diff --git a/src/sbd-inquisitor.c b/src/sbd-inquisitor.c index 664e7d9..c459cc6 100644 --- a/src/sbd-inquisitor.c +++ b/src/sbd-inquisitor.c @@ -1,834 +1,950 @@ /* * Copyright (C) 2013 Lars Marowsky-Bree * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #include "sbd.h" #define LOCKSTRLEN 11 static struct servants_list_item *servants_leader = NULL; int check_pcmk = 0; int servant_count = 0; int servant_restart_interval = 5; int servant_restart_count = 1; int start_mode = 0; char* pidfile = NULL; void recruit_servant(const char *devname, pid_t pid) { struct servants_list_item *s = servants_leader; struct servants_list_item *newbie; newbie = malloc(sizeof(*newbie)); if (!newbie) { fprintf(stderr, "malloc failed in recruit_servant.\n"); exit(1); } memset(newbie, 0, sizeof(*newbie)); newbie->devname = strdup(devname); newbie->pid = pid; newbie->first_start = 1; if (!s) { servants_leader = newbie; } else { while (s->next) s = s->next; s->next = newbie; } servant_count++; } int assign_servant(const char* devname, functionp_t functionp, int mode, const void* argp) { pid_t pid = 0; int rc = 0; pid = fork(); if (pid == 0) { /* child */ maximize_priority(); rc = (*functionp)(devname, mode, argp); if (rc == -1) exit(1); else exit(0); } else if (pid != -1) { /* parent */ return pid; } else { cl_log(LOG_ERR,"Failed to fork servant"); exit(1); } } struct servants_list_item *lookup_servant_by_dev(const char *devname) { struct servants_list_item *s; for (s = servants_leader; s; s = s->next) { if (strncasecmp(s->devname, devname, strlen(s->devname))) break; } return s; } struct servants_list_item *lookup_servant_by_pid(pid_t pid) { struct servants_list_item *s; for (s = servants_leader; s; s = s->next) { if (s->pid == pid) break; } return s; } int check_all_dead(void) { struct servants_list_item *s; int r = 0; union sigval svalue; for (s = servants_leader; s; s = s->next) { if (s->pid != 0) { r = sigqueue(s->pid, 0, svalue); if (r == -1 && errno == ESRCH) continue; return 0; } } return 1; } void servant_start(struct servants_list_item *s) { int r = 0; union sigval svalue; if (s->pid != 0) { r = sigqueue(s->pid, 0, svalue); if ((r != -1 || errno != ESRCH)) return; } s->restarts++; if (strcmp("pcmk",s->devname) == 0) { DBGLOG(LOG_INFO, "Starting Pacemaker servant"); s->pid = assign_servant(s->devname, servant_pcmk, start_mode, NULL); } else { #if SUPPORT_SHARED_DISK DBGLOG(LOG_INFO, "Starting servant for device %s", s->devname); s->pid = assign_servant(s->devname, servant, start_mode, s); #else cl_log(LOG_ERR, "Shared disk functionality not supported"); return; #endif } clock_gettime(CLOCK_MONOTONIC, &s->t_started); return; } void servants_start(void) { struct servants_list_item *s; for (s = servants_leader; s; s = s->next) { s->restarts = 0; servant_start(s); } } void servants_kill(void) { struct servants_list_item *s; union sigval svalue; for (s = servants_leader; s; s = s->next) { if (s->pid != 0) sigqueue(s->pid, SIGKILL, svalue); } } inline void cleanup_servant_by_pid(pid_t pid) { struct servants_list_item* s; s = lookup_servant_by_pid(pid); if (s) { cl_log(LOG_WARNING, "Servant for %s (pid: %i) has terminated", s->devname, s->pid); s->pid = 0; } else { /* This most likely is a stray signal from somewhere, or * a SIGCHLD for a process that has previously * explicitly disconnected. */ DBGLOG(LOG_INFO, "cleanup_servant: Nothing known about pid %i", pid); } } int inquisitor_decouple(void) { pid_t ppid = getppid(); union sigval signal_value; /* During start-up, we only arm the watchdog once we've got * quorum at least once. */ if (watchdog_use) { if (watchdog_init() < 0) { return -1; } } if (ppid > 1) { sigqueue(ppid, SIG_LIVENESS, signal_value); } return 0; } static int sbd_lock_running(long pid) { int rc = 0; long mypid; int running = 0; char proc_path[PATH_MAX], exe_path[PATH_MAX], myexe_path[PATH_MAX]; /* check if pid is running */ if (kill(pid, 0) < 0 && errno == ESRCH) { goto bail; } #ifndef HAVE_PROC_PID return 1; #endif /* check to make sure pid hasn't been reused by another process */ snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", pid); rc = readlink(proc_path, exe_path, PATH_MAX-1); if(rc < 0) { cl_perror("Could not read from %s", proc_path); goto bail; } exe_path[rc] = 0; mypid = (unsigned long) getpid(); snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", mypid); rc = readlink(proc_path, myexe_path, PATH_MAX-1); if(rc < 0) { cl_perror("Could not read from %s", proc_path); goto bail; } myexe_path[rc] = 0; if(strcmp(exe_path, myexe_path) == 0) { running = 1; } bail: return running; } static int sbd_lock_pidfile(const char *filename) { char lf_name[256], tf_name[256], buf[LOCKSTRLEN+1]; int fd; long pid, mypid; int rc; struct stat sbuf; if (filename == NULL) { errno = EFAULT; return -1; } mypid = (unsigned long) getpid(); snprintf(lf_name, sizeof(lf_name), "%s",filename); snprintf(tf_name, sizeof(tf_name), "%s.%lu", filename, mypid); if ((fd = open(lf_name, O_RDONLY)) >= 0) { if (fstat(fd, &sbuf) >= 0 && sbuf.st_size < LOCKSTRLEN) { sleep(1); /* if someone was about to create one, * give'm a sec to do so * Though if they follow our protocol, * this won't happen. They should really * put the pid in, then link, not the * other way around. */ } if (read(fd, buf, sizeof(buf)) < 1) { /* lockfile empty -> rm it and go on */; } else { if (sscanf(buf, "%lu", &pid) < 1) { /* lockfile screwed up -> rm it and go on */ } else { if (pid > 1 && (getpid() != pid) && sbd_lock_running(pid)) { /* is locked by existing process * -> give up */ close(fd); return -1; } else { /* stale lockfile -> rm it and go on */ } } } unlink(lf_name); close(fd); } if ((fd = open(tf_name, O_CREAT | O_WRONLY | O_EXCL, 0644)) < 0) { /* Hmmh, why did we fail? Anyway, nothing we can do about it */ return -3; } /* Slight overkill with the %*d format ;-) */ snprintf(buf, sizeof(buf), "%*lu\n", LOCKSTRLEN-1, mypid); if (write(fd, buf, LOCKSTRLEN) != LOCKSTRLEN) { /* Again, nothing we can do about this */ rc = -3; close(fd); goto out; } close(fd); switch (link(tf_name, lf_name)) { case 0: if (stat(tf_name, &sbuf) < 0) { /* something weird happened */ rc = -3; break; } if (sbuf.st_nlink < 2) { /* somehow, it didn't get through - NFS trouble? */ rc = -2; break; } rc = 0; break; case EEXIST: rc = -1; break; default: rc = -3; } out: unlink(tf_name); return rc; } /* * Unlock a file (remove its lockfile) * do we need to check, if its (still) ours? No, IMHO, if someone else * locked our line, it's his fault -tho * returns 0 on success * <0 if some failure occured */ static int sbd_unlock_pidfile(const char *filename) { char lf_name[256]; if (filename == NULL) { errno = EFAULT; return -1; } snprintf(lf_name, sizeof(lf_name), "%s", filename); return