diff --git a/src/sbd-inquisitor.c b/src/sbd-inquisitor.c
index c1f8369..cdfa226 100644
--- a/src/sbd-inquisitor.c
+++ b/src/sbd-inquisitor.c
@@ -1,807 +1,807 @@
/*
* Copyright (C) 2013 Lars Marowsky-Bree <lmb@suse.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "sbd.h"
#define LOCKSTRLEN 11
struct servants_list_item *servants_leader = NULL;
extern int check_pcmk;
int servant_count = 0;
int servant_restart_interval = 5;
int servant_restart_count = 1;
int start_mode = 0;
char* pidfile = NULL;
void recruit_servant(const char *devname, pid_t pid)
{
struct servants_list_item *s = servants_leader;
struct servants_list_item *newbie;
newbie = malloc(sizeof(*newbie));
if (!newbie) {
fprintf(stderr, "malloc failed in recruit_servant.\n");
exit(1);
}
memset(newbie, 0, sizeof(*newbie));
newbie->devname = strdup(devname);
newbie->pid = pid;
newbie->first_start = 1;
if (!s) {
servants_leader = newbie;
} else {
while (s->next)
s = s->next;
s->next = newbie;
}
servant_count++;
}
-int assign_servant(const char* devname, functionp_t functionp, const void* argp)
+int assign_servant(const char* devname, functionp_t functionp, int mode, const void* argp)
{
pid_t pid = 0;
int rc = 0;
pid = fork();
if (pid == 0) { /* child */
maximize_priority();
- rc = (*functionp)(devname, argp);
+ rc = (*functionp)(devname, mode, argp);
if (rc == -1)
exit(1);
else
exit(0);
} else if (pid != -1) { /* parent */
return pid;
} else {
cl_log(LOG_ERR,"Failed to fork servant");
exit(1);
}
}
struct servants_list_item *lookup_servant_by_dev(const char *devname)
{
struct servants_list_item *s;
for (s = servants_leader; s; s = s->next) {
if (strncasecmp(s->devname, devname, strlen(s->devname)))
break;
}
return s;
}
struct servants_list_item *lookup_servant_by_pid(pid_t pid)
{
struct servants_list_item *s;
for (s = servants_leader; s; s = s->next) {
if (s->pid == pid)
break;
}
return s;
}
int check_all_dead(void)
{
struct servants_list_item *s;
int r = 0;
union sigval svalue;
for (s = servants_leader; s; s = s->next) {
if (s->pid != 0) {
r = sigqueue(s->pid, 0, svalue);
if (r == -1 && errno == ESRCH)
continue;
return 0;
}
}
return 1;
}
void servant_start(struct servants_list_item *s)
{
int r = 0;
union sigval svalue;
if (s->pid != 0) {
r = sigqueue(s->pid, 0, svalue);
if ((r != -1 || errno != ESRCH))
return;
}
s->restarts++;
if (strcmp("pcmk",s->devname) == 0) {
DBGLOG(LOG_INFO, "Starting Pacemaker servant");
- s->pid = assign_servant(s->devname, servant_pcmk, NULL);
+ s->pid = assign_servant(s->devname, servant_pcmk, start_mode, NULL);
} else {
DBGLOG(LOG_INFO, "Starting servant for device %s",
s->devname);
- s->pid = assign_servant(s->devname, servant, s);
+ s->pid = assign_servant(s->devname, servant, start_mode, s);
}
clock_gettime(CLOCK_MONOTONIC, &s->t_started);
return;
}
void servants_start(void)
{
struct servants_list_item *s;
for (s = servants_leader; s; s = s->next) {
s->restarts = 0;
servant_start(s);
}
}
void servants_kill(void)
{
struct servants_list_item *s;
union sigval svalue;
for (s = servants_leader; s; s = s->next) {
if (s->pid != 0)
sigqueue(s->pid, SIGKILL, svalue);
}
}
inline void cleanup_servant_by_pid(pid_t pid)
{
struct servants_list_item* s;
s = lookup_servant_by_pid(pid);
if (s) {
cl_log(LOG_WARNING, "Servant for %s (pid: %i) has terminated",
s->devname, s->pid);
s->pid = 0;
} else {
/* This most likely is a stray signal from somewhere, or
* a SIGCHLD for a process that has previously
* explicitly disconnected. */
DBGLOG(LOG_INFO, "cleanup_servant: Nothing known about pid %i",
pid);
}
}
int inquisitor_decouple(void)
{
pid_t ppid = getppid();
union sigval signal_value;
/* During start-up, we only arm the watchdog once we've got
* quorum at least once. */
if (watchdog_use) {
if (watchdog_init() < 0) {
return -1;
}
}
if (ppid > 1) {
sigqueue(ppid, SIG_LIVENESS, signal_value);
}
return 0;
}
static int sbd_lock_running(long pid)
{
int rc = 0;
long mypid;
int running = 0;
char proc_path[PATH_MAX], exe_path[PATH_MAX], myexe_path[PATH_MAX];
/* check if pid is running */
if (kill(pid, 0) < 0 && errno == ESRCH) {
goto bail;
}
#ifndef HAVE_PROC_PID
return 1;
#endif
/* check to make sure pid hasn't been reused by another process */
snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", pid);
rc = readlink(proc_path, exe_path, PATH_MAX-1);
if(rc < 0) {
cl_perror("Could not read from %s", proc_path);
goto bail;
}
exe_path[rc] = 0;
mypid = (unsigned long) getpid();
snprintf(proc_path, sizeof(proc_path), "/proc/%lu/exe", mypid);
rc = readlink(proc_path, myexe_path, PATH_MAX-1);
if(rc < 0) {
cl_perror("Could not read from %s", proc_path);
goto bail;
}
myexe_path[rc] = 0;
if(strcmp(exe_path, myexe_path) == 0) {
running = 1;
}
bail:
return running;
}
static int
sbd_lock_pidfile(const char *filename)
{
char lf_name[256], tf_name[256], buf[LOCKSTRLEN+1];
int fd;
long pid, mypid;
int rc;
struct stat sbuf;
if (filename == NULL) {
errno = EFAULT;
return -1;
}
mypid = (unsigned long) getpid();
snprintf(lf_name, sizeof(lf_name), "%s",filename);
snprintf(tf_name, sizeof(tf_name), "%s.%lu",
filename, mypid);
if ((fd = open(lf_name, O_RDONLY)) >= 0) {
if (fstat(fd, &sbuf) >= 0 && sbuf.st_size < LOCKSTRLEN) {
sleep(1); /* if someone was about to create one,
* give'm a sec to do so
* Though if they follow our protocol,
* this won't happen. They should really
* put the pid in, then link, not the
* other way around.
