Page MenuHomeClusterLabs Projects
Files F2825262

No OneTemporary
Actions

View Options

diff --git a/daemons/attrd/attrd_utils.c b/daemons/attrd/attrd_utils.c
index 6fc16bedce..8bd3cd5d2a 100644
--- a/daemons/attrd/attrd_utils.c
+++ b/daemons/attrd/attrd_utils.c
@@ -1,285 +1,273 @@
/*
* Copyright 2004-2020 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 <stdbool.h>
#include <errno.h>
#include <glib.h>
#include <regex.h>
#include <sys/types.h>
#include <crm/crm.h>
#include <crm/common/ipcs_internal.h>
#include <crm/common/mainloop.h>
#include "pacemaker-attrd.h"
cib_t *the_cib = NULL;
static bool shutting_down = FALSE;
static GMainLoop *mloop = NULL;
/*!
* \internal
* \brief Check whether we're currently shutting down
*
* \return TRUE if shutting down, FALSE otherwise
*/
gboolean
attrd_shutting_down()
{
return shutting_down;
}
/*!
* \internal
* \brief Exit (using mainloop or not, as appropriate)
*
* \param[in] nsig Ignored
*/
void
attrd_shutdown(int nsig)
{
// Tell various functions not to do anthing
shutting_down = TRUE;
// Don't respond to signals while shutting down
mainloop_destroy_signal(SIGTERM);
mainloop_destroy_signal(SIGCHLD);
mainloop_destroy_signal(SIGPIPE);
mainloop_destroy_signal(SIGUSR1);
mainloop_destroy_signal(SIGUSR2);
mainloop_destroy_signal(SIGTRAP);
if ((mloop == NULL) || !g_main_loop_is_running(mloop)) {
/* If there's no main loop active, just exit. This should be possible
* only if we get SIGTERM in brief windows at start-up and shutdown.
*/
crm_exit(CRM_EX_OK);
} else {
g_main_loop_quit(mloop);
g_main_loop_unref(mloop);
}
}
/*!
* \internal
* \brief Create a main loop for attrd
*/
void
attrd_init_mainloop()
{
mloop = g_main_loop_new(NULL, FALSE);
}
/*!
* \internal
* \brief Run attrd main loop
*/
void
attrd_run_mainloop()
{
g_main_loop_run(mloop);
}
/*!
* \internal
* \brief Accept a new client IPC connection
*
* \param[in] c New connection
* \param[in] uid Client user id
* \param[in] gid Client group id
*
* \return pcmk_ok on success, -errno otherwise
*/
static int32_t
attrd_ipc_accept(qb_ipcs_connection_t *c, uid_t uid, gid_t gid)
{
crm_trace("New client connection %p", c);
if (shutting_down) {
crm_info("Ignoring new connection from pid %d during shutdown",
pcmk__client_pid(c));
return -EPERM;
}
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return pcmk_ok;
}
-/*!
- * \internal
- * \brief Callback for successful client connection
- *
- * \param[in] c New connection
- */
-static void
-attrd_ipc_created(qb_ipcs_connection_t *c)
-{
- crm_trace("Client connection %p accepted", c);
-}
-
/*!
* \internal
* \brief Destroy a client IPC connection
*
* \param[in] c Connection to destroy
*
* \return FALSE (i.e. do not re-run this callback)
*/
static int32_t
attrd_ipc_closed(qb_ipcs_connection_t *c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
crm_trace("Ignoring request to clean up unknown connection %p", c);
} else {
crm_trace("Cleaning up closed client connection %p", c);
pcmk__free_client(client);
}
return FALSE;
}
/*!
* \internal
* \brief Destroy a client IPC connection
*
* \param[in] c Connection to destroy
*
* \note We handle a destroyed connection the same as a closed one,
* but we need a separate handler because the return type is different.
*/
static void
attrd_ipc_destroy(qb_ipcs_connection_t *c)
{
crm_trace("Destroying client connection %p", c);
attrd_ipc_closed(c);
}
/*!
* \internal
* \brief Set up attrd IPC communication
*
* \param[out] ipcs Will be set to newly allocated server connection
* \param[in] dispatch_fn Handler for new messages on connection
*/
void
attrd_init_ipc(qb_ipcs_service_t **ipcs, qb_ipcs_msg_process_fn dispatch_fn)
{
static struct qb_ipcs_service_handlers ipc_callbacks = {
.connection_accept = attrd_ipc_accept,
- .connection_created = attrd_ipc_created,
+ .connection_created = NULL,
.msg_process = NULL,
.connection_closed = attrd_ipc_closed,
.connection_destroyed = attrd_ipc_destroy
};
ipc_callbacks.msg_process = dispatch_fn;
attrd_ipc_server_init(ipcs, &ipc_callbacks);
}
void
attrd_cib_disconnect()
{
CRM_CHECK(the_cib != NULL, return);
the_cib->cmds->signoff(the_cib);
cib_delete(the_cib);
the_cib = NULL;
}
/* strlen("value") */
#define plus_plus_len (5)
/*!
* \internal
* \brief Check whether an attribute value should be expanded
*
* \param[in] value Attribute value to check
*
* \return TRUE if value needs expansion, FALSE otherwise
*/
gboolean
attrd_value_needs_expansion(const char *value)
{
return ((strlen(value) >= (plus_plus_len + 2))
&& (value[plus_plus_len] == '+')
&& ((value[plus_plus_len + 1] == '+')
|| (value[plus_plus_len + 1] == '=')));
}
/*!
* \internal
* \brief Expand an increment expression into an integer
*
* \param[in] value Attribute increment expression to expand
* \param[in] old_value Previous value of attribute
*
* \return Expanded value
*/
int
attrd_expand_value(const char *value, const char *old_value)
{
int offset = 1;
int int_value = char2score(old_value);
if (value[plus_plus_len + 1] != '+') {
const char *offset_s = value + (plus_plus_len + 2);
offset = char2score(offset_s);
}
int_value += offset;
if (int_value > INFINITY) {
int_value = INFINITY;
}
return int_value;
}
/*!
* \internal
* \brief Create regular expression matching failure-related attributes
*
* \param[out] regex Where to store created regular expression
* \param[in] rsc Name of resource to clear (or NULL for all)
* \param[in] op Operation to clear if rsc is specified (or NULL for all)
* \param[in] interval_ms Interval of operation to clear if op is specified
*
* \return pcmk_ok on success, -EINVAL if arguments are invalid
*
* \note The caller is responsible for freeing the result with regfree().
*/
int
attrd_failure_regex(regex_t *regex, const char *rsc, const char *op,
guint interval_ms)
{
char *pattern = NULL;
int rc;
/* Create a pattern that matches desired attributes */
if (rsc == NULL) {
pattern = strdup(ATTRD_RE_CLEAR_ALL);
} else if (op == NULL) {
pattern = crm_strdup_printf(ATTRD_RE_CLEAR_ONE, rsc);
} else {
pattern = crm_strdup_printf(ATTRD_RE_CLEAR_OP, rsc, op, interval_ms);
}
/* Compile pattern into regular expression */
crm_trace("Clearing attributes matching %s", pattern);
rc = regcomp(regex, pattern, REG_EXTENDED|REG_NOSUB);
free(pattern);
return (rc == 0)? pcmk_ok : -EINVAL;
}
diff --git a/daemons/based/based_callbacks.c b/daemons/based/based_callbacks.c
index 200bce902a..87ed31c9b4 100644
--- a/daemons/based/based_callbacks.c
+++ b/daemons/based/based_callbacks.c
@@ -1,1557 +1,1551 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <stdio.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h> /* U64T ~ PRIu64 */
#include <crm/crm.h>
#include <crm/cib.h>
#include <crm/msg_xml.h>
#include <crm/cluster/internal.h>
#include <crm/common/xml.h>
#include <crm/common/remote_internal.h>
#include <pacemaker-based.h>
#define EXIT_ESCALATION_MS 10000
static unsigned long cib_local_bcast_num = 0;
typedef struct cib_local_notify_s {
xmlNode *notify_src;
char *client_id;
gboolean from_peer;
gboolean sync_reply;
} cib_local_notify_t;
int next_client_id = 0;
gboolean legacy_mode = FALSE;
qb_ipcs_service_t *ipcs_ro = NULL;
qb_ipcs_service_t *ipcs_rw = NULL;
qb_ipcs_service_t *ipcs_shm = NULL;
void send_cib_replace(const xmlNode * sync_request, const char *host);
static void cib_process_request(xmlNode* request, gboolean force_synchronous,
gboolean privileged,
pcmk__client_t *cib_client);
static int cib_process_command(xmlNode *request, xmlNode **reply,
xmlNode **cib_diff, gboolean privileged);
gboolean cib_common_callback(qb_ipcs_connection_t * c, void *data, size_t size,
gboolean privileged);
gboolean cib_legacy_mode(void)
{
return legacy_mode;
}
static int32_t
cib_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
if (cib_shutdown_flag) {
crm_info("Ignoring new IPC client [%d] during shutdown",
pcmk__client_pid(c));
return -EPERM;
}
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
-static void
-cib_ipc_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection %p", c);
-}
-
static int32_t
cib_ipc_dispatch_rw(qb_ipcs_connection_t * c, void *data, size_t size)
{
pcmk__client_t *client = pcmk__find_client(c);
crm_trace("%p message from %s", c, client->id);
return cib_common_callback(c, data, size, TRUE);
}
static int32_t
cib_ipc_dispatch_ro(qb_ipcs_connection_t * c, void *data, size_t size)
{
pcmk__client_t *client = pcmk__find_client(c);
crm_trace("%p message from %s", c, client->id);
return cib_common_callback(c, data, size, FALSE);
}
/* Error code means? */
static int32_t
cib_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p", c);
pcmk__free_client(client);
return 0;
}
static void
cib_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
cib_ipc_closed(c);
if (cib_shutdown_flag) {
cib_shutdown(0);
}
}
struct qb_ipcs_service_handlers ipc_ro_callbacks = {
.connection_accept = cib_ipc_accept,
- .connection_created = cib_ipc_created,
+ .connection_created = NULL,
.msg_process = cib_ipc_dispatch_ro,
.connection_closed = cib_ipc_closed,
.connection_destroyed = cib_ipc_destroy
};
struct qb_ipcs_service_handlers ipc_rw_callbacks = {
.connection_accept = cib_ipc_accept,
- .connection_created = cib_ipc_created,
+ .connection_created = NULL,
.msg_process = cib_ipc_dispatch_rw,
.connection_closed = cib_ipc_closed,
.connection_destroyed = cib_ipc_destroy
};
void
cib_common_callback_worker(uint32_t id, uint32_t flags, xmlNode * op_request,
pcmk__client_t *cib_client, gboolean privileged)
{
const char *op = crm_element_value(op_request, F_CIB_OPERATION);
if (crm_str_eq(op, CRM_OP_REGISTER, TRUE)) {
if (flags & crm_ipc_client_response) {
xmlNode *ack = create_xml_node(NULL, __FUNCTION__);
crm_xml_add(ack, F_CIB_OPERATION, CRM_OP_REGISTER);
crm_xml_add(ack, F_CIB_CLIENTID, cib_client->id);
pcmk__ipc_send_xml(cib_client, id, ack, flags);
cib_client->request_id = 0;
free_xml(ack);
}
return;
} else if (crm_str_eq(op, T_CIB_NOTIFY, TRUE)) {
/* Update the notify filters for this client */
int on_off = 0;
long long bit = 0;
const char *type = crm_element_value(op_request, F_CIB_NOTIFY_TYPE);
crm_element_value_int(op_request, F_CIB_NOTIFY_ACTIVATE, &on_off);
crm_debug("Setting %s callbacks for %s (%s): %s",
type, cib_client->name, cib_client->id, on_off ? "on" : "off");
if (safe_str_eq(type, T_CIB_POST_NOTIFY)) {
bit = cib_notify_post;
} else if (safe_str_eq(type, T_CIB_PRE_NOTIFY)) {
bit = cib_notify_pre;
} else if (safe_str_eq(type, T_CIB_UPDATE_CONFIRM)) {
bit = cib_notify_confirm;
} else if (safe_str_eq(type, T_CIB_DIFF_NOTIFY)) {
bit = cib_notify_diff;
} else if (safe_str_eq(type, T_CIB_REPLACE_NOTIFY)) {
bit = cib_notify_replace;
}
if (on_off) {
set_bit(cib_client->options, bit);
} else {
clear_bit(cib_client->options, bit);
}
if (flags & crm_ipc_client_response) {
/* TODO - include rc */
pcmk__ipc_send_ack(cib_client, id, flags, "ack");
}
return;
}
cib_process_request(op_request, FALSE, privileged, cib_client);
}
int32_t
cib_common_callback(qb_ipcs_connection_t * c, void *data, size_t size, gboolean privileged)
{
uint32_t id = 0;
uint32_t flags = 0;
int call_options = 0;
pcmk__client_t *cib_client = pcmk__find_client(c);
xmlNode *op_request = pcmk__client_data2xml(cib_client, data, size, &id,
&flags);
if (op_request) {
crm_element_value_int(op_request, F_CIB_CALLOPTS, &call_options);
}
if (op_request == NULL) {
crm_trace("Invalid message from %p", c);
pcmk__ipc_send_ack(cib_client, id, flags, "nack");
return 0;
} else if(cib_client == NULL) {
crm_trace("Invalid client %p", c);
return 0;
}
if (is_set(call_options, cib_sync_call)) {
CRM_LOG_ASSERT(flags & crm_ipc_client_response);
CRM_LOG_ASSERT(cib_client->request_id == 0); /* This means the client has two synchronous events in-flight */
cib_client->request_id = id; /* Reply only to the last one */
}
if (cib_client->name == NULL) {
const char *value = crm_element_value(op_request, F_CIB_CLIENTNAME);
if (value == NULL) {
cib_client->name = crm_itoa(cib_client->pid);
} else {
cib_client->name = strdup(value);
if (crm_is_daemon_name(value)) {
set_bit(cib_client->options, cib_is_daemon);
}
}
}
/* Allow cluster daemons more leeway before being evicted */
if (is_set(cib_client->options, cib_is_daemon)) {
const char *qmax = cib_config_lookup("cluster-ipc-limit");
if (pcmk__set_client_queue_max(cib_client, qmax)) {
crm_trace("IPC threshold for %s[%u] is now %u",
cib_client->name, cib_client->pid, cib_client->queue_max);
}
}
crm_xml_add(op_request, F_CIB_CLIENTID, cib_client->id);
crm_xml_add(op_request, F_CIB_CLIENTNAME, cib_client->name);
#if ENABLE_ACL
CRM_LOG_ASSERT(cib_client->user != NULL);
crm_acl_get_set_user(op_request, F_CIB_USER, cib_client->user);
#endif
crm_log_xml_trace(op_request, "Client[inbound]");
cib_common_callback_worker(id, flags, op_request, cib_client, privileged);
free_xml(op_request);
return 0;
}
static uint64_t ping_seq = 0;
static char *ping_digest = NULL;
static bool ping_modified_since = FALSE;
int sync_our_cib(xmlNode * request, gboolean all);
static gboolean
cib_digester_cb(gpointer data)
{
if (cib_is_master) {
char buffer[32];
xmlNode *ping = create_xml_node(NULL, "ping");
ping_seq++;
free(ping_digest);
ping_digest = NULL;
ping_modified_since = FALSE;
snprintf(buffer, 32, "%" U64T, ping_seq);
crm_trace("Requesting peer digests (%s)", buffer);
crm_xml_add(ping, F_TYPE, "cib");
crm_xml_add(ping, F_CIB_OPERATION, CRM_OP_PING);
crm_xml_add(ping, F_CIB_PING_ID, buffer);
crm_xml_add(ping, XML_ATTR_CRM_VERSION, CRM_FEATURE_SET);
send_cluster_message(NULL, crm_msg_cib, ping, TRUE);
free_xml(ping);
}
return FALSE;
}
static void
process_ping_reply(xmlNode *reply)
{
uint64_t seq = 0;
const char *host = crm_element_value(reply, F_ORIG);
xmlNode *pong = get_message_xml(reply, F_CIB_CALLDATA);
const char *seq_s = crm_element_value(pong, F_CIB_PING_ID);
const char *digest = crm_element_value(pong, XML_ATTR_DIGEST);
if (seq_s) {
seq = (uint64_t) crm_parse_ll(seq_s, NULL);
}
if(digest == NULL) {
crm_trace("Ignoring ping reply %s from %s with no digest", seq_s, host);
} else if(seq != ping_seq) {
crm_trace("Ignoring out of sequence ping reply %s from %s", seq_s, host);
} else if(ping_modified_since) {
crm_trace("Ignoring ping reply %s from %s: cib updated since", seq_s, host);
} else {
const char *version = crm_element_value(pong, XML_ATTR_CRM_VERSION);
if(ping_digest == NULL) {
crm_trace("Calculating new digest");
ping_digest = calculate_xml_versioned_digest(the_cib, FALSE, TRUE, version);
}
crm_trace("Processing ping reply %s from %s (%s)", seq_s, host, digest);
if(safe_str_eq(ping_digest, digest) == FALSE) {
xmlNode *remote_cib = get_message_xml(pong, F_CIB_CALLDATA);
crm_notice("Local CIB %s.%s.%s.%s differs from %s: %s.%s.%s.%s %p",
crm_element_value(the_cib, XML_ATTR_GENERATION_ADMIN),
crm_element_value(the_cib, XML_ATTR_GENERATION),
crm_element_value(the_cib, XML_ATTR_NUMUPDATES),
ping_digest, host,
remote_cib?crm_element_value(remote_cib, XML_ATTR_GENERATION_ADMIN):"_",
remote_cib?crm_element_value(remote_cib, XML_ATTR_GENERATION):"_",
remote_cib?crm_element_value(remote_cib, XML_ATTR_NUMUPDATES):"_",
digest, remote_cib);
if(remote_cib && remote_cib->children) {
/* Additional debug */
xml_calculate_changes(the_cib, remote_cib);
xml_log_changes(LOG_INFO, __FUNCTION__, remote_cib);
crm_trace("End of differences");
}
free_xml(remote_cib);
sync_our_cib(reply, FALSE);
}
}
}
static void
do_local_notify(xmlNode * notify_src, const char *client_id,
gboolean sync_reply, gboolean from_peer)
{
int rid = 0;
int call_id = 0;
pcmk__client_t *client_obj = NULL;
CRM_ASSERT(notify_src && client_id);
crm_element_value_int(notify_src, F_CIB_CALLID, &call_id);
client_obj = pcmk__find_client_by_id(client_id);
if (client_obj == NULL) {
crm_debug("Could not send response %d: client %s not found",
call_id, client_id);
return;
}
if (sync_reply) {
if (client_obj->ipcs) {
CRM_LOG_ASSERT(client_obj->request_id);
rid = client_obj->request_id;
client_obj->request_id = 0;
crm_trace("Sending response %d to %s %s",
rid, client_obj->name,
from_peer ? "(originator of delegated request)" : "");
} else {
crm_trace("Sending response [call %d] to %s %s",
call_id, client_obj->name, from_peer ? "(originator of delegated request)" : "");
}
} else {
crm_trace("Sending event %d to %s %s",
call_id, client_obj->name, from_peer ? "(originator of delegated request)" : "");
}
switch (client_obj->kind) {
case PCMK__CLIENT_IPC:
{
int rc = pcmk__ipc_send_xml(client_obj, rid, notify_src,
(sync_reply? crm_ipc_flags_none
: crm_ipc_server_event));
if (rc != pcmk_rc_ok) {
crm_warn("%s reply to %s failed: %s " CRM_XS " rc=%d",
(sync_reply? "Synchronous" : "Asynchronous"),
client_obj->name, pcmk_rc_str(rc), rc);
}
}
break;
#ifdef HAVE_GNUTLS_GNUTLS_H
case PCMK__CLIENT_TLS:
#endif
case PCMK__CLIENT_TCP:
pcmk__remote_send_xml(client_obj->remote, notify_src);
break;
default:
crm_err("Unknown transport %d for %s", client_obj->kind, client_obj->name);
}
}
static void
local_notify_destroy_callback(gpointer data)
{
cib_local_notify_t *notify = data;
free_xml(notify->notify_src);
free(notify->client_id);
free(notify);
}
static void
check_local_notify(int bcast_id)
{
cib_local_notify_t *notify = NULL;
if (!local_notify_queue) {
return;
}
notify = g_hash_table_lookup(local_notify_queue, GINT_TO_POINTER(bcast_id));
if (notify) {
do_local_notify(notify->notify_src, notify->client_id, notify->sync_reply,
notify->from_peer);
g_hash_table_remove(local_notify_queue, GINT_TO_POINTER(bcast_id));
}
}
static void
queue_local_notify(xmlNode * notify_src, const char *client_id, gboolean sync_reply,
gboolean from_peer)
{
cib_local_notify_t *notify = calloc(1, sizeof(cib_local_notify_t));
notify->notify_src = notify_src;
notify->client_id = strdup(client_id);
notify->sync_reply = sync_reply;
notify->from_peer = from_peer;
if (!local_notify_queue) {
local_notify_queue = g_hash_table_new_full(g_direct_hash,
g_direct_equal, NULL,
local_notify_destroy_callback);
}
g_hash_table_insert(local_notify_queue, GINT_TO_POINTER(cib_local_bcast_num), notify);
}
static void
parse_local_options_v1(pcmk__client_t *cib_client, int call_type,
int call_options, const char *host, const char *op,
gboolean *local_notify, gboolean *needs_reply,
gboolean *process, gboolean *needs_forward)
{
if (cib_op_modifies(call_type)
&& !(call_options & cib_inhibit_bcast)) {
/* we need to send an update anyway */
*needs_reply = TRUE;
} else {
*needs_reply = FALSE;
}
if (host == NULL && (call_options & cib_scope_local)) {
crm_trace("Processing locally scoped %s op from %s", op, cib_client->name);
*local_notify = TRUE;
} else if (host == NULL && cib_is_master) {
crm_trace("Processing master %s op locally from %s", op, cib_client->name);
*local_notify = TRUE;
} else if (safe_str_eq(host, cib_our_uname)) {
crm_trace("Processing locally addressed %s op from %s", op, cib_client->name);
*local_notify = TRUE;
} else if (stand_alone) {
*needs_forward = FALSE;
*local_notify = TRUE;
*process = TRUE;
} else {
crm_trace("%s op from %s needs to be forwarded to %s",
op, cib_client->name, host ? host : "the master instance");
*needs_forward = TRUE;
*process = FALSE;
}
}
static void
parse_local_options_v2(pcmk__client_t *cib_client, int call_type,
int call_options, const char *host, const char *op,
gboolean *local_notify, gboolean *needs_reply,
gboolean *process, gboolean *needs_forward)
{
if (cib_op_modifies(call_type)) {
if(safe_str_eq(op, CIB_OP_MASTER) || safe_str_eq(op, CIB_OP_SLAVE)) {
/* Always handle these locally */
*process = TRUE;
*needs_reply = FALSE;
*local_notify = TRUE;
*needs_forward = FALSE;
return;
} else {
/* Redirect all other updates via CPG */
*needs_reply = TRUE;
*needs_forward = TRUE;
*process = FALSE;
crm_trace("%s op from %s needs to be forwarded to %s",
op, cib_client->name, host ? host : "the master instance");
return;
}
}
*process = TRUE;
*needs_reply = FALSE;
*local_notify = TRUE;
*needs_forward = FALSE;
if (stand_alone) {
crm_trace("Processing %s op from %s (stand-alone)", op, cib_client->name);
} else if (host == NULL) {
crm_trace("Processing unaddressed %s op from %s", op, cib_client->name);
} else if (safe_str_eq(host, cib_our_uname)) {
crm_trace("Processing locally addressed %s op from %s", op, cib_client->name);
} else {
crm_trace("%s op from %s needs to be forwarded to %s", op, cib_client->name, host);
*needs_forward = TRUE;
*process = FALSE;
}
}
static void
parse_local_options(pcmk__client_t *cib_client, int call_type,
int call_options, const char *host, const char *op,
gboolean *local_notify, gboolean *needs_reply,
gboolean *process, gboolean *needs_forward)
{
if(cib_legacy_mode()) {
parse_local_options_v1(cib_client, call_type, call_options, host,
op, local_notify, needs_reply, process, needs_forward);
} else {
parse_local_options_v2(cib_client, call_type, call_options, host,
op, local_notify, needs_reply, process, needs_forward);
}
}
static gboolean
parse_peer_options_v1(int call_type, xmlNode * request,
gboolean * local_notify, gboolean * needs_reply, gboolean * process,
gboolean * needs_forward)
{
const char *op = NULL;
const char *host = NULL;
const char *delegated = NULL;
const char *originator = crm_element_value(request, F_ORIG);
const char *reply_to = crm_element_value(request, F_CIB_ISREPLY);
const char *update = crm_element_value(request, F_CIB_GLOBAL_UPDATE);
gboolean is_reply = safe_str_eq(reply_to, cib_our_uname);
if (crm_is_true(update)) {
*needs_reply = FALSE;
if (is_reply) {
*local_notify = TRUE;
crm_trace("Processing global/peer update from %s"
" that originated from us", originator);
} else {
crm_trace("Processing global/peer update from %s", originator);
}
return TRUE;
}
op = crm_element_value(request, F_CIB_OPERATION);
crm_trace("Processing %s request sent by %s", op, originator);
if (safe_str_eq(op, "cib_shutdown_req")) {
/* Always process these */
*local_notify = FALSE;
if (reply_to == NULL || is_reply) {
*process = TRUE;
}
if (is_reply) {
*needs_reply = FALSE;
}
return *process;
}
if (is_reply && safe_str_eq(op, CRM_OP_PING)) {
process_ping_reply(request);
return FALSE;
}
if (is_reply) {
crm_trace("Forward reply sent from %s to local clients", originator);
*process = FALSE;
*needs_reply = FALSE;
*local_notify = TRUE;
return TRUE;
}
host = crm_element_value(request, F_CIB_HOST);
if (host != NULL && safe_str_eq(host, cib_our_uname)) {
crm_trace("Processing %s request sent to us from %s", op, originator);
return TRUE;
} else if(is_reply == FALSE && safe_str_eq(op, CRM_OP_PING)) {
crm_trace("Processing %s request sent to %s by %s", op, host?host:"everyone", originator);
*needs_reply = TRUE;
return TRUE;
} else if (host == NULL && cib_is_master == TRUE) {
crm_trace("Processing %s request sent to master instance from %s", op, originator);
return TRUE;
}
delegated = crm_element_value(request, F_CIB_DELEGATED);
if (delegated != NULL) {
crm_trace("Ignoring msg for master instance");
} else if (host != NULL) {
/* this is for a specific instance and we're not it */
crm_trace("Ignoring msg for instance on %s", crm_str(host));
} else if (reply_to == NULL && cib_is_master == FALSE) {
/* this is for the master instance and we're not it */
crm_trace("Ignoring reply to %s", crm_str(reply_to));
} else if (safe_str_eq(op, "cib_shutdown_req")) {
if (reply_to != NULL) {
crm_debug("Processing %s from %s", op, originator);
*needs_reply = FALSE;
} else {
crm_debug("Processing %s reply from %s", op, originator);
}
return TRUE;
} else {
crm_err("Nothing for us to do?");
crm_log_xml_err(request, "Peer[inbound]");
}
return FALSE;
}
static gboolean
parse_peer_options_v2(int call_type, xmlNode * request,
gboolean * local_notify, gboolean * needs_reply, gboolean * process,
gboolean * needs_forward)
{
const char *host = NULL;
const char *delegated = crm_element_value(request, F_CIB_DELEGATED);
const char *op = crm_element_value(request, F_CIB_OPERATION);
const char *originator = crm_element_value(request, F_ORIG);
const char *reply_to = crm_element_value(request, F_CIB_ISREPLY);
const char *update = crm_element_value(request, F_CIB_GLOBAL_UPDATE);
gboolean is_reply = safe_str_eq(reply_to, cib_our_uname);
if(safe_str_eq(op, CIB_OP_REPLACE)) {
/* sync_our_cib() sets F_CIB_ISREPLY */
if (reply_to) {
delegated = reply_to;
}
goto skip_is_reply;
} else if(safe_str_eq(op, CIB_OP_SYNC)) {
} else if (is_reply && safe_str_eq(op, CRM_OP_PING)) {
process_ping_reply(request);
return FALSE;
} else if (safe_str_eq(op, CIB_OP_UPGRADE)) {
/* Only the DC (node with the oldest software) should process
* this operation if F_CIB_SCHEMA_MAX is unset
*
* If the DC is happy it will then send out another
* CIB_OP_UPGRADE which will tell all nodes to do the actual
* upgrade.
