diff options
Diffstat (limited to 'collectors/ebpf.plugin/ebpf_fd.c')
| -rw-r--r-- | collectors/ebpf.plugin/ebpf_fd.c | 419 |
1 files changed, 371 insertions, 48 deletions
diff --git a/collectors/ebpf.plugin/ebpf_fd.c b/collectors/ebpf.plugin/ebpf_fd.c index b4e577dad..30b7f22ce 100644 --- a/collectors/ebpf.plugin/ebpf_fd.c +++ b/collectors/ebpf.plugin/ebpf_fd.c @@ -29,15 +29,301 @@ struct config fd_config = { .first_section = NULL, .last_section = NULL, .mutex .index = {.avl_tree = { .root = NULL, .compar = appconfig_section_compare }, .rwlock = AVL_LOCK_INITIALIZER } }; -struct netdata_static_thread fd_thread = {"FD KERNEL", NULL, NULL, 1, NULL, - NULL, NULL}; -static enum ebpf_threads_status ebpf_fd_exited = NETDATA_THREAD_EBPF_RUNNING; +struct netdata_static_thread fd_thread = {"FD KERNEL", + .config_section = NULL, + .config_name = NULL, + .env_name = NULL, + .enabled = 1, + .thread = NULL, + .init_routine = NULL, + .start_routine = NULL}; + static netdata_idx_t fd_hash_values[NETDATA_FD_COUNTER]; static netdata_idx_t *fd_values = NULL; netdata_fd_stat_t *fd_vector = NULL; netdata_fd_stat_t **fd_pid = NULL; +netdata_ebpf_targets_t fd_targets[] = { {.name = "open", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = "close", .mode = EBPF_LOAD_TRAMPOLINE}, + {.name = NULL, .mode = EBPF_LOAD_TRAMPOLINE}}; + +#ifdef LIBBPF_MAJOR_VERSION +#include "includes/fd.skel.h" // BTF code + +static struct fd_bpf *bpf_obj = NULL; + +/** + * Disable probe + * + * Disable all probes to use exclusively another method. + * + * @param obj is the main structure for bpf objects +*/ +static inline void ebpf_fd_disable_probes(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sys_open_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_sys_open_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_kprobe, false); + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } +} + +/* + * Disable specific probe + * + * Disable probes according the kernel version + * + * @param obj is the main structure for bpf objects + */ +static inline void ebpf_disable_specific_probes(struct fd_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_kprobe, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_close_fd_kretprobe, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_kprobe, false); + } +} + +/* + * Disable trampoline + * + * Disable all trampoline to use exclusively another method. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_disable_trampoline(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_sys_open_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_sys_open_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); +} + +/* + * Disable specific trampoline + * + * Disable trampoline according to kernel version. + * + * @param obj is the main structure for bpf objects. + */ +static inline void ebpf_disable_specific_trampoline(struct fd_bpf *obj) +{ + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_autoload(obj->progs.netdata___close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata___close_fd_fexit, false); + } else { + bpf_program__set_autoload(obj->progs.netdata_close_fd_fentry, false); + bpf_program__set_autoload(obj->progs.netdata_close_fd_fexit, false); + } +} + +/** + * Set trampoline target + * + * Set the targets we will monitor. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_set_trampoline_target(struct fd_bpf *obj) +{ + bpf_program__set_attach_target(obj->progs.netdata_sys_open_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + bpf_program__set_attach_target(obj->progs.netdata_sys_open_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + bpf_program__set_attach_target(obj->progs.netdata_release_task_fd_fentry, 0, EBPF_COMMON_FNCT_CLEAN_UP); + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + bpf_program__set_attach_target( + obj->progs.netdata_close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + bpf_program__set_attach_target(obj->progs.netdata_close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + } else { + bpf_program__set_attach_target( + obj->progs.netdata___close_fd_fentry, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + bpf_program__set_attach_target( + obj->progs.netdata___close_fd_fexit, 0, fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + } +} + +/** + * Mount Attach Probe + * + * Attach probes to target + * + * @param obj is the main structure for bpf objects. + * + * @return It returns 0 on success and -1 otherwise. + */ +static int ebpf_fd_attach_probe(struct fd_bpf *obj) +{ + obj->links.netdata_sys_open_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_sys_open_kprobe, false, + fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + int ret = libbpf_get_error(obj->links.netdata_sys_open_kprobe); + if (ret) + return -1; + + obj->links.netdata_sys_open_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_sys_open_kretprobe, true, + fd_targets[NETDATA_FD_SYSCALL_OPEN].name); + ret = libbpf_get_error(obj->links.netdata_sys_open_kretprobe); + if (ret) + return -1; + + obj->links.netdata_release_task_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_release_task_fd_kprobe, + false, + EBPF_COMMON_FNCT_CLEAN_UP); + ret = libbpf_get_error(obj->links.netdata_release_task_fd_kprobe); + if (ret) + return -1; + + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + obj->links.netdata_close_fd_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata_close_fd_kretprobe, true, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata_close_fd_kretprobe); + if (ret) + return -1; + + obj->links.netdata_close_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata_close_fd_kprobe, false, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata_close_fd_kprobe); + if (ret) + return -1; + } else { + obj->links.netdata___close_fd_kretprobe = bpf_program__attach_kprobe(obj->progs.netdata___close_fd_kretprobe, + true, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata___close_fd_kretprobe); + if (ret) + return -1; + + obj->links.netdata___close_fd_kprobe = bpf_program__attach_kprobe(obj->progs.netdata___close_fd_kprobe, + false, + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name); + ret = libbpf_get_error(obj->links.netdata___close_fd_kprobe); + if (ret) + return -1; + } + + return 0; +} + +/** + * Set target values + * + * Set pointers used to laod data. + */ +static void ebpf_fd_set_target_values() +{ + static char *close_targets[] = {"close_fd", "__close_fd"}; + static char *open_targets[] = {"do_sys_openat2", "do_sys_open"}; + if (running_on_kernel >= NETDATA_EBPF_KERNEL_5_11) { + fd_targets[NETDATA_FD_SYSCALL_OPEN].name = open_targets[0]; + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name = close_targets[0]; + } else { + fd_targets[NETDATA_FD_SYSCALL_OPEN].name = open_targets[1]; + fd_targets[NETDATA_FD_SYSCALL_CLOSE].name = close_targets[1]; + } +} + +/** + * Set hash tables + * + * Set the values for maps according the value given by kernel. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_fd_set_hash_tables(struct fd_bpf *obj) +{ + fd_maps[NETDATA_FD_GLOBAL_STATS].map_fd = bpf_map__fd(obj->maps.tbl_fd_global); + fd_maps[NETDATA_FD_PID_STATS].map_fd = bpf_map__fd(obj->maps.tbl_fd_pid); + fd_maps[NETDATA_FD_CONTROLLER].map_fd = bpf_map__fd(obj->maps.fd_ctrl); +} + +/** + * Adjust Map Size + * + * Resize maps according input from users. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + */ +static void ebpf_fd_adjust_map_size(struct fd_bpf *obj, ebpf_module_t *em) +{ + ebpf_update_map_size(obj->maps.tbl_fd_pid, &fd_maps[NETDATA_FD_PID_STATS], + em, bpf_map__name(obj->maps.tbl_fd_pid)); +} + +/** + * Disable Release Task + * + * Disable release task when apps is not enabled. + * + * @param obj is the main structure for bpf objects. + */ +static void ebpf_fd_disable_release_task(struct fd_bpf *obj) +{ + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_kprobe, false); + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); +} + +/** + * Load and attach + * + * Load and attach the eBPF code in kernel. + * + * @param obj is the main structure for bpf objects. + * @param em structure with configuration + * + * @return it returns 0 on succes and -1 otherwise + */ +static inline int ebpf_fd_load_and_attach(struct fd_bpf *obj, ebpf_module_t *em) +{ + netdata_ebpf_targets_t *mt = em->targets; + netdata_ebpf_program_loaded_t test = mt[NETDATA_FD_SYSCALL_OPEN].mode; + + ebpf_fd_set_target_values(); + if (test == EBPF_LOAD_TRAMPOLINE) { + ebpf_fd_disable_probes(obj); + ebpf_disable_specific_trampoline(obj); + + ebpf_set_trampoline_target(obj); + // TODO: Remove this in next PR, because this specific trampoline has an error. + bpf_program__set_autoload(obj->progs.netdata_release_task_fd_fentry, false); + } else { + ebpf_disable_trampoline(obj); + ebpf_disable_specific_probes(obj); + } + + ebpf_fd_adjust_map_size(obj, em); + + if (!em->apps_charts && !em->cgroup_charts) + ebpf_fd_disable_release_task(obj); + + int ret = fd_bpf__load(obj); + if (ret) { + return ret; + } + + ret = (test == EBPF_LOAD_TRAMPOLINE) ? fd_bpf__attach(obj) : ebpf_fd_attach_probe(obj); + if (!ret) { + ebpf_fd_set_hash_tables(obj); + + ebpf_update_controller(fd_maps[NETDATA_CACHESTAT_CTRL].map_fd, em); + } + + return ret; +} +#endif + /***************************************************************** * * FUNCTIONS TO CLOSE THE THREAD @@ -45,6 +331,38 @@ netdata_fd_stat_t **fd_pid = NULL; *****************************************************************/ /** + * FD Free + * + * Cleanup variables after child threads to stop + * + * @param ptr thread data. + */ +static void ebpf_fd_free(ebpf_module_t *em) +{ + pthread_mutex_lock(&ebpf_exit_cleanup); + if (em->thread->enabled == NETDATA_THREAD_EBPF_RUNNING) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPING; + pthread_mutex_unlock(&ebpf_exit_cleanup); + return; + } + pthread_mutex_unlock(&ebpf_exit_cleanup); + + ebpf_cleanup_publish_syscall(fd_publish_aggregated); + freez(fd_thread.thread); + freez(fd_values); + freez(fd_vector); + +#ifdef LIBBPF_MAJOR_VERSION + if (bpf_obj) + fd_bpf__destroy(bpf_obj); +#endif + + pthread_mutex_lock(&ebpf_exit_cleanup); + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; + pthread_mutex_unlock(&ebpf_exit_cleanup); +} + +/** * FD Exit * * Cancel child thread and exit. @@ -54,12 +372,8 @@ netdata_fd_stat_t **fd_pid = NULL; static void ebpf_fd_exit(void *ptr) { ebpf_module_t *em = (ebpf_module_t *)ptr; - if (!em->enabled) { - em->enabled = NETDATA_MAIN_THREAD_EXITED; - return; - } - - ebpf_fd_exited = NETDATA_THREAD_EBPF_STOPPING; + netdata_thread_cancel(*fd_thread.thread); + ebpf_fd_free(em); } /** @@ -70,16 +384,7 @@ static void ebpf_fd_exit(void *ptr) static void ebpf_fd_cleanup(void *ptr) { ebpf_module_t *em = (ebpf_module_t *)ptr; - if (ebpf_fd_exited != NETDATA_THREAD_EBPF_STOPPED) - return; - - ebpf_cleanup_publish_syscall(fd_publish_aggregated); - freez(fd_thread.thread); - freez(fd_values); - freez(fd_vector); - - fd_thread.enabled = NETDATA_MAIN_THREAD_EXITED; - em->enabled = NETDATA_MAIN_THREAD_EXITED; + ebpf_fd_free(em); } /***************************************************************** @@ -153,17 +458,12 @@ void *ebpf_fd_read_hash(void *ptr) ebpf_module_t *em = (ebpf_module_t *)ptr; usec_t step = NETDATA_FD_SLEEP_MS * em->update_every; - while (ebpf_fd_exited == NETDATA_THREAD_EBPF_RUNNING) { - usec_t dt = heartbeat_next(&hb, step); - (void)dt; - if (ebpf_fd_exited == NETDATA_THREAD_EBPF_STOPPING) - break; + while (!ebpf_exit_plugin) { + (void)heartbeat_next(&hb, step); read_global_table(); } - ebpf_fd_exited = NETDATA_THREAD_EBPF_STOPPED; - netdata_thread_cleanup_pop(1); return NULL; } @@ -549,20 +849,15 @@ static void ebpf_create_systemd_fd_charts(ebpf_module_t *em) * Send collected data to Netdata. * * @param em the main collector structure - * - * @return It returns the status for chart creation, if it is necessary to remove a specific dimension zero is returned - * otherwise function returns 1 to avoid chart recreation */ -static int ebpf_send_systemd_fd_charts(ebpf_module_t *em) +static void ebpf_send_systemd_fd_charts(ebpf_module_t *em) { - int ret = 1; ebpf_cgroup_target_t *ect; write_begin_chart(NETDATA_SERVICE_FAMILY, NETDATA_SYSCALL_APPS_FILE_OPEN); for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { if (unlikely(ect->systemd) && unlikely(ect->updated)) { write_chart_dimension(ect->name, ect->publish_systemd_fd.open_call); - } else if (unlikely(ect->systemd)) - ret = 0; + } } write_end_chart(); @@ -593,8 +888,6 @@ static int ebpf_send_systemd_fd_charts(ebpf_module_t *em) } write_end_chart(); } - - return ret; } /** @@ -615,13 +908,11 @@ static void ebpf_fd_send_cgroup_data(ebpf_module_t *em) int has_systemd = shm_ebpf_cgroup.header->systemd_enabled; if (has_systemd) { - static int systemd_charts = 0; - if (!systemd_charts) { + if (send_cgroup_chart) { ebpf_create_systemd_fd_charts(em); - systemd_charts = 1; } - systemd_charts = ebpf_send_systemd_fd_charts(em); + ebpf_send_systemd_fd_charts(em); } for (ect = ebpf_cgroup_pids; ect ; ect = ect->next) { @@ -810,6 +1101,40 @@ static void ebpf_fd_allocate_global_vectors(int apps) fd_values = callocz((size_t)ebpf_nprocs, sizeof(netdata_idx_t)); } +/* + * Load BPF + * + * Load BPF files. + * + * @param em the structure with configuration + */ +static int ebpf_fd_load_bpf(ebpf_module_t *em) +{ + int ret = 0; + ebpf_adjust_apps_cgroup(em, em->targets[NETDATA_FD_SYSCALL_OPEN].mode); + if (em->load & EBPF_LOAD_LEGACY) { + em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); + if (!em->probe_links) { + em->enabled = CONFIG_BOOLEAN_NO; + ret = -1; + } + } +#ifdef LIBBPF_MAJOR_VERSION + else { + bpf_obj = fd_bpf__open(); + if (!bpf_obj) + ret = -1; + else + ret = ebpf_fd_load_and_attach(bpf_obj, em); + } +#endif + + if (ret) + error("%s %s", EBPF_DEFAULT_ERROR_MSG, em->thread_name); + + return ret; +} + /** * Directory Cache thread * @@ -826,17 +1151,16 @@ void *ebpf_fd_thread(void *ptr) ebpf_module_t *em = (ebpf_module_t *)ptr; em->maps = fd_maps; - if (!em->enabled) +#ifdef LIBBPF_MAJOR_VERSION + ebpf_adjust_thread_load(em, default_btf); +#endif + if (ebpf_fd_load_bpf(em)) { + em->thread->enabled = NETDATA_THREAD_EBPF_STOPPED; goto endfd; + } ebpf_fd_allocate_global_vectors(em->apps_charts); - em->probe_links = ebpf_load_program(ebpf_plugin_dir, em, running_on_kernel, isrh, &em->objects); - if (!em->probe_links) { - em->enabled = CONFIG_BOOLEAN_NO; - goto endfd; - } - int algorithms[NETDATA_FD_SYSCALL_END] = { NETDATA_EBPF_INCREMENTAL_IDX, NETDATA_EBPF_INCREMENTAL_IDX }; @@ -852,8 +1176,7 @@ void *ebpf_fd_thread(void *ptr) fd_collector(em); endfd: - if (!em->enabled) - ebpf_update_disabled_plugin_stats(em); + ebpf_update_disabled_plugin_stats(em); netdata_thread_cleanup_pop(1); return NULL; |