*/
}
if (read(fd, buf, sizeof(buf)) < 1) {
/* lockfile empty -> rm it and go on */;
} else {
if (sscanf(buf, "%lu", &pid) < 1) {
/* lockfile screwed up -> rm it and go on */
} else {
if (pid > 1 && (getpid() != pid)
&& sbd_lock_running(pid)) {
/* is locked by existing process
* -> give up */
close(fd);
return -1;
} else {
/* stale lockfile -> rm it and go on */
}
}
}
unlink(lf_name);
close(fd);
}
if ((fd = open(tf_name, O_CREAT | O_WRONLY | O_EXCL, 0644)) < 0) {
/* Hmmh, why did we fail? Anyway, nothing we can do about it */
return -3;
}
/* Slight overkill with the %*d format ;-) */
snprintf(buf, sizeof(buf), "%*lu\n", LOCKSTRLEN-1, mypid);
if (write(fd, buf, LOCKSTRLEN) != LOCKSTRLEN) {
/* Again, nothing we can do about this */
rc = -3;
close(fd);
goto out;
}
close(fd);
switch (link(tf_name, lf_name)) {
case 0:
if (stat(tf_name, &sbuf) < 0) {
/* something weird happened */
rc = -3;
break;
}
if (sbuf.st_nlink < 2) {
/* somehow, it didn't get through - NFS trouble? */
rc = -2;
break;
}
rc = 0;
break;
case EEXIST:
rc = -1;
break;
default:
rc = -3;
}
out:
unlink(tf_name);
return rc;
}
/*
* Unlock a file (remove its lockfile)
* do we need to check, if its (still) ours? No, IMHO, if someone else
* locked our line, it's his fault -tho
* returns 0 on success
* <0 if some failure occured
*/
static int
sbd_unlock_pidfile(const char *filename)
{
char lf_name[256];
if (filename == NULL) {
errno = EFAULT;
return -1;
}
snprintf(lf_name, sizeof(lf_name), "%s", filename);
return unlink(lf_name);
}
void inquisitor_child(void)
{
int sig, pid;
sigset_t procmask;
siginfo_t sinfo;
int status;
struct timespec timeout;
int exiting = 0;
int decoupled = 0;
int pcmk_healthy = 0;
int pcmk_override = 0;
time_t latency;
struct timespec t_last_tickle, t_now;
struct servants_list_item* s;
if (debug_mode) {
cl_log(LOG_ERR, "DEBUG MODE IS ACTIVE - DO NOT RUN IN PRODUCTION!");
}
set_proc_title("sbd: inquisitor");
if (pidfile) {
if (sbd_lock_pidfile(pidfile) < 0) {
exit(1);
}
}
sigemptyset(&procmask);
sigaddset(&procmask, SIGCHLD);
sigaddset(&procmask, SIG_LIVENESS);
sigaddset(&procmask, SIG_EXITREQ);
sigaddset(&procmask, SIG_TEST);
sigaddset(&procmask, SIG_IO_FAIL);
sigaddset(&procmask, SIG_PCMK_UNHEALTHY);
sigaddset(&procmask, SIG_RESTART);
sigaddset(&procmask, SIGUSR1);
sigaddset(&procmask, SIGUSR2);
sigprocmask(SIG_BLOCK, &procmask, NULL);
servants_start();
timeout.tv_sec = timeout_loop;
timeout.tv_nsec = 0;
clock_gettime(CLOCK_MONOTONIC, &t_last_tickle);
while (1) {
int good_servants = 0;
sig = sigtimedwait(&procmask, &sinfo, &timeout);
clock_gettime(CLOCK_MONOTONIC, &t_now);
if (sig == SIG_EXITREQ) {
servants_kill();
watchdog_close(true);
exiting = 1;
} else if (sig == SIGCHLD) {
while ((pid = waitpid(-1, &status, WNOHANG))) {
if (pid == -1 && errno == ECHILD) {
break;
} else {
cleanup_servant_by_pid(pid);
}
}
} else if (sig == SIG_PCMK_UNHEALTHY) {
s = lookup_servant_by_pid(sinfo.si_pid);
if (s && strcmp(s->devname, "pcmk") == 0) {
if (pcmk_healthy != 0) {
cl_log(LOG_WARNING, "Pacemaker health check: UNHEALTHY");
}
pcmk_healthy = 0;
clock_gettime(CLOCK_MONOTONIC, &s->t_last);
} else {
cl_log(LOG_WARNING, "Ignoring SIG_PCMK_UNHEALTHY from unknown source");
}
} else if (sig == SIG_IO_FAIL) {
s = lookup_servant_by_pid(sinfo.si_pid);
if (s) {
DBGLOG(LOG_INFO, "Servant for %s requests to be disowned",
s->devname);
cleanup_servant_by_pid(sinfo.si_pid);
}
} else if (sig == SIG_LIVENESS) {
s = lookup_servant_by_pid(sinfo.si_pid);
if (s) {
if (strcmp(s->devname, "pcmk") == 0) {
if (pcmk_healthy != 1) {
cl_log(LOG_INFO, "Pacemaker health check: OK");
}
pcmk_healthy = 1;
};
s->first_start = 0;
clock_gettime(CLOCK_MONOTONIC, &s->t_last);
}
} else if (sig == SIG_TEST) {
} else if (sig == SIGUSR1) {
if (exiting)
continue;
servants_start();
}
if (exiting) {
if (check_all_dead()) {
if (pidfile) {
sbd_unlock_pidfile(pidfile);
}
exit(0);
} else
continue;
}
good_servants = 0;
for (s = servants_leader; s; s = s->next) {
int age = t_now.tv_sec - s->t_last.tv_sec;
if (!s->t_last.tv_sec)
continue;
if (age < (int)(timeout_io+timeout_loop)) {
if (strcmp(s->devname, "pcmk") != 0) {
good_servants++;
}
s->outdated = 0;
} else if (!s->outdated) {
if (strcmp(s->devname, "pcmk") == 0) {
/* If the state is outdated, we
* override the last reported
* state */
pcmk_healthy = 0;
cl_log(LOG_WARNING, "Pacemaker state outdated (age: %d)",
age);
} else if (!s->restart_blocked) {
cl_log(LOG_WARNING, "Servant for %s outdated (age: %d)",
s->devname, age);
}
s->outdated = 1;
}
}
if (quorum_read(good_servants) || pcmk_healthy) {
if (!decoupled) {
if (inquisitor_decouple() < 0) {
servants_kill();
exiting = 1;
continue;
} else {
decoupled = 1;
}
}
if (!quorum_read(good_servants)) {
if (!pcmk_override) {
cl_log(LOG_WARNING, "Majority of devices lost - surviving on pacemaker");
pcmk_override = 1; /* Just to ensure the message is only logged once */
}
} else {
pcmk_override = 0;
}
watchdog_tickle();
clock_gettime(CLOCK_MONOTONIC, &t_last_tickle);
}
/* Note that this can actually be negative, since we set
* last_tickle after we set now. */
latency = t_now.tv_sec - t_last_tickle.tv_sec;
if (timeout_watchdog && (latency > (int)timeout_watchdog)) {
if (!decoupled) {
/* We're still being watched by our
* parent. We don't fence, but exit. */
cl_log(LOG_ERR, "SBD: Not enough votes to proceed. Aborting start-up.");
servants_kill();
exiting = 1;
continue;
}
if (debug_mode < 2) {
/* At level 2 or above, we do nothing, but expect
* things to eventually return to
* normal. */
do_reset();
} else {
cl_log(LOG_ERR, "SBD: DEBUG MODE: Would have fenced due to timeout!");
}
}
if (timeout_watchdog_warn && (latency > (int)timeout_watchdog_warn)) {
