static void
systemd_exec_result(DBusMessage *reply, svc_action_t *op)
{
DBusError error;
if (pcmk_dbus_find_error((void*)&error, reply, &error)) {
/* ignore "already started" or "not running" errors */
if (!systemd_mask_error(op, error.name)) {
crm_err("Could not issue %s for %s: %s", op->action, op->rsc, error.message);
}
dbus_error_free(&error);
} else {
if(!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
crm_warn("Call to %s passed but return type was unexpected", op->action);
op->rc = PCMK_OCF_OK;
} else {
const char *path = NULL;
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
crm_info("Call to %s passed: %s", op->action, path);
op->rc = PCMK_OCF_OK;
}
}
operation_finalize(op);
}
static void
systemd_unit_check(const char *name, const char *state, void *userdata)
{
svc_action_t * op = userdata;
crm_trace("Resource %s has %s='%s'", op->rsc, name, state);
if(state == NULL) {
op->rc = PCMK_OCF_NOT_RUNNING;
} else if (g_strcmp0(state, "active") == 0) {
op->rc = PCMK_OCF_OK;
} else if (g_strcmp0(state, "activating") == 0) {
op->rc = PCMK_OCF_PENDING;
} else if (g_strcmp0(state, "deactivating") == 0) {
op->rc = PCMK_OCF_PENDING;
} else {
op->rc = PCMK_OCF_NOT_RUNNING;
}
if (op->synchronous == FALSE) {
if (op->opaque->pending) {
dbus_pending_call_unref(op->opaque->pending);
}
op->opaque->pending = NULL;
operation_finalize(op);
}
}
static const char *
systemd_loadunit_result(DBusMessage *reply, svc_action_t * op)
{
const char *path = NULL;
DBusError error;
if (pcmk_dbus_find_error((void*)&path, reply, &error)) {
if(op && !systemd_mask_error(op, error.name)) {
crm_err("Could not load systemd unit %s for %s: %s",
op->agent, op->id, error.message);
}
dbus_error_free(&error);
} else if(pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
}
if(op) {
if (path) {
systemd_unit_exec_with_unit(op, path);
} else if (op->synchronous == FALSE) {
operation_finalize(op);
}
}
return path;
}
gboolean
systemd_unit_exec_with_unit(svc_action_t * op, const char *unit)
{
const char *method = op->action;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
CRM_ASSERT(unit);
if (safe_str_eq(op->action, "monitor") || safe_str_eq(method, "status")) {
DBusPendingCall *pending = NULL;
char *state;
state = pcmk_dbus_get_property(systemd_proxy, BUS_NAME, unit,
BUS_NAME ".Unit", "ActiveState",
op->synchronous?NULL:systemd_unit_check,
op, op->synchronous?NULL:&pending, op->timeout);
if (op->synchronous) {
systemd_unit_check("ActiveState", state, op);
free(state);
return op->rc == PCMK_OCF_OK;
} else if (pending) {
services_set_op_pending(op, pending);
return TRUE;
} else {
return operation_finalize(op);
}
} else if (g_strcmp0(method, "start") == 0) {
FILE *file_strm = NULL;
char *override_dir = crm_strdup_printf("%s/%s.service.d", SYSTEMD_OVERRIDE_ROOT, op->agent);
char *override_file = crm_strdup_printf("%s/%s.service.d/50-pacemaker.conf", SYSTEMD_OVERRIDE_ROOT, op->agent);
method = "StartUnit";
crm_build_path(override_dir, 0755);
file_strm = fopen(override_file, "w");
if (file_strm != NULL) {
/* TODO: Insert the start timeout in too */
char *override = crm_strdup_printf(
"[Unit]\n"
"Description=Cluster Controlled %s\n"
"Before=pacemaker.service\n"
"\n"
"[Service]\n"
"Restart=no\n",
op->agent);
int rc = fprintf(file_strm, "%s\n", override);
free(override);
if (rc < 0) {
crm_perror(LOG_ERR, "Cannot write to systemd override file %s", override_file);
}
} else {
static void
operation_finished(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)
{
svc_action_t *op = mainloop_child_userdata(p);
char *prefix = g_strdup_printf("%s:%d", op->id, op->pid);
mainloop_clear_child_userdata(p);
op->status = PCMK_LRM_OP_DONE;
CRM_ASSERT(op->pid == pid);
if (op->opaque->stderr_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
dispatch_stderr(op);
}
if (op->opaque->stdout_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
dispatch_stdout(op);
}
if (signo) {
if (mainloop_child_timeout(p)) {
crm_warn("%s - timed out after %dms", prefix, op->timeout);
op->status = PCMK_LRM_OP_TIMEOUT;
op->rc = PCMK_OCF_TIMEOUT;
} else {
crm_warn("%s - terminated with signal %d", prefix, signo);
op->status = PCMK_LRM_OP_ERROR;
op->rc = PCMK_OCF_SIGNAL;
}
} else {
op->rc = exitcode;
crm_debug("%s - exited with rc=%d", prefix, exitcode);
}
g_free(prefix);
prefix = g_strdup_printf("%s:%d:stderr", op->id, op->pid);
crm_log_output(LOG_NOTICE, prefix, op->stderr_data);
g_free(prefix);
prefix = g_strdup_printf("%s:%d:stdout", op->id, op->pid);
crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
