diff --git a/lib/services/services.c b/lib/services/services.c
index 1800a6b999..03da623435 100644
--- a/lib/services/services.c
+++ b/lib/services/services.c
@@ -1,1316 +1,1310 @@
/*
* Copyright 2010-2021 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 <stdio.h>
#include <errno.h>
#include <unistd.h>
#include <dirent.h>
#include <fcntl.h>
#include <crm/crm.h>
#include <crm/common/mainloop.h>
#include <crm/services.h>
#include <crm/services_internal.h>
#include <crm/stonith-ng.h>
#include <crm/msg_xml.h>
#include "services_private.h"
#include "services_ocf.h"
#include "services_lsb.h"
#if SUPPORT_UPSTART
# include <upstart.h>
#endif
#if SUPPORT_SYSTEMD
# include <systemd.h>
#endif
#if SUPPORT_NAGIOS
# include <services_nagios.h>
#endif
/* TODO: Develop a rollover strategy */
static int operations = 0;
static GHashTable *recurring_actions = NULL;
/* ops waiting to run async because of conflicting active
* pending ops */
static GList *blocked_ops = NULL;
/* ops currently active (in-flight) */
static GList *inflight_ops = NULL;
static void handle_blocked_ops(void);
/*!
* \brief Find first service class that can provide a specified agent
*
* \param[in] agent Name of agent to search for
*
* \return Service class if found, NULL otherwise
*
* \note The priority is LSB, then systemd, then upstart. It would be preferable
* to put systemd first, but LSB merely requires a file existence check,
* while systemd requires contacting D-Bus.
*/
const char *
resources_find_service_class(const char *agent)
{
if (services__lsb_agent_exists(agent)) {
return PCMK_RESOURCE_CLASS_LSB;
}
#if SUPPORT_SYSTEMD
if (systemd_unit_exists(agent)) {
return PCMK_RESOURCE_CLASS_SYSTEMD;
}
#endif
#if SUPPORT_UPSTART
if (upstart_job_exists(agent)) {
return PCMK_RESOURCE_CLASS_UPSTART;
}
#endif
return NULL;
}
static inline void
init_recurring_actions(void)
{
if (recurring_actions == NULL) {
recurring_actions = pcmk__strkey_table(NULL, NULL);
}
}
/*!
* \internal
* \brief Check whether op is in-flight systemd or upstart op
*
* \param[in] op Operation to check
*
* \return TRUE if op is in-flight systemd or upstart op
*/
static inline gboolean
inflight_systemd_or_upstart(svc_action_t *op)
{
return pcmk__strcase_any_of(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD,
PCMK_RESOURCE_CLASS_UPSTART, NULL) &&
g_list_find(inflight_ops, op) != NULL;
}
/*!
* \internal
* \brief Expand "service" alias to an actual resource class
*
* \param[in] rsc Resource name (for logging only)
* \param[in] standard Resource class as configured
* \param[in] agent Agent name to look for
*
* \return Newly allocated string with actual resource class
*
* \note The caller is responsible for calling free() on the result.
*/
static char *
expand_resource_class(const char *rsc, const char *standard, const char *agent)
{
char *expanded_class = NULL;
if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0) {
const char *found_class = resources_find_service_class(agent);
if (found_class) {
crm_debug("Found %s agent %s for %s", found_class, agent, rsc);
expanded_class = strdup(found_class);
} else {
crm_info("Assuming resource class lsb for agent %s for %s",
agent, rsc);
expanded_class = strdup(PCMK_RESOURCE_CLASS_LSB);
}
} else {
expanded_class = strdup(standard);
}
CRM_ASSERT(expanded_class);
return expanded_class;
}
/*!
