diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c
index f5a729741f..9c6fcf4c76 100644
--- a/daemons/fenced/pacemaker-fenced.c
+++ b/daemons/fenced/pacemaker-fenced.c
@@ -1,666 +1,666 @@
/*
* Copyright 2009-2025 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/utsname.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h> // PRIu32, PRIx32
#include <crm/crm.h>
#include <crm/common/cmdline_internal.h>
#include <crm/common/ipc.h>
#include <crm/common/ipc_internal.h>
#include <crm/common/output_internal.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/common/mainloop.h>
#include <crm/cib/internal.h>
#include <pacemaker-fenced.h>
#define SUMMARY "daemon for executing fencing devices in a Pacemaker cluster"
// @TODO This should be guint
long long stonith_watchdog_timeout_ms = 0;
GList *stonith_watchdog_targets = NULL;
static GMainLoop *mainloop = NULL;
gboolean stonith_shutdown_flag = FALSE;
static qb_ipcs_service_t *ipcs = NULL;
static pcmk__output_t *out = NULL;
pcmk__supported_format_t formats[] = {
PCMK__SUPPORTED_FORMAT_NONE,
PCMK__SUPPORTED_FORMAT_TEXT,
PCMK__SUPPORTED_FORMAT_XML,
{ NULL, NULL, NULL }
};
static struct {
gboolean stand_alone;
gchar **log_files;
} options;
crm_exit_t exit_code = CRM_EX_OK;
static void stonith_cleanup(void);
static int32_t
st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
if (stonith_shutdown_flag) {
crm_info("Ignoring new client [%d] during shutdown",
pcmk__client_pid(c));
return -ECONNREFUSED;
}
if (pcmk__new_client(c, uid, gid) == NULL) {
return -ENOMEM;
}
return 0;
}
/* Exit code means? */
static int32_t
st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
uint32_t call_options = st_opt_none;
xmlNode *request = NULL;
pcmk__client_t *c = pcmk__find_client(qbc);
const char *op = NULL;
int rc = pcmk_rc_ok;
if (c == NULL) {
crm_info("Invalid client: %p", qbc);
return 0;
}
request = pcmk__client_data2xml(c, data, &id, &flags);
if (request == NULL) {
pcmk__ipc_send_ack(c, id, flags, PCMK__XE_NACK, NULL, CRM_EX_PROTOCOL);
return 0;
}
op = crm_element_value(request, PCMK__XA_CRM_TASK);
if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) {
crm_xml_add(request, PCMK__XA_T, PCMK__VALUE_STONITH_NG);
crm_xml_add(request, PCMK__XA_ST_OP, op);
crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id);
crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c));
crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, fenced_get_local_node());
pcmk__cluster_send_message(NULL, pcmk_ipc_fenced, request);
pcmk__xml_free(request);
return 0;
}
if (c->name == NULL) {
const char *value = crm_element_value(request, PCMK__XA_ST_CLIENTNAME);
c->name = crm_strdup_printf("%s.%u", pcmk__s(value, "unknown"), c->pid);
}
rc = pcmk__xe_get_flags(request, PCMK__XA_ST_CALLOPT, &call_options,
st_opt_none);
if (rc != pcmk_rc_ok) {
crm_warn("Couldn't parse options from IPC request: %s",
pcmk_rc_str(rc));
}
crm_trace("Flags %#08" PRIx32 "/%#08x for command %" PRIu32
" from client %s", flags, call_options, id, pcmk__client_name(c));
if (pcmk_is_set(call_options, st_opt_sync_call)) {
pcmk__assert(pcmk_is_set(flags, crm_ipc_client_response));
CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */
c->request_id = id; /* Reply only to the last one */
}
crm_xml_add(request, PCMK__XA_ST_CLIENTID, c->id);
crm_xml_add(request, PCMK__XA_ST_CLIENTNAME, pcmk__client_name(c));
crm_xml_add(request, PCMK__XA_ST_CLIENTNODE, fenced_get_local_node());
crm_log_xml_trace(request, "ipc-received");
stonith_command(c, id, flags, request, NULL);
pcmk__xml_free(request);
return 0;
}
/* Error code means? */
static int32_t
st_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p closed", c);
pcmk__free_client(client);
/* 0 means: yes, go ahead and destroy the connection */
return 0;
}
static void
st_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p destroyed", c);
st_ipc_closed(c);
}
static void
