 23 /* dof_xlator_t */ +#define DOF_SECT_XLEXPORT 24 /* dof_xlator_t */ +#define DOF_SECT_PREXPORT 25 /* dof_secidx_t array (exported objs) */ +#define DOF_SECT_PRENOFFS 26 /* uint32_t array (enabled offsets) */ + +#define DOF_SECF_LOAD 1 /* section should be loaded */ + +#define DOF_SEC_ISLOADABLE(x) \ + (((x) == DOF_SECT_ECBDESC) || ((x) == DOF_SECT_PROBEDESC) || \ + ((x) == DOF_SECT_ACTDESC) || ((x) == DOF_SECT_DIFOHDR) || \ + ((x) == DOF_SECT_DIF) || ((x) == DOF_SECT_STRTAB) || \ + ((x) == DOF_SECT_VARTAB) || ((x) == DOF_SECT_RELTAB) || \ + ((x) == DOF_SECT_TYPTAB) || ((x) == DOF_SECT_URELHDR) || \ + ((x) == DOF_SECT_KRELHDR) || ((x) == DOF_SECT_OPTDESC) || \ + ((x) == DOF_SECT_PROVIDER) || ((x) == DOF_SECT_PROBES) || \ + ((x) == DOF_SECT_PRARGS) || ((x) == DOF_SECT_PROFFS) || \ + ((x) == DOF_SECT_INTTAB) || ((x) == DOF_SECT_XLTAB) || \ + ((x) == DOF_SECT_XLMEMBERS) || ((x) == DOF_SECT_XLIMPORT) || \ + ((x) == DOF_SECT_XLIMPORT) || ((x) == DOF_SECT_XLEXPORT) || \ + ((x) == DOF_SECT_PREXPORT) || ((x) == DOF_SECT_PRENOFFS)) + +typedef struct dof_ecbdesc { + dof_secidx_t dofe_probes; /* link to DOF_SECT_PROBEDESC */ + dof_secidx_t dofe_pred; /* link to DOF_SECT_DIFOHDR */ + dof_secidx_t dofe_actions; /* link to DOF_SECT_ACTDESC */ + uint32_t dofe_pad; /* reserved for future use */ + uint64_t dofe_uarg; /* user-supplied library argument */ +} dof_ecbdesc_t; + +typedef struct dof_probedesc { + dof_secidx_t dofp_strtab; /* link to DOF_SECT_STRTAB section */ + dof_stridx_t dofp_provider; /* provider string */ + dof_stridx_t dofp_mod; /* module string */ + dof_stridx_t dofp_func; /* function string */ + dof_stridx_t dofp_name; /* name string */ + uint32_t dofp_id; /* probe identifier (or zero) */ +} dof_probedesc_t; + +typedef struct dof_actdesc { + dof_secidx_t dofa_difo; /* link to DOF_SECT_DIFOHDR */ + dof_secidx_t dofa_strtab; /* link to DOF_SECT_STRTAB section */ + uint32_t dofa_kind; /* action kind (DTRACEACT_* constant) */ + uint32_t dofa_ntuple; /* number of subsequent tuple actions */ + uint64_t dofa_arg; /* kind-specific argument */ + uint64_t dofa_uarg; /* user-supplied argument */ +} dof_actdesc_t; + +typedef struct dof_difohdr { + dtrace_diftype_t dofd_rtype; /* return type for this fragment */ + dof_secidx_t dofd_links[1]; /* variable length array of indices */ +} dof_difohdr_t; + +typedef struct dof_relohdr { + dof_secidx_t dofr_strtab; /* link to DOF_SECT_STRTAB for names */ + dof_secidx_t dofr_relsec; /* link to DOF_SECT_RELTAB for relos */ + dof_secidx_t dofr_tgtsec; /* link to section we are relocating */ +} dof_relohdr_t; + +typedef struct dof_relodesc { + dof_stridx_t dofr_name; /* string name of relocation symbol */ + uint32_t dofr_type; /* relo type (DOF_RELO_* constant) */ + uint64_t dofr_offset; /* byte offset for relocation */ + uint64_t dofr_data; /* additional type-specific data */ +} dof_relodesc_t; + +#define DOF_RELO_NONE 0 /* empty relocation entry */ +#define DOF_RELO_SETX 1 /* relocate setx value */ + +typedef struct dof_optdesc { + uint32_t dofo_option; /* option identifier */ + dof_secidx_t dofo_strtab; /* string table, if string option */ + uint64_t dofo_value; /* option value or string index */ +} dof_optdesc_t; + +typedef uint32_t dof_attr_t; /* encoded stability attributes */ + +#define DOF_ATTR(n, d, c) (((n) << 24) | ((d) << 16) | ((c) << 8)) +#define DOF_ATTR_NAME(a) (((a) >> 24) & 0xff) +#define DOF_ATTR_DATA(a) (((a) >> 16) & 0xff) +#define DOF_ATTR_CLASS(a) (((a) >> 8) & 0xff) + +typedef struct dof_provider { + dof_secidx_t dofpv_strtab; /* link to DOF_SECT_STRTAB section */ + dof_secidx_t dofpv_probes; /* link to DOF_SECT_PROBES section */ + dof_secidx_t dofpv_prargs; /* link to DOF_SECT_PRARGS section */ + dof_secidx_t dofpv_proffs; /* link to DOF_SECT_PROFFS section */ + dof_stridx_t dofpv_name; /* provider name string */ + dof_attr_t dofpv_provattr; /* provider attributes */ + dof_attr_t dofpv_modattr; /* module attributes */ + dof_attr_t dofpv_funcattr; /* function attributes */ + dof_attr_t dofpv_nameattr; /* name attributes */ + dof_attr_t dofpv_argsattr; /* args attributes */ + dof_secidx_t dofpv_prenoffs; /* link to DOF_SECT_PRENOFFS section */ +} dof_provider_t; + +typedef struct dof_probe { + uint64_t dofpr_addr; /* probe base address or offset */ + dof_stridx_t dofpr_func; /* probe function string */ + dof_stridx_t dofpr_name; /* probe name string */ + dof_stridx_t dofpr_nargv; /* native argument type strings */ + dof_stridx_t dofpr_xargv; /* translated argument type strings */ + uint32_t dofpr_argidx; /* index of first argument mapping */ + uint32_t dofpr_offidx; /* index of first offset entry */ + uint8_t dofpr_nargc; /* native argument count */ + uint8_t dofpr_xargc; /* translated argument count */ + uint16_t dofpr_noffs; /* number of offset entries for probe */ + uint32_t dofpr_enoffidx; /* index of first is-enabled offset */ + uint16_t dofpr_nenoffs; /* number of is-enabled offsets */ + uint16_t dofpr_pad1; /* reserved for future use */ + uint32_t dofpr_pad2; /* reserved for future use */ +} dof_probe_t; + +typedef struct dof_xlator { + dof_secidx_t dofxl_members; /* link to DOF_SECT_XLMEMBERS section */ + dof_secidx_t dofxl_strtab; /* link to DOF_SECT_STRTAB section */ + dof_stridx_t dofxl_argv; /* input parameter type strings */ + uint32_t dofxl_argc; /* input parameter list length */ + dof_stridx_t dofxl_type; /* output type string name */ + dof_attr_t dofxl_attr; /* output stability attributes */ +} dof_xlator_t; + +typedef struct dof_xlmember { + dof_secidx_t dofxm_difo; /* member link to DOF_SECT_DIFOHDR */ + dof_stridx_t dofxm_name; /* member name */ + dtrace_diftype_t dofxm_type; /* member type */ +} dof_xlmember_t; + +typedef struct dof_xlref { + dof_secidx_t dofxr_xlator; /* link to DOF_SECT_XLATORS section */ + uint32_t dofxr_member; /* index of referenced dof_xlmember */ + uint32_t dofxr_argn; /* index of argument for DIF_OP_XLARG */ +} dof_xlref_t; + +/* + * DTrace Intermediate Format Object (DIFO) + * + * A DIFO is used to store the compiled DIF for a D expression, its return + * type, and its string and variable tables. The string table is a single + * buffer of character data into which sets instructions and variable + * references can reference strings using a byte offset. The variable table + * is an array of dtrace_difv_t structures that describe the name and type of + * each variable and the id used in the DIF code. This structure is described + * above in the DIF section of this header file. The DIFO is used at both + * user-level (in the library) and in the kernel, but the structure is never + * passed between the two: the DOF structures form the only interface. As a + * result, the definition can change depending on the presence of _KERNEL. + */ +typedef struct dtrace_difo { + dif_instr_t *dtdo_buf; /* instruction buffer */ + uint64_t *dtdo_inttab; /* integer table (optional) */ + char *dtdo_strtab; /* string table (optional) */ + dtrace_difv_t *dtdo_vartab; /* variable table (optional) */ + uint_t dtdo_len; /* length of instruction buffer */ + uint_t dtdo_intlen; /* length of integer table */ + uint_t dtdo_strlen; /* length of string table */ + uint_t dtdo_varlen; /* length of variable table */ + dtrace_diftype_t dtdo_rtype; /* return type */ + uint_t dtdo_refcnt; /* owner reference count */ + uint_t dtdo_destructive; /* invokes destructive subroutines */ +#if !defined(_KERNEL) || defined(IN_RING3) + dof_relodesc_t *dtdo_kreltab; /* kernel relocations */ + dof_relodesc_t *dtdo_ureltab; /* user relocations */ + struct dt_node **dtdo_xlmtab; /* translator references */ + uint_t dtdo_krelen; /* length of krelo table */ + uint_t dtdo_urelen; /* length of urelo table */ + uint_t dtdo_xlmlen; /* length of translator table */ +#endif +} dtrace_difo_t; + +/* + * DTrace Enabling Description Structures + * + * When DTrace is tracking the description of a DTrace enabling entity (probe, + * predicate, action, ECB, record, etc.), it does so in a description + * structure. These structures all end in "desc", and are used at both + * user-level and in the kernel -- but (with the exception of + * dtrace_probedesc_t) they are never passed between them. Typically, + * user-level will use the description structures when assembling an enabling. + * It will then distill those description structures into a DOF object (see + * above), and send it into the kernel. The kernel will again use the + * description structures to create a description of the enabling as it reads + * the DOF. When the description is complete, the enabling will be actually + * created -- turning it into the structures that represent the enabling + * instead of merely describing it. Not surprisingly, the description + * structures bear a strong resemblance to the DOF structures that act as their + * conduit. + */ +struct dtrace_predicate; + +typedef struct dtrace_probedesc { + dtrace_id_t dtpd_id; /* probe identifier */ + char dtpd_provider[DTRACE_PROVNAMELEN]; /* probe provider name */ + char dtpd_mod[DTRACE_MODNAMELEN]; /* probe module name */ + char dtpd_func[DTRACE_FUNCNAMELEN]; /* probe function name */ + char dtpd_name[DTRACE_NAMELEN]; /* probe name */ +} dtrace_probedesc_t; + +typedef struct dtrace_repldesc { + dtrace_probedesc_t dtrpd_match; /* probe descr. to match */ + dtrace_probedesc_t dtrpd_create; /* probe descr. to create */ +} dtrace_repldesc_t; + +typedef struct dtrace_preddesc { + dtrace_difo_t *dtpdd_difo; /* pointer to DIF object */ + struct dtrace_predicate *dtpdd_predicate; /* pointer to predicate */ +} dtrace_preddesc_t; + +typedef struct dtrace_actdesc { + dtrace_difo_t *dtad_difo; /* pointer to DIF object */ + struct dtrace_actdesc *dtad_next; /* next action */ + dtrace_actkind_t dtad_kind; /* kind of action */ + uint32_t dtad_ntuple; /* number in tuple */ + uint64_t dtad_arg; /* action argument */ + uint64_t dtad_uarg; /* user argument */ + int dtad_refcnt; /* reference count */ +} dtrace_actdesc_t; + +typedef struct dtrace_ecbdesc { + dtrace_actdesc_t *dted_action; /* action description(s) */ + dtrace_preddesc_t dted_pred; /* predicate description */ + dtrace_probedesc_t dted_probe; /* probe description */ + uint64_t dted_uarg; /* library argument */ + int dted_refcnt; /* reference count */ +} dtrace_ecbdesc_t; + +/* + * DTrace Metadata Description Structures + * + * DTrace separates the trace data stream from the metadata stream. The only + * metadata tokens placed in the data stream are enabled probe identifiers + * (EPIDs) or (in the case of aggregations) aggregation identifiers. In order + * to determine the structure of the data, DTrace consumers pass the token to + * the kernel, and receive in return a corresponding description of the enabled + * probe (via the dtrace_eprobedesc structure) or the aggregation (via the + * dtrace_aggdesc structure). Both of these structures are expressed in terms + * of record descriptions (via the dtrace_recdesc structure) that describe the + * exact structure of the data. Some record descriptions may also contain a + * format identifier; this additional bit of metadata can be retrieved from the + * kernel, for which a format description is returned via the dtrace_fmtdesc + * structure. Note that all four of these structures must be bitness-neutral + * to allow for a 32-bit DTrace consumer on a 64-bit kernel. + */ +typedef struct dtrace_recdesc { + dtrace_actkind_t dtrd_action; /* kind of action */ + uint32_t dtrd_size; /* size of record */ + uint32_t dtrd_offset; /* offset in ECB's data */ + uint16_t dtrd_alignment; /* required alignment */ + uint16_t dtrd_format; /* format, if any */ + uint64_t dtrd_arg; /* action argument */ + uint64_t dtrd_uarg; /* user argument */ +} dtrace_recdesc_t; + +typedef struct dtrace_eprobedesc { + dtrace_epid_t dtepd_epid; /* enabled probe ID */ + dtrace_id_t dtepd_probeid; /* probe ID */ + uint64_t dtepd_uarg; /* library argument */ + uint32_t dtepd_size; /* total size */ + int dtepd_nrecs; /* number of records */ + dtrace_recdesc_t dtepd_rec[1]; /* records themselves */ +} dtrace_eprobedesc_t; + +typedef struct dtrace_aggdesc { + DTRACE_PTR(char, dtagd_name); /* not filled in by kernel */ + dtrace_aggvarid_t dtagd_varid; /* not filled in by kernel */ + int dtagd_flags; /* not filled in by kernel */ + dtrace_aggid_t dtagd_id; /* aggregation ID */ + dtrace_epid_t dtagd_epid; /* enabled probe ID */ + uint32_t dtagd_size; /* size in bytes */ + int dtagd_nrecs; /* number of records */ + uint32_t dtagd_pad; /* explicit padding */ + dtrace_recdesc_t dtagd_rec[1]; /* record descriptions */ +} dtrace_aggdesc_t; + +typedef struct dtrace_fmtdesc { + DTRACE_PTR(char, dtfd_string); /* format string */ + int dtfd_length; /* length of format string */ + uint16_t dtfd_format; /* format identifier */ +} dtrace_fmtdesc_t; + +#define DTRACE_SIZEOF_EPROBEDESC(desc) \ + (sizeof (dtrace_eprobedesc_t) + ((desc)->dtepd_nrecs ? \ + (((desc)->dtepd_nrecs - 1) * sizeof (dtrace_recdesc_t)) : 0)) + +#define DTRACE_SIZEOF_AGGDESC(desc) \ + (sizeof (dtrace_aggdesc_t) + ((desc)->dtagd_nrecs ? \ + (((desc)->dtagd_nrecs - 1) * sizeof (dtrace_recdesc_t)) : 0)) + +/* + * DTrace Option Interface + * + * Run-time DTrace options are set and retrieved via DOF_SECT_OPTDESC sections + * in a DOF image. The dof_optdesc structure contains an option identifier and + * an option value. The valid option identifiers are found below; the mapping + * between option identifiers and option identifying strings is maintained at + * user-level. Note that the value of DTRACEOPT_UNSET is such that all of the + * following are potentially valid option values: all positive integers, zero + * and negative one. Some options (notably "bufpolicy" and "bufresize") take + * predefined tokens as their values; these are defined with + * DTRACEOPT_{option}_{token}. + */ +#define DTRACEOPT_BUFSIZE 0 /* buffer size */ +#define DTRACEOPT_BUFPOLICY 1 /* buffer policy */ +#define DTRACEOPT_DYNVARSIZE 2 /* dynamic variable size */ +#define DTRACEOPT_AGGSIZE 3 /* aggregation size */ +#define DTRACEOPT_SPECSIZE 4 /* speculation size */ +#define DTRACEOPT_NSPEC 5 /* number of speculations */ +#define DTRACEOPT_STRSIZE 6 /* string size */ +#define DTRACEOPT_CLEANRATE 7 /* dynvar cleaning rate */ +#define DTRACEOPT_CPU 8 /* CPU to trace */ +#define DTRACEOPT_BUFRESIZE 9 /* buffer resizing policy */ +#define DTRACEOPT_GRABANON 10 /* grab anonymous state, if any */ +#define DTRACEOPT_FLOWINDENT 11 /* indent function entry/return */ +#define DTRACEOPT_QUIET 12 /* only output explicitly traced data */ +#define DTRACEOPT_STACKFRAMES 13 /* number of stack frames */ +#define DTRACEOPT_USTACKFRAMES 14 /* number of user stack frames */ +#define DTRACEOPT_AGGRATE 15 /* aggregation snapshot rate */ +#define DTRACEOPT_SWITCHRATE 16 /* buffer switching rate */ +#define DTRACEOPT_STATUSRATE 17 /* status rate */ +#define DTRACEOPT_DESTRUCTIVE 18 /* destructive actions allowed */ +#define DTRACEOPT_STACKINDENT 19 /* output indent for stack traces */ +#define DTRACEOPT_RAWBYTES 20 /* always print bytes in raw form */ +#define DTRACEOPT_JSTACKFRAMES 21 /* number of jstack() frames */ +#define DTRACEOPT_JSTACKSTRSIZE 22 /* size of jstack() string table */ +#define DTRACEOPT_AGGSORTKEY 23 /* sort aggregations by key */ +#define DTRACEOPT_AGGSORTREV 24 /* reverse-sort aggregations */ +#define DTRACEOPT_AGGSORTPOS 25 /* agg. position to sort on */ +#define DTRACEOPT_AGGSORTKEYPOS 26 /* agg. key position to sort on */ +#define DTRACEOPT_MAX 27 /* number of options */ + +#define DTRACEOPT_UNSET (dtrace_optval_t)-2 /* unset option */ + +#define DTRACEOPT_BUFPOLICY_RING 0 /* ring buffer */ +#define DTRACEOPT_BUFPOLICY_FILL 1 /* fill buffer, then stop */ +#define DTRACEOPT_BUFPOLICY_SWITCH 2 /* switch buffers */ + +#define DTRACEOPT_BUFRESIZE_AUTO 0 /* automatic resizing */ +#define DTRACEOPT_BUFRESIZE_MANUAL 1 /* manual resizing */ + +/* + * DTrace Buffer Interface + * + * In order to get a snapshot of the principal or aggregation buffer, + * user-level passes a buffer description to the kernel with the dtrace_bufdesc + * structure. This describes which CPU user-level is interested in, and + * where user-level wishes the kernel to snapshot the buffer to (the + * dtbd_data field). The kernel uses the same structure to pass back some + * information regarding the buffer: the size of data actually copied out, the + * number of drops, the number of errors, and the offset of the oldest record. + * If the buffer policy is a "switch" policy, taking a snapshot of the + * principal buffer has the additional effect of switching the active and + * inactive buffers. Taking a snapshot of the aggregation buffer _always_ has + * the additional effect of switching the active and inactive buffers. + */ +typedef struct dtrace_bufdesc { + uint64_t dtbd_size; /* size of buffer */ + uint32_t dtbd_cpu; /* CPU or DTRACE_CPUALL */ + uint32_t dtbd_errors; /* number of errors */ + uint64_t dtbd_drops; /* number of drops */ + DTRACE_PTR(char, dtbd_data); /* data */ + uint64_t dtbd_oldest; /* offset of oldest record */ +} dtrace_bufdesc_t; + +/* + * DTrace Status + * + * The status of DTrace is relayed via the dtrace_status structure. This + * structure contains members to count drops other than the capacity drops + * available via the buffer interface (see above). This consists of dynamic + * drops (including capacity dynamic drops, rinsing drops and dirty drops), and + * speculative drops (including capacity speculative drops, drops due to busy + * speculative buffers and drops due to unavailable speculative buffers). + * Additionally, the status structure contains a field to indicate the number + * of "fill"-policy buffers have been filled and a boolean field to indicate + * that exit() has been called. If the dtst_exiting field is non-zero, no + * further data will be generated until tracing is stopped (at which time any + * enablings of the END action will be processed); if user-level sees that + * this field is non-zero, tracing should be stopped as soon as possible. + */ +typedef struct dtrace_status { + uint64_t dtst_dyndrops; /* dynamic drops */ + uint64_t dtst_dyndrops_rinsing; /* dyn drops due to rinsing */ + uint64_t dtst_dyndrops_dirty; /* dyn drops due to dirty */ + uint64_t dtst_specdrops; /* speculative drops */ + uint64_t dtst_specdrops_busy; /* spec drops due to busy */ + uint64_t dtst_specdrops_unavail; /* spec drops due to unavail */ + uint64_t dtst_errors; /* total errors */ + uint64_t dtst_filled; /* number of filled bufs */ + uint64_t dtst_stkstroverflows; /* stack string tab overflows */ + uint64_t dtst_dblerrors; /* errors in ERROR probes */ + char dtst_killed; /* non-zero if killed */ + char dtst_exiting; /* non-zero if exit() called */ + char dtst_pad[6]; /* pad out to 64-bit align */ +} dtrace_status_t; + +/* + * DTrace Configuration + * + * User-level may need to understand some elements of the kernel DTrace + * configuration in order to generate correct DIF. This information is + * conveyed via the dtrace_conf structure. + */ +typedef struct dtrace_conf { + uint_t dtc_difversion; /* supported DIF version */ + uint_t dtc_difintregs; /* # of DIF integer registers */ + uint_t dtc_diftupregs; /* # of DIF tuple registers */ + uint_t dtc_ctfmodel; /* CTF data model */ + uint_t dtc_pad[8]; /* reserved for future use */ +} dtrace_conf_t; + +/* + * DTrace Faults + * + * The constants below DTRACEFLT_LIBRARY indicate probe processing faults; + * constants at or above DTRACEFLT_LIBRARY indicate faults in probe + * postprocessing at user-level. Probe processing faults induce an ERROR + * probe and are replicated in unistd.d to allow users' ERROR probes to decode + * the error condition using thse symbolic labels. + */ +#define DTRACEFLT_UNKNOWN 0 /* Unknown fault */ +#define DTRACEFLT_BADADDR 1 /* Bad address */ +#define DTRACEFLT_BADALIGN 2 /* Bad alignment */ +#define DTRACEFLT_ILLOP 3 /* Illegal operation */ +#define DTRACEFLT_DIVZERO 4 /* Divide-by-zero */ +#define DTRACEFLT_NOSCRATCH 5 /* Out of scratch space */ +#define DTRACEFLT_KPRIV 6 /* Illegal kernel access */ +#define DTRACEFLT_UPRIV 7 /* Illegal user access */ +#define DTRACEFLT_TUPOFLOW 8 /* Tuple stack overflow */ +#define DTRACEFLT_BADSTACK 9 /* Bad stack */ + +#define DTRACEFLT_LIBRARY 1000 /* Library-level fault */ + +/* + * DTrace Argument Types + * + * Because it would waste both space and time, argument types do not reside + * with the probe. In order to determine argument types for args[X] + * variables, the D compiler queries for argument types on a probe-by-probe + * basis. (This optimizes for the common case that arguments are either not + * used or used in an untyped fashion.) Typed arguments are specified with a + * string of the type name in the dtragd_native member of the argument + * description structure. Typed arguments may be further translated to types + * of greater stability; the provider indicates such a translated argument by + * filling in the dtargd_xlate member with the string of the translated type. + * Finally, the provider may indicate which argument value a given argument + * maps to by setting the dtargd_mapping member -- allowing a single argument + * to map to multiple args[X] variables. + */ +typedef struct dtrace_argdesc { + dtrace_id_t dtargd_id; /* probe identifier */ + int dtargd_ndx; /* arg number (-1 iff none) */ + int dtargd_mapping; /* value mapping */ + char dtargd_native[DTRACE_ARGTYPELEN]; /* native type name */ + char dtargd_xlate[DTRACE_ARGTYPELEN]; /* translated type name */ +} dtrace_argdesc_t; + +/* + * DTrace Stability Attributes + * + * Each DTrace provider advertises the name and data stability of each of its + * probe description components, as well as its architectural dependencies. + * The D compiler can query the provider attributes (dtrace_pattr_t below) in + * order to compute the properties of an input program and report them. + */ +typedef uint8_t dtrace_stability_t; /* stability code (see attributes(5)) */ +typedef uint8_t dtrace_class_t; /* architectural dependency class */ + +#define DTRACE_STABILITY_INTERNAL 0 /* private to DTrace itself */ +#define DTRACE_STABILITY_PRIVATE 1 /* private to Sun (see docs) */ +#define DTRACE_STABILITY_OBSOLETE 2 /* scheduled for removal */ +#define DTRACE_STABILITY_EXTERNAL 3 /* not controlled by Sun */ +#define DTRACE_STABILITY_UNSTABLE 4 /* new or rapidly changing */ +#define DTRACE_STABILITY_EVOLVING 5 /* less rapidly changing */ +#define DTRACE_STABILITY_STABLE 6 /* mature interface from Sun */ +#define DTRACE_STABILITY_STANDARD 7 /* industry standard */ +#define DTRACE_STABILITY_MAX 7 /* maximum valid stability */ + +#define DTRACE_CLASS_UNKNOWN 0 /* unknown architectural dependency */ +#define DTRACE_CLASS_CPU 1 /* CPU-module-specific */ +#define DTRACE_CLASS_PLATFORM 2 /* platform-specific (uname -i) */ +#define DTRACE_CLASS_GROUP 3 /* hardware-group-specific (uname -m) */ +#define DTRACE_CLASS_ISA 4 /* ISA-specific (uname -p) */ +#define DTRACE_CLASS_COMMON 5 /* common to all systems */ +#define DTRACE_CLASS_MAX 5 /* maximum valid class */ + +#define DTRACE_PRIV_NONE 0x0000 +#define DTRACE_PRIV_KERNEL 0x0001 +#define DTRACE_PRIV_USER 0x0002 +#define DTRACE_PRIV_PROC 0x0004 +#define DTRACE_PRIV_OWNER 0x0008 +#define DTRACE_PRIV_ZONEOWNER 0x0010 + +#define DTRACE_PRIV_ALL \ + (DTRACE_PRIV_KERNEL | DTRACE_PRIV_USER | \ + DTRACE_PRIV_PROC | DTRACE_PRIV_OWNER | DTRACE_PRIV_ZONEOWNER) + +typedef struct dtrace_ppriv { + uint32_t dtpp_flags; /* privilege flags */ + uid_t dtpp_uid; /* user ID */ + zoneid_t dtpp_zoneid; /* zone ID */ +} dtrace_ppriv_t; + +typedef struct dtrace_attribute { + dtrace_stability_t dtat_name; /* entity name stability */ + dtrace_stability_t dtat_data; /* entity data stability */ + dtrace_class_t dtat_class; /* entity data dependency */ +} dtrace_attribute_t; + +typedef struct dtrace_pattr { + dtrace_attribute_t dtpa_provider; /* provider attributes */ + dtrace_attribute_t dtpa_mod; /* module attributes */ + dtrace_attribute_t dtpa_func; /* function attributes */ + dtrace_attribute_t dtpa_name; /* name attributes */ + dtrace_attribute_t dtpa_args; /* args[] attributes */ +} dtrace_pattr_t; + +typedef struct dtrace_providerdesc { + char dtvd_name[DTRACE_PROVNAMELEN]; /* provider name */ + dtrace_pattr_t dtvd_attr; /* stability attributes */ + dtrace_ppriv_t dtvd_priv; /* privileges required */ +} dtrace_providerdesc_t; + +/* + * DTrace Pseudodevice Interface + * + * DTrace is controlled through ioctl(2)'s to the in-kernel dtrace:dtrace + * pseudodevice driver. These ioctls comprise the user-kernel interface to + * DTrace. + */ +#define DTRACEIOC (('d' << 24) | ('t' << 16) | ('r' << 8)) +#define DTRACEIOC_PROVIDER (DTRACEIOC | 1) /* provider query */ +#define DTRACEIOC_PROBES (DTRACEIOC | 2) /* probe query */ +#define DTRACEIOC_BUFSNAP (DTRACEIOC | 4) /* snapshot buffer */ +#define DTRACEIOC_PROBEMATCH (DTRACEIOC | 5) /* match probes */ +#define DTRACEIOC_ENABLE (DTRACEIOC | 6) /* enable probes */ +#define DTRACEIOC_AGGSNAP (DTRACEIOC | 7) /* snapshot agg. */ +#define DTRACEIOC_EPROBE (DTRACEIOC | 8) /* get eprobe desc. */ +#define DTRACEIOC_PROBEARG (DTRACEIOC | 9) /* get probe arg */ +#define DTRACEIOC_CONF (DTRACEIOC | 10) /* get config. */ +#define DTRACEIOC_STATUS (DTRACEIOC | 11) /* get status */ +#define DTRACEIOC_GO (DTRACEIOC | 12) /* start tracing */ +#define DTRACEIOC_STOP (DTRACEIOC | 13) /* stop tracing */ +#define DTRACEIOC_AGGDESC (DTRACEIOC | 15) /* get agg. desc. */ +#define DTRACEIOC_FORMAT (DTRACEIOC | 16) /* get format str */ +#define DTRACEIOC_DOFGET (DTRACEIOC | 17) /* get DOF */ +#define DTRACEIOC_REPLICATE (DTRACEIOC | 18) /* replicate enab */ + +/* + * DTrace Helpers + * + * In general, DTrace establishes probes in processes and takes actions on + * processes without knowing their specific user-level structures. Instead of + * existing in the framework, process-specific knowledge is contained by the + * enabling D program -- which can apply process-specific knowledge by making + * appropriate use of DTrace primitives like copyin() and copyinstr() to + * operate on user-level data. However, there may exist some specific probes + * of particular semantic relevance that the application developer may wish to + * explicitly export. For example, an application may wish to export a probe + * at the point that it begins and ends certain well-defined transactions. In + * addition to providing probes, programs may wish to offer assistance for + * certain actions. For example, in highly dynamic environments (e.g., Java), + * it may be difficult to obtain a stack trace in terms of meaningful symbol + * names (the translation from instruction addresses to corresponding symbol + * names may only be possible in situ); these environments may wish to define + * a series of actions to be applied in situ to obtain a meaningful stack + * trace. + * + * These two mechanisms -- user-level statically defined tracing and assisting + * DTrace actions -- are provided via DTrace _helpers_. Helpers are specified + * via DOF, but unlike enabling DOF, helper DOF may contain definitions of + * providers, probes and their arguments. If a helper wishes to provide + * action assistance, probe descriptions and corresponding DIF actions may be + * specified in the helper DOF. For such helper actions, however, the probe + * description describes the specific helper: all DTrace helpers have the + * provider name "dtrace" and the module name "helper", and the name of the + * helper is contained in the function name (for example, the ustack() helper + * is named "ustack"). Any helper-specific name may be contained in the name + * (for example, if a helper were to have a constructor, it might be named + * "dtrace:helper:<helper>:init"). Helper actions are only called when the + * action that they are helping is taken. Helper actions may only return DIF + * expressions, and may only call the following subroutines: + * + * alloca() <= Allocates memory out of the consumer's scratch space + * bcopy() <= Copies memory to scratch space + * copyin() <= Copies memory from user-level into consumer's scratch + * copyinto() <= Copies memory into a specific location in scratch + * copyinstr() <= Copies a string into a specific location in scratch + * + * Helper actions may only access the following built-in variables: + * + * curthread <= Current kthread_t pointer + * tid <= Current thread identifier + * pid <= Current process identifier + * ppid <= Parent process identifier + * uid <= Current user ID + * gid <= Current group ID + * execname <= Current executable name + * zonename <= Current zone name + * + * Helper actions may not manipulate or allocate dynamic variables, but they + * may have clause-local and statically-allocated global variables. The + * helper action variable state is specific to the helper action -- variables + * used by the helper action may not be accessed outside of the helper + * action, and the helper action may not access variables that like outside + * of it. Helper actions may not load from kernel memory at-large; they are + * restricting to loading current user state (via copyin() and variants) and + * scratch space. As with probe enablings, helper actions are executed in + * program order. The result of the helper action is the result of the last + * executing helper expression. + * + * Helpers -- composed of either providers/probes or probes/actions (or both) + * -- are added by opening the "helper" minor node, and issuing an ioctl(2) + * (DTRACEHIOC_ADDDOF) that specifies the dof_helper_t structure. This + * encapsulates the name and base address of the user-level library or + * executable publishing the helpers and probes as well as the DOF that + * contains the definitions of those helpers and probes. + * + * The DTRACEHIOC_ADD and DTRACEHIOC_REMOVE are left in place for legacy + * helpers and should no longer be used. No other ioctls are valid on the + * helper minor node. + */ +#define DTRACEHIOC (('d' << 24) | ('t' << 16) | ('h' << 8)) +#define DTRACEHIOC_ADD (DTRACEHIOC | 1) /* add helper */ +#define DTRACEHIOC_REMOVE (DTRACEHIOC | 2) /* remove helper */ +#define DTRACEHIOC_ADDDOF (DTRACEHIOC | 3) /* add helper DOF */ + +typedef struct dof_helper { + char dofhp_mod[DTRACE_MODNAMELEN]; /* executable or library name */ + uint64_t dofhp_addr; /* base address of object */ + uint64_t dofhp_dof; /* address of helper DOF */ +} dof_helper_t; + +#define DTRACEMNR_DTRACE "dtrace" /* node for DTrace ops */ +#define DTRACEMNR_HELPER "helper" /* node for helpers */ +#define DTRACEMNRN_DTRACE 0 /* minor for DTrace ops */ +#define DTRACEMNRN_HELPER 1 /* minor for helpers */ +#define DTRACEMNRN_CLONE 2 /* first clone minor */ + +#if defined(_KERNEL) || defined(IN_RING0) + +/* + * DTrace Provider API + * + * The following functions are implemented by the DTrace framework and are + * used to implement separate in-kernel DTrace providers. Common functions + * are provided in uts/common/os/dtrace.c. ISA-dependent subroutines are + * defined in uts/<isa>/dtrace/dtrace_asm.s or uts/<isa>/dtrace/dtrace_isa.c. + * + * The provider API has two halves: the API that the providers consume from + * DTrace, and the API that providers make available to DTrace. + * + * 1 Framework-to-Provider API + * + * 1.1 Overview + * + * The Framework-to-Provider API is represented by the dtrace_pops structure + * that the provider passes to the framework when registering itself. This + * structure consists of the following members: + * + * dtps_provide() <-- Provide all probes, all modules + * dtps_provide_module() <-- Provide all probes in specified module + * dtps_enable() <-- Enable specified probe + * dtps_disable() <-- Disable specified probe + * dtps_suspend() <-- Suspend specified probe + * dtps_resume() <-- Resume specified probe + * dtps_getargdesc() <-- Get the argument description for args[X] + * dtps_getargval() <-- Get the value for an argX or args[X] variable + * dtps_usermode() <-- Find out if the probe was fired in user mode + * dtps_destroy() <-- Destroy all state associated with this probe + * + * 1.2 void dtps_provide(void *arg, const dtrace_probedesc_t *spec) + * + * 1.2.1 Overview + * + * Called to indicate that the provider should provide all probes. If the + * specified description is non-NULL, dtps_provide() is being called because + * no probe matched a specified probe -- if the provider has the ability to + * create custom probes, it may wish to create a probe that matches the + * specified description. + * + * 1.2.