* \internal
* \brief Create a simple svc_action_t instance
*
* \return Newly allocated instance (or NULL if not enough memory)
*/
static svc_action_t *
new_action(void)
{
svc_action_t *op = calloc(1, sizeof(svc_action_t));
if (op == NULL) {
return NULL;
}
op->opaque = calloc(1, sizeof(svc_action_private_t));
if (op->opaque == NULL) {
free(op);
return NULL;
}
// Initialize result
services__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_UNKNOWN, NULL);
return op;
}
static bool
required_argument_missing(uint32_t ra_caps, const char *name,
const char *standard, const char *provider,
const char *agent, const char *action)
{
if (pcmk__str_empty(name)) {
crm_info("Cannot create operation without resource name (bug?)");
return true;
}
if (pcmk__str_empty(standard)) {
crm_info("Cannot create operation for %s without resource class (bug?)",
name);
return true;
}
if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider)
&& pcmk__str_empty(provider)) {
crm_info("Cannot create operation for %s resource %s "
"without provider (bug?)", standard, name);
return true;
}
if (pcmk__str_empty(agent)) {
crm_info("Cannot create operation for %s without agent name (bug?)",
name);
return true;
}
if (pcmk__str_empty(action)) {
crm_info("Cannot create operation for %s without action name (bug?)",
name);
return true;
}
return false;
}
// \return Standard Pacemaker return code (pcmk_rc_ok or ENOMEM)
static int
copy_action_arguments(svc_action_t *op, uint32_t ra_caps, const char *name,
const char *standard, const char *provider,
const char *agent, const char *action)
{
op->rsc = strdup(name);
if (op->rsc == NULL) {
return ENOMEM;
}
op->agent = strdup(agent);
if (op->agent == NULL) {
return ENOMEM;
}
op->standard = expand_resource_class(name, standard, agent);
if (op->standard == NULL) {
return ENOMEM;
}
if (pcmk_is_set(ra_caps, pcmk_ra_cap_status)
&& pcmk__str_eq(action, "monitor", pcmk__str_casei)) {
action = "status";
}
op->action = strdup(action);
if (op->action == NULL) {
return ENOMEM;
}
if (pcmk_is_set(ra_caps, pcmk_ra_cap_provider)) {
op->provider = strdup(provider);
if (op->provider == NULL) {
return ENOMEM;
}
}
return pcmk_rc_ok;
}
svc_action_t *
services__create_resource_action(const char *name, const char *standard,
const char *provider, const char *agent,
const char *action, guint interval_ms, int timeout,
GHashTable *params, enum svc_action_flags flags)
{
svc_action_t *op = NULL;
uint32_t ra_caps = pcmk_get_ra_caps(standard);
int rc = pcmk_rc_ok;
op = new_action();
if (op == NULL) {
crm_crit("Cannot prepare action: %s", strerror(ENOMEM));
if (params != NULL) {
g_hash_table_destroy(params);
}
return NULL;
}
op->interval_ms = interval_ms;
op->timeout = timeout;
op->flags = flags;
op->sequence = ++operations;
// Take ownership of params
if (pcmk_is_set(ra_caps, pcmk_ra_cap_params)) {
op->params = params;
} else if (params != NULL) {
g_hash_table_destroy(params);
params = NULL;
}
if (required_argument_missing(ra_caps, name, standard, provider, agent,
action)) {
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR_FATAL,
"Required agent or action information missing");
return op;
}
op->id = pcmk__op_key(name, action, interval_ms);
if (copy_action_arguments(op, ra_caps, name, standard, provider, agent,
action) != pcmk_rc_ok) {
crm_crit("Cannot prepare %s action for %s: %s",
action, name, strerror(ENOMEM));
services__handle_exec_error(op, ENOMEM);
return op;
}
if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
rc = services__ocf_prepare(op);
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
rc = services__lsb_prepare(op);
#if SUPPORT_SYSTEMD
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
- op->opaque->exec = strdup("systemd-dbus");
- if (op->opaque->exec == NULL) {
- crm_crit("Cannot prepare %s action for %s: %s",
- action, name, strerror(ENOMEM));
- services__handle_exec_error(op, ENOMEM);
- return op;
- }
+ rc = services__systemd_prepare(op);
#endif
#if SUPPORT_UPSTART
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
op->opaque->exec = strdup("upstart-dbus");
if (op->opaque->exec == NULL) {
crm_crit("Cannot prepare %s action for %s: %s",
