diff --git a/src/sbd-common.c b/src/sbd-common.c index 3abf75f..249899d 100644 --- a/src/sbd-common.c +++ b/src/sbd-common.c @@ -1,1355 +1,1362 @@ /* * 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 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 program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #include "sbd.h" #include #include #ifdef __GLIBC__ #include #endif #include #include #include #include #include #include #include #ifdef _POSIX_MEMLOCK # include #endif /* Tunable defaults: */ unsigned long timeout_watchdog = SBD_WATCHDOG_TIMEOUT_DEFAULT; int timeout_msgwait = 2 * SBD_WATCHDOG_TIMEOUT_DEFAULT; unsigned long timeout_watchdog_warn = calculate_timeout_watchdog_warn(SBD_WATCHDOG_TIMEOUT_DEFAULT); bool do_calculate_timeout_watchdog_warn = true; int timeout_allocate = 2; int timeout_loop = 1; int timeout_io = 3; int timeout_startup = 120; int watchdog_use = 1; int watchdog_set_timeout = 1; unsigned long timeout_watchdog_crashdump = 0; int skip_rt = 0; int debug = 0; int debug_mode = 0; char *watchdogdev = NULL; bool watchdogdev_is_default = false; char * local_uname; /* Global, non-tunable variables: */ int sector_size = 0; int watchdogfd = -1; int servant_health = 0; /*const char *devname;*/ const char *cmdname; void usage(void) { fprintf(stderr, "Shared storage fencing tool.\n" "Syntax:\n" " %s \n" "Options:\n" "-d Block device to use (mandatory; can be specified up to 3 times)\n" "-h Display this help.\n" "-n Set local node name; defaults to uname -n (optional)\n" "\n" "-R Do NOT enable realtime priority (debugging only)\n" "-W Use watchdog (recommended) (watch only)\n" "-w Specify watchdog device (optional) (watch only)\n" "-T Do NOT initialize the watchdog timeout (watch only)\n" "-S <0|1> Set start mode if the node was previously fenced (watch only)\n" "-p Write pidfile to the specified path (watch only)\n" "-v|-vv|-vvv Enable verbose|debug|debug-library logging (optional)\n" "\n" "-1 Set watchdog timeout to N seconds (optional, create only)\n" "-2 Set slot allocation timeout to N seconds (optional, create only)\n" "-3 Set daemon loop timeout to N seconds (optional, create only)\n" "-4 Set msgwait timeout to N seconds (optional, create only)\n" "-5 Warn if loop latency exceeds threshold (optional, watch only)\n" " (default is 3, set to 0 to disable)\n" "-C Watchdog timeout to set before crashdumping\n" " (def: 0s = disable gracefully, optional)\n" "-I Async IO read timeout (defaults to 3 * loop timeout, optional)\n" "-s Timeout to wait for devices to become available (def: 120s)\n" "-t Dampening delay before faulty servants are restarted (optional)\n" " (default is 5, set to 0 to disable)\n" "-F # of failures before a servant is considered faulty (optional)\n" " (default is 1, set to 0 to disable)\n" "-P Check Pacemaker quorum and node health (optional, watch only)\n" "-Z Enable trace mode. WARNING: UNSAFE FOR PRODUCTION!\n" "-r Set timeout-action to comma-separated combination of\n" " noflush|flush plus reboot|crashdump|off (default is flush,reboot)\n" "Commands:\n" #if SUPPORT_SHARED_DISK "create initialize N slots on - OVERWRITES DEVICE!\n" "list List all allocated slots on device, and messages.\n" "dump Dump meta-data header from device.\n" "allocate \n" " Allocate a slot for node (optional)\n" "message (test|reset|off|crashdump|clear|exit)\n" " Writes the specified message to node's slot.\n" #endif "watch Loop forever, monitoring own slot\n" "query-watchdog Check for available watchdog-devices and print some info\n" "test-watchdog Test the watchdog-device selected.\n" " Attention: This will arm the watchdog and have your system reset\n" " in case your watchdog is working properly!\n" , cmdname); } #define MAX_WATCHDOGS 64 #define SYS_CLASS_WATCHDOG "/sys/class/watchdog" #define SYS_CHAR_DEV_DIR "/sys/dev/char" #define WATCHDOG_NODEDIR "/dev/" static bool is_watchdog(dev_t device) { static int num_watchdog_devs = 0; static dev_t watchdog_devs[MAX_WATCHDOGS]; struct dirent *entry; int i; /* populate on first call */ if (num_watchdog_devs == 0) { DIR *dp; watchdog_devs[0] = makedev(10,130); num_watchdog_devs = 1; /* get additional devices from /sys/class/watchdog */ dp = opendir(SYS_CLASS_WATCHDOG); if (dp) { while ((entry = readdir(dp))) { if (entry->d_type == DT_LNK) { FILE *file; char entry_name[NAME_MAX+sizeof(SYS_CLASS_WATCHDOG)+5]; snprintf(entry_name, sizeof(entry_name), SYS_CLASS_WATCHDOG "/%s/dev", entry->d_name); file = fopen(entry_name, "r"); if (file) { int major, minor; if (fscanf(file, "%d:%d", &major, &minor) == 2) { watchdog_devs[num_watchdog_devs++] = makedev(major, minor); } fclose(file); if (num_watchdog_devs == MAX_WATCHDOGS) { break; } } } } closedir(dp); } } for (i=0; i < num_watchdog_devs; i++) { if (device == watchdog_devs[i]) { return true; } } return false; } static int watchdog_init_interval_fd(int wdfd, int timeout) { if (ioctl(wdfd, WDIOC_SETTIMEOUT, &timeout) < 0) { cl_perror( "WDIOC_SETTIMEOUT" ": Failed to set watchdog timer to %u seconds.", timeout); cl_log(LOG_CRIT, "Please validate your watchdog configuration!"); cl_log(LOG_CRIT, "Choose a different watchdog driver or specify -T to skip this if you are completely sure."); return -1; } return 0; } int watchdog_init_interval(void) { if (watchdogfd < 0) { return 0; } if (watchdog_set_timeout == 0) { cl_log(LOG_INFO, "NOT setting watchdog timeout on explicit user request!"); return 0; } if (watchdog_init_interval_fd(watchdogfd, timeout_watchdog) < 0) { return -1; } cl_log(LOG_INFO, "Set watchdog timeout to %u seconds.", (int) timeout_watchdog); return 0; } static int watchdog_tickle_fd(int wdfd, char *wddev) { if (write(wdfd, "", 1) != 1) { cl_perror("Watchdog write failure: %s!", wddev); return -1; } return 0; } int watchdog_tickle(void) { if (watchdogfd >= 0) { return watchdog_tickle_fd(watchdogfd, watchdogdev); } return 0; } static int watchdog_init_fd(char *wddev, int timeout) { int wdfd; wdfd = open(wddev, O_WRONLY); if (wdfd >= 0) { if (((timeout >= 0) && (watchdog_init_interval_fd(wdfd, timeout) < 0)) || (watchdog_tickle_fd(wdfd, wddev) < 0)) { close(wdfd); return -1; } } else { struct stat statbuf; if(!stat(wddev, &statbuf) && S_ISCHR(statbuf.st_mode) && is_watchdog(statbuf.st_rdev)) { cl_perror("Cannot open watchdog device '%s'", wddev); } else { cl_perror("Seems as if '%s' isn't a valid watchdog-device", wddev); } return -1; } return wdfd; } int watchdog_init(void) { if (watchdogfd < 0 && watchdogdev != NULL) { int timeout = timeout_watchdog; if (watchdog_set_timeout == 0) { cl_log(LOG_INFO, "NOT setting watchdog timeout on explicit user request!"); timeout = -1; } watchdogfd = watchdog_init_fd(watchdogdev, timeout); if (watchdogfd >= 0) { cl_log(LOG_NOTICE, "Using watchdog device '%s'", watchdogdev); if (watchdog_set_timeout) { cl_log(LOG_INFO, "Set watchdog timeout to %u seconds.", (int) timeout_watchdog); } } else { return -1; } } return 0; } static void watchdog_close_fd(int wdfd, char *wddev, bool disarm) { if (disarm) { int r; int flags = WDIOS_DISABLECARD;; /* Explicitly disarm it */ r = ioctl(wdfd, WDIOC_SETOPTIONS, &flags); if (r < 0) { cl_perror("Failed to disable hardware watchdog %s", wddev); } /* To be sure, use magic close logic, too */ for (;;) { if (write(wdfd, "V", 1) > 0) { break; } cl_perror("Cannot disable watchdog device %s", wddev); } } if (close(wdfd) < 0) { cl_perror("Watchdog close(%d) failed", wdfd); } } void watchdog_close(bool disarm) { if (watchdogfd < 0) { return; } watchdog_close_fd(watchdogfd, watchdogdev, disarm); watchdogfd = -1; } struct watchdog_list_item { dev_t dev; char *dev_node; char *dev_ident; char *dev_driver; pid_t busy_pid; char *busy_name; struct