g_free(prefix);
operation_finalize(op);
}
static void
upstart_async_dispatch(DBusPendingCall *pending, void *user_data)
{
DBusError error;
DBusMessage *reply = NULL;
svc_action_t *op = user_data;
dbus_error_init(&error);
if(pending) {
reply = dbus_pending_call_steal_reply(pending);
}
if (pcmk_dbus_find_error(pending, reply, &error)) {
/* ignore "already started" or "not running" errors */
if (!upstart_mask_error(op, error.name)) {
crm_err("%s for %s: %s", op->action, op->rsc, error.message);
}
dbus_error_free(&error);
} else if (!g_strcmp0(op->action, "stop")) {
/* No return vaue */
op->rc = PCMK_OCF_OK;
} else {
if(!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
crm_warn("Call to %s passed but return type was unexpected", op->action);
op->rc = PCMK_OCF_OK;
} else {
const char *path = NULL;
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
crm_info("Call to %s passed: %s", op->action, path);
op->rc = PCMK_OCF_OK;
}
}
CRM_LOG_ASSERT(pending == op->opaque->pending);
services_set_op_pending(op, NULL);
operation_finalize(op);
if(reply) {
dbus_message_unref(reply);
}
}
static void
systemd_unit_exec_done(GObject * source_object, GAsyncResult * res, gpointer user_data)
{
GError *error = NULL;
GVariant *_ret = NULL;
svc_action_t *op = user_data;
GDBusProxy *proxy = G_DBUS_PROXY(source_object);
/* Obtain rc and stderr/out */
_ret = g_dbus_proxy_call_finish(proxy, res, &error);
if (error) {
/* ignore "already started" or "not running" errors */
crm_trace("Could not issue %s for %s: %s", op->action, op->rsc, error->message);
if (strstr(error->message, "systemd1.LoadFailed")
|| strstr(error->message, "systemd1.InvalidName")) {
if (safe_str_eq(op->action, "stop")) {
crm_trace("Masking Stop failure for %s: unknown services are stopped", op->rsc);
op->rc = PCMK_EXECRA_OK;
} else {
op->rc = PCMK_EXECRA_NOT_INSTALLED;
}
} else {
crm_err("Could not issue %s for %s: %s", op->action, op->rsc, error->message);
}
g_error_free(error);
} else if(g_variant_is_of_type (_ret, G_VARIANT_TYPE("(o)"))) {
char *path = NULL;
g_variant_get(_ret, "(o)", &path);
crm_info("Call to %s passed: type '%s' %s", op->action, g_variant_get_type_string(_ret),
path);
op->rc = PCMK_EXECRA_OK;
} else {
crm_err("Call to %s passed but return type was '%s' not '(o)'", op->action, g_variant_get_type_string(_ret));
op->rc = PCMK_EXECRA_OK;
}
operation_finalize(op);
if (_ret) {
g_variant_unref(_ret);
}
}
static void
upstart_job_check(const char *name, const char *state, void *userdata)
{
svc_action_t * op = userdata;
if (state && g_strcmp0(state, "running") == 0) {
op->rc = PCMK_OCF_OK;
/* } else if (g_strcmp0(state, "activating") == 0) { */
/* op->rc = PCMK_OCF_PENDING; */
} else {
op->rc = PCMK_OCF_NOT_RUNNING;
}
if (op->synchronous == FALSE) {
services_set_op_pending(op, NULL);
operation_finalize(op);
}
}
static void
systemd_unit_check(const char *name, const char *state, void *userdata)
{
svc_action_t * op = userdata;
CRM_ASSERT(state != NULL);
if (g_strcmp0(state, "active") == 0) {
op->rc = PCMK_OCF_OK;
} else if (g_strcmp0(state, "activating") == 0) {
op->rc = PCMK_OCF_PENDING;
} else {
op->rc = PCMK_OCF_NOT_RUNNING;
}
if (op->synchronous == FALSE) {
operation_finalize(op);
}
}
void
handle_blocked_ops(void)
{
GList *executed_ops = NULL;
GList *gIter = NULL;
svc_action_t *op = NULL;
gboolean res = FALSE;
if (processing_blocked_ops) {
/* avoid nested calling of this function */
return;
}
processing_blocked_ops = TRUE;
/* n^2 operation here, but blocked ops are incredibly rare. this list
* will be empty 99% of the time. */
for (gIter = blocked_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
if (is_op_blocked(op->rsc)) {
continue;
}
executed_ops = g_list_append(executed_ops, op);
res = action_async_helper(op);
if (res == FALSE) {
op->status = PCMK_LRM_OP_ERROR;
/* this can cause this function to be called recursively
* which is why we have processing_blocked_ops static variable */
operation_finalize(op);
}
}
for (gIter = executed_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
blocked_ops = g_list_remove(blocked_ops, op);
}
g_list_free(executed_ops);
processing_blocked_ops = FALSE;
}
/* Returns FALSE if 'op' should be free'd by the caller */
gboolean
services_os_action_execute(svc_action_t * op, gboolean synchronous)
{
int stdout_fd[2];
int stderr_fd[2];
sigset_t mask;
sigset_t old_mask;
struct stat st;
if (pipe(stdout_fd) < 0) {
crm_err("pipe() failed");
}
if (pipe(stderr_fd) < 0) {
crm_err("pipe() failed");
}