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is a pointer to a probe description that the provider may + * wish to consider when creating custom probes. The provider is expected to + * call back into the DTrace framework via dtrace_probe_create() to create + * any necessary probes. dtps_provide() may be called even if the provider + * has made available all probes; the provider should check the return value + * of dtrace_probe_create() to handle this case. Note that the provider need + * not implement both dtps_provide() and dtps_provide_module(); see + * "Arguments and Notes" for dtrace_register(), below. + * + * 1.2.3 Return value + * + * None. + * + * 1.2.4 Caller's context + * + * dtps_provide() is typically called from open() or ioctl() context, but may + * be called from other contexts as well. The DTrace framework is locked in + * such a way that providers may not register or unregister. This means that + * the provider may not call any DTrace API that affects its registration with + * the framework, including dtrace_register(), dtrace_unregister(), + * dtrace_invalidate(), and dtrace_condense(). However, the context is such + * that the provider may (and indeed, is expected to) call probe-related + * DTrace routines, including dtrace_probe_create(), dtrace_probe_lookup(), + * and dtrace_probe_arg(). + * + * 1.3 void dtps_provide_module(void *arg, struct modctl *mp) + * + * 1.3.1 Overview + * + * Called to indicate that the provider should provide all probes in the + * specified module. + * + * 1.3.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is a pointer to a modctl structure that indicates the + * module for which probes should be created. + * + * 1.3.3 Return value + * + * None. + * + * 1.3.4 Caller's context + * + * dtps_provide_module() may be called from open() or ioctl() context, but + * may also be called from a module loading context. mod_lock is held, and + * the DTrace framework is locked in such a way that providers may not + * register or unregister. This means that the provider may not call any + * DTrace API that affects its registration with the framework, including + * dtrace_register(), dtrace_unregister(), dtrace_invalidate(), and + * dtrace_condense(). However, the context is such that the provider may (and + * indeed, is expected to) call probe-related DTrace routines, including + * dtrace_probe_create(), dtrace_probe_lookup(), and dtrace_probe_arg(). Note + * that the provider need not implement both dtps_provide() and + * dtps_provide_module(); see "Arguments and Notes" for dtrace_register(), + * below. + * + * 1.4 int dtps_enable(void *arg, dtrace_id_t id, void *parg) + * + * 1.4.1 Overview + * + * Called to enable the specified probe. + * + * 1.4.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be enabled. The third + * argument is the probe argument as passed to dtrace_probe_create(). + * dtps_enable() will be called when a probe transitions from not being + * enabled at all to having one or more ECB. The number of ECBs associated + * with the probe may change without subsequent calls into the provider. + * When the number of ECBs drops to zero, the provider will be explicitly + * told to disable the probe via dtps_disable(). dtrace_probe() should never + * be called for a probe identifier that hasn't been explicitly enabled via + * dtps_enable(). + * + * 1.4.3 Return value + * + * On success, dtps_enable() should return 0. On failure, -1 should be + * returned. + * + * 1.4.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. cpu_lock is held. mod_lock is not held and may not + * be acquired. + * + * 1.5 void dtps_disable(void *arg, dtrace_id_t id, void *parg) + * + * 1.5.1 Overview + * + * Called to disable the specified probe. + * + * 1.5.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be disabled. The third + * argument is the probe argument as passed to dtrace_probe_create(). + * dtps_disable() will be called when a probe transitions from being enabled + * to having zero ECBs. dtrace_probe() should never be called for a probe + * identifier that has been explicitly enabled via dtps_disable(). + * + * 1.5.3 Return value + * + * None. + * + * 1.5.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. cpu_lock is held. mod_lock is not held and may not + * be acquired. + * + * 1.6 void dtps_suspend(void *arg, dtrace_id_t id, void *parg) + * + * 1.6.1 Overview + * + * Called to suspend the specified enabled probe. This entry point is for + * providers that may need to suspend some or all of their probes when CPUs + * are being powered on or when the boot monitor is being entered for a + * prolonged period of time. + * + * 1.6.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be suspended. The + * third argument is the probe argument as passed to dtrace_probe_create(). + * dtps_suspend will only be called on an enabled probe. Providers that + * provide a dtps_suspend entry point will want to take roughly the action + * that it takes for dtps_disable. + * + * 1.6.3 Return value + * + * None. + * + * 1.6.4 Caller's context + * + * Interrupts are disabled. The DTrace framework is in a state such that the + * specified probe cannot be disabled or destroyed for the duration of + * dtps_suspend(). As interrupts are disabled, the provider is afforded + * little latitude; the provider is expected to do no more than a store to + * memory. + * + * 1.7 void dtps_resume(void *arg, dtrace_id_t id, void *parg) + * + * 1.7.1 Overview + * + * Called to resume the specified enabled probe. This entry point is for + * providers that may need to resume some or all of their probes after the + * completion of an event that induced a call to dtps_suspend(). + * + * 1.7.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be resumed. The + * third argument is the probe argument as passed to dtrace_probe_create(). + * dtps_resume will only be called on an enabled probe. Providers that + * provide a dtps_resume entry point will want to take roughly the action + * that it takes for dtps_enable. + * + * 1.7.3 Return value + * + * None. + * + * 1.7.4 Caller's context + * + * Interrupts are disabled. The DTrace framework is in a state such that the + * specified probe cannot be disabled or destroyed for the duration of + * dtps_resume(). As interrupts are disabled, the provider is afforded + * little latitude; the provider is expected to do no more than a store to + * memory. + * + * 1.8 void dtps_getargdesc(void *arg, dtrace_id_t id, void *parg, + * dtrace_argdesc_t *desc) + * + * 1.8.1 Overview + * + * Called to retrieve the argument description for an args[X] variable. + * + * 1.8.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * fourth argument is a pointer to the argument description. This + * description is both an input and output parameter: it contains the + * index of the desired argument in the dtargd_ndx field, and expects + * the other fields to be filled in upon return. If there is no argument + * corresponding to the specified index, the dtargd_ndx field should be set + * to DTRACE_ARGNONE. + * + * 1.8.3 Return value + * + * None. The dtargd_ndx, dtargd_native, dtargd_xlate and dtargd_mapping + * members of the dtrace_argdesc_t structure are all output values. + * + * 1.8.4 Caller's context + * + * dtps_getargdesc() is called from ioctl() context. mod_lock is held, and + * the DTrace framework is locked in such a way that providers may not + * register or unregister. This means that the provider may not call any + * DTrace API that affects its registration with the framework, including + * dtrace_register(), dtrace_unregister(), dtrace_invalidate(), and + * dtrace_condense(). + * + * 1.9 uint64_t dtps_getargval(void *arg, dtrace_id_t id, void *parg, + * int argno, int aframes) + * + * 1.9.1 Overview + * + * Called to retrieve a value for an argX or args[X] variable. + * + * 1.9.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * fourth argument is the number of the argument (the X in the example in + * 1.9.1). The fifth argument is the number of stack frames that were used + * to get from the actual place in the code that fired the probe to + * dtrace_probe() itself, the so-called artificial frames. This argument may + * be used to descend an appropriate number of frames to find the correct + * values. If this entry point is left NULL, the dtrace_getarg() built-in + * function is used. + * + * 1.9.3 Return value + * + * The value of the argument. + * + * 1.9.4 Caller's context + * + * This is called from within dtrace_probe() meaning that interrupts + * are disabled. No locks should be taken within this entry point. + * + * 1.10 int dtps_usermode(void *arg, dtrace_id_t id, void *parg) + * + * 1.10.1 Overview + * + * Called to determine if the probe was fired in a user context. + * + * 1.10.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the current probe. The third + * argument is the probe argument as passed to dtrace_probe_create(). This + * entry point must not be left NULL for providers whose probes allow for + * mixed mode tracing, that is to say those probes that can fire during + * kernel- _or_ user-mode execution + * + * 1.10.3 Return value + * + * A boolean value. + * + * 1.10.4 Caller's context + * + * This is called from within dtrace_probe() meaning that interrupts + * are disabled. No locks should be taken within this entry point. + * + * 1.11 void dtps_destroy(void *arg, dtrace_id_t id, void *parg) + * + * 1.11.1 Overview + * + * Called to destroy the specified probe. + * + * 1.11.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_register(). The + * second argument is the identifier of the probe to be destroyed. The third + * argument is the probe argument as passed to dtrace_probe_create(). The + * provider should free all state associated with the probe. The framework + * guarantees that dtps_destroy() is only called for probes that have either + * been disabled via dtps_disable() or were never enabled via dtps_enable(). + * Once dtps_disable() has been called for a probe, no further call will be + * made specifying the probe. + * + * 1.11.3 Return value + * + * None. + * + * 1.11.4 Caller's context + * + * The DTrace framework is locked in such a way that it may not be called + * back into at all. mod_lock is held. cpu_lock is not held, and may not be + * acquired. + * + * + * 2 Provider-to-Framework API + * + * 2.1 Overview + * + * The Provider-to-Framework API provides the mechanism for the provider to + * register itself with the DTrace framework, to create probes, to lookup + * probes and (most importantly) to fire probes. The Provider-to-Framework + * consists of: + * + * dtrace_register() <-- Register a provider with the DTrace framework + * dtrace_unregister() <-- Remove a provider's DTrace registration + * dtrace_invalidate() <-- Invalidate the specified provider + * dtrace_condense() <-- Remove a provider's unenabled probes + * dtrace_attached() <-- Indicates whether or not DTrace has attached + * dtrace_probe_create() <-- Create a DTrace probe + * dtrace_probe_lookup() <-- Lookup a DTrace probe based on its name + * dtrace_probe_arg() <-- Return the probe argument for a specific probe + * dtrace_probe() <-- Fire the specified probe + * + * 2.2 int dtrace_register(const char *name, const dtrace_pattr_t *pap, + * uint32_t priv, cred_t *cr, const dtrace_pops_t *pops, void *arg, + * dtrace_provider_id_t *idp) + * + * 2.2.1 Overview + * + * dtrace_register() registers the calling provider with the DTrace + * framework. It should generally be called by DTrace providers in their + * attach(9E) entry point. + * + * 2.2.2 Arguments and Notes + * + * The first argument is the name of the provider. The second argument is a + * pointer to the stability attributes for the provider. The third argument + * is the privilege flags for the provider, and must be some combination of: + * + * DTRACE_PRIV_NONE <= All users may enable probes from this provider + * + * DTRACE_PRIV_PROC <= Any user with privilege of PRIV_DTRACE_PROC may + * enable probes from this provider + * + * DTRACE_PRIV_USER <= Any user with privilege of PRIV_DTRACE_USER may + * enable probes from this provider + * + * DTRACE_PRIV_KERNEL <= Any user with privilege of PRIV_DTRACE_KERNEL + * may enable probes from this provider + * + * DTRACE_PRIV_OWNER <= This flag places an additional constraint on + * the privilege requirements above. These probes + * require either (a) a user ID matching the user + * ID of the cred passed in the fourth argument + * or (b) the PRIV_PROC_OWNER privilege. + * + * DTRACE_PRIV_ZONEOWNER<= This flag places an additional constraint on + * the privilege requirements above. These probes + * require either (a) a zone ID matching the zone + * ID of the cred passed in the fourth argument + * or (b) the PRIV_PROC_ZONE privilege. + * + * Note that these flags designate the _visibility_ of the probes, not + * the conditions under which they may or may not fire. + * + * The fourth argument is the credential that is associated with the + * provider. This argument should be NULL if the privilege flags don't + * include DTRACE_PRIV_OWNER or DTRACE_PRIV_ZONEOWNER. If non-NULL, the + * framework stashes the uid and zoneid represented by this credential + * for use at probe-time, in implicit predicates. These limit visibility + * of the probes to users and/or zones which have sufficient privilege to + * access them. + * + * The fifth argument is a DTrace provider operations vector, which provides + * the implementation for the Framework-to-Provider API. (See Section 1, + * above.) This must be non-NULL, and each member must be non-NULL. The + * exceptions to this are (1) the dtps_provide() and dtps_provide_module() + * members (if the provider so desires, _one_ of these members may be left + * NULL -- denoting that the provider only implements the other) and (2) + * the dtps_suspend() and dtps_resume() members, which must either both be + * NULL or both be non-NULL. + * + * The sixth argument is a cookie to be specified as the first argument for + * each function in the Framework-to-Provider API. This argument may have + * any value. + * + * The final argument is a pointer to dtrace_provider_id_t. If + * dtrace_register() successfully completes, the provider identifier will be + * stored in the memory pointed to be this argument. This argument must be + * non-NULL. + * + * 2.2.3 Return value + * + * On success, dtrace_register() returns 0 and stores the new provider's + * identifier into the memory pointed to by the idp argument. On failure, + * dtrace_register() returns an errno: + * + * EINVAL The arguments passed to dtrace_register() were somehow invalid. + * This may because a parameter that must be non-NULL was NULL, + * because the name was invalid (either empty or an illegal + * provider name) or because the attributes were invalid. + * + * No other failure code is returned. + * + * 2.2.4 Caller's context + * + * dtrace_register() may induce calls to dtrace_provide(); the provider must + * hold no locks across dtrace_register() that may also be acquired by + * dtrace_provide(). cpu_lock and mod_lock must not be held. + * + * 2.3 int dtrace_unregister(dtrace_provider_t id) + * + * 2.3.1 Overview + * + * Unregisters the specified provider from the DTrace framework. It should + * generally be called by DTrace providers in their detach(9E) entry point. + * + * 2.3.2 Arguments and Notes + * + * The only argument is the provider identifier, as returned from a + * successful call to dtrace_register(). As a result of calling + * dtrace_unregister(), the DTrace framework will call back into the provider + * via the dtps_destroy() entry point. Once dtrace_unregister() successfully + * completes, however, the DTrace framework will no longer make calls through + * the Framework-to-Provider API. + * + * 2.3.3 Return value + * + * On success, dtrace_unregister returns 0. On failure, dtrace_unregister() + * returns an errno: + * + * EBUSY There are currently processes that have the DTrace pseudodevice + * open, or there exists an anonymous enabling that hasn't yet + * been claimed. + * + * No other failure code is returned. + * + * 2.3.4 Caller's context + * + * Because a call to dtrace_unregister() may induce calls through the + * Framework-to-Provider API, the caller may not hold any lock across + * dtrace_register() that is also acquired in any of the Framework-to- + * Provider API functions. Additionally, mod_lock may not be held. + * + * 2.4 void dtrace_invalidate(dtrace_provider_id_t id) + * + * 2.4.1 Overview + * + * Invalidates the specified provider. All subsequent probe lookups for the + * specified provider will fail, but its probes will not be removed. + * + * 2.4.2 Arguments and note + * + * The only argument is the provider identifier, as returned from a + * successful call to dtrace_register(). In general, a provider's probes + * always remain valid; dtrace_invalidate() is a mechanism for invalidating + * an entire provider, regardless of whether or not probes are enabled or + * not. Note that dtrace_invalidate() will _not_ prevent already enabled + * probes from firing -- it will merely prevent any new enablings of the + * provider's probes. + * + * 2.5 int dtrace_condense(dtrace_provider_id_t id) + * + * 2.5.1 Overview + * + * Removes all the unenabled probes for the given provider. This function is + * not unlike dtrace_unregister(), except that it doesn't remove the + * provider just as many of its associated probes as it can. + * + * 2.5.2 Arguments and Notes + * + * As with dtrace_unregister(), the sole argument is the provider identifier + * as returned from a successful call to dtrace_register(). As a result of + * calling dtrace_condense(), the DTrace framework will call back into the + * given provider's dtps_destroy() entry point for each of the provider's + * unenabled probes. + * + * 2.5.3 Return value + * + * Currently, dtrace_condense() always returns 0. However, consumers of this + * function should check the return value as appropriate; its behavior may + * change in the future. + * + * 2.5.4 Caller's context + * + * As with dtrace_unregister(), the caller may not hold any lock across + * dtrace_condense() that is also acquired in the provider's entry points. + * Also, mod_lock may not be held. + * + * 2.6 int dtrace_attached() + * + * 2.6.1 Overview + * + * Indicates whether or not DTrace has attached. + * + * 2.6.2 Arguments and Notes + * + * For most providers, DTrace makes initial contact beyond registration. + * That is, once a provider has registered with DTrace, it waits to hear + * from DTrace to create probes. However, some providers may wish to + * proactively create probes without first being told by DTrace to do so. + * If providers wish to do this, they must first call dtrace_attached() to + * determine if DTrace itself has attached. If dtrace_attached() returns 0, + * the provider must not make any other Provider-to-Framework API call. + * + * 2.6.3 Return value + * + * dtrace_attached() returns 1 if DTrace has attached, 0 otherwise. + * + * 2.7 int dtrace_probe_create(dtrace_provider_t id, const char *mod, + * const char *func, const char *name, int aframes, void *arg) + * + * 2.7.1 Overview + * + * Creates a probe with specified module name, function name, and name. + * + * 2.7.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second, third, and fourth + * arguments are the module name, function name, and probe name, + * respectively. Of these, module name and function name may both be NULL + * (in which case the probe is considered to be unanchored), or they may both + * be non-NULL. The name must be non-NULL, and must point to a non-empty + * string. + * + * The fifth argument is the number of artificial stack frames that will be + * found on the stack when dtrace_probe() is called for the new probe. These + * artificial frames will be automatically be pruned should the stack() or + * stackdepth() functions be called as part of one of the probe's ECBs. If + * the parameter doesn't add an artificial frame, this parameter should be + * zero. + * + * The final argument is a probe argument that will be passed back to the + * provider when a probe-specific operation is called. (e.g., via + * dtps_enable(), dtps_disable(), etc.) + * + * Note that it is up to the provider to be sure that the probe that it + * creates does not already exist -- if the provider is unsure of the probe's + * existence, it should assure its absence with dtrace_probe_lookup() before + * calling dtrace_probe_create(). + * + * 2.7.3 Return value + * + * dtrace_probe_create() always succeeds, and always returns the identifier + * of the newly-created probe. + * + * 2.7.4 Caller's context + * + * While dtrace_probe_create() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may be called from other + * non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.8 dtrace_id_t dtrace_probe_lookup(dtrace_provider_t id, const char *mod, + * const char *func, const char *name) + * + * 2.8.1 Overview + * + * Looks up a probe based on provdider and one or more of module name, + * function name and probe name. + * + * 2.8.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second, third, and fourth + * arguments are the module name, function name, and probe name, + * respectively. Any of these may be NULL; dtrace_probe_lookup() will return + * the identifier of the first probe that is provided by the specified + * provider and matches all of the non-NULL matching criteria. + * dtrace_probe_lookup() is generally used by a provider to be check the + * existence of a probe before creating it with dtrace_probe_create(). + * + * 2.8.3 Return value + * + * If the probe exists, returns its identifier. If the probe does not exist, + * return DTRACE_IDNONE. + * + * 2.8.4 Caller's context + * + * While dtrace_probe_lookup() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may also be called from + * other non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.9 void *dtrace_probe_arg(dtrace_provider_t id, dtrace_id_t probe) + * + * 2.9.1 Overview + * + * Returns the probe argument associated with the specified probe. + * + * 2.9.2 Arguments and Notes + * + * The first argument is the provider identifier, as returned from a + * successful call to dtrace_register(). The second argument is a probe + * identifier, as returned from dtrace_probe_lookup() or + * dtrace_probe_create(). This is useful if a probe has multiple + * provider-specific components to it: the provider can create the probe + * once with provider-specific state, and then add to the state by looking + * up the probe based on probe identifier. + * + * 2.9.3 Return value + * + * Returns the argument associated with the specified probe. If the + * specified probe does not exist, or if the specified probe is not provided + * by the specified provider, NULL is returned. + * + * 2.9.4 Caller's context + * + * While dtrace_probe_arg() is generally expected to be called from + * dtps_provide() and/or dtps_provide_module(), it may also be called from + * other non-DTrace contexts. Neither cpu_lock nor mod_lock may be held. + * + * 2.10 void dtrace_probe(dtrace_id_t probe, uintptr_t arg0, uintptr_t arg1, + * uintptr_t arg2, uintptr_t arg3, uintptr_t arg4) + * + * 2.10.1 Overview + * + * The epicenter of DTrace: fires the specified probes with the specified + * arguments. + * + * 2.10.2 Arguments and Notes + * + * The first argument is a probe identifier as returned by + * dtrace_probe_create() or dtrace_probe_lookup(). The second through sixth + * arguments are the values to which the D variables "arg0" through "arg4" + * will be mapped. + * + * dtrace_probe() should be called whenever the specified probe has fired -- + * however the provider defines it. + * + * 2.10.3 Return value + * + * None. + * + * 2.10.4 Caller's context + * + * dtrace_probe() may be called in virtually any context: kernel, user, + * interrupt, high-level interrupt, with arbitrary adaptive locks held, with + * dispatcher locks held, with interrupts disabled, etc. The only latitude + * that must be afforded to DTrace is the ability to make calls within + * itself (and to its in-kernel subroutines) and the ability to access + * arbitrary (but mapped) memory. On some platforms, this constrains + * context. For example, on UltraSPARC, dtrace_probe() cannot be called + * from any context in which TL is greater than zero. dtrace_probe() may + * also not be called from any routine which may be called by dtrace_probe() + * -- which includes functions in the DTrace framework and some in-kernel + * DTrace subroutines. All such functions "dtrace_"; providers that + * instrument the kernel arbitrarily should be sure to not instrument these + * routines. + */ +typedef struct dtrace_pops { + void (*dtps_provide)(void *arg, const dtrace_probedesc_t *spec); + void (*dtps_provide_module)(void *arg, struct modctl *mp); + int (*dtps_enable)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_disable)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_suspend)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_resume)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_getargdesc)(void *arg, dtrace_id_t id, void *parg, + dtrace_argdesc_t *desc); + uint64_t (*dtps_getargval)(void *arg, dtrace_id_t id, void *parg, + int argno, int aframes); + int (*dtps_usermode)(void *arg, dtrace_id_t id, void *parg); + void (*dtps_destroy)(void *arg, dtrace_id_t id, void *parg); +} dtrace_pops_t; + +typedef uintptr_t dtrace_provider_id_t; + +extern int dtrace_register(const char *, const dtrace_pattr_t *, uint32_t, + cred_t *, const dtrace_pops_t *, void *, dtrace_provider_id_t *); +extern int dtrace_unregister(dtrace_provider_id_t); +extern int dtrace_condense(dtrace_provider_id_t); +extern void dtrace_invalidate(dtrace_provider_id_t); +extern dtrace_id_t dtrace_probe_lookup(dtrace_provider_id_t, const char *, + const char *, const char *); +extern dtrace_id_t dtrace_probe_create(dtrace_provider_id_t, const char *, + const char *, const char *, int, void *); +extern void *dtrace_probe_arg(dtrace_provider_id_t, dtrace_id_t); +extern void dtrace_probe(dtrace_id_t, uintptr_t arg0, uintptr_t arg1, + uintptr_t arg2, uintptr_t arg3, uintptr_t arg4); + +/* + * DTrace Meta Provider API + * + * The following functions are implemented by the DTrace framework and are + * used to implement meta providers. Meta providers plug into the DTrace + * framework and are used to instantiate new providers on the fly. At + * present, there is only one type of meta provider and only one meta + * provider may be registered with the DTrace framework at a time. The + * sole meta provider type provides user-land static tracing facilities + * by taking meta probe descriptions and adding a corresponding provider + * into the DTrace framework. + * + * 1 Framework-to-Provider + * + * 1.1 Overview + * + * The Framework-to-Provider API is represented by the dtrace_mops structure + * that the meta provider passes to the framework when registering itself as + * a meta provider. This structure consists of the following members: + * + * dtms_create_probe() <-- Add a new probe to a created provider + * dtms_provide_pid() <-- Create a new provider for a given process + * dtms_remove_pid() <-- Remove a previously created provider + * + * 1.2 void dtms_create_probe(void *arg, void *parg, + * dtrace_helper_probedesc_t *probedesc); + * + * 1.2.1 Overview + * + * Called by the DTrace framework to create a new probe in a provider + * created by this meta provider. + * + * 1.2.