restarter.c revision 53f3aea0943e36e5fed2615ad5f9fd1f17de51d2
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * CDDL HEADER START
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * The contents of this file are subject to the terms of the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Common Development and Distribution License (the "License").
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * You may not use this file except in compliance with the License.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * See the License for the specific language governing permissions
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * and limitations under the License.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * When distributing Covered Code, include this CDDL HEADER in each
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If applicable, add the following below this CDDL HEADER, with the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * fields enclosed by brackets "[]" replaced with your own identifying
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * information: Portions Copyright [yyyy] [name of copyright owner]
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * CDDL HEADER END
66cd0f60c3182913d379abb730ae755bf6367126Kacheong Poon * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
633fc3a6eed35d918db16925b7048d7a2e28064aSebastien Roy * Use is subject to license terms.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * restarter.c - service manipulation
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This component manages services whose restarter is svc.startd, the standard
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * restarter. It translates restarter protocol events from the graph engine
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * into actions on processes, as a delegated restarter would do.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * The master restarter manages a number of always-running threads:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - restarter event thread: events from the graph engine
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - timeout thread: thread to fire queued timeouts
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - contract thread: thread to handle contract events
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - wait thread: thread to handle wait-based services
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * The other threads are created as-needed:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - per-instance method threads
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - per-instance event processing threads
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * The interaction of all threads must result in the following conditions
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * being satisfied (on a per-instance basis):
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - restarter events must be processed in order
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - method execution must be serialized
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - instance delete must be held until outstanding methods are complete
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - contract events shouldn't be processed while a method is running
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - timeouts should fire even when a method is running
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Service instances are represented by restarter_inst_t's and are kept in the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_list list.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Service States
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * The current state of a service instance is kept in
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * restarter_inst_t->ri_i.i_state. If transition to a new state could take
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * some time, then before we effect the transition we set
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * restarter_inst_t->ri_i.i_next_state to the target state, and afterwards we
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * rotate i_next_state to i_state and set i_next_state to
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * RESTARTER_STATE_NONE. So usually i_next_state is _NONE when ri_lock is not
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * held. The exception is when we launch methods, which are done with
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * a separate thread. To keep any other threads from grabbing ri_lock before
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * method_thread() does, we set ri_method_thread to the thread id of the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * method thread, and when it is nonzero any thread with a different thread id
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * waits on ri_method_cv.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Method execution is serialized by blocking on ri_method_cv in
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * inst_lookup_by_id() and waiting for a 0 value of ri_method_thread. This
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * also prevents the instance structure from being deleted until all
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * outstanding operations such as method_thread() have finished.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Lock ordering:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * dgraph_lock [can be held when taking:]
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * dictionary->dict_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * st->st_load_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * wait_info_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ru->restarter_update_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * restarter_queue->rpeq_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_list.ril_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * inst->ri_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * st->st_configd_live_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_list.ril_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * graph_queue->gpeq_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * gu->gu_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * st->st_configd_live_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * dictionary->dict_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * inst->ri_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * graph_queue->gpeq_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * gu->gu_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * tu->tu_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * tq->tq_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * inst->ri_queue_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * wait_info_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * bp->cb_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * single_user_thread_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * wait_info_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * gu_freeze_lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * logbuf_mutex nests inside pretty much everything.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstatic uu_list_pool_t *restarter_instance_pool;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstatic restarter_instance_list_t instance_list;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_instance_compare(const void *lc_arg, const void *rc_arg,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon int lc_id = ((const restarter_inst_t *)lc_arg)->ri_id;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon !pthread_equal(inst->ri_method_thread, pthread_self())) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_warn("%s:%d: Bad flags 0x%x.\n", __FILE__, __LINE__, flags);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Fails with ECONNABORTED or ECANCELED.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Pooncheck_contract(restarter_inst_t *inst, boolean_t primary,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon ctidp = primary ? &inst->ri_i.i_primary_ctid :
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon fd = contract_open(*ctidp, NULL, "status", O_RDONLY);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = restarter_remove_contract(scf_inst, *ctidp, primary ?