/* Fail fast */
if(stat(op->opaque->exec, &st) != 0) {
int rc = errno;
crm_warn("Cannot execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (synchronous) {
sigemptyset(&mask);
sigaddset(&mask, SIGCHLD);
sigemptyset(&old_mask);
if (sigprocmask(SIG_BLOCK, &mask, &old_mask) < 0) {
crm_perror(LOG_ERR, "sigprocmask() failed");
}
}
op->pid = fork();
switch (op->pid) {
case -1:
{
int rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
close(stderr_fd[0]);
close(stderr_fd[1]);
crm_err("Could not execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
return FALSE;
}
case 0: /* Child */
close(stdout_fd[0]);
close(stderr_fd[0]);
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
crm_err("dup2() failed (stdout)");
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
crm_err("dup2() failed (stderr)");
}
close(stderr_fd[1]);
}
action_launch_child(op);
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
op->opaque->stdout_fd = stdout_fd[0];
set_fd_opts(op->opaque->stdout_fd, O_NONBLOCK);
op->opaque->stderr_fd = stderr_fd[0];
set_fd_opts(op->opaque->stderr_fd, O_NONBLOCK);
if (synchronous) {
action_synced_wait(op, mask);
if (sigismember(&old_mask, SIGCHLD) == 0) {
if (sigprocmask(SIG_UNBLOCK, &mask, NULL) < 0) {
crm_perror(LOG_ERR, "sigprocmask() to unblocked failed");
}
}
} else {
crm_trace("Async waiting for %d - %s", op->pid, op->opaque->exec);
mainloop_child_add_with_flags(op->pid,
op->timeout,
op->id,
op,
(op->flags & SVC_ACTION_LEAVE_GROUP) ? mainloop_leave_pid_group : 0,
operation_finished);
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);
}
return TRUE;
}
gboolean
systemd_unit_exec_with_unit(svc_action_t * op, const char *unit)
{
const char *method = op->action;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
CRM_ASSERT(unit);
if (unit == NULL) {
crm_debug("Could not obtain unit named '%s'", op->agent);
op->rc = PCMK_OCF_NOT_INSTALLED;
op->status = PCMK_LRM_OP_NOT_INSTALLED;
goto cleanup;
}
if (safe_str_eq(op->action, "monitor") || safe_str_eq(method, "status")) {
char *state = pcmk_dbus_get_property(systemd_proxy, BUS_NAME, unit, BUS_NAME ".Unit", "ActiveState",
op->synchronous?NULL:systemd_unit_check, op);
if (op->synchronous) {
systemd_unit_check("ActiveState", state, op);
free(state);
return op->rc == PCMK_OCF_OK;
}
return TRUE;
} else if (g_strcmp0(method, "start") == 0) {
FILE *file_strm = NULL;
char *override_dir = g_strdup_printf("%s/%s", SYSTEMD_OVERRIDE_ROOT, unit);
char *override_file = g_strdup_printf("%s/%s/50-pacemaker.conf", SYSTEMD_OVERRIDE_ROOT, unit);
method = "StartUnit";
crm_build_path(override_dir, 0755);
file_strm = fopen(override_file, "w");
if (file_strm != NULL) {
int rc = fprintf(file_strm, "[Service]\nRestart=no");
if (rc < 0) {
crm_perror(LOG_ERR, "Cannot write to systemd override file %s", override_file);
}
} else {
crm_err("Cannot open systemd override file %s for writing", override_file);
}
if (file_strm != NULL) {
fflush(file_strm);
fclose(file_strm);
}
systemd_daemon_reload();
free(override_file);
free(override_dir);
} else if (g_strcmp0(method, "stop") == 0) {
char *override_file = g_strdup_printf("%s/%s/50-pacemaker.conf", SYSTEMD_OVERRIDE_ROOT, unit);
method = "StopUnit";
unlink(override_file);
free(override_file);
systemd_daemon_reload();
} else if (g_strcmp0(method, "restart") == 0) {
method = "RestartUnit";
} else {
op->rc = PCMK_OCF_UNIMPLEMENT_FEATURE;
goto cleanup;
}
crm_debug("Calling %s for %s: %s", method, op->rsc, unit);
msg = systemd_new_method(BUS_NAME".Manager", method);
CRM_ASSERT(msg != NULL);
/* (ss) */
{
const char *replace_s = "replace";
char *name = systemd_service_name(op->agent);
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 == FALSE) {
return pcmk_dbus_send(msg, systemd_proxy, systemd_async_dispatch, op);
} else {
DBusError error;
reply = pcmk_dbus_send_recv(msg, systemd_proxy, &error);
systemd_exec_result(reply, op);
if(reply) {
dbus_message_unref(reply);
}
}
if(msg) {
dbus_message_unref(msg);
}
cleanup:
if (op->synchronous == FALSE) {
operation_finalize(op);
return TRUE;
}
return op->rc == PCMK_OCF_OK;
}
static void
operation_finished(mainloop_child_t *p, int status, int signo, int exitcode)
{
char *next = NULL;
char *offset = NULL;
svc_action_t *op = mainloop_get_child_userdata(p);
pid_t pid = mainloop_get_child_pid(p);
mainloop_clear_child_userdata(p);
op->status = PCMK_LRM_OP_DONE;
CRM_ASSERT(op->pid == pid);
if (op->opaque->stderr_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
dispatch_stderr(op);
}
