Page MenuHomeClusterLabs Projects
Files F4624617

No OneTemporary
Actions

View Options

diff --git a/lib/services/services_linux.c b/lib/services/services_linux.c
index f280065c01..bbe7d5d220 100644
--- a/lib/services/services_linux.c
+++ b/lib/services/services_linux.c
@@ -1,1434 +1,1448 @@
/*
* Copyright 2010-2022 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>
#ifndef _GNU_SOURCE
# define _GNU_SOURCE
#endif
#include <sys/types.h>
#include <sys/stat.h>
#include <sys/wait.h>
#include <errno.h>
#include <unistd.h>
#include <dirent.h>
#include <grp.h>
#include <string.h>
#include <sys/time.h>
#include <sys/resource.h>
#include "crm/crm.h"
#include "crm/common/mainloop.h"
#include "crm/services.h"
#include "crm/services_internal.h"
#include "services_private.h"
static void close_pipe(int fildes[]);
/* We have two alternative ways of handling SIGCHLD when synchronously waiting
* for spawned processes to complete. Both rely on polling a file descriptor to
* discover SIGCHLD events.
*
* If sys/signalfd.h is available (e.g. on Linux), we call signalfd() to
* generate the file descriptor. Otherwise, we use the "self-pipe trick"
* (opening a pipe and writing a byte to it when SIGCHLD is received).
*/
#ifdef HAVE_SYS_SIGNALFD_H
// signalfd() implementation
#include <sys/signalfd.h>
// Everything needed to manage SIGCHLD handling
struct sigchld_data_s {
sigset_t mask; // Signals to block now (including SIGCHLD)
sigset_t old_mask; // Previous set of blocked signals
};
// Initialize SIGCHLD data and prepare for use
static bool
sigchld_setup(struct sigchld_data_s *data)
{
sigemptyset(&(data->mask));
sigaddset(&(data->mask), SIGCHLD);
sigemptyset(&(data->old_mask));
// Block SIGCHLD (saving previous set of blocked signals to restore later)
if (sigprocmask(SIG_BLOCK, &(data->mask), &(data->old_mask)) < 0) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=sigprocmask", pcmk_rc_str(errno));
return false;
}
return true;
}
// Get a file descriptor suitable for polling for SIGCHLD events
static int
sigchld_open(struct sigchld_data_s *data)
{
int fd;
CRM_CHECK(data != NULL, return -1);
fd = signalfd(-1, &(data->mask), SFD_NONBLOCK);
if (fd < 0) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=signalfd", pcmk_rc_str(errno));
}
return fd;
}
// Close a file descriptor returned by sigchld_open()
static void
sigchld_close(int fd)
{
if (fd > 0) {
close(fd);
}
}
// Return true if SIGCHLD was received from polled fd
static bool
sigchld_received(int fd)
{
struct signalfd_siginfo fdsi;
ssize_t s;
if (fd < 0) {
return false;
}
s = read(fd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=read", pcmk_rc_str(errno));
} else if (fdsi.ssi_signo == SIGCHLD) {
return true;
}
return false;
}
// Do anything needed after done waiting for SIGCHLD
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the original set of blocked signals
if ((sigismember(&(data->old_mask), SIGCHLD) == 0)
&& (sigprocmask(SIG_UNBLOCK, &(data->mask), NULL) < 0)) {
crm_warn("Could not clean up after child process completion: %s",
pcmk_rc_str(errno));
}
}
#else // HAVE_SYS_SIGNALFD_H not defined
// Self-pipe implementation (see above for function descriptions)
struct sigchld_data_s {
int pipe_fd[2]; // Pipe file descriptors
struct sigaction sa; // Signal handling info (with SIGCHLD)
struct sigaction old_sa; // Previous signal handling info
};
// We need a global to use in the signal handler
volatile struct sigchld_data_s *last_sigchld_data = NULL;
static void
sigchld_handler()
{
// We received a SIGCHLD, so trigger pipe polling
if ((last_sigchld_data != NULL)
&& (last_sigchld_data->pipe_fd[1] >= 0)
&& (write(last_sigchld_data->pipe_fd[1], "", 1) == -1)) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=write", pcmk_rc_str(errno));
}
}
static bool
sigchld_setup(struct sigchld_data_s *data)
{
int rc;
data->pipe_fd[0] = data->pipe_fd[1] = -1;
if (pipe(data->pipe_fd) == -1) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=pipe", pcmk_rc_str(errno));
return false;
}
rc = pcmk__set_nonblocking(data->pipe_fd[0]);
if (rc != pcmk_rc_ok) {
crm_info("Could not set pipe input non-blocking: %s " CRM_XS " rc=%d",
pcmk_rc_str(rc), rc);
}
rc = pcmk__set_nonblocking(data->pipe_fd[1]);
if (rc != pcmk_rc_ok) {
crm_info("Could not set pipe output non-blocking: %s " CRM_XS " rc=%d",
pcmk_rc_str(rc), rc);
}
// Set SIGCHLD handler
data->sa.sa_handler = sigchld_handler;
data->sa.sa_flags = 0;
sigemptyset(&(data->sa.sa_mask));
if (sigaction(SIGCHLD, &(data->sa), &(data->old_sa)) < 0) {
crm_info("Wait for child process completion failed: %s "
CRM_XS " source=sigaction", pcmk_rc_str(errno));
}
// Remember data for use in signal handler
last_sigchld_data = data;
return true;
}
static int
sigchld_open(struct sigchld_data_s *data)
{
CRM_CHECK(data != NULL, return -1);
return data->pipe_fd[0];
}
static void
sigchld_close(int fd)
{
// Pipe will be closed in sigchld_cleanup()
return;
}
static bool
sigchld_received(int fd)
{
char ch;
if (fd < 0) {
return false;
}
// Clear out the self-pipe
while (read(fd, &ch, 1) == 1) /*omit*/;
return true;
}
static void
sigchld_cleanup(struct sigchld_data_s *data)
{
// Restore the previous SIGCHLD handler
if (sigaction(SIGCHLD, &(data->old_sa), NULL) < 0) {
crm_warn("Could not clean up after child process completion: %s",
pcmk_rc_str(errno));
}
close_pipe(data->pipe_fd);
}
#endif
/*!
* \internal
* \brief Close the two file descriptors of a pipe
*
* \param[in] fildes Array of file descriptors opened by pipe()
*/
static void
close_pipe(int fildes[])
{
if (fildes[0] >= 0) {
close(fildes[0]);
fildes[0] = -1;
}
if (fildes[1] >= 0) {
close(fildes[1]);
fildes[1] = -1;
}
}
static gboolean
svc_read_output(int fd, svc_action_t * op, bool is_stderr)
{
char *data = NULL;
int rc = 0, len = 0;
char buf[500];
static const size_t buf_read_len = sizeof(buf) - 1;
if (fd < 0) {
crm_trace("No fd for %s", op->id);
return FALSE;
}
if (is_stderr && op->stderr_data) {
len = strlen(op->stderr_data);
data = op->stderr_data;
crm_trace("Reading %s stderr into offset %d", op->id, len);
} else if (is_stderr == FALSE && op->stdout_data) {
len = strlen(op->stdout_data);
data = op->stdout_data;
crm_trace("Reading %s stdout into offset %d", op->id, len);
} else {
crm_trace("Reading %s %s into offset %d", op->id, is_stderr?"stderr":"stdout", len);
}
do {
rc = read(fd, buf, buf_read_len);
if (rc > 0) {
buf[rc] = 0;
crm_trace("Got %d chars: %.80s", rc, buf);
data = pcmk__realloc(data, len + rc + 1);
len += sprintf(data + len, "%s", buf);
} else if (errno != EINTR) {
/* error or EOF
* Cleanup happens in pipe_done()
*/
rc = FALSE;
break;
}
} while (rc == buf_read_len || rc < 0);
if (is_stderr) {
op->stderr_data = data;
} else {
op->stdout_data = data;
}
return rc;
}
static int
dispatch_stdout(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
static int
dispatch_stderr(gpointer userdata)
{
svc_action_t *op = (svc_action_t *) userdata;
return svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
static void
pipe_out_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
crm_trace("%p", op);
op->opaque->stdout_gsource = NULL;
if (op->opaque->stdout_fd > STDOUT_FILENO) {
close(op->opaque->stdout_fd);
}
op->opaque->stdout_fd = -1;
}
static void
pipe_err_done(gpointer user_data)
{
svc_action_t *op = (svc_action_t *) user_data;
op->opaque->stderr_gsource = NULL;
if (op->opaque->stderr_fd > STDERR_FILENO) {
close(op->opaque->stderr_fd);
}
op->opaque->stderr_fd = -1;
}
static struct mainloop_fd_callbacks stdout_callbacks = {
.dispatch = dispatch_stdout,
.destroy = pipe_out_done,
};
static struct mainloop_fd_callbacks stderr_callbacks = {
.dispatch = dispatch_stderr,
.destroy = pipe_err_done,
};
static void
set_ocf_env(const char *key, const char *value, gpointer user_data)
{
if (setenv(key, value, 1) != 0) {
crm_perror(LOG_ERR, "setenv failed for key:%s and value:%s", key, value);
}
}
static void
set_ocf_env_with_prefix(gpointer key, gpointer value, gpointer user_data)
{
char buffer[500];
snprintf(buffer, sizeof(buffer), strcmp(key, "OCF_CHECK_LEVEL") != 0 ? "OCF_RESKEY_%s" : "%s", (char *)key);
set_ocf_env(buffer, value, user_data);
}
static void
set_alert_env(gpointer key, gpointer value, gpointer user_data)
{
int rc;
if (value != NULL) {
rc = setenv(key, value, 1);
} else {
rc = unsetenv(key);
}
if (rc < 0) {
crm_perror(LOG_ERR, "setenv %s=%s",
(char*)key, (value? (char*)value : ""));
} else {
crm_trace("setenv %s=%s", (char*)key, (value? (char*)value : ""));
}
}
/*!
* \internal
* \brief Add environment variables suitable for an action
*
* \param[in] op Action to use
*/
static void
add_action_env_vars(const svc_action_t *op)
{
void (*env_setter)(gpointer, gpointer, gpointer) = NULL;
if (op->agent == NULL) {
env_setter = set_alert_env; /* we deal with alert handler */
} else if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) {
env_setter = set_ocf_env_with_prefix;
}
if (env_setter != NULL && op->params != NULL) {
g_hash_table_foreach(op->params, env_setter, NULL);
}
if (env_setter == NULL || env_setter == set_alert_env) {
return;
}
set_ocf_env("OCF_RA_VERSION_MAJOR", PCMK_OCF_MAJOR_VERSION, NULL);
set_ocf_env("OCF_RA_VERSION_MINOR", PCMK_OCF_MINOR_VERSION, NULL);
set_ocf_env("OCF_ROOT", OCF_ROOT_DIR, NULL);
set_ocf_env("OCF_EXIT_REASON_PREFIX", PCMK_OCF_REASON_PREFIX, NULL);
if (op->rsc) {
set_ocf_env("OCF_RESOURCE_INSTANCE", op->rsc, NULL);
}
if (op->agent != NULL) {
set_ocf_env("OCF_RESOURCE_TYPE", op->agent, NULL);
}
/* Notes: this is not added to specification yet. Sept 10,2004 */
if (op->provider != NULL) {
set_ocf_env("OCF_RESOURCE_PROVIDER", op->provider, NULL);
}
}
static void
pipe_in_single_parameter(gpointer key, gpointer value, gpointer user_data)
{
svc_action_t *op = user_data;
char *buffer = crm_strdup_printf("%s=%s\n", (char *)key, (char *) value);
int ret, total = 0, len = strlen(buffer);
do {
errno = 0;
ret = write(op->opaque->stdin_fd, buffer + total, len - total);
if (ret > 0) {
total += ret;
}
} while ((errno == EINTR) && (total < len));
free(buffer);
}
/*!
* \internal
* \brief Pipe parameters in via stdin for action
*
* \param[in] op Action to use
*/
static void
pipe_in_action_stdin_parameters(const svc_action_t *op)
{
crm_debug("sending args");
if (op->params) {
g_hash_table_foreach(op->params, pipe_in_single_parameter, (gpointer) op);
}
}
gboolean
recurring_action_timer(gpointer data)
{
svc_action_t *op = data;
crm_debug("Scheduling another invocation of %s", op->id);
/* Clean out the old result */
free(op->stdout_data);
op->stdout_data = NULL;
free(op->stderr_data);
op->stderr_data = NULL;
op->opaque->repeat_timer = 0;
services_action_async(op, NULL);
return FALSE;
}
/*!
* \internal
* \brief Finalize handling of an asynchronous operation
*
* Given a completed asynchronous operation, cancel or reschedule it as
* appropriate if recurring, call its callback if registered, stop tracking it,
* and clean it up.
*
* \param[in,out] op Operation to finalize
*
* \return Standard Pacemaker return code
* \retval EINVAL Caller supplied NULL or invalid \p op
* \retval EBUSY Uncanceled recurring action has only been cleaned up
* \retval pcmk_rc_ok Action has been freed
*
* \note If the return value is not pcmk_rc_ok, the caller is responsible for
* freeing the action.
*/
int
services__finalize_async_op(svc_action_t *op)
{
CRM_CHECK((op != NULL) && !(op->synchronous), return EINVAL);
if (op->interval_ms != 0) {
// Recurring operations must be either cancelled or rescheduled
if (op->cancel) {
services__set_cancelled(op);
cancel_recurring_action(op);
} else {
op->opaque->repeat_timer = g_timeout_add(op->interval_ms,
recurring_action_timer,
(void *) op);
}
}
if (op->opaque->callback != NULL) {
op->opaque->callback(op);
}
// Stop tracking the operation (as in-flight or blocked)
op->pid = 0;
services_untrack_op(op);
if ((op->interval_ms != 0) && !(op->cancel)) {
// Do not free recurring actions (they will get freed when cancelled)
services_action_cleanup(op);
return EBUSY;
}
services_action_free(op);
return pcmk_rc_ok;
}
static void
close_op_input(svc_action_t *op)
{
if (op->opaque->stdin_fd >= 0) {
close(op->opaque->stdin_fd);
}
}
static void
finish_op_output(svc_action_t *op, bool is_stderr)
{
mainloop_io_t **source;
int fd;
if (is_stderr) {
source = &(op->opaque->stderr_gsource);
fd = op->opaque->stderr_fd;
} else {
source = &(op->opaque->stdout_gsource);
fd = op->opaque->stdout_fd;
}
if (op->synchronous || *source) {
crm_trace("Finish reading %s[%d] %s",
op->id, op->pid, (is_stderr? "stdout" : "stderr"));
svc_read_output(fd, op, is_stderr);
if (op->synchronous) {
close(fd);
} else {
mainloop_del_fd(*source);
*source = NULL;
}
}
}
// Log an operation's stdout and stderr
static void
log_op_output(svc_action_t *op)
{
char *prefix = crm_strdup_printf("%s[%d] error output", op->id, op->pid);
/* The library caller has better context to know how important the output
* is, so log it at info and debug severity here. They can log it again at
* higher severity if appropriate.
*/
crm_log_output(LOG_INFO, prefix, op->stderr_data);
strcpy(prefix + strlen(prefix) - strlen("error output"), "output");
crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
free(prefix);
}
// Truncate exit reasons at this many characters
#define EXIT_REASON_MAX_LEN 128
static void
parse_exit_reason_from_stderr(svc_action_t *op)
{
const char *reason_start = NULL;
const char *reason_end = NULL;
const int prefix_len = strlen(PCMK_OCF_REASON_PREFIX);
if ((op->stderr_data == NULL) ||
// Only OCF agents have exit reasons in stderr
!pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_none)) {
return;
}
// Find the last occurrence of the magic string indicating an exit reason
for (const char *cur = strstr(op->stderr_data, PCMK_OCF_REASON_PREFIX);
cur != NULL; cur = strstr(cur, PCMK_OCF_REASON_PREFIX)) {
cur += prefix_len; // Skip over magic string
reason_start = cur;
}
if ((reason_start == NULL) || (reason_start[0] == '\n')
|| (reason_start[0] == '\0')) {
return; // No or empty exit reason
}
// Exit reason goes to end of line (or end of output)
reason_end = strchr(reason_start, '\n');
if (reason_end == NULL) {
reason_end = reason_start + strlen(reason_start);
}
// Limit size of exit reason to something reasonable
if (reason_end > (reason_start + EXIT_REASON_MAX_LEN)) {
reason_end = reason_start + EXIT_REASON_MAX_LEN;
}
free(op->opaque->exit_reason);
op->opaque->exit_reason = strndup(reason_start, reason_end - reason_start);
}
/*!