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon RESTARTER_CONTRACT_PRIMARY : RESTARTER_CONTRACT_TRANSIENT);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_die("Repository backend access denied.\n");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_error(LOG_INFO, "Could not remove unusable contract id %ld "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "for %s from repository.\n", *ctidp, inst->ri_i.i_fmri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstatic int stop_instance(scf_handle_t *, restarter_inst_t *, stop_cause_t);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * int restarter_insert_inst(scf_handle_t *, char *)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If the inst is already in the restarter list, return its id. If the inst
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * is not in the restarter list, initialize a restarter_inst_t, initialize its
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * states, insert it into the list, and return 0.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ENOENT - name is not in the repository
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_insert_inst(scf_handle_t *h, const char *name)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * We don't use inst_lookup_by_name() here because we want the lookup
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * & insert to be atomic.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst = uu_list_find(instance_list.ril_instance_list, &id, NULL,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Allocate an instance */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst = startd_zalloc(sizeof (restarter_inst_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_utmpx_prefix = startd_alloc(max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_fmri = startd_alloc(strlen(name) + 1);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) strcpy((char *)inst->ri_i.i_fmri, name);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_queue = startd_list_create(restarter_queue_pool, inst, 0);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * id shouldn't be -1 since we use the same dictionary as graph.c, but
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * just in case.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_id = (id != -1 ? id : dict_insert(name));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon special_online_hooks_get(name, &inst->ri_pre_online_hook,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon &inst->ri_post_online_hook, &inst->ri_post_offline_hook);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_C_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (scf_handle_decode_fmri(h, name, NULL, scf_svc, scf_inst, NULL,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If there's no running snapshot, then we execute using the editing
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * snapshot. Pending snapshots will be taken later.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if ((scf_service_get_name(scf_svc, svc_name, max_scf_name_size) < 0) ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (scf_instance_get_name(scf_inst, inst_name, max_scf_name_size) <
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) snprintf(logfilebuf, PATH_MAX, "%s:%s", svc_name, inst_name);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (*c == '/')
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) snprintf(inst->ri_logstem, PATH_MAX, "%s%s", logfilebuf,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If the restarter group is missing, use uninit/none. Otherwise,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * we're probably being restarted & don't want to mess up the states
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * that are there.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = scf_instance_get_pg(scf_inst, SCF_PG_RESTARTER, pg);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (r != 0) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This shouldn't happen since the graph engine should
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * have initialized the state to uninitialized/none if
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * there was no restarter pg. In case somebody
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * deleted it, though....
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = libscf_read_states(pg, &state, &next_state);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (r != 0) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Force next_state to _NONE since we
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * don't look for method processes.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Inform the restarter of our state without
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * changing the STIME in the repository.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (libscf_get_startd_properties(scf_inst, snap, &inst->ri_flags,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This is odd, because the graph engine should have required
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the general property group. So we'll just use default
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * flags in anticipation of the graph engine sending us
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * REMOVE_INSTANCE when it finds out that the general property
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * group has been deleted.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (libscf_get_template_values(scf_inst, snap,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon &inst->ri_common_name, &inst->ri_C_common_name)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (libscf_read_method_ids(h, scf_inst, inst->ri_i.i_fmri,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon &inst->ri_i.i_primary_ctid, &inst->ri_i.i_transient_ctid,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon contract_hash_store(inst->ri_i.i_primary_ctid, inst->ri_id);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (check_contract(inst, B_TRUE, scf_inst)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (check_contract(inst, B_FALSE, scf_inst)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* No more failures we live through, so add it to the list. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_init(&inst->ri_lock, &mutex_attrs);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_init(&inst->ri_queue_lock, &mutex_attrs);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_init(&inst->ri_method_cv, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_node_init(inst, &inst->ri_link, restarter_instance_pool);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_insert(instance_list.ril_instance_list, inst, idx);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon ret = wait_register(start_pid, inst->ri_i.i_fmri, 0, 1);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Implication: if we can't reregister the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance, we will start another one. Two
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instances may or may not result in a resource
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: couldn't reregister %ld for wait\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Leading PID has exited.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, inst, state,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s is a %s-style service\n", name,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: inserted instance into restarter list\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free((void *)inst->ri_i.i_fmri, strlen(inst->ri_i.i_fmri) + 1);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_C_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_utmpx_prefix, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Must drop the instance lock so we can pick up the instance_list