watchdog_list_item *next; }; struct link_list_item { char *dev_node; char *link_name; struct link_list_item *next; }; static struct watchdog_list_item *watchdog_list = NULL; static int watchdog_list_items = 0; static void watchdog_populate_list(void) { struct dirent *entry; char entry_name[sizeof(WATCHDOG_NODEDIR)+NAME_MAX]; DIR *dp; char buf[NAME_MAX+sizeof(WATCHDOG_NODEDIR)] = ""; struct link_list_item *link_list = NULL; if (watchdog_list != NULL) { return; } /* search for watchdog nodes in /dev */ dp = opendir(WATCHDOG_NODEDIR); if (dp) { /* first go for links and memorize them */ while ((entry = readdir(dp))) { if (entry->d_type == DT_LNK) { int len; snprintf(entry_name, sizeof(entry_name), WATCHDOG_NODEDIR "%s", entry->d_name); /* realpath(entry_name, buf) unfortunately does a stat on * target so we can't really use it to check if links stay * within /dev without triggering e.g. AVC-logs (with * SELinux policy that just allows stat within /dev). * Without canonicalization that doesn't actually touch the * filesystem easily available introduce some limitations * for simplicity: * - just simple path without '..' * - just one level of symlinks (avoid e.g. loop-checking) */ len = readlink(entry_name, buf, sizeof(buf) - 1); if ((len < 1) || (len > sizeof(buf) - sizeof(WATCHDOG_NODEDIR) -1 - 1)) { continue; } buf[len] = '\0'; if (buf[0] != '/') { memmove(&buf[sizeof(WATCHDOG_NODEDIR)-1], buf, len+1); memcpy(buf, WATCHDOG_NODEDIR, sizeof(WATCHDOG_NODEDIR)-1); len += sizeof(WATCHDOG_NODEDIR)-1; } if (strstr(buf, "/../") || strncmp(WATCHDOG_NODEDIR, buf, sizeof(WATCHDOG_NODEDIR)-1)) { continue; } else { /* just memorize to avoid statting the target - SELinux */ struct link_list_item *lli = calloc(1, sizeof(struct link_list_item)); if (lli == NULL) { break; } lli->dev_node = strdup(buf); lli->link_name = strdup(entry_name); if ((lli->dev_node == NULL) || (lli->link_name == NULL)) { free(lli->dev_node); free(lli->link_name); free(lli); break; } lli->next = link_list; link_list = lli; } } } rewinddir(dp); while ((entry = readdir(dp))) { if (entry->d_type == DT_CHR) { struct stat statbuf; snprintf(entry_name, sizeof(entry_name), WATCHDOG_NODEDIR "%s", entry->d_name); if(!stat(entry_name, &statbuf) && S_ISCHR(statbuf.st_mode) && is_watchdog(statbuf.st_rdev)) { int wdfd; struct watchdog_list_item *wdg = calloc(1, sizeof(struct watchdog_list_item)); int len; struct link_list_item *tmp_list = NULL; if (wdg == NULL) { break; } wdg->dev = statbuf.st_rdev; wdg->dev_node = strdup(entry_name); if (wdg->dev_node == NULL) { free(wdg); break; } wdg->next = watchdog_list; watchdog_list = wdg; watchdog_list_items++; wdfd = watchdog_init_fd(entry_name, -1); if (wdfd >= 0) { struct watchdog_info ident; ident.identity[0] = '\0'; ioctl(wdfd, WDIOC_GETSUPPORT, &ident); watchdog_close_fd(wdfd, entry_name, true); if (ident.identity[0]) { wdg->dev_ident = strdup((char *) ident.identity); } } snprintf(entry_name, sizeof(entry_name), SYS_CHAR_DEV_DIR "/%d:%d/device/driver", major(wdg->dev), minor(wdg->dev)); len = readlink(entry_name, buf, sizeof(buf) - 1); if (len > 0) { buf[len] = '\0'; wdg->dev_driver = strdup(basename(buf)); } else if ((wdg->dev_ident) && (strcmp(wdg->dev_ident, "Software Watchdog") == 0)) { wdg->dev_driver = strdup("softdog"); } /* create dupes if we have memorized links * to this node */ for (tmp_list = link_list; tmp_list; tmp_list = tmp_list->next) { if (!strcmp(tmp_list->dev_node, wdg->dev_node)) { struct watchdog_list_item *dupe_wdg = calloc(1, sizeof(struct watchdog_list_item)); if (dupe_wdg == NULL) { break; } /* as long as we never purge watchdog_list * there is no need to dupe strings */ *dupe_wdg = *wdg; dupe_wdg->dev_node = strdup(tmp_list->link_name); if (dupe_wdg->dev_node == NULL) { free(dupe_wdg); break; } dupe_wdg->next = watchdog_list; watchdog_list = dupe_wdg; watchdog_list_items++; } /* for performance reasons we could remove * the link_list entry */ } } } } closedir(dp); } /* cleanup link list */ while (link_list) { struct link_list_item *tmp_list = link_list; link_list = link_list->next; free(tmp_list->dev_node); free(tmp_list->link_name); free(tmp_list); } } static void watchdog_checkbusy() { DIR *dproc; struct dirent *entry; dproc = opendir("/proc"); if (!dproc) { /* no proc directory to search through */ return; } while ((entry = readdir(dproc)) != NULL) { pid_t local_pid; char *leftover; DIR *dpid; char procpath[NAME_MAX+10] = { 0 }; if (entry->d_name[0] == '.') { continue; } local_pid = strtol(entry->d_name, &leftover, 10); if (leftover[0] != '\0') continue; snprintf(procpath, sizeof(procpath), "/proc/%s/fd", entry->d_name); dpid = opendir(procpath); if (!dpid) { /* silently continue - might be just a race */ continue; } while ((entry = readdir(dpid)) != NULL) { struct watchdog_list_item *wdg; char entry_name[sizeof(procpath)+NAME_MAX+1] = { 0 }; char buf[NAME_MAX+1] = { 0 }; int len; if (entry->d_type != DT_LNK) { continue; } snprintf(entry_name, sizeof(entry_name), "%s/%s", procpath, entry->d_name); len = readlink(entry_name, buf, sizeof(buf) - 1); if (len < 1) { continue; } buf[len] = '\0'; for (wdg = watchdog_list; wdg != NULL; wdg = wdg->next) { if (!strcmp(buf, wdg->dev_node)) { char name[16]; FILE *file; wdg->busy_pid = local_pid; snprintf(procpath, sizeof(procpath), "/proc/%d/status", local_pid); file = fopen(procpath, "r"); if (file) { if (fscanf(file, "Name:\t%15[a-zA-Z0-9 _-]", name) == 1) { wdg->busy_name = strdup(name); } fclose(file); } } } } closedir(dpid); } closedir(dproc); return; } int watchdog_info(void) { struct watchdog_list_item *wdg; int wdg_cnt = 0; watchdog_populate_list(); watchdog_checkbusy(); printf("\nDiscovered %d watchdog devices:\n", watchdog_list_items); for (wdg = watchdog_list; wdg != NULL; wdg = wdg->next) { wdg_cnt++; if (wdg->busy_pid) { printf("\n[%d] %s\nIdentity: Busy: PID %d (%s)\nDriver: %s\n", wdg_cnt, wdg->dev_node, wdg->busy_pid, wdg->busy_name?wdg->busy_name:"", wdg->dev_driver?wdg->dev_driver:""); } else { printf("\n[%d] %s\nIdentity: %s\nDriver: %s\n", wdg_cnt, wdg->dev_node, wdg->dev_ident?wdg->dev_ident: "Error: device hogged via alias major/minor?", wdg->dev_driver?wdg->dev_driver:""); } if ((wdg->dev_driver) && (strcmp(wdg->dev_driver, "softdog") == 0)) { printf("CAUTION: Not recommended for use with sbd.\n"); } } return 0; } int watchdog_test(void) { int i; if ((watchdog_set_timeout == 0) || !watchdog_use) { printf("\nWatchdog is disabled - aborting test!!!\n"); return 0; } if (watchdogdev_is_default) { watchdog_populate_list(); if (watchdog_list_items > 1) { printf("\nError: Multiple watchdog devices discovered.\n" " Use -w or SBD_WATCHDOG_DEV to specify\n" " which device to reset the system with\n"); watchdog_info(); return -1; } } if ((isatty(fileno(stdin)))) { char buffer[16]; printf("\nWARNING: This operation is expected to force-reboot this system\n" " without following any shutdown procedures.\n\n" "Proceed? [NO/Proceed] "); if ((fgets(buffer, 16, stdin) == NULL) || strcmp(buffer, "Proceed\n")) { printf("\nAborting watchdog test!!!\n"); return 0; } printf("\n"); } printf("Initializing %s with a reset countdown of %d seconds ...\n", watchdogdev, (int) timeout_watchdog); if ((watchdog_init() < 0) || (watchdog_init_interval() < 0)) { printf("Failed to initialize watchdog!!!\n"); watchdog_info(); return -1; } printf("\n"); printf("NOTICE: The watchdog device is expected to reset the system\n" " in %d seconds. If system remains active beyond that time,\n" " watchdog may not be functional.\n\n", (int) timeout_watchdog); for (i=timeout_watchdog; i>1; i--) { printf("Reset countdown ... %d seconds\n", i); sleep(1); } for (i=2; i>0; i--) { printf("System expected to reset any moment ...