action, name, strerror(ENOMEM));
services__handle_exec_error(op, ENOMEM);
return op;
}
#endif
#if SUPPORT_NAGIOS
} else if (strcasecmp(op->standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
op->opaque->exec = pcmk__full_path(op->agent, NAGIOS_PLUGIN_DIR);
op->opaque->args[0] = strdup(op->opaque->exec);
if (op->opaque->args[0] == NULL) {
crm_crit("Cannot prepare %s action for %s: %s",
action, name, strerror(ENOMEM));
services__handle_exec_error(op, ENOMEM);
return op;
}
if (pcmk__str_eq(op->action, "monitor", pcmk__str_casei) && (op->interval_ms == 0)) {
/* Invoke --version for a nagios probe */
op->opaque->args[1] = strdup("--version");
if (op->opaque->args[1] == NULL) {
crm_crit("Cannot prepare %s action for %s: %s",
action, name, strerror(ENOMEM));
services__handle_exec_error(op, ENOMEM);
return op;
}
} else if (op->params) {
GHashTableIter iter;
char *key = NULL;
char *value = NULL;
int index = 1; // 0 is already set to executable name
g_hash_table_iter_init(&iter, op->params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
if (index > (PCMK__NELEM(op->opaque->args) - 2)) {
crm_info("Cannot prepare %s action for %s: Too many parameters",
action, name);
services__set_result(op, NAGIOS_STATE_UNKNOWN,
PCMK_EXEC_ERROR_HARD,
"Too many parameters");
break;
}
if (pcmk__str_eq(key, XML_ATTR_CRM_VERSION, pcmk__str_casei) || strstr(key, CRM_META "_")) {
continue;
}
op->opaque->args[index++] = crm_strdup_printf("--%s", key);
op->opaque->args[index++] = strdup(value);
if (op->opaque->args[index - 1] == NULL) {
crm_crit("Cannot prepare %s action for %s: %s",
action, name, strerror(ENOMEM));
services__handle_exec_error(op, ENOMEM);
return op;
}
}
}
// Nagios actions don't need to keep the parameters
if (op->params != NULL) {
g_hash_table_destroy(op->params);
op->params = NULL;
}
#endif
} else {
crm_err("Unknown resource standard: %s", op->standard);
rc = ENOENT;
}
if (rc != pcmk_rc_ok) {
crm_err("Cannot prepare %s operation for %s: %s",
action, name, strerror(rc));
services__handle_exec_error(op, rc);
}
return op;
}
svc_action_t *
resources_action_create(const char *name, const char *standard,
const char *provider, const char *agent,
const char *action, guint interval_ms, int timeout,
GHashTable *params, enum svc_action_flags flags)
{
svc_action_t *op = services__create_resource_action(name, standard,
provider, agent, action, interval_ms, timeout,
params, flags);
if (op == NULL || op->rc != 0) {
services_action_free(op);
return NULL;
} else {
// Preserve public API backward compatibility
op->rc = PCMK_OCF_OK;
op->status = PCMK_EXEC_DONE;
return op;
}
}
svc_action_t *
services_action_create_generic(const char *exec, const char *args[])
{
svc_action_t *op = new_action();
CRM_ASSERT(op != NULL);
op->opaque->exec = strdup(exec);
op->opaque->args[0] = strdup(exec);
if ((op->opaque->exec == NULL) || (op->opaque->args[0] == NULL)) {
crm_crit("Cannot prepare action for '%s': %s", exec, strerror(ENOMEM));
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
strerror(ENOMEM));
return op;
}
if (args == NULL) {
return op;
}
for (int cur_arg = 1; args[cur_arg - 1] != NULL; cur_arg++) {
if (cur_arg == PCMK__NELEM(op->opaque->args)) {
crm_info("Cannot prepare action for '%s': Too many arguments",
exec);
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR_HARD, "Too many arguments");
break;
}
op->opaque->args[cur_arg] = strdup(args[cur_arg - 1]);
if (op->opaque->args[cur_arg] == NULL) {
crm_crit("Cannot prepare action for '%s': %s",
exec, strerror(ENOMEM));
services__set_result(op, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
strerror(ENOMEM));
break;
}
}
return op;
}
/*!
* \brief Create an alert agent action
*
* \param[in] id Alert ID
* \param[in] exec Path to alert agent executable
* \param[in] timeout Action timeout
* \param[in] params Parameters to use with action
* \param[in] sequence Action sequence number
* \param[in] cb_data Data to pass to callback function
*
* \return New action on success, NULL on error
* \note It is the caller's responsibility to free cb_data.
* The caller should not free params explicitly.
