diff --git a/exec/main.c b/exec/main.c index 85c74ee7..0ca5634a 100644 --- a/exec/main.c +++ b/exec/main.c @@ -1,1474 +1,1407 @@ /* * Copyright (c) 2002-2006 MontaVista Software, Inc. * Copyright (c) 2006-2012 Red Hat, Inc. * * All rights reserved. * * Author: Steven Dake (sdake@redhat.com) * * This software licensed under BSD license, the text of which follows: * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * - Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * - Neither the name of the MontaVista Software, Inc. nor the names of its * contributors may be used to endorse or promote products derived from this * software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ /** * \mainpage Corosync * * This is the doxygen generated developer documentation for the Corosync * project. For more information about Corosync, please see the project * web site, corosync.org. * * \section license License * * This software licensed under BSD license, the text of which follows: * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * - Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * - Neither the name of the MontaVista Software, Inc. nor the names of its * contributors may be used to endorse or promote products derived from this * software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "quorum.h" #include "totemsrp.h" #include "logconfig.h" #include "totemconfig.h" #include "main.h" #include "sync.h" #include "timer.h" #include "util.h" #include "apidef.h" #include "service.h" #include "schedwrk.h" #ifdef HAVE_SMALL_MEMORY_FOOTPRINT #define IPC_LOGSYS_SIZE 1024*64 #else #define IPC_LOGSYS_SIZE 8192*128 #endif LOGSYS_DECLARE_SYSTEM ("corosync", LOGSYS_MODE_OUTPUT_STDERR | LOGSYS_MODE_OUTPUT_SYSLOG, LOG_DAEMON, LOG_INFO); LOGSYS_DECLARE_SUBSYS ("MAIN"); #define SERVER_BACKLOG 5 static int sched_priority = 0; static unsigned int service_count = 32; static struct totem_logging_configuration totem_logging_configuration; static struct corosync_api_v1 *api = NULL; static int sync_in_process = 1; static qb_loop_t *corosync_poll_handle; struct sched_param global_sched_param; static corosync_timer_handle_t corosync_stats_timer_handle; static const char *corosync_lock_file = LOCALSTATEDIR"/run/corosync.pid"; static int ip_version = AF_INET; qb_loop_t *cs_poll_handle_get (void) { return (corosync_poll_handle); } int cs_poll_dispatch_add (qb_loop_t * handle, int fd, int events, void *data, int (*dispatch_fn) (int fd, int revents, void *data)) { return qb_loop_poll_add(handle, QB_LOOP_MED, fd, events, data, dispatch_fn); } int cs_poll_dispatch_delete(qb_loop_t * handle, int fd) { return qb_loop_poll_del(handle, fd); } void corosync_state_dump (void) { int i; for (i = 0; i < SERVICES_COUNT_MAX; i++) { if (corosync_service[i] && corosync_service[i]->exec_dump_fn) { corosync_service[i]->exec_dump_fn (); } } } static void corosync_blackbox_write_to_file (void) { char fname[PATH_MAX]; char fdata_fname[PATH_MAX]; char time_str[PATH_MAX]; struct tm cur_time_tm; time_t cur_time_t; ssize_t res; cur_time_t = time(NULL); localtime_r(&cur_time_t, &cur_time_tm); strftime(time_str, PATH_MAX, "%Y-%m-%dT%H:%M:%S", &cur_time_tm); snprintf(fname, PATH_MAX, "%s/fdata-%s-%lld", get_run_dir(), time_str, (long long int)getpid()); if ((res = qb_log_blackbox_write_to_file(fname)) < 0) { LOGSYS_PERROR(-res, LOGSYS_LEVEL_ERROR, "Can't store blackbox file"); } snprintf(fdata_fname, sizeof(fdata_fname), "%s/fdata", get_run_dir()); unlink(fdata_fname); if (symlink(fname, fdata_fname) == -1) { log_printf(LOGSYS_LEVEL_ERROR, "Can't create symlink to '%s' for corosync blackbox file '%s'", fname, fdata_fname); } } static void unlink_all_completed (void) { api->timer_delete (corosync_stats_timer_handle); qb_loop_stop (corosync_poll_handle); icmap_fini(); } void corosync_shutdown_request (void) { corosync_service_unlink_all (api, unlink_all_completed); } static int32_t sig_diag_handler (int num, void *data) { corosync_state_dump (); return 0; } static int32_t sig_exit_handler (int num, void *data) { log_printf(LOGSYS_LEVEL_NOTICE, "Node was shut down by a signal"); corosync_service_unlink_all (api, unlink_all_completed); return 0; } static void sigsegv_handler (int num) { (void)signal (SIGSEGV, SIG_DFL); corosync_blackbox_write_to_file (); qb_log_fini(); raise (SIGSEGV); } /* * QB wrapper for real signal handler */ static int32_t sig_segv_handler (int num, void *data) { sigsegv_handler(num); return 0; } static void sigabrt_handler (int num) { (void)signal (SIGABRT, SIG_DFL); corosync_blackbox_write_to_file (); qb_log_fini(); raise (SIGABRT); } /* * QB wrapper for real signal handler */ static int32_t sig_abrt_handler (int num, void *data) { sigabrt_handler(num); return 0; } #define LOCALHOST_IP inet_addr("127.0.0.1") static void *corosync_group_handle; static struct totempg_group corosync_group = { .group = "a", .group_len = 1 }; static void serialize_lock (void) { } static void serialize_unlock (void) { } static void corosync_sync_completed (void) { log_printf (LOGSYS_LEVEL_NOTICE, "Completed service synchronization, ready to provide service."); sync_in_process = 0; cs_ipcs_sync_state_changed(sync_in_process); cs_ipc_allow_connections(1); /* * Inform totem to start using new message queue again */ totempg_trans_ack(); } static int corosync_sync_callbacks_retrieve ( int service_id, struct sync_callbacks *callbacks) { if (corosync_service[service_id] == NULL) { return (-1); } if (callbacks == NULL) { return (0); } callbacks->name = corosync_service[service_id]->name; callbacks->sync_init = corosync_service[service_id]->sync_init; callbacks->sync_process = corosync_service[service_id]->sync_process; callbacks->sync_activate = corosync_service[service_id]->sync_activate; callbacks->sync_abort = corosync_service[service_id]->sync_abort; return (0); } static struct memb_ring_id corosync_ring_id; static void member_object_joined (unsigned int nodeid) { char member_ip[ICMAP_KEYNAME_MAXLEN]; char member_join_count[ICMAP_KEYNAME_MAXLEN]; char member_status[ICMAP_KEYNAME_MAXLEN]; snprintf(member_ip, ICMAP_KEYNAME_MAXLEN, "runtime.totem.pg.mrp.srp.members.%u.ip", nodeid); snprintf(member_join_count, ICMAP_KEYNAME_MAXLEN, "runtime.totem.pg.mrp.srp.members.%u.join_count", nodeid); snprintf(member_status, ICMAP_KEYNAME_MAXLEN, "runtime.totem.pg.mrp.srp.members.%u.status", nodeid); if (icmap_get(member_ip, NULL, NULL, NULL) == CS_OK) { icmap_inc(member_join_count); icmap_set_string(member_status, "joined"); } else { icmap_set_string(member_ip, (char*)api->totem_ifaces_print (nodeid)); icmap_set_uint32(member_join_count, 1); icmap_set_string(member_status, "joined"); } log_printf (LOGSYS_LEVEL_DEBUG, "Member joined: %s", api->totem_ifaces_print (nodeid)); } static void member_object_left (unsigned int nodeid) { char member_status[ICMAP_KEYNAME_MAXLEN]; snprintf(member_status, ICMAP_KEYNAME_MAXLEN, "runtime.totem.pg.mrp.srp.members.%u.status", nodeid); icmap_set_string(member_status, "left"); log_printf (LOGSYS_LEVEL_DEBUG, "Member left: %s", api->totem_ifaces_print (nodeid)); } static void confchg_fn ( enum totem_configuration_type configuration_type, const unsigned int *member_list, size_t member_list_entries, const unsigned int *left_list, size_t left_list_entries, const unsigned int *joined_list, size_t joined_list_entries, const struct memb_ring_id *ring_id) { int i; int abort_activate = 0; if (sync_in_process == 1) { abort_activate = 1; } sync_in_process = 1; cs_ipcs_sync_state_changed(sync_in_process); memcpy (&corosync_ring_id, ring_id, sizeof (struct memb_ring_id)); for (i = 0; i < left_list_entries; i++) { member_object_left (left_list[i]); } for (i = 0; i < joined_list_entries; i++) { member_object_joined (joined_list[i]); } /* * Call configuration change for all services */ for (i = 0; i < service_count; i++) { if (corosync_service[i] && corosync_service[i]->confchg_fn) { corosync_service[i]->confchg_fn (configuration_type, member_list, member_list_entries, left_list, left_list_entries, joined_list, joined_list_entries, ring_id); } } if (abort_activate) { sync_abort (); } if (configuration_type == TOTEM_CONFIGURATION_TRANSITIONAL) { sync_save_transitional (member_list, member_list_entries, ring_id); } if (configuration_type == TOTEM_CONFIGURATION_REGULAR) { sync_start (member_list, member_list_entries, ring_id); } } static void priv_drop (void) { return; /* TODO: we are still not dropping privs */ } static void corosync_tty_detach (void) { int devnull; /* * Disconnect from TTY if this is not a debug run */ switch (fork ()) { case -1: corosync_exit_error (COROSYNC_DONE_FORK); break; case 0: /* * child which is disconnected, run this process */ break; default: exit (0); break; } /* Create new session */ (void)setsid(); /* * Map stdin/out/err to /dev/null. */ devnull = open("/dev/null", O_RDWR); if (devnull == -1) { corosync_exit_error (COROSYNC_DONE_STD_TO_NULL_REDIR); } if (dup2(devnull, 0) < 0 || dup2(devnull, 1) < 0 || dup2(devnull, 2) < 0) { close(devnull); corosync_exit_error (COROSYNC_DONE_STD_TO_NULL_REDIR); } close(devnull); } static void corosync_mlockall (void) { int res; struct rlimit rlimit; rlimit.rlim_cur = RLIM_INFINITY; rlimit.rlim_max = RLIM_INFINITY; #ifndef RLIMIT_MEMLOCK #define RLIMIT_MEMLOCK RLIMIT_VMEM #endif setrlimit (RLIMIT_MEMLOCK, &rlimit); res = mlockall (MCL_CURRENT | MCL_FUTURE); if (res == -1) { LOGSYS_PERROR (errno, LOGSYS_LEVEL_WARNING, "Could not lock memory of service to avoid page faults"); }; } static void corosync_totem_stats_updater (void *data) { totempg_stats_t * stats; uint32_t total_mtt_rx_token; uint32_t total_backlog_calc; uint32_t total_token_holdtime; int t, prev, i; int32_t token_count; char key_name[ICMAP_KEYNAME_MAXLEN]; stats = api->totem_get_stats(); icmap_set_uint32("runtime.totem.pg.msg_reserved", stats->msg_reserved); icmap_set_uint32("runtime.totem.pg.msg_queue_avail", stats->msg_queue_avail); icmap_set_uint64("runtime.totem.pg.mrp.srp.orf_token_tx", stats->mrp->srp->orf_token_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.orf_token_rx", stats->mrp->srp->orf_token_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_merge_detect_tx", stats->mrp->srp->memb_merge_detect_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_merge_detect_rx", stats->mrp->srp->memb_merge_detect_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_join_tx", stats->mrp->srp->memb_join_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_join_rx", stats->mrp->srp->memb_join_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.mcast_tx", stats->mrp->srp->mcast_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.mcast_retx", stats->mrp->srp->mcast_retx); icmap_set_uint64("runtime.totem.pg.mrp.srp.mcast_rx", stats->mrp->srp->mcast_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_commit_token_tx", stats->mrp->srp->memb_commit_token_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.memb_commit_token_rx", stats->mrp->srp->memb_commit_token_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.token_hold_cancel_tx", stats->mrp->srp->token_hold_cancel_tx); icmap_set_uint64("runtime.totem.pg.mrp.srp.token_hold_cancel_rx", stats->mrp->srp->token_hold_cancel_rx); icmap_set_uint64("runtime.totem.pg.mrp.srp.operational_entered", stats->mrp->srp->operational_entered); icmap_set_uint64("runtime.totem.pg.mrp.srp.operational_token_lost", stats->mrp->srp->operational_token_lost); icmap_set_uint64("runtime.totem.pg.mrp.srp.gather_entered", stats->mrp->srp->gather_entered); icmap_set_uint64("runtime.totem.pg.mrp.srp.gather_token_lost", stats->mrp->srp->gather_token_lost); icmap_set_uint64("runtime.totem.pg.mrp.srp.commit_entered", stats->mrp->srp->commit_entered); icmap_set_uint64("runtime.totem.pg.mrp.srp.commit_token_lost", stats->mrp->srp->commit_token_lost); icmap_set_uint64("runtime.totem.pg.mrp.srp.recovery_entered", stats->mrp->srp->recovery_entered); icmap_set_uint64("runtime.totem.pg.mrp.srp.recovery_token_lost", stats->mrp->srp->recovery_token_lost); icmap_set_uint64("runtime.totem.pg.mrp.srp.consensus_timeouts", stats->mrp->srp->consensus_timeouts); icmap_set_uint64("runtime.totem.pg.mrp.srp.rx_msg_dropped", stats->mrp->srp->rx_msg_dropped); icmap_set_uint32("runtime.totem.pg.mrp.srp.continuous_gather", stats->mrp->srp->continuous_gather); icmap_set_uint32("runtime.totem.pg.mrp.srp.continuous_sendmsg_failures", stats->mrp->srp->continuous_sendmsg_failures); icmap_set_uint8("runtime.totem.pg.mrp.srp.firewall_enabled_or_nic_failure", stats->mrp->srp->continuous_gather > MAX_NO_CONT_GATHER ? 1 : 0); if (stats->mrp->srp->continuous_gather > MAX_NO_CONT_GATHER || stats->mrp->srp->continuous_sendmsg_failures > MAX_NO_CONT_SENDMSG_FAILURES) { log_printf (LOGSYS_LEVEL_WARNING, "Totem is unable to form a cluster because of an " "operating system or network fault. The most common " "cause of this message is that the local firewall is " "configured improperly."); icmap_set_uint8("runtime.totem.pg.mrp.srp.firewall_enabled_or_nic_failure", 1); } else { icmap_set_uint8("runtime.totem.pg.mrp.srp.firewall_enabled_or_nic_failure", 0); } for (i = 0; i < stats->mrp->srp->rrp->interface_count; i++) { snprintf(key_name, ICMAP_KEYNAME_MAXLEN, "runtime.totem.pg.mrp.rrp.%u.faulty", i); icmap_set_uint8(key_name, stats->mrp->srp->rrp->faulty[i]); } total_mtt_rx_token = 0; total_token_holdtime = 0; total_backlog_calc = 0; token_count = 0; t = stats->mrp->srp->latest_token; while (1) { if (t == 0) prev = TOTEM_TOKEN_STATS_MAX - 1; else prev = t - 1; if (prev == stats->mrp->srp->earliest_token) break; /* if tx == 0, then dropped token (not ours) */ if (stats->mrp->srp->token[t].tx != 0 || (stats->mrp->srp->token[t].rx - stats->mrp->srp->token[prev].rx) > 0 ) { total_mtt_rx_token += (stats->mrp->srp->token[t].rx - stats->mrp->srp->token[prev].rx); total_token_holdtime += (stats->mrp->srp->token[t].tx - stats->mrp->srp->token[t].rx); total_backlog_calc += stats->mrp->srp->token[t].backlog_calc; token_count++; } t = prev; } if (token_count) { icmap_set_uint32("runtime.totem.pg.mrp.srp.mtt_rx_token", (total_mtt_rx_token / token_count)); icmap_set_uint32("runtime.totem.pg.mrp.srp.avg_token_workload", (total_token_holdtime / token_count)); icmap_set_uint32("runtime.totem.pg.mrp.srp.avg_backlog_calc", (total_backlog_calc / token_count)); } cs_ipcs_stats_update(); api->timer_add_duration (1500 * MILLI_2_NANO_SECONDS, NULL, corosync_totem_stats_updater, &corosync_stats_timer_handle); } -static void totem_dynamic_notify( - int32_t event, - const char *key_name, - struct icmap_notify_value new_val, - struct icmap_notify_value old_val, - void *user_data) -{ - int res; - unsigned int ring_no; - unsigned int member_no; - struct totem_ip_address member; - int add_new_member = 0; - int remove_old_member = 0; - char tmp_str[ICMAP_KEYNAME_MAXLEN]; - - res = sscanf(key_name, "nodelist.node.%u.ring%u%s", &member_no, &ring_no, tmp_str); - if (res != 3) - return ; - - if (strcmp(tmp_str, "_addr") != 0) { - return; - } - - if (event == ICMAP_TRACK_ADD && new_val.type == ICMAP_VALUETYPE_STRING) { - add_new_member = 1; - } - - if (event == ICMAP_TRACK_DELETE && old_val.type == ICMAP_VALUETYPE_STRING) { - remove_old_member = 1; - } - - if (event == ICMAP_TRACK_MODIFY && new_val.type == ICMAP_VALUETYPE_STRING && - old_val.type == ICMAP_VALUETYPE_STRING) { - add_new_member = 1; - remove_old_member = 1; - } - - if (remove_old_member) { - log_printf(LOGSYS_LEVEL_DEBUG, - "removing dynamic member %s for ring %u", (char *)old_val.data, ring_no); - if (totemip_parse(&member, (char *)old_val.data, ip_version) == 0) { - totempg_member_remove (&member, ring_no); - } - } - - if (add_new_member) { - log_printf(LOGSYS_LEVEL_DEBUG, - "adding dynamic member %s for ring %u", (char *)new_val.data, ring_no); - if (totemip_parse(&member, (char *)new_val.data, ip_version) == 0) { - totempg_member_add (&member, ring_no); - } - } -} - -static void corosync_totem_dynamic_init (void) -{ - icmap_track_t icmap_track = NULL; - - icmap_track_add("nodelist.node.", - ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX, - totem_dynamic_notify, - NULL, - &icmap_track); -} - static void corosync_totem_stats_init (void) { icmap_set_uint32("runtime.totem.pg.mrp.srp.mtt_rx_token", 0); icmap_set_uint32("runtime.totem.pg.mrp.srp.avg_token_workload", 0); icmap_set_uint32("runtime.totem.pg.mrp.srp.avg_backlog_calc", 0); /* start stats timer */ api->timer_add_duration (1500 * MILLI_2_NANO_SECONDS, NULL, corosync_totem_stats_updater, &corosync_stats_timer_handle); } - static void deliver_fn ( unsigned int nodeid, const void *msg, unsigned int msg_len, int endian_conversion_required) { const struct qb_ipc_request_header *header; int32_t service; int32_t fn_id; uint32_t id; header = msg; if (endian_conversion_required) { id = swab32 (header->id); } else { id = header->id; } /* * Call the proper executive handler */ service = id >> 16; fn_id = id & 0xffff; if (!corosync_service[service]) { return; } if (fn_id >= corosync_service[service]->exec_engine_count) { log_printf(LOGSYS_LEVEL_WARNING, "discarded unknown message %d for service %d (max id %d)", fn_id, service, corosync_service[service]->exec_engine_count); return; } icmap_fast_inc(service_stats_rx[service][fn_id]); if (endian_conversion_required) { assert(corosync_service[service]->exec_engine[fn_id].exec_endian_convert_fn != NULL); corosync_service[service]->exec_engine[fn_id].exec_endian_convert_fn ((void *)msg); } corosync_service[service]->exec_engine[fn_id].exec_handler_fn (msg, nodeid); } int main_mcast ( const struct iovec *iovec, unsigned int iov_len, unsigned int guarantee) { const struct qb_ipc_request_header *req = iovec->iov_base; int32_t service; int32_t fn_id; service = req->id >> 16; fn_id = req->id & 0xffff; if (corosync_service[service]) { icmap_fast_inc(service_stats_tx[service][fn_id]); } return (totempg_groups_mcast_joined (corosync_group_handle, iovec, iov_len, guarantee)); } static void corosync_ring_id_create_or_load ( struct memb_ring_id *memb_ring_id, const struct totem_ip_address *addr) { int fd; int res = 0; char filename[PATH_MAX]; snprintf (filename, sizeof(filename), "%s/ringid_%s", get_run_dir(), totemip_print (addr)); fd = open (filename, O_RDONLY, 0700); /* * If file can be opened and read, read the ring id */ if (fd != -1) { res = read (fd, &memb_ring_id->seq, sizeof (uint64_t)); close (fd); } /* * If file could not be opened or read, create a new ring id */ if ((fd == -1) || (res != sizeof (uint64_t))) { memb_ring_id->seq = 0; umask(0); fd = open (filename, O_CREAT|O_RDWR, 0700); if (fd != -1) { res = write (fd, &memb_ring_id->seq, sizeof (uint64_t)); close (fd); if (res == -1) { LOGSYS_PERROR (errno, LOGSYS_LEVEL_ERROR, "Couldn't write ringid file '%s'", filename); corosync_exit_error (COROSYNC_DONE_STORE_RINGID); } } else { LOGSYS_PERROR (errno, LOGSYS_LEVEL_ERROR, "Couldn't create ringid file '%s'", filename); corosync_exit_error (COROSYNC_DONE_STORE_RINGID); } } totemip_copy(&memb_ring_id->rep, addr); assert (!totemip_zero_check(&memb_ring_id->rep)); } static void corosync_ring_id_store ( const struct memb_ring_id *memb_ring_id, const struct totem_ip_address *addr) { char filename[PATH_MAX]; int fd; int res; snprintf (filename, sizeof(filename), "%s/ringid_%s", get_run_dir(), totemip_print (addr)); fd = open (filename, O_WRONLY, 0700); if (fd == -1) { fd = open (filename, O_CREAT|O_RDWR, 0700); } if (fd == -1) { LOGSYS_PERROR(errno, LOGSYS_LEVEL_ERROR, "Couldn't store new ring id %llx to stable storage", memb_ring_id->seq); corosync_exit_error (COROSYNC_DONE_STORE_RINGID); } log_printf (LOGSYS_LEVEL_DEBUG, "Storing new sequence id for ring %llx", memb_ring_id->seq); res = write (fd, &memb_ring_id->seq, sizeof(memb_ring_id->seq)); close (fd); if (res != sizeof(memb_ring_id->seq)) { LOGSYS_PERROR(errno, LOGSYS_LEVEL_ERROR, "Couldn't store new ring id %llx to stable storage", memb_ring_id->seq); corosync_exit_error (COROSYNC_DONE_STORE_RINGID); } } static qb_loop_timer_handle recheck_the_q_level_timer; void corosync_recheck_the_q_level(void *data) { totempg_check_q_level(corosync_group_handle); if (cs_ipcs_q_level_get() == TOTEM_Q_LEVEL_CRITICAL) { qb_loop_timer_add(cs_poll_handle_get(), QB_LOOP_MED, 1*QB_TIME_NS_IN_MSEC, NULL, corosync_recheck_the_q_level, &recheck_the_q_level_timer); } } struct sending_allowed_private_data_struct { int reserved_msgs; }; int corosync_sending_allowed ( unsigned int service, unsigned int id, const void *msg, void *sending_allowed_private_data) { struct sending_allowed_private_data_struct *pd = (struct sending_allowed_private_data_struct *)sending_allowed_private_data; struct iovec reserve_iovec; struct qb_ipc_request_header *header = (struct qb_ipc_request_header *)msg; int sending_allowed; reserve_iovec.iov_base = (char *)header; reserve_iovec.iov_len = header->size; pd->reserved_msgs = totempg_groups_joined_reserve ( corosync_group_handle, &reserve_iovec, 1); if (pd->reserved_msgs == -1) { return -EINVAL; } sending_allowed = QB_FALSE; if (corosync_quorum_is_quorate() == 1 || corosync_service[service]->allow_inquorate == CS_LIB_ALLOW_INQUORATE) { // we are quorate // now check flow control if (corosync_service[service]->lib_engine[id].flow_control == CS_LIB_FLOW_CONTROL_NOT_REQUIRED) { sending_allowed = QB_TRUE; } else if (pd->reserved_msgs && sync_in_process == 0) { sending_allowed = QB_TRUE; } else if (pd->reserved_msgs == 0) { return -ENOBUFS; } else /* (sync_in_process) */ { return -EINPROGRESS; } } else { return -EHOSTUNREACH; } return (sending_allowed); } void corosync_sending_allowed_release (void *sending_allowed_private_data) { struct sending_allowed_private_data_struct *pd = (struct sending_allowed_private_data_struct *)sending_allowed_private_data; if (pd->reserved_msgs == -1) { return; } totempg_groups_joined_release (pd->reserved_msgs); } int message_source_is_local (const mar_message_source_t *source) { int ret = 0; assert (source != NULL); if (source->nodeid == totempg_my_nodeid_get ()) { ret = 1; } return ret; } void message_source_set ( mar_message_source_t *source, void *conn) { assert ((source != NULL) && (conn != NULL)); memset (source, 0, sizeof (mar_message_source_t)); source->nodeid = totempg_my_nodeid_get (); source->conn = conn; } struct scheduler_pause_timeout_data { struct totem_config *totem_config; qb_loop_timer_handle handle; unsigned long long tv_prev; unsigned long long max_tv_diff; }; static void timer_function_scheduler_timeout (void *data) { struct scheduler_pause_timeout_data *timeout_data = (struct scheduler_pause_timeout_data *)data; unsigned long long tv_current; unsigned long long tv_diff; tv_current = qb_util_nano_current_get (); if (timeout_data->tv_prev == 0) { /* * Initial call -> just pretent everything is ok */ timeout_data->tv_prev = tv_current; timeout_data->max_tv_diff = 0; } tv_diff = tv_current - timeout_data->tv_prev; timeout_data->tv_prev = tv_current; if (tv_diff > timeout_data->max_tv_diff) { log_printf (LOGSYS_LEVEL_WARNING, "Corosync main process was not scheduled for %0.4f ms " "(threshold is %0.4f ms). Consider token timeout increase.", (float)tv_diff / QB_TIME_NS_IN_MSEC, (float)timeout_data->max_tv_diff / QB_TIME_NS_IN_MSEC); } /* * Set next threshold, because token_timeout can change */ timeout_data->max_tv_diff = timeout_data->totem_config->token_timeout * QB_TIME_NS_IN_MSEC * 0.8; qb_loop_timer_add (corosync_poll_handle, QB_LOOP_MED, timeout_data->totem_config->token_timeout * QB_TIME_NS_IN_MSEC / 3, timeout_data, timer_function_scheduler_timeout, &timeout_data->handle); } static void corosync_setscheduler (void) { #if defined(HAVE_PTHREAD_SETSCHEDPARAM) && defined(HAVE_SCHED_GET_PRIORITY_MAX) && defined(HAVE_SCHED_SETSCHEDULER) int res; sched_priority = sched_get_priority_max (SCHED_RR); if (sched_priority != -1) { global_sched_param.sched_priority = sched_priority; res = sched_setscheduler (0, SCHED_RR, &global_sched_param); if (res == -1) { LOGSYS_PERROR(errno, LOGSYS_LEVEL_WARNING, "Could not set SCHED_RR at priority %d", global_sched_param.sched_priority); global_sched_param.sched_priority = 0; #ifdef HAVE_QB_LOG_THREAD_PRIORITY_SET qb_log_thread_priority_set (SCHED_OTHER, 0); #endif } else { /* * Turn on SCHED_RR in logsys system */ #ifdef HAVE_QB_LOG_THREAD_PRIORITY_SET res = qb_log_thread_priority_set (SCHED_RR, sched_priority); #else res = -1; #endif if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Could not set logsys thread priority." " Can't continue because of priority inversions."); corosync_exit_error (COROSYNC_DONE_LOGSETUP); } } } else { LOGSYS_PERROR (errno, LOGSYS_LEVEL_WARNING, "Could not get maximum scheduler priority"); sched_priority = 0; } #else log_printf(LOGSYS_LEVEL_WARNING, "The Platform is missing process priority setting features. Leaving at default."); #endif } static void _logsys_log_printf(int level, int subsys, const char *function_name, const char *file_name, int file_line, const char *format, ...) __attribute__((format(printf, 6, 7))); static void _logsys_log_printf(int level, int subsys, const char *function_name, const char *file_name, int file_line, const char *format, ...) { va_list ap; va_start(ap, format); qb_log_from_external_source_va(function_name, file_name, format, level, file_line, subsys, ap); va_end(ap); } static void fplay_key_change_notify_fn ( int32_t event, const char *key_name, struct icmap_notify_value new_val, struct icmap_notify_value old_val, void *user_data) { if (strcmp(key_name, "runtime.blackbox.dump_flight_data") == 0) { fprintf(stderr,"Writetofile\n"); corosync_blackbox_write_to_file (); } if (strcmp(key_name, "runtime.blackbox.dump_state") == 0) { fprintf(stderr,"statefump\n"); corosync_state_dump (); } } static void corosync_fplay_control_init (void) { icmap_track_t track = NULL; icmap_set_string("runtime.blackbox.dump_flight_data", "no"); icmap_set_string("runtime.blackbox.dump_state", "no"); icmap_track_add("runtime.blackbox.dump_flight_data", ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY, fplay_key_change_notify_fn, NULL, &track); icmap_track_add("runtime.blackbox.dump_state", ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY, fplay_key_change_notify_fn, NULL, &track); } /* * Set RO flag for keys, which ether doesn't make sense to change by user (statistic) * or which when changed are not reflected by runtime (totem.crypto_cipher, ...). * * Also some RO keys cannot be determined in this stage, so they are set later in * other functions (like nodelist.local_node_pos, ...) */ static void set_icmap_ro_keys_flag (void) { /* * Set RO flag for all keys of internal configuration and runtime statistics */ icmap_set_ro_access("internal_configuration.", CS_TRUE, CS_TRUE); icmap_set_ro_access("runtime.connections.", CS_TRUE, CS_TRUE); icmap_set_ro_access("runtime.totem.", CS_TRUE, CS_TRUE); icmap_set_ro_access("runtime.services.", CS_TRUE, CS_TRUE); icmap_set_ro_access("runtime.config.", CS_TRUE, CS_TRUE); /* * Set RO flag for constrete keys of configuration which can't be changed * during runtime */ icmap_set_ro_access("totem.crypto_cipher", CS_FALSE, CS_TRUE); icmap_set_ro_access("totem.crypto_hash", CS_FALSE, CS_TRUE); icmap_set_ro_access("totem.secauth", CS_FALSE, CS_TRUE); icmap_set_ro_access("totem.ip_version", CS_FALSE, CS_TRUE); icmap_set_ro_access("totem.rrp_mode", CS_FALSE, CS_TRUE); icmap_set_ro_access("totem.netmtu", CS_FALSE, CS_TRUE); icmap_set_ro_access("qb.ipc_type", CS_FALSE, CS_TRUE); icmap_set_ro_access("config.reload_in_progress", CS_FALSE, CS_TRUE); icmap_set_ro_access("config.totemconfig_reload_in_progress", CS_FALSE, CS_TRUE); } static void main_service_ready (void) { int res; /* * This must occur after totempg is initialized because "this_ip" must be set */ res = corosync_service_defaults_link_and_init (api); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Could not initialize default services"); corosync_exit_error (COROSYNC_DONE_INIT_SERVICES); } cs_ipcs_init(); corosync_totem_stats_init (); corosync_fplay_control_init (); - corosync_totem_dynamic_init (); sync_init ( corosync_sync_callbacks_retrieve, corosync_sync_completed); } static enum e_corosync_done corosync_flock (const char *lockfile, pid_t pid) { struct flock lock; enum e_corosync_done err; char pid_s[17]; int fd_flag; int lf; err = COROSYNC_DONE_EXIT; lf = open (lockfile, O_WRONLY | O_CREAT, 0640); if (lf == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't create lock file."); return (COROSYNC_DONE_AQUIRE_LOCK); } retry_fcntl: lock.l_type = F_WRLCK; lock.l_start = 0; lock.l_whence = SEEK_SET; lock.l_len = 0; if (fcntl (lf, F_SETLK, &lock) == -1) { switch (errno) { case EINTR: goto retry_fcntl; break; case EAGAIN: case EACCES: log_printf (LOGSYS_LEVEL_ERROR, "Another Corosync instance is already running."); err = COROSYNC_DONE_ALREADY_RUNNING; goto error_close; break; default: log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't aquire lock. Error was %s", strerror(errno)); err = COROSYNC_DONE_AQUIRE_LOCK; goto error_close; break; } } if (ftruncate (lf, 0) == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't truncate lock file. Error was %s", strerror (errno)); err = COROSYNC_DONE_AQUIRE_LOCK; goto error_close_unlink; } memset (pid_s, 0, sizeof (pid_s)); snprintf (pid_s, sizeof (pid_s) - 1, "%u\n", pid); retry_write: if (write (lf, pid_s, strlen (pid_s)) != strlen (pid_s)) { if (errno == EINTR) { goto retry_write; } else { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't write pid to lock file. " "Error was %s", strerror (errno)); err = COROSYNC_DONE_AQUIRE_LOCK; goto error_close_unlink; } } if ((fd_flag = fcntl (lf, F_GETFD, 0)) == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't get close-on-exec flag from lock file. " "Error was %s", strerror (errno)); err = COROSYNC_DONE_AQUIRE_LOCK; goto error_close_unlink; } fd_flag |= FD_CLOEXEC; if (fcntl (lf, F_SETFD, fd_flag) == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't set close-on-exec flag to lock file. " "Error was %s", strerror (errno)); err = COROSYNC_DONE_AQUIRE_LOCK; goto error_close_unlink; } return (err); error_close_unlink: unlink (lockfile); error_close: close (lf); return (err); } int main (int argc, char **argv, char **envp) { const char *error_string; struct totem_config totem_config; int res, ch; int background, setprio, testonly; struct stat stat_out; enum e_corosync_done flock_err; uint64_t totem_config_warnings; struct scheduler_pause_timeout_data scheduler_pause_timeout_data; /* default configuration */ background = 1; setprio = 1; testonly = 0; while ((ch = getopt (argc, argv, "fprtv")) != EOF) { switch (ch) { case 'f': background = 0; break; case 'p': setprio = 0; break; case 'r': setprio = 1; break; case 't': testonly = 1; break; case 'v': printf ("Corosync Cluster Engine, version '%s'\n", VERSION); printf ("Copyright (c) 2006-2009 Red Hat, Inc.