stonith_peer_callback(xmlNode * msg, void *private_data)
{
const char *remote_peer = crm_element_value(msg, PCMK__XA_SRC);
const char *op = crm_element_value(msg, PCMK__XA_ST_OP);
if (pcmk__str_eq(op, STONITH_OP_POKE, pcmk__str_none)) {
return;
}
crm_log_xml_trace(msg, "Peer[inbound]");
stonith_command(NULL, 0, 0, msg, remote_peer);
}
#if SUPPORT_COROSYNC
static void
handle_cpg_message(cpg_handle_t handle, const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
xmlNode *xml = NULL;
const char *from = NULL;
char *data = pcmk__cpg_message_data(handle, nodeid, pid, msg, &from);
if(data == NULL) {
return;
}
xml = pcmk__xml_parse(data);
if (xml == NULL) {
crm_err("Invalid XML: '%.120s'", data);
free(data);
return;
}
crm_xml_add(xml, PCMK__XA_SRC, from);
stonith_peer_callback(xml, NULL);
pcmk__xml_free(xml);
free(data);
}
static void
stonith_peer_cs_destroy(gpointer user_data)
{
crm_crit("Lost connection to cluster layer, shutting down");
stonith_shutdown(0);
}
#endif
void
do_local_reply(const xmlNode *notify_src, pcmk__client_t *client,
int call_options)
{
/* send callback to originating child */
int local_rc = pcmk_rc_ok;
int rid = 0;
uint32_t ipc_flags = crm_ipc_server_event;
if (pcmk_is_set(call_options, st_opt_sync_call)) {
CRM_LOG_ASSERT(client->request_id);
rid = client->request_id;
client->request_id = 0;
ipc_flags = crm_ipc_flags_none;
}
local_rc = pcmk__ipc_send_xml(client, rid, notify_src, ipc_flags);
if (local_rc == pcmk_rc_ok) {
crm_trace("Sent response %d to client %s",
rid, pcmk__client_name(client));
} else {
crm_warn("%synchronous reply to client %s failed: %s",
(pcmk_is_set(call_options, st_opt_sync_call)? "S" : "As"),
pcmk__client_name(client), pcmk_rc_str(local_rc));
}
}
uint64_t
get_stonith_flag(const char *name)
{
if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_FENCE, pcmk__str_none)) {
return fenced_nf_fence_result;
} else if (pcmk__str_eq(name, STONITH_OP_DEVICE_ADD, pcmk__str_casei)) {
return fenced_nf_device_registered;
} else if (pcmk__str_eq(name, STONITH_OP_DEVICE_DEL, pcmk__str_casei)) {
- return st_callback_device_del;
+ return fenced_nf_device_removed;
} else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY,
pcmk__str_none)) {
return st_callback_notify_history;
} else if (pcmk__str_eq(name, PCMK__VALUE_ST_NOTIFY_HISTORY_SYNCED,
pcmk__str_none)) {
return st_callback_notify_history_synced;
}
return fenced_nf_none;
}
static void
stonith_notify_client(gpointer key, gpointer value, gpointer user_data)
{
const xmlNode *update_msg = user_data;
pcmk__client_t *client = value;
const char *type = NULL;
CRM_CHECK(client != NULL, return);
CRM_CHECK(update_msg != NULL, return);
type = crm_element_value(update_msg, PCMK__XA_SUBT);
CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return);
if (client->ipcs == NULL) {
crm_trace("Skipping client with NULL channel");
return;
}
if (pcmk_is_set(client->flags, get_stonith_flag(type))) {
int rc = pcmk__ipc_send_xml(client, 0, update_msg,
crm_ipc_server_event);
if (rc != pcmk_rc_ok) {
crm_warn("%s notification of client %s failed: %s "
QB_XS " id=%.8s rc=%d", type, pcmk__client_name(client),
pcmk_rc_str(rc), client->id, rc);
} else {
crm_trace("Sent %s notification to client %s",
type, pcmk__client_name(client));
}
}
}
void
do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout)
{
pcmk__client_t *client = NULL;
xmlNode *notify_data = NULL;
if (!timeout || !call_id || !client_id) {
return;
}
client = pcmk__find_client_by_id(client_id);
if (!client) {
return;
}
notify_data = pcmk__xe_create(NULL, PCMK__XE_ST_ASYNC_TIMEOUT_VALUE);
crm_xml_add(notify_data, PCMK__XA_T, PCMK__VALUE_ST_ASYNC_TIMEOUT_VALUE);
crm_xml_add(notify_data, PCMK__XA_ST_CALLID, call_id);
crm_xml_add_int(notify_data, PCMK__XA_ST_TIMEOUT, timeout);
crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id);
if (client) {
pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event);
}
pcmk__xml_free(notify_data);
}
/*!
