diff --git a/daemons/fenced/fenced_cib.c b/daemons/fenced/fenced_cib.c
index 6aac039e35..4737d46dea 100644
--- a/daemons/fenced/fenced_cib.c
+++ b/daemons/fenced/fenced_cib.c
@@ -1,893 +1,885 @@
/*
* Copyright 2009-2023 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 <stdbool.h>
#include <stdio.h>
#include <libxml/tree.h>
#include <libxml/xpath.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/cluster/internal.h>
#include <crm/cib.h>
#include <crm/cib/internal.h>
#include <crm/pengine/status.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include <pacemaker-fenced.h>
#define rsc_name(x) x->clone_name?x->clone_name:x->id
extern pcmk_scheduler_t *fenced_data_set;
static xmlNode *local_cib = NULL;
static cib_t *cib_api = NULL;
static bool have_cib_devices = FALSE;
-static const unsigned long long data_set_flags = pcmk_sched_location_only
- |pcmk_sched_no_compat
- |pcmk_sched_no_counts;
/*!
* \internal
* \brief Check whether a node has a specific attribute name/value
*
* \param[in] node Name of node to check
* \param[in] name Name of an attribute to look for
* \param[in] value The value the named attribute needs to be set to in order to be considered a match
*
* \return TRUE if the locally cached CIB has the specified node attribute
*/
gboolean
node_has_attr(const char *node, const char *name, const char *value)
{
GString *xpath = NULL;
xmlNode *match;
CRM_CHECK((local_cib != NULL) && (node != NULL) && (name != NULL)
&& (value != NULL), return FALSE);
/* Search for the node's attributes in the CIB. While the schema allows
* multiple sets of instance attributes, and allows instance attributes to
* use id-ref to reference values elsewhere, that is intended for resources,
* so we ignore that here.
*/
xpath = g_string_sized_new(256);
pcmk__g_strcat(xpath,
"//" XML_CIB_TAG_NODES "/" XML_CIB_TAG_NODE
"[@" XML_ATTR_UNAME "='", node, "']/" XML_TAG_ATTR_SETS
"/" XML_CIB_TAG_NVPAIR
"[@" XML_NVPAIR_ATTR_NAME "='", name, "' "
"and @" XML_NVPAIR_ATTR_VALUE "='", value, "']", NULL);
match = get_xpath_object((const char *) xpath->str, local_cib, LOG_NEVER);
g_string_free(xpath, TRUE);
return (match != NULL);
}
static void
add_topology_level(xmlNode *match)
{
char *desc = NULL;
pcmk__action_result_t result = PCMK__UNKNOWN_RESULT;
CRM_CHECK(match != NULL, return);
fenced_register_level(match, &desc, &result);
fenced_send_level_notification(STONITH_OP_LEVEL_ADD, &result, desc);
pcmk__reset_result(&result);
free(desc);
}
static void
topology_remove_helper(const char *node, int level)
{
char *desc = NULL;
pcmk__action_result_t result = PCMK__UNKNOWN_RESULT;
xmlNode *data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL);
crm_xml_add(data, F_STONITH_ORIGIN, __func__);
crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level);
crm_xml_add(data, XML_ATTR_STONITH_TARGET, node);
fenced_unregister_level(data, &desc, &result);
fenced_send_level_notification(STONITH_OP_LEVEL_DEL, &result, desc);
pcmk__reset_result(&result);
free_xml(data);
free(desc);
}
static void
remove_topology_level(xmlNode *match)
{
int index = 0;
char *key = NULL;
CRM_CHECK(match != NULL, return);