\n"); logsys_system_fini(); return EXIT_SUCCESS; break; default: fprintf(stderr, \ "usage:\n"\ " -f : Start application in foreground.\n"\ " -p : Do not set process priority.\n"\ " -t : Test configuration and exit.\n"\ " -r : Set round robin realtime scheduling (default).\n"\ " -v : Display version and SVN revision of Corosync and exit.\n"); logsys_system_fini(); return EXIT_FAILURE; } } /* * Set round robin realtime scheduling with priority 99 * Lock all memory to avoid page faults which may interrupt * application healthchecking */ if (setprio) { corosync_setscheduler (); } corosync_mlockall (); /* * Other signals are registered later via qb_loop_signal_add */ (void)signal (SIGSEGV, sigsegv_handler); (void)signal (SIGABRT, sigabrt_handler); #if MSG_NOSIGNAL != 0 (void)signal (SIGPIPE, SIG_IGN); #endif if (icmap_init() != CS_OK) { log_printf (LOGSYS_LEVEL_ERROR, "Corosync Executive couldn't initialize configuration component."); corosync_exit_error (COROSYNC_DONE_ICMAP); } set_icmap_ro_keys_flag(); /* * Initialize the corosync_api_v1 definition */ api = apidef_get (); res = coroparse_configparse(icmap_get_global_map(), &error_string); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); corosync_exit_error (COROSYNC_DONE_MAINCONFIGREAD); } res = corosync_log_config_read (&error_string); if (res == -1) { /* * if we are here, we _must_ flush the logsys queue * and try to inform that we couldn't read the config. * this is a desperate attempt before certain death * and there is no guarantee that we can print to stderr * nor that logsys is sending the messages where we expect. */ log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); fprintf(stderr, "%s", error_string); syslog (LOGSYS_LEVEL_ERROR, "%s", error_string); corosync_exit_error (COROSYNC_DONE_LOGCONFIGREAD); } if (!testonly) { log_printf (LOGSYS_LEVEL_NOTICE, "Corosync Cluster Engine ('%s'): started and ready to provide service.", VERSION); log_printf (LOGSYS_LEVEL_INFO, "Corosync built-in features:" PACKAGE_FEATURES ""); } /* * Make sure required directory is present */ res = stat (get_run_dir(), &stat_out); if ((res == -1) || (res == 0 && !S_ISDIR(stat_out.st_mode))) { log_printf (LOGSYS_LEVEL_ERROR, "Required directory not present %s. Please create it.", get_run_dir()); corosync_exit_error (COROSYNC_DONE_DIR_NOT_PRESENT); } res = chdir(get_run_dir()); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "Cannot chdir to run directory %s. " "Please make sure it has correct context and rights.", get_run_dir()); corosync_exit_error (COROSYNC_DONE_DIR_NOT_PRESENT); } res = totem_config_read (&totem_config, &error_string, &totem_config_warnings); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); corosync_exit_error (COROSYNC_DONE_MAINCONFIGREAD); } if (totem_config_warnings & TOTEM_CONFIG_WARNING_MEMBERS_IGNORED) { log_printf (LOGSYS_LEVEL_WARNING, "member section is used together with nodelist. Members ignored."); } if (totem_config_warnings & TOTEM_CONFIG_WARNING_MEMBERS_DEPRECATED) { log_printf (LOGSYS_LEVEL_WARNING, "member section is deprecated."); } if (totem_config_warnings & TOTEM_CONFIG_WARNING_TOTEM_NODEID_IGNORED) { log_printf (LOGSYS_LEVEL_WARNING, "nodeid appears both in totem section and nodelist. Nodelist one is used."); } if (totem_config_warnings != 0) { log_printf (LOGSYS_LEVEL_WARNING, "Please migrate config file to nodelist."); } res = totem_config_keyread (&totem_config, &error_string); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); corosync_exit_error (COROSYNC_DONE_MAINCONFIGREAD); } res = totem_config_validate (&totem_config, &error_string); if (res == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); corosync_exit_error (COROSYNC_DONE_MAINCONFIGREAD); } if (testonly) { corosync_exit_error (COROSYNC_DONE_EXIT); } ip_version = totem_config.ip_version; totem_config.totem_memb_ring_id_create_or_load = corosync_ring_id_create_or_load; totem_config.totem_memb_ring_id_store = corosync_ring_id_store; totem_config.totem_logging_configuration = totem_logging_configuration; totem_config.totem_logging_configuration.log_subsys_id = _logsys_subsys_create("TOTEM", "totem," "totemmrp.c,totemrrp.c,totemip.c,totemconfig.c,totemcrypto.c,totemsrp.c," "totempg.c,totemiba.c,totemudp.c,totemudpu.c,totemnet.c"); totem_config.totem_logging_configuration.log_level_security = LOGSYS_LEVEL_WARNING; totem_config.totem_logging_configuration.log_level_error = LOGSYS_LEVEL_ERROR; totem_config.totem_logging_configuration.log_level_warning = LOGSYS_LEVEL_WARNING; totem_config.totem_logging_configuration.log_level_notice = LOGSYS_LEVEL_NOTICE; totem_config.totem_logging_configuration.log_level_debug = LOGSYS_LEVEL_DEBUG; totem_config.totem_logging_configuration.log_level_trace = LOGSYS_LEVEL_TRACE; totem_config.totem_logging_configuration.log_printf = _logsys_log_printf; logsys_config_apply(); /* * Now we are fully initialized. */ if (background) { corosync_tty_detach (); } corosync_poll_handle = qb_loop_create (); memset(&scheduler_pause_timeout_data, 0, sizeof(scheduler_pause_timeout_data)); scheduler_pause_timeout_data.totem_config = &totem_config; timer_function_scheduler_timeout (&scheduler_pause_timeout_data); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_LOW, SIGUSR2, NULL, sig_diag_handler, NULL); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_HIGH, SIGINT, NULL, sig_exit_handler, NULL); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_HIGH, SIGSEGV, NULL, sig_segv_handler, NULL); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_HIGH, SIGABRT, NULL, sig_abrt_handler, NULL); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_HIGH, SIGQUIT, NULL, sig_exit_handler, NULL); qb_loop_signal_add(corosync_poll_handle, QB_LOOP_HIGH, SIGTERM, NULL, sig_exit_handler, NULL); if (logsys_thread_start() != 0) { log_printf (LOGSYS_LEVEL_ERROR, "Can't initialize log thread"); corosync_exit_error (COROSYNC_DONE_LOGCONFIGREAD); } if ((flock_err = corosync_flock (corosync_lock_file, getpid ())) != COROSYNC_DONE_EXIT) { corosync_exit_error (flock_err); } /* * if totempg_initialize doesn't have root priveleges, it cannot * bind to a specific interface. This only matters if * there is more then one interface in a system, so * in this case, only a warning is printed */ /* * Join multicast group and setup delivery * and configuration change functions */ totempg_initialize ( corosync_poll_handle, &totem_config); totempg_service_ready_register ( main_service_ready); totempg_groups_initialize ( &corosync_group_handle, deliver_fn, confchg_fn); totempg_groups_join ( corosync_group_handle, &corosync_group, 1); /* * Drop root privleges to user 'corosync' * TODO: Don't really need full root capabilities; * needed capabilities are: * CAP_NET_RAW (bindtodevice) * CAP_SYS_NICE (setscheduler) * CAP_IPC_LOCK (mlockall) */ priv_drop (); schedwrk_init ( serialize_lock, serialize_unlock); /* * Start main processing loop */ qb_loop_run (corosync_poll_handle); /* * Exit was requested */ totempg_finalize (); /* * free the loop resources */ qb_loop_destroy (corosync_poll_handle); /* * free up the icmap */ /* * Remove pid lock file */ unlink (corosync_lock_file); corosync_exit_error (COROSYNC_DONE_EXIT); return EXIT_SUCCESS; } diff --git a/exec/totemconfig.c b/exec/totemconfig.c index 2acee2a9..b678752d 100644 --- a/exec/totemconfig.c +++ b/exec/totemconfig.c @@ -1,1404 +1,1535 @@ /* * Copyright (c) 2002-2005 MontaVista Software, Inc. * Copyright (c) 2006-2013 Red Hat, Inc. * * All rights reserved. * * Author: Steven Dake (sdake@redhat.com) * Jan Friesse (jfriesse@redhat.com) * * This software licensed under BSD license, the text of which follows: * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions are met: * * - Redistributions of source code must retain the above copyright notice, * this list of conditions and the following disclaimer. * - Redistributions in binary form must reproduce the above copyright notice, * this list of conditions and the following disclaimer in the documentation * and/or other materials provided with the distribution. * - Neither the name of the MontaVista Software, Inc. nor the names of its * contributors may be used to endorse or promote products derived from this * software without specific prior written permission. * * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF * THE POSSIBILITY OF SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "util.h" #include "totemconfig.h" #define TOKEN_RETRANSMITS_BEFORE_LOSS_CONST 4 #define TOKEN_TIMEOUT 1000 #define TOKEN_COEFFICIENT 650 #define JOIN_TIMEOUT 50 #define MERGE_TIMEOUT 200 #define DOWNCHECK_TIMEOUT 1000 #define FAIL_TO_RECV_CONST 2500 #define SEQNO_UNCHANGED_CONST 30 #define MINIMUM_TIMEOUT (int)(1000/HZ)*3 #define MAX_NETWORK_DELAY 50 #define WINDOW_SIZE 50 #define MAX_MESSAGES 17 #define MISS_COUNT_CONST 5 #define RRP_PROBLEM_COUNT_TIMEOUT 2000 #define RRP_PROBLEM_COUNT_THRESHOLD_DEFAULT 10 #define RRP_PROBLEM_COUNT_THRESHOLD_MIN 2 #define RRP_AUTORECOVERY_CHECK_TIMEOUT 1000 #define DEFAULT_PORT 5405 static char error_string_response[512]; static void add_totem_config_notification(struct totem_config *totem_config); /* All the volatile parameters are uint32s, luckily */ static uint32_t *totem_get_param_by_name(struct totem_config *totem_config, const char *param_name) { if (strcmp(param_name, "totem.token") == 0) return &totem_config->token_timeout; if (strcmp(param_name, "totem.token_retransmit") == 0) return &totem_config->token_retransmit_timeout; if (strcmp(param_name, "totem.hold") == 0) return &totem_config->token_hold_timeout; if (strcmp(param_name, "totem.token_retransmits_before_loss_const") == 0) return &totem_config->token_retransmits_before_loss_const; if (strcmp(param_name, "totem.join") == 0) return &totem_config->join_timeout; if (strcmp(param_name, "totem.send_join") == 0) return &totem_config->send_join_timeout; if (strcmp(param_name, "totem.consensus") == 0) return &totem_config->consensus_timeout; if (strcmp(param_name, "totem.merge") == 0) return &totem_config->merge_timeout; if (strcmp(param_name, "totem.downcheck") == 0) return &totem_config->downcheck_timeout; if (strcmp(param_name, "totem.fail_recv_const") == 0) return &totem_config->fail_to_recv_const; if (strcmp(param_name, "totem.seqno_unchanged_const") == 0) return &totem_config->seqno_unchanged_const; if (strcmp(param_name, "totem.rrp_token_expired_timeout") == 0) return &totem_config->rrp_token_expired_timeout; if (strcmp(param_name, "totem.rrp_problem_count_timeout") == 0) return &totem_config->rrp_problem_count_timeout; if (strcmp(param_name, "totem.rrp_problem_count_threshold") == 0) return &totem_config->rrp_problem_count_threshold; if (strcmp(param_name, "totem.rrp_problem_count_mcast_threshold") == 0) return &totem_config->rrp_problem_count_mcast_threshold; if (strcmp(param_name, "totem.rrp_autorecovery_check_timeout") == 0) return &totem_config->rrp_autorecovery_check_timeout; if (strcmp(param_name, "totem.heartbeat_failures_allowed") == 0) return &totem_config->heartbeat_failures_allowed; if (strcmp(param_name, "totem.max_network_delay") == 0) return &totem_config->max_network_delay; if (strcmp(param_name, "totem.window_size") == 0) return &totem_config->window_size; if (strcmp(param_name, "totem.max_messages") == 0) return &totem_config->max_messages; if (strcmp(param_name, "totem.miss_count_const") == 0) return &totem_config->miss_count_const; return NULL; } /* * Read key_name from icmap. If key is not found or key_name == delete_key or if allow_zero is false * and readed value is zero, default value is used and stored into totem_config. */ static void totem_volatile_config_set_value (struct totem_config *totem_config, const char *key_name, const char *deleted_key, unsigned int default_value, int allow_zero_value) { char runtime_key_name[ICMAP_KEYNAME_MAXLEN]; if (icmap_get_uint32(key_name, totem_get_param_by_name(totem_config, key_name)) != CS_OK || (deleted_key != NULL && strcmp(deleted_key, key_name) == 0) || (!allow_zero_value && *totem_get_param_by_name(totem_config, key_name) == 0)) { *totem_get_param_by_name(totem_config, key_name) = default_value; } /* * Store totem_config value to cmap runtime section */ strcpy(runtime_key_name, "runtime.config."); strcat(runtime_key_name, key_name); icmap_set_uint32(runtime_key_name, *totem_get_param_by_name(totem_config, key_name)); } /* * Read and validate config values from cmap and store them into totem_config. If key doesn't exists, * default value is stored. deleted_key is name of key beeing processed by delete operation * from cmap. It is considered as non existing even if it can be read. Can be NULL. */ static void totem_volatile_config_read (struct totem_config *totem_config, const char *deleted_key) { uint32_t u32; totem_volatile_config_set_value(totem_config, "totem.token_retransmits_before_loss_const", deleted_key, TOKEN_RETRANSMITS_BEFORE_LOSS_CONST, 0); totem_volatile_config_set_value(totem_config, "totem.token", deleted_key, TOKEN_TIMEOUT, 0); if (totem_config->interface_count > 0 && totem_config->interfaces[0].member_count > 2) { u32 = TOKEN_COEFFICIENT; icmap_get_uint32("totem.token_coefficient", &u32); totem_config->token_timeout += (totem_config->interfaces[0].member_count - 2) * u32; /* * Store totem_config value to cmap runtime section */ icmap_set_uint32("runtime.config.totem.token", totem_config->token_timeout); } totem_volatile_config_set_value(totem_config, "totem.max_network_delay", deleted_key, MAX_NETWORK_DELAY, 0); totem_volatile_config_set_value(totem_config, "totem.window_size", deleted_key, WINDOW_SIZE, 0); totem_volatile_config_set_value(totem_config, "totem.max_messages", deleted_key, MAX_MESSAGES, 0); totem_volatile_config_set_value(totem_config, "totem.miss_count_const", deleted_key, MISS_COUNT_CONST, 0); totem_volatile_config_set_value(totem_config, "totem.token_retransmit", deleted_key, (int)(totem_config->token_timeout / (totem_config->token_retransmits_before_loss_const + 0.2)), 0); totem_volatile_config_set_value(totem_config, "totem.hold", deleted_key, (int)(totem_config->token_retransmit_timeout * 0.8 - (1000/HZ)), 0); totem_volatile_config_set_value(totem_config, "totem.join", deleted_key, JOIN_TIMEOUT, 0); totem_volatile_config_set_value(totem_config, "totem.consensus", deleted_key, (int)(float)(1.2 * totem_config->token_timeout), 0); totem_volatile_config_set_value(totem_config, "totem.merge", deleted_key, MERGE_TIMEOUT, 0); totem_volatile_config_set_value(totem_config, "totem.downcheck", deleted_key, DOWNCHECK_TIMEOUT, 0); totem_volatile_config_set_value(totem_config, "totem.fail_recv_const", deleted_key, FAIL_TO_RECV_CONST, 0); totem_volatile_config_set_value(totem_config, "totem.seqno_unchanged_const", deleted_key, SEQNO_UNCHANGED_CONST, 0); totem_volatile_config_set_value(totem_config, "totem.send_join", deleted_key, 0, 1); totem_volatile_config_set_value(totem_config, "totem.rrp_problem_count_timeout", deleted_key, RRP_PROBLEM_COUNT_TIMEOUT, 0); totem_volatile_config_set_value(totem_config, "totem.rrp_problem_count_threshold", deleted_key, RRP_PROBLEM_COUNT_THRESHOLD_DEFAULT, 0); totem_volatile_config_set_value(totem_config, "totem.rrp_problem_count_mcast_threshold", deleted_key, totem_config->rrp_problem_count_threshold * 10, 0); totem_volatile_config_set_value(totem_config, "totem.rrp_token_expired_timeout", deleted_key, totem_config->token_retransmit_timeout, 0); totem_volatile_config_set_value(totem_config, "totem.rrp_autorecovery_check_timeout", deleted_key, RRP_AUTORECOVERY_CHECK_TIMEOUT, 0); totem_volatile_config_set_value(totem_config, "totem.heartbeat_failures_allowed", deleted_key, 0, 1); } static int totem_volatile_config_validate ( struct totem_config *totem_config, const char **error_string) { static char local_error_reason[512]; const char *error_reason = local_error_reason; if (totem_config->max_network_delay < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The max_network_delay parameter (%d ms) may not be less than (%d ms).", totem_config->max_network_delay, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->token_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The token timeout parameter (%d ms) may not be less than (%d ms).", totem_config->token_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->token_retransmit_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The token retransmit timeout parameter (%d ms) may not be less than (%d ms).", totem_config->token_retransmit_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->token_hold_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The token hold timeout parameter (%d ms) may not be less than (%d ms).", totem_config->token_hold_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->join_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The join timeout parameter (%d ms) may not be less than (%d ms).", totem_config->join_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->consensus_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The consensus timeout parameter (%d ms) may not be less than (%d ms).", totem_config->consensus_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->consensus_timeout < totem_config->join_timeout) { snprintf (local_error_reason, sizeof(local_error_reason), "The consensus timeout parameter (%d ms) may not be less than join timeout (%d ms).", totem_config->consensus_timeout, totem_config->join_timeout); goto parse_error; } if (totem_config->merge_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The merge timeout parameter (%d ms) may not be less than (%d ms).", totem_config->merge_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->downcheck_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The downcheck timeout parameter (%d ms) may not be less than (%d ms).", totem_config->downcheck_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->rrp_problem_count_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The RRP problem count timeout parameter (%d ms) may not be less than (%d ms).", totem_config->rrp_problem_count_timeout, MINIMUM_TIMEOUT); goto parse_error; } if (totem_config->rrp_problem_count_threshold < RRP_PROBLEM_COUNT_THRESHOLD_MIN) { snprintf (local_error_reason, sizeof(local_error_reason), "The RRP problem count threshold (%d problem count) may not be less than (%d problem count).", totem_config->rrp_problem_count_threshold, RRP_PROBLEM_COUNT_THRESHOLD_MIN); goto parse_error; } if (totem_config->rrp_problem_count_mcast_threshold < RRP_PROBLEM_COUNT_THRESHOLD_MIN) { snprintf (local_error_reason, sizeof(local_error_reason), "The RRP multicast problem count threshold (%d problem count) may not be less than (%d problem count).", totem_config->rrp_problem_count_mcast_threshold, RRP_PROBLEM_COUNT_THRESHOLD_MIN); goto parse_error; } if (totem_config->rrp_token_expired_timeout < MINIMUM_TIMEOUT) { snprintf (local_error_reason, sizeof(local_error_reason), "The RRP token expired timeout parameter (%d ms) may not be less than (%d ms).", totem_config->rrp_token_expired_timeout, MINIMUM_TIMEOUT); goto parse_error; } return 0; parse_error: snprintf (error_string_response, sizeof(error_string_response), "parse error in config: %s\n", error_reason); *error_string = error_string_response; return (-1); } static int totem_get_crypto(struct totem_config *totem_config) { char *str; const char *tmp_cipher; const char *tmp_hash; tmp_hash = "sha1"; tmp_cipher = "aes256"; if (icmap_get_string("totem.secauth", &str) == CS_OK) { if (strcmp (str, "off") == 0) { tmp_hash = "none"; tmp_cipher = "none"; } free(str); } if (icmap_get_string("totem.crypto_cipher", &str) == CS_OK) { if (strcmp(str, "none") == 0) { tmp_cipher = "none"; } if (strcmp(str, "aes256") == 0) { tmp_cipher = "aes256"; } if (strcmp(str, "aes192") == 0) { tmp_cipher = "aes192"; } if (strcmp(str, "aes128") == 0) { tmp_cipher = "aes128"; } if (strcmp(str, "3des") == 0) { tmp_cipher = "3des"; } free(str); } if (icmap_get_string("totem.crypto_hash", &str) == CS_OK) { if (strcmp(str, "none") == 0) { tmp_hash = "none"; } if (strcmp(str, "md5") == 0) { tmp_hash = "md5"; } if (strcmp(str, "sha1") == 0) { tmp_hash = "sha1"; } if (strcmp(str, "sha256") == 0) { tmp_hash = "sha256"; } if (strcmp(str, "sha384") == 0) { tmp_hash = "sha384"; } if (strcmp(str, "sha512") == 0) { tmp_hash = "sha512"; } free(str); } if ((strcmp(tmp_cipher, "none") != 0) && (strcmp(tmp_hash, "none") == 0)) { return -1; } free(totem_config->crypto_cipher_type); free(totem_config->crypto_hash_type); totem_config->crypto_cipher_type = strdup(tmp_cipher); totem_config->crypto_hash_type = strdup(tmp_hash); return 0; } static int totem_config_get_ip_version(void) { int res; char *str; res = AF_INET; if (icmap_get_string("totem.ip_version", &str) == CS_OK) { if (strcmp(str, "ipv4") == 0) { res = AF_INET; } if (strcmp(str, "ipv6") == 0) { res = AF_INET6; } free(str); } return (res); } static uint16_t generate_cluster_id (const char *cluster_name) { int i; int value = 0; for (i = 0; i < strlen(cluster_name); i++) { value <<= 1; value += cluster_name[i]; } return (value & 0xFFFF); } static int get_cluster_mcast_addr ( const char *cluster_name, const struct totem_ip_address *bindnet, unsigned int ringnumber, int ip_version, struct totem_ip_address *res) { uint16_t clusterid; char addr[INET6_ADDRSTRLEN + 1]; int err; if (cluster_name == NULL) { return (-1); } clusterid = generate_cluster_id(cluster_name) + ringnumber; memset (res, 0, sizeof(*res)); switch (bindnet->family) { case AF_INET: snprintf(addr, sizeof(addr), "239.192.