Page MenuHomeClusterLabs Projects

No OneTemporary

diff --git a/daemons/controld/controld_cib.c b/daemons/controld/controld_cib.c
index b07b66c1ff..9887314dd0 100644
--- a/daemons/controld/controld_cib.c
+++ b/daemons/controld/controld_cib.c
@@ -1,1068 +1,1072 @@
/*
* 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 General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <unistd.h> /* sleep */
#include <crm/common/alerts_internal.h>
#include <crm/common/xml.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/lrmd_internal.h>
#include <pacemaker-controld.h>
// Call ID of the most recent in-progress CIB resource update (or 0 if none)
static int pending_rsc_update = 0;
/*!
* \internal
* \brief Respond to a dropped CIB connection
*
* \param[in] user_data CIB connection that dropped
*/
static void
handle_cib_disconnect(gpointer user_data)
{
CRM_LOG_ASSERT(user_data == controld_globals.cib_conn);
controld_trigger_fsa();
controld_globals.cib_conn->state = cib_disconnected;
if (pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
// @TODO This should trigger a reconnect, not a shutdown
crm_crit("Lost connection to the CIB manager, shutting down");
register_fsa_input(C_FSA_INTERNAL, I_ERROR, NULL);
controld_clear_fsa_input_flags(R_CIB_CONNECTED);
} else { // Expected
crm_info("Disconnected from the CIB manager");
}
}
static void
do_cib_updated(const char *event, xmlNode * msg)
{
const xmlNode *patchset = NULL;
const char *client_name = NULL;
crm_debug("Received CIB diff notification: DC=%s", pcmk__btoa(AM_I_DC));
if (cib__get_notify_patchset(msg, &patchset) != pcmk_rc_ok) {
return;
}
if (cib__element_in_patchset(patchset, XML_CIB_TAG_ALERTS)
|| cib__element_in_patchset(patchset, XML_CIB_TAG_CRMCONFIG)) {
controld_trigger_config();
}
if (!AM_I_DC) {
// We're not in control of the join sequence
return;
}
client_name = crm_element_value(msg, F_CIB_CLIENTNAME);
if (!cib__client_triggers_refresh(client_name)) {
// The CIB is still accurate
return;
}
if (cib__element_in_patchset(patchset, XML_CIB_TAG_NODES)
|| cib__element_in_patchset(patchset, XML_CIB_TAG_STATUS)) {
/* An unsafe client modified the nodes or status section. Ensure the
* node list is up-to-date, and start the join process again so we get
* everyone's current resource history.
*/
if (client_name == NULL) {
client_name = crm_element_value(msg, F_CIB_CLIENTID);
}
crm_notice("Populating nodes and starting an election after %s event "
"triggered by %s",
event, pcmk__s(client_name, "(unidentified client)"));
populate_cib_nodes(node_update_quick|node_update_all, __func__);
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
}
}
void
controld_disconnect_cib_manager(void)
{
cib_t *cib_conn = controld_globals.cib_conn;
CRM_ASSERT(cib_conn != NULL);
crm_debug("Disconnecting from the CIB manager");
controld_clear_fsa_input_flags(R_CIB_CONNECTED);
cib_conn->cmds->del_notify_callback(cib_conn, T_CIB_DIFF_NOTIFY,
do_cib_updated);
cib_free_callbacks(cib_conn);
if (cib_conn->state != cib_disconnected) {
cib_conn->cmds->set_secondary(cib_conn,
cib_scope_local|cib_discard_reply);
cib_conn->cmds->signoff(cib_conn);
}
}
/* A_CIB_STOP, A_CIB_START, O_CIB_RESTART */
void
do_cib_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input, fsa_data_t * msg_data)
{
static int cib_retries = 0;
cib_t *cib_conn = controld_globals.cib_conn;
void (*dnotify_fn) (gpointer user_data) = handle_cib_disconnect;
void (*update_cb) (const char *event, xmlNodePtr msg) = do_cib_updated;
int rc = pcmk_ok;
CRM_ASSERT(cib_conn != NULL);
if (pcmk_is_set(action, A_CIB_STOP)) {
if ((cib_conn->state != cib_disconnected)
&& (pending_rsc_update != 0)) {
crm_info("Waiting for resource update %d to complete",
pending_rsc_update);
crmd_fsa_stall(FALSE);
return;
}
controld_disconnect_cib_manager();
}
if (!pcmk_is_set(action, A_CIB_START)) {
return;
}
if (cur_state == S_STOPPING) {
crm_err("Ignoring request to connect to the CIB manager after "
"shutdown");
return;
}
rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
cib_command_nonblocking);
if (rc != pcmk_ok) {
// A short wait that usually avoids stalling the FSA
sleep(1);
rc = cib_conn->cmds->signon(cib_conn, CRM_SYSTEM_CRMD,
cib_command_nonblocking);
}
if (rc != pcmk_ok) {
crm_info("Could not connect to the CIB manager: %s", pcmk_strerror(rc));
} else if (cib_conn->cmds->set_connection_dnotify(cib_conn,
dnotify_fn) != pcmk_ok) {
crm_err("Could not set dnotify callback");
} else if (cib_conn->cmds->add_notify_callback(cib_conn,
T_CIB_DIFF_NOTIFY,
update_cb) != pcmk_ok) {
crm_err("Could not set CIB notification callback (update)");
} else {
controld_set_fsa_input_flags(R_CIB_CONNECTED);
cib_retries = 0;
}
if (!pcmk_is_set(controld_globals.fsa_input_register, R_CIB_CONNECTED)) {
cib_retries++;
if (cib_retries < 30) {
crm_warn("Couldn't complete CIB registration %d times... "
"pause and retry", cib_retries);
controld_start_wait_timer();
crmd_fsa_stall(FALSE);
} else {
crm_err("Could not complete CIB registration %d times... "
"hard error", cib_retries);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
}
#define MIN_CIB_OP_TIMEOUT (30)
/*!
* \internal
* \brief Get the timeout (in seconds) that should be used with CIB operations
*
* \return The maximum of 30 seconds, the value of the PCMK_cib_timeout
* environment variable, or 10 seconds times one more than the number of
* nodes in the cluster.
*/
unsigned int
cib_op_timeout(void)
{
// @COMPAT: Drop env_timeout at 3.0.0
static int env_timeout = -1;
unsigned int calculated_timeout = 0;
if (env_timeout == -1) {
const char *env = pcmk__env_option(PCMK__ENV_CIB_TIMEOUT);
pcmk__scan_min_int(env, &env_timeout, MIN_CIB_OP_TIMEOUT);
crm_trace("Minimum CIB op timeout: %ds (environment: %s)",
env_timeout, (env? env : "none"));
}
calculated_timeout = 1 + crm_active_peers();
if (crm_remote_peer_cache) {
calculated_timeout += g_hash_table_size(crm_remote_peer_cache);
}
calculated_timeout *= 10;
calculated_timeout = QB_MAX(calculated_timeout, env_timeout);
crm_trace("Calculated timeout: %us", calculated_timeout);
if (controld_globals.cib_conn) {
controld_globals.cib_conn->call_timeout = calculated_timeout;
}
return calculated_timeout;
}
/*!
