diff --git a/daemons/execd/execd_commands.c b/daemons/execd/execd_commands.c
index 30fdaf5fb1..4669950f71 100644
--- a/daemons/execd/execd_commands.c
+++ b/daemons/execd/execd_commands.c
@@ -1,1953 +1,1955 @@
/*
* Copyright 2012-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <crm/fencing/internal.h>
#include <glib.h>
// Check whether we have a high-resolution monotonic clock
#undef PCMK__TIME_USE_CGT
#if HAVE_DECL_CLOCK_MONOTONIC && defined(CLOCK_MONOTONIC)
# define PCMK__TIME_USE_CGT
# include <time.h> /* clock_gettime */
#endif
#include <unistd.h>
#include <crm/crm.h>
#include <crm/fencing/internal.h>
#include <crm/services.h>
#include <crm/services_internal.h>
#include <crm/common/mainloop.h>
#include <crm/common/ipc.h>
#include <crm/common/ipc_internal.h>
#include <crm/common/xml.h>
#include "pacemaker-execd.h"
GHashTable *rsc_list = NULL;
typedef struct lrmd_cmd_s {
int timeout;
guint interval_ms;
int start_delay;
int timeout_orig;
int call_id;
int call_opts;
/* Timer ids, must be removed on cmd destruction. */
int delay_id;
int stonith_recurring_id;
int rsc_deleted;
int service_flags;
char *client_id;
char *origin;
char *rsc_id;
char *action;
char *real_action;
char *userdata_str;
pcmk__action_result_t result;
/* We can track operation queue time and run time, to be saved with the CIB
* resource history (and displayed in cluster status). We need
* high-resolution monotonic time for this purpose, so we use
* clock_gettime(CLOCK_MONOTONIC, ...) (if available, otherwise this feature
* is disabled).
*
* However, we also need epoch timestamps for recording the time the command
* last ran and the time its return value last changed, for use in time
* displays (as opposed to interval calculations). We keep time_t values for
* this purpose.
*
* The last run time is used for both purposes, so we keep redundant
* monotonic and epoch values for this. Technically the two could represent
* different times, but since time_t has only second resolution and the
* values are used for distinct purposes, that is not significant.
*/
#ifdef PCMK__TIME_USE_CGT
/* Recurring and systemd operations may involve more than one executor
* command per operation, so they need info about the original and the most
* recent.
*/
struct timespec t_first_run; // When op first ran
struct timespec t_run; // When op most recently ran
struct timespec t_first_queue; // When op was first queued
struct timespec t_queue; // When op was most recently queued
#endif
time_t epoch_last_run; // Epoch timestamp of when op last ran
time_t epoch_rcchange; // Epoch timestamp of when rc last changed
bool first_notify_sent;
int last_notify_rc;
int last_notify_op_status;
int last_pid;
GHashTable *params;
} lrmd_cmd_t;
static void cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc);
static gboolean execute_resource_action(gpointer user_data);
static void cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id);
#ifdef PCMK__TIME_USE_CGT
/*!
* \internal
* \brief Check whether a struct timespec has been set
*
* \param[in] timespec Time to check
*
* \return true if timespec has been set (i.e. is nonzero), false otherwise
*/
static inline bool
time_is_set(const struct timespec *timespec)
{
return (timespec != NULL) &&
((timespec->tv_sec != 0) || (timespec->tv_nsec != 0));
}
/*
* \internal
* \brief Set a timespec (and its original if unset) to the current time
*
* \param[out] t_current Where to store current time
* \param[out] t_orig Where to copy t_current if unset
*/
static void
get_current_time(struct timespec *t_current, struct timespec *t_orig)
{
clock_gettime(CLOCK_MONOTONIC, t_current);
if ((t_orig != NULL) && !time_is_set(t_orig)) {
*t_orig = *t_current;
}
}
/*!
* \internal
* \brief Return difference between two times in milliseconds
*
* \param[in] now More recent time (or NULL to use current time)
* \param[in] old Earlier time
*
* \return milliseconds difference (or 0 if old is NULL or unset)
*
* \note Can overflow on 32bit machines when the differences is around
* 24 days or more.
*/
static int
time_diff_ms(const struct timespec *now, const struct timespec *old)
{
int diff_ms = 0;
if (time_is_set(old)) {
struct timespec local_now = { 0, };
if (now == NULL) {
clock_gettime(CLOCK_MONOTONIC, &local_now);
now = &local_now;
}
diff_ms = (now->tv_sec - old->tv_sec) * 1000
+ (now->tv_nsec - old->tv_nsec) / 1000000;
}
return diff_ms;
}
/*!
* \internal
* \brief Reset a command's operation times to their original values.
*
* Reset a command's run and queued timestamps to the timestamps of the original
* command, so we report the entire time since then and not just the time since
* the most recent command (for recurring and systemd operations).
*
* \param[in,out] cmd Executor command object to reset
*
* \note It's not obvious what the queued time should be for a systemd
* start/stop operation, which might go like this:
* initial command queued 5ms, runs 3s
* monitor command queued 10ms, runs 10s
* monitor command queued 10ms, runs 10s
* Is the queued time for that operation 5ms, 10ms or 25ms? The current
* implementation will report 5ms. If it's 25ms, then we need to
* subtract 20ms from the total exec time so as not to count it twice.
* We can implement that later if it matters to anyone ...
*/
static void
cmd_original_times(lrmd_cmd_t * cmd)
{
cmd->t_run = cmd->t_first_run;
cmd->t_queue = cmd->t_first_queue;
}
#endif
static inline bool
action_matches(const lrmd_cmd_t *cmd, const char *action, guint interval_ms)
{
return (cmd->interval_ms == interval_ms)
&& pcmk__str_eq(cmd->action, action, pcmk__str_casei);
}
/*!
* \internal
* \brief Log the result of an asynchronous command
*
* \param[in] cmd Command to log result for
* \param[in] exec_time_ms Execution time in milliseconds, if known
* \param[in] queue_time_ms Queue time in milliseconds, if known
*/
static void
log_finished(const lrmd_cmd_t *cmd, int exec_time_ms, int queue_time_ms)
{
int log_level = LOG_INFO;
GString *str = g_string_sized_new(100); // reasonable starting size
if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
log_level = LOG_DEBUG;
}
g_string_append_printf(str, "%s %s (call %d",
cmd->rsc_id, cmd->action, cmd->call_id);
if (cmd->last_pid != 0) {
g_string_append_printf(str, ", PID %d", cmd->last_pid);
}
if (cmd->result.execution_status == PCMK_EXEC_DONE) {
g_string_append_printf(str, ") exited with status %d",
cmd->result.exit_status);
} else {
pcmk__g_strcat(str, ") could not be executed: ",
pcmk_exec_status_str(cmd->result.execution_status),
NULL);
}
if (cmd->result.exit_reason != NULL) {
pcmk__g_strcat(str, " (", cmd->result.exit_reason, ")", NULL);
}
#ifdef PCMK__TIME_USE_CGT
pcmk__g_strcat(str, " (execution time ",
pcmk__readable_interval(exec_time_ms), NULL);
if (queue_time_ms > 0) {
pcmk__g_strcat(str, " after being queued ",
pcmk__readable_interval(queue_time_ms), NULL);
}
g_string_append_c(str, ')');
#endif
do_crm_log(log_level, "%s", str->str);
g_string_free(str, TRUE);
}
static void
log_execute(lrmd_cmd_t * cmd)
{
int log_level = LOG_INFO;
if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)) {
log_level = LOG_DEBUG;
}
do_crm_log(log_level, "executing - rsc:%s action:%s call_id:%d",
cmd->rsc_id, cmd->action, cmd->call_id);
}
static const char *
normalize_action_name(lrmd_rsc_t * rsc, const char *action)
{
if (pcmk__str_eq(action, PCMK_ACTION_MONITOR, pcmk__str_casei) &&
pcmk_is_set(pcmk_get_ra_caps(rsc->class), pcmk_ra_cap_status)) {
return PCMK_ACTION_STATUS;
}
return action;
}
static lrmd_rsc_t *
build_rsc_from_xml(xmlNode * msg)
{
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR);
lrmd_rsc_t *rsc = NULL;
rsc = calloc(1, sizeof(lrmd_rsc_t));
crm_element_value_int(msg, PCMK__XA_LRMD_CALLOPT, &rsc->call_opts);
rsc->rsc_id = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ID);
rsc->class = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_CLASS);
rsc->provider = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_PROVIDER);
rsc->type = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_TYPE);
rsc->work = mainloop_add_trigger(G_PRIORITY_HIGH, execute_resource_action,
rsc);
// Initialize fence device probes (to return "not running")
pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR,
PCMK_EXEC_NO_FENCE_DEVICE, NULL);
return rsc;
}
static lrmd_cmd_t *
create_lrmd_cmd(xmlNode *msg, pcmk__client_t *client)
{
int call_options = 0;
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, msg, LOG_ERR);
lrmd_cmd_t *cmd = NULL;
cmd = calloc(1, sizeof(lrmd_cmd_t));
crm_element_value_int(msg, PCMK__XA_LRMD_CALLOPT, &call_options);
cmd->call_opts = call_options;
cmd->client_id = strdup(client->id);
crm_element_value_int(msg, PCMK__XA_LRMD_CALLID, &cmd->call_id);
crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &cmd->interval_ms);
crm_element_value_int(rsc_xml, PCMK__XA_LRMD_TIMEOUT, &cmd->timeout);
crm_element_value_int(rsc_xml, F_LRMD_RSC_START_DELAY, &cmd->start_delay);
cmd->timeout_orig = cmd->timeout;
cmd->origin = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_ORIGIN);
cmd->action = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ACTION);
cmd->userdata_str = crm_element_value_copy(rsc_xml,
PCMK__XA_LRMD_RSC_USERDATA_STR);
cmd->rsc_id = crm_element_value_copy(rsc_xml, PCMK__XA_LRMD_RSC_ID);
cmd->params = xml2list(rsc_xml);
if (pcmk__str_eq(g_hash_table_lookup(cmd->params, "CRM_meta_on_fail"),
PCMK_VALUE_BLOCK, pcmk__str_casei)) {
crm_debug("Setting flag to leave pid group on timeout and "
"only kill action pid for " PCMK__OP_FMT,
cmd->rsc_id, cmd->action, cmd->interval_ms);
cmd->service_flags = pcmk__set_flags_as(__func__, __LINE__,
LOG_TRACE, "Action",
cmd->action, 0,
SVC_ACTION_LEAVE_GROUP,
"SVC_ACTION_LEAVE_GROUP");
}
return cmd;
}
static void
stop_recurring_timer(lrmd_cmd_t *cmd)
{
if (cmd) {
if (cmd->stonith_recurring_id) {
g_source_remove(cmd->stonith_recurring_id);
}
cmd->stonith_recurring_id = 0;
}
}
static void
free_lrmd_cmd(lrmd_cmd_t * cmd)
{
stop_recurring_timer(cmd);
if (cmd->delay_id) {
g_source_remove(cmd->delay_id);
}
if (cmd->params) {
g_hash_table_destroy(cmd->params);
}
pcmk__reset_result(&(cmd->result));
free(cmd->origin);
free(cmd->action);
free(cmd->real_action);
free(cmd->userdata_str);
free(cmd->rsc_id);
free(cmd->client_id);
free(cmd);
}
static gboolean
stonith_recurring_op_helper(gpointer data)
{
lrmd_cmd_t *cmd = data;
lrmd_rsc_t *rsc;
cmd->stonith_recurring_id = 0;
if (!cmd->rsc_id) {
return FALSE;
}
rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id);
CRM_ASSERT(rsc != NULL);
/* take it out of recurring_ops list, and put it in the pending ops
* to be executed */
rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd);
rsc->pending_ops = g_list_append(rsc->pending_ops, cmd);
#ifdef PCMK__TIME_USE_CGT
get_current_time(&(cmd->t_queue), &(cmd->t_first_queue));
#endif
mainloop_set_trigger(rsc->work);
return FALSE;
}
static inline void
start_recurring_timer(lrmd_cmd_t *cmd)
{
if (cmd && (cmd->interval_ms > 0)) {
cmd->stonith_recurring_id = g_timeout_add(cmd->interval_ms,
stonith_recurring_op_helper,
cmd);
}
}
static gboolean
start_delay_helper(gpointer data)
{
lrmd_cmd_t *cmd = data;
lrmd_rsc_t *rsc = NULL;
cmd->delay_id = 0;
rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL;
if (rsc) {
mainloop_set_trigger(rsc->work);
}
return FALSE;
}
/*!
* \internal
* \brief Check whether a list already contains the equivalent of a given action
*
* \param[in] action_list List to search
* \param[in] cmd Action to search for
*/
static lrmd_cmd_t *
find_duplicate_action(const GList *action_list, const lrmd_cmd_t *cmd)
{
for (const GList *item = action_list; item != NULL; item = item->next) {
lrmd_cmd_t *dup = item->data;
if (action_matches(cmd, dup->action, dup->interval_ms)) {
return dup;
}
}
return NULL;
}
static bool
merge_recurring_duplicate(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd)
{
lrmd_cmd_t * dup = NULL;
bool dup_pending = true;
if (cmd->interval_ms == 0) {
return false;
}
// Search for a duplicate of this action (in-flight or not)
dup = find_duplicate_action(rsc->pending_ops, cmd);
if (dup == NULL) {
dup_pending = false;
dup = find_duplicate_action(rsc->recurring_ops, cmd);
if (dup == NULL) {
return false;
}
}
/* Do not merge fencing monitors marked for cancellation, so we can reply to
* the cancellation separately.
*/
if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH,
pcmk__str_casei)
&& (dup->result.execution_status == PCMK_EXEC_CANCELLED)) {
return false;
}
/* This should not occur. If it does, we need to investigate how something
* like this is possible in the controller.
*/
crm_warn("Duplicate recurring op entry detected (" PCMK__OP_FMT
"), merging with previous op entry",
rsc->rsc_id, normalize_action_name(rsc, dup->action),
dup->interval_ms);
// Merge new action's call ID and user data into existing action
dup->first_notify_sent = false;
free(dup->userdata_str);
dup->userdata_str = cmd->userdata_str;
cmd->userdata_str = NULL;
dup->call_id = cmd->call_id;
free_lrmd_cmd(cmd);
cmd = NULL;
/* If dup is not pending, that means it has already executed at least once
* and is waiting in the interval. In that case, stop waiting and initiate
* a new instance now.
*/
if (!dup_pending) {
if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH,
pcmk__str_casei)) {
stop_recurring_timer(dup);
stonith_recurring_op_helper(dup);
} else {
services_action_kick(rsc->rsc_id,
normalize_action_name(rsc, dup->action),
dup->interval_ms);
}
}
return true;
}
static void
schedule_lrmd_cmd(lrmd_rsc_t * rsc, lrmd_cmd_t * cmd)
{
CRM_CHECK(cmd != NULL, return);
CRM_CHECK(rsc != NULL, return);
crm_trace("Scheduling %s on %s", cmd->action, rsc->rsc_id);
if (merge_recurring_duplicate(rsc, cmd)) {
// Equivalent of cmd has already been scheduled
return;
}
/* The controller expects the executor to automatically cancel
* recurring operations before a resource stops.
*/
if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
cancel_all_recurring(rsc, NULL);
}
rsc->pending_ops = g_list_append(rsc->pending_ops, cmd);
#ifdef PCMK__TIME_USE_CGT
get_current_time(&(cmd->t_queue), &(cmd->t_first_queue));
#endif
mainloop_set_trigger(rsc->work);
if (cmd->start_delay) {
cmd->delay_id = g_timeout_add(cmd->start_delay, start_delay_helper, cmd);
}
}
static xmlNode *
create_lrmd_reply(const char *origin, int rc, int call_id)
{
xmlNode *reply = create_xml_node(NULL, T_LRMD_REPLY);
crm_xml_add(reply, PCMK__XA_LRMD_ORIGIN, origin);
crm_xml_add_int(reply, PCMK__XA_LRMD_RC, rc);
crm_xml_add_int(reply, PCMK__XA_LRMD_CALLID, call_id);
return reply;
}
static void
send_client_notify(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *update_msg = user_data;
pcmk__client_t *client = value;
int rc;
int log_level = LOG_WARNING;
const char *msg = NULL;
CRM_CHECK(client != NULL, return);
if (client->name == NULL) {
crm_trace("Skipping notification to client without name");
return;
}
if (pcmk_is_set(client->flags, pcmk__client_to_proxy)) {
/* We only want to notify clients of the executor IPC API. If we are
* running as Pacemaker Remote, we may have clients proxied to other
* IPC services in the cluster, so skip those.
*/
crm_trace("Skipping executor API notification to client %s",
pcmk__client_name(client));
return;
}
rc = lrmd_server_send_notify(client, update_msg);
if (rc == pcmk_rc_ok) {
return;
}
switch (rc) {
case ENOTCONN:
case EPIPE: // Client exited without waiting for notification
log_level = LOG_INFO;
msg = "Disconnected";
break;
default:
msg = pcmk_rc_str(rc);
break;
}
do_crm_log(log_level, "Could not notify client %s: %s " CRM_XS " rc=%d",
pcmk__client_name(client), msg, rc);
}
static void
send_cmd_complete_notify(lrmd_cmd_t * cmd)
{
xmlNode *notify = NULL;
int exec_time = 0;
int queue_time = 0;
#ifdef PCMK__TIME_USE_CGT
exec_time = time_diff_ms(NULL, &(cmd->t_run));
queue_time = time_diff_ms(&cmd->t_run, &(cmd->t_queue));
#endif
log_finished(cmd, exec_time, queue_time);
/* If the originator requested to be notified only for changes in recurring
* operation results, skip the notification if the result hasn't changed.