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is the provider cookie for the associated provider; + * this is obtained from the return value of dtms_provide_pid(). The third + * argument is the helper probe description. + * + * 1.2.3 Return value + * + * None + * + * 1.2.4 Caller's context + * + * dtms_create_probe() is called from either ioctl() or module load context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context is + * such that the provider may (and is expected to) call provider-related + * DTrace provider APIs including dtrace_probe_create(). + * + * 1.3 void *dtms_provide_pid(void *arg, dtrace_meta_provider_t *mprov, + * pid_t pid) + * + * 1.3.1 Overview + * + * Called by the DTrace framework to instantiate a new provider given the + * description of the provider and probes in the mprov argument. The + * meta provider should call dtrace_register() to insert the new provider + * into the DTrace framework. + * + * 1.3.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is a pointer to a structure describing the new + * helper provider. The third argument is the process identifier for + * process associated with this new provider. Note that the name of the + * provider as passed to dtrace_register() should be the contatenation of + * the dtmpb_provname member of the mprov argument and the processs + * identifier as a string. + * + * 1.3.3 Return value + * + * The cookie for the provider that the meta provider creates. This is + * the same value that it passed to dtrace_register(). + * + * 1.3.4 Caller's context + * + * dtms_provide_pid() is called from either ioctl() or module load context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context + * is such that the provider may -- and is expected to -- call + * provider-related DTrace provider APIs including dtrace_register(). + * + * 1.4 void dtms_remove_pid(void *arg, dtrace_meta_provider_t *mprov, + * pid_t pid) + * + * 1.4.1 Overview + * + * Called by the DTrace framework to remove a provider that had previously + * been instantiated via the dtms_provide_pid() entry point. The meta + * provider need not remove the provider immediately, but this entry + * point indicates that the provider should be removed as soon as possible + * using the dtrace_unregister() API. + * + * 1.4.2 Arguments and notes + * + * The first argument is the cookie as passed to dtrace_meta_register(). + * The second argument is a pointer to a structure describing the helper + * provider. The third argument is the process identifier for process + * associated with this new provider. + * + * 1.4.3 Return value + * + * None + * + * 1.4.4 Caller's context + * + * dtms_remove_pid() is called from either ioctl() or exit() context. + * The DTrace framework is locked in such a way that meta providers may not + * register or unregister. This means that the meta provider cannot call + * dtrace_meta_register() or dtrace_meta_unregister(). However, the context + * is such that the provider may -- and is expected to -- call + * provider-related DTrace provider APIs including dtrace_unregister(). + */ +typedef struct dtrace_helper_probedesc { + char *dthpb_mod; /* probe module */ + char *dthpb_func; /* probe function */ + char *dthpb_name; /* probe name */ + uint64_t dthpb_base; /* base address */ + uint32_t *dthpb_offs; /* offsets array */ + uint32_t *dthpb_enoffs; /* is-enabled offsets array */ + uint32_t dthpb_noffs; /* offsets count */ + uint32_t dthpb_nenoffs; /* is-enabled offsets count */ + uint8_t *dthpb_args; /* argument mapping array */ + uint8_t dthpb_xargc; /* translated argument count */ + uint8_t dthpb_nargc; /* native argument count */ + char *dthpb_xtypes; /* translated types strings */ + char *dthpb_ntypes; /* native types strings */ +} dtrace_helper_probedesc_t; + +typedef struct dtrace_helper_provdesc { + char *dthpv_provname; /* provider name */ + dtrace_pattr_t dthpv_pattr; /* stability attributes */ +} dtrace_helper_provdesc_t; + +typedef struct dtrace_mops { + void (*dtms_create_probe)(void *, void *, dtrace_helper_probedesc_t *); + void *(*dtms_provide_pid)(void *, dtrace_helper_provdesc_t *, pid_t); + void (*dtms_remove_pid)(void *, dtrace_helper_provdesc_t *, pid_t); +} dtrace_mops_t; + +typedef uintptr_t dtrace_meta_provider_id_t; + +extern int dtrace_meta_register(const char *, const dtrace_mops_t *, void *, + dtrace_meta_provider_id_t *); +extern int dtrace_meta_unregister(dtrace_meta_provider_id_t); + +/* + * DTrace Kernel Hooks + * + * The following functions are implemented by the base kernel and form a set of + * hooks used by the DTrace framework. DTrace hooks are implemented in either + * uts/common/os/dtrace_subr.c, an ISA-specific assembly file, or in a + * uts/<platform>/os/dtrace_subr.c corresponding to each hardware platform. + */ + +typedef enum dtrace_vtime_state { + DTRACE_VTIME_INACTIVE = 0, /* No DTrace, no TNF */ + DTRACE_VTIME_ACTIVE, /* DTrace virtual time, no TNF */ + DTRACE_VTIME_INACTIVE_TNF, /* No DTrace, TNF active */ + DTRACE_VTIME_ACTIVE_TNF /* DTrace virtual time _and_ TNF */ +} dtrace_vtime_state_t; + +extern dtrace_vtime_state_t dtrace_vtime_active; +extern void dtrace_vtime_switch(kthread_t *next); +extern void dtrace_vtime_enable_tnf(void); +extern void dtrace_vtime_disable_tnf(void); +extern void dtrace_vtime_enable(void); +extern void dtrace_vtime_disable(void); + +struct regs; + +extern int (*dtrace_pid_probe_ptr)(struct regs *); +extern int (*dtrace_return_probe_ptr)(struct regs *); +extern void (*dtrace_fasttrap_fork_ptr)(proc_t *, proc_t *); +extern void (*dtrace_fasttrap_exec_ptr)(proc_t *); +extern void (*dtrace_fasttrap_exit_ptr)(proc_t *); +extern void dtrace_fasttrap_fork(proc_t *, proc_t *); + +typedef uintptr_t dtrace_icookie_t; +typedef void (*dtrace_xcall_t)(void *); + +extern dtrace_icookie_t dtrace_interrupt_disable(void); +extern void dtrace_interrupt_enable(dtrace_icookie_t); + +extern void dtrace_membar_producer(void); +extern void dtrace_membar_consumer(void); + +extern void (*dtrace_cpu_init)(processorid_t); +extern void (*dtrace_modload)(struct modctl *); +extern void (*dtrace_modunload)(struct modctl *); +extern void (*dtrace_helpers_cleanup)(VBDTVOID); +extern void (*dtrace_helpers_fork)(proc_t *parent, proc_t *child); +extern void (*dtrace_cpustart_init)(VBDTVOID); +extern void (*dtrace_cpustart_fini)(VBDTVOID); + +extern void (*dtrace_debugger_init)(VBDTVOID); +extern void (*dtrace_debugger_fini)(VBDTVOID); +extern dtrace_cacheid_t dtrace_predcache_id; + +extern hrtime_t dtrace_gethrtime(void); +extern void dtrace_sync(void); +extern void dtrace_toxic_ranges(void (*)(uintptr_t, uintptr_t)); +extern void dtrace_xcall(processorid_t, dtrace_xcall_t, void *); +#ifdef VBOX +extern void dtrace_vpanic(const char *, va_list); +#else +extern void dtrace_vpanic(const char *, __va_list); +#endif +extern void dtrace_panic(const char *, ...); + +extern int dtrace_safe_defer_signal(void); +extern void dtrace_safe_synchronous_signal(void); + +extern int dtrace_mach_aframes(void); + +#if defined(__i386) || defined(__amd64) +extern int dtrace_instr_size(uchar_t *instr); +extern int dtrace_instr_size_isa(uchar_t *, model_t, int *); +extern void dtrace_invop_add(int (*)(uintptr_t, uintptr_t *, uintptr_t)); +extern void dtrace_invop_remove(int (*)(uintptr_t, uintptr_t *, uintptr_t)); +extern void dtrace_invop_callsite(void); +#endif + +#ifdef __sparc +extern int dtrace_blksuword32(uintptr_t, uint32_t *, int); +extern void dtrace_getfsr(uint64_t *); +#endif + +#ifndef VBOX +# define VBDT_GET_CPUID() (CPU->cpu_id) +# define VBDT_GET_PROC() (ttoproc(curthread)) +#else +# define VBDT_GET_CPUID() (RTMpCpuId()) +# define VBDT_GET_PROC() (VBoxDtGetCurrentProc()) +proc_t *VBoxDtGetCurrentProc(void); +#endif + +#define DTRACE_CPUFLAG_ISSET(flag) \ + (cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags & (flag)) + +#define DTRACE_CPUFLAG_SET(flag) \ + (cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags |= (flag)) + +#define DTRACE_CPUFLAG_CLEAR(flag) \ + (cpu_core[VBDT_GET_CPUID()].cpuc_dtrace_flags &= ~(flag)) + +#endif /* _KERNEL || IN_RING0 */ + +#endif /* _ASM */ + +#if defined(__i386) || defined(__amd64) + +#define DTRACE_INVOP_PUSHL_EBP 1 +#define DTRACE_INVOP_POPL_EBP 2 +#define DTRACE_INVOP_LEAVE 3 +#define DTRACE_INVOP_NOP 4 +#define DTRACE_INVOP_RET 5 + +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DTRACE_H */ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace_impl.h b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace_impl.h new file mode 100644 index 00000000..a0e53d5a --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/common/sys/dtrace_impl.h @@ -0,0 +1,1304 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#ifndef _SYS_DTRACE_IMPL_H +#define _SYS_DTRACE_IMPL_H + +#ifndef VBOX +#pragma ident "%Z%%M% %I% %E% SMI" +#endif + +#ifdef __cplusplus +extern "C" { +#endif + +/* + * DTrace Dynamic Tracing Software: Kernel Implementation Interfaces + * + * Note: The contents of this file are private to the implementation of the + * Solaris system and DTrace subsystem and are subject to change at any time + * without notice. Applications and drivers using these interfaces will fail + * to run on future releases. These interfaces should not be used for any + * purpose except those expressly outlined in dtrace(7D) and libdtrace(3LIB). + * Please refer to the "Solaris Dynamic Tracing Guide" for more information. + */ + +#include <sys/dtrace.h> + +/* + * DTrace Implementation Constants and Typedefs + */ +#define DTRACE_MAXPROPLEN 128 +#define DTRACE_DYNVAR_CHUNKSIZE 256 + +struct dtrace_probe; +struct dtrace_ecb; +struct dtrace_predicate; +struct dtrace_action; +struct dtrace_provider; +struct dtrace_state; + +typedef struct dtrace_probe dtrace_probe_t; +typedef struct dtrace_ecb dtrace_ecb_t; +typedef struct dtrace_predicate dtrace_predicate_t; +typedef struct dtrace_action dtrace_action_t; +typedef struct dtrace_provider dtrace_provider_t; +typedef struct dtrace_meta dtrace_meta_t; +typedef struct dtrace_state dtrace_state_t; +typedef uint32_t dtrace_optid_t; +typedef uint32_t dtrace_specid_t; +typedef uint64_t dtrace_genid_t; + +/* + * DTrace Probes + * + * The probe is the fundamental unit of the DTrace architecture. Probes are + * created by DTrace providers, and managed by the DTrace framework. A probe + * is identified by a unique <provider, module, function, name> tuple, and has + * a unique probe identifier assigned to it. (Some probes are not associated + * with a specific point in text; these are called _unanchored probes_ and have + * no module or function associated with them.) Probes are represented as a + * dtrace_probe structure. To allow quick lookups based on each element of the + * probe tuple, probes are hashed by each of provider, module, function and + * name. (If a lookup is performed based on a regular expression, a + * dtrace_probekey is prepared, and a linear search is performed.) Each probe + * is additionally pointed to by a linear array indexed by its identifier. The + * identifier is the provider's mechanism for indicating to the DTrace + * framework that a probe has fired: the identifier is passed as the first + * argument to dtrace_probe(), where it is then mapped into the corresponding + * dtrace_probe structure. From the dtrace_probe structure, dtrace_probe() can + * iterate over the probe's list of enabling control blocks; see "DTrace + * Enabling Control Blocks", below.) + */ +struct dtrace_probe { + dtrace_id_t dtpr_id; /* probe identifier */ + dtrace_ecb_t *dtpr_ecb; /* ECB list; see below */ + dtrace_ecb_t *dtpr_ecb_last; /* last ECB in list */ + void *dtpr_arg; /* provider argument */ + dtrace_cacheid_t dtpr_predcache; /* predicate cache ID */ + int dtpr_aframes; /* artificial frames */ + dtrace_provider_t *dtpr_provider; /* pointer to provider */ + char *dtpr_mod; /* probe's module name */ + char *dtpr_func; /* probe's function name */ + char *dtpr_name; /* probe's name */ + dtrace_probe_t *dtpr_nextmod; /* next in module hash */ + dtrace_probe_t *dtpr_prevmod; /* previous in module hash */ + dtrace_probe_t *dtpr_nextfunc; /* next in function hash */ + dtrace_probe_t *dtpr_prevfunc; /* previous in function hash */ + dtrace_probe_t *dtpr_nextname; /* next in name hash */ + dtrace_probe_t *dtpr_prevname; /* previous in name hash */ + dtrace_genid_t dtpr_gen; /* probe generation ID */ +}; + +typedef int dtrace_probekey_f(const char *, const char *, int); + +typedef struct dtrace_probekey { + const char *dtpk_prov; /* provider name to match */ + dtrace_probekey_f *dtpk_pmatch; /* provider matching function */ + const char *dtpk_mod; /* module name to match */ + dtrace_probekey_f *dtpk_mmatch; /* module matching function */ + const char *dtpk_func; /* func name to match */ + dtrace_probekey_f *dtpk_fmatch; /* func matching function */ + const char *dtpk_name; /* name to match */ + dtrace_probekey_f *dtpk_nmatch; /* name matching function */ + dtrace_id_t dtpk_id; /* identifier to match */ +} dtrace_probekey_t; + +typedef struct dtrace_hashbucket { + struct dtrace_hashbucket *dthb_next; /* next on hash chain */ + dtrace_probe_t *dthb_chain; /* chain of probes */ + int dthb_len; /* number of probes here */ +} dtrace_hashbucket_t; + +typedef struct dtrace_hash { + dtrace_hashbucket_t **dth_tab; /* hash table */ + int dth_size; /* size of hash table */ + int dth_mask; /* mask to index into table */ + int dth_nbuckets; /* total number of buckets */ + uintptr_t dth_nextoffs; /* offset of next in probe */ + uintptr_t dth_prevoffs; /* offset of prev in probe */ + uintptr_t dth_stroffs; /* offset of str in probe */ +} dtrace_hash_t; + +/* + * DTrace Enabling Control Blocks + * + * When a provider wishes to fire a probe, it calls into dtrace_probe(), + * passing the probe identifier as the first argument. As described above, + * dtrace_probe() maps the identifier into a pointer to a dtrace_probe_t + * structure. This structure contains information about the probe, and a + * pointer to the list of Enabling Control Blocks (ECBs). Each ECB points to + * DTrace consumer state, and contains an optional predicate, and a list of + * actions. (Shown schematically below.) The ECB abstraction allows a single + * probe to be multiplexed across disjoint consumers, or across disjoint + * enablings of a single probe within one consumer. + * + * Enabling Control Block + * dtrace_ecb_t + * +------------------------+ + * | dtrace_epid_t ---------+--------------> Enabled Probe ID (EPID) + * | dtrace_state_t * ------+--------------> State associated with this ECB + * | dtrace_predicate_t * --+---------+ + * | dtrace_action_t * -----+----+ | + * | dtrace_ecb_t * ---+ | | | Predicate (if any) + * +-------------------+----+ | | dtrace_predicate_t + * | | +---> +--------------------+ + * | | | dtrace_difo_t * ---+----> DIFO + * | | +--------------------+ + * | | + * Next ECB | | Action + * (if any) | | dtrace_action_t + * : +--> +-------------------+ + * : | dtrace_actkind_t -+------> kind + * v | dtrace_difo_t * --+------> DIFO (if any) + * | dtrace_recdesc_t -+------> record descr. + * | dtrace_action_t * +------+ + * +-------------------+ | + * | Next action + * +-------------------------------+ (if any) + * | + * | Action + * | dtrace_action_t + * +--> +-------------------+ + * | dtrace_actkind_t -+------> kind + * | dtrace_difo_t * --+------> DIFO (if any) + * | dtrace_action_t * +------+ + * +-------------------+ | + * | Next action + * +-------------------------------+ (if any) + * | + * : + * v + * + * + * dtrace_probe() iterates over the ECB list. If the ECB needs less space + * than is available in the principal buffer, the ECB is processed: if the + * predicate is non-NULL, the DIF object is executed. If the result is + * non-zero, the action list is processed, with each action being executed + * accordingly. When the action list has been completely executed, processing + * advances to the next ECB. processing advances to the next ECB. If the + * result is non-zero; For each ECB, it first determines the The ECB + * abstraction allows disjoint consumers to multiplex on single probes. + */ +struct dtrace_ecb { + dtrace_epid_t dte_epid; /* enabled probe ID */ + uint32_t dte_alignment; /* required alignment */ + size_t dte_needed; /* bytes needed */ + size_t dte_size; /* total size of payload */ + dtrace_predicate_t *dte_predicate; /* predicate, if any */ + dtrace_action_t *dte_action; /* actions, if any */ + dtrace_ecb_t *dte_next; /* next ECB on probe */ + dtrace_state_t *dte_state; /* pointer to state */ + uint32_t dte_cond; /* security condition */ + dtrace_probe_t *dte_probe; /* pointer to probe */ + dtrace_action_t *dte_action_last; /* last action on ECB */ + uint64_t dte_uarg; /* library argument */ +}; + +struct dtrace_predicate { + dtrace_difo_t *dtp_difo; /* DIF object */ + dtrace_cacheid_t dtp_cacheid; /* cache identifier */ + int dtp_refcnt; /* reference count */ +}; + +struct dtrace_action { + dtrace_actkind_t dta_kind; /* kind of action */ + uint16_t dta_intuple; /* boolean: in aggregation */ + uint32_t dta_refcnt; /* reference count */ + dtrace_difo_t *dta_difo; /* pointer to DIFO */ + dtrace_recdesc_t dta_rec; /* record description */ + dtrace_action_t *dta_prev; /* previous action */ + dtrace_action_t *dta_next; /* next action */ +}; + +typedef struct dtrace_aggregation { + dtrace_action_t dtag_action; /* action; must be first */ + dtrace_aggid_t dtag_id; /* identifier */ + dtrace_ecb_t *dtag_ecb; /* corresponding ECB */ + dtrace_action_t *dtag_first; /* first action in tuple */ + uint32_t dtag_base; /* base of aggregation */ + uint8_t dtag_hasarg; /* boolean: has argument */ + uint64_t dtag_initial; /* initial value */ + void (*dtag_aggregate)(uint64_t *, uint64_t, uint64_t); +} dtrace_aggregation_t; + +/* + * DTrace Buffers + * + * Principal buffers, aggregation buffers, and speculative buffers are all + * managed with the dtrace_buffer structure. By default, this structure + * includes twin data buffers -- dtb_tomax and dtb_xamot -- that serve as the + * active and passive buffers, respectively. For speculative buffers, + * dtb_xamot will be NULL; for "ring" and "fill" buffers, dtb_xamot will point + * to a scratch buffer. For all buffer types, the dtrace_buffer structure is + * always allocated on a per-CPU basis; a single dtrace_buffer structure is + * never shared among CPUs. (That is, there is never true sharing of the + * dtrace_buffer structure; to prevent false sharing of the structure, it must + * always be aligned to the coherence granularity -- generally 64 bytes.) + * + * One of the critical design decisions of DTrace is that a given ECB always + * stores the same quantity and type of data. This is done to assure that the + * only metadata required for an ECB's traced data is the EPID. That is, from + * the EPID, the consumer can determine the data layout. (The data buffer + * layout is shown schematically below.) By assuring that one can determine + * data layout from the EPID, the metadata stream can be separated from the + * data stream -- simplifying the data stream enormously. + * + * base of data buffer ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * | data, cont. | + * +------+--------------------+------+ + * | EPID | data | | + * +------+--------------------+ | + * | || | + * | || | + * | \/ | + * : : + * . . + * . . + * . . + * : : + * | | + * limit of data buffer ---> +----------------------------------+ + * + * When evaluating an ECB, dtrace_probe() determines if the ECB's needs of the + * principal buffer (both scratch and payload) exceed the available space. If + * the ECB's needs exceed available space (and if the principal buffer policy + * is the default "switch" policy), the ECB is dropped, the buffer's drop count + * is incremented, and processing advances to the next ECB. If the ECB's needs + * can be met with the available space, the ECB is processed, but the offset in + * the principal buffer is only advanced if the ECB completes processing + * without error. + * + * When a buffer is to be switched (either because the buffer is the principal + * buffer with a "switch" policy or because it is an aggregation buffer), a + * cross call is issued to the CPU associated with the buffer. In the cross + * call context, interrupts are disabled, and the active and the inactive + * buffers are atomically switched. This involves switching the data pointers, + * copying the various state fields (offset, drops, errors, etc.) into their + * inactive equivalents, and clearing the state fields. Because interrupts are + * disabled during this procedure, the switch is guaranteed to appear atomic to + * dtrace_probe(). + * + * DTrace Ring Buffering + * + * To process a ring buffer correctly, one must know the oldest valid record. + * Processing starts at the oldest record in the buffer and continues until + * the end of the buffer is reached. Processing then resumes starting with + * the record stored at offset 0 in the buffer, and continues until the + * youngest record is processed. If trace records are of a fixed-length, + * determining the oldest record is trivial: + * + * - If the ring buffer has not wrapped, the oldest record is the record + * stored at offset 0. + * + * - If the ring buffer has wrapped, the oldest record is the record stored + * at the current offset. + * + * With variable length records, however, just knowing the current offset + * doesn't suffice for determining the oldest valid record: assuming that one + * allows for arbitrary data, one has no way of searching forward from the + * current offset to find the oldest valid record. (That is, one has no way + * of separating data from metadata.) It would be possible to simply refuse to + * process any data in the ring buffer between the current offset and the + * limit, but this leaves (potentially) an enormous amount of otherwise valid + * data unprocessed. + * + * To effect ring buffering, we track two offsets in the buffer: the current + * offset and the _wrapped_ offset. If a request is made to reserve some + * amount of data, and the buffer has wrapped, the wrapped offset is + * incremented until the wrapped offset minus the current offset is greater + * than or equal to the reserve request. This is done by repeatedly looking + * up the ECB corresponding to the EPID at the current wrapped offset, and + * incrementing the wrapped offset by the size of the data payload + * corresponding to that ECB. If this offset is greater than or equal to the + * limit of the data buffer, the wrapped offset is set to 0. Thus, the + * current offset effectively "chases" the wrapped offset around the buffer. + * Schematically: + * + * base of data buffer ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * | data, cont. | + * +------+---------------------------+ + * | EPID | data | + * current offset ---> +------+---------------------------+ + * | invalid data | + * wrapped offset ---> +------+--------------------+------+ + * | EPID | data | EPID | + * +------+--------+------+----+------+ + * | data | EPID | data | + * +---------------+------+-----------+ + * : : + * . . + * . ... valid data ... . + * . . + * : : + * +------+-------------+------+------+ + * | EPID | data | EPID | data | + * +------+------------++------+------+ + * | data, cont. | leftover | + * limit of data buffer ---> +-------------------+--------------+ + * + * If the amount of requested buffer space exceeds the amount of space + * available between the current offset and the end of the buffer: + * + * (1) all words in the data buffer between the current offset and the limit + * of the data buffer (marked "leftover", above) are set to + * DTRACE_EPIDNONE + * + * (2) the wrapped offset is set to zero + * + * (3) the iteration process described above occurs until the wrapped offset + * is greater than the amount of desired space. + * + * The wrapped offset is implemented by (re-)using the inactive offset. + * In a "switch" buffer policy, the inactive offset stores the offset in + * the inactive buffer; in a "ring" buffer policy, it stores the wrapped + * offset. + * + * DTrace Scratch Buffering + * + * Some ECBs may wish to allocate dynamically-sized temporary scratch memory. + * To accommodate such requests easily, scratch memory may be allocated in + * the buffer beyond the current offset plus the needed memory of the current + * ECB. If there isn't sufficient room in the buffer for the requested amount + * of scratch space, the allocation fails and an error is generated. Scratch + * memory is tracked in the dtrace_mstate_t and is automatically freed when + * the ECB ceases processing. Note that ring buffers cannot allocate their + * scratch from the principal buffer -- lest they needlessly overwrite older, + * valid data. Ring buffers therefore have their own dedicated scratch buffer + * from which scratch is allocated. + */ +#define DTRACEBUF_RING 0x0001 /* bufpolicy set to "ring" */ +#define DTRACEBUF_FILL 0x0002 /* bufpolicy set to "fill" */ +#define DTRACEBUF_NOSWITCH 0x0004 /* do not switch buffer */ +#define DTRACEBUF_WRAPPED 0x0008 /* ring buffer has wrapped */ +#define DTRACEBUF_DROPPED 0x0010 /* drops occurred */ +#define DTRACEBUF_ERROR 0x0020 /* errors occurred */ +#define DTRACEBUF_FULL 0x0040 /* "fill" buffer is full */ +#define DTRACEBUF_CONSUMED 0x0080 /* buffer has been consumed */ +#define DTRACEBUF_INACTIVE 0x0100 /* buffer is not yet active */ + +typedef struct dtrace_buffer { + uint64_t dtb_offset; /* current offset in buffer */ + uint64_t dtb_size; /* size of buffer */ + uint32_t dtb_flags; /* flags */ + uint32_t dtb_drops; /* number of drops */ + caddr_t dtb_tomax; /* active buffer */ + caddr_t dtb_xamot; /* inactive buffer */ + uint32_t dtb_xamot_flags; /* inactive flags */ + uint32_t dtb_xamot_drops; /* drops in inactive buffer */ + uint64_t dtb_xamot_offset; /* offset in inactive buffer */ + uint32_t dtb_errors; /* number of errors */ + uint32_t dtb_xamot_errors; /* errors in inactive buffer */ +#ifndef _LP64 + uint64_t dtb_pad1; +#endif +} dtrace_buffer_t; + +/* + * DTrace Aggregation Buffers + * + * Aggregation buffers use much of the same mechanism as described above + * ("DTrace Buffers"). However, because an aggregation is fundamentally a + * hash, there exists dynamic metadata associated with an aggregation buffer + * that is not associated with other kinds of buffers. This aggregation + * metadata is _only_ relevant for the in-kernel implementation of + * aggregations; it is not actually relevant to user-level consumers. To do + * this, we allocate dynamic aggregation data (hash keys and hash buckets) + * starting below the _limit_ of the buffer, and we allocate data from the + * _base_ of the buffer. When the aggregation buffer is copied out, _only_ the + * data is copied out; the metadata is simply discarded. Schematically, + * aggregation buffers look like: + * + * base of data buffer ---> +-------+------+-----------+-------+ + * | aggid | key | value | aggid | + * +-------+------+-----------+-------+ + * | key | + * +-------+-------+-----+------------+ + * | value | aggid | key | value | + * +-------+------++-----+------+-----+ + * | aggid | key | value | | + * +-------+------+-------------+ | + * | || | + * | || | + * | \/ | + * : : + * . . + * . . + * . . + * : : + * | /\ | + * | || +------------+ + * | || | | + * +---------------------+ | + * | hash keys | + * | (dtrace_aggkey structures) | + * | | + * +----------------------------------+ + * | hash buckets | + * | (dtrace_aggbuffer structure) | + * | | + * limit of data buffer ---> +----------------------------------+ + * + * + * As implied above, just as