unlink(lf_name); } int quorum_read(int good_servants) { if (servant_count >= 3) return (good_servants > servant_count/2); else return (good_servants >= 1); } void inquisitor_child(void) { int sig, pid; sigset_t procmask; siginfo_t sinfo; int status; struct timespec timeout; int exiting = 0; int decoupled = 0; int pcmk_healthy = 0; int pcmk_override = 0; time_t latency; struct timespec t_last_tickle, t_now; struct servants_list_item* s; if (debug_mode) { cl_log(LOG_ERR, "DEBUG MODE IS ACTIVE - DO NOT RUN IN PRODUCTION!"); } set_proc_title("sbd: inquisitor"); if (pidfile) { if (sbd_lock_pidfile(pidfile) < 0) { exit(1); } } sigemptyset(&procmask); sigaddset(&procmask, SIGCHLD); + sigaddset(&procmask, SIGTERM); sigaddset(&procmask, SIG_LIVENESS); sigaddset(&procmask, SIG_EXITREQ); sigaddset(&procmask, SIG_TEST); sigaddset(&procmask, SIG_IO_FAIL); sigaddset(&procmask, SIG_PCMK_UNHEALTHY); sigaddset(&procmask, SIG_RESTART); sigaddset(&procmask, SIGUSR1); sigaddset(&procmask, SIGUSR2); sigprocmask(SIG_BLOCK, &procmask, NULL); servants_start(); timeout.tv_sec = timeout_loop; timeout.tv_nsec = 0; clock_gettime(CLOCK_MONOTONIC, &t_last_tickle); while (1) { int good_servants = 0; sig = sigtimedwait(&procmask, &sinfo, &timeout); clock_gettime(CLOCK_MONOTONIC, &t_now); - if (sig == SIG_EXITREQ) { + if (sig == SIG_EXITREQ || sig == SIGTERM) { servants_kill(); watchdog_close(true); exiting = 1; } else if (sig == SIGCHLD) { while ((pid = waitpid(-1, &status, WNOHANG))) { if (pid == -1 && errno == ECHILD) { break; } else { cleanup_servant_by_pid(pid); } } } else if (sig == SIG_PCMK_UNHEALTHY) { s = lookup_servant_by_pid(sinfo.si_pid); if (s && strcmp(s->devname, "pcmk") == 0) { if (pcmk_healthy != 0) { cl_log(LOG_WARNING, "Pacemaker health check: UNHEALTHY"); } pcmk_healthy = 0; clock_gettime(CLOCK_MONOTONIC, &s->t_last); } else { cl_log(LOG_WARNING, "Ignoring SIG_PCMK_UNHEALTHY from unknown source"); } } else if (sig == SIG_IO_FAIL) { s = lookup_servant_by_pid(sinfo.si_pid); if (s) { DBGLOG(LOG_INFO, "Servant for %s requests to be disowned", s->devname); cleanup_servant_by_pid(sinfo.si_pid); } } else if (sig == SIG_LIVENESS) { s = lookup_servant_by_pid(sinfo.si_pid); if (s) { if (strcmp(s->devname, "pcmk") == 0) { if (pcmk_healthy != 1) { cl_log(LOG_INFO, "Pacemaker health check: OK"); } pcmk_healthy = 1; }; s->first_start = 0; clock_gettime(CLOCK_MONOTONIC, &s->t_last); } } else if (sig == SIG_TEST) { } else if (sig == SIGUSR1) { if (exiting) continue; servants_start(); } if (exiting) { if (check_all_dead()) { if (pidfile) { sbd_unlock_pidfile(pidfile); } exit(0); } else continue; } good_servants = 0; for (s = servants_leader; s; s = s->next) { int age = t_now.tv_sec - s->t_last.tv_sec; if (!s->t_last.tv_sec) continue; if (age < (int)(timeout_io+timeout_loop)) { if (strcmp(s->devname, "pcmk") != 0) { good_servants++; } s->outdated = 0; } else if (!s->outdated) { if (strcmp(s->devname, "pcmk") == 0) { /* If the state is outdated, we * override the last reported * state */ pcmk_healthy = 0; cl_log(LOG_WARNING, "Pacemaker state outdated (age: %d)", age); } else if (!s->restart_blocked) { cl_log(LOG_WARNING, "Servant for %s outdated (age: %d)", s->devname, age); } s->outdated = 1; } } if (quorum_read(good_servants) || pcmk_healthy) { if (!decoupled) { if (inquisitor_decouple() < 0) { servants_kill(); exiting = 1; continue; } else { decoupled = 1; } } if (!quorum_read(good_servants)) { if (!pcmk_override) { cl_log(LOG_WARNING, "Majority of devices lost - surviving on pacemaker"); pcmk_override = 1; /* Just to ensure the message is only logged once */ } } else { pcmk_override = 0; } watchdog_tickle(); clock_gettime(CLOCK_MONOTONIC, &t_last_tickle); } /* Note that this can actually be negative, since we set * last_tickle after we set now. */ latency = t_now.tv_sec - t_last_tickle.tv_sec; if (timeout_watchdog && (latency > (int)timeout_watchdog)) { if (!decoupled) { /* We're still being watched by our * parent. We don't fence, but exit. */ cl_log(LOG_ERR, "SBD: Not enough votes to proceed. Aborting start-up."); servants_kill(); exiting = 1; continue; } if (debug_mode < 2) { /* At level 2 or above, we do nothing, but expect * things to eventually return to * normal. */ do_reset(); } else { cl_log(LOG_ERR, "SBD: DEBUG MODE: Would have fenced due to timeout!"); } } if (timeout_watchdog_warn && (latency > (int)timeout_watchdog_warn)) { cl_log(LOG_WARNING, "Latency: No liveness for %d s exceeds threshold of %d s (healthy servants: %d)", (int)latency, (int)timeout_watchdog_warn, good_servants); } for (s = servants_leader; s; s = s->next) { int age = t_now.tv_sec - s->t_started.tv_sec; if (age > servant_restart_interval) { s->restarts = 0; s->restart_blocked = 0; } if (servant_restart_count && (s->restarts >= servant_restart_count) && !s->restart_blocked) { if (servant_restart_count > 1) { cl_log(LOG_WARNING, "Max retry count (%d) reached: not restarting servant for %s", (int)servant_restart_count, s->devname); } s->restart_blocked = 1; } if (!s->restart_blocked) { servant_start(s); } } } /* not reached */ exit(0); } int inquisitor(void) { int sig, pid, inquisitor_pid; int status; sigset_t procmask; siginfo_t sinfo; /* Where's the best place for sysrq init ?*/ sysrq_init(); sigemptyset(&procmask); sigaddset(&procmask, SIGCHLD); sigaddset(&procmask, SIG_LIVENESS); sigprocmask(SIG_BLOCK, &procmask, NULL); inquisitor_pid = make_daemon(); if (inquisitor_pid == 0) { inquisitor_child(); } /* We're the parent. Wait for a happy signal from our child * before we proceed - we either get "SIG_LIVENESS" when the * inquisitor has completed the first successful round, or * ECHLD when it exits with an error. */ while (1) { sig = sigwaitinfo(&procmask, &sinfo); if (sig == SIGCHLD) { while ((pid = waitpid(-1, &status, WNOHANG))) { if (pid == -1 && errno == ECHILD) { break; } /* We got here because the inquisitor * did not succeed. */ return -1; } } else if (sig == SIG_LIVENESS) { /* Inquisitor started up properly. */ return 0; } else { fprintf(stderr, "Nobody expected the spanish inquisition!