* \internal
* \brief Process the completion of an asynchronous child process
*
* \param[in] p Child process that completed
* \param[in] pid Process ID of child
* \param[in] core (unused)
* \param[in] signo Signal that interrupted child, if any
* \param[in] exitcode Exit status of child process
*/
static void
async_action_complete(mainloop_child_t *p, pid_t pid, int core, int signo,
int exitcode)
{
svc_action_t *op = mainloop_child_userdata(p);
mainloop_clear_child_userdata(p);
CRM_CHECK(op->pid == pid,
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, "Bug in mainloop handling");
return);
/* Depending on the priority the mainloop gives the stdout and stderr
* file descriptors, this function could be called before everything has
* been read from them, so force a final read now.
*/
finish_op_output(op, true);
finish_op_output(op, false);
close_op_input(op);
if (signo == 0) {
crm_debug("%s[%d] exited with status %d", op->id, op->pid, exitcode);
services__set_result(op, exitcode, PCMK_EXEC_DONE, NULL);
log_op_output(op);
parse_exit_reason_from_stderr(op);
} else if (mainloop_child_timeout(p)) {
const char *what = NULL;
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_STONITH,
pcmk__str_none)) {
what = "Fence agent";
} else if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_ALERT,
pcmk__str_none)) {
what = "Alert agent";
} else if (op->standard != NULL) {
what = "Resource agent";
} else {
what = "Process";
}
crm_info("%s[%d] timed out after %dms", op->id, op->pid, op->timeout);
services__format_result(op, services__generic_error(op),
PCMK_EXEC_TIMEOUT,
"%s did not complete in time", what);
} else if (op->cancel) {
/* If an in-flight recurring operation was killed because it was
* cancelled, don't treat that as a failure.
*/
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_CANCELLED, NULL);
} else {
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
- services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
- "Process interrupted by signal");
+ services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
+ "Process interrupted by %s signal",
+ strsignal(signo));
}
services__finalize_async_op(op);
}
/*!
* \internal
* \brief Return agent standard's exit status for "generic error"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for errors in general.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__generic_error(svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, "status", pcmk__str_casei)) {
return PCMK_LSB_STATUS_UNKNOWN;
}
#if SUPPORT_NAGIOS
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
return NAGIOS_STATE_UNKNOWN;
}
#endif
return PCMK_OCF_UNKNOWN_ERROR;
}
/*!
* \internal
* \brief Return agent standard's exit status for "not installed"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "not installed" errors.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__not_installed_error(svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, "status", pcmk__str_casei)) {
return PCMK_LSB_STATUS_NOT_INSTALLED;
}
#if SUPPORT_NAGIOS
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
return NAGIOS_STATE_UNKNOWN;
}
#endif
return PCMK_OCF_NOT_INSTALLED;
}
/*!
* \internal
* \brief Return agent standard's exit status for "insufficient privileges"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "insufficient privileges" errors.
*
* \param[in] op Action that error is for
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__authorization_error(svc_action_t *op)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, "status", pcmk__str_casei)) {
return PCMK_LSB_STATUS_INSUFFICIENT_PRIV;
}
#if SUPPORT_NAGIOS
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
return NAGIOS_INSUFFICIENT_PRIV;
}
#endif
return PCMK_OCF_INSUFFICIENT_PRIV;
}
/*!
* \internal
* \brief Return agent standard's exit status for "not configured"
*
* When returning an internal error for an action, a value that is appropriate
* to the action's agent standard must be used. This function returns a value
* appropriate for "not configured" errors.
*
* \param[in] op Action that error is for
* \param[in] is_fatal Whether problem is cluster-wide instead of only local
*
* \return Exit status appropriate to agent standard
* \note Actions without a standard will get PCMK_OCF_UNKNOWN_ERROR.
*/
int
services__configuration_error(svc_action_t *op, bool is_fatal)
{
if ((op == NULL) || (op->standard == NULL)) {
return PCMK_OCF_UNKNOWN_ERROR;
}
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)
&& pcmk__str_eq(op->action, "status", pcmk__str_casei)) {
return PCMK_LSB_NOT_CONFIGURED;
}
#if SUPPORT_NAGIOS
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
return NAGIOS_STATE_UNKNOWN;
}
#endif
return is_fatal? PCMK_OCF_NOT_CONFIGURED : PCMK_OCF_INVALID_PARAM;
}
/*!
* \internal
* \brief Set operation rc and status per errno from stat(), fork() or execvp()
*
* \param[in,out] op Operation to set rc and status for
* \param[in] error Value of errno after system call
*
* \return void
*/
void
services__handle_exec_error(svc_action_t * op, int error)
{
+ const char *name = op->opaque->exec;
+
+ if (name == NULL) {
+ name = op->agent;
+ if (name == NULL) {
+ name = op->id;
+ }
+ }
+
switch (error) { /* see execve(2), stat(2) and fork(2) */
case ENOENT: /* No such file or directory */
case EISDIR: /* Is a directory */
case ENOTDIR: /* Path component is not a directory */
case EINVAL: /* Invalid executable format */
case ENOEXEC: /* Invalid executable format */
- services__set_result(op, services__not_installed_error(op),
- PCMK_EXEC_NOT_INSTALLED, pcmk_rc_str(error));
+ services__format_result(op, services__not_installed_error(op),
+ PCMK_EXEC_NOT_INSTALLED, "%s: %s",
+ name, pcmk_rc_str(error));
break;
case EACCES: /* permission denied (various errors) */
case EPERM: /* permission denied (various errors) */
- services__set_result(op, services__authorization_error(op),
- PCMK_EXEC_ERROR, pcmk_rc_str(error));
+ services__format_result(op, services__authorization_error(op),
+ PCMK_EXEC_ERROR, "%s: %s",
+ name, pcmk_rc_str(error));
break;
default:
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, pcmk_rc_str(error));
}
}
/*!
* \internal
* \brief Exit a child process that failed before executing agent
*
* \param[in] op Action that failed
* \param[in] exit_status Exit status code to use
* \param[in] exit_reason Exit reason to output if for OCF agent
*/
static void
exit_child(svc_action_t *op, int exit_status, const char *exit_reason)
{
if ((op != NULL) && (exit_reason != NULL)
&& pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_OCF,
pcmk__str_none)) {
fprintf(stderr, PCMK_OCF_REASON_PREFIX "%s\n", exit_reason);
}
_exit(exit_status);
}
static void
action_launch_child(svc_action_t *op)
{
int rc;
/* SIGPIPE is ignored (which is different from signal blocking) by the gnutls library.
* Depending on the libqb version in use, libqb may set SIGPIPE to be ignored as well.
* We do not want this to be inherited by the child process. By resetting this the signal
* to the default behavior, we avoid some potential odd problems that occur during OCF
* scripts when SIGPIPE is ignored by the environment. */
signal(SIGPIPE, SIG_DFL);
#if defined(HAVE_SCHED_SETSCHEDULER)
if (sched_getscheduler(0) != SCHED_OTHER) {
struct sched_param sp;
memset(&sp, 0, sizeof(sp));
sp.sched_priority = 0;
if (sched_setscheduler(0, SCHED_OTHER, &sp) == -1) {
crm_info("Could not reset scheduling policy for %s", op->id);
}
}
#endif
if (setpriority(PRIO_PROCESS, 0, 0) == -1) {
crm_info("Could not reset process priority for %s", op->id);
}
/* Man: The call setpgrp() is equivalent to setpgid(0,0)
* _and_ compiles on BSD variants too
* need to investigate if it works the same too.
*/
setpgid(0, 0);
pcmk__close_fds_in_child(false);
/* It would be nice if errors in this function could be reported as
* execution status (for example, PCMK_EXEC_NO_SECRETS for the secrets error
* below) instead of exit status. However, we've already forked, so
* exit status is all we have. At least for OCF actions, we can output an
* exit reason for the parent to parse.
*/
#if SUPPORT_CIBSECRETS
rc = pcmk__substitute_secrets(op->rsc, op->params);
if (rc != pcmk_rc_ok) {
if (pcmk__str_eq(op->action, "stop", pcmk__str_casei)) {
crm_info("Proceeding with stop operation for %s "
"despite being unable to load CIB secrets (%s)",
op->rsc, pcmk_rc_str(rc));
} else {
crm_err("Considering %s unconfigured "
"because unable to load CIB secrets: %s",
op->rsc, pcmk_rc_str(rc));
exit_child(op, services__configuration_error(op, false),
"Unable to load CIB secrets");
}
}
#endif
add_action_env_vars(op);
/* Become the desired user */
if (op->opaque->uid && (geteuid() == 0)) {
// If requested, set effective group
if (op->opaque->gid && (setgid(op->opaque->gid) < 0)) {
crm_err("Considering %s unauthorized because could not set "
"child group to %d: %s",
op->id, op->opaque->gid, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not set group for child process");
}
// Erase supplementary group list
// (We could do initgroups() if we kept a copy of the username)
if (setgroups(0, NULL) < 0) {
crm_err("Considering %s unauthorized because could not "
"clear supplementary groups: %s", op->id, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not clear supplementary groups for child process");
}
// Set effective user
if (setuid(op->opaque->uid) < 0) {
crm_err("Considering %s unauthorized because could not set user "
"to %d: %s", op->id, op->opaque->uid, strerror(errno));
exit_child(op, services__authorization_error(op),
"Could not set user for child process");
}
}
// Execute the agent (doesn't return if successful)
execvp(op->opaque->exec, op->opaque->args);
// An earlier stat() should have avoided most possible errors
rc = errno;
services__handle_exec_error(op, rc);
crm_err("Unable to execute %s: %s", op->id, strerror(rc));
exit_child(op, op->rc, "Child process was unable to execute file");
}
/*!
* \internal
* \brief Wait for synchronous action to complete, and set its result
*
* \param[in] op Action to wait for
* \param[in] data Child signal data
*/
static void
wait_for_sync_result(svc_action_t *op, struct sigchld_data_s *data)
{
int status = 0;
int timeout = op->timeout;
time_t start = time(NULL);
struct pollfd fds[3];
int wait_rc = 0;
const char *wait_reason = NULL;
fds[0].fd = op->opaque->stdout_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
fds[1].fd = op->opaque->stderr_fd;
fds[1].events = POLLIN;
fds[1].revents = 0;
fds[2].fd = sigchld_open(data);
fds[2].events = POLLIN;
fds[2].revents = 0;
crm_trace("Waiting for %s[%d]", op->id, op->pid);
do {
int poll_rc = poll(fds, 3, timeout);
wait_reason = NULL;
if (poll_rc > 0) {
if (fds[0].revents & POLLIN) {
svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
if (fds[1].revents & POLLIN) {
svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
if ((fds[2].revents & POLLIN) && sigchld_received(fds[2].fd)) {
wait_rc = waitpid(op->pid, &status, WNOHANG);
if ((wait_rc > 0) || ((wait_rc < 0) && (errno == ECHILD))) {
// Child process exited or doesn't exist
break;
} else if (wait_rc < 0) {
wait_reason = pcmk_rc_str(errno);
crm_info("Wait for completion of %s[%d] failed: %s "
CRM_XS " source=waitpid",
op->id, op->pid, wait_reason);
wait_rc = 0; // Act as if process is still running
}
}
} else if (poll_rc == 0) {
// Poll timed out with no descriptors ready
timeout = 0;
break;
} else if ((poll_rc < 0) && (errno != EINTR)) {
wait_reason = pcmk_rc_str(errno);
crm_info("Wait for completion of %s[%d] failed: %s "
CRM_XS " source=poll", op->id, op->pid, wait_reason);
break;
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
crm_trace("Stopped waiting for %s[%d]", op->id, op->pid);
finish_op_output(op, true);
finish_op_output(op, false);
close_op_input(op);
sigchld_close(fds[2].fd);
if (wait_rc <= 0) {
if ((op->timeout > 0) && (timeout <= 0)) {
services__set_result(op, services__generic_error(op),
PCMK_EXEC_TIMEOUT,
"Process did not exit within specified timeout");
crm_info("%s[%d] timed out after %dms",
op->id, op->pid, op->timeout);
} else {
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR, wait_reason);
}
/* If only child hasn't been successfully waited for, yet.
This is to limit killing wrong target a bit more. */
if ((wait_rc == 0) && (waitpid(op->pid, &status, WNOHANG) == 0)) {
if (kill(op->pid, SIGKILL)) {
crm_warn("Could not kill rogue child %s[%d]: %s",
op->id, op->pid, pcmk_rc_str(errno));
}
/* Safe to skip WNOHANG here as we sent non-ignorable signal. */
while ((waitpid(op->pid, &status, 0) == (pid_t) -1)
&& (errno == EINTR)) {
/* keep waiting */;
}
}
} else if (WIFEXITED(status)) {
services__set_result(op, WEXITSTATUS(status), PCMK_EXEC_DONE, NULL);
parse_exit_reason_from_stderr(op);
crm_info("%s[%d] exited with status %d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
- services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR,
- "Process interrupted by signal");
+ services__format_result(op, services__generic_error(op),
+ PCMK_EXEC_ERROR,
+ "Process interrupted by %s signal",
+ strsignal(signo));
crm_info("%s[%d] terminated with signal %d (%s)",
op->id, op->pid, signo, strsignal(signo));
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
crm_warn("%s[%d] dumped core", op->id, op->pid);
}
#endif
} else {
// Shouldn't be possible to get here
services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR,
"Unable to wait for child to complete");
}
}
/*!
* \internal
* \brief Execute an action whose standard uses executable files
*
* \param[in] op Action to execute
*
* \return Standard Pacemaker return value
* \retval EBUSY Recurring operation could not be initiated
* \retval pcmk_rc_error Synchronous action failed
* \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action
* should not be freed (because it's pending or because
* it failed to execute and was already freed)
*
* \note If the return value for an asynchronous action is not pcmk_rc_ok, the
* caller is responsible for freeing the action.