* \internal
* \brief Notify relevant IPC clients of a fencing operation result
*
* \param[in] type Notification type
* \param[in] result Result of fencing operation (assume success if NULL)
* \param[in] data If not NULL, add to notification as call data
*/
void
fenced_send_notification(const char *type, const pcmk__action_result_t *result,
xmlNode *data)
{
/* TODO: Standardize the contents of data */
xmlNode *update_msg = pcmk__xe_create(NULL, PCMK__XE_NOTIFY);
CRM_LOG_ASSERT(type != NULL);
crm_xml_add(update_msg, PCMK__XA_T, PCMK__VALUE_ST_NOTIFY);
crm_xml_add(update_msg, PCMK__XA_SUBT, type);
crm_xml_add(update_msg, PCMK__XA_ST_OP, type);
stonith__xe_set_result(update_msg, result);
if (data != NULL) {
xmlNode *wrapper = pcmk__xe_create(update_msg, PCMK__XE_ST_CALLDATA);
pcmk__xml_copy(wrapper, data);
}
crm_trace("Notifying clients");
pcmk__foreach_ipc_client(stonith_notify_client, update_msg);
pcmk__xml_free(update_msg);
crm_trace("Notify complete");
}
/*!
* \internal
* \brief Send notifications for a configuration change to subscribed clients
*
* \param[in] op Notification type (\c STONITH_OP_DEVICE_ADD,
* \c STONITH_OP_DEVICE_DEL, \c STONITH_OP_LEVEL_ADD, or
* \c STONITH_OP_LEVEL_DEL)
* \param[in] result Operation result
* \param[in] desc Description of what changed (either device ID or string
* representation of level
* (<tt><target>[<level_index>]</tt>))
*/
void
fenced_send_config_notification(const char *op,
const pcmk__action_result_t *result,
const char *desc)
{
xmlNode *notify_data = pcmk__xe_create(NULL, op);
crm_xml_add(notify_data, PCMK__XA_ST_DEVICE_ID, desc);
fenced_send_notification(op, result, notify_data);
pcmk__xml_free(notify_data);
}
/*!
* \internal
* \brief Check whether a node does watchdog-fencing
*
* \param[in] node Name of node to check
*
* \return TRUE if node found in stonith_watchdog_targets
* or stonith_watchdog_targets is empty indicating
* all nodes are doing watchdog-fencing
*/
gboolean
node_does_watchdog_fencing(const char *node)
{
return ((stonith_watchdog_targets == NULL) ||
pcmk__str_in_list(node, stonith_watchdog_targets, pcmk__str_casei));
}
void
stonith_shutdown(int nsig)
{
crm_info("Terminating with %d clients", pcmk__ipc_client_count());
stonith_shutdown_flag = TRUE;
if (mainloop != NULL && g_main_loop_is_running(mainloop)) {
g_main_loop_quit(mainloop);
}
}
static void
stonith_cleanup(void)
{
fenced_cib_cleanup();
if (ipcs) {
qb_ipcs_destroy(ipcs);
}
pcmk__cluster_destroy_node_caches();
pcmk__client_cleanup();
free_stonith_remote_op_list();
free_topology_list();
fenced_free_device_table();
free_metadata_cache();
fenced_unregister_handlers();
}
struct qb_ipcs_service_handlers ipc_callbacks = {
.connection_accept = st_ipc_accept,
.connection_created = NULL,
.msg_process = st_ipc_dispatch,
.connection_closed = st_ipc_closed,
.connection_destroyed = st_ipc_destroy
};
/*!