\n"); continue; } } /* not reached */ return -1; } + +static int +parse_device_line(const char *line) +{ + int lpc = 0; + int last = 0; + int max = 0; + int found = 0; + + if(line) { + max = strlen(line); + } + + if (max <= 0) { + return found; + } + + crm_trace("Processing %d bytes: [%s]", max, line); + /* Skip initial whitespace */ + for (lpc = 0; lpc <= max && isspace(line[lpc]); lpc++) { + last = lpc + 1; + } + + /* Now the actual content */ + for (lpc = 0; lpc <= max; lpc++) { + int a_space = isspace(line[lpc]); + + if (a_space && lpc < max && isspace(line[lpc + 1])) { + /* fast-forward to the end of the spaces */ + + } else if (a_space || line[lpc] == ';' || line[lpc] == 0) { + int rc = 1; + char *entry = NULL; + + if (lpc != last) { + entry = calloc(1, 1 + lpc - last); + rc = sscanf(line + last, "%[^;]", entry); + } + + if (entry == NULL) { + /* Skip */ + } else if (rc != 1) { + crm_warn("Could not parse (%d %d): %s", last, lpc, line + last); + } else { + crm_trace("Adding '%s'", entry); + recruit_servant(entry, 0); + found++; + } + + free(entry); + last = lpc + 1; + } + } + return found; +} + int main(int argc, char **argv, char **envp) { int exit_status = 0; int c; int w = 0; int qb_facility; + const char *value = NULL; + int start_delay = 0; if ((cmdname = strrchr(argv[0], '/')) == NULL) { cmdname = argv[0]; } else { ++cmdname; } qb_facility = qb_log_facility2int("daemon"); qb_log_init(cmdname, qb_facility, LOG_ERR); qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_TRUE); qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_FALSE); sbd_get_uname(); + value = getenv("SBD_DEVICE"); + if(value) { +#if SUPPORT_SHARED_DISK + int devices = parse_device_line(value); + if(devices > 0) { + fprintf(stderr, "Invalid device line: %s\n", value); + exit_status = -2; + goto out; + } +#else + fprintf(stderr, "Shared disk functionality not supported\n"); + exit_status = -2; + goto out; +#endif + } + + value = getenv("SBD_PACEMAKER"); + if(value) { + check_pcmk = crm_is_true(value); + } + + value = getenv("SBD_STARTMODE"); + if(value) { + start_mode = crm_int_helper(value, NULL); + cl_log(LOG_INFO, "Start mode set to: %d", (int)start_mode); + } + + value = getenv("SBD_WATCHDOG_DEV"); + if(value) { + watchdogdev = value; + } + + value = getenv("SBD_PIDFILE"); + if(value) { + pidfile = strdup(value); + cl_log(LOG_INFO, "pidfile set to %s", pidfile); + } + + value = getenv("SBD_DELAY_START"); + if(value) { + start_delay = crm_is_true(value); + } + while ((c = getopt(argc, argv, "C:DPRTWZhvw:d:n:p:1:2:3:4:5:t:I:F:S:s:")) != -1) { switch (c) { case 'D': break; case 'Z': debug_mode++; cl_log(LOG_INFO, "Debug mode now at level %d", (int)debug_mode); break; case 'R': skip_rt = 1; cl_log(LOG_INFO, "Realtime mode deactivated."); break; case 'S': start_mode = atoi(optarg); cl_log(LOG_INFO, "Start mode set to: %d", (int)start_mode); break; case 's': timeout_startup = atoi(optarg); cl_log(LOG_INFO, "Start timeout set to: %d", (int)timeout_startup); break; case 'v': debug = 1; cl_log(LOG_INFO, "Verbose mode enabled."); break; case 'T': watchdog_set_timeout = 0; cl_log(LOG_INFO, "Setting watchdog timeout disabled; using defaults."); break; case 'W': w++; break; case 'w': - watchdogdev = strdup(optarg); + watchdogdev = optarg; break; case 'd': #if SUPPORT_SHARED_DISK recruit_servant(optarg, 0); #else fprintf(stderr, "Shared disk functionality not supported\n"); exit_status = -2; goto out; #endif break; case 'P': check_pcmk = 1; break; case 'n': local_uname = strdup(optarg); cl_log(LOG_INFO, "Overriding local hostname to %s", local_uname); break; case 'p': pidfile = strdup(optarg); cl_log(LOG_INFO, "pidfile set to %s", pidfile); break; case 'C': timeout_watchdog_crashdump = atoi(optarg); cl_log(LOG_INFO, "Setting crashdump watchdog timeout to %d", (int)timeout_watchdog_crashdump); break; case '1': timeout_watchdog = atoi(optarg); break; case '2': timeout_allocate = atoi(optarg); break; case '3': timeout_loop = atoi(optarg); break; case '4': timeout_msgwait = atoi(optarg); break; case '5': timeout_watchdog_warn = atoi(optarg); cl_log(LOG_INFO, "Setting latency warning to %d", (int)timeout_watchdog_warn); break; case 't': servant_restart_interval = atoi(optarg); cl_log(LOG_INFO, "Setting servant restart interval to %d", (int)servant_restart_interval); break; case 'I': timeout_io = atoi(optarg); cl_log(LOG_INFO, "Setting IO timeout to %d", (int)timeout_io); break; case 'F': servant_restart_count = atoi(optarg); cl_log(LOG_INFO, "Servant restart count set to %d", (int)servant_restart_count); break; case 'h': usage(); return (0); default: exit_status = -2; goto out; break; } } if (w > 0) { - watchdog_use = w % 2; - } + watchdog_use = w % 2; + + } else if(watchdogdev == NULL || strcmp(watchdogdev, "/dev/null") == 0) { + watchdog_use = 0; + } if (watchdog_use) { cl_log(LOG_INFO, "Watchdog enabled."); } else { cl_log(LOG_INFO, "Watchdog disabled."); } if (servant_count < 1 || servant_count > 3) { fprintf(stderr, "You must specify 1 to 3 devices via the -d option.\n"); exit_status = -1; goto out; } /* There must at least be one command following the options: */ if ((argc - optind) < 1) { fprintf(stderr, "Not enough arguments.\n"); exit_status = -2; goto out; } if (init_set_proc_title(argc, argv, envp) < 0) { fprintf(stderr, "Allocation of proc title failed.\n"); exit_status = -1; goto out; } maximize_priority(); #if SUPPORT_SHARED_DISK if (strcmp(argv[optind], "create") == 0) { exit_status = init_devices(servants_leader); } else if (strcmp(argv[optind], "dump") == 0) { exit_status = dump_headers(servants_leader); } else if (strcmp(argv[optind], "allocate") == 0) { exit_status = allocate_slots(argv[optind + 1], servants_leader); } else if (strcmp(argv[optind], "list") == 0) { exit_status = list_slots(servants_leader); } else if (strcmp(argv[optind], "message") == 0) { exit_status = messenger(argv[optind + 1], argv[optind + 2], servants_leader); } else if (strcmp(argv[optind], "ping") == 0) { exit_status = ping_via_slots(argv[optind + 1], servants_leader); } else if (strcmp(argv[optind], "watch") == 0) { - open_any_device(servants_leader); + if(servant_count > 0) { + /* If no devices are specified, its not an error to be unable to find one */ + open_any_device(servants_leader); + } /* We only want this to have an effect during watch right now; * pinging and fencing would be too confused */ if (check_pcmk) { recruit_servant("pcmk", 0); servant_count--; } + if(start_delay) { + unsigned long delay = get_first_msgwait(servants_leader); + + sleep(delay); + } + exit_status = inquisitor(); } else { exit_status = -2; } #else if (strcmp(argv[optind], "watch") == 0) { + /* sleep $(sbd -d "$SBD_DEVICE" dump | grep -m 1 msgwait | awk '{print $4}') 2>/dev/null */ + /* We only want this to have an effect during watch right now; * pinging and fencing would be too confused */ if (check_pcmk) { recruit_servant("pcmk", 0); servant_count--; } exit_status = inquisitor(); } #endif out: if (exit_status < 0) { if (exit_status == -2) { usage(); } else { fprintf(stderr, "sbd failed; please check the logs.