we assure that ECBs always store a constant + * amount of data, we assure that a given aggregation -- identified by its + * aggregation ID -- always stores data of a constant quantity and type. + * As with EPIDs, this allows the aggregation ID to serve as the metadata for a + * given record. + * + * Note that the size of the dtrace_aggkey structure must be sizeof (uintptr_t) + * aligned. (If this the structure changes such that this becomes false, an + * assertion will fail in dtrace_aggregate().) + */ +typedef struct dtrace_aggkey { + uint32_t dtak_hashval; /* hash value */ + uint32_t dtak_action:4; /* action -- 4 bits */ + uint32_t dtak_size:28; /* size -- 28 bits */ + caddr_t dtak_data; /* data pointer */ + struct dtrace_aggkey *dtak_next; /* next in hash chain */ +} dtrace_aggkey_t; + +typedef struct dtrace_aggbuffer { + uintptr_t dtagb_hashsize; /* number of buckets */ + uintptr_t dtagb_free; /* free list of keys */ + dtrace_aggkey_t **dtagb_hash; /* hash table */ +} dtrace_aggbuffer_t; + +/* + * DTrace Speculations + * + * Speculations have a per-CPU buffer and a global state. Once a speculation + * buffer has been comitted or discarded, it cannot be reused until all CPUs + * have taken the same action (commit or discard) on their respective + * speculative buffer. However, because DTrace probes may execute in arbitrary + * context, other CPUs cannot simply be cross-called at probe firing time to + * perform the necessary commit or discard. The speculation states thus + * optimize for the case that a speculative buffer is only active on one CPU at + * the time of a commit() or discard() -- for if this is the case, other CPUs + * need not take action, and the speculation is immediately available for + * reuse. If the speculation is active on multiple CPUs, it must be + * asynchronously cleaned -- potentially leading to a higher rate of dirty + * speculative drops. The speculation states are as follows: + * + * DTRACESPEC_INACTIVE <= Initial state; inactive speculation + * DTRACESPEC_ACTIVE <= Allocated, but not yet speculatively traced to + * DTRACESPEC_ACTIVEONE <= Speculatively traced to on one CPU + * DTRACESPEC_ACTIVEMANY <= Speculatively traced to on more than one CPU + * DTRACESPEC_COMMITTING <= Currently being commited on one CPU + * DTRACESPEC_COMMITTINGMANY <= Currently being commited on many CPUs + * DTRACESPEC_DISCARDING <= Currently being discarded on many CPUs + * + * The state transition diagram is as follows: + * + * +----------------------------------------------------------+ + * | | + * | +------------+ | + * | +-------------------| COMMITTING |<-----------------+ | + * | | +------------+ | | + * | | copied spec. ^ commit() on | | discard() on + * | | into principal | active CPU | | active CPU + * | | | commit() | | + * V V | | | + * +----------+ +--------+ +-----------+ + * | INACTIVE |---------------->| ACTIVE |--------------->| ACTIVEONE | + * +----------+ speculation() +--------+ speculate() +-----------+ + * ^ ^ | | | + * | | | discard() | | + * | | asynchronously | discard() on | | speculate() + * | | cleaned V inactive CPU | | on inactive + * | | +------------+ | | CPU + * | +-------------------| DISCARDING |<-----------------+ | + * | +------------+ | + * | asynchronously ^ | + * | copied spec. | discard() | + * | into principal +------------------------+ | + * | | V + * +----------------+ commit() +------------+ + * | COMMITTINGMANY |<----------------------------------| ACTIVEMANY | + * +----------------+ +------------+ + */ +typedef enum dtrace_speculation_state { + DTRACESPEC_INACTIVE = 0, + DTRACESPEC_ACTIVE, + DTRACESPEC_ACTIVEONE, + DTRACESPEC_ACTIVEMANY, + DTRACESPEC_COMMITTING, + DTRACESPEC_COMMITTINGMANY, + DTRACESPEC_DISCARDING +} dtrace_speculation_state_t; + +typedef struct dtrace_speculation { + dtrace_speculation_state_t dtsp_state; /* current speculation state */ + int dtsp_cleaning; /* non-zero if being cleaned */ + dtrace_buffer_t *dtsp_buffer; /* speculative buffer */ +} dtrace_speculation_t; + +/* + * DTrace Dynamic Variables + * + * The dynamic variable problem is obviously decomposed into two subproblems: + * allocating new dynamic storage, and freeing old dynamic storage. The + * presence of the second problem makes the first much more complicated -- or + * rather, the absence of the second renders the first trivial. This is the + * case with aggregations, for which there is effectively no deallocation of + * dynamic storage. (Or more accurately, all dynamic storage is deallocated + * when a snapshot is taken of the aggregation.) As DTrace dynamic variables + * allow for both dynamic allocation and dynamic deallocation, the + * implementation of dynamic variables is quite a bit more complicated than + * that of their aggregation kin. + * + * We observe that allocating new dynamic storage is tricky only because the + * size can vary -- the allocation problem is much easier if allocation sizes + * are uniform. We further observe that in D, the size of dynamic variables is + * actually _not_ dynamic -- dynamic variable sizes may be determined by static + * analysis of DIF text. (This is true even of putatively dynamically-sized + * objects like strings and stacks, the sizes of which are dictated by the + * "stringsize" and "stackframes" variables, respectively.) We exploit this by + * performing this analysis on all DIF before enabling any probes. For each + * dynamic load or store, we calculate the dynamically-allocated size plus the + * size of the dtrace_dynvar structure plus the storage required to key the + * data. For all DIF, we take the largest value and dub it the _chunksize_. + * We then divide dynamic memory into two parts: a hash table that is wide + * enough to have every chunk in its own bucket, and a larger region of equal + * chunksize units. Whenever we wish to dynamically allocate a variable, we + * always allocate a single chunk of memory. Depending on the uniformity of + * allocation, this will waste some amount of memory -- but it eliminates the + * non-determinism inherent in traditional heap fragmentation. + * + * Dynamic objects are allocated by storing a non-zero value to them; they are + * deallocated by storing a zero value to them. Dynamic variables are + * complicated enormously by being shared between CPUs. In particular, + * consider the following scenario: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | allocates dynamic object a[123] | | | + * | by storing the value 345 to it | | | + * | ---------> | + * | | | wishing to load from object | + * | | | a[123], performs lookup in | + * | | | dynamic variable space | + * | <--------- | + * | deallocates object a[123] by | | | + * | storing 0 to it | | | + * | | | | + * | allocates dynamic object b[567] | | performs load from a[123] | + * | by storing the value 789 to it | | | + * : : : : + * . . . . + * + * This is obviously a race in the D program, but there are nonetheless only + * two valid values for CPU B's load from a[123]: 345 or 0. Most importantly, + * CPU B may _not_ see the value 789 for a[123]. + * + * There are essentially two ways to deal with this: + * + * (1) Explicitly spin-lock variables. That is, if CPU B wishes to load + * from a[123], it needs to lock a[123] and hold the lock for the + * duration that it wishes to manipulate it. + * + * (2) Avoid reusing freed chunks until it is known that no CPU is referring + * to them. + * + * The implementation of (1) is rife with complexity, because it requires the + * user of a dynamic variable to explicitly decree when they are done using it. + * Were all variables by value, this perhaps wouldn't be debilitating -- but + * dynamic variables of non-scalar types are tracked by reference. That is, if + * a dynamic variable is, say, a string, and that variable is to be traced to, + * say, the principal buffer, the DIF emulation code returns to the main + * dtrace_probe() loop a pointer to the underlying storage, not the contents of + * the storage. Further, code calling on DIF emulation would have to be aware + * that the DIF emulation has returned a reference to a dynamic variable that + * has been potentially locked. The variable would have to be unlocked after + * the main dtrace_probe() loop is finished with the variable, and the main + * dtrace_probe() loop would have to be careful to not call any further DIF + * emulation while the variable is locked to avoid deadlock. More generally, + * if one were to implement (1), DIF emulation code dealing with dynamic + * variables could only deal with one dynamic variable at a time (lest deadlock + * result). To sum, (1) exports too much subtlety to the users of dynamic + * variables -- increasing maintenance burden and imposing serious constraints + * on future DTrace development. + * + * The implementation of (2) is also complex, but the complexity is more + * manageable. We need to be sure that when a variable is deallocated, it is + * not placed on a traditional free list, but rather on a _dirty_ list. Once a + * variable is on a dirty list, it cannot be found by CPUs performing a + * subsequent lookup of the variable -- but it may still be in use by other + * CPUs. To assure that all CPUs that may be seeing the old variable have + * cleared out of probe context, a dtrace_sync() can be issued. Once the + * dtrace_sync() has completed, it can be known that all CPUs are done + * manipulating the dynamic variable -- the dirty list can be atomically + * appended to the free list. Unfortunately, there's a slight hiccup in this + * mechanism: dtrace_sync() may not be issued from probe context. The + * dtrace_sync() must be therefore issued asynchronously from non-probe + * context. For this we rely on the DTrace cleaner, a cyclic that runs at the + * "cleanrate" frequency. To ease this implementation, we define several chunk + * lists: + * + * - Dirty. Deallocated chunks, not yet cleaned. Not available. + * + * - Rinsing. Formerly dirty chunks that are currently being asynchronously + * cleaned. Not available, but will be shortly. Dynamic variable + * allocation may not spin or block for availability, however. + * + * - Clean. Clean chunks, ready for allocation -- but not on the free list. + * + * - Free. Available for allocation. + * + * Moreover, to avoid absurd contention, _each_ of these lists is implemented + * on a per-CPU basis. This is only for performance, not correctness; chunks + * may be allocated from another CPU's free list. The algorithm for allocation + * then is this: + * + * (1) Attempt to atomically allocate from current CPU's free list. If list + * is non-empty and allocation is successful, allocation is complete. + * + * (2) If the clean list is non-empty, atomically move it to the free list, + * and reattempt (1). + * + * (3) If the dynamic variable space is in the CLEAN state, look for free + * and clean lists on other CPUs by setting the current CPU to the next + * CPU, and reattempting (1). If the next CPU is the current CPU (that + * is, if all CPUs have been checked), atomically switch the state of + * the dynamic variable space based on the following: + * + * - If no free chunks were found and no dirty chunks were found, + * atomically set the state to EMPTY. + * + * - If dirty chunks were found, atomically set the state to DIRTY. + * + * - If rinsing chunks were found, atomically set the state to RINSING. + * + * (4) Based on state of dynamic variable space state, increment appropriate + * counter to indicate dynamic drops (if in EMPTY state) vs. dynamic + * dirty drops (if in DIRTY state) vs. dynamic rinsing drops (if in + * RINSING state). Fail the allocation. + * + * The cleaning cyclic operates with the following algorithm: for all CPUs + * with a non-empty dirty list, atomically move the dirty list to the rinsing + * list. Perform a dtrace_sync(). For all CPUs with a non-empty rinsing list, + * atomically move the rinsing list to the clean list. Perform another + * dtrace_sync(). By this point, all CPUs have seen the new clean list; the + * state of the dynamic variable space can be restored to CLEAN. + * + * There exist two final races that merit explanation. The first is a simple + * allocation race: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | allocates dynamic object a[123] | | allocates dynamic object a[123] | + * | by storing the value 345 to it | | by storing the value 567 to it | + * | | | | + * : : : : + * . . . . + * + * Again, this is a race in the D program. It can be resolved by having a[123] + * hold the value 345 or a[123] hold the value 567 -- but it must be true that + * a[123] have only _one_ of these values. (That is, the racing CPUs may not + * put the same element twice on the same hash chain.) This is resolved + * simply: before the allocation is undertaken, the start of the new chunk's + * hash chain is noted. Later, after the allocation is complete, the hash + * chain is atomically switched to point to the new element. If this fails + * (because of either concurrent allocations or an allocation concurrent with a + * deletion), the newly allocated chunk is deallocated to the dirty list, and + * the whole process of looking up (and potentially allocating) the dynamic + * variable is reattempted. + * + * The final race is a simple deallocation race: + * + * CPU A CPU B + * +---------------------------------+ +---------------------------------+ + * | | | | + * | deallocates dynamic object | | deallocates dynamic object | + * | a[123] by storing the value 0 | | a[123] by storing the value 0 | + * | to it | | to it | + * | | | | + * : : : : + * . . . . + * + * Once again, this is a race in the D program, but it is one that we must + * handle without corrupting the underlying data structures. Because + * deallocations require the deletion of a chunk from the middle of a hash + * chain, we cannot use a single-word atomic operation to remove it. For this, + * we add a spin lock to the hash buckets that is _only_ used for deallocations + * (allocation races are handled as above). Further, this spin lock is _only_ + * held for the duration of the delete; before control is returned to the DIF + * emulation code, the hash bucket is unlocked. + */ +typedef struct dtrace_key { + uint64_t dttk_value; /* data value or data pointer */ + uint64_t dttk_size; /* 0 if by-val, >0 if by-ref */ +} dtrace_key_t; + +typedef struct dtrace_tuple { + uint32_t dtt_nkeys; /* number of keys in tuple */ + uint32_t dtt_pad; /* padding */ + dtrace_key_t dtt_key[1]; /* array of tuple keys */ +} dtrace_tuple_t; + +typedef struct dtrace_dynvar { + uint64_t dtdv_hashval; /* hash value -- 0 if free */ + struct dtrace_dynvar *dtdv_next; /* next on list or hash chain */ + void *dtdv_data; /* pointer to data */ + dtrace_tuple_t dtdv_tuple; /* tuple key */ +} dtrace_dynvar_t; + +typedef enum dtrace_dynvar_op { + DTRACE_DYNVAR_ALLOC, + DTRACE_DYNVAR_NOALLOC, + DTRACE_DYNVAR_DEALLOC +} dtrace_dynvar_op_t; + +typedef struct dtrace_dynhash { + dtrace_dynvar_t *dtdh_chain; /* hash chain for this bucket */ + uintptr_t dtdh_lock; /* deallocation lock */ +#ifdef _LP64 + uintptr_t dtdh_pad[6]; /* pad to avoid false sharing */ +#else + uintptr_t dtdh_pad[14]; /* pad to avoid false sharing */ +#endif +} dtrace_dynhash_t; + +typedef struct dtrace_dstate_percpu { + dtrace_dynvar_t *dtdsc_free; /* free list for this CPU */ + dtrace_dynvar_t *dtdsc_dirty; /* dirty list for this CPU */ + dtrace_dynvar_t *dtdsc_rinsing; /* rinsing list for this CPU */ + dtrace_dynvar_t *dtdsc_clean; /* clean list for this CPU */ + uint64_t dtdsc_drops; /* number of capacity drops */ + uint64_t dtdsc_dirty_drops; /* number of dirty drops */ + uint64_t dtdsc_rinsing_drops; /* number of rinsing drops */ +#ifdef _LP64 + uint64_t dtdsc_pad; /* pad to avoid false sharing */ +#else + uint64_t dtdsc_pad[2]; /* pad to avoid false sharing */ +#endif +} dtrace_dstate_percpu_t; + +typedef enum dtrace_dstate_state { + DTRACE_DSTATE_CLEAN = 0, + DTRACE_DSTATE_EMPTY, + DTRACE_DSTATE_DIRTY, + DTRACE_DSTATE_RINSING +} dtrace_dstate_state_t; + +typedef struct dtrace_dstate { + void *dtds_base; /* base of dynamic var. space */ + size_t dtds_size; /* size of dynamic var. space */ + size_t dtds_hashsize; /* number of buckets in hash */ + size_t dtds_chunksize; /* size of each chunk */ + dtrace_dynhash_t *dtds_hash; /* pointer to hash table */ + dtrace_dstate_state_t dtds_state; /* current dynamic var. state */ + dtrace_dstate_percpu_t *dtds_percpu; /* per-CPU dyn. var. state */ +} dtrace_dstate_t; + +/* + * DTrace Variable State + * + * The DTrace variable state tracks user-defined variables in its dtrace_vstate + * structure. Each DTrace consumer has exactly one dtrace_vstate structure, + * but some dtrace_vstate structures may exist without a corresponding DTrace + * consumer (see "DTrace Helpers", below). As described in <sys/dtrace.h>, + * user-defined variables can have one of three scopes: + * + * DIFV_SCOPE_GLOBAL => global scope + * DIFV_SCOPE_THREAD => thread-local scope (i.e. "self->" variables) + * DIFV_SCOPE_LOCAL => clause-local scope (i.e. "this->" variables) + * + * The variable state tracks variables by both their scope and their allocation + * type: + * + * - The dtvs_globals and dtvs_locals members each point to an array of + * dtrace_statvar structures. These structures contain both the variable + * metadata (dtrace_difv structures) and the underlying storage for all + * statically allocated variables, including statically allocated + * DIFV_SCOPE_GLOBAL variables and all DIFV_SCOPE_LOCAL variables. + * + * - The dtvs_tlocals member points to an array of dtrace_difv structures for + * DIFV_SCOPE_THREAD variables. As such, this array tracks _only_ the + * variable metadata for DIFV_SCOPE_THREAD variables; the underlying storage + * is allocated out of the dynamic variable space. + * + * - The dtvs_dynvars member is the dynamic variable state associated with the + * variable state. The dynamic variable state (described in "DTrace Dynamic + * Variables", above) tracks all DIFV_SCOPE_THREAD variables and all + * dynamically-allocated DIFV_SCOPE_GLOBAL variables. + */ +typedef struct dtrace_statvar { + uint64_t dtsv_data; /* data or pointer to it */ + size_t dtsv_size; /* size of pointed-to data */ + int dtsv_refcnt; /* reference count */ + dtrace_difv_t dtsv_var; /* variable metadata */ +} dtrace_statvar_t; + +typedef struct dtrace_vstate { + dtrace_state_t *dtvs_state; /* back pointer to state */ + dtrace_statvar_t **dtvs_globals; /* statically-allocated glbls */ + int dtvs_nglobals; /* number of globals */ + dtrace_difv_t *dtvs_tlocals; /* thread-local metadata */ + int dtvs_ntlocals; /* number of thread-locals */ + dtrace_statvar_t **dtvs_locals; /* clause-local data */ + int dtvs_nlocals; /* number of clause-locals */ + dtrace_dstate_t dtvs_dynvars; /* dynamic variable state */ +} dtrace_vstate_t; + +/* + * DTrace Machine State + * + * In the process of processing a fired probe, DTrace needs to track and/or + * cache some per-CPU state associated with that particular firing. This is + * state that is always discarded after the probe firing has completed, and + * much of it is not specific to any DTrace consumer, remaining valid across + * all ECBs. This state is tracked in the dtrace_mstate structure. + */ +#define DTRACE_MSTATE_ARGS 0x00000001 +#define DTRACE_MSTATE_PROBE 0x00000002 +#define DTRACE_MSTATE_EPID 0x00000004 +#define DTRACE_MSTATE_TIMESTAMP 0x00000008 +#define DTRACE_MSTATE_STACKDEPTH 0x00000010 +#define DTRACE_MSTATE_CALLER 0x00000020 +#define DTRACE_MSTATE_IPL 0x00000040 +#define DTRACE_MSTATE_FLTOFFS 0x00000080 +#define DTRACE_MSTATE_WALLTIMESTAMP 0x00000100 +#define DTRACE_MSTATE_USTACKDEPTH 0x00000200 +#define DTRACE_MSTATE_UCALLER 0x00000400 + +typedef struct dtrace_mstate { + uintptr_t dtms_scratch_base; /* base of scratch space */ + uintptr_t dtms_scratch_ptr; /* current scratch pointer */ + size_t dtms_scratch_size; /* scratch size */ + uint32_t dtms_present; /* variables that are present */ + uint64_t dtms_arg[5]; /* cached arguments */ + dtrace_epid_t dtms_epid; /* current EPID */ + uint64_t dtms_timestamp; /* cached timestamp */ + hrtime_t dtms_walltimestamp; /* cached wall timestamp */ + int dtms_stackdepth; /* cached stackdepth */ + int dtms_ustackdepth; /* cached ustackdepth */ + struct dtrace_probe *dtms_probe; /* current probe */ + uintptr_t dtms_caller; /* cached caller */ + uint64_t dtms_ucaller; /* cached user-level caller */ + int dtms_ipl; /* cached interrupt pri lev */ + int dtms_fltoffs; /* faulting DIFO offset */ + uintptr_t dtms_strtok; /* saved strtok() pointer */ + uint32_t dtms_access; /* memory access rights */ + dtrace_difo_t *dtms_difo; /* current dif object */ +} dtrace_mstate_t; + +#define DTRACE_COND_OWNER 0x1 +#define DTRACE_COND_USERMODE 0x2 +#define DTRACE_COND_ZONEOWNER 0x4 + +#define DTRACE_PROBEKEY_MAXDEPTH 8 /* max glob recursion depth */ + +/* + * Access flag used by dtrace_mstate.dtms_access. + */ +#define DTRACE_ACCESS_KERNEL 0x1 /* the priv to read kmem */ + + +/* + * DTrace Activity + * + * Each DTrace consumer is in one of several states, which (for purposes of + * avoiding yet-another overloading of the noun "state") we call the current + * _activity_. The activity transitions on dtrace_go() (from DTRACIOCGO), on + * dtrace_stop() (from DTRACIOCSTOP) and on the exit() action. Activities may + * only transition in one direction; the activity transition diagram is a + * directed acyclic graph. The activity transition diagram is as follows: + * + * + * +----------+ +--------+ +--------+ + * | INACTIVE |------------------>| WARMUP |------------------>| ACTIVE | + * +----------+ dtrace_go(), +--------+ dtrace_go(), +--------+ + * before BEGIN | after BEGIN | | | + * | | | | + * exit() action | | | | + * from BEGIN ECB | | | | + * | | | | + * v | | | + * +----------+ exit() action | | | + * +-----------------------------| DRAINING |<-------------------+ | | + * | +----------+ | | + * | | | | + * | dtrace_stop(), | | | + * | before END | | | + * | | | | + * | v | | + * | +---------+ +----------+ | | + * | | STOPPED |<----------------| COOLDOWN |<----------------------+ | + * | +---------+ dtrace_stop(), +----------+ dtrace_stop(), | + * | after END before END | + * | | + * | +--------+ | + * +----------------------------->| KILLED |<--------------------------+ + * deadman timeout or +--------+ deadman timeout or + * killed consumer killed consumer + * + * Note that once a DTrace consumer has stopped tracing, there is no way to + * restart it; if a DTrace consumer wishes to restart tracing, it must reopen + * the DTrace pseudodevice. + */ +typedef enum dtrace_activity { + DTRACE_ACTIVITY_INACTIVE = 0, /* not yet running */ + DTRACE_ACTIVITY_WARMUP, /* while starting */ + DTRACE_ACTIVITY_ACTIVE, /* running */ + DTRACE_ACTIVITY_DRAINING, /* before stopping */ + DTRACE_ACTIVITY_COOLDOWN, /* while stopping */ + DTRACE_ACTIVITY_STOPPED, /* after stopping */ + DTRACE_ACTIVITY_KILLED /* killed */ +} dtrace_activity_t; + +/* + * DTrace Helper Implementation + * + * A description of the helper architecture may be found in <sys/dtrace.h>. + * Each process contains a pointer to its helpers in its p_dtrace_helpers + * member. This is a pointer to a dtrace_helpers structure, which contains an + * array of pointers to dtrace_helper structures, helper variable state (shared + * among a process's helpers) and a generation count. (The generation count is + * used to provide an identifier when a helper is added so that it may be + * subsequently removed.) The dtrace_helper structure is self-explanatory, + * containing pointers to the objects needed to execute the helper. Note that + * helpers are _duplicated_ across fork(2), and destroyed on exec(2). No more + * than dtrace_helpers_max are allowed per-process. + */ +#define DTRACE_HELPER_ACTION_USTACK 0 +#define DTRACE_NHELPER_ACTIONS 1 + +typedef struct dtrace_helper_action { + int dtha_generation; /* helper action generation */ + int dtha_nactions; /* number of actions */ + dtrace_difo_t *dtha_predicate; /* helper action predicate */ + dtrace_difo_t **dtha_actions; /* array of actions */ + struct dtrace_helper_action *dtha_next; /* next helper action */ +} dtrace_helper_action_t; + +typedef struct dtrace_helper_provider { + int dthp_generation; /* helper provider generation */ + uint32_t dthp_ref; /* reference count */ + dof_helper_t dthp_prov; /* DOF w/ provider and probes */ +} dtrace_helper_provider_t; + +typedef struct dtrace_helpers { + dtrace_helper_action_t **dthps_actions; /* array of helper actions */ + dtrace_vstate_t dthps_vstate; /* helper action var. state */ + dtrace_helper_provider_t **dthps_provs; /* array of providers */ + uint_t dthps_nprovs; /* count of providers */ + uint_t dthps_maxprovs; /* provider array size */ + int dthps_generation; /* current generation */ + pid_t dthps_pid; /* pid of associated proc */ + int dthps_deferred; /* helper in deferred list */ + struct dtrace_helpers *dthps_next; /* next pointer */ + struct dtrace_helpers *dthps_prev; /* prev pointer */ +} dtrace_helpers_t; + +/* + * DTrace Helper Action Tracing + * + * Debugging helper actions can be arduous. To ease the development and + * debugging of helpers, DTrace contains a tracing-framework-within-a-tracing- + * framework: helper tracing. If dtrace_helptrace_enabled is non-zero (which + * it is by default on DEBUG kernels), all helper activity will be traced to a + * global, in-kernel ring buffer. Each entry includes a pointer to the specific + * helper, the location within the helper, and a trace of all local variables. + * The ring buffer may be displayed in a human-readable format with the + * ::dtrace_helptrace mdb(1) dcmd. + */ +#define DTRACE_HELPTRACE_NEXT (-1) +#define DTRACE_HELPTRACE_DONE (-2) +#define DTRACE_HELPTRACE_ERR (-3) + +typedef struct dtrace_helptrace { + dtrace_helper_action_t *dtht_helper; /* helper action */ + int dtht_where; /* where in helper action */ + int dtht_nlocals; /* number of locals */ + int dtht_fault; /* type of fault (if any) */ + int dtht_fltoffs; /* DIF offset */ + uint64_t dtht_illval; /* faulting value */ + uint64_t dtht_locals[1]; /* local variables */ +} dtrace_helptrace_t; + +/* + * DTrace Credentials + * + * In probe context, we have limited flexibility to examine the credentials + * of the DTrace consumer that created a particular enabling. We use + * the Least Privilege interfaces to cache the consumer's cred pointer and + * some facts about that credential in a dtrace_cred_t structure. These + * can limit the consumer's breadth of visibility and what actions the + * consumer may take. + */ +#define DTRACE_CRV_ALLPROC 0x01 +#define DTRACE_CRV_KERNEL 0x02 +#define DTRACE_CRV_ALLZONE 0x04 + +#define DTRACE_CRV_ALL (DTRACE_CRV_ALLPROC | DTRACE_CRV_KERNEL | \ + DTRACE_CRV_ALLZONE) + +#define DTRACE_CRA_PROC 0x0001 +#define DTRACE_CRA_PROC_CONTROL 0x0002 +#define DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER 0x0004 +#define DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE 0x0008 +#define DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG 0x0010 +#define DTRACE_CRA_KERNEL 0x0020 +#define DTRACE_CRA_KERNEL_DESTRUCTIVE 0x0040 + +#define DTRACE_CRA_ALL (DTRACE_CRA_PROC | \ + DTRACE_CRA_PROC_CONTROL | \ + DTRACE_CRA_PROC_DESTRUCTIVE_ALLUSER | \ + DTRACE_CRA_PROC_DESTRUCTIVE_ALLZONE | \ + DTRACE_CRA_PROC_DESTRUCTIVE_CREDCHG | \ + DTRACE_CRA_KERNEL | \ + DTRACE_CRA_KERNEL_DESTRUCTIVE) + +typedef struct dtrace_cred { + cred_t *dcr_cred; + uint8_t dcr_destructive; + uint8_t dcr_visible; + uint16_t dcr_action; +} dtrace_cred_t; + +/* + * DTrace Consumer State + * + * Each DTrace consumer has an associated dtrace_state structure that contains + * its in-kernel DTrace state -- including options, credentials, statistics and + * pointers to ECBs, buffers, speculations and formats. A dtrace_state + * structure is also allocated for anonymous enablings. When anonymous state + * is grabbed, the grabbing consumers dts_anon pointer is set to the grabbed + * dtrace_state structure. + */ +struct dtrace_state { +#ifndef VBOX + dev_t dts_dev; /* device */ +#endif + int dts_necbs; /* total number of ECBs */ + dtrace_ecb_t **dts_ecbs; /* array of ECBs */ + dtrace_epid_t dts_epid; /* next EPID to allocate */ + size_t dts_needed; /* greatest needed space */ + struct dtrace_state *dts_anon; /* anon. state, if grabbed */ + dtrace_activity_t dts_activity; /* current activity */ + dtrace_vstate_t dts_vstate; /* variable state */ + dtrace_buffer_t *dts_buffer; /* principal buffer */ + dtrace_buffer_t *dts_aggbuffer; /* aggregation buffer */ + dtrace_speculation_t *dts_speculations; /* speculation array */ + int dts_nspeculations; /* number of speculations */ + int dts_naggregations; /* number of aggregations */ + dtrace_aggregation_t **dts_aggregations; /* aggregation array */ + vmem_t *dts_aggid_arena; /* arena for aggregation IDs */ + uint64_t dts_errors; /* total number of errors */ + uint32_t dts_speculations_busy; /* number of spec. busy */ + uint32_t dts_speculations_unavail; /* number of spec unavail */ + uint32_t dts_stkstroverflows; /* stack string tab overflows */ + uint32_t dts_dblerrors; /* errors in ERROR probes */ + uint32_t dts_reserve; /* space reserved for END */ + hrtime_t dts_laststatus; /* time of last status */ + cyclic_id_t dts_cleaner; /* cleaning cyclic */ + cyclic_id_t dts_deadman; /* deadman cyclic */ + hrtime_t dts_alive; /* time last alive */ + char dts_speculates; /* boolean: has speculations */ + char dts_destructive; /* boolean: has dest. actions */ + int dts_nformats; /* number of formats */ + char **dts_formats; /* format string array */ + dtrace_optval_t dts_options[DTRACEOPT_MAX]; /* options */ + dtrace_cred_t dts_cred; /* credentials */ + size_t dts_nretained; /* number of retained enabs */ +}; + +struct dtrace_provider { + dtrace_pattr_t dtpv_attr; /* provider attributes */ + dtrace_ppriv_t dtpv_priv; /* provider privileges */ + dtrace_pops_t dtpv_pops; /* provider operations */ + char *dtpv_name; /* provider name */ + void *dtpv_arg; /* provider argument */ + uint_t dtpv_defunct; /* boolean: defunct provider */ + struct dtrace_provider *dtpv_next; /* next provider */ +}; + +struct dtrace_meta { + dtrace_mops_t dtm_mops; /* meta provider operations */ + char *dtm_name; /* meta provider name */ + void *dtm_arg; /* meta provider user arg */ + uint64_t dtm_count; /* no. of associated provs. */ +}; + +/* + * DTrace Enablings + * + * A dtrace_enabling structure is used to track a collection of ECB + * descriptions -- before they have been turned into actual ECBs. This is + * created as a result of DOF processing, and is generally used to generate + * ECBs immediately thereafter. However, enablings are also generally + * retained should the probes they describe be created at a later time; as + * each new module or provider registers with the framework, the retained + * enablings are reevaluated, with any new match resulting in new ECBs. To + * prevent probes from being matched more than once, the enabling tracks the + * last probe generation matched, and only matches probes from subsequent + * generations. + */ +typedef struct dtrace_enabling { + dtrace_ecbdesc_t **dten_desc; /* all ECB descriptions */ + int dten_ndesc; /* number of ECB descriptions */ + int dten_maxdesc; /* size of ECB array */ + dtrace_vstate_t *dten_vstate; /* associated variable state */ + dtrace_genid_t dten_probegen; /* matched probe generation */ + dtrace_ecbdesc_t *dten_current; /* current ECB description */ + int dten_error; /* current error value */ + int dten_primed; /* boolean: set if primed */ + struct dtrace_enabling *dten_prev; /* previous enabling */ + struct dtrace_enabling *dten_next; /* next enabling */ +} dtrace_enabling_t; + +/* + * DTrace Anonymous Enablings + * + * Anonymous enablings are DTrace enablings that are not associated with a + * controlling process, but rather derive their enabling from DOF stored as + * properties in the dtrace.conf file. If there is an anonymous enabling, a + * DTrace consumer state and enabling are created on attach. The state may be + * subsequently grabbed by the first consumer specifying the "grabanon" + * option. As long as an anonymous DTrace enabling exists, dtrace(7D) will + * refuse to unload. + */ +typedef struct dtrace_anon { + dtrace_state_t *dta_state; /* DTrace consumer state */ + dtrace_enabling_t *dta_enabling; /* pointer to enabling */ + processorid_t dta_beganon; /* which CPU BEGIN ran on */ +} dtrace_anon_t; + +/* + * DTrace Error Debugging + */ +#ifdef DEBUG +#define DTRACE_ERRDEBUG +#endif + +#ifdef DTRACE_ERRDEBUG + +typedef struct dtrace_errhash { + const char *dter_msg; /* error message */ + int dter_count; /* number of times seen */ +} dtrace_errhash_t; + +#define DTRACE_ERRHASHSZ 256 /* must be > number of err msgs */ + +#endif /* DTRACE_ERRDEBUG */ + +/* + * DTrace Toxic Ranges + * + * DTrace supports safe loads from probe context; if the address turns out to + * be invalid, a bit will be set by the kernel indicating that DTrace + * encountered a memory error, and DTrace will propagate the error to the user + * accordingly. However, there may exist some regions of memory in which an + * arbitrary load can change system state, and from which it is impossible to + * recover from such a load after it has been attempted. Examples of this may + * include memory in which programmable I/O registers are mapped (for which a + * read may have some implications for the device) or (in the specific case of + * UltraSPARC-I and -II) the virtual address hole. The platform is required + * to make DTrace aware of these toxic ranges; DTrace will then check that + * target addresses are not in a toxic range before attempting to issue a + * safe load. + */ +typedef struct dtrace_toxrange { + uintptr_t dtt_base; /* base of toxic range */ + uintptr_t dtt_limit; /* limit of toxic range */ +} dtrace_toxrange_t; + +extern uint64_t dtrace_getarg(int, int); +extern greg_t dtrace_getfp(void); +extern int dtrace_getipl(void); +extern uintptr_t dtrace_caller(int); +extern uint32_t dtrace_cas32(uint32_t *, uint32_t, uint32_t); +extern void *dtrace_casptr(void *, void *, void *); +extern void dtrace_copyin(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyinstr(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyout(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +extern void dtrace_copyoutstr(uintptr_t, uintptr_t, size_t, + volatile uint16_t *); +extern void dtrace_getpcstack(pc_t *, int, int, uint32_t *); +extern ulong_t dtrace_getreg(struct regs *, uint_t); +extern int dtrace_getstackdepth(int); +extern void dtrace_getupcstack(uint64_t *, int); +extern void dtrace_getufpstack(uint64_t *, uint64_t *, int); +extern int dtrace_getustackdepth(void); +extern uintptr_t dtrace_fulword(void *); +extern uint8_t dtrace_fuword8(void *); +extern uint16_t dtrace_fuword16(void *); +extern uint32_t dtrace_fuword32(void *); +extern uint64_t dtrace_fuword64(void *); +extern void dtrace_probe_error(dtrace_state_t *, dtrace_epid_t, int, int, + int, uintptr_t); +extern int dtrace_assfail(const char *, const char *, int); +extern int dtrace_attached(void); +extern hrtime_t dtrace_gethrestime(void); + +#ifdef __sparc +extern void dtrace_flush_windows(void); +extern void dtrace_flush_user_windows(void); +extern uint_t dtrace_getotherwin(void); +extern uint_t dtrace_getfprs(void); +#else +extern void dtrace_copy(uintptr_t, uintptr_t, size_t); +extern void dtrace_copystr(uintptr_t, uintptr_t, size_t, volatile uint16_t *); +#endif + +#ifndef VBOX /* got our own assert thing */ +/* + * DTrace Assertions + * + * DTrace calls ASSERT from probe context. To assure that a failed ASSERT + * does not induce a markedly more catastrophic failure (e.g., one from which + * a dump cannot be gleaned), DTrace must define its own ASSERT to be one that + * may safely be called from probe context. This header file must thus be + * included by any DTrace component that calls ASSERT from probe context, and + * _only_ by those components. (The only exception to this is kernel + * debugging infrastructure at user-level that doesn't depend on calling + * ASSERT.) + */ +#undef ASSERT +#ifdef DEBUG +#define ASSERT(EX) ((void)((EX) || \ + dtrace_assfail(#EX, __FILE__, __LINE__))) +#else +#define ASSERT(X) ((void)0) +#endif +#endif /* !VBOX */ + +#ifdef __cplusplus +} +#endif + +#endif /* _SYS_DTRACE_IMPL_H */ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/dtrace_subr.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/dtrace_subr.c new file mode 100644 index 00000000..baf79197 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/i86pc/os/dtrace_subr.c @@ -0,0 +1,400 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#include <sys/dtrace.h> +#include <sys/fasttrap.h> +#include <sys/x_call.h> +#include <sys/cmn_err.h> +#include <sys/trap.h> +#include <sys/psw.h> +#include <sys/privregs.h> +#include <sys/machsystm.h> +#include <vm/seg_kmem.h> + +typedef struct dtrace_invop_hdlr { + int (*dtih_func)(uintptr_t, uintptr_t *, uintptr_t); + struct dtrace_invop_hdlr *dtih_next; +} dtrace_invop_hdlr_t; + +dtrace_invop_hdlr_t *dtrace_invop_hdlr; + +int +dtrace_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax) +{ + dtrace_invop_hdlr_t *hdlr; + int rval; + + for (hdlr = dtrace_invop_hdlr; hdlr != NULL; hdlr = hdlr->dtih_next) { + if ((rval = hdlr->dtih_func(addr, stack, eax)) != 0) + return (rval); + } + + return (0); +} + +void +dtrace_invop_add(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr; + + hdlr = kmem_alloc(sizeof (dtrace_invop_hdlr_t), KM_SLEEP); + hdlr->dtih_func = func; + hdlr->dtih_next = dtrace_invop_hdlr; + dtrace_invop_hdlr = hdlr; +} + +void +dtrace_invop_remove(int (*func)(uintptr_t, uintptr_t *, uintptr_t)) +{ + dtrace_invop_hdlr_t *hdlr = dtrace_invop_hdlr, *prev = NULL; + + for (;;) { + if (hdlr == NULL) + panic("attempt to remove non-existent invop handler"); + + if (hdlr->dtih_func == func) + break; + + prev = hdlr; + hdlr = hdlr->dtih_next; + } + + if (prev == NULL) { + ASSERT(dtrace_invop_hdlr == hdlr); + dtrace_invop_hdlr = hdlr->dtih_next; + } else { + ASSERT(dtrace_invop_hdlr != hdlr); + prev->dtih_next = hdlr->dtih_next; + } + + kmem_free(hdlr, sizeof (dtrace_invop_hdlr_t)); +} + +int +dtrace_getipl(void) +{ + return (CPU->cpu_pri); +} + +/*ARGSUSED*/ +void +dtrace_toxic_ranges(void (*func)(uintptr_t base, uintptr_t limit)) +{ +#ifdef __amd64 + extern uintptr_t toxic_addr; + extern size_t toxic_size; + + (*func)(0, _userlimit); + + if (hole_end > hole_start) + (*func)(hole_start, hole_end); + (*func)(toxic_addr, toxic_addr + toxic_size); +#else + extern void *device_arena_contains(void *, size_t, size_t *); + caddr_t vaddr; + size_t len; + + for (vaddr = (caddr_t)kernelbase; vaddr < (caddr_t)KERNEL_TEXT; + vaddr += len) { + len = (caddr_t)KERNEL_TEXT - vaddr; + vaddr = device_arena_contains(vaddr, len, &len); + if (vaddr == NULL) + break; + (*func)((uintptr_t)vaddr, (uintptr_t)vaddr + len); + } +#endif + (*func)(0, _userlimit); +} + +static int +dtrace_xcall_func(dtrace_xcall_t func, void *arg) +{ + (*func)(arg); + + return (0); +} + +/*ARGSUSED*/ +void +dtrace_xcall(processorid_t cpu, dtrace_xcall_t func, void *arg) +{ + cpuset_t set; + + CPUSET_ZERO(set); + + if (cpu == DTRACE_CPUALL) { + CPUSET_ALL(set); + } else { + CPUSET_ADD(set, cpu); + } + + kpreempt_disable(); + xc_sync((xc_arg_t)func, (xc_arg_t)arg, 0, CPUSET2BV(set), + (xc_func_t)dtrace_xcall_func); + kpreempt_enable(); +} + +void +dtrace_sync_func(void) +{} + +void +dtrace_sync(void) +{ + dtrace_xcall(DTRACE_CPUALL, (dtrace_xcall_t)dtrace_sync_func, NULL); +} + +int (*dtrace_pid_probe_ptr)(struct regs *); +int (*dtrace_return_probe_ptr)(struct regs *); + +void +dtrace_user_probe(struct regs *rp, caddr_t addr, processorid_t cpuid) +{ + krwlock_t *rwp; + proc_t *p = curproc; + extern void trap(struct regs *, caddr_t, processorid_t); + + if (USERMODE(rp->r_cs) || (rp->r_ps & PS_VM)) { + if (curthread->t_cred != p->p_cred) { + cred_t *oldcred = curthread->t_cred; + /* + * DTrace accesses t_cred in probe context. t_cred + * must always be either NULL, or point to a valid, + * allocated cred structure. + */ + curthread->t_cred = crgetcred(); + crfree(oldcred); + } + } + + if (rp->r_trapno == T_DTRACE_RET) { + uint8_t step = curthread->t_dtrace_step; + uint8_t ret = curthread->t_dtrace_ret; + uintptr_t npc = curthread->t_dtrace_npc; + + if (curthread->t_dtrace_ast) { + aston(curthread); + curthread->t_sig_check = 1; + } + + /* + * Clear all user tracing flags. + */ + curthread->t_dtrace_ft = 0; + + /* + * If we weren't expecting to take a return probe trap, kill + * the process as though it had just executed an unassigned + * trap instruction. + */ + if (step == 0) { + tsignal(curthread, SIGILL); + return; + } + + /* + * If we hit this trap unrelated to a return probe, we're + * just here to reset the AST flag since we deferred a signal + * until after we logically single-stepped the instruction we + * copied out. + */ + if (ret == 0) { + rp->r_pc = npc; + return; + } + + /* + * We need to wait until after we've called the + * dtrace_return_probe_ptr function pointer to set %pc. + */ + rwp = &CPU->cpu_ft_lock; + rw_enter(rwp, RW_READER); + if (dtrace_return_probe_ptr != NULL) + (void) (*dtrace_return_probe_ptr)(rp); + rw_exit(rwp); + rp->r_pc = npc; + + } else if (rp->r_trapno == T_BPTFLT) { + uint8_t instr, instr2; + caddr_t linearpc; + rwp = &CPU->cpu_ft_lock; + + /* + * The DTrace fasttrap provider uses the breakpoint trap + * (int 3). We let DTrace take the first crack at handling + * this trap; if it's not a probe that DTrace knowns about, + * we call into the trap() routine to handle it like a + * breakpoint placed by a conventional debugger. + */ + rw_enter(rwp, RW_READER); + if (dtrace_pid_probe_ptr != NULL && + (*dtrace_pid_probe_ptr)(rp) == 0) { + rw_exit(rwp); + return; + } + rw_exit(rwp); + + if (dtrace_linear_pc(rp, p, &linearpc) != 0) { + trap(rp, addr, cpuid); + return; + } + + /* + * If the instruction that caused the breakpoint trap doesn't + * look like an int 3 anymore, it may be that this tracepoint + * was removed just after the user thread executed it. In + * that case, return to user land to retry the instuction. + * Note that we assume the length of the instruction to retry + * is 1 byte because that's the length of FASTTRAP_INSTR. + * We check for r_pc > 0 and > 2 so that we don't have to + * deal with segment wraparound. + */ + if (rp->r_pc > 0 && fuword8(linearpc - 1, &instr) == 0 && + instr != FASTTRAP_INSTR && + (instr != 3 || (rp->r_pc >= 2 && + (fuword8(linearpc - 2, &instr2) != 0 || instr2 != 0xCD)))) { + rp->r_pc--; + return; + } + + trap(rp, addr, cpuid); + + } else { + trap(rp, addr, cpuid); + } +} + +void +dtrace_safe_synchronous_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. If the instruction + * we copied out caused a synchonous trap, reset the pc back to its + * original value and turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + } else if (rp->r_pc == t->t_dtrace_scrpc || + rp->r_pc == t->t_dtrace_astpc) { + rp->r_pc = t->t_dtrace_pc; + t->t_dtrace_ft = 0; + } +} + +int +dtrace_safe_defer_signal(void) +{ + kthread_t *t = curthread; + struct regs *rp = lwptoregs(ttolwp(t)); + size_t isz = t->t_dtrace_npc - t->t_dtrace_pc; + + ASSERT(t->t_dtrace_on); + + /* + * If we're not in the range of scratch addresses, we're not actually + * tracing user instructions so turn off the flags. + */ + if (rp->r_pc < t->t_dtrace_scrpc || + rp->r_pc > t->t_dtrace_astpc + isz) { + t->t_dtrace_ft = 0; + return (0); + } + + /* + * If we've executed the original instruction, but haven't performed + * the jmp back to t->t_dtrace_npc or the clean up of any registers + * used to emulate %rip-relative instructions in 64-bit mode, do that + * here and take the signal right away. We detect this condition by + * seeing if the program counter is the range [scrpc + isz, astpc). + */ + if (t->t_dtrace_astpc - rp->r_pc < + t->t_dtrace_astpc - t->t_dtrace_scrpc - isz) { +#ifdef __amd64 + /* + * If there is a scratch register and we're on the + * instruction immediately after the modified instruction, + * restore the value of that scratch register. + */ + if (t->t_dtrace_reg != 0 && + rp->r_pc == t->t_dtrace_scrpc + isz) { + switch (t->t_dtrace_reg) { + case REG_RAX: + rp->r_rax = t->t_dtrace_regv; + break; + case REG_RCX: + rp->r_rcx = t->t_dtrace_regv; + break; + case REG_R8: + rp->r_r8 = t->t_dtrace_regv; + break; + case REG_R9: + rp->r_r9 = t->t_dtrace_regv; + break; + } + } +#endif + rp->r_pc = t->t_dtrace_npc; + t->t_dtrace_ft = 0; + return (0); + } + + /* + * Otherwise, make sure we'll return to the kernel after executing + * the copied out instruction and defer the signal. + */ + if (!t->t_dtrace_step) { + ASSERT(rp->r_pc < t->t_dtrace_astpc); + rp->r_pc += t->t_dtrace_astpc - t->t_dtrace_scrpc; + t->t_dtrace_step = 1; + } + + t->t_dtrace_ast = 1; + + return (1); +} + +/* + * Additional artificial frames for the machine type. For i86pc, we're already + * accounted for, so return 0. On the hypervisor, we have an additional frame + * (xen_callback_handler). + */ +int +dtrace_mach_aframes(void) +{ +#ifdef __xpv + return (1); +#else + return (0); +#endif +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile new file mode 100644 index 00000000..5cd153e3 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile @@ -0,0 +1,76 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License (the "License"). +# You may not use this file except in compliance with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2008 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +UTSBASE = ../.. + +MODULE = dtrace +OBJECTS = $(DTRACE_OBJS:%=$(OBJS_DIR)/%) +LINTS = $(DTRACE_OBJS:%.o=$(LINTS_DIR)/%.ln) +ROOTMODULE = $(ROOT_DRV_DIR)/$(MODULE) +CONF_SRCDIR = $(UTSBASE)/common/dtrace + +include $(UTSBASE)/intel/Makefile.intel + +# +# For now, disable these lint checks; maintainers should endeavor +# to investigate and remove these for maximum lint coverage. +# Please do not carry these forward to new Makefiles. +# +LINTTAGS += -erroff=E_SUSPICIOUS_COMPARISON +LINTTAGS += -erroff=E_BAD_PTR_CAST_ALIGN +LINTTAGS += -erroff=E_SUPPRESSION_DIRECTIVE_UNUSED +LINTTAGS += -erroff=E_STATIC_UNUSED +LINTTAGS += -erroff=E_PTRDIFF_OVERFLOW +LINTTAGS += -erroff=E_ASSIGN_NARROW_CONV + +ALL_TARGET = $(BINARY) $(SRC_CONFILE) +LINT_TARGET = $(MODULE).lint +INSTALL_TARGET = $(BINARY) $(ROOTMODULE) $(ROOT_CONFFILE) +AS_INC_PATH += -I$(DSF_DIR)/$(OBJS_DIR) + +ASSYM_H = $(DSF_DIR)/$(OBJS_DIR)/assym.h + +.KEEP_STATE: + +def: $(DEF_DEPS) + +all: $(ALL_DEPS) + +clean: $(CLEAN_DEPS) + +clobber: $(CLOBBER_DEPS) + +lint: $(LINT_DEPS) + +modlintlib: $(MODLINTLIB_DEPS) + +clean.lint: $(CLEAN_LINT_DEPS) + +install: $(INSTALL_DEPS) + +$(BINARY): $(ASSYM_H) + +include $(UTSBASE)/intel/Makefile.targ diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile.kup b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile.kup new file mode 100644 index 00000000..e69de29b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/Makefile.kup diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_asm.s b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_asm.s new file mode 100644 index 00000000..47b981d1 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_asm.s @@ -0,0 +1,457 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/asm_linkage.h> +#include <sys/regset.h> + +#if defined(lint) +#include <sys/dtrace_impl.h> +#else +#include "assym.h" +#endif + +#if defined(lint) || defined(__lint) + +greg_t +dtrace_getfp(void) +{ return (0); } + +#else /* lint */ + +#if defined(__amd64) + + ENTRY_NP(dtrace_getfp) + movq %rbp, %rax + ret + SET_SIZE(dtrace_getfp) + +#elif defined(__i386) + + ENTRY_NP(dtrace_getfp) + movl %ebp, %eax + ret + SET_SIZE(dtrace_getfp) + +#endif /* __i386 */ +#endif /* lint */ + + +#if defined(lint) || defined(__lint) + +uint32_t +dtrace_cas32(uint32_t *target, uint32_t cmp, uint32_t new) +{ + uint32_t old; + + if ((old = *target) == cmp) + *target = new; + return (old); +} + +void * +dtrace_casptr(void *target, void *cmp, void *new) +{ + void *old; + + if ((old = *(void **)target) == cmp) + *(void **)target = new; + return (old); +} + +#else /* lint */ + +#if defined(__amd64) + + ENTRY(dtrace_cas32) + movl %esi, %eax + lock + cmpxchgl %edx, (%rdi) + ret + SET_SIZE(dtrace_cas32) + + ENTRY(dtrace_casptr) + movq %rsi, %rax + lock + cmpxchgq %rdx, (%rdi) + ret + SET_SIZE(dtrace_casptr) + +#elif defined(__i386) + + ENTRY(dtrace_cas32) + ALTENTRY(dtrace_casptr) + movl 4(%esp), %edx + movl 8(%esp), %eax + movl 12(%esp), %ecx + lock + cmpxchgl %ecx, (%edx) + ret + SET_SIZE(dtrace_casptr) + SET_SIZE(dtrace_cas32) + +#endif /* __i386 */ +#endif /* lint */ + +#if defined(lint) + +/*ARGSUSED*/ +uintptr_t +dtrace_caller(int aframes) +{ + return (0); +} + +#else /* lint */ + +#if defined(__amd64) + ENTRY(dtrace_caller) + movq $-1, %rax + ret + SET_SIZE(dtrace_caller) + +#elif defined(__i386) + + ENTRY(dtrace_caller) + movl $-1, %eax + ret + SET_SIZE(dtrace_caller) + +#endif /* __i386 */ +#endif /* lint */ + +#if defined(lint) + +/*ARGSUSED*/ +void +dtrace_copy(uintptr_t src, uintptr_t dest, size_t size) +{} + +#else + +#if defined(__amd64) + + ENTRY(dtrace_copy) + pushq %rbp + movq %rsp, %rbp + + xchgq %rdi, %rsi /* make %rsi source, %rdi dest */ + movq %rdx, %rcx /* load count */ + repz /* repeat for count ... */ + smovb /* move from %ds:rsi to %ed:rdi */ + leave + ret + SET_SIZE(dtrace_copy) + +#elif defined(__i386) + + ENTRY(dtrace_copy) + pushl %ebp + movl %esp, %ebp + pushl %esi + pushl %edi + + movl 8(%ebp), %esi / Load source address + movl 12(%ebp), %edi / Load destination address + movl 16(%ebp), %ecx / Load count + repz / Repeat for count... + smovb / move from %ds:si to %es:di + + popl %edi + popl %esi + movl %ebp, %esp + popl %ebp + ret + SET_SIZE(dtrace_copy) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +void +dtrace_copystr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{} + +#else + +#if defined(__amd64) + + ENTRY(dtrace_copystr) + pushq %rbp + movq %rsp, %rbp + +0: + movb (%rdi), %al /* load from source */ + movb %al, (%rsi) /* store to destination */ + addq $1, %rdi /* increment source pointer */ + addq $1, %rsi /* increment destination pointer */ + subq $1, %rdx /* decrement remaining count */ + cmpb $0, %al + je 2f + testq $0xfff, %rdx /* test if count is 4k-aligned */ + jnz 1f /* if not, continue with copying */ + testq $CPU_DTRACE_BADADDR, (%rcx) /* load and test dtrace flags */ + jnz 2f +1: + cmpq $0, %rdx + jne 0b +2: + leave + ret + + SET_SIZE(dtrace_copystr) + +#elif defined(__i386) + + ENTRY(dtrace_copystr) + + pushl %ebp / Setup stack frame + movl %esp, %ebp + pushl %ebx / Save registers + + movl 8(%ebp), %ebx / Load source address + movl 12(%ebp), %edx / Load destination address + movl 16(%ebp), %ecx / Load count + +0: + movb (%ebx), %al / Load from source + movb %al, (%edx) / Store to destination + incl %ebx / Increment source pointer + incl %edx / Increment destination pointer + decl %ecx / Decrement remaining count + cmpb $0, %al + je 2f + testl $0xfff, %ecx / Check if count is 4k-aligned + jnz 1f + movl 20(%ebp), %eax / load flags pointer + testl $CPU_DTRACE_BADADDR, (%eax) / load and test dtrace flags + jnz 2f +1: + cmpl $0, %ecx + jne 0b + +2: + popl %ebx + movl %ebp, %esp + popl %ebp + ret + + SET_SIZE(dtrace_copystr) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +uintptr_t +dtrace_fulword(void *addr) +{ return (0); } + +#else +#if defined(__amd64) + + ENTRY(dtrace_fulword) + movq (%rdi), %rax + ret + SET_SIZE(dtrace_fulword) + +#elif defined(__i386) + + ENTRY(dtrace_fulword) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + SET_SIZE(dtrace_fulword) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +uint8_t +dtrace_fuword8_nocheck(void *addr) +{ return (0); } + +#else +#if defined(__amd64) + + ENTRY(dtrace_fuword8_nocheck) + xorq %rax, %rax + movb (%rdi), %al + ret + SET_SIZE(dtrace_fuword8_nocheck) + +#elif defined(__i386) + + ENTRY(dtrace_fuword8_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzbl (%ecx), %eax + ret + SET_SIZE(dtrace_fuword8_nocheck) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +uint16_t +dtrace_fuword16_nocheck(void *addr) +{ return (0); } + +#else +#if defined(__amd64) + + ENTRY(dtrace_fuword16_nocheck) + xorq %rax, %rax + movw (%rdi), %ax + ret + SET_SIZE(dtrace_fuword16_nocheck) + +#elif defined(__i386) + + ENTRY(dtrace_fuword16_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movzwl (%ecx), %eax + ret + SET_SIZE(dtrace_fuword16_nocheck) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +uint32_t +dtrace_fuword32_nocheck(void *addr) +{ return (0); } + +#else +#if defined(__amd64) + + ENTRY(dtrace_fuword32_nocheck) + xorq %rax, %rax + movl (%rdi), %eax + ret + SET_SIZE(dtrace_fuword32_nocheck) + +#elif defined(__i386) + + ENTRY(dtrace_fuword32_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + movl (%ecx), %eax + ret + SET_SIZE(dtrace_fuword32_nocheck) + +#endif /* __i386 */ +#endif + +#if defined(lint) + +/*ARGSUSED*/ +uint64_t +dtrace_fuword64_nocheck(void *addr) +{ return (0); } + +#else +#if defined(__amd64) + + ENTRY(dtrace_fuword64_nocheck) + movq (%rdi), %rax + ret + SET_SIZE(dtrace_fuword64_nocheck) + +#elif defined(__i386) + + ENTRY(dtrace_fuword64_nocheck) + movl 4(%esp), %ecx + xorl %eax, %eax + xorl %edx, %edx + movl (%ecx), %eax + movl 4(%ecx), %edx + ret + SET_SIZE(dtrace_fuword64_nocheck) + +#endif /* __i386 */ +#endif + +#if defined(lint) || defined(__lint) + +/*ARGSUSED*/ +void +dtrace_probe_error(dtrace_state_t *state, dtrace_epid_t epid, int which, + int fault, int fltoffs, uintptr_t illval) +{} + +#else /* lint */ +#if defined(__amd64) + + ENTRY(dtrace_probe_error) + pushq %rbp + movq %rsp, %rbp + subq $0x8, %rsp + movq %r9, (%rsp) + movq %r8, %r9 + movq %rcx, %r8 + movq %rdx, %rcx + movq %rsi, %rdx + movq %rdi, %rsi + movl dtrace_probeid_error(%rip), %edi + call dtrace_probe + addq $0x8, %rsp + leave + ret + SET_SIZE(dtrace_probe_error) + +#elif defined(__i386) + + ENTRY(dtrace_probe_error) + pushl %ebp + movl %esp, %ebp + pushl 0x1c(%ebp) + pushl 0x18(%ebp) + pushl 0x14(%ebp) + pushl 0x10(%ebp) + pushl 0xc(%ebp) + pushl 0x8(%ebp) + pushl dtrace_probeid_error + call dtrace_probe + movl %ebp, %esp + popl %ebp + ret + SET_SIZE(dtrace_probe_error) + +#endif /* __i386 */ +#endif diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_isa.