\n"); sleep(1); } for (i=5; i>0; i--) { printf("System should have reset ...\n"); sleep(1); } printf("Error: The watchdog device has failed to reboot the system,\n" " and it may not be suitable for usage with sbd.\n"); /* test should trigger a reboot thus returning is actually bad */ return -1; } /* This duplicates some code from linux/ioprio.h since these are not included * even in linux-kernel-headers. Sucks. See also * /usr/src/linux/Documentation/block/ioprio.txt and ioprio_set(2) */ extern int sys_ioprio_set(int, int, int); int ioprio_set(int which, int who, int ioprio); inline int ioprio_set(int which, int who, int ioprio) { return syscall(__NR_ioprio_set, which, who, ioprio); } enum { IOPRIO_CLASS_NONE, IOPRIO_CLASS_RT, IOPRIO_CLASS_BE, IOPRIO_CLASS_IDLE, }; enum { IOPRIO_WHO_PROCESS = 1, IOPRIO_WHO_PGRP, IOPRIO_WHO_USER, }; #define IOPRIO_BITS (16) #define IOPRIO_CLASS_SHIFT (13) #define IOPRIO_PRIO_MASK ((1UL << IOPRIO_CLASS_SHIFT) - 1) #define IOPRIO_PRIO_CLASS(mask) ((mask) >> IOPRIO_CLASS_SHIFT) #define IOPRIO_PRIO_DATA(mask) ((mask) & IOPRIO_PRIO_MASK) #define IOPRIO_PRIO_VALUE(class, data) (((class) << IOPRIO_CLASS_SHIFT) | data) static void sbd_stack_hogger(unsigned char * inbuf, int kbytes) { unsigned char buf[1024]; if(kbytes <= 0) { return; } if (inbuf == NULL) { memset(buf, HOG_CHAR, sizeof(buf)); } else { memcpy(buf, inbuf, sizeof(buf)); } if (kbytes > 0) { sbd_stack_hogger(buf, kbytes-1); } return; } static void sbd_malloc_hogger(int kbytes) { int j; void**chunks; int chunksize = 1024; if(kbytes <= 0) { return; } /* * We could call mallopt(M_MMAP_MAX, 0) to disable it completely, * but we've already called mlockall() * * We could also call mallopt(M_TRIM_THRESHOLD, -1) to prevent malloc * from giving memory back to the system, but we've already called * mlockall(MCL_FUTURE), so there's no need. */ chunks = malloc(kbytes * sizeof(void *)); if (chunks == NULL) { cl_log(LOG_WARNING, "Could not preallocate chunk array"); return; } for (j=0; j < kbytes; ++j) { chunks[j] = malloc(chunksize); if (chunks[j] == NULL) { cl_log(LOG_WARNING, "Could not preallocate block %d", j); } else { memset(chunks[j], 0, chunksize); } } for (j=0; j < kbytes; ++j) { free(chunks[j]); } free(chunks); } static void sbd_memlock(int stackgrowK, int heapgrowK) { #ifdef _POSIX_MEMLOCK /* * We could call setrlimit(RLIMIT_MEMLOCK,...) with a large * number, but the mcp runs as root and mlock(2) says: * * Since Linux 2.6.9, no limits are placed on the amount of memory * that a privileged process may lock, and this limit instead * governs the amount of memory that an unprivileged process may * lock. */ if (mlockall(MCL_CURRENT|MCL_FUTURE) >= 0) { cl_log(LOG_INFO, "Locked ourselves in memory"); /* Now allocate some extra pages (MCL_FUTURE will ensure they stay around) */ sbd_malloc_hogger(heapgrowK); sbd_stack_hogger(NULL, stackgrowK); } else { cl_perror("Unable to lock ourselves into memory"); } #else cl_log(LOG_ERR, "Unable to lock ourselves into memory"); #endif } static int get_realtime_budget(void) { FILE *f; char fname[PATH_MAX]; int res = -1, lnum = 0, num; char *cgroup = NULL, *namespecs = NULL; snprintf(fname, PATH_MAX, "/proc/%jd/cgroup", (intmax_t)getpid()); f = fopen(fname, "rt"); if (f == NULL) { cl_log(LOG_WARNING, "Can't open cgroup file for pid=%jd", (intmax_t)getpid()); goto exit_res; } while( (num = fscanf(f, "%d:%m[^:]:%m[^\n]\n", &lnum, &namespecs, &cgroup)) !=EOF ) { if (namespecs && strstr(namespecs, "cpuacct")) { free(namespecs); break; } if (cgroup) { free(cgroup); cgroup = NULL; } if (namespecs) { free(namespecs); namespecs = NULL; } /* not to get stuck if format changes */ if ((num < 3) && ((fscanf(f, "%*[^\n]") == EOF) || (fscanf(f, "\n") == EOF))) { break; } } fclose(f); if (cgroup == NULL) { cl_log(LOG_WARNING, "Failed getting cgroup for pid=%jd", (intmax_t)getpid()); goto exit_res; } snprintf(fname, PATH_MAX, "/sys/fs/cgroup/cpu%s/cpu.rt_runtime_us", cgroup); f = fopen(fname, "rt"); if (f == NULL) { cl_log(LOG_WARNING, "cpu.rt_runtime_us existed for root-slice but " "doesn't for '%s'", cgroup); goto exit_res; } if (fscanf(f, "%d", &res) != 1) { cl_log(LOG_WARNING, "failed reading rt-budget from %s", fname); } else { cl_log(LOG_INFO, "slice='%s' has rt-budget=%d", cgroup, res); } fclose(f); exit_res: if (cgroup) { free(cgroup); } return res; } /* stolen from corosync */ static int sbd_move_to_root_cgroup(bool enforce_root_cgroup) { FILE *f; int res = -1; /* * /sys/fs/cgroup is hardcoded, because most of Linux distributions are now * using systemd and systemd uses hardcoded path of cgroup mount point. * * This feature is expected to be removed as soon as systemd gets support * for managing RT configuration. */ f = fopen("/sys/fs/cgroup/cpu/cpu.rt_runtime_us", "rt"); if (f == NULL) { cl_log(LOG_DEBUG, "cpu.rt_runtime_us doesn't exist -> " "system without cgroup or with disabled CONFIG_RT_GROUP_SCHED"); res = 0; goto exit_res; } fclose(f); if ((!enforce_root_cgroup) && (get_realtime_budget() > 0)) { cl_log(LOG_DEBUG, "looks as if we have rt-budget in the slice we are " "-> skip moving to root-slice"); res = 0; goto exit_res; } f = fopen("/sys/fs/cgroup/cpu/tasks", "w"); if (f == NULL) { cl_log(LOG_WARNING, "Can't open cgroups tasks file for writing"); goto exit_res; } if (fprintf(f, "%jd\n", (intmax_t)getpid()) <= 0) { cl_log(LOG_WARNING, "Can't write sbd pid into cgroups tasks file"); goto close_and_exit_res; } close_and_exit_res: if (fclose(f) != 0) { cl_log(LOG_WARNING, "Can't close cgroups tasks file"); goto exit_res; } exit_res: return (res); } void sbd_make_realtime(int priority, int stackgrowK, int heapgrowK) { if(priority < 0) { return; } do { #ifdef SCHED_RR if (move_to_root_cgroup) { sbd_move_to_root_cgroup(enforce_moving_to_root_cgroup); } { int pmin = sched_get_priority_min(SCHED_RR); int pmax = sched_get_priority_max(SCHED_RR); struct sched_param sp; int pcurrent; if (priority == 0) { priority = pmax; } else if (priority < pmin) { priority = pmin; } else if (priority > pmax) { priority = pmax; } if (sched_getparam(0, &sp) < 0) { cl_perror("Unable to get scheduler priority"); } else if ((pcurrent = sched_getscheduler(0)) < 0) { cl_perror("Unable to get scheduler policy"); } else if ((pcurrent == SCHED_RR) && (sp.sched_priority >= priority)) { cl_log(LOG_INFO, "Stay with priority (%d) for policy SCHED_RR", sp.sched_priority); break; } else { memset(&sp, 0, sizeof(sp)); sp.sched_priority = priority; if (sched_setscheduler(0, SCHED_RR, &sp) < 0) { cl_perror( "Unable to set scheduler policy to SCHED_RR priority %d", priority); } else { cl_log(LOG_INFO, "Scheduler policy is now SCHED_RR priority %d", priority); break; } } } #else cl_log(LOG_ERR, "System does not support updating the scheduler policy"); #endif #ifdef PRIO_PGRP if (setpriority(PRIO_PGRP, 0, INT_MIN) < 0) { cl_perror("Unable to raise the scheduler priority"); } else { cl_log(LOG_INFO, "Scheduler priority raised to the maximum"); } #else cl_perror("System does not support setting the scheduler priority"); #endif } while (0); sbd_memlock(heapgrowK, stackgrowK); } void maximize_priority(void) { if (skip_rt) { cl_log(LOG_INFO, "Not elevating to realtime (-R specified)."); return; } sbd_make_realtime(0, 256, 256); if (ioprio_set(IOPRIO_WHO_PROCESS, getpid(), IOPRIO_PRIO_VALUE(IOPRIO_CLASS_RT, 1)) != 0) { cl_perror("ioprio_set() call failed."); } } void sysrq_init(void) { FILE* procf; int c; procf = fopen("/proc/sys/kernel/sysrq", "r"); if (!procf) { cl_perror("cannot open /proc/sys/kernel/sysrq for read."); return; } if (fscanf(procf, "%d", &c) != 1) { cl_perror("Parsing sysrq failed"); c = 