diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h
index a91a4bec50..a354f0cb32 100644
--- a/include/crm/common/internal.h
+++ b/include/crm/common/internal.h
@@ -1,359 +1,360 @@
/*
* Copyright 2015-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_INTERNAL__H
#define PCMK__CRM_COMMON_INTERNAL__H
#include <unistd.h> // pid_t, getpid()
#include <stdbool.h> // bool
#include <stdint.h> // uint8_t, uint64_t
#include <glib.h> // guint, GList, GHashTable
#include <libxml/tree.h> // xmlNode
#include <crm/common/logging.h> // do_crm_log_unlikely(), etc.
#include <crm/common/mainloop.h> // mainloop_io_t, struct ipc_client_callbacks
#include <crm/common/strings.h> // crm_strdup_printf()
#include <crm/common/actions_internal.h>
#include <crm/common/digest_internal.h>
#include <crm/common/health_internal.h>
#include <crm/common/io_internal.h>
#include <crm/common/iso8601_internal.h>
#include <crm/common/results_internal.h>
#include <crm/common/messages_internal.h>
#include <crm/common/nvpair_internal.h>
#include <crm/common/scores_internal.h>
#include <crm/common/strings_internal.h>
#include <crm/common/acl_internal.h>
#ifdef __cplusplus
extern "C" {
#endif
/* This says whether the current application is a Pacemaker daemon or not,
* and is used to change default logging settings such as whether to log to
* stderr, etc., as well as a few other details such as whether blackbox signal
* handling is enabled.
*
* It is set when logging is initialized, and does not need to be set directly.
*/
extern bool pcmk__is_daemon;
// Number of elements in a statically defined array
#define PCMK__NELEM(a) ((int) (sizeof(a)/sizeof(a[0])) )
#if PCMK__ENABLE_CIBSECRETS
/* internal CIB utilities (from cib_secrets.c) */
int pcmk__substitute_secrets(const char *rsc_id, GHashTable *params);
#endif
/* internal main loop utilities (from mainloop.c) */
int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata,
const struct ipc_client_callbacks *callbacks,
mainloop_io_t **source);
guint pcmk__mainloop_timer_get_period(const mainloop_timer_t *timer);
/* internal node-related XML utilities (from nodes.c) */
/*!
* \internal
* \brief Add local node name and ID to an XML node
*
* \param[in,out] request XML node to modify
* \param[in] node The local node's name
* \param[in] nodeid The local node's ID (can be 0)
*/
void pcmk__xe_add_node(xmlNode *xml, const char *node, int nodeid);
/* internal name/value utilities (from nvpair.c) */
int pcmk__scan_nvpair(const char *input, char **name, char **value);
char *pcmk__format_nvpair(const char *name, const char *value,
const char *units);
/* internal procfs utilities (from procfs.c) */
pid_t pcmk__procfs_pid_of(const char *name);
unsigned int pcmk__procfs_num_cores(void);
int pcmk__procfs_pid2path(pid_t pid, char path[], size_t path_size);
bool pcmk__procfs_has_pids(void);
+void pcmk__sysrq_trigger(char t);
/* internal functions related to process IDs (from pid.c) */
/*!
* \internal
* \brief Check whether process exists (by PID and optionally executable path)
*
* \param[in] pid PID of process to check
* \param[in] daemon If not NULL, path component to match with procfs entry
*
* \return Standard Pacemaker return code
* \note Particular return codes of interest include pcmk_rc_ok for alive,
* ESRCH for process is not alive (verified by kill and/or executable path
* match), EACCES for caller unable or not allowed to check. A result of
* "alive" is less reliable when \p daemon is not provided or procfs is
* not available, since there is no guarantee that the PID has not been
* recycled for another process.
* \note This function cannot be used to verify \e authenticity of the process.
*/
int pcmk__pid_active(pid_t pid, const char *daemon);
int pcmk__read_pidfile(const char *filename, pid_t *pid);
int pcmk__pidfile_matches(const char *filename, pid_t expected_pid,
const char *expected_name, pid_t *pid);
int pcmk__lock_pidfile(const char *filename, const char *name);
// bitwise arithmetic utilities
/*!