*/
if (cmd->first_notify_sent
&& pcmk_is_set(cmd->call_opts, lrmd_opt_notify_changes_only)
&& (cmd->last_notify_rc == cmd->result.exit_status)
&& (cmd->last_notify_op_status == cmd->result.execution_status)) {
return;
}
cmd->first_notify_sent = true;
cmd->last_notify_rc = cmd->result.exit_status;
cmd->last_notify_op_status = cmd->result.execution_status;
notify = create_xml_node(NULL, T_LRMD_NOTIFY);
crm_xml_add(notify, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add_int(notify, PCMK__XA_LRMD_TIMEOUT, cmd->timeout);
crm_xml_add_ms(notify, F_LRMD_RSC_INTERVAL, cmd->interval_ms);
crm_xml_add_int(notify, F_LRMD_RSC_START_DELAY, cmd->start_delay);
crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_RC, cmd->result.exit_status);
crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_OP_STATUS,
cmd->result.execution_status);
crm_xml_add_int(notify, PCMK__XA_LRMD_CALLID, cmd->call_id);
crm_xml_add_int(notify, F_LRMD_RSC_DELETED, cmd->rsc_deleted);
crm_xml_add_ll(notify, PCMK__XA_LRMD_RUN_TIME,
(long long) cmd->epoch_last_run);
crm_xml_add_ll(notify, PCMK__XA_LRMD_RCCHANGE_TIME,
(long long) cmd->epoch_rcchange);
#ifdef PCMK__TIME_USE_CGT
crm_xml_add_int(notify, PCMK__XA_LRMD_EXEC_TIME, exec_time);
crm_xml_add_int(notify, PCMK__XA_LRMD_QUEUE_TIME, queue_time);
#endif
crm_xml_add(notify, PCMK__XA_LRMD_OP, LRMD_OP_RSC_EXEC);
crm_xml_add(notify, PCMK__XA_LRMD_RSC_ID, cmd->rsc_id);
if(cmd->real_action) {
crm_xml_add(notify, PCMK__XA_LRMD_RSC_ACTION, cmd->real_action);
} else {
crm_xml_add(notify, PCMK__XA_LRMD_RSC_ACTION, cmd->action);
}
crm_xml_add(notify, PCMK__XA_LRMD_RSC_USERDATA_STR, cmd->userdata_str);
crm_xml_add(notify, F_LRMD_RSC_EXIT_REASON, cmd->result.exit_reason);
if (cmd->result.action_stderr != NULL) {
- crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->result.action_stderr);
+ crm_xml_add(notify, PCMK__XA_LRMD_RSC_OUTPUT,
+ cmd->result.action_stderr);
} else if (cmd->result.action_stdout != NULL) {
- crm_xml_add(notify, F_LRMD_RSC_OUTPUT, cmd->result.action_stdout);
+ crm_xml_add(notify, PCMK__XA_LRMD_RSC_OUTPUT,
+ cmd->result.action_stdout);
}
if (cmd->params) {
char *key = NULL;
char *value = NULL;
GHashTableIter iter;
xmlNode *args = create_xml_node(notify, PCMK__XE_ATTRIBUTES);
g_hash_table_iter_init(&iter, cmd->params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
hash2smartfield((gpointer) key, (gpointer) value, args);
}
}
if ((cmd->client_id != NULL)
&& pcmk_is_set(cmd->call_opts, lrmd_opt_notify_orig_only)) {
pcmk__client_t *client = pcmk__find_client_by_id(cmd->client_id);
if (client != NULL) {
send_client_notify(client->id, client, notify);
}
} else {
pcmk__foreach_ipc_client(send_client_notify, notify);
}
free_xml(notify);
}
static void
send_generic_notify(int rc, xmlNode * request)
{
if (pcmk__ipc_client_count() != 0) {
int call_id = 0;
xmlNode *notify = NULL;
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR);
const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
const char *op = crm_element_value(request, PCMK__XA_LRMD_OP);
crm_element_value_int(request, PCMK__XA_LRMD_CALLID, &call_id);
notify = create_xml_node(NULL, T_LRMD_NOTIFY);
crm_xml_add(notify, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add_int(notify, PCMK__XA_LRMD_RC, rc);
crm_xml_add_int(notify, PCMK__XA_LRMD_CALLID, call_id);
crm_xml_add(notify, PCMK__XA_LRMD_OP, op);
crm_xml_add(notify, PCMK__XA_LRMD_RSC_ID, rsc_id);
pcmk__foreach_ipc_client(send_client_notify, notify);
free_xml(notify);
}
}
static void
cmd_reset(lrmd_cmd_t * cmd)
{
cmd->last_pid = 0;
#ifdef PCMK__TIME_USE_CGT
memset(&cmd->t_run, 0, sizeof(cmd->t_run));
memset(&cmd->t_queue, 0, sizeof(cmd->t_queue));
#endif
cmd->epoch_last_run = 0;
pcmk__reset_result(&(cmd->result));
cmd->result.execution_status = PCMK_EXEC_DONE;
}
static void
cmd_finalize(lrmd_cmd_t * cmd, lrmd_rsc_t * rsc)
{
crm_trace("Resource operation rsc:%s action:%s completed (%p %p)", cmd->rsc_id, cmd->action,
rsc ? rsc->active : NULL, cmd);
if (rsc && (rsc->active == cmd)) {
rsc->active = NULL;
mainloop_set_trigger(rsc->work);
}
if (!rsc) {
cmd->rsc_deleted = 1;
}
/* reset original timeout so client notification has correct information */
cmd->timeout = cmd->timeout_orig;
send_cmd_complete_notify(cmd);
if ((cmd->interval_ms != 0)
&& (cmd->result.execution_status == PCMK_EXEC_CANCELLED)) {
if (rsc) {
rsc->recurring_ops = g_list_remove(rsc->recurring_ops, cmd);
rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd);
}
free_lrmd_cmd(cmd);
} else if (cmd->interval_ms == 0) {
if (rsc) {
rsc->pending_ops = g_list_remove(rsc->pending_ops, cmd);
}
free_lrmd_cmd(cmd);
} else {
/* Clear all the values pertaining just to the last iteration of a recurring op. */
cmd_reset(cmd);
}
}
struct notify_new_client_data {
xmlNode *notify;
pcmk__client_t *new_client;
};
static void
notify_one_client(gpointer key, gpointer value, gpointer user_data)
{
pcmk__client_t *client = value;
struct notify_new_client_data *data = user_data;
if (!pcmk__str_eq(client->id, data->new_client->id, pcmk__str_casei)) {
send_client_notify(key, (gpointer) client, (gpointer) data->notify);
}
}
void
notify_of_new_client(pcmk__client_t *new_client)
{
struct notify_new_client_data data;
data.new_client = new_client;
data.notify = create_xml_node(NULL, T_LRMD_NOTIFY);
crm_xml_add(data.notify, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data.notify, PCMK__XA_LRMD_OP, LRMD_OP_NEW_CLIENT);
pcmk__foreach_ipc_client(notify_one_client, &data);
free_xml(data.notify);
}
void
client_disconnect_cleanup(const char *client_id)
{
GHashTableIter iter;
lrmd_rsc_t *rsc = NULL;
char *key = NULL;
g_hash_table_iter_init(&iter, rsc_list);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & rsc)) {
if (pcmk_all_flags_set(rsc->call_opts, lrmd_opt_drop_recurring)) {
/* This client is disconnecting, drop any recurring operations
* it may have initiated on the resource */
cancel_all_recurring(rsc, client_id);
}
}
}
static void
action_complete(svc_action_t * action)
{
lrmd_rsc_t *rsc;
lrmd_cmd_t *cmd = action->cb_data;
enum ocf_exitcode code;
#ifdef PCMK__TIME_USE_CGT
const char *rclass = NULL;
bool goagain = false;
#endif
if (!cmd) {
crm_err("Completed executor action (%s) does not match any known operations",
action->id);
return;
}
#ifdef PCMK__TIME_USE_CGT
if (cmd->result.exit_status != action->rc) {
cmd->epoch_rcchange = time(NULL);
}
#endif
cmd->last_pid = action->pid;
// Cast variable instead of function return to keep compilers happy
code = services_result2ocf(action->standard, cmd->action, action->rc);
pcmk__set_result(&(cmd->result), (int) code,
action->status, services__exit_reason(action));
rsc = cmd->rsc_id ? g_hash_table_lookup(rsc_list, cmd->rsc_id) : NULL;
#ifdef PCMK__TIME_USE_CGT
if (rsc && pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_SERVICE, pcmk__str_casei)) {
rclass = resources_find_service_class(rsc->type);
} else if(rsc) {
rclass = rsc->class;
}
if (pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_SYSTEMD, pcmk__str_casei)) {
if (pcmk__result_ok(&(cmd->result))
&& pcmk__strcase_any_of(cmd->action, PCMK_ACTION_START,
PCMK_ACTION_STOP, NULL)) {
/* systemd returns from start and stop actions after the action
* begins, not after it completes. We have to jump through a few
* hoops so that we don't report 'complete' to the rest of pacemaker
* until it's actually done.
*/
goagain = true;
cmd->real_action = cmd->action;
cmd->action = strdup(PCMK_ACTION_MONITOR);
} else if (cmd->real_action != NULL) {
// This is follow-up monitor to check whether start/stop completed
if (cmd->result.execution_status == PCMK_EXEC_PENDING) {
goagain = true;
} else if (pcmk__result_ok(&(cmd->result))
&& pcmk__str_eq(cmd->real_action, PCMK_ACTION_STOP,
pcmk__str_casei)) {
goagain = true;
} else {
int time_sum = time_diff_ms(NULL, &(cmd->t_first_run));
int timeout_left = cmd->timeout_orig - time_sum;
crm_debug("%s systemd %s is now complete (elapsed=%dms, "
"remaining=%dms): %s (%d)",
cmd->rsc_id, cmd->real_action, time_sum, timeout_left,
services_ocf_exitcode_str(cmd->result.exit_status),
cmd->result.exit_status);
cmd_original_times(cmd);
// Monitors may return "not running", but start/stop shouldn't
if ((cmd->result.execution_status == PCMK_EXEC_DONE)
&& (cmd->result.exit_status == PCMK_OCF_NOT_RUNNING)) {
if (pcmk__str_eq(cmd->real_action, PCMK_ACTION_START,
pcmk__str_casei)) {
cmd->result.exit_status = PCMK_OCF_UNKNOWN_ERROR;
} else if (pcmk__str_eq(cmd->real_action, PCMK_ACTION_STOP,
pcmk__str_casei)) {
cmd->result.exit_status = PCMK_OCF_OK;
}
}
}
}
}
#endif
#if SUPPORT_NAGIOS
if (rsc && pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)) {
if (action_matches(cmd, PCMK_ACTION_MONITOR, 0)
&& pcmk__result_ok(&(cmd->result))) {
/* Successfully executed --version for the nagios plugin */
cmd->result.exit_status = PCMK_OCF_NOT_RUNNING;
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)
&& !pcmk__result_ok(&(cmd->result))) {
#ifdef PCMK__TIME_USE_CGT
goagain = true;
#endif
}
}
#endif
#ifdef PCMK__TIME_USE_CGT
if (goagain) {
int time_sum = time_diff_ms(NULL, &(cmd->t_first_run));
int timeout_left = cmd->timeout_orig - time_sum;
int delay = cmd->timeout_orig / 10;
if(delay >= timeout_left && timeout_left > 20) {
delay = timeout_left/2;
}
delay = QB_MIN(2000, delay);
if (delay < timeout_left) {
cmd->start_delay = delay;
cmd->timeout = timeout_left;
if (pcmk__result_ok(&(cmd->result))) {
crm_debug("%s %s may still be in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)",
cmd->rsc_id, cmd->real_action, time_sum, timeout_left, delay);
} else if (cmd->result.execution_status == PCMK_EXEC_PENDING) {
crm_info("%s %s is still in progress: re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)",
cmd->rsc_id, cmd->action, time_sum, timeout_left, delay);
} else {
crm_notice("%s %s failed '%s' (%d): re-scheduling (elapsed=%dms, remaining=%dms, start_delay=%dms)",
cmd->rsc_id, cmd->action,
services_ocf_exitcode_str(cmd->result.exit_status),
cmd->result.exit_status, time_sum, timeout_left,
delay);
}
cmd_reset(cmd);
if(rsc) {
rsc->active = NULL;
}
schedule_lrmd_cmd(rsc, cmd);
/* Don't finalize cmd, we're not done with it yet */
return;
} else {
crm_notice("Giving up on %s %s (rc=%d): timeout (elapsed=%dms, remaining=%dms)",
cmd->rsc_id,
(cmd->real_action? cmd->real_action : cmd->action),
cmd->result.exit_status, time_sum, timeout_left);
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_TIMEOUT,
"Investigate reason for timeout, and adjust "
"configured operation timeout if necessary");
cmd_original_times(cmd);
}
}
#endif
pcmk__set_result_output(&(cmd->result), services__grab_stdout(action),
services__grab_stderr(action));
cmd_finalize(cmd, rsc);
}
/*!
* \internal
* \brief Process the result of a fence device action (start, stop, or monitor)
*
* \param[in,out] cmd Fence device action that completed
* \param[in] exit_status Fencer API exit status for action
* \param[in] execution_status Fencer API execution status for action
* \param[in] exit_reason Human-friendly detail, if action failed
*/
static void
stonith_action_complete(lrmd_cmd_t *cmd, int exit_status,
enum pcmk_exec_status execution_status,
const char *exit_reason)
{
// This can be NULL if resource was removed before command completed
lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, cmd->rsc_id);
// Simplify fencer exit status to uniform exit status
if (exit_status != CRM_EX_OK) {
exit_status = PCMK_OCF_UNKNOWN_ERROR;
}
if (cmd->result.execution_status == PCMK_EXEC_CANCELLED) {
/* An in-flight fence action was cancelled. The execution status is
* already correct, so don't overwrite it.
*/
execution_status = PCMK_EXEC_CANCELLED;
} else {
/* Some execution status codes have specific meanings for the fencer
* that executor clients may not expect, so map them to a simple error
* status.
*/
switch (execution_status) {
case PCMK_EXEC_NOT_CONNECTED:
case PCMK_EXEC_INVALID:
execution_status = PCMK_EXEC_ERROR;
break;
case PCMK_EXEC_NO_FENCE_DEVICE:
/* This should be possible only for probes in practice, but
* interpret for all actions to be safe.
*/
if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_none)) {
exit_status = PCMK_OCF_NOT_RUNNING;
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP,
pcmk__str_none)) {
exit_status = PCMK_OCF_OK;
} else {
exit_status = PCMK_OCF_NOT_INSTALLED;
}
execution_status = PCMK_EXEC_ERROR;
break;
case PCMK_EXEC_NOT_SUPPORTED:
exit_status = PCMK_OCF_UNIMPLEMENT_FEATURE;
break;
default:
break;
}
}
pcmk__set_result(&cmd->result, exit_status, execution_status, exit_reason);
// Certain successful actions change the known state of the resource
if ((rsc != NULL) && pcmk__result_ok(&(cmd->result))) {
if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) {
pcmk__set_result(&rsc->fence_probe_result, CRM_EX_OK,
PCMK_EXEC_DONE, NULL); // "running"
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP,
pcmk__str_casei)) {
pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR,
PCMK_EXEC_NO_FENCE_DEVICE, NULL); // "not running"
}
}
/* The recurring timer should not be running at this point in any case, but
* as a failsafe, stop it if it is.
*/
stop_recurring_timer(cmd);
/* Reschedule this command if appropriate. If a recurring command is *not*
* rescheduled, its status must be PCMK_EXEC_CANCELLED, otherwise it will
* not be removed from recurring_ops by cmd_finalize().
*/
if (rsc && (cmd->interval_ms > 0)
&& (cmd->result.execution_status != PCMK_EXEC_CANCELLED)) {
start_recurring_timer(cmd);
}
cmd_finalize(cmd, rsc);
}
static void
lrmd_stonith_callback(stonith_t * stonith, stonith_callback_data_t * data)
{
if ((data == NULL) || (data->userdata == NULL)) {
crm_err("Ignoring fence action result: "
"Invalid callback arguments (bug?)");
} else {
stonith_action_complete((lrmd_cmd_t *) data->userdata,
stonith__exit_status(data),
stonith__execution_status(data),
stonith__exit_reason(data));
}
}
void
stonith_connection_failed(void)
{
GHashTableIter iter;
lrmd_rsc_t *rsc = NULL;
crm_warn("Connection to fencer lost (any pending operations for "
"fence devices will be considered failed)");
g_hash_table_iter_init(&iter, rsc_list);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &rsc)) {
if (!pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH,
pcmk__str_none)) {
continue;
}
/* If we registered this fence device, we don't know whether the
* fencer still has the registration or not. Cause future probes to
* return an error until the resource is stopped or started
* successfully. This is especially important if the controller also
* went away (possibly due to a cluster layer restart) and won't
* receive our client notification of any monitors finalized below.
*/
if (rsc->fence_probe_result.execution_status == PCMK_EXEC_DONE) {
pcmk__set_result(&rsc->fence_probe_result, CRM_EX_ERROR,
PCMK_EXEC_NOT_CONNECTED,
"Lost connection to fencer");
}
// Consider any active, pending, or recurring operations as failed
for (GList *op = rsc->recurring_ops; op != NULL; op = op->next) {
lrmd_cmd_t *cmd = op->data;
/* This won't free a recurring op but instead restart its timer.
* If cmd is rsc->active, this will set rsc->active to NULL, so we
* don't have to worry about finalizing it a second time below.
*/
stonith_action_complete(cmd,
CRM_EX_ERROR, PCMK_EXEC_NOT_CONNECTED,
"Lost connection to fencer");
}
if (rsc->active != NULL) {
rsc->pending_ops = g_list_prepend(rsc->pending_ops, rsc->active);
}
while (rsc->pending_ops != NULL) {
// This will free the op and remove it from rsc->pending_ops
stonith_action_complete((lrmd_cmd_t *) rsc->pending_ops->data,
CRM_EX_ERROR, PCMK_EXEC_NOT_CONNECTED,
"Lost connection to fencer");
}
}
}
/*!
* \internal
* \brief Execute a stonith resource "start" action
*
* Start a stonith resource by registering it with the fencer.
* (Stonith agents don't have a start command.)
*
* \param[in,out] stonith_api Connection to fencer
* \param[in] rsc Stonith resource to start
* \param[in] cmd Start command to execute
*
* \return pcmk_ok on success, -errno otherwise
*/
static int
execd_stonith_start(stonith_t *stonith_api, const lrmd_rsc_t *rsc,
const lrmd_cmd_t *cmd)
{
char *key = NULL;
char *value = NULL;
stonith_key_value_t *device_params = NULL;
int rc = pcmk_ok;
// Convert command parameters to stonith API key/values
if (cmd->params) {
GHashTableIter iter;
g_hash_table_iter_init(&iter, cmd->params);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & value)) {
device_params = stonith_key_value_add(device_params, key, value);
}
}
/* The fencer will automatically register devices via CIB notifications
* when the CIB changes, but to avoid a possible race condition between
* the fencer receiving the notification and the executor requesting that
* resource, the executor registers the device as well. The fencer knows how
* to handle duplicate registrations.
*/
rc = stonith_api->cmds->register_device(stonith_api, st_opt_sync_call,
cmd->rsc_id, rsc->provider,
rsc->type, device_params);
stonith_key_value_freeall(device_params, 1, 1);
return rc;
}
/*!
* \internal
* \brief Execute a stonith resource "stop" action
*
* Stop a stonith resource by unregistering it with the fencer.
* (Stonith agents don't have a stop command.)