if (op->opaque->stdout_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
dispatch_stdout(op);
}
if (signo) {
if (mainloop_get_child_timeout(p)) {
crm_warn("%s:%d - timed out after %dms", op->id, op->pid,
op->timeout);
op->status = PCMK_LRM_OP_TIMEOUT;
op->rc = PCMK_OCF_TIMEOUT;
} else {
crm_warn("%s:%d - terminated with signal %d", op->id, op->pid,
signo);
op->status = PCMK_LRM_OP_ERROR;
op->rc = PCMK_OCF_SIGNAL;
}
} else {
op->rc = exitcode;
crm_debug("%s:%d - exited with rc=%d", op->id, op->pid, exitcode);
if (op->stdout_data) {
next = op->stdout_data;
do {
offset = next;
next = strchrnul(offset, '\n');
crm_debug("%s:%d [ %.*s ]", op->id, op->pid,
(int) (next - offset), offset);
if (next[0] != 0) {
next++;
}
} while (next != NULL && next[0] != 0);
}
if (op->stderr_data) {
next = op->stderr_data;
do {
offset = next;
next = strchrnul(offset, '\n');
crm_notice("%s:%d [ %.*s ]", op->id, op->pid,
(int) (next - offset), offset);
if (next[0] != 0) {
next++;
}
} while (next != NULL && next[0] != 0);
}
}
op->pid = 0;
operation_finalize(op);
}
gboolean
systemd_unit_exec_with_unit(svc_action_t * op, const char *unit)
{
const char *method = op->action;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
CRM_ASSERT(unit);
if (safe_str_eq(op->action, "monitor") || safe_str_eq(method, "status")) {
DBusPendingCall *pending = NULL;
char *state;
state = systemd_get_property(unit, "ActiveState",
(op->synchronous? NULL : systemd_unit_check),
op, (op->synchronous? NULL : &pending),
op->timeout);
if (op->synchronous) {
systemd_unit_check("ActiveState", state, op);
free(state);
return op->rc == PCMK_OCF_OK;
} else if (pending) {
services_set_op_pending(op, pending);
return TRUE;
} else {
return operation_finalize(op);
}
} else if (g_strcmp0(method, "start") == 0) {
FILE *file_strm = NULL;
char *override_dir = crm_strdup_printf("%s/%s.service.d", SYSTEMD_OVERRIDE_ROOT, op->agent);
char *override_file = crm_strdup_printf("%s/%s.service.d/50-pacemaker.conf", SYSTEMD_OVERRIDE_ROOT, op->agent);
mode_t orig_umask;
method = "StartUnit";
crm_build_path(override_dir, 0755);
/* Ensure the override file is world-readable. This is not strictly
* necessary, but it avoids a systemd warning in the logs.
*/
orig_umask = umask(S_IWGRP | S_IWOTH);
file_strm = fopen(override_file, "w");
umask(orig_umask);
if (file_strm != NULL) {
/* TODO: Insert the start timeout in too */
char *override = crm_strdup_printf(
"[Unit]\n"
"Description=Cluster Controlled %s\n"
"Before=pacemaker.service\n"
"\n"
"[Service]\n"
"Restart=no\n",
op->agent);
int rc = fprintf(file_strm, "%s\n", override);
free(override);
if (rc < 0) {
crm_perror(LOG_ERR, "Cannot write to systemd override file %s", override_file);
}
} else {
static void
action_synced_wait(svc_action_t * op, sigset_t *mask)
{
int status = 0;
int timeout = op->timeout;
int sfd = -1;
time_t start = -1;
struct pollfd fds[3];
int wait_rc = 0;
#ifdef HAVE_SYS_SIGNALFD_H
sfd = signalfd(-1, mask, SFD_NONBLOCK);
if (sfd < 0) {
crm_perror(LOG_ERR, "signalfd() failed");
}
#else
sfd = sigchld_pipe[0];
#endif
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 = sfd;
fds[2].events = POLLIN;
fds[2].revents = 0;
crm_trace("Waiting for %d", op->pid);
start = time(NULL);
do {
int poll_rc = poll(fds, 3, timeout);
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) {
#ifdef HAVE_SYS_SIGNALFD_H
struct signalfd_siginfo fdsi;
ssize_t s;
s = read(sfd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
crm_perror(LOG_ERR, "Read from signal fd %d failed", sfd);
} else if (fdsi.ssi_signo == SIGCHLD) {
#else
if (1) {
/* Clear out the sigchld pipe. */
char ch;
while (read(sfd, &ch, 1) == 1);
#endif
wait_rc = waitpid(op->pid, &status, WNOHANG);
if (wait_rc < 0){
crm_perror(LOG_ERR, "waitpid() for %d failed", op->pid);
} else if (wait_rc > 0) {
break;
}
}
}
} else if (poll_rc == 0) {
timeout = 0;
break;
} else if (poll_rc < 0) {
if (errno != EINTR) {
crm_perror(LOG_ERR, "poll() failed");
break;
}
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
crm_trace("Child done: %d", op->pid);
if (wait_rc <= 0) {
int killrc = kill(op->pid, SIGKILL);
op->rc = PCMK_OCF_UNKNOWN_ERROR;
if (op->timeout > 0 && timeout <= 0) {
op->status = PCMK_LRM_OP_TIMEOUT;
crm_warn("%s:%d - timed out after %dms", op->id, op->pid, op->timeout);
} else {
op->status = PCMK_LRM_OP_ERROR;
}
if (killrc && errno != ESRCH) {
crm_err("kill(%d, KILL) failed: %d", op->pid, errno);
}
/*
* From sigprocmask(2):
* It is not possible to block SIGKILL or SIGSTOP. Attempts to do so are silently ignored.