\n"); } return (1); } return (0); } diff --git a/src/sbd-md.c b/src/sbd-md.c index ef65657..0e756b7 100644 --- a/src/sbd-md.c +++ b/src/sbd-md.c @@ -1,1228 +1,1254 @@ /* * Copyright (C) 2013 Lars Marowsky-Bree * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #include "sbd.h" #define SBD_MSG_EMPTY 0x00 #define SBD_MSG_TEST 0x01 #define SBD_MSG_RESET 0x02 #define SBD_MSG_OFF 0x03 #define SBD_MSG_EXIT 0x04 #define SBD_MSG_CRASHDUMP 0x05 #define SLOT_TO_SECTOR(slot) (1+slot*2) #define MBOX_TO_SECTOR(mbox) (2+mbox*2) extern int servant_count; static int servant_inform_parent = 0; /* These have to match the values in the header of the partition */ static char sbd_magic[8] = "SBD_SBD_"; static char sbd_version = 0x02; struct slot_msg_arg_t { const char* name; const char* msg; }; static signed char cmd2char(const char *cmd) { if (strcmp("clear", cmd) == 0) { return SBD_MSG_EMPTY; } else if (strcmp("test", cmd) == 0) { return SBD_MSG_TEST; } else if (strcmp("reset", cmd) == 0) { return SBD_MSG_RESET; } else if (strcmp("off", cmd) == 0) { return SBD_MSG_OFF; } else if (strcmp("exit", cmd) == 0) { return SBD_MSG_EXIT; } else if (strcmp("crashdump", cmd) == 0) { return SBD_MSG_CRASHDUMP; } return -1; } static const char* char2cmd(const char cmd) { switch (cmd) { case SBD_MSG_EMPTY: return "clear"; break; case SBD_MSG_TEST: return "test"; break; case SBD_MSG_RESET: return "reset"; break; case SBD_MSG_OFF: return "off"; break; case SBD_MSG_EXIT: return "exit"; break; case SBD_MSG_CRASHDUMP: return "crashdump"; break; default: return "undefined"; break; } } static void close_device(struct sbd_context *st) { close(st->devfd); free(st); } static struct sbd_context * open_device(const char* devname, int loglevel) { struct sbd_context *st; if (!devname) return NULL; st = malloc(sizeof(struct sbd_context)); if (!st) return NULL; memset(st, 0, sizeof(struct sbd_context)); if (io_setup(1, &st->ioctx) != 0) { cl_perror("io_setup failed"); free(st); return NULL; } st->devfd = open(devname, O_SYNC|O_RDWR|O_DIRECT); if (st->devfd == -1) { if (loglevel == LOG_DEBUG) { DBGLOG(loglevel, "Opening device %s failed.", devname); } else { cl_log(loglevel, "Opening device %s failed.", devname); } free(st); return NULL; } ioctl(st->devfd, BLKSSZGET, §or_size); if (sector_size == 0) { cl_perror("Get sector size failed.\n"); close_device(st); return NULL; } return st; } static void * sector_alloc(void) { void *x; x = valloc(sector_size); if (!x) { exit(1); } memset(x, 0, sector_size); return x; } static int sector_io(struct sbd_context *st, int sector, void *data, int rw) { struct timespec timeout; struct io_event event; struct iocb *ios[1] = { &st->io }; long r; timeout.tv_sec = timeout_io; timeout.tv_nsec = 0; memset(&st->io, 0, sizeof(struct iocb)); if (rw) { io_prep_pwrite(&st->io, st->devfd, data, sector_size, sector_size * sector); } else { io_prep_pread(&st->io, st->devfd, data, sector_size, sector_size * sector); } if (io_submit(st->ioctx, 1, ios) != 1) { cl_log(LOG_ERR, "Failed to submit IO request! (rw=%d)", rw); return -1; } errno = 0; r = io_getevents(st->ioctx, 1L, 1L, &event, &timeout); if (r < 0 ) { cl_log(LOG_ERR, "Failed to retrieve IO events (rw=%d)", rw); return -1; } else if (r < 1L) { cl_log(LOG_INFO, "Cancelling IO request due to timeout (rw=%d)", rw); r = io_cancel(st->ioctx, ios[0], &event); if (r) { DBGLOG(LOG_INFO, "Could not cancel IO request (rw=%d)", rw); /* Doesn't really matter, debugging information. */ } return -1; } else if (r > 1L) { cl_log(LOG_ERR, "More than one IO was returned (r=%ld)", r); return -1; } /* IO is happy */ if (event.res == sector_size) { return 0; } else { cl_log(LOG_ERR, "Short IO (rw=%d, res=%lu, sector_size=%d)", rw, event.res, sector_size); return -1; } } static int sector_write(struct sbd_context *st, int sector, void *data) { return sector_io(st, sector, data, 1); } static int sector_read(struct sbd_context *st, int sector, void *data) { return sector_io(st, sector, data, 0); } static int slot_read(struct sbd_context *st, int slot, struct sector_node_s *s_node) { return sector_read(st, SLOT_TO_SECTOR(slot), s_node); } static int slot_write(struct sbd_context *st, int slot, struct sector_node_s *s_node) { return sector_write(st, SLOT_TO_SECTOR(slot), s_node); } static int mbox_write(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox) { return sector_write(st, MBOX_TO_SECTOR(mbox), s_mbox); } static int mbox_read(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox) { return sector_read(st, MBOX_TO_SECTOR(mbox), s_mbox); } static int mbox_write_verify(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox) { void *data; int rc = 0; if (sector_write(st, MBOX_TO_SECTOR(mbox), s_mbox) < 0) return -1; data = sector_alloc(); if (sector_read(st, MBOX_TO_SECTOR(mbox), data) < 0) { rc = -1; goto out; } if (memcmp(s_mbox, data, sector_size) != 0) { cl_log(LOG_ERR, "Write verification failed!"); rc = -1; goto out; } rc = 0; out: free(data); return rc; } static int header_write(struct sbd_context *st, struct sector_header_s *s_header) { s_header->sector_size = htonl(s_header->sector_size); s_header->timeout_watchdog = htonl(s_header->timeout_watchdog); s_header->timeout_allocate = htonl(s_header->timeout_allocate); s_header->timeout_loop = htonl(s_header->timeout_loop); s_header->timeout_msgwait = htonl(s_header->timeout_msgwait); return sector_write(st, 0, s_header); } static int header_read(struct sbd_context *st, struct sector_header_s *s_header) { if (sector_read(st, 0, s_header) < 0) return -1; s_header->sector_size = ntohl(s_header->sector_size); s_header->timeout_watchdog = ntohl(s_header->timeout_watchdog); s_header->timeout_allocate = ntohl(s_header->timeout_allocate); s_header->timeout_loop = ntohl(s_header->timeout_loop); s_header->timeout_msgwait = ntohl(s_header->timeout_msgwait); /* This sets the global defaults: */ timeout_watchdog = s_header->timeout_watchdog; timeout_allocate = s_header->timeout_allocate; timeout_loop = s_header->timeout_loop; timeout_msgwait = s_header->timeout_msgwait; return 0; } static int valid_header(const struct sector_header_s *s_header) { if (memcmp(s_header->magic, sbd_magic, sizeof(s_header->magic)) != 0) { cl_log(LOG_ERR, "Header magic does not match."); return -1; } if (s_header->version != sbd_version) { cl_log(LOG_ERR, "Header version does not match."); return -1; } if (s_header->sector_size != sector_size) { cl_log(LOG_ERR, "Header sector size does not match."); return -1; } return 0; } static struct sector_header_s * header_get(struct sbd_context *st) { struct sector_header_s *s_header; s_header = sector_alloc(); if (header_read(st, s_header) < 0) { cl_log(LOG_ERR, "Unable to read header from device %d", st->devfd); return NULL; } if (valid_header(s_header) < 0) { cl_log(LOG_ERR, "header on device %d is not valid.", st->devfd); return NULL; } /* cl_log(LOG_INFO, "Found version %d header with %d slots", s_header->version, s_header->slots); */ return s_header; } static int header_dump(struct sbd_context *st) { struct sector_header_s *s_header; char uuid[37]; s_header = header_get(st); if (s_header == NULL) return -1; printf("Header version : %u.