*
* \param[in,out] stonith_api Connection to fencer
* \param[in] rsc Stonith resource to stop
*
* \return pcmk_ok on success, -errno otherwise
*/
static inline int
execd_stonith_stop(stonith_t *stonith_api, const lrmd_rsc_t *rsc)
{
/* @TODO Failure would indicate a problem communicating with fencer;
* perhaps we should try reconnecting and retrying a few times?
*/
return stonith_api->cmds->remove_device(stonith_api, st_opt_sync_call,
rsc->rsc_id);
}
/*!
* \internal
* \brief Initiate a stonith resource agent recurring "monitor" action
*
* \param[in,out] stonith_api Connection to fencer
* \param[in,out] rsc Stonith resource to monitor
* \param[in] cmd Monitor command being executed
*
* \return pcmk_ok if monitor was successfully initiated, -errno otherwise
*/
static inline int
execd_stonith_monitor(stonith_t *stonith_api, lrmd_rsc_t *rsc, lrmd_cmd_t *cmd)
{
int rc = stonith_api->cmds->monitor(stonith_api, 0, cmd->rsc_id,
cmd->timeout / 1000);
rc = stonith_api->cmds->register_callback(stonith_api, rc, 0, 0, cmd,
"lrmd_stonith_callback",
lrmd_stonith_callback);
if (rc == TRUE) {
rsc->active = cmd;
rc = pcmk_ok;
} else {
rc = -pcmk_err_generic;
}
return rc;
}
static void
execute_stonith_action(lrmd_rsc_t *rsc, lrmd_cmd_t *cmd)
{
int rc = 0;
bool do_monitor = FALSE;
stonith_t *stonith_api = get_stonith_connection();
if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR, pcmk__str_casei)
&& (cmd->interval_ms == 0)) {
// Probes don't require a fencer connection
stonith_action_complete(cmd, rsc->fence_probe_result.exit_status,
rsc->fence_probe_result.execution_status,
rsc->fence_probe_result.exit_reason);
return;
} else if (stonith_api == NULL) {
stonith_action_complete(cmd, PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_NOT_CONNECTED,
"No connection to fencer");
return;
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_START, pcmk__str_casei)) {
rc = execd_stonith_start(stonith_api, rsc, cmd);
if (rc == pcmk_ok) {
do_monitor = TRUE;
}
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
rc = execd_stonith_stop(stonith_api, rsc);
} else if (pcmk__str_eq(cmd->action, PCMK_ACTION_MONITOR,
pcmk__str_casei)) {
do_monitor = TRUE;
} else {
stonith_action_complete(cmd, PCMK_OCF_UNIMPLEMENT_FEATURE,
PCMK_EXEC_ERROR,
"Invalid fence device action (bug?)");
return;
}
if (do_monitor) {
rc = execd_stonith_monitor(stonith_api, rsc, cmd);
if (rc == pcmk_ok) {
// Don't clean up yet, we will find out result of the monitor later
return;
}
}
stonith_action_complete(cmd,
((rc == pcmk_ok)? CRM_EX_OK : CRM_EX_ERROR),
stonith__legacy2status(rc),
((rc == -pcmk_err_generic)? NULL : pcmk_strerror(rc)));
}
static void
execute_nonstonith_action(lrmd_rsc_t *rsc, lrmd_cmd_t *cmd)
{
svc_action_t *action = NULL;
GHashTable *params_copy = NULL;
CRM_ASSERT(rsc);
CRM_ASSERT(cmd);
crm_trace("Creating action, resource:%s action:%s class:%s provider:%s agent:%s",
rsc->rsc_id, cmd->action, rsc->class, rsc->provider, rsc->type);
#if SUPPORT_NAGIOS
/* Recurring operations are cancelled anyway for a stop operation */
if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_NAGIOS, pcmk__str_casei)
&& pcmk__str_eq(cmd->action, PCMK_ACTION_STOP, pcmk__str_casei)) {
cmd->result.exit_status = PCMK_OCF_OK;
cmd_finalize(cmd, rsc);
return;
}
#endif
params_copy = pcmk__str_table_dup(cmd->params);
action = services__create_resource_action(rsc->rsc_id, rsc->class, rsc->provider,
rsc->type,
normalize_action_name(rsc, cmd->action),
cmd->interval_ms, cmd->timeout,
params_copy, cmd->service_flags);
if (action == NULL) {
pcmk__set_result(&(cmd->result), PCMK_OCF_UNKNOWN_ERROR,
PCMK_EXEC_ERROR, strerror(ENOMEM));
cmd_finalize(cmd, rsc);
return;
}
if (action->rc != PCMK_OCF_UNKNOWN) {
pcmk__set_result(&(cmd->result), action->rc, action->status,
services__exit_reason(action));
services_action_free(action);
cmd_finalize(cmd, rsc);
return;
}
action->cb_data = cmd;
if (services_action_async(action, action_complete)) {
/* The services library has taken responsibility for the action. It
* could be pending, blocked, or merged into a duplicate recurring
* action, in which case the action callback (action_complete())
* will be called when the action completes, otherwise the callback has
* already been called.
*
* action_complete() calls cmd_finalize() which can free cmd, so cmd
* cannot be used here.
*/
} else {
/* This is a recurring action that is not being cancelled and could not
* be initiated. It has been rescheduled, and the action callback
* (action_complete()) has been called, which in this case has already
* called cmd_finalize(), which in this case should only reset (not
* free) cmd.
*/
pcmk__set_result(&(cmd->result), action->rc, action->status,
services__exit_reason(action));
services_action_free(action);
}
}
static gboolean
execute_resource_action(gpointer user_data)
{
lrmd_rsc_t *rsc = (lrmd_rsc_t *) user_data;
lrmd_cmd_t *cmd = NULL;
CRM_CHECK(rsc != NULL, return FALSE);
if (rsc->active) {
crm_trace("%s is still active", rsc->rsc_id);
return TRUE;
}
if (rsc->pending_ops) {
GList *first = rsc->pending_ops;
cmd = first->data;
if (cmd->delay_id) {
crm_trace
("Command %s %s was asked to run too early, waiting for start_delay timeout of %dms",
cmd->rsc_id, cmd->action, cmd->start_delay);
return TRUE;
}
rsc->pending_ops = g_list_remove_link(rsc->pending_ops, first);
g_list_free_1(first);
#ifdef PCMK__TIME_USE_CGT
get_current_time(&(cmd->t_run), &(cmd->t_first_run));
#endif
cmd->epoch_last_run = time(NULL);
}
if (!cmd) {
crm_trace("Nothing further to do for %s", rsc->rsc_id);
return TRUE;
}
rsc->active = cmd; /* only one op at a time for a rsc */
if (cmd->interval_ms) {
rsc->recurring_ops = g_list_append(rsc->recurring_ops, cmd);
}
log_execute(cmd);
if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
execute_stonith_action(rsc, cmd);
} else {
execute_nonstonith_action(rsc, cmd);
}
return TRUE;
}
void
free_rsc(gpointer data)
{
GList *gIter = NULL;
lrmd_rsc_t *rsc = data;
int is_stonith = pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH,
pcmk__str_casei);
gIter = rsc->pending_ops;
while (gIter != NULL) {
GList *next = gIter->next;
lrmd_cmd_t *cmd = gIter->data;
/* command was never executed */
cmd->result.execution_status = PCMK_EXEC_CANCELLED;
cmd_finalize(cmd, NULL);
gIter = next;
}
/* frees list, but not list elements. */
g_list_free(rsc->pending_ops);
gIter = rsc->recurring_ops;
while (gIter != NULL) {
GList *next = gIter->next;
lrmd_cmd_t *cmd = gIter->data;
if (is_stonith) {
cmd->result.execution_status = PCMK_EXEC_CANCELLED;
/* If a stonith command is in-flight, just mark it as cancelled;
* it is not safe to finalize/free the cmd until the stonith api
* says it has either completed or timed out.
*/
if (rsc->active != cmd) {
cmd_finalize(cmd, NULL);
}
} else {
/* This command is already handed off to service library,
* let service library cancel it and tell us via the callback
* when it is cancelled. The rsc can be safely destroyed
* even if we are waiting for the cancel result */
services_action_cancel(rsc->rsc_id,
normalize_action_name(rsc, cmd->action),
cmd->interval_ms);
}
gIter = next;
}
/* frees list, but not list elements. */
g_list_free(rsc->recurring_ops);
free(rsc->rsc_id);
free(rsc->class);
free(rsc->provider);
free(rsc->type);
mainloop_destroy_trigger(rsc->work);
free(rsc);
}
static int
process_lrmd_signon(pcmk__client_t *client, xmlNode *request, int call_id,
xmlNode **reply)
{
int rc = pcmk_ok;
time_t now = time(NULL);
const char *protocol_version =
crm_element_value(request, PCMK__XA_LRMD_PROTOCOL_VERSION);
const char *start_state = pcmk__env_option(PCMK__ENV_NODE_START_STATE);
if (compare_version(protocol_version, LRMD_COMPATIBLE_PROTOCOL) < 0) {
crm_err("Cluster API version must be greater than or equal to %s, not %s",
LRMD_COMPATIBLE_PROTOCOL, protocol_version);
rc = -EPROTO;
}
if (pcmk__xe_attr_is_true(request, PCMK__XA_LRMD_IS_IPC_PROVIDER)) {
#ifdef PCMK__COMPILE_REMOTE
if ((client->remote != NULL)
&& pcmk_is_set(client->flags,
pcmk__client_tls_handshake_complete)) {
const char *op = crm_element_value(request, PCMK__XA_LRMD_OP);
// This is a remote connection from a cluster node's controller
ipc_proxy_add_provider(client);
/* If this was a register operation, also ask for new schema files but
* only if it's supported by the protocol version.
*/
if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none) &&
LRMD_SUPPORTS_SCHEMA_XFER(protocol_version)) {
remoted_request_cib_schema_files();
}
} else {
rc = -EACCES;
}
#else
rc = -EPROTONOSUPPORT;
#endif
}
*reply = create_lrmd_reply(__func__, rc, call_id);
crm_xml_add(*reply, PCMK__XA_LRMD_OP, CRM_OP_REGISTER);
crm_xml_add(*reply, PCMK__XA_LRMD_CLIENTID, client->id);
crm_xml_add(*reply, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION);
crm_xml_add_ll(*reply, PCMK__XA_UPTIME, now - start_time);
if (start_state) {
crm_xml_add(*reply, PCMK__XA_NODE_START_STATE, start_state);
}
return rc;
}
static int
process_lrmd_rsc_register(pcmk__client_t *client, uint32_t id, xmlNode *request)
{
int rc = pcmk_ok;
lrmd_rsc_t *rsc = build_rsc_from_xml(request);
lrmd_rsc_t *dup = g_hash_table_lookup(rsc_list, rsc->rsc_id);
if (dup &&
pcmk__str_eq(rsc->class, dup->class, pcmk__str_casei) &&
pcmk__str_eq(rsc->provider, dup->provider, pcmk__str_casei) && pcmk__str_eq(rsc->type, dup->type, pcmk__str_casei)) {
crm_notice("Ignoring duplicate registration of '%s'", rsc->rsc_id);
free_rsc(rsc);
return rc;
}
g_hash_table_replace(rsc_list, rsc->rsc_id, rsc);
crm_info("Cached agent information for '%s'", rsc->rsc_id);
return rc;
}
static xmlNode *
process_lrmd_get_rsc_info(xmlNode *request, int call_id)
{
int rc = pcmk_ok;
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR);
const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
xmlNode *reply = NULL;
lrmd_rsc_t *rsc = NULL;
if (rsc_id == NULL) {
rc = -ENODEV;
} else {
rsc = g_hash_table_lookup(rsc_list, rsc_id);
if (rsc == NULL) {
crm_info("Agent information for '%s' not in cache", rsc_id);
rc = -ENODEV;
}
}
reply = create_lrmd_reply(__func__, rc, call_id);
if (rsc) {
crm_xml_add(reply, PCMK__XA_LRMD_RSC_ID, rsc->rsc_id);
crm_xml_add(reply, PCMK__XA_LRMD_CLASS, rsc->class);
crm_xml_add(reply, PCMK__XA_LRMD_PROVIDER, rsc->provider);
crm_xml_add(reply, PCMK__XA_LRMD_TYPE, rsc->type);
}
return reply;
}
static int
process_lrmd_rsc_unregister(pcmk__client_t *client, uint32_t id,
xmlNode *request)
{
int rc = pcmk_ok;
lrmd_rsc_t *rsc = NULL;
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR);
const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
if (!rsc_id) {
return -ENODEV;
}
rsc = g_hash_table_lookup(rsc_list, rsc_id);
if (rsc == NULL) {
crm_info("Ignoring unregistration of resource '%s', which is not registered",
rsc_id);
return pcmk_ok;
}
if (rsc->active) {
/* let the caller know there are still active ops on this rsc to watch for */
crm_trace("Operation (%p) still in progress for unregistered resource %s",
rsc->active, rsc_id);
rc = -EINPROGRESS;
}
g_hash_table_remove(rsc_list, rsc_id);
return rc;
}
static int
process_lrmd_rsc_exec(pcmk__client_t *client, uint32_t id, xmlNode *request)
{
lrmd_rsc_t *rsc = NULL;
lrmd_cmd_t *cmd = NULL;
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR);
const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
int call_id;
if (!rsc_id) {
return -EINVAL;
}
if (!(rsc = g_hash_table_lookup(rsc_list, rsc_id))) {
crm_info("Resource '%s' not found (%d active resources)",
rsc_id, g_hash_table_size(rsc_list));
return -ENODEV;
}
cmd = create_lrmd_cmd(request, client);
call_id = cmd->call_id;
/* Don't reference cmd after handing it off to be scheduled.
* The cmd could get merged and freed. */
schedule_lrmd_cmd(rsc, cmd);
return call_id;
}
static int
cancel_op(const char *rsc_id, const char *action, guint interval_ms)
{
GList *gIter = NULL;
lrmd_rsc_t *rsc = g_hash_table_lookup(rsc_list, rsc_id);
/* How to cancel an action.
* 1. Check pending ops list, if it hasn't been handed off
* to the service library or stonith recurring list remove
* it there and that will stop it.
* 2. If it isn't in the pending ops list, then it's either a
* recurring op in the stonith recurring list, or the service
* library's recurring list. Stop it there
* 3. If not found in any lists, then this operation has either
* been executed already and is not a recurring operation, or
* never existed.
*/
if (!rsc) {
return -ENODEV;
}
for (gIter = rsc->pending_ops; gIter != NULL; gIter = gIter->next) {
lrmd_cmd_t *cmd = gIter->data;
if (action_matches(cmd, action, interval_ms)) {
cmd->result.execution_status = PCMK_EXEC_CANCELLED;
cmd_finalize(cmd, rsc);
return pcmk_ok;
}
}
if (pcmk__str_eq(rsc->class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
/* The service library does not handle stonith operations.
* We have to handle recurring stonith operations ourselves. */
for (gIter = rsc->recurring_ops; gIter != NULL; gIter = gIter->next) {
lrmd_cmd_t *cmd = gIter->data;
if (action_matches(cmd, action, interval_ms)) {
cmd->result.execution_status = PCMK_EXEC_CANCELLED;
if (rsc->active != cmd) {
cmd_finalize(cmd, rsc);
}
return pcmk_ok;
}
}
} else if (services_action_cancel(rsc_id,
normalize_action_name(rsc, action),
interval_ms) == TRUE) {
/* The service library will tell the action_complete callback function
* this action was cancelled, which will destroy the cmd and remove
* it from the recurring_op list. Do not do that in this function
* if the service library says it cancelled it. */
return pcmk_ok;
}
return -EOPNOTSUPP;
}
static void
cancel_all_recurring(lrmd_rsc_t * rsc, const char *client_id)
{
GList *cmd_list = NULL;
GList *cmd_iter = NULL;
/* Notice a copy of each list is created when concat is called.
* This prevents odd behavior from occurring when the cmd_list
* is iterated through later on. It is possible the cancel_op
* function may end up modifying the recurring_ops and pending_ops
* lists. If we did not copy those lists, our cmd_list iteration
* could get messed up.*/
if (rsc->recurring_ops) {
cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->recurring_ops));
}
if (rsc->pending_ops) {
cmd_list = g_list_concat(cmd_list, g_list_copy(rsc->pending_ops));
}
if (!cmd_list) {
return;
}
for (cmd_iter = cmd_list; cmd_iter; cmd_iter = cmd_iter->next) {
lrmd_cmd_t *cmd = cmd_iter->data;
if (cmd->interval_ms == 0) {
continue;
}
if (client_id && !pcmk__str_eq(cmd->client_id, client_id, pcmk__str_casei)) {
continue;
}
cancel_op(rsc->rsc_id, cmd->action, cmd->interval_ms);
}
/* frees only the copied list data, not the cmds */
g_list_free(cmd_list);
}
static int
process_lrmd_rsc_cancel(pcmk__client_t *client, uint32_t id, xmlNode *request)
{
xmlNode *rsc_xml = get_xpath_object("//" F_LRMD_RSC, request, LOG_ERR);
const char *rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
const char *action = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ACTION);
guint interval_ms = 0;
crm_element_value_ms(rsc_xml, F_LRMD_RSC_INTERVAL, &interval_ms);
if (!rsc_id || !action) {
return -EINVAL;
}
return cancel_op(rsc_id, action, interval_ms);
}
static void
add_recurring_op_xml(xmlNode *reply, lrmd_rsc_t *rsc)
{
xmlNode *rsc_xml = create_xml_node(reply, F_LRMD_RSC);
crm_xml_add(rsc_xml, PCMK__XA_LRMD_RSC_ID, rsc->rsc_id);
for (GList *item = rsc->recurring_ops; item != NULL; item = item->next) {
lrmd_cmd_t *cmd = item->data;
xmlNode *op_xml = create_xml_node(rsc_xml, T_LRMD_RSC_OP);
crm_xml_add(op_xml, PCMK__XA_LRMD_RSC_ACTION,
pcmk__s(cmd->real_action, cmd->action));
crm_xml_add_ms(op_xml, F_LRMD_RSC_INTERVAL, cmd->interval_ms);
crm_xml_add_int(op_xml, PCMK__XA_LRMD_TIMEOUT, cmd->timeout_orig);
}
}
static xmlNode *
process_lrmd_get_recurring(xmlNode *request, int call_id)
{
int rc = pcmk_ok;
const char *rsc_id = NULL;
lrmd_rsc_t *rsc = NULL;
xmlNode *reply = NULL;
xmlNode *rsc_xml = NULL;
// Resource ID is optional
rsc_xml = first_named_child(request, PCMK__XA_LRMD_CALLDATA);
if (rsc_xml) {
rsc_xml = first_named_child(rsc_xml, F_LRMD_RSC);
}
if (rsc_xml) {
rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
}
// If resource ID is specified, resource must exist
if (rsc_id != NULL) {
rsc = g_hash_table_lookup(rsc_list, rsc_id);
if (rsc == NULL) {
crm_info("Resource '%s' not found (%d active resources)",
rsc_id, g_hash_table_size(rsc_list));
rc = -ENODEV;
}
}
reply = create_lrmd_reply(__func__, rc, call_id);
// If resource ID is not specified, check all resources
if (rsc_id == NULL) {
GHashTableIter iter;
char *key = NULL;
g_hash_table_iter_init(&iter, rsc_list);
while (g_hash_table_iter_next(&iter, (gpointer *) &key,
(gpointer *) &rsc)) {
add_recurring_op_xml(reply, rsc);
}
} else if (rsc) {
add_recurring_op_xml(reply, rsc);
}
return reply;
}
void
process_lrmd_message(pcmk__client_t *client, uint32_t id, xmlNode *request)
{
int rc = pcmk_ok;
int call_id = 0;
const char *op = crm_element_value(request, PCMK__XA_LRMD_OP);
int do_reply = 0;
int do_notify = 0;
xmlNode *reply = NULL;
/* Certain IPC commands may be done only by privileged users (i.e. root or
* hacluster), because they would otherwise provide a means of bypassing
* ACLs.