*
* This makes it safe to skip WNOHANG here
*/
waitpid(op->pid, &status, 0);
} else if (WIFEXITED(status)) {
op->status = PCMK_LRM_OP_DONE;
op->rc = WEXITSTATUS(status);
crm_info("Managed %s process %d exited with rc=%d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
op->status = PCMK_LRM_OP_ERROR;
crm_err("Managed %s process %d exited with signal=%d", op->id, op->pid, signo);
}
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
crm_err("Managed %s process %d dumped core", op->id, op->pid);
}
#endif
svc_read_output(op->opaque->stdout_fd, op, FALSE);
svc_read_output(op->opaque->stderr_fd, op, TRUE);
close(op->opaque->stdout_fd);
close(op->opaque->stderr_fd);
#ifdef HAVE_SYS_SIGNALFD_H
close(sfd);
#endif
}
/* For an asynchronous 'op', returns FALSE if 'op' should be free'd by the caller */
/* For a synchronous 'op', returns FALSE if 'op' fails */
gboolean
services_os_action_execute(svc_action_t * op, gboolean synchronous)
{
int stdout_fd[2];
int stderr_fd[2];
struct stat st;
sigset_t *pmask;
#ifdef HAVE_SYS_SIGNALFD_H
sigset_t mask;
sigset_t old_mask;
#define sigchld_cleanup() { \
if (sigismember(&old_mask, SIGCHLD) == 0) { \
if (sigprocmask(SIG_UNBLOCK, &mask, NULL) < 0) { \
crm_perror(LOG_ERR, "sigprocmask() failed to unblock sigchld"); \
} \
} \
}
#else
struct sigaction sa;
struct sigaction old_sa;
#define sigchld_cleanup() { \
if (sigaction(SIGCHLD, &old_sa, NULL) < 0) { \
crm_perror(LOG_ERR, "sigaction() failed to remove sigchld handler"); \
} \
close(sigchld_pipe[0]); \
close(sigchld_pipe[1]); \
}
#endif
if (pipe(stdout_fd) < 0) {
crm_err("pipe() failed");
}
if (pipe(stderr_fd) < 0) {
crm_err("pipe() failed");
}
/* Fail fast */
if(stat(op->opaque->exec, &st) != 0) {
int rc = errno;
crm_warn("Cannot execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (synchronous) {
#ifdef HAVE_SYS_SIGNALFD_H
sigemptyset(&mask);
sigaddset(&mask, SIGCHLD);
sigemptyset(&old_mask);
if (sigprocmask(SIG_BLOCK, &mask, &old_mask) < 0) {
crm_perror(LOG_ERR, "sigprocmask() failed to block sigchld");
}
pmask = &mask;
#else
if(pipe(sigchld_pipe) == -1) {
crm_perror(LOG_ERR, "pipe() failed");
}
set_fd_opts(sigchld_pipe[0], O_NONBLOCK);
set_fd_opts(sigchld_pipe[1], O_NONBLOCK);
sa.sa_handler = sigchld_handler;
sa.sa_flags = 0;
sigemptyset(&sa.sa_mask);
if (sigaction(SIGCHLD, &sa, &old_sa) < 0) {
crm_perror(LOG_ERR, "sigaction() failed to set sigchld handler");
}
pmask = NULL;
#endif
}
op->pid = fork();
switch (op->pid) {
case -1:
{
int rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
close(stderr_fd[0]);
close(stderr_fd[1]);
crm_err("Could not execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
sigchld_cleanup();
return FALSE;
}
case 0: /* Child */
close(stdout_fd[0]);
close(stderr_fd[0]);
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
crm_err("dup2() failed (stdout)");
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
crm_err("dup2() failed (stderr)");
}
close(stderr_fd[1]);
}
if (synchronous) {
sigchld_cleanup();
}
action_launch_child(op);
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
op->opaque->stdout_fd = stdout_fd[0];
set_fd_opts(op->opaque->stdout_fd, O_NONBLOCK);
op->opaque->stderr_fd = stderr_fd[0];
set_fd_opts(op->opaque->stderr_fd, O_NONBLOCK);
if (synchronous) {
action_synced_wait(op, pmask);
sigchld_cleanup();
} else {
crm_trace("Async waiting for %d - %s", op->pid, op->opaque->exec);
mainloop_child_add_with_flags(op->pid,
op->timeout,
op->id,
op,
(op->flags & SVC_ACTION_LEAVE_GROUP) ? mainloop_leave_pid_group : 0,
operation_finished);
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 TRUE;
}
gboolean
systemd_unit_exec(svc_action_t * op, gboolean synchronous)
{
char *unit = NULL;
GError *error = NULL;
gboolean pass = FALSE;
GVariant *_ret = NULL;
const char *action = op->action;