key = stonith_level_key(match, fenced_target_by_unknown);
crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index);
topology_remove_helper(key, index);
free(key);
}
static void
register_fencing_topology(xmlXPathObjectPtr xpathObj)
{
int max = numXpathResults(xpathObj), lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *match = getXpathResult(xpathObj, lpc);
remove_topology_level(match);
add_topology_level(match);
}
}
/* Fencing
<diff crm_feature_set="3.0.6">
<diff-removed>
<fencing-topology>
<fencing-level id="f-p1.1" target="pcmk-1" index="1" devices="poison-pill" __crm_diff_marker__="removed:top"/>
<fencing-level id="f-p1.2" target="pcmk-1" index="2" devices="power" __crm_diff_marker__="removed:top"/>
<fencing-level devices="disk,network" id="f-p2.1"/>
</fencing-topology>
</diff-removed>
<diff-added>
<fencing-topology>
<fencing-level id="f-p.1" target="pcmk-1" index="1" devices="poison-pill" __crm_diff_marker__="added:top"/>
<fencing-level id="f-p2.1" target="pcmk-2" index="1" devices="disk,something"/>
<fencing-level id="f-p3.1" target="pcmk-2" index="2" devices="power" __crm_diff_marker__="added:top"/>
</fencing-topology>
</diff-added>
</diff>
*/
void
fencing_topology_init(void)
{
xmlXPathObjectPtr xpathObj = NULL;
const char *xpath = "//" XML_TAG_FENCING_LEVEL;
crm_trace("Full topology refresh");
free_topology_list();
init_topology_list();
/* Grab everything */
xpathObj = xpath_search(local_cib, xpath);
register_fencing_topology(xpathObj);
freeXpathObject(xpathObj);
}
static void
remove_cib_device(xmlXPathObjectPtr xpathObj)
{
int max = numXpathResults(xpathObj), lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
const char *rsc_id = NULL;
const char *standard = NULL;
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if(match != NULL) {
standard = crm_element_value(match, XML_AGENT_ATTR_CLASS);
}
if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
continue;
}
rsc_id = crm_element_value(match, XML_ATTR_ID);
stonith_device_remove(rsc_id, true);
}
}
static void
update_stonith_watchdog_timeout_ms(xmlNode *cib)
{
long timeout_ms = 0;
xmlNode *stonith_watchdog_xml = NULL;
const char *value = NULL;
stonith_watchdog_xml = get_xpath_object("//nvpair[@name='stonith-watchdog-timeout']",
cib, LOG_NEVER);
if (stonith_watchdog_xml) {
value = crm_element_value(stonith_watchdog_xml, XML_NVPAIR_ATTR_VALUE);
}
if (value) {
timeout_ms = crm_get_msec(value);
}
if (timeout_ms < 0) {
timeout_ms = pcmk__auto_watchdog_timeout();
}
stonith_watchdog_timeout_ms = timeout_ms;
}
/*!
* \internal
* \brief Check whether our uname is in a resource's allowed node list
*
* \param[in] rsc Resource to check
*
* \return Pointer to node object if found, NULL otherwise
*/
static pcmk_node_t *
our_node_allowed_for(const pcmk_resource_t *rsc)
{
GHashTableIter iter;
pcmk_node_t *node = NULL;
if (rsc && stonith_our_uname) {
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
if (node && strcmp(node->details->uname, stonith_our_uname) == 0) {
break;
}
node = NULL;
}
}
return node;
}
/*!
* \internal
* \brief If a resource or any of its children are STONITH devices, update their
* definitions given a cluster working set.