* \internal
* \brief Get CIB call options to use local scope if primary is unavailable
*
* \return CIB call options
*/
int
crmd_cib_smart_opt(void)
{
int call_opt = cib_none;
if ((controld_globals.fsa_state == S_ELECTION)
|| (controld_globals.fsa_state == S_PENDING)) {
crm_info("Sending update to local CIB in state: %s",
fsa_state2string(controld_globals.fsa_state));
cib__set_call_options(call_opt, "update", cib_scope_local);
}
return call_opt;
}
static void
cib_delete_callback(xmlNode *msg, int call_id, int rc, xmlNode *output,
void *user_data)
{
char *desc = user_data;
if (rc == 0) {
crm_debug("Deletion of %s (via CIB call %d) succeeded", desc, call_id);
} else {
crm_warn("Deletion of %s (via CIB call %d) failed: %s " CRM_XS " rc=%d",
desc, call_id, pcmk_strerror(rc), rc);
}
}
// Searches for various portions of node_state to delete
// Match a particular node's node_state (takes node name 1x)
#define XPATH_NODE_STATE "//" XML_CIB_TAG_STATE \
"[@" PCMK_XA_UNAME "='%s']"
// Node's lrm section (name 1x)
#define XPATH_NODE_LRM XPATH_NODE_STATE "/" XML_CIB_TAG_LRM
/* Node's lrm_rsc_op entries and lrm_resource entries without unexpired lock
* (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x)
*/
#define XPATH_NODE_LRM_UNLOCKED XPATH_NODE_STATE "//" XML_LRM_TAG_RSC_OP \
"|" XPATH_NODE_STATE \
"//" XML_LRM_TAG_RESOURCE \
"[not(@" PCMK_OPT_SHUTDOWN_LOCK ") " \
"or " PCMK_OPT_SHUTDOWN_LOCK "<%lld]"
// Node's transient_attributes section (name 1x)
#define XPATH_NODE_ATTRS XPATH_NODE_STATE "/" XML_TAG_TRANSIENT_NODEATTRS
// Everything under node_state (name 1x)
#define XPATH_NODE_ALL XPATH_NODE_STATE "/*"
/* Unlocked history + transient attributes
* (name 2x, (seconds_since_epoch - PCMK_OPT_SHUTDOWN_LOCK_LIMIT) 1x, name 1x)
*/
#define XPATH_NODE_ALL_UNLOCKED XPATH_NODE_LRM_UNLOCKED "|" XPATH_NODE_ATTRS
/*!
* \internal
* \brief Get the XPath and description of a node state section to be deleted
*
* \param[in] uname Desired node
* \param[in] section Subsection of node_state to be deleted
* \param[out] xpath Where to store XPath of \p section
* \param[out] desc If not \c NULL, where to store description of \p section
*/
void
controld_node_state_deletion_strings(const char *uname,
enum controld_section_e section,
char **xpath, char **desc)
{
const char *desc_pre = NULL;
// Shutdown locks that started before this time are expired
long long expire = (long long) time(NULL)
- controld_globals.shutdown_lock_limit;
switch (section) {
case controld_section_lrm:
*xpath = crm_strdup_printf(XPATH_NODE_LRM, uname);
desc_pre = "resource history";
break;
case controld_section_lrm_unlocked:
*xpath = crm_strdup_printf(XPATH_NODE_LRM_UNLOCKED,
uname, uname, expire);
desc_pre = "resource history (other than shutdown locks)";
break;
case controld_section_attrs:
*xpath = crm_strdup_printf(XPATH_NODE_ATTRS, uname);
desc_pre = "transient attributes";
break;
case controld_section_all:
*xpath = crm_strdup_printf(XPATH_NODE_ALL, uname);
desc_pre = "all state";
break;
case controld_section_all_unlocked:
*xpath = crm_strdup_printf(XPATH_NODE_ALL_UNLOCKED,
uname, uname, expire, uname);
desc_pre = "all state (other than shutdown locks)";
break;
default:
// We called this function incorrectly
CRM_ASSERT(false);
break;
}
if (desc != NULL) {
*desc = crm_strdup_printf("%s for node %s", desc_pre, uname);
}
}
/*!
* \internal
* \brief Delete subsection of a node's CIB node_state
*
* \param[in] uname Desired node
* \param[in] section Subsection of node_state to delete
* \param[in] options CIB call options to use
*/
void
controld_delete_node_state(const char *uname, enum controld_section_e section,
int options)
{
cib_t *cib = controld_globals.cib_conn;
char *xpath = NULL;
char *desc = NULL;
int cib_rc = pcmk_ok;
CRM_ASSERT((uname != NULL) && (cib != NULL));
controld_node_state_deletion_strings(uname, section, &xpath, &desc);
cib__set_call_options(options, "node state deletion",
cib_xpath|cib_multiple);
cib_rc = cib->cmds->remove(cib, xpath, NULL, options);
fsa_register_cib_callback(cib_rc, desc, cib_delete_callback);
crm_info("Deleting %s (via CIB call %d) " CRM_XS " xpath=%s",
desc, cib_rc, xpath);
// CIB library handles freeing desc
free(xpath);
}
// Takes node name and resource ID
#define XPATH_RESOURCE_HISTORY "//" XML_CIB_TAG_STATE \
"[@" PCMK_XA_UNAME "='%s']/" \
XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \
"/" XML_LRM_TAG_RESOURCE \
"[@" PCMK_XA_ID "='%s']"
// @TODO could add "and @PCMK_OPT_SHUTDOWN_LOCK" to limit to locks
/*!
* \internal
* \brief Clear resource history from CIB for a given resource and node
*
* \param[in] rsc_id ID of resource to be cleared
* \param[in] node Node whose resource history should be cleared
* \param[in] user_name ACL user name to use
* \param[in] call_options CIB call options
*
* \return Standard Pacemaker return code
*/
int
controld_delete_resource_history(const char *rsc_id, const char *node,
const char *user_name, int call_options)
{
char *desc = NULL;
char *xpath = NULL;
int rc = pcmk_rc_ok;
cib_t *cib = controld_globals.cib_conn;
CRM_CHECK((rsc_id != NULL) && (node != NULL), return EINVAL);
desc = crm_strdup_printf("resource history for %s on %s", rsc_id, node);
if (cib == NULL) {
crm_err("Unable to clear %s: no CIB connection", desc);
free(desc);
return ENOTCONN;
}
// Ask CIB to delete the entry
xpath = crm_strdup_printf(XPATH_RESOURCE_HISTORY, node, rsc_id);
cib->cmds->set_user(cib, user_name);
rc = cib->cmds->remove(cib, xpath, NULL, call_options|cib_xpath);
cib->cmds->set_user(cib, NULL);
if (rc < 0) {
rc = pcmk_legacy2rc(rc);
crm_err("Could not delete resource status of %s on %s%s%s: %s "
CRM_XS " rc=%d", rsc_id, node,
(user_name? " for user " : ""), (user_name? user_name : ""),
pcmk_rc_str(rc), rc);
free(desc);
free(xpath);
return rc;
}
if (pcmk_is_set(call_options, cib_sync_call)) {
if (pcmk_is_set(call_options, cib_dryrun)) {
crm_debug("Deletion of %s would succeed", desc);
} else {
crm_debug("Deletion of %s succeeded", desc);
}
free(desc);
} else {
crm_info("Clearing %s (via CIB call %d) " CRM_XS " xpath=%s",
desc, rc, xpath);
fsa_register_cib_callback(rc, desc, cib_delete_callback);
// CIB library handles freeing desc
}
free(xpath);
return pcmk_rc_ok;
}
/*!