%u\n", s_header->version, s_header->minor_version); if (s_header->minor_version > 0) { uuid_unparse_lower(s_header->uuid, uuid); printf("UUID : %s\n", uuid); } printf("Number of slots : %u\n", s_header->slots); printf("Sector size : %lu\n", (unsigned long)s_header->sector_size); printf("Timeout (watchdog) : %lu\n", (unsigned long)s_header->timeout_watchdog); printf("Timeout (allocate) : %lu\n", (unsigned long)s_header->timeout_allocate); printf("Timeout (loop) : %lu\n", (unsigned long)s_header->timeout_loop); printf("Timeout (msgwait) : %lu\n", (unsigned long)s_header->timeout_msgwait); return 0; } static int init_device(struct sbd_context *st) { struct sector_header_s *s_header; struct sector_node_s *s_node; struct sector_mbox_s *s_mbox; struct stat s; char uuid[37]; int i; int rc = 0; s_header = sector_alloc(); s_node = sector_alloc(); s_mbox = sector_alloc(); memcpy(s_header->magic, sbd_magic, sizeof(s_header->magic)); s_header->version = sbd_version; s_header->slots = 255; s_header->sector_size = sector_size; s_header->timeout_watchdog = timeout_watchdog; s_header->timeout_allocate = timeout_allocate; s_header->timeout_loop = timeout_loop; s_header->timeout_msgwait = timeout_msgwait; s_header->minor_version = 1; uuid_generate(s_header->uuid); uuid_unparse_lower(s_header->uuid, uuid); fstat(st->devfd, &s); /* printf("st_size = %ld, st_blksize = %ld, st_blocks = %ld\n", s.st_size, s.st_blksize, s.st_blocks); */ cl_log(LOG_INFO, "Creating version %d.%d header on device %d (uuid: %s)", s_header->version, s_header->minor_version, st->devfd, uuid); fprintf(stdout, "Creating version %d.%d header on device %d (uuid: %s)\n", s_header->version, s_header->minor_version, st->devfd, uuid); if (header_write(st, s_header) < 0) { rc = -1; goto out; } cl_log(LOG_INFO, "Initializing %d slots on device %d", s_header->slots, st->devfd); fprintf(stdout, "Initializing %d slots on device %d\n", s_header->slots, st->devfd); for (i=0;i < s_header->slots;i++) { if (slot_write(st, i, s_node) < 0) { rc = -1; goto out; } if (mbox_write(st, i, s_mbox) < 0) { rc = -1; goto out; } } out: free(s_node); free(s_header); free(s_mbox); return(rc); } /* Check if there already is a slot allocated to said name; returns the * slot number. If not found, returns -1. * This is necessary because slots might not be continuous. */ static int slot_lookup(struct sbd_context *st, const struct sector_header_s *s_header, const char *name) { struct sector_node_s *s_node = NULL; int i; int rc = -1; if (!name) { cl_log(LOG_ERR, "slot_lookup(): No name specified.\n"); goto out; } s_node = sector_alloc(); for (i=0; i < s_header->slots; i++) { if (slot_read(st, i, s_node) < 0) { rc = -2; goto out; } if (s_node->in_use != 0) { if (strncasecmp(s_node->name, name, sizeof(s_node->name)) == 0) { DBGLOG(LOG_INFO, "%s owns slot %d", name, i); rc = i; goto out; } } } out: free(s_node); return rc; } static int slot_unused(struct sbd_context *st, const struct sector_header_s *s_header) { struct sector_node_s *s_node; int i; int rc = -1; s_node = sector_alloc(); for (i=0; i < s_header->slots; i++) { if (slot_read(st, i, s_node) < 0) { rc = -1; goto out; } if (s_node->in_use == 0) { rc = i; goto out; } } out: free(s_node); return rc; } static int slot_allocate(struct sbd_context *st, const char *name) { struct sector_header_s *s_header = NULL; struct sector_node_s *s_node = NULL; struct sector_mbox_s *s_mbox = NULL; int i; int rc = 0; if (!name) { cl_log(LOG_ERR, "slot_allocate(): No name specified.\n"); fprintf(stderr, "slot_allocate(): No name specified.\n"); rc = -1; goto out; } s_header = header_get(st); if (!s_header) { rc = -1; goto out; } s_node = sector_alloc(); s_mbox = sector_alloc(); while (1) { i = slot_lookup(st, s_header, name); if ((i >= 0) || (i == -2)) { /* -1 is "no slot found", in which case we * proceed to allocate a new one. * -2 is "read error during lookup", in which * case we error out too * >= 0 is "slot already allocated" */ rc = i; goto out; } i = slot_unused(st, s_header); if (i >= 0) { cl_log(LOG_INFO, "slot %d is unused - trying to own", i); fprintf(stdout, "slot %d is unused - trying to own\n", i); memset(s_node, 0, sizeof(*s_node)); s_node->in_use = 1; strncpy(s_node->name, name, sizeof(s_node->name)); if (slot_write(st, i, s_node) < 0) { rc = -1; goto out; } sleep(timeout_allocate); } else { cl_log(LOG_ERR, "No more free slots."); fprintf(stderr, "No more free slots.\n"); rc = -1; goto out; } } out: free(s_node); free(s_header); free(s_mbox); return(rc); } static int slot_list(struct sbd_context *st) { struct sector_header_s *s_header = NULL; struct sector_node_s *s_node = NULL; struct sector_mbox_s *s_mbox = NULL; int i; int rc = 0; s_header = header_get(st); if (!s_header) { rc = -1; goto out; } s_node = sector_alloc(); s_mbox = sector_alloc(); for (i=0; i < s_header->slots; i++) { if (slot_read(st, i, s_node) < 0) { rc = -1; goto out; } if (s_node->in_use > 0) { if (mbox_read(st, i, s_mbox) < 0) { rc = -1; goto out; } printf("%d\t%s\t%s\t%s\n", i, s_node->name, char2cmd(s_mbox->cmd), s_mbox->from); } } out: free(s_node); free(s_header); free(s_mbox); return rc; } static int slot_msg(struct sbd_context *st, const char *name, const char *cmd) { struct sector_header_s *s_header = NULL; struct sector_mbox_s *s_mbox = NULL; int mbox; int rc = 0; char uuid[37]; if (!name || !cmd) { cl_log(LOG_ERR, "slot_msg(): No recipient / cmd specified.\n"); rc = -1; goto out; } s_header = header_get(st); if (!s_header) { rc = -1; goto out; } if (strcmp(name, "LOCAL") == 0) { name = local_uname; } if (s_header->minor_version > 0) { uuid_unparse_lower(s_header->uuid, uuid); cl_log(LOG_INFO, "Device UUID: %s", uuid); } mbox = slot_lookup(st, s_header, name); if (mbox < 0) { cl_log(LOG_ERR, "slot_msg(): No slot found for %s.", name); rc = -1; goto out; } s_mbox = sector_alloc(); s_mbox->cmd = cmd2char(cmd); if (s_mbox->cmd < 0) { cl_log(LOG_ERR, "slot_msg(): Invalid command %s.", cmd); rc = -1; goto out; } strncpy(s_mbox->from, local_uname, sizeof(s_mbox->from)-1); cl_log(LOG_INFO, "Writing %s to node slot %s", cmd, name); if (mbox_write_verify(st, mbox, s_mbox) < -1) { rc = -1; goto out; } if (strcasecmp(cmd, "exit") != 0) { cl_log(LOG_INFO, "Messaging delay: %d", (int)timeout_msgwait); sleep(timeout_msgwait); } cl_log(LOG_INFO, "%s successfully delivered to %s", cmd, name); out: free(s_mbox); free(s_header); return rc; } static int slot_ping(struct sbd_context *st, const char *name) { struct sector_header_s *s_header = NULL; struct sector_mbox_s *s_mbox = NULL; int mbox; int waited = 0; int rc = 0; if (!name) { cl_log(LOG_ERR, "slot_ping(): No recipient specified.