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * lock & remove the instance.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon rip = uu_list_find(instance_list.ril_instance_list, &id, NULL, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_remove(instance_list.ril_instance_list, ri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: deleted instance from restarter list\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * We can lock the instance without holding the instance_list lock
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * since we removed the instance from the list.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon contract_hash_remove(ri->ri_i.i_primary_ctid);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon while (ri->ri_method_thread != 0 || ri->ri_method_waiters > 0)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_wait(&ri->ri_method_cv, &ri->ri_lock);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon while ((e = uu_list_teardown(ri->ri_queue, &cookie)) != NULL)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(e, sizeof (*e));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free((void *)ri->ri_i.i_fmri, strlen(ri->ri_i.i_fmri) + 1);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(ri->ri_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(ri->ri_C_common_name, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(ri->ri_utmpx_prefix, max_scf_value_size);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_destroy(&ri->ri_queue_lock);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_is_wait_style()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Returns 1 if the given instance is a "wait-style" service instance.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_WAIT);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_is_transient_style()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Returns 1 if the given instance is a transient service instance.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Pooninstance_is_transient_style(restarter_inst_t *inst)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon return ((inst->ri_flags & RINST_STYLE_MASK) == RINST_TRANSIENT);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance_in_transition()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Returns 1 if instance is in transition, 0 if not
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (inst->ri_i.i_next_state == RESTARTER_STATE_NONE)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * returns 1 if instance is already started, 0 if not
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (inst->ri_i.i_state == RESTARTER_STATE_ONLINE ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_state == RESTARTER_STATE_DEGRADED)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * 0 - success
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ECONNRESET - success, but h was rebound
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_instance_update_states(scf_handle_t *h, restarter_inst_t *ri,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon restarter_instance_state_t new_state_next, restarter_error_t err, char *aux)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon e = _restarter_commit_states(h, &ri->ri_i, new_state, new_state_next,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Like startd_alloc(). */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_error(LOG_NOTICE, "Could not commit state change for %s "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "to repository: %s.\n", ri->ri_i.i_fmri, strerror(e));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* FALLTHROUGH */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon states = startd_alloc(sizeof (protocol_states_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon graph_protocol_send_event(ri->ri_i.i_fmri, GRAPH_UPDATE_STATE_CHANGE,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_mark_pending_snapshot(const char *fmri, uint_t flag)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon assert(flag == RINST_RETAKE_RUNNING || flag == RINST_RETAKE_START);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_take_pending_snapshots(scf_handle_t *h)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon for (inst = uu_list_first(instance_list.ril_instance_list);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst = uu_list_next(instance_list.ril_instance_list, inst)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This is where we'd check inst->ri_method_thread and if it
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * were nonzero we'd wait in anticipation of another thread
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * executing a method for inst. Doing so with the instance_list
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * locked, though, leads to deadlock. Since taking a snapshot
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * during that window won't hurt anything, we'll just continue.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) libscf_fmri_get_instance(h, fmri, &sinst);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon rsnap = libscf_get_or_make_running_snapshot(sinst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (r = libscf_snapshots_poststart(h, fmri,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon/* ARGSUSED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * int stop_instance()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Stop the instance identified by the instance given as the second argument,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * for the cause stated.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * 0 - success
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * -1 - inst is in transition
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstop_instance(scf_handle_t *local_handle, restarter_inst_t *inst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon const char *cp;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon cp = "process received fatal signal from outside the service";
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon cp = "process killed due to uncorrectable hardware error";
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) fprintf(stderr, "Unknown cause %d at %s:%d.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Services in the disabled and maintenance state are ignored */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_state == RESTARTER_STATE_DISABLED) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: stop_instance -> is maint/disabled\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Already stopped instances are left alone */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "Restarter: %s is already stopped.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* requeue event by returning -1 */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Restarter: Not stopping %s, in transition.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(inst, B_TRUE, "Stopping because %s.", cp);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(re == RERR_FAULT ? LOG_INFO : LOG_DEBUG,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: Instance stopping because %s.\n", inst->ri_i.i_fmri, cp);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (instance_is_wait_style(inst) && cause == RSTOP_EXIT) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * No need to stop instance, as child has exited; remove
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * contract and move the instance to the offline state.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (err = restarter_instance_update_states(local_handle,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst, inst->ri_i.i_state, RESTARTER_STATE_OFFLINE, re,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon bad_error("restarter_instance_update_states", err);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) update_fault_count(inst, FAULT_COUNT_RESET);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (err = restarter_instance_update_states(local_handle,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst, inst->ri_i.i_next_state, RESTARTER_STATE_NONE, re,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon bad_error("restarter_instance_update_states", err);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon } else if (instance_is_wait_style(inst) && re == RERR_RESTART) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Stopping a wait service through means other than the pid