* \internal
* \brief Set specified flags in a flag group
*
* \param[in] function Function name of caller
* \param[in] line Line number of caller
* \param[in] log_level Log a message at this level
* \param[in] flag_type Label describing this flag group (for logging)
* \param[in] target Name of object whose flags these are (for logging)
* \param[in] flag_group Flag group being manipulated
* \param[in] flags Which flags in the group should be set
* \param[in] flags_str Readable equivalent of \p flags (for logging)
*
* \return Possibly modified flag group
*/
static inline uint64_t
pcmk__set_flags_as(const char *function, int line, uint8_t log_level,
const char *flag_type, const char *target,
uint64_t flag_group, uint64_t flags, const char *flags_str)
{
uint64_t result = flag_group | flags;
if (result != flag_group) {
do_crm_log_unlikely(log_level,
"%s flags %#.8llx (%s) for %s set by %s:%d",
((flag_type == NULL)? "Group of" : flag_type),
(unsigned long long) flags,
((flags_str == NULL)? "flags" : flags_str),
((target == NULL)? "target" : target),
function, line);
}
return result;
}
/*!
* \internal
* \brief Clear specified flags in a flag group
*
* \param[in] function Function name of caller
* \param[in] line Line number of caller
* \param[in] log_level Log a message at this level
* \param[in] flag_type Label describing this flag group (for logging)
* \param[in] target Name of object whose flags these are (for logging)
* \param[in] flag_group Flag group being manipulated
* \param[in] flags Which flags in the group should be cleared
* \param[in] flags_str Readable equivalent of \p flags (for logging)
*
* \return Possibly modified flag group
*/
static inline uint64_t
pcmk__clear_flags_as(const char *function, int line, uint8_t log_level,
const char *flag_type, const char *target,
uint64_t flag_group, uint64_t flags, const char *flags_str)
{
uint64_t result = flag_group & ~flags;
if (result != flag_group) {
do_crm_log_unlikely(log_level,
"%s flags %#.8llx (%s) for %s cleared by %s:%d",
((flag_type == NULL)? "Group of" : flag_type),
(unsigned long long) flags,
((flags_str == NULL)? "flags" : flags_str),
((target == NULL)? "target" : target),
function, line);
}
return result;
}
/*!
* \internal
* \brief Get readable string for whether specified flags are set
*
* \param[in] flag_group Group of flags to check
* \param[in] flags Which flags in \p flag_group should be checked
*
* \return "true" if all \p flags are set in \p flag_group, otherwise "false"
*/
static inline const char *
pcmk__flag_text(uint64_t flag_group, uint64_t flags)
{
return pcmk__btoa(pcmk_all_flags_set(flag_group, flags));
}
// miscellaneous utilities (from utils.c)
void pcmk__daemonize(const char *name, const char *pidfile);
void pcmk__panic(const char *reason);
pid_t pcmk__locate_sbd(void);
void pcmk__sleep_ms(unsigned int ms);
guint pcmk__create_timer(guint interval_ms, GSourceFunc fn, gpointer data);
guint pcmk__timeout_ms2s(guint timeout_ms);
extern int pcmk__score_red;
extern int pcmk__score_green;
extern int pcmk__score_yellow;
/*!
* \internal
* \brief Allocate new zero-initialized memory, asserting on failure
*
* \param[in] file File where \p function is located
* \param[in] function Calling function
* \param[in] line Line within \p file
* \param[in] nmemb Number of elements to allocate memory for
* \param[in] size Size of each element
*
* \return Newly allocated memory of of size <tt>nmemb * size</tt> (guaranteed
* not to be \c NULL)
*
* \note The caller is responsible for freeing the return value using \c free().
*/
static inline void *
pcmk__assert_alloc_as(const char *file, const char *function, uint32_t line,
size_t nmemb, size_t size)
{
void *ptr = calloc(nmemb, size);
if (ptr == NULL) {
crm_abort(file, function, line, "Out of memory", FALSE, TRUE);
crm_exit(CRM_EX_OSERR);
}
return ptr;
}
/*!