*
* Except this time F_CIB_SCHEMA_MAX will be set which puts a
* limit on how far newer nodes will go
*/
const char *max = crm_element_value(request, F_CIB_SCHEMA_MAX);
const char *upgrade_rc = crm_element_value(request, F_CIB_UPGRADE_RC);
crm_trace("Parsing %s operation%s for %s with max=%s and upgrade_rc=%s",
op, (is_reply? " reply" : ""),
(cib_is_master? "master" : "slave"),
(max? max : "none"), (upgrade_rc? upgrade_rc : "none"));
if (upgrade_rc != NULL) {
// Our upgrade request was rejected by DC, notify clients of result
crm_xml_add(request, F_CIB_RC, upgrade_rc);
} else if ((max == NULL) && cib_is_master) {
/* We are the DC, check if this upgrade is allowed */
goto skip_is_reply;
} else if(max) {
/* Ok, go ahead and upgrade to 'max' */
goto skip_is_reply;
} else {
// Ignore broadcast client requests when we're not DC
return FALSE;
}
} else if (crm_is_true(update)) {
crm_info("Detected legacy %s global update from %s", op, originator);
send_sync_request(NULL);
legacy_mode = TRUE;
return FALSE;
} else if (is_reply && cib_op_modifies(call_type)) {
crm_trace("Ignoring legacy %s reply sent from %s to local clients", op, originator);
return FALSE;
} else if (safe_str_eq(op, "cib_shutdown_req")) {
/* Legacy handling */
crm_debug("Legacy handling of %s message from %s", op, originator);
*local_notify = FALSE;
if (reply_to == NULL) {
*process = TRUE;
}
return *process;
}
if(is_reply) {
crm_trace("Handling %s reply sent from %s to local clients", op, originator);
*process = FALSE;
*needs_reply = FALSE;
*local_notify = TRUE;
return TRUE;
}
skip_is_reply:
*process = TRUE;
*needs_reply = FALSE;
if(safe_str_eq(delegated, cib_our_uname)) {
*local_notify = TRUE;
} else {
*local_notify = FALSE;
}
host = crm_element_value(request, F_CIB_HOST);
if (host != NULL && safe_str_eq(host, cib_our_uname)) {
crm_trace("Processing %s request sent to us from %s", op, originator);
*needs_reply = TRUE;
return TRUE;
} else if (host != NULL) {
/* this is for a specific instance and we're not it */
crm_trace("Ignoring %s operation for instance on %s", op, crm_str(host));
return FALSE;
} else if(is_reply == FALSE && safe_str_eq(op, CRM_OP_PING)) {
*needs_reply = TRUE;
}
crm_trace("Processing %s request sent to everyone by %s/%s on %s %s", op,
crm_element_value(request, F_CIB_CLIENTNAME),
crm_element_value(request, F_CIB_CALLID),
originator, (*local_notify)?"(notify)":"");
return TRUE;
}
static gboolean
parse_peer_options(int call_type, xmlNode * request,
gboolean * local_notify, gboolean * needs_reply, gboolean * process,
gboolean * needs_forward)
{
/* TODO: What happens when an update comes in after node A
* requests the CIB from node B, but before it gets the reply (and
* sends out the replace operation)
*/
if(cib_legacy_mode()) {
return parse_peer_options_v1(
call_type, request, local_notify, needs_reply, process, needs_forward);
} else {
return parse_peer_options_v2(
call_type, request, local_notify, needs_reply, process, needs_forward);
}
}
static void
forward_request(xmlNode * request, pcmk__client_t *cib_client, int call_options)
{
const char *op = crm_element_value(request, F_CIB_OPERATION);
const char *host = crm_element_value(request, F_CIB_HOST);
crm_xml_add(request, F_CIB_DELEGATED, cib_our_uname);
if (host != NULL) {
crm_trace("Forwarding %s op to %s", op, host);
send_cluster_message(crm_get_peer(0, host), crm_msg_cib, request, FALSE);
} else {
crm_trace("Forwarding %s op to master instance", op);
send_cluster_message(NULL, crm_msg_cib, request, FALSE);
}
/* Return the request to its original state */
xml_remove_prop(request, F_CIB_DELEGATED);
if (call_options & cib_discard_reply) {
crm_trace("Client not interested in reply");
}
}
static gboolean
send_peer_reply(xmlNode * msg, xmlNode * result_diff, const char *originator, gboolean broadcast)
{
CRM_ASSERT(msg != NULL);
if (broadcast) {
/* this (successful) call modified the CIB _and_ the
* change needs to be broadcast...
* send via HA to other nodes
*/
int diff_add_updates = 0;
int diff_add_epoch = 0;
int diff_add_admin_epoch = 0;
int diff_del_updates = 0;
int diff_del_epoch = 0;
int diff_del_admin_epoch = 0;
const char *digest = NULL;
int format = 1;
CRM_LOG_ASSERT(result_diff != NULL);
digest = crm_element_value(result_diff, XML_ATTR_DIGEST);
crm_element_value_int(result_diff, "format", &format);
cib_diff_version_details(result_diff,
&diff_add_admin_epoch, &diff_add_epoch, &diff_add_updates,
&diff_del_admin_epoch, &diff_del_epoch, &diff_del_updates);
crm_trace("Sending update diff %d.%d.%d -> %d.%d.%d %s",
diff_del_admin_epoch, diff_del_epoch, diff_del_updates,
diff_add_admin_epoch, diff_add_epoch, diff_add_updates, digest);
crm_xml_add(msg, F_CIB_ISREPLY, originator);
crm_xml_add(msg, F_CIB_GLOBAL_UPDATE, XML_BOOLEAN_TRUE);
crm_xml_add(msg, F_CIB_OPERATION, CIB_OP_APPLY_DIFF);
crm_xml_add(msg, F_CIB_USER, CRM_DAEMON_USER);
if (format == 1) {
CRM_ASSERT(digest != NULL);
}
add_message_xml(msg, F_CIB_UPDATE_DIFF, result_diff);
crm_log_xml_explicit(msg, "copy");
return send_cluster_message(NULL, crm_msg_cib, msg, TRUE);
} else if (originator != NULL) {
/* send reply via HA to originating node */
crm_trace("Sending request result to %s only", originator);
crm_xml_add(msg, F_CIB_ISREPLY, originator);
return send_cluster_message(crm_get_peer(0, originator), crm_msg_cib, msg, FALSE);
}
return FALSE;
}
static void
cib_process_request(xmlNode *request, gboolean force_synchronous,
gboolean privileged, pcmk__client_t *cib_client)
{
int call_type = 0;
int call_options = 0;
gboolean process = TRUE;
gboolean is_update = TRUE;
gboolean from_peer = TRUE;
gboolean needs_reply = TRUE;
gboolean local_notify = FALSE;
gboolean needs_forward = FALSE;
gboolean global_update = crm_is_true(crm_element_value(request, F_CIB_GLOBAL_UPDATE));
xmlNode *op_reply = NULL;
xmlNode *result_diff = NULL;
int rc = pcmk_ok;
const char *op = crm_element_value(request, F_CIB_OPERATION);
const char *originator = crm_element_value(request, F_ORIG);
const char *host = crm_element_value(request, F_CIB_HOST);
const char *target = NULL;
const char *call_id = crm_element_value(request, F_CIB_CALLID);
const char *client_id = crm_element_value(request, F_CIB_CLIENTID);
const char *client_name = crm_element_value(request, F_CIB_CLIENTNAME);
const char *reply_to = crm_element_value(request, F_CIB_ISREPLY);
if (cib_client) {
from_peer = FALSE;
}
crm_element_value_int(request, F_CIB_CALLOPTS, &call_options);
if (force_synchronous) {
call_options |= cib_sync_call;
}
if (host != NULL && strlen(host) == 0) {
host = NULL;
}
if (host) {
target = host;
} else if (call_options & cib_scope_local) {
target = "local host";
} else {
target = "master";
}
if (from_peer) {
crm_trace("Processing peer %s operation from %s/%s on %s intended for %s (reply=%s)",
op, client_name, call_id, originator, target, reply_to);
} else {
crm_xml_add(request, F_ORIG, cib_our_uname);
crm_trace("Processing local %s operation from %s/%s intended for %s", op, client_name, call_id, target);
}
rc = cib_get_operation_id(op, &call_type);
if (rc != pcmk_ok) {
/* TODO: construct error reply? */
crm_err("Pre-processing of command failed: %s", pcmk_strerror(rc));
return;
}
if (from_peer == FALSE) {
parse_local_options(cib_client, call_type, call_options, host, op,
&local_notify, &needs_reply, &process, &needs_forward);
} else if (parse_peer_options(call_type, request, &local_notify,
&needs_reply, &process, &needs_forward) == FALSE) {
return;
}
is_update = cib_op_modifies(call_type);
if (call_options & cib_discard_reply) {
needs_reply = is_update;
local_notify = FALSE;
}
if (needs_forward) {
const char *host = crm_element_value(request, F_CIB_HOST);
const char *section = crm_element_value(request, F_CIB_SECTION);
int log_level = LOG_INFO;
if (safe_str_eq(op, CRM_OP_NOOP)) {
log_level = LOG_DEBUG;
}
do_crm_log(log_level,
"Forwarding %s operation for section %s to %s (origin=%s/%s/%s)",
op,
section ? section : "'all'",
host ? host : cib_legacy_mode() ? "master" : "all",
originator ? originator : "local",
client_name, call_id);
forward_request(request, cib_client, call_options);
return;
}
if (cib_status != pcmk_ok) {
const char *call = crm_element_value(request, F_CIB_CALLID);
rc = cib_status;
crm_err("Operation ignored, cluster configuration is invalid."
" Please repair and restart: %s", pcmk_strerror(cib_status));
op_reply = create_xml_node(NULL, "cib-reply");
crm_xml_add(op_reply, F_TYPE, T_CIB);
crm_xml_add(op_reply, F_CIB_OPERATION, op);
crm_xml_add(op_reply, F_CIB_CALLID, call);
crm_xml_add(op_reply, F_CIB_CLIENTID, client_id);
crm_xml_add_int(op_reply, F_CIB_CALLOPTS, call_options);
crm_xml_add_int(op_reply, F_CIB_RC, rc);
crm_trace("Attaching reply output");
add_message_xml(op_reply, F_CIB_CALLDATA, the_cib);
crm_log_xml_explicit(op_reply, "cib:reply");
} else if (process) {
time_t finished = 0;
time_t now = time(NULL);
int level = LOG_INFO;
const char *section = crm_element_value(request, F_CIB_SECTION);
rc = cib_process_command(request, &op_reply, &result_diff, privileged);
if (is_update == FALSE) {
level = LOG_TRACE;
} else if (global_update) {
switch (rc) {
case pcmk_ok:
level = LOG_INFO;
break;
case -pcmk_err_old_data:
case -pcmk_err_diff_resync:
case -pcmk_err_diff_failed:
level = LOG_TRACE;
break;
default:
level = LOG_ERR;
}
} else if (rc != pcmk_ok && is_update) {
level = LOG_WARNING;
}
do_crm_log(level,
"Completed %s operation for section %s: %s (rc=%d, origin=%s/%s/%s, version=%s.%s.%s)",
op, section ? section : "'all'", pcmk_strerror(rc), rc,
originator ? originator : "local", client_name, call_id,
the_cib ? crm_element_value(the_cib, XML_ATTR_GENERATION_ADMIN) : "0",
the_cib ? crm_element_value(the_cib, XML_ATTR_GENERATION) : "0",
the_cib ? crm_element_value(the_cib, XML_ATTR_NUMUPDATES) : "0");
finished = time(NULL);
if ((finished - now) > 3) {
crm_trace("%s operation took %lds to complete", op, (long)(finished - now));
crm_write_blackbox(0, NULL);
}
if (op_reply == NULL && (needs_reply || local_notify)) {
crm_err("Unexpected NULL reply to message");
crm_log_xml_err(request, "null reply");
needs_reply = FALSE;
local_notify = FALSE;
}
}
/* from now on we are the server */
if(is_update && cib_legacy_mode() == FALSE) {
crm_trace("Completed pre-sync update from %s/%s/%s%s",
originator ? originator : "local", client_name, call_id,
local_notify?" with local notification":"");
} else if (needs_reply == FALSE || stand_alone) {
/* nothing more to do...