*/
bool allowed = pcmk_is_set(client->flags, pcmk__client_privileged);
crm_trace("Processing %s operation from %s", op, client->id);
crm_element_value_int(request, PCMK__XA_LRMD_CALLID, &call_id);
if (pcmk__str_eq(op, CRM_OP_IPC_FWD, pcmk__str_none)) {
#ifdef PCMK__COMPILE_REMOTE
if (allowed) {
ipc_proxy_forward_client(client, request);
} else {
rc = -EACCES;
}
#else
rc = -EPROTONOSUPPORT;
#endif
do_reply = 1;
} else if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none)) {
rc = process_lrmd_signon(client, request, call_id, &reply);
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_RSC_REG, pcmk__str_none)) {
if (allowed) {
rc = process_lrmd_rsc_register(client, id, request);
do_notify = 1;
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_RSC_INFO, pcmk__str_none)) {
if (allowed) {
reply = process_lrmd_get_rsc_info(request, call_id);
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_RSC_UNREG, pcmk__str_none)) {
if (allowed) {
rc = process_lrmd_rsc_unregister(client, id, request);
/* don't notify anyone about failed un-registers */
if (rc == pcmk_ok || rc == -EINPROGRESS) {
do_notify = 1;
}
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_RSC_EXEC, pcmk__str_none)) {
if (allowed) {
rc = process_lrmd_rsc_exec(client, id, request);
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_RSC_CANCEL, pcmk__str_none)) {
if (allowed) {
rc = process_lrmd_rsc_cancel(client, id, request);
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_POKE, pcmk__str_none)) {
do_notify = 1;
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_CHECK, pcmk__str_none)) {
if (allowed) {
xmlNode *data = get_message_xml(request, PCMK__XA_LRMD_CALLDATA);
const char *timeout = NULL;
CRM_LOG_ASSERT(data != NULL);
timeout = crm_element_value(data, PCMK__XA_LRMD_WATCHDOG);
pcmk__valid_stonith_watchdog_timeout(timeout);
} else {
rc = -EACCES;
}
} else if (pcmk__str_eq(op, LRMD_OP_ALERT_EXEC, pcmk__str_none)) {
if (allowed) {
rc = process_lrmd_alert_exec(client, id, request);
} else {
rc = -EACCES;
}
do_reply = 1;
} else if (pcmk__str_eq(op, LRMD_OP_GET_RECURRING, pcmk__str_none)) {
if (allowed) {
reply = process_lrmd_get_recurring(request, call_id);
} else {
rc = -EACCES;
}
do_reply = 1;
} else {
rc = -EOPNOTSUPP;
do_reply = 1;
crm_err("Unknown IPC request '%s' from client %s",
op, pcmk__client_name(client));
}
if (rc == -EACCES) {
crm_warn("Rejecting IPC request '%s' from unprivileged client %s",
op, pcmk__client_name(client));
}
crm_debug("Processed %s operation from %s: rc=%d, reply=%d, notify=%d",
op, client->id, rc, do_reply, do_notify);
if (do_reply) {
int send_rc = pcmk_rc_ok;
if (reply == NULL) {
reply = create_lrmd_reply(__func__, rc, call_id);
}
send_rc = lrmd_server_send_reply(client, id, reply);
free_xml(reply);
if (send_rc != pcmk_rc_ok) {
crm_warn("Reply to client %s failed: %s " CRM_XS " rc=%d",
pcmk__client_name(client), pcmk_rc_str(send_rc), send_rc);
}
}
if (do_notify) {
send_generic_notify(rc, request);
}
}
diff --git a/include/crm/common/xml_names_internal.h b/include/crm/common/xml_names_internal.h
index 617ae75d1e..3479db87e6 100644
--- a/include/crm/common/xml_names_internal.h
+++ b/include/crm/common/xml_names_internal.h
@@ -1,326 +1,327 @@
/*
* Copyright 2004-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#ifndef PCMK__CRM_COMMON_XML_NAMES_INTERNAL__H
# define PCMK__CRM_COMMON_XML_NAMES_INTERNAL__H
#ifdef __cplusplus
extern "C" {
#endif
/*
* XML element names used only by internal code
*/
#define PCMK__XE_ACK "ack"
#define PCMK__XE_ATTRIBUTES "attributes"
#define PCMK__XE_CIB_CALLBACK "cib-callback"
#define PCMK__XE_CIB_COMMAND "cib_command"
#define PCMK__XE_CIB_REPLY "cib-reply"
#define PCMK__XE_CIB_RESULT "cib_result"
#define PCMK__XE_CIB_TRANSACTION "cib_transaction"
#define PCMK__XE_COPY "copy"
#define PCMK__XE_CRM_EVENT "crm_event"
#define PCMK__XE_CRM_XML "crm_xml"
#define PCMK__XE_DIV "div"
#define PCMK__XE_DOWNED "downed"
#define PCMK__XE_EXIT_NOTIFICATION "exit-notification"
#define PCMK__XE_FAILED "failed"
#define PCMK__XE_FAILED_UPDATE "failed_update"
#define PCMK__XE_GENERATION_TUPLE "generation_tuple"
#define PCMK__XE_LRM "lrm"
#define PCMK__XE_LRM_RESOURCE "lrm_resource"
#define PCMK__XE_LRM_RESOURCES "lrm_resources"
#define PCMK__XE_LRM_RSC_OP "lrm_rsc_op"
#define PCMK__XE_MAINTENANCE "maintenance"
#define PCMK__XE_META "meta"
#define PCMK__XE_NACK "nack"
#define PCMK__XE_NODE_STATE "node_state"
#define PCMK__XE_NOTIFY "notify"
#define PCMK__XE_OPTIONS "options"
#define PCMK__XE_PARAM "param"
#define PCMK__XE_PING "ping"
#define PCMK__XE_PING_RESPONSE "ping_response"
#define PCMK__XE_PSEUDO_EVENT "pseudo_event"
#define PCMK__XE_RSC_OP "rsc_op"
#define PCMK__XE_SHUTDOWN "shutdown"
#define PCMK__XE_SPAN "span"
#define PCMK__XE_ST_ASYNC_TIMEOUT_VALUE "st-async-timeout-value"
#define PCMK__XE_ST_DEVICE_ACTION "st_device_action"
#define PCMK__XE_ST_DEVICE_ID "st_device_id"
#define PCMK__XE_ST_HISTORY "st_history"
#define PCMK__XE_ST_REPLY "st-reply"
#define PCMK__XE_STONITH_COMMAND "stonith_command"
#define PCMK__XE_TICKET_STATE "ticket_state"
#define PCMK__XE_TRANSIENT_ATTRIBUTES "transient_attributes"
#define PCMK__XE_TRANSITION_GRAPH "transition_graph"
#define PCMK__XE_XPATH_QUERY "xpath-query"
#define PCMK__XE_XPATH_QUERY_PATH "xpath-query-path"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XE_ACL_USER "acl_user"
/* @COMPAT Deprecate somehow. It's undocumented and behaves the same as
* PCMK__XE_CIB in places where it's recognized.
*/
#define PCMK__XE_ALL "all"
// @COMPAT Deprecated since 2.1.8
#define PCMK__XE_CIB_GENERATION "cib_generation"
// @COMPAT Deprecated since 2.1.8
#define PCMK__XE_CIB_UPDATE "cib_update"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_ADDED "diff-added"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XE_DIFF_REMOVED "diff-removed"
// @COMPAT Deprecated since 1.0.8 (commit 4cb100f)
#define PCMK__XE_LIFETIME "lifetime"
/* @COMPAT Deprecated since 2.0.0; alias for <clone> with PCMK_META_PROMOTABLE
* set to "true"
*/
#define PCMK__XE_PROMOTABLE_LEGACY "master"
// @COMPAT Support for rkt is deprecated since 2.1.8
#define PCMK__XE_RKT "rkt"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XE_ROLE_REF "role_ref"
/*
* XML attribute names used only by internal code
*/
#define PCMK__XA_ATTR_CLEAR_INTERVAL "attr_clear_interval"
#define PCMK__XA_ATTR_CLEAR_OPERATION "attr_clear_operation"
#define PCMK__XA_ATTR_DAMPENING "attr_dampening"
#define PCMK__XA_ATTR_HOST "attr_host"
#define PCMK__XA_ATTR_HOST_ID "attr_host_id"
#define PCMK__XA_ATTR_IS_PRIVATE "attr_is_private"
#define PCMK__XA_ATTR_IS_REMOTE "attr_is_remote"
#define PCMK__XA_ATTR_NAME "attr_name"
#define PCMK__XA_ATTR_REGEX "attr_regex"
#define PCMK__XA_ATTR_RESOURCE "attr_resource"
#define PCMK__XA_ATTR_SECTION "attr_section"
#define PCMK__XA_ATTR_SET "attr_set"
#define PCMK__XA_ATTR_SET_TYPE "attr_set_type"
#define PCMK__XA_ATTR_SYNC_POINT "attr_sync_point"
#define PCMK__XA_ATTR_USER "attr_user"
#define PCMK__XA_ATTR_VALUE "attr_value"
#define PCMK__XA_ATTR_VERSION "attr_version"
#define PCMK__XA_ATTR_WRITER "attr_writer"
#define PCMK__XA_ATTRD_IS_FORCE_WRITE "attrd_is_force_write"
#define PCMK__XA_CALL_ID "call-id"
#define PCMK__XA_CIB_CALLDATA "cib_calldata"
#define PCMK__XA_CIB_CALLID "cib_callid"
#define PCMK__XA_CIB_CALLOPT "cib_callopt"
#define PCMK__XA_CIB_CLIENTID "cib_clientid"
#define PCMK__XA_CIB_CLIENTNAME "cib_clientname"
#define PCMK__XA_CIB_DELEGATED_FROM "cib_delegated_from"
#define PCMK__XA_CIB_HOST "cib_host"
#define PCMK__XA_CIB_ISREPLYTO "cib_isreplyto"
#define PCMK__XA_CIB_NOTIFY_ACTIVATE "cib_notify_activate"
#define PCMK__XA_CIB_NOTIFY_TYPE "cib_notify_type"
#define PCMK__XA_CIB_OP "cib_op"
#define PCMK__XA_CIB_PING_ID "cib_ping_id"
#define PCMK__XA_CIB_RC "cib_rc"
#define PCMK__XA_CIB_SCHEMA_MAX "cib_schema_max"
#define PCMK__XA_CIB_SECTION "cib_section"
#define PCMK__XA_CIB_UPDATE "cib_update"
#define PCMK__XA_CIB_UPDATE_RESULT "cib_update_result"
#define PCMK__XA_CIB_UPGRADE_RC "cib_upgrade_rc"
#define PCMK__XA_CIB_USER "cib_user"
#define PCMK__XA_CLIENT_NAME "client_name"
#define PCMK__XA_CLIENT_UUID "client_uuid"
#define PCMK__XA_CONFIG_ERRORS "config-errors"
#define PCMK__XA_CONFIG_WARNINGS "config-warnings"
#define PCMK__XA_CONFIRM "confirm"
#define PCMK__XA_CONNECTION_HOST "connection_host"
#define PCMK__XA_CONTENT "content"
#define PCMK__XA_CRMD_STATE "crmd_state"
#define PCMK__XA_CRM_HOST_TO "crm_host_to"
#define PCMK__XA_CRM_LIMIT_MAX "crm-limit-max"
#define PCMK__XA_CRM_LIMIT_MODE "crm-limit-mode"
#define PCMK__XA_CRM_SUBSYSTEM "crm_subsystem"
#define PCMK__XA_CRM_SYS_FROM "crm_sys_from"
#define PCMK__XA_CRM_SYS_TO "crm_sys_to"
#define PCMK__XA_CRM_TASK "crm_task"
#define PCMK__XA_CRM_TGRAPH_IN "crm-tgraph-in"
#define PCMK__XA_CRM_USER "crm_user"
#define PCMK__XA_DC_LEAVING "dc-leaving"
#define PCMK__XA_DIGEST "digest"
#define PCMK__XA_ELECTION_AGE_SEC "election-age-sec"
#define PCMK__XA_ELECTION_AGE_NANO_SEC "election-age-nano-sec"
#define PCMK__XA_ELECTION_ID "election-id"
#define PCMK__XA_ELECTION_OWNER "election-owner"
#define PCMK__XA_GRANTED "granted"
#define PCMK__XA_GRAPH_ERRORS "graph-errors"
#define PCMK__XA_GRAPH_WARNINGS "graph-warnings"
#define PCMK__XA_HIDDEN "hidden"
#define PCMK__XA_HTTP_EQUIV "http-equiv"
#define PCMK__XA_IN_CCM "in_ccm"
#define PCMK__XA_JOIN "join"
#define PCMK__XA_JOIN_ID "join_id"
#define PCMK__XA_LINE "line"
#define PCMK__XA_LONG_ID "long-id"
#define PCMK__XA_LRMD_CALLDATA "lrmd_calldata"
#define PCMK__XA_LRMD_CALLID "lrmd_callid"
#define PCMK__XA_LRMD_CALLOPT "lrmd_callopt"
#define PCMK__XA_LRMD_CLASS "lrmd_class"
#define PCMK__XA_LRMD_CLIENTID "lrmd_clientid"
#define PCMK__XA_LRMD_CLIENTNAME "lrmd_clientname"
#define PCMK__XA_LRMD_EXEC_OP_STATUS "lrmd_exec_op_status"
#define PCMK__XA_LRMD_EXEC_RC "lrmd_exec_rc"
#define PCMK__XA_LRMD_EXEC_TIME "lrmd_exec_time"
#define PCMK__XA_LRMD_IS_IPC_PROVIDER "lrmd_is_ipc_provider"
#define PCMK__XA_LRMD_OP "lrmd_op"
#define PCMK__XA_LRMD_ORIGIN "lrmd_origin"
#define PCMK__XA_LRMD_PROTOCOL_VERSION "lrmd_protocol_version"
#define PCMK__XA_LRMD_PROVIDER "lrmd_provider"
#define PCMK__XA_LRMD_QUEUE_TIME "lrmd_queue_time"
#define PCMK__XA_LRMD_RC "lrmd_rc"
#define PCMK__XA_LRMD_RCCHANGE_TIME "lrmd_rcchange_time"
#define PCMK__XA_LRMD_REMOTE_MSG_ID "lrmd_remote_msg_id"
#define PCMK__XA_LRMD_REMOTE_MSG_TYPE "lrmd_remote_msg_type"
#define PCMK__XA_LRMD_RSC_ACTION "lrmd_rsc_action"
#define PCMK__XA_LRMD_RSC_ID "lrmd_rsc_id"
+#define PCMK__XA_LRMD_RSC_OUTPUT "lrmd_rsc_output"
#define PCMK__XA_LRMD_RSC_USERDATA_STR "lrmd_rsc_userdata_str"
#define PCMK__XA_LRMD_RUN_TIME "lrmd_run_time"
#define PCMK__XA_LRMD_TIMEOUT "lrmd_timeout"
#define PCMK__XA_LRMD_TYPE "lrmd_type"
#define PCMK__XA_LRMD_WATCHDOG "lrmd_watchdog"
#define PCMK__XA_MAJOR_VERSION "major_version"
#define PCMK__XA_MINOR_VERSION "minor_version"
#define PCMK__XA_MODE "mode"
#define PCMK__XA_MOON "moon"
#define PCMK__XA_NAMESPACE "namespace"
#define PCMK__XA_NODE_FENCED "node_fenced"
#define PCMK__XA_NODE_IN_MAINTENANCE "node_in_maintenance"
#define PCMK__XA_NODE_START_STATE "node_start_state"
#define PCMK__XA_NODE_STATE "node_state"
#define PCMK__XA_OP_DIGEST "op-digest"
#define PCMK__XA_OP_FORCE_RESTART "op-force-restart"
#define PCMK__XA_OP_RESTART_DIGEST "op-restart-digest"
#define PCMK__XA_OP_SECURE_DIGEST "op-secure-digest"
#define PCMK__XA_OP_SECURE_PARAMS "op-secure-params"
#define PCMK__XA_OP_STATUS "op-status"
#define PCMK__XA_OPERATION_KEY "operation_key"
#define PCMK__XA_ORIGINAL_CIB_OP "original_cib_op"
#define PCMK__XA_PACEMAKERD_STATE "pacemakerd_state"
#define PCMK__XA_PASSWORD "password"
#define PCMK__XA_PRIORITY "priority"
#define PCMK__XA_RC_CODE "rc-code"
#define PCMK__XA_REAP "reap"
/* Actions to be executed on Pacemaker Remote nodes are routed through the
* controller on the cluster node hosting the remote connection. That cluster
* node is considered the router node for the action.