* \internal
* \brief Callback for peer status changes
*
* \param[in] type What changed
* \param[in] node What peer had the change
* \param[in] data Previous value of what changed
*/
static void
st_peer_update_callback(enum pcmk__node_update type, pcmk__node_status_t *node,
const void *data)
{
if ((type != pcmk__node_update_processes)
&& !pcmk_is_set(node->flags, pcmk__node_status_remote)) {
/*
* This is a hack until we can send to a nodeid and/or we fix node name lookups
* These messages are ignored in stonith_peer_callback()
*/
xmlNode *query = pcmk__xe_create(NULL, PCMK__XE_STONITH_COMMAND);
crm_xml_add(query, PCMK__XA_T, PCMK__VALUE_STONITH_NG);
crm_xml_add(query, PCMK__XA_ST_OP, STONITH_OP_POKE);
crm_debug("Broadcasting our uname because of node %" PRIu32,
node->cluster_layer_id);
pcmk__cluster_send_message(NULL, pcmk_ipc_fenced, query);
pcmk__xml_free(query);
}
}
/* @COMPAT Deprecated since 2.1.8. Use pcmk_list_fence_attrs() or
* crm_resource --list-options=fencing instead of querying daemon metadata.
*
* NOTE: pcs (as of at least 0.11.8) uses this
*/
static int
fencer_metadata(void)
{
const char *name = PCMK__SERVER_FENCED;
const char *desc_short = N_("Instance attributes available for all "
"\"stonith\"-class resources");
const char *desc_long = N_("Instance attributes available for all "
"\"stonith\"-class resources and used by "
"Pacemaker's fence daemon");
return pcmk__daemon_metadata(out, name, desc_short, desc_long,
pcmk__opt_fencing);
}
static GOptionEntry entries[] = {
{ "stand-alone", 's', G_OPTION_FLAG_NONE, G_OPTION_ARG_NONE,
&options.stand_alone, N_("Intended for use in regression testing only"),
NULL },
{ "logfile", 'l', G_OPTION_FLAG_NONE, G_OPTION_ARG_FILENAME_ARRAY,
&options.log_files, N_("Send logs to the additional named logfile"), NULL },
{ NULL }
};
static GOptionContext *
build_arg_context(pcmk__common_args_t *args, GOptionGroup **group)
{
GOptionContext *context = NULL;
context = pcmk__build_arg_context(args, "text (default), xml", group, NULL);
pcmk__add_main_args(context, entries);
return context;
}
int
main(int argc, char **argv)
{
int rc = pcmk_rc_ok;
pcmk_cluster_t *cluster = NULL;
crm_ipc_t *old_instance = NULL;
GError *error = NULL;
GOptionGroup *output_group = NULL;
pcmk__common_args_t *args = pcmk__new_common_args(SUMMARY);
gchar **processed_args = pcmk__cmdline_preproc(argv, "l");
GOptionContext *context = build_arg_context(args, &output_group);
crm_log_preinit(NULL, argc, argv);
pcmk__register_formats(output_group, formats);
if (!g_option_context_parse_strv(context, &processed_args, &error)) {
exit_code = CRM_EX_USAGE;
goto done;
}
rc = pcmk__output_new(&out, args->output_ty, args->output_dest, argv);
if (rc != pcmk_rc_ok) {
exit_code = CRM_EX_ERROR;
g_set_error(&error, PCMK__EXITC_ERROR, exit_code,
"Error creating output format %s: %s",
args->output_ty, pcmk_rc_str(rc));
goto done;
}
if (args->version) {
out->version(out, false);
goto done;
}
if ((g_strv_length(processed_args) >= 2)
&& pcmk__str_eq(processed_args[1], "metadata", pcmk__str_none)) {
rc = fencer_metadata();
if (rc != pcmk_rc_ok) {
exit_code = CRM_EX_FATAL;
g_set_error(&error, PCMK__EXITC_ERROR, exit_code,
"Unable to display metadata: %s", pcmk_rc_str(rc));
}
goto done;
}