* \internal
* \brief Build XML and string of parameters meeting some criteria, for digest
*
* \param[in] op Executor event with parameter table to use
* \param[in] metadata Parsed meta-data for executed resource agent
* \param[in] param_type Flag used for selection criteria
* \param[out] result Will be set to newly created XML with selected
* parameters as attributes
*
* \return Newly allocated space-separated string of parameter names
* \note Selection criteria varies by param_type: for the restart digest, we
* want parameters that are *not* marked reloadable (OCF 1.1) or that
* *are* marked unique (pre-1.1), for both string and XML results; for the
* secure digest, we want parameters that *are* marked private for the
* string, but parameters that are *not* marked private for the XML.
* \note It is the caller's responsibility to free the string return value with
* \p g_string_free() and the XML result with \p free_xml().
*/
static GString *
build_parameter_list(const lrmd_event_data_t *op,
const struct ra_metadata_s *metadata,
enum ra_param_flags_e param_type, xmlNode **result)
{
GString *list = NULL;
*result = create_xml_node(NULL, XML_TAG_PARAMS);
/* Consider all parameters only except private ones to be consistent with
* what scheduler does with calculate_secure_digest().
*/
if (param_type == ra_param_private
&& compare_version(controld_globals.dc_version, "3.16.0") >= 0) {
g_hash_table_foreach(op->params, hash2field, *result);
pcmk__filter_op_for_digest(*result);
}
for (GList *iter = metadata->ra_params; iter != NULL; iter = iter->next) {
struct ra_param_s *param = (struct ra_param_s *) iter->data;
bool accept_for_list = false;
bool accept_for_xml = false;
switch (param_type) {
case ra_param_reloadable:
accept_for_list = !pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = accept_for_list;
break;
case ra_param_unique:
accept_for_list = pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = accept_for_list;
break;
case ra_param_private:
accept_for_list = pcmk_is_set(param->rap_flags, param_type);
accept_for_xml = !accept_for_list;
break;
}
if (accept_for_list) {
crm_trace("Attr %s is %s", param->rap_name, ra_param_flag2text(param_type));
if (list == NULL) {
// We will later search for " WORD ", so start list with a space
pcmk__add_word(&list, 256, " ");
}
pcmk__add_word(&list, 0, param->rap_name);
} else {
crm_trace("Rejecting %s for %s", param->rap_name, ra_param_flag2text(param_type));
}
if (accept_for_xml) {
const char *v = g_hash_table_lookup(op->params, param->rap_name);
if (v != NULL) {
crm_trace("Adding attr %s=%s to the xml result", param->rap_name, v);
crm_xml_add(*result, param->rap_name, v);
}
} else {
crm_trace("Removing attr %s from the xml result", param->rap_name);
xml_remove_prop(*result, param->rap_name);
}
}
if (list != NULL) {
// We will later search for " WORD ", so end list with a space
pcmk__add_word(&list, 0, " ");
}
return list;
}
static void
append_restart_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
GString *list = NULL;
char *digest = NULL;
xmlNode *restart = NULL;
CRM_LOG_ASSERT(op->params != NULL);
if (op->interval_ms > 0) {
/* monitors are not reloadable */
return;
}
if (pcmk_is_set(metadata->ra_flags, ra_supports_reload_agent)) {
- // Add parameters not marked reloadable to the "op-force-restart" list
+ /* Add parameters not marked reloadable to the PCMK__XA_OP_FORCE_RESTART
+ * list
+ */
list = build_parameter_list(op, metadata, ra_param_reloadable,
&restart);
} else if (pcmk_is_set(metadata->ra_flags, ra_supports_legacy_reload)) {
/* @COMPAT pre-OCF-1.1 resource agents
*
* Before OCF 1.1, Pacemaker abused "unique=0" to indicate
* reloadability. Add any parameters with unique="1" to the
- * "op-force-restart" list.
+ * PCMK__XA_OP_FORCE_RESTART list.
*/
list = build_parameter_list(op, metadata, ra_param_unique, &restart);
} else {
// Resource does not support agent reloads
return;
}
digest = calculate_operation_digest(restart, version);
- /* Add "op-force-restart" and "op-restart-digest" to indicate the resource supports reload,
- * no matter if it actually supports any parameters with unique="1"). */
- crm_xml_add(update, XML_LRM_ATTR_OP_RESTART,
+ /* Add PCMK__XA_OP_FORCE_RESTART and "op-restart-digest" to indicate the
+ * resource supports reload, no matter if it actually supports any
+ * reloadable parameters
+ */
+ crm_xml_add(update, PCMK__XA_OP_FORCE_RESTART,
(list == NULL)? "" : (const char *) list->str);
crm_xml_add(update, XML_LRM_ATTR_RESTART_DIGEST, digest);
if ((list != NULL) && (list->len > 0)) {
crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
} else {
crm_trace("%s: %s", op->rsc_id, digest);
}
if (list != NULL) {
g_string_free(list, TRUE);
}
free_xml(restart);
free(digest);
}
static void
append_secure_list(lrmd_event_data_t *op, struct ra_metadata_s *metadata,
xmlNode *update, const char *version)
{
GString *list = NULL;
char *digest = NULL;
xmlNode *secure = NULL;
CRM_LOG_ASSERT(op->params != NULL);
/*
* To keep XML_LRM_ATTR_OP_SECURE short, we want it to contain the
* secure parameters but XML_LRM_ATTR_SECURE_DIGEST to be based on
* the insecure ones
*/
list = build_parameter_list(op, metadata, ra_param_private, &secure);
if (list != NULL) {
digest = calculate_operation_digest(secure, version);
crm_xml_add(update, XML_LRM_ATTR_OP_SECURE, (const char *) list->str);
crm_xml_add(update, XML_LRM_ATTR_SECURE_DIGEST, digest);
crm_trace("%s: %s, %s", op->rsc_id, digest, (const char *) list->str);
g_string_free(list, TRUE);
} else {
crm_trace("%s: no secure parameters", op->rsc_id);
}
free_xml(secure);
free(digest);
}
/*!
* \internal
* \brief Create XML for a resource history entry
*
* \param[in] func Function name of caller
* \param[in,out] parent XML to add entry to
* \param[in] rsc Affected resource
* \param[in,out] op Action to add an entry for (or NULL to do nothing)
* \param[in] node_name Node where action occurred
*/
void
controld_add_resource_history_xml_as(const char *func, xmlNode *parent,
const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op,
const char *node_name)
{
int target_rc = 0;
xmlNode *xml_op = NULL;
struct ra_metadata_s *metadata = NULL;
const char *caller_version = NULL;
lrm_state_t *lrm_state = NULL;
if (op == NULL) {
return;
}
target_rc = rsc_op_expected_rc(op);
caller_version = g_hash_table_lookup(op->params, PCMK_XA_CRM_FEATURE_SET);
CRM_CHECK(caller_version != NULL, caller_version = CRM_FEATURE_SET);
xml_op = pcmk__create_history_xml(parent, op, caller_version, target_rc,
controld_globals.our_nodename, func);
if (xml_op == NULL) {
return;
}
if ((rsc == NULL) || (op->params == NULL)
|| !crm_op_needs_metadata(rsc->standard, op->op_type)) {
crm_trace("No digests needed for %s action on %s (params=%p rsc=%p)",
op->op_type, op->rsc_id, op->params, rsc);
return;
}
lrm_state = lrm_state_find(node_name);
if (lrm_state == NULL) {
crm_warn("Cannot calculate digests for operation " PCMK__OP_FMT
" because we have no connection to executor for %s",
op->rsc_id, op->op_type, op->interval_ms, node_name);
return;
}
/* Ideally the metadata is cached, and the agent is just a fallback.