*/
#define PCMK__XA_ROUTER_NODE "router_node"
#define PCMK__XA_RSC_ID "rsc-id"
#define PCMK__XA_RSC_PROVIDES "rsc_provides"
#define PCMK__XA_SCHEMA "schema"
#define PCMK__XA_SCHEMAS "schemas"
#define PCMK__XA_SRC "src"
#define PCMK__XA_ST_ACTION_DISALLOWED "st_action_disallowed"
#define PCMK__XA_ST_ACTION_TIMEOUT "st_action_timeout"
#define PCMK__XA_ST_AVAILABLE_DEVICES "st-available-devices"
#define PCMK__XA_ST_CALLID "st_callid"
#define PCMK__XA_ST_CALLDATA "st_calldata"
#define PCMK__XA_ST_CALLOPT "st_callopt"
#define PCMK__XA_ST_CLIENTID "st_clientid"
#define PCMK__XA_ST_CLIENTNAME "st_clientname"
#define PCMK__XA_ST_CLIENTNODE "st_clientnode"
#define PCMK__XA_ST_DATE "st_date"
#define PCMK__XA_ST_DATE_NSEC "st_date_nsec"
#define PCMK__XA_ST_DELAY "st_delay"
#define PCMK__XA_ST_DELAY_BASE "st_delay_base"
#define PCMK__XA_ST_DELAY_MAX "st_delay_max"
#define PCMK__XA_ST_DELEGATE "st_delegate"
#define PCMK__XA_ST_DEVICE_ACTION "st_device_action"
#define PCMK__XA_ST_DEVICE_ID "st_device_id"
#define PCMK__XA_ST_DEVICE_SUPPORT_FLAGS "st_device_support_flags"
#define PCMK__XA_ST_DIFFERENTIAL "st_differential"
#define PCMK__XA_ST_MONITOR_VERIFIED "st_monitor_verified"
#define PCMK__XA_ST_NOTIFY_ACTIVATE "st_notify_activate"
#define PCMK__XA_ST_NOTIFY_DEACTIVATE "st_notify_deactivate"
#define PCMK__XA_ST_OP "st_op"
#define PCMK__XA_ST_OP_MERGED "st_op_merged"
#define PCMK__XA_ST_ORIGIN "st_origin"
#define PCMK__XA_ST_OUTPUT "st_output"
#define PCMK__XA_ST_RC "st_rc"
#define PCMK__XA_ST_REMOTE_OP "st_remote_op"
#define PCMK__XA_ST_REMOTE_OP_RELAY "st_remote_op_relay"
#define PCMK__XA_ST_REQUIRED "st_required"
#define PCMK__XA_ST_STATE "st_state"
#define PCMK__XA_ST_TARGET "st_target"
#define PCMK__XA_ST_TIMEOUT "st_timeout"
#define PCMK__XA_ST_TOLERANCE "st_tolerance"
#define PCMK__XA_SUBT "subt" // subtype
#define PCMK__XA_T "t" // type
#define PCMK__XA_TRANSITION_KEY "transition-key"
#define PCMK__XA_TRANSITION_MAGIC "transition-magic"
#define PCMK__XA_UPTIME "uptime"
// @COMPAT Deprecated since 2.1.8
#define PCMK__XA_CIB_OBJECT "cib_object"
// @COMPAT Deprecated since 2.1.8
#define PCMK__XA_CIB_OBJECT_TYPE "cib_object_type"
// @COMPAT Deprecated since 1.1.12; used with legacy CIB updates
#define PCMK__XA_CIB_LOCAL_NOTIFY_ID "cib_local_notify_id"
// @COMPAT Deprecated since 1.1.12; used with legacy CIB updates
#define PCMK__XA_CIB_UPDATE_DIFF "cib_update_diff"
// @COMPAT Used only with v1 patchsets
#define PCMK__XA_CRM_DIFF_MARKER "__crm_diff_marker__"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_FIRST_INSTANCE "first-instance"
// @COMPAT Deprecated since 2.1.7
#define PCMK__XA_ORDERING "ordering"
// @COMPAT Deprecated alias for PCMK_XA_PROMOTED_MAX since 2.0.0
#define PCMK__XA_PROMOTED_MAX_LEGACY "masters"
// @COMPAT Deprecated alias for PCMK_XA_PROMOTED_ONLY since 2.0.0
#define PCMK__XA_PROMOTED_ONLY_LEGACY "master_only"
// @COMPAT Deprecated since 1.1.12
#define PCMK__XA_REF "ref"
// @COMPAT Deprecated since 2.1.6
#define PCMK__XA_REPLACE "replace"
// @COMPAT Deprecated alias for \c PCMK_XA_AUTOMATIC since 1.1.14
#define PCMK__XA_REQUIRED "required"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_RSC_INSTANCE "rsc-instance"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_THEN_INSTANCE "then-instance"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_WITH_RSC_INSTANCE "with-rsc-instance"
#ifdef __cplusplus
}
#endif
#endif // PCMK__CRM_COMMON_XML_NAMES_INTERNAL__H
diff --git a/lib/lrmd/lrmd_client.c b/lib/lrmd/lrmd_client.c
index 6d86592679..a403640d32 100644
--- a/lib/lrmd/lrmd_client.c
+++ b/lib/lrmd/lrmd_client.c
@@ -1,2576 +1,2576 @@
/*
* Copyright 2012-2024 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU Lesser General Public License
* version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <unistd.h>
#include <stdlib.h>
#include <stdio.h>
#include <stdint.h> // uint32_t, uint64_t
#include <stdarg.h>
#include <string.h>
#include <ctype.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/wait.h>
#include <glib.h>
#include <dirent.h>
#include <crm/crm.h>
#include <crm/lrmd.h>
#include <crm/lrmd_internal.h>
#include <crm/services.h>
#include <crm/services_internal.h>
#include <crm/common/mainloop.h>
#include <crm/common/ipc_internal.h>
#include <crm/common/remote_internal.h>
#include <crm/common/xml.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h> // stonith__*
#ifdef HAVE_GNUTLS_GNUTLS_H
# include <gnutls/gnutls.h>
#endif
#include <sys/socket.h>
#include <netinet/in.h>
#include <netinet/ip.h>
#include <arpa/inet.h>
#include <netdb.h>
#define MAX_TLS_RECV_WAIT 10000
CRM_TRACE_INIT_DATA(lrmd);
static int lrmd_api_disconnect(lrmd_t * lrmd);
static int lrmd_api_is_connected(lrmd_t * lrmd);
/* IPC proxy functions */
int lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg);
static void lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg);
void lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg));
#ifdef HAVE_GNUTLS_GNUTLS_H
# define LRMD_CLIENT_HANDSHAKE_TIMEOUT 5000 /* 5 seconds */
gnutls_psk_client_credentials_t psk_cred_s;
static void lrmd_tls_disconnect(lrmd_t * lrmd);
static int global_remote_msg_id = 0;
static void lrmd_tls_connection_destroy(gpointer userdata);
#endif
typedef struct lrmd_private_s {
uint64_t type;
char *token;
mainloop_io_t *source;
/* IPC parameters */
crm_ipc_t *ipc;
pcmk__remote_t *remote;
/* Extra TLS parameters */
char *remote_nodename;
#ifdef HAVE_GNUTLS_GNUTLS_H
char *server;
int port;
gnutls_psk_client_credentials_t psk_cred_c;
/* while the async connection is occurring, this is the id
* of the connection timeout timer. */
int async_timer;
int sock;
/* since tls requires a round trip across the network for a
* request/reply, there are times where we just want to be able
* to send a request from the client and not wait around (or even care
* about) what the reply is. */
int expected_late_replies;
GList *pending_notify;
crm_trigger_t *process_notify;
#endif
lrmd_event_callback callback;
/* Internal IPC proxy msg passing for remote guests */
void (*proxy_callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg);
void *proxy_callback_userdata;
char *peer_version;
} lrmd_private_t;
static lrmd_list_t *
lrmd_list_add(lrmd_list_t * head, const char *value)
{
lrmd_list_t *p, *end;
p = calloc(1, sizeof(lrmd_list_t));
p->val = strdup(value);
end = head;
while (end && end->next) {
end = end->next;
}
if (end) {
end->next = p;
} else {
head = p;
}
return head;
}
void
lrmd_list_freeall(lrmd_list_t * head)
{
lrmd_list_t *p;
while (head) {
char *val = (char *)head->val;
p = head->next;
free(val);
free(head);
head = p;
}
}
lrmd_key_value_t *
lrmd_key_value_add(lrmd_key_value_t * head, const char *key, const char *value)
{
lrmd_key_value_t *p, *end;
p = calloc(1, sizeof(lrmd_key_value_t));
p->key = strdup(key);
p->value = strdup(value);
end = head;
while (end && end->next) {
end = end->next;
}
if (end) {
end->next = p;
} else {
head = p;
}
return head;
}
void
lrmd_key_value_freeall(lrmd_key_value_t * head)
{
lrmd_key_value_t *p;
while (head) {
p = head->next;
free(head->key);
free(head->value);
free(head);
head = p;
}
}
/*!
* \brief Create a new lrmd_event_data_t object
*
* \param[in] rsc_id ID of resource involved in event
* \param[in] task Action name
* \param[in] interval_ms Action interval
*
* \return Newly allocated and initialized lrmd_event_data_t
* \note This functions asserts on memory errors, so the return value is
* guaranteed to be non-NULL. The caller is responsible for freeing the
* result with lrmd_free_event().
*/
lrmd_event_data_t *
lrmd_new_event(const char *rsc_id, const char *task, guint interval_ms)
{
lrmd_event_data_t *event = calloc(1, sizeof(lrmd_event_data_t));
CRM_ASSERT(event != NULL);
pcmk__str_update((char **) &event->rsc_id, rsc_id);
pcmk__str_update((char **) &event->op_type, task);
event->interval_ms = interval_ms;
return event;
}
lrmd_event_data_t *
lrmd_copy_event(lrmd_event_data_t * event)
{
lrmd_event_data_t *copy = NULL;
copy = calloc(1, sizeof(lrmd_event_data_t));
copy->type = event->type;
pcmk__str_update((char **) ©->rsc_id, event->rsc_id);
pcmk__str_update((char **) ©->op_type, event->op_type);
pcmk__str_update((char **) ©->user_data, event->user_data);
copy->call_id = event->call_id;
copy->timeout = event->timeout;
copy->interval_ms = event->interval_ms;
copy->start_delay = event->start_delay;
copy->rsc_deleted = event->rsc_deleted;
copy->rc = event->rc;
copy->op_status = event->op_status;
pcmk__str_update((char **) ©->output, event->output);
copy->t_run = event->t_run;
copy->t_rcchange = event->t_rcchange;
copy->exec_time = event->exec_time;
copy->queue_time = event->queue_time;
copy->connection_rc = event->connection_rc;
copy->params = pcmk__str_table_dup(event->params);
pcmk__str_update((char **) ©->remote_nodename, event->remote_nodename);
pcmk__str_update((char **) ©->exit_reason, event->exit_reason);
return copy;
}
/*!
* \brief Free an executor event
*
* \param[in,out] Executor event object to free
*/
void
lrmd_free_event(lrmd_event_data_t *event)
{
if (event == NULL) {
return;
}
// @TODO Why are these const char *?
free((void *) event->rsc_id);
free((void *) event->op_type);
free((void *) event->user_data);
free((void *) event->remote_nodename);
lrmd__reset_result(event);
if (event->params != NULL) {
g_hash_table_destroy(event->params);
}
free(event);
}
static void
lrmd_dispatch_internal(lrmd_t * lrmd, xmlNode * msg)
{
const char *type;
const char *proxy_session = crm_element_value(msg, F_LRMD_IPC_SESSION);
lrmd_private_t *native = lrmd->lrmd_private;
lrmd_event_data_t event = { 0, };
if (proxy_session != NULL) {
/* this is proxy business */
lrmd_internal_proxy_dispatch(lrmd, msg);
return;
} else if (!native->callback) {
/* no callback set */
crm_trace("notify event received but client has not set callback");
return;
}
event.remote_nodename = native->remote_nodename;
type = crm_element_value(msg, PCMK__XA_LRMD_OP);
crm_element_value_int(msg, PCMK__XA_LRMD_CALLID, &event.call_id);
event.rsc_id = crm_element_value(msg, PCMK__XA_LRMD_RSC_ID);
if (pcmk__str_eq(type, LRMD_OP_RSC_REG, pcmk__str_none)) {
event.type = lrmd_event_register;
} else if (pcmk__str_eq(type, LRMD_OP_RSC_UNREG, pcmk__str_none)) {
event.type = lrmd_event_unregister;
} else if (pcmk__str_eq(type, LRMD_OP_RSC_EXEC, pcmk__str_none)) {
time_t epoch = 0;
crm_element_value_int(msg, PCMK__XA_LRMD_TIMEOUT, &event.timeout);
crm_element_value_ms(msg, F_LRMD_RSC_INTERVAL, &event.interval_ms);
crm_element_value_int(msg, F_LRMD_RSC_START_DELAY, &event.start_delay);
crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_RC, (int *) &event.rc);
crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_OP_STATUS,
&event.op_status);
crm_element_value_int(msg, F_LRMD_RSC_DELETED, &event.rsc_deleted);
crm_element_value_epoch(msg, PCMK__XA_LRMD_RUN_TIME, &epoch);
event.t_run = (unsigned int) epoch;
crm_element_value_epoch(msg, PCMK__XA_LRMD_RCCHANGE_TIME, &epoch);
event.t_rcchange = (unsigned int) epoch;
crm_element_value_int(msg, PCMK__XA_LRMD_EXEC_TIME,
(int *) &event.exec_time);
crm_element_value_int(msg, PCMK__XA_LRMD_QUEUE_TIME,
(int *) &event.queue_time);
event.op_type = crm_element_value(msg, PCMK__XA_LRMD_RSC_ACTION);
event.user_data = crm_element_value(msg,
PCMK__XA_LRMD_RSC_USERDATA_STR);
event.type = lrmd_event_exec_complete;
/* output and exit_reason may be freed by a callback */
- event.output = crm_element_value_copy(msg, F_LRMD_RSC_OUTPUT);
+ event.output = crm_element_value_copy(msg, PCMK__XA_LRMD_RSC_OUTPUT);
lrmd__set_result(&event, event.rc, event.op_status,
crm_element_value(msg, F_LRMD_RSC_EXIT_REASON));
event.params = xml2list(msg);
} else if (pcmk__str_eq(type, LRMD_OP_NEW_CLIENT, pcmk__str_none)) {
event.type = lrmd_event_new_client;
} else if (pcmk__str_eq(type, LRMD_OP_POKE, pcmk__str_none)) {
event.type = lrmd_event_poke;
} else {
return;
}
crm_trace("op %s notify event received", type);
native->callback(&event);
if (event.params) {
g_hash_table_destroy(event.params);
}
lrmd__reset_result(&event);
}
// \return Always 0, to indicate that IPC mainloop source should be kept
static int
lrmd_ipc_dispatch(const char *buffer, ssize_t length, gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
if (native->callback != NULL) {
xmlNode *msg = string2xml(buffer);
lrmd_dispatch_internal(lrmd, msg);
free_xml(msg);
}
return 0;
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_free_xml(gpointer userdata)
{
free_xml((xmlNode *) userdata);
}
static bool
remote_executor_connected(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
return (native->remote->tls_session != NULL);
}
/*!