// Open additional log files
pcmk__add_logfiles(options.log_files, out);
crm_log_init(NULL, LOG_INFO + args->verbosity, TRUE,
(args->verbosity > 0), argc, argv, FALSE);
crm_notice("Starting Pacemaker fencer");
old_instance = crm_ipc_new("stonith-ng", 0);
if (old_instance == NULL) {
/* crm_ipc_new() will have already logged an error message with
* crm_err()
*/
exit_code = CRM_EX_FATAL;
goto done;
}
if (pcmk__connect_generic_ipc(old_instance) == pcmk_rc_ok) {
// IPC endpoint already up
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_crit("Aborting start-up because another fencer instance is "
"already active");
goto done;
} else {
// Not up or not authentic, we'll proceed either way
crm_ipc_destroy(old_instance);
old_instance = NULL;
}
mainloop_add_signal(SIGTERM, stonith_shutdown);
pcmk__cluster_init_node_caches();
rc = fenced_scheduler_init();
if (rc != pcmk_rc_ok) {
exit_code = CRM_EX_FATAL;
g_set_error(&error, PCMK__EXITC_ERROR, exit_code,
"Error initializing scheduler data: %s", pcmk_rc_str(rc));
goto done;
}
cluster = pcmk_cluster_new();
#if SUPPORT_COROSYNC
if (pcmk_get_cluster_layer() == pcmk_cluster_layer_corosync) {
pcmk_cluster_set_destroy_fn(cluster, stonith_peer_cs_destroy);
pcmk_cpg_set_deliver_fn(cluster, handle_cpg_message);
pcmk_cpg_set_confchg_fn(cluster, pcmk__cpg_confchg_cb);
}
#endif // SUPPORT_COROSYNC
pcmk__cluster_set_status_callback(&st_peer_update_callback);
if (pcmk_cluster_connect(cluster) != pcmk_rc_ok) {
exit_code = CRM_EX_FATAL;
crm_crit("Cannot sign in to the cluster... terminating");
goto done;
}
fenced_set_local_node(cluster->priv->node_name);
if (!options.stand_alone) {
setup_cib();
}
fenced_init_device_table();
init_topology_list();
pcmk__serve_fenced_ipc(&ipcs, &ipc_callbacks);
// Create the mainloop and run it...
mainloop = g_main_loop_new(NULL, FALSE);
crm_notice("Pacemaker fencer successfully started and accepting connections");
g_main_loop_run(mainloop);
done:
g_strfreev(processed_args);
pcmk__free_arg_context(context);
g_strfreev(options.log_files);
stonith_cleanup();
pcmk_cluster_free(cluster);
fenced_scheduler_cleanup();
pcmk__output_and_clear_error(&error, out);
if (out != NULL) {
out->finish(out, exit_code, true, NULL);
pcmk__output_free(out);
}
pcmk__unregister_formats();
crm_exit(exit_code);
}
diff --git a/daemons/fenced/pacemaker-fenced.h b/daemons/fenced/pacemaker-fenced.h
index 13e4a43d64..80d16d1cff 100644
--- a/daemons/fenced/pacemaker-fenced.h
+++ b/daemons/fenced/pacemaker-fenced.h
@@ -1,401 +1,404 @@
/*
* Copyright 2009-2025 the Pacemaker project contributors
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <stdint.h> // uint32_t, uint64_t
#include <libxml/tree.h> // xmlNode
#include <crm/common/mainloop.h>
#include <crm/cluster.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
/*!
* \internal
* \brief Check whether target has already been fenced recently
*
* \param[in] tolerance Number of seconds to look back in time
* \param[in] target Name of node to search for
* \param[in] action Action we want to match
*
* \return TRUE if an equivalent fencing operation took place in the last
* \p tolerance seconds, FALSE otherwise
*/
gboolean stonith_check_fence_tolerance(int tolerance, const char *target, const char *action);
/*!