*
* @TODO Go through all callers and ensure they get metadata asynchronously
* first.
*/
metadata = controld_get_rsc_metadata(lrm_state, rsc,
controld_metadata_from_agent
|controld_metadata_from_cache);
if (metadata == NULL) {
return;
}
crm_trace("Including additional digests for %s:%s:%s",
rsc->standard, rsc->provider, rsc->type);
append_restart_list(op, metadata, xml_op, caller_version);
append_secure_list(op, metadata, xml_op, caller_version);
return;
}
/*!
* \internal
* \brief Record an action as pending in the CIB, if appropriate
*
* \param[in] node_name Node where the action is pending
* \param[in] rsc Resource that action is for
* \param[in,out] op Pending action
*
* \return true if action was recorded in CIB, otherwise false
*/
bool
controld_record_pending_op(const char *node_name, const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op)
{
const char *record_pending = NULL;
CRM_CHECK((node_name != NULL) && (rsc != NULL) && (op != NULL),
return false);
// Never record certain operation types as pending
if ((op->op_type == NULL) || (op->params == NULL)
|| !controld_action_is_recordable(op->op_type)) {
return false;
}
// Check action's PCMK_META_RECORD_PENDING meta-attribute (defaults to true)
record_pending = crm_meta_value(op->params, PCMK_META_RECORD_PENDING);
if ((record_pending != NULL) && !crm_is_true(record_pending)) {
return false;
}
op->call_id = -1;
op->t_run = time(NULL);
op->t_rcchange = op->t_run;
lrmd__set_result(op, PCMK_OCF_UNKNOWN, PCMK_EXEC_PENDING, NULL);
crm_debug("Recording pending %s-interval %s for %s on %s in the CIB",
pcmk__readable_interval(op->interval_ms), op->op_type, op->rsc_id,
node_name);
controld_update_resource_history(node_name, rsc, op, 0);
return true;
}
static void
cib_rsc_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
switch (rc) {
case pcmk_ok:
case -pcmk_err_diff_failed:
case -pcmk_err_diff_resync:
crm_trace("Resource history update completed (call=%d rc=%d)",
call_id, rc);
break;
default:
if (call_id > 0) {
crm_warn("Resource history update %d failed: %s "
CRM_XS " rc=%d", call_id, pcmk_strerror(rc), rc);
} else {
crm_warn("Resource history update failed: %s " CRM_XS " rc=%d",
pcmk_strerror(rc), rc);
}
}
if (call_id == pending_rsc_update) {
pending_rsc_update = 0;
controld_trigger_fsa();
}
}
/* Only successful stops, and probes that found the resource inactive, get locks
* recorded in the history. This ensures the resource stays locked to the node
* until it is active there again after the node comes back up.
*/
static bool
should_preserve_lock(lrmd_event_data_t *op)
{
if (!pcmk_is_set(controld_globals.flags, controld_shutdown_lock_enabled)) {
return false;
}
if (!strcmp(op->op_type, PCMK_ACTION_STOP) && (op->rc == PCMK_OCF_OK)) {
return true;
}
if (!strcmp(op->op_type, PCMK_ACTION_MONITOR)
&& (op->rc == PCMK_OCF_NOT_RUNNING)) {
return true;
}
return false;
}
/*!
* \internal
* \brief Request a CIB update
*
* \param[in] section Section of CIB to update
* \param[in] data New XML of CIB section to update
* \param[in] options CIB call options
* \param[in] callback If not \c NULL, set this as the operation callback
*
* \return Standard Pacemaker return code
*
* \note If \p callback is \p cib_rsc_callback(), the CIB update's call ID is
* stored in \p pending_rsc_update on success.
*/
int
controld_update_cib(const char *section, xmlNode *data, int options,
void (*callback)(xmlNode *, int, int, xmlNode *, void *))
{
cib_t *cib = controld_globals.cib_conn;
int cib_rc = -ENOTCONN;
CRM_ASSERT(data != NULL);
if (cib != NULL) {
cib_rc = cib->cmds->modify(cib, section, data, options);
if (cib_rc >= 0) {
crm_debug("Submitted CIB update %d for %s section",
cib_rc, section);
}
}
if (callback == NULL) {
if (cib_rc < 0) {
crm_err("Failed to update CIB %s section: %s",
section, pcmk_rc_str(pcmk_legacy2rc(cib_rc)));
}
} else {
if ((cib_rc >= 0) && (callback == cib_rsc_callback)) {
/* Checking for a particular callback is a little hacky, but it
* didn't seem worth adding an output argument for cib_rc for just
* one use case.
*/
pending_rsc_update = cib_rc;
}
fsa_register_cib_callback(cib_rc, NULL, callback);
}
return (cib_rc >= 0)? pcmk_rc_ok : pcmk_legacy2rc(cib_rc);
}
/*!
* \internal
* \brief Update resource history entry in CIB
*
* \param[in] node_name Node where action occurred
* \param[in] rsc Resource that action is for
* \param[in,out] op Action to record
* \param[in] lock_time If nonzero, when resource was locked to node
*
* \note On success, the CIB update's call ID will be stored in
* pending_rsc_update.
*/
void
controld_update_resource_history(const char *node_name,
const lrmd_rsc_info_t *rsc,
lrmd_event_data_t *op, time_t lock_time)
{
xmlNode *update = NULL;
xmlNode *xml = NULL;
int call_opt = crmd_cib_smart_opt();
const char *node_id = NULL;
const char *container = NULL;
CRM_CHECK((node_name != NULL) && (op != NULL), return);
if (rsc == NULL) {
crm_warn("Resource %s no longer exists in the executor", op->rsc_id);
controld_ack_event_directly(NULL, NULL, rsc, op, op->rsc_id);
return;
}
// <status>
update = create_xml_node(NULL, XML_CIB_TAG_STATUS);
// <node_state ...>
xml = create_xml_node(update, XML_CIB_TAG_STATE);
if (pcmk__str_eq(node_name, controld_globals.our_nodename,
pcmk__str_casei)) {
node_id = controld_globals.our_uuid;
} else {
node_id = node_name;
pcmk__xe_set_bool_attr(xml, XML_NODE_IS_REMOTE, true);
}
crm_xml_add(xml, PCMK_XA_ID, node_id);
crm_xml_add(xml, PCMK_XA_UNAME, node_name);
crm_xml_add(xml, PCMK_XA_CRM_DEBUG_ORIGIN, __func__);
// <lrm ...>
xml = create_xml_node(xml, XML_CIB_TAG_LRM);
crm_xml_add(xml, PCMK_XA_ID, node_id);
// <lrm_resources>
xml = create_xml_node(xml, XML_LRM_TAG_RESOURCES);
// <lrm_resource ...>
xml = create_xml_node(xml, XML_LRM_TAG_RESOURCE);
crm_xml_add(xml, PCMK_XA_ID, op->rsc_id);
crm_xml_add(xml, PCMK_XA_CLASS, rsc->standard);
crm_xml_add(xml, PCMK_XA_PROVIDER, rsc->provider);
crm_xml_add(xml, PCMK_XA_TYPE, rsc->type);
if (lock_time != 0) {
/* Actions on a locked resource should either preserve the lock by
* recording it with the action result, or clear it.