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_isa.c new file mode 100644 index 00000000..6b5a4e1b --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/dtrace_isa.c @@ -0,0 +1,746 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2007 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/dtrace_impl.h> +#include <sys/stack.h> +#include <sys/frame.h> +#include <sys/cmn_err.h> +#include <sys/privregs.h> +#include <sys/sysmacros.h> + +extern uintptr_t kernelbase; + +int dtrace_ustackdepth_max = 2048; + +void +dtrace_getpcstack(pc_t *pcstack, int pcstack_limit, int aframes, + uint32_t *intrpc) +{ + struct frame *fp = (struct frame *)dtrace_getfp(); + struct frame *nextfp, *minfp, *stacktop; + int depth = 0; + int on_intr, last = 0; + uintptr_t pc; + uintptr_t caller = CPU->cpu_dtrace_caller; + + if ((on_intr = CPU_ON_INTR(CPU)) != 0) + stacktop = (struct frame *)(CPU->cpu_intr_stack + SA(MINFRAME)); + else + stacktop = (struct frame *)curthread->t_stk; + minfp = fp; + + aframes++; + + if (intrpc != NULL && depth < pcstack_limit) + pcstack[depth++] = (pc_t)intrpc; + + while (depth < pcstack_limit) { + nextfp = (struct frame *)fp->fr_savfp; + pc = fp->fr_savpc; + + if (nextfp <= minfp || nextfp >= stacktop) { + if (on_intr) { + /* + * Hop from interrupt stack to thread stack. + */ + stacktop = (struct frame *)curthread->t_stk; + minfp = (struct frame *)curthread->t_stkbase; + on_intr = 0; + continue; + } + + /* + * This is the last frame we can process; indicate + * that we should return after processing this frame. + */ + last = 1; + } + + if (aframes > 0) { + if (--aframes == 0 && caller != NULL) { + /* + * We've just run out of artificial frames, + * and we have a valid caller -- fill it in + * now. + */ + ASSERT(depth < pcstack_limit); + pcstack[depth++] = (pc_t)caller; + caller = NULL; + } + } else { + if (depth < pcstack_limit) + pcstack[depth++] = (pc_t)pc; + } + + if (last) { + while (depth < pcstack_limit) + pcstack[depth++] = NULL; + return; + } + + fp = nextfp; + minfp = fp; + } +} + +static int +dtrace_getustack_common(uint64_t *pcstack, int pcstack_limit, uintptr_t pc, + uintptr_t sp) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + uintptr_t oldcontext = lwp->lwp_oldcontext; + uintptr_t oldsp; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags; + size_t s1, s2; + int ret = 0; + + ASSERT(pcstack == NULL || pcstack_limit > 0); + ASSERT(dtrace_ustackdepth_max > 0); + + if (p->p_model == DATAMODEL_NATIVE) { + s1 = sizeof (struct frame) + 2 * sizeof (long); + s2 = s1 + sizeof (siginfo_t); + } else { + s1 = sizeof (struct frame32) + 3 * sizeof (int); + s2 = s1 + sizeof (siginfo32_t); + } + + while (pc != 0) { + /* + * We limit the number of times we can go around this + * loop to account for a circular stack. + */ + if (ret++ >= dtrace_ustackdepth_max) { + *flags |= CPU_DTRACE_BADSTACK; + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = sp; + break; + } + + if (pcstack != NULL) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + } + + if (sp == 0) + break; + + oldsp = sp; + + if (oldcontext == sp + s1 || oldcontext == sp + s2) { + if (p->p_model == DATAMODEL_NATIVE) { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fulword(&gregs[REG_FP]); + pc = dtrace_fulword(&gregs[REG_PC]); + + oldcontext = dtrace_fulword(&ucp->uc_link); + } else { + ucontext32_t *ucp = (ucontext32_t *)oldcontext; + greg32_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fuword32(&gregs[EBP]); + pc = dtrace_fuword32(&gregs[EIP]); + + oldcontext = dtrace_fuword32(&ucp->uc_link); + } + } else { + if (p->p_model == DATAMODEL_NATIVE) { + struct frame *fr = (struct frame *)sp; + + pc = dtrace_fulword(&fr->fr_savpc); + sp = dtrace_fulword(&fr->fr_savfp); + } else { + struct frame32 *fr = (struct frame32 *)sp; + + pc = dtrace_fuword32(&fr->fr_savpc); + sp = dtrace_fuword32(&fr->fr_savfp); + } + } + + if (sp == oldsp) { + *flags |= CPU_DTRACE_BADSTACK; + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = sp; + break; + } + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + + return (ret); +} + +void +dtrace_getupcstack(uint64_t *pcstack, int pcstack_limit) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp; + int n; + + ASSERT(DTRACE_CPUFLAG_ISSET(CPU_DTRACE_NOFAULT)); + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = rp->r_pc; + sp = rp->r_fp; + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + n = dtrace_getustack_common(pcstack, pcstack_limit, pc, sp); + ASSERT(n >= 0); + ASSERT(n <= pcstack_limit); + + pcstack += n; + pcstack_limit -= n; + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = NULL; +} + +int +dtrace_getustackdepth(void) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp; + int n = 0; + + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + return (0); + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_FAULT)) + return (-1); + + pc = rp->r_pc; + sp = rp->r_fp; + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + n++; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + n += dtrace_getustack_common(NULL, 0, pc, sp); + + return (n); +} + +void +dtrace_getufpstack(uint64_t *pcstack, uint64_t *fpstack, int pcstack_limit) +{ + klwp_t *lwp = ttolwp(curthread); + proc_t *p = curproc; + struct regs *rp; + uintptr_t pc, sp, oldcontext; + volatile uint16_t *flags = + (volatile uint16_t *)&cpu_core[CPU->cpu_id].cpuc_dtrace_flags; + size_t s1, s2; + + if (*flags & CPU_DTRACE_FAULT) + return; + + if (pcstack_limit <= 0) + return; + + /* + * If there's no user context we still need to zero the stack. + */ + if (lwp == NULL || p == NULL || (rp = lwp->lwp_regs) == NULL) + goto zero; + + *pcstack++ = (uint64_t)p->p_pid; + pcstack_limit--; + + if (pcstack_limit <= 0) + return; + + pc = rp->r_pc; + sp = rp->r_fp; + oldcontext = lwp->lwp_oldcontext; + + if (p->p_model == DATAMODEL_NATIVE) { + s1 = sizeof (struct frame) + 2 * sizeof (long); + s2 = s1 + sizeof (siginfo_t); + } else { + s1 = sizeof (struct frame32) + 3 * sizeof (int); + s2 = s1 + sizeof (siginfo32_t); + } + + if (DTRACE_CPUFLAG_ISSET(CPU_DTRACE_ENTRY)) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = 0; + pcstack_limit--; + if (pcstack_limit <= 0) + return; + + if (p->p_model == DATAMODEL_NATIVE) + pc = dtrace_fulword((void *)rp->r_sp); + else + pc = dtrace_fuword32((void *)rp->r_sp); + } + + while (pc != 0) { + *pcstack++ = (uint64_t)pc; + *fpstack++ = sp; + pcstack_limit--; + if (pcstack_limit <= 0) + break; + + if (sp == 0) + break; + + if (oldcontext == sp + s1 || oldcontext == sp + s2) { + if (p->p_model == DATAMODEL_NATIVE) { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fulword(&gregs[REG_FP]); + pc = dtrace_fulword(&gregs[REG_PC]); + + oldcontext = dtrace_fulword(&ucp->uc_link); + } else { + ucontext_t *ucp = (ucontext_t *)oldcontext; + greg_t *gregs = ucp->uc_mcontext.gregs; + + sp = dtrace_fuword32(&gregs[EBP]); + pc = dtrace_fuword32(&gregs[EIP]); + + oldcontext = dtrace_fuword32(&ucp->uc_link); + } + } else { + if (p->p_model == DATAMODEL_NATIVE) { + struct frame *fr = (struct frame *)sp; + + pc = dtrace_fulword(&fr->fr_savpc); + sp = dtrace_fulword(&fr->fr_savfp); + } else { + struct frame32 *fr = (struct frame32 *)sp; + + pc = dtrace_fuword32(&fr->fr_savpc); + sp = dtrace_fuword32(&fr->fr_savfp); + } + } + + /* + * This is totally bogus: if we faulted, we're going to clear + * the fault and break. This is to deal with the apparently + * broken Java stacks on x86. + */ + if (*flags & CPU_DTRACE_FAULT) { + *flags &= ~CPU_DTRACE_FAULT; + break; + } + } + +zero: + while (pcstack_limit-- > 0) + *pcstack++ = NULL; +} + +/*ARGSUSED*/ +uint64_t +dtrace_getarg(int arg, int aframes) +{ + uintptr_t val; + struct frame *fp = (struct frame *)dtrace_getfp(); + uintptr_t *stack; + int i; +#if defined(__amd64) + /* + * A total of 6 arguments are passed via registers; any argument with + * index of 5 or lower is therefore in a register. + */ + int inreg = 5; +#endif + + for (i = 1; i <= aframes; i++) { + fp = (struct frame *)(fp->fr_savfp); + + if (fp->fr_savpc == (pc_t)dtrace_invop_callsite) { +#if !defined(__amd64) + /* + * If we pass through the invalid op handler, we will + * use the pointer that it passed to the stack as the + * second argument to dtrace_invop() as the pointer to + * the stack. When using this stack, we must step + * beyond the EIP/RIP that was pushed when the trap was + * taken -- hence the "+ 1" below. + */ + stack = ((uintptr_t **)&fp[1])[1] + 1; +#else + /* + * In the case of amd64, we will use the pointer to the + * regs structure that was pushed when we took the + * trap. To get this structure, we must increment + * beyond the frame structure, and then again beyond + * the calling RIP stored in dtrace_invop(). If the + * argument that we're seeking is passed on the stack, + * we'll pull the true stack pointer out of the saved + * registers and decrement our argument by the number + * of arguments passed in registers; if the argument + * we're seeking is passed in regsiters, we can just + * load it directly. + */ + struct regs *rp = (struct regs *)((uintptr_t)&fp[1] + + sizeof (uintptr_t)); + + if (arg <= inreg) { + stack = (uintptr_t *)&rp->r_rdi; + } else { + stack = (uintptr_t *)(rp->r_rsp); + arg -= inreg; + } +#endif + goto load; + } + + } + + /* + * We know that we did not come through a trap to get into + * dtrace_probe() -- the provider simply called dtrace_probe() + * directly. As this is the case, we need to shift the argument + * that we're looking for: the probe ID is the first argument to + * dtrace_probe(), so the argument n will actually be found where + * one would expect to find argument (n + 1). + */ + arg++; + +#if defined(__amd64) + if (arg <= inreg) { + /* + * This shouldn't happen. If the argument is passed in a + * register then it should have been, well, passed in a + * register... + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + arg -= (inreg + 1); +#endif + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[arg]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return (val); +} + +/*ARGSUSED*/ +int +dtrace_getstackdepth(int aframes) +{ + struct frame *fp = (struct frame *)dtrace_getfp(); + struct frame *nextfp, *minfp, *stacktop; + int depth = 0; + int on_intr; + + if ((on_intr = CPU_ON_INTR(CPU)) != 0) + stacktop = (struct frame *)(CPU->cpu_intr_stack + SA(MINFRAME)); + else + stacktop = (struct frame *)curthread->t_stk; + minfp = fp; + + aframes++; + + for (;;) { + depth++; + + nextfp = (struct frame *)fp->fr_savfp; + + if (nextfp <= minfp || nextfp >= stacktop) { + if (on_intr) { + /* + * Hop from interrupt stack to thread stack. + */ + stacktop = (struct frame *)curthread->t_stk; + minfp = (struct frame *)curthread->t_stkbase; + on_intr = 0; + continue; + } + break; + } + + fp = nextfp; + minfp = fp; + } + + if (depth <= aframes) + return (0); + + return (depth - aframes); +} + +ulong_t +dtrace_getreg(struct regs *rp, uint_t reg) +{ +#if defined(__amd64) + int regmap[] = { + REG_GS, /* GS */ + REG_FS, /* FS */ + REG_ES, /* ES */ + REG_DS, /* DS */ + REG_RDI, /* EDI */ + REG_RSI, /* ESI */ + REG_RBP, /* EBP */ + REG_RSP, /* ESP */ + REG_RBX, /* EBX */ + REG_RDX, /* EDX */ + REG_RCX, /* ECX */ + REG_RAX, /* EAX */ + REG_TRAPNO, /* TRAPNO */ + REG_ERR, /* ERR */ + REG_RIP, /* EIP */ + REG_CS, /* CS */ + REG_RFL, /* EFL */ + REG_RSP, /* UESP */ + REG_SS /* SS */ + }; + + if (reg <= SS) { + if (reg >= sizeof (regmap) / sizeof (int)) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + reg = regmap[reg]; + } else { + reg -= SS + 1; + } + + switch (reg) { + case REG_RDI: + return (rp->r_rdi); + case REG_RSI: + return (rp->r_rsi); + case REG_RDX: + return (rp->r_rdx); + case REG_RCX: + return (rp->r_rcx); + case REG_R8: + return (rp->r_r8); + case REG_R9: + return (rp->r_r9); + case REG_RAX: + return (rp->r_rax); + case REG_RBX: + return (rp->r_rbx); + case REG_RBP: + return (rp->r_rbp); + case REG_R10: + return (rp->r_r10); + case REG_R11: + return (rp->r_r11); + case REG_R12: + return (rp->r_r12); + case REG_R13: + return (rp->r_r13); + case REG_R14: + return (rp->r_r14); + case REG_R15: + return (rp->r_r15); + case REG_DS: + return (rp->r_ds); + case REG_ES: + return (rp->r_es); + case REG_FS: + return (rp->r_fs); + case REG_GS: + return (rp->r_gs); + case REG_TRAPNO: + return (rp->r_trapno); + case REG_ERR: + return (rp->r_err); + case REG_RIP: + return (rp->r_rip); + case REG_CS: + return (rp->r_cs); + case REG_SS: + return (rp->r_ss); + case REG_RFL: + return (rp->r_rfl); + case REG_RSP: + return (rp->r_rsp); + default: + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + +#else + if (reg > SS) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + return ((&rp->r_gs)[reg]); +#endif +} + +static int +dtrace_copycheck(uintptr_t uaddr, uintptr_t kaddr, size_t size) +{ + ASSERT(kaddr >= kernelbase && kaddr + size >= kaddr); + + if (uaddr + size >= kernelbase || uaddr + size < uaddr) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = uaddr; + return (0); + } + + return (1); +} + +/*ARGSUSED*/ +void +dtrace_copyin(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(uaddr, kaddr, size); +} + +/*ARGSUSED*/ +void +dtrace_copyout(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copy(kaddr, uaddr, size); +} + +void +dtrace_copyinstr(uintptr_t uaddr, uintptr_t kaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(uaddr, kaddr, size, flags); +} + +void +dtrace_copyoutstr(uintptr_t kaddr, uintptr_t uaddr, size_t size, + volatile uint16_t *flags) +{ + if (dtrace_copycheck(uaddr, kaddr, size)) + dtrace_copystr(kaddr, uaddr, size, flags); +} + +uint8_t +dtrace_fuword8(void *uaddr) +{ + extern uint8_t dtrace_fuword8_nocheck(void *); + if ((uintptr_t)uaddr >= _userlimit) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword8_nocheck(uaddr)); +} + +uint16_t +dtrace_fuword16(void *uaddr) +{ + extern uint16_t dtrace_fuword16_nocheck(void *); + if ((uintptr_t)uaddr >= _userlimit) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword16_nocheck(uaddr)); +} + +uint32_t +dtrace_fuword32(void *uaddr) +{ + extern uint32_t dtrace_fuword32_nocheck(void *); + if ((uintptr_t)uaddr >= _userlimit) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword32_nocheck(uaddr)); +} + +uint64_t +dtrace_fuword64(void *uaddr) +{ + extern uint64_t dtrace_fuword64_nocheck(void *); + if ((uintptr_t)uaddr >= _userlimit) { + DTRACE_CPUFLAG_SET(CPU_DTRACE_BADADDR); + cpu_core[CPU->cpu_id].cpuc_dtrace_illval = (uintptr_t)uaddr; + return (0); + } + return (dtrace_fuword64_nocheck(uaddr)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap.conf new file mode 100644 index 00000000..a25f883f --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap.conf @@ -0,0 +1,39 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2004 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="fasttrap" parent="pseudo" instance=0; + +# fasttrap-max-probes is the limit on the number of tracepoints created +# by DTrace's pid provider. This value should be increased if DTrace is +# unable to create the desired probes because the limit has been hit. +#fasttrap-max-probes=250000; + +# fasttrap-hash-size determines the size of the hash table used to store +# enabled DTrace pid provider tracepoints. If there are many enabled +# tracepoints or many tracepoints hash to the same value, increasing this +# variable can improve the performance of executing a traced instruction. +#fasttrap-hash-size=16384; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap_isa.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap_isa.c new file mode 100644 index 00000000..c4b58453 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fasttrap_isa.c @@ -0,0 +1,1745 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ + +/* + * Copyright 2008 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + +#pragma ident "%Z%%M% %I% %E% SMI" + +#include <sys/fasttrap_isa.h> +#include <sys/fasttrap_impl.h> +#include <sys/dtrace.h> +#include <sys/dtrace_impl.h> +#include <sys/cmn_err.h> +#include <sys/regset.h> +#include <sys/privregs.h> +#include <sys/segments.h> +#include <sys/x86_archext.h> +#include <sys/sysmacros.h> +#include <sys/trap.h> +#include <sys/archsystm.h> + +/* + * Lossless User-Land Tracing on x86 + * --------------------------------- + * + * The execution of most instructions is not dependent on the address; for + * these instructions it is sufficient to copy them into the user process's + * address space and execute them. To effectively single-step an instruction + * in user-land, we copy out the following sequence of instructions to scratch + * space in the user thread's ulwp_t structure. + * + * We then set the program counter (%eip or %rip) to point to this scratch + * space. Once execution resumes, the original instruction is executed and + * then control flow is redirected to what was originally the subsequent + * instruction. If the kernel attemps to deliver a signal while single- + * stepping, the signal is deferred and the program counter is moved into the + * second sequence of instructions. The second sequence ends in a trap into + * the kernel where the deferred signal is then properly handled and delivered. + * + * For instructions whose execute is position dependent, we perform simple + * emulation. These instructions are limited to control transfer + * instructions in 32-bit mode, but in 64-bit mode there's the added wrinkle + * of %rip-relative addressing that means that almost any instruction can be + * position dependent. For all the details on how we emulate generic + * instructions included %rip-relative instructions, see the code in + * fasttrap_pid_probe() below where we handle instructions of type + * FASTTRAP_T_COMMON (under the header: Generic Instruction Tracing). + */ + +#define FASTTRAP_MODRM_MOD(modrm) (((modrm) >> 6) & 0x3) +#define FASTTRAP_MODRM_REG(modrm) (((modrm) >> 3) & 0x7) +#define FASTTRAP_MODRM_RM(modrm) ((modrm) & 0x7) +#define FASTTRAP_MODRM(mod, reg, rm) (((mod) << 6) | ((reg) << 3) | (rm)) + +#define FASTTRAP_SIB_SCALE(sib) (((sib) >> 6) & 0x3) +#define FASTTRAP_SIB_INDEX(sib) (((sib) >> 3) & 0x7) +#define FASTTRAP_SIB_BASE(sib) ((sib) & 0x7) + +#define FASTTRAP_REX_W(rex) (((rex) >> 3) & 1) +#define FASTTRAP_REX_R(rex) (((rex) >> 2) & 1) +#define FASTTRAP_REX_X(rex) (((rex) >> 1) & 1) +#define FASTTRAP_REX_B(rex) ((rex) & 1) +#define FASTTRAP_REX(w, r, x, b) \ + (0x40 | ((w) << 3) | ((r) << 2) | ((x) << 1) | (b)) + +/* + * Single-byte op-codes. + */ +#define FASTTRAP_PUSHL_EBP 0x55 + +#define FASTTRAP_JO 0x70 +#define FASTTRAP_JNO 0x71 +#define FASTTRAP_JB 0x72 +#define FASTTRAP_JAE 0x73 +#define FASTTRAP_JE 0x74 +#define FASTTRAP_JNE 0x75 +#define FASTTRAP_JBE 0x76 +#define FASTTRAP_JA 0x77 +#define FASTTRAP_JS 0x78 +#define FASTTRAP_JNS 0x79 +#define FASTTRAP_JP 0x7a +#define FASTTRAP_JNP 0x7b +#define FASTTRAP_JL 0x7c +#define FASTTRAP_JGE 0x7d +#define FASTTRAP_JLE 0x7e +#define FASTTRAP_JG 0x7f + +#define FASTTRAP_NOP 0x90 + +#define FASTTRAP_MOV_EAX 0xb8 +#define FASTTRAP_MOV_ECX 0xb9 + +#define FASTTRAP_RET16 0xc2 +#define FASTTRAP_RET 0xc3 + +#define FASTTRAP_LOOPNZ 0xe0 +#define FASTTRAP_LOOPZ 0xe1 +#define FASTTRAP_LOOP 0xe2 +#define FASTTRAP_JCXZ 0xe3 + +#define FASTTRAP_CALL 0xe8 +#define FASTTRAP_JMP32 0xe9 +#define FASTTRAP_JMP8 0xeb + +#define FASTTRAP_INT3 0xcc +#define FASTTRAP_INT 0xcd + +#define FASTTRAP_2_BYTE_OP 0x0f +#define FASTTRAP_GROUP5_OP 0xff + +/* + * Two-byte op-codes (second byte only). + */ +#define FASTTRAP_0F_JO 0x80 +#define FASTTRAP_0F_JNO 0x81 +#define FASTTRAP_0F_JB 0x82 +#define FASTTRAP_0F_JAE 0x83 +#define FASTTRAP_0F_JE 0x84 +#define FASTTRAP_0F_JNE 0x85 +#define FASTTRAP_0F_JBE 0x86 +#define FASTTRAP_0F_JA 0x87 +#define FASTTRAP_0F_JS 0x88 +#define FASTTRAP_0F_JNS 0x89 +#define FASTTRAP_0F_JP 0x8a +#define FASTTRAP_0F_JNP 0x8b +#define FASTTRAP_0F_JL 0x8c +#define FASTTRAP_0F_JGE 0x8d +#define FASTTRAP_0F_JLE 0x8e +#define FASTTRAP_0F_JG 0x8f + +#define FASTTRAP_EFLAGS_OF 0x800 +#define FASTTRAP_EFLAGS_DF 0x400 +#define FASTTRAP_EFLAGS_SF 0x080 +#define FASTTRAP_EFLAGS_ZF 0x040 +#define FASTTRAP_EFLAGS_AF 0x010 +#define FASTTRAP_EFLAGS_PF 0x004 +#define FASTTRAP_EFLAGS_CF 0x001 + +/* + * Instruction prefixes. + */ +#define FASTTRAP_PREFIX_OPERAND 0x66 +#define FASTTRAP_PREFIX_ADDRESS 0x67 +#define FASTTRAP_PREFIX_CS 0x2E +#define FASTTRAP_PREFIX_DS 0x3E +#define FASTTRAP_PREFIX_ES 0x26 +#define FASTTRAP_PREFIX_FS 0x64 +#define FASTTRAP_PREFIX_GS 0x65 +#define FASTTRAP_PREFIX_SS 0x36 +#define FASTTRAP_PREFIX_LOCK 0xF0 +#define FASTTRAP_PREFIX_REP 0xF3 +#define FASTTRAP_PREFIX_REPNE 0xF2 + +#define FASTTRAP_NOREG 0xff + +/* + * Map between instruction register encodings and the kernel constants which + * correspond to indicies into struct regs. + */ +#ifdef __amd64 +static const uint8_t regmap[16] = { + REG_RAX, REG_RCX, REG_RDX, REG_RBX, REG_RSP, REG_RBP, REG_RSI, REG_RDI, + REG_R8, REG_R9, REG_R10, REG_R11, REG_R12, REG_R13, REG_R14, REG_R15, +}; +#else +static const uint8_t regmap[8] = { + EAX, ECX, EDX, EBX, UESP, EBP, ESI, EDI +}; +#endif + +static ulong_t fasttrap_getreg(struct regs *, uint_t); + +static uint64_t +fasttrap_anarg(struct regs *rp, int function_entry, int argno) +{ + uint64_t value; + int shift = function_entry ? 1 : 0; + +#ifdef __amd64 + if (curproc->p_model == DATAMODEL_LP64) { + uintptr_t *stack; + + /* + * In 64-bit mode, the first six arguments are stored in + * registers. + */ + if (argno < 6) + return ((&rp->r_rdi)[argno]); + + stack = (uintptr_t *)rp->r_sp; + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + value = dtrace_fulword(&stack[argno - 6 + shift]); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | CPU_DTRACE_BADADDR); + } else { +#endif + uint32_t *stack = (uint32_t *)rp->r_sp; + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + value = dtrace_fuword32(&stack[argno + shift]); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | CPU_DTRACE_BADADDR); +#ifdef __amd64 + } +#endif + + return (value); +} + +/*ARGSUSED*/ +int +fasttrap_tracepoint_init(proc_t *p, fasttrap_tracepoint_t *tp, uintptr_t pc, + fasttrap_probe_type_t type) +{ + uint8_t instr[FASTTRAP_MAX_INSTR_SIZE + 10]; + size_t len = FASTTRAP_MAX_INSTR_SIZE; + size_t first = MIN(len, PAGESIZE - (pc & PAGEOFFSET)); + uint_t start = 0; + int rmindex, size; + uint8_t seg, rex = 0; + + /* + * Read the instruction at the given address out of the process's + * address space. We don't have to worry about a debugger + * changing this instruction before we overwrite it with our trap + * instruction since P_PR_LOCK is set. Since instructions can span + * pages, we potentially read the instruction in two parts. If the + * second part fails, we just zero out that part of the instruction. + */ + if (uread(p, &instr[0], first, pc) != 0) + return (-1); + if (len > first && + uread(p, &instr[first], len - first, pc + first) != 0) { + bzero(&instr[first], len - first); + len = first; + } + + /* + * If the disassembly fails, then we have a malformed instruction. + */ + if ((size = dtrace_instr_size_isa(instr, p->p_model, &rmindex)) <= 0) + return (-1); + + /* + * Make sure the disassembler isn't completely broken. + */ + ASSERT(-1 <= rmindex && rmindex < size); + + /* + * If the computed size is greater than the number of bytes read, + * then it was a malformed instruction possibly because it fell on a + * page boundary and the subsequent page was missing or because of + * some malicious user. + */ + if (size > len) + return (-1); + + tp->ftt_size = (uint8_t)size; + tp->ftt_segment = FASTTRAP_SEG_NONE; + + /* + * Find the start of the instruction's opcode by processing any + * legacy prefixes. + */ + for (;;) { + seg = 0; + switch (instr[start]) { + case FASTTRAP_PREFIX_SS: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_GS: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_FS: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_ES: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_DS: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_CS: + seg++; + RT_FALL_THRU(); + case FASTTRAP_PREFIX_OPERAND: + case FASTTRAP_PREFIX_ADDRESS: + case FASTTRAP_PREFIX_LOCK: + case FASTTRAP_PREFIX_REP: + case FASTTRAP_PREFIX_REPNE: + if (seg != 0) { + /* + * It's illegal for an instruction to specify + * two segment prefixes -- give up on this + * illegal instruction. + */ + if (tp->ftt_segment != FASTTRAP_SEG_NONE) + return (-1); + + tp->ftt_segment = seg; + } + start++; + continue; + } + break; + } + +#ifdef __amd64 + /* + * Identify the REX prefix on 64-bit processes. + */ + if (p->p_model == DATAMODEL_LP64 && (instr[start] & 0xf0) == 0x40) + rex = instr[start++]; +#endif + + /* + * Now that we're pretty sure that the instruction is okay, copy the + * valid part to the tracepoint. + */ + bcopy(instr, tp->ftt_instr, FASTTRAP_MAX_INSTR_SIZE); + + tp->ftt_type = FASTTRAP_T_COMMON; + if (instr[start] == FASTTRAP_2_BYTE_OP) { + switch (instr[start + 1]) { + case FASTTRAP_0F_JO: + case FASTTRAP_0F_JNO: + case FASTTRAP_0F_JB: + case FASTTRAP_0F_JAE: + case FASTTRAP_0F_JE: + case FASTTRAP_0F_JNE: + case FASTTRAP_0F_JBE: + case FASTTRAP_0F_JA: + case FASTTRAP_0F_JS: + case FASTTRAP_0F_JNS: + case FASTTRAP_0F_JP: + case FASTTRAP_0F_JNP: + case FASTTRAP_0F_JL: + case FASTTRAP_0F_JGE: + case FASTTRAP_0F_JLE: + case FASTTRAP_0F_JG: + tp->ftt_type = FASTTRAP_T_JCC; + tp->ftt_code = (instr[start + 1] & 0x0f) | FASTTRAP_JO; + tp->ftt_dest = pc + tp->ftt_size + + /* LINTED - alignment */ + *(int32_t *)&instr[start + 2]; + break; + } + } else if (instr[start] == FASTTRAP_GROUP5_OP) { + uint_t mod = FASTTRAP_MODRM_MOD(instr[start + 1]); + uint_t reg = FASTTRAP_MODRM_REG(instr[start + 1]); + uint_t rm = FASTTRAP_MODRM_RM(instr[start + 1]); + + if (reg == 2 || reg == 4) { + uint_t i, sz; + + if (reg == 2) + tp->ftt_type = FASTTRAP_T_CALL; + else + tp->ftt_type = FASTTRAP_T_JMP; + + if (mod == 3) + tp->ftt_code = 2; + else + tp->ftt_code = 1; + + ASSERT(p->p_model == DATAMODEL_LP64 || rex == 0); + + /* + * See AMD x86-64 Architecture Programmer's Manual + * Volume 3, Section 1.2.7, Table 1-12, and + * Appendix A.3.1, Table A-15. + */ + if (mod != 3 && rm == 4) { + uint8_t sib = instr[start + 2]; + uint_t index = FASTTRAP_SIB_INDEX(sib); + uint_t base = FASTTRAP_SIB_BASE(sib); + + tp->ftt_scale = FASTTRAP_SIB_SCALE(sib); + + tp->ftt_index = (index == 4) ? + FASTTRAP_NOREG : + regmap[index | (FASTTRAP_REX_X(rex) << 3)]; + tp->ftt_base = (mod == 0 && base == 5) ? + FASTTRAP_NOREG : + regmap[base | (FASTTRAP_REX_B(rex) << 3)]; + + i = 3; + sz = mod == 1 ? 1 : 4; + } else { + /* + * In 64-bit mode, mod == 0 and r/m == 5 + * denotes %rip-relative addressing; in 32-bit + * mode, the base register isn't used. In both + * modes, there is a 32-bit operand. + */ + if (mod == 0 && rm == 5) { +#ifdef __amd64 + if (p->p_model == DATAMODEL_LP64) + tp->ftt_base = REG_RIP; + else +#endif + tp->ftt_base = FASTTRAP_NOREG; + sz = 4; + } else { + uint8_t base = rm | + (FASTTRAP_REX_B(rex) << 3); + + tp->ftt_base = regmap[base]; + sz = mod == 1 ? 1 : mod == 2 ? 4 : 0; + } + tp->ftt_index = FASTTRAP_NOREG; + i = 2; + } + + if (sz == 1) { + tp->ftt_dest = *(int8_t *)&instr[start + i]; + } else if (sz == 4) { + /* LINTED - alignment */ + tp->ftt_dest = *(int32_t *)&instr[start + i]; + } else { + tp->ftt_dest = 0; + } + } + } else { + switch (instr[start]) { + case FASTTRAP_RET: + tp->ftt_type = FASTTRAP_T_RET; + break; + + case FASTTRAP_RET16: + tp->ftt_type = FASTTRAP_T_RET16; + /* LINTED - alignment */ + tp->ftt_dest = *(uint16_t *)&instr[start + 1]; + break; + + case FASTTRAP_JO: + case FASTTRAP_JNO: + case FASTTRAP_JB: + case FASTTRAP_JAE: + case FASTTRAP_JE: + case FASTTRAP_JNE: + case FASTTRAP_JBE: + case FASTTRAP_JA: + case FASTTRAP_JS: + case FASTTRAP_JNS: + case FASTTRAP_JP: + case FASTTRAP_JNP: + case FASTTRAP_JL: + case FASTTRAP_JGE: + case FASTTRAP_JLE: + case FASTTRAP_JG: + tp->ftt_type = FASTTRAP_T_JCC; + tp->ftt_code = instr[start]; + tp->ftt_dest = pc + tp->ftt_size + + (int8_t)instr[start + 1]; + break; + + case FASTTRAP_LOOPNZ: + case FASTTRAP_LOOPZ: + case FASTTRAP_LOOP: + tp->ftt_type = FASTTRAP_T_LOOP; + tp->ftt_code = instr[start]; + tp->ftt_dest = pc + tp->ftt_size + + (int8_t)instr[start + 1]; + break; + + case FASTTRAP_JCXZ: + tp->ftt_type = FASTTRAP_T_JCXZ; + tp->ftt_dest = pc + tp->ftt_size + + (int8_t)instr[start + 1]; + break; + + case FASTTRAP_CALL: + tp->ftt_type = FASTTRAP_T_CALL; + tp->ftt_dest = pc + tp->ftt_size + + /* LINTED - alignment */ + *(int32_t *)&instr[start + 1]; + tp->ftt_code = 0; + break; + + case FASTTRAP_JMP32: + tp->ftt_type = FASTTRAP_T_JMP; + tp->ftt_dest = pc + tp->ftt_size + + /* LINTED - alignment */ + *(int32_t *)&instr[start + 1]; + break; + case FASTTRAP_JMP8: + tp->ftt_type = FASTTRAP_T_JMP; + tp->ftt_dest = pc + tp->ftt_size + + (int8_t)instr[start + 1]; + break; + + case FASTTRAP_PUSHL_EBP: + if (start == 0) + tp->ftt_type = FASTTRAP_T_PUSHL_EBP; + break; + + case FASTTRAP_NOP: +#ifdef __amd64 + ASSERT(p->p_model == DATAMODEL_LP64 || rex == 0); + + /* + * On amd64 we have to be careful not to confuse a nop + * (actually xchgl %eax, %eax) with an instruction using + * the same opcode, but that does something different + * (e.g. xchgl %r8d, %eax or xcghq %r8, %rax). + */ + if (FASTTRAP_REX_B(rex) == 0) +#endif + tp->ftt_type = FASTTRAP_T_NOP; + break; + + case FASTTRAP_INT3: + /* + * The pid provider shares the int3 trap with debugger + * breakpoints so we can't instrument them. + */ + ASSERT(instr[start] == FASTTRAP_INSTR); + return (-1); + + case FASTTRAP_INT: + /* + * Interrupts seem like they could be traced with + * no negative implications, but it's possible that + * a thread could be redirected by the trap handling + * code which would eventually return to the + * instruction after the interrupt. If the interrupt + * were in our scratch space, the subsequent + * instruction might be overwritten before we return. + * Accordingly we refuse to instrument any interrupt. + */ + return (-1); + } + } + +#ifdef __amd64 + if (p->p_model == DATAMODEL_LP64 && tp->ftt_type == FASTTRAP_T_COMMON) { + /* + * If the process is 64-bit and the instruction type is still + * FASTTRAP_T_COMMON -- meaning we're going to copy it out an + * execute it -- we need to watch for %rip-relative + * addressing mode. See the portion of fasttrap_pid_probe() + * below where we handle tracepoints with type + * FASTTRAP_T_COMMON for how we emulate instructions that + * employ %rip-relative addressing. + */ + if (rmindex != -1) { + uint_t mod = FASTTRAP_MODRM_MOD(instr[rmindex]); + uint_t reg = FASTTRAP_MODRM_REG(instr[rmindex]); + uint_t rm = FASTTRAP_MODRM_RM(instr[rmindex]); + + ASSERT(rmindex > start); + + if (mod == 0 && rm == 5) { + /* + * We need to be sure to avoid other + * registers used by this instruction. While + * the reg field may determine the op code + * rather than denoting a register, assuming + * that it denotes a register is always safe. + * We leave the REX field intact and use + * whatever value's there for simplicity. + */ + if (reg != 0) { + tp->ftt_ripmode = FASTTRAP_RIP_1 | + (FASTTRAP_RIP_X * + FASTTRAP_REX_B(rex)); + rm = 0; + } else { + tp->ftt_ripmode = FASTTRAP_RIP_2 | + (FASTTRAP_RIP_X * + FASTTRAP_REX_B(rex)); + rm = 1; + } + + tp->ftt_modrm = tp->ftt_instr[rmindex]; + tp->ftt_instr[rmindex] = + FASTTRAP_MODRM(2, reg, rm); + } + } + } +#endif + + return (0); +} + +int +fasttrap_tracepoint_install(proc_t *p, fasttrap_tracepoint_t *tp) +{ + fasttrap_instr_t instr = FASTTRAP_INSTR; + + if (uwrite(p, &instr, 1, tp->ftt_pc) != 0) + return (-1); + + return (0); +} + +int +fasttrap_tracepoint_remove(proc_t *p, fasttrap_tracepoint_t *tp) +{ + uint8_t instr; + + /* + * Distinguish between read or write failures and a changed + * instruction. + */ + if (uread(p, &instr, 1, tp->ftt_pc) != 0) + return (0); + if (instr != FASTTRAP_INSTR) + return (0); + if (uwrite(p, &tp->ftt_instr[0], 1, tp->ftt_pc) != 0) + return (-1); + + return (0); +} + +#ifdef __amd64 +static uintptr_t +fasttrap_fulword_noerr(const void *uaddr) +{ + uintptr_t ret; + + if (fasttrap_fulword(uaddr, &ret) == 0) + return (ret); + + return (0); +} +#endif + +static uint32_t +fasttrap_fuword32_noerr(const void *uaddr) +{ + uint32_t ret; + + if (fasttrap_fuword32(uaddr, &ret) == 0) + return (ret); + + return (0); +} + +static void +fasttrap_return_common(struct regs *rp, uintptr_t pc, pid_t pid, + uintptr_t new_pc) +{ + fasttrap_tracepoint_t *tp; + fasttrap_bucket_t *bucket; + fasttrap_id_t *id; + kmutex_t *pid_mtx; + + pid_mtx = &cpu_core[CPU->cpu_id].cpuc_pid_lock; + mutex_enter(pid_mtx); + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (pid == tp->ftt_pid && pc == tp->ftt_pc && + tp->ftt_proc->ftpc_acount != 0) + break; + } + + /* + * Don't sweat it if we can't find the tracepoint again; unlike + * when we're in fasttrap_pid_probe(), finding the tracepoint here + * is not essential to the correct execution of the process. + */ + if (tp == NULL) { + mutex_exit(pid_mtx); + return; + } + + for (id = tp->ftt_retids; id != NULL; id = id->fti_next) { + /* + * If there's a branch that could act as a return site, we + * need to trace it, and check here if the program counter is + * external to the function. + */ + if (tp->ftt_type != FASTTRAP_T_RET && + tp->ftt_type != FASTTRAP_T_RET16 && + new_pc - id->fti_probe->ftp_faddr < + id->fti_probe->ftp_fsize) + continue; + + dtrace_probe(id->fti_probe->ftp_id, + pc - id->fti_probe->ftp_faddr, + rp->r_r0, rp->r_r1, 0, 0); + } + + mutex_exit(pid_mtx); +} + +static void +fasttrap_sigsegv(proc_t *p, kthread_t *t, uintptr_t addr) +{ + sigqueue_t *sqp = kmem_zalloc(sizeof (sigqueue_t), KM_SLEEP); + + sqp->sq_info.si_signo = SIGSEGV; + sqp->sq_info.si_code = SEGV_MAPERR; + sqp->sq_info.si_addr = (caddr_t)addr; + + mutex_enter(&p->p_lock); + sigaddqa(p, t, sqp); + mutex_exit(&p->p_lock); + + if (t != NULL) + aston(t); +} + +#ifdef __amd64 +static void +fasttrap_usdt_args64(fasttrap_probe_t *probe, struct regs *rp, int argc, + uintptr_t *argv) +{ + int i, x, cap = MIN(argc, probe->ftp_nargs); + uintptr_t *stack = (uintptr_t *)rp->r_sp; + + for (i = 0; i < cap; i++) { + x = probe->ftp_argmap[i]; + + if (x < 6) + argv[i] = (&rp->r_rdi)[x]; + else + argv[i] = fasttrap_fulword_noerr(&stack[x]); + } + + for (; i < argc; i++) { + argv[i] = 0; + } +} +#endif + +static void +fasttrap_usdt_args32(fasttrap_probe_t *probe, struct regs *rp, int argc, + uint32_t *argv) +{ + int i, x, cap = MIN(argc, probe->ftp_nargs); + uint32_t *stack = (uint32_t *)rp->r_sp; + + for (i = 0; i < cap; i++) { + x = probe->ftp_argmap[i]; + + argv[i] = fasttrap_fuword32_noerr(&stack[x]); + } + + for (; i < argc; i++) { + argv[i] = 0; + } +} + +static int +fasttrap_do_seg(fasttrap_tracepoint_t *tp, struct regs *rp, uintptr_t *addr) +{ + proc_t *p = curproc; + user_desc_t *desc; + uint16_t sel, ndx, type; + uintptr_t limit; + + switch (tp->ftt_segment) { + case FASTTRAP_SEG_CS: + sel = rp->r_cs; + break; + case FASTTRAP_SEG_DS: + sel = rp->r_ds; + break; + case FASTTRAP_SEG_ES: + sel = rp->r_es; + break; + case FASTTRAP_SEG_FS: + sel = rp->r_fs; + break; + case FASTTRAP_SEG_GS: + sel = rp->r_gs; + break; + case FASTTRAP_SEG_SS: + sel = rp->r_ss; + break; + } + + /* + * Make sure the given segment register specifies a user priority + * selector rather than a kernel selector. + */ + if (!SELISUPL(sel)) + return (-1); + + ndx = SELTOIDX(sel); + + /* + * Check the bounds and grab the descriptor out of the specified + * descriptor table. + */ + if (SELISLDT(sel)) { + if (ndx > p->p_ldtlimit) + return (-1); + + desc = p->p_ldt + ndx; + + } else { + if (ndx >= NGDT) + return (-1); + + desc = cpu_get_gdt() + ndx; + } + + /* + * The descriptor must have user privilege level and it must be + * present in memory. + */ + if (desc->usd_dpl != SEL_UPL || desc->usd_p != 1) + return (-1); + + type = desc->usd_type; + + /* + * If the S bit in the type field is not set, this descriptor can + * only be used in system context. + */ + if ((type & 0x10) != 0x10) + return (-1); + + limit = USEGD_GETLIMIT(desc) * (desc->usd_gran ? PAGESIZE : 1); + + if (tp->ftt_segment == FASTTRAP_SEG_CS) { + /* + * The code/data bit and readable bit must both be set. + */ + if ((type & 0xa) != 0xa) + return (-1); + + if (*addr > limit) + return (-1); + } else { + /* + * The code/data bit must be clear. + */ + if ((type & 0x8) != 0) + return (-1); + + /* + * If the expand-down bit is clear, we just check the limit as + * it would naturally be applied. Otherwise, we need to check + * that the address is the range [limit + 1 .. 