* \internal
* \brief TLS dispatch function (for both trigger and file descriptor sources)
*
* \param[in,out] userdata API connection
*
* \return Always return a nonnegative value, which as a file descriptor
* dispatch function means keep the mainloop source, and as a
* trigger dispatch function, 0 means remove the trigger from the
* mainloop while 1 means keep it (and job completed)
*/
static int
lrmd_tls_dispatch(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *xml = NULL;
int rc = pcmk_rc_ok;
if (!remote_executor_connected(lrmd)) {
crm_trace("TLS dispatch triggered after disconnect");
return 0;
}
crm_trace("TLS dispatch triggered");
/* First check if there are any pending notifies to process that came
* while we were waiting for replies earlier. */
if (native->pending_notify) {
GList *iter = NULL;
crm_trace("Processing pending notifies");
for (iter = native->pending_notify; iter; iter = iter->next) {
lrmd_dispatch_internal(lrmd, iter->data);
}
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
/* Next read the current buffer and see if there are any messages to handle. */
switch (pcmk__remote_ready(native->remote, 0)) {
case pcmk_rc_ok:
rc = pcmk__read_remote_message(native->remote, -1);
xml = pcmk__remote_message_xml(native->remote);
break;
case ETIME:
// Nothing to read, check if a full message is already in buffer
xml = pcmk__remote_message_xml(native->remote);
break;
default:
rc = ENOTCONN;
break;
}
while (xml) {
const char *msg_type = crm_element_value(xml,
PCMK__XA_LRMD_REMOTE_MSG_TYPE);
if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) {
lrmd_dispatch_internal(lrmd, xml);
} else if (pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) {
if (native->expected_late_replies > 0) {
native->expected_late_replies--;
} else {
int reply_id = 0;
crm_element_value_int(xml, PCMK__XA_LRMD_CALLID, &reply_id);
/* if this happens, we want to know about it */
crm_err("Got outdated Pacemaker Remote reply %d", reply_id);
}
}
free_xml(xml);
xml = pcmk__remote_message_xml(native->remote);
}
if (rc == ENOTCONN) {
crm_info("Lost %s executor connection while reading data",
(native->remote_nodename? native->remote_nodename : "local"));
lrmd_tls_disconnect(lrmd);
return 0;
}
return 1;
}
#endif
/* Not used with mainloop */
int
lrmd_poll(lrmd_t * lrmd, int timeout)
{
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
return crm_ipc_ready(native->ipc);
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
if (native->pending_notify) {
return 1;
} else {
int rc = pcmk__remote_ready(native->remote, 0);
switch (rc) {
case pcmk_rc_ok:
return 1;
case ETIME:
return 0;
default:
return pcmk_rc2legacy(rc);
}
}
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
return -EPROTONOSUPPORT;
}
}
/* Not used with mainloop */
bool
lrmd_dispatch(lrmd_t * lrmd)
{
lrmd_private_t *private = NULL;
CRM_ASSERT(lrmd != NULL);
private = lrmd->lrmd_private;
switch (private->type) {
case pcmk__client_ipc:
while (crm_ipc_ready(private->ipc)) {
if (crm_ipc_read(private->ipc) > 0) {
const char *msg = crm_ipc_buffer(private->ipc);
lrmd_ipc_dispatch(msg, strlen(msg), lrmd);
}
}
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
lrmd_tls_dispatch(lrmd);
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
private->type);
}
if (lrmd_api_is_connected(lrmd) == FALSE) {
crm_err("Connection closed");
return FALSE;
}
return TRUE;
}
static xmlNode *
lrmd_create_op(const char *token, const char *op, xmlNode *data, int timeout,
enum lrmd_call_options options)
{
xmlNode *op_msg = create_xml_node(NULL, "lrmd_command");
CRM_CHECK(op_msg != NULL, return NULL);
CRM_CHECK(token != NULL, return NULL);
crm_xml_add(op_msg, PCMK__XA_T, T_LRMD);
crm_xml_add(op_msg, PCMK__XA_LRMD_OP, op);
crm_xml_add_int(op_msg, PCMK__XA_LRMD_TIMEOUT, timeout);
crm_xml_add_int(op_msg, PCMK__XA_LRMD_CALLOPT, options);
if (data != NULL) {
add_message_xml(op_msg, PCMK__XA_LRMD_CALLDATA, data);
}
crm_trace("Created executor %s command with call options %.8lx (%d)",
op, (long)options, options);
return op_msg;
}
static void
lrmd_ipc_connection_destroy(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
crm_info("Disconnected from local executor");
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
crm_info("Disconnected from remote executor on %s",
native->remote_nodename);
break;
#endif
default:
crm_err("Unsupported executor connection type %d (bug?)",
native->type);
}
/* Prevent these from being cleaned up in lrmd_api_disconnect() */
native->ipc = NULL;
native->source = NULL;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.type = lrmd_event_disconnect;
event.remote_nodename = native->remote_nodename;
native->callback(&event);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_tls_connection_destroy(gpointer userdata)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
crm_info("TLS connection destroyed");
if (native->remote->tls_session) {
gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
}
if (native->psk_cred_c) {
gnutls_psk_free_client_credentials(native->psk_cred_c);
}
if (native->sock) {
close(native->sock);
}
if (native->process_notify) {
mainloop_destroy_trigger(native->process_notify);
native->process_notify = NULL;
}
if (native->pending_notify) {
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
free(native->remote->buffer);
free(native->remote->start_state);
native->remote->buffer = NULL;
native->remote->start_state = NULL;
native->source = 0;
native->sock = 0;
native->psk_cred_c = NULL;
native->remote->tls_session = NULL;
native->sock = 0;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.remote_nodename = native->remote_nodename;
event.type = lrmd_event_disconnect;
native->callback(&event);
}
return;
}
// \return Standard Pacemaker return code
int
lrmd__remote_send_xml(pcmk__remote_t *session, xmlNode *msg, uint32_t id,
const char *msg_type)
{
crm_xml_add_int(msg, PCMK__XA_LRMD_REMOTE_MSG_ID, id);
crm_xml_add(msg, PCMK__XA_LRMD_REMOTE_MSG_TYPE, msg_type);
return pcmk__remote_send_xml(session, msg);
}
// \return Standard Pacemaker return code
static int
read_remote_reply(lrmd_t *lrmd, int total_timeout, int expected_reply_id,
xmlNode **reply)
{
lrmd_private_t *native = lrmd->lrmd_private;
time_t start = time(NULL);
const char *msg_type = NULL;
int reply_id = 0;
int remaining_timeout = 0;
int rc = pcmk_rc_ok;
/* A timeout of 0 here makes no sense. We have to wait a period of time
* for the response to come back. If -1 or 0, default to 10 seconds. */
if (total_timeout <= 0 || total_timeout > MAX_TLS_RECV_WAIT) {
total_timeout = MAX_TLS_RECV_WAIT;
}
for (*reply = NULL; *reply == NULL; ) {
*reply = pcmk__remote_message_xml(native->remote);
if (*reply == NULL) {
/* read some more off the tls buffer if we still have time left. */
if (remaining_timeout) {
remaining_timeout = total_timeout - ((time(NULL) - start) * 1000);
} else {
remaining_timeout = total_timeout;
}
if (remaining_timeout <= 0) {
return ETIME;
}
rc = pcmk__read_remote_message(native->remote, remaining_timeout);
if (rc != pcmk_rc_ok) {
return rc;
}
*reply = pcmk__remote_message_xml(native->remote);
if (*reply == NULL) {
return ENOMSG;
}
}
crm_element_value_int(*reply, PCMK__XA_LRMD_REMOTE_MSG_ID, &reply_id);
msg_type = crm_element_value(*reply, PCMK__XA_LRMD_REMOTE_MSG_TYPE);
if (!msg_type) {
crm_err("Empty msg type received while waiting for reply");
free_xml(*reply);
*reply = NULL;
} else if (pcmk__str_eq(msg_type, "notify", pcmk__str_casei)) {
/* got a notify while waiting for reply, trigger the notify to be processed later */
crm_info("queueing notify");
native->pending_notify = g_list_append(native->pending_notify, *reply);
if (native->process_notify) {
crm_info("notify trigger set.");
mainloop_set_trigger(native->process_notify);
}
*reply = NULL;
} else if (!pcmk__str_eq(msg_type, "reply", pcmk__str_casei)) {
/* msg isn't a reply, make some noise */
crm_err("Expected a reply, got %s", msg_type);
free_xml(*reply);
*reply = NULL;
} else if (reply_id != expected_reply_id) {
if (native->expected_late_replies > 0) {
native->expected_late_replies--;
} else {
crm_err("Got outdated reply, expected id %d got id %d", expected_reply_id, reply_id);
}
free_xml(*reply);
*reply = NULL;
}
}
if (native->remote->buffer && native->process_notify) {
mainloop_set_trigger(native->process_notify);
}
return rc;
}
// \return Standard Pacemaker return code
static int
send_remote_message(lrmd_t *lrmd, xmlNode *msg)
{
int rc = pcmk_rc_ok;
lrmd_private_t *native = lrmd->lrmd_private;
global_remote_msg_id++;
if (global_remote_msg_id <= 0) {
global_remote_msg_id = 1;
}
rc = lrmd__remote_send_xml(native->remote, msg, global_remote_msg_id,
"request");
if (rc != pcmk_rc_ok) {
crm_err("Disconnecting because TLS message could not be sent to "
"Pacemaker Remote: %s", pcmk_rc_str(rc));
lrmd_tls_disconnect(lrmd);
}
return rc;
}
static int
lrmd_tls_send_recv(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply)
{
int rc = 0;
xmlNode *xml = NULL;
if (!remote_executor_connected(lrmd)) {
return -ENOTCONN;
}
rc = send_remote_message(lrmd, msg);
if (rc != pcmk_rc_ok) {
return pcmk_rc2legacy(rc);
}
rc = read_remote_reply(lrmd, timeout, global_remote_msg_id, &xml);
if (rc != pcmk_rc_ok) {
crm_err("Disconnecting remote after request %d reply not received: %s "
CRM_XS " rc=%d timeout=%dms",
global_remote_msg_id, pcmk_rc_str(rc), rc, timeout);
lrmd_tls_disconnect(lrmd);
}
if (reply) {
*reply = xml;
} else {
free_xml(xml);
}
return pcmk_rc2legacy(rc);
}
#endif
static int
lrmd_send_xml(lrmd_t * lrmd, xmlNode * msg, int timeout, xmlNode ** reply)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
rc = crm_ipc_send(native->ipc, msg, crm_ipc_client_response, timeout, reply);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
rc = lrmd_tls_send_recv(lrmd, msg, timeout, reply);
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
rc = -EPROTONOSUPPORT;
}
return rc;
}
static int
lrmd_send_xml_no_reply(lrmd_t * lrmd, xmlNode * msg)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
rc = crm_ipc_send(native->ipc, msg, crm_ipc_flags_none, 0, NULL);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
rc = send_remote_message(lrmd, msg);
if (rc == pcmk_rc_ok) {
/* we don't want to wait around for the reply, but
* since the request/reply protocol needs to behave the same
* as libqb, a reply will eventually come later anyway. */
native->expected_late_replies++;
}
rc = pcmk_rc2legacy(rc);
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
rc = -EPROTONOSUPPORT;
}
return rc;
}
static int
lrmd_api_is_connected(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
return crm_ipc_connected(native->ipc);
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
return remote_executor_connected(lrmd);
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
return 0;
}
}
/*!
* \internal
* \brief Send a prepared API command to the executor
*
* \param[in,out] lrmd Existing connection to the executor
* \param[in] op Name of API command to send
* \param[in] data Command data XML to add to the sent command
* \param[out] output_data If expecting a reply, it will be stored here
* \param[in] timeout Timeout in milliseconds (if 0, defaults to
* a sensible value per the type of connection,
* standard vs. pacemaker remote);
* also propagated to the command XML
* \param[in] call_options Call options to pass to server when sending
* \param[in] expect_reply If TRUE, wait for a reply from the server;
* must be TRUE for IPC (as opposed to TLS) clients
*
* \return pcmk_ok on success, -errno on error
*/
static int
lrmd_send_command(lrmd_t *lrmd, const char *op, xmlNode *data,
xmlNode **output_data, int timeout,
enum lrmd_call_options options, gboolean expect_reply)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *op_msg = NULL;
xmlNode *op_reply = NULL;
if (!lrmd_api_is_connected(lrmd)) {
return -ENOTCONN;
}
if (op == NULL) {
crm_err("No operation specified");
return -EINVAL;
}
CRM_CHECK(native->token != NULL,;
);
crm_trace("Sending %s op to executor", op);
op_msg = lrmd_create_op(native->token, op, data, timeout, options);
if (op_msg == NULL) {
return -EINVAL;
}
if (expect_reply) {
rc = lrmd_send_xml(lrmd, op_msg, timeout, &op_reply);
} else {
rc = lrmd_send_xml_no_reply(lrmd, op_msg);
goto done;
}
if (rc < 0) {
crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%d): %d", op, timeout, rc);
goto done;
} else if(op_reply == NULL) {
rc = -ENOMSG;
goto done;
}
rc = pcmk_ok;
crm_trace("%s op reply received", op);
if (crm_element_value_int(op_reply, PCMK__XA_LRMD_RC, &rc) != 0) {
rc = -ENOMSG;
goto done;
}
crm_log_xml_trace(op_reply, "Reply");
if (output_data) {
*output_data = op_reply;
op_reply = NULL; /* Prevent subsequent free */
}
done:
if (lrmd_api_is_connected(lrmd) == FALSE) {
crm_err("Executor disconnected");
}
free_xml(op_msg);
free_xml(op_reply);
return rc;
}
static int
lrmd_api_poke_connection(lrmd_t * lrmd)
{
int rc;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
rc = lrmd_send_command(lrmd, LRMD_OP_POKE, data, NULL, 0, 0,
(native->type == pcmk__client_ipc));
free_xml(data);
return rc < 0 ? rc : pcmk_ok;
}
// \return Standard Pacemaker return code
int
lrmd__validate_remote_settings(lrmd_t *lrmd, GHashTable *hash)
{
int rc = pcmk_rc_ok;
const char *value;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *data = create_xml_node(NULL, PCMK__XA_LRMD_OP);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
value = g_hash_table_lookup(hash, PCMK_OPT_STONITH_WATCHDOG_TIMEOUT);
if ((value) &&
(stonith__watchdog_fencing_enabled_for_node(native->remote_nodename))) {
crm_xml_add(data, PCMK__XA_LRMD_WATCHDOG, value);
}
rc = lrmd_send_command(lrmd, LRMD_OP_CHECK, data, NULL, 0, 0,
(native->type == pcmk__client_ipc));
free_xml(data);
return (rc < 0)? pcmk_legacy2rc(rc) : pcmk_rc_ok;
}
static int
lrmd_handshake(lrmd_t * lrmd, const char *name)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
xmlNode *reply = NULL;
xmlNode *hello = create_xml_node(NULL, "lrmd_command");
crm_xml_add(hello, PCMK__XA_T, T_LRMD);
crm_xml_add(hello, PCMK__XA_LRMD_OP, CRM_OP_REGISTER);
crm_xml_add(hello, PCMK__XA_LRMD_CLIENTNAME, name);
crm_xml_add(hello, PCMK__XA_LRMD_PROTOCOL_VERSION, LRMD_PROTOCOL_VERSION);
/* advertise that we are a proxy provider */
if (native->proxy_callback) {
pcmk__xe_set_bool_attr(hello, PCMK__XA_LRMD_IS_IPC_PROVIDER, true);
}
rc = lrmd_send_xml(lrmd, hello, -1, &reply);
if (rc < 0) {
crm_perror(LOG_DEBUG, "Couldn't complete registration with the executor API: %d", rc);
rc = -ECOMM;
} else if (reply == NULL) {
crm_err("Did not receive registration reply");
rc = -EPROTO;
} else {
const char *version = crm_element_value(reply,
PCMK__XA_LRMD_PROTOCOL_VERSION);
const char *msg_type = crm_element_value(reply, PCMK__XA_LRMD_OP);
const char *tmp_ticket = crm_element_value(reply,
PCMK__XA_LRMD_CLIENTID);
const char *start_state = crm_element_value(reply, PCMK__XA_NODE_START_STATE);
long long uptime = -1;
crm_element_value_int(reply, PCMK__XA_LRMD_RC, &rc);
/* The remote executor may add its uptime to the XML reply, which is
* useful in handling transient attributes when the connection to the
* remote node unexpectedly drops. If no parameter is given, just
* default to -1.
*/
crm_element_value_ll(reply, PCMK__XA_UPTIME, &uptime);
native->remote->uptime = uptime;
if (start_state) {
native->remote->start_state = strdup(start_state);
}
if (rc == -EPROTO) {
crm_err("Executor protocol version mismatch between client (%s) and server (%s)",
LRMD_PROTOCOL_VERSION, version);
crm_log_xml_err(reply, "Protocol Error");
} else if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) {
crm_err("Invalid registration message: %s", msg_type);
crm_log_xml_err(reply, "Bad reply");
rc = -EPROTO;
} else if (tmp_ticket == NULL) {
crm_err("No registration token provided");
crm_log_xml_err(reply, "Bad reply");
rc = -EPROTO;
} else {
crm_trace("Obtained registration token: %s", tmp_ticket);
native->token = strdup(tmp_ticket);
native->peer_version = strdup(version?version:"1.0"); /* Included since 1.1 */
rc = pcmk_ok;
}
}
free_xml(reply);
free_xml(hello);
if (rc != pcmk_ok) {
lrmd_api_disconnect(lrmd);
}
return rc;
}
static int
lrmd_ipc_connect(lrmd_t * lrmd, int *fd)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
struct ipc_client_callbacks lrmd_callbacks = {
.dispatch = lrmd_ipc_dispatch,
.destroy = lrmd_ipc_connection_destroy
};
crm_info("Connecting to executor");
if (fd) {
/* No mainloop */
native->ipc = crm_ipc_new(CRM_SYSTEM_LRMD, 0);
if (native->ipc != NULL) {
rc = pcmk__connect_generic_ipc(native->ipc);
if (rc == pcmk_rc_ok) {
rc = pcmk__ipc_fd(native->ipc, fd);
}
if (rc != pcmk_rc_ok) {
crm_err("Connection to executor failed: %s", pcmk_rc_str(rc));
rc = -ENOTCONN;
}
}
} else {
native->source = mainloop_add_ipc_client(CRM_SYSTEM_LRMD, G_PRIORITY_HIGH, 0, lrmd, &lrmd_callbacks);
native->ipc = mainloop_get_ipc_client(native->source);
}
if (native->ipc == NULL) {
crm_debug("Could not connect to the executor API");
rc = -ENOTCONN;
}
return rc;
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
copy_gnutls_datum(gnutls_datum_t *dest, gnutls_datum_t *source)
{
CRM_ASSERT((dest != NULL) && (source != NULL) && (source->data != NULL));
dest->data = gnutls_malloc(source->size);
CRM_ASSERT(dest->data);
memcpy(dest->data, source->data, source->size);
dest->size = source->size;
}
static void
clear_gnutls_datum(gnutls_datum_t *datum)
{
gnutls_free(datum->data);
datum->data = NULL;
datum->size = 0;
}
#define KEY_READ_LEN 256 // Chunk size for reading key from file
// \return Standard Pacemaker return code
static int
read_gnutls_key(const char *location, gnutls_datum_t *key)
{
FILE *stream = NULL;
size_t buf_len = KEY_READ_LEN;
if ((location == NULL) || (key == NULL)) {
return EINVAL;
}
stream = fopen(location, "r");
if (stream == NULL) {
return errno;
}
key->data = gnutls_malloc(buf_len);
key->size = 0;
while (!feof(stream)) {
int next = fgetc(stream);
if (next == EOF) {
if (!feof(stream)) {
crm_warn("Pacemaker Remote key read was partially successful "
"(copy in memory may be corrupted)");
}
break;
}
if (key->size == buf_len) {
buf_len = key->size + KEY_READ_LEN;
key->data = gnutls_realloc(key->data, buf_len);
CRM_ASSERT(key->data);
}
key->data[key->size++] = (unsigned char) next;
}
fclose(stream);
if (key->size == 0) {
clear_gnutls_datum(key);
return ENOKEY;
}
return pcmk_rc_ok;
}
// Cache the most recently used Pacemaker Remote authentication key
struct key_cache_s {
time_t updated; // When cached key was read (valid for 1 minute)
const char *location; // Where cached key was read from
gnutls_datum_t key; // Cached key
};
static bool
key_is_cached(struct key_cache_s *key_cache)
{
return key_cache->updated != 0;
}
static bool
key_cache_expired(struct key_cache_s *key_cache)
{
return (time(NULL) - key_cache->updated) >= 60;
}
static void
clear_key_cache(struct key_cache_s *key_cache)
{
clear_gnutls_datum(&(key_cache->key));
if ((key_cache->updated != 0) || (key_cache->location != NULL)) {
key_cache->updated = 0;
key_cache->location = NULL;
crm_debug("Cleared Pacemaker Remote key cache");
}
}
static void
get_cached_key(struct key_cache_s *key_cache, gnutls_datum_t *key)
{
copy_gnutls_datum(key, &(key_cache->key));
crm_debug("Using cached Pacemaker Remote key from %s",
pcmk__s(key_cache->location, "unknown location"));
}
static void
cache_key(struct key_cache_s *key_cache, gnutls_datum_t *key,
const char *location)
{
key_cache->updated = time(NULL);
key_cache->location = location;
copy_gnutls_datum(&(key_cache->key), key);
crm_debug("Using (and cacheing) Pacemaker Remote key from %s",
pcmk__s(location, "unknown location"));
}
/*!
* \internal
* \brief Get Pacemaker Remote authentication key from file or cache
*
* \param[in] location Path to key file to try (this memory must
* persist across all calls of this function)
* \param[out] key Key from location or cache
*
* \return Standard Pacemaker return code
*/
static int
get_remote_key(const char *location, gnutls_datum_t *key)
{
static struct key_cache_s key_cache = { 0, };
int rc = pcmk_rc_ok;
if ((location == NULL) || (key == NULL)) {
return EINVAL;
}
if (key_is_cached(&key_cache)) {
if (key_cache_expired(&key_cache)) {
clear_key_cache(&key_cache);
} else {
get_cached_key(&key_cache, key);
return pcmk_rc_ok;
}
}
rc = read_gnutls_key(location, key);
if (rc != pcmk_rc_ok) {
return rc;
}
cache_key(&key_cache, key, location);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Initialize the Pacemaker Remote authentication key
*
* Try loading the Pacemaker Remote authentication key from cache if available,
* otherwise from these locations, in order of preference: the value of the
* PCMK_authkey_location environment variable, if set; the Pacemaker default key
* file location; or (for historical reasons) /etc/corosync/authkey.