cl_log(LOG_WARNING,
"Latency: No liveness for %d s exceeds threshold of %d s (healthy servants: %d)",
(int)latency, (int)timeout_watchdog_warn, good_servants);
}
for (s = servants_leader; s; s = s->next) {
int age = t_now.tv_sec - s->t_started.tv_sec;
if (age > servant_restart_interval) {
s->restarts = 0;
s->restart_blocked = 0;
}
if (servant_restart_count
&& (s->restarts >= servant_restart_count)
&& !s->restart_blocked) {
if (servant_restart_count > 1) {
cl_log(LOG_WARNING, "Max retry count (%d) reached: not restarting servant for %s",
(int)servant_restart_count, s->devname);
}
s->restart_blocked = 1;
}
if (!s->restart_blocked) {
servant_start(s);
}
}
}
/* not reached */
exit(0);
}
int inquisitor(void)
{
int sig, pid, inquisitor_pid;
int status;
sigset_t procmask;
siginfo_t sinfo;
/* Where's the best place for sysrq init ?*/
sysrq_init();
sigemptyset(&procmask);
sigaddset(&procmask, SIGCHLD);
sigaddset(&procmask, SIG_LIVENESS);
sigprocmask(SIG_BLOCK, &procmask, NULL);
inquisitor_pid = make_daemon();
if (inquisitor_pid == 0) {
inquisitor_child();
}
/* We're the parent. Wait for a happy signal from our child
* before we proceed - we either get "SIG_LIVENESS" when the
* inquisitor has completed the first successful round, or
* ECHLD when it exits with an error. */
while (1) {
sig = sigwaitinfo(&procmask, &sinfo);
if (sig == SIGCHLD) {
while ((pid = waitpid(-1, &status, WNOHANG))) {
if (pid == -1 && errno == ECHILD) {
break;
}
/* We got here because the inquisitor
* did not succeed. */
return -1;
}
} else if (sig == SIG_LIVENESS) {
/* Inquisitor started up properly. */
return 0;
} else {
fprintf(stderr, "Nobody expected the spanish inquisition!\n");
continue;
}
}
/* not reached */
return -1;
}
int main(int argc, char **argv, char **envp)
{
int exit_status = 0;
int c;
int w = 0;
int qb_facility;
if ((cmdname = strrchr(argv[0], '/')) == NULL) {
cmdname = argv[0];
} else {
++cmdname;
}
qb_facility = qb_log_facility2int("daemon");
qb_log_init(cmdname, qb_facility, LOG_ERR);
qb_log_ctl(QB_LOG_SYSLOG, QB_LOG_CONF_ENABLED, QB_TRUE);
qb_log_ctl(QB_LOG_STDERR, QB_LOG_CONF_ENABLED, QB_FALSE);
sbd_get_uname();
while ((c = getopt(argc, argv, "C:DPRTWZhvw:d:n:p:1:2:3:4:5:t:I:F:S:s:")) != -1) {
switch (c) {
case 'D':
break;
case 'Z':
debug_mode++;
cl_log(LOG_INFO, "Debug mode now at level %d", (int)debug_mode);
break;
case 'R':
skip_rt = 1;
cl_log(LOG_INFO, "Realtime mode deactivated.");
break;
case 'S':
start_mode = atoi(optarg);
cl_log(LOG_INFO, "Start mode set to: %d", (int)start_mode);
break;
case 's':
timeout_startup = atoi(optarg);
cl_log(LOG_INFO, "Start timeout set to: %d", (int)timeout_startup);
break;
case 'v':
debug = 1;
cl_log(LOG_INFO, "Verbose mode enabled.");
break;
case 'T':
watchdog_set_timeout = 0;
cl_log(LOG_INFO, "Setting watchdog timeout disabled; using defaults.");
break;
case 'W':
w++;
break;
case 'w':
watchdogdev = strdup(optarg);
break;
case 'd':
recruit_servant(optarg, 0);
break;
case 'P':
check_pcmk = 1;
break;
case 'n':
local_uname = strdup(optarg);
cl_log(LOG_INFO, "Overriding local hostname to %s", local_uname);
break;
case 'p':
pidfile = strdup(optarg);
cl_log(LOG_INFO, "pidfile set to %s", pidfile);
break;
case 'C':
timeout_watchdog_crashdump = atoi(optarg);
cl_log(LOG_INFO, "Setting crashdump watchdog timeout to %d",
(int)timeout_watchdog_crashdump);
break;
case '1':
timeout_watchdog = atoi(optarg);
break;
case '2':
timeout_allocate = atoi(optarg);
break;
case '3':
timeout_loop = atoi(optarg);
break;
case '4':
timeout_msgwait = atoi(optarg);
break;
case '5':
timeout_watchdog_warn = atoi(optarg);
cl_log(LOG_INFO, "Setting latency warning to %d",
(int)timeout_watchdog_warn);
break;
case 't':
servant_restart_interval = atoi(optarg);
cl_log(LOG_INFO, "Setting servant restart interval to %d",
(int)servant_restart_interval);
break;
case 'I':
timeout_io = atoi(optarg);
cl_log(LOG_INFO, "Setting IO timeout to %d",
(int)timeout_io);
break;
case 'F':
servant_restart_count = atoi(optarg);
cl_log(LOG_INFO, "Servant restart count set to %d",
(int)servant_restart_count);
break;
case 'h':
usage();
return (0);
default:
exit_status = -2;
goto out;
break;
}
}
if (w > 0) {
watchdog_use = w % 2;
}
if (watchdog_use) {
cl_log(LOG_INFO, "Watchdog enabled.");
} else {
cl_log(LOG_INFO, "Watchdog disabled.");
}
if (servant_count < 1 || servant_count > 3) {
fprintf(stderr, "You must specify 1 to 3 devices via the -d option.\n");
exit_status = -1;
goto out;
}
/* There must at least be one command following the options: */
if ((argc - optind) < 1) {
fprintf(stderr, "Not enough arguments.\n");
exit_status = -2;
goto out;
}
if (init_set_proc_title(argc, argv, envp) < 0) {
fprintf(stderr, "Allocation of proc title failed.\n");
exit_status = -1;
goto out;
}
maximize_priority();
if (strcmp(argv[optind], "create") == 0) {
- exit_status = init_devices();
+ exit_status = init_devices(servants_leader);
} else if (strcmp(argv[optind], "dump") == 0) {
- exit_status = dump_headers();
+ exit_status = dump_headers(servants_leader);
} else if (strcmp(argv[optind], "allocate") == 0) {
- exit_status = allocate_slots(argv[optind + 1]);
+ exit_status = allocate_slots(argv[optind + 1], servants_leader);
} else if (strcmp(argv[optind], "list") == 0) {
- exit_status = list_slots();
+ exit_status = list_slots(servants_leader);
} else if (strcmp(argv[optind], "message") == 0) {
- exit_status = messenger(argv[optind + 1], argv[optind + 2]);
+ 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]);
+ exit_status = ping_via_slots(argv[optind + 1], servants_leader);
} else if (strcmp(argv[optind], "watch") == 0) {
- open_any_device();
+ open_any_device(servants_leader);