*/
int
services__execute_file(svc_action_t *op)
{
int stdout_fd[2];
int stderr_fd[2];
int stdin_fd[2] = {-1, -1};
int rc;
struct stat st;
struct sigchld_data_s data;
// Catch common failure conditions early
if (stat(op->opaque->exec, &st) != 0) {
rc = errno;
crm_info("Cannot execute '%s': %s " CRM_XS " stat rc=%d",
op->opaque->exec, pcmk_strerror(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pipe(stdout_fd) < 0) {
rc = errno;
crm_info("Cannot execute '%s': %s " CRM_XS " pipe(stdout) rc=%d",
op->opaque->exec, pcmk_strerror(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pipe(stderr_fd) < 0) {
rc = errno;
close_pipe(stdout_fd);
crm_info("Cannot execute '%s': %s " CRM_XS " pipe(stderr) rc=%d",
op->opaque->exec, pcmk_strerror(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
if (pcmk_is_set(pcmk_get_ra_caps(op->standard), pcmk_ra_cap_stdin)) {
if (pipe(stdin_fd) < 0) {
rc = errno;
close_pipe(stdout_fd);
close_pipe(stderr_fd);
crm_info("Cannot execute '%s': %s " CRM_XS " pipe(stdin) rc=%d",
op->opaque->exec, pcmk_strerror(rc), rc);
services__handle_exec_error(op, rc);
goto done;
}
}
if (op->synchronous && !sigchld_setup(&data)) {
close_pipe(stdin_fd);
close_pipe(stdout_fd);
close_pipe(stderr_fd);
sigchld_cleanup(&data);
services__set_result(op, services__generic_error(op), PCMK_EXEC_ERROR,
"Could not manage signals for child process");
goto done;
}
op->pid = fork();
switch (op->pid) {
case -1:
rc = errno;
close_pipe(stdin_fd);
close_pipe(stdout_fd);
close_pipe(stderr_fd);
crm_info("Cannot execute '%s': %s " CRM_XS " fork rc=%d",
op->opaque->exec, pcmk_strerror(rc), rc);
services__handle_exec_error(op, rc);
if (op->synchronous) {
sigchld_cleanup(&data);
}
goto done;
break;
case 0: /* Child */
close(stdout_fd[0]);
close(stderr_fd[0]);
if (stdin_fd[1] >= 0) {
close(stdin_fd[1]);
}
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
crm_warn("Can't redirect output from '%s': %s "
CRM_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
crm_warn("Can't redirect error output from '%s': %s "
CRM_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stderr_fd[1]);
}
if ((stdin_fd[0] >= 0) &&
(STDIN_FILENO != stdin_fd[0])) {
if (dup2(stdin_fd[0], STDIN_FILENO) != STDIN_FILENO) {
crm_warn("Can't redirect input to '%s': %s "
CRM_XS " errno=%d",
op->opaque->exec, pcmk_rc_str(errno), errno);
}
close(stdin_fd[0]);
}
if (op->synchronous) {
sigchld_cleanup(&data);
}
action_launch_child(op);
CRM_ASSERT(0); /* action_launch_child is effectively noreturn */
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
if (stdin_fd[0] >= 0) {
close(stdin_fd[0]);
}
op->opaque->stdout_fd = stdout_fd[0];
rc = pcmk__set_nonblocking(op->opaque->stdout_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' output non-blocking: %s "
CRM_XS " rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
}
op->opaque->stderr_fd = stderr_fd[0];
rc = pcmk__set_nonblocking(op->opaque->stderr_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' error output non-blocking: %s "
CRM_XS " rc=%d",
op->opaque->exec, pcmk_rc_str(rc), rc);
}
op->opaque->stdin_fd = stdin_fd[1];
if (op->opaque->stdin_fd >= 0) {
// using buffer behind non-blocking-fd here - that could be improved
// as long as no other standard uses stdin_fd assume stonith
rc = pcmk__set_nonblocking(op->opaque->stdin_fd);
if (rc != pcmk_rc_ok) {
crm_info("Could not set '%s' input non-blocking: %s "
CRM_XS " fd=%d,rc=%d", op->opaque->exec,
pcmk_rc_str(rc), op->opaque->stdin_fd, rc);
}
pipe_in_action_stdin_parameters(op);
// as long as we are handling parameters directly in here just close
close(op->opaque->stdin_fd);
op->opaque->stdin_fd = -1;
}
// after fds are setup properly and before we plug anything into mainloop
if (op->opaque->fork_callback) {
op->opaque->fork_callback(op);
}
if (op->synchronous) {
wait_for_sync_result(op, &data);
sigchld_cleanup(&data);
goto done;
}
crm_trace("Waiting async for '%s'[%d]", op->opaque->exec, op->pid);
mainloop_child_add_with_flags(op->pid, op->timeout, op->id, op,
pcmk_is_set(op->flags, SVC_ACTION_LEAVE_GROUP)? mainloop_leave_pid_group : 0,
async_action_complete);
op->opaque->stdout_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stdout_fd, op,
&stdout_callbacks);
op->opaque->stderr_gsource = mainloop_add_fd(op->id,
G_PRIORITY_LOW,
op->opaque->stderr_fd, op,
&stderr_callbacks);
services_add_inflight_op(op);
return pcmk_rc_ok;
done:
if (op->synchronous) {
return (op->rc == PCMK_OCF_OK)? pcmk_rc_ok : pcmk_rc_error;
} else {
return services__finalize_async_op(op);
}
}
GList *
services_os_get_single_directory_list(const char *root, gboolean files, gboolean executable)
{
GList *list = NULL;
struct dirent **namelist;
int entries = 0, lpc = 0;
char buffer[PATH_MAX];
entries = scandir(root, &namelist, NULL, alphasort);
if (entries <= 0) {
return list;
}
for (lpc = 0; lpc < entries; lpc++) {
struct stat sb;
if ('.' == namelist[lpc]->d_name[0]) {
free(namelist[lpc]);
continue;
}
snprintf(buffer, sizeof(buffer), "%s/%s", root, namelist[lpc]->d_name);
if (stat(buffer, &sb)) {
continue;
}
if (S_ISDIR(sb.st_mode)) {
if (files) {
free(namelist[lpc]);
continue;
}
} else if (S_ISREG(sb.st_mode)) {
if (files == FALSE) {
free(namelist[lpc]);
continue;
} else if (executable
&& (sb.st_mode & S_IXUSR) == 0
&& (sb.st_mode & S_IXGRP) == 0 && (sb.st_mode & S_IXOTH) == 0) {
free(namelist[lpc]);
continue;
}
}
list = g_list_append(list, strdup(namelist[lpc]->d_name));
free(namelist[lpc]);
}
free(namelist);
return list;
}
GList *
services_os_get_directory_list(const char *root, gboolean files, gboolean executable)
{
GList *result = NULL;
char *dirs = strdup(root);
char *dir = NULL;
if (pcmk__str_empty(dirs)) {
free(dirs);
return result;
}
for (dir = strtok(dirs, ":"); dir != NULL; dir = strtok(NULL, ":")) {
GList *tmp = services_os_get_single_directory_list(dir, files, executable);
if (tmp) {
result = g_list_concat(result, tmp);
}
}
free(dirs);
return result;
}
diff --git a/lib/services/systemd.c b/lib/services/systemd.c
index bed1861077..9883e5b2ae 100644
--- a/lib/services/systemd.c
+++ b/lib/services/systemd.c
@@ -1,1088 +1,1094 @@
/*
* Copyright 2012-2022 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 <crm/crm.h>
#include <crm/services.h>
#include <crm/services_internal.h>
#include <crm/common/mainloop.h>
#include <sys/stat.h>
#include <gio/gio.h>
#include <services_private.h>
#include <systemd.h>
#include <dbus/dbus.h>
#include <pcmk-dbus.h>
static void invoke_unit_by_path(svc_action_t *op, const char *unit);
#define BUS_NAME "org.freedesktop.systemd1"
#define BUS_NAME_MANAGER BUS_NAME ".Manager"
#define BUS_NAME_UNIT BUS_NAME ".Unit"
#define BUS_PATH "/org/freedesktop/systemd1"
/*!
* \internal
* \brief Prepare a systemd action
*
* \param[in] op Action to prepare
*
* \return Standard Pacemaker return code
*/
int
services__systemd_prepare(svc_action_t *op)
{
op->opaque->exec = strdup("systemd-dbus");
if (op->opaque->exec == NULL) {
return ENOMEM;
}
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Map a systemd result to a standard OCF result
*
* \param[in] exit_status Systemd result
*
* \return Standard OCF result
*/
enum ocf_exitcode
services__systemd2ocf(int exit_status)
{
// This library uses OCF codes for systemd actions
return (enum ocf_exitcode) exit_status;
}
static inline DBusMessage *
systemd_new_method(const char *method)
{
crm_trace("Calling: %s on " BUS_NAME_MANAGER, method);
return dbus_message_new_method_call(BUS_NAME, BUS_PATH, BUS_NAME_MANAGER,
method);
}
/*
* Functions to manage a static DBus connection
*/
static DBusConnection* systemd_proxy = NULL;
static inline DBusPendingCall *
systemd_send(DBusMessage *msg,
void(*done)(DBusPendingCall *pending, void *user_data),
void *user_data, int timeout)
{
return pcmk_dbus_send(msg, systemd_proxy, done, user_data, timeout);
}
static inline DBusMessage *
systemd_send_recv(DBusMessage *msg, DBusError *error, int timeout)
{
return pcmk_dbus_send_recv(msg, systemd_proxy, error, timeout);
}
/*!
* \internal
* \brief Send a method to systemd without arguments, and wait for reply
*
* \param[in] method Method to send
*
* \return Systemd reply on success, NULL (and error will be logged) otherwise
*
* \note The caller must call dbus_message_unref() on the reply after
* handling it.
*/
static DBusMessage *
systemd_call_simple_method(const char *method)
{
DBusMessage *msg = systemd_new_method(method);
DBusMessage *reply = NULL;
DBusError error;
/* Don't call systemd_init() here, because that calls this */
CRM_CHECK(systemd_proxy, return NULL);
if (msg == NULL) {
crm_err("Could not create message to send %s to systemd", method);
return NULL;
}
dbus_error_init(&error);
reply = systemd_send_recv(msg, &error, DBUS_TIMEOUT_USE_DEFAULT);
dbus_message_unref(msg);
if (dbus_error_is_set(&error)) {
crm_err("Could not send %s to systemd: %s (%s)",
method, error.message, error.name);
dbus_error_free(&error);
return NULL;
} else if (reply == NULL) {
crm_err("Could not send %s to systemd: no reply received", method);
return NULL;
}
return reply;
}
static gboolean
systemd_init(void)
{
static int need_init = 1;
// https://dbus.freedesktop.org/doc/api/html/group__DBusConnection.html
if (systemd_proxy
&& dbus_connection_get_is_connected(systemd_proxy) == FALSE) {
crm_warn("Connection to System DBus is closed. Reconnecting...");
pcmk_dbus_disconnect(systemd_proxy);
systemd_proxy = NULL;
need_init = 1;
}
if (need_init) {
need_init = 0;
systemd_proxy = pcmk_dbus_connect();
}
if (systemd_proxy == NULL) {
return FALSE;
}
return TRUE;
}
static inline char *
systemd_get_property(const char *unit, const char *name,
void (*callback)(const char *name, const char *value, void *userdata),
void *userdata, DBusPendingCall **pending, int timeout)
{
return systemd_proxy?
pcmk_dbus_get_property(systemd_proxy, BUS_NAME, unit, BUS_NAME_UNIT,
name, callback, userdata, pending, timeout)
: NULL;
}
void
systemd_cleanup(void)
{
if (systemd_proxy) {
pcmk_dbus_disconnect(systemd_proxy);
systemd_proxy = NULL;
}
}
/*
* end of systemd_proxy functions
*/
/*!
* \internal
* \brief Check whether a file name represents a manageable systemd unit
*
* \param[in] name File name to check
*
* \return Pointer to "dot" before filename extension if so, NULL otherwise
*/
static const char *
systemd_unit_extension(const char *name)
{
if (name) {
const char *dot = strrchr(name, '.');
if (dot && (!strcmp(dot, ".service")
|| !strcmp(dot, ".socket")
|| !strcmp(dot, ".mount")
|| !strcmp(dot, ".timer")
|| !strcmp(dot, ".path"))) {
return dot;
}
}
return NULL;
}
static char *
systemd_service_name(const char *name, bool add_instance_name)
{
const char *dot = NULL;
if (pcmk__str_empty(name)) {
return NULL;
}
/* Services that end with an @ sign are systemd templates. They expect an
* instance name to follow the service name. If no instance name was
* provided, just add "pacemaker" to the string as the instance name. It
* doesn't seem to matter for purposes of looking up whether a service
* exists or not.
*
* A template can be specified either with or without the unit extension,
* so this block handles both cases.
*/
dot = systemd_unit_extension(name);
if (dot) {
if (dot != name && *(dot-1) == '@') {
char *s = NULL;
if (asprintf(&s, "%.*spacemaker%s", (int) (dot-name), name, dot) == -1) {
/* If asprintf fails, just return name. */
return strdup(name);
}
return s;
} else {
return strdup(name);
}
} else if (add_instance_name && *(name+strlen(name)-1) == '@') {
return crm_strdup_printf("%spacemaker.service", name);
} else {
return crm_strdup_printf("%s.service", name);
}
}
static void
systemd_daemon_reload_complete(DBusPendingCall *pending, void *user_data)
{
DBusError error;
DBusMessage *reply = NULL;
unsigned int reload_count = GPOINTER_TO_UINT(user_data);
dbus_error_init(&error);
if(pending) {
reply = dbus_pending_call_steal_reply(pending);
}
if (pcmk_dbus_find_error(pending, reply, &error)) {
crm_warn("Could not issue systemd reload %d: %s",
reload_count, error.message);
dbus_error_free(&error);
} else {
crm_trace("Reload %d complete", reload_count);
}
if(pending) {
dbus_pending_call_unref(pending);
}
if(reply) {
dbus_message_unref(reply);
}
}
static bool
systemd_daemon_reload(int timeout)
{
static unsigned int reload_count = 0;
DBusMessage *msg = systemd_new_method("Reload");
reload_count++;
CRM_ASSERT(msg != NULL);
systemd_send(msg, systemd_daemon_reload_complete,
GUINT_TO_POINTER(reload_count), timeout);
dbus_message_unref(msg);
return TRUE;
}
/*!
* \internal
* \brief Set an action result based on a method error
*
* \param[in] op Action to set result for
* \param[in] error Method error
*/
static void
set_result_from_method_error(svc_action_t *op, const DBusError *error)
{
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Unable to invoke systemd DBus method");
if (strstr(error->name, "org.freedesktop.systemd1.InvalidName")
|| strstr(error->name, "org.freedesktop.systemd1.LoadFailed")
|| strstr(error->name, "org.freedesktop.systemd1.NoSuchUnit")) {
if (pcmk__str_eq(op->action, "stop", pcmk__str_casei)) {
crm_trace("Masking systemd stop failure (%s) for %s "
"because unknown service can be considered stopped",
error->name, crm_str(op->rsc));
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
return;
}
- services__set_result(op, PCMK_OCF_NOT_INSTALLED,
- PCMK_EXEC_NOT_INSTALLED, "systemd unit not found");
+ services__format_result(op, PCMK_OCF_NOT_INSTALLED,
+ PCMK_EXEC_NOT_INSTALLED,
+ "systemd unit %s not found", op->agent);
}
crm_info("DBus request for %s of systemd unit %s for resource %s failed: %s",
op->action, op->agent, crm_str(op->rsc), error->message);
}
/*!
* \internal
* \brief Extract unit path from LoadUnit reply, and execute action
*
* \param[in] reply LoadUnit reply
* \param[in] op Action to execute (or NULL to just return path)
*
* \return DBus object path for specified unit if successful (only valid for
* lifetime of \p reply), otherwise NULL
*/
static const char *
execute_after_loadunit(DBusMessage *reply, svc_action_t *op)
{
const char *path = NULL;
DBusError error;
/* path here is not used other than as a non-NULL flag to indicate that a
* request was indeed sent
*/
if (pcmk_dbus_find_error((void *) &path, reply, &error)) {
if (op != NULL) {
set_result_from_method_error(op, &error);
}
dbus_error_free(&error);
} else if (!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH,
__func__, __LINE__)) {
if (op != NULL) {
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"systemd DBus method had unexpected reply");
crm_info("Could not load systemd unit %s for %s: "
"DBus reply has unexpected type", op->agent, op->id);
} else {
crm_info("Could not load systemd unit: "
"DBus reply has unexpected type");
}
} else {
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
}
if (op != NULL) {
if (path != NULL) {
invoke_unit_by_path(op, path);
} else if (!(op->synchronous)) {
- services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
- "No DBus object found for systemd unit");
+ services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
+ "No DBus object found for systemd unit %s",
+ op->agent);
services__finalize_async_op(op);
}
}
return path;
}
/*!
* \internal
* \brief Execute a systemd action after its LoadUnit completes
*
* \param[in] pending If not NULL, DBus call associated with LoadUnit request
* \param[in] user_data Action to execute
*/
static void
loadunit_completed(DBusPendingCall *pending, void *user_data)
{
DBusMessage *reply = NULL;
svc_action_t *op = user_data;
crm_trace("LoadUnit result for %s arrived", op->id);
// Grab the reply
if (pending != NULL) {
reply = dbus_pending_call_steal_reply(pending);
}
// The call is no longer pending
CRM_LOG_ASSERT(pending == op->opaque->pending);
services_set_op_pending(op, NULL);
// Execute the desired action based on the reply
execute_after_loadunit(reply, user_data);
if (reply != NULL) {
dbus_message_unref(reply);
}
}
/*!