* this was a non-originating slave update
*/
crm_trace("Completed slave update");
} else if (cib_legacy_mode() &&
rc == pcmk_ok && result_diff != NULL && !(call_options & cib_inhibit_bcast)) {
gboolean broadcast = FALSE;
cib_local_bcast_num++;
crm_xml_add_int(request, F_CIB_LOCAL_NOTIFY_ID, cib_local_bcast_num);
broadcast = send_peer_reply(request, result_diff, originator, TRUE);
if (broadcast && client_id && local_notify && op_reply) {
/* If we have been asked to sync the reply,
* and a bcast msg has gone out, we queue the local notify
* until we know the bcast message has been received */
local_notify = FALSE;
crm_trace("Queuing local %ssync notification for %s",
(call_options & cib_sync_call) ? "" : "a-", client_id);
queue_local_notify(op_reply, client_id, (call_options & cib_sync_call), from_peer);
op_reply = NULL; /* the reply is queued, so don't free here */
}
} else if (call_options & cib_discard_reply) {
crm_trace("Caller isn't interested in reply");
} else if (from_peer) {
if (is_update == FALSE || result_diff == NULL) {
crm_trace("Request not broadcast: R/O call");
} else if (call_options & cib_inhibit_bcast) {
crm_trace("Request not broadcast: inhibited");
} else if (rc != pcmk_ok) {
crm_trace("Request not broadcast: call failed: %s", pcmk_strerror(rc));
} else {
crm_trace("Directing reply to %s", originator);
}
send_peer_reply(op_reply, result_diff, originator, FALSE);
}
if (local_notify && client_id) {
crm_trace("Performing local %ssync notification for %s",
(call_options & cib_sync_call) ? "" : "a-", client_id);
if (process == FALSE) {
do_local_notify(request, client_id, call_options & cib_sync_call, from_peer);
} else {
do_local_notify(op_reply, client_id, call_options & cib_sync_call, from_peer);
}
}
free_xml(op_reply);
free_xml(result_diff);
return;
}
static int
cib_process_command(xmlNode * request, xmlNode ** reply, xmlNode ** cib_diff, gboolean privileged)
{
xmlNode *input = NULL;
xmlNode *output = NULL;
xmlNode *result_cib = NULL;
xmlNode *current_cib = NULL;
int call_type = 0;
int call_options = 0;
const char *op = NULL;
const char *section = NULL;
const char *call_id = crm_element_value(request, F_CIB_CALLID);
int rc = pcmk_ok;
int rc2 = pcmk_ok;
gboolean send_r_notify = FALSE;
gboolean global_update = FALSE;
gboolean config_changed = FALSE;
gboolean manage_counters = TRUE;
static mainloop_timer_t *digest_timer = NULL;
CRM_ASSERT(cib_status == pcmk_ok);
if(digest_timer == NULL) {
digest_timer = mainloop_timer_add("digester", 5000, FALSE, cib_digester_cb, NULL);
}
*reply = NULL;
*cib_diff = NULL;
current_cib = the_cib;
/* Start processing the request... */
op = crm_element_value(request, F_CIB_OPERATION);
crm_element_value_int(request, F_CIB_CALLOPTS, &call_options);
rc = cib_get_operation_id(op, &call_type);
if (rc == pcmk_ok && privileged == FALSE) {
rc = cib_op_can_run(call_type, call_options, privileged, global_update);
}
rc2 = cib_op_prepare(call_type, request, &input, &section);
if (rc == pcmk_ok) {
rc = rc2;
}
if (rc != pcmk_ok) {
crm_trace("Call setup failed: %s", pcmk_strerror(rc));
goto done;
} else if (cib_op_modifies(call_type) == FALSE) {
rc = cib_perform_op(op, call_options, cib_op_func(call_type), TRUE,
section, request, input, FALSE, &config_changed,
current_cib, &result_cib, NULL, &output);
CRM_CHECK(result_cib == NULL, free_xml(result_cib));
goto done;
}
/* Handle a valid write action */
global_update = crm_is_true(crm_element_value(request, F_CIB_GLOBAL_UPDATE));
if (global_update) {
/* legacy code */
manage_counters = FALSE;
call_options |= cib_force_diff;
crm_trace("Global update detected");
CRM_CHECK(call_type == 3 || call_type == 4, crm_err("Call type: %d", call_type);
crm_log_xml_err(request, "bad op"));
}
if (rc == pcmk_ok) {
ping_modified_since = TRUE;
if (call_options & cib_inhibit_bcast) {
/* skip */
crm_trace("Skipping update: inhibit broadcast");
manage_counters = FALSE;
}
if (is_not_set(call_options, cib_dryrun) && safe_str_eq(section, XML_CIB_TAG_STATUS)) {
/* Copying large CIBs accounts for a huge percentage of our CIB usage */
call_options |= cib_zero_copy;
} else {
clear_bit(call_options, cib_zero_copy);
}
/* result_cib must not be modified after cib_perform_op() returns */
rc = cib_perform_op(op, call_options, cib_op_func(call_type), FALSE,
section, request, input, manage_counters, &config_changed,
current_cib, &result_cib, cib_diff, &output);
if (manage_counters == FALSE) {
int format = 1;
/* Legacy code
* If the diff is NULL at this point, it's because nothing changed
*/
if (*cib_diff) {
crm_element_value_int(*cib_diff, "format", &format);
}
if (format == 1) {
config_changed = cib_config_changed(NULL, NULL, cib_diff);
}
}
/* Always write to disk for replace ops,
* this also negates the need to detect ordering changes
*/
if (crm_str_eq(CIB_OP_REPLACE, op, TRUE)) {
config_changed = TRUE;
}
}
if (rc == pcmk_ok && is_not_set(call_options, cib_dryrun)) {
crm_trace("Activating %s->%s%s%s",
crm_element_value(current_cib, XML_ATTR_NUMUPDATES),
crm_element_value(result_cib, XML_ATTR_NUMUPDATES),
(is_set(call_options, cib_zero_copy)? " zero-copy" : ""),
(config_changed? " changed" : ""));
if(is_not_set(call_options, cib_zero_copy)) {
rc = activateCibXml(result_cib, config_changed, op);
crm_trace("Activated %s (%d)",
crm_element_value(current_cib, XML_ATTR_NUMUPDATES), rc);
}
if (rc == pcmk_ok && cib_internal_config_changed(*cib_diff)) {
cib_read_config(config_hash, result_cib);
}
if (crm_str_eq(CIB_OP_REPLACE, op, TRUE)) {
if (section == NULL) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_TAG_CIB)) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_CIB_TAG_NODES)) {
send_r_notify = TRUE;
} else if (safe_str_eq(section, XML_CIB_TAG_STATUS)) {
send_r_notify = TRUE;
}
} else if (crm_str_eq(CIB_OP_ERASE, op, TRUE)) {
send_r_notify = TRUE;
}
mainloop_timer_stop(digest_timer);
mainloop_timer_start(digest_timer);
} else if (rc == -pcmk_err_schema_validation) {
CRM_ASSERT(is_not_set(call_options, cib_zero_copy));
if (output != NULL) {
crm_log_xml_info(output, "cib:output");
free_xml(output);
}
output = result_cib;
} else {
crm_trace("Not activating %d %d %s", rc, is_set(call_options, cib_dryrun), crm_element_value(result_cib, XML_ATTR_NUMUPDATES));
if(is_not_set(call_options, cib_zero_copy)) {
free_xml(result_cib);
}
}
if ((call_options & (cib_inhibit_notify|cib_dryrun)) == 0) {
const char *client = crm_element_value(request, F_CIB_CLIENTNAME);
crm_trace("Sending notifications %d", is_set(call_options, cib_dryrun));
cib_diff_notify(call_options, client, call_id, op, input, rc, *cib_diff);
}
if (send_r_notify) {
const char *origin = crm_element_value(request, F_ORIG);
cib_replace_notify(origin, the_cib, rc, *cib_diff);
}
xml_log_patchset(LOG_TRACE, "cib:diff", *cib_diff);
done:
if ((call_options & cib_discard_reply) == 0) {
const char *caller = crm_element_value(request, F_CIB_CLIENTID);
*reply = create_xml_node(NULL, "cib-reply");
crm_xml_add(*reply, F_TYPE, T_CIB);
crm_xml_add(*reply, F_CIB_OPERATION, op);
crm_xml_add(*reply, F_CIB_CALLID, call_id);
crm_xml_add(*reply, F_CIB_CLIENTID, caller);
crm_xml_add_int(*reply, F_CIB_CALLOPTS, call_options);
crm_xml_add_int(*reply, F_CIB_RC, rc);
if (output != NULL) {
crm_trace("Attaching reply output");
add_message_xml(*reply, F_CIB_CALLDATA, output);
}
crm_log_xml_explicit(*reply, "cib:reply");
}
crm_trace("cleanup");
if (cib_op_modifies(call_type) == FALSE && output != current_cib) {
free_xml(output);
output = NULL;
}
if (call_type >= 0) {
cib_op_cleanup(call_type, call_options, &input, &output);
}
crm_trace("done");
return rc;
}
void
cib_peer_callback(xmlNode * msg, void *private_data)
{
const char *reason = NULL;
const char *originator = crm_element_value(msg, F_ORIG);
if (cib_legacy_mode() && (originator == NULL || crm_str_eq(originator, cib_our_uname, TRUE))) {
/* message is from ourselves */
int bcast_id = 0;
if (!(crm_element_value_int(msg, F_CIB_LOCAL_NOTIFY_ID, &bcast_id))) {
check_local_notify(bcast_id);
}
return;
} else if (crm_peer_cache == NULL) {
reason = "membership not established";
goto bail;
}
if (crm_element_value(msg, F_CIB_CLIENTNAME) == NULL) {
crm_xml_add(msg, F_CIB_CLIENTNAME, originator);
}
/* crm_log_xml_trace("Peer[inbound]", msg); */
cib_process_request(msg, FALSE, TRUE, NULL);
return;
bail:
if (reason) {
const char *seq = crm_element_value(msg, F_SEQ);
const char *op = crm_element_value(msg, F_CIB_OPERATION);
crm_warn("Discarding %s message (%s) from %s: %s", op, seq, originator, reason);
}
}
static gboolean
cib_force_exit(gpointer data)
{
crm_notice("Forcing exit!");
terminate_cib(__FUNCTION__, CRM_EX_ERROR);
return FALSE;
}
static void
disconnect_remote_client(gpointer key, gpointer value, gpointer user_data)
{
pcmk__client_t *a_client = value;
crm_err("Disconnecting %s... Not implemented", crm_str(a_client->name));
}
void
cib_shutdown(int nsig)
{
struct qb_ipcs_stats srv_stats;
if (cib_shutdown_flag == FALSE) {
int disconnects = 0;
qb_ipcs_connection_t *c = NULL;
cib_shutdown_flag = TRUE;
c = qb_ipcs_connection_first_get(ipcs_rw);
while (c != NULL) {
qb_ipcs_connection_t *last = c;
c = qb_ipcs_connection_next_get(ipcs_rw, last);
crm_debug("Disconnecting r/w client %p...", last);
qb_ipcs_disconnect(last);
qb_ipcs_connection_unref(last);
disconnects++;
}
c = qb_ipcs_connection_first_get(ipcs_ro);
while (c != NULL) {
qb_ipcs_connection_t *last = c;
c = qb_ipcs_connection_next_get(ipcs_ro, last);
crm_debug("Disconnecting r/o client %p...", last);
qb_ipcs_disconnect(last);
qb_ipcs_connection_unref(last);
disconnects++;
}
c = qb_ipcs_connection_first_get(ipcs_shm);
while (c != NULL) {
qb_ipcs_connection_t *last = c;
c = qb_ipcs_connection_next_get(ipcs_shm, last);
crm_debug("Disconnecting non-blocking r/w client %p...", last);
qb_ipcs_disconnect(last);
qb_ipcs_connection_unref(last);
disconnects++;
}
disconnects += pcmk__ipc_client_count();
crm_debug("Disconnecting %d remote clients", pcmk__ipc_client_count());
pcmk__foreach_ipc_client(disconnect_remote_client, NULL);
crm_info("Disconnected %d clients", disconnects);
}
qb_ipcs_stats_get(ipcs_rw, &srv_stats, QB_FALSE);
if (pcmk__ipc_client_count() == 0) {
crm_info("All clients disconnected (%d)", srv_stats.active_connections);
initiate_exit();
} else {
crm_info("Waiting on %d clients to disconnect (%d)",
pcmk__ipc_client_count(), srv_stats.active_connections);
}
}
void
initiate_exit(void)
{
int active = 0;
xmlNode *leaving = NULL;
active = crm_active_peers();
if (active < 2) {
terminate_cib(__FUNCTION__, 0);
return;
}
crm_info("Sending disconnect notification to %d peers...", active);
leaving = create_xml_node(NULL, "exit-notification");
crm_xml_add(leaving, F_TYPE, "cib");
crm_xml_add(leaving, F_CIB_OPERATION, "cib_shutdown_req");
send_cluster_message(NULL, crm_msg_cib, leaving, TRUE);
free_xml(leaving);
g_timeout_add(EXIT_ESCALATION_MS, cib_force_exit, NULL);
}
extern int remote_fd;
extern int remote_tls_fd;
/*!
* \internal
* \brief Close remote sockets, free the global CIB and quit
*
* \param[in] caller Name of calling function (for log message)
* \param[in] fast If -1, skip disconnect; if positive, exit that
*/
void
terminate_cib(const char *caller, int fast)
{
crm_info("%s: Exiting%s...", caller,
(fast > 0)? " fast" : mainloop ? " from mainloop" : "");
if (remote_fd > 0) {
close(remote_fd);
remote_fd = 0;
}
if (remote_tls_fd > 0) {
close(remote_tls_fd);
remote_tls_fd = 0;
}
uninitializeCib();
if (fast > 0) {
/* Quit fast on error */
cib_ipc_servers_destroy(ipcs_ro, ipcs_rw, ipcs_shm);
crm_exit(fast);
} else if ((mainloop != NULL) && g_main_loop_is_running(mainloop)) {
/* Quit via returning from the main loop. If fast == -1, we skip the
* disconnect here, and it will be done when the main loop returns
* (this allows the peer status callback to avoid messing with the
* peer caches).
*/
if (fast == 0) {
crm_cluster_disconnect(&crm_cluster);
}
g_main_loop_quit(mainloop);
} else {
/* Quit via clean exit. Even the peer status callback can disconnect
* here, because we're not returning control to the caller. */
crm_cluster_disconnect(&crm_cluster);
cib_ipc_servers_destroy(ipcs_ro, ipcs_rw, ipcs_shm);
crm_exit(CRM_EX_OK);
}
}
diff --git a/daemons/controld/controld_control.c b/daemons/controld/controld_control.c
index 83a27deb7d..1a1c8c44bf 100644
--- a/daemons/controld/controld_control.c
+++ b/daemons/controld/controld_control.c
@@ -1,780 +1,774 @@
/*
* Copyright 2004-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/pengine/rules.h>
#include <crm/cluster/internal.h>
#include <crm/cluster/election.h>
#include <crm/common/ipcs_internal.h>
#include <pacemaker-controld.h>
qb_ipcs_service_t *ipcs = NULL;
#if SUPPORT_COROSYNC
extern gboolean crm_connect_corosync(crm_cluster_t * cluster);
#endif
void crm_shutdown(int nsig);
gboolean crm_read_options(gpointer user_data);
gboolean fsa_has_quorum = FALSE;
crm_trigger_t *fsa_source = NULL;
crm_trigger_t *config_read = NULL;
bool no_quorum_suicide_escalation = FALSE;
bool controld_shutdown_lock_enabled = false;
/* A_HA_CONNECT */
void
do_ha_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)
{
gboolean registered = FALSE;
static crm_cluster_t *cluster = NULL;
if (cluster == NULL) {
cluster = calloc(1, sizeof(crm_cluster_t));
}
if (action & A_HA_DISCONNECT) {
crm_cluster_disconnect(cluster);
crm_info("Disconnected from the cluster");
set_bit(fsa_input_register, R_HA_DISCONNECTED);
}
if (action & A_HA_CONNECT) {
crm_set_status_callback(&peer_update_callback);
crm_set_autoreap(FALSE);
if (is_corosync_cluster()) {
#if SUPPORT_COROSYNC
registered = crm_connect_corosync(cluster);
#endif
}
if (registered == TRUE) {
controld_election_init(cluster->uname);
fsa_our_uname = cluster->uname;
fsa_our_uuid = cluster->uuid;
if(cluster->uuid == NULL) {
crm_err("Could not obtain local uuid");
registered = FALSE;
}
}
if (registered == FALSE) {
set_bit(fsa_input_register, R_HA_DISCONNECTED);
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
return;
}
populate_cib_nodes(node_update_none, __FUNCTION__);
clear_bit(fsa_input_register, R_HA_DISCONNECTED);
crm_info("Connected to the cluster");
}
if (action & ~(A_HA_CONNECT | A_HA_DISCONNECT)) {
crm_err("Unexpected action %s in %s", fsa_action2string(action), __FUNCTION__);
}
}
/* A_SHUTDOWN */
void
do_shutdown(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)
{
/* just in case */
set_bit(fsa_input_register, R_SHUTDOWN);
controld_disconnect_fencer(FALSE);
}
/* A_SHUTDOWN_REQ */
void
do_shutdown_req(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)
{
xmlNode *msg = NULL;
set_bit(fsa_input_register, R_SHUTDOWN);
crm_info("Sending shutdown request to all peers (DC is %s)",
(fsa_our_dc? fsa_our_dc : "not set"));
msg = create_request(CRM_OP_SHUTDOWN_REQ, NULL, NULL, CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
/* set_bit(fsa_input_register, R_STAYDOWN); */
if (send_cluster_message(NULL, crm_msg_crmd, msg, TRUE) == FALSE) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
free_xml(msg);
}
extern char *max_generation_from;
extern xmlNode *max_generation_xml;
extern GHashTable *resource_history;
extern GHashTable *voted;
void
crmd_fast_exit(crm_exit_t exit_code)
{
if (is_set(fsa_input_register, R_STAYDOWN)) {
crm_warn("Inhibiting respawn "CRM_XS" remapping exit code %d to %d",
exit_code, CRM_EX_FATAL);
exit_code = CRM_EX_FATAL;
} else if ((exit_code == CRM_EX_OK)
&& is_set(fsa_input_register, R_IN_RECOVERY)) {
crm_err("Could not recover from internal error");
exit_code = CRM_EX_ERROR;
}
crm_exit(exit_code);
}
crm_exit_t
crmd_exit(crm_exit_t exit_code)
{
GListPtr gIter = NULL;
GMainLoop *mloop = crmd_mainloop;
static bool in_progress = FALSE;
if (in_progress && (exit_code == CRM_EX_OK)) {
crm_debug("Exit is already in progress");
return exit_code;
} else if(in_progress) {
crm_notice("Error during shutdown process, exiting now with status %d (%s)",
exit_code, crm_exit_str(exit_code));
crm_write_blackbox(SIGTRAP, NULL);
crmd_fast_exit(exit_code);
}
in_progress = TRUE;
crm_trace("Preparing to exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
/* Suppress secondary errors resulting from us disconnecting everything */
set_bit(fsa_input_register, R_HA_DISCONNECTED);
/* Close all IPC servers and clients to ensure any and all shared memory files are cleaned up */
if(ipcs) {
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs);
ipcs = NULL;
}
controld_close_attrd_ipc();
pe_subsystem_free();
controld_disconnect_fencer(TRUE);
if ((exit_code == CRM_EX_OK) && (crmd_mainloop == NULL)) {
crm_debug("No mainloop detected");
exit_code = CRM_EX_ERROR;
}
/* On an error, just get out.
*
* Otherwise, make the effort to have mainloop exit gracefully so
* that it (mostly) cleans up after itself and valgrind has less
* to report on - allowing real errors stand out
*/
if (exit_code != CRM_EX_OK) {
crm_notice("Forcing immediate exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
crm_write_blackbox(SIGTRAP, NULL);
crmd_fast_exit(exit_code);
}
/* Clean up as much memory as possible for valgrind */
for (gIter = fsa_message_queue; gIter != NULL; gIter = gIter->next) {
fsa_data_t *fsa_data = gIter->data;
crm_info("Dropping %s: [ state=%s cause=%s origin=%s ]",
fsa_input2string(fsa_data->fsa_input),
fsa_state2string(fsa_state),
fsa_cause2string(fsa_data->fsa_cause), fsa_data->origin);
delete_fsa_input(fsa_data);
}
clear_bit(fsa_input_register, R_MEMBERSHIP);
g_list_free(fsa_message_queue); fsa_message_queue = NULL;
metadata_cache_fini();
controld_election_fini();
/* Tear down the CIB manager connection, but don't free it yet -- it could
* be used when we drain the mainloop later.
*/
cib_free_callbacks(fsa_cib_conn);
fsa_cib_conn->cmds->signoff(fsa_cib_conn);
verify_stopped(fsa_state, LOG_WARNING);
clear_bit(fsa_input_register, R_LRM_CONNECTED);
lrm_state_destroy_all();
/* This basically will not work, since mainloop has a reference to it */
mainloop_destroy_trigger(fsa_source); fsa_source = NULL;
mainloop_destroy_trigger(config_read); config_read = NULL;
mainloop_destroy_trigger(transition_trigger); transition_trigger = NULL;
pcmk__client_cleanup();
crm_peer_destroy();
controld_free_fsa_timers();
te_cleanup_stonith_history_sync(NULL, TRUE);
controld_free_sched_timer();
free(fsa_our_dc_version); fsa_our_dc_version = NULL;
free(fsa_our_uname); fsa_our_uname = NULL;
free(fsa_our_uuid); fsa_our_uuid = NULL;
free(fsa_our_dc); fsa_our_dc = NULL;
free(fsa_cluster_name); fsa_cluster_name = NULL;
free(te_uuid); te_uuid = NULL;
free(failed_stop_offset); failed_stop_offset = NULL;
free(failed_start_offset); failed_start_offset = NULL;
free(max_generation_from); max_generation_from = NULL;
free_xml(max_generation_xml); max_generation_xml = NULL;
mainloop_destroy_signal(SIGPIPE);
mainloop_destroy_signal(SIGUSR1);
mainloop_destroy_signal(SIGTERM);
mainloop_destroy_signal(SIGTRAP);
/* leave SIGCHLD engaged as we might still want to drain some service-actions */
if (mloop) {
GMainContext *ctx = g_main_loop_get_context(crmd_mainloop);
/* Don't re-enter this block */
crmd_mainloop = NULL;
/* no signals on final draining anymore */
mainloop_destroy_signal(SIGCHLD);
crm_trace("Draining mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
{
int lpc = 0;
while((g_main_context_pending(ctx) && lpc < 10)) {
lpc++;
crm_trace("Iteration %d", lpc);
g_main_context_dispatch(ctx);
}
}
crm_trace("Closing mainloop %d %d", g_main_loop_is_running(mloop), g_main_context_pending(ctx));
g_main_loop_quit(mloop);
/* Won't do anything yet, since we're inside it now */
g_main_loop_unref(mloop);
} else {
mainloop_destroy_signal(SIGCHLD);
}
cib_delete(fsa_cib_conn);
fsa_cib_conn = NULL;
throttle_fini();
/* Graceful */
crm_trace("Done preparing for exit with status %d (%s)",
exit_code, crm_exit_str(exit_code));
return exit_code;
}
/* A_EXIT_0, A_EXIT_1 */
void
do_exit(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)
{
crm_exit_t exit_code = CRM_EX_OK;
int log_level = LOG_INFO;
const char *exit_type = "gracefully";
if (action & A_EXIT_1) {
log_level = LOG_ERR;
exit_type = "forcefully";
exit_code = CRM_EX_ERROR;
}
verify_stopped(cur_state, LOG_ERR);
do_crm_log(log_level, "Performing %s - %s exiting the controller",
fsa_action2string(action), exit_type);
crm_info("[%s] stopped (%d)", crm_system_name, exit_code);
crmd_exit(exit_code);
}
static void sigpipe_ignore(int nsig) { return; }
/* A_STARTUP */
void
do_startup(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)
{
crm_debug("Registering Signal Handlers");
mainloop_add_signal(SIGTERM, crm_shutdown);
mainloop_add_signal(SIGPIPE, sigpipe_ignore);
fsa_source = mainloop_add_trigger(G_PRIORITY_HIGH, crm_fsa_trigger, NULL);
config_read = mainloop_add_trigger(G_PRIORITY_HIGH, crm_read_options, NULL);
transition_trigger = mainloop_add_trigger(G_PRIORITY_LOW, te_graph_trigger, NULL);
crm_debug("Creating CIB manager and executor objects");
fsa_cib_conn = cib_new();
lrm_state_init_local();
if (controld_init_fsa_timers() == FALSE) {
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
}
}
static int32_t
crmd_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
-static void
-crmd_ipc_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection %p", c);
-}
-
static int32_t
crmd_ipc_dispatch(qb_ipcs_connection_t * c, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
pcmk__client_t *client = pcmk__find_client(c);
xmlNode *msg = pcmk__client_data2xml(client, data, size, &id, &flags);
crm_trace("Invoked: %s", pcmk__client_name(client));
pcmk__ipc_send_ack(client, id, flags, "ack");
if (msg == NULL) {
return 0;
}
#if ENABLE_ACL
CRM_ASSERT(client->user != NULL);
crm_acl_get_set_user(msg, F_CRM_USER, client->user);
#endif
crm_trace("Processing IPC message from %s", pcmk__client_name(client));
crm_log_xml_trace(msg, "controller[inbound]");
crm_xml_add(msg, F_CRM_SYS_FROM, client->id);
if (controld_authorize_ipc_message(msg, client, NULL)) {
route_message(C_IPC_MESSAGE, msg);
}
trigger_fsa(fsa_source);
free_xml(msg);
return 0;
}
static int32_t
crmd_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client) {
crm_trace("Disconnecting %sregistered client %s (%p/%p)",
(client->userdata? "" : "un"), pcmk__client_name(client),
c, client);
free(client->userdata);
pcmk__free_client(client);
trigger_fsa(fsa_source);
}
return 0;
}
static void
crmd_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
crmd_ipc_closed(c);
}
/* A_STOP */
void
do_stop(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)
{
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs); ipcs = NULL;
register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
}
/* A_STARTED */
void
do_started(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 struct qb_ipcs_service_handlers crmd_callbacks = {
.connection_accept = crmd_ipc_accept,
- .connection_created = crmd_ipc_created,
+ .connection_created = NULL,
.msg_process = crmd_ipc_dispatch,
.connection_closed = crmd_ipc_closed,
.connection_destroyed = crmd_ipc_destroy
};
if (cur_state != S_STARTING) {
crm_err("Start cancelled... %s", fsa_state2string(cur_state));
return;
} else if (is_set(fsa_input_register, R_MEMBERSHIP) == FALSE) {
crm_info("Delaying start, no membership data (%.16llx)", R_MEMBERSHIP);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_LRM_CONNECTED) == FALSE) {
crm_info("Delaying start, not connected to executor (%.16llx)", R_LRM_CONNECTED);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_CIB_CONNECTED) == FALSE) {
crm_info("Delaying start, CIB not connected (%.16llx)", R_CIB_CONNECTED);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_READ_CONFIG) == FALSE) {
crm_info("Delaying start, Config not read (%.16llx)", R_READ_CONFIG);
crmd_fsa_stall(TRUE);
return;
} else if (is_set(fsa_input_register, R_PEER_DATA) == FALSE) {
crm_info("Delaying start, No peer data (%.16llx)", R_PEER_DATA);
crmd_fsa_stall(TRUE);
return;
}
crm_debug("Init server comms");
ipcs = crmd_ipc_server_init(&crmd_callbacks);
if (ipcs == NULL) {
crm_err("Failed to create IPC server: shutting down and inhibiting respawn");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
} else {
crm_notice("Pacemaker controller successfully started and accepting connections");
}
controld_trigger_fencer_connect();
clear_bit(fsa_input_register, R_STARTING);
register_fsa_input(msg_data->fsa_cause, I_PENDING, NULL);
}
/* A_RECOVER */
void
do_recover(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)
{
set_bit(fsa_input_register, R_IN_RECOVERY);
crm_warn("Fast-tracking shutdown in response to errors");
register_fsa_input(C_FSA_INTERNAL, I_TERMINATE, NULL);
}
/* *INDENT-OFF* */
static pe_cluster_option crmd_opts[] = {
/* name, old-name, validate, values, default, short description, long description */
{ "dc-version", NULL, "string", NULL, "none", NULL,
"Version of Pacemaker on the cluster's DC.",
"Includes the hash which identifies the exact changeset it was built from. Used for diagnostic purposes."