*
* \param[in,out] rsc Resource to check
* \param[in,out] data_set Cluster working set with device information
*/
static void
cib_device_update(pcmk_resource_t *rsc, pcmk_scheduler_t *data_set)
{
pcmk_node_t *node = NULL;
const char *value = NULL;
const char *rclass = NULL;
pcmk_node_t *parent = NULL;
/* If this is a complex resource, check children rather than this resource itself. */
if(rsc->children) {
GList *gIter = NULL;
for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) {
cib_device_update(gIter->data, data_set);
if(pe_rsc_is_clone(rsc)) {
crm_trace("Only processing one copy of the clone %s", rsc->id);
break;
}
}
return;
}
/* We only care about STONITH resources. */
rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS);
if (!pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
return;
}
/* If this STONITH resource is disabled, remove it. */
if (pe__resource_is_disabled(rsc)) {
crm_info("Device %s has been disabled", rsc->id);
return;
}
/* if watchdog-fencing is disabled handle any watchdog-fence
resource as if it was disabled
*/
if ((stonith_watchdog_timeout_ms <= 0) &&
pcmk__str_eq(rsc->id, STONITH_WATCHDOG_ID, pcmk__str_none)) {
crm_info("Watchdog-fencing disabled thus handling "
"device %s as disabled", rsc->id);
return;
}
/* Check whether our node is allowed for this resource (and its parent if in a group) */
node = our_node_allowed_for(rsc);
if (rsc->parent && (rsc->parent->variant == pcmk_rsc_variant_group)) {
parent = our_node_allowed_for(rsc->parent);
}
if(node == NULL) {
/* Our node is disallowed, so remove the device */
GHashTableIter iter;
crm_info("Device %s has been disabled on %s: unknown", rsc->id, stonith_our_uname);
g_hash_table_iter_init(&iter, rsc->allowed_nodes);
while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) {
crm_trace("Available: %s = %d", pe__node_name(node), node->weight);
}
return;
} else if(node->weight < 0 || (parent && parent->weight < 0)) {
/* Our node (or its group) is disallowed by score, so remove the device */
int score = (node->weight < 0)? node->weight : parent->weight;
crm_info("Device %s has been disabled on %s: score=%s",
rsc->id, stonith_our_uname, pcmk_readable_score(score));
return;
} else {
/* Our node is allowed, so update the device information */
int rc;
xmlNode *data;
GHashTable *rsc_params = NULL;
GHashTableIter gIter;
stonith_key_value_t *params = NULL;
const char *name = NULL;
const char *agent = crm_element_value(rsc->xml, XML_EXPR_ATTR_TYPE);
const char *rsc_provides = NULL;
crm_debug("Device %s is allowed on %s: score=%d", rsc->id, stonith_our_uname, node->weight);
rsc_params = pe_rsc_params(rsc, node, data_set);
get_meta_attributes(rsc->meta, rsc, node, data_set);
rsc_provides = g_hash_table_lookup(rsc->meta, PCMK_STONITH_PROVIDES);
g_hash_table_iter_init(&gIter, rsc_params);
while (g_hash_table_iter_next(&gIter, (gpointer *) & name, (gpointer *) & value)) {
if (!name || !value) {
continue;
}
params = stonith_key_value_add(params, name, value);
crm_trace(" %s=%s", name, value);
}
data = create_device_registration_xml(rsc_name(rsc), st_namespace_any,
agent, params, rsc_provides);
stonith_key_value_freeall(params, 1, 1);
rc = stonith_device_register(data, TRUE);
CRM_ASSERT(rc == pcmk_ok);
free_xml(data);
}
}
/*!