* \internal
* \brief Flags for \c fenced_device_t configuration, state, and support
*/
enum fenced_device_flags {
//! This flag has no effect
fenced_df_none = UINT32_C(0),
//! Device supports list action
fenced_df_supports_list = (UINT32_C(1) << 0),
//! Device supports on action
fenced_df_supports_on = (UINT32_C(1) << 1),
//! Device supports reboot action
fenced_df_supports_reboot = (UINT32_C(1) << 2),
//! Device supports status action
fenced_df_supports_status = (UINT32_C(1) << 3),
//! Device is automatically used to unfence newly joined nodes
fenced_df_auto_unfence = (UINT32_C(1) << 4),
//! Device has run a successful list, status, or monitor action on this node
fenced_df_verified = (UINT32_C(1) << 5),
//! Device has been registered via the stonith API
fenced_df_api_registered = (UINT32_C(1) << 6),
//! Device has been registered via the fencer's CIB diff callback
fenced_df_cib_registered = (UINT32_C(1) << 7),
//! Device has not yet been re-registered after a CIB change
fenced_df_dirty = (UINT32_C(1) << 8),
};
/*!
* \internal
* \brief Set flags for a fencing device
*
* \param[in,out] device Device whose flags to set (\c fenced_device_t)
* \param[in] set_flags Group of <tt>enum fenced_device_flags</tt> to set
*/
#define fenced_device_set_flags(device, set_flags) do { \
pcmk__assert((device) != NULL); \
(device)->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \
"Fence device", (device)->id, \
(device)->flags, set_flags, \
#set_flags); \
} while (0)
/*!
* \internal
* \brief Clear flags for a fencing device
*
* \param[in,out] device Device whose flags to clear (\c fenced_device_t)
* \param[in] clear_flags Group of <tt>enum fenced_device_flags</tt> to
* clear
*/
#define fenced_device_clear_flags(device, clear_flags) do { \
pcmk__assert((device) != NULL); \
(device)->flags = pcmk__clear_flags_as(__func__, __LINE__, \
LOG_TRACE, "Fence device", \
(device)->id, \
(device)->flags, \
clear_flags, #clear_flags); \
} while (0)
/*!
* \internal
* \brief Flags for fencer client notification types
*/
enum fenced_notify_flags {
//! This flag has no effect
fenced_nf_none = UINT32_C(0),
//! Notify about fencing operation results
fenced_nf_fence_result = (UINT32_C(1) << 0),
// @TODO Consider notifying about device registrations via the CIB
//! Notify about fencing device registrations via the fencer API
fenced_nf_device_registered = (UINT32_C(1) << 1),
- st_callback_device_del = (UINT32_C(1) << 4),
+ // @TODO Consider notifying about device removals via the CIB
+ //! Notify about fencing device removals via the fencer API
+ fenced_nf_device_removed = (UINT32_C(1) << 2),
+
st_callback_notify_history = (UINT32_C(1) << 5),
st_callback_notify_history_synced = (UINT32_C(1) << 6),
};
typedef struct {
char *id;
char *agent;
char *namespace;
/*! list of actions that must execute on the target node. Used for unfencing */
GString *on_target_actions;
GList *targets;
time_t targets_age;
uint32_t flags; // Group of enum fenced_device_flags
GHashTable *params;
GHashTable *aliases;
GList *pending_ops;
mainloop_timer_t *timer;
crm_trigger_t *work;
xmlNode *agent_metadata;
const char *default_host_arg;
} fenced_device_t;
/* These values are used to index certain arrays by "phase". Usually an
* operation has only one "phase", so phase is always zero. However, some
* reboots are remapped to "off" then "on", in which case "reboot" will be
* phase 0, "off" will be phase 1 and "on" will be phase 2.