\n"); rc = -1; goto out; } s_header = header_get(st); if (!s_header) { rc = -1; goto out; } if (strcmp(name, "LOCAL") == 0) { name = local_uname; } mbox = slot_lookup(st, s_header, name); if (mbox < 0) { cl_log(LOG_ERR, "slot_msg(): No slot found for %s.", name); rc = -1; goto out; } s_mbox = sector_alloc(); s_mbox->cmd = SBD_MSG_TEST; strncpy(s_mbox->from, local_uname, sizeof(s_mbox->from)-1); DBGLOG(LOG_DEBUG, "Pinging node %s", name); if (mbox_write(st, mbox, s_mbox) < -1) { rc = -1; goto out; } rc = -1; while (waited <= timeout_msgwait) { if (mbox_read(st, mbox, s_mbox) < 0) break; if (s_mbox->cmd != SBD_MSG_TEST) { rc = 0; break; } sleep(1); waited++; } if (rc == 0) { cl_log(LOG_DEBUG, "%s successfully pinged.", name); } else { cl_log(LOG_ERR, "%s failed to ping.", name); } out: free(s_mbox); free(s_header); return rc; } int init_devices(struct servants_list_item *servants) { int rc = 0; struct sbd_context *st; struct servants_list_item *s; for (s = servants; s; s = s->next) { fprintf(stdout, "Initializing device %s\n", s->devname); st = open_device(s->devname, LOG_ERR); if (!st) { return -1; } rc = init_device(st); close_device(st); if (rc == -1) { fprintf(stderr, "Failed to init device %s\n", s->devname); return rc; } fprintf(stdout, "Device %s is initialized.\n", s->devname); } return 0; } static int slot_msg_wrapper(const char* devname, int mode, const void* argp) { int rc = 0; struct sbd_context *st; const struct slot_msg_arg_t* arg = (const struct slot_msg_arg_t*)argp; st = open_device(devname, LOG_WARNING); if (!st) return -1; cl_log(LOG_INFO, "Delivery process handling %s", devname); rc = slot_msg(st, arg->name, arg->msg); close_device(st); return rc; } static int slot_ping_wrapper(const char* devname, int mode, const void* argp) { int rc = 0; const char* name = (const char*)argp; struct sbd_context *st; st = open_device(devname, LOG_WARNING); if (!st) return -1; rc = slot_ping(st, name); close_device(st); return rc; } int allocate_slots(const char *name, struct servants_list_item *servants) { int rc = 0; struct sbd_context *st; struct servants_list_item *s; for (s = servants; s; s = s->next) { fprintf(stdout, "Trying to allocate slot for %s on device %s.\n", name, s->devname); st = open_device(s->devname, LOG_WARNING); if (!st) { return -1; } rc = slot_allocate(st, name); close_device(st); if (rc < 0) return rc; fprintf(stdout, "Slot for %s has been allocated on %s.\n", name, s->devname); } return 0; } int list_slots(struct servants_list_item *servants) { int rc = 0; struct servants_list_item *s; struct sbd_context *st; for (s = servants; s; s = s->next) { st = open_device(s->devname, LOG_WARNING); if (!st) { fprintf(stdout, "== disk %s unreadable!\n", s->devname); continue; } rc = slot_list(st); close_device(st); if (rc == -1) { fprintf(stdout, "== Slots on disk %s NOT dumped\n", s->devname); } } return 0; } int ping_via_slots(const char *name, struct servants_list_item *servants) { int sig = 0; pid_t pid = 0; int status = 0; int servants_finished = 0; sigset_t procmask; siginfo_t sinfo; struct servants_list_item *s; sigemptyset(&procmask); sigaddset(&procmask, SIGCHLD); sigprocmask(SIG_BLOCK, &procmask, NULL); for (s = servants; s; s = s->next) { s->pid = assign_servant(s->devname, &slot_ping_wrapper, 0, (const void*)name); } while (servants_finished < servant_count) { sig = sigwaitinfo(&procmask, &sinfo); if (sig == SIGCHLD) { while ((pid = wait(&status))) { if (pid == -1 && errno == ECHILD) { break; } else { s = lookup_servant_by_pid(pid); if (s) { servants_finished++; } } } } } return 0; } int quorum_write(int good_servants) { return (good_servants > servant_count/2); } int messenger(const char *name, const char *msg, struct servants_list_item *servants) { int sig = 0; pid_t pid = 0; int status = 0; int servants_finished = 0; int successful_delivery = 0; sigset_t procmask; siginfo_t sinfo; struct servants_list_item *s; struct slot_msg_arg_t slot_msg_arg = {name, msg}; sigemptyset(&procmask); sigaddset(&procmask, SIGCHLD); sigprocmask(SIG_BLOCK, &procmask, NULL); for (s = servants; s; s = s->next) { s->pid = assign_servant(s->devname, &slot_msg_wrapper, 0, &slot_msg_arg); } while (!(quorum_write(successful_delivery) || (servants_finished == servant_count))) { sig = sigwaitinfo(&procmask, &sinfo); if (sig == SIGCHLD) { while ((pid = waitpid(-1, &status, WNOHANG))) { if (pid == -1 && errno == ECHILD) { break; } else { servants_finished++; if (WIFEXITED(status) && WEXITSTATUS(status) == 0) { DBGLOG(LOG_INFO, "Process %d succeeded.", (int)pid); successful_delivery++; } else { cl_log(LOG_WARNING, "Process %d failed to deliver!", (int)pid); } } } } } if (quorum_write(successful_delivery)) { cl_log(LOG_INFO, "Message successfully delivered."); return 0; } else { cl_log(LOG_ERR, "Message is not delivered via more then a half of devices"); return -1; } } +unsigned long +get_first_msgwait(struct servants_list_item *servants) +{ + unsigned long msgwait = 0; + struct servants_list_item *s = servants; + + for (s = servants; s; s = s->next) { + struct sbd_context *st; + struct sector_header_s *s_header; + st = open_device(s->devname, LOG_WARNING); + if (!st) { + continue; + } + + s_header = header_get(st); + if (s_header != NULL) { + msgwait = (unsigned long)s_header->timeout_msgwait; + close_device(st); + return msgwait; + } + + close_device(st); + } + return msgwait; +} + int dump_headers(struct servants_list_item *servants) { int rc = 0; struct servants_list_item *s = servants; struct sbd_context *st; for (s = servants; s; s = s->next) { fprintf(stdout, "==Dumping header on disk %s\n", s->devname); st = open_device(s->devname, LOG_WARNING); if (!st) { fprintf(stdout, "== disk %s unreadable!