/* We only want this to have an effect during watch right now;
* pinging and fencing would be too confused */
if (check_pcmk) {
recruit_servant("pcmk", 0);
servant_count--;
}
exit_status = inquisitor();
} else {
exit_status = -2;
}
out:
if (exit_status < 0) {
if (exit_status == -2) {
usage();
} else {
fprintf(stderr, "sbd failed; please check the logs.\n");
}
return (1);
}
return (0);
}
diff --git a/src/sbd-md.c b/src/sbd-md.c
index e536276..bb961a0 100644
--- a/src/sbd-md.c
+++ b/src/sbd-md.c
@@ -1,555 +1,552 @@
/*
* Copyright (C) 2013 Lars Marowsky-Bree <lmb@suse.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include "sbd.h"
-extern struct servants_list_item *servants_leader;
extern int servant_count;
-extern int start_mode;
+static int servant_inform_parent = 0;
-static int servant_inform_parent = 0;
-
-int init_devices()
+int init_devices(struct servants_list_item *servants)
{
int rc = 0;
struct sbd_context *st;
struct servants_list_item *s;
- for (s = servants_leader; s; s = s->next) {
+ for (s = servants; s; s = s->next) {
fprintf(stdout, "Initializing device %s\n",
s->devname);
st = open_device(s->devname, LOG_ERR);
if (!st) {
return -1;
}
rc = init_device(st);
close_device(st);
if (rc == -1) {
fprintf(stderr, "Failed to init device %s\n", s->devname);
return rc;
}
fprintf(stdout, "Device %s is initialized.\n", s->devname);
}
return 0;
}
-int slot_msg_wrapper(const char* devname, const void* argp)
+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;
}
-int slot_ping_wrapper(const char* devname, const void* argp)
+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)
+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_leader; s; s = s->next) {
+ 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()
+int list_slots(struct servants_list_item *servants)
{
int rc = 0;
struct servants_list_item *s;
struct sbd_context *st;
- for (s = servants_leader; s; s = s->next) {
+ for (s = servants; s; s = s->next) {
st = open_device(s->devname, LOG_WARNING);
if (!st) {
fprintf(stdout, "== disk %s unreadable!\n", s->devname);
continue;
}
rc = slot_list(st);
close_device(st);
if (rc == -1) {
fprintf(stdout, "== Slots on disk %s NOT dumped\n", s->devname);
}
}
return 0;
}
-int ping_via_slots(const char *name)
+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_leader; s; s = s->next) {
- s->pid = assign_servant(s->devname, &slot_ping_wrapper, (const void*)name);
+ for (s = servants; s; s = s->next) {
+ s->pid = assign_servant(s->devname, &slot_ping_wrapper, 0, (const void*)name);
}
while (servants_finished < servant_count) {
sig = sigwaitinfo(&procmask, &sinfo);
if (sig == SIGCHLD) {
while ((pid = wait(&status))) {
if (pid == -1 && errno == ECHILD) {
break;
} else {
s = lookup_servant_by_pid(pid);
if (s) {
servants_finished++;
}
}
}
}
}
return 0;
}
int quorum_write(int good_servants)
{
return (good_servants > servant_count/2);
}
int quorum_read(int good_servants)
{
if (servant_count >= 3)
return (good_servants > servant_count/2);
else
return (good_servants >= 1);
}
-int messenger(const char *name, const char *msg)
+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_leader; s; s = s->next) {
- s->pid = assign_servant(s->devname, &slot_msg_wrapper, &slot_msg_arg);
+ for (s = servants; s; s = s->next) {
+ s->pid = assign_servant(s->devname, &slot_msg_wrapper, 0, &slot_msg_arg);
}
while (!(quorum_write(successful_delivery) ||
(servants_finished == servant_count))) {
sig = sigwaitinfo(&procmask, &sinfo);
if (sig == SIGCHLD) {
while ((pid = waitpid(-1, &status, WNOHANG))) {
if (pid == -1 && errno == ECHILD) {
break;
} else {
servants_finished++;
if (WIFEXITED(status)
&& WEXITSTATUS(status) == 0) {
DBGLOG(LOG_INFO, "Process %d succeeded.",
(int)pid);
successful_delivery++;
} else {
cl_log(LOG_WARNING, "Process %d failed to deliver!",
(int)pid);
}
}
}
}
}
if (quorum_write(successful_delivery)) {
cl_log(LOG_INFO, "Message successfully delivered.");
return 0;
} else {
cl_log(LOG_ERR, "Message is not delivered via more then a half of devices");
return -1;
}
}
-int dump_headers(void)
+int dump_headers(struct servants_list_item *servants)
{
int rc = 0;
- struct servants_list_item *s = servants_leader;
+ struct servants_list_item *s = servants;
struct sbd_context *st;
- for (s = servants_leader; s; s = s->next) {
+ for (s = servants; s; s = s->next) {
fprintf(stdout, "==Dumping header on disk %s\n", s->devname);
st = open_device(s->devname, LOG_WARNING);
if (!st) {
fprintf(stdout, "== disk %s unreadable!\n", s->devname);
continue;
}
rc = header_dump(st);
close_device(st);
if (rc == -1) {
fprintf(stdout, "==Header on disk %s NOT dumped\n", s->devname);
} else {
fprintf(stdout, "==Header on disk %s is dumped\n", s->devname);
}
}
return rc;
}
-void open_any_device(void)
+void open_any_device(struct servants_list_item *servants)
{
struct sector_header_s *hdr_cur = NULL;
struct timespec t_0;
int t_wait = 0;
clock_gettime(CLOCK_MONOTONIC, &t_0);
while (!hdr_cur && t_wait < timeout_startup) {
struct timespec t_now;
struct servants_list_item* s;
- for (s = servants_leader; s; s = s->next) {
+ for (s = servants; s; s = s->next) {
struct sbd_context *st = open_device(s->devname, LOG_DEBUG);
if (!st)
continue;
hdr_cur = header_get(st);
close_device(st);
if (hdr_cur)
break;
}
clock_gettime(CLOCK_MONOTONIC, &t_now);
t_wait = t_now.tv_sec - t_0.tv_sec;
if (!hdr_cur) {
sleep(timeout_loop);
}
}
if (hdr_cur) {
timeout_watchdog = hdr_cur->timeout_watchdog;
timeout_allocate = hdr_cur->timeout_allocate;
timeout_loop = hdr_cur->timeout_loop;
timeout_msgwait = hdr_cur->timeout_msgwait;
} else {
cl_log(LOG_ERR, "No devices were available at start-up within %i seconds.",
timeout_startup);
exit(1);
}