char *name = systemd_service_name(op->agent);
op->rc = PCMK_EXECRA_UNKNOWN_ERROR;
CRM_ASSERT(systemd_init());
crm_debug("Performing %ssynchronous %s op on systemd unit %s named '%s'",
synchronous ? "" : "a", op->action, op->agent, op->rsc);
if (safe_str_eq(op->action, "meta-data")) {
op->stdout_data = systemd_unit_metadata(op->agent);
op->rc = PCMK_EXECRA_OK;
goto cleanup;
}
pass = systemd_unit_by_name(systemd_proxy, op->agent, &unit, NULL, &error);
if (error || pass == FALSE) {
crm_debug("Could not obtain unit named '%s': %s", op->agent,
error ? error->message : "unknown");
if (error && strstr(error->message, "systemd1.NoSuchUnit")) {
op->rc = PCMK_EXECRA_NOT_INSTALLED;
}
g_error_free(error);
goto cleanup;
}
if (safe_str_eq(op->action, "monitor") || safe_str_eq(action, "status")) {
char *state = systemd_unit_property(unit, BUS_NAME ".Unit", "ActiveState");
if (g_strcmp0(state, "active") == 0) {
op->rc = PCMK_EXECRA_OK;
} else {
op->rc = PCMK_EXECRA_NOT_RUNNING;
}
free(state);
goto cleanup;
} else if (g_strcmp0(action, "start") == 0) {
action = "StartUnit";
} else if (g_strcmp0(action, "stop") == 0) {
action = "StopUnit";
} else if (g_strcmp0(action, "restart") == 0) {
action = "RestartUnit";
} else {
op->rc = PCMK_EXECRA_UNIMPLEMENT_FEATURE;
goto cleanup;
}
crm_debug("Calling %s for %s: %s", action, op->rsc, unit);
if (synchronous == FALSE) {
g_dbus_proxy_call(systemd_proxy, action, g_variant_new("(ss)", name, "replace"),
G_DBUS_CALL_FLAGS_NONE, op->timeout, NULL, systemd_unit_exec_done, op);
free(unit);
free(name);
return TRUE;
}
_ret = g_dbus_proxy_call_sync(systemd_proxy, action, g_variant_new("(ss)", name, "replace"),
G_DBUS_CALL_FLAGS_NONE, op->timeout, NULL, &error);
if (error) {
/* ignore "already started" or "not running" errors */
if (safe_str_eq(op->action, "stop")
&& strstr(error->message, "systemd1.InvalidName")) {
crm_trace("Masking Stop failure for %s: unknown services are stopped", op->rsc);
op->rc = PCMK_EXECRA_OK;
} else {
crm_err("Could not issue %s for %s: %s (%s)", action, op->rsc, error->message, unit);
}
g_error_free(error);
} else if(g_variant_is_of_type (_ret, G_VARIANT_TYPE("(o)"))) {
char *path = NULL;
g_variant_get(_ret, "(o)", &path);
crm_info("Call to %s passed: type '%s' %s", op->action, g_variant_get_type_string(_ret),
path);
op->rc = PCMK_EXECRA_OK;
} else {
crm_err("Call to %s passed but return type was '%s' not '(o)'", op->action, g_variant_get_type_string(_ret));
op->rc = PCMK_EXECRA_OK;
}
cleanup:
free(unit);
free(name);
if (_ret) {
g_variant_unref(_ret);
}
if (synchronous == FALSE) {
operation_finalize(op);
return TRUE;
}
return op->rc == PCMK_EXECRA_OK;
}
/* For a synchronous 'op', returns FALSE if 'op' fails */
gboolean
upstart_job_exec(svc_action_t * op, gboolean synchronous)
{
char *job = NULL;
int arg_wait = TRUE;
const char *arg_env = "pacemaker=1";
const char *action = op->action;
DBusError error;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
DBusMessageIter iter, array_iter;
op->rc = PCMK_OCF_UNKNOWN_ERROR;
CRM_ASSERT(upstart_init());
if (safe_str_eq(op->action, "meta-data")) {
op->stdout_data = upstart_job_metadata(op->agent);
op->rc = PCMK_OCF_OK;
goto cleanup;
}
if(!upstart_job_by_name(op->agent, &job, op->timeout)) {
crm_debug("Could not obtain job named '%s' to %s", op->agent, action);
if (!g_strcmp0(action, "stop")) {
op->rc = PCMK_OCF_OK;
} else {
op->rc = PCMK_OCF_NOT_INSTALLED;
op->status = PCMK_LRM_OP_NOT_INSTALLED;
}
goto cleanup;
}
if (safe_str_eq(op->action, "monitor") || safe_str_eq(action, "status")) {
char *path = get_first_instance(job, op->timeout);
op->rc = PCMK_OCF_NOT_RUNNING;
if(path) {
DBusPendingCall *pending = NULL;
char *state = pcmk_dbus_get_property(
upstart_proxy, BUS_NAME, path, UPSTART_06_API ".Instance", "state",
op->synchronous?NULL:upstart_job_check, op,