* \internal
* \brief Allocate new zero-initialized memory, asserting on failure
*
* \param[in] nmemb Number of elements to allocate memory for
* \param[in] size Size of each element
*
* \return Newly allocated memory of of size <tt>nmemb * size</tt> (guaranteed
* not to be \c NULL)
*
* \note The caller is responsible for freeing the return value using \c free().
*/
#define pcmk__assert_alloc(nmemb, size) \
pcmk__assert_alloc_as(__FILE__, __func__, __LINE__, nmemb, size)
/*!
* \internal
* \brief Resize a dynamically allocated memory block
*
* \param[in] ptr Memory block to resize (or NULL to allocate new memory)
* \param[in] size New size of memory block in bytes (must be > 0)
*
* \return Pointer to resized memory block
*
* \note This asserts on error, so the result is guaranteed to be non-NULL
* (which is the main advantage of this over directly using realloc()).
*/
static inline void *
pcmk__realloc(void *ptr, size_t size)
{
void *new_ptr;
// realloc(p, 0) can replace free(p) but this wrapper can't
pcmk__assert(size > 0);
new_ptr = realloc(ptr, size);
if (new_ptr == NULL) {
free(ptr);
abort();
}
return new_ptr;
}
static inline char *
pcmk__getpid_s(void)
{
return crm_strdup_printf("%lu", (unsigned long) getpid());
}
// More efficient than g_list_length(list) == 1
static inline bool
pcmk__list_of_1(GList *list)
{
return list && (list->next == NULL);
}
// More efficient than g_list_length(list) > 1
static inline bool
pcmk__list_of_multiple(GList *list)
{
return list && (list->next != NULL);
}
/* convenience functions for failure-related node attributes */
#define PCMK__FAIL_COUNT_PREFIX "fail-count"
#define PCMK__LAST_FAILURE_PREFIX "last-failure"
/*!
* \internal
* \brief Generate a failure-related node attribute name for a resource
*
* \param[in] prefix Start of attribute name
* \param[in] rsc_id Resource name
* \param[in] op Operation name
* \param[in] interval_ms Operation interval
*
* \return Newly allocated string with attribute name
*
* \note Failure attributes are named like PREFIX-RSC#OP_INTERVAL (for example,
* "fail-count-myrsc#monitor_30000"). The '#' is used because it is not
* a valid character in a resource ID, to reliably distinguish where the
* operation name begins. The '_' is used simply to be more comparable to
* action labels like "myrsc_monitor_30000".
*/
static inline char *
pcmk__fail_attr_name(const char *prefix, const char *rsc_id, const char *op,
guint interval_ms)
{
CRM_CHECK(prefix && rsc_id && op, return NULL);
return crm_strdup_printf("%s-%s#%s_%u", prefix, rsc_id, op, interval_ms);
}
static inline char *
pcmk__failcount_name(const char *rsc_id, const char *op, guint interval_ms)
{
return pcmk__fail_attr_name(PCMK__FAIL_COUNT_PREFIX, rsc_id, op,
interval_ms);
}
static inline char *
pcmk__lastfailure_name(const char *rsc_id, const char *op, guint interval_ms)
{
return pcmk__fail_attr_name(PCMK__LAST_FAILURE_PREFIX, rsc_id, op,
interval_ms);
}
// internal resource agent functions (from agents.c)
int pcmk__effective_rc(int rc);
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_INTERNAL__H
diff --git a/lib/common/procfs.c b/lib/common/procfs.c
index 4053a50d06..91853aafe7 100644
--- a/lib/common/procfs.c
+++ b/lib/common/procfs.c
@@ -1,233 +1,255 @@
/*
* Copyright 2015-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <dirent.h>
#include <ctype.h>
#if HAVE_LINUX_PROCFS
/*!