* \internal
* \brief Update all STONITH device definitions based on current CIB
*/
static void
cib_devices_update(void)
{
GHashTableIter iter;
stonith_device_t *device = NULL;
crm_info("Updating devices to version %s.%s.%s",
crm_element_value(local_cib, XML_ATTR_GENERATION_ADMIN),
crm_element_value(local_cib, XML_ATTR_GENERATION),
crm_element_value(local_cib, XML_ATTR_NUMUPDATES));
- if (fenced_data_set->now != NULL) {
- crm_time_free(fenced_data_set->now);
- fenced_data_set->now = NULL;
- }
- fenced_data_set->localhost = stonith_our_uname;
- pcmk__schedule_actions(local_cib, data_set_flags, fenced_data_set);
+ fenced_scheduler_run(local_cib);
g_hash_table_iter_init(&iter, device_list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) {
if (device->cib_registered) {
device->dirty = TRUE;
}
}
/* have list repopulated if cib has a watchdog-fencing-resource
TODO: keep a cached list for queries happening while we are refreshing
*/
g_list_free_full(stonith_watchdog_targets, free);
stonith_watchdog_targets = NULL;
g_list_foreach(fenced_data_set->resources, (GFunc) cib_device_update, fenced_data_set);
g_hash_table_iter_init(&iter, device_list);
while (g_hash_table_iter_next(&iter, NULL, (void **)&device)) {
if (device->dirty) {
g_hash_table_iter_remove(&iter);
}
}
fenced_data_set->input = NULL; // Wasn't a copy, so don't let API free it
pe_reset_working_set(fenced_data_set);
}
static void
update_cib_stonith_devices_v1(const char *event, xmlNode * msg)
{
const char *reason = "none";
gboolean needs_update = FALSE;
xmlXPathObjectPtr xpath_obj = NULL;
/* process new constraints */
xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_CONS_TAG_RSC_LOCATION);
if (numXpathResults(xpath_obj) > 0) {
int max = numXpathResults(xpath_obj), lpc = 0;
/* Safest and simplest to always recompute */
needs_update = TRUE;
reason = "new location constraint";
for (lpc = 0; lpc < max; lpc++) {
xmlNode *match = getXpathResult(xpath_obj, lpc);
crm_log_xml_trace(match, "new constraint");
}
}
freeXpathObject(xpath_obj);
/* process deletions */
xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_RESOURCE);
if (numXpathResults(xpath_obj) > 0) {
remove_cib_device(xpath_obj);
}
freeXpathObject(xpath_obj);
/* process additions */
xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_RESOURCE);
if (numXpathResults(xpath_obj) > 0) {
int max = numXpathResults(xpath_obj), lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
const char *rsc_id = NULL;
const char *standard = NULL;
xmlNode *match = getXpathResult(xpath_obj, lpc);
rsc_id = crm_element_value(match, XML_ATTR_ID);
standard = crm_element_value(match, XML_AGENT_ATTR_CLASS);
if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) {
continue;
}
crm_trace("Fencing resource %s was added or modified", rsc_id);
reason = "new resource";
needs_update = TRUE;
}
}
freeXpathObject(xpath_obj);
if(needs_update) {
crm_info("Updating device list from CIB: %s", reason);
cib_devices_update();
}
}
static void
update_cib_stonith_devices_v2(const char *event, xmlNode * msg)
{
xmlNode *change = NULL;
char *reason = NULL;
bool needs_update = FALSE;
xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT);
for (change = pcmk__xml_first_child(patchset); change != NULL;
change = pcmk__xml_next(change)) {
const char *op = crm_element_value(change, XML_DIFF_OP);
const char *xpath = crm_element_value(change, XML_DIFF_PATH);
const char *shortpath = NULL;
if ((op == NULL) ||
(strcmp(op, "move") == 0) ||