op->synchronous?NULL:&pending, op->timeout);
free(job);
free(path);
if(op->synchronous) {
upstart_job_check("state", state, op);
free(state);
return op->rc == PCMK_OCF_OK;
} else if (pending) {
services_set_op_pending(op, pending);
services_add_inflight_op(op);
return TRUE;
}
return FALSE;
}
goto cleanup;
} else if (!g_strcmp0(action, "start")) {
action = "Start";
} else if (!g_strcmp0(action, "stop")) {
action = "Stop";
} else if (!g_strcmp0(action, "restart")) {
action = "Restart";
} else {
op->rc = PCMK_OCF_UNIMPLEMENT_FEATURE;
goto cleanup;
}
crm_debug("Calling %s for %s on %s", action, op->rsc, job);
msg = dbus_message_new_method_call(BUS_NAME, // target for the method call
job, // object to call on
UPSTART_JOB_IFACE, // interface to call on
action); // method name
CRM_ASSERT(msg != NULL);
dbus_message_iter_init_append (msg, &iter);
CRM_LOG_ASSERT(dbus_message_iter_open_container (&iter,
DBUS_TYPE_ARRAY,
DBUS_TYPE_STRING_AS_STRING,
&array_iter));
CRM_LOG_ASSERT(dbus_message_iter_append_basic (&array_iter, DBUS_TYPE_STRING, &arg_env));
CRM_LOG_ASSERT(dbus_message_iter_close_container (&iter, &array_iter));
CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_BOOLEAN, &arg_wait, DBUS_TYPE_INVALID));
if (op->synchronous == FALSE) {
DBusPendingCall* pending = pcmk_dbus_send(msg, upstart_proxy, upstart_async_dispatch, op, op->timeout);
free(job);
if(pending) {
services_set_op_pending(op, pending);
services_add_inflight_op(op);
return TRUE;
}
return FALSE;
}
dbus_error_init(&error);
reply = pcmk_dbus_send_recv(msg, upstart_proxy, &error, op->timeout);
if(error.name) {
if(!upstart_mask_error(op, error.name)) {
crm_err("Could not issue %s for %s: %s (%s)", action, op->rsc, error.name, job);
}
} else if (!g_strcmp0(op->action, "stop")) {
/* No return vaue */
op->rc = PCMK_OCF_OK;
} else if(!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
crm_warn("Call to %s passed but return type was unexpected", op->action);
op->rc = PCMK_OCF_OK;
} else {
const char *path = NULL;
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
crm_info("Call to %s passed: %s", op->action, path);
op->rc = PCMK_OCF_OK;
}
cleanup:
free(job);
if(msg) {
dbus_message_unref(msg);
}
if(reply) {
dbus_message_unref(reply);
}
if (op->synchronous == FALSE) {
return operation_finalize(op);
}
return op->rc == PCMK_OCF_OK;
}
/* Returns FALSE if 'op' should be free'd by the caller */
gboolean
services_os_action_execute(svc_action_t * op, gboolean synchronous)
{
int lpc;
int stdout_fd[2];
int stderr_fd[2];
sigset_t mask;
sigset_t old_mask;
struct stat st;
if (pipe(stdout_fd) < 0) {
crm_err("pipe() failed");
}
if (pipe(stderr_fd) < 0) {
crm_err("pipe() failed");
}
/* Fail fast */
if(stat(op->opaque->exec, &st) != 0) {
int rc = errno;
crm_warn("Cannot execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
return FALSE;
}
if (synchronous) {
sigemptyset(&mask);
sigaddset(&mask, SIGCHLD);
sigemptyset(&old_mask);
if (sigprocmask(SIG_BLOCK, &mask, &old_mask) < 0) {
crm_perror(LOG_ERR, "sigprocmask() failed");
}
}
op->pid = fork();
switch (op->pid) {
case -1:
{
int rc = errno;
close(stdout_fd[0]);
close(stdout_fd[1]);
close(stderr_fd[0]);
close(stderr_fd[1]);
crm_err("Could not execute '%s': %s (%d)", op->opaque->exec, pcmk_strerror(rc), rc);
services_handle_exec_error(op, rc);
if (!synchronous) {
return operation_finalize(op);
}
return FALSE;
}
case 0: /* Child */
/* 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);
close(stdout_fd[0]);
close(stderr_fd[0]);
if (STDOUT_FILENO != stdout_fd[1]) {
if (dup2(stdout_fd[1], STDOUT_FILENO) != STDOUT_FILENO) {
crm_err("dup2() failed (stdout)");
}
close(stdout_fd[1]);
}
if (STDERR_FILENO != stderr_fd[1]) {
if (dup2(stderr_fd[1], STDERR_FILENO) != STDERR_FILENO) {
crm_err("dup2() failed (stderr)");
}
close(stderr_fd[1]);
}