*/
svc_action_t *
services_alert_create(const char *id, const char *exec, int timeout,
GHashTable *params, int sequence, void *cb_data)
{
svc_action_t *action = services_action_create_generic(exec, NULL);
action->timeout = timeout;
action->id = strdup(id);
action->params = params;
action->sequence = sequence;
action->cb_data = cb_data;
return action;
}
/*!
* \brief Set the user and group that an action will execute as
*
* \param[in,out] action Action to modify
* \param[in] user Name of user to execute action as
* \param[in] group Name of group to execute action as
*
* \return pcmk_ok on success, -errno otherwise
*
* \note This will have no effect unless the process executing the action runs
* as root, and the action is not a systemd or upstart action.
* We could implement this for systemd by adding User= and Group= to
* [Service] in the override file, but that seems more likely to cause
* problems than be useful.
*/
int
services_action_user(svc_action_t *op, const char *user)
{
CRM_CHECK((op != NULL) && (user != NULL), return -EINVAL);
return crm_user_lookup(user, &(op->opaque->uid), &(op->opaque->gid));
}
/*!
* \brief Execute an alert agent action
*
* \param[in] action Action to execute
* \param[in] cb Function to call when action completes
*
* \return TRUE if the library will free action, FALSE otherwise
*
* \note If this function returns FALSE, it is the caller's responsibility to
* free the action with services_action_free().
*/
gboolean
services_alert_async(svc_action_t *action, void (*cb)(svc_action_t *op))
{
action->synchronous = false;
action->opaque->callback = cb;
return services__execute_file(action) == pcmk_rc_ok;
}
#if SUPPORT_DBUS
/*!
* \internal
* \brief Update operation's pending DBus call, unreferencing old one if needed
*
* \param[in,out] op Operation to modify
* \param[in] pending Pending call to set
*/
void
services_set_op_pending(svc_action_t *op, DBusPendingCall *pending)
{
if (op->opaque->pending && (op->opaque->pending != pending)) {
if (pending) {
crm_info("Lost pending %s DBus call (%p)", op->id, op->opaque->pending);
} else {
crm_trace("Done with pending %s DBus call (%p)", op->id, op->opaque->pending);
}
dbus_pending_call_unref(op->opaque->pending);
}
op->opaque->pending = pending;
if (pending) {
crm_trace("Updated pending %s DBus call (%p)", op->id, pending);
} else {
crm_trace("Cleared pending %s DBus call", op->id);
}
}
#endif
void
services_action_cleanup(svc_action_t * op)
{
if ((op == NULL) || (op->opaque == NULL)) {
return;
}
#if SUPPORT_DBUS
if(op->opaque->timerid != 0) {
crm_trace("Removing timer for call %s to %s", op->action, op->rsc);
g_source_remove(op->opaque->timerid);
op->opaque->timerid = 0;
}
if(op->opaque->pending) {
if (dbus_pending_call_get_completed(op->opaque->pending)) {
// This should never be the case
crm_warn("Result of %s op %s was unhandled",
op->standard, op->id);
} else {
crm_debug("Will ignore any result of canceled %s op %s",
op->standard, op->id);
}
dbus_pending_call_cancel(op->opaque->pending);
services_set_op_pending(op, NULL);
}
#endif
if (op->opaque->stderr_gsource) {
mainloop_del_fd(op->opaque->stderr_gsource);
op->opaque->stderr_gsource = NULL;
}
if (op->opaque->stdout_gsource) {
mainloop_del_fd(op->opaque->stdout_gsource);
op->opaque->stdout_gsource = NULL;
}
}
void
services_action_free(svc_action_t * op)
{
unsigned int i;
if (op == NULL) {
return;
}
/* The operation should be removed from all tracking lists by this point.
* If it's not, we have a bug somewhere, so bail. That may lead to a
* memory leak, but it's better than a use-after-free segmentation fault.
*/
CRM_CHECK(g_list_find(inflight_ops, op) == NULL, return);
CRM_CHECK(g_list_find(blocked_ops, op) == NULL, return);
CRM_CHECK((recurring_actions == NULL)
|| (g_hash_table_lookup(recurring_actions, op->id) == NULL),
return);
services_action_cleanup(op);
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
free(op->id);
free(op->opaque->exec);
for (i = 0; i < PCMK__NELEM(op->opaque->args); i++) {
free(op->opaque->args[i]);
}
free(op->opaque->exit_reason);
free(op->opaque);
free(op->rsc);
free(op->action);
free(op->standard);
free(op->agent);
free(op->provider);
free(op->stdout_data);
free(op->stderr_data);
if (op->params) {
g_hash_table_destroy(op->params);
op->params = NULL;
}
free(op);
}
gboolean
cancel_recurring_action(svc_action_t * op)
{
crm_info("Cancelling %s operation %s", op->standard, op->id);
if (recurring_actions) {
g_hash_table_remove(recurring_actions, op->id);
}
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
return TRUE;
}
/*!