strstr(xpath, "/"XML_CIB_TAG_STATUS)) {
continue;
} else if (pcmk__str_eq(op, "delete", pcmk__str_casei) && strstr(xpath, "/"XML_CIB_TAG_RESOURCE)) {
const char *rsc_id = NULL;
char *search = NULL;
char *mutable = NULL;
if (strstr(xpath, XML_TAG_ATTR_SETS) ||
strstr(xpath, XML_TAG_META_SETS)) {
needs_update = TRUE;
pcmk__str_update(&reason,
"(meta) attribute deleted from resource");
break;
}
pcmk__str_update(&mutable, xpath);
rsc_id = strstr(mutable, "primitive[@" XML_ATTR_ID "=\'");
if (rsc_id != NULL) {
rsc_id += strlen("primitive[@" XML_ATTR_ID "=\'");
search = strchr(rsc_id, '\'');
}
if (search != NULL) {
*search = 0;
stonith_device_remove(rsc_id, true);
/* watchdog_device_update called afterwards
to fall back to implicit definition if needed */
} else {
crm_warn("Ignoring malformed CIB update (resource deletion)");
}
free(mutable);
} else if (strstr(xpath, "/"XML_CIB_TAG_RESOURCES) ||
strstr(xpath, "/"XML_CIB_TAG_CONSTRAINTS) ||
strstr(xpath, "/"XML_CIB_TAG_RSCCONFIG)) {
shortpath = strrchr(xpath, '/'); CRM_ASSERT(shortpath);
reason = crm_strdup_printf("%s %s", op, shortpath+1);
needs_update = TRUE;
break;
}
}
if(needs_update) {
crm_info("Updating device list from CIB: %s", reason);
cib_devices_update();
} else {
crm_trace("No updates for device list found in CIB");
}
free(reason);
}
static void
update_cib_stonith_devices(const char *event, xmlNode * msg)
{
int format = 1;
xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT);
CRM_ASSERT(patchset);
crm_element_value_int(patchset, PCMK_XA_FORMAT, &format);
switch(format) {
case 1:
update_cib_stonith_devices_v1(event, msg);
break;
case 2:
update_cib_stonith_devices_v2(event, msg);
break;
default:
crm_warn("Unknown patch format: %d", format);
}
}
static void
watchdog_device_update(void)
{
if (stonith_watchdog_timeout_ms > 0) {
if (!g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID) &&
!stonith_watchdog_targets) {
/* getting here watchdog-fencing enabled, no device there yet
and reason isn't stonith_watchdog_targets preventing that
*/
int rc;
xmlNode *xml;
xml = create_device_registration_xml(
STONITH_WATCHDOG_ID,
st_namespace_internal,
STONITH_WATCHDOG_AGENT,
NULL, /* stonith_device_register will add our
own name as PCMK_STONITH_HOST_LIST param
so we can skip that here
*/
NULL);
rc = stonith_device_register(xml, TRUE);
free_xml(xml);
if (rc != pcmk_ok) {
rc = pcmk_legacy2rc(rc);
exit_code = CRM_EX_FATAL;
crm_crit("Cannot register watchdog pseudo fence agent: %s",
pcmk_rc_str(rc));
stonith_shutdown(0);
}
}
} else if (g_hash_table_lookup(device_list, STONITH_WATCHDOG_ID) != NULL) {
/* be silent if no device - todo parameter to stonith_device_remove */
stonith_device_remove(STONITH_WATCHDOG_ID, true);
}
}
/*!
* \internal
* \brief Query the full CIB
*
* \return Standard Pacemaker return code
*/
static int
fenced_query_cib(void)
{
int rc = pcmk_ok;
crm_trace("Re-requesting full CIB");
rc = cib_api->cmds->query(cib_api, NULL, &local_cib,
cib_scope_local|cib_sync_call);
rc = pcmk_legacy2rc(rc);
if (rc == pcmk_rc_ok) {
CRM_ASSERT(local_cib != NULL);
} else {
crm_err("Couldn't retrieve the CIB: %s " CRM_XS " rc=%d",
pcmk_rc_str(rc), rc);
}
return rc;
}
static void
remove_fencing_topology(xmlXPathObjectPtr xpathObj)
{
int max = numXpathResults(xpathObj), lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *match = getXpathResult(xpathObj, lpc);
CRM_LOG_ASSERT(match != NULL);
if (match && crm_element_value(match, XML_DIFF_MARKER)) {
/* Deletion */
int index = 0;