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * exiting should keep wait_thread() from restarting the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * service, by removing it from the wait list.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * We cannot remove it right now otherwise the process will
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * end up <defunct> so mark it to be ignored.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (err = restarter_instance_update_states(local_handle, inst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_state, inst->ri_i.i_enabled ? RESTARTER_STATE_OFFLINE :
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon bad_error("restarter_instance_update_states", err);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_method_thread = startd_thread_create(method_thread, info);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ENOENT - fmri is not in instance_list
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * 0 - success
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ECONNRESET - success, though handle was rebound
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * -1 - instance is in transition
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstop_instance_fmri(scf_handle_t *h, const char *fmri, uint_t flags)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonunmaintain_instance(scf_handle_t *h, restarter_inst_t *rip,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon const char *cp;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (rip->ri_i.i_state != RESTARTER_STATE_MAINT) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Ignoring maintenance off command because %s is not in the "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) fprintf(stderr, "Uncaught case for %d at %s:%d.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(rip, B_TRUE, "Leaving maintenance because %s.",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: Instance leaving maintenance because "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, rip, RESTARTER_STATE_UNINIT,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If we did ADMIN_MAINT_ON_IMMEDIATE, then there might still be
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * a primary contract.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon switch (r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Must have been deleted. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = restarter_remove_contract(inst, ctid, RESTARTER_CONTRACT_PRIMARY);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* NOTREACHED */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Could not remove contract id %lu for %s (%s).\n", ctid,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * enable_inst()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Set inst->ri_i.i_enabled. Expects 'e' to be _ENABLE, _DISABLE, or
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * _ADMIN_DISABLE. If the event is _ENABLE and inst is uninitialized or
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * disabled, move it to offline. If the event is _DISABLE or
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * _ADMIN_DISABLE, make sure inst will move to disabled.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * 0 - success
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ECONNRESET - h was rebound
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonenable_inst(scf_handle_t *h, restarter_inst_t *inst, restarter_event_type_t e)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon assert(e == RESTARTER_EVENT_TYPE_ADMIN_DISABLE ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * B_FALSE: Don't log an error if the log_instance()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * fails because it will fail on the miniroot before
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * install-discovery runs.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: Instance enabled.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, inst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon RESTARTER_STATE_OFFLINE, RESTARTER_STATE_NONE,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Not changing state of %s for enable command.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon return (r == ECONNRESET ? 0 : r);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_m_inst = safe_scf_instance_create(h);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* B_FALSE: See log_instance(..., "Enabled."); above */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: Instance disabled.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, inst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon RESTARTER_STATE_DISABLED, RESTARTER_STATE_NONE,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * We only want to pull the instance out of maintenance
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * if the disable is on adminstrative request. The
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * graph engine sends _DISABLE events whenever a
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * service isn't in the disabled state, and we don't
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * want to pull the service out of maintenance if,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * for example, it is there due to a dependency cycle.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon unmaintain_instance(h, inst, RUNMAINT_DISABLE);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) fprintf(stderr, "Restarter instance %s has "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "unknown state %d.\n", inst->ri_i.i_fmri, state);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstart_instance(scf_handle_t *local_handle, restarter_inst_t *inst)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: trying to start instance\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Services in the disabled and maintenance state are ignored */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (inst->ri_i.i_state == RESTARTER_STATE_MAINT ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_state == RESTARTER_STATE_DISABLED ||
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: start_instance -> is maint/disabled\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Already started instances are left alone */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: start_instance -> is already started\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: starting instance.\n", inst->ri_i.i_fmri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(local_handle, inst,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_i.i_state, RESTARTER_STATE_ONLINE, RERR_NONE, "none");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon inst->ri_method_thread = startd_thread_create(method_thread, info);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonevent_from_tty(scf_handle_t *h, restarter_inst_t *rip)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst))
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonmaintain_instance(scf_handle_t *h, restarter_inst_t *rip, int immediate,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon const char *aux)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(rip, B_TRUE, "Stopping for maintenance due to %s.", aux);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: stopping for maintenance due to %s.\n",
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson /* Services in the maintenance state are ignored */
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson if (rip->ri_i.i_state == RESTARTER_STATE_MAINT) {
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson "%s: maintain_instance -> is already in maintenance\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If aux state is "service_request" and
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson * restarter_actions/auxiliary_fmri property is set with a valid fmri,
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson * copy the fmri to restarter/auxiliary_fmri so svcs -x can use.