0; } fclose(procf); if (c == 1) return; /* 8 for debugging dumps of processes, 128 for reboot/poweroff */ c |= 136; procf = fopen("/proc/sys/kernel/sysrq", "w"); if (!procf) { cl_perror("cannot open /proc/sys/kernel/sysrq for writing"); return; } fprintf(procf, "%d", c); fclose(procf); return; } void sysrq_trigger(char t) { FILE *procf; procf = fopen("/proc/sysrq-trigger", "a"); if (!procf) { cl_perror("Opening sysrq-trigger failed."); return; } cl_log(LOG_INFO, "sysrq-trigger: %c\n", t); fprintf(procf, "%c\n", t); fclose(procf); return; } static void do_exit(char kind, bool do_flush) { /* TODO: Turn debug_mode into a bit field? Delay + kdump for example */ const char *reason = NULL; if (kind == 'c') { cl_log(LOG_NOTICE, "Initiating kdump"); } else if (debug_mode == 1) { cl_log(LOG_WARNING, "Initiating kdump instead of panicking the node (debug mode)"); kind = 'c'; } if (debug_mode == 2) { cl_log(LOG_WARNING, "Shutting down SBD instead of panicking the node (debug mode)"); watchdog_close(true); exit(0); } if (debug_mode == 3) { /* Give the system some time to flush logs to disk before rebooting. */ cl_log(LOG_WARNING, "Delaying node panic by 10s (debug mode)"); watchdog_close(true); sync(); sleep(10); } switch(kind) { case 'b': reason = "reboot"; break; case 'c': reason = "crashdump"; break; case 'o': reason = "off"; break; default: reason = "unknown"; break; } cl_log(LOG_EMERG, "Rebooting system: %s", reason); if (do_flush) { sync(); } if (kind == 'c') { if (timeout_watchdog_crashdump) { if (timeout_watchdog != timeout_watchdog_crashdump) { timeout_watchdog = timeout_watchdog_crashdump; watchdog_init_interval(); } watchdog_close(false); } else { watchdog_close(true); } sysrq_trigger(kind); } else { watchdog_close(false); sysrq_trigger(kind); if (reboot((kind == 'o')?RB_POWER_OFF:RB_AUTOBOOT) < 0) { cl_perror("%s failed", (kind == 'o')?"Poweroff":"Reboot"); } } exit(1); } void do_crashdump(void) { do_exit('c', true); } void do_reset(void) { do_exit('b', true); } void do_off(void) { do_exit('o', true); } void do_timeout_action(void) { do_exit(timeout_sysrq_char, do_flush); } /* * Change directory to the directory our core file needs to go in * Call after you establish the userid you're running under. */ int sbd_cdtocoredir(void) { int rc; static const char *dir = NULL; if (dir == NULL) { dir = CRM_CORE_DIR; } if ((rc=chdir(dir)) < 0) { int errsave = errno; cl_perror("Cannot chdir to [%s]", dir); errno = errsave; } return rc; } pid_t make_daemon(void) { pid_t pid; const char * devnull = "/dev/null"; pid = fork(); if (pid < 0) { cl_log(LOG_ERR, "%s: could not start daemon\n", cmdname); cl_perror("fork"); exit(1); }else if (pid > 0) { return pid; } qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_FALSE); /* This is the child; ensure privileges have not been lost. */ maximize_priority(); sysrq_init(); umask(022); close(0); (void)open(devnull, O_RDONLY); close(1); (void)open(devnull, O_WRONLY); close(2); (void)open(devnull, O_WRONLY); sbd_cdtocoredir(); return 0; } void sbd_get_uname(void) { struct utsname uname_buf; int i; if (uname(&uname_buf) < 0) { cl_perror("uname() failed?"); exit(1); } local_uname = strdup(uname_buf.nodename); for (i = 0; i < strlen(local_uname); i++) local_uname[i] = tolower(local_uname[i]); } #define FMT_MAX 256 void sbd_set_format_string(int method, const char *daemon) { int offset = 0; char fmt[FMT_MAX]; struct utsname res; switch(method) { case QB_LOG_STDERR: break; case QB_LOG_SYSLOG: if(daemon && strcmp(daemon, "sbd") != 0) { offset += snprintf(fmt + offset, FMT_MAX - offset, "%10s: ", daemon); } break; default: /* When logging to a file */ if (uname(&res) == 0) { offset += snprintf(fmt + offset, FMT_MAX - offset, "%%t [%d] %s %10s: ", getpid(), res.nodename, daemon); } else { offset += snprintf(fmt + offset, FMT_MAX - offset, "%%t [%d] %10s: ", getpid(), daemon); } } if (debug && method >= QB_LOG_STDERR) { offset += snprintf(fmt + offset, FMT_MAX - offset, "(%%-12f:%%5l %%g) %%-7p: %%n: "); } else { offset += snprintf(fmt + offset, FMT_MAX - offset, "%%g %%-7p: %%n: "); } if (method == QB_LOG_SYSLOG) { offset += snprintf(fmt + offset, FMT_MAX - offset, "%%b"); } else { offset += snprintf(fmt + offset, FMT_MAX - offset, "\t%%b"); } if(offset > 0) { qb_log_format_set(method, fmt); } } +int sigqueue_zero(pid_t pid, int sig) +{ +union sigval signal_value; + + memset(&signal_value, 0, sizeof(signal_value)); + + return sigqueue(pid, sig, signal_value); +} + void notify_parent(void) { pid_t ppid; - union sigval signal_value; - memset(&signal_value, 0, sizeof(signal_value)); ppid = getppid(); if (ppid == 1) { /* Our parent died unexpectedly. Triggering * self-fence. */ cl_log(LOG_WARNING, "Our parent is dead."); do_timeout_action(); } switch (servant_health) { case pcmk_health_pending: case pcmk_health_shutdown: case pcmk_health_transient: DBGLOG(LOG_DEBUG, "Not notifying parent: state transient (%d)", servant_health); break; case pcmk_health_unknown: case pcmk_health_unclean: case pcmk_health_noquorum: DBGLOG(LOG_WARNING, "Notifying parent: UNHEALTHY (%d)", servant_health); - sigqueue(ppid, SIG_PCMK_UNHEALTHY, signal_value); + sigqueue_zero(ppid, SIG_PCMK_UNHEALTHY); break; case pcmk_health_online: DBGLOG(LOG_DEBUG, "Notifying parent: healthy"); - sigqueue(ppid, SIG_LIVENESS, signal_value); + sigqueue_zero(ppid, SIG_LIVENESS); break; default: DBGLOG(LOG_WARNING, "Notifying parent: UNHEALTHY %d", servant_health); - sigqueue(ppid, SIG_PCMK_UNHEALTHY, signal_value); + sigqueue_zero(ppid, SIG_PCMK_UNHEALTHY); break; } } void set_servant_health(enum pcmk_health state, int level, char const *format, ...) { if (servant_health != state) { va_list ap; int len = 0; char *string = NULL; servant_health = state; va_start(ap, format); len = vasprintf (&string, format, ap); if(len > 0) { cl_log(level, "%s", string); } va_end(ap); free(string); } } bool sbd_is_disk(struct servants_list_item *servant) { if ((servant != NULL) && (servant->devname != NULL) && (servant->devname[0] == '/')) { return true; } return false; } bool sbd_is_cluster(struct servants_list_item *servant) { if ((servant != NULL) && (servant->devname != NULL) && (strcmp("cluster", servant->devname) == 0)) { return true; } return false; } bool sbd_is_pcmk(struct servants_list_item *servant) { if ((servant != NULL) && (servant->devname != NULL) && (strcmp("pcmk", servant->devname) == 0)) { return true; } return false; } diff --git a/src/sbd-inquisitor.c b/src/sbd-inquisitor.c index d917cd1..051801b 100644 --- a/src/sbd-inquisitor.c +++ b/src/sbd-inquisitor.c @@ -1,1408 +1,1405 @@ /* * 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 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 program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. */ #include #include "sbd.h" #define LOCKSTRLEN 11 static struct servants_list_item *servants_leader = NULL; int disk_priority = 1; int check_pcmk = 1; int check_cluster = 1; int has_check_pcmk_env = false; int disk_count = 0; int servant_count = 0; int servant_restart_interval = 5; int servant_restart_count = 1; int start_mode = 0; char* pidfile = NULL; bool do_flush = true; char timeout_sysrq_char = 'b'; bool move_to_root_cgroup = true; bool enforce_moving_to_root_cgroup = false; bool sync_resource_startup = false; int parse_device_line(const char *line); static int sanitize_numeric_option_value(const char *value) { char *end = NULL; long int result = -1; if (value == NULL) { return -1; } errno = 0; result = strtol(value, &end, 10); if (result <= INT_MIN || result >= INT_MAX || errno != 0) { result = -1; } else if (*end != '\0') { result = -1; } return (int)result; } static const char * sanitize_option_value(const char *value) { size_t max = 0; size_t lpc = 0; if (value == NULL) { return NULL; } max = strlen(value); for (lpc = 0; lpc < max; lpc++) { if (!isspace(value[lpc])) { break; } } return (strlen(value + lpc) > 0 ? (value + lpc) : NULL); } static const char * get_env_option(const char *option) { const char *value = getenv(option); return sanitize_option_value(value); } static int recruit_servant(const char *devname, pid_t pid) { struct servants_list_item *s = servants_leader; struct servants_list_item *newbie; if (lookup_servant_by_dev(devname)) { cl_log(LOG_DEBUG, "Servant %s already exists", devname); return 0; } newbie = malloc(sizeof(*newbie)); if (newbie) { memset(newbie, 0, sizeof(*newbie)); newbie->devname = strdup(devname); newbie->pid = pid; newbie->first_start = 1; } if (!newbie || !newbie->devname) { fprintf(stderr, "heap allocation failed in recruit_servant.