* \internal
* \brief Get process ID and name associated with a /proc directory entry
*
* \param[in] entry Directory entry (must be result of readdir() on /proc)
* \param[out] name If not NULL, a char[16] to hold the process name
* \param[out] pid If not NULL, will be set to process ID of entry
*
* \return Standard Pacemaker return code
* \note This should be called only on Linux systems, as not all systems that
* support /proc store process names and IDs in the same way. The kernel
* limits the process name to the first 15 characters (plus terminator).
* It would be nice if there were a public kernel API constant for that
* limit, but there isn't.
*/
static int
pcmk__procfs_process_info(const struct dirent *entry, char *name, pid_t *pid)
{
int fd, local_pid;
FILE *file;
struct stat statbuf;
char procpath[128] = { 0 };
/* We're only interested in entries whose name is a PID,
* so skip anything non-numeric or that is too long.
*
* 114 = 128 - strlen("/proc/") - strlen("/status") - 1
*/
local_pid = atoi(entry->d_name);
if ((local_pid <= 0) || (strlen(entry->d_name) > 114)) {
return -1;
}
if (pid) {
*pid = (pid_t) local_pid;
}
/* Get this entry's file information */
strcpy(procpath, "/proc/");
strcat(procpath, entry->d_name);
fd = open(procpath, O_RDONLY);
if (fd < 0 ) {
return -1;
}
if (fstat(fd, &statbuf) < 0) {
close(fd);
return -1;
}
close(fd);
/* We're only interested in subdirectories */
if (!S_ISDIR(statbuf.st_mode)) {
return -1;
}
/* Read the first entry ("Name:") from the process's status file.
* We could handle the valgrind case if we parsed the cmdline file
* instead, but that's more of a pain than it's worth.
*/
if (name != NULL) {
strcat(procpath, "/status");
file = fopen(procpath, "r");
if (!file) {
return -1;
}
if (fscanf(file, "Name:\t%15[^\n]", name) != 1) {
fclose(file);
return -1;
}
name[15] = 0;
fclose(file);
}
return 0;
}
#endif // HAVE_LINUX_PROCFS
/*!
* \internal
* \brief Return process ID of a named process
*
* \param[in] name Process name (as used in /proc/.../status)
*
* \return Process ID of named process if running, 0 otherwise
*
* \note This will return 0 if the process is being run via valgrind.
* This should be called only on Linux systems.
*/
pid_t
pcmk__procfs_pid_of(const char *name)
{
#if HAVE_LINUX_PROCFS
DIR *dp;
struct dirent *entry;
pid_t pid = 0;
char entry_name[64] = { 0 };
dp = opendir("/proc");
if (dp == NULL) {
crm_notice("Can not read /proc directory to track existing components");
return 0;
}
while ((entry = readdir(dp)) != NULL) {
if ((pcmk__procfs_process_info(entry, entry_name, &pid) == pcmk_rc_ok)
&& pcmk__str_eq(entry_name, name, pcmk__str_casei)
&& (pcmk__pid_active(pid, NULL) == pcmk_rc_ok)) {
crm_info("Found %s active as process %lld", name, (long long) pid);
break;
}
pid = 0;
}
closedir(dp);
return pid;
#else
return 0;
#endif // HAVE_LINUX_PROCFS
}
/*!
* \internal
* \brief Calculate number of logical CPU cores from procfs
*
* \return Number of cores (or 1 if unable to determine)
*/
unsigned int
pcmk__procfs_num_cores(void)
{
#if HAVE_LINUX_PROCFS
int cores = 0;
FILE *stream = NULL;
/* Parse /proc/stat instead of /proc/cpuinfo because it's smaller */
stream = fopen("/proc/stat", "r");
if (stream == NULL) {
crm_perror(LOG_INFO, "Could not open /proc/stat");
} else {
char buffer[2048];
while (fgets(buffer, sizeof(buffer), stream)) {
if (pcmk__starts_with(buffer, "cpu") && isdigit(buffer[3])) {
++cores;
}
}
fclose(stream);
}
return cores? cores : 1;
#else
return 1;
#endif // HAVE_LINUX_PROCFS
}
/*!