},
{ "cluster-infrastructure", NULL, "string", NULL, "corosync", NULL,
"The messaging stack on which Pacemaker is currently running.",
"Used for informational and diagnostic purposes." },
{ "cluster-name", NULL, "string", NULL, NULL, NULL,
"An arbitrary name for the cluster",
"This optional value is mostly for users' convenience as desired "
"in administration, but may also be used in Pacemaker configuration "
"rules via the #cluster-name node attribute, and by higher-level tools "
"and resource agents."
},
{ XML_CONFIG_ATTR_DC_DEADTIME, NULL, "time", NULL, "20s", &check_time,
"How long to wait for a response from other nodes during startup.",
"The \"correct\" value will depend on the speed/load of your network and the type of switches used."
},
{ XML_CONFIG_ATTR_RECHECK, NULL, "time",
"Zero disables polling. Positive values are an interval in seconds (unless other SI units are specified. eg. 5min)",
"15min", &check_timer,
"Polling interval for time based changes to options, resource parameters and constraints.",
"The Cluster is primarily event driven, however the configuration can have elements that change based on time."
" To ensure these changes take effect, we can optionally poll the cluster's status for changes."
},
{ "load-threshold", NULL, "percentage", NULL, "80%", &check_utilization,
"The maximum amount of system resources that should be used by nodes in the cluster",
"The cluster will slow down its recovery process when the amount of system resources used"
" (currently CPU) approaches this limit",
},
{ "node-action-limit", NULL, "integer", NULL, "0", &check_number,
"The maximum number of jobs that can be scheduled per node. Defaults to 2x cores"},
{ XML_CONFIG_ATTR_FENCE_REACTION, NULL, "string", NULL, "stop", NULL,
"How a cluster node should react if notified of its own fencing",
"A cluster node may receive notification of its own fencing if fencing "
"is misconfigured, or if fabric fencing is in use that doesn't cut "
"cluster communication. Allowed values are \"stop\" to attempt to "
"immediately stop pacemaker and stay stopped, or \"panic\" to attempt "
"to immediately reboot the local node, falling back to stop on failure."
},
{ XML_CONFIG_ATTR_ELECTION_FAIL, NULL, "time", NULL, "2min", &check_timer,
"*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug."
},
{ XML_CONFIG_ATTR_FORCE_QUIT, NULL, "time", NULL, "20min", &check_timer,
"*** Advanced Use Only ***.", "If need to adjust this value, it probably indicates the presence of a bug."
},
{
"join-integration-timeout", "crmd-integration-timeout",
"time", NULL, "3min", &check_timer,
"*** Advanced Use Only ***",
"If need to adjust this value, it probably indicates the presence of a bug"
},
{
"join-finalization-timeout", "crmd-finalization-timeout",
"time", NULL, "30min", &check_timer,
"*** Advanced Use Only ***",
"If you need to adjust this value, it probably indicates the presence of a bug"
},
{
"transition-delay", "crmd-transition-delay",
"time", NULL, "0s", &check_timer,
"*** Advanced Use Only *** Enabling this option will slow down cluster recovery under all conditions",
"Delay cluster recovery for the configured interval to allow for additional/related events to occur.\n"
"Useful if your configuration is sensitive to the order in which ping updates arrive."
},
{ "stonith-watchdog-timeout", NULL, "time", NULL, NULL, &check_sbd_timeout,
"How long to wait before we can assume nodes are safely down", NULL
},
{ "stonith-max-attempts",NULL,"integer",NULL,"10",&check_positive_number,
"How many times stonith can fail before it will no longer be attempted on a target"
},
// Already documented in libpe_status (other values must be kept identical)
{ "no-quorum-policy", NULL, "enum", "stop, freeze, ignore, suicide", "stop", &check_quorum, NULL, NULL },
{ XML_CONFIG_ATTR_SHUTDOWN_LOCK, NULL, "boolean", NULL, "false", &check_boolean, NULL, NULL },
};
/* *INDENT-ON* */
void
crmd_metadata(void)
{
config_metadata("pacemaker-controld", "1.0",
"controller properties",
"Cluster properties used by Pacemaker's controller,"
" formerly known as crmd",
crmd_opts, DIMOF(crmd_opts));
}
static void
verify_crmd_options(GHashTable * options)
{
verify_all_options(options, crmd_opts, DIMOF(crmd_opts));
}
static const char *
crmd_pref(GHashTable * options, const char *name)
{
return get_cluster_pref(options, crmd_opts, DIMOF(crmd_opts), name);
}
static void
config_query_callback(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
const char *value = NULL;
GHashTable *config_hash = NULL;
crm_time_t *now = crm_time_new(NULL);
xmlNode *crmconfig = NULL;
xmlNode *alerts = NULL;
if (rc != pcmk_ok) {
fsa_data_t *msg_data = NULL;
crm_err("Local CIB query resulted in an error: %s", pcmk_strerror(rc));
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
if (rc == -EACCES || rc == -pcmk_err_schema_validation) {
crm_err("The cluster is mis-configured - shutting down and staying down");
set_bit(fsa_input_register, R_STAYDOWN);
}
goto bail;
}
crmconfig = output;
if ((crmconfig) &&
(crm_element_name(crmconfig)) &&
(strcmp(crm_element_name(crmconfig), XML_CIB_TAG_CRMCONFIG) != 0)) {
crmconfig = first_named_child(crmconfig, XML_CIB_TAG_CRMCONFIG);
}
if (!crmconfig) {
fsa_data_t *msg_data = NULL;
crm_err("Local CIB query for " XML_CIB_TAG_CRMCONFIG " section failed");
register_fsa_error(C_FSA_INTERNAL, I_ERROR, NULL);
goto bail;
}
crm_debug("Call %d : Parsing CIB options", call_id);
config_hash = crm_str_table_new();
pe_unpack_nvpairs(crmconfig, crmconfig, XML_CIB_TAG_PROPSET, NULL,
config_hash, CIB_OPTIONS_FIRST, FALSE, now, NULL);
verify_crmd_options(config_hash);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_DC_DEADTIME);
election_trigger->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "node-action-limit"); /* Also checks migration-limit */
throttle_update_job_max(value);
value = crmd_pref(config_hash, "load-threshold");
if(value) {
throttle_set_load_target(strtof(value, NULL) / 100.0);
}
value = crmd_pref(config_hash, "no-quorum-policy");
if (safe_str_eq(value, "suicide") && pcmk_locate_sbd()) {
no_quorum_suicide_escalation = TRUE;
}
set_fence_reaction(crmd_pref(config_hash, XML_CONFIG_ATTR_FENCE_REACTION));
value = crmd_pref(config_hash,"stonith-max-attempts");
update_stonith_max_attempts(value);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_FORCE_QUIT);
shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
crm_debug("Shutdown escalation occurs if DC has not responded to request in %ums",
shutdown_escalation_timer->period_ms);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_ELECTION_FAIL);
controld_set_election_period(value);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_RECHECK);
recheck_interval_ms = crm_parse_interval_spec(value);
crm_debug("Re-run scheduler after %dms of inactivity", recheck_interval_ms);
value = crmd_pref(config_hash, "transition-delay");
transition_timer->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "join-integration-timeout");
integration_timer->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, "join-finalization-timeout");
finalization_timer->period_ms = crm_parse_interval_spec(value);
value = crmd_pref(config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK);
controld_shutdown_lock_enabled = crm_is_true(value);
free(fsa_cluster_name);
fsa_cluster_name = NULL;
value = g_hash_table_lookup(config_hash, "cluster-name");
if (value) {
fsa_cluster_name = strdup(value);
}
alerts = first_named_child(output, XML_CIB_TAG_ALERTS);
crmd_unpack_alerts(alerts);
set_bit(fsa_input_register, R_READ_CONFIG);
crm_trace("Triggering FSA: %s", __FUNCTION__);
mainloop_set_trigger(fsa_source);
g_hash_table_destroy(config_hash);
bail:
crm_time_free(now);
}
gboolean
crm_read_options(gpointer user_data)
{
int call_id =
fsa_cib_conn->cmds->query(fsa_cib_conn,
"//" XML_CIB_TAG_CRMCONFIG " | //" XML_CIB_TAG_ALERTS,
NULL, cib_xpath | cib_scope_local);
fsa_register_cib_callback(call_id, FALSE, NULL, config_query_callback);
crm_trace("Querying the CIB... call %d", call_id);
return TRUE;
}
/* A_READCONFIG */
void
do_read_config(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)
{
throttle_init();
mainloop_set_trigger(config_read);
}
void
crm_shutdown(int nsig)
{
if (crmd_mainloop != NULL && g_main_loop_is_running(crmd_mainloop)) {
if (is_set(fsa_input_register, R_SHUTDOWN)) {
crm_err("Escalating the shutdown");
register_fsa_input_before(C_SHUTDOWN, I_ERROR, NULL);
} else {
set_bit(fsa_input_register, R_SHUTDOWN);
register_fsa_input(C_SHUTDOWN, I_SHUTDOWN, NULL);
if (shutdown_escalation_timer->period_ms == 0) {
const char *value = crmd_pref(NULL, XML_CONFIG_ATTR_FORCE_QUIT);
shutdown_escalation_timer->period_ms = crm_parse_interval_spec(value);
}
/* can't rely on this... */
crm_notice("Shutting down cluster resource manager " CRM_XS
" limit=%ums", shutdown_escalation_timer->period_ms);
controld_start_timer(shutdown_escalation_timer);
}
} else {
crm_info("exit from shutdown");
crmd_exit(CRM_EX_OK);
}
}
diff --git a/daemons/execd/remoted_proxy.c b/daemons/execd/remoted_proxy.c
index bfe0202999..dda111e242 100644
--- a/daemons/execd/remoted_proxy.c
+++ b/daemons/execd/remoted_proxy.c
@@ -1,456 +1,450 @@
/*
* Copyright 2012-2020 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 <unistd.h>
#include "pacemaker-execd.h"
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/services.h>
#include <crm/common/mainloop.h>
#include <crm/common/ipc.h>
#include <crm/common/ipcs_internal.h>
#include <crm/cib/internal.h>
#include <crm/fencing/internal.h>
static qb_ipcs_service_t *cib_ro = NULL;
static qb_ipcs_service_t *cib_rw = NULL;
static qb_ipcs_service_t *cib_shm = NULL;
static qb_ipcs_service_t *attrd_ipcs = NULL;
static qb_ipcs_service_t *crmd_ipcs = NULL;
static qb_ipcs_service_t *stonith_ipcs = NULL;
// An IPC provider is a cluster node controller connecting as a client
static GList *ipc_providers = NULL;
/* ipc clients == things like cibadmin, crm_resource, connecting locally */
static GHashTable *ipc_clients = NULL;
/*!
* \internal
* \brief Get an IPC proxy provider
*
* \return Pointer to a provider if one exists, NULL otherwise
*
* \note Grab the first provider, which is the most recent connection. That way,
* if we haven't yet timed out an old, failed connection, we don't try to
* use it.
*/
pcmk__client_t *
ipc_proxy_get_provider()
{
return ipc_providers? (pcmk__client_t *) (ipc_providers->data) : NULL;
}
/*!
* \internal
* \brief Accept a client connection on a proxy IPC server
*
* \param[in] c Client's IPC connection
* \param[in] uid Client's user ID
* \param[in] gid Client's group ID
* \param[in] ipc_channel Name of IPC server to proxy
*
* \return pcmk_ok on success, -errno on error
*/
static int32_t
ipc_proxy_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid, const char *ipc_channel)
{
pcmk__client_t *client;
pcmk__client_t *ipc_proxy = ipc_proxy_get_provider();
xmlNode *msg;
if (ipc_proxy == NULL) {
crm_warn("Cannot proxy IPC connection from uid %d gid %d to %s "
"because not connected to cluster", uid, gid, ipc_channel);
return -EREMOTEIO;
}
/* This new client is a local IPC client on a Pacemaker Remote controlled
* node, needing to access cluster node IPC services.
*/
client = pcmk__new_client(c, uid, gid);
if (client == NULL) {
return -EREMOTEIO;
}
/* This ipc client is bound to a single ipc provider. If the
* provider goes away, this client is disconnected */
client->userdata = strdup(ipc_proxy->id);
client->name = crm_strdup_printf("proxy-%s-%d-%.8s", ipc_channel, client->pid, client->id);
g_hash_table_insert(ipc_clients, client->id, client);
msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
crm_xml_add(msg, F_LRMD_IPC_OP, LRMD_IPC_OP_NEW);
crm_xml_add(msg, F_LRMD_IPC_IPC_SERVER, ipc_channel);
crm_xml_add(msg, F_LRMD_IPC_SESSION, client->id);
lrmd_server_send_notify(ipc_proxy, msg);
free_xml(msg);
crm_debug("Accepted IPC proxy connection (session ID %s) "
"from uid %d gid %d on channel %s",
client->id, uid, gid, ipc_channel);
return 0;
}
static int32_t
crmd_proxy_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
return ipc_proxy_accept(c, uid, gid, CRM_SYSTEM_CRMD);
}
static int32_t
attrd_proxy_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
return ipc_proxy_accept(c, uid, gid, T_ATTRD);
}
static int32_t
stonith_proxy_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
return ipc_proxy_accept(c, uid, gid, "stonith-ng");
}
static int32_t
cib_proxy_accept_rw(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
return ipc_proxy_accept(c, uid, gid, CIB_CHANNEL_RW);
}
static int32_t
cib_proxy_accept_ro(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
return ipc_proxy_accept(c, uid, gid, CIB_CHANNEL_RO);
}
-static void
-ipc_proxy_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection %p", c);
-}
-
void
ipc_proxy_forward_client(pcmk__client_t *ipc_proxy, xmlNode *xml)
{
const char *session = crm_element_value(xml, F_LRMD_IPC_SESSION);
const char *msg_type = crm_element_value(xml, F_LRMD_IPC_OP);
xmlNode *msg = get_message_xml(xml, F_LRMD_IPC_MSG);
pcmk__client_t *ipc_client;
int rc = pcmk_rc_ok;
/* If the IPC provider is acknowledging our shutdown request,
* defuse the short exit timer to give the cluster time to
* stop any resources we're running.
*/
if (safe_str_eq(msg_type, LRMD_IPC_OP_SHUTDOWN_ACK)) {
handle_shutdown_ack();
return;
}
if (safe_str_eq(msg_type, LRMD_IPC_OP_SHUTDOWN_NACK)) {
handle_shutdown_nack();
return;
}
ipc_client = pcmk__find_client_by_id(session);
if (ipc_client == NULL) {
xmlNode *msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
crm_xml_add(msg, F_LRMD_IPC_OP, LRMD_IPC_OP_DESTROY);
crm_xml_add(msg, F_LRMD_IPC_SESSION, session);
lrmd_server_send_notify(ipc_proxy, msg);
free_xml(msg);
return;
}
/* This is an event or response from the ipc provider
* going to the local ipc client.
*
* Looking at the chain of events.
*
* -----remote node----------------|---- cluster node ------
* ipc_client <--1--> this code
* <--2--> pacemaker-controld:remote_proxy_cb/remote_proxy_relay_event()
* <--3--> ipc server
*
* This function is receiving a msg from connection 2
* and forwarding it to connection 1.
*/
if (safe_str_eq(msg_type, LRMD_IPC_OP_EVENT)) {
crm_trace("Sending event to %s", ipc_client->id);
rc = pcmk__ipc_send_xml(ipc_client, 0, msg, crm_ipc_server_event);
} else if (safe_str_eq(msg_type, LRMD_IPC_OP_RESPONSE)) {
int msg_id = 0;
crm_element_value_int(xml, F_LRMD_IPC_MSG_ID, &msg_id);
crm_trace("Sending response to %d - %s", ipc_client->request_id, ipc_client->id);
rc = pcmk__ipc_send_xml(ipc_client, msg_id, msg, FALSE);
CRM_LOG_ASSERT(msg_id == ipc_client->request_id);
ipc_client->request_id = 0;
} else if (safe_str_eq(msg_type, LRMD_IPC_OP_DESTROY)) {
qb_ipcs_disconnect(ipc_client->ipcs);
} else {
crm_err("Unknown ipc proxy msg type %s" , msg_type);
}
if (rc != pcmk_rc_ok) {
crm_warn("Could not proxy IPC to client %s: %s " CRM_XS " rc=%d",
ipc_client->id, pcmk_rc_str(rc), rc);
}
}
static int32_t
ipc_proxy_dispatch(qb_ipcs_connection_t * c, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
pcmk__client_t *client = pcmk__find_client(c);
pcmk__client_t *ipc_proxy = pcmk__find_client_by_id(client->userdata);
xmlNode *request = NULL;
xmlNode *msg = NULL;
if (!ipc_proxy) {
qb_ipcs_disconnect(client->ipcs);
return 0;
}
/* This is a request from the local ipc client going
* to the ipc provider.
*
* Looking at the chain of events.
*
* -----remote node----------------|---- cluster node ------
* ipc_client <--1--> this code
* <--2--> pacemaker-controld:remote_proxy_dispatch_internal()
* <--3--> ipc server
*
* This function is receiving a request from connection
* 1 and forwarding it to connection 2.
*/
request = pcmk__client_data2xml(client, data, size, &id, &flags);
if (!request) {
return 0;
}
CRM_CHECK(client != NULL, crm_err("Invalid client");
free_xml(request); return FALSE);
CRM_CHECK(client->id != NULL, crm_err("Invalid client: %p", client);
free_xml(request); return FALSE);
/* This ensures that synced request/responses happen over the event channel
* in the controller, allowing the controller to process the messages async.
*/
set_bit(flags, crm_ipc_proxied);
client->request_id = id;
msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
crm_xml_add(msg, F_LRMD_IPC_OP, LRMD_IPC_OP_REQUEST);
crm_xml_add(msg, F_LRMD_IPC_SESSION, client->id);
crm_xml_add(msg, F_LRMD_IPC_CLIENT, pcmk__client_name(client));
crm_xml_add(msg, F_LRMD_IPC_USER, client->user);
crm_xml_add_int(msg, F_LRMD_IPC_MSG_ID, id);
crm_xml_add_int(msg, F_LRMD_IPC_MSG_FLAGS, flags);
add_message_xml(msg, F_LRMD_IPC_MSG, request);
lrmd_server_send_notify(ipc_proxy, msg);
free_xml(request);
free_xml(msg);
return 0;
}
/*!
* \internal
* \brief Notify a proxy provider that we wish to shut down
*
* \return 0 on success, -1 on error
*/
int
ipc_proxy_shutdown_req(pcmk__client_t *ipc_proxy)
{
xmlNode *msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
int rc;
crm_xml_add(msg, F_LRMD_IPC_OP, LRMD_IPC_OP_SHUTDOWN_REQ);
/* We don't really have a session, but the controller needs this attribute
* to recognize this as proxy communication.