\n"); exit(1); } /* some sanity-check on our newbie */ if (sbd_is_disk(newbie)) { cl_log(LOG_INFO, "Monitoring %s", devname); disk_count++; } else if (sbd_is_pcmk(newbie) || sbd_is_cluster(newbie)) { /* alive just after pcmk and cluster servants have shown up */ newbie->outdated = 1; } else { /* toss our newbie */ cl_log(LOG_ERR, "Refusing to recruit unrecognized servant %s", devname); free((void *) newbie->devname); free(newbie); return -1; } if (!s) { servants_leader = newbie; } else { while (s->next) s = s->next; s->next = newbie; } servant_count++; return 0; } 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(); sbd_set_format_string(QB_LOG_SYSLOG, devname); 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 (strcasecmp(s->devname, devname) == 0) 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); + r = sigqueue_zero(s->pid, 0); 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); + r = sigqueue_zero(s->pid, 0); if ((r != -1 || errno != ESRCH)) return; } s->restarts++; if (sbd_is_disk(s)) { #if SUPPORT_SHARED_DISK DBGLOG(LOG_INFO, "Starting servant for device %s", s->devname); s->pid = assign_servant(s->devname, servant_md, start_mode, s); #else cl_log(LOG_ERR, "Shared disk functionality not supported"); return; #endif } else if(sbd_is_pcmk(s)) { DBGLOG(LOG_INFO, "Starting Pacemaker servant"); s->pid = assign_servant(s->devname, servant_pcmk, start_mode, NULL); } else if(sbd_is_cluster(s)) { DBGLOG(LOG_INFO, "Starting Cluster servant"); s->pid = assign_servant(s->devname, servant_cluster, start_mode, NULL); } else { cl_log(LOG_ERR, "Unrecognized servant: %s", s->devname); } 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); + if (s->pid != 0) { + sigqueue_zero(s->pid, SIGKILL); + } } } static 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); + sigqueue_zero(ppid, SIG_LIVENESS); } 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, "%ld", &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 cluster_alive(bool all) { int alive = 1; struct servants_list_item* s; if(servant_count == disk_count) { return 0; } for (s = servants_leader; s; s = s->next) { if (sbd_is_cluster(s) || sbd_is_pcmk(s)) { if(s->outdated) { alive = 0; } else if(all == false) { return 1; } } } return alive; } int quorum_read(int good_servants) { if (disk_count > 2) return (good_servants > disk_count/2); else return (good_servants > 0); } 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 cluster_appeared = 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 %d IS ACTIVE - DO NOT RUN IN PRODUCTION!", debug_mode); } 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_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) { bool tickle = 0; bool can_detach = 0; int good_servants = 0; sig = sigtimedwait(&procmask, &sinfo, &timeout); clock_gettime(CLOCK_MONOTONIC, &t_now); 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 { s = lookup_servant_by_pid(pid); if (sbd_is_disk(s)) { if (WIFEXITED(status)) { switch(WEXITSTATUS(status)) { case EXIT_MD_SERVANT_IO_FAIL: DBGLOG(LOG_INFO, "Servant for %s requests to be disowned", s->devname); break; case EXIT_MD_SERVANT_REQUEST_RESET: cl_log(LOG_WARNING, "%s requested a reset", s->devname); do_reset(); break; case EXIT_MD_SERVANT_REQUEST_SHUTOFF: cl_log(LOG_WARNING, "%s requested a shutoff", s->devname); do_off(); break; case EXIT_MD_SERVANT_REQUEST_CRASHDUMP: cl_log(LOG_WARNING, "%s requested a crashdump", s->devname); do_crashdump(); break; default: break; } } } else if (sbd_is_pcmk(s)) { if (WIFEXITED(status)) { switch(WEXITSTATUS(status)) { case EXIT_PCMK_SERVANT_GRACEFUL_SHUTDOWN: DBGLOG(LOG_INFO, "PCMK-Servant has exited gracefully"); /* revert to state prior to pacemaker-detection */ s->restarts = 0; s->restart_blocked = 0; cluster_appeared = 0; s->outdated = 1; s->t_last.tv_sec = 0; break; default: break; } } } cleanup_servant_by_pid(pid); } } } else if (sig == SIG_PCMK_UNHEALTHY) { s = lookup_servant_by_pid(sinfo.si_pid); if (sbd_is_cluster(s) || sbd_is_pcmk(s)) { if (s->outdated == 0) { cl_log(LOG_WARNING, "%s health check: UNHEALTHY", s->devname); } s->t_last.tv_sec = 1; } else { cl_log(LOG_WARNING, "Ignoring SIG_PCMK_UNHEALTHY from unknown source"); } } else if (sig == SIG_LIVENESS) { s = lookup_servant_by_pid(sinfo.si_pid); if (s) { 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 (sbd_is_disk(s)) { good_servants++; } if (s->outdated) { cl_log(LOG_NOTICE, "Servant %s is healthy (age: %d)", s->devname, age); } s->outdated = 0; } else if (!s->outdated) { if (!s->restart_blocked) { cl_log(LOG_WARNING, "Servant %s is outdated (age: %d)", s->devname, age); } s->outdated = 1; } } if(disk_count == 0) { /* NO disks, everything is up to the cluster */ if(cluster_alive(true)) { /* We LIVE! */ if(cluster_appeared == false) { cl_log(LOG_INFO, "Active cluster detected"); } tickle = 1; can_detach = 1; cluster_appeared = 1; } else if(cluster_alive(false)) { if(!decoupled) { /* On the way up, detach and arm the watchdog */ cl_log(LOG_INFO, "Partial cluster detected, detaching"); } can_detach = 1; tickle = !cluster_appeared; } else if(!decoupled) { /* Stay alive until the cluster comes up */ tickle = !cluster_appeared; } } else if(disk_priority == 1 || servant_count == disk_count) { if (quorum_read(good_servants)) { /* There are disks and we're connected to the majority of them */ tickle = 1; can_detach = 1; pcmk_override = 0; } else if (servant_count > disk_count && cluster_alive(true)) { tickle = 1; if(!pcmk_override) { cl_log(LOG_WARNING, "Majority of devices lost - surviving on pacemaker"); pcmk_override = 1; /* Only log this message once */ } } } else if(cluster_alive(true) && quorum_read(good_servants)) { /* Both disk and cluster servants are healthy */ tickle = 1; can_detach = 1; cluster_appeared = 1; } else if(quorum_read(good_servants)) { /* The cluster takes priority but only once * connected for the first time. * * Until then, we tickle based on disk quorum. */ can_detach = 1; tickle = !cluster_appeared; } /* cl_log(LOG_DEBUG, "Tickle: q=%d, g=%d, p=%d, s=%d", */ /* quorum_read(good_servants), good_servants, tickle, disk_count); */ if(tickle) { watchdog_tickle(); clock_gettime(CLOCK_MONOTONIC, &t_last_tickle); } if (!decoupled && can_detach) { /* We only do this at the point either the disk or * cluster servants become healthy */ cl_log(LOG_DEBUG, "Decoupling"); if (inquisitor_decouple() < 0) { servants_kill(); exiting = 1; continue; } else { decoupled = 1; } } /* 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_timeout_action(); } 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 %ds exceeds watchdog warning timeout of %ds (healthy servants: %d)", (int)latency, (int)timeout_watchdog_warn, good_servants); if (debug_mode && watchdog_use) { /* In debug mode, trigger a reset before the watchdog can panic the machine */ do_timeout_action(); } } 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; } int parse_device_line(const char *line) { size_t lpc = 0; size_t last = 0; size_t max = 0; int found = 0; bool skip_space = true; int space_run = 0; if (!line) { return 0; } max = strlen(line); cl_log(LOG_DEBUG, "Processing %d bytes: [%s]", (int) max, line); for (lpc = 0; lpc <= max; lpc++) { if (isspace(line[lpc])) { if (skip_space) { last = lpc + 1; } else { space_run++; } continue; } skip_space = false; if (line[lpc] == ';' || line[lpc] == 0) { int rc = 0; char *entry = calloc(1, 1 + lpc - last); if (entry) { rc = sscanf(line + last, "%[^;]", entry); } else { fprintf(stderr, "Heap allocation failed parsing device-line.