free(hdr_cur);
return;
}
/*
::-::-::-::-::-::-::-::-::-::-::-::-::
Begin disk based servant code
::-::-::-::-::-::-::-::-::-::-::-::-::
*/
static int servant_check_timeout_inconsistent(struct sector_header_s *hdr)
{
if (timeout_watchdog != hdr->timeout_watchdog) {
cl_log(LOG_WARNING, "watchdog timeout: %d versus %d on this device",
(int)timeout_watchdog, (int)hdr->timeout_watchdog);
return -1;
}
if (timeout_allocate != hdr->timeout_allocate) {
cl_log(LOG_WARNING, "allocate timeout: %d versus %d on this device",
(int)timeout_allocate, (int)hdr->timeout_allocate);
return -1;
}
if (timeout_loop != hdr->timeout_loop) {
cl_log(LOG_WARNING, "loop timeout: %d versus %d on this device",
(int)timeout_loop, (int)hdr->timeout_loop);
return -1;
}
if (timeout_msgwait != hdr->timeout_msgwait) {
cl_log(LOG_WARNING, "msgwait timeout: %d versus %d on this device",
(int)timeout_msgwait, (int)hdr->timeout_msgwait);
return -1;
}
return 0;
}
/* This is a bit hackish, but the easiest way to rewire all process
* exits to send the desired signal to the parent. */
void servant_exit(void)
{
pid_t ppid;
union sigval signal_value;
ppid = getppid();
if (servant_inform_parent) {
memset(&signal_value, 0, sizeof(signal_value));
sigqueue(ppid, SIG_IO_FAIL, signal_value);
}
}
-int servant(const char *diskname, const void* argp)
+int servant(const char *diskname, int mode, const void* argp)
{
struct sector_mbox_s *s_mbox = NULL;
struct sector_node_s *s_node = NULL;
struct sector_header_s *s_header = NULL;
int mbox;
int rc = 0;
time_t t0, t1, latency;
union sigval signal_value;
sigset_t servant_masks;
struct sbd_context *st;
pid_t ppid;
char uuid[37];
const struct servants_list_item *s = argp;
if (!diskname) {
cl_log(LOG_ERR, "Empty disk name %s.", diskname);
return -1;
}
cl_log(LOG_INFO, "Servant starting for device %s", diskname);
/* Block most of the signals */
sigfillset(&servant_masks);
sigdelset(&servant_masks, SIGKILL);
sigdelset(&servant_masks, SIGFPE);
sigdelset(&servant_masks, SIGILL);
sigdelset(&servant_masks, SIGSEGV);
sigdelset(&servant_masks, SIGBUS);
sigdelset(&servant_masks, SIGALRM);
/* FIXME: check error */
sigprocmask(SIG_SETMASK, &servant_masks, NULL);
atexit(servant_exit);
servant_inform_parent = 1;
st = open_device(diskname, LOG_WARNING);
if (!st) {
return -1;
}
s_header = header_get(st);
if (!s_header) {
cl_log(LOG_ERR, "Not a valid header on %s", diskname);
return -1;
}
if (servant_check_timeout_inconsistent(s_header) < 0) {
cl_log(LOG_ERR, "Timeouts on %s do not match first device",
diskname);
return -1;
}
if (s_header->minor_version > 0) {
uuid_unparse_lower(s_header->uuid, uuid);
cl_log(LOG_INFO, "Device %s uuid: %s", diskname, uuid);
}
mbox = slot_allocate(st, local_uname);
if (mbox < 0) {
cl_log(LOG_ERR,
"No slot allocated, and automatic allocation failed for disk %s.",
diskname);
rc = -1;
goto out;
}
s_node = sector_alloc();
if (slot_read(st, mbox, s_node) < 0) {
cl_log(LOG_ERR, "Unable to read node entry on %s",
diskname);
exit(1);
}
DBGLOG(LOG_INFO, "Monitoring slot %d on disk %s", mbox, diskname);
if (s_header->minor_version == 0) {
set_proc_title("sbd: watcher: %s - slot: %d", diskname, mbox);
} else {
set_proc_title("sbd: watcher: %s - slot: %d - uuid: %s",
diskname, mbox, uuid);
}
s_mbox = sector_alloc();
if (s->first_start) {
- if (start_mode > 0) {
+ if (mode > 0) {
if (mbox_read(st, mbox, s_mbox) < 0) {
cl_log(LOG_ERR, "mbox read failed during start-up in servant.");
rc = -1;
goto out;
}
if (s_mbox->cmd != SBD_MSG_EXIT &&
s_mbox->cmd != SBD_MSG_EMPTY) {
/* Not a clean stop. Abort start-up */
cl_log(LOG_WARNING, "Found fencing message - aborting start-up. Manual intervention required!");
ppid = getppid();
sigqueue(ppid, SIG_EXITREQ, signal_value);
rc = 0;
goto out;
}
}
DBGLOG(LOG_INFO, "First servant start - zeroing inbox");
memset(s_mbox, 0, sizeof(*s_mbox));
if (mbox_write(st, mbox, s_mbox) < 0) {
rc = -1;
goto out;
}
}
memset(&signal_value, 0, sizeof(signal_value));
while (1) {
struct sector_header_s *s_header_retry = NULL;
struct sector_node_s *s_node_retry = NULL;
t0 = time(NULL);
sleep(timeout_loop);
ppid = getppid();
if (ppid == 1) {
/* Our parent died unexpectedly. Triggering
* self-fence. */
do_reset();
}
/* These attempts are, by definition, somewhat racy. If
* the device is wiped out or corrupted between here and
* us reading our mbox, there is nothing we can do about
* that. But at least we tried. */
s_header_retry = header_get(st);
if (!s_header_retry) {
cl_log(LOG_ERR, "No longer found a valid header on %s", diskname);
exit(1);
}
if (memcmp(s_header, s_header_retry, sizeof(*s_header)) != 0) {
cl_log(LOG_ERR, "Header on %s changed since start-up!", diskname);
exit(1);
}
free(s_header_retry);
s_node_retry = sector_alloc();
if (slot_read(st, mbox, s_node_retry) < 0) {
cl_log(LOG_ERR, "slot read failed in servant.");
exit(1);
}
if (memcmp(s_node, s_node_retry, sizeof(*s_node)) != 0) {
cl_log(LOG_ERR, "Node entry on %s changed since start-up!", diskname);
exit(1);
}
free(s_node_retry);
if (mbox_read(st, mbox, s_mbox) < 0) {
cl_log(LOG_ERR, "mbox read failed in servant.");
exit(1);
}
if (s_mbox->cmd > 0) {
cl_log(LOG_INFO,
"Received command %s from %s on disk %s",
char2cmd(s_mbox->cmd), s_mbox->from, diskname);
switch (s_mbox->cmd) {
case SBD_MSG_TEST:
memset(s_mbox, 0, sizeof(*s_mbox));
mbox_write(st, mbox, s_mbox);
sigqueue(ppid, SIG_TEST, signal_value);
break;
case SBD_MSG_RESET:
do_reset();
break;
case SBD_MSG_OFF:
do_off();
break;
case SBD_MSG_EXIT:
sigqueue(ppid, SIG_EXITREQ, signal_value);
break;
case SBD_MSG_CRASHDUMP:
do_crashdump();
break;
default:
/* FIXME:
An "unknown" message might result
from a partial write.
log it and clear the slot.