* \brief Cancel a recurring action
*
* \param[in] name Name of resource that operation is for
* \param[in] action Name of operation to cancel
* \param[in] interval_ms Interval of operation to cancel
*
* \return TRUE if action was successfully cancelled, FALSE otherwise
*/
gboolean
services_action_cancel(const char *name, const char *action, guint interval_ms)
{
gboolean cancelled = FALSE;
char *id = pcmk__op_key(name, action, interval_ms);
svc_action_t *op = NULL;
/* We can only cancel a recurring action */
init_recurring_actions();
op = g_hash_table_lookup(recurring_actions, id);
if (op == NULL) {
goto done;
}
// Tell services__finalize_async_op() not to reschedule the operation
op->cancel = TRUE;
/* Stop tracking it as a recurring operation, and stop its repeat timer */
cancel_recurring_action(op);
/* If the op has a PID, it's an in-flight child process, so kill it.
*
* Whether the kill succeeds or fails, the main loop will send the op to
* async_action_complete() (and thus services__finalize_async_op()) when the
* process goes away.
*/
if (op->pid != 0) {
crm_info("Terminating in-flight op %s[%d] early because it was cancelled",
id, op->pid);
cancelled = mainloop_child_kill(op->pid);
if (cancelled == FALSE) {
crm_err("Termination of %s[%d] failed", id, op->pid);
}
goto done;
}
#if SUPPORT_DBUS
// In-flight systemd and upstart ops don't have a pid
if (inflight_systemd_or_upstart(op)) {
inflight_ops = g_list_remove(inflight_ops, op);
/* This will cause any result that comes in later to be discarded, so we
* don't call the callback and free the operation twice.
*/
services_action_cleanup(op);
}
#endif
/* The rest of this is essentially equivalent to
* services__finalize_async_op(), minus the handle_blocked_ops() call.
*/
// Report operation as cancelled
services__set_result(op, op->rc, PCMK_EXEC_CANCELLED, NULL);
if (op->opaque->callback) {
op->opaque->callback(op);
}
blocked_ops = g_list_remove(blocked_ops, op);
services_action_free(op);
cancelled = TRUE;
// @TODO Initiate handle_blocked_ops() asynchronously
done:
free(id);
return cancelled;
}
gboolean
services_action_kick(const char *name, const char *action, guint interval_ms)
{
svc_action_t * op = NULL;
char *id = pcmk__op_key(name, action, interval_ms);
init_recurring_actions();
op = g_hash_table_lookup(recurring_actions, id);
free(id);
if (op == NULL) {
return FALSE;
}
if (op->pid || inflight_systemd_or_upstart(op)) {
return TRUE;
} else {
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
recurring_action_timer(op);
return TRUE;
}
}
/*!
* \internal
* \brief Add a new recurring operation, checking for duplicates
*
* \param[in] op Operation to add
*
* \return TRUE if duplicate found (and reschedule), FALSE otherwise
*/
static gboolean
handle_duplicate_recurring(svc_action_t * op)
{
svc_action_t * dup = NULL;
/* check for duplicates */
dup = g_hash_table_lookup(recurring_actions, op->id);
if (dup && (dup != op)) {
/* update user data */
if (op->opaque->callback) {
dup->opaque->callback = op->opaque->callback;
dup->cb_data = op->cb_data;
op->cb_data = NULL;
}
/* immediately execute the next interval */
if (dup->pid != 0) {
if (op->opaque->repeat_timer) {
g_source_remove(op->opaque->repeat_timer);
op->opaque->repeat_timer = 0;
}
recurring_action_timer(dup);
}
/* free the duplicate */
services_action_free(op);
return TRUE;
}
return FALSE;
}
/*!