*/
if (!should_preserve_lock(op)) {
lock_time = 0;
}
crm_xml_add_ll(xml, PCMK_OPT_SHUTDOWN_LOCK, (long long) lock_time);
}
if (op->params != NULL) {
container = g_hash_table_lookup(op->params,
CRM_META "_" PCMK__META_CONTAINER);
if (container != NULL) {
crm_trace("Resource %s is a part of container resource %s",
op->rsc_id, container);
crm_xml_add(xml, PCMK__META_CONTAINER, container);
}
}
// <lrm_resource_op ...> (possibly more than one)
controld_add_resource_history_xml(xml, rsc, op, node_name);
/* Update CIB asynchronously. Even if it fails, the resource state should be
* discovered during the next election. Worst case, the node is wrongly
* fenced for running a resource it isn't.
*/
crm_log_xml_trace(update, __func__);
controld_update_cib(XML_CIB_TAG_STATUS, update, call_opt, cib_rsc_callback);
free_xml(update);
}
/*!
* \internal
* \brief Erase an LRM history entry from the CIB, given the operation data
*
* \param[in] op Operation whose history should be deleted
*/
void
controld_delete_action_history(const lrmd_event_data_t *op)
{
xmlNode *xml_top = NULL;
CRM_CHECK(op != NULL, return);
xml_top = create_xml_node(NULL, XML_LRM_TAG_RSC_OP);
crm_xml_add_int(xml_top, PCMK__XA_CALL_ID, op->call_id);
crm_xml_add(xml_top, PCMK__XA_TRANSITION_KEY, op->user_data);
if (op->interval_ms > 0) {
char *op_id = pcmk__op_key(op->rsc_id, op->op_type, op->interval_ms);
/* Avoid deleting last_failure too (if it was a result of this recurring op failing) */
crm_xml_add(xml_top, PCMK_XA_ID, op_id);
free(op_id);
}
crm_debug("Erasing resource operation history for " PCMK__OP_FMT " (call=%d)",
op->rsc_id, op->op_type, op->interval_ms, op->call_id);
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn,
XML_CIB_TAG_STATUS, xml_top,
cib_none);
crm_log_xml_trace(xml_top, "op:cancel");
free_xml(xml_top);
}
/* Define xpath to find LRM resource history entry by node and resource */
#define XPATH_HISTORY \
"/" XML_TAG_CIB "/" XML_CIB_TAG_STATUS \
"/" XML_CIB_TAG_STATE "[@" PCMK_XA_UNAME "='%s']" \
"/" XML_CIB_TAG_LRM "/" XML_LRM_TAG_RESOURCES \
"/" XML_LRM_TAG_RESOURCE "[@" PCMK_XA_ID "='%s']" \
"/" XML_LRM_TAG_RSC_OP
/* ... and also by operation key */
#define XPATH_HISTORY_ID XPATH_HISTORY "[@" PCMK_XA_ID "='%s']"
/* ... and also by operation key and operation call ID */
#define XPATH_HISTORY_CALL XPATH_HISTORY \
"[@" PCMK_XA_ID "='%s' and @" PCMK__XA_CALL_ID "='%d']"
/* ... and also by operation key and original operation key */
#define XPATH_HISTORY_ORIG XPATH_HISTORY \
"[@" PCMK_XA_ID "='%s' and @" PCMK__XA_OPERATION_KEY "='%s']"
/*!
* \internal
* \brief Delete a last_failure resource history entry from the CIB
*
* \param[in] rsc_id Name of resource to clear history for
* \param[in] node Name of node to clear history for
* \param[in] action If specified, delete only if this was failed action
* \param[in] interval_ms If \p action is specified, it has this interval
*/
void
controld_cib_delete_last_failure(const char *rsc_id, const char *node,
const char *action, guint interval_ms)
{
char *xpath = NULL;
char *last_failure_key = NULL;
CRM_CHECK((rsc_id != NULL) && (node != NULL), return);
// Generate XPath to match desired entry
last_failure_key = pcmk__op_key(rsc_id, "last_failure", 0);
if (action == NULL) {
xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id,
last_failure_key);
} else {
char *action_key = pcmk__op_key(rsc_id, action, interval_ms);
xpath = crm_strdup_printf(XPATH_HISTORY_ORIG, node, rsc_id,
last_failure_key, action_key);
free(action_key);
}
free(last_failure_key);
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
NULL, cib_xpath);
free(xpath);
}
/*!
* \internal
* \brief Delete resource history entry from the CIB, given operation key
*
* \param[in] rsc_id Name of resource to clear history for
* \param[in] node Name of node to clear history for
* \param[in] key Operation key of operation to clear history for
* \param[in] call_id If specified, delete entry only if it has this call ID
*/
void
controld_delete_action_history_by_key(const char *rsc_id, const char *node,
const char *key, int call_id)
{
char *xpath = NULL;
CRM_CHECK((rsc_id != NULL) && (node != NULL) && (key != NULL), return);
if (call_id > 0) {
xpath = crm_strdup_printf(XPATH_HISTORY_CALL, node, rsc_id, key,
call_id);
} else {
xpath = crm_strdup_printf(XPATH_HISTORY_ID, node, rsc_id, key);
}
controld_globals.cib_conn->cmds->remove(controld_globals.cib_conn, xpath,
NULL, cib_xpath);
free(xpath);
}
diff --git a/include/crm_internal.h b/include/crm_internal.h
index 3960652613..c14fe2195b 100644
--- a/include/crm_internal.h
+++ b/include/crm_internal.h
@@ -1,197 +1,198 @@
/*
* Copyright 2006-2024 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 CRM_INTERNAL__H
# define CRM_INTERNAL__H
# ifndef PCMK__CONFIG_H
# define PCMK__CONFIG_H
# include <config.h>
# endif
# include <portability.h>
/* Our minimum glib dependency is 2.42. Define that as both the minimum and
* maximum glib APIs that are allowed (i.e. APIs that were already deprecated
* in 2.42, and APIs introduced after 2.42, cannot be used by Pacemaker code).
*/
#define GLIB_VERSION_MIN_REQUIRED GLIB_VERSION_2_42
#define GLIB_VERSION_MAX_ALLOWED GLIB_VERSION_2_42
# include <glib.h>
# include <stdbool.h>
# include <libxml/tree.h>
/* Public API headers can guard including deprecated API headers with this
* symbol, thus preventing internal code (which includes this header) from using
* deprecated APIs, while still allowing external code to use them by default.