* \internal
* \brief Get the executable path corresponding to a process ID
*
* \param[in] pid Process ID to check
* \param[out] path Where to store executable path
* \param[in] path_size Size of \p path in characters (ideally PATH_MAX)
*
* \return Standard Pacemaker error code (as possible errno values from
* readlink())
*/
int
pcmk__procfs_pid2path(pid_t pid, char path[], size_t path_size)
{
#if HAVE_LINUX_PROCFS
char procfs_exe_path[PATH_MAX];
ssize_t link_rc;
if (snprintf(procfs_exe_path, PATH_MAX, "/proc/%lld/exe",
(long long) pid) >= PATH_MAX) {
return ENAMETOOLONG; // Truncated (shouldn't be possible in practice)
}
link_rc = readlink(procfs_exe_path, path, path_size - 1);
if (link_rc < 0) {
return errno;
} else if (link_rc >= (path_size - 1)) {
return ENAMETOOLONG;
}
path[link_rc] = '\0';
return pcmk_rc_ok;
#else
return EOPNOTSUPP;
#endif // HAVE_LINUX_PROCFS
}
/*!
* \internal
* \brief Check whether process ID information is available from procfs
*
* \return true if process ID information is available, otherwise false
*/
bool
pcmk__procfs_has_pids(void)
{
#if HAVE_LINUX_PROCFS
static bool have_pids = false;
static bool checked = false;
if (!checked) {
char path[PATH_MAX];
have_pids = pcmk__procfs_pid2path(getpid(), path, sizeof(path)) == pcmk_rc_ok;
checked = true;
}
return have_pids;
#else
return false;
#endif // HAVE_LINUX_PROCFS
}
+
+/*!
+ * \internal
+ * \brief Trigger a sysrq command if supported on current platform
+ *
+ * \param[in] t Sysrq command to trigger
+ */
+void
+pcmk__sysrq_trigger(char t)
+{
+#if HAVE_LINUX_PROCFS
+ // Root can always write here, regardless of kernel.sysrq value
+ FILE *procf = fopen("/proc/sysrq-trigger", "a");
+
+ if (procf == NULL) {
+ crm_warn("Could not open sysrq-trigger: %s", strerror(errno));
+ } else {
+ fprintf(procf, "%c\n", t);
+ fclose(procf);
+ }
+#endif // HAVE_LINUX_PROCFS
+}
diff --git a/lib/common/watchdog.c b/lib/common/watchdog.c
index 6196e571a9..e64487a9ad 100644
--- a/lib/common/watchdog.c
+++ b/lib/common/watchdog.c
@@ -1,315 +1,293 @@
/*
* Copyright 2013-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sched.h>
#include <sys/ioctl.h>
#include <sys/reboot.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <ctype.h>
#include <dirent.h>
#include <signal.h>
static pid_t sbd_pid = 0;
-/*!
- * \internal
- * \brief Trigger a sysrq command if supported on current platform
- *
- * \param[in] t Sysrq command to trigger
- */
-static void
-sysrq_trigger(char t)
-{
-#if HAVE_LINUX_PROCFS
- // Root can always write here, regardless of kernel.sysrq value
- FILE *procf = fopen("/proc/sysrq-trigger", "a");
-
- if (procf == NULL) {
- crm_warn("Could not open sysrq-trigger: %s", strerror(errno));
- } else {
- fprintf(procf, "%c\n", t);
- fclose(procf);
- }
-#endif // HAVE_LINUX_PROCFS
-}
-
/*!
* \internal
* \brief Tell pacemakerd to panic the local host
*
* \param[in] ppid Process ID of parent process
*/
static void
panic_local_nonroot(pid_t ppid)
{
if (ppid > 1) { // pacemakerd is still our parent
crm_emerg("Escalating panic to " PCMK__SERVER_PACEMAKERD "[%lld]",
(long long) ppid);
} else { // Signal (non-parent) pacemakerd if possible
#if HAVE_LINUX_PROCFS
ppid = pcmk__procfs_pid_of(PCMK__SERVER_PACEMAKERD);
if (ppid > 0) {
union sigval signal_value;
crm_emerg("Signaling " PCMK__SERVER_PACEMAKERD "[%lld] to panic",
(long long) ppid);
memset(&signal_value, 0, sizeof(signal_value));
if (sigqueue(ppid, SIGQUIT, signal_value) < 0) {
crm_emerg("Exiting after signal failure: %s", strerror(errno));
}
} else {
#endif
crm_emerg("Exiting with no known " PCMK__SERVER_PACEMAKERD
"process");
#if HAVE_LINUX_PROCFS
}
#endif
}
crm_exit(CRM_EX_PANIC);
}
/*!