*/
cl_log(LOG_ERR, "Unknown message on disk %s",
diskname);
memset(s_mbox, 0, sizeof(*s_mbox));
mbox_write(st, mbox, s_mbox);
break;
}
}
sigqueue(ppid, SIG_LIVENESS, signal_value);
t1 = time(NULL);
latency = t1 - t0;
if (timeout_watchdog_warn && (latency > timeout_watchdog_warn)) {
cl_log(LOG_WARNING,
"Latency: %d exceeded threshold %d on disk %s",
(int)latency, (int)timeout_watchdog_warn,
diskname);
} else if (debug) {
DBGLOG(LOG_INFO, "Latency: %d on disk %s", (int)latency,
diskname);
}
}
out:
free(s_mbox);
close_device(st);
if (rc == 0) {
servant_inform_parent = 0;
}
return rc;
}
diff --git a/src/sbd-pacemaker.c b/src/sbd-pacemaker.c
index 77dac9f..819a394 100644
--- a/src/sbd-pacemaker.c
+++ b/src/sbd-pacemaker.c
@@ -1,515 +1,515 @@
/*
* Copyright (C) 2013 Lars Marowsky-Bree <lmb@suse.com>
*
* Based on crm_mon.c, which was:
* Copyright (C) 2004 Andrew Beekhof <andrew@beekhof.net>
*
* 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 library; if not, write to the Free Software
* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
*/
/* TODO list:
*
* - Trying to shutdown a node if no devices are up will fail, since SBD
* currently uses a message via the disk to achieve this.
*
* - Shutting down cluster nodes while the majority of devices is down
* will eventually take the cluster below the quorum threshold, at which
* time the remaining cluster nodes will all immediately suicide.
*
*/
#include "sbd.h"
#include <sys/param.h>
#include <crm/crm.h>
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <libgen.h>
#include <sys/utsname.h>
#include <config.h>
#include <crm_config.h>
#include <crm/msg_xml.h>
#include <crm/common/util.h>
#include <crm/common/xml.h>
#include <crm/common/ipc.h>
#include <crm/common/mainloop.h>
#ifdef CHECK_AIS
# include <crm/cluster.h>
#endif
#include <crm/cib.h>
#include <crm/pengine/status.h>
static void clean_up(int rc);
static void crm_diff_update(const char *event, xmlNode * msg);
static void mon_refresh_state(void);
static int cib_connect(gboolean full);
static void set_pcmk_health(int healthy);
static void notify_parent(void);
static GMainLoop *mainloop = NULL;
static guint timer_id_reconnect = 0;
static guint timer_id_notify = 0;
static int reconnect_msec = 1000;
static int pcmk_healthy = 0;
static int cib_connected = 0;
#ifdef CHECK_AIS
static guint timer_id_ais = 0;
static enum cluster_type_e cluster_stack = pcmk_cluster_unknown;
static struct timespec t_last_quorum;
static int check_ais = 0;
#endif
#define LOGONCE(state, lvl, fmt, args...) do { \
if (last_state != state) { \
cl_log(lvl, fmt, ##args); \
last_state = state; \
} \
} while(0)
static cib_t *cib = NULL;
static xmlNode *current_cib = NULL;
static long last_refresh = 0;
static gboolean
mon_timer_reconnect(gpointer data)
{
int rc = 0;
if (timer_id_reconnect > 0) {
g_source_remove(timer_id_reconnect);
}
rc = cib_connect(TRUE);
if (rc != 0) {
cl_log(LOG_WARNING, "CIB reconnect failed: %d", rc);
timer_id_reconnect = g_timeout_add(reconnect_msec, mon_timer_reconnect, NULL);
} else {
cl_log(LOG_INFO, "CIB reconnect successful");
}
return FALSE;
}
static void
mon_cib_connection_destroy(gpointer user_data)
{
if (cib) {
cl_log(LOG_WARNING, "Disconnected from CIB");
cib->cmds->signoff(cib);
set_pcmk_health(2);
timer_id_reconnect = g_timeout_add(reconnect_msec, mon_timer_reconnect, NULL);
}
cib_connected = 0;
return;
}
static gboolean
mon_timer_notify(gpointer data)
{
static int counter = 0;
int counter_max = timeout_watchdog / timeout_loop;
if (timer_id_notify > 0) {
g_source_remove(timer_id_notify);
}
if (cib_connected) {
if (counter == counter_max) {
free_xml(current_cib);
current_cib = get_cib_copy(cib);
mon_refresh_state();
counter = 0;
} else {
cib->cmds->noop(cib, 0);
notify_parent();
counter++;
}
}
timer_id_notify = g_timeout_add(timeout_loop * 1000, mon_timer_notify, NULL);
return FALSE;
}
/*
* Mainloop signal handler.