\n"); exit(1); } if (rc != 1) { cl_log(LOG_WARNING, "Could not parse: '%s'", line + last); } else { entry[strlen(entry)-space_run] = '\0'; cl_log(LOG_DEBUG, "Adding '%s'", entry); if (recruit_servant(entry, 0) != 0) { free(entry); // sbd should refuse to start if any of the configured device names is invalid. return -1; } found++; } free(entry); skip_space = true; last = lpc + 1; } space_run = 0; } return found; } #define SBD_SOURCE_FILES "sbd-cluster.c,sbd-common.c,sbd-inquisitor.c,sbd-md.c,sbd-pacemaker.c,setproctitle.c" static void sbd_log_filter_ctl(const char *files, uint8_t priority) { if (files == NULL) { files = SBD_SOURCE_FILES; } qb_log_filter_ctl(QB_LOG_SYSLOG, QB_LOG_FILTER_ADD, QB_LOG_FILTER_FILE, files, priority); qb_log_filter_ctl(QB_LOG_STDERR, QB_LOG_FILTER_ADD, QB_LOG_FILTER_FILE, files, priority); } int arg_enabled(int arg_count) { return arg_count % 2; } int main(int argc, char **argv, char **envp) { int exit_status = 0; int c; int W_count = 0; int c_count = 0; int P_count = 0; int qb_facility; const char *value = NULL; bool delay_start = false; long delay = 0; char *timeout_action = NULL; if ((cmdname = strrchr(argv[0], '/')) == NULL) { cmdname = argv[0]; } else { ++cmdname; } watchdogdev = strdup("/dev/watchdog"); watchdogdev_is_default = true; qb_facility = qb_log_facility2int("daemon"); qb_log_init(cmdname, qb_facility, LOG_WARNING); sbd_set_format_string(QB_LOG_SYSLOG, "sbd"); 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_log_filter_ctl(NULL, LOG_NOTICE); sbd_get_uname(); value = get_env_option("SBD_PACEMAKER"); if(value) { check_pcmk = crm_is_true(value); check_cluster = crm_is_true(value); has_check_pcmk_env = true; } cl_log(LOG_INFO, "SBD_PACEMAKER set to: %d (%s)", (int)check_pcmk, value?value:"default"); value = get_env_option("SBD_STARTMODE"); if(value == NULL) { } else if(strcmp(value, "clean") == 0) { start_mode = 1; } else if(strcmp(value, "always") == 0) { start_mode = 0; } cl_log(LOG_INFO, "Start mode set to: %d (%s)", (int)start_mode, value?value:"default"); value = get_env_option("SBD_WATCHDOG_DEV"); if(value) { free(watchdogdev); watchdogdev = strdup(value); watchdogdev_is_default = false; } /* SBD_WATCHDOG has been dropped from sbd.sysconfig example. * This is for backward compatibility. */ value = get_env_option("SBD_WATCHDOG"); if(value) { watchdog_use = crm_is_true(value); } value = get_env_option("SBD_WATCHDOG_TIMEOUT"); if(value) { timeout_watchdog = crm_get_msec(value) / 1000; } value = get_env_option("SBD_PIDFILE"); if(value) { pidfile = strdup(value); cl_log(LOG_INFO, "pidfile set to %s", pidfile); } value = get_env_option("SBD_DELAY_START"); if(value) { delay_start = crm_is_true(value); if (!delay_start) { delay = crm_get_msec(value) / 1000; if (delay > 0) { delay_start = true; } } } value = get_env_option("SBD_TIMEOUT_ACTION"); if(value) { timeout_action = strdup(value); } value = get_env_option("SBD_MOVE_TO_ROOT_CGROUP"); if(value) { move_to_root_cgroup = crm_is_true(value); if (move_to_root_cgroup) { enforce_moving_to_root_cgroup = true; } else { if (strcmp(value, "auto") == 0) { move_to_root_cgroup = true; } } } while ((c = getopt(argc, argv, "czC:DPRTWZhvw:d:n:p:1:2:3:4:5:t:I:F:S:s:r:")) != -1) { int sanitized_num_optarg = 0; /* Call it before checking optarg for NULL to make coverity happy */ const char *sanitized_optarg = sanitize_option_value(optarg); if (optarg && ((sanitized_optarg == NULL) || (strchr("SsC12345tIF", c) && (sanitized_num_optarg = sanitize_numeric_option_value(sanitized_optarg)) < 0))) { fprintf(stderr, "Invalid value \"%s\" for option -%c\n", optarg, c); exit_status = -2; goto out; } 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 = sanitized_num_optarg; cl_log(LOG_INFO, "Start mode set to: %d", (int)start_mode); break; case 's': timeout_startup = sanitized_num_optarg; cl_log(LOG_INFO, "Start timeout set to: %d", (int)timeout_startup); break; case 'v': debug++; if(debug == 1) { sbd_log_filter_ctl(NULL, LOG_INFO); cl_log(LOG_INFO, "Verbose mode enabled."); } else if(debug == 2) { sbd_log_filter_ctl(NULL, LOG_DEBUG); cl_log(LOG_INFO, "Debug mode enabled."); } else if(debug == 3) { /* Go nuts, turn on pacemaker's logging too */ sbd_log_filter_ctl("*", LOG_DEBUG); cl_log(LOG_INFO, "Debug library mode enabled."); } break; case 'T': watchdog_set_timeout = 0; cl_log(LOG_INFO, "Setting watchdog timeout disabled; using defaults."); break; case 'W': W_count++; break; case 'w': free(watchdogdev); watchdogdev = strdup(sanitized_optarg); watchdogdev_is_default = false; cl_log(LOG_NOTICE, "Using watchdog device '%s'", watchdogdev); break; case 'd': #if SUPPORT_SHARED_DISK if (recruit_servant(sanitized_optarg, 0) != 0) { fprintf(stderr, "Invalid device: %s\n", optarg); exit_status = -1; goto out; } #else fprintf(stderr, "Shared disk functionality not supported\n"); exit_status = -2; goto out; #endif break; case 'c': c_count++; break; case 'P': P_count++; break; case 'z': disk_priority = 0; break; case 'n': local_uname = strdup(sanitized_optarg); cl_log(LOG_INFO, "Overriding local hostname to %s", local_uname); break; case 'p': pidfile = strdup(sanitized_optarg); cl_log(LOG_INFO, "pidfile set to %s", pidfile); break; case 'C': timeout_watchdog_crashdump = sanitized_num_optarg; cl_log(LOG_INFO, "Setting crashdump watchdog timeout to %d", (int)timeout_watchdog_crashdump); break; case '1': timeout_watchdog = sanitized_num_optarg; break; case '2': timeout_allocate = sanitized_num_optarg; break; case '3': timeout_loop = sanitized_num_optarg; break; case '4': timeout_msgwait = sanitized_num_optarg; break; case '5': timeout_watchdog_warn = sanitized_num_optarg; do_calculate_timeout_watchdog_warn = false; cl_log(LOG_INFO, "Setting latency warning to %d", (int)timeout_watchdog_warn); break; case 't': servant_restart_interval = sanitized_num_optarg; cl_log(LOG_INFO, "Setting servant restart interval to %d", (int)servant_restart_interval); break; case 'I': timeout_io = sanitized_num_optarg; cl_log(LOG_INFO, "Setting IO timeout to %d", (int)timeout_io); break; case 'F': servant_restart_count = sanitized_num_optarg; cl_log(LOG_INFO, "Servant restart count set to %d", (int)servant_restart_count); break; case 'r': if (timeout_action) { free(timeout_action); } timeout_action = strdup(sanitized_optarg); break; case 'h': usage(); goto out; break; default: exit_status = -2; goto out; break; } } if (disk_count == 0) { /* if we already have disks from commandline then it is probably undesirable to add those from environment (general rule cmdline has precedence) */ value = get_env_option("SBD_DEVICE"); if ((value) && strlen(value)) { #if SUPPORT_SHARED_DISK int devices = parse_device_line(value); if(devices < 1) { fprintf(stderr, "Invalid device line: %s\n", value); exit_status = -1; goto out; } #else fprintf(stderr, "Shared disk functionality not supported\n"); exit_status = -2; goto out; #endif } } if (watchdogdev == NULL || strcmp(watchdogdev, "/dev/null") == 0) { watchdog_use = 0; } else if (W_count > 0) { watchdog_use = arg_enabled(W_count); } if (watchdog_use) { cl_log(LOG_INFO, "Watchdog enabled."); } else { cl_log(LOG_INFO, "Watchdog disabled."); } if (c_count > 0) { check_cluster = arg_enabled(c_count); } if (P_count > 0) { int check_pcmk_arg = arg_enabled(P_count); if (has_check_pcmk_env && check_pcmk_arg != check_pcmk) { cl_log(LOG_WARNING, "Pacemaker integration is %s: " "SBD_PACEMAKER=%s is overridden by %s option. " "It's recommended to only use SBD_PACEMAKER.", check_pcmk_arg? "enabled" : "disabled", check_pcmk? "yes" : "no", check_pcmk_arg? "-P" : "-PP"); } check_pcmk = check_pcmk_arg; } if ((disk_count > 0) && (strlen(local_uname) > SECTOR_NAME_MAX)) { fprintf(stderr, "Node name mustn't be longer than %d chars.