char *target = stonith_level_key(match, fenced_target_by_unknown);
crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index);
if (target == NULL) {
crm_err("Invalid fencing target in element %s", ID(match));
} else if (index <= 0) {
crm_err("Invalid level for %s in element %s", target, ID(match));
} else {
topology_remove_helper(target, index);
}
/* } else { Deal with modifications during the 'addition' stage */
}
}
}
static void
update_fencing_topology(const char *event, xmlNode * msg)
{
int format = 1;
const char *xpath;
xmlXPathObjectPtr xpathObj = NULL;
xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT);
CRM_ASSERT(patchset);
crm_element_value_int(patchset, PCMK_XA_FORMAT, &format);
if(format == 1) {
/* Process deletions (only) */
xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_TAG_FENCING_LEVEL;
xpathObj = xpath_search(msg, xpath);
remove_fencing_topology(xpathObj);
freeXpathObject(xpathObj);
/* Process additions and changes */
xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_TAG_FENCING_LEVEL;
xpathObj = xpath_search(msg, xpath);
register_fencing_topology(xpathObj);
freeXpathObject(xpathObj);
} else if(format == 2) {
xmlNode *change = NULL;
int add[] = { 0, 0, 0 };
int del[] = { 0, 0, 0 };
xml_patch_versions(patchset, add, del);
for (change = pcmk__xml_first_child(patchset); change != NULL;
change = pcmk__xml_next(change)) {
const char *op = crm_element_value(change, XML_DIFF_OP);
const char *xpath = crm_element_value(change, XML_DIFF_PATH);
if(op == NULL) {
continue;
} else if(strstr(xpath, "/" XML_TAG_FENCING_LEVEL) != NULL) {
/* Change to a specific entry */
crm_trace("Handling %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath);
if(strcmp(op, "move") == 0) {
continue;
} else if(strcmp(op, "create") == 0) {
add_topology_level(change->children);
} else if(strcmp(op, "modify") == 0) {
xmlNode *match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
remove_topology_level(match->children);
add_topology_level(match->children);
}
} else if(strcmp(op, "delete") == 0) {
/* Nuclear option, all we have is the path and an id... not enough to remove a specific entry */
crm_info("Re-initializing fencing topology after %s operation %d.%d.%d for %s",
op, add[0], add[1], add[2], xpath);
fencing_topology_init();
return;
}
} else if (strstr(xpath, "/" XML_TAG_FENCING_TOPOLOGY) != NULL) {
/* Change to the topology in general */
crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s",
op, add[0], add[1], add[2], xpath);
fencing_topology_init();
return;
} else if (strstr(xpath, "/" XML_CIB_TAG_CONFIGURATION)) {
/* Changes to the whole config section, possibly including the topology as a whild */
if(first_named_child(change, XML_TAG_FENCING_TOPOLOGY) == NULL) {
crm_trace("Nothing for us in %s operation %d.%d.%d for %s.",
op, add[0], add[1], add[2], xpath);
} else if(strcmp(op, "delete") == 0 || strcmp(op, "create") == 0) {
crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s.",
op, add[0], add[1], add[2], xpath);
fencing_topology_init();
return;
}
} else {
crm_trace("Nothing for us in %s operation %d.%d.%d for %s",
op, add[0], add[1], add[2], xpath);
}
}
} else {
crm_warn("Unknown patch format: %d", format);
}
}
static void
update_cib_cache_cb(const char *event, xmlNode * msg)
{
long timeout_ms_saved = stonith_watchdog_timeout_ms;
bool need_full_refresh = false;
if(!have_cib_devices) {
crm_trace("Skipping updates until we get a full dump");
return;
} else if(msg == NULL) {
crm_trace("Missing %s update", event);