/* close all descriptors except stdin/out/err and channels to logd */
for (lpc = getdtablesize() - 1; lpc > STDERR_FILENO; lpc--) {
close(lpc);
}
#if SUPPORT_CIBSECRETS
if (replace_secret_params(op->rsc, op->params) < 0) {
/* replacing secrets failed! */
if (safe_str_eq(op->action,"stop")) {
/* don't fail on stop! */
crm_info("proceeding with the stop operation for %s", op->rsc);
} else {
crm_err("failed to get secrets for %s, "
"considering resource not configured", op->rsc);
_exit(PCMK_OCF_NOT_CONFIGURED);
}
}
#endif
/* Setup environment correctly */
add_OCF_env_vars(op);
/* execute the RA */
execvp(op->opaque->exec, op->opaque->args);
/* Most cases should have been already handled by stat() */
services_handle_exec_error(op, errno);
_exit(op->rc);
}
/* Only the parent reaches here */
close(stdout_fd[1]);
close(stderr_fd[1]);
op->opaque->stdout_fd = stdout_fd[0];
set_fd_opts(op->opaque->stdout_fd, O_NONBLOCK);
op->opaque->stderr_fd = stderr_fd[0];
set_fd_opts(op->opaque->stderr_fd, O_NONBLOCK);
if (synchronous) {
#ifndef HAVE_SYS_SIGNALFD_H
CRM_ASSERT(FALSE);
#else
int status = 0;
int timeout = op->timeout;
int sfd = -1;
time_t start = -1;
struct pollfd fds[3];
int wait_rc = 0;
sfd = signalfd(-1, &mask, 0);
if (sfd < 0) {
crm_perror(LOG_ERR, "signalfd() failed");
}
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 = sfd;
fds[2].events = POLLIN;
fds[2].revents = 0;
crm_trace("Waiting for %d", op->pid);
start = time(NULL);
do {
int poll_rc = poll(fds, 3, timeout);
if (poll_rc > 0) {
if (fds[0].revents & POLLIN) {
read_output(op->opaque->stdout_fd, op);
}
if (fds[1].revents & POLLIN) {
read_output(op->opaque->stderr_fd, op);
}
if (fds[2].revents & POLLIN) {
struct signalfd_siginfo fdsi;
ssize_t s;
s = read(sfd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
crm_perror(LOG_ERR, "Read from signal fd %d failed", sfd);
} else if (fdsi.ssi_signo == SIGCHLD) {
wait_rc = waitpid(op->pid, &status, WNOHANG);
if (wait_rc < 0){
crm_perror(LOG_ERR, "waitpid() for %d failed", op->pid);
} else if (wait_rc > 0) {
break;
}
}
}
} else if (poll_rc == 0) {
timeout = 0;
break;
} else if (poll_rc < 0) {
if (errno != EINTR) {
crm_perror(LOG_ERR, "poll() failed");
break;
}
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
crm_trace("Child done: %d", op->pid);
if (wait_rc <= 0) {
int killrc = kill(op->pid, SIGKILL);
op->rc = PCMK_OCF_UNKNOWN_ERROR;
if (op->timeout > 0 && timeout <= 0) {
op->status = PCMK_LRM_OP_TIMEOUT;
crm_warn("%s:%d - timed out after %dms", op->id, op->pid, op->timeout);
} else {
op->status = PCMK_LRM_OP_ERROR;
}
if (killrc && errno != ESRCH) {
crm_err("kill(%d, KILL) failed: %d", op->pid, errno);
}
/*
* From sigprocmask(2):
* It is not possible to block SIGKILL or SIGSTOP. Attempts to do so are silently ignored.
*
* This makes it safe to skip WNOHANG here
*/
waitpid(op->pid, &status, 0);
} else if (WIFEXITED(status)) {
op->status = PCMK_LRM_OP_DONE;
op->rc = WEXITSTATUS(status);
crm_info("Managed %s process %d exited with rc=%d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
op->status = PCMK_LRM_OP_ERROR;
crm_err("Managed %s process %d exited with signal=%d", op->id, op->pid, signo);
}
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
crm_err("Managed %s process %d dumped core", op->id, op->pid);
}
#endif
read_output(op->opaque->stdout_fd, op);
read_output(op->opaque->stderr_fd, op);
close(op->opaque->stdout_fd);
close(op->opaque->stderr_fd);
close(sfd);
if (sigismember(&old_mask, SIGCHLD) == 0) {
if (sigprocmask(SIG_UNBLOCK, &mask, NULL) < 0) {
crm_perror(LOG_ERR, "sigprocmask() to unblocked failed");
}
}
#endif
} else {
crm_trace("Async waiting for %d - %s", op->pid, op->opaque->exec);
mainloop_child_add(op->pid, op->timeout, op->id, op, operation_finished);
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);
}
return TRUE;
}