* \internal
* \brief Panic the local host (if root) or tell pacemakerd to do so
*/
static void
panic_local(void)
{
const char *full_panic_action = pcmk__env_option(PCMK__ENV_PANIC_ACTION);
const char *panic_action = full_panic_action;
int reboot_cmd = RB_AUTOBOOT; // Default panic action is reboot
if (geteuid() != 0) { // Non-root caller such as the controller
panic_local_nonroot(getppid());
return;
}
if (pcmk__starts_with(full_panic_action, "sync-")) {
panic_action += sizeof("sync-") - 1;
sync();
}
if (pcmk__str_empty(full_panic_action)
|| pcmk__str_eq(panic_action, PCMK_VALUE_REBOOT, pcmk__str_none)) {
- sysrq_trigger('b');
+ pcmk__sysrq_trigger('b');
} else if (pcmk__str_eq(panic_action, PCMK_VALUE_CRASH, pcmk__str_none)) {
- sysrq_trigger('c');
+ pcmk__sysrq_trigger('c');
} else if (pcmk__str_eq(panic_action, PCMK_VALUE_OFF, pcmk__str_none)) {
- sysrq_trigger('o');
+ pcmk__sysrq_trigger('o');
#ifdef RB_POWER_OFF
reboot_cmd = RB_POWER_OFF;
#elif defined(RB_POWEROFF)
reboot_cmd = RB_POWEROFF;
#endif
} else {
crm_warn("Using default '" PCMK_VALUE_REBOOT "' for local option PCMK_"
PCMK__ENV_PANIC_ACTION " because '%s' is not a valid value",
full_panic_action);
- sysrq_trigger('b');
+ pcmk__sysrq_trigger('b');
}
// sysrq failed or is not supported on this platform, so fall back to reboot
reboot(reboot_cmd);
// Even reboot failed, nothing left to do but exit
crm_emerg("Exiting after reboot failed: %s", strerror(errno));
if (getppid() > 1) { // pacemakerd is parent process
crm_exit(CRM_EX_PANIC);
} else { // This is pacemakerd, or an orphaned subdaemon
crm_exit(CRM_EX_FATAL);
}
}
/*!
* \internal
* \brief Tell sbd to kill the local host, then exit
*/
static void
panic_sbd(void)
{
union sigval signal_value;
pid_t ppid = getppid();
memset(&signal_value, 0, sizeof(signal_value));
/* TODO: Arrange for a slightly less brutal option? */
if(sigqueue(sbd_pid, SIGKILL, signal_value) < 0) {
crm_emerg("Panicking directly because couldn't signal sbd");
panic_local();
}
if(ppid > 1) {
/* child daemon */
crm_exit(CRM_EX_PANIC);
} else {
/* pacemakerd or orphan child */
crm_exit(CRM_EX_FATAL);
}
}
/*!
* \internal
* \brief Panic the local host
*
* Panic the local host either by sbd (if running), directly, or by asking
* pacemakerd. If trace logging this function, exit instead.
*
* \param[in] reason Why panic is needed (for logging only)
*/
void
pcmk__panic(const char *reason)
{
if (pcmk__locate_sbd() > 1) {
crm_emerg("Signaling sbd[%lld] to panic the system: %s",
(long long) sbd_pid, reason);
panic_sbd();
} else {
crm_emerg("Panicking the system directly: %s", reason);
panic_local();
}
}
/*!
