From 4f5791ebd03eaec1c7da0865a383175b05102712 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Sun, 5 May 2024 19:47:29 +0200 Subject: Adding upstream version 2:4.17.12+dfsg. Signed-off-by: Daniel Baumann --- source4/dsdb/kcc/kcc_periodic.c | 825 ++++++++++++++++++++++++++++++++++++++++ 1 file changed, 825 insertions(+) create mode 100644 source4/dsdb/kcc/kcc_periodic.c (limited to 'source4/dsdb/kcc/kcc_periodic.c') diff --git a/source4/dsdb/kcc/kcc_periodic.c b/source4/dsdb/kcc/kcc_periodic.c new file mode 100644 index 0000000..428f21d --- /dev/null +++ b/source4/dsdb/kcc/kcc_periodic.c @@ -0,0 +1,825 @@ +/* + Unix SMB/CIFS Implementation. + KCC service periodic handling + + Copyright (C) Andrew Tridgell 2009 + based on repl service code + + This program is free software; you can redistribute it and/or modify + it under the terms of the GNU General Public License as published by + the Free Software Foundation; either version 3 of the License, or + (at your option) any later version. + + This program is distributed in the hope that it will be useful, + but WITHOUT ANY WARRANTY; without even the implied warranty of + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + GNU General Public License for more details. + + You should have received a copy of the GNU General Public License + along with this program. If not, see . + +*/ + +#include "includes.h" +#include "lib/events/events.h" +#include "dsdb/samdb/samdb.h" +#include "auth/auth.h" +#include "samba/service.h" +#include "lib/messaging/irpc.h" +#include "dsdb/kcc/kcc_connection.h" +#include "dsdb/kcc/kcc_service.h" +#include +#include "../lib/util/dlinklist.h" +#include "librpc/gen_ndr/ndr_misc.h" +#include "librpc/gen_ndr/ndr_drsuapi.h" +#include "librpc/gen_ndr/ndr_drsblobs.h" +#include "librpc/gen_ndr/ndr_irpc_c.h" +#include "param/param.h" +#include "dsdb/common/util.h" + +/* + * see if two repsFromToBlob blobs are for the same source DSA + */ +static bool kccsrv_same_source_dsa(struct repsFromToBlob *r1, struct repsFromToBlob *r2) +{ + return GUID_equal(&r1->ctr.ctr1.source_dsa_obj_guid, + &r2->ctr.ctr1.source_dsa_obj_guid); +} + +/* + * see if a repsFromToBlob is in a list + */ +static bool reps_in_list(struct repsFromToBlob *r, struct repsFromToBlob *reps, uint32_t count) +{ + uint32_t i; + for (i=0; ictr.ctr1; + struct GUID invocation_id = r1->source_dsa_invocation_id; + unsigned int i, j; + TALLOC_CTX *tmp_ctx; + + /* we are expecting only version 1 */ + SMB_ASSERT(r->version == 1); + + tmp_ctx = talloc_new(p); + if (!tmp_ctx) { + return false; + } + + for (i=0; icount; i++) { + struct ldb_message *msg = res->msgs[i]; + struct ldb_message_element *el; + struct ldb_dn *dn; + + struct GUID id2 = samdb_result_guid(msg, "invocationID"); + if (GUID_all_zero(&id2) || + !GUID_equal(&invocation_id, &id2)) { + continue; + } + + el = ldb_msg_find_element(msg, "msDS-hasMasterNCs"); + if (!el || el->num_values == 0) { + el = ldb_msg_find_element(msg, "hasMasterNCs"); + if (!el || el->num_values == 0) { + continue; + } + } + for (j=0; jnum_values; j++) { + dn = ldb_dn_from_ldb_val(tmp_ctx, service->samdb, &el->values[j]); + if (!ldb_dn_validate(dn)) { + talloc_free(dn); + continue; + } + if (ldb_dn_compare(dn, p->dn) == 0) { + DEBUG(5,("%s %s match on %s in %s\n", + r1->other_info->dns_name, + el->name, + ldb_dn_get_linearized(dn), + ldb_dn_get_linearized(msg->dn))); + talloc_free(tmp_ctx); + return