return;
}
/* Maintain a local copy of the CIB so that we have full access
* to device definitions, location constraints, and node attributes
*/
if (local_cib != NULL) {
int rc = pcmk_ok;
xmlNode *patchset = NULL;
crm_element_value_int(msg, F_CIB_RC, &rc);
if (rc != pcmk_ok) {
return;
}
patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT);
rc = xml_apply_patchset(local_cib, patchset, TRUE);
switch (rc) {
case pcmk_ok:
case -pcmk_err_old_data:
break;
case -pcmk_err_diff_resync:
case -pcmk_err_diff_failed:
crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc);
free_xml(local_cib);
local_cib = NULL;
break;
default:
crm_warn("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc);
free_xml(local_cib);
local_cib = NULL;
}
}
if (local_cib == NULL) {
if (fenced_query_cib() != pcmk_rc_ok) {
return;
}
need_full_refresh = true;
}
pcmk__refresh_node_caches_from_cib(local_cib);
update_stonith_watchdog_timeout_ms(local_cib);
if (timeout_ms_saved != stonith_watchdog_timeout_ms) {
need_full_refresh = true;
}
if (need_full_refresh) {
fencing_topology_init();
cib_devices_update();
} else {
// Partial refresh
update_fencing_topology(event, msg);
update_cib_stonith_devices(event, msg);
}
watchdog_device_update();
}
static void
init_cib_cache_cb(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
crm_info("Updating device list from CIB");
have_cib_devices = TRUE;
local_cib = copy_xml(output);
pcmk__refresh_node_caches_from_cib(local_cib);
update_stonith_watchdog_timeout_ms(local_cib);
fencing_topology_init();
cib_devices_update();
watchdog_device_update();
}
static void
cib_connection_destroy(gpointer user_data)
{
if (stonith_shutdown_flag) {
crm_info("Connection to the CIB manager closed");
return;
} else {
crm_crit("Lost connection to the CIB manager, shutting down");
}
if (cib_api) {
cib_api->cmds->signoff(cib_api);
}
stonith_shutdown(0);
}
/*!
* \internal
* \brief Disconnect from CIB manager
*/
void
fenced_cib_cleanup(void)
{
if (cib_api != NULL) {
cib_api->cmds->del_notify_callback(cib_api, T_CIB_DIFF_NOTIFY,
update_cib_cache_cb);
cib__clean_up_connection(&cib_api);
}
free_xml(local_cib);
local_cib = NULL;
}
void
setup_cib(void)
{
int rc, retries = 0;
cib_api = cib_new();
if (cib_api == NULL) {
crm_err("No connection to the CIB manager");
return;
}
do {
sleep(retries);
rc = cib_api->cmds->signon(cib_api, CRM_SYSTEM_STONITHD, cib_command);
} while (rc == -ENOTCONN && ++retries < 5);
if (rc != pcmk_ok) {
crm_err("Could not connect to the CIB manager: %s (%d)", pcmk_strerror(rc), rc);
} else if (pcmk_ok !=
cib_api->cmds->add_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb)) {
crm_err("Could not set CIB notification callback");
} else {
rc = cib_api->cmds->query(cib_api, NULL, NULL, cib_scope_local);
cib_api->cmds->register_callback(cib_api, rc, 120, FALSE, NULL, "init_cib_cache_cb",
init_cib_cache_cb);
cib_api->cmds->set_connection_dnotify(cib_api, cib_connection_destroy);
crm_info("Watching for fencing topology changes");
}
}
diff --git a/daemons/fenced/fenced_scheduler.c b/daemons/fenced/fenced_scheduler.c
index 3e9aa32e13..7d4f5aafae 100644
--- a/daemons/fenced/fenced_scheduler.c
+++ b/daemons/fenced/fenced_scheduler.c
@@ -1,71 +1,92 @@
/*
* Copyright 2009-2023 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 <stdio.h>
#include <errno.h>
#include <crm/pengine/status.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include <pacemaker-fenced.h>
pcmk_scheduler_t *fenced_data_set = NULL;
/*!