* \internal
* \brief Execute an action appropriately according to its standard
*
* \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 already was or is
* pending)
*
* \note If the return value for an asynchronous action is not pcmk_rc_ok, the
* caller is responsible for freeing the action.
*/
static int
execute_action(svc_action_t *op)
{
#if SUPPORT_UPSTART
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_UPSTART,
pcmk__str_casei)) {
return services__execute_upstart(op);
}
#endif
#if SUPPORT_SYSTEMD
if (pcmk__str_eq(op->standard, PCMK_RESOURCE_CLASS_SYSTEMD,
pcmk__str_casei)) {
return services__execute_systemd(op);
}
#endif
return services__execute_file(op);
}
void
services_add_inflight_op(svc_action_t * op)
{
if (op == NULL) {
return;
}
CRM_ASSERT(op->synchronous == FALSE);
/* keep track of ops that are in-flight to avoid collisions in the same namespace */
if (op->rsc) {
inflight_ops = g_list_append(inflight_ops, op);
}
}
/*!
* \internal
* \brief Stop tracking an operation that completed
*
* \param[in] op Operation to stop tracking
*/
void
services_untrack_op(svc_action_t *op)
{
/* Op is no longer in-flight or blocked */
inflight_ops = g_list_remove(inflight_ops, op);
blocked_ops = g_list_remove(blocked_ops, op);
/* Op is no longer blocking other ops, so check if any need to run */
handle_blocked_ops();
}
gboolean
services_action_async_fork_notify(svc_action_t * op,
void (*action_callback) (svc_action_t *),
void (*action_fork_callback) (svc_action_t *))
{
op->synchronous = false;
if (action_callback) {
op->opaque->callback = action_callback;
}
if (action_fork_callback) {
op->opaque->fork_callback = action_fork_callback;
}
if (op->interval_ms > 0) {
init_recurring_actions();
if (handle_duplicate_recurring(op) == TRUE) {
/* entry rescheduled, dup freed */
/* exit early */
return TRUE;
}
g_hash_table_replace(recurring_actions, op->id, op);
}
if (!pcmk_is_set(op->flags, SVC_ACTION_NON_BLOCKED)
&& op->rsc && is_op_blocked(op->rsc)) {
blocked_ops = g_list_append(blocked_ops, op);
return TRUE;
}
return execute_action(op) == pcmk_rc_ok;
}
gboolean
services_action_async(svc_action_t * op,
void (*action_callback) (svc_action_t *))
{
return services_action_async_fork_notify(op, action_callback, NULL);
}
static gboolean processing_blocked_ops = FALSE;
gboolean
is_op_blocked(const char *rsc)
{
GList *gIter = NULL;
svc_action_t *op = NULL;
for (gIter = inflight_ops; gIter != NULL; gIter = gIter->next) {
op = gIter->data;
if (pcmk__str_eq(op->rsc, rsc, pcmk__str_casei)) {
return TRUE;
}
}
return FALSE;
}
static void
handle_blocked_ops(void)
{
GList *executed_ops = NULL;
GList *gIter = NULL;
svc_action_t *op = NULL;
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);
if (execute_action(op) != pcmk_rc_ok) {
/* this can cause this function to be called recursively
* which is why we have processing_blocked_ops static variable */
services__finalize_async_op(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;
}
/*!
* \internal
* \brief Execute a meta-data action appropriately to standard
*
* \param[in] op Meta-data action to execute
*
* \return Standard Pacemaker return code
*/
static int
execute_metadata_action(svc_action_t *op)
{
const char *class = op->standard;
if (op->agent == NULL) {
crm_err("meta-data requested without specifying agent");
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR_FATAL, "Agent not specified");
return EINVAL;
}
if (class == NULL) {
crm_err("meta-data requested for agent %s without specifying class",
op->agent);
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR_FATAL,
"Agent standard not specified");
return EINVAL;
}
if (!strcmp(class, PCMK_RESOURCE_CLASS_SERVICE)) {
class = resources_find_service_class(op->agent);
}
if (class == NULL) {
crm_err("meta-data requested for %s, but could not determine class",
op->agent);
services__set_result(op, services__generic_error(op),
PCMK_EXEC_ERROR_HARD,
"Agent standard could not be determined");
return EINVAL;
}
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) {
return pcmk_legacy2rc(services__get_lsb_metadata(op->agent,
&op->stdout_data));
}
#if SUPPORT_NAGIOS
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
return pcmk_legacy2rc(services__get_nagios_metadata(op->agent,
&op->stdout_data));
}
#endif
return execute_action(op);
}
gboolean
services_action_sync(svc_action_t * op)
{
gboolean rc = TRUE;
if (op == NULL) {
crm_trace("No operation to execute");
return FALSE;
}
op->synchronous = true;
if (pcmk__str_eq(op->action, "meta-data", pcmk__str_casei)) {
/* Synchronous meta-data operations are handled specially. Since most
* resource classes don't provide any meta-data, it has to be
* synthesized from available information about the agent.