*/
static void
mon_shutdown(int nsig)
{
clean_up(0);
}
static int
cib_connect(gboolean full)
{
int rc = 0;
CRM_CHECK(cib != NULL, return -EINVAL);
cib_connected = 0;
crm_xml_init();
if (cib->state != cib_connected_query && cib->state != cib_connected_command) {
rc = cib->cmds->signon(cib, crm_system_name, cib_query);
if (rc != 0) {
return rc;
}
current_cib = get_cib_copy(cib);
mon_refresh_state();
if (full) {
if (rc == 0) {
rc = cib->cmds->set_connection_dnotify(cib, mon_cib_connection_destroy);
if (rc == -EPROTONOSUPPORT) {
/* Notification setup failed, won't be able to reconnect after failure */
rc = 0;
}
}
if (rc == 0) {
cib->cmds->del_notify_callback(cib, T_CIB_DIFF_NOTIFY, crm_diff_update);
rc = cib->cmds->add_notify_callback(cib, T_CIB_DIFF_NOTIFY, crm_diff_update);
}
if (rc != 0) {
/* Notification setup failed, could not monitor CIB actions */
clean_up(-rc);
}
}
}
if (!rc) {
cib_connected = 1;
}
return rc;
}
#ifdef CHECK_AIS
static gboolean
mon_timer_ais(gpointer data)
{
if (timer_id_ais > 0) {
g_source_remove(timer_id_ais);
}
send_cluster_text(crm_class_quorum, NULL, TRUE, NULL, crm_msg_ais);
/* The timer is set in the response processing */
return FALSE;
}
static void
ais_membership_destroy(gpointer user_data)
{
cl_log(LOG_ERR, "AIS connection terminated - corosync down?");
#if SUPPORT_PLUGIN
ais_fd_sync = -1;
#endif
/* TODO: Is recovery even worth it here? After all, this means
* that corosync died ... */
exit(1);
}
static void
ais_membership_dispatch(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
uint32_t kind = 0;
const char *from = NULL;
char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from);
if (!data) {
return;
}
free(data);
data = NULL;
if (kind != crm_class_quorum) {
return;
}
DBGLOG(LOG_INFO, "AIS quorum state: %d", (int)crm_have_quorum);
clock_gettime(CLOCK_MONOTONIC, &t_last_quorum);
timer_id_ais = g_timeout_add(timeout_loop * 1000, mon_timer_ais, NULL);
return;
}
#endif
static void
compute_status(pe_working_set_t * data_set)
{
static int updates = 0;
static int last_state = 0;
int healthy = 0;
node_t *dc = NULL;
struct timespec t_now;
updates++;
dc = data_set->dc_node;
clock_gettime(CLOCK_MONOTONIC, &t_now);
if (dc == NULL) {
LOGONCE(1, LOG_INFO, "We don't have a DC right now.");
healthy = 2;
goto out;
} else {
const char *cib_quorum = crm_element_value(data_set->input, XML_ATTR_HAVE_QUORUM);
if (crm_is_true(cib_quorum)) {
DBGLOG(LOG_INFO, "CIB: We have quorum!");
} else {
LOGONCE(3, LOG_WARNING, "CIB: We do NOT have quorum!");
goto out;
}
}
#ifdef CHECK_AIS
if (check_ais) {
int quorum_age = t_now.tv_sec - t_last_quorum.tv_sec;
if (quorum_age > (int)(timeout_io+timeout_loop)) {
if (t_last_quorum.tv_sec != 0)
LOGONCE(2, LOG_WARNING, "AIS: Quorum outdated!");
goto out;
}
if (crm_have_quorum) {
DBGLOG(LOG_INFO, "AIS: We have quorum!");
} else {
LOGONCE(8, LOG_WARNING, "AIS: We do NOT have quorum!");
goto out;
}
}
#endif
node_t *node = pe_find_node(data_set->nodes, local_uname);
if (node->details->unclean) {
LOGONCE(4, LOG_WARNING, "Node state: UNCLEAN");
} else if (node->details->pending) {
LOGONCE(5, LOG_WARNING, "Node state: pending");
healthy = 2;
} else if (node->details->online) {
LOGONCE(6, LOG_INFO, "Node state: online");
healthy = 1;
} else {
LOGONCE(7, LOG_WARNING, "Node state: UNKNOWN");
}
out:
set_pcmk_health(healthy);
return;
}
static void
set_pcmk_health(int healthy)
{
pcmk_healthy = healthy;
notify_parent();
}
static 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_reset();
}
switch (pcmk_healthy) {
case 2:
DBGLOG(LOG_INFO, "Not notifying parent: state transient");
break;
case 1:
DBGLOG(LOG_INFO, "Notifying parent: healthy");
sigqueue(ppid, SIG_LIVENESS, signal_value);
break;
case 0:
default:
DBGLOG(LOG_WARNING, "Notifying parent: UNHEALTHY");
sigqueue(ppid, SIG_PCMK_UNHEALTHY, signal_value);
break;
}
}
static void
crm_diff_update(const char *event, xmlNode * msg)
{
int rc = -1;
const char *op = NULL;
if (current_cib != NULL) {
xmlNode *cib_last = current_cib;
current_cib = NULL;
rc = cib_apply_patch_event(msg, cib_last, ¤t_cib, LOG_DEBUG);
free_xml(cib_last);
switch(rc) {
case -pcmk_err_diff_resync:
case -pcmk_err_diff_failed:
crm_warn("[%s] %s Patch aborted: %s (%d)", event, op, pcmk_strerror(rc), rc);
case pcmk_ok:
break;
default:
crm_notice("[%s] %s ABORTED: %s (%d)", event, op, pcmk_strerror(rc), rc);
break;
}
}
if (current_cib == NULL) {
current_cib = get_cib_copy(cib);
}
mon_refresh_state();
}
static void
mon_refresh_state(void)
{
xmlNode *cib_copy = copy_xml(current_cib);
pe_working_set_t data_set;
last_refresh = time(NULL);
if (cli_config_update(&cib_copy, NULL, FALSE) == FALSE) {
cl_log(LOG_WARNING, "cli_config_update() failed - forcing reconnect to CIB");
if (cib) {
cib->cmds->signoff(cib);
}
} else {
set_working_set_defaults(&data_set);
data_set.input = cib_copy;
cluster_status(&data_set);
compute_status(&data_set);
cleanup_calculations(&data_set);
}
return;
}
static void
clean_up(int rc)
{
if (cib != NULL) {
cib->cmds->signoff(cib);
cib_delete(cib);
cib = NULL;
}
if (rc >= 0) {
exit(rc);
}
return;
}
int
-servant_pcmk(const char *diskname, const void* argp)
+servant_pcmk(const char *diskname, int mode, const void* argp)
{
int exit_code = 0;
crm_cluster_t crm_cluster;
cl_log(LOG_INFO, "Monitoring Pacemaker health");
set_proc_title("sbd: watcher: Pacemaker");
/* We don't want any noisy crm messages */
set_crm_log_level(LOG_CRIT);
#ifdef CHECK_AIS
cluster_stack = get_cluster_type();
if (cluster_stack != pcmk_cluster_classic_ais) {
check_ais = 0;
} else {
check_ais = 1;
cl_log(LOG_INFO, "Legacy plug-in detected, AIS quorum check enabled");
if(is_openais_cluster()) {
crm_cluster.destroy = ais_membership_destroy;
crm_cluster.cpg.cpg_deliver_fn = ais_membership_dispatch;
/* crm_cluster.cpg.cpg_confchg_fn = pcmk_cpg_membership; TODO? */
crm_cluster.cpg.cpg_confchg_fn = NULL;
}
while (!crm_cluster_connect(&crm_cluster)) {
cl_log(LOG_INFO, "Waiting to sign in with cluster ...");
sleep(reconnect_msec / 1000);
}
}
#endif
if (current_cib == NULL) {
cib = cib_new();
do {
exit_code = cib_connect(TRUE);
if (exit_code != 0) {
sleep(reconnect_msec / 1000);
}
} while (exit_code == -ENOTCONN);
if (exit_code != 0) {
clean_up(-exit_code);
}
}
mainloop = g_main_new(FALSE);
mainloop_add_signal(SIGTERM, mon_shutdown);
mainloop_add_signal(SIGINT, mon_shutdown);
timer_id_notify = g_timeout_add(timeout_loop * 1000, mon_timer_notify, NULL);
#ifdef CHECK_AIS
if (check_ais) {
timer_id_ais = g_timeout_add(timeout_loop * 1000, mon_timer_ais, NULL);
}
#endif
g_main_run(mainloop);
g_main_destroy(mainloop);
clean_up(0);
return 0; /* never reached */
}
diff --git a/src/sbd.h b/src/sbd.h
index a402e64..7f19f55 100644
--- a/src/sbd.h
+++ b/src/sbd.h
@@ -1,225 +1,224 @@
/*
* Copyright (C) 2013 Lars Marowsky-Bree <lmb@suse.com>
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This software is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* General Public License for more details.