\n", s->devname); continue; } rc = header_dump(st); close_device(st); if (rc == -1) { fprintf(stdout, "==Header on disk %s NOT dumped\n", s->devname); } else { fprintf(stdout, "==Header on disk %s is dumped\n", s->devname); } } return rc; } void open_any_device(struct servants_list_item *servants) { struct sector_header_s *hdr_cur = NULL; struct timespec t_0; int t_wait = 0; clock_gettime(CLOCK_MONOTONIC, &t_0); while (!hdr_cur && t_wait < timeout_startup) { struct timespec t_now; struct servants_list_item* s; for (s = servants; s; s = s->next) { struct sbd_context *st = open_device(s->devname, LOG_DEBUG); if (!st) continue; hdr_cur = header_get(st); close_device(st); if (hdr_cur) break; } clock_gettime(CLOCK_MONOTONIC, &t_now); t_wait = t_now.tv_sec - t_0.tv_sec; if (!hdr_cur) { sleep(timeout_loop); } } if (hdr_cur) { timeout_watchdog = hdr_cur->timeout_watchdog; timeout_allocate = hdr_cur->timeout_allocate; timeout_loop = hdr_cur->timeout_loop; timeout_msgwait = hdr_cur->timeout_msgwait; } else { cl_log(LOG_ERR, "No devices were available at start-up within %i seconds.", timeout_startup); exit(1); } free(hdr_cur); return; } /* ::-::-::-::-::-::-::-::-::-::-::-::-:: Begin disk based servant code ::-::-::-::-::-::-::-::-::-::-::-::-:: */ static int servant_check_timeout_inconsistent(struct sector_header_s *hdr) { if (timeout_watchdog != hdr->timeout_watchdog) { cl_log(LOG_WARNING, "watchdog timeout: %d versus %d on this device", (int)timeout_watchdog, (int)hdr->timeout_watchdog); return -1; } if (timeout_allocate != hdr->timeout_allocate) { cl_log(LOG_WARNING, "allocate timeout: %d versus %d on this device", (int)timeout_allocate, (int)hdr->timeout_allocate); return -1; } if (timeout_loop != hdr->timeout_loop) { cl_log(LOG_WARNING, "loop timeout: %d versus %d on this device", (int)timeout_loop, (int)hdr->timeout_loop); return -1; } if (timeout_msgwait != hdr->timeout_msgwait) { cl_log(LOG_WARNING, "msgwait timeout: %d versus %d on this device", (int)timeout_msgwait, (int)hdr->timeout_msgwait); return -1; } return 0; } /* This is a bit hackish, but the easiest way to rewire all process * exits to send the desired signal to the parent. */ void servant_exit(void) { pid_t ppid; union sigval signal_value; ppid = getppid(); if (servant_inform_parent) { memset(&signal_value, 0, sizeof(signal_value)); sigqueue(ppid, SIG_IO_FAIL, signal_value); } } int servant(const char *diskname, int mode, const void* argp) { struct sector_mbox_s *s_mbox = NULL; struct sector_node_s *s_node = NULL; struct sector_header_s *s_header = NULL; int mbox; int rc = 0; time_t t0, t1, latency; union sigval signal_value; sigset_t servant_masks; struct sbd_context *st; pid_t ppid; char uuid[37]; const struct servants_list_item *s = argp; if (!diskname) { cl_log(LOG_ERR, "Empty disk name %s.", diskname); return -1; } cl_log(LOG_INFO, "Servant starting for device %s", diskname); /* Block most of the signals */ sigfillset(&servant_masks); sigdelset(&servant_masks, SIGKILL); sigdelset(&servant_masks, SIGFPE); sigdelset(&servant_masks, SIGILL); sigdelset(&servant_masks, SIGSEGV); sigdelset(&servant_masks, SIGBUS); sigdelset(&servant_masks, SIGALRM); /* FIXME: check error */ sigprocmask(SIG_SETMASK, &servant_masks, NULL); atexit(servant_exit); servant_inform_parent = 1; st = open_device(diskname, LOG_WARNING); if (!st) { return -1; } s_header = header_get(st); if (!s_header) { cl_log(LOG_ERR, "Not a valid header on %s", diskname); return -1; } if (servant_check_timeout_inconsistent(s_header) < 0) { cl_log(LOG_ERR, "Timeouts on %s do not match first device", diskname); return -1; } if (s_header->minor_version > 0) { uuid_unparse_lower(s_header->uuid, uuid); cl_log(LOG_INFO, "Device %s uuid: %s", diskname, uuid); } mbox = slot_allocate(st, local_uname); if (mbox < 0) { cl_log(LOG_ERR, "No slot allocated, and automatic allocation failed for disk %s.", diskname); rc = -1; goto out; } s_node = sector_alloc(); if (slot_read(st, mbox, s_node) < 0) { cl_log(LOG_ERR, "Unable to read node entry on %s", diskname); exit(1); } DBGLOG(LOG_INFO, "Monitoring slot %d on disk %s", mbox, diskname); if (s_header->minor_version == 0) { set_proc_title("sbd: watcher: %s - slot: %d", diskname, mbox); } else { set_proc_title("sbd: watcher: %s - slot: %d - uuid: %s", diskname, mbox, uuid); } s_mbox = sector_alloc(); if (s->first_start) { if (mode > 0) { if (mbox_read(st, mbox, s_mbox) < 0) { cl_log(LOG_ERR, "mbox read failed during start-up in servant."); rc = -1; goto out; } if (s_mbox->cmd != SBD_MSG_EXIT && s_mbox->cmd != SBD_MSG_EMPTY) { /* Not a clean stop. Abort start-up */ cl_log(LOG_WARNING, "Found fencing message - aborting start-up. Manual intervention required!"); ppid = getppid(); sigqueue(ppid, SIG_EXITREQ, signal_value); rc = 0; goto out; } } DBGLOG(LOG_INFO, "First servant start - zeroing inbox"); memset(s_mbox, 0, sizeof(*s_mbox)); if (mbox_write(st, mbox, s_mbox) < 0) { rc = -1; goto out; } } memset(&signal_value, 0, sizeof(signal_value)); while (1) { struct sector_header_s *s_header_retry = NULL; struct sector_node_s *s_node_retry = NULL; t0 = time(NULL); sleep(timeout_loop); ppid = getppid(); if (ppid == 1) { /* Our parent died unexpectedly. Triggering * self-fence. */ do_reset(); } /* These attempts are, by definition, somewhat racy. If * the device is wiped out or corrupted between here and * us reading our mbox, there is nothing we can do about * that. But at least we tried. */ s_header_retry = header_get(st); if (!s_header_retry) { cl_log(LOG_ERR, "No longer found a valid header on %s", diskname); exit(1); } if (memcmp(s_header, s_header_retry, sizeof(*s_header)) != 0) { cl_log(LOG_ERR, "Header on %s changed since start-up!", diskname); exit(1); } free(s_header_retry); s_node_retry = sector_alloc(); if (slot_read(st, mbox, s_node_retry) < 0) { cl_log(LOG_ERR, "slot read failed in servant."); exit(1); } if (memcmp(s_node, s_node_retry, sizeof(*s_node)) != 0) { cl_log(LOG_ERR, "Node entry on %s changed since start-up!", diskname); exit(1); } free(s_node_retry); if (mbox_read(st, mbox, s_mbox) < 0) { cl_log(LOG_ERR, "mbox read failed in servant."); exit(1); } if (s_mbox->cmd > 0) { cl_log(LOG_INFO, "Received command %s from %s on disk %s", char2cmd(s_mbox->cmd), s_mbox->from, diskname); switch (s_mbox->cmd) { case SBD_MSG_TEST: memset(s_mbox, 0, sizeof(*s_mbox)); mbox_write(st, mbox, s_mbox); sigqueue(ppid, SIG_TEST, signal_value); break; case SBD_MSG_RESET: do_reset(); break; case SBD_MSG_OFF: do_off(); break; case SBD_MSG_EXIT: sigqueue(ppid, SIG_EXITREQ, signal_value); break; case SBD_MSG_CRASHDUMP: do_crashdump(); break; default: /* FIXME: An "unknown" message might result from a partial write. log it and clear the slot. */ cl_log(LOG_ERR, "Unknown message on disk %s", diskname); memset(s_mbox, 0, sizeof(*s_mbox)); mbox_write(st, mbox, s_mbox); break; } } sigqueue(ppid, SIG_LIVENESS, signal_value); t1 = time(NULL); latency = t1 - t0; if (timeout_watchdog_warn && (latency > timeout_watchdog_warn)) { cl_log(LOG_WARNING, "Latency: %d exceeded threshold %d on disk %s", (int)latency, (int)timeout_watchdog_warn, diskname); } else if (debug) { DBGLOG(LOG_INFO, "Latency: %d on disk %s", (int)latency, diskname); } } out: free(s_mbox); close_device(st); if (rc == 0) { servant_inform_parent = 0; } return rc; } diff --git a/src/sbd.h b/src/sbd.h index 1384c81..1a30758 100644 --- a/src/sbd.h +++ b/src/sbd.h @@ -1,172 +1,173 @@ /* * Copyright (C) 2013 Lars Marowsky-Bree * * This program is free software; you can redistribute it and/or * modify it under the terms of the