*/
#define PCMK_ALLOW_DEPRECATED 0
# include <crm/lrmd.h>
# include <crm/common/logging.h>
# include <crm/common/logging_internal.h>
# include <crm/common/ipc_internal.h>
# include <crm/common/options_internal.h>
# include <crm/common/output_internal.h>
# include <crm/common/xml_internal.h>
# include <crm/common/internal.h>
# include <locale.h>
# include <gettext.h>
#define N_(String) (String)
#ifdef ENABLE_NLS
# define _(String) gettext(String)
#else
# define _(String) (String)
#endif
/*
* XML element names used only by internal code
*/
// @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 2.0.0; alias for <clone> with PCMK_META_PROMOTABLE
* set to "true"
*/
#define PCMK__XE_PROMOTABLE_LEGACY "master"
/*
* XML attribute names used only by internal code
*/
#define PCMK__XA_ATTR_DAMPENING "attr_dampening"
#define PCMK__XA_ATTR_FORCE "attrd_is_force_write"
#define PCMK__XA_ATTR_INTERVAL "attr_clear_interval"
#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_NODE_ID "attr_host_id"
#define PCMK__XA_ATTR_NODE_NAME "attr_host"
#define PCMK__XA_ATTR_OPERATION "attr_clear_operation"
#define PCMK__XA_ATTR_PATTERN "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_UUID "attr_key"
#define PCMK__XA_ATTR_VALUE "attr_value"
#define PCMK__XA_ATTR_VERSION "attr_version"
#define PCMK__XA_ATTR_WRITER "attr_writer"
#define PCMK__XA_CALL_ID "call-id"
#define PCMK__XA_CONFIG_ERRORS "config-errors"
#define PCMK__XA_CONFIG_WARNINGS "config-warnings"
#define PCMK__XA_CONFIRM "confirm"
#define PCMK__XA_CONN_HOST "connection_host"
#define PCMK__XA_CRMD "crmd"
#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_EXPECTED "expected"
#define PCMK__XA_FILE "file"
#define PCMK__XA_GRAPH_ERRORS "graph-errors"
#define PCMK__XA_GRAPH_WARNINGS "graph-warnings"
#define PCMK__XA_IN_CCM "in_ccm"
#define PCMK__XA_JOIN "join"
#define PCMK__XA_JOIN_ID "join_id"
#define PCMK__XA_LONG_ID "long-id"
#define PCMK__XA_MODE "mode"
#define PCMK__XA_NODE_START_STATE "node_start_state"
#define PCMK__XA_OBJECT_TYPE "object_type"
#define PCMK__XA_OPERATION_KEY "operation_key"
#define PCMK__XA_OP_DIGEST "op-digest"
+#define PCMK__XA_OP_FORCE_RESTART "op-force-restart"
#define PCMK__XA_OP_STATUS "op-status"
#define PCMK__XA_PACEMAKERD_STATE "pacemakerd_state"
#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_SCHEMA "schema"
#define PCMK__XA_SCHEMAS "schemas"
#define PCMK__XA_SRC "src"
#define PCMK__XA_SUBT "subt" // subtype
#define PCMK__XA_T "t" // type
#define PCMK__XA_TASK "task"
#define PCMK__XA_TRANSITION_KEY "transition-key"
#define PCMK__XA_TRANSITION_MAGIC "transition-magic"
#define PCMK__XA_UPTIME "uptime"
// @COMPAT Deprecated since 2.1.5
#define PCMK__XA_FIRST_INSTANCE "first-instance"
// @COMPAT Deprecated since 2.1.6
#define PCMK__XA_REPLACE "replace"
// @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"
/*
* IPC service names that are only used internally
*/
# define PCMK__SERVER_BASED_RO "cib_ro"
# define PCMK__SERVER_BASED_RW "cib_rw"
# define PCMK__SERVER_BASED_SHM "cib_shm"
/*
* IPC commands that can be sent to Pacemaker daemons
*/
#define PCMK__ATTRD_CMD_PEER_REMOVE "peer-remove"
#define PCMK__ATTRD_CMD_UPDATE "update"
#define PCMK__ATTRD_CMD_UPDATE_BOTH "update-both"
#define PCMK__ATTRD_CMD_UPDATE_DELAY "update-delay"
#define PCMK__ATTRD_CMD_QUERY "query"
#define PCMK__ATTRD_CMD_REFRESH "refresh"
#define PCMK__ATTRD_CMD_FLUSH "flush"
#define PCMK__ATTRD_CMD_SYNC "sync"
#define PCMK__ATTRD_CMD_SYNC_RESPONSE "sync-response"
#define PCMK__ATTRD_CMD_CLEAR_FAILURE "clear-failure"
#define PCMK__ATTRD_CMD_CONFIRM "confirm"
#define PCMK__CONTROLD_CMD_NODES "list-nodes"
#endif /* CRM_INTERNAL__H */
diff --git a/lib/pengine/pe_digest.c b/lib/pengine/pe_digest.c
index e53136b829..a702cc7434 100644
--- a/lib/pengine/pe_digest.c
+++ b/lib/pengine/pe_digest.c
@@ -1,609 +1,609 @@
/*
* 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.
*/
#include <crm_internal.h>
#include <glib.h>
#include <stdbool.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/xml.h>
#include <crm/common/xml_internal.h>
#include <crm/pengine/internal.h>
#include "pe_status_private.h"
extern bool pcmk__is_daemon;
/*!
* \internal
* \brief Free an operation digest cache entry
*
* \param[in,out] ptr Pointer to cache entry to free
*
* \note The argument is a gpointer so this can be used as a hash table
* free function.
*/
void
pe__free_digests(gpointer ptr)
{
pcmk__op_digest_t *data = ptr;
if (data != NULL) {
free_xml(data->params_all);
free_xml(data->params_secure);
free_xml(data->params_restart);
free(data->digest_all_calc);
free(data->digest_restart_calc);
free(data->digest_secure_calc);
free(data);
}
}
// Return true if XML attribute name is not substring of a given string
static bool
attr_not_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) == NULL) {
crm_trace("Filtering %s (not found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
// Return true if XML attribute name is substring of a given string
static bool
attr_in_string(xmlAttrPtr a, void *user_data)
{
bool filter = false;
char *name = crm_strdup_printf(" %s ", (const char *) a->name);
if (strstr((const char *) user_data, name) != NULL) {
crm_trace("Filtering %s (found in '%s')",
(const char *) a->name, (const char *) user_data);
filter = true;
}
free(name);
return filter;
}
/*!
* \internal
* \brief Add digest of all parameters to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in,out] rsc Resource that action was for
* \param[in] node Node action was performed on
* \param[in] params Resource parameters evaluated for node
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] xml_op Unused
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value table to override resource parameters
* \param[in,out] scheduler Scheduler data
*/
static void
calculate_main_digest(pcmk__op_digest_t *data, pcmk_resource_t *rsc,
const pcmk_node_t *node, GHashTable *params,
const char *task, guint *interval_ms,
const xmlNode *xml_op, const char *op_version,
GHashTable *overrides, pcmk_scheduler_t *scheduler)
{
xmlNode *action_config = NULL;
data->params_all = create_xml_node(NULL, XML_TAG_PARAMS);
/* REMOTE_CONTAINER_HACK: Allow Pacemaker Remote nodes to run containers
* that themselves are Pacemaker Remote nodes
*/
(void) pe__add_bundle_remote_name(rsc, scheduler, data->params_all,
PCMK_REMOTE_RA_ADDR);
if (overrides != NULL) {
// If interval was overridden, reset it
const char *meta_name = CRM_META "_" PCMK_META_INTERVAL;
const char *interval_s = g_hash_table_lookup(overrides, meta_name);
if (interval_s != NULL) {
long long value_ll;
if ((pcmk__scan_ll(interval_s, &value_ll, 0LL) == pcmk_rc_ok)
&& (value_ll >= 0) && (value_ll <= G_MAXUINT)) {
*interval_ms = (guint) value_ll;
}
}
// Add overrides to list of all parameters
g_hash_table_foreach(overrides, hash2field, data->params_all);
}
// Add provided instance parameters
g_hash_table_foreach(params, hash2field, data->params_all);
// Find action configuration XML in CIB
action_config = pcmk__find_action_config(rsc, task, *interval_ms, true);
/* Add action-specific resource instance attributes to the digest list.