* \internal
* \brief Execute a systemd action, given the unit name
*
* \param[in] arg_name Unit name (possibly shortened, i.e. without ".service")
* \param[in] op Action to execute (if NULL, just get the object path)
* \param[out] path If non-NULL and \p op is NULL or synchronous, where to
* store DBus object path for specified unit
*
* \return Standard Pacemaker return code (for NULL \p op, pcmk_rc_ok means unit
* was found; for synchronous actions, pcmk_rc_ok means unit was
* executed, with the actual result stored in \p op; for asynchronous
* actions, pcmk_rc_ok means action was initiated)
* \note It is the caller's responsibility to free the path.
*/
static int
invoke_unit_by_name(const char *arg_name, svc_action_t *op, char **path)
{
DBusMessage *msg;
DBusMessage *reply = NULL;
DBusPendingCall *pending = NULL;
char *name = NULL;
if (!systemd_init()) {
if (op != NULL) {
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"No DBus connection");
}
return ENOTCONN;
}
/* Create a LoadUnit DBus method (equivalent to GetUnit if already loaded),
* which makes the unit usable via further DBus methods.
*
* <method name="LoadUnit">
* <arg name="name" type="s" direction="in"/>
* <arg name="unit" type="o" direction="out"/>
* </method>
*/
msg = systemd_new_method("LoadUnit");
CRM_ASSERT(msg != NULL);
// Add the (expanded) unit name as the argument
name = systemd_service_name(arg_name, op == NULL || pcmk__str_eq(op->action, "meta-data", pcmk__str_none));
CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &name,
DBUS_TYPE_INVALID));
free(name);
if ((op == NULL) || op->synchronous) {
// For synchronous ops, wait for a reply and extract the result
const char *unit = NULL;
int rc = pcmk_rc_ok;
reply = systemd_send_recv(msg, NULL,
(op? op->timeout : DBUS_TIMEOUT_USE_DEFAULT));
dbus_message_unref(msg);
unit = execute_after_loadunit(reply, op);
if (unit == NULL) {
rc = ENOENT;
if (path != NULL) {
*path = NULL;
}
} else if (path != NULL) {
*path = strdup(unit);
if (*path == NULL) {
rc = ENOMEM;
}
}
if (reply != NULL) {
dbus_message_unref(reply);
}
return rc;
}
// For asynchronous ops, initiate the LoadUnit call and return
pending = systemd_send(msg, loadunit_completed, op, op->timeout);
if (pending == NULL) {
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Unable to send DBus message");
dbus_message_unref(msg);
return ECOMM;
}
// LoadUnit was successfully initiated
services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
services_set_op_pending(op, pending);
dbus_message_unref(msg);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Compare two strings alphabetically (case-insensitive)
*
* \param[in] a First string to compare
* \param[in] b Second string to compare
*
* \return 0 if strings are equal, -1 if a < b, 1 if a > b
*
* \note Usable as a GCompareFunc with g_list_sort().
* NULL is considered less than non-NULL.
*/
static gint
sort_str(gconstpointer a, gconstpointer b)
{
if (!a && !b) {
return 0;
} else if (!a) {
return -1;
} else if (!b) {
return 1;
}
return strcasecmp(a, b);
}
GList *
systemd_unit_listall(void)
{
int nfiles = 0;
GList *units = NULL;
DBusMessageIter args;
DBusMessageIter unit;
DBusMessageIter elem;
DBusMessage *reply = NULL;
if (systemd_init() == FALSE) {
return NULL;
}
/*
" <method name=\"ListUnitFiles\">\n" \
" <arg name=\"files\" type=\"a(ss)\" direction=\"out\"/>\n" \
" </method>\n" \
*/
reply = systemd_call_simple_method("ListUnitFiles");
if (reply == NULL) {
return NULL;
}
if (!dbus_message_iter_init(reply, &args)) {
crm_err("Could not list systemd unit files: systemd reply has no arguments");
dbus_message_unref(reply);
return NULL;
}
if (!pcmk_dbus_type_check(reply, &args, DBUS_TYPE_ARRAY,
__func__, __LINE__)) {
crm_err("Could not list systemd unit files: systemd reply has invalid arguments");
dbus_message_unref(reply);
return NULL;
}
dbus_message_iter_recurse(&args, &unit);
for (; dbus_message_iter_get_arg_type(&unit) != DBUS_TYPE_INVALID;
dbus_message_iter_next(&unit)) {
DBusBasicValue value;
const char *match = NULL;
char *unit_name = NULL;
char *basename = NULL;
if(!pcmk_dbus_type_check(reply, &unit, DBUS_TYPE_STRUCT, __func__, __LINE__)) {
crm_warn("Skipping systemd reply argument with unexpected type");
continue;
}
dbus_message_iter_recurse(&unit, &elem);
if(!pcmk_dbus_type_check(reply, &elem, DBUS_TYPE_STRING, __func__, __LINE__)) {
crm_warn("Skipping systemd reply argument with no string");
continue;
}
dbus_message_iter_get_basic(&elem, &value);
if (value.str == NULL) {
crm_debug("ListUnitFiles reply did not provide a string");
continue;
}
crm_trace("DBus ListUnitFiles listed: %s", value.str);
match = systemd_unit_extension(value.str);
if (match == NULL) {
// This is not a unit file type we know how to manage
crm_debug("ListUnitFiles entry '%s' is not supported as resource",
value.str);
continue;
}
// ListUnitFiles returns full path names, we just want base name
basename = strrchr(value.str, '/');
if (basename) {
basename = basename + 1;
} else {
basename = value.str;
}
if (!strcmp(match, ".service")) {
// Service is the "default" unit type, so strip it
unit_name = strndup(basename, match - basename);
} else {
unit_name = strdup(basename);
}
nfiles++;
units = g_list_prepend(units, unit_name);
}
dbus_message_unref(reply);
crm_trace("Found %d manageable systemd unit files", nfiles);
units = g_list_sort(units, sort_str);
return units;
}
gboolean
systemd_unit_exists(const char *name)
{
char *path = NULL;
char *state = NULL;
/* Note: Makes a blocking dbus calls
* Used by resources_find_service_class() when resource class=service
*/
if ((invoke_unit_by_name(name, NULL, &path) != pcmk_rc_ok)
|| (path == NULL)) {
return FALSE;
}
/* A successful LoadUnit is not sufficient to determine the unit's
* existence; it merely means the LoadUnit request received a reply.
* We must make another blocking call to check the LoadState property.
*/
state = systemd_get_property(path, "LoadState", NULL, NULL, NULL,
DBUS_TIMEOUT_USE_DEFAULT);
free(path);
if (pcmk__str_any_of(state, "loaded", "masked", NULL)) {
free(state);
return TRUE;
}
free(state);
return FALSE;
}
#define METADATA_FORMAT \
"<?xml version=\"1.0\"?>\n" \
"<!DOCTYPE resource-agent SYSTEM \"ra-api-1.dtd\">\n" \
"<resource-agent name=\"%s\" version=\"" PCMK_DEFAULT_AGENT_VERSION "\">\n" \
" <version>1.1</version>\n" \
" <longdesc lang=\"en\">\n" \
" %s\n" \
" </longdesc>\n" \
" <shortdesc lang=\"en\">systemd unit file for %s</shortdesc>\n" \
" <parameters/>\n" \
" <actions>\n" \
" <action name=\"start\" timeout=\"100\" />\n" \
" <action name=\"stop\" timeout=\"100\" />\n" \
" <action name=\"status\" timeout=\"100\" />\n" \
" <action name=\"monitor\" timeout=\"100\" interval=\"60\"/>\n" \
" <action name=\"meta-data\" timeout=\"5\" />\n" \
" </actions>\n" \
" <special tag=\"systemd\"/>\n" \
"</resource-agent>\n"
static char *
systemd_unit_metadata(const char *name, int timeout)
{
char *meta = NULL;
char *desc = NULL;
char *path = NULL;
char *escaped = NULL;
if (invoke_unit_by_name(name, NULL, &path) == pcmk_rc_ok) {
/* TODO: Worth a making blocking call for? Probably not. Possibly if cached. */
desc = systemd_get_property(path, "Description", NULL, NULL, NULL,
timeout);
} else {
desc = crm_strdup_printf("Systemd unit file for %s", name);
}
escaped = crm_xml_escape(desc);
meta = crm_strdup_printf(METADATA_FORMAT, name, escaped, name);
free(desc);
free(path);
free(escaped);
return meta;
}
/*!
* \internal
* \brief Determine result of method from reply
*
* \param[in] reply Reply to start, stop, or restart request
* \param[in] op Action that was executed
*/
static void
process_unit_method_reply(DBusMessage *reply, svc_action_t *op)
{
DBusError error;
/* The first use of error here is not used other than as a non-NULL flag to
* indicate that a request was indeed sent
*/
if (pcmk_dbus_find_error((void *) &error, reply, &error)) {
set_result_from_method_error(op, &error);
dbus_error_free(&error);
} else if (!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH,
__func__, __LINE__)) {
crm_info("DBus request for %s of %s succeeded but "
"return type was unexpected", op->action, crm_str(op->rsc));
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE,
"systemd DBus method had unexpected reply");
} else {
const char *path = NULL;
dbus_message_get_args(reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
crm_debug("DBus request for %s of %s using %s succeeded",
op->action, crm_str(op->rsc), path);
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
}
}
/*!
* \internal
* \brief Process the completion of an asynchronous unit start, stop, or restart
*
* \param[in] pending If not NULL, DBus call associated with request
* \param[in] user_data Action that was executed
*/
static void
unit_method_complete(DBusPendingCall *pending, void *user_data)
{
DBusMessage *reply = NULL;
svc_action_t *op = user_data;
crm_trace("Result for %s arrived", op->id);
// Grab the reply
if (pending != NULL) {
reply = dbus_pending_call_steal_reply(pending);
}
// The call is no longer pending
CRM_LOG_ASSERT(pending == op->opaque->pending);
services_set_op_pending(op, NULL);
// Determine result and finalize action
process_unit_method_reply(reply, op);
services__finalize_async_op(op);
if (reply != NULL) {
dbus_message_unref(reply);
}
}
#define SYSTEMD_OVERRIDE_ROOT "/run/systemd/system/"
/* When the cluster manages a systemd resource, we create a unit file override
* to order the service "before" pacemaker. The "before" relationship won't
* actually be used, since systemd won't ever start the resource -- we're
* interested in the reverse shutdown ordering it creates, to ensure that
* systemd doesn't stop the resource at shutdown while pacemaker is still
* running.
*
* @TODO Add start timeout
*/
#define SYSTEMD_OVERRIDE_TEMPLATE \
"[Unit]\n" \
"Description=Cluster Controlled %s\n" \
"Before=pacemaker.service pacemaker_remote.service\n" \
"\n" \
"[Service]\n" \
"Restart=no\n"
// Temporarily use rwxr-xr-x umask when opening a file for writing
static FILE *
create_world_readable(const char *filename)
{
mode_t orig_umask = umask(S_IWGRP | S_IWOTH);
FILE *fp = fopen(filename, "w");
umask(orig_umask);
return fp;
}
static void
create_override_dir(const char *agent)
{
char *override_dir = crm_strdup_printf(SYSTEMD_OVERRIDE_ROOT
"/%s.service.d", agent);
int rc = pcmk__build_path(override_dir, 0755);
if (rc != pcmk_rc_ok) {
crm_warn("Could not create systemd override directory %s: %s",
override_dir, pcmk_rc_str(rc));
}
free(override_dir);
}
static char *
get_override_filename(const char *agent)
{
return crm_strdup_printf(SYSTEMD_OVERRIDE_ROOT
"/%s.service.d/50-pacemaker.conf", agent);
}
static void
systemd_create_override(const char *agent, int timeout)
{
FILE *file_strm = NULL;
char *override_file = get_override_filename(agent);
create_override_dir(agent);
/* Ensure the override file is world-readable. This is not strictly
* necessary, but it avoids a systemd warning in the logs.
*/
file_strm = create_world_readable(override_file);
if (file_strm == NULL) {
crm_err("Cannot open systemd override file %s for writing",
override_file);
} else {
char *override = crm_strdup_printf(SYSTEMD_OVERRIDE_TEMPLATE, agent);
int rc = fprintf(file_strm, "%s\n", override);
free(override);
if (rc < 0) {
crm_perror(LOG_WARNING, "Cannot write to systemd override file %s",
override_file);
}
fflush(file_strm);
fclose(file_strm);
systemd_daemon_reload(timeout);
}
free(override_file);
}
static void
systemd_remove_override(const char *agent, int timeout)
{
char *override_file = get_override_filename(agent);
int rc = unlink(override_file);
if (rc < 0) {
// Stop may be called when already stopped, which is fine
crm_perror(LOG_DEBUG, "Cannot remove systemd override file %s",
override_file);
} else {
systemd_daemon_reload(timeout);
}
free(override_file);
}
/*!
* \internal
* \brief Parse result of systemd status check
*
* Set a status action's exit status and execution status based on a DBus
* property check result, and finalize the action if asynchronous.
*
* \param[in] name DBus interface name for property that was checked
* \param[in] state Property value
* \param[in] userdata Status action that check was done for
*/
static void
parse_status_result(const char *name, const char *state, void *userdata)
{
svc_action_t *op = userdata;
crm_trace("Resource %s has %s='%s'",
crm_str(op->rsc), name, crm_str(state));
if (pcmk__str_eq(state, "active", pcmk__str_none)) {
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
} else if (pcmk__str_eq(state, "reloading", pcmk__str_none)) {
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
} else if (pcmk__str_eq(state, "activating", pcmk__str_none)) {
services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
} else if (pcmk__str_eq(state, "deactivating", pcmk__str_none)) {
services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
} else {
services__set_result(op, PCMK_OCF_NOT_RUNNING, PCMK_EXEC_DONE, state);
}
if (!(op->synchronous)) {
services_set_op_pending(op, NULL);
services__finalize_async_op(op);
}
}
/*!
* \internal
* \brief Invoke a systemd unit, given its DBus object path
*
* \param[in] op Action to execute
* \param[in] unit DBus object path of systemd unit to invoke
*/
static void
invoke_unit_by_path(svc_action_t *op, const char *unit)
{
const char *method = NULL;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
if (pcmk__str_any_of(op->action, "monitor", "status", NULL)) {
DBusPendingCall *pending = NULL;
char *state;
state = systemd_get_property(unit, "ActiveState",
(op->synchronous? NULL : parse_status_result),
op, (op->synchronous? NULL : &pending),
op->timeout);
if (op->synchronous) {
parse_status_result("ActiveState", state, op);
free(state);
} else if (pending == NULL) { // Could not get ActiveState property
- services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
- "Could not get unit state from DBus");
+ services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
+ "Could not get state for unit %s from DBus",
+ op->agent);
services__finalize_async_op(op);
} else {
services_set_op_pending(op, pending);
}
return;
} else if (pcmk__str_eq(op->action, "start", pcmk__str_none)) {
method = "StartUnit";
systemd_create_override(op->agent, op->timeout);
} else if (pcmk__str_eq(op->action, "stop", pcmk__str_none)) {
method = "StopUnit";
systemd_remove_override(op->agent, op->timeout);
} else if (pcmk__str_eq(op->action, "restart", pcmk__str_none)) {
method = "RestartUnit";
} else {
- services__set_result(op, PCMK_OCF_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR,
- "Action not implemented for systemd resources");
+ services__format_result(op, PCMK_OCF_UNIMPLEMENT_FEATURE,
+ PCMK_EXEC_ERROR,
+ "Action %s not implemented "
+ "for systemd resources", crm_str(op->action));
if (!(op->synchronous)) {
services__finalize_async_op(op);
}
return;
}
crm_trace("Calling %s for unit path %s named %s",
method, unit, crm_str(op->rsc));
msg = systemd_new_method(method);
CRM_ASSERT(msg != NULL);
/* (ss) */
{
const char *replace_s = "replace";
char *name = systemd_service_name(op->agent, pcmk__str_eq(op->action, "meta-data", pcmk__str_none));
CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &name, DBUS_TYPE_INVALID));
CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_STRING, &replace_s, DBUS_TYPE_INVALID));
free(name);
}
if (op->synchronous) {
reply = systemd_send_recv(msg, NULL, op->timeout);
dbus_message_unref(msg);
process_unit_method_reply(reply, op);
if (reply != NULL) {
dbus_message_unref(reply);
}
} else {
DBusPendingCall *pending = systemd_send(msg, unit_method_complete, op,
op->timeout);
dbus_message_unref(msg);
if (pending == NULL) {
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Unable to send DBus message");
services__finalize_async_op(op);
} else {
services_set_op_pending(op, pending);
}
}
}
static gboolean
systemd_timeout_callback(gpointer p)
{
svc_action_t * op = p;
op->opaque->timerid = 0;
crm_info("%s action for systemd unit %s named '%s' timed out",
op->action, op->agent, op->rsc);
- services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT,
- "Systemd unit action did not complete in time");
+ services__format_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_TIMEOUT,
+ "%s action for systemd unit %s "
+ "did not complete in time", op->action, op->agent);
services__finalize_async_op(op);
return FALSE;
}
/*!