%d.%d", clusterid >> 8, clusterid % 0xFF); break; case AF_INET6: snprintf(addr, sizeof(addr), "ff15::%x", clusterid); break; default: /* * Unknown family */ return (-1); } err = totemip_parse (res, addr, ip_version); return (err); } static int find_local_node_in_nodelist(struct totem_config *totem_config) { icmap_iter_t iter; const char *iter_key; int res = 0; unsigned int node_pos; int local_node_pos = -1; struct totem_ip_address bind_addr; int interface_up, interface_num; char tmp_key[ICMAP_KEYNAME_MAXLEN]; char *node_addr_str; struct totem_ip_address node_addr; res = totemip_iface_check(&totem_config->interfaces[0].bindnet, &bind_addr, &interface_up, &interface_num, totem_config->clear_node_high_bit); if (res == -1) { return (-1); } iter = icmap_iter_init("nodelist.node."); while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) { res = sscanf(iter_key, "nodelist.node.%u.%s", &node_pos, tmp_key); if (res != 2) { continue; } if (strcmp(tmp_key, "ring0_addr") != 0) { continue; } snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.ring0_addr", node_pos); if (icmap_get_string(tmp_key, &node_addr_str) != CS_OK) { continue; } res = totemip_parse (&node_addr, node_addr_str, totem_config->ip_version); free(node_addr_str); if (res == -1) { continue ; } if (totemip_equal(&bind_addr, &node_addr)) { local_node_pos = node_pos; } } icmap_iter_finalize(iter); return (local_node_pos); } -static void put_nodelist_members_to_config(struct totem_config *totem_config) +/* + * Compute difference between two set of totem interface arrays. set1 and set2 + * are changed so for same ring, ip existing in both set1 and set2 are cleared + * (set to 0), and ips which are only in set1 or set2 remains untouched. + * totempg_node_add/remove is called. + */ +static void compute_interfaces_diff(int interface_count, + struct totem_interface *set1, + struct totem_interface *set2) +{ + int ring_no, set1_pos, set2_pos; + struct totem_ip_address empty_ip_address; + + memset(&empty_ip_address, 0, sizeof(empty_ip_address)); + + for (ring_no = 0; ring_no < interface_count; ring_no++) { + for (set1_pos = 0; set1_pos < set1[ring_no].member_count; set1_pos++) { + for (set2_pos = 0; set2_pos < set2[ring_no].member_count; set2_pos++) { + /* + * For current ring_no remove all set1 items existing + * in set2 + */ + if (memcmp(&set1[ring_no].member_list[set1_pos], + &set2[ring_no].member_list[set2_pos], + sizeof(struct totem_ip_address)) == 0) { + memset(&set1[ring_no].member_list[set1_pos], 0, + sizeof(struct totem_ip_address)); + memset(&set2[ring_no].member_list[set2_pos], 0, + sizeof(struct totem_ip_address)); + } + } + } + } + + for (ring_no = 0; ring_no < interface_count; ring_no++) { + for (set1_pos = 0; set1_pos < set1[ring_no].member_count; set1_pos++) { + /* + * All items which remained in set1 doesn't exists in set2 any longer so + * node has to be removed. + */ + if (memcmp(&set1[ring_no].member_list[set1_pos], &empty_ip_address, sizeof(empty_ip_address)) != 0) { + log_printf(LOGSYS_LEVEL_DEBUG, + "removing dynamic member %s for ring %u", + totemip_print(&set1[ring_no].member_list[set1_pos]), + ring_no); + + totempg_member_remove(&set1[ring_no].member_list[set1_pos], ring_no); + } + } + for (set2_pos = 0; set2_pos < set2[ring_no].member_count; set2_pos++) { + /* + * All items which remained in set2 doesn't existed in set1 so this is no node + * and has to be added. + */ + if (memcmp(&set2[ring_no].member_list[set2_pos], &empty_ip_address, sizeof(empty_ip_address)) != 0) { + log_printf(LOGSYS_LEVEL_DEBUG, + "adding dynamic member %s for ring %u", + totemip_print(&set2[ring_no].member_list[set2_pos]), + ring_no); + + totempg_member_add(&set2[ring_no].member_list[set2_pos], ring_no); + } + } + } +} + +static void put_nodelist_members_to_config(struct totem_config *totem_config, int reload) { icmap_iter_t iter, iter2; const char *iter_key, *iter_key2; int res = 0; unsigned int node_pos; char tmp_key[ICMAP_KEYNAME_MAXLEN]; char tmp_key2[ICMAP_KEYNAME_MAXLEN]; char *node_addr_str; int member_count; unsigned int ringnumber = 0; int i, j; + struct totem_interface *orig_interfaces = NULL; + struct totem_interface *new_interfaces = NULL; + + if (reload) { + /* + * We need to compute diff only for reload. Also for initial configuration + * not all totem structures are initialized so corosync will crash during + * member_add/remove + */ + orig_interfaces = malloc (sizeof (struct totem_interface) * INTERFACE_MAX); + assert(orig_interfaces != NULL); + new_interfaces = malloc (sizeof (struct totem_interface) * INTERFACE_MAX); + assert(new_interfaces != NULL); + + memcpy(orig_interfaces, totem_config->interfaces, sizeof (struct totem_interface) * INTERFACE_MAX); + } /* Clear out nodelist so we can put the new one in if needed */ for (i = 0; i < totem_config->interface_count; i++) { for (j = 0; j < PROCESSOR_COUNT_MAX; j++) { memset(&totem_config->interfaces[i].member_list[j], 0, sizeof(struct totem_ip_address)); } totem_config->interfaces[i].member_count = 0; } iter = icmap_iter_init("nodelist.node."); while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) { res = sscanf(iter_key, "nodelist.node.%u.%s", &node_pos, tmp_key); if (res != 2) { continue; } if (strcmp(tmp_key, "ring0_addr") != 0) { continue; } snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.", node_pos); iter2 = icmap_iter_init(tmp_key); while ((iter_key2 = icmap_iter_next(iter2, NULL, NULL)) != NULL) { res = sscanf(iter_key2, "nodelist.node.%u.ring%u%s", &node_pos, &ringnumber, tmp_key2); if (res != 3 || strcmp(tmp_key2, "_addr") != 0) { continue; } if (icmap_get_string(iter_key2, &node_addr_str) != CS_OK) { continue; } member_count = totem_config->interfaces[ringnumber].member_count; res = totemip_parse(&totem_config->interfaces[ringnumber].member_list[member_count], node_addr_str, totem_config->ip_version); if (res != -1) { totem_config->interfaces[ringnumber].member_count++; } free(node_addr_str); } icmap_iter_finalize(iter2); } icmap_iter_finalize(iter); + + if (reload) { + memcpy(new_interfaces, totem_config->interfaces, sizeof (struct totem_interface) * INTERFACE_MAX); + + compute_interfaces_diff(totem_config->interface_count, orig_interfaces, new_interfaces); + + free(new_interfaces); + free(orig_interfaces); + } +} + +static void nodelist_dynamic_notify( + int32_t event, + const char *key_name, + struct icmap_notify_value new_val, + struct icmap_notify_value old_val, + void *user_data) +{ + int res; + unsigned int ring_no; + unsigned int member_no; + char tmp_str[ICMAP_KEYNAME_MAXLEN]; + uint8_t reloading; + struct totem_config *totem_config = (struct totem_config *)user_data; + + /* + * If a full reload is in progress then don't do anything until it's done and + * can reconfigure it all atomically + */ + if (icmap_get_uint8("config.totemconfig_reload_in_progress", &reloading) == CS_OK && reloading) { + return ; + } + + res = sscanf(key_name, "nodelist.node.%u.ring%u%s", &member_no, &ring_no, tmp_str); + if (res != 3) + return ; + + if (strcmp(tmp_str, "_addr") != 0) { + return; + } + + put_nodelist_members_to_config(totem_config, 1); } + /* * Tries to find node (node_pos) in config nodelist which address matches any * local interface. Address can be stored in ring0_addr or if ipaddr_key_prefix is not NULL * key with prefix ipaddr_key is used (there can be multiuple of them) * This function differs * from find_local_node_in_nodelist because it doesn't need bindnetaddr, * but doesn't work when bind addr is network address (so IP must be exact * match). * * Returns 1 on success (address was found, node_pos is then correctly set) or 0 on failure. */ int totem_config_find_local_addr_in_nodelist(const char *ipaddr_key_prefix, unsigned int *node_pos) { struct list_head addrs; struct totem_ip_if_address *if_addr; icmap_iter_t iter, iter2; const char *iter_key, *iter_key2; struct list_head *list; const char *ipaddr_key; int ip_version; struct totem_ip_address node_addr; char *node_addr_str; int node_found = 0; int res = 0; char tmp_key[ICMAP_KEYNAME_MAXLEN]; if (totemip_getifaddrs(&addrs) == -1) { return 0; } ip_version = totem_config_get_ip_version(); iter = icmap_iter_init("nodelist.node."); while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) { res = sscanf(iter_key, "nodelist.node.%u.%s", node_pos, tmp_key); if (res != 2) { continue; } if (strcmp(tmp_key, "ring0_addr") != 0) { continue; } if (icmap_get_string(iter_key, &node_addr_str) != CS_OK) { continue ; } free(node_addr_str); /* * ring0_addr found -> let's iterate thru ipaddr_key_prefix */ snprintf(tmp_key, sizeof(tmp_key), "nodelist.node.%u.