*/
crm_xml_add(msg, F_LRMD_IPC_SESSION, "0");
rc = (lrmd_server_send_notify(ipc_proxy, msg) != pcmk_rc_ok)? -1 : 0;
free_xml(msg);
return rc;
}
static int32_t
ipc_proxy_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
pcmk__client_t *ipc_proxy;
if (client == NULL) {
return 0;
}
ipc_proxy = pcmk__find_client_by_id(client->userdata);
crm_trace("Connection %p", c);
if (ipc_proxy) {
xmlNode *msg = create_xml_node(NULL, T_LRMD_IPC_PROXY);
crm_xml_add(msg, F_LRMD_IPC_OP, LRMD_IPC_OP_DESTROY);
crm_xml_add(msg, F_LRMD_IPC_SESSION, client->id);
lrmd_server_send_notify(ipc_proxy, msg);
free_xml(msg);
}
g_hash_table_remove(ipc_clients, client->id);
free(client->userdata);
client->userdata = NULL;
pcmk__free_client(client);
return 0;
}
static void
ipc_proxy_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
ipc_proxy_closed(c);
}
static struct qb_ipcs_service_handlers crmd_proxy_callbacks = {
.connection_accept = crmd_proxy_accept,
- .connection_created = ipc_proxy_created,
+ .connection_created = NULL,
.msg_process = ipc_proxy_dispatch,
.connection_closed = ipc_proxy_closed,
.connection_destroyed = ipc_proxy_destroy
};
static struct qb_ipcs_service_handlers attrd_proxy_callbacks = {
.connection_accept = attrd_proxy_accept,
- .connection_created = ipc_proxy_created,
+ .connection_created = NULL,
.msg_process = ipc_proxy_dispatch,
.connection_closed = ipc_proxy_closed,
.connection_destroyed = ipc_proxy_destroy
};
static struct qb_ipcs_service_handlers stonith_proxy_callbacks = {
.connection_accept = stonith_proxy_accept,
- .connection_created = ipc_proxy_created,
+ .connection_created = NULL,
.msg_process = ipc_proxy_dispatch,
.connection_closed = ipc_proxy_closed,
.connection_destroyed = ipc_proxy_destroy
};
static struct qb_ipcs_service_handlers cib_proxy_callbacks_ro = {
.connection_accept = cib_proxy_accept_ro,
- .connection_created = ipc_proxy_created,
+ .connection_created = NULL,
.msg_process = ipc_proxy_dispatch,
.connection_closed = ipc_proxy_closed,
.connection_destroyed = ipc_proxy_destroy
};
static struct qb_ipcs_service_handlers cib_proxy_callbacks_rw = {
.connection_accept = cib_proxy_accept_rw,
- .connection_created = ipc_proxy_created,
+ .connection_created = NULL,
.msg_process = ipc_proxy_dispatch,
.connection_closed = ipc_proxy_closed,
.connection_destroyed = ipc_proxy_destroy
};
void
ipc_proxy_add_provider(pcmk__client_t *ipc_proxy)
{
// Prepending ensures the most recent connection is always first
ipc_providers = g_list_prepend(ipc_providers, ipc_proxy);
}
void
ipc_proxy_remove_provider(pcmk__client_t *ipc_proxy)
{
GHashTableIter iter;
pcmk__client_t *ipc_client = NULL;
char *key = NULL;
GList *remove_these = NULL;
GListPtr gIter = NULL;
ipc_providers = g_list_remove(ipc_providers, ipc_proxy);
g_hash_table_iter_init(&iter, ipc_clients);
while (g_hash_table_iter_next(&iter, (gpointer *) & key, (gpointer *) & ipc_client)) {
const char *proxy_id = ipc_client->userdata;
if (safe_str_eq(proxy_id, ipc_proxy->id)) {
crm_info("ipc proxy connection for client %s pid %d destroyed because cluster node disconnected.",
ipc_client->id, ipc_client->pid);
/* we can't remove during the iteration, so copy items
* to a list we can destroy later */
remove_these = g_list_append(remove_these, ipc_client);
}
}
for (gIter = remove_these; gIter != NULL; gIter = gIter->next) {
ipc_client = gIter->data;
// Disconnection callback will free the client here
qb_ipcs_disconnect(ipc_client->ipcs);
}
/* just frees the list, not the elements in the list */
g_list_free(remove_these);
}
void
ipc_proxy_init(void)
{
ipc_clients = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, NULL);
cib_ipc_servers_init(&cib_ro,
&cib_rw,
&cib_shm,
&cib_proxy_callbacks_ro,
&cib_proxy_callbacks_rw);
attrd_ipc_server_init(&attrd_ipcs, &attrd_proxy_callbacks);
stonith_ipc_server_init(&stonith_ipcs, &stonith_proxy_callbacks);
crmd_ipcs = crmd_ipc_server_init(&crmd_proxy_callbacks);
if (crmd_ipcs == NULL) {
crm_err("Failed to create controller: exiting and inhibiting respawn");
crm_warn("Verify pacemaker and pacemaker_remote are not both enabled");
crm_exit(CRM_EX_FATAL);
}
}
void
ipc_proxy_cleanup(void)
{
if (ipc_providers) {
g_list_free(ipc_providers);
ipc_providers = NULL;
}
if (ipc_clients) {
g_hash_table_destroy(ipc_clients);
ipc_clients = NULL;
}
cib_ipc_servers_destroy(cib_ro, cib_rw, cib_shm);
qb_ipcs_destroy(attrd_ipcs);
qb_ipcs_destroy(stonith_ipcs);
qb_ipcs_destroy(crmd_ipcs);
cib_ro = NULL;
cib_rw = NULL;
cib_shm = NULL;
}
diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c
index bff2efd082..5274f599a6 100644
--- a/daemons/fenced/pacemaker-fenced.c
+++ b/daemons/fenced/pacemaker-fenced.c
@@ -1,1529 +1,1523 @@
/*
* Copyright 2009-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* This source code is licensed under the GNU General Public License version 2
* or later (GPLv2+) WITHOUT ANY WARRANTY.
*/
#include <crm_internal.h>
#include <sys/param.h>
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <sys/utsname.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h> /* U32T ~ PRIu32, X32T ~ PRIx32 */
#include <crm/crm.h>
#include <crm/msg_xml.h>
#include <crm/common/ipc.h>
#include <crm/common/ipcs_internal.h>
#include <crm/cluster/internal.h>
#include <crm/stonith-ng.h>
#include <crm/fencing/internal.h>
#include <crm/common/xml.h>
#include <crm/common/mainloop.h>
#include <crm/cib/internal.h>
#include <crm/pengine/status.h>
#include <pacemaker-internal.h>
#include <pacemaker-fenced.h>
char *stonith_our_uname = NULL;
char *stonith_our_uuid = NULL;
long stonith_watchdog_timeout_ms = 0;
static GMainLoop *mainloop = NULL;
gboolean stand_alone = FALSE;
static gboolean no_cib_connect = FALSE;
static gboolean stonith_shutdown_flag = FALSE;
static qb_ipcs_service_t *ipcs = NULL;
static xmlNode *local_cib = NULL;
static pe_working_set_t *fenced_data_set = NULL;
static cib_t *cib_api = NULL;
static void *cib_library = NULL;
static void stonith_shutdown(int nsig);
static void stonith_cleanup(void);
static int32_t
st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
if (stonith_shutdown_flag) {
crm_info("Ignoring new client [%d] during shutdown",
pcmk__client_pid(c));
return -EPERM;
}
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
-static void
-st_ipc_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection created for %p", c);
-}
-
/* Exit code means? */
static int32_t
st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
int call_options = 0;
xmlNode *request = NULL;
pcmk__client_t *c = pcmk__find_client(qbc);
const char *op = NULL;
if (c == NULL) {
crm_info("Invalid client: %p", qbc);
return 0;
}
request = pcmk__client_data2xml(c, data, size, &id, &flags);
if (request == NULL) {
pcmk__ipc_send_ack(c, id, flags, "nack");
return 0;
}
op = crm_element_value(request, F_CRM_TASK);
if(safe_str_eq(op, CRM_OP_RM_NODE_CACHE)) {
crm_xml_add(request, F_TYPE, T_STONITH_NG);
crm_xml_add(request, F_STONITH_OPERATION, op);
crm_xml_add(request, F_STONITH_CLIENTID, c->id);
crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c));
crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname);
send_cluster_message(NULL, crm_msg_stonith_ng, request, FALSE);
free_xml(request);
return 0;
}
if (c->name == NULL) {
const char *value = crm_element_value(request, F_STONITH_CLIENTNAME);
if (value == NULL) {
value = "unknown";
}
c->name = crm_strdup_printf("%s.%u", value, c->pid);
}
crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options);
crm_trace("Flags %" X32T "/%u for command %" U32T " from %s",
flags, call_options, id, pcmk__client_name(c));
if (is_set(call_options, st_opt_sync_call)) {
CRM_ASSERT(flags & crm_ipc_client_response);
CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */
c->request_id = id; /* Reply only to the last one */
}
crm_xml_add(request, F_STONITH_CLIENTID, c->id);
crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c));
crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname);
crm_log_xml_trace(request, "Client[inbound]");
stonith_command(c, id, flags, request, NULL);
free_xml(request);
return 0;
}
/* Error code means? */
static int32_t
st_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p closed", c);
pcmk__free_client(client);
/* 0 means: yes, go ahead and destroy the connection */
return 0;
}
static void
st_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p destroyed", c);
st_ipc_closed(c);
}
static void
stonith_peer_callback(xmlNode * msg, void *private_data)
{
const char *remote_peer = crm_element_value(msg, F_ORIG);
const char *op = crm_element_value(msg, F_STONITH_OPERATION);
if (crm_str_eq(op, "poke", TRUE)) {
return;
}
crm_log_xml_trace(msg, "Peer[inbound]");
stonith_command(NULL, 0, 0, msg, remote_peer);
}
#if SUPPORT_COROSYNC
static void
stonith_peer_ais_callback(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
uint32_t kind = 0;
xmlNode *xml = NULL;
const char *from = NULL;
char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from);
if(data == NULL) {
return;
}
if (kind == crm_class_cluster) {
xml = string2xml(data);
if (xml == NULL) {
crm_err("Invalid XML: '%.120s'", data);
free(data);
return;
}
crm_xml_add(xml, F_ORIG, from);
/* crm_xml_add_int(xml, F_SEQ, wrapper->id); */
stonith_peer_callback(xml, NULL);
}
free_xml(xml);
free(data);
return;
}
static void
stonith_peer_cs_destroy(gpointer user_data)
{
crm_crit("Lost connection to cluster layer, shutting down");
stonith_shutdown(0);
}
#endif
void
do_local_reply(xmlNode * notify_src, const char *client_id, gboolean sync_reply, gboolean from_peer)
{
/* send callback to originating child */
pcmk__client_t *client_obj = NULL;
int local_rc = pcmk_rc_ok;
crm_trace("Sending response");
client_obj = pcmk__find_client_by_id(client_id);
crm_trace("Sending callback to request originator");
if (client_obj == NULL) {
local_rc = EPROTO;
crm_trace("No client to sent the response to. F_STONITH_CLIENTID not set.");
} else {
int rid = 0;
if (sync_reply) {
CRM_LOG_ASSERT(client_obj->request_id);
rid = client_obj->request_id;
client_obj->request_id = 0;
crm_trace("Sending response %d to %s %s",
rid, client_obj->name, from_peer ? "(originator of delegated request)" : "");
} else {
crm_trace("Sending an event to %s %s",
client_obj->name, from_peer ? "(originator of delegated request)" : "");
}
local_rc = pcmk__ipc_send_xml(client_obj, rid, notify_src,
(sync_reply? crm_ipc_flags_none
: crm_ipc_server_event));
}
if ((local_rc != pcmk_rc_ok) && (client_obj != NULL)) {
crm_warn("%s reply to %s failed: %s",
(sync_reply? "Synchronous" : "Asynchronous"),
(client_obj? client_obj->name : "unknown client"),
pcmk_rc_str(local_rc));
}
}
long long
get_stonith_flag(const char *name)
{
if (safe_str_eq(name, T_STONITH_NOTIFY_FENCE)) {
return st_callback_notify_fence;
} else if (safe_str_eq(name, STONITH_OP_DEVICE_ADD)) {
return st_callback_device_add;
} else if (safe_str_eq(name, STONITH_OP_DEVICE_DEL)) {
return st_callback_device_del;
} else if (safe_str_eq(name, T_STONITH_NOTIFY_HISTORY)) {
return st_callback_notify_history;
} else if (safe_str_eq(name, T_STONITH_NOTIFY_HISTORY_SYNCED)) {
return st_callback_notify_history_synced;
}
return st_callback_unknown;
}
static void
stonith_notify_client(gpointer key, gpointer value, gpointer user_data)
{
xmlNode *update_msg = user_data;
pcmk__client_t *client = value;
const char *type = NULL;
CRM_CHECK(client != NULL, return);
CRM_CHECK(update_msg != NULL, return);
type = crm_element_value(update_msg, F_SUBTYPE);
CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return);
if (client->ipcs == NULL) {
crm_trace("Skipping client with NULL channel");
return;
}
if (client->options & get_stonith_flag(type)) {
int rc = pcmk__ipc_send_xml(client, 0, update_msg,
crm_ipc_server_event|crm_ipc_server_error);
if (rc != pcmk_rc_ok) {
crm_warn("%s notification of client %s failed: %s "
CRM_XS " id=%.8s rc=%d", type, pcmk__client_name(client),
pcmk_rc_str(rc), client->id, rc);
} else {
crm_trace("Sent %s notification to client %s.%.6s", type,
pcmk__client_name(client), client->id);
}
}
}
void
do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout)
{
pcmk__client_t *client = NULL;
xmlNode *notify_data = NULL;
if (!timeout || !call_id || !client_id) {
return;
}
client = pcmk__find_client_by_id(client_id);
if (!client) {
return;
}
notify_data = create_xml_node(NULL, T_STONITH_TIMEOUT_VALUE);
crm_xml_add(notify_data, F_TYPE, T_STONITH_TIMEOUT_VALUE);
crm_xml_add(notify_data, F_STONITH_CALLID, call_id);
crm_xml_add_int(notify_data, F_STONITH_TIMEOUT, timeout);
crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id);
if (client) {
pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event);
}
free_xml(notify_data);
}
void
do_stonith_notify(int options, const char *type, int result, xmlNode * data)
{
/* TODO: Standardize the contents of data */
xmlNode *update_msg = create_xml_node(NULL, "notify");
CRM_CHECK(type != NULL,;);
crm_xml_add(update_msg, F_TYPE, T_STONITH_NOTIFY);
crm_xml_add(update_msg, F_SUBTYPE, type);
crm_xml_add(update_msg, F_STONITH_OPERATION, type);
crm_xml_add_int(update_msg, F_STONITH_RC, result);
if (data != NULL) {
add_message_xml(update_msg, F_STONITH_CALLDATA, data);
}
crm_trace("Notifying clients");
pcmk__foreach_ipc_client(stonith_notify_client, update_msg);
free_xml(update_msg);
crm_trace("Notify complete");
}
static void
do_stonith_notify_config(int options, const char *op, int rc,
const char *desc, int active)
{
xmlNode *notify_data = create_xml_node(NULL, op);
CRM_CHECK(notify_data != NULL, return);
crm_xml_add(notify_data, F_STONITH_DEVICE, desc);
crm_xml_add_int(notify_data, F_STONITH_ACTIVE, active);
do_stonith_notify(options, op, rc, notify_data);
free_xml(notify_data);
}
void
do_stonith_notify_device(int options, const char *op, int rc, const char *desc)
{
do_stonith_notify_config(options, op, rc, desc, g_hash_table_size(device_list));
}
void
do_stonith_notify_level(int options, const char *op, int rc, const char *desc)
{
do_stonith_notify_config(options, op, rc, desc, g_hash_table_size(topology));
}
static void
topology_remove_helper(const char *node, int level)
{
int rc;
char *desc = NULL;
xmlNode *data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL);
crm_xml_add(data, F_STONITH_ORIGIN, __FUNCTION__);
crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level);
crm_xml_add(data, XML_ATTR_STONITH_TARGET, node);
rc = stonith_level_remove(data, &desc);
do_stonith_notify_level(0, STONITH_OP_LEVEL_DEL, rc, desc);
free_xml(data);
free(desc);
}
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 (safe_str_neq(standard, PCMK_RESOURCE_CLASS_STONITH)) {
continue;
}
rsc_id = crm_element_value(match, XML_ATTR_ID);
stonith_device_remove(rsc_id, TRUE);
}
}
static void
handle_topology_change(xmlNode *match, bool remove)
{
int rc;
char *desc = NULL;
CRM_CHECK(match != NULL, return);
crm_trace("Updating %s", ID(match));
if(remove) {
int index = 0;
char *key = stonith_level_key(match, -1);
crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index);
topology_remove_helper(key, index);
free(key);
}
rc = stonith_level_register(match, &desc);
do_stonith_notify_level(0, STONITH_OP_LEVEL_ADD, rc, desc);
free(desc);
}
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, -1);
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
register_fencing_topology(xmlXPathObjectPtr xpathObj)
{
int max = numXpathResults(xpathObj), lpc = 0;
for (lpc = 0; lpc < max; lpc++) {
xmlNode *match = getXpathResult(xpathObj, lpc);
handle_topology_change(match, TRUE);
}
}
/* 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>
*/
static void
fencing_topology_init()
{
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);
}
#define rsc_name(x) x->clone_name?x->clone_name:x->id
/*!
* \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 node_t *
our_node_allowed_for(resource_t *rsc)
{
GHashTableIter iter;
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] rsc Resource to check
* \param[in] data_set Cluster working set with device information
*/
static void cib_device_update(resource_t *rsc, pe_working_set_t *data_set)
{
node_t *node = NULL;
const char *value = NULL;
const char *rclass = NULL;
node_t *parent = NULL;
gboolean remove = TRUE;
/* If this is a complex resource, check children rather than this resource itself.
* TODO: Mark each installed device and remove if untouched when this process finishes.
*/
if(rsc->children) {
GListPtr 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 (safe_str_neq(rclass, PCMK_RESOURCE_CLASS_STONITH)) {
return;
}
/* If this STONITH resource is disabled, just remove it. */
if (pe__resource_is_disabled(rsc)) {
crm_info("Device %s has been disabled", rsc->id);
goto update_done;
}
/* 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 == pe_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", node->details->uname, node->weight);
}
goto update_done;
} else if(node->weight < 0 || (parent && parent->weight < 0)) {
/* Our node (or its group) is disallowed by score, so remove the device */
char *score = score2char((node->weight < 0) ? node->weight : parent->weight);
crm_info("Device %s has been disabled on %s: score=%s", rsc->id, stonith_our_uname, score);
free(score);
goto update_done;
} else {
/* Our node is allowed, so update the device information */
int rc;
xmlNode *data;
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);
get_rsc_attributes(rsc->parameters, rsc, node, data_set);
get_meta_attributes(rsc->meta, rsc, node, data_set);
rsc_provides = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_PROVIDES);
g_hash_table_iter_init(&gIter, rsc->parameters);
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);
}
remove = FALSE;
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, NULL, TRUE);
CRM_ASSERT(rc == pcmk_ok);
free_xml(data);
}
update_done:
if(remove && g_hash_table_lookup(device_list, rsc_name(rsc))) {
stonith_device_remove(rsc_name(rsc), TRUE);
}
}
/*!
* \internal
* \brief Update all STONITH device definitions based on current CIB
*/
static void
cib_devices_update(void)
{
GListPtr gIter = 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));
CRM_ASSERT(fenced_data_set != NULL);
fenced_data_set->input = local_cib;
fenced_data_set->now = crm_time_new(NULL);
fenced_data_set->flags |= pe_flag_quick_location;
fenced_data_set->localhost = stonith_our_uname;
cluster_status(fenced_data_set);
pcmk__schedule_actions(fenced_data_set, NULL, NULL);
for (gIter = fenced_data_set->resources; gIter != NULL; gIter = gIter->next) {
cib_device_update(gIter->data, fenced_data_set);
}
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_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 = __xml_first_child(patchset); change != NULL; change = __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 (safe_str_eq(op, "delete") && 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;
reason = strdup("(meta) attribute deleted from resource");
break;
}
mutable = strdup(xpath);
rsc_id = strstr(mutable, "primitive[@id=\'");
if (rsc_id != NULL) {
rsc_id += strlen("primitive[@id=\'");
search = strchr(rsc_id, '\'');
}
if (search != NULL) {
*search = 0;
stonith_device_remove(rsc_id, TRUE);
} 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 the 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_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 (safe_str_neq(standard, PCMK_RESOURCE_CLASS_STONITH)) {
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 the cib: %s", reason);
cib_devices_update();
}
}
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, "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);
}
}
/* Needs to hold node name + attribute name + attribute value + 75 */
#define XPATH_MAX 512
/*!
* \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)
{
char xpath[XPATH_MAX];
xmlNode *match;
int n;
CRM_CHECK(local_cib != 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.
*/
n = snprintf(xpath, XPATH_MAX, "//" XML_CIB_TAG_NODES
"/" XML_CIB_TAG_NODE "[@uname='%s']/" XML_TAG_ATTR_SETS
"/" XML_CIB_TAG_NVPAIR "[@name='%s' and @value='%s']",
node, name, value);
match = get_xpath_object(xpath, local_cib, LOG_NEVER);
CRM_CHECK(n < XPATH_MAX, return FALSE);
return (match != NULL);
}
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, "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 = __xml_first_child(patchset); change != NULL; change = __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) {
handle_topology_change(change->children, FALSE);
} else if(strcmp(op, "modify") == 0) {
xmlNode *match = first_named_child(change, XML_DIFF_RESULT);
if(match) {
handle_topology_change(match->children, TRUE);
}
} 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 bool have_cib_devices = FALSE;
static void
update_cib_cache_cb(const char *event, xmlNode * msg)
{
int rc = pcmk_ok;
xmlNode *stonith_enabled_xml = NULL;
xmlNode *stonith_watchdog_xml = NULL;
const char *stonith_enabled_s = NULL;
static gboolean stonith_enabled_saved = TRUE;
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);
xml_log_patchset(LOG_TRACE, "Config update", patchset);
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) {
crm_trace("Re-requesting the full cib");
rc = cib_api->cmds->query(cib_api, NULL, &local_cib, cib_scope_local | cib_sync_call);
if(rc != pcmk_ok) {
crm_err("Couldn't retrieve the CIB: %s (%d)", pcmk_strerror(rc), rc);
return;
}
CRM_ASSERT(local_cib != NULL);
stonith_enabled_saved = FALSE; /* Trigger a full refresh below */
}
crm_peer_caches_refresh(local_cib);
stonith_enabled_xml = get_xpath_object("//nvpair[@name='stonith-enabled']",
local_cib, LOG_NEVER);
if (stonith_enabled_xml) {
stonith_enabled_s = crm_element_value(stonith_enabled_xml, XML_NVPAIR_ATTR_VALUE);
}
if (stonith_enabled_s == NULL || crm_is_true(stonith_enabled_s)) {
long timeout_ms = 0;
const char *value = NULL;
stonith_watchdog_xml = get_xpath_object("//nvpair[@name='stonith-watchdog-timeout']",
local_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 = crm_auto_watchdog_timeout();
}
if(timeout_ms != stonith_watchdog_timeout_ms) {
crm_notice("New watchdog timeout %lds (was %lds)", timeout_ms/1000, stonith_watchdog_timeout_ms/1000);
stonith_watchdog_timeout_ms = timeout_ms;
}
} else {
stonith_watchdog_timeout_ms = 0;
}
if (stonith_enabled_s && crm_is_true(stonith_enabled_s) == FALSE) {
crm_trace("Ignoring cib updates while stonith is disabled");
stonith_enabled_saved = FALSE;
return;
} else if (stonith_enabled_saved == FALSE) {
crm_info("Updating stonith device and topology lists now that stonith is enabled");
stonith_enabled_saved = TRUE;
fencing_topology_init();
cib_devices_update();
} else {
update_fencing_topology(event, msg);
update_cib_stonith_devices(event, msg);
}
}
static void
init_cib_cache_cb(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data)
{
crm_info("Updating device list from the cib: init");
have_cib_devices = TRUE;
local_cib = copy_xml(output);
crm_peer_caches_refresh(local_cib);
fencing_topology_init();
cib_devices_update();
}
static void
stonith_shutdown(int nsig)
{
crm_info("Terminating with %d clients", pcmk__ipc_client_count());
stonith_shutdown_flag = TRUE;
if (mainloop != NULL && g_main_loop_is_running(mainloop)) {
g_main_loop_quit(mainloop);
} else {
stonith_cleanup();
crm_exit(CRM_EX_OK);
}
}
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);
}
static void
stonith_cleanup(void)
{
if (cib_api) {
cib_api->cmds->signoff(cib_api);
}
if (ipcs) {
qb_ipcs_destroy(ipcs);
}
crm_peer_destroy();
pcmk__client_cleanup();
free_stonith_remote_op_list();
free_topology_list();
free_device_list();
free_metadata_cache();
free(stonith_our_uname);
stonith_our_uname = NULL;
free_xml(local_cib);
local_cib = NULL;
}
/* *INDENT-OFF* */
static struct crm_option long_options[] = {
{"stand-alone", 0, 0, 's'},
{"stand-alone-w-cpg", 0, 0, 'c'},
{"logfile", 1, 0, 'l'},
{"verbose", 0, 0, 'V'},
{"version", 0, 0, '$'},
{"help", 0, 0, '?'},
{0, 0, 0, 0}
};
/* *INDENT-ON* */
static void
setup_cib(void)
{
int rc, retries = 0;
static cib_t *(*cib_new_fn) (void) = NULL;
if (cib_new_fn == NULL) {
cib_new_fn = find_library_function(&cib_library, CIB_LIBRARY, "cib_new", TRUE);
}
if (cib_new_fn != NULL) {
cib_api = (*cib_new_fn) ();
}
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 stonith topology changes");
}
}
struct qb_ipcs_service_handlers ipc_callbacks = {
.connection_accept = st_ipc_accept,
- .connection_created = st_ipc_created,
+ .connection_created = NULL,
.msg_process = st_ipc_dispatch,
.connection_closed = st_ipc_closed,
.connection_destroyed = st_ipc_destroy
};
/*!