true; + } + talloc_free(dn); + } + } + talloc_free(tmp_ctx); + return false; +} + +struct kccsrv_notify_drepl_server_state { + struct dreplsrv_refresh r; +}; + +static void kccsrv_notify_drepl_server_done(struct tevent_req *subreq); + +/** + * Force dreplsrv to update its state as topology is changed + */ +static void kccsrv_notify_drepl_server(struct kccsrv_service *s, + TALLOC_CTX *mem_ctx) +{ + struct kccsrv_notify_drepl_server_state *state; + struct dcerpc_binding_handle *irpc_handle; + struct tevent_req *subreq; + + state = talloc_zero(s, struct kccsrv_notify_drepl_server_state); + if (state == NULL) { + return; + } + + irpc_handle = irpc_binding_handle_by_name(state, s->task->msg_ctx, + "dreplsrv", &ndr_table_irpc); + if (irpc_handle == NULL) { + /* dreplsrv is not running yet */ + TALLOC_FREE(state); + return; + } + + subreq = dcerpc_dreplsrv_refresh_r_send(state, s->task->event_ctx, + irpc_handle, &state->r); + if (subreq == NULL) { + TALLOC_FREE(state); + return; + } + tevent_req_set_callback(subreq, kccsrv_notify_drepl_server_done, state); +} + +static void kccsrv_notify_drepl_server_done(struct tevent_req *subreq) +{ + struct kccsrv_notify_drepl_server_state *state = + tevent_req_callback_data(subreq, + struct kccsrv_notify_drepl_server_state); + + dcerpc_dreplsrv_refresh_r_recv(subreq, state); + TALLOC_FREE(subreq); + + /* we don't care about errors */ + TALLOC_FREE(state); +} + +uint32_t kccsrv_replica_flags(struct kccsrv_service *s) +{ + if (s->am_rodc) { + return DRSUAPI_DRS_INIT_SYNC | + DRSUAPI_DRS_PER_SYNC | + DRSUAPI_DRS_ADD_REF | + DRSUAPI_DRS_SPECIAL_SECRET_PROCESSING | + DRSUAPI_DRS_NONGC_RO_REP; + } + return DRSUAPI_DRS_INIT_SYNC | + DRSUAPI_DRS_PER_SYNC | + DRSUAPI_DRS_ADD_REF | + DRSUAPI_DRS_WRIT_REP; +} + +/* + * add any missing repsFrom structures to our partitions + */ +NTSTATUS kccsrv_add_repsFrom(struct kccsrv_service *s, TALLOC_CTX *mem_ctx, + struct repsFromToBlob *reps, uint32_t count, + struct ldb_result *res) +{ + struct dsdb_ldb_dn_list_node *p; + bool notify_dreplsrv = false; + uint32_t replica_flags = kccsrv_replica_flags(s); + + /* update the repsFrom on all partitions */ + for (p=s->partitions; p; p=p->next) { + struct repsFromToBlob *our_reps; + uint32_t our_count; + WERROR werr; + uint32_t i, j; + bool modified = false; + + werr = dsdb_loadreps(s->samdb, mem_ctx, p->dn, "repsFrom", &our_reps, &our_count); + if (!W_ERROR_IS_OK(werr)) { + DEBUG(0,(__location__ ": Failed to load repsFrom from %s - %s\n", + ldb_dn_get_linearized(p->dn), ldb_errstring(s->samdb))); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + /* see if the entry already exists */ + for (i=0; idns_name)); + } + } + + /* remove any stale ones */ + for (i=0; idns_name)); + memmove(&our_reps[i], &our_reps[i+1], (our_count-(i+1))*sizeof(our_reps[0])); + our_count--; + i--; + modified = true; + } + } + + if (modified) { + werr = dsdb_savereps(s->samdb, mem_ctx, p->dn, "repsFrom", our_reps, our_count); + if (!W_ERROR_IS_OK(werr)) { + DEBUG(0,(__location__ ": Failed to save repsFrom to %s - %s\n", + ldb_dn_get_linearized(p->dn), ldb_errstring(s->samdb))); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + /* dreplsrv should refresh its state */ + notify_dreplsrv = true; + } + + /* remove stale repsTo entries */ + modified = false; + werr = dsdb_loadreps(s->samdb, mem_ctx, p->dn, "repsTo", &our_reps, &our_count); + if (!W_ERROR_IS_OK(werr)) { + DEBUG(0,(__location__ ": Failed to load repsTo from %s - %s\n", + ldb_dn_get_linearized(p->dn), ldb_errstring(s->samdb))); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + /* remove any stale ones */ + for (i=0; idns_name)); + memmove(&our_reps[i], &our_reps[i+1], (our_count-(i+1))*sizeof(our_reps[0])); + our_count--; + i--; + modified = true; + } + } + + if (modified) { + werr = dsdb_savereps(s->samdb, mem_ctx, p->dn, "repsTo", our_reps, our_count); + if (!W_ERROR_IS_OK(werr)) { + DEBUG(0,(__location__ ": Failed to save repsTo to %s - %s\n", + ldb_dn_get_linearized(p->dn), ldb_errstring(s->samdb))); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + /* dreplsrv should refresh its state */ + notify_dreplsrv = true; + } + } + + /* notify dreplsrv toplogy has changed */ + if (notify_dreplsrv) { + kccsrv_notify_drepl_server(s, mem_ctx); + } + + return NT_STATUS_OK; + +} + + +/* + form a unique list of DNs from a search result and a given set of attributes + */ +static int kccsrv_dn_list(struct ldb_context *ldb, struct ldb_result *res, + TALLOC_CTX *mem_ctx, + const char **attrs, + struct ldb_dn ***dn_list, int *dn_count) +{ + int i; + struct ldb_dn **nc_list = NULL; + int nc_count = 0; + + nc_list = talloc_array(mem_ctx, struct ldb_dn *, 0); + if (nc_list == NULL) { + return LDB_ERR_OPERATIONS_ERROR; + } + + /* gather up a list of all NCs in this forest */ + for (i=0; icount; i++) { + struct ldb_message *msg = res->msgs[i]; + int j; + for (j=0; attrs[j]; j++) { + struct ldb_message_element *el; + int k; + + el = ldb_msg_find_element(msg, attrs[j]); + if (el == NULL) continue; + for (k=0; knum_values; k++) { + struct ldb_dn *dn; + dn = ldb_dn_from_ldb_val(nc_list, ldb, &el->values[k]); + if (dn != NULL) { + int l; + for (l=0; lsamdb, res, tmp_ctx, attrs1, &nc_list, &nc_count); + if (ret != LDB_SUCCESS) { + DEBUG(1,("Failed to get NC list for GC update - %s\n", ldb_errstring(s->samdb))); + talloc_free(tmp_ctx); + return ret; + } + + /* get a list of what NCs we are already replicating */ + ret = dsdb_search_dn(s->samdb, tmp_ctx, &res2, samdb_ntds_settings_dn(s->samdb, tmp_ctx), attrs2, 0); + if (ret != LDB_SUCCESS) { + DEBUG(1,("Failed to get our NC list attributes for GC update - %s\n", ldb_errstring(s->samdb))); + talloc_free(tmp_ctx); + return ret; + } + + ret = kccsrv_dn_list(s->samdb, res2, tmp_ctx, attrs2, &our_nc_list, &our_nc_count); + if (ret != LDB_SUCCESS) { + DEBUG(1,("Failed to get our NC list for GC update - %s\n", ldb_errstring(s->samdb))); + talloc_free(tmp_ctx); + return ret; + } + + msg = ldb_msg_new(tmp_ctx); + if (msg == NULL) { + talloc_free(tmp_ctx); + return LDB_ERR_OPERATIONS_ERROR; + } + msg->dn = res2->msgs[0]->dn; + + /* see if we are missing any */ + for (i=0; inum_elements == 0) { + /* none to add */ + talloc_free(tmp_ctx); + return LDB_SUCCESS; + } + + if (s->am_rodc) { + DEBUG(5, ("%d partial replica should be added but we are RODC so we skip\n", msg->num_elements)); + talloc_free(tmp_ctx); + return LDB_SUCCESS; + } + + msg->elements[0].flags = LDB_FLAG_MOD_ADD; + + ret = dsdb_modify(s->samdb, msg, 0); + if (ret != LDB_SUCCESS) { + DEBUG(0,("Failed to add hasPartialReplicaNCs - %s\n", + ldb_errstring(s->samdb))); + } + + talloc_free(tmp_ctx); + return ret; +} + + +/* + this is the core of our initial simple KCC + We just add a repsFrom entry for all DCs we find that have nTDSDSA + objects, except for