\n", SECTOR_NAME_MAX); fprintf(stderr, "If uname is longer define a name to be used by sbd.\n"); exit_status = -1; goto out; } if (disk_count > 3) { fprintf(stderr, "You can specify up 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; } if (timeout_action) { char *p[2]; int i; char c; int nrflags = sscanf(timeout_action, "%m[a-z],%m[a-z]%c", &p[0], &p[1], &c); bool parse_error = (nrflags < 1) || (nrflags > 2); for (i = 0; (i < nrflags) && (i < 2); i++) { if (!strcmp(p[i], "reboot")) { timeout_sysrq_char = 'b'; } else if (!strcmp(p[i], "crashdump")) { timeout_sysrq_char = 'c'; } else if (!strcmp(p[i], "off")) { timeout_sysrq_char = 'o'; } else if (!strcmp(p[i], "flush")) { do_flush = true; } else if (!strcmp(p[i], "noflush")) { do_flush = false; } else { parse_error = true; } free(p[i]); } if (parse_error) { fprintf(stderr, "Failed to parse timeout-action \"%s\".\n", timeout_action); exit_status = -1; goto out; } } if (strcmp(argv[optind], "watch") == 0) { value = get_env_option("SBD_SYNC_RESOURCE_STARTUP"); sync_resource_startup = crm_is_true(value?value:SBD_SYNC_RESOURCE_STARTUP_DEFAULT); #if !USE_PACEMAKERD_API if (sync_resource_startup) { fprintf(stderr, "Failed to sync resource-startup as " "SBD was built against pacemaker not supporting pacemakerd-API.\n"); exit_status = -1; goto out; } #else if (check_pcmk && !sync_resource_startup) { cl_log(LOG_WARNING, "SBD built against pacemaker supporting " "pacemakerd-API. Should think about enabling " "SBD_SYNC_RESOURCE_STARTUP."); } else if (!check_pcmk && sync_resource_startup) { fprintf(stderr, "Set SBD_PACEMAKER=yes to allow resource startup syncing. " "Otherwise explicitly set SBD_SYNC_RESOURCE_STARTUP=no if to intentionally disable.\n"); exit_status = -1; goto out; } #endif } #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 #endif if (strcmp(argv[optind], "query-watchdog") == 0) { exit_status = watchdog_info(); } else if (strcmp(argv[optind], "test-watchdog") == 0) { exit_status = watchdog_test(); } else if (strcmp(argv[optind], "watch") == 0) { /* sleep $(sbd $SBD_DEVICE_ARGS dump | grep -m 1 msgwait | awk '{print $4}') 2>/dev/null */ const char *delay_source = delay ? "SBD_DELAY_START" : ""; #if SUPPORT_SHARED_DISK if(disk_count > 0) { /* If no devices are specified, its not an error to be unable to find one */ open_any_device(servants_leader); if (delay_start && delay <= 0) { delay = get_first_msgwait(servants_leader); if (delay > 0) { delay_source = "msgwait"; } else { cl_log(LOG_WARNING, "No 'msgwait' value from disk, using '2 * watchdog-timeout' for 'delay' starting"); } } } #endif /* Re-calculate timeout_watchdog_warn based on any timeout_watchdog from: * SBD_WATCHDOG_TIMEOUT, -1 option or on-disk setting read with open_any_device() */ if (do_calculate_timeout_watchdog_warn) { timeout_watchdog_warn = calculate_timeout_watchdog_warn(timeout_watchdog); } if (delay_start) { /* diskless mode or disk read issues causing get_first_msgwait() to return a 0 for delay */ if (delay <= 0) { delay = 2 * timeout_watchdog; delay_source = "watchdog-timeout * 2"; } cl_log(LOG_DEBUG, "Delay start (yes), (delay: %ld), (delay source: %s)", delay, delay_source); sleep((unsigned long) delay); } else { cl_log(LOG_DEBUG, "Delay start (no)"); } /* We only want this to have an effect during watch right now; * pinging and fencing would be too confused */ cl_log(LOG_INFO, "Turning on pacemaker checks: %d", check_pcmk); if (check_pcmk) { recruit_servant("pcmk", 0); #if SUPPORT_PLUGIN check_cluster = 1; #endif } cl_log(LOG_INFO, "Turning on cluster checks: %d", check_cluster); if (check_cluster) { recruit_servant("cluster", 0); } cl_log(LOG_NOTICE, "%s flush + write \'%c\' to sysrq in case of timeout", do_flush?"Do":"Skip", timeout_sysrq_char); exit_status = inquisitor(); } else { exit_status = -2; } out: if (timeout_action) { free(timeout_action); } 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 2a237ad..71c317f 100644 --- a/src/sbd-md.c +++ b/src/sbd-md.c @@ -1,1288 +1,1285 @@ /* * 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 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 program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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 disk_count; /* 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) { if (!st) { return; } if (st->ioctx) { io_destroy(st->ioctx); } if (st->devfd >= 0) { close(st->devfd); } free(st->buffer); free(st); } static struct sbd_context * open_device(const char* devname, int loglevel) { struct sbd_context *st; if (!devname) return NULL; st = calloc(1, sizeof(struct sbd_context)); if (!st) { return NULL; } st->devfd = -1; if (io_setup(1, &st->ioctx) != 0) { cl_perror("io_setup failed"); goto out; } 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); } goto out; } ioctl(st->devfd, BLKSSZGET, §or_size); if (sector_size == 0) { cl_perror("Get sector size failed.\n"); goto out; } if (posix_memalign(&st->buffer, sector_size, sector_size)) { cl_perror("Couldn't allocate sector-buffer."); goto out; } return st; out: close_device(st); return NULL; } static void * sector_alloc(void) { void *x; x = calloc(1, sector_size); if (!x) { exit(1); } 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) { memcpy(st->buffer, data, sector_size); io_prep_pwrite(&st->io, st->devfd, st->buffer, sector_size, (long long) sector_size * sector); } else { memset(st->buffer, 0, sector_size); io_prep_pread(&st->io, st->devfd, st->buffer, sector_size, (long long) 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, r=%ld)", rw, r); 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) { if (!rw) { memcpy(data, st->buffer, 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); free(s_header); return NULL; } if (valid_header(s_header) < 0) { cl_log(LOG_ERR, "header on device %d is not valid.", st->devfd); free(s_header); 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); free(s_header); 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; 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); 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_mbox); free(s_node); free(s_header); 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, SECTOR_NAME_MAX) == 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, SECTOR_NAME_MAX); 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_mbox); free(s_node); free(s_header); 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_mbox); free(s_node); free(s_header); 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, SECTOR_NAME_MAX); 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, SECTOR_NAME_MAX); 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); } fprintf(stdout, "Did you check sbd service down on all nodes before? If not do so now and restart afterwards.