* \internal
* \brief Return the process ID of sbd (or 0 if it is not running)
*/
pid_t
pcmk__locate_sbd(void)
{
char *pidfile = NULL;
char *sbd_path = NULL;
int rc;
if(sbd_pid > 1) {
return sbd_pid;
}
/* Look for the pid file */
pidfile = crm_strdup_printf(PCMK__RUN_DIR "/sbd.pid");
sbd_path = crm_strdup_printf("%s/sbd", SBIN_DIR);
/* Read the pid file */
rc = pcmk__pidfile_matches(pidfile, 0, sbd_path, &sbd_pid);
if (rc == pcmk_rc_ok) {
crm_trace("SBD detected at pid %lld (via PID file %s)",
(long long) sbd_pid, pidfile);
#if HAVE_LINUX_PROCFS
} else {
/* Fall back to /proc for systems that support it */
sbd_pid = pcmk__procfs_pid_of("sbd");
crm_trace("SBD detected at pid %lld (via procfs)",
(long long) sbd_pid);
#endif // HAVE_LINUX_PROCFS
}
if(sbd_pid < 0) {
sbd_pid = 0;
crm_trace("SBD not detected");
}
free(pidfile);
free(sbd_path);
return sbd_pid;
}
long
pcmk__get_sbd_watchdog_timeout(void)
{
static long sbd_timeout = -2;
if (sbd_timeout == -2) {
sbd_timeout = crm_get_msec(getenv("SBD_WATCHDOG_TIMEOUT"));
}
return sbd_timeout;
}
bool
pcmk__get_sbd_sync_resource_startup(void)
{
static int sync_resource_startup = PCMK__SBD_SYNC_DEFAULT;
static bool checked_sync_resource_startup = false;
if (!checked_sync_resource_startup) {
const char *sync_env = getenv("SBD_SYNC_RESOURCE_STARTUP");
if (sync_env == NULL) {
crm_trace("Defaulting to %sstart-up synchronization with sbd",
(PCMK__SBD_SYNC_DEFAULT? "" : "no "));
} else if (crm_str_to_boolean(sync_env, &sync_resource_startup) < 0) {
crm_warn("Defaulting to %sstart-up synchronization with sbd "
"because environment value '%s' is invalid",
(PCMK__SBD_SYNC_DEFAULT? "" : "no "), sync_env);
}
checked_sync_resource_startup = true;
}
return sync_resource_startup != 0;
}
long
pcmk__auto_stonith_watchdog_timeout(void)
{
long sbd_timeout = pcmk__get_sbd_watchdog_timeout();
return (sbd_timeout <= 0)? 0 : (2 * sbd_timeout);
}
bool
pcmk__valid_stonith_watchdog_timeout(const char *value)
{
/* @COMPAT At a compatibility break, accept either negative values or a
* specific string like "auto" (but not both) to mean "auto-calculate the
* timeout." Reject other values that aren't parsable as timeouts.
*/
long st_timeout = value? crm_get_msec(value) : 0;
if (st_timeout < 0) {
st_timeout = pcmk__auto_stonith_watchdog_timeout();
crm_debug("Using calculated value %ld for "
PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " (%s)",
st_timeout, value);
}
if (st_timeout == 0) {
crm_debug("Watchdog may be enabled but "
PCMK_OPT_STONITH_WATCHDOG_TIMEOUT " is disabled (%s)",
value? value : "default");
} else if (pcmk__locate_sbd() == 0) {
crm_emerg("Shutting down: " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" configured (%s) but SBD not active",
pcmk__s(value, "auto"));
crm_exit(CRM_EX_FATAL);
return false;
} else {
long sbd_timeout = pcmk__get_sbd_watchdog_timeout();
if (st_timeout < sbd_timeout) {
crm_emerg("Shutting down: " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" (%s) too short (must be >%ldms)",
value, sbd_timeout);
crm_exit(CRM_EX_FATAL);
return false;
}
crm_info("Watchdog configured with " PCMK_OPT_STONITH_WATCHDOG_TIMEOUT
" %s and SBD timeout %ldms",
value, sbd_timeout);
}
return true;
}