* \internal
* \brief Execute a systemd action
*
* \param[in] op Action to execute
*
* \return Standard Pacemaker return code
* \retval EBUSY Recurring operation could not be initiated
* \retval pcmk_rc_error Synchronous action failed
* \retval pcmk_rc_ok Synchronous action succeeded, or asynchronous action
* should not be freed (because it's pending or because
* it failed to execute and was already freed)
*
* \note If the return value for an asynchronous action is not pcmk_rc_ok, the
* caller is responsible for freeing the action.
*/
int
services__execute_systemd(svc_action_t *op)
{
CRM_ASSERT(op != NULL);
if ((op->action == NULL) || (op->agent == NULL)) {
services__set_result(op, PCMK_OCF_NOT_CONFIGURED, PCMK_EXEC_ERROR_FATAL,
"Bug in action caller");
goto done;
}
if (!systemd_init()) {
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"No DBus connection");
goto done;
}
crm_debug("Performing %ssynchronous %s op on systemd unit %s named '%s'",
(op->synchronous? "" : "a"), op->action, op->agent,
crm_str(op->rsc));
if (pcmk__str_eq(op->action, "meta-data", pcmk__str_casei)) {
op->stdout_data = systemd_unit_metadata(op->agent, op->timeout);
services__set_result(op, PCMK_OCF_OK, PCMK_EXEC_DONE, NULL);
goto done;
}
/* invoke_unit_by_name() should always override these values, which are here
* just as a fail-safe in case there are any code paths that neglect to
*/
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Bug in service library");
if (invoke_unit_by_name(op->agent, op, NULL) == pcmk_rc_ok) {
op->opaque->timerid = g_timeout_add(op->timeout + 5000,
systemd_timeout_callback, op);
services_add_inflight_op(op);
return pcmk_rc_ok;
}
done:
if (op->synchronous) {
return (op->rc == PCMK_OCF_OK)? pcmk_rc_ok : pcmk_rc_error;
} else {
return services__finalize_async_op(op);
}
}
diff --git a/tools/crm_resource_runtime.c b/tools/crm_resource_runtime.c
index b877beea69..8c61d3b1b1 100644
--- a/tools/crm_resource_runtime.c
+++ b/tools/crm_resource_runtime.c
@@ -1,2011 +1,2012 @@
/*
- * Copyright 2004-2021 the Pacemaker project contributors
+ * Copyright 2004-2022 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm_resource.h>
#include <crm/common/ipc_controld.h>
#include <crm/common/lists_internal.h>
#include <crm/services_internal.h>
resource_checks_t *
cli_check_resource(pe_resource_t *rsc, char *role_s, char *managed)
{
pe_resource_t *parent = uber_parent(rsc);
resource_checks_t *rc = calloc(1, sizeof(resource_checks_t));
if (role_s) {
enum rsc_role_e role = text2role(role_s);
if (role == RSC_ROLE_STOPPED) {
rc->flags |= rsc_remain_stopped;
} else if (pcmk_is_set(parent->flags, pe_rsc_promotable) &&
(role == RSC_ROLE_UNPROMOTED)) {
rc->flags |= rsc_unpromotable;
}
}
if (managed && !crm_is_true(managed)) {
rc->flags |= rsc_unmanaged;
}
if (rsc->lock_node) {
rc->lock_node = rsc->lock_node->details->uname;
}
rc->rsc = rsc;
return rc;
}
static GList *
build_node_info_list(pe_resource_t *rsc)
{
GList *retval = NULL;
for (GList *iter = rsc->children; iter != NULL; iter = iter->next) {
pe_resource_t *child = (pe_resource_t *) iter->data;
for (GList *iter2 = child->running_on; iter2 != NULL; iter2 = iter2->next) {
pe_node_t *node = (pe_node_t *) iter2->data;
node_info_t *ni = calloc(1, sizeof(node_info_t));
ni->node_name = node->details->uname;
ni->promoted = pcmk_is_set(rsc->flags, pe_rsc_promotable) &&
child->fns->state(child, TRUE) == RSC_ROLE_PROMOTED;
retval = g_list_prepend(retval, ni);
}
}
return retval;
}
GList *
cli_resource_search(pe_resource_t *rsc, const char *requested_name,
pe_working_set_t *data_set)
{
GList *retval = NULL;
pe_resource_t *parent = uber_parent(rsc);
if (pe_rsc_is_clone(rsc)) {
retval = build_node_info_list(rsc);
/* The anonymous clone children's common ID is supplied */
} else if (pe_rsc_is_clone(parent)
&& !pcmk_is_set(rsc->flags, pe_rsc_unique)
&& rsc->clone_name
&& pcmk__str_eq(requested_name, rsc->clone_name, pcmk__str_casei)
&& !pcmk__str_eq(requested_name, rsc->id, pcmk__str_casei)) {
retval = build_node_info_list(parent);
} else if (rsc->running_on != NULL) {
for (GList *iter = rsc->running_on; iter != NULL; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
node_info_t *ni = calloc(1, sizeof(node_info_t));
ni->node_name = node->details->uname;
ni->promoted = (rsc->fns->state(rsc, TRUE) == RSC_ROLE_PROMOTED);
retval = g_list_prepend(retval, ni);
}
}
return retval;
}
#define XPATH_MAX 1024
// \return Standard Pacemaker return code
static int
find_resource_attr(pcmk__output_t *out, cib_t * the_cib, const char *attr,
const char *rsc, const char *attr_set_type, const char *set_name,
const char *attr_id, const char *attr_name, char **value)
{
int offset = 0;
int rc = pcmk_rc_ok;
xmlNode *xml_search = NULL;
char *xpath_string = NULL;
const char *xpath_base = NULL;
if(value) {
*value = NULL;
}
if(the_cib == NULL) {
return ENOTCONN;
}
xpath_base = pcmk_cib_xpath_for(XML_CIB_TAG_RESOURCES);
if (xpath_base == NULL) {
crm_err(XML_CIB_TAG_RESOURCES " CIB element not known (bug?)");
return ENOMSG;
}
xpath_string = calloc(1, XPATH_MAX);
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "%s",
xpath_base);
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "//*[@id=\"%s\"]", rsc);
if (attr_set_type) {
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "/%s", attr_set_type);
if (set_name) {
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "[@id=\"%s\"]", set_name);
}
}
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "//nvpair[");
if (attr_id) {
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "@id=\"%s\"", attr_id);
}
if (attr_name) {
if (attr_id) {
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, " and ");
}
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "@name=\"%s\"", attr_name);
}
offset += snprintf(xpath_string + offset, XPATH_MAX - offset, "]");
CRM_LOG_ASSERT(offset > 0);
rc = the_cib->cmds->query(the_cib, xpath_string, &xml_search,
cib_sync_call | cib_scope_local | cib_xpath);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
goto done;
}
crm_log_xml_debug(xml_search, "Match");
if (xml_has_children(xml_search)) {
xmlNode *child = NULL;
rc = ENOTUNIQ;
out->info(out, "Multiple attributes match name=%s", attr_name);
for (child = pcmk__xml_first_child(xml_search); child != NULL;
child = pcmk__xml_next(child)) {
out->info(out, " Value: %s \t(id=%s)",
crm_element_value(child, XML_NVPAIR_ATTR_VALUE), ID(child));
}
out->spacer(out);
} else if(value) {
const char *tmp = crm_element_value(xml_search, attr);
if (tmp) {
*value = strdup(tmp);
}
}
done:
free(xpath_string);
free_xml(xml_search);
return rc;
}
/* PRIVATE. Use the find_matching_attr_resources instead. */
static void
find_matching_attr_resources_recursive(pcmk__output_t *out, GList/* <pe_resource_t*> */ ** result,
pe_resource_t * rsc, const char * rsc_id,
const char * attr_set, const char * attr_set_type,
const char * attr_id, const char * attr_name,
cib_t * cib, const char * cmd, int depth)
{
int rc = pcmk_rc_ok;
char *lookup_id = clone_strip(rsc->id);
char *local_attr_id = NULL;
/* visit the children */
for(GList *gIter = rsc->children; gIter; gIter = gIter->next) {
find_matching_attr_resources_recursive(out, result, (pe_resource_t*)gIter->data,
rsc_id, attr_set, attr_set_type,
attr_id, attr_name, cib, cmd, depth+1);
/* do it only once for clones */
if(pe_clone == rsc->variant) {
break;
}
}
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
/* Post-order traversal.
* The root is always on the list and it is the last item. */
if((0 == depth) || (pcmk_rc_ok == rc)) {
/* push the head */
*result = g_list_append(*result, rsc);
}
free(local_attr_id);
free(lookup_id);
}
/* The result is a linearized pre-ordered tree of resources. */
static GList/*<pe_resource_t*>*/ *
find_matching_attr_resources(pcmk__output_t *out, pe_resource_t * rsc,
const char * rsc_id, const char * attr_set,
const char * attr_set_type, const char * attr_id,
const char * attr_name, cib_t * cib, const char * cmd,
gboolean force)
{
int rc = pcmk_rc_ok;
char *lookup_id = NULL;
char *local_attr_id = NULL;
GList * result = NULL;
/* If --force is used, update only the requested resource (clone or primitive).
* Otherwise, if the primitive has the attribute, use that.
* Otherwise use the clone. */
if(force == TRUE) {
return g_list_append(result, rsc);
}
if(rsc->parent && pe_clone == rsc->parent->variant) {
int rc = pcmk_rc_ok;
char *local_attr_id = NULL;
rc = find_resource_attr(out, cib, XML_ATTR_ID, rsc_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
free(local_attr_id);
if(rc != pcmk_rc_ok) {
rsc = rsc->parent;
out->info(out, "Performing %s of '%s' on '%s', the parent of '%s'",
cmd, attr_name, rsc->id, rsc_id);
}
return g_list_append(result, rsc);
} else if(rsc->parent == NULL && rsc->children && pe_clone == rsc->variant) {
pe_resource_t *child = rsc->children->data;
if(child->variant == pe_native) {
lookup_id = clone_strip(child->id); /* Could be a cloned group! */
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
if(rc == pcmk_rc_ok) {
rsc = child;
out->info(out, "A value for '%s' already exists in child '%s', performing %s on that instead of '%s'",
attr_name, lookup_id, cmd, rsc_id);
}
free(local_attr_id);
free(lookup_id);
}
return g_list_append(result, rsc);
}
/* If the resource is a group ==> children inherit the attribute if defined. */
find_matching_attr_resources_recursive(out, &result, rsc, rsc_id, attr_set,
attr_set_type, attr_id, attr_name,
cib, cmd, 0);
return result;
}
// \return Standard Pacemaker return code
int
cli_resource_update_attribute(pe_resource_t *rsc, const char *requested_name,
const char *attr_set, const char *attr_set_type,
const char *attr_id, const char *attr_name,
const char *attr_value, gboolean recursive,
cib_t *cib, int cib_options,
pe_working_set_t *data_set, gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
static bool need_init = TRUE;
char *local_attr_id = NULL;
char *local_attr_set = NULL;
GList/*<pe_resource_t*>*/ *resources = NULL;
const char *common_attr_id = attr_id;
if (attr_id == NULL && force == FALSE) {
find_resource_attr (out, cib, XML_ATTR_ID, uber_parent(rsc)->id, NULL,
NULL, NULL, attr_name, NULL);
}
if (pcmk__str_eq(attr_set_type, XML_TAG_ATTR_SETS, pcmk__str_casei)) {
if (force == FALSE) {
rc = find_resource_attr(out, cib, XML_ATTR_ID, uber_parent(rsc)->id,
XML_TAG_META_SETS, attr_set, attr_id,
attr_name, &local_attr_id);
if (rc == pcmk_rc_ok && !out->is_quiet(out)) {
out->err(out, "WARNING: There is already a meta attribute for '%s' called '%s' (id=%s)",
uber_parent(rsc)->id, attr_name, local_attr_id);
out->err(out, " Delete '%s' first or use the force option to override",
local_attr_id);
}
free(local_attr_id);
if (rc == pcmk_rc_ok) {
return ENOTUNIQ;
}
}
resources = g_list_append(resources, rsc);
} else {
resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type,
attr_id, attr_name, cib, "update", force);
}
/* If either attr_set or attr_id is specified,
* one clearly intends to modify a single resource.