*
* services_action_async() doesn't treat meta-data actions specially, so
* it will result in an error for classes that don't support the action.
*/
rc = (execute_metadata_action(op) == pcmk_rc_ok);
} else {
rc = (execute_action(op) == pcmk_rc_ok);
}
crm_trace(" > " PCMK__OP_FMT ": %s = %d",
op->rsc, op->action, op->interval_ms, op->opaque->exec, op->rc);
if (op->stdout_data) {
crm_trace(" > stdout: %s", op->stdout_data);
}
if (op->stderr_data) {
crm_trace(" > stderr: %s", op->stderr_data);
}
return rc;
}
GList *
get_directory_list(const char *root, gboolean files, gboolean executable)
{
return services_os_get_directory_list(root, files, executable);
}
GList *
resources_list_standards(void)
{
GList *standards = NULL;
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_OCF));
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_LSB));
standards = g_list_append(standards, strdup(PCMK_RESOURCE_CLASS_SERVICE));
#if SUPPORT_SYSTEMD
{
GList *agents = systemd_unit_listall();
if (agents != NULL) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_SYSTEMD));
g_list_free_full(agents, free);
}
}
#endif
#if SUPPORT_UPSTART
{
GList *agents = upstart_job_listall();
if (agents != NULL) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_UPSTART));
g_list_free_full(agents, free);
}
}
#endif
#if SUPPORT_NAGIOS
{
GList *agents = services__list_nagios_agents();
if (agents != NULL) {
standards = g_list_append(standards,
strdup(PCMK_RESOURCE_CLASS_NAGIOS));
g_list_free_full(agents, free);
}
}
#endif
return standards;
}
GList *
resources_list_providers(const char *standard)
{
if (pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider)) {
return resources_os_list_ocf_providers();
}
return NULL;
}
GList *
resources_list_agents(const char *standard, const char *provider)
{
if ((standard == NULL)
|| (strcasecmp(standard, PCMK_RESOURCE_CLASS_SERVICE) == 0)) {
GList *tmp1;
GList *tmp2;
GList *result = services__list_lsb_agents();
if (standard == NULL) {
tmp1 = result;
tmp2 = resources_os_list_ocf_agents(NULL);
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
}
#if SUPPORT_SYSTEMD
tmp1 = result;
tmp2 = systemd_unit_listall();
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
#endif
#if SUPPORT_UPSTART
tmp1 = result;
tmp2 = upstart_job_listall();
if (tmp2) {
result = g_list_concat(tmp1, tmp2);
}
#endif
return result;
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_OCF) == 0) {
return resources_os_list_ocf_agents(provider);
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_LSB) == 0) {
return services__list_lsb_agents();
#if SUPPORT_SYSTEMD
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_SYSTEMD) == 0) {
return systemd_unit_listall();
#endif
#if SUPPORT_UPSTART
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_UPSTART) == 0) {
return upstart_job_listall();
#endif
#if SUPPORT_NAGIOS
} else if (strcasecmp(standard, PCMK_RESOURCE_CLASS_NAGIOS) == 0) {
return services__list_nagios_agents();
#endif
}
return NULL;
}
gboolean
resources_agent_exists(const char *standard, const char *provider, const char *agent)
{
GList *standards = NULL;
GList *providers = NULL;
GList *iter = NULL;
gboolean rc = FALSE;
gboolean has_providers = FALSE;
standards = resources_list_standards();
for (iter = standards; iter != NULL; iter = iter->next) {
if (pcmk__str_eq(iter->data, standard, pcmk__str_none)) {
rc = TRUE;
break;
}
}
if (rc == FALSE) {
goto done;
}
rc = FALSE;
has_providers = pcmk_is_set(pcmk_get_ra_caps(standard), pcmk_ra_cap_provider);
if (has_providers == TRUE && provider != NULL) {
providers = resources_list_providers(standard);
for (iter = providers; iter != NULL; iter = iter->next) {
if (pcmk__str_eq(iter->data, provider, pcmk__str_none)) {
rc = TRUE;
break;
}
}
} else if (has_providers == FALSE && provider == NULL) {
rc = TRUE;
}
if (rc == FALSE) {
goto done;
}
if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) {
if (services__lsb_agent_exists(agent)) {
rc = TRUE;
#if SUPPORT_SYSTEMD
} else if (systemd_unit_exists(agent)) {
rc = TRUE;
#endif
#if SUPPORT_UPSTART
} else if (upstart_job_exists(agent)) {