\n"); 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) { int rv = 0; st = open_device(s->devname, LOG_WARNING); if (!st) { rc = -1; fprintf(stderr, "== disk %s unreadable!\n", s->devname); continue; } rv = slot_list(st); close_device(st); if (rv == -1) { rc = -1; fprintf(stderr, "== Slots on disk %s NOT dumped\n", s->devname); } } return rc; } 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) { if(sbd_is_disk(s)) { s->pid = assign_servant(s->devname, &slot_ping_wrapper, 0, (const void*)name); } } while (servants_finished < disk_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 (sbd_is_disk(s)) { servants_finished++; } } } } } return 0; } int quorum_write(int good_servants) { return (good_servants > disk_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 == disk_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); free(s_header); 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) { int rv; fprintf(stdout, "==Dumping header on disk %s\n", s->devname); st = open_device(s->devname, LOG_WARNING); if (st) { rv = header_dump(st); close_device(st); } else { fprintf(stderr, "== disk %s unreadable!\n", s->devname); rv = -1; } if (rv == -1) { rc = -1; fprintf(stderr, "==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; bool logged_once = false; 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) { if (logged_once == false) { cl_log(LOG_WARNING, "Failed to open %s. " "Trying any other configured devices, " "otherwise retrying every %ds within %ds", s->devname, timeout_loop, timeout_startup); logged_once = true; } continue; } hdr_cur = header_get(st); close_device(st); if (hdr_cur) { break; } else { if (logged_once == false) { cl_log(LOG_WARNING, "Failed to read header from %s. " "Trying any other configured devices, " "otherwise retrying every %ds within %ds", s->devname, timeout_loop, timeout_startup); logged_once = true; } } } 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; } int servant_md(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; 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); st = open_device(diskname, LOG_WARNING); if (!st) { exit(EXIT_MD_SERVANT_IO_FAIL); } s_header = header_get(st); if (!s_header) { cl_log(LOG_ERR, "Not a valid header on %s", diskname); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } if (servant_check_timeout_inconsistent(s_header) < 0) { cl_log(LOG_ERR, "Timeouts on %s do not match first device", diskname); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } 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 = EXIT_MD_SERVANT_IO_FAIL; 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); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } cl_log(LOG_NOTICE, "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 = EXIT_MD_SERVANT_IO_FAIL; 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); + sigqueue_zero(ppid, SIG_EXITREQ); 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 = EXIT_MD_SERVANT_IO_FAIL; 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_timeout_action(); } /* 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); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } if (memcmp(s_header, s_header_retry, sizeof(*s_header)) != 0) { cl_log(LOG_ERR, "Header on %s changed since start-up!", diskname); free(s_header_retry); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } 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."); free(s_node_retry); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } if (memcmp(s_node, s_node_retry, sizeof(*s_node)) != 0) { cl_log(LOG_ERR, "Node entry on %s changed since start-up!", diskname); free(s_node_retry); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } free(s_node_retry); if (mbox_read(st, mbox, s_mbox) < 0) { cl_log(LOG_ERR, "mbox read failed in servant."); rc = EXIT_MD_SERVANT_IO_FAIL; goto out; } if (s_mbox->cmd > 0) { cl_log(LOG_NOTICE, "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); + sigqueue_zero(ppid, SIG_TEST); break; case SBD_MSG_RESET: rc = EXIT_MD_SERVANT_REQUEST_RESET; goto out; case SBD_MSG_OFF: rc = EXIT_MD_SERVANT_REQUEST_SHUTOFF; goto out; case SBD_MSG_EXIT: - sigqueue(ppid, SIG_EXITREQ, signal_value); + sigqueue_zero(ppid, SIG_EXITREQ); break; case SBD_MSG_CRASHDUMP: rc = EXIT_MD_SERVANT_REQUEST_CRASHDUMP; goto out; 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); + sigqueue_zero(ppid, SIG_LIVENESS); t1 = time(NULL); latency = t1 - t0; if (timeout_watchdog_warn && (latency > timeout_watchdog_warn)) { cl_log(LOG_WARNING, "Latency: %ds exceeded watchdog warning timeout %ds on disk %s", (int)latency, (int)timeout_watchdog_warn, diskname); } else if (debug) { DBGLOG(LOG_DEBUG, "Latency: %ds on disk %s", (int)latency, diskname); } } out: free(s_node); free(s_mbox); free(s_header); close_device(st); exit(rc); } diff --git a/src/sbd.h b/src/sbd.h index bbdc6f1..7874e8d 100644 --- a/src/sbd.h +++ b/src/sbd.h @@ -1,219 +1,220 @@ /* * 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 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 program; if not, write to the Free Software Foundation, Inc., * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 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 #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_PCMK_UNHEALTHY (SIGRTMIN + 5) /* FIXME: should add dynamic check of SIG_XX >= SIGRTMAX */ /* exit status for disk-servant */ #define EXIT_MD_SERVANT_IO_FAIL 20 #define EXIT_MD_SERVANT_REQUEST_RESET 21 #define EXIT_MD_SERVANT_REQUEST_SHUTOFF 22 #define EXIT_MD_SERVANT_REQUEST_CRASHDUMP 23 /* exit status for pcmk-servant */ #define EXIT_PCMK_SERVANT_GRACEFUL_SHUTDOWN 30 #define HOG_CHAR 0xff #define SECTOR_NAME_MAX 63 /* 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[SECTOR_NAME_MAX+1]; }; struct sector_node_s { /* slots will be created with in_use == 0 */ char in_use; char name[SECTOR_NAME_MAX+1]; }; 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 *buffer; }; enum pcmk_health { pcmk_health_unknown, pcmk_health_pending, pcmk_health_transient, pcmk_health_unclean, pcmk_health_shutdown, pcmk_health_online, pcmk_health_noquorum, }; 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); int watchdog_info(void); int watchdog_test(void); void sysrq_trigger(char t); void do_crashdump(void); void do_reset(void); void do_off(void); void do_timeout_action(void); pid_t make_daemon(void); void maximize_priority(void); void sbd_get_uname(void); void sbd_set_format_string(int method, const char *daemon); +int sigqueue_zero(pid_t pid, int sig); void notify_parent(void); /* Tunable defaults: */ extern unsigned long timeout_watchdog; extern unsigned long timeout_watchdog_warn; extern bool do_calculate_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 char *watchdogdev; extern bool watchdogdev_is_default; extern char* local_uname; extern bool do_flush; extern char timeout_sysrq_char; extern bool move_to_root_cgroup; extern bool enforce_moving_to_root_cgroup; extern bool sync_resource_startup; /* 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_md(const char *diskname, int mode, const void* argp); #endif int servant_pcmk(const char *diskname, int mode, const void* argp); int servant_cluster(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) extern int servant_health; void set_servant_health(enum pcmk_health state, int level, char const *format, ...) __attribute__ ((__format__ (__printf__, 3, 4))); bool sbd_is_disk(struct servants_list_item *servant); bool sbd_is_pcmk(struct servants_list_item *servant); bool sbd_is_cluster(struct servants_list_item *servant); #define calculate_timeout_watchdog_warn(timeout) \ (timeout < 5 ? 2 : \ (timeout < (ULONG_MAX / 3) ? \ (((unsigned long) timeout) * 3 / 5) : (((unsigned long) timeout) / 5 * 3)))