GNU General Public * License as published by the Free Software Foundation; either * version 2.1 of the License, or (at your option) any later version. * * This software is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU * General Public License for more details. * * You should have received a copy of the GNU General Public * License along with this library; if not, write to the Free Software * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ #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 /* signals reserved for multi-disk sbd */ #define SIG_LIVENESS (SIGRTMIN + 1) /* report liveness of the disk */ #define SIG_EXITREQ (SIGRTMIN + 2) /* exit request to inquisitor */ #define SIG_TEST (SIGRTMIN + 3) /* trigger self test */ #define SIG_RESTART (SIGRTMIN + 4) /* trigger restart of all failed disk */ #define SIG_IO_FAIL (SIGRTMIN + 5) /* the IO child requests to be considered failed */ #define SIG_PCMK_UNHEALTHY (SIGRTMIN + 6) /* FIXME: should add dynamic check of SIG_XX >= SIGRTMAX */ #define HOG_CHAR 0xff #define HA_COREDIR "/var/lib/heartbeat/cores" /* Sector data types */ struct sector_header_s { char magic[8]; unsigned char version; unsigned char slots; /* Caveat: stored in network byte-order */ uint32_t sector_size; uint32_t timeout_watchdog; uint32_t timeout_allocate; uint32_t timeout_loop; uint32_t timeout_msgwait; /* Minor version for extensions to the core data set: * compatible and optional values. */ unsigned char minor_version; uuid_t uuid; /* 16 bytes */ }; struct sector_mbox_s { signed char cmd; char from[64]; }; struct sector_node_s { /* slots will be created with in_use == 0 */ char in_use; char name[64]; }; struct servants_list_item { const char* devname; pid_t pid; int restarts; int restart_blocked; int outdated; int first_start; struct timespec t_last, t_started; struct servants_list_item *next; }; struct sbd_context { int devfd; io_context_t ioctx; struct iocb io; }; void usage(void); int watchdog_init_interval(void); int watchdog_tickle(void); int watchdog_init(void); void sysrq_init(void); void watchdog_close(bool disarm); void sysrq_trigger(char t); void do_crashdump(void); void do_reset(void); void do_off(void); pid_t make_daemon(void); void maximize_priority(void); void sbd_get_uname(void); /* Tunable defaults: */ extern unsigned long timeout_watchdog; extern unsigned long timeout_watchdog_warn; extern unsigned long timeout_watchdog_crashdump; extern int timeout_allocate; extern int timeout_loop; extern int timeout_msgwait; extern int timeout_io; extern int timeout_startup; extern int watchdog_use; extern int watchdog_set_timeout; extern int skip_rt; extern int debug; extern int debug_mode; extern const char *watchdogdev; extern char* local_uname; /* Global, non-tunable variables: */ extern int sector_size; extern int watchdogfd; extern const char* cmdname; typedef int (*functionp_t)(const char* devname, int mode, const void* argp); int assign_servant(const char* devname, functionp_t functionp, int mode, const void* argp); #if SUPPORT_SHARED_DISK void open_any_device(struct servants_list_item *servants); int init_devices(struct servants_list_item *servants); int allocate_slots(const char *name, struct servants_list_item *servants); int list_slots(struct servants_list_item *servants); int ping_via_slots(const char *name, struct servants_list_item *servants); int dump_headers(struct servants_list_item *servants); +unsigned long get_first_msgwait(struct servants_list_item *servants); int messenger(const char *name, const char *msg, struct servants_list_item *servants); int servant(const char *diskname, int mode, const void* argp); #endif int servant_pcmk(const char *diskname, int mode, const void* argp); struct servants_list_item *lookup_servant_by_dev(const char *devname); struct servants_list_item *lookup_servant_by_pid(pid_t pid); int init_set_proc_title(int argc, char *argv[], char *envp[]); void set_proc_title(const char *fmt,...); #define cl_log(level, fmt, args...) qb_log_from_external_source( __func__, __FILE__, fmt, level, __LINE__, 0, ##args) # define cl_perror(fmt, args...) do { \ const char *err = strerror(errno); \ cl_log(LOG_ERR, fmt ": %s (%d)", ##args, err, errno); \ } while(0) #define DBGLOG(lvl, fmt, args...) do { \ if (debug > 0) cl_log(lvl, fmt, ##args); \ } while(0) diff --git a/src/sbd.service b/src/sbd.service index 83eaebc..7363d9a 100644 --- a/src/sbd.service +++ b/src/sbd.service @@ -1,23 +1,25 @@ [Unit] Description=Shared-storage based fencing daemon Before=pacemaker.service corosync.service After=systemd-modules-load.service iscsi.service PartOf=corosync.service RefuseManualStop=true RefuseManualStart=true [Service] Type=forking -ExecStart=/usr/share/sbd/sbd.sh start -ExecStop=/usr/share/sbd/sbd.sh stop -PIDFile=/var/run/sbd.pid +EnvironmentFile=-@sysconfdir@/sysconfig/sbd +ExecStart=@sbindir@/sbd $SBD_OPTS watch +ExecStop=kill -TERM $MAINPID +PIDFile=${SBD_PIDFILE} + # Could this benefit from exit codes for restart? # Does this need to be set to msgwait * 1.2? # TimeoutSec= # If SBD crashes, it'll very likely suicide immediately due to the # hardware watchdog. But one can always try. Restart=on-abort [Install] -RequiredBy=pacemaker.service +RequiredBy=corosync.service diff --git a/src/sbd.sysconfig b/src/sbd.sysconfig index 743f99d..e97613a 100644 --- a/src/sbd.sysconfig +++ b/src/sbd.sysconfig @@ -1,52 +1,58 @@ ## Type: string ## Default: "" # # SBD_DEVICE specifies the devices to use for exchanging sbd messages # and to monitor. If specifying more than one path, use ";" as # separator. # -SBD_DEVICE="" +#SBD_DEVICE="" ## Type: yesno ## Default: yes # # Whether to enable the pacemaker integration. # -SBD_PACEMAKER= +SBD_PACEMAKER=yes ## Type: list(always,clean) ## Default: always # # Specify the start mode for sbd. Setting this to "clean" will only # allow sbd to start if it was not previously fenced. See the -S option # in the man page. # -SBD_STARTMODE= +SBD_STARTMODE=clean ## Type: yesno ## Default: no # # Whether to delay after starting sbd on boot for "msgwait" seconds. # This may be necessary if your cluster nodes reboot so fast that the # other nodes are still waiting in the fence acknowledgement phase. # This is an occasional issue with virtual machines. # # This option may be ignored at a later point, once pacemaker handles # this case better. # -SBD_DELAY_START= +SBD_DELAY_START=no ## Type: yesno ## Default: yes # # Whether to use a watchdog. # -SBD_WATCHDOG= +SBD_WATCHDOG_DEV=/dev/watchdog + +## Type: string +## Default: /var/run/sbd.pid +# +# Where to store the PID of the active process +# +SBD_PIDFILE=/var/run/sbd.pid ## Type: string ## Default: "" # # Additional options for starting sbd # SBD_OPTS= -