*
* \param[out] key Where to store key
*
* \return Standard Pacemaker return code
*/
int
lrmd__init_remote_key(gnutls_datum_t *key)
{
static const char *env_location = NULL;
static bool need_env = true;
int env_rc = pcmk_rc_ok;
int default_rc = pcmk_rc_ok;
int alt_rc = pcmk_rc_ok;
bool env_is_default = false;
bool env_is_fallback = false;
if (need_env) {
env_location = pcmk__env_option(PCMK__ENV_AUTHKEY_LOCATION);
need_env = false;
}
// Try location in environment variable, if set
if (env_location != NULL) {
env_rc = get_remote_key(env_location, key);
if (env_rc == pcmk_rc_ok) {
return pcmk_rc_ok;
}
env_is_default = !strcmp(env_location, DEFAULT_REMOTE_KEY_LOCATION);
env_is_fallback = !strcmp(env_location, ALT_REMOTE_KEY_LOCATION);
/* @TODO It would be more secure to fail, rather than fall back to the
* default, if an explicitly set key location is not readable, and it
* would be better to never use the Corosync location as a fallback.
* However, that would break any deployments currently working with the
* fallbacks.
*/
}
// Try default location, if environment wasn't explicitly set to it
if (env_is_default) {
default_rc = env_rc;
} else {
default_rc = get_remote_key(DEFAULT_REMOTE_KEY_LOCATION, key);
}
// Try fallback location, if environment wasn't set to it and default failed
if (env_is_fallback) {
alt_rc = env_rc;
} else if (default_rc != pcmk_rc_ok) {
alt_rc = get_remote_key(ALT_REMOTE_KEY_LOCATION, key);
}
// We have all results, so log and return
if ((env_rc != pcmk_rc_ok) && (default_rc != pcmk_rc_ok)
&& (alt_rc != pcmk_rc_ok)) { // Environment set, everything failed
crm_warn("Could not read Pacemaker Remote key from %s (%s%s%s%s%s): %s",
env_location,
env_is_default? "" : "or default location ",
env_is_default? "" : DEFAULT_REMOTE_KEY_LOCATION,
!env_is_default && !env_is_fallback? " " : "",
env_is_fallback? "" : "or fallback location ",
env_is_fallback? "" : ALT_REMOTE_KEY_LOCATION,
pcmk_rc_str(env_rc));
return ENOKEY;
}
if (env_rc != pcmk_rc_ok) { // Environment set but failed, using a default
crm_warn("Could not read Pacemaker Remote key from %s "
"(using %s location %s instead): %s",
env_location,
(default_rc == pcmk_rc_ok)? "default" : "fallback",
(default_rc == pcmk_rc_ok)? DEFAULT_REMOTE_KEY_LOCATION : ALT_REMOTE_KEY_LOCATION,
pcmk_rc_str(env_rc));
return pcmk_rc_ok;
}
if ((default_rc != pcmk_rc_ok) && (alt_rc != pcmk_rc_ok)) {
// Environment unset, defaults failed
crm_warn("Could not read Pacemaker Remote key from default location %s"
" (or fallback location %s): %s",
DEFAULT_REMOTE_KEY_LOCATION, ALT_REMOTE_KEY_LOCATION,
pcmk_rc_str(default_rc));
return ENOKEY;
}
return pcmk_rc_ok; // Environment variable unset, a default worked
}
static void
lrmd_gnutls_global_init(void)
{
static int gnutls_init = 0;
if (!gnutls_init) {
crm_gnutls_global_init();
}
gnutls_init = 1;
}
#endif
static void
report_async_connection_result(lrmd_t * lrmd, int rc)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->callback) {
lrmd_event_data_t event = { 0, };
event.type = lrmd_event_connect;
event.remote_nodename = native->remote_nodename;
event.connection_rc = rc;
native->callback(&event);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static inline int
lrmd__tls_client_handshake(pcmk__remote_t *remote)
{
return pcmk__tls_client_handshake(remote, LRMD_CLIENT_HANDSHAKE_TIMEOUT);
}
/*!
* \internal
* \brief Add trigger and file descriptor mainloop sources for TLS
*
* \param[in,out] lrmd API connection with established TLS session
* \param[in] do_handshake Whether to perform executor handshake
*
* \return Standard Pacemaker return code
*/
static int
add_tls_to_mainloop(lrmd_t *lrmd, bool do_handshake)
{
lrmd_private_t *native = lrmd->lrmd_private;
int rc = pcmk_rc_ok;
char *name = crm_strdup_printf("pacemaker-remote-%s:%d",
native->server, native->port);
struct mainloop_fd_callbacks tls_fd_callbacks = {
.dispatch = lrmd_tls_dispatch,
.destroy = lrmd_tls_connection_destroy,
};
native->process_notify = mainloop_add_trigger(G_PRIORITY_HIGH,
lrmd_tls_dispatch, lrmd);
native->source = mainloop_add_fd(name, G_PRIORITY_HIGH, native->sock, lrmd,
&tls_fd_callbacks);
/* Async connections lose the client name provided by the API caller, so we
* have to use our generated name here to perform the executor handshake.
*
* @TODO Keep track of the caller-provided name. Perhaps we should be using
* that name in this function instead of generating one anyway.
*/
if (do_handshake) {
rc = lrmd_handshake(lrmd, name);
rc = pcmk_legacy2rc(rc);
}
free(name);
return rc;
}
static void
lrmd_tcp_connect_cb(void *userdata, int rc, int sock)
{
lrmd_t *lrmd = userdata;
lrmd_private_t *native = lrmd->lrmd_private;
gnutls_datum_t psk_key = { NULL, 0 };
native->async_timer = 0;
if (rc != pcmk_rc_ok) {
lrmd_tls_connection_destroy(lrmd);
crm_info("Could not connect to Pacemaker Remote at %s:%d: %s "
CRM_XS " rc=%d",
native->server, native->port, pcmk_rc_str(rc), rc);
report_async_connection_result(lrmd, pcmk_rc2legacy(rc));
return;
}
/* The TCP connection was successful, so establish the TLS connection.
* @TODO make this async to avoid blocking code in client
*/
native->sock = sock;
rc = lrmd__init_remote_key(&psk_key);
if (rc != pcmk_rc_ok) {
crm_info("Could not connect to Pacemaker Remote at %s:%d: %s "
CRM_XS " rc=%d",
native->server, native->port, pcmk_rc_str(rc), rc);
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, pcmk_rc2legacy(rc));
return;
}
gnutls_psk_allocate_client_credentials(&native->psk_cred_c);
gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW);
gnutls_free(psk_key.data);
native->remote->tls_session = pcmk__new_tls_session(sock, GNUTLS_CLIENT,
GNUTLS_CRD_PSK,
native->psk_cred_c);
if (native->remote->tls_session == NULL) {
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, -EPROTO);
return;
}
if (lrmd__tls_client_handshake(native->remote) != pcmk_rc_ok) {
crm_warn("Disconnecting after TLS handshake with Pacemaker Remote server %s:%d failed",
native->server, native->port);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = NULL;
lrmd_tls_connection_destroy(lrmd);
report_async_connection_result(lrmd, -EKEYREJECTED);
return;
}
crm_info("TLS connection to Pacemaker Remote server %s:%d succeeded",
native->server, native->port);
rc = add_tls_to_mainloop(lrmd, true);
report_async_connection_result(lrmd, pcmk_rc2legacy(rc));
}
static int
lrmd_tls_connect_async(lrmd_t * lrmd, int timeout /*ms */ )
{
int rc;
int timer_id = 0;
lrmd_private_t *native = lrmd->lrmd_private;
lrmd_gnutls_global_init();
native->sock = -1;
rc = pcmk__connect_remote(native->server, native->port, timeout, &timer_id,
&(native->sock), lrmd, lrmd_tcp_connect_cb);
if (rc != pcmk_rc_ok) {
crm_warn("Pacemaker Remote connection to %s:%d failed: %s "
CRM_XS " rc=%d",
native->server, native->port, pcmk_rc_str(rc), rc);
return pcmk_rc2legacy(rc);
}
native->async_timer = timer_id;
return pcmk_ok;
}
static int
lrmd_tls_connect(lrmd_t * lrmd, int *fd)
{
int rc;
lrmd_private_t *native = lrmd->lrmd_private;
gnutls_datum_t psk_key = { NULL, 0 };
lrmd_gnutls_global_init();
native->sock = -1;
rc = pcmk__connect_remote(native->server, native->port, 0, NULL,
&(native->sock), NULL, NULL);
if (rc != pcmk_rc_ok) {
crm_warn("Pacemaker Remote connection to %s:%d failed: %s "
CRM_XS " rc=%d",
native->server, native->port, pcmk_rc_str(rc), rc);
lrmd_tls_connection_destroy(lrmd);
return -ENOTCONN;
}
rc = lrmd__init_remote_key(&psk_key);
if (rc != pcmk_rc_ok) {
lrmd_tls_connection_destroy(lrmd);
return pcmk_rc2legacy(rc);
}
gnutls_psk_allocate_client_credentials(&native->psk_cred_c);
gnutls_psk_set_client_credentials(native->psk_cred_c, DEFAULT_REMOTE_USERNAME, &psk_key, GNUTLS_PSK_KEY_RAW);
gnutls_free(psk_key.data);
native->remote->tls_session = pcmk__new_tls_session(native->sock, GNUTLS_CLIENT,
GNUTLS_CRD_PSK,
native->psk_cred_c);
if (native->remote->tls_session == NULL) {
lrmd_tls_connection_destroy(lrmd);
return -EPROTO;
}
if (lrmd__tls_client_handshake(native->remote) != pcmk_rc_ok) {
crm_err("Session creation for %s:%d failed", native->server, native->port);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = NULL;
lrmd_tls_connection_destroy(lrmd);
return -EKEYREJECTED;
}
crm_info("Client TLS connection established with Pacemaker Remote server %s:%d", native->server,
native->port);
if (fd) {
*fd = native->sock;
} else {
add_tls_to_mainloop(lrmd, false);
}
return pcmk_ok;
}
#endif
static int
lrmd_api_connect(lrmd_t * lrmd, const char *name, int *fd)
{
int rc = -ENOTCONN;
lrmd_private_t *native = lrmd->lrmd_private;
switch (native->type) {
case pcmk__client_ipc:
rc = lrmd_ipc_connect(lrmd, fd);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
rc = lrmd_tls_connect(lrmd, fd);
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
rc = -EPROTONOSUPPORT;
}
if (rc == pcmk_ok) {
rc = lrmd_handshake(lrmd, name);
}
return rc;
}
static int
lrmd_api_connect_async(lrmd_t * lrmd, const char *name, int timeout)
{
int rc = pcmk_ok;
lrmd_private_t *native = lrmd->lrmd_private;
CRM_CHECK(native && native->callback, return -EINVAL);
switch (native->type) {
case pcmk__client_ipc:
/* fake async connection with ipc. it should be fast
* enough that we gain very little from async */
rc = lrmd_api_connect(lrmd, name, NULL);
if (!rc) {
report_async_connection_result(lrmd, rc);
}
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
rc = lrmd_tls_connect_async(lrmd, timeout);
if (rc) {
/* connection failed, report rc now */
report_async_connection_result(lrmd, rc);
}
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
rc = -EPROTONOSUPPORT;
}
return rc;
}
static void
lrmd_ipc_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->source != NULL) {
/* Attached to mainloop */
mainloop_del_ipc_client(native->source);
native->source = NULL;
native->ipc = NULL;
} else if (native->ipc) {
/* Not attached to mainloop */
crm_ipc_t *ipc = native->ipc;
native->ipc = NULL;
crm_ipc_close(ipc);
crm_ipc_destroy(ipc);
}
}
#ifdef HAVE_GNUTLS_GNUTLS_H
static void
lrmd_tls_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->remote->tls_session) {
gnutls_bye(*native->remote->tls_session, GNUTLS_SHUT_RDWR);
gnutls_deinit(*native->remote->tls_session);
gnutls_free(native->remote->tls_session);
native->remote->tls_session = 0;
}
if (native->async_timer) {
g_source_remove(native->async_timer);
native->async_timer = 0;
}
if (native->source != NULL) {
/* Attached to mainloop */
mainloop_del_ipc_client(native->source);
native->source = NULL;
} else if (native->sock) {
close(native->sock);
native->sock = 0;
}
if (native->pending_notify) {
g_list_free_full(native->pending_notify, lrmd_free_xml);
native->pending_notify = NULL;
}
}
#endif
static int
lrmd_api_disconnect(lrmd_t * lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
int rc = pcmk_ok;
switch (native->type) {
case pcmk__client_ipc:
crm_debug("Disconnecting from local executor");
lrmd_ipc_disconnect(lrmd);
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case pcmk__client_tls:
crm_debug("Disconnecting from remote executor on %s",
native->remote_nodename);
lrmd_tls_disconnect(lrmd);
break;
#endif
default:
crm_err("Unsupported executor connection type (bug?): %d",
native->type);
rc = -EPROTONOSUPPORT;
}
free(native->token);
native->token = NULL;
free(native->peer_version);
native->peer_version = NULL;
return rc;
}
static int
lrmd_api_register_rsc(lrmd_t * lrmd,
const char *rsc_id,
const char *class,
const char *provider, const char *type, enum lrmd_call_options options)
{
int rc = pcmk_ok;
xmlNode *data = NULL;
if (!class || !type || !rsc_id) {
return -EINVAL;
}
if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)
&& (provider == NULL)) {
return -EINVAL;
}
data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
crm_xml_add(data, PCMK__XA_LRMD_CLASS, class);
crm_xml_add(data, PCMK__XA_LRMD_PROVIDER, provider);
crm_xml_add(data, PCMK__XA_LRMD_TYPE, type);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_REG, data, NULL, 0, options, TRUE);
free_xml(data);
return rc;
}
static int
lrmd_api_unregister_rsc(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_UNREG, data, NULL, 0, options, TRUE);
free_xml(data);
return rc;
}
lrmd_rsc_info_t *
lrmd_new_rsc_info(const char *rsc_id, const char *standard,
const char *provider, const char *type)
{
lrmd_rsc_info_t *rsc_info = calloc(1, sizeof(lrmd_rsc_info_t));
CRM_ASSERT(rsc_info);
pcmk__str_update(&rsc_info->id, rsc_id);
pcmk__str_update(&rsc_info->standard, standard);
pcmk__str_update(&rsc_info->provider, provider);
pcmk__str_update(&rsc_info->type, type);
return rsc_info;
}
lrmd_rsc_info_t *
lrmd_copy_rsc_info(lrmd_rsc_info_t * rsc_info)
{
return lrmd_new_rsc_info(rsc_info->id, rsc_info->standard,
rsc_info->provider, rsc_info->type);
}
void
lrmd_free_rsc_info(lrmd_rsc_info_t * rsc_info)
{
if (!rsc_info) {
return;
}
free(rsc_info->id);
free(rsc_info->type);
free(rsc_info->standard);
free(rsc_info->provider);
free(rsc_info);
}
static lrmd_rsc_info_t *
lrmd_api_get_rsc_info(lrmd_t * lrmd, const char *rsc_id, enum lrmd_call_options options)
{
lrmd_rsc_info_t *rsc_info = NULL;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
xmlNode *output = NULL;
const char *class = NULL;
const char *provider = NULL;
const char *type = NULL;
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
lrmd_send_command(lrmd, LRMD_OP_RSC_INFO, data, &output, 0, options, TRUE);
free_xml(data);
if (!output) {
return NULL;
}
class = crm_element_value(output, PCMK__XA_LRMD_CLASS);
provider = crm_element_value(output, PCMK__XA_LRMD_PROVIDER);
type = crm_element_value(output, PCMK__XA_LRMD_TYPE);
if (!class || !type) {
free_xml(output);
return NULL;
} else if (pcmk_is_set(pcmk_get_ra_caps(class), pcmk_ra_cap_provider)
&& !provider) {
free_xml(output);
return NULL;
}
rsc_info = lrmd_new_rsc_info(rsc_id, class, provider, type);
free_xml(output);
return rsc_info;
}
void
lrmd_free_op_info(lrmd_op_info_t *op_info)
{
if (op_info) {
free(op_info->rsc_id);
free(op_info->action);
free(op_info->interval_ms_s);
free(op_info->timeout_ms_s);
free(op_info);
}
}
static int
lrmd_api_get_recurring_ops(lrmd_t *lrmd, const char *rsc_id, int timeout_ms,
enum lrmd_call_options options, GList **output)
{
xmlNode *data = NULL;
xmlNode *output_xml = NULL;
int rc = pcmk_ok;
if (output == NULL) {
return -EINVAL;
}
*output = NULL;
// Send request
if (rsc_id) {
data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
}
rc = lrmd_send_command(lrmd, LRMD_OP_GET_RECURRING, data, &output_xml,
timeout_ms, options, TRUE);
if (data) {
free_xml(data);
}
// Process reply
if ((rc != pcmk_ok) || (output_xml == NULL)) {
return rc;
}
for (xmlNode *rsc_xml = first_named_child(output_xml, F_LRMD_RSC);
(rsc_xml != NULL) && (rc == pcmk_ok);
rsc_xml = crm_next_same_xml(rsc_xml)) {
rsc_id = crm_element_value(rsc_xml, PCMK__XA_LRMD_RSC_ID);
if (rsc_id == NULL) {
crm_err("Could not parse recurring operation information from executor");
continue;
}
for (xmlNode *op_xml = first_named_child(rsc_xml, T_LRMD_RSC_OP);
op_xml != NULL; op_xml = crm_next_same_xml(op_xml)) {
lrmd_op_info_t *op_info = calloc(1, sizeof(lrmd_op_info_t));
if (op_info == NULL) {
rc = -ENOMEM;
break;
}
op_info->rsc_id = strdup(rsc_id);
op_info->action = crm_element_value_copy(op_xml,
PCMK__XA_LRMD_RSC_ACTION);
op_info->interval_ms_s = crm_element_value_copy(op_xml,
F_LRMD_RSC_INTERVAL);
op_info->timeout_ms_s =
crm_element_value_copy(op_xml, PCMK__XA_LRMD_TIMEOUT);
*output = g_list_prepend(*output, op_info);
}
}
free_xml(output_xml);
return rc;
}
static void
lrmd_api_set_callback(lrmd_t * lrmd, lrmd_event_callback callback)
{
lrmd_private_t *native = lrmd->lrmd_private;
native->callback = callback;
}
void
lrmd_internal_set_proxy_callback(lrmd_t * lrmd, void *userdata, void (*callback)(lrmd_t *lrmd, void *userdata, xmlNode *msg))
{
lrmd_private_t *native = lrmd->lrmd_private;
native->proxy_callback = callback;
native->proxy_callback_userdata = userdata;
}
void
lrmd_internal_proxy_dispatch(lrmd_t *lrmd, xmlNode *msg)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->proxy_callback) {
crm_log_xml_trace(msg, "PROXY_INBOUND");
native->proxy_callback(lrmd, native->proxy_callback_userdata, msg);
}
}
int
lrmd_internal_proxy_send(lrmd_t * lrmd, xmlNode *msg)
{
if (lrmd == NULL) {
return -ENOTCONN;
}
crm_xml_add(msg, PCMK__XA_LRMD_OP, CRM_OP_IPC_FWD);
crm_log_xml_trace(msg, "PROXY_OUTBOUND");