* It is the last item on the resource list.*/
for(GList *gIter = (attr_set||attr_id) ? g_list_last(resources) : resources
; gIter; gIter = gIter->next) {
char *lookup_id = NULL;
xmlNode *xml_top = NULL;
xmlNode *xml_obj = NULL;
local_attr_id = NULL;
local_attr_set = NULL;
rsc = (pe_resource_t*)gIter->data;
attr_id = common_attr_id;
lookup_id = clone_strip(rsc->id); /* Could be a cloned group! */
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
if (rc == pcmk_rc_ok) {
crm_debug("Found a match for name=%s: id=%s", attr_name, local_attr_id);
attr_id = local_attr_id;
} else if (rc != ENXIO) {
free(lookup_id);
free(local_attr_id);
g_list_free(resources);
return rc;
} else {
const char *tag = crm_element_name(rsc->xml);
if (attr_set == NULL) {
local_attr_set = crm_strdup_printf("%s-%s", lookup_id,
attr_set_type);
attr_set = local_attr_set;
}
if (attr_id == NULL) {
local_attr_id = crm_strdup_printf("%s-%s", attr_set, attr_name);
attr_id = local_attr_id;
}
xml_top = create_xml_node(NULL, tag);
crm_xml_add(xml_top, XML_ATTR_ID, lookup_id);
xml_obj = create_xml_node(xml_top, attr_set_type);
crm_xml_add(xml_obj, XML_ATTR_ID, attr_set);
}
xml_obj = crm_create_nvpair_xml(xml_obj, attr_id, attr_name, attr_value);
if (xml_top == NULL) {
xml_top = xml_obj;
}
crm_log_xml_debug(xml_top, "Update");
rc = cib->cmds->modify(cib, XML_CIB_TAG_RESOURCES, xml_top, cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Set '%s' option: id=%s%s%s%s%s value=%s", lookup_id, local_attr_id,
attr_set ? " set=" : "", attr_set ? attr_set : "",
attr_name ? " name=" : "", attr_name ? attr_name : "", attr_value);
}
free_xml(xml_top);
free(lookup_id);
free(local_attr_id);
free(local_attr_set);
if(recursive && pcmk__str_eq(attr_set_type, XML_TAG_META_SETS, pcmk__str_casei)) {
GList *lpc = NULL;
if(need_init) {
need_init = FALSE;
pcmk__unpack_constraints(data_set);
pe__clear_resource_flags_on_all(data_set, pe_rsc_allocating);
}
crm_debug("Looking for dependencies %p", rsc->rsc_cons_lhs);
pe__set_resource_flags(rsc, pe_rsc_allocating);
for (lpc = rsc->rsc_cons_lhs; lpc != NULL; lpc = lpc->next) {
pcmk__colocation_t *cons = (pcmk__colocation_t *) lpc->data;
crm_debug("Checking %s %d", cons->id, cons->score);
if ((cons->score > 0)
&& !pcmk_is_set(cons->dependent->flags, pe_rsc_allocating)) {
/* Don't get into colocation loops */
crm_debug("Setting %s=%s for dependent resource %s",
attr_name, attr_value, cons->dependent->id);
cli_resource_update_attribute(cons->dependent,
cons->dependent->id, NULL,
attr_set_type, NULL,
attr_name, attr_value,
recursive, cib, cib_options,
data_set, force);
}
}
}
}
g_list_free(resources);
return rc;
}
// \return Standard Pacemaker return code
int
cli_resource_delete_attribute(pe_resource_t *rsc, const char *requested_name,
const char *attr_set, const char *attr_set_type,
const char *attr_id, const char *attr_name,
cib_t *cib, int cib_options,
pe_working_set_t *data_set, gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
GList/*<pe_resource_t*>*/ *resources = NULL;
if (attr_id == NULL && force == FALSE) {
find_resource_attr(out, cib, XML_ATTR_ID, uber_parent(rsc)->id, NULL,
NULL, NULL, attr_name, NULL);
}
if(pcmk__str_eq(attr_set_type, XML_TAG_META_SETS, pcmk__str_casei)) {
resources = find_matching_attr_resources(out, rsc, requested_name, attr_set, attr_set_type,
attr_id, attr_name, cib, "delete", force);
} else {
resources = g_list_append(resources, rsc);
}
for(GList *gIter = resources; gIter; gIter = gIter->next) {
char *lookup_id = NULL;
xmlNode *xml_obj = NULL;
char *local_attr_id = NULL;
rsc = (pe_resource_t*)gIter->data;
lookup_id = clone_strip(rsc->id);
rc = find_resource_attr(out, cib, XML_ATTR_ID, lookup_id, attr_set_type,
attr_set, attr_id, attr_name, &local_attr_id);
if (rc == ENXIO) {
free(lookup_id);
rc = pcmk_rc_ok;
continue;
} else if (rc != pcmk_rc_ok) {
free(lookup_id);
g_list_free(resources);
return rc;
}
if (attr_id == NULL) {
attr_id = local_attr_id;
}
xml_obj = crm_create_nvpair_xml(NULL, attr_id, attr_name, NULL);
crm_log_xml_debug(xml_obj, "Delete");
CRM_ASSERT(cib);
rc = cib->cmds->remove(cib, XML_CIB_TAG_RESOURCES, xml_obj, cib_options);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
out->info(out, "Deleted '%s' option: id=%s%s%s%s%s", lookup_id, local_attr_id,
attr_set ? " set=" : "", attr_set ? attr_set : "",
attr_name ? " name=" : "", attr_name ? attr_name : "");
}
free(lookup_id);
free_xml(xml_obj);
free(local_attr_id);
}
g_list_free(resources);
return rc;
}
// \return Standard Pacemaker return code
static int
send_lrm_rsc_op(pcmk_ipc_api_t *controld_api, bool do_fail_resource,
const char *host_uname, const char *rsc_id, pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
const char *router_node = host_uname;
const char *rsc_api_id = NULL;
const char *rsc_long_id = NULL;
const char *rsc_class = NULL;
const char *rsc_provider = NULL;
const char *rsc_type = NULL;
bool cib_only = false;
pe_resource_t *rsc = pe_find_resource(data_set->resources, rsc_id);
if (rsc == NULL) {
out->err(out, "Resource %s not found", rsc_id);
return ENXIO;
} else if (rsc->variant != pe_native) {
out->err(out, "We can only process primitive resources, not %s", rsc_id);
return EINVAL;
}
rsc_class = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
rsc_provider = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER),
rsc_type = crm_element_value(rsc->xml, XML_ATTR_TYPE);
if ((rsc_class == NULL) || (rsc_type == NULL)) {
out->err(out, "Resource %s does not have a class and type", rsc_id);
return EINVAL;
}
{
pe_node_t *node = pe_find_node(data_set->nodes, host_uname);
if (node == NULL) {
out->err(out, "Node %s not found", host_uname);
return pcmk_rc_node_unknown;
}
if (!(node->details->online)) {
if (do_fail_resource) {
out->err(out, "Node %s is not online", host_uname);
return ENOTCONN;
} else {
cib_only = true;
}
}
if (!cib_only && pe__is_guest_or_remote_node(node)) {
node = pe__current_node(node->details->remote_rsc);
if (node == NULL) {
out->err(out, "No cluster connection to Pacemaker Remote node %s detected",
host_uname);
return ENOTCONN;
}
router_node = node->details->uname;
}
}
if (rsc->clone_name) {
rsc_api_id = rsc->clone_name;
rsc_long_id = rsc->id;
} else {
rsc_api_id = rsc->id;
}
if (do_fail_resource) {
return pcmk_controld_api_fail(controld_api, host_uname, router_node,
rsc_api_id, rsc_long_id,
rsc_class, rsc_provider, rsc_type);
} else {
return pcmk_controld_api_refresh(controld_api, host_uname, router_node,
rsc_api_id, rsc_long_id, rsc_class,
rsc_provider, rsc_type, cib_only);
}
}
/*!
* \internal
* \brief Get resource name as used in failure-related node attributes
*
* \param[in] rsc Resource to check
*
* \return Newly allocated string containing resource's fail name
* \note The caller is responsible for freeing the result.
*/
static inline char *
rsc_fail_name(pe_resource_t *rsc)
{
const char *name = (rsc->clone_name? rsc->clone_name : rsc->id);
return pcmk_is_set(rsc->flags, pe_rsc_unique)? strdup(name) : clone_strip(name);
}
// \return Standard Pacemaker return code
static int
clear_rsc_history(pcmk_ipc_api_t *controld_api, const char *host_uname,
const char *rsc_id, pe_working_set_t *data_set)
{
int rc = pcmk_rc_ok;
/* Erase the resource's entire LRM history in the CIB, even if we're only
* clearing a single operation's fail count. If we erased only entries for a
* single operation, we might wind up with a wrong idea of the current
* resource state, and we might not re-probe the resource.
*/
rc = send_lrm_rsc_op(controld_api, false, host_uname, rsc_id, data_set);
if (rc != pcmk_rc_ok) {
return rc;
}
crm_trace("Processing %d mainloop inputs",
pcmk_controld_api_replies_expected(controld_api));
while (g_main_context_iteration(NULL, FALSE)) {
crm_trace("Processed mainloop input, %d still remaining",
pcmk_controld_api_replies_expected(controld_api));
}
return rc;
}
// \return Standard Pacemaker return code
static int
clear_rsc_failures(pcmk__output_t *out, pcmk_ipc_api_t *controld_api,
const char *node_name, const char *rsc_id, const char *operation,
const char *interval_spec, pe_working_set_t *data_set)
{
int rc = pcmk_rc_ok;
const char *failed_value = NULL;
const char *failed_id = NULL;
const char *interval_ms_s = NULL;
GHashTable *rscs = NULL;
GHashTableIter iter;
/* Create a hash table to use as a set of resources to clean. This lets us
* clean each resource only once (per node) regardless of how many failed
* operations it has.
*/
rscs = pcmk__strkey_table(NULL, NULL);
// Normalize interval to milliseconds for comparison to history entry
if (operation) {
interval_ms_s = crm_strdup_printf("%u",
crm_parse_interval_spec(interval_spec));
}
for (xmlNode *xml_op = pcmk__xml_first_child(data_set->failed);
xml_op != NULL;
xml_op = pcmk__xml_next(xml_op)) {
failed_id = crm_element_value(xml_op, XML_LRM_ATTR_RSCID);
if (failed_id == NULL) {
// Malformed history entry, should never happen
continue;
}
// No resource specified means all resources match
if (rsc_id) {
pe_resource_t *fail_rsc = pe_find_resource_with_flags(data_set->resources,
failed_id,
pe_find_renamed|pe_find_anon);
if (!fail_rsc || !pcmk__str_eq(rsc_id, fail_rsc->id, pcmk__str_casei)) {
continue;
}
}
// Host name should always have been provided by this point
failed_value = crm_element_value(xml_op, XML_ATTR_UNAME);
if (!pcmk__str_eq(node_name, failed_value, pcmk__str_casei)) {
continue;
}
// No operation specified means all operations match
if (operation) {
failed_value = crm_element_value(xml_op, XML_LRM_ATTR_TASK);
if (!pcmk__str_eq(operation, failed_value, pcmk__str_casei)) {
continue;
}
// Interval (if operation was specified) defaults to 0 (not all)
failed_value = crm_element_value(xml_op, XML_LRM_ATTR_INTERVAL_MS);
if (!pcmk__str_eq(interval_ms_s, failed_value, pcmk__str_casei)) {
continue;
}
}
g_hash_table_add(rscs, (gpointer) failed_id);
}
g_hash_table_iter_init(&iter, rscs);
while (g_hash_table_iter_next(&iter, (gpointer *) &failed_id, NULL)) {
crm_debug("Erasing failures of %s on %s", failed_id, node_name);
rc = clear_rsc_history(controld_api, node_name, failed_id, data_set);
if (rc != pcmk_rc_ok) {
return rc;
}
}
g_hash_table_destroy(rscs);
return rc;
}
// \return Standard Pacemaker return code
static int
clear_rsc_fail_attrs(pe_resource_t *rsc, const char *operation,
const char *interval_spec, pe_node_t *node)
{
int rc = pcmk_rc_ok;
int attr_options = pcmk__node_attr_none;
char *rsc_name = rsc_fail_name(rsc);
if (pe__is_guest_or_remote_node(node)) {
attr_options |= pcmk__node_attr_remote;
}
rc = pcmk__node_attr_request_clear(NULL, node->details->uname, rsc_name,
operation, interval_spec, NULL,
attr_options);
free(rsc_name);
return rc;
}
// \return Standard Pacemaker return code
int
cli_resource_delete(pcmk_ipc_api_t *controld_api, const char *host_uname,
pe_resource_t *rsc, const char *operation,
const char *interval_spec, bool just_failures,
pe_working_set_t *data_set, gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
pe_node_t *node = NULL;
if (rsc == NULL) {
return ENXIO;
} else if (rsc->children) {
GList *lpc = NULL;
for (lpc = rsc->children; lpc != NULL; lpc = lpc->next) {
pe_resource_t *child = (pe_resource_t *) lpc->data;
rc = cli_resource_delete(controld_api, host_uname, child, operation,
interval_spec, just_failures, data_set,
force);
if (rc != pcmk_rc_ok) {
return rc;
}
}
return pcmk_rc_ok;
} else if (host_uname == NULL) {
GList *lpc = NULL;
GList *nodes = g_hash_table_get_values(rsc->known_on);
if(nodes == NULL && force) {
nodes = pcmk__copy_node_list(data_set->nodes, false);
} else if(nodes == NULL && rsc->exclusive_discover) {
GHashTableIter iter;
pe_node_t *node = NULL;
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void**)&node)) {
if(node->weight >= 0) {
nodes = g_list_prepend(nodes, node);
}
}
} else if(nodes == NULL) {
nodes = g_hash_table_get_values(rsc->allowed_nodes);
}
for (lpc = nodes; lpc != NULL; lpc = lpc->next) {
node = (pe_node_t *) lpc->data;
if (node->details->online) {
rc = cli_resource_delete(controld_api, node->details->uname,
rsc, operation, interval_spec,
just_failures, data_set, force);
}
if (rc != pcmk_rc_ok) {
g_list_free(nodes);
return rc;
}
}
g_list_free(nodes);
return pcmk_rc_ok;
}
node = pe_find_node(data_set->nodes, host_uname);
if (node == NULL) {
out->err(out, "Unable to clean up %s because node %s not found",
rsc->id, host_uname);
return ENODEV;
}
if (!node->details->rsc_discovery_enabled) {
out->err(out, "Unable to clean up %s because resource discovery disabled on %s",
rsc->id, host_uname);
return EOPNOTSUPP;
}
if (controld_api == NULL) {
out->err(out, "Dry run: skipping clean-up of %s on %s due to CIB_file",
rsc->id, host_uname);
return pcmk_rc_ok;
}
rc = clear_rsc_fail_attrs(rsc, operation, interval_spec, node);
if (rc != pcmk_rc_ok) {
out->err(out, "Unable to clean up %s failures on %s: %s",
rsc->id, host_uname, pcmk_rc_str(rc));
return rc;
}
if (just_failures) {
rc = clear_rsc_failures(out, controld_api, host_uname, rsc->id, operation,
interval_spec, data_set);
} else {
rc = clear_rsc_history(controld_api, host_uname, rsc->id, data_set);
}
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned %s failures on %s, but unable to clean history: %s",
rsc->id, host_uname, pcmk_strerror(rc));
} else {
out->info(out, "Cleaned up %s on %s", rsc->id, host_uname);
}
return rc;
}
// \return Standard Pacemaker return code
int
cli_cleanup_all(pcmk_ipc_api_t *controld_api, const char *node_name,
const char *operation, const char *interval_spec,
pe_working_set_t *data_set)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
int attr_options = pcmk__node_attr_none;
const char *display_name = node_name? node_name : "all nodes";
if (controld_api == NULL) {
out->info(out, "Dry run: skipping clean-up of %s due to CIB_file",
display_name);
return rc;
}
if (node_name) {
pe_node_t *node = pe_find_node(data_set->nodes, node_name);
if (node == NULL) {
out->err(out, "Unknown node: %s", node_name);
return ENXIO;
}
if (pe__is_guest_or_remote_node(node)) {
attr_options |= pcmk__node_attr_remote;
}
}
rc = pcmk__node_attr_request_clear(NULL, node_name, NULL, operation,
interval_spec, NULL, attr_options);
if (rc != pcmk_rc_ok) {
out->err(out, "Unable to clean up all failures on %s: %s",
display_name, pcmk_rc_str(rc));
return rc;
}
if (node_name) {
rc = clear_rsc_failures(out, controld_api, node_name, NULL,
operation, interval_spec, data_set);
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned all resource failures on %s, but unable to clean history: %s",
node_name, pcmk_strerror(rc));
return rc;
}
} else {
for (GList *iter = data_set->nodes; iter; iter = iter->next) {
pe_node_t *node = (pe_node_t *) iter->data;
rc = clear_rsc_failures(out, controld_api, node->details->uname, NULL,
operation, interval_spec, data_set);
if (rc != pcmk_rc_ok) {
out->err(out, "Cleaned all resource failures on all nodes, but unable to clean history: %s",
pcmk_strerror(rc));
return rc;
}
}
}
out->info(out, "Cleaned up all resources on %s", display_name);
return rc;
}
int
cli_resource_check(pcmk__output_t *out, cib_t * cib_conn, pe_resource_t *rsc)
{
char *role_s = NULL;
char *managed = NULL;
pe_resource_t *parent = uber_parent(rsc);
int rc = pcmk_rc_no_output;
resource_checks_t *checks = NULL;
find_resource_attr(out, cib_conn, XML_NVPAIR_ATTR_VALUE, parent->id,
NULL, NULL, NULL, XML_RSC_ATTR_MANAGED, &managed);
find_resource_attr(out, cib_conn, XML_NVPAIR_ATTR_VALUE, parent->id,
NULL, NULL, NULL, XML_RSC_ATTR_TARGET_ROLE, &role_s);
checks = cli_check_resource(rsc, role_s, managed);
if (checks->flags != 0 || checks->lock_node != NULL) {
rc = out->message(out, "resource-check-list", checks);
}
free(role_s);
free(managed);
free(checks);
return rc;
}
// \return Standard Pacemaker return code
int
cli_resource_fail(pcmk_ipc_api_t *controld_api, const char *host_uname,
const char *rsc_id, pe_working_set_t *data_set)
{
crm_notice("Failing %s on %s", rsc_id, host_uname);
return send_lrm_rsc_op(controld_api, true, host_uname, rsc_id, data_set);
}
static GHashTable *
generate_resource_params(pe_resource_t *rsc, pe_node_t *node,
pe_working_set_t *data_set)
{
GHashTable *params = NULL;
GHashTable *meta = NULL;
GHashTable *combined = NULL;
GHashTableIter iter;
char *key = NULL;
char *value = NULL;
combined = pcmk__strkey_table(free, free);
params = pe_rsc_params(rsc, node, data_set);
if (params != NULL) {
g_hash_table_iter_init(&iter, params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
g_hash_table_insert(combined, strdup(key), strdup(value));
}
}
meta = pcmk__strkey_table(free, free);
get_meta_attributes(meta, rsc, node, data_set);
if (meta != NULL) {
g_hash_table_iter_init(&iter, meta);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
char *crm_name = crm_meta_name(key);
g_hash_table_insert(combined, crm_name, strdup(value));
}
g_hash_table_destroy(meta);
}
return combined;
}
bool resource_is_running_on(pe_resource_t *rsc, const char *host)
{
bool found = TRUE;
GList *hIter = NULL;
GList *hosts = NULL;
if(rsc == NULL) {
return FALSE;
}
rsc->fns->location(rsc, &hosts, TRUE);
for (hIter = hosts; host != NULL && hIter != NULL; hIter = hIter->next) {
pe_node_t *node = (pe_node_t *) hIter->data;
if(strcmp(host, node->details->uname) == 0) {
crm_trace("Resource %s is running on %s\n", rsc->id, host);
goto done;
} else if(strcmp(host, node->details->id) == 0) {
crm_trace("Resource %s is running on %s\n", rsc->id, host);
goto done;
}
}
if(host != NULL) {
crm_trace("Resource %s is not running on: %s\n", rsc->id, host);
found = FALSE;
} else if(host == NULL && hosts == NULL) {
crm_trace("Resource %s is not running\n", rsc->id);
found = FALSE;
}
done:
g_list_free(hosts);
return found;
}
/*!