*
* You should have received a copy of the GNU General Public
* License along with this library; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include <arpa/inet.h>
#include <asm/unistd.h>
#include <ctype.h>
#include <errno.h>
#include <fcntl.h>
#include <libaio.h>
#include <linux/fs.h>
#include <linux/types.h>
#include <linux/watchdog.h>
#include <malloc.h>
#include <signal.h>
#include <stdio.h>
#include <stdlib.h>
#include <stdbool.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/ptrace.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/utsname.h>
#include <sys/wait.h>
#include <syslog.h>
#include <time.h>
#include <unistd.h>
#include <uuid/uuid.h>
#include <qb/qblog.h>
/* signals reserved for multi-disk sbd */
#define SIG_LIVENESS (SIGRTMIN + 1) /* report liveness of the disk */
#define SIG_EXITREQ (SIGRTMIN + 2) /* exit request to inquisitor */
#define SIG_TEST (SIGRTMIN + 3) /* trigger self test */
#define SIG_RESTART (SIGRTMIN + 4) /* trigger restart of all failed disk */
#define SIG_IO_FAIL (SIGRTMIN + 5) /* the IO child requests to be considered failed */
#define SIG_PCMK_UNHEALTHY (SIGRTMIN + 6)
/* FIXME: should add dynamic check of SIG_XX >= SIGRTMAX */
#define HOG_CHAR 0xff
#define HA_COREDIR "/var/lib/heartbeat/cores"
/* Sector data types */
struct sector_header_s {
char magic[8];
unsigned char version;
unsigned char slots;
/* Caveat: stored in network byte-order */
uint32_t sector_size;
uint32_t timeout_watchdog;
uint32_t timeout_allocate;
uint32_t timeout_loop;
uint32_t timeout_msgwait;
/* Minor version for extensions to the core data set:
* compatible and optional values. */
unsigned char minor_version;
uuid_t uuid; /* 16 bytes */
};
struct sector_mbox_s {
signed char cmd;
char from[64];
};
struct sector_node_s {
/* slots will be created with in_use == 0 */
char in_use;
char name[64];
};
struct servants_list_item {
const char* devname;
pid_t pid;
int restarts;
int restart_blocked;
int outdated;
int first_start;
struct timespec t_last, t_started;
struct servants_list_item *next;
};
struct sbd_context {
int devfd;
io_context_t ioctx;
struct iocb io;
};
#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)
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);
struct sbd_context *open_device(const char* devname, int loglevel);
-void open_any_device(void);
+void open_any_device(struct servants_list_item *servants);
void close_device(struct sbd_context *st);
signed char cmd2char(const char *cmd);
void * sector_alloc(void);
const char* char2cmd(const char cmd);
int sector_write(struct sbd_context *st, int sector, void *data);
int sector_read(struct sbd_context *st, int sector, void *data);
int slot_read(struct sbd_context *st, int slot, struct sector_node_s *s_node);
int slot_write(struct sbd_context *st, int slot, struct sector_node_s *s_node);
int mbox_write(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox);
int mbox_read(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox);
int mbox_write_verify(struct sbd_context *st, int mbox, struct sector_mbox_s *s_mbox);
/* After a call to header_write(), certain data fields will have been
* converted to on-disk byte-order; the header should not be accessed
* afterwards anymore! */
int header_write(struct sbd_context *st, struct sector_header_s *s_header);
int header_read(struct sbd_context *st, struct sector_header_s *s_header);
int valid_header(const struct sector_header_s *s_header);
struct sector_header_s * header_get(struct sbd_context *st);
int init_device(struct sbd_context *st);
int slot_lookup(struct sbd_context *st, const struct sector_header_s *s_header, const char *name);
int slot_unused(struct sbd_context *st, const struct sector_header_s *s_header);
int slot_allocate(struct sbd_context *st, const char *name);
int slot_list(struct sbd_context *st);
int slot_ping(struct sbd_context *st, const char *name);
int slot_msg(struct sbd_context *st, const char *name, const char *cmd);
int header_dump(struct sbd_context *st);
void sysrq_trigger(char t);
void do_crashdump(void);
void do_reset(void);
void do_off(void);
pid_t make_daemon(void);
void maximize_priority(void);
void sbd_get_uname(void);
/* Tunable defaults: */
extern unsigned long timeout_watchdog;
extern unsigned long timeout_watchdog_warn;
extern unsigned long timeout_watchdog_crashdump;
extern int timeout_allocate;
extern int timeout_loop;
extern int timeout_msgwait;
extern int timeout_io;
extern int timeout_startup;
extern int watchdog_use;
extern int watchdog_set_timeout;
extern int skip_rt;
extern int debug;
extern int debug_mode;
extern const char *watchdogdev;
extern char* local_uname;
/* Global, non-tunable variables: */
extern int sector_size;
extern int watchdogfd;
extern const char* cmdname;
-typedef int (*functionp_t)(const char* devname, const void* argp);
+typedef int (*functionp_t)(const char* devname, int mode, const void* argp);
-int assign_servant(const char* devname, functionp_t functionp, const void* argp);
-int init_devices(void);
+int assign_servant(const char* devname, functionp_t functionp, int mode, const void* argp);
+int init_devices(struct servants_list_item *servants);
struct slot_msg_arg_t {
const char* name;
const char* msg;
};
-int slot_msg_wrapper(const char* devname, const void* argp);
-int slot_ping_wrapper(const char* devname, const void* argp);
-int allocate_slots(const char *name);
-int list_slots(void);
-int ping_via_slots(const char *name);
-int dump_headers(void);
+
+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);
int check_all_dead(void);
void servant_exit(void);
-int servant(const char *diskname, const void* argp);
+int servant(const char *diskname, int mode, const void* argp);
void recruit_servant(const char *devname, pid_t pid);
struct servants_list_item *lookup_servant_by_dev(const char *devname);
struct servants_list_item *lookup_servant_by_pid(pid_t pid);
void servants_kill(void);
void servants_start(void);
void servant_start(struct servants_list_item *s);
void inquisitor_child(void);
int inquisitor(void);
int inquisitor_decouple(void);
-int messenger(const char *name, const char *msg);
+int messenger(const char *name, const char *msg, struct servants_list_item *servants);
void cleanup_servant_by_pid(pid_t pid);
int quorum_write(int good_servants);
int quorum_read(int good_servants);
int pcmk_have_quorum(void);
-int servant_pcmk(const char *diskname, const void* argp);
+int servant_pcmk(const char *diskname, int mode, const void* argp);
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)