ourselves + */ +NTSTATUS kccsrv_simple_update(struct kccsrv_service *s, TALLOC_CTX *mem_ctx) +{ + struct ldb_result *res; + unsigned int i; + int ret; + const char *attrs[] = { "objectGUID", "invocationID", "msDS-hasMasterNCs", "hasMasterNCs", "msDS-HasDomainNCs", NULL }; + struct repsFromToBlob *reps = NULL; + uint32_t count = 0; + struct kcc_connection_list *ntds_conn, *dsa_conn; + + ret = dsdb_search(s->samdb, mem_ctx, &res, s->config_dn, LDB_SCOPE_SUBTREE, + attrs, DSDB_SEARCH_SHOW_EXTENDED_DN, "objectClass=nTDSDSA"); + if (ret != LDB_SUCCESS) { + DEBUG(0,(__location__ ": Failed nTDSDSA search - %s\n", ldb_errstring(s->samdb))); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + if (samdb_is_gc(s->samdb)) { + kccsrv_gc_update(s, res); + } + + /* get the current list of connections */ + ntds_conn = kccsrv_find_connections(s, mem_ctx); + + dsa_conn = talloc_zero(mem_ctx, struct kcc_connection_list); + + for (i=0; icount; i++) { + struct repsFromTo1 *r1; + struct GUID ntds_guid, invocation_id; + + ntds_guid = samdb_result_guid(res->msgs[i], "objectGUID"); + if (GUID_equal(&ntds_guid, &s->ntds_guid)) { + /* don't replicate with ourselves */ + continue; + } + + invocation_id = samdb_result_guid(res->msgs[i], "invocationID"); + + reps = talloc_realloc(mem_ctx, reps, struct repsFromToBlob, count+1); + NT_STATUS_HAVE_NO_MEMORY(reps); + + ZERO_STRUCT(reps[count]); + reps[count].version = 1; + r1 = &reps[count].ctr.ctr1; + + r1->other_info = talloc_zero(reps, struct repsFromTo1OtherInfo); + r1->other_info->dns_name = samdb_ntds_msdcs_dns_name(s->samdb, reps, &ntds_guid); + r1->source_dsa_obj_guid = ntds_guid; + r1->source_dsa_invocation_id = invocation_id; + r1->replica_flags = kccsrv_replica_flags(s); + memset(r1->schedule, 0x11, sizeof(r1->schedule)); + + dsa_conn->servers = talloc_realloc(dsa_conn, dsa_conn->servers, + struct kcc_connection, + dsa_conn->count + 1); + NT_STATUS_HAVE_NO_MEMORY(dsa_conn->servers); + dsa_conn->servers[dsa_conn->count].dsa_guid = r1->source_dsa_obj_guid; + dsa_conn->count++; + + count++; + } + + kccsrv_apply_connections(s, ntds_conn, dsa_conn); + + return kccsrv_add_repsFrom(s, mem_ctx, reps, count, res); +} + + +static void kccsrv_periodic_run(struct kccsrv_service *service); + +static void kccsrv_periodic_handler_te(struct tevent_context *ev, struct tevent_timer *te, + struct timeval t, void *ptr) +{ + struct kccsrv_service *service = talloc_get_type(ptr, struct kccsrv_service); + WERROR status; + + service->periodic.te = NULL; + + kccsrv_periodic_run(service); + + status = kccsrv_periodic_schedule(service, service->periodic.interval); + if (!W_ERROR_IS_OK(status)) { + task_server_terminate(service->task, win_errstr(status), true); + return; + } +} + +WERROR kccsrv_periodic_schedule(struct kccsrv_service *service, uint32_t next_interval) +{ + TALLOC_CTX *tmp_mem; + struct tevent_timer *new_te; + struct timeval next_time; + + /* prevent looping */ + if (next_interval == 0) next_interval = 1; + + next_time = timeval_current_ofs(next_interval, 50); + + if (service->periodic.te) { + /* + * if the timestamp of the new event is higher, + * as current next we don't need to reschedule + */ + if (timeval_compare(&next_time, &service->periodic.next_event) > 0) { + return WERR_OK; + } + } + + /* reset the next scheduled timestamp */ + service->periodic.next_event = next_time; + + new_te = tevent_add_timer(service->task->event_ctx, service, + service->periodic.next_event, + kccsrv_periodic_handler_te, service); + W_ERROR_HAVE_NO_MEMORY(new_te); + + tmp_mem = talloc_new(service); + DEBUG(4,("kccsrv_periodic_schedule(%u) %sscheduled for: %s\n", + next_interval, + (service->periodic.te?"re":""), + nt_time_string(tmp_mem, timeval_to_nttime(&next_time)))); + talloc_free(tmp_mem); + + talloc_free(service->periodic.te); + service->periodic.te = new_te; + + return WERR_OK; +} + +/* + * Check to see if any dns entries need scavenging. This only occurs if aging + * is enabled in general ("zone scavenging" lpcfg) and on the zone + * (zone->fAging is true). + */ +static NTSTATUS kccsrv_dns_zone_scavenging( + struct kccsrv_service *s, + TALLOC_CTX *mem_ctx) +{ + + time_t current_time = time(NULL); + time_t dns_scavenge_interval; + NTSTATUS status; + char *error_string = NULL; + + /* + * Only perform zone scavenging if it's been enabled. + * (it still might be disabled on all zones). + */ + if (!lpcfg_dns_zone_scavenging(s->task->lp_ctx)) { + DBG_INFO("DNS scavenging not enabled"); + return NT_STATUS_OK; + } + + dns_scavenge_interval = lpcfg_parm_int(s->task->lp_ctx, + NULL, + "dnsserver", + "scavenging_interval", + 2 * 60 * 60); + if ((current_time - s->last_dns_scavenge) > dns_scavenge_interval) { + s->last_dns_scavenge = current_time; + status = dns_tombstone_records(mem_ctx, s->samdb, + &error_string); + if (!NT_STATUS_IS_OK(status)) { + const char *err = NULL; + if (error_string != NULL) { + err = error_string; + } else { + err = nt_errstr(status); + } + DBG_ERR("DNS record scavenging process failed: %s", + err); + return status; + } + } + DBG_INFO("Successfully tombstoned stale DNS records"); + return NT_STATUS_OK; +} +/* + * check to see if any dns tombstones should be deleted. This is not optional + * ([MS-DNSP] "DsTombstoneInterval") -- stale tombstones are useless clutter. + * + * Windows does it daily at 2am; we do it roughly daily at an uncontrolled + * time. + */ +static NTSTATUS kccsrv_dns_zone_tombstone_deletion(struct kccsrv_service *s, + TALLOC_CTX *mem_ctx) +{ + time_t current_time = time(NULL); + NTSTATUS status; + char *error_string = NULL; + time_t dns_collection_interval = + lpcfg_parm_int(s->task->lp_ctx, + NULL, + "dnsserver", + "tombstone_collection_interval", + 24 * 60 * 60); + + if ((current_time - s->last_dns_tombstone_collection) > + dns_collection_interval) { + s->last_dns_tombstone_collection = current_time; + status = dns_delete_tombstones(mem_ctx, s->samdb, + &error_string); + if (!NT_STATUS_IS_OK(status)) { + const char *err = NULL; + if (error_string != NULL) { + err = error_string; + } else { + err = nt_errstr(status); + } + DBG_ERR("DNS tombstone deletion failed: %s", err); + return status; + } + } + DBG_INFO("Successfully deleted DNS tombstones"); + return NT_STATUS_OK; +} + +/* + check to see if any deleted objects need scavenging + */ +static NTSTATUS kccsrv_check_deleted(struct kccsrv_service *s, TALLOC_CTX *mem_ctx) +{ + time_t current_time = time(NULL); + time_t interval = lpcfg_parm_int( + s->task->lp_ctx, NULL, "kccsrv", "check_deleted_interval", 86400); + uint32_t tombstoneLifetime; + int ret; + unsigned int num_objects_removed = 0; + unsigned int num_links_removed = 0; + NTSTATUS status; + char *error_string = NULL; + + if (current_time - s->last_deleted_check < interval) { + return NT_STATUS_OK; + } + + ret = dsdb_tombstone_lifetime(s->samdb, &tombstoneLifetime); + if (ret != LDB_SUCCESS) { + DEBUG(1,(__location__ ": Failed to get tombstone lifetime\n")); + return NT_STATUS_INTERNAL_DB_CORRUPTION; + } + + s->last_deleted_check = current_time; + + status = dsdb_garbage_collect_tombstones(mem_ctx, s->samdb, + s->partitions, + current_time, tombstoneLifetime, + &num_objects_removed, + &num_links_removed, + &error_string); + + if (NT_STATUS_IS_OK(status)) { + DEBUG(5, ("garbage_collect_tombstones: Removed %u tombstone objects " + "and %u tombstone links successfully\n", + num_objects_removed, num_links_removed)); + } else { + DEBUG(2, ("garbage_collect_tombstones: Failure removing tombstone " + "objects and links after removing %u tombstone objects " + "and %u tombstone links successfully: %s\n", + num_objects_removed, num_links_removed, + error_string ? error_string : nt_errstr(status))); + } + return status; +} + +static void kccsrv_periodic_run(struct kccsrv_service *service) +{ + TALLOC_CTX *mem_ctx; + NTSTATUS status; + + DEBUG(4,("kccsrv_periodic_run(): update\n")); + + mem_ctx = talloc_new(service); + + if (service->samba_kcc_code) + status = kccsrv_samba_kcc(service); + else { + status = kccsrv_simple_update(service, mem_ctx); + if (!NT_STATUS_IS_OK(status)) + DEBUG(0,("kccsrv_simple_update failed - %s\n", + nt_errstr(status))); + } + + status = kccsrv_check_deleted(service, mem_ctx); + if (!NT_STATUS_IS_OK(status)) { + DEBUG(0,("kccsrv_check_deleted failed - %s\n", nt_errstr(status))); + } + status = kccsrv_dns_zone_scavenging(service, mem_ctx); + if (!NT_STATUS_IS_OK(status)) { + DBG_ERR("kccsrv_dns_zone_aging failed - %s\n", + nt_errstr(status)); + } + status = kccsrv_dns_zone_tombstone_deletion(service, mem_ctx); + if (!NT_STATUS_IS_OK(status)) { + DBG_ERR("kccsrv_dns_zone_tombstone_scavenging failed - %s\n", + nt_errstr(status)); + } + talloc_free(mem_ctx); +} + +/* Called when samba_kcc script has finished + */ +static void samba_kcc_done(struct tevent_req *subreq) +{ + struct kccsrv_service *service = + tevent_req_callback_data(subreq, struct kccsrv_service); + int rc; + int sys_errno; + + service->periodic.subreq = NULL; + + rc = samba_runcmd_recv(subreq, &sys_errno); + TALLOC_FREE(subreq); + + if (rc != 0) + service->periodic.status = + map_nt_error_from_unix_common(sys_errno); + else + service->periodic.status = NT_STATUS_OK; + + if (!NT_STATUS_IS_OK(service->periodic.status)) + DEBUG(0,(__location__ ": Failed samba_kcc - %s\n", + nt_errstr(service->periodic.status))); + else + DEBUG(3,("Completed samba_kcc OK\n")); +} + +/* Invocation of the samba_kcc python script for replication + * topology generation. + */ +NTSTATUS kccsrv_samba_kcc(struct kccsrv_service *service) +{ + NTSTATUS status = NT_STATUS_OK; + const char * const *samba_kcc_command = + lpcfg_samba_kcc_command(service->task->lp_ctx); + + /* kill any existing child */ + TALLOC_FREE(service->periodic.subreq); + + DEBUG(2, ("Calling samba_kcc script\n")); + service->periodic.subreq = samba_runcmd_send(service, + service->task->event_ctx, + timeval_current_ofs(40, 0), + 2, 0, samba_kcc_command, NULL); + + if (service->periodic.subreq == NULL) { + status = NT_STATUS_NO_MEMORY; + goto xerror; + } + tevent_req_set_callback(service->periodic.subreq, + samba_kcc_done, service); + +xerror: + if (!NT_STATUS_IS_OK(status)) + DEBUG(0,(__location__ ": failed - %s\n", nt_errstr(status))); + return status; +} -- cgit v1.2.3