* \internal
* \brief Callback for peer status changes
*
* \param[in] type What changed
* \param[in] node What peer had the change
* \param[in] data Previous value of what changed
*/
static void
st_peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data)
{
if ((type != crm_status_processes) && !is_set(node->flags, crm_remote_node)) {
/*
* This is a hack until we can send to a nodeid and/or we fix node name lookups
* These messages are ignored in stonith_peer_callback()
*/
xmlNode *query = create_xml_node(NULL, "stonith_command");
crm_xml_add(query, F_XML_TAGNAME, "stonith_command");
crm_xml_add(query, F_TYPE, T_STONITH_NG);
crm_xml_add(query, F_STONITH_OPERATION, "poke");
crm_debug("Broadcasting our uname because of node %u", node->id);
send_cluster_message(NULL, crm_msg_stonith_ng, query, FALSE);
free_xml(query);
}
}
int
main(int argc, char **argv)
{
int flag;
int lpc = 0;
int argerr = 0;
int option_index = 0;
crm_cluster_t cluster;
const char *actions[] = { "reboot", "off", "on", "list", "monitor", "status" };
crm_ipc_t *old_instance = NULL;
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "mode [options]", long_options,
"Provides a summary of cluster's current state."
"\n\nOutputs varying levels of detail in a number of different formats.\n");
while (1) {
flag = crm_get_option(argc, argv, &option_index);
if (flag == -1) {
break;
}
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'l':
crm_add_logfile(optarg);
break;
case 's':
stand_alone = TRUE;
break;
case 'c':
stand_alone = FALSE;
no_cib_connect = TRUE;
break;
case '$':
case '?':
crm_help(flag, CRM_EX_OK);
break;
default:
++argerr;
break;
}
}
if (argc - optind == 1 && safe_str_eq("metadata", argv[optind])) {
printf("<?xml version=\"1.0\"?><!DOCTYPE resource-agent SYSTEM \"ra-api-1.dtd\">\n");
printf("<resource-agent name=\"pacemaker-fenced\">\n");
printf(" <version>1.0</version>\n");
printf(" <longdesc lang=\"en\">Instance attributes available for all \"stonith\"-class resources"
" and used by Pacemaker's fence daemon, formerly known as stonithd</longdesc>\n");
printf(" <shortdesc lang=\"en\">Instance attributes available for all \"stonith\"-class resources</shortdesc>\n");
printf(" <parameters>\n");
#if 0
// priority is not implemented yet
printf(" <parameter name=\"priority\" unique=\"0\">\n");
printf(" <shortdesc lang=\"en\">Devices that are not in a topology "
"are tried in order of highest to lowest integer priority</shortdesc>\n");
printf(" <content type=\"integer\" default=\"0\"/>\n");
printf(" </parameter>\n");
#endif
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_HOSTARG);
printf
(" <shortdesc lang=\"en\">Advanced use only: An alternate parameter to supply instead of 'port'</shortdesc>\n");
printf
(" <longdesc lang=\"en\">Some devices do not support the standard 'port' parameter or may provide additional ones.\n"
"Use this to specify an alternate, device-specific, parameter that should indicate the machine to be fenced.\n"
"A value of 'none' can be used to tell the cluster not to supply any additional parameters.\n"
" </longdesc>\n");
printf(" <content type=\"string\" default=\"port\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_HOSTMAP);
printf
(" <shortdesc lang=\"en\">A mapping of host names to ports numbers for devices that do not support host names.</shortdesc>\n");
printf
(" <longdesc lang=\"en\">Eg. node1:1;node2:2,3 would tell the cluster to use port 1 for node1 and ports 2 and 3 for node2</longdesc>\n");
printf(" <content type=\"string\" default=\"\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_HOSTLIST);
printf
(" <shortdesc lang=\"en\">A list of machines controlled by this device (Optional unless %s=static-list).</shortdesc>\n",
STONITH_ATTR_HOSTCHECK);
printf(" <content type=\"string\" default=\"\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_HOSTCHECK);
printf
(" <shortdesc lang=\"en\">How to determine which machines are controlled by the device.</shortdesc>\n");
printf(" <longdesc lang=\"en\">Allowed values: dynamic-list "
"(query the device via the 'list' command), static-list "
"(check the " STONITH_ATTR_HOSTLIST " attribute), status "
"(query the device via the 'status' command), none (assume "
"every device can fence every machine)</longdesc>\n");
printf(" <content type=\"string\" default=\"dynamic-list\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_DELAY_MAX);
printf
(" <shortdesc lang=\"en\">Enable a random delay for stonith actions and specify the maximum of random delay.</shortdesc>\n");
printf
(" <longdesc lang=\"en\">This prevents double fencing when using slow devices such as sbd.\n"
"Use this to enable a random delay for stonith actions.\n"
"The overall delay is derived from this random delay value adding a static delay so that the sum is kept below the maximum delay.</longdesc>\n");
printf(" <content type=\"time\" default=\"0s\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_DELAY_BASE);
printf
(" <shortdesc lang=\"en\">Enable a base delay for stonith actions and specify base delay value.</shortdesc>\n");
printf
(" <longdesc lang=\"en\">This prevents double fencing when different delays are configured on the nodes.\n"
"Use this to enable a static delay for stonith actions.\n"
"The overall delay is derived from a random delay value adding this static delay so that the sum is kept below the maximum delay.</longdesc>\n");
printf(" <content type=\"time\" default=\"0s\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"%s\" unique=\"0\">\n", STONITH_ATTR_ACTION_LIMIT);
printf
(" <shortdesc lang=\"en\">The maximum number of actions can be performed in parallel on this device</shortdesc>\n");
printf
(" <longdesc lang=\"en\">Cluster property concurrent-fencing=true needs to be configured first.\n"
"Then use this to specify the maximum number of actions can be performed in parallel on this device. -1 is unlimited.</longdesc>\n");
printf(" <content type=\"integer\" default=\"1\"/>\n");
printf(" </parameter>\n");
for (lpc = 0; lpc < DIMOF(actions); lpc++) {
printf(" <parameter name=\"pcmk_%s_action\" unique=\"0\">\n", actions[lpc]);
printf
(" <shortdesc lang=\"en\">Advanced use only: An alternate command to run instead of '%s'</shortdesc>\n",
actions[lpc]);
printf
(" <longdesc lang=\"en\">Some devices do not support the standard commands or may provide additional ones.\n"
"Use this to specify an alternate, device-specific, command that implements the '%s' action.</longdesc>\n",
actions[lpc]);
printf(" <content type=\"string\" default=\"%s\"/>\n", actions[lpc]);
printf(" </parameter>\n");
printf(" <parameter name=\"pcmk_%s_timeout\" unique=\"0\">\n", actions[lpc]);
printf
(" <shortdesc lang=\"en\">Advanced use only: Specify an alternate timeout to use for %s actions instead of stonith-timeout</shortdesc>\n",
actions[lpc]);
printf
(" <longdesc lang=\"en\">Some devices need much more/less time to complete than normal.\n"
"Use this to specify an alternate, device-specific, timeout for '%s' actions.</longdesc>\n",
actions[lpc]);
printf(" <content type=\"time\" default=\"60s\"/>\n");
printf(" </parameter>\n");
printf(" <parameter name=\"pcmk_%s_retries\" unique=\"0\">\n", actions[lpc]);
printf
(" <shortdesc lang=\"en\">Advanced use only: The maximum number of times to retry the '%s' command within the timeout period</shortdesc>\n",
actions[lpc]);
printf(" <longdesc lang=\"en\">Some devices do not support multiple connections."
" Operations may 'fail' if the device is busy with another task so Pacemaker will automatically retry the operation, if there is time remaining."
" Use this option to alter the number of times Pacemaker retries '%s' actions before giving up."
"</longdesc>\n", actions[lpc]);
printf(" <content type=\"integer\" default=\"2\"/>\n");
printf(" </parameter>\n");
}
printf(" </parameters>\n");
printf("</resource-agent>\n");
return CRM_EX_OK;
}
if (optind != argc) {
++argerr;
}
if (argerr) {
crm_help('?', CRM_EX_USAGE);
}
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
crm_notice("Starting Pacemaker fencer");
old_instance = crm_ipc_new("stonith-ng", 0);
if (crm_ipc_connect(old_instance)) {
/* IPC end-point already up */
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_err("pacemaker-fenced is already active, aborting startup");
crm_exit(CRM_EX_OK);
} else {
/* not up or not authentic, we'll proceed either way */
crm_ipc_destroy(old_instance);
old_instance = NULL;
}
mainloop_add_signal(SIGTERM, stonith_shutdown);
crm_peer_init();
fenced_data_set = pe_new_working_set();
CRM_ASSERT(fenced_data_set != NULL);
set_bit(fenced_data_set->flags, pe_flag_no_counts);
set_bit(fenced_data_set->flags, pe_flag_no_compat);
if (stand_alone == FALSE) {
if (is_corosync_cluster()) {
#if SUPPORT_COROSYNC
cluster.destroy = stonith_peer_cs_destroy;
cluster.cpg.cpg_deliver_fn = stonith_peer_ais_callback;
cluster.cpg.cpg_confchg_fn = pcmk_cpg_membership;
#endif
}
crm_set_status_callback(&st_peer_update_callback);
if (crm_cluster_connect(&cluster) == FALSE) {
crm_crit("Cannot sign in to the cluster... terminating");
crm_exit(CRM_EX_FATAL);
}
stonith_our_uname = cluster.uname;
stonith_our_uuid = cluster.uuid;
if (no_cib_connect == FALSE) {
setup_cib();
}
} else {
stonith_our_uname = strdup("localhost");
}
init_device_list();
init_topology_list();
if(stonith_watchdog_timeout_ms > 0) {
int rc;
xmlNode *xml;
stonith_key_value_t *params = NULL;
params = stonith_key_value_add(params, STONITH_ATTR_HOSTLIST, stonith_our_uname);
xml = create_device_registration_xml("watchdog", st_namespace_internal,
STONITH_WATCHDOG_AGENT, params,
NULL);
stonith_key_value_freeall(params, 1, 1);
rc = stonith_device_register(xml, NULL, FALSE);
free_xml(xml);
if (rc != pcmk_ok) {
crm_crit("Cannot register watchdog pseudo fence agent");
crm_exit(CRM_EX_FATAL);
}
}
stonith_ipc_server_init(&ipcs, &ipc_callbacks);
/* Create the mainloop and run it... */
mainloop = g_main_loop_new(NULL, FALSE);
crm_notice("Pacemaker fencer successfully started and accepting connections");
g_main_loop_run(mainloop);
stonith_cleanup();
pe_free_working_set(fenced_data_set);
crm_exit(CRM_EX_OK);
}
diff --git a/daemons/pacemakerd/pacemakerd.c b/daemons/pacemakerd/pacemakerd.c
index 642f5c4f1c..70c044b990 100644
--- a/daemons/pacemakerd/pacemakerd.c
+++ b/daemons/pacemakerd/pacemakerd.c
@@ -1,1445 +1,1439 @@
/*
* Copyright 2010-2020 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 "pacemakerd.h"
#include <pwd.h>
#include <grp.h>
#include <poll.h>
#include <stdio.h>
#include <stdbool.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/time.h>
#include <sys/resource.h>
#include <sys/reboot.h>
#include <crm/crm.h> /* indirectly: CRM_EX_* */
#include <crm/cib/internal.h> /* cib_channel_ro */
#include <crm/msg_xml.h>
#include <crm/common/ipcs_internal.h>
#include <crm/common/mainloop.h>
#include <crm/cluster/internal.h>
#include <crm/cluster.h>
#include <crm/common/ipc_internal.h> /* PCMK__SPECIAL_PID*, ... */
#ifdef SUPPORT_COROSYNC
#include <corosync/cfg.h>
#endif
#include <dirent.h>
#include <ctype.h>
static gboolean pcmk_quorate = FALSE;
static gboolean fatal_error = FALSE;
static GMainLoop *mainloop = NULL;
static bool global_keep_tracking = false;
#define PCMK_PROCESS_CHECK_INTERVAL 5
static const char *local_name = NULL;
static uint32_t local_nodeid = 0;
static crm_trigger_t *shutdown_trigger = NULL;
static const char *pid_file = PCMK_RUN_DIR "/pacemaker.pid";
typedef struct pcmk_child_s {
pid_t pid;
long flag;
int start_seq;
int respawn_count;
gboolean respawn;
const char *name;
const char *uid;
const char *command;
const char *endpoint; /* IPC server name */
gboolean active_before_startup;
} pcmk_child_t;
/* Index into the array below */
#define PCMK_CHILD_CONTROLD 3
static pcmk_child_t pcmk_children[] = {
{
0, crm_proc_none, 0, 0, FALSE, "none",
NULL, NULL
},
{
0, crm_proc_execd, 3, 0, TRUE, "pacemaker-execd",
NULL, CRM_DAEMON_DIR "/pacemaker-execd",
CRM_SYSTEM_LRMD
},
{
0, crm_proc_based, 1, 0, TRUE, "pacemaker-based",
CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-based",
CIB_CHANNEL_RO
},
{
0, crm_proc_controld, 6, 0, TRUE, "pacemaker-controld",
CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-controld",
CRM_SYSTEM_CRMD
},
{
0, crm_proc_attrd, 4, 0, TRUE, "pacemaker-attrd",
CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-attrd",
T_ATTRD
},
{
0, crm_proc_schedulerd, 5, 0, TRUE, "pacemaker-schedulerd",
CRM_DAEMON_USER, CRM_DAEMON_DIR "/pacemaker-schedulerd",
CRM_SYSTEM_PENGINE
},
{
0, crm_proc_fenced, 2, 0, TRUE, "pacemaker-fenced",
NULL, CRM_DAEMON_DIR "/pacemaker-fenced",
"stonith-ng"
},
};
static gboolean check_active_before_startup_processes(gpointer user_data);
static int child_liveness(pcmk_child_t *child);
static gboolean start_child(pcmk_child_t * child);
static gboolean update_node_processes(uint32_t id, const char *uname,
uint32_t procs);
void update_process_clients(pcmk__client_t *client);
static uint32_t
get_process_list(void)
{
int lpc = 0;
uint32_t procs = crm_get_cluster_proc();
for (lpc = 0; lpc < SIZEOF(pcmk_children); lpc++) {
if (pcmk_children[lpc].pid != 0) {
procs |= pcmk_children[lpc].flag;
}
}
return procs;
}
static void
pcmk_process_exit(pcmk_child_t * child)
{
child->pid = 0;
child->active_before_startup = FALSE;
/* Broadcast the fact that one of our processes died ASAP
*
* Try to get some logging of the cause out first though
* because we're probably about to get fenced
*
* Potentially do this only if respawn_count > N
* to allow for local recovery
*/
update_node_processes(local_nodeid, NULL, get_process_list());
child->respawn_count += 1;
if (child->respawn_count > MAX_RESPAWN) {
crm_err("Child respawn count exceeded by %s", child->name);
child->respawn = FALSE;
}
if (shutdown_trigger) {
/* resume step-wise shutdown (returned TRUE yields no parallelizing) */
mainloop_set_trigger(shutdown_trigger);
/* intended to speed up propagating expected lay-off of the daemons? */
update_node_processes(local_nodeid, NULL, get_process_list());
} else if (!child->respawn) {
/* nothing to do */
} else if (crm_is_true(getenv("PCMK_fail_fast"))) {
crm_err("Rebooting system because of %s", child->name);
pcmk_panic(__FUNCTION__);
} else if (child_liveness(child) == pcmk_rc_ok) {
crm_warn("One-off suppressing strict respawning of a child process %s,"
" appears alright per %s IPC end-point",
child->name, child->endpoint);
/* need to monitor how it evolves, and start new process if badly */
child->active_before_startup = TRUE;
if (!global_keep_tracking) {
global_keep_tracking = true;
g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL,
check_active_before_startup_processes, NULL);
}
} else {
crm_notice("Respawning failed child process: %s", child->name);
start_child(child);
}
}
static void pcmk_exit_with_cluster(int exitcode)
{
#ifdef SUPPORT_COROSYNC
corosync_cfg_handle_t cfg_handle;
cs_error_t err;
if (exitcode == CRM_EX_FATAL) {
crm_info("Asking Corosync to shut down");
err = corosync_cfg_initialize(&cfg_handle, NULL);
if (err != CS_OK) {
crm_warn("Unable to open handle to corosync to close it down. err=%d", err);
}
err = corosync_cfg_try_shutdown(cfg_handle, COROSYNC_CFG_SHUTDOWN_FLAG_IMMEDIATE);
if (err != CS_OK) {
crm_warn("Corosync shutdown failed. err=%d", err);
}
corosync_cfg_finalize(cfg_handle);
}
#endif
crm_exit(exitcode);
}
static void
pcmk_child_exit(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)
{
pcmk_child_t *child = mainloop_child_userdata(p);
const char *name = mainloop_child_name(p);
if (signo) {
do_crm_log(((signo == SIGKILL)? LOG_WARNING : LOG_ERR),
"%s[%d] terminated with signal %d (core=%d)",
name, pid, signo, core);
} else {
switch(exitcode) {
case CRM_EX_OK:
crm_info("%s[%d] exited with status %d (%s)",
name, pid, exitcode, crm_exit_str(exitcode));
break;
case CRM_EX_FATAL:
crm_warn("Shutting cluster down because %s[%d] had fatal failure",
name, pid);
child->respawn = FALSE;
fatal_error = TRUE;
pcmk_shutdown(SIGTERM);
break;
case CRM_EX_PANIC:
do_crm_log_always(LOG_EMERG,
"%s[%d] instructed the machine to reset",
name, pid);
child->respawn = FALSE;
fatal_error = TRUE;
pcmk_panic(__FUNCTION__);
pcmk_shutdown(SIGTERM);
break;
default:
crm_err("%s[%d] exited with status %d (%s)",
name, pid, exitcode, crm_exit_str(exitcode));
break;
}
}
pcmk_process_exit(child);
}
static gboolean
stop_child(pcmk_child_t * child, int signal)
{
if (signal == 0) {
signal = SIGTERM;
}
/* why to skip PID of 1?
- FreeBSD ~ how untrackable process behind IPC is masqueraded as
- elsewhere: how "init" task is designated; in particular, in systemd
arrangement of socket-based activation, this is pretty real */
if (child->command == NULL || child->pid == PCMK__SPECIAL_PID) {
crm_debug("Nothing to do for child \"%s\" (process %lld)",
child->name, (long long) PCMK__SPECIAL_PID_AS_0(child->pid));
return TRUE;
}
if (child->pid <= 0) {
crm_trace("Client %s not running", child->name);
return TRUE;
}
errno = 0;
if (kill(child->pid, signal) == 0) {
crm_notice("Stopping %s "CRM_XS" sent signal %d to process %lld",
child->name, signal, (long long) child->pid);
} else {
crm_err("Could not stop %s (process %lld) with signal %d: %s",
child->name, (long long) child->pid, signal, strerror(errno));
}
return TRUE;
}
static char *opts_default[] = { NULL, NULL };
static char *opts_vgrind[] = { NULL, NULL, NULL, NULL, NULL };
/* TODO once libqb is taught to juggle with IPC end-points carried over as
bare file descriptor (https://github.com/ClusterLabs/libqb/issues/325)
it shall hand over these descriptors here if/once they are successfully
pre-opened in (presumably) child_liveness(), to avoid any remaining
room for races */
static gboolean
start_child(pcmk_child_t * child)
{
uid_t uid = 0;
gid_t gid = 0;
gboolean use_valgrind = FALSE;
gboolean use_callgrind = FALSE;
const char *devnull = "/dev/null";
const char *env_valgrind = getenv("PCMK_valgrind_enabled");
const char *env_callgrind = getenv("PCMK_callgrind_enabled");
child->active_before_startup = FALSE;
if (child->command == NULL) {
crm_info("Nothing to do for child \"%s\"", child->name);
return TRUE;
}
if (env_callgrind != NULL && crm_is_true(env_callgrind)) {
use_callgrind = TRUE;
use_valgrind = TRUE;
} else if (env_callgrind != NULL && strstr(env_callgrind, child->name)) {
use_callgrind = TRUE;
use_valgrind = TRUE;
} else if (env_valgrind != NULL && crm_is_true(env_valgrind)) {
use_valgrind = TRUE;
} else if (env_valgrind != NULL && strstr(env_valgrind, child->name)) {
use_valgrind = TRUE;
}
if (use_valgrind && strlen(VALGRIND_BIN) == 0) {
crm_warn("Cannot enable valgrind for %s:"
" The location of the valgrind binary is unknown", child->name);
use_valgrind = FALSE;
}
if (child->uid) {
if (crm_user_lookup(child->uid, &uid, &gid) < 0) {
crm_err("Invalid user (%s) for %s: not found", child->uid, child->name);
return FALSE;
}
crm_info("Using uid=%u and group=%u for process %s", uid, gid, child->name);
}
child->pid = fork();
CRM_ASSERT(child->pid != -1);
if (child->pid > 0) {
/* parent */
mainloop_child_add(child->pid, 0, child->name, child, pcmk_child_exit);
crm_info("Forked child %lld for process %s%s",
(long long) child->pid, child->name,
use_valgrind ? " (valgrind enabled: " VALGRIND_BIN ")" : "");
update_node_processes(local_nodeid, NULL, get_process_list());
return TRUE;
} else {
/* Start a new session */
(void)setsid();
/* Setup the two alternate arg arrays */
opts_vgrind[0] = strdup(VALGRIND_BIN);
if (use_callgrind) {
opts_vgrind[1] = strdup("--tool=callgrind");
opts_vgrind[2] = strdup("--callgrind-out-file=" CRM_STATE_DIR "/callgrind.out.%p");
opts_vgrind[3] = strdup(child->command);
opts_vgrind[4] = NULL;
} else {
opts_vgrind[1] = strdup(child->command);
opts_vgrind[2] = NULL;
opts_vgrind[3] = NULL;
opts_vgrind[4] = NULL;
}
opts_default[0] = strdup(child->command);
if(gid) {
// Whether we need root group access to talk to cluster layer
bool need_root_group = TRUE;
if (is_corosync_cluster()) {
/* Corosync clusters can drop root group access, because we set
* uidgid.gid.${gid}=1 via CMAP, which allows these processes to
* connect to corosync.