%s", *node_pos, (ipaddr_key_prefix != NULL ? ipaddr_key_prefix : "ring0_addr")); iter2 = icmap_iter_init(tmp_key); while ((iter_key2 = icmap_iter_next(iter2, NULL, NULL)) != NULL) { /* * ring0_addr must be exact match, not prefix */ ipaddr_key = (ipaddr_key_prefix != NULL ? iter_key2 : tmp_key); if (icmap_get_string(ipaddr_key, &node_addr_str) != CS_OK) { continue ; } if (totemip_parse(&node_addr, node_addr_str, ip_version) == -1) { free(node_addr_str); continue ; } free(node_addr_str); /* * Try to match ip with if_addrs */ node_found = 0; for (list = addrs.next; list != &addrs; list = list->next) { if_addr = list_entry(list, struct totem_ip_if_address, list); if (totemip_equal(&node_addr, &if_addr->ip_addr)) { node_found = 1; break; } } if (node_found) { break ; } } icmap_iter_finalize(iter2); if (node_found) { break ; } } icmap_iter_finalize(iter); totemip_freeifaddrs(&addrs); return (node_found); } static void config_convert_nodelist_to_interface(struct totem_config *totem_config) { int res = 0; unsigned int node_pos; char tmp_key[ICMAP_KEYNAME_MAXLEN]; char tmp_key2[ICMAP_KEYNAME_MAXLEN]; char *node_addr_str; unsigned int ringnumber = 0; icmap_iter_t iter; const char *iter_key; if (totem_config_find_local_addr_in_nodelist(NULL, &node_pos)) { /* * We found node, so create interface section */ snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.", node_pos); iter = icmap_iter_init(tmp_key); while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) { res = sscanf(iter_key, "nodelist.node.%u.ring%u%s", &node_pos, &ringnumber, tmp_key2); if (res != 3 || strcmp(tmp_key2, "_addr") != 0) { continue ; } if (icmap_get_string(iter_key, &node_addr_str) != CS_OK) { continue; } snprintf(tmp_key2, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.bindnetaddr", ringnumber); icmap_set_string(tmp_key2, node_addr_str); free(node_addr_str); } icmap_iter_finalize(iter); } } extern int totem_config_read ( struct totem_config *totem_config, const char **error_string, uint64_t *warnings) { int res = 0; char *str; unsigned int ringnumber = 0; int member_count = 0; icmap_iter_t iter, member_iter; const char *iter_key; const char *member_iter_key; char ringnumber_key[ICMAP_KEYNAME_MAXLEN]; char tmp_key[ICMAP_KEYNAME_MAXLEN]; uint8_t u8; uint16_t u16; char *cluster_name = NULL; int i; int local_node_pos; int nodeid_set; *warnings = 0; memset (totem_config, 0, sizeof (struct totem_config)); totem_config->interfaces = malloc (sizeof (struct totem_interface) * INTERFACE_MAX); if (totem_config->interfaces == 0) { *error_string = "Out of memory trying to allocate ethernet interface storage area"; return -1; } memset (totem_config->interfaces, 0, sizeof (struct totem_interface) * INTERFACE_MAX); strcpy (totem_config->rrp_mode, "none"); icmap_get_uint32("totem.version", (uint32_t *)&totem_config->version); if (totem_get_crypto(totem_config) != 0) { *error_string = "crypto_cipher requires crypto_hash with value other than none"; return -1; } if (icmap_get_string("totem.rrp_mode", &str) == CS_OK) { if (strlen(str) >= TOTEM_RRP_MODE_BYTES) { *error_string = "totem.rrp_mode is too long"; free(str); return -1; } strcpy (totem_config->rrp_mode, str); free(str); } icmap_get_uint32("totem.nodeid", &totem_config->node_id); totem_config->clear_node_high_bit = 0; if (icmap_get_string("totem.clear_node_high_bit", &str) == CS_OK) { if (strcmp (str, "yes") == 0) { totem_config->clear_node_high_bit = 1; } free(str); } icmap_get_uint32("totem.threads", &totem_config->threads); icmap_get_uint32("totem.netmtu", &totem_config->net_mtu); if (icmap_get_string("totem.cluster_name", &cluster_name) != CS_OK) { cluster_name = NULL; } totem_config->ip_version = totem_config_get_ip_version(); if (icmap_get_string("totem.interface.0.bindnetaddr", &str) != CS_OK) { /* * We were not able to find ring 0 bindnet addr. Try to use nodelist informations */ config_convert_nodelist_to_interface(totem_config); } else { free(str); } /* * Broadcast option is global but set in interface section, * so reset before processing interfaces. */ totem_config->broadcast_use = 0; iter = icmap_iter_init("totem.interface."); while ((iter_key = icmap_iter_next(iter, NULL, NULL)) != NULL) { res = sscanf(iter_key, "totem.interface.%[^.].%s", ringnumber_key, tmp_key); if (res != 2) { continue; } if (strcmp(tmp_key, "bindnetaddr") != 0) { continue; } member_count = 0; ringnumber = atoi(ringnumber_key); if (ringnumber >= INTERFACE_MAX) { free(cluster_name); snprintf (error_string_response, sizeof(error_string_response), "parse error in config: interface ring number %u is bigger than allowed maximum %u\n", ringnumber, INTERFACE_MAX - 1); *error_string = error_string_response; return -1; } /* * Get the bind net address */ if (icmap_get_string(iter_key, &str) == CS_OK) { res = totemip_parse (&totem_config->interfaces[ringnumber].bindnet, str, totem_config->interfaces[ringnumber].mcast_addr.family); free(str); } /* * Get interface multicast address */ snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.mcastaddr", ringnumber); if (icmap_get_string(tmp_key, &str) == CS_OK) { res = totemip_parse (&totem_config->interfaces[ringnumber].mcast_addr, str, totem_config->ip_version); free(str); } else { /* * User not specified address -> autogenerate one from cluster_name key * (if available) */ res = get_cluster_mcast_addr (cluster_name, &totem_config->interfaces[ringnumber].bindnet, ringnumber, totem_config->ip_version, &totem_config->interfaces[ringnumber].mcast_addr); } snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.broadcast", ringnumber); if (icmap_get_string(tmp_key, &str) == CS_OK) { if (strcmp (str, "yes") == 0) { totem_config->broadcast_use = 1; } free(str); } /* * Get mcast port */ snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.mcastport", ringnumber); if (icmap_get_uint16(tmp_key, &totem_config->interfaces[ringnumber].ip_port) != CS_OK) { if (totem_config->broadcast_use) { totem_config->interfaces[ringnumber].ip_port = DEFAULT_PORT + (2 * ringnumber); } else { totem_config->interfaces[ringnumber].ip_port = DEFAULT_PORT; } } /* * Get the TTL */ totem_config->interfaces[ringnumber].ttl = 1; snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.ttl", ringnumber); if (icmap_get_uint8(tmp_key, &u8) == CS_OK) { totem_config->interfaces[ringnumber].ttl = u8; } snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.member.", ringnumber); member_iter = icmap_iter_init(tmp_key); while ((member_iter_key = icmap_iter_next(member_iter, NULL, NULL)) != NULL) { if (member_count == 0) { if (icmap_get_string("nodelist.node.0.ring0_addr", &str) == CS_OK) { free(str); *warnings |= TOTEM_CONFIG_WARNING_MEMBERS_IGNORED; break; } else { *warnings |= TOTEM_CONFIG_WARNING_MEMBERS_DEPRECATED; } } if (icmap_get_string(member_iter_key, &str) == CS_OK) { res = totemip_parse (&totem_config->interfaces[ringnumber].member_list[member_count++], str, totem_config->ip_version); } } icmap_iter_finalize(member_iter); totem_config->interfaces[ringnumber].member_count = member_count; totem_config->interface_count++; } icmap_iter_finalize(iter); /* * Use broadcast is global, so if set, make sure to fill mcast addr correctly */ if (totem_config->broadcast_use) { for (ringnumber = 0; ringnumber < totem_config->interface_count; ringnumber++) { totemip_parse (&totem_config->interfaces[ringnumber].mcast_addr, "255.255.255.255", 0); } } /* * Store automatically generated items back to icmap */ for (i = 0; i < totem_config->interface_count; i++) { snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.mcastaddr", i); if (icmap_get_string(tmp_key, &str) == CS_OK) { free(str); } else { str = (char *)totemip_print(&totem_config->interfaces[i].mcast_addr); icmap_set_string(tmp_key, str); } snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "totem.interface.%u.mcastport", i); if (icmap_get_uint16(tmp_key, &u16) != CS_OK) { icmap_set_uint16(tmp_key, totem_config->interfaces[i].ip_port); } } totem_config->transport_number = TOTEM_TRANSPORT_UDP; if (icmap_get_string("totem.transport", &str) == CS_OK) { if (strcmp (str, "udpu") == 0) { totem_config->transport_number = TOTEM_TRANSPORT_UDPU; } if (strcmp (str, "iba") == 0) { totem_config->transport_number = TOTEM_TRANSPORT_RDMA; } free(str); } free(cluster_name); /* * Check existence of nodelist */ if (icmap_get_string("nodelist.node.0.ring0_addr", &str) == CS_OK) { free(str); /* * find local node */ local_node_pos = find_local_node_in_nodelist(totem_config); if (local_node_pos != -1) { icmap_set_uint32("nodelist.local_node_pos", local_node_pos); snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.nodeid", local_node_pos); nodeid_set = (totem_config->node_id != 0); if (icmap_get_uint32(tmp_key, &totem_config->node_id) == CS_OK && nodeid_set) { *warnings |= TOTEM_CONFIG_WARNING_TOTEM_NODEID_IGNORED; } /* * Make localnode ring0_addr read only, so we can be sure that local * node never changes. If rebinding to other IP would be in future * supported, this must be changed and handled properly! */ snprintf(tmp_key, ICMAP_KEYNAME_MAXLEN, "nodelist.node.%u.ring0_addr", local_node_pos); icmap_set_ro_access(tmp_key, 0, 1); icmap_set_ro_access("nodelist.local_node_pos", 0, 1); } - put_nodelist_members_to_config(totem_config); + put_nodelist_members_to_config(totem_config, 0); } /* * Get things that might change in the future (and can depend on totem_config->interfaces); */ totem_volatile_config_read(totem_config, NULL); icmap_set_uint8("config.totemconfig_reload_in_progress", 0); add_totem_config_notification(totem_config); return 0; } int totem_config_validate ( struct totem_config *totem_config, const char **error_string) { static char local_error_reason[512]; char parse_error[512]; const char *error_reason = local_error_reason; int i, j; unsigned int interface_max = INTERFACE_MAX; unsigned int port1, port2; if (totem_config->interface_count == 0) { error_reason = "No interfaces defined"; goto parse_error; } for (i = 0; i < totem_config->interface_count; i++) { /* * Some error checking of parsed data to make sure its valid */ struct totem_ip_address null_addr; memset (&null_addr, 0, sizeof (struct totem_ip_address)); if ((totem_config->transport_number == 0) && memcmp (&totem_config->interfaces[i].mcast_addr, &null_addr, sizeof (struct totem_ip_address)) == 0) { error_reason = "No multicast address specified"; goto parse_error; } if (totem_config->interfaces[i].ip_port == 0) { error_reason = "No multicast port specified"; goto parse_error; } if (totem_config->interfaces[i].ttl > 255) { error_reason = "Invalid TTL (should be 0..255)"; goto parse_error; } if (totem_config->transport_number != TOTEM_TRANSPORT_UDP && totem_config->interfaces[i].ttl != 1) { error_reason = "Can only set ttl on multicast transport types"; goto parse_error; } if (totem_config->interfaces[i].mcast_addr.family == AF_INET6 && totem_config->node_id == 0) { error_reason = "An IPV6 network requires that a node ID be specified."; goto parse_error; } if (totem_config->broadcast_use == 0 && totem_config->transport_number == TOTEM_TRANSPORT_UDP) { if (totem_config->interfaces[i].mcast_addr.family != totem_config->interfaces[i].bindnet.family) { error_reason = "Multicast address family does not match bind address family"; goto parse_error; } if (totemip_is_mcast (&totem_config->interfaces[i].mcast_addr) != 0) { error_reason = "mcastaddr is not a correct multicast address."; goto parse_error; } } if (totem_config->interfaces[0].bindnet.family != totem_config->interfaces[i].bindnet.family) { error_reason = "Not all bind address belong to the same IP family"; goto parse_error; } /* * Ensure mcast address/port differs */ if (totem_config->transport_number == TOTEM_TRANSPORT_UDP) { for (j = i + 1; j < totem_config->interface_count; j++) { port1 = totem_config->interfaces[i].ip_port; port2 = totem_config->interfaces[j].ip_port; if (totemip_equal(&totem_config->interfaces[i].mcast_addr, &totem_config->interfaces[j].mcast_addr) && (((port1 > port2 ? port1 : port2) - (port1 < port2 ? port1 : port2)) <= 1)) { error_reason = "Interfaces multicast address/port pair must differ"; goto parse_error; } } } } if (totem_config->version != 2) { error_reason = "This totem parser can only parse version 2 configurations."; goto parse_error; } if (totem_volatile_config_validate(totem_config, error_string) == -1) { return (-1); } /* * RRP values validation */ if (strcmp (totem_config->rrp_mode, "none") && strcmp (totem_config->rrp_mode, "active") && strcmp (totem_config->rrp_mode, "passive")) { snprintf (local_error_reason, sizeof(local_error_reason), "The RRP mode \"%s\" specified is invalid. It must be none, active, or passive.\n", totem_config->rrp_mode); goto parse_error; } if (strcmp (totem_config->rrp_mode, "none") == 0) { interface_max = 1; } if (interface_max < totem_config->interface_count) { snprintf (parse_error, sizeof(parse_error), "%d is too many configured interfaces for the rrp_mode setting %s.", totem_config->interface_count, totem_config->rrp_mode); error_reason = parse_error; goto parse_error; } if (totem_config->net_mtu == 0) { totem_config->net_mtu = 1500; } return 0; parse_error: snprintf (error_string_response, sizeof(error_string_response), "parse error in config: %s\n", error_reason); *error_string = error_string_response; return (-1); } static int read_keyfile ( const char *key_location, struct totem_config *totem_config, const char **error_string) { int fd; int res; ssize_t expected_key_len = sizeof (totem_config->private_key); int saved_errno; char error_str[100]; const char *error_ptr; fd = open (key_location, O_RDONLY); if (fd == -1) { error_ptr = qb_strerror_r(errno, error_str, sizeof(error_str)); snprintf (error_string_response, sizeof(error_string_response), "Could not open %s: %s\n", key_location, error_ptr); goto parse_error; } res = read (fd, totem_config->private_key, expected_key_len); saved_errno = errno; close (fd); if (res == -1) { error_ptr = qb_strerror_r (saved_errno, error_str, sizeof(error_str)); snprintf (error_string_response, sizeof(error_string_response), "Could not read %s: %s\n", key_location, error_ptr); goto parse_error; } totem_config->private_key_len = expected_key_len; if (res != expected_key_len) { snprintf (error_string_response, sizeof(error_string_response), "Could only read %d bits of 1024 bits from %s.\n", res * 8, key_location); goto parse_error; } return 0; parse_error: *error_string = error_string_response; return (-1); } int totem_config_keyread ( struct totem_config *totem_config, const char **error_string) { int got_key = 0; char *key_location = NULL; int res; size_t key_len; memset (totem_config->private_key, 0, 128); totem_config->private_key_len = 128; if (strcmp(totem_config->crypto_cipher_type, "none") == 0 && strcmp(totem_config->crypto_hash_type, "none") == 0) { return (0); } /* cmap may store the location of the key file */ if (icmap_get_string("totem.keyfile", &key_location) == CS_OK) { res = read_keyfile(key_location, totem_config, error_string); free(key_location); if (res) { goto key_error; } got_key = 1; } else { /* Or the key itself may be in the cmap */ if (icmap_get("totem.key", NULL, &key_len, NULL) == CS_OK) { if (key_len > sizeof (totem_config->private_key)) { sprintf(error_string_response, "key is too long"); goto key_error; } if (icmap_get("totem.key", totem_config->private_key, &key_len, NULL) == CS_OK) { totem_config->private_key_len = key_len; got_key = 1; } else { sprintf(error_string_response, "can't store private key"); goto key_error; } } } /* In desperation we read the default filename */ if (!got_key) { const char *filename = getenv("COROSYNC_TOTEM_AUTHKEY_FILE"); if (!filename) filename = COROSYSCONFDIR "/authkey"; res = read_keyfile(filename, totem_config, error_string); if (res) goto key_error; } return (0); key_error: *error_string = error_string_response; return (-1); } static void debug_dump_totem_config(const struct totem_config *totem_config) { log_printf(LOGSYS_LEVEL_DEBUG, "Token Timeout (%d ms) retransmit timeout (%d ms)", totem_config->token_timeout, totem_config->token_retransmit_timeout); log_printf(LOGSYS_LEVEL_DEBUG, "token hold (%d ms) retransmits before loss (%d retrans)", totem_config->token_hold_timeout, totem_config->token_retransmits_before_loss_const); log_printf(LOGSYS_LEVEL_DEBUG, "join (%d ms) send_join (%d ms) consensus (%d ms) merge (%d ms)", totem_config->join_timeout, totem_config->send_join_timeout, totem_config->consensus_timeout, totem_config->merge_timeout); log_printf(LOGSYS_LEVEL_DEBUG, "downcheck (%d ms) fail to recv const (%d msgs)", totem_config->downcheck_timeout, totem_config->fail_to_recv_const); log_printf(LOGSYS_LEVEL_DEBUG, "seqno unchanged const (%d rotations) Maximum network MTU %d", totem_config->seqno_unchanged_const, totem_config->net_mtu); log_printf(LOGSYS_LEVEL_DEBUG, "window size per rotation (%d messages) maximum messages per rotation (%d messages)", totem_config->window_size, totem_config->max_messages); log_printf(LOGSYS_LEVEL_DEBUG, "missed count const (%d messages)", totem_config->miss_count_const); log_printf(LOGSYS_LEVEL_DEBUG, "RRP token expired timeout (%d ms)", totem_config->rrp_token_expired_timeout); log_printf(LOGSYS_LEVEL_DEBUG, "RRP token problem counter (%d ms)", totem_config->rrp_problem_count_timeout); log_printf(LOGSYS_LEVEL_DEBUG, "RRP threshold (%d problem count)", totem_config->rrp_problem_count_threshold); log_printf(LOGSYS_LEVEL_DEBUG, "RRP multicast threshold (%d problem count)", totem_config->rrp_problem_count_mcast_threshold); log_printf(LOGSYS_LEVEL_DEBUG, "RRP automatic recovery check timeout (%d ms)", totem_config->rrp_autorecovery_check_timeout); log_printf(LOGSYS_LEVEL_DEBUG, "RRP mode set to %s.", totem_config->rrp_mode); log_printf(LOGSYS_LEVEL_DEBUG, "heartbeat_failures_allowed (%d)", totem_config->heartbeat_failures_allowed); log_printf(LOGSYS_LEVEL_DEBUG, "max_network_delay (%d ms)", totem_config->max_network_delay); } static void totem_change_notify( int32_t event, const char *key_name, struct icmap_notify_value new_val, struct icmap_notify_value old_val, void *user_data) { struct totem_config *totem_config = (struct totem_config *)user_data; uint32_t *param; uint8_t reloading; const char *deleted_key = NULL; const char *error_string; /* * If a full reload is in progress then don't do anything until it's done and * can reconfigure it all atomically */ if (icmap_get_uint8("config.reload_in_progress", &reloading) == CS_OK && reloading) return; param = totem_get_param_by_name((struct totem_config *)user_data, key_name); /* * Process change only if changed key is found in totem_config (-> param is not NULL) * or for special key token_coefficient. token_coefficient key is not stored in * totem_config, but it is used for computation of token timeout. */ if (!param && strcmp(key_name, "totem.token_coefficient") != 0) return; /* * Values other than UINT32 are not supported, or needed (yet) */ switch (event) { case ICMAP_TRACK_DELETE: deleted_key = key_name; break; case ICMAP_TRACK_ADD: case ICMAP_TRACK_MODIFY: deleted_key = NULL; break; default: break; } totem_volatile_config_read (totem_config, deleted_key); log_printf(LOGSYS_LEVEL_DEBUG, "Totem related config key changed. Dumping actual totem config."); debug_dump_totem_config(totem_config); if (totem_volatile_config_validate(totem_config, &error_string) == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); /* * TODO: Consider corosync exit and/or load defaults for volatile * values. For now, log error seems to be enough */ } } static void totem_reload_notify( int32_t event, const char *key_name, struct icmap_notify_value new_val, struct icmap_notify_value old_val, void *user_data) { struct totem_config *totem_config = (struct totem_config *)user_data; uint32_t local_node_pos; const char *error_string; /* Reload has completed */ if (*(uint8_t *)new_val.data == 0) { - put_nodelist_members_to_config (totem_config); + put_nodelist_members_to_config (totem_config, 1); totem_volatile_config_read (totem_config, NULL); log_printf(LOGSYS_LEVEL_DEBUG, "Configuration reloaded. Dumping actual totem config."); debug_dump_totem_config(totem_config); if (totem_volatile_config_validate(totem_config, &error_string) == -1) { log_printf (LOGSYS_LEVEL_ERROR, "%s", error_string); /* * TODO: Consider corosync exit and/or load defaults for volatile * values. For now, log error seems to be enough */ } /* Reinstate the local_node_pos */ local_node_pos = find_local_node_in_nodelist(totem_config); if (local_node_pos != -1) { icmap_set_uint32("nodelist.local_node_pos", local_node_pos); } icmap_set_uint8("config.totemconfig_reload_in_progress", 0); } else { icmap_set_uint8("config.totemconfig_reload_in_progress", 1); } } static void add_totem_config_notification(struct totem_config *totem_config) { icmap_track_t icmap_track; icmap_track_add("totem.", ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX, totem_change_notify, totem_config, &icmap_track); icmap_track_add("config.reload_in_progress", ICMAP_TRACK_ADD | ICMAP_TRACK_MODIFY, totem_reload_notify, totem_config, &icmap_track); + + icmap_track_add("nodelist.node.", + ICMAP_TRACK_ADD | ICMAP_TRACK_DELETE | ICMAP_TRACK_MODIFY | ICMAP_TRACK_PREFIX, + nodelist_dynamic_notify, + (void *)totem_config, + &icmap_track); }