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson if (strcmp(aux, "service_request") == 0 && libscf_fmri_get_instance(h,
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson if (restarter_inst_validate_ractions_aux_fmri(scf_inst) == 0) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "restarter_inst_set_aux_fmri failed: ",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "restarter_inst_validate_ractions_aux_fmri "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "restarter_inst_reset_aux_fmri failed: ",
b1cd7879d8fc9fb80c59139db15960e8e6d6bf93Anders Persson rip->ri_m_inst = safe_scf_instance_create(h);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, rip,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon RESTARTER_STATE_MAINT, RESTARTER_STATE_NONE, RERR_RESTART,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, rip, rip->ri_i.i_state,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon RESTARTER_STATE_MAINT, RERR_NONE, (char *)aux);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon rip->ri_method_thread = startd_thread_create(method_thread, info);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrefresh_instance(scf_handle_t *h, restarter_inst_t *rip)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(rip, B_TRUE, "Rereading configuration.");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: rereading configuration.\n",
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson r = libscf_fmri_get_instance(h, rip->ri_i.i_fmri, &inst);
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson /* Must have been deleted. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = libscf_get_startd_properties(inst, snap, &rip->ri_flags,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s is a %s-style service\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon rip->ri_i.i_fmri, service_style(rip->ri_flags));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Succeed in anticipation of REMOVE_INSTANCE. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Refresh does not change the state. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) restarter_instance_update_states(h, rip,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon rip->ri_i.i_state, rip->ri_i.i_state, RERR_NONE, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonconst char *event_names[] = { "INVALID", "ADD_INSTANCE", "REMOVE_INSTANCE",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "ENABLE", "DISABLE", "ADMIN_DEGRADED", "ADMIN_REFRESH",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "ADMIN_RESTART", "ADMIN_MAINT_OFF", "ADMIN_MAINT_ON",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "ADMIN_MAINT_ON_IMMEDIATE", "STOP", "START", "DEPENDENCY_CYCLE",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * void *restarter_process_events()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Called in a separate thread to process the events on an instance's
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * queue. Empties the queue completely, and tries to keep the thread
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * around for a little while after the queue is empty to save on
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * startup costs.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* grab the queue lock */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon while ((event = uu_list_first(rip->ri_queue)) != NULL) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* drop the queue lock */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Grab the inst lock -- this waits until any outstanding
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * method finishes running.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Getting deleted in the middle isn't an error. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* process the event */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) stop_instance(h, inst, RSTOP_DEPENDENCY);
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson maintain_instance(h, inst, 0, "dependency_cycle");
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson case RESTARTER_EVENT_TYPE_INVALID_DEPENDENCY:
3e95bd4ab92abca814bd28e854607d1975c7dc88Anders Persson maintain_instance(h, inst, 0, "invalid_dependency");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "service_request");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "administrative_request");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "service_request");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "administrative_request");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s command (for %s) unimplemented.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon event_names[event->riq_type], inst->ri_i.i_fmri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Not restarting %s; not running.\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Stop the instance. If it can be restarted,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the graph engine will send a new event.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Aborting.\n", __FILE__, __LINE__, event->riq_type);
9cd928fe5e3ea4e05f64cfb380beb54b2623e7dcAlan Maguire /* grab the queue lock */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* delete the event */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(event, sizeof (restarter_instance_qentry_t));
e5e7971ff24064af923af2464e78fa6d2fb28deaErik Nordmark * Try to preserve the thread for a little while for future use.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_reltimedwait_np(&rip->ri_queue_cv,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon case RESTARTER_EVENT_TYPE_ADMIN_MAINT_ON_IMMEDIATE:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonrestarter_queue_event(restarter_inst_t *ri, restarter_protocol_event_t *e)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon qe = startd_zalloc(sizeof (restarter_instance_qentry_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_node_init(qe, &qe->riq_link, restarter_queue_pool);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon r = uu_list_insert_before(ri->ri_queue, NULL, qe);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * void *restarter_event_thread()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Handle incoming graph events by placing them on a per-instance
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * queue. We can't lock the main part of the instance structure, so
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * just modify the seprarately locked event queue portion.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This is a new thread, and thus, gets its own handle
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * to the repository.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /*CONSTCOND*/
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_wait(&ru->restarter_update_cv,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon while ((e = restarter_event_dequeue()) != NULL) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * ADD_INSTANCE is special: there's likely no