0xffff] or + * [limit + 1 ... 0xffffffff] depending on if the default + * operand size bit is set. + */ + if ((type & 0x4) == 0) { + if (*addr > limit) + return (-1); + } else if (desc->usd_def32) { + if (*addr < limit + 1 || 0xffff < *addr) + return (-1); + } else { + if (*addr < limit + 1 || 0xffffffff < *addr) + return (-1); + } + } + + *addr += USEGD_GETBASE(desc); + + return (0); +} + +int +fasttrap_pid_probe(struct regs *rp) +{ + proc_t *p = curproc; + uintptr_t pc = rp->r_pc - 1, new_pc = 0; + fasttrap_bucket_t *bucket; + kmutex_t *pid_mtx; + fasttrap_tracepoint_t *tp, tp_local; + pid_t pid; + dtrace_icookie_t cookie; + uint_t is_enabled = 0; + + /* + * It's possible that a user (in a veritable orgy of bad planning) + * could redirect this thread's flow of control before it reached the + * return probe fasttrap. In this case we need to kill the process + * since it's in a unrecoverable state. + */ + if (curthread->t_dtrace_step) { + ASSERT(curthread->t_dtrace_on); + fasttrap_sigtrap(p, curthread, pc); + return (0); + } + + /* + * Clear all user tracing flags. + */ + curthread->t_dtrace_ft = 0; + curthread->t_dtrace_pc = 0; + curthread->t_dtrace_npc = 0; + curthread->t_dtrace_scrpc = 0; + curthread->t_dtrace_astpc = 0; +#ifdef __amd64 + curthread->t_dtrace_regv = 0; +#endif + + /* + * Treat a child created by a call to vfork(2) as if it were its + * parent. We know that there's only one thread of control in such a + * process: this one. + */ + while (p->p_flag & SVFORK) { + p = p->p_parent; + } + + pid = p->p_pid; + pid_mtx = &cpu_core[CPU->cpu_id].cpuc_pid_lock; + mutex_enter(pid_mtx); + bucket = &fasttrap_tpoints.fth_table[FASTTRAP_TPOINTS_INDEX(pid, pc)]; + + /* + * Lookup the tracepoint that the process just hit. + */ + for (tp = bucket->ftb_data; tp != NULL; tp = tp->ftt_next) { + if (pid == tp->ftt_pid && pc == tp->ftt_pc && + tp->ftt_proc->ftpc_acount != 0) + break; + } + + /* + * If we couldn't find a matching tracepoint, either a tracepoint has + * been inserted without using the pid<pid> ioctl interface (see + * fasttrap_ioctl), or somehow we have mislaid this tracepoint. + */ + if (tp == NULL) { + mutex_exit(pid_mtx); + return (-1); + } + + /* + * Set the program counter to the address of the traced instruction + * so that it looks right in ustack() output. + */ + rp->r_pc = pc; + + if (tp->ftt_ids != NULL) { + fasttrap_id_t *id; + +#ifdef __amd64 + if (p->p_model == DATAMODEL_LP64) { + for (id = tp->ftt_ids; id != NULL; id = id->fti_next) { + fasttrap_probe_t *probe = id->fti_probe; + + if (id->fti_ptype == DTFTP_ENTRY) { + /* + * We note that this was an entry + * probe to help ustack() find the + * first caller. + */ + cookie = dtrace_interrupt_disable(); + DTRACE_CPUFLAG_SET(CPU_DTRACE_ENTRY); + dtrace_probe(probe->ftp_id, rp->r_rdi, + rp->r_rsi, rp->r_rdx, rp->r_rcx, + rp->r_r8); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_ENTRY); + dtrace_interrupt_enable(cookie); + } else if (id->fti_ptype == DTFTP_IS_ENABLED) { + /* + * Note that in this case, we don't + * call dtrace_probe() since it's only + * an artificial probe meant to change + * the flow of control so that it + * encounters the true probe. + */ + is_enabled = 1; + } else if (probe->ftp_argmap == NULL) { + dtrace_probe(probe->ftp_id, rp->r_rdi, + rp->r_rsi, rp->r_rdx, rp->r_rcx, + rp->r_r8); + } else { + uintptr_t t[5]; + + fasttrap_usdt_args64(probe, rp, + sizeof (t) / sizeof (t[0]), t); + + dtrace_probe(probe->ftp_id, t[0], t[1], + t[2], t[3], t[4]); + } + } + } else { +#endif + uintptr_t s0, s1, s2, s3, s4, s5; + uint32_t *stack = (uint32_t *)rp->r_sp; + + /* + * In 32-bit mode, all arguments are passed on the + * stack. If this is a function entry probe, we need + * to skip the first entry on the stack as it + * represents the return address rather than a + * parameter to the function. + */ + s0 = fasttrap_fuword32_noerr(&stack[0]); + s1 = fasttrap_fuword32_noerr(&stack[1]); + s2 = fasttrap_fuword32_noerr(&stack[2]); + s3 = fasttrap_fuword32_noerr(&stack[3]); + s4 = fasttrap_fuword32_noerr(&stack[4]); + s5 = fasttrap_fuword32_noerr(&stack[5]); + + for (id = tp->ftt_ids; id != NULL; id = id->fti_next) { + fasttrap_probe_t *probe = id->fti_probe; + + if (id->fti_ptype == DTFTP_ENTRY) { + /* + * We note that this was an entry + * probe to help ustack() find the + * first caller. + */ + cookie = dtrace_interrupt_disable(); + DTRACE_CPUFLAG_SET(CPU_DTRACE_ENTRY); + dtrace_probe(probe->ftp_id, s1, s2, + s3, s4, s5); + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_ENTRY); + dtrace_interrupt_enable(cookie); + } else if (id->fti_ptype == DTFTP_IS_ENABLED) { + /* + * Note that in this case, we don't + * call dtrace_probe() since it's only + * an artificial probe meant to change + * the flow of control so that it + * encounters the true probe. + */ + is_enabled = 1; + } else if (probe->ftp_argmap == NULL) { + dtrace_probe(probe->ftp_id, s0, s1, + s2, s3, s4); + } else { + uint32_t t[5]; + + fasttrap_usdt_args32(probe, rp, + sizeof (t) / sizeof (t[0]), t); + + dtrace_probe(probe->ftp_id, t[0], t[1], + t[2], t[3], t[4]); + } + } +#ifdef __amd64 + } +#endif + } + + /* + * We're about to do a bunch of work so we cache a local copy of + * the tracepoint to emulate the instruction, and then find the + * tracepoint again later if we need to light up any return probes. + */ + tp_local = *tp; + mutex_exit(pid_mtx); + tp = &tp_local; + + /* + * Set the program counter to appear as though the traced instruction + * had completely executed. This ensures that fasttrap_getreg() will + * report the expected value for REG_RIP. + */ + rp->r_pc = pc + tp->ftt_size; + + /* + * If there's an is-enabled probe connected to this tracepoint it + * means that there was a 'xorl %eax, %eax' or 'xorq %rax, %rax' + * instruction that was placed there by DTrace when the binary was + * linked. As this probe is, in fact, enabled, we need to stuff 1 + * into %eax or %rax. Accordingly, we can bypass all the instruction + * emulation logic since we know the inevitable result. It's possible + * that a user could construct a scenario where the 'is-enabled' + * probe was on some other instruction, but that would be a rather + * exotic way to shoot oneself in the foot. + */ + if (is_enabled) { + rp->r_r0 = 1; + new_pc = rp->r_pc; + goto done; + } + + /* + * We emulate certain types of instructions to ensure correctness + * (in the case of position dependent instructions) or optimize + * common cases. The rest we have the thread execute back in user- + * land. + */ + switch (tp->ftt_type) { + case FASTTRAP_T_RET: + case FASTTRAP_T_RET16: + { + uintptr_t dst; + uintptr_t addr; + int ret; + + /* + * We have to emulate _every_ facet of the behavior of a ret + * instruction including what happens if the load from %esp + * fails; in that case, we send a SIGSEGV. + */ +#ifdef __amd64 + if (p->p_model == DATAMODEL_NATIVE) { +#endif + ret = fasttrap_fulword((void *)rp->r_sp, &dst); + addr = rp->r_sp + sizeof (uintptr_t); +#ifdef __amd64 + } else { + uint32_t dst32; + ret = fasttrap_fuword32((void *)rp->r_sp, &dst32); + dst = dst32; + addr = rp->r_sp + sizeof (uint32_t); + } +#endif + + if (ret == -1) { + fasttrap_sigsegv(p, curthread, rp->r_sp); + new_pc = pc; + break; + } + + if (tp->ftt_type == FASTTRAP_T_RET16) + addr += tp->ftt_dest; + + rp->r_sp = addr; + new_pc = dst; + break; + } + + case FASTTRAP_T_JCC: + { + uint_t taken; + + switch (tp->ftt_code) { + case FASTTRAP_JO: + taken = (rp->r_ps & FASTTRAP_EFLAGS_OF) != 0; + break; + case FASTTRAP_JNO: + taken = (rp->r_ps & FASTTRAP_EFLAGS_OF) == 0; + break; + case FASTTRAP_JB: + taken = (rp->r_ps & FASTTRAP_EFLAGS_CF) != 0; + break; + case FASTTRAP_JAE: + taken = (rp->r_ps & FASTTRAP_EFLAGS_CF) == 0; + break; + case FASTTRAP_JE: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) != 0; + break; + case FASTTRAP_JNE: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) == 0; + break; + case FASTTRAP_JBE: + taken = (rp->r_ps & FASTTRAP_EFLAGS_CF) != 0 || + (rp->r_ps & FASTTRAP_EFLAGS_ZF) != 0; + break; + case FASTTRAP_JA: + taken = (rp->r_ps & FASTTRAP_EFLAGS_CF) == 0 && + (rp->r_ps & FASTTRAP_EFLAGS_ZF) == 0; + break; + case FASTTRAP_JS: + taken = (rp->r_ps & FASTTRAP_EFLAGS_SF) != 0; + break; + case FASTTRAP_JNS: + taken = (rp->r_ps & FASTTRAP_EFLAGS_SF) == 0; + break; + case FASTTRAP_JP: + taken = (rp->r_ps & FASTTRAP_EFLAGS_PF) != 0; + break; + case FASTTRAP_JNP: + taken = (rp->r_ps & FASTTRAP_EFLAGS_PF) == 0; + break; + case FASTTRAP_JL: + taken = ((rp->r_ps & FASTTRAP_EFLAGS_SF) == 0) != + ((rp->r_ps & FASTTRAP_EFLAGS_OF) == 0); + break; + case FASTTRAP_JGE: + taken = ((rp->r_ps & FASTTRAP_EFLAGS_SF) == 0) == + ((rp->r_ps & FASTTRAP_EFLAGS_OF) == 0); + break; + case FASTTRAP_JLE: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) != 0 || + ((rp->r_ps & FASTTRAP_EFLAGS_SF) == 0) != + ((rp->r_ps & FASTTRAP_EFLAGS_OF) == 0); + break; + case FASTTRAP_JG: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) == 0 && + ((rp->r_ps & FASTTRAP_EFLAGS_SF) == 0) == + ((rp->r_ps & FASTTRAP_EFLAGS_OF) == 0); + break; + + } + + if (taken) + new_pc = tp->ftt_dest; + else + new_pc = pc + tp->ftt_size; + break; + } + + case FASTTRAP_T_LOOP: + { + uint_t taken; +#ifdef __amd64 + greg_t cx = rp->r_rcx--; +#else + greg_t cx = rp->r_ecx--; +#endif + + switch (tp->ftt_code) { + case FASTTRAP_LOOPNZ: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) == 0 && + cx != 0; + break; + case FASTTRAP_LOOPZ: + taken = (rp->r_ps & FASTTRAP_EFLAGS_ZF) != 0 && + cx != 0; + break; + case FASTTRAP_LOOP: + taken = (cx != 0); + break; + } + + if (taken) + new_pc = tp->ftt_dest; + else + new_pc = pc + tp->ftt_size; + break; + } + + case FASTTRAP_T_JCXZ: + { +#ifdef __amd64 + greg_t cx = rp->r_rcx; +#else + greg_t cx = rp->r_ecx; +#endif + + if (cx == 0) + new_pc = tp->ftt_dest; + else + new_pc = pc + tp->ftt_size; + break; + } + + case FASTTRAP_T_PUSHL_EBP: + { + int ret; + uintptr_t addr; +#ifdef __amd64 + if (p->p_model == DATAMODEL_NATIVE) { +#endif + addr = rp->r_sp - sizeof (uintptr_t); + ret = fasttrap_sulword((void *)addr, rp->r_fp); +#ifdef __amd64 + } else { + addr = rp->r_sp - sizeof (uint32_t); + ret = fasttrap_suword32((void *)addr, + (uint32_t)rp->r_fp); + } +#endif + + if (ret == -1) { + fasttrap_sigsegv(p, curthread, addr); + new_pc = pc; + break; + } + + rp->r_sp = addr; + new_pc = pc + tp->ftt_size; + break; + } + + case FASTTRAP_T_NOP: + new_pc = pc + tp->ftt_size; + break; + + case FASTTRAP_T_JMP: + case FASTTRAP_T_CALL: + if (tp->ftt_code == 0) { + new_pc = tp->ftt_dest; + } else { + uintptr_t value, addr = tp->ftt_dest; + + if (tp->ftt_base != FASTTRAP_NOREG) + addr += fasttrap_getreg(rp, tp->ftt_base); + if (tp->ftt_index != FASTTRAP_NOREG) + addr += fasttrap_getreg(rp, tp->ftt_index) << + tp->ftt_scale; + + if (tp->ftt_code == 1) { + /* + * If there's a segment prefix for this + * instruction, we'll need to check permissions + * and bounds on the given selector, and adjust + * the address accordingly. + */ + if (tp->ftt_segment != FASTTRAP_SEG_NONE && + fasttrap_do_seg(tp, rp, &addr) != 0) { + fasttrap_sigsegv(p, curthread, addr); + new_pc = pc; + break; + } + +#ifdef __amd64 + if (p->p_model == DATAMODEL_NATIVE) { +#endif + if (fasttrap_fulword((void *)addr, + &value) == -1) { + fasttrap_sigsegv(p, curthread, + addr); + new_pc = pc; + break; + } + new_pc = value; +#ifdef __amd64 + } else { + uint32_t value32; + addr = (uintptr_t)(uint32_t)addr; + if (fasttrap_fuword32((void *)addr, + &value32) == -1) { + fasttrap_sigsegv(p, curthread, + addr); + new_pc = pc; + break; + } + new_pc = value32; + } +#endif + } else { + new_pc = addr; + } + } + + /* + * If this is a call instruction, we need to push the return + * address onto the stack. If this fails, we send the process + * a SIGSEGV and reset the pc to emulate what would happen if + * this instruction weren't traced. + */ + if (tp->ftt_type == FASTTRAP_T_CALL) { + int ret; + uintptr_t addr; +#ifdef __amd64 + if (p->p_model == DATAMODEL_NATIVE) { + addr = rp->r_sp - sizeof (uintptr_t); + ret = fasttrap_sulword((void *)addr, + pc + tp->ftt_size); + } else { +#endif + addr = rp->r_sp - sizeof (uint32_t); + ret = fasttrap_suword32((void *)addr, + (uint32_t)(pc + tp->ftt_size)); +#ifdef __amd64 + } +#endif + + if (ret == -1) { + fasttrap_sigsegv(p, curthread, addr); + new_pc = pc; + break; + } + + rp->r_sp = addr; + } + + break; + + case FASTTRAP_T_COMMON: + { + uintptr_t addr; +#if defined(__amd64) + uint8_t scratch[2 * FASTTRAP_MAX_INSTR_SIZE + 22]; +#else + uint8_t scratch[2 * FASTTRAP_MAX_INSTR_SIZE + 7]; +#endif + uint_t i = 0; + klwp_t *lwp = ttolwp(curthread); + + /* + * Compute the address of the ulwp_t and step over the + * ul_self pointer. The method used to store the user-land + * thread pointer is very different on 32- and 64-bit + * kernels. + */ +#if defined(__amd64) + if (p->p_model == DATAMODEL_LP64) { + addr = lwp->lwp_pcb.pcb_fsbase; + addr += sizeof (void *); + } else { + addr = lwp->lwp_pcb.pcb_gsbase; + addr += sizeof (caddr32_t); + } +#else + addr = USEGD_GETBASE(&lwp->lwp_pcb.pcb_gsdesc); + addr += sizeof (void *); +#endif + + /* + * Generic Instruction Tracing + * --------------------------- + * + * This is the layout of the scratch space in the user-land + * thread structure for our generated instructions. + * + * 32-bit mode bytes + * ------------------------ ----- + * a: <original instruction> <= 15 + * jmp <pc + tp->ftt_size> 5 + * b: <original instrction> <= 15 + * int T_DTRACE_RET 2 + * ----- + * <= 37 + * + * 64-bit mode bytes + * ------------------------ ----- + * a: <original instruction> <= 15 + * jmp 0(%rip) 6 + * <pc + tp->ftt_size> 8 + * b: <original instruction> <= 15 + * int T_DTRACE_RET 2 + * ----- + * <= 46 + * + * The %pc is set to a, and curthread->t_dtrace_astpc is set + * to b. If we encounter a signal on the way out of the + * kernel, trap() will set %pc to curthread->t_dtrace_astpc + * so that we execute the original instruction and re-enter + * the kernel rather than redirecting to the next instruction. + * + * If there are return probes (so we know that we're going to + * need to reenter the kernel after executing the original + * instruction), the scratch space will just contain the + * original instruction followed by an interrupt -- the same + * data as at b. + * + * %rip-relative Addressing + * ------------------------ + * + * There's a further complication in 64-bit mode due to %rip- + * relative addressing. While this is clearly a beneficial + * architectural decision for position independent code, it's + * hard not to see it as a personal attack against the pid + * provider since before there was a relatively small set of + * instructions to emulate; with %rip-relative addressing, + * almost every instruction can potentially depend on the + * address at which it's executed. Rather than emulating + * the broad spectrum of instructions that can now be + * position dependent, we emulate jumps and others as in + * 32-bit mode, and take a different tack for instructions + * using %rip-relative addressing. + * + * For every instruction that uses the ModRM byte, the + * in-kernel disassembler reports its location. We use the + * ModRM byte to identify that an instruction uses + * %rip-relative addressing and to see what other registers + * the instruction uses. To emulate those instructions, + * we modify the instruction to be %rax-relative rather than + * %rip-relative (or %rcx-relative if the instruction uses + * %rax; or %r8- or %r9-relative if the REX.B is present so + * we don't have to rewrite the REX prefix). We then load + * the value that %rip would have been into the scratch + * register and generate an instruction to reset the scratch + * register back to its original value. The instruction + * sequence looks like this: + * + * 64-mode %rip-relative bytes + * ------------------------ ----- + * a: <modified instruction> <= 15 + * movq $<value>, %<scratch> 6 + * jmp 0(%rip) 6 + * <pc + tp->ftt_size> 8 + * b: <modified instruction> <= 15 + * int T_DTRACE_RET 2 + * ----- + * 52 + * + * We set curthread->t_dtrace_regv so that upon receiving + * a signal we can reset the value of the scratch register. + */ + + ASSERT(tp->ftt_size < FASTTRAP_MAX_INSTR_SIZE); + + curthread->t_dtrace_scrpc = addr; + bcopy(tp->ftt_instr, &scratch[i], tp->ftt_size); + i += tp->ftt_size; + +#ifdef __amd64 + if (tp->ftt_ripmode != 0) { + greg_t *reg; + + ASSERT(p->p_model == DATAMODEL_LP64); + ASSERT(tp->ftt_ripmode & + (FASTTRAP_RIP_1 | FASTTRAP_RIP_2)); + + /* + * If this was a %rip-relative instruction, we change + * it to be either a %rax- or %rcx-relative + * instruction (depending on whether those registers + * are used as another operand; or %r8- or %r9- + * relative depending on the value of REX.B). We then + * set that register and generate a movq instruction + * to reset the value. + */ + if (tp->ftt_ripmode & FASTTRAP_RIP_X) + scratch[i++] = FASTTRAP_REX(1, 0, 0, 1); + else + scratch[i++] = FASTTRAP_REX(1, 0, 0, 0); + + if (tp->ftt_ripmode & FASTTRAP_RIP_1) + scratch[i++] = FASTTRAP_MOV_EAX; + else + scratch[i++] = FASTTRAP_MOV_ECX; + + switch (tp->ftt_ripmode) { + case FASTTRAP_RIP_1: + reg = &rp->r_rax; + curthread->t_dtrace_reg = REG_RAX; + break; + case FASTTRAP_RIP_2: + reg = &rp->r_rcx; + curthread->t_dtrace_reg = REG_RCX; + break; + case FASTTRAP_RIP_1 | FASTTRAP_RIP_X: + reg = &rp->r_r8; + curthread->t_dtrace_reg = REG_R8; + break; + case FASTTRAP_RIP_2 | FASTTRAP_RIP_X: + reg = &rp->r_r9; + curthread->t_dtrace_reg = REG_R9; + break; + } + + /* LINTED - alignment */ + *(uint64_t *)&scratch[i] = *reg; + curthread->t_dtrace_regv = *reg; + *reg = pc + tp->ftt_size; + i += sizeof (uint64_t); + } +#endif + + /* + * Generate the branch instruction to what would have + * normally been the subsequent instruction. In 32-bit mode, + * this is just a relative branch; in 64-bit mode this is a + * %rip-relative branch that loads the 64-bit pc value + * immediately after the jmp instruction. + */ +#ifdef __amd64 + if (p->p_model == DATAMODEL_LP64) { + scratch[i++] = FASTTRAP_GROUP5_OP; + scratch[i++] = FASTTRAP_MODRM(0, 4, 5); + /* LINTED - alignment */ + *(uint32_t *)&scratch[i] = 0; + i += sizeof (uint32_t); + /* LINTED - alignment */ + *(uint64_t *)&scratch[i] = pc + tp->ftt_size; + i += sizeof (uint64_t); + } else { +#endif + /* + * Set up the jmp to the next instruction; note that + * the size of the traced instruction cancels out. + */ + scratch[i++] = FASTTRAP_JMP32; + /* LINTED - alignment */ + *(uint32_t *)&scratch[i] = pc - addr - 5; + i += sizeof (uint32_t); +#ifdef __amd64 + } +#endif + + curthread->t_dtrace_astpc = addr + i; + bcopy(tp->ftt_instr, &scratch[i], tp->ftt_size); + i += tp->ftt_size; + scratch[i++] = FASTTRAP_INT; + scratch[i++] = T_DTRACE_RET; + + ASSERT(i <= sizeof (scratch)); + + if (fasttrap_copyout(scratch, (char *)addr, i)) { + fasttrap_sigtrap(p, curthread, pc); + new_pc = pc; + break; + } + + if (tp->ftt_retids != NULL) { + curthread->t_dtrace_step = 1; + curthread->t_dtrace_ret = 1; + new_pc = curthread->t_dtrace_astpc; + } else { + new_pc = curthread->t_dtrace_scrpc; + } + + curthread->t_dtrace_pc = pc; + curthread->t_dtrace_npc = pc + tp->ftt_size; + curthread->t_dtrace_on = 1; + break; + } + + default: + panic("fasttrap: mishandled an instruction"); + } + +done: + /* + * If there were no return probes when we first found the tracepoint, + * we should feel no obligation to honor any return probes that were + * subsequently enabled -- they'll just have to wait until the next + * time around. + */ + if (tp->ftt_retids != NULL) { + /* + * We need to wait until the results of the instruction are + * apparent before invoking any return probes. If this + * instruction was emulated we can just call + * fasttrap_return_common(); if it needs to be executed, we + * need to wait until the user thread returns to the kernel. + */ + if (tp->ftt_type != FASTTRAP_T_COMMON) { + /* + * Set the program counter to the address of the traced + * instruction so that it looks right in ustack() + * output. We had previously set it to the end of the + * instruction to simplify %rip-relative addressing. + */ + rp->r_pc = pc; + + fasttrap_return_common(rp, pc, pid, new_pc); + } else { + ASSERT(curthread->t_dtrace_ret != 0); + ASSERT(curthread->t_dtrace_pc == pc); + ASSERT(curthread->t_dtrace_scrpc != 0); + ASSERT(new_pc == curthread->t_dtrace_astpc); + } + } + + rp->r_pc = new_pc; + + return (0); +} + +int +fasttrap_return_probe(struct regs *rp) +{ + proc_t *p = curproc; + uintptr_t pc = curthread->t_dtrace_pc; + uintptr_t npc = curthread->t_dtrace_npc; + + curthread->t_dtrace_pc = 0; + curthread->t_dtrace_npc = 0; + curthread->t_dtrace_scrpc = 0; + curthread->t_dtrace_astpc = 0; + + /* + * Treat a child created by a call to vfork(2) as if it were its + * parent. We know that there's only one thread of control in such a + * process: this one. + */ + while (p->p_flag & SVFORK) { + p = p->p_parent; + } + + /* + * We set rp->r_pc to the address of the traced instruction so + * that it appears to dtrace_probe() that we're on the original + * instruction, and so that the user can't easily detect our + * complex web of lies. dtrace_return_probe() (our caller) + * will correctly set %pc after we return. + */ + rp->r_pc = pc; + + fasttrap_return_common(rp, pc, p->p_pid, npc); + + return (0); +} + +/*ARGSUSED*/ +uint64_t +fasttrap_pid_getarg(void *arg, dtrace_id_t id, void *parg, int argno, + int aframes) +{ + return (fasttrap_anarg(ttolwp(curthread)->lwp_regs, 1, argno)); +} + +/*ARGSUSED*/ +uint64_t +fasttrap_usdt_getarg(void *arg, dtrace_id_t id, void *parg, int argno, + int aframes) +{ + return (fasttrap_anarg(ttolwp(curthread)->lwp_regs, 0, argno)); +} + +static ulong_t +fasttrap_getreg(struct regs *rp, uint_t reg) +{ +#ifdef __amd64 + switch (reg) { + case REG_R15: return (rp->r_r15); + case REG_R14: return (rp->r_r14); + case REG_R13: return (rp->r_r13); + case REG_R12: return (rp->r_r12); + case REG_R11: return (rp->r_r11); + case REG_R10: return (rp->r_r10); + case REG_R9: return (rp->r_r9); + case REG_R8: return (rp->r_r8); + case REG_RDI: return (rp->r_rdi); + case REG_RSI: return (rp->r_rsi); + case REG_RBP: return (rp->r_rbp); + case REG_RBX: return (rp->r_rbx); + case REG_RDX: return (rp->r_rdx); + case REG_RCX: return (rp->r_rcx); + case REG_RAX: return (rp->r_rax); + case REG_TRAPNO: return (rp->r_trapno); + case REG_ERR: return (rp->r_err); + case REG_RIP: return (rp->r_rip); + case REG_CS: return (rp->r_cs); + case REG_RFL: return (rp->r_rfl); + case REG_RSP: return (rp->r_rsp); + case REG_SS: return (rp->r_ss); + case REG_FS: return (rp->r_fs); + case REG_GS: return (rp->r_gs); + case REG_DS: return (rp->r_ds); + case REG_ES: return (rp->r_es); + case REG_FSBASE: return (rdmsr(MSR_AMD_FSBASE)); + case REG_GSBASE: return (rdmsr(MSR_AMD_GSBASE)); + } + + panic("dtrace: illegal register constant"); + /*NOTREACHED*/ +#else + if (reg >= _NGREG) + panic("dtrace: illegal register constant"); + + return (((greg_t *)&rp->r_gs)[reg]); +#endif +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.c new file mode 100644 index 00000000..b9353bd3 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.c @@ -0,0 +1,849 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/modctl.h> +#include <sys/dtrace.h> +#include <sys/kobj.h> +#include <sys/stat.h> +#include <sys/ddi.h> +#include <sys/sunddi.h> +#include <sys/conf.h> + +#define FBT_PUSHL_EBP 0x55 +#define FBT_MOVL_ESP_EBP0_V0 0x8b +#define FBT_MOVL_ESP_EBP1_V0 0xec +#define FBT_MOVL_ESP_EBP0_V1 0x89 +#define FBT_MOVL_ESP_EBP1_V1 0xe5 +#define FBT_REX_RSP_RBP 0x48 + +#define FBT_POPL_EBP 0x5d +#define FBT_RET 0xc3 +#define FBT_RET_IMM16 0xc2 +#define FBT_LEAVE 0xc9 + +#ifdef __amd64 +#define FBT_PATCHVAL 0xcc +#else +#define FBT_PATCHVAL 0xf0 +#endif + +#define FBT_ENTRY "entry" +#define FBT_RETURN "return" +#define FBT_ADDR2NDX(addr) ((((uintptr_t)(addr)) >> 4) & fbt_probetab_mask) +#define FBT_PROBETAB_SIZE 0x8000 /* 32k entries -- 128K total */ + +typedef struct fbt_probe { + struct fbt_probe *fbtp_hashnext; + uint8_t *fbtp_patchpoint; + int8_t fbtp_rval; + uint8_t fbtp_patchval; + uint8_t fbtp_savedval; + uintptr_t fbtp_roffset; + dtrace_id_t fbtp_id; + char *fbtp_name; + struct modctl *fbtp_ctl; + int fbtp_loadcnt; + int fbtp_symndx; + int fbtp_primary; + struct fbt_probe *fbtp_next; +} fbt_probe_t; + +static dev_info_t *fbt_devi; +static dtrace_provider_id_t fbt_id; +static fbt_probe_t **fbt_probetab; +static int fbt_probetab_size; +static int fbt_probetab_mask; +static int fbt_verbose = 0; + +static int +fbt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t rval) +{ + uintptr_t stack0, stack1, stack2, stack3, stack4; + fbt_probe_t *fbt = fbt_probetab[FBT_ADDR2NDX(addr)]; + + for (; fbt != NULL; fbt = fbt->fbtp_hashnext) { + if ((uintptr_t)fbt->fbtp_patchpoint == addr) { + if (fbt->fbtp_roffset == 0) { + int i = 0; + /* + * When accessing the arguments on the stack, + * we must protect against accessing beyond + * the stack. We can safely set NOFAULT here + * -- we know that interrupts are already + * disabled. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + CPU->cpu_dtrace_caller = stack[i++]; +#ifdef __amd64 + /* + * On amd64, stack[0] contains the dereferenced + * stack pointer, stack[1] contains savfp, + * stack[2] contains savpc. We want to step + * over these entries. + */ + i += 2; +#endif + stack0 = stack[i++]; + stack1 = stack[i++]; + stack2 = stack[i++]; + stack3 = stack[i++]; + stack4 = stack[i++]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); + + dtrace_probe(fbt->fbtp_id, stack0, stack1, + stack2, stack3, stack4); + + CPU->cpu_dtrace_caller = NULL; + } else { +#ifdef __amd64 + /* + * On amd64, we instrument the ret, not the + * leave. We therefore need to set the caller + * to assure that the top frame of a stack() + * action is correct. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + CPU->cpu_dtrace_caller = stack[0]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); +#endif + + dtrace_probe(fbt->fbtp_id, fbt->fbtp_roffset, + rval, 0, 0, 0); + CPU->cpu_dtrace_caller = NULL; + } + + return (fbt->fbtp_rval); + } + } + + return (0); +} + +/*ARGSUSED*/ +static void +fbt_provide_module(void *arg, struct modctl *ctl) +{ + struct module *mp = ctl->mod_mp; + char *str = mp->strings; + int nsyms = mp->nsyms; + Shdr *symhdr = mp->symhdr; + char *modname = ctl->mod_modname; + char *name; + fbt_probe_t *fbt, *retfbt; + size_t symsize; + int i, size; + + /* + * Employees of dtrace and their families are ineligible. Void + * where prohibited. + */ + if (strcmp(modname, "dtrace") == 0) + return; + + if (ctl->mod_requisites != NULL) { + struct modctl_list *list; + + list = (struct modctl_list *)ctl->mod_requisites; + + for (; list != NULL; list = list->modl_next) { + if (strcmp(list->modl_modp->mod_modname, "dtrace") == 0) + return; + } + } + + /* + * KMDB is ineligible for instrumentation -- it may execute in + * any context, including probe context. + */ + if (strcmp(modname, "kmdbmod") == 0) + return; + + if (str == NULL || symhdr == NULL || symhdr->sh_addr == NULL) { + /* + * If this module doesn't (yet) have its string or symbol + * table allocated, clear out. + */ + return; + } + + symsize = symhdr->sh_entsize; + + if (mp->fbt_nentries) { + /* + * This module has some FBT entries allocated; we're afraid + * to screw with it. + */ + return; + } + + for (i = 1; i < nsyms; i++) { + uint8_t *instr, *limit; + Sym *sym = (Sym *)(symhdr->sh_addr + i * symsize); + int j; + + if (ELF_ST_TYPE(sym->st_info) != STT_FUNC) + continue; + + /* + * Weak symbols are not candidates. This could be made to + * work (where weak functions and their underlying function + * appear as two disjoint probes), but it's not simple. + */ + if (ELF_ST_BIND(sym->st_info) == STB_WEAK) + continue; + + name = str + sym->st_name; + + if (strstr(name, "dtrace_") == name && + strstr(name, "dtrace_safe_") != name) { + /* + * Anything beginning with "dtrace_" may be called + * from probe context unless it explitly indicates + * that it won't be called from probe context by + * using the prefix "dtrace_safe_". + */ + continue; + } + + if (strstr(name, "kdi_") == name || + strstr(name, "_kdi_") != NULL) { + /* + * Any function name beginning with "kdi_" or + * containing the string "_kdi_" is a part of the + * kernel debugger interface and may be called in + * arbitrary context -- including probe context. + */ + continue; + } + + /* + * Due to 4524008, _init and _fini may have a bloated st_size. + * While this bug was fixed quite some time ago, old drivers + * may be lurking. We need to develop a better solution to + * this problem, such that correct _init and _fini functions + * (the vast majority) may be correctly traced. One solution + * may be to scan through the entire symbol table to see if + * any symbol overlaps with _init. If none does, set a bit in + * the module structure that this module has correct _init and + * _fini sizes. This will cause some pain the first time a + * module is scanned, but at least it would be O(N) instead of + * O(N log N)... + */ + if (strcmp(name, "_init") == 0) + continue; + + if (strcmp(name, "_fini") == 0) + continue; + + /* + * In order to be eligible, the function must begin with the + * following sequence: + * + * pushl %esp + * movl %esp, %ebp + * + * Note that there are two variants of encodings that generate + * the movl; we must check for both. For 64-bit, we would + * normally insist that a function begin with the following + * sequence: + * + * pushq %rbp + * movq %rsp, %rbp + * + * However, the compiler for 64-bit often splits these two + * instructions -- and the first instruction in the function + * is often not the pushq. As a result, on 64-bit we look + * for any "pushq %rbp" in the function and we instrument + * this with a breakpoint instruction. + */ + instr = (uint8_t *)sym->st_value; + limit = (uint8_t *)(sym->st_value + sym->st_size); + +#ifdef __amd64 + while (instr < limit) { + if (*instr == FBT_PUSHL_EBP) + break; + + if ((size = dtrace_instr_size(instr)) <= 0) + break; + + instr += size; + } + + if (instr >= limit || *instr != FBT_PUSHL_EBP) { + /* + * We either don't save the frame pointer in this + * function, or we ran into some disassembly + * screw-up. Either way, we bail. + */ + continue; + } +#else + if (instr[0] != FBT_PUSHL_EBP) + continue; + + if (!(instr[1] == FBT_MOVL_ESP_EBP0_V0 && + instr[2] == FBT_MOVL_ESP_EBP1_V0) && + !(instr[1] == FBT_MOVL_ESP_EBP0_V1 && + instr[2] == FBT_MOVL_ESP_EBP1_V1)) + continue; +#endif + + fbt = kmem_zalloc(sizeof (fbt_probe_t), KM_SLEEP); + fbt->fbtp_name = name; + fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, + name, FBT_ENTRY, 3, fbt); + fbt->fbtp_patchpoint = instr; + fbt->fbtp_ctl = ctl; + fbt->fbtp_loadcnt = ctl->mod_loadcnt; + fbt->fbtp_rval = DTRACE_INVOP_PUSHL_EBP; + fbt->fbtp_savedval = *instr; + fbt->fbtp_patchval = FBT_PATCHVAL; + + fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; + fbt->fbtp_symndx = i; + fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; + + mp->fbt_nentries++; + + retfbt = NULL; +again: + if (instr >= limit) + continue; + + /* + * If this disassembly fails, then we've likely walked off into + * a jump table or some other unsuitable area. Bail out of the + * disassembly now. + */ + if ((size = dtrace_instr_size(instr)) <= 0) + continue; + +#ifdef __amd64 + /* + * We only instrument "ret" on amd64 -- we don't yet instrument + * ret imm16, largely because the compiler doesn't seem to + * (yet) emit them in the kernel... + */ + if (*instr != FBT_RET) { + instr += size; + goto again; + } +#else + if (!(size == 1 && + (*instr == FBT_POPL_EBP || *instr == FBT_LEAVE) && + (*(instr + 1) == FBT_RET || + *(instr + 1) == FBT_RET_IMM16))) { + instr += size; + goto again; + } +#endif + + /* + * We (desperately) want to avoid erroneously instrumenting a + * jump table, especially given that our markers are pretty + * short: two bytes on x86, and just one byte on amd64. To + * determine if we're looking at a true instruction sequence + * or an inline jump table that happens to contain the same + * byte sequences, we resort to some heuristic sleeze: we + * treat this instruction as being contained within a pointer, + * and see if that pointer points to within the body of the + * function. If it does, we refuse to instrument it. + */ + for (j = 0; j < sizeof (uintptr_t); j++) { + uintptr_t check = (uintptr_t)instr - j; + uint8_t *ptr; + + if (check < sym->st_value) + break; + + if (check + sizeof (uintptr_t) > (uintptr_t)limit) + continue; + + ptr = *(uint8_t **)check; + + if (ptr >= (uint8_t *)sym->st_value && ptr < limit) { + instr += size; + goto again; + } + } + + /* + * We have a winner! + */ + fbt = kmem_zalloc(sizeof (fbt_probe_t), KM_SLEEP); + fbt->fbtp_name = name; + + if (retfbt == NULL) { + fbt->fbtp_id = dtrace_probe_create(fbt_id, modname, + name, FBT_RETURN, 3, fbt); + } else { + retfbt->fbtp_next = fbt; + fbt->fbtp_id = retfbt->fbtp_id; + } + + retfbt = fbt; + fbt->fbtp_patchpoint = instr; + fbt->fbtp_ctl = ctl; + fbt->fbtp_loadcnt = ctl->mod_loadcnt; + +#ifndef __amd64 + if (*instr == FBT_POPL_EBP) { + fbt->fbtp_rval = DTRACE_INVOP_POPL_EBP; + } else { + ASSERT(*instr == FBT_LEAVE); + fbt->fbtp_rval = DTRACE_INVOP_LEAVE; + } + fbt->fbtp_roffset = + (uintptr_t)(instr - (uint8_t *)sym->st_value) + 1; + +#else + ASSERT(*instr == FBT_RET); + fbt->fbtp_rval = DTRACE_INVOP_RET; + fbt->fbtp_roffset = + (uintptr_t)(instr - (uint8_t *)sym->st_value); +#endif + + fbt->fbtp_savedval = *instr; + fbt->fbtp_patchval = FBT_PATCHVAL; + fbt->fbtp_hashnext = fbt_probetab[FBT_ADDR2NDX(instr)]; + fbt->fbtp_symndx = i; + fbt_probetab[FBT_ADDR2NDX(instr)] = fbt; + + mp->fbt_nentries++; + + instr += size; + goto again; + } +} + +/*ARGSUSED*/ +static void +fbt_destroy(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg, *next, *hash, *last; + struct modctl *ctl = fbt->fbtp_ctl; + int ndx; + + do { + if (ctl != NULL && ctl->mod_loadcnt == fbt->fbtp_loadcnt) { + if ((ctl->mod_loadcnt == fbt->fbtp_loadcnt && + ctl->mod_loaded)) { + ((struct module *) + (ctl->mod_mp))->fbt_nentries--; + } + } + + /* + * Now we need to remove this probe from the fbt_probetab. + */ + ndx = FBT_ADDR2NDX(fbt->fbtp_patchpoint); + last = NULL; + hash = fbt_probetab[ndx]; + + while (hash != fbt) { + ASSERT(hash != NULL); + last = hash; + hash = hash->fbtp_hashnext; + } + + if (last != NULL) { + last->fbtp_hashnext = fbt->fbtp_hashnext; + } else { + fbt_probetab[ndx] = fbt->fbtp_hashnext; + } + + next = fbt->fbtp_next; + kmem_free(fbt, sizeof (fbt_probe_t)); + + fbt = next; + } while (fbt != NULL); +} + +/*ARGSUSED*/ +static int +fbt_enable(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + struct modctl *ctl = fbt->fbtp_ctl; + + ctl->mod_nenabled++; + + if (!ctl->mod_loaded) { + if (fbt_verbose) { + cmn_err(CE_NOTE, "fbt is failing for probe %s " + "(module %s unloaded)", + fbt->fbtp_name, ctl->mod_modname); + } + + return (0); + } + + /* + * Now check that our modctl has the expected load count. If it + * doesn't, this module must have been unloaded and reloaded -- and + * we're not going to touch it. + */ + if (ctl->mod_loadcnt != fbt->fbtp_loadcnt) { + if (fbt_verbose) { + cmn_err(CE_NOTE, "fbt is failing for probe %s " + "(module %s reloaded)", + fbt->fbtp_name, ctl->mod_modname); + } + + return (0); + } + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_patchval; + + return (0); +} + +/*ARGSUSED*/ +static void +fbt_disable(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + struct modctl *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->mod_nenabled > 0); + ctl->mod_nenabled--; + + if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_savedval; +} + +/*ARGSUSED*/ +static void +fbt_suspend(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + struct modctl *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->mod_nenabled > 0); + + if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_savedval; +} + +/*ARGSUSED*/ +static void +fbt_resume(void *arg, dtrace_id_t id, void *parg) +{ + fbt_probe_t *fbt = parg; + struct modctl *ctl = fbt->fbtp_ctl; + + ASSERT(ctl->mod_nenabled > 0); + + if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt)) + return; + + for (; fbt != NULL; fbt = fbt->fbtp_next) + *fbt->fbtp_patchpoint = fbt->fbtp_patchval; +} + +/*ARGSUSED*/ +static void +fbt_getargdesc(void *arg, dtrace_id_t id, void *parg, dtrace_argdesc_t *desc) +{ + fbt_probe_t *fbt = parg; + struct modctl *ctl = fbt->fbtp_ctl; + struct module *mp = ctl->mod_mp; + ctf_file_t *fp = NULL, *pfp; + ctf_funcinfo_t f; + int error; + ctf_id_t argv[32], type; + int argc = sizeof (argv) / sizeof (ctf_id_t); + const char *parent; + + if (!ctl->mod_loaded || (ctl->mod_loadcnt != fbt->fbtp_loadcnt)) + goto err; + + if (fbt->fbtp_roffset != 0 && desc->dtargd_ndx == 0) { + (void) strcpy(desc->dtargd_native, "int"); + return; + } + + if ((fp = ctf_modopen(mp, &error)) == NULL) { + /* + * We have no CTF information for this module -- and therefore + * no args[] information. + */ + goto err; + } + + /* + * If we have a parent container, we must manually import it. + */ + if ((parent = ctf_parent_name(fp)) != NULL) { + struct modctl *mp = &modules; + struct modctl *mod = NULL; + + /* + * We must iterate over all modules to find the module that + * is our parent. + */ + do { + if (strcmp(mp->mod_modname, parent) == 0) { + mod = mp; + break; + } + } while ((mp = mp->mod_next) != &modules); + + if (mod == NULL) + goto err; + + if ((pfp = ctf_modopen(mod->mod_mp, &error)) == NULL) { + goto err; + } + + if (ctf_import(fp, pfp) != 0) { + ctf_close(pfp); + goto err; + } + + ctf_close(pfp); + } + + if (ctf_func_info(fp, fbt->fbtp_symndx, &f) == CTF_ERR) + goto err; + + if (fbt->fbtp_roffset != 0) { + if (desc->dtargd_ndx > 1) + goto err; + + ASSERT(desc->dtargd_ndx == 1); + type = f.ctc_return; + } else { + if (desc->dtargd_ndx + 1 > f.ctc_argc) + goto err; + + if (ctf_func_args(fp, fbt->fbtp_symndx, argc, argv) == CTF_ERR) + goto err; + + type = argv[desc->dtargd_ndx]; + } + + if (ctf_type_name(fp, type, desc->dtargd_native, + DTRACE_ARGTYPELEN) != NULL) { + ctf_close(fp); + return; + } +err: + if (fp != NULL) + ctf_close(fp); + + desc->dtargd_ndx = DTRACE_ARGNONE; +} + +static dtrace_pattr_t fbt_attr = { +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN }, +{ DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_ISA }, +{ DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_ISA }, +}; + +static dtrace_pops_t fbt_pops = { + NULL, + fbt_provide_module, + fbt_enable, + fbt_disable, + fbt_suspend, + fbt_resume, + fbt_getargdesc, + NULL, + NULL, + fbt_destroy +}; + +static void +fbt_cleanup(dev_info_t *devi) +{ + dtrace_invop_remove(fbt_invop); + ddi_remove_minor_node(devi, NULL); + kmem_free(fbt_probetab, fbt_probetab_size * sizeof (fbt_probe_t *)); + fbt_probetab = NULL; + fbt_probetab_mask = 0; +} + +static int +fbt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + switch (cmd) { + case DDI_ATTACH: + break; + case DDI_RESUME: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (fbt_probetab_size == 0) + fbt_probetab_size = FBT_PROBETAB_SIZE; + + fbt_probetab_mask = fbt_probetab_size - 1; + fbt_probetab = + kmem_zalloc(fbt_probetab_size * sizeof (fbt_probe_t *), KM_SLEEP); + + dtrace_invop_add(fbt_invop); + + if (ddi_create_minor_node(devi, "fbt", S_IFCHR, 0, + DDI_PSEUDO, NULL) == DDI_FAILURE || + dtrace_register("fbt", &fbt_attr, DTRACE_PRIV_KERNEL, NULL, + &fbt_pops, NULL, &fbt_id) != 0) { + fbt_cleanup(devi); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + fbt_devi = devi; + + return (DDI_SUCCESS); +} + +static int +fbt_detach(dev_info_t *devi, ddi_detach_cmd_t cmd) +{ + switch (cmd) { + case DDI_DETACH: + break; + case DDI_SUSPEND: + return (DDI_SUCCESS); + default: + return (DDI_FAILURE); + } + + if (dtrace_unregister(fbt_id) != 0) + return (DDI_FAILURE); + + fbt_cleanup(devi); + + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +fbt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)fbt_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +fbt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops fbt_cb_ops = { + fbt_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops fbt_ops = { + DEVO_REV, /* devo_rev */ + 0, /* refcnt */ + fbt_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + fbt_attach, /* attach */ + fbt_detach, /* detach */ + nodev, /* reset */ + &fbt_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Function Boundary Tracing", /* name of module */ + &fbt_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.conf new file mode 100644 index 00000000..8e7cbe1e --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/fbt.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#ident "%Z%%M% %I% %E% SMI" + +name="fbt" parent="pseudo" instance=0; diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.c b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.c new file mode 100644 index 00000000..38be2233 --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.c @@ -0,0 +1,557 @@ +/* + * CDDL HEADER START + * + * The contents of this file are subject to the terms of the + * Common Development and Distribution License (the "License"). + * You may not use this file except in compliance with the License. + * + * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE + * or http://www.opensolaris.org/os/licensing. + * See the License for the specific language governing permissions + * and limitations under the License. + * + * When distributing Covered Code, include this CDDL HEADER in each + * file and include the License file at usr/src/OPENSOLARIS.LICENSE. + * If applicable, add the following below this CDDL HEADER, with the + * fields enclosed by brackets "[]" replaced with your own identifying + * information: Portions Copyright [yyyy] [name of copyright owner] + * + * CDDL HEADER END + */ +/* + * Copyright 2009 Sun Microsystems, Inc. All rights reserved. + * Use is subject to license terms. + */ + + +#include <sys/modctl.h> +#include <sys/sunddi.h> +#include <sys/dtrace.h> +#include <sys/kobj.h> +#include <sys/stat.h> +#include <sys/conf.h> +#include <vm/seg_kmem.h> +#include <sys/stack.h> +#include <sys/frame.h> +#include <sys/dtrace_impl.h> +#include <sys/cmn_err.h> +#include <sys/sysmacros.h> +#include <sys/privregs.h> +#include <sys/sdt_impl.h> + +#define SDT_PATCHVAL 0xf0 +#define SDT_ADDR2NDX(addr) ((((uintptr_t)(addr)) >> 4) & sdt_probetab_mask) +#define SDT_PROBETAB_SIZE 0x1000 /* 4k entries -- 16K total */ + +static dev_info_t *sdt_devi; +static int sdt_verbose = 0; +static sdt_probe_t **sdt_probetab; +static int sdt_probetab_size; +static int sdt_probetab_mask; + +/*ARGSUSED*/ +static int +sdt_invop(uintptr_t addr, uintptr_t *stack, uintptr_t eax) +{ + uintptr_t stack0, stack1, stack2, stack3, stack4; + int i = 0; + sdt_probe_t *sdt = sdt_probetab[SDT_ADDR2NDX(addr)]; + +#ifdef __amd64 + /* + * On amd64, stack[0] contains the dereferenced stack pointer, + * stack[1] contains savfp, stack[2] contains savpc. We want + * to step over these entries. + */ + i += 3; +#endif + + for (; sdt != NULL; sdt = sdt->sdp_hashnext) { + if ((uintptr_t)sdt->sdp_patchpoint == addr) { + /* + * When accessing the arguments on the stack, we must + * protect against accessing beyond the stack. We can + * safely set NOFAULT here -- we know that interrupts + * are already disabled. + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + stack0 = stack[i++]; + stack1 = stack[i++]; + stack2 = stack[i++]; + stack3 = stack[i++]; + stack4 = stack[i++]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT | + CPU_DTRACE_BADADDR); + + dtrace_probe(sdt->sdp_id, stack0, stack1, + stack2, stack3, stack4); + + return (DTRACE_INVOP_NOP); + } + } + + return (0); +} + +/*ARGSUSED*/ +static void +sdt_provide_module(void *arg, struct modctl *ctl) +{ + struct module *mp = ctl->mod_mp; + char *modname = ctl->mod_modname; + sdt_probedesc_t *sdpd; + sdt_probe_t *sdp, *old; + sdt_provider_t *prov; + int len; + + /* + * One for all, and all for one: if we haven't yet registered all of + * our providers, we'll refuse to provide anything. + */ + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (prov->sdtp_id == DTRACE_PROVNONE) + return; + } + + if (mp->sdt_nprobes != 0 || (sdpd = mp->sdt_probes) == NULL) + return; + + for (sdpd = mp->sdt_probes; sdpd != NULL; sdpd = sdpd->sdpd_next) { + char *name = sdpd->sdpd_name, *func, *nname; + int i, j; + sdt_provider_t *prov; + ulong_t offs; + dtrace_id_t id; + + for (prov = sdt_providers; prov->sdtp_prefix != NULL; prov++) { + char *prefix = prov->sdtp_prefix; + + if (strncmp(name, prefix, strlen(prefix)) == 0) { + name += strlen(prefix); + break; + } + } + + nname = kmem_alloc(len = strlen(name) + 1, KM_SLEEP); + + for (i = 0, j = 0; name[j] != '\0'; i++) { + if (name[j] == '_' && name[j + 1] == '_') { + nname[i] = '-'; + j += 2; + } else { + nname[i] = name[j++]; + } + } + + nname[i] = '\0'; + + sdp = kmem_zalloc(sizeof (sdt_probe_t), KM_SLEEP); + sdp->sdp_loadcnt = ctl->mod_loadcnt; + sdp->sdp_ctl = ctl; + sdp->sdp_name = nname; + sdp->sdp_namelen = len; + sdp->sdp_provider = prov; + + func = kobj_searchsym(mp, sdpd->sdpd_offset, &offs); + + if (func == NULL) + func = "<unknown>"; + + /* + * We have our provider. Now create the probe. + */ + if ((id = dtrace_probe_lookup(prov->sdtp_id, modname, + func, nname)) != DTRACE_IDNONE) { + old = dtrace_probe_arg(prov->sdtp_id, id); + ASSERT(old != NULL); + + sdp->sdp_next = old->sdp_next; + sdp->sdp_id = id; + old->sdp_next = sdp; + } else { + sdp->sdp_id = dtrace_probe_create(prov->sdtp_id, + modname, func, nname, 3, sdp); + + mp->sdt_nprobes++; + } + + sdp->sdp_hashnext = + sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)]; + sdt_probetab[SDT_ADDR2NDX(sdpd->sdpd_offset)] = sdp; + + sdp->sdp_patchval = SDT_PATCHVAL; + sdp->sdp_patchpoint = (uint8_t *)sdpd->sdpd_offset; + sdp->sdp_savedval = *sdp->sdp_patchpoint; + } +} + +/*ARGSUSED*/ +static void +sdt_destroy(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg, *old, *last, *hash; + struct modctl *ctl = sdp->sdp_ctl; + int ndx; + + if (ctl != NULL && ctl->mod_loadcnt == sdp->sdp_loadcnt) { + if ((ctl->mod_loadcnt == sdp->sdp_loadcnt && + ctl->mod_loaded)) { + ((struct module *)(ctl->mod_mp))->sdt_nprobes--; + } + } + + while (sdp != NULL) { + old = sdp; + + /* + * Now we need to remove this probe from the sdt_probetab. + */ + ndx = SDT_ADDR2NDX(sdp->sdp_patchpoint); + last = NULL; + hash = sdt_probetab[ndx]; + + while (hash != sdp) { + ASSERT(hash != NULL); + last = hash; + hash = hash->sdp_hashnext; + } + + if (last != NULL) { + last->sdp_hashnext = sdp->sdp_hashnext; + } else { + sdt_probetab[ndx] = sdp->sdp_hashnext; + } + + kmem_free(sdp->sdp_name, sdp->sdp_namelen); + sdp = sdp->sdp_next; + kmem_free(old, sizeof (sdt_probe_t)); + } +} + +/*ARGSUSED*/ +static int +sdt_enable(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg; + struct modctl *ctl = sdp->sdp_ctl; + + ctl->mod_nenabled++; + + /* + * If this module has disappeared since we discovered its probes, + * refuse to enable it. + */ + if (!ctl->mod_loaded) { + if (sdt_verbose) { + cmn_err(CE_NOTE, "sdt is failing for probe %s " + "(module %s unloaded)", + sdp->sdp_name, ctl->mod_modname); + } + goto err; + } + + /* + * Now check that our modctl has the expected load count. If it + * doesn't, this module must have been unloaded and reloaded -- and + * we're not going to touch it. + */ + if (ctl->mod_loadcnt != sdp->sdp_loadcnt) { + if (sdt_verbose) { + cmn_err(CE_NOTE, "sdt is failing for probe %s " + "(module %s reloaded)", + sdp->sdp_name, ctl->mod_modname); + } + goto err; + } + + while (sdp != NULL) { + *sdp->sdp_patchpoint = sdp->sdp_patchval; + sdp = sdp->sdp_next; + } +err: + return (0); +} + +/*ARGSUSED*/ +static void +sdt_disable(void *arg, dtrace_id_t id, void *parg) +{ + sdt_probe_t *sdp = parg; + struct modctl *ctl = sdp->sdp_ctl; + + ctl->mod_nenabled--; + + if (!ctl->mod_loaded || ctl->mod_loadcnt != sdp->sdp_loadcnt) + goto err; + + while (sdp != NULL) { + *sdp->sdp_patchpoint = sdp->sdp_savedval; + sdp = sdp->sdp_next; + } + +err: + ; +} + +/*ARGSUSED*/ +uint64_t +sdt_getarg(void *arg, dtrace_id_t id, void *parg, int argno, int aframes) +{ + uintptr_t val; + struct frame *fp = (struct frame *)dtrace_getfp(); + uintptr_t *stack; + int i; +#if defined(__amd64) + /* + * A total of 6 arguments are passed via registers; any argument with + * index of 5 or lower is therefore in a register. + */ + int inreg = 5; +#endif + + for (i = 1; i <= aframes; i++) { + fp = (struct frame *)(fp->fr_savfp); + + if (fp->fr_savpc == (pc_t)dtrace_invop_callsite) { +#if !defined(__amd64) + /* + * If we pass through the invalid op handler, we will + * use the pointer that it passed to the stack as the + * second argument to dtrace_invop() as the pointer to + * the stack. + */ + stack = ((uintptr_t **)&fp[1])[1]; +#else + /* + * In the case of amd64, we will use the pointer to the + * regs structure that was pushed when we took the + * trap. To get this structure, we must increment + * beyond the frame structure. If the argument that + * we're seeking is passed on the stack, we'll pull + * the true stack pointer out of the saved registers + * and decrement our argument by the number of + * arguments passed in registers; if the argument + * we're seeking is passed in regsiters, we can just + * load it directly. + */ + struct regs *rp = (struct regs *)((uintptr_t)&fp[1] + + sizeof (uintptr_t)); + + if (argno <= inreg) { + stack = (uintptr_t *)&rp->r_rdi; + } else { + stack = (uintptr_t *)(rp->r_rsp); + argno -= (inreg + 1); + } +#endif + goto load; + } + } + + /* + * We know that we did not come through a trap to get into + * dtrace_probe() -- the provider simply called dtrace_probe() + * directly. As this is the case, we need to shift the argument + * that we're looking for: the probe ID is the first argument to + * dtrace_probe(), so the argument n will actually be found where + * one would expect to find argument (n + 1). + */ + argno++; + +#if defined(__amd64) + if (argno <= inreg) { + /* + * This shouldn't happen. If the argument is passed in a + * register then it should have been, well, passed in a + * register... + */ + DTRACE_CPUFLAG_SET(CPU_DTRACE_ILLOP); + return (0); + } + + argno -= (inreg + 1); +#endif + stack = (uintptr_t *)&fp[1]; + +load: + DTRACE_CPUFLAG_SET(CPU_DTRACE_NOFAULT); + val = stack[argno]; + DTRACE_CPUFLAG_CLEAR(CPU_DTRACE_NOFAULT); + + return (val); +} + +static dtrace_pops_t sdt_pops = { + NULL, + sdt_provide_module, + sdt_enable, + sdt_disable, + NULL, + NULL, + sdt_getargdesc, + sdt_getarg, + NULL, + sdt_destroy +}; + +/*ARGSUSED*/ +static int +sdt_attach(dev_info_t *devi, ddi_attach_cmd_t cmd) +{ + sdt_provider_t *prov; + + if (ddi_create_minor_node(devi, "sdt", S_IFCHR, + 0, DDI_PSEUDO, NULL) == DDI_FAILURE) { + cmn_err(CE_NOTE, "/dev/sdt couldn't create minor node"); + ddi_remove_minor_node(devi, NULL); + return (DDI_FAILURE); + } + + ddi_report_dev(devi); + sdt_devi = devi; + + if (sdt_probetab_size == 0) + sdt_probetab_size = SDT_PROBETAB_SIZE; + + sdt_probetab_mask = sdt_probetab_size - 1; + sdt_probetab = + kmem_zalloc(sdt_probetab_size * sizeof (sdt_probe_t *), KM_SLEEP); + dtrace_invop_add(sdt_invop); + + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (dtrace_register(prov->sdtp_name, prov->sdtp_attr, + DTRACE_PRIV_KERNEL, NULL, + &sdt_pops, prov, &prov->sdtp_id) != 0) { + cmn_err(CE_WARN, "failed to register sdt provider %s", + prov->sdtp_name); + } + } + + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +sdt_detach(dev_info_t *dip, ddi_detach_cmd_t cmd) +{ + sdt_provider_t *prov; + + switch (cmd) { + case DDI_DETACH: + break; + + case DDI_SUSPEND: + return (DDI_SUCCESS); + + default: + return (DDI_FAILURE); + } + + for (prov = sdt_providers; prov->sdtp_name != NULL; prov++) { + if (prov->sdtp_id != DTRACE_PROVNONE) { + if (dtrace_unregister(prov->sdtp_id) != 0) + return (DDI_FAILURE); + + prov->sdtp_id = DTRACE_PROVNONE; + } + } + + dtrace_invop_remove(sdt_invop); + kmem_free(sdt_probetab, sdt_probetab_size * sizeof (sdt_probe_t *)); + + return (DDI_SUCCESS); +} + +/*ARGSUSED*/ +static int +sdt_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result) +{ + int error; + + switch (infocmd) { + case DDI_INFO_DEVT2DEVINFO: + *result = (void *)sdt_devi; + error = DDI_SUCCESS; + break; + case DDI_INFO_DEVT2INSTANCE: + *result = (void *)0; + error = DDI_SUCCESS; + break; + default: + error = DDI_FAILURE; + } + return (error); +} + +/*ARGSUSED*/ +static int +sdt_open(dev_t *devp, int flag, int otyp, cred_t *cred_p) +{ + return (0); +} + +static struct cb_ops sdt_cb_ops = { + sdt_open, /* open */ + nodev, /* close */ + nulldev, /* strategy */ + nulldev, /* print */ + nodev, /* dump */ + nodev, /* read */ + nodev, /* write */ + nodev, /* ioctl */ + nodev, /* devmap */ + nodev, /* mmap */ + nodev, /* segmap */ + nochpoll, /* poll */ + ddi_prop_op, /* cb_prop_op */ + 0, /* streamtab */ + D_NEW | D_MP /* Driver compatibility flag */ +}; + +static struct dev_ops sdt_ops = { + DEVO_REV, /* devo_rev, */ + 0, /* refcnt */ + sdt_info, /* get_dev_info */ + nulldev, /* identify */ + nulldev, /* probe */ + sdt_attach, /* attach */ + sdt_detach, /* detach */ + nodev, /* reset */ + &sdt_cb_ops, /* driver operations */ + NULL, /* bus operations */ + nodev, /* dev power */ + ddi_quiesce_not_needed, /* quiesce */ +}; + +/* + * Module linkage information for the kernel. + */ +static struct modldrv modldrv = { + &mod_driverops, /* module type (this is a pseudo driver) */ + "Statically Defined Tracing", /* name of module */ + &sdt_ops, /* driver ops */ +}; + +static struct modlinkage modlinkage = { + MODREV_1, + (void *)&modldrv, + NULL +}; + +int +_init(void) +{ + return (mod_install(&modlinkage)); +} + +int +_info(struct modinfo *modinfop) +{ + return (mod_info(&modlinkage, modinfop)); +} + +int +_fini(void) +{ + return (mod_remove(&modlinkage)); +} diff --git a/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.conf b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.conf new file mode 100644 index 00000000..bde2149c --- /dev/null +++ b/src/VBox/ExtPacks/VBoxDTrace/onnv/uts/intel/dtrace/sdt.conf @@ -0,0 +1,28 @@ +# +# CDDL HEADER START +# +# The contents of this file are subject to the terms of the +# Common Development and Distribution License, Version 1.0 only +# (the "License"). You may not use this file except in compliance +# with the License. +# +# You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE +# or http://www.opensolaris.org/os/licensing. +# See the License for the specific language governing permissions +# and limitations under the License. +# +# When distributing Covered Code, include this CDDL HEADER in each +# file and include the License file at usr/src/OPENSOLARIS.LICENSE. +# If applicable, add the following below this CDDL HEADER, with the +# fields enclosed by brackets "[]" replaced with your own identifying +# information: Portions Copyright [yyyy] [name of copyright owner] +# +# CDDL HEADER END +# +# +# Copyright 2003 Sun Microsystems, Inc. All rights reserved. +# Use is subject to license terms. +# +#pragma ident "%Z%%M% %I% %E% SMI" + +name="sdt" parent="pseudo" instance=0; |