*/
enum st_remap_phase {
st_phase_requested = 0,
st_phase_off = 1,
st_phase_on = 2,
st_phase_max = 3
};
typedef struct remote_fencing_op_s {
/* @TODO Abstract the overlap with async_command_t (some members have
* different names for the same thing), which should allow reducing
* duplication in some functions
*/
/* The unique id associated with this operation */
char *id;
/*! The node this operation will fence */
char *target;
/*! The fencing action to perform on the target. (reboot, on, off) */
char *action;
/*! When was the fencing action recorded (seconds since epoch) */
time_t created;
/*! Marks if the final notifications have been sent to local stonith clients. */
gboolean notify_sent;
/*! The number of query replies received */
guint replies;
/*! The number of query replies expected */
guint replies_expected;
/*! Does this node own control of this operation */
gboolean owner;
/*! After query is complete, This the high level timer that expires the entire operation */
guint op_timer_total;
/*! This timer expires the current fencing request. Many fencing
* requests may exist in a single operation */
guint op_timer_one;
/*! This timer expires the query request sent out to determine
* what nodes are contain what devices, and who those devices can fence */
guint query_timer;
/*! This is the default timeout to use for each fencing device if no
* custom timeout is received in the query. */
gint base_timeout;
/*! This is the calculated total timeout an operation can take before
* expiring. This is calculated by adding together all the timeout
* values associated with the devices this fencing operation may call */
gint total_timeout;
/*!
* Fencing delay (in seconds) requested by API client (used by controller to
* implement \c PCMK_OPT_PRIORITY_FENCING_DELAY). A value of -1 means
* disable all configured delays.
*/
int client_delay;
/*! Delegate is the node being asked to perform a fencing action
* on behalf of the node that owns the remote operation. Some operations
* will involve multiple delegates. This value represents the final delegate
* that is used. */
char *delegate;
/*! The point at which the remote operation completed */
time_t completed;
//! Group of enum stonith_call_options associated with this operation
uint32_t call_options;
/*! The current state of the remote operation. This indicates
* what stage the op is in, query, exec, done, duplicate, failed. */
enum op_state state;
/*! The node that owns the remote operation */
char *originator;
/*! The local client id that initiated the fencing request */
char *client_id;
/*! The client's call_id that initiated the fencing request */
int client_callid;
/*! The name of client that initiated the fencing request */
char *client_name;
/*! List of the received query results for all the nodes in the cpg group */
GList *query_results;
/*! The original request that initiated the remote stonith operation */
xmlNode *request;
/*! The current topology level being executed */
guint level;
/*! The current operation phase being executed */
enum st_remap_phase phase;
/*! Devices with automatic unfencing (always run if "on" requested, never if remapped) */
GList *automatic_list;
/*! List of all devices at the currently executing topology level */
GList *devices_list;
/*! Current entry in the topology device list */
GList *devices;
/*! List of duplicate operations attached to this operation. Once this operation
* completes, the duplicate operations will be closed out as well. */
GList *duplicates;
/*! The point at which the remote operation completed(nsec) */
long long completed_nsec;
/*! The (potentially intermediate) result of the operation */
pcmk__action_result_t result;
} remote_fencing_op_t;
void fenced_broadcast_op_result(const remote_fencing_op_t *op, bool op_merged);
// How the user specified the target of a topology level
enum fenced_target_by {
fenced_target_by_unknown = -1, // Invalid or not yet parsed
fenced_target_by_name, // By target name
fenced_target_by_pattern, // By a pattern matching target names
fenced_target_by_attribute, // By a node attribute/value on target
};
/*
* Complex fencing requirements are specified via fencing topologies.
* A topology consists of levels; each level is a list of fencing devices.
* Topologies are stored in a hash table by node name. When a node needs to be
* fenced, if it has an entry in the topology table, the levels are tried
* sequentially, and the devices in each level are tried sequentially.
* Fencing is considered successful as soon as any level succeeds;
* a level is considered successful if all its devices succeed.
* Essentially, all devices at a given level are "and-ed" and the
* levels are "or-ed".
*
* This structure is used for the topology table entries.
* Topology levels start from 1, so levels[0] is unused and always NULL.