*
* If this is a one-time action with action-specific instance attributes,
* enforce a restart instead of reload-agent in case the main digest doesn't
* match, even if the restart digest does. This ensures any changes of the
* action-specific parameters get applied for this specific action, and
* digests calculated for the resulting history will be correct. Default the
* result to RSC_DIGEST_RESTART for the case where the main digest doesn't
* match.
*/
params = pcmk__unpack_action_rsc_params(action_config, node->details->attrs,
scheduler);
if ((*interval_ms == 0) && (g_hash_table_size(params) > 0)) {
data->rc = pcmk__digest_restart;
}
g_hash_table_foreach(params, hash2field, data->params_all);
g_hash_table_destroy(params);
// Add action meta-attributes
params = pcmk__unpack_action_meta(rsc, node, task, *interval_ms,
action_config);
g_hash_table_foreach(params, hash2metafield, data->params_all);
g_hash_table_destroy(params);
pcmk__filter_op_for_digest(data->params_all);
data->digest_all_calc = calculate_operation_digest(data->params_all,
op_version);
}
// Return true if XML attribute name is a Pacemaker-defined fencing parameter
static bool
is_fence_param(xmlAttrPtr attr, void *user_data)
{
return pcmk_stonith_param((const char *) attr->name);
}
/*!
* \internal
* \brief Add secure digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] rsc Resource that action was for
* \param[in] params Resource parameters evaluated for node
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
* \param[in] overrides Key/value hash table to override resource parameters
*/
static void
calculate_secure_digest(pcmk__op_digest_t *data, const pcmk_resource_t *rsc,
GHashTable *params, const xmlNode *xml_op,
const char *op_version, GHashTable *overrides)
{
const char *class = crm_element_value(rsc->xml, PCMK_XA_CLASS);
const char *secure_list = NULL;
bool old_version = (compare_version(op_version, "3.16.0") < 0);
if (xml_op == NULL) {
secure_list = " passwd password user ";
} else {
secure_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_SECURE);
}
if (old_version) {
data->params_secure = create_xml_node(NULL, XML_TAG_PARAMS);
if (overrides != NULL) {
g_hash_table_foreach(overrides, hash2field, data->params_secure);
}
g_hash_table_foreach(params, hash2field, data->params_secure);
} else {
// Start with a copy of all parameters
data->params_secure = copy_xml(data->params_all);
}
if (secure_list != NULL) {
pcmk__xe_remove_matching_attrs(data->params_secure, attr_in_string,
(void *) secure_list);
}
if (old_version
&& pcmk_is_set(pcmk_get_ra_caps(class),
pcmk_ra_cap_fence_params)) {
/* For stonith resources, Pacemaker adds special parameters,
* but these are not listed in fence agent meta-data, so with older
* versions of DC, the controller will not hash them. That means we have
* to filter them out before calculating our hash for comparison.
*/
pcmk__xe_remove_matching_attrs(data->params_secure, is_fence_param,
NULL);
}
pcmk__filter_op_for_digest(data->params_secure);
/* CRM_meta_timeout *should* be part of a digest for recurring operations.
* However, with older versions of DC, the controller does not add timeout
* to secure digests, because it only includes parameters declared by the
* resource agent.
* Remove any timeout that made it this far, to match.
*/
if (old_version) {
xml_remove_prop(data->params_secure, CRM_META "_" PCMK_META_TIMEOUT);
}
data->digest_secure_calc = calculate_operation_digest(data->params_secure,
op_version);
}
/*!
* \internal
* \brief Add restart digest to a digest cache entry
*
* \param[out] data Digest cache entry to modify
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] op_version CRM feature set to use for digest calculation
*
* \note This function doesn't need to handle overrides because it starts with
* data->params_all, which already has overrides applied.
*/
static void
calculate_restart_digest(pcmk__op_digest_t *data, const xmlNode *xml_op,
const char *op_version)
{
const char *value = NULL;
// We must have XML of resource operation history
if (xml_op == NULL) {
return;
}
// And the history must have a restart digest to compare against
if (crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST) == NULL) {
return;
}
// Start with a copy of all parameters
data->params_restart = copy_xml(data->params_all);
// Then filter out reloadable parameters, if any
- value = crm_element_value(xml_op, XML_LRM_ATTR_OP_RESTART);
+ value = crm_element_value(xml_op, PCMK__XA_OP_FORCE_RESTART);
if (value != NULL) {
pcmk__xe_remove_matching_attrs(data->params_restart, attr_not_in_string,
(void *) value);
}
value = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
data->digest_restart_calc = calculate_operation_digest(data->params_restart,
value);
}
/*!
* \internal
* \brief Create a new digest cache entry with calculated digests
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in,out] interval_ms Action's interval (will be reset if in overrides)
* \param[in] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] overrides Key/value table to override resource parameters
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to new digest cache entry (or NULL on memory error)
* \note It is the caller's responsibility to free the result using
* pe__free_digests().
*/
pcmk__op_digest_t *
pe__calculate_digests(pcmk_resource_t *rsc, const char *task,
guint *interval_ms, const pcmk_node_t *node,
const xmlNode *xml_op, GHashTable *overrides,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = calloc(1, sizeof(pcmk__op_digest_t));
const char *op_version = NULL;
GHashTable *params = NULL;
if (data == NULL) {
pcmk__sched_err("Could not allocate memory for operation digest");
return NULL;
}
data->rc = pcmk__digest_match;
if (xml_op != NULL) {
op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
}
if (op_version == NULL && scheduler != NULL && scheduler->input != NULL) {
op_version = crm_element_value(scheduler->input,
PCMK_XA_CRM_FEATURE_SET);
}
if (op_version == NULL) {
op_version = CRM_FEATURE_SET;
}
params = pe_rsc_params(rsc, node, scheduler);
calculate_main_digest(data, rsc, node, params, task, interval_ms, xml_op,
op_version, overrides, scheduler);
if (calc_secure) {
calculate_secure_digest(data, rsc, params, xml_op, op_version,
overrides);
}
calculate_restart_digest(data, xml_op, op_version);
return data;
}
/*!
* \internal
* \brief Calculate action digests and store in node's digest cache
*
* \param[in,out] rsc Resource that action was for
* \param[in] task Name of action performed
* \param[in] interval_ms Action's interval
* \param[in,out] node Node action was performed on
* \param[in] xml_op XML of operation in CIB status (if available)
* \param[in] calc_secure Whether to calculate secure digest
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry
*/
static pcmk__op_digest_t *
rsc_action_digest(pcmk_resource_t *rsc, const char *task, guint interval_ms,
pcmk_node_t *node, const xmlNode *xml_op,
bool calc_secure, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
char *key = pcmk__op_key(rsc->id, task, interval_ms);
data = g_hash_table_lookup(node->details->digest_cache, key);
if (data == NULL) {
data = pe__calculate_digests(rsc, task, &interval_ms, node, xml_op,
NULL, calc_secure, scheduler);
CRM_ASSERT(data != NULL);
g_hash_table_insert(node->details->digest_cache, strdup(key), data);
}
free(key);
return data;
}
/*!