*/
need_root_group = FALSE;
}
// Drop root group access if not needed
if (!need_root_group && (setgid(gid) < 0)) {
crm_perror(LOG_ERR, "Could not set group to %d", gid);
}
/* Initialize supplementary groups to only those always granted to
* the user, plus haclient (so we can access IPC).
*/
if (initgroups(child->uid, gid) < 0) {
crm_err("Cannot initialize groups for %s: %s (%d)", child->uid, pcmk_strerror(errno), errno);
}
}
if (uid && setuid(uid) < 0) {
crm_perror(LOG_ERR, "Could not set user to %d (%s)", uid, child->uid);
}
pcmk__close_fds_in_child(true);
(void)open(devnull, O_RDONLY); /* Stdin: fd 0 */
(void)open(devnull, O_WRONLY); /* Stdout: fd 1 */
(void)open(devnull, O_WRONLY); /* Stderr: fd 2 */
if (use_valgrind) {
(void)execvp(VALGRIND_BIN, opts_vgrind);
} else {
(void)execvp(child->command, opts_default);
}
crm_perror(LOG_ERR, "FATAL: Cannot exec %s", child->command);
crm_exit(CRM_EX_FATAL);
}
return TRUE; /* never reached */
}
static gboolean
escalate_shutdown(gpointer data)
{
pcmk_child_t *child = data;
if (child->pid == PCMK__SPECIAL_PID) {
pcmk_process_exit(child);
} else if (child->pid != 0) {
/* Use SIGSEGV instead of SIGKILL to create a core so we can see what it was up to */
crm_err("Child %s not terminating in a timely manner, forcing", child->name);
stop_child(child, SIGSEGV);
}
return FALSE;
}
#define SHUTDOWN_ESCALATION_PERIOD 180000 /* 3m */
static gboolean
pcmk_shutdown_worker(gpointer user_data)
{
static int phase = 0;
static time_t next_log = 0;
static int max = SIZEOF(pcmk_children);
int lpc = 0;
if (phase == 0) {
crm_notice("Shutting down Pacemaker");
phase = max;
}
for (; phase > 0; phase--) {
/* Don't stop anything with start_seq < 1 */
for (lpc = max - 1; lpc >= 0; lpc--) {
pcmk_child_t *child = &(pcmk_children[lpc]);
if (phase != child->start_seq) {
continue;
}
if (child->pid != 0) {
time_t now = time(NULL);
if (child->respawn) {
if (child->pid == PCMK__SPECIAL_PID) {
crm_warn("The process behind %s IPC cannot be"
" terminated, so either wait the graceful"
" period of %ld s for its native termination"
" if it vitally depends on some other daemons"
" going down in a controlled way already,"
" or locate and kill the correct %s process"
" on your own; set PCMK_fail_fast=1 to avoid"
" this altogether next time around",
child->name, (long) SHUTDOWN_ESCALATION_PERIOD,
child->command);
}
next_log = now + 30;
child->respawn = FALSE;
stop_child(child, SIGTERM);
if (phase < pcmk_children[PCMK_CHILD_CONTROLD].start_seq) {
g_timeout_add(SHUTDOWN_ESCALATION_PERIOD,
escalate_shutdown, child);
}
} else if (now >= next_log) {
next_log = now + 30;
crm_notice("Still waiting for %s to terminate "
CRM_XS " pid=%lld seq=%d",
child->name, (long long) child->pid,
child->start_seq);
}
return TRUE;
}
/* cleanup */
crm_debug("%s confirmed stopped", child->name);
child->pid = 0;
}
}
/* send_cluster_id(); */
crm_notice("Shutdown complete");
{
const char *delay = pcmk__env_option("shutdown_delay");
if(delay) {
sync();
sleep(crm_get_msec(delay) / 1000);
}
}
g_main_loop_quit(mainloop);
if (fatal_error) {
crm_notice("Shutting down and staying down after fatal error");
pcmk_exit_with_cluster(CRM_EX_FATAL);
}
return TRUE;
}
static void
pcmk_ignore(int nsig)
{
crm_info("Ignoring signal %s (%d)", strsignal(nsig), nsig);
}
static void
pcmk_sigquit(int nsig)
{
pcmk_panic(__FUNCTION__);
}
void
pcmk_shutdown(int nsig)
{
if (shutdown_trigger == NULL) {
shutdown_trigger = mainloop_add_trigger(G_PRIORITY_HIGH, pcmk_shutdown_worker, NULL);
}
mainloop_set_trigger(shutdown_trigger);
}
static int32_t
pcmk_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
-static void
-pcmk_ipc_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection %p", c);
-}
-
/* Exit code means? */
static int32_t
pcmk_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
const char *task = NULL;
pcmk__client_t *c = pcmk__find_client(qbc);
xmlNode *msg = pcmk__client_data2xml(c, data, size, &id, &flags);
pcmk__ipc_send_ack(c, id, flags, "ack");
if (msg == NULL) {
return 0;
}
task = crm_element_value(msg, F_CRM_TASK);
if (crm_str_eq(task, CRM_OP_QUIT, TRUE)) {
/* Time to quit */
crm_notice("Shutting down in response to ticket %s (%s)",
crm_element_value(msg, F_CRM_REFERENCE), crm_element_value(msg, F_CRM_ORIGIN));
pcmk_shutdown(15);
} else if (crm_str_eq(task, CRM_OP_RM_NODE_CACHE, TRUE)) {
/* Send to everyone */
struct iovec *iov;
int id = 0;
const char *name = NULL;
crm_element_value_int(msg, XML_ATTR_ID, &id);
name = crm_element_value(msg, XML_ATTR_UNAME);
crm_notice("Instructing peers to remove references to node %s/%u", name, id);
iov = calloc(1, sizeof(struct iovec));
iov->iov_base = dump_xml_unformatted(msg);
iov->iov_len = 1 + strlen(iov->iov_base);
send_cpg_iov(iov);
} else {
update_process_clients(c);
}
free_xml(msg);
return 0;
}
/* Error code means? */
static int32_t
pcmk_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p", c);
pcmk__free_client(client);
return 0;
}
static void
pcmk_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
pcmk_ipc_closed(c);
}
struct qb_ipcs_service_handlers mcp_ipc_callbacks = {
.connection_accept = pcmk_ipc_accept,
- .connection_created = pcmk_ipc_created,
+ .connection_created = NULL,
.msg_process = pcmk_ipc_dispatch,
.connection_closed = pcmk_ipc_closed,
.connection_destroyed = pcmk_ipc_destroy
};
static void
send_xml_to_client(gpointer key, gpointer value, gpointer user_data)
{
pcmk__ipc_send_xml((pcmk__client_t *) value, 0, (xmlNode *) user_data,
crm_ipc_server_event);
}
/*!
* \internal
* \brief Send an XML message with process list of all known peers to client(s)
*
* \param[in] client Send message to this client, or all clients if NULL
*/
void
update_process_clients(pcmk__client_t *client)
{
GHashTableIter iter;
crm_node_t *node = NULL;
xmlNode *update = create_xml_node(NULL, "nodes");
if (is_corosync_cluster()) {
crm_xml_add_int(update, "quorate", pcmk_quorate);
}
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) & node)) {
xmlNode *xml = create_xml_node(update, "node");
crm_xml_add_int(xml, "id", node->id);
crm_xml_add(xml, "uname", node->uname);
crm_xml_add(xml, "state", node->state);
crm_xml_add_int(xml, "processes", node->processes);
}
if(client) {
crm_trace("Sending process list to client %s", client->id);
send_xml_to_client(NULL, client, update);
} else {
crm_trace("Sending process list to %d clients",
pcmk__ipc_client_count());
pcmk__foreach_ipc_client(send_xml_to_client, update);
}
free_xml(update);
}
/*!
* \internal
* \brief Send a CPG message with local node's process list to all peers
*/
static void
update_process_peers(void)
{
/* Do nothing for corosync-2 based clusters */
struct iovec *iov = calloc(1, sizeof(struct iovec));
CRM_ASSERT(iov);
if (local_name) {
iov->iov_base = crm_strdup_printf("<node uname=\"%s\" proclist=\"%u\"/>",
local_name, get_process_list());
} else {
iov->iov_base = crm_strdup_printf("<node proclist=\"%u\"/>",
get_process_list());
}
iov->iov_len = strlen(iov->iov_base) + 1;
crm_trace("Sending %s", (char*) iov->iov_base);
send_cpg_iov(iov);
}
/*!
* \internal
* \brief Update a node's process list, notifying clients and peers if needed
*
* \param[in] id Node ID of affected node
* \param[in] uname Uname of affected node
* \param[in] procs Affected node's process list mask
*
* \return TRUE if the process list changed, FALSE otherwise
*/
static gboolean
update_node_processes(uint32_t id, const char *uname, uint32_t procs)
{
gboolean changed = FALSE;
crm_node_t *node = crm_get_peer(id, uname);
if (procs != 0) {
if (procs != node->processes) {
crm_debug("Node %s now has process list: %.32x (was %.32x)",
node->uname, procs, node->processes);
node->processes = procs;
changed = TRUE;
/* If local node's processes have changed, notify clients/peers */
if (id == local_nodeid) {
update_process_clients(NULL);
update_process_peers();
}
} else {
crm_trace("Node %s still has process list: %.32x", node->uname, procs);
}
}
return changed;
}
/* *INDENT-OFF* */
static struct crm_option long_options[] = {
/* Top-level Options */
{"help", 0, 0, '?', "\tThis text"},
{"version", 0, 0, '$', "\tVersion information" },
{"verbose", 0, 0, 'V', "\tIncrease debug output"},
{"shutdown", 0, 0, 'S', "\tInstruct Pacemaker to shutdown on this machine"},
{"features", 0, 0, 'F', "\tDisplay the full version and list of features Pacemaker was built with"},
{"-spacer-", 1, 0, '-', "\nAdditional Options:"},
{"foreground", 0, 0, 'f', "\t(Ignored) Pacemaker always runs in the foreground"},
{"pid-file", 1, 0, 'p', "\t(Ignored) Daemon pid file location"},
{"standby", 0, 0, 's', "\tStart node in standby state"},
{NULL, 0, 0, 0}
};
/* *INDENT-ON* */
static void
mcp_chown(const char *path, uid_t uid, gid_t gid)
{
int rc = chown(path, uid, gid);
if (rc < 0) {
crm_warn("Cannot change the ownership of %s to user %s and gid %d: %s",
path, CRM_DAEMON_USER, gid, pcmk_strerror(errno));
}
}
/*!
* \internal
* \brief Check the liveness of the child based on IPC name and PID if tracked
*
* \param[inout] child Child tracked data
*
* \return Standard Pacemaker return code
*
* \note Return codes of particular interest include pcmk_rc_ipc_unresponsive
* indicating that no trace of IPC liveness was detected,
* pcmk_rc_ipc_unauthorized indicating that the IPC endpoint is blocked by
* an unauthorized process, and pcmk_rc_ipc_pid_only indicating that
* the child is up by PID but not IPC end-point (possibly starting).
* \note This function doesn't modify any of \p child members but \c pid,
* and is not actively toying with processes as such but invoking
* \c stop_child in one particular case (there's for some reason
* a different authentic holder of the IPC end-point).
*/
static int
child_liveness(pcmk_child_t *child)
{
uid_t cl_uid = 0;
gid_t cl_gid = 0;
const uid_t root_uid = 0;
const gid_t root_gid = 0;
const uid_t *ref_uid;
const gid_t *ref_gid;
int rc = pcmk_rc_ipc_unresponsive;
pid_t ipc_pid = 0;
if (child->endpoint == NULL
&& (child->pid <= 0 || child->pid == PCMK__SPECIAL_PID)) {
crm_err("Cannot track child %s for missing both API end-point and PID",
child->name);
rc = EINVAL; // Misuse of function when child is not trackable
} else if (child->endpoint != NULL) {
int legacy_rc = pcmk_ok;
if (child->uid == NULL) {
ref_uid = &root_uid;
ref_gid = &root_gid;
} else {
ref_uid = &cl_uid;
ref_gid = &cl_gid;
legacy_rc = pcmk_daemon_user(&cl_uid, &cl_gid);
}
if (legacy_rc < 0) {
rc = pcmk_legacy2rc(legacy_rc);
crm_err("Could not find user and group IDs for user %s: %s "
CRM_XS " rc=%d", CRM_DAEMON_USER, pcmk_rc_str(rc), rc);
} else {
rc = pcmk__ipc_is_authentic_process_active(child->endpoint,
*ref_uid, *ref_gid,
&ipc_pid);
if ((rc == pcmk_rc_ok) || (rc == pcmk_rc_ipc_unresponsive)) {
if (child->pid <= 0) {
/* If rc is pcmk_rc_ok, ipc_pid is nonzero and this
* initializes a new child. If rc is
* pcmk_rc_ipc_unresponsive, ipc_pid is zero, and we will
* investigate further.
*/
child->pid = ipc_pid;
} else if ((ipc_pid != 0) && (child->pid != ipc_pid)) {
/* An unexpected (but authorized) process is responding to
* IPC. Investigate further.
*/
rc = pcmk_rc_ipc_unresponsive;
}
}
}
}
if (rc == pcmk_rc_ipc_unresponsive) {
/* If we get here, a child without IPC is being tracked, no IPC liveness
* has been detected, or IPC liveness has been detected with an
* unexpected (but authorized) process. This is safe on FreeBSD since
* the only change possible from a proper child's PID into "special" PID
* of 1 behind more loosely related process.
*/
int ret = pcmk__pid_active(child->pid, child->name);
if (ipc_pid && ((ret != pcmk_rc_ok)
|| ipc_pid == PCMK__SPECIAL_PID
|| (pcmk__pid_active(ipc_pid,
child->name) == pcmk_rc_ok))) {
/* An unexpected (but authorized) process was detected at the IPC
* endpoint, and either it is active, or the child we're tracking is
* not.
*/
if (ret == pcmk_rc_ok) {
/* The child we're tracking is active. Kill it, and adopt the
* detected process. This assumes that our children don't fork
* (thus getting a different PID owning the IPC), but rather the
* tracking got out of sync because of some means external to
* Pacemaker, and adopting the detected process is better than
* killing it and possibly having to spawn a new child.
*/
/* not possessing IPC, afterall (what about corosync CPG?) */
stop_child(child, SIGKILL);
}
rc = pcmk_rc_ok;
child->pid = ipc_pid;
} else if (ret == pcmk_rc_ok) {
// Our tracked child's PID was found active, but not its IPC
rc = pcmk_rc_ipc_pid_only;
} else if ((child->pid == 0) && (ret == EINVAL)) {
// FreeBSD can return EINVAL
rc = pcmk_rc_ipc_unresponsive;
} else {
switch (ret) {
case EACCES:
rc = pcmk_rc_ipc_unauthorized;
break;
case ESRCH:
rc = pcmk_rc_ipc_unresponsive;
break;
default:
rc = ret;
break;
}
}
}
return rc;
}
static gboolean
check_active_before_startup_processes(gpointer user_data)
{
int start_seq = 1, lpc = 0;
static int max = SIZEOF(pcmk_children);
gboolean keep_tracking = FALSE;
for (start_seq = 1; start_seq < max; start_seq++) {
for (lpc = 0; lpc < max; lpc++) {
if (pcmk_children[lpc].active_before_startup == FALSE) {
/* we are already tracking it as a child process. */
continue;
} else if (start_seq != pcmk_children[lpc].start_seq) {
continue;
} else {
int rc = child_liveness(&pcmk_children[lpc]);
switch (rc) {
case pcmk_rc_ok:
break;
case pcmk_rc_ipc_unresponsive:
case pcmk_rc_ipc_pid_only: // This case: it was previously OK
if (pcmk_children[lpc].respawn == TRUE) {
crm_err("%s[%lld] terminated%s", pcmk_children[lpc].name,
(long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
} else {
/* orderly shutdown */
crm_notice("%s[%lld] terminated%s", pcmk_children[lpc].name,
(long long) PCMK__SPECIAL_PID_AS_0(pcmk_children[lpc].pid),
(rc == pcmk_rc_ipc_pid_only)? " as IPC server" : "");
}
pcmk_process_exit(&(pcmk_children[lpc]));
continue;
default:
crm_exit(CRM_EX_FATAL);
break; /* static analysis/noreturn */
}
}
/* at least one of the processes found at startup
* is still going, so keep this recurring timer around */
keep_tracking = TRUE;
}
}
global_keep_tracking = keep_tracking;
return keep_tracking;
}
/*!
* \internal
* \brief Initial one-off check of the pre-existing "child" processes
*
* With "child" process, we mean the subdaemon that defines an API end-point
* (all of them do as of the comment) -- the possible complement is skipped
* as it is deemed it has no such shared resources to cause conflicts about,
* hence it can presumably be started anew without hesitation.
* If that won't hold true in the future, the concept of a shared resource
* will have to be generalized beyond the API end-point.
*
* For boundary cases that the "child" is still starting (IPC end-point is yet
* to be witnessed), or more rarely (practically FreeBSD only), when there's
* a pre-existing "untrackable" authentic process, we give the situation some
* time to possibly unfold in the right direction, meaning that said socket
* will appear or the unattainable process will disappear per the observable
* IPC, respectively.
*
* \return Standard Pacemaker return code
*
* \note Since this gets run at the very start, \c respawn_count fields
* for particular children get temporarily overloaded with "rounds
* of waiting" tracking, restored once we are about to finish with
* success (i.e. returning value >=0) and will remain unrestored
* otherwise. One way to suppress liveness detection logic for
* particular child is to set the said value to a negative number.
*/
#define WAIT_TRIES 4 /* together with interleaved sleeps, worst case ~ 1s */
static int
find_and_track_existing_processes(void)
{
bool tracking = false;
bool wait_in_progress;
int rc;
size_t i, rounds;
for (rounds = 1; rounds <= WAIT_TRIES; rounds++) {
wait_in_progress = false;
for (i = 0; i < SIZEOF(pcmk_children); i++) {
if ((pcmk_children[i].endpoint == NULL)
|| (pcmk_children[i].respawn_count < 0)) {
continue;
}
rc = child_liveness(&pcmk_children[i]);
if (rc == pcmk_rc_ipc_unresponsive) {
/* As a speculation, don't give up if there are more rounds to
* come for other reasons, but don't artificially wait just
* because of this, since we would preferably start ASAP.
*/
continue;
}
pcmk_children[i].respawn_count = rounds;
switch (rc) {
case pcmk_rc_ok:
if (pcmk_children[i].pid == PCMK__SPECIAL_PID) {
if (crm_is_true(getenv("PCMK_fail_fast"))) {
crm_crit("Cannot reliably track pre-existing"
" authentic process behind %s IPC on this"
" platform and PCMK_fail_fast requested",
pcmk_children[i].endpoint);
return EOPNOTSUPP;
} else if (pcmk_children[i].respawn_count == WAIT_TRIES) {
crm_notice("Assuming pre-existing authentic, though"
" on this platform untrackable, process"
" behind %s IPC is stable (was in %d"
" previous samples) so rather than"
" bailing out (PCMK_fail_fast not"
" requested), we just switch to a less"
" optimal IPC liveness monitoring"
" (not very suitable for heavy load)",
pcmk_children[i].name, WAIT_TRIES - 1);
crm_warn("The process behind %s IPC cannot be"
" terminated, so the overall shutdown"
" will get delayed implicitly (%ld s),"
" which serves as a graceful period for"
" its native termination if it vitally"
" depends on some other daemons going"
" down in a controlled way already",
pcmk_children[i].name,
(long) SHUTDOWN_ESCALATION_PERIOD);
} else {
wait_in_progress = true;
crm_warn("Cannot reliably track pre-existing"
" authentic process behind %s IPC on this"
" platform, can still disappear in %d"
" attempt(s)", pcmk_children[i].endpoint,
WAIT_TRIES - pcmk_children[i].respawn_count);
continue;
}
}
crm_notice("Tracking existing %s process (pid=%lld)",
pcmk_children[i].name,
(long long) PCMK__SPECIAL_PID_AS_0(
pcmk_children[i].pid));
pcmk_children[i].respawn_count = -1; /* 0~keep watching */
pcmk_children[i].active_before_startup = TRUE;
tracking = true;
break;
case pcmk_rc_ipc_pid_only:
if (pcmk_children[i].respawn_count == WAIT_TRIES) {
crm_crit("%s IPC end-point for existing authentic"
" process %lld did not (re)appear",
pcmk_children[i].endpoint,
(long long) PCMK__SPECIAL_PID_AS_0(
pcmk_children[i].pid));
return rc;
}
wait_in_progress = true;
crm_warn("Cannot find %s IPC end-point for existing"
" authentic process %lld, can still (re)appear"
" in %d attempts (?)",
pcmk_children[i].endpoint,
(long long) PCMK__SPECIAL_PID_AS_0(
pcmk_children[i].pid),
WAIT_TRIES - pcmk_children[i].respawn_count);
continue;
default:
crm_crit("Checked liveness of %s: %s " CRM_XS " rc=%d",
pcmk_children[i].name, pcmk_rc_str(rc), rc);
return rc;
}
}
if (!wait_in_progress) {
break;
}
(void) poll(NULL, 0, 250); /* a bit for changes to possibly happen */
}
for (i = 0; i < SIZEOF(pcmk_children); i++) {
pcmk_children[i].respawn_count = 0; /* restore pristine state */
}
if (tracking) {
g_timeout_add_seconds(PCMK_PROCESS_CHECK_INTERVAL,
check_active_before_startup_processes, NULL);
}
return pcmk_rc_ok;
}
static void
init_children_processes(void)
{
int start_seq = 1, lpc = 0;
static int max = SIZEOF(pcmk_children);
/* start any children that have not been detected */
for (start_seq = 1; start_seq < max; start_seq++) {
/* don't start anything with start_seq < 1 */
for (lpc = 0; lpc < max; lpc++) {
if (pcmk_children[lpc].pid != 0) {
/* we are already tracking it */
continue;
}
if (start_seq == pcmk_children[lpc].start_seq) {
start_child(&(pcmk_children[lpc]));
}
}
}
/* From this point on, any daemons being started will be due to
* respawning rather than node start.
*
* This may be useful for the daemons to know
*/
setenv("PCMK_respawned", "true", 1);
}
static void
mcp_cpg_destroy(gpointer user_data)
{
crm_crit("Lost connection to cluster layer, shutting down");
crm_exit(CRM_EX_DISCONNECT);
}
/*!