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance structure yet, so we need to handle the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * addition synchronously.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (restarter_insert_inst(h, e->rpe_inst) != 0)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Lookup the instance, locking only the event queue.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Can't grab ri_lock here because it might be held
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * by a long-running method.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Ignoring %s command for unknown service "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Keep ADMIN events from filling up the queue. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "queue overflow. Dropping administrative "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "%s: Instance event "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "queue overflow. Dropping administrative "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Now add the event to the instance queue. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Start a thread if one isn't already
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Signal the existing thread that there's
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * a new event.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Unreachable for now -- there's currently no graceful cleanup
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * called on exit().
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Since ri_lock isn't held by the contract id lookup, this
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * instance may have been restarted and now be in a new
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * contract, making the old contract no longer valid for this
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * void contract_action()
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Take action on contract events.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Pooncontract_action(scf_handle_t *h, restarter_inst_t *inst, ctid_t id,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If startd has stopped this contract, there is no need to
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * stop it again.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if ((type & (CT_PR_EV_EMPTY | CT_PR_EV_CORE | CT_PR_EV_SIGNAL
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * There shouldn't be other events, since that's not how we set
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the terms. Thus, just log an error and drive on.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: contract %ld received unexpected critical event "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * We ignore all events; if they impact the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * process we're monitoring, then the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * wait_thread will stop the instance.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "%s: ignoring contract event on wait-style service\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * A CT_PR_EV_EMPTY event is an RSTOP_EXIT request.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * void *restarter_contract_event_thread(void *)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Listens to the process contract bundle for critical events, taking action
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * on events from contracts we know we are responsible for.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Await graph load completion. That is, stop here, until we've scanned
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the repository for contract - instance associations.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon while (!(st->st_load_complete && st->st_load_instances == 0))
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_wait(&st->st_load_cv, &st->st_load_lock);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This is a new thread, and thus, gets its own handle
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * to the repository.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if ((local_handle = libscf_handle_create_bound(SCF_VERSION)) == NULL)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_die("Unable to bind a new repository handle: %s\n",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon fd = open64(CTFS_ROOT "/process/pbundle", O_RDONLY);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Make sure we get all events (including those generated by configd
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * before this thread was started).
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Error reading next contract event: %s",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Fetch cookie. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if ((sfd = contract_open(ctid, "process", "status", O_RDONLY))
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (err = ct_status_read(sfd, CTD_COMMON, &status)) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_WARNING, "Could not get status for "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "Received event %d for ctid %ld "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * svc.configd(1M) restart handling performed by the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * fork_configd_thread. We don't acknowledge, as that thread
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * will do so.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This can happen for two reasons:
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - method_run() has not yet stored the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the contract into the internal hash table.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * - we receive an EMPTY event for an abandoned
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If there is any contract in the process of
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * being stored into the hash table then re-read
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the event later.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Reset event %d for unknown "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* don't go too fast */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Do not call contract_to_inst() again if first
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * call succeeded.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * This can happen if we receive an EMPTY
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * event for an abandoned contract.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Received event %d for unknown contract id "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Received event %d for contract id "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon contract_action(local_handle, inst, ctid, type);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon efd = contract_open(ct_event_get_ctid(ev), "process", "ctl",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /*NOTREACHED*/
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Timeout queue, processed by restarter_timeouts_event_thread().