* \internal
* \brief Initialize scheduler data for fencer purposes
*
* \return Standard Pacemaker return code
*/
int
fenced_scheduler_init(void)
{
pcmk__output_t *logger = NULL;
int rc = pcmk__log_output_new(&logger);
if (rc != pcmk_rc_ok) {
return rc;
}
fenced_data_set = pe_new_working_set();
if (fenced_data_set == NULL) {
pcmk__output_free(logger);
return ENOMEM;
}
pe__register_messages(logger);
pcmk__register_lib_messages(logger);
pcmk__output_set_log_level(logger, LOG_TRACE);
fenced_data_set->priv = logger;
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Free all scheduler-related resources
*/
void
fenced_scheduler_cleanup(void)
{
if (fenced_data_set != NULL) {
pcmk__output_t *logger = fenced_data_set->priv;
if (logger != NULL) {
logger->finish(logger, CRM_EX_OK, true, NULL);
pcmk__output_free(logger);
fenced_data_set->priv = NULL;
}
pe_free_working_set(fenced_data_set);
fenced_data_set = NULL;
}
}
+
+/*!
+ * \internal
+ * \brief Run the scheduler for fencer purposes
+ *
+ * \param[in] cib Cluster's current CIB
+ */
+void
+fenced_scheduler_run(xmlNode *cib)
+{
+ CRM_CHECK((cib != NULL) && (fenced_data_set != NULL), return);
+
+ if (fenced_data_set->now != NULL) {
+ crm_time_free(fenced_data_set->now);
+ fenced_data_set->now = NULL;
+ }
+ fenced_data_set->localhost = stonith_our_uname;
+ pcmk__schedule_actions(cib, pcmk_sched_location_only
+ |pcmk_sched_no_compat
+ |pcmk_sched_no_counts, fenced_data_set);
+}
diff --git a/daemons/fenced/pacemaker-fenced.h b/daemons/fenced/pacemaker-fenced.h
index c2acf184bd..220978a5a6 100644
--- a/daemons/fenced/pacemaker-fenced.h
+++ b/daemons/fenced/pacemaker-fenced.h
@@ -1,331 +1,334 @@
/*
* Copyright 2009-2023 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);
typedef struct stonith_device_s {
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;
gboolean has_attr_map;
// Whether target's nodeid should be passed as a parameter to the agent
gboolean include_nodeid;
/* whether the cluster should automatically unfence nodes with the device */
gboolean automatic_unfencing;
guint priority;
uint32_t flags; // Group of enum st_device_flags
GHashTable *params;
GHashTable *aliases;
GList *pending_ops;
mainloop_timer_t *timer;
crm_trigger_t *work;
xmlNode *agent_metadata;
/*! A verified device is one that has contacted the
* agent successfully to perform a monitor operation */
gboolean verified;
gboolean cib_registered;
gboolean api_registered;
gboolean dirty;
} stonith_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 {
/* 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 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);
// Fencer-specific client flags
enum st_client_flags {
st_callback_unknown = UINT64_C(0),
st_callback_notify_fence = (UINT64_C(1) << 0),
st_callback_device_add = (UINT64_C(1) << 2),
st_callback_device_del = (UINT64_C(1) << 4),
st_callback_notify_history = (UINT64_C(1) << 5),
st_callback_notify_history_synced = (UINT64_C(1) << 6)
};
// 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_MAX];
} stonith_topology_t;
void stonith_shutdown(int nsig);
void init_device_list(void);
void free_device_list(void);
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 stonith_device_register(xmlNode *msg, gboolean 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_device_notification(const char *op,
const pcmk__action_result_t *result,
const char *desc);
void fenced_send_level_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(crm_node_t *peer);
void set_fencing_completed(remote_fencing_op_t * op);
int fenced_handle_manual_confirmation(const pcmk__client_t *client,
xmlNode *msg);
void fencer_metadata(void);
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_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 st_device_supports_on if \p action is "on", otherwise
* st_device_supports_none
*/
static inline uint32_t
fenced_support_flag(const char *action)
{
if (pcmk__str_eq(action, PCMK_ACTION_ON, pcmk__str_none)) {
return st_device_supports_on;
}
return st_device_supports_none;
}
extern char *stonith_our_uname;
extern gboolean stand_alone;
extern GHashTable *device_list;
extern GHashTable *topology;
extern 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;