* \internal
* \brief Process a CPG message (process list or manual peer cache removal)
*
* \param[in] handle CPG connection (ignored)
* \param[in] groupName CPG group name (ignored)
* \param[in] nodeid ID of affected node
* \param[in] pid Process ID (ignored)
* \param[in] msg CPG XML message
* \param[in] msg_len Length of msg in bytes (ignored)
*/
static void
mcp_cpg_deliver(cpg_handle_t handle,
const struct cpg_name *groupName,
uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len)
{
xmlNode *xml = string2xml(msg);
const char *task = crm_element_value(xml, F_CRM_TASK);
crm_trace("Received CPG message (%s): %.200s",
(task? task : "process list"), (char*)msg);
if (task == NULL) {
if (nodeid == local_nodeid) {
crm_debug("Ignoring message with local node's process list");
} else {
uint32_t procs = 0;
const char *uname = crm_element_value(xml, "uname");
crm_element_value_int(xml, "proclist", (int *)&procs);
if (update_node_processes(nodeid, uname, procs)) {
update_process_clients(NULL);
}
}
} else if (crm_str_eq(task, CRM_OP_RM_NODE_CACHE, TRUE)) {
int id = 0;
const char *name = NULL;
crm_element_value_int(xml, XML_ATTR_ID, &id);
name = crm_element_value(xml, XML_ATTR_UNAME);
reap_crm_member(id, name);
}
if (xml != NULL) {
free_xml(xml);
}
}
static void
mcp_cpg_membership(cpg_handle_t handle,
const struct cpg_name *groupName,
const struct cpg_address *member_list, size_t member_list_entries,
const struct cpg_address *left_list, size_t left_list_entries,
const struct cpg_address *joined_list, size_t joined_list_entries)
{
/* Update peer cache if needed */
pcmk_cpg_membership(handle, groupName, member_list, member_list_entries,
left_list, left_list_entries,
joined_list, joined_list_entries);
/* Always broadcast our own presence after any membership change */
update_process_peers();
}
static gboolean
mcp_quorum_callback(unsigned long long seq, gboolean quorate)
{
pcmk_quorate = quorate;
return TRUE;
}
static void
mcp_quorum_destroy(gpointer user_data)
{
crm_info("connection lost");
}
int
main(int argc, char **argv)
{
int rc;
int flag;
int argerr = 0;
int option_index = 0;
gboolean shutdown = FALSE;
uid_t pcmk_uid = 0;
gid_t pcmk_gid = 0;
struct rlimit cores;
crm_ipc_t *old_instance = NULL;
qb_ipcs_service_t *ipcs = NULL;
static crm_cluster_t cluster;
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "mode [options]", long_options, "Start/Stop Pacemaker\n");
mainloop_add_signal(SIGHUP, pcmk_ignore);
mainloop_add_signal(SIGQUIT, pcmk_sigquit);
while (1) {
flag = crm_get_option(argc, argv, &option_index);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'f':
/* Legacy */
break;
case 'p':
pid_file = optarg;
break;
case 's':
pcmk__set_env_option("node_start_state", "standby");
break;
case '$':
case '?':
crm_help(flag, CRM_EX_OK);
break;
case 'S':
shutdown = TRUE;
break;
case 'F':
printf("Pacemaker %s (Build: %s)\n Supporting v%s: %s\n", PACEMAKER_VERSION, BUILD_VERSION,
CRM_FEATURE_SET, CRM_FEATURES);
crm_exit(CRM_EX_OK);
default:
printf("Argument code 0%o (%c) is not (?yet?) supported\n", flag, flag);
++argerr;
break;
}
}
if (optind < argc) {
printf("non-option ARGV-elements: ");
while (optind < argc)
printf("%s ", argv[optind++]);
printf("\n");
}
if (argerr) {
crm_help('?', CRM_EX_USAGE);
}
setenv("LC_ALL", "C", 1);
pcmk__set_env_option("mcp", "true");
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
crm_debug("Checking for existing Pacemaker instance");
old_instance = crm_ipc_new(CRM_SYSTEM_MCP, 0);
(void) crm_ipc_connect(old_instance);
if (shutdown) {
crm_debug("Shutting down existing Pacemaker instance by request");
while (crm_ipc_connected(old_instance)) {
xmlNode *cmd =
create_request(CRM_OP_QUIT, NULL, NULL, CRM_SYSTEM_MCP, CRM_SYSTEM_MCP, NULL);
crm_debug(".");
crm_ipc_send(old_instance, cmd, 0, 0, NULL);
free_xml(cmd);
sleep(2);
}
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_exit(CRM_EX_OK);
} else if (crm_ipc_connected(old_instance)) {
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
crm_err("Aborting start-up because active Pacemaker instance found");
crm_exit(CRM_EX_FATAL);
}
crm_ipc_close(old_instance);
crm_ipc_destroy(old_instance);
if (mcp_read_config() == FALSE) {
crm_notice("Could not obtain corosync config data, exiting");
crm_exit(CRM_EX_UNAVAILABLE);
}
// OCF shell functions and cluster-glue need facility under different name
{
const char *facility = pcmk__env_option("logfacility");
if (facility && safe_str_neq(facility, "none")) {
setenv("HA_LOGFACILITY", facility, 1);
}
}
crm_notice("Starting Pacemaker %s "CRM_XS" build=%s features:%s",
PACEMAKER_VERSION, BUILD_VERSION, CRM_FEATURES);
mainloop = g_main_loop_new(NULL, FALSE);
rc = getrlimit(RLIMIT_CORE, &cores);
if (rc < 0) {
crm_perror(LOG_ERR, "Cannot determine current maximum core size.");
} else {
if (cores.rlim_max == 0 && geteuid() == 0) {
cores.rlim_max = RLIM_INFINITY;
} else {
crm_info("Maximum core file size is: %lu", (unsigned long)cores.rlim_max);
}
cores.rlim_cur = cores.rlim_max;
rc = setrlimit(RLIMIT_CORE, &cores);
if (rc < 0) {
crm_perror(LOG_ERR,
"Core file generation will remain disabled."
" Core files are an important diagnostic tool, so"
" please consider enabling them by default.");
}
}
if (pcmk_daemon_user(&pcmk_uid, &pcmk_gid) < 0) {
crm_err("Cluster user %s does not exist, aborting Pacemaker startup", CRM_DAEMON_USER);
crm_exit(CRM_EX_NOUSER);
}
// Used by some resource agents
if ((mkdir(CRM_STATE_DIR, 0750) < 0) && (errno != EEXIST)) {
crm_warn("Could not create " CRM_STATE_DIR ": %s", pcmk_strerror(errno));
} else {
mcp_chown(CRM_STATE_DIR, pcmk_uid, pcmk_gid);
}
/* Used to store core/blackbox/scheduler/cib files in */
crm_build_path(CRM_PACEMAKER_DIR, 0750);
mcp_chown(CRM_PACEMAKER_DIR, pcmk_uid, pcmk_gid);
/* Used to store core files in */
crm_build_path(CRM_CORE_DIR, 0750);
mcp_chown(CRM_CORE_DIR, pcmk_uid, pcmk_gid);
/* Used to store blackbox dumps in */
crm_build_path(CRM_BLACKBOX_DIR, 0750);
mcp_chown(CRM_BLACKBOX_DIR, pcmk_uid, pcmk_gid);
// Used to store scheduler inputs in
crm_build_path(PE_STATE_DIR, 0750);
mcp_chown(PE_STATE_DIR, pcmk_uid, pcmk_gid);
/* Used to store the cluster configuration */
crm_build_path(CRM_CONFIG_DIR, 0750);
mcp_chown(CRM_CONFIG_DIR, pcmk_uid, pcmk_gid);
// Don't build CRM_RSCTMP_DIR, pacemaker-execd will do it
ipcs = mainloop_add_ipc_server(CRM_SYSTEM_MCP, QB_IPC_NATIVE, &mcp_ipc_callbacks);
if (ipcs == NULL) {
crm_err("Couldn't start IPC server");
crm_exit(CRM_EX_OSERR);
}
/* Allows us to block shutdown */
if (cluster_connect_cfg(&local_nodeid) == FALSE) {
crm_err("Couldn't connect to Corosync's CFG service");
crm_exit(CRM_EX_PROTOCOL);
}
if(pcmk_locate_sbd() > 0) {
setenv("PCMK_watchdog", "true", 1);
} else {
setenv("PCMK_watchdog", "false", 1);
}
switch (find_and_track_existing_processes()) {
case pcmk_rc_ok:
break;
case pcmk_rc_ipc_unauthorized:
crm_exit(CRM_EX_CANTCREAT);
default:
crm_exit(CRM_EX_FATAL);
};
cluster.destroy = mcp_cpg_destroy;
cluster.cpg.cpg_deliver_fn = mcp_cpg_deliver;
cluster.cpg.cpg_confchg_fn = mcp_cpg_membership;
crm_set_autoreap(FALSE);
rc = pcmk_ok;
if (cluster_connect_cpg(&cluster) == FALSE) {
crm_err("Couldn't connect to Corosync's CPG service");
rc = -ENOPROTOOPT;
} else if (cluster_connect_quorum(mcp_quorum_callback, mcp_quorum_destroy)
== FALSE) {
rc = -ENOTCONN;
} else {
local_name = get_local_node_name();
update_node_processes(local_nodeid, local_name, get_process_list());
mainloop_add_signal(SIGTERM, pcmk_shutdown);
mainloop_add_signal(SIGINT, pcmk_shutdown);
init_children_processes();
crm_notice("Pacemaker daemon successfully started and accepting connections");
g_main_loop_run(mainloop);
}
if (ipcs) {
crm_trace("Closing IPC server");
mainloop_del_ipc_server(ipcs);
ipcs = NULL;
}
g_main_loop_unref(mainloop);
cluster_disconnect_cpg(&cluster);
cluster_disconnect_cfg();
crm_exit(crm_errno2exit(rc));
}
diff --git a/daemons/schedulerd/pacemaker-schedulerd.c b/daemons/schedulerd/pacemaker-schedulerd.c
index 88a17b2fd3..e4cd9a4a7f 100644
--- a/daemons/schedulerd/pacemaker-schedulerd.c
+++ b/daemons/schedulerd/pacemaker-schedulerd.c
@@ -1,360 +1,354 @@
/*
* Copyright 2004-2020 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 <crm/crm.h>
#include <stdio.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <stdlib.h>
#include <errno.h>
#include <fcntl.h>
#include <libxml/parser.h>
#include <crm/common/ipcs_internal.h>
#include <crm/common/mainloop.h>
#include <crm/pengine/internal.h>
#include <pacemaker-internal.h>
#include <crm/msg_xml.h>
#define OPTARGS "hVc"
static GMainLoop *mainloop = NULL;
static qb_ipcs_service_t *ipcs = NULL;
static pe_working_set_t *sched_data_set = NULL;
#define get_series() was_processing_error?1:was_processing_warning?2:3
typedef struct series_s {
const char *name;
const char *param;
int wrap;
} series_t;
series_t series[] = {
{"pe-unknown", "_do_not_match_anything_", -1},
{"pe-error", "pe-error-series-max", -1},
{"pe-warn", "pe-warn-series-max", 200},
{"pe-input", "pe-input-series-max", 400},
};
void pengine_shutdown(int nsig);
static gboolean
process_pe_message(xmlNode *msg, xmlNode *xml_data, pcmk__client_t *sender)
{
static char *last_digest = NULL;
static char *filename = NULL;
const char *sys_to = crm_element_value(msg, F_CRM_SYS_TO);
const char *op = crm_element_value(msg, F_CRM_TASK);
const char *ref = crm_element_value(msg, F_CRM_REFERENCE);
crm_trace("Processing %s op (ref=%s)...", op, ref);
if (op == NULL) {
/* error */
} else if (strcasecmp(op, CRM_OP_HELLO) == 0) {
/* ignore */
} else if (safe_str_eq(crm_element_value(msg, F_CRM_MSG_TYPE), XML_ATTR_RESPONSE)) {
/* ignore */
} else if (sys_to == NULL || strcasecmp(sys_to, CRM_SYSTEM_PENGINE) != 0) {
crm_trace("Bad sys-to %s", crm_str(sys_to));
return FALSE;
} else if (strcasecmp(op, CRM_OP_PECALC) == 0) {
unsigned int seq;
int series_id = 0;
int series_wrap = 0;
char *digest = NULL;
const char *value = NULL;
time_t execution_date = time(NULL);
xmlNode *converted = NULL;
xmlNode *reply = NULL;
gboolean is_repoke = FALSE;
gboolean process = TRUE;
crm_config_error = FALSE;
crm_config_warning = FALSE;
was_processing_error = FALSE;
was_processing_warning = FALSE;
if (sched_data_set == NULL) {
sched_data_set = pe_new_working_set();
CRM_ASSERT(sched_data_set != NULL);
set_bit(sched_data_set->flags, pe_flag_no_counts);
set_bit(sched_data_set->flags, pe_flag_no_compat);
}
digest = calculate_xml_versioned_digest(xml_data, FALSE, FALSE, CRM_FEATURE_SET);
converted = copy_xml(xml_data);
if (cli_config_update(&converted, NULL, TRUE) == FALSE) {
sched_data_set->graph = create_xml_node(NULL, XML_TAG_GRAPH);
crm_xml_add_int(sched_data_set->graph, "transition_id", 0);
crm_xml_add_int(sched_data_set->graph, "cluster-delay", 0);
process = FALSE;
free(digest);
} else if (safe_str_eq(digest, last_digest)) {
crm_info("Input has not changed since last time, not saving to disk");
is_repoke = TRUE;
free(digest);
} else {
free(last_digest);
last_digest = digest;
}
if (process) {
pcmk__schedule_actions(sched_data_set, converted, NULL);
}
series_id = get_series();
series_wrap = series[series_id].wrap;
value = pe_pref(sched_data_set->config_hash, series[series_id].param);
if (value != NULL) {
series_wrap = (int) crm_parse_ll(value, NULL);
if (errno != 0) {
series_wrap = series[series_id].wrap;
}
} else {
crm_config_warn("No value specified for cluster"
" preference: %s", series[series_id].param);
}
if (pcmk__read_series_sequence(PE_STATE_DIR, series[series_id].name,
&seq) != pcmk_rc_ok) {
// @TODO maybe handle errors better ...
seq = 0;
}
crm_trace("Series %s: wrap=%d, seq=%u, pref=%s",
series[series_id].name, series_wrap, seq, value);
sched_data_set->input = NULL;
reply = create_reply(msg, sched_data_set->graph);
CRM_ASSERT(reply != NULL);
if (is_repoke == FALSE) {
free(filename);
filename = pcmk__series_filename(PE_STATE_DIR,
series[series_id].name, seq, true);
}
crm_xml_add(reply, F_CRM_TGRAPH_INPUT, filename);
crm_xml_add_int(reply, "graph-errors", was_processing_error);
crm_xml_add_int(reply, "graph-warnings", was_processing_warning);
crm_xml_add_int(reply, "config-errors", crm_config_error);
crm_xml_add_int(reply, "config-warnings", crm_config_warning);
if (pcmk__ipc_send_xml(sender, 0, reply,
crm_ipc_server_event) != pcmk_rc_ok) {
int graph_file_fd = 0;
char *graph_file = NULL;
umask(S_IWGRP | S_IWOTH | S_IROTH);
graph_file = crm_strdup_printf("%s/pengine.graph.XXXXXX",
PE_STATE_DIR);
graph_file_fd = mkstemp(graph_file);
crm_err("Couldn't send transition graph to peer, writing to %s instead",
graph_file);
crm_xml_add(reply, F_CRM_TGRAPH, graph_file);
write_xml_fd(sched_data_set->graph, graph_file, graph_file_fd, FALSE);
free(graph_file);
free_xml(first_named_child(reply, F_CRM_DATA));
CRM_ASSERT(pcmk__ipc_send_xml(sender, 0, reply,
crm_ipc_server_event) == pcmk_rc_ok);
}
free_xml(reply);
pe_reset_working_set(sched_data_set);
pcmk__log_transition_summary(filename);
if (is_repoke == FALSE && series_wrap != 0) {
unlink(filename);
crm_xml_add_ll(xml_data, "execution-date", (long long) execution_date);
write_xml_file(xml_data, filename, TRUE);
pcmk__write_series_sequence(PE_STATE_DIR, series[series_id].name,
++seq, series_wrap);
} else {
crm_trace("Not writing out %s: %d & %d", filename, is_repoke, series_wrap);
}
free_xml(converted);
}
return TRUE;
}
static int32_t
pe_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid)
{
crm_trace("Connection %p", c);
if (pcmk__new_client(c, uid, gid) == NULL) {
return -EIO;
}
return 0;
}
-static void
-pe_ipc_created(qb_ipcs_connection_t * c)
-{
- crm_trace("Connection %p", c);
-}
-
gboolean process_pe_message(xmlNode *msg, xmlNode *xml_data,
pcmk__client_t *sender);
static int32_t
pe_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size)
{
uint32_t id = 0;
uint32_t flags = 0;
pcmk__client_t *c = pcmk__find_client(qbc);
xmlNode *msg = pcmk__client_data2xml(c, data, size, &id, &flags);
pcmk__ipc_send_ack(c, id, flags, "ack");
if (msg != NULL) {
xmlNode *data_xml = get_message_xml(msg, F_CRM_DATA);
process_pe_message(msg, data_xml, c);
free_xml(msg);
}
return 0;
}
/* Error code means? */
static int32_t
pe_ipc_closed(qb_ipcs_connection_t * c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client == NULL) {
return 0;
}
crm_trace("Connection %p", c);
pcmk__free_client(client);
return 0;
}
static void
pe_ipc_destroy(qb_ipcs_connection_t * c)
{
crm_trace("Connection %p", c);
pe_ipc_closed(c);
}
struct qb_ipcs_service_handlers ipc_callbacks = {
.connection_accept = pe_ipc_accept,
- .connection_created = pe_ipc_created,
+ .connection_created = NULL,
.msg_process = pe_ipc_dispatch,
.connection_closed = pe_ipc_closed,
.connection_destroyed = pe_ipc_destroy
};
/* *INDENT-OFF* */
static struct crm_option long_options[] = {
/* Top-level Options */
{"help", 0, 0, '?', "\tThis text"},
{"verbose", 0, 0, 'V', "\tIncrease debug output"},
{0, 0, 0, 0}
};
/* *INDENT-ON* */
int
main(int argc, char **argv)
{
int flag;
int index = 0;
int argerr = 0;
crm_log_preinit(NULL, argc, argv);
crm_set_options(NULL, "[options]",
long_options, "Daemon for calculating the cluster's response to events");
mainloop_add_signal(SIGTERM, pengine_shutdown);
while (1) {
flag = crm_get_option(argc, argv, &index);
if (flag == -1)
break;
switch (flag) {
case 'V':
crm_bump_log_level(argc, argv);
break;
case 'h': /* Help message */
crm_help('?', CRM_EX_OK);
break;
default:
++argerr;
break;
}
}
if (argc - optind == 1 && safe_str_eq("metadata", argv[optind])) {
pe_metadata();
return CRM_EX_OK;
}
if (optind > argc) {
++argerr;
}
if (argerr) {
crm_help('?', CRM_EX_USAGE);
}
crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE);
crm_notice("Starting Pacemaker scheduler");
if (pcmk__daemon_can_write(PE_STATE_DIR, NULL) == FALSE) {
crm_err("Terminating due to bad permissions on " PE_STATE_DIR);
fprintf(stderr,
"ERROR: Bad permissions on " PE_STATE_DIR " (see logs for details)\n");
fflush(stderr);
return CRM_EX_FATAL;
}
ipcs = mainloop_add_ipc_server(CRM_SYSTEM_PENGINE, QB_IPC_SHM, &ipc_callbacks);
if (ipcs == NULL) {
crm_err("Failed to create IPC server: shutting down and inhibiting respawn");
crm_exit(CRM_EX_FATAL);
}
/* Create the mainloop and run it... */
mainloop = g_main_loop_new(NULL, FALSE);
crm_notice("Pacemaker scheduler successfully started and accepting connections");
g_main_loop_run(mainloop);
pe_free_working_set(sched_data_set);
crm_info("Exiting %s", crm_system_name);
crm_exit(CRM_EX_OK);
}
void
pengine_shutdown(int nsig)
{
mainloop_del_ipc_server(ipcs);
pe_free_working_set(sched_data_set);
crm_exit(CRM_EX_OK);
}
diff --git a/maint/mocked/based.c b/maint/mocked/based.c
index 59cc0d94e6..fa5797f0f5 100644
--- a/maint/mocked/based.c
+++ b/maint/mocked/based.c
@@ -1,338 +1,331 @@
/*
* Copyright 2019-2020 the Pacemaker project contributors
*
* The version control history for this file may have further details.
*
* Licensed under the GNU General Public License version 2 or later (GPLv2+).
*/
/*
* Clean room attempt (admittedly with lot of code borrowed or inspired from
* the full-blown daemon), minimalistic implementation of based daemon, with
* only important aspects being implemented at the moment.
*
* Hopefully easy to adapt for variety of purposes.
*
* NOTE: currently, only cib_rw API end-point is opened, future refinements
* as new modules are added should conditionalize per what the module
* indicates in the context (which is intentionally very loose data glue
* between the skeleton and modules themselves (like CGI variables so
* to say, but more structurally predestined so as to avoid complexities
* of hash table lookups etc.)
*/
#include <crm_internal.h>
#if 0
#include "crm/common/ipcs_internal.h" /* pcmk__client_t */
#include "crm/common/xml.h" /* crm_xml_add */
#endif
#include "crm/msg_xml.h" /* F_SUBTYPE */
#include "daemons/based/pacemaker-based.h" /* cib_notify_diff */
#include <qb/qbipcs.h> /* qb_ipcs_connection_t */
#include "based.h"
/* direct global access violated in one case only
- mock_based_ipc_accept adds a reference to it to crm_cient_t->userdata */
mock_based_context_t mock_based_context;
/* see based/based_callbacks.c:cib_ipc_accept */
static int32_t
mock_based_ipc_accept(qb_ipcs_connection_t *c, uid_t uid, gid_t gid)
{
int32_t ret = 0;
pcmk__client_t *cib_client;
crm_trace("Connection %p", c);
cib_client = pcmk__new_client(c, uid, gid);
if (cib_client == NULL) {
ret = -EIO;
}
cib_client->userdata = &mock_based_context;
return ret;
}
-/* see based/based_callbacks.c:cib_ipc_created */
-static void
-mock_based_ipc_created(qb_ipcs_connection_t *c)
-{
- crm_trace("Connection %p", c);
-}
-
/* see based/based_callbacks.c:cib_ipc_closed */
static int32_t
mock_based_ipc_closed(qb_ipcs_connection_t *c)
{
pcmk__client_t *client = pcmk__find_client(c);
if (client != NULL) {
crm_trace("Connection %p", c);
pcmk__free_client(client);
}
return 0;
}
/* see based/based_callbacks.c:cib_ipc_destroy */
static void
mock_based_ipc_destroy(qb