rc = TRUE;
#endif
} else {
rc = FALSE;
}
} else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_OCF, pcmk__str_casei)) {
rc = services__ocf_agent_exists(provider, agent);
} else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_LSB, pcmk__str_casei)) {
rc = services__lsb_agent_exists(agent);
#if SUPPORT_SYSTEMD
} else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) {
rc = systemd_unit_exists(agent);
#endif
#if SUPPORT_UPSTART
} else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_UPSTART, pcmk__str_casei)) {
rc = upstart_job_exists(agent);
#endif
#if SUPPORT_NAGIOS
} else if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
rc = services__nagios_agent_exists(agent);
#endif
} else {
rc = FALSE;
}
done:
g_list_free(standards);
g_list_free(providers);
return rc;
}
/*!
* \internal
* \brief Set the result of an action
*
* \param[out] action Where to set action result
* \param[in] agent_status Exit status to set
* \param[in] exec_status Execution status to set
* \param[in] reason Human-friendly description of event to set
*/
void
services__set_result(svc_action_t *action, int agent_status,
enum pcmk_exec_status exec_status, const char *reason)
{
if (action == NULL) {
return;
}
action->rc = agent_status;
action->status = exec_status;
if (!pcmk__str_eq(action->opaque->exit_reason, reason,
pcmk__str_none)) {
free(action->opaque->exit_reason);
action->opaque->exit_reason = (reason == NULL)? NULL : strdup(reason);
}
}
/*!
* \internal
* \brief Get the exit reason of an action
*
* \param[in] action Action to check
*
* \return Action's exit reason (or NULL if none)
*/
const char *
services__exit_reason(svc_action_t *action)
{
return action->opaque->exit_reason;
}
diff --git a/lib/services/systemd.c b/lib/services/systemd.c
index d3e2eecac4..f832f02494 100644
--- a/lib/services/systemd.c
+++ b/lib/services/systemd.c
@@ -1,1019 +1,1037 @@
/*
* Copyright 2012-2021 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;
+}
+
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)
{
if (name == NULL) {
return NULL;
}
if (systemd_unit_extension(name)) {
return strdup(name);
}
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_err("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");
}
crm_err("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_err("Could not load systemd unit %s for %s: "
"DBus reply has unexpected type", op->agent, op->id);
} else {
crm_err("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__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 return value if non-NULL.
*/
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);
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);
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;
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);
}
meta = crm_strdup_printf(METADATA_FORMAT, name, desc, name);
free(desc);
free(path);
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_warn("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__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");
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);
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_warn("%s operation on 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 action did not complete within specified timeout");
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 already was or is
* pending)
*
* \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/lib/services/systemd.h b/lib/services/systemd.h
index 12ed69f40f..6d5dea9dee 100644
--- a/lib/services/systemd.h
+++ b/lib/services/systemd.h
@@ -1,24 +1,27 @@
/*
* Copyright 2012-2021 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef SYSTEMD__H
# define SYSTEMD__H
# include <glib.h>
# include "crm/services.h"
G_GNUC_INTERNAL GList *systemd_unit_listall(void);
+G_GNUC_INTERNAL
+int services__systemd_prepare(svc_action_t *op);
+
G_GNUC_INTERNAL
int services__execute_systemd(svc_action_t *op);
G_GNUC_INTERNAL gboolean systemd_unit_exists(const gchar * name);
G_GNUC_INTERNAL void systemd_cleanup(void);
#endif /* SYSTEMD__H */