*/
typedef struct stonith_topology_s {
enum fenced_target_by kind; // How target was specified
/*! Node name regex or attribute name=value for which topology applies */
char *target;
char *target_value;
char *target_pattern;
char *target_attribute;
/*! Names of fencing devices at each topology level */
GList *levels[ST__LEVEL_COUNT];
} stonith_topology_t;
void stonith_shutdown(int nsig);
void fenced_init_device_table(void);
void fenced_free_device_table(void);
bool fenced_has_watchdog_device(void);
void fenced_foreach_device(GHFunc fn, gpointer user_data);
void fenced_foreach_device_remove(GHRFunc fn);
void init_topology_list(void);
void free_topology_list(void);
void free_stonith_remote_op_list(void);
void init_stonith_remote_op_hash_table(GHashTable **table);
void free_metadata_cache(void);
void fenced_unregister_handlers(void);
uint64_t get_stonith_flag(const char *name);
void stonith_command(pcmk__client_t *client, uint32_t id, uint32_t flags,
xmlNode *op_request, const char *remote_peer);
int fenced_device_register(const xmlNode *dev, bool from_cib);
void stonith_device_remove(const char *id, bool from_cib);
char *stonith_level_key(const xmlNode *msg, enum fenced_target_by);
void fenced_register_level(xmlNode *msg, char **desc,
pcmk__action_result_t *result);
void fenced_unregister_level(xmlNode *msg, char **desc,
pcmk__action_result_t *result);
stonith_topology_t *find_topology_for_host(const char *host);
void do_local_reply(const xmlNode *notify_src, pcmk__client_t *client,
int call_options);
xmlNode *fenced_construct_reply(const xmlNode *request, xmlNode *data,
const pcmk__action_result_t *result);
void
do_stonith_async_timeout_update(const char *client, const char *call_id, int timeout);
void fenced_send_notification(const char *type,
const pcmk__action_result_t *result,
xmlNode *data);
void fenced_send_config_notification(const char *op,
const pcmk__action_result_t *result,
const char *desc);
remote_fencing_op_t *initiate_remote_stonith_op(const pcmk__client_t *client,
xmlNode *request,
gboolean manual_ack);
void fenced_process_fencing_reply(xmlNode *msg);
int process_remote_stonith_query(xmlNode * msg);
void *create_remote_stonith_op(const char *client, xmlNode * request, gboolean peer);
void stonith_fence_history(xmlNode *msg, xmlNode **output,
const char *remote_peer, int options);
void stonith_fence_history_trim(void);
bool fencing_peer_active(pcmk__node_status_t *peer);
void set_fencing_completed(remote_fencing_op_t * op);
int fenced_handle_manual_confirmation(const pcmk__client_t *client,
xmlNode *msg);
const char *fenced_device_reboot_action(const char *device_id);
bool fenced_device_supports_on(const char *device_id);
gboolean node_has_attr(const char *node, const char *name, const char *value);
gboolean node_does_watchdog_fencing(const char *node);
void fencing_topology_init(void);
void setup_cib(void);
void fenced_cib_cleanup(void);
int fenced_scheduler_init(void);
void fenced_set_local_node(const char *node_name);
const char *fenced_get_local_node(void);
void fenced_scheduler_cleanup(void);
void fenced_scheduler_run(xmlNode *cib);
static inline void
fenced_set_protocol_error(pcmk__action_result_t *result)
{
pcmk__set_result(result, CRM_EX_PROTOCOL, PCMK_EXEC_INVALID,
"Fencer API request missing required information (bug?)");
}
/*!
* \internal
* \brief Get the device flag to use with a given action when searching devices
*
* \param[in] action Action to check
*
* \return \c fenced_df_supports_on if \p action is "on", otherwise
* \c fenced_df_none
*/
static inline uint32_t
fenced_support_flag(const char *action)
{
if (pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none)) {
return fenced_df_supports_on;
}
return fenced_df_none;
}
extern GHashTable *topology;
extern long long stonith_watchdog_timeout_ms;
extern GList *stonith_watchdog_targets;
extern GHashTable *stonith_remote_op_list;
extern crm_exit_t exit_code;
extern gboolean stonith_shutdown_flag;