* \internal
* \brief Create a list of all resources active on host from a given list
*
* \param[in] host Name of host to check whether resources are active
* \param[in] rsc_list List of resources to check
*
* \return New list of resources from list that are active on host
*/
static GList *
get_active_resources(const char *host, GList *rsc_list)
{
GList *rIter = NULL;
GList *active = NULL;
for (rIter = rsc_list; rIter != NULL; rIter = rIter->next) {
pe_resource_t *rsc = (pe_resource_t *) rIter->data;
/* Expand groups to their members, because if we're restarting a member
* other than the first, we can't otherwise tell which resources are
* stopping and starting.
*/
if (rsc->variant == pe_group) {
active = g_list_concat(active,
get_active_resources(host, rsc->children));
} else if (resource_is_running_on(rsc, host)) {
active = g_list_append(active, strdup(rsc->id));
}
}
return active;
}
static void dump_list(GList *items, const char *tag)
{
int lpc = 0;
GList *item = NULL;
for (item = items; item != NULL; item = item->next) {
crm_trace("%s[%d]: %s", tag, lpc, (char*)item->data);
lpc++;
}
}
static void display_list(pcmk__output_t *out, GList *items, const char *tag)
{
GList *item = NULL;
for (item = items; item != NULL; item = item->next) {
out->info(out, "%s%s", tag, (const char *)item->data);
}
}
/*!
* \internal
* \brief Upgrade XML to latest schema version and use it as working set input
*
* This also updates the working set timestamp to the current time.
*
* \param[in] data_set Working set instance to update
* \param[in] xml XML to use as input
*
* \return Standard Pacemaker return code
* \note On success, caller is responsible for freeing memory allocated for
* data_set->now.
* \todo This follows the example of other callers of cli_config_update()
* and returns ENOKEY ("Required key not available") if that fails,
* but perhaps pcmk_rc_schema_validation would be better in that case.
*/
int
update_working_set_xml(pe_working_set_t *data_set, xmlNode **xml)
{
if (cli_config_update(xml, NULL, FALSE) == FALSE) {
return ENOKEY;
}
data_set->input = *xml;
data_set->now = crm_time_new(NULL);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Update a working set's XML input based on a CIB query
*
* \param[in] data_set Data set instance to initialize
* \param[in] cib Connection to the CIB manager
*
* \return Standard Pacemaker return code
* \note On success, caller is responsible for freeing memory allocated for
* data_set->input and data_set->now.
*/
static int
update_working_set_from_cib(pcmk__output_t *out, pe_working_set_t * data_set,
cib_t *cib)
{
xmlNode *cib_xml_copy = NULL;
int rc = pcmk_rc_ok;
rc = cib->cmds->query(cib, NULL, &cib_xml_copy, cib_scope_local | cib_sync_call);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not obtain the current CIB: %s (%d)", pcmk_strerror(rc), rc);
return rc;
}
rc = update_working_set_xml(data_set, &cib_xml_copy);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not upgrade the current CIB XML");
free_xml(cib_xml_copy);
return rc;
}
return rc;
}
// \return Standard Pacemaker return code
static int
update_dataset(cib_t *cib, pe_working_set_t * data_set, bool simulate)
{
char *pid = NULL;
char *shadow_file = NULL;
cib_t *shadow_cib = NULL;
int rc = pcmk_rc_ok;
pcmk__output_t *out = data_set->priv;
pe_reset_working_set(data_set);
rc = update_working_set_from_cib(out, data_set, cib);
if (rc != pcmk_rc_ok) {
return rc;
}
if(simulate) {
bool prev_quiet = false;
pid = pcmk__getpid_s();
shadow_cib = cib_shadow_new(pid);
shadow_file = get_shadow_file(pid);
if (shadow_cib == NULL) {
out->err(out, "Could not create shadow cib: '%s'", pid);
rc = ENXIO;
goto done;
}
rc = write_xml_file(data_set->input, shadow_file, FALSE);
if (rc < 0) {
out->err(out, "Could not populate shadow cib: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
rc = shadow_cib->cmds->signon(shadow_cib, crm_system_name, cib_command);
rc = pcmk_legacy2rc(rc);
if (rc != pcmk_rc_ok) {
out->err(out, "Could not connect to shadow cib: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
pcmk__schedule_actions(data_set, data_set->input, NULL);
prev_quiet = out->is_quiet(out);
out->quiet = true;
pcmk__simulate_transition(data_set, shadow_cib, NULL);
out->quiet = prev_quiet;
rc = update_dataset(shadow_cib, data_set, FALSE);
} else {
cluster_status(data_set);
}
done:
/* Do not free data_set->input here, we need rsc->xml to be valid later on */
cib_delete(shadow_cib);
free(pid);
if(shadow_file) {
unlink(shadow_file);
free(shadow_file);
}
return rc;
}
static int
max_delay_for_resource(pe_working_set_t * data_set, pe_resource_t *rsc)
{
int delay = 0;
int max_delay = 0;
if(rsc && rsc->children) {
GList *iter = NULL;
for(iter = rsc->children; iter; iter = iter->next) {
pe_resource_t *child = (pe_resource_t *)iter->data;
delay = max_delay_for_resource(data_set, child);
if(delay > max_delay) {
double seconds = delay / 1000.0;
crm_trace("Calculated new delay of %.1fs due to %s", seconds, child->id);
max_delay = delay;
}
}
} else if(rsc) {
char *key = crm_strdup_printf("%s_%s_0", rsc->id, RSC_STOP);
pe_action_t *stop = custom_action(rsc, key, RSC_STOP, NULL, TRUE, FALSE, data_set);
const char *value = g_hash_table_lookup(stop->meta, XML_ATTR_TIMEOUT);
long long result_ll;
if ((pcmk__scan_ll(value, &result_ll, -1LL) == pcmk_rc_ok)
&& (result_ll >= 0) && (result_ll <= INT_MAX)) {
max_delay = (int) result_ll;
} else {
max_delay = -1;
}
pe_free_action(stop);
}
return max_delay;
}
static int
max_delay_in(pe_working_set_t * data_set, GList *resources)
{
int max_delay = 0;
GList *item = NULL;
for (item = resources; item != NULL; item = item->next) {
int delay = 0;
pe_resource_t *rsc = pe_find_resource(data_set->resources, (const char *)item->data);
delay = max_delay_for_resource(data_set, rsc);
if(delay > max_delay) {
double seconds = delay / 1000.0;
crm_trace("Calculated new delay of %.1fs due to %s", seconds, rsc->id);
max_delay = delay;
}
}
return 5 + (max_delay / 1000);
}
#define waiting_for_starts(d, r, h) ((d != NULL) || \
(!resource_is_running_on((r), (h))))
/*!
* \internal
* \brief Restart a resource (on a particular host if requested).
*
* \param[in] rsc The resource to restart
* \param[in] host The host to restart the resource on (or NULL for all)
* \param[in] timeout_ms Consider failed if actions do not complete in this time
* (specified in milliseconds, but a two-second
* granularity is actually used; if 0, a timeout will be
* calculated based on the resource timeout)
* \param[in] cib Connection to the CIB manager
*
* \return Standard Pacemaker return code (exits on certain failures)
*/
int
cli_resource_restart(pcmk__output_t *out, pe_resource_t *rsc, const char *host,
const char *move_lifetime, int timeout_ms, cib_t *cib,
int cib_options, gboolean promoted_role_only, gboolean force)
{
int rc = pcmk_rc_ok;
int lpc = 0;
int before = 0;
int step_timeout_s = 0;
int sleep_interval = 2;
int timeout = timeout_ms / 1000;
bool stop_via_ban = FALSE;
char *rsc_id = NULL;
char *orig_target_role = NULL;
GList *list_delta = NULL;
GList *target_active = NULL;
GList *current_active = NULL;
GList *restart_target_active = NULL;
pe_working_set_t *data_set = NULL;
if (!resource_is_running_on(rsc, host)) {
const char *id = rsc->clone_name?rsc->clone_name:rsc->id;
if(host) {
out->err(out, "%s is not running on %s and so cannot be restarted", id, host);
} else {
out->err(out, "%s is not running anywhere and so cannot be restarted", id);
}
return ENXIO;
}
rsc_id = strdup(rsc->id);
if ((pe_rsc_is_clone(rsc) || pe_bundle_replicas(rsc)) && host) {
stop_via_ban = TRUE;
}
/*
grab full cib
determine originally active resources
disable or ban
poll cib and watch for affected resources to get stopped
without --timeout, calculate the stop timeout for each step and wait for that
if we hit --timeout or the service timeout, re-enable or un-ban, report failure and indicate which resources we couldn't take down
if everything stopped, re-enable or un-ban
poll cib and watch for affected resources to get started
without --timeout, calculate the start timeout for each step and wait for that
if we hit --timeout or the service timeout, report (different) failure and indicate which resources we couldn't bring back up
report success
Optimizations:
- use constraints to determine ordered list of affected resources
- Allow a --no-deps option (aka. --force-restart)
*/
data_set = pe_new_working_set();
if (data_set == NULL) {
crm_perror(LOG_ERR, "Could not allocate working set");
rc = ENOMEM;
goto done;
}
data_set->priv = out;
pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat);
rc = update_dataset(cib, data_set, FALSE);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not get new resource list: %s (%d)", pcmk_strerror(rc), rc);
goto done;
}
restart_target_active = get_active_resources(host, data_set->resources);
current_active = get_active_resources(host, data_set->resources);
dump_list(current_active, "Origin");
if (stop_via_ban) {
/* Stop the clone or bundle instance by banning it from the host */
out->quiet = true;
rc = cli_resource_ban(out, rsc_id, host, move_lifetime, NULL, cib,
cib_options, promoted_role_only);
} else {
/* Stop the resource by setting target-role to Stopped.
* Remember any existing target-role so we can restore it later
* (though it only makes any difference if it's Unpromoted).
*/
char *lookup_id = clone_strip(rsc->id);
find_resource_attr(out, cib, XML_NVPAIR_ATTR_VALUE, lookup_id, NULL, NULL,
NULL, XML_RSC_ATTR_TARGET_ROLE, &orig_target_role);
free(lookup_id);
rc = cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE,
RSC_STOPPED, FALSE, cib, cib_options,
data_set, force);
}
if(rc != pcmk_rc_ok) {
out->err(out, "Could not set target-role for %s: %s (%d)", rsc_id, pcmk_strerror(rc), rc);
if (current_active) {
g_list_free_full(current_active, free);
}
if (restart_target_active) {
g_list_free_full(restart_target_active, free);
}
goto done;
}
rc = update_dataset(cib, data_set, TRUE);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources would be stopped");
goto failure;
}
target_active = get_active_resources(host, data_set->resources);
dump_list(target_active, "Target");
list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp);
out->info(out, "Waiting for %d resources to stop:", g_list_length(list_delta));
display_list(out, list_delta, " * ");
step_timeout_s = timeout / sleep_interval;
while (list_delta != NULL) {
before = g_list_length(list_delta);
if(timeout_ms == 0) {
step_timeout_s = max_delay_in(data_set, list_delta) / sleep_interval;
}
/* We probably don't need the entire step timeout */
for(lpc = 0; (lpc < step_timeout_s) && (list_delta != NULL); lpc++) {
sleep(sleep_interval);
if(timeout) {
timeout -= sleep_interval;
crm_trace("%ds remaining", timeout);
}
rc = update_dataset(cib, data_set, FALSE);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources were stopped");
goto failure;
}
if (current_active) {
g_list_free_full(current_active, free);
}
current_active = get_active_resources(host, data_set->resources);
g_list_free(list_delta);
list_delta = pcmk__subtract_lists(current_active, target_active, (GCompareFunc) strcmp);
dump_list(current_active, "Current");
dump_list(list_delta, "Delta");
}
crm_trace("%d (was %d) resources remaining", g_list_length(list_delta), before);
if(before == g_list_length(list_delta)) {
/* aborted during stop phase, print the contents of list_delta */
out->err(out, "Could not complete shutdown of %s, %d resources remaining", rsc_id, g_list_length(list_delta));
display_list(out, list_delta, " * ");
rc = ETIME;
goto failure;
}
}
if (stop_via_ban) {
rc = cli_resource_clear(rsc_id, host, NULL, cib, cib_options, TRUE, force);
} else if (orig_target_role) {
rc = cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE,
orig_target_role, FALSE, cib,
cib_options, data_set, force);
free(orig_target_role);
orig_target_role = NULL;
} else {
rc = cli_resource_delete_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS,
NULL, XML_RSC_ATTR_TARGET_ROLE, cib,
cib_options, data_set, force);
}
if(rc != pcmk_rc_ok) {
out->err(out, "Could not unset target-role for %s: %s (%d)", rsc_id, pcmk_strerror(rc), rc);
goto done;
}
if (target_active) {
g_list_free_full(target_active, free);
}
target_active = restart_target_active;
list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp);
out->info(out, "Waiting for %d resources to start again:", g_list_length(list_delta));
display_list(out, list_delta, " * ");
step_timeout_s = timeout / sleep_interval;
while (waiting_for_starts(list_delta, rsc, host)) {
before = g_list_length(list_delta);
if(timeout_ms == 0) {
step_timeout_s = max_delay_in(data_set, list_delta) / sleep_interval;
}
/* We probably don't need the entire step timeout */
for (lpc = 0; (lpc < step_timeout_s) && waiting_for_starts(list_delta, rsc, host); lpc++) {
sleep(sleep_interval);
if(timeout) {
timeout -= sleep_interval;
crm_trace("%ds remaining", timeout);
}
rc = update_dataset(cib, data_set, FALSE);
if(rc != pcmk_rc_ok) {
out->err(out, "Could not determine which resources were started");
goto failure;
}
if (current_active) {
g_list_free_full(current_active, free);
}
/* It's OK if dependent resources moved to a different node,
* so we check active resources on all nodes.