return lrmd_send_xml_no_reply(lrmd, msg);
}
static int
stonith_get_metadata(const char *provider, const char *type, char **output)
{
int rc = pcmk_ok;
stonith_t *stonith_api = stonith_api_new();
if (stonith_api == NULL) {
crm_err("Could not get fence agent meta-data: API memory allocation failed");
return -ENOMEM;
}
rc = stonith_api->cmds->metadata(stonith_api, st_opt_sync_call, type,
provider, output, 0);
if ((rc == pcmk_ok) && (*output == NULL)) {
rc = -EIO;
}
stonith_api->cmds->free(stonith_api);
return rc;
}
static int
lrmd_api_get_metadata(lrmd_t *lrmd, const char *standard, const char *provider,
const char *type, char **output,
enum lrmd_call_options options)
{
return lrmd->cmds->get_metadata_params(lrmd, standard, provider, type,
output, options, NULL);
}
static int
lrmd_api_get_metadata_params(lrmd_t *lrmd, const char *standard,
const char *provider, const char *type,
char **output, enum lrmd_call_options options,
lrmd_key_value_t *params)
{
svc_action_t *action = NULL;
GHashTable *params_table = NULL;
if (!standard || !type) {
lrmd_key_value_freeall(params);
return -EINVAL;
}
if (pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
lrmd_key_value_freeall(params);
return stonith_get_metadata(provider, type, output);
}
params_table = pcmk__strkey_table(free, free);
for (const lrmd_key_value_t *param = params; param; param = param->next) {
pcmk__insert_dup(params_table, param->key, param->value);
}
action = services__create_resource_action(type, standard, provider, type,
PCMK_ACTION_META_DATA, 0,
PCMK_DEFAULT_METADATA_TIMEOUT_MS,
params_table, 0);
lrmd_key_value_freeall(params);
if (action == NULL) {
return -ENOMEM;
}
if (action->rc != PCMK_OCF_UNKNOWN) {
services_action_free(action);
return -EINVAL;
}
if (!services_action_sync(action)) {
crm_err("Failed to retrieve meta-data for %s:%s:%s",
standard, provider, type);
services_action_free(action);
return -EIO;
}
if (!action->stdout_data) {
crm_err("Failed to receive meta-data for %s:%s:%s",
standard, provider, type);
services_action_free(action);
return -EIO;
}
*output = strdup(action->stdout_data);
services_action_free(action);
return pcmk_ok;
}
static int
lrmd_api_exec(lrmd_t *lrmd, const char *rsc_id, const char *action,
const char *userdata, guint interval_ms,
int timeout, /* ms */
int start_delay, /* ms */
enum lrmd_call_options options, lrmd_key_value_t * params)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
xmlNode *args = create_xml_node(data, PCMK__XE_ATTRIBUTES);
lrmd_key_value_t *tmp = NULL;
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action);
crm_xml_add(data, PCMK__XA_LRMD_RSC_USERDATA_STR, userdata);
crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms);
crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout);
crm_xml_add_int(data, F_LRMD_RSC_START_DELAY, start_delay);
for (tmp = params; tmp; tmp = tmp->next) {
hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args);
}
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_EXEC, data, NULL, timeout, options, TRUE);
free_xml(data);
lrmd_key_value_freeall(params);
return rc;
}
/* timeout is in ms */
static int
lrmd_api_exec_alert(lrmd_t *lrmd, const char *alert_id, const char *alert_path,
int timeout, lrmd_key_value_t *params)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_ALERT);
xmlNode *args = create_xml_node(data, PCMK__XE_ATTRIBUTES);
lrmd_key_value_t *tmp = NULL;
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, F_LRMD_ALERT_ID, alert_id);
crm_xml_add(data, F_LRMD_ALERT_PATH, alert_path);
crm_xml_add_int(data, PCMK__XA_LRMD_TIMEOUT, timeout);
for (tmp = params; tmp; tmp = tmp->next) {
hash2smartfield((gpointer) tmp->key, (gpointer) tmp->value, args);
}
rc = lrmd_send_command(lrmd, LRMD_OP_ALERT_EXEC, data, NULL, timeout,
lrmd_opt_notify_orig_only, TRUE);
free_xml(data);
lrmd_key_value_freeall(params);
return rc;
}
static int
lrmd_api_cancel(lrmd_t *lrmd, const char *rsc_id, const char *action,
guint interval_ms)
{
int rc = pcmk_ok;
xmlNode *data = create_xml_node(NULL, F_LRMD_RSC);
crm_xml_add(data, PCMK__XA_LRMD_ORIGIN, __func__);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ACTION, action);
crm_xml_add(data, PCMK__XA_LRMD_RSC_ID, rsc_id);
crm_xml_add_ms(data, F_LRMD_RSC_INTERVAL, interval_ms);
rc = lrmd_send_command(lrmd, LRMD_OP_RSC_CANCEL, data, NULL, 0, 0, TRUE);
free_xml(data);
return rc;
}
static int
list_stonith_agents(lrmd_list_t ** resources)
{
int rc = 0;
stonith_t *stonith_api = stonith_api_new();
stonith_key_value_t *stonith_resources = NULL;
stonith_key_value_t *dIter = NULL;
if (stonith_api == NULL) {
crm_err("Could not list fence agents: API memory allocation failed");
return -ENOMEM;
}
stonith_api->cmds->list_agents(stonith_api, st_opt_sync_call, NULL,
&stonith_resources, 0);
stonith_api->cmds->free(stonith_api);
for (dIter = stonith_resources; dIter; dIter = dIter->next) {
rc++;
if (resources) {
*resources = lrmd_list_add(*resources, dIter->value);
}
}
stonith_key_value_freeall(stonith_resources, 1, 0);
return rc;
}
static int
lrmd_api_list_agents(lrmd_t * lrmd, lrmd_list_t ** resources, const char *class,
const char *provider)
{
int rc = 0;
int stonith_count = 0; // Initially, whether to include stonith devices
if (pcmk__str_eq(class, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
stonith_count = 1;
} else {
GList *gIter = NULL;
GList *agents = resources_list_agents(class, provider);
for (gIter = agents; gIter != NULL; gIter = gIter->next) {
*resources = lrmd_list_add(*resources, (const char *)gIter->data);
rc++;
}
g_list_free_full(agents, free);
if (!class) {
stonith_count = 1;
}
}
if (stonith_count) {
// Now, if stonith devices are included, how many there are
stonith_count = list_stonith_agents(resources);
if (stonith_count > 0) {
rc += stonith_count;
}
}
if (rc == 0) {
crm_notice("No agents found for class %s", class);
rc = -EPROTONOSUPPORT;
}
return rc;
}
static bool
does_provider_have_agent(const char *agent, const char *provider, const char *class)
{
bool found = false;
GList *agents = NULL;
GList *gIter2 = NULL;
agents = resources_list_agents(class, provider);
for (gIter2 = agents; gIter2 != NULL; gIter2 = gIter2->next) {
if (pcmk__str_eq(agent, gIter2->data, pcmk__str_casei)) {
found = true;
}
}
g_list_free_full(agents, free);
return found;
}
static int
lrmd_api_list_ocf_providers(lrmd_t * lrmd, const char *agent, lrmd_list_t ** providers)
{
int rc = pcmk_ok;
char *provider = NULL;
GList *ocf_providers = NULL;
GList *gIter = NULL;
ocf_providers = resources_list_providers(PCMK_RESOURCE_CLASS_OCF);
for (gIter = ocf_providers; gIter != NULL; gIter = gIter->next) {
provider = gIter->data;
if (!agent || does_provider_have_agent(agent, provider,
PCMK_RESOURCE_CLASS_OCF)) {
*providers = lrmd_list_add(*providers, (const char *)gIter->data);
rc++;
}
}
g_list_free_full(ocf_providers, free);
return rc;
}
static int
lrmd_api_list_standards(lrmd_t * lrmd, lrmd_list_t ** supported)
{
int rc = 0;
GList *standards = NULL;
GList *gIter = NULL;
standards = resources_list_standards();
for (gIter = standards; gIter != NULL; gIter = gIter->next) {
*supported = lrmd_list_add(*supported, (const char *)gIter->data);
rc++;
}
if (list_stonith_agents(NULL) > 0) {
*supported = lrmd_list_add(*supported, PCMK_RESOURCE_CLASS_STONITH);
rc++;
}
g_list_free_full(standards, free);
return rc;
}
/*!
* \internal
* \brief Create an executor API object
*
* \param[out] api Will be set to newly created API object (it is the
* caller's responsibility to free this value with
* lrmd_api_delete() if this function succeeds)
* \param[in] nodename If the object will be used for a remote connection,
* the node name to use in cluster for remote executor
* \param[in] server If the object will be used for a remote connection,
* the resolvable host name to connect to
* \param[in] port If the object will be used for a remote connection,
* port number on \p server to connect to
*
* \return Standard Pacemaker return code
* \note If the caller leaves one of \p nodename or \p server NULL, the other's
* value will be used for both. If the caller leaves both NULL, an API
* object will be created for a local executor connection.
*/
int
lrmd__new(lrmd_t **api, const char *nodename, const char *server, int port)
{
lrmd_private_t *pvt = NULL;
if (api == NULL) {
return EINVAL;
}
*api = NULL;
// Allocate all memory needed
*api = calloc(1, sizeof(lrmd_t));
if (*api == NULL) {
return ENOMEM;
}
pvt = calloc(1, sizeof(lrmd_private_t));
if (pvt == NULL) {
lrmd_api_delete(*api);
*api = NULL;
return ENOMEM;
}
(*api)->lrmd_private = pvt;
// @TODO Do we need to do this for local connections?
pvt->remote = calloc(1, sizeof(pcmk__remote_t));
(*api)->cmds = calloc(1, sizeof(lrmd_api_operations_t));
if ((pvt->remote == NULL) || ((*api)->cmds == NULL)) {
lrmd_api_delete(*api);
*api = NULL;
return ENOMEM;
}
// Set methods
(*api)->cmds->connect = lrmd_api_connect;
(*api)->cmds->connect_async = lrmd_api_connect_async;
(*api)->cmds->is_connected = lrmd_api_is_connected;
(*api)->cmds->poke_connection = lrmd_api_poke_connection;
(*api)->cmds->disconnect = lrmd_api_disconnect;
(*api)->cmds->register_rsc = lrmd_api_register_rsc;
(*api)->cmds->unregister_rsc = lrmd_api_unregister_rsc;
(*api)->cmds->get_rsc_info = lrmd_api_get_rsc_info;
(*api)->cmds->get_recurring_ops = lrmd_api_get_recurring_ops;
(*api)->cmds->set_callback = lrmd_api_set_callback;
(*api)->cmds->get_metadata = lrmd_api_get_metadata;
(*api)->cmds->exec = lrmd_api_exec;
(*api)->cmds->cancel = lrmd_api_cancel;
(*api)->cmds->list_agents = lrmd_api_list_agents;
(*api)->cmds->list_ocf_providers = lrmd_api_list_ocf_providers;
(*api)->cmds->list_standards = lrmd_api_list_standards;
(*api)->cmds->exec_alert = lrmd_api_exec_alert;
(*api)->cmds->get_metadata_params = lrmd_api_get_metadata_params;
if ((nodename == NULL) && (server == NULL)) {
pvt->type = pcmk__client_ipc;
} else {
#ifdef HAVE_GNUTLS_GNUTLS_H
if (nodename == NULL) {
nodename = server;
} else if (server == NULL) {
server = nodename;
}
pvt->type = pcmk__client_tls;
pvt->remote_nodename = strdup(nodename);
pvt->server = strdup(server);
if ((pvt->remote_nodename == NULL) || (pvt->server == NULL)) {
lrmd_api_delete(*api);
*api = NULL;
return ENOMEM;
}
pvt->port = port;
if (pvt->port == 0) {
pvt->port = crm_default_remote_port();
}
#else
crm_err("Cannot communicate with Pacemaker Remote "
"because GnuTLS is not enabled for this build");
lrmd_api_delete(*api);
*api = NULL;
return EOPNOTSUPP;
#endif
}
return pcmk_rc_ok;
}
lrmd_t *
lrmd_api_new(void)
{
lrmd_t *api = NULL;
CRM_ASSERT(lrmd__new(&api, NULL, NULL, 0) == pcmk_rc_ok);
return api;
}
lrmd_t *
lrmd_remote_api_new(const char *nodename, const char *server, int port)
{
lrmd_t *api = NULL;
CRM_ASSERT(lrmd__new(&api, nodename, server, port) == pcmk_rc_ok);
return api;
}
void
lrmd_api_delete(lrmd_t * lrmd)
{
if (lrmd == NULL) {
return;
}
if (lrmd->cmds != NULL) { // Never NULL, but make static analysis happy
if (lrmd->cmds->disconnect != NULL) { // Also never really NULL
lrmd->cmds->disconnect(lrmd); // No-op if already disconnected
}
free(lrmd->cmds);
}
if (lrmd->lrmd_private != NULL) {
lrmd_private_t *native = lrmd->lrmd_private;
#ifdef HAVE_GNUTLS_GNUTLS_H
free(native->server);
#endif
free(native->remote_nodename);
free(native->remote);
free(native->token);
free(native->peer_version);
free(lrmd->lrmd_private);
}
free(lrmd);
}
struct metadata_cb {
void (*callback)(int pid, const pcmk__action_result_t *result,
void *user_data);
void *user_data;
};
/*!
* \internal
* \brief Process asynchronous metadata completion
*
* \param[in,out] action Metadata action that completed
*/
static void
metadata_complete(svc_action_t *action)
{
struct metadata_cb *metadata_cb = (struct metadata_cb *) action->cb_data;
pcmk__action_result_t result = PCMK__UNKNOWN_RESULT;
pcmk__set_result(&result, action->rc, action->status,
services__exit_reason(action));
pcmk__set_result_output(&result, action->stdout_data, action->stderr_data);
metadata_cb->callback(0, &result, metadata_cb->user_data);
result.action_stdout = NULL; // Prevent free, because action owns it
result.action_stderr = NULL; // Prevent free, because action owns it
pcmk__reset_result(&result);
free(metadata_cb);
}
/*!
* \internal
* \brief Retrieve agent metadata asynchronously
*
* \param[in] rsc Resource agent specification
* \param[in] callback Function to call with result (this will always be
* called, whether by this function directly or later
* via the main loop, and on success the metadata will
* be in its result argument's action_stdout)
* \param[in,out] user_data User data to pass to callback
*
* \return Standard Pacemaker return code
* \note This function is not a lrmd_api_operations_t method because it does not
* need an lrmd_t object and does not go through the executor, but
* executes the agent directly.
*/
int
lrmd__metadata_async(const lrmd_rsc_info_t *rsc,
void (*callback)(int pid,
const pcmk__action_result_t *result,
void *user_data),
void *user_data)
{
svc_action_t *action = NULL;
struct metadata_cb *metadata_cb = NULL;
pcmk__action_result_t result = PCMK__UNKNOWN_RESULT;
CRM_CHECK(callback != NULL, return EINVAL);
if ((rsc == NULL) || (rsc->standard == NULL) || (rsc->type == NULL)) {
pcmk__set_result(&result, PCMK_OCF_NOT_CONFIGURED,
PCMK_EXEC_ERROR_FATAL,
"Invalid resource specification");
callback(0, &result, user_data);
pcmk__reset_result(&result);
return EINVAL;
}
if (strcmp(rsc->standard, PCMK_RESOURCE_CLASS_STONITH) == 0) {
return stonith__metadata_async(rsc->type,
PCMK_DEFAULT_METADATA_TIMEOUT_MS / 1000,
callback, user_data);
}
action = services__create_resource_action(pcmk__s(rsc->id, rsc->type),
rsc->standard, rsc->provider,
rsc->type,
PCMK_ACTION_META_DATA, 0,
PCMK_DEFAULT_METADATA_TIMEOUT_MS,
NULL, 0);
if (action == NULL) {
pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Out of memory");
callback(0, &result, user_data);
pcmk__reset_result(&result);
return ENOMEM;
}
if (action->rc != PCMK_OCF_UNKNOWN) {
pcmk__set_result(&result, action->rc, action->status,
services__exit_reason(action));
callback(0, &result, user_data);
pcmk__reset_result(&result);
services_action_free(action);
return EINVAL;
}
action->cb_data = calloc(1, sizeof(struct metadata_cb));
if (action->cb_data == NULL) {
services_action_free(action);
pcmk__set_result(&result, PCMK_OCF_UNKNOWN_ERROR, PCMK_EXEC_ERROR,
"Out of memory");
callback(0, &result, user_data);
pcmk__reset_result(&result);
return ENOMEM;
}
metadata_cb = (struct metadata_cb *) action->cb_data;
metadata_cb->callback = callback;
metadata_cb->user_data = user_data;
if (!services_action_async(action, metadata_complete)) {
services_action_free(action);
return pcmk_rc_error; // @TODO Derive from action->rc and ->status
}
// The services library has taken responsibility for action
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Set the result of an executor event
*
* \param[in,out] event Executor event to set
* \param[in] rc OCF exit status of event
* \param[in] op_status Executor status of event
* \param[in] exit_reason Human-friendly description of event
*/
void
lrmd__set_result(lrmd_event_data_t *event, enum ocf_exitcode rc, int op_status,
const char *exit_reason)
{
if (event == NULL) {
return;
}
event->rc = rc;
event->op_status = op_status;
pcmk__str_update((char **) &event->exit_reason, exit_reason);
}
/*!
* \internal
* \brief Clear an executor event's exit reason, output, and error output
*
* \param[in,out] event Executor event to reset
*/
void
lrmd__reset_result(lrmd_event_data_t *event)
{
if (event == NULL) {
return;
}
free((void *) event->exit_reason);
event->exit_reason = NULL;
free((void *) event->output);
event->output = NULL;
}
/*!
* \internal
* \brief Get the uptime of a remote resource connection
*
* When the cluster connects to a remote resource, part of that resource's
* handshake includes the uptime of the remote resource's connection. This
* uptime is stored in the lrmd_t object.
*
* \return The connection's uptime, or -1 if unknown
*/
time_t
lrmd__uptime(lrmd_t *lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->remote == NULL) {
return -1;
} else {
return native->remote->uptime;
}
}
const char *
lrmd__node_start_state(lrmd_t *lrmd)
{
lrmd_private_t *native = lrmd->lrmd_private;
if (native->remote == NULL) {
return NULL;
} else {
return native->remote->start_state;
}
}