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poontypedef struct timeout_update {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poonstatic const char *timeout_ovr_svcs[] = {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/network/initial:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/network/service:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/rmtmpfiles:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/network/loopback:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/network/physical:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/device/local:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/metainit:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/filesystem/usr:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/filesystem/minimal:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc:/system/filesystem/local:default",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon for (i = 0; timeout_ovr_svcs[i] != NULL; ++i) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (strcmp(inst->ri_i.i_fmri, timeout_ovr_svcs[i]) == 0) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(inst, B_TRUE, "Timeout override by "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "svc.startd. Using infinite timeout.");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poontimeout_compare(const void *lc_arg, const void *rc_arg, void *private)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon hrtime_t t1 = ((const timeout_entry_t *)lc_arg)->te_timeout;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon hrtime_t t2 = ((const timeout_entry_t *)rc_arg)->te_timeout;
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon timeouts = startd_zalloc(sizeof (timeout_queue_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_init(&timeouts->tq_lock, &mutex_attrs);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon timeout_pool = startd_list_pool_create("timeouts",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon sizeof (timeout_entry_t), offsetof(timeout_entry_t, te_link),
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon timeouts->tq_list = startd_list_create(timeout_pool,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon tu = startd_zalloc(sizeof (timeout_update_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_init(&tu->tu_lock, &mutex_attrs);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poontimeout_insert(restarter_inst_t *inst, ctid_t cid, uint64_t timeout_sec)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * If we overflow LLONG_MAX, we're never timing out anyways, so
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * just return.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon if (timeout_sec >= (LLONG_MAX - now) / 1000000000LL) {
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance(inst, B_TRUE, "timeout_seconds too large, "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "treating as infinite.");
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* hrtime is in nanoseconds. Convert timeout_sec. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon entry = startd_alloc(sizeof (timeout_entry_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon entry->te_fmri = safe_strdup(inst->ri_i.i_fmri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon entry->te_logstem = safe_strdup(inst->ri_logstem);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* Insert the calculated timeout time onto the queue. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) uu_list_find(timeouts->tq_list, entry, NULL, &idx);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_node_init(entry, &entry->te_link, timeout_pool);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_insert(timeouts->tq_list, entry, idx);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poontimeout_remove(restarter_inst_t *inst, ctid_t cid)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon uu_list_remove(timeouts->tq_list, inst->ri_timeout);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon startd_free(inst->ri_timeout, sizeof (timeout_entry_t));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Walk through the (sorted) timeouts list. While the timeout
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * at the head of the list is <= the current time, kill the
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_WARNING, "%s: Method or service exit timed "
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "out. Killing contract %ld.\n", e->te_fmri, e->te_ctid);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_instance_fmri(e->te_fmri, e->te_logstem, B_TRUE,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "Method or service exit timed out. Killing contract %ld.",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) contract_kill(e->te_ctid, SIGKILL, e->te_fmri);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * void *restarter_timeouts_event_thread(void *)
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Responsible for monitoring the method timeouts. This thread must
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * be started before any methods are called.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * Timeouts are entered on a priority queue, which is processed by
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * this thread. As timeouts are specified in seconds, we'll do
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * the necessary processing every second, as long as the queue
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * is not empty.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /*CONSTCOND*/
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * As long as the timeout list isn't empty, process it
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon * every second.
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon /* The list is empty, wait until we have more timeouts. */
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_cond_wait(&tu->tu_cv, &tu->tu_lock);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) startd_thread_create(restarter_timeouts_event_thread, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) startd_thread_create(restarter_event_thread, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) startd_thread_create(restarter_contracts_event_thread, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) startd_thread_create(wait_thread, NULL);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon restarter_instance_pool = startd_list_pool_create("restarter_instances",
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon sizeof (restarter_inst_t), offsetof(restarter_inst_t,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon ri_link), restarter_instance_compare, UU_LIST_POOL_DEBUG);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) memset(&instance_list, 0, sizeof (instance_list));
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon (void) pthread_mutex_init(&instance_list.ril_lock, &mutex_attrs);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon instance_list.ril_instance_list = startd_list_create(
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon restarter_instance_pool, &instance_list, UU_LIST_SORTED);
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon restarter_queue_pool = startd_list_pool_create(
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon "restarter_instance_queue", sizeof (restarter_instance_qentry_t),
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon offsetof(restarter_instance_qentry_t, riq_link), NULL,
721fffe35d40e548a5a58dc53a2ec9c6762172d9Kacheong Poon log_framework(LOG_DEBUG, "Initialized restarter\n");