*/
current_active = get_active_resources(NULL, data_set->resources);
g_list_free(list_delta);
list_delta = pcmk__subtract_lists(target_active, current_active, (GCompareFunc) strcmp);
dump_list(current_active, "Current");
dump_list(list_delta, "Delta");
}
if(before == g_list_length(list_delta)) {
/* aborted during start phase, print the contents of list_delta */
out->err(out, "Could not complete restart of %s, %d resources remaining", rsc_id, g_list_length(list_delta));
display_list(out, list_delta, " * ");
rc = ETIME;
goto failure;
}
}
rc = pcmk_rc_ok;
goto done;
failure:
if (stop_via_ban) {
cli_resource_clear(rsc_id, host, NULL, cib, cib_options, TRUE, force);
} else if (orig_target_role) {
cli_resource_update_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS, NULL,
XML_RSC_ATTR_TARGET_ROLE, orig_target_role,
FALSE, cib, cib_options, data_set, force);
free(orig_target_role);
} else {
cli_resource_delete_attribute(rsc, rsc_id, NULL, XML_TAG_META_SETS, NULL,
XML_RSC_ATTR_TARGET_ROLE, cib, cib_options,
data_set, force);
}
done:
if (list_delta) {
g_list_free(list_delta);
}
if (current_active) {
g_list_free_full(current_active, free);
}
if (target_active && (target_active != restart_target_active)) {
g_list_free_full(target_active, free);
}
if (restart_target_active) {
g_list_free_full(restart_target_active, free);
}
free(rsc_id);
pe_free_working_set(data_set);
return rc;
}
static inline bool action_is_pending(pe_action_t *action)
{
if (pcmk_any_flags_set(action->flags, pe_action_optional|pe_action_pseudo)
|| !pcmk_is_set(action->flags, pe_action_runnable)
|| pcmk__str_eq("notify", action->task, pcmk__str_casei)) {
return false;
}
return true;
}
/*!
* \internal
* \brief Return TRUE if any actions in a list are pending
*
* \param[in] actions List of actions to check
*
* \return TRUE if any actions in the list are pending, FALSE otherwise
*/
static bool
actions_are_pending(GList *actions)
{
GList *action;
for (action = actions; action != NULL; action = action->next) {
pe_action_t *a = (pe_action_t *)action->data;
if (action_is_pending(a)) {
crm_notice("Waiting for %s (flags=%#.8x)", a->uuid, a->flags);
return TRUE;
}
}
return FALSE;
}
static void
print_pending_actions(pcmk__output_t *out, GList *actions)
{
GList *action;
out->info(out, "Pending actions:");
for (action = actions; action != NULL; action = action->next) {
pe_action_t *a = (pe_action_t *) action->data;
if (!action_is_pending(a)) {
continue;
}
if (a->node) {
out->info(out, "\tAction %d: %s\ton %s", a->id, a->uuid, a->node->details->uname);
} else {
out->info(out, "\tAction %d: %s", a->id, a->uuid);
}
}
}
/* For --wait, timeout (in seconds) to use if caller doesn't specify one */
#define WAIT_DEFAULT_TIMEOUT_S (60 * 60)
/* For --wait, how long to sleep between cluster state checks */
#define WAIT_SLEEP_S (2)
/*!
* \internal
* \brief Wait until all pending cluster actions are complete
*
* This waits until either the CIB's transition graph is idle or a timeout is
* reached.
*
* \param[in] timeout_ms Consider failed if actions do not complete in this time
* (specified in milliseconds, but one-second granularity
* is actually used; if 0, a default will be used)
* \param[in] cib Connection to the CIB manager
*
* \return Standard Pacemaker return code
*/
int
wait_till_stable(pcmk__output_t *out, int timeout_ms, cib_t * cib)
{
pe_working_set_t *data_set = NULL;
int rc = pcmk_rc_ok;
int timeout_s = timeout_ms? ((timeout_ms + 999) / 1000) : WAIT_DEFAULT_TIMEOUT_S;
time_t expire_time = time(NULL) + timeout_s;
time_t time_diff;
bool printed_version_warning = out->is_quiet(out); // i.e. don't print if quiet
data_set = pe_new_working_set();
if (data_set == NULL) {
return ENOMEM;
}
pe__set_working_set_flags(data_set, pe_flag_no_counts|pe_flag_no_compat);
do {
/* Abort if timeout is reached */
time_diff = expire_time - time(NULL);
if (time_diff > 0) {
crm_info("Waiting up to %ld seconds for cluster actions to complete", time_diff);
} else {
print_pending_actions(out, data_set->actions);
pe_free_working_set(data_set);
return ETIME;
}
if (rc == pcmk_rc_ok) { /* this avoids sleep on first loop iteration */
sleep(WAIT_SLEEP_S);
}
/* Get latest transition graph */
pe_reset_working_set(data_set);
rc = update_working_set_from_cib(out, data_set, cib);
if (rc != pcmk_rc_ok) {
pe_free_working_set(data_set);
return rc;
}
pcmk__schedule_actions(data_set, data_set->input, NULL);
if (!printed_version_warning) {
/* If the DC has a different version than the local node, the two
* could come to different conclusions about what actions need to be
* done. Warn the user in this case.
*
* @TODO A possible long-term solution would be to reimplement the
* wait as a new controller operation that would be forwarded to the
* DC. However, that would have potential problems of its own.
*/
const char *dc_version = g_hash_table_lookup(data_set->config_hash,
"dc-version");
if (!pcmk__str_eq(dc_version, PACEMAKER_VERSION "-" BUILD_VERSION, pcmk__str_casei)) {
out->info(out, "warning: wait option may not work properly in "
"mixed-version cluster");
printed_version_warning = TRUE;
}
}
} while (actions_are_pending(data_set->actions));
pe_free_working_set(data_set);
return rc;
}
static const char *
get_action(const char *rsc_action) {
const char *action = NULL;
if (pcmk__str_eq(rsc_action, "validate", pcmk__str_casei)) {
action = "validate-all";
} else if (pcmk__str_eq(rsc_action, "force-check", pcmk__str_casei)) {
action = "monitor";
} else if (pcmk__strcase_any_of(rsc_action, "force-start", "force-stop",
"force-demote", "force-promote", NULL)) {
action = rsc_action+6;
} else {
action = rsc_action;
}
return action;
}
/*!
* \brief Set up environment variables as expected by resource agents
*
* When the cluster executes resource agents, it adds certain environment
* variables (directly or via resource meta-attributes) expected by some
* resource agents. Add the essential ones that many resource agents expect, so
* the behavior is the same for command-line execution.
*
* \param[in] params Resource parameters that will be passed to agent
* \param[in] timeout_ms Action timeout (in milliseconds)
* \param[in] check_level OCF check level
* \param[in] verbosity Verbosity level
*/
static void
set_agent_environment(GHashTable *params, int timeout_ms, int check_level,
int verbosity)
{
g_hash_table_insert(params, strdup("CRM_meta_timeout"),
crm_strdup_printf("%d", timeout_ms));
g_hash_table_insert(params, strdup(XML_ATTR_CRM_VERSION),
strdup(CRM_FEATURE_SET));
if (check_level >= 0) {
char *level = crm_strdup_printf("%d", check_level);
setenv("OCF_CHECK_LEVEL", level, 1);
free(level);
}
setenv("HA_debug", (verbosity > 0)? "1" : "0", 1);
if (verbosity > 1) {
setenv("OCF_TRACE_RA", "1", 1);
}
/* A resource agent using the standard ocf-shellfuncs library will not print
* messages to stderr if it doesn't have a controlling terminal (e.g. if
* crm_resource is called via script or ssh). This forces it to do so.
*/
setenv("OCF_TRACE_FILE", "/dev/stderr", 0);
}
/*!
* \internal
* \brief Apply command-line overrides to resource parameters
*
* \param[in] params Parameters to be passed to agent
* \param[in] overrides Parameters to override (or NULL if none)
*/
static void
apply_overrides(GHashTable *params, GHashTable *overrides)
{
if (overrides != NULL) {
GHashTableIter iter;
char *name = NULL;
char *value = NULL;
g_hash_table_iter_init(&iter, overrides);
while (g_hash_table_iter_next(&iter, (gpointer *) &name,
(gpointer *) &value)) {
g_hash_table_replace(params, strdup(name), strdup(value));
}
}
}
crm_exit_t
cli_resource_execute_from_params(pcmk__output_t *out, const char *rsc_name,
const char *rsc_class, const char *rsc_prov,
const char *rsc_type, const char *rsc_action,
GHashTable *params, GHashTable *override_hash,
int timeout_ms, int resource_verbose, gboolean force,
int check_level)
{
const char *class = rsc_class;
const char *action = get_action(rsc_action);
crm_exit_t exit_code = CRM_EX_OK;
svc_action_t *op = NULL;
// If no timeout was provided, use the same default as the cluster
if (timeout_ms == 0) {
timeout_ms = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S);
}
set_agent_environment(params, timeout_ms, check_level, resource_verbose);
apply_overrides(params, override_hash);
op = services__create_resource_action(rsc_name? rsc_name : "test",
rsc_class, rsc_prov, rsc_type, action,
0, timeout_ms, params, 0);
if (op == NULL) {
out->err(out, "Could not execute %s using %s%s%s:%s: %s",
action, rsc_class, (rsc_prov? ":" : ""),
(rsc_prov? rsc_prov : ""), rsc_type, strerror(ENOMEM));
g_hash_table_destroy(params);
return CRM_EX_OSERR;
}
if (pcmk__str_eq(rsc_class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) {
class = resources_find_service_class(rsc_type);
}
if (!pcmk__strcase_any_of(class, PCMK_RESOURCE_CLASS_OCF,
PCMK_RESOURCE_CLASS_LSB, NULL)) {
- services__set_result(op, CRM_EX_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR,
- "Manual execution of this standard is unsupported");
+ services__format_result(op, CRM_EX_UNIMPLEMENT_FEATURE, PCMK_EXEC_ERROR,
+ "Manual execution of the %s standard "
+ "is unsupported", crm_str(class));
}
if (op->rc != PCMK_OCF_UNKNOWN) {
exit_code = op->rc;
goto done;
}
services_action_sync(op);
// Map results to OCF codes for consistent reporting to user
{
enum ocf_exitcode ocf_code = services_result2ocf(class, action, op->rc);
// Cast variable instead of function return to keep compilers happy
exit_code = (crm_exit_t) ocf_code;
}
done:
out->message(out, "resource-agent-action", resource_verbose, rsc_class,
rsc_prov, rsc_type, rsc_name, rsc_action, override_hash,
exit_code, op->status, services__exit_reason(op),
op->stdout_data, op->stderr_data);
services_action_free(op);
return exit_code;
}
crm_exit_t
cli_resource_execute(pe_resource_t *rsc, const char *requested_name,
const char *rsc_action, GHashTable *override_hash,
int timeout_ms, cib_t * cib, pe_working_set_t *data_set,
int resource_verbose, gboolean force, int check_level)
{
pcmk__output_t *out = data_set->priv;
crm_exit_t exit_code = CRM_EX_OK;
const char *rid = NULL;
const char *rtype = NULL;
const char *rprov = NULL;
const char *rclass = NULL;
GHashTable *params = NULL;
if (pcmk__strcase_any_of(rsc_action, "force-start", "force-demote",
"force-promote", NULL)) {
if(pe_rsc_is_clone(rsc)) {
GList *nodes = cli_resource_search(rsc, requested_name, data_set);
if(nodes != NULL && force == FALSE) {
out->err(out, "It is not safe to %s %s here: the cluster claims it is already active",
rsc_action, rsc->id);
out->err(out, "Try setting target-role=Stopped first or specifying "
"the force option");
return CRM_EX_UNSAFE;
}
g_list_free_full(nodes, free);
}
}
if(pe_rsc_is_clone(rsc)) {
/* Grab the first child resource in the hope it's not a group */
rsc = rsc->children->data;
}
if(rsc->variant == pe_group) {
out->err(out, "Sorry, the %s option doesn't support group resources", rsc_action);
return CRM_EX_UNIMPLEMENT_FEATURE;
} else if (rsc->variant == pe_container || pe_rsc_is_bundled(rsc)) {
out->err(out, "Sorry, the %s option doesn't support bundled resources", rsc_action);
return CRM_EX_UNIMPLEMENT_FEATURE;
}
rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
rprov = crm_element_value(rsc->xml, XML_AGENT_ATTR_PROVIDER);
rtype = crm_element_value(rsc->xml, XML_ATTR_TYPE);
params = generate_resource_params(rsc, NULL /* @TODO use local node */,
data_set);
if (timeout_ms == 0) {
timeout_ms = pe_get_configured_timeout(rsc, get_action(rsc_action), data_set);
}
rid = pe_rsc_is_anon_clone(rsc->parent)? requested_name : rsc->id;
exit_code = cli_resource_execute_from_params(out, rid, rclass, rprov, rtype, rsc_action,
params, override_hash, timeout_ms,
resource_verbose, force, check_level);
return exit_code;
}
// \return Standard Pacemaker return code
int
cli_resource_move(pe_resource_t *rsc, const char *rsc_id, const char *host_name,
const char *move_lifetime, cib_t *cib, int cib_options,
pe_working_set_t *data_set, gboolean promoted_role_only,
gboolean force)
{
pcmk__output_t *out = data_set->priv;
int rc = pcmk_rc_ok;
unsigned int count = 0;
pe_node_t *current = NULL;
pe_node_t *dest = pe_find_node(data_set->nodes, host_name);
bool cur_is_dest = FALSE;
if (dest == NULL) {
return pcmk_rc_node_unknown;
}
if (promoted_role_only && !pcmk_is_set(rsc->flags, pe_rsc_promotable)) {
pe_resource_t *p = uber_parent(rsc);
if (pcmk_is_set(p->flags, pe_rsc_promotable)) {
out->info(out, "Using parent '%s' for move instead of '%s'.", rsc->id, rsc_id);
rsc_id = p->id;
rsc = p;
} else {
out->info(out, "Ignoring master option: %s is not promotable", rsc_id);
promoted_role_only = FALSE;
}
}
current = pe__find_active_requires(rsc, &count);
if (pcmk_is_set(rsc->flags, pe_rsc_promotable)) {
GList *iter = NULL;
unsigned int promoted_count = 0;
pe_node_t *promoted_node = NULL;
for(iter = rsc->children; iter; iter = iter->next) {
pe_resource_t *child = (pe_resource_t *)iter->data;
enum rsc_role_e child_role = child->fns->state(child, TRUE);
if (child_role == RSC_ROLE_PROMOTED) {
rsc = child;
promoted_node = pe__current_node(child);
promoted_count++;
}
}
if (promoted_role_only || (promoted_count != 0)) {
count = promoted_count;
current = promoted_node;
}
}
if (count > 1) {
if (pe_rsc_is_clone(rsc)) {
current = NULL;
} else {
return pcmk_rc_multiple;
}
}
if (current && (current->details == dest->details)) {
cur_is_dest = TRUE;
if (force) {
crm_info("%s is already %s on %s, reinforcing placement with location constraint.",
rsc_id, promoted_role_only?"promoted":"active", dest->details->uname);
} else {
return pcmk_rc_already;
}
}
/* Clear any previous prefer constraints across all nodes. */
cli_resource_clear(rsc_id, NULL, data_set->nodes, cib, cib_options, FALSE, force);
/* Clear any previous ban constraints on 'dest'. */
cli_resource_clear(rsc_id, dest->details->uname, data_set->nodes, cib,
cib_options, TRUE, force);
/* Record an explicit preference for 'dest' */
rc = cli_resource_prefer(out, rsc_id, dest->details->uname, move_lifetime,
cib, cib_options, promoted_role_only);
crm_trace("%s%s now prefers node %s%s",
rsc->id, (promoted_role_only? " (promoted)" : ""),
dest->details->uname, force?"(forced)":"");
/* only ban the previous location if current location != destination location.
* it is possible to use -M to enforce a location without regard of where the
* resource is currently located */
if(force && (cur_is_dest == FALSE)) {
/* Ban the original location if possible */
if(current) {
(void)cli_resource_ban(out, rsc_id, current->details->uname, move_lifetime,
NULL, cib, cib_options, promoted_role_only);
} else if(count > 1) {
out->info(out, "Resource '%s' is currently %s in %d locations. "
"One may now move to %s",
rsc_id, (promoted_role_only? "promoted" : "active"),
count, dest->details->uname);
out->info(out, "To prevent '%s' from being %s at a specific location, "
"specify a node.",
rsc_id, (promoted_role_only? "promoted" : "active"));
} else {
crm_trace("Not banning %s from its current location: not active", rsc_id);
}
}
return rc;
}

File Metadata

Mime Type
text/x-diff
Expires
Tue, Jul 8, 6:37 PM (2 h, 27 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
2002721
Default Alt Text
(151 KB)

Event Timeline