* \internal
* \brief Calculate operation digests and compare against an XML history entry
*
* \param[in,out] rsc Resource to check
* \param[in] xml_op Resource history XML
* \param[in,out] node Node to use for digest calculation
* \param[in,out] scheduler Scheduler data
*
* \return Pointer to node's digest cache entry, with comparison result set
*/
pcmk__op_digest_t *
rsc_action_digest_cmp(pcmk_resource_t *rsc, const xmlNode *xml_op,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
pcmk__op_digest_t *data = NULL;
guint interval_ms = 0;
const char *op_version;
const char *task = crm_element_value(xml_op, PCMK_XA_OPERATION);
const char *digest_all;
const char *digest_restart;
CRM_ASSERT(node != NULL);
op_version = crm_element_value(xml_op, PCMK_XA_CRM_FEATURE_SET);
digest_all = crm_element_value(xml_op, PCMK__XA_OP_DIGEST);
digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST);
crm_element_value_ms(xml_op, PCMK_META_INTERVAL, &interval_ms);
data = rsc_action_digest(rsc, task, interval_ms, node, xml_op,
pcmk_is_set(scheduler->flags,
pcmk_sched_sanitized),
scheduler);
if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) {
pcmk__rsc_info(rsc,
"Parameters to %ums-interval %s action for %s on %s "
"changed: hash was %s vs. now %s (restart:%s) %s",
interval_ms, task, rsc->id, pe__node_name(node),
pcmk__s(digest_restart, "missing"),
data->digest_restart_calc, op_version,
crm_element_value(xml_op, PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_restart;
} else if (digest_all == NULL) {
/* it is unknown what the previous op digest was */
data->rc = pcmk__digest_unknown;
} else if (strcmp(digest_all, data->digest_all_calc) != 0) {
/* Given a non-recurring operation with extra parameters configured,
* in case that the main digest doesn't match, even if the restart
* digest matches, enforce a restart rather than a reload-agent anyway.
* So that it ensures any changes of the extra parameters get applied
* for this specific operation, and the digests calculated for the
* resulting lrm_rsc_op will be correct.
* Preserve the implied rc pcmk__digest_restart for the case that the
* main digest doesn't match.
*/
if ((interval_ms == 0) && (data->rc == pcmk__digest_restart)) {
pcmk__rsc_info(rsc,
"Parameters containing extra ones to %ums-interval"
" %s action for %s on %s "
"changed: hash was %s vs. now %s (restart:%s) %s",
interval_ms, task, rsc->id, pe__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc, op_version,
crm_element_value(xml_op,
PCMK__XA_TRANSITION_MAGIC));
} else {
pcmk__rsc_info(rsc,
"Parameters to %ums-interval %s action for %s on %s "
"changed: hash was %s vs. now %s (%s:%s) %s",
interval_ms, task, rsc->id, pe__node_name(node),
pcmk__s(digest_all, "missing"),
data->digest_all_calc,
(interval_ms > 0)? "reschedule" : "reload",
op_version,
crm_element_value(xml_op,
PCMK__XA_TRANSITION_MAGIC));
data->rc = pcmk__digest_mismatch;
}
} else {
data->rc = pcmk__digest_match;
}
return data;
}
/*!
* \internal
* \brief Create an unfencing summary for use in special node attribute
*
* Create a string combining a fence device's resource ID, agent type, and
* parameter digest (whether for all parameters or just non-private parameters).
* This can be stored in a special node attribute, allowing us to detect changes
* in either the agent type or parameters, to know whether unfencing must be
* redone or can be safely skipped when the device's history is cleaned.
*
* \param[in] rsc_id Fence device resource ID
* \param[in] agent_type Fence device agent
* \param[in] param_digest Fence device parameter digest
*
* \return Newly allocated string with unfencing digest
* \note The caller is responsible for freeing the result.
*/
static inline char *
create_unfencing_summary(const char *rsc_id, const char *agent_type,
const char *param_digest)
{
return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest);
}
/*!
* \internal
* \brief Check whether a node can skip unfencing
*
* Check whether a fence device's current definition matches a node's
* stored summary of when it was last unfenced by the device.
*
* \param[in] rsc_id Fence device's resource ID
* \param[in] agent Fence device's agent type
* \param[in] digest_calc Fence device's current parameter digest
* \param[in] node_summary Value of node's special unfencing node attribute
* (a comma-separated list of unfencing summaries for
* all devices that have unfenced this node)
*
* \return TRUE if digest matches, FALSE otherwise
*/
static bool
unfencing_digest_matches(const char *rsc_id, const char *agent,
const char *digest_calc, const char *node_summary)
{
bool matches = FALSE;
if (rsc_id && agent && digest_calc && node_summary) {
char *search_secure = create_unfencing_summary(rsc_id, agent,
digest_calc);
/* The digest was calculated including the device ID and agent,
* so there is no risk of collision using strstr().
*/
matches = (strstr(node_summary, search_secure) != NULL);
crm_trace("Calculated unfencing digest '%s' %sfound in '%s'",
search_secure, matches? "" : "not ", node_summary);
free(search_secure);
}
return matches;
}
/* Magic string to use as action name for digest cache entries used for
* unfencing checks. This is not a real action name (i.e. "on"), so
* pcmk__check_action_config() won't confuse these entries with real actions.
*/
#define STONITH_DIGEST_TASK "stonith-on"
/*!
* \internal
* \brief Calculate fence device digests and digest comparison result
*
* \param[in,out] rsc Fence device resource
* \param[in] agent Fence device's agent type
* \param[in,out] node Node with digest cache to use
* \param[in,out] scheduler Scheduler data
*
* \return Node's digest cache entry
*/
pcmk__op_digest_t *
pe__compare_fencing_digest(pcmk_resource_t *rsc, const char *agent,
pcmk_node_t *node, pcmk_scheduler_t *scheduler)
{
const char *node_summary = NULL;
// Calculate device's current parameter digests
pcmk__op_digest_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, 0U,
node, NULL, TRUE, scheduler);
// Check whether node has special unfencing summary node attribute
node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_ALL);
if (node_summary == NULL) {
data->rc = pcmk__digest_unknown;
return data;
}
// Check whether full parameter digest matches
if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc,
node_summary)) {
data->rc = pcmk__digest_match;
return data;
}
// Check whether secure parameter digest matches
node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_SECURE);
if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc,
node_summary)) {
data->rc = pcmk__digest_match;
if (!pcmk__is_daemon && scheduler->priv != NULL) {
pcmk__output_t *out = scheduler->priv;
out->info(out, "Only 'private' parameters to %s "
"for unfencing %s changed", rsc->id,
pe__node_name(node));
}
return data;
}
// Parameters don't match
data->rc = pcmk__digest_mismatch;
if (pcmk_is_set(scheduler->flags, pcmk_sched_sanitized)
&& (data->digest_secure_calc != NULL)) {
if (scheduler->priv != NULL) {
pcmk__output_t *out = scheduler->priv;
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
out->info(out, "Parameters to %s for unfencing "
"%s changed, try '%s'", rsc->id,
pe__node_name(node), digest);
free(digest);
} else if (!pcmk__is_daemon) {
char *digest = create_unfencing_summary(rsc->id, agent,
data->digest_secure_calc);
printf("Parameters to %s for unfencing %s changed, try '%s'\n",
rsc->id, pe__node_name(node), digest);
free(digest);
}
}
return data;
}

File Metadata

Mime Type
text/x-diff
Expires
Thu, Jun 26, 7:47 PM (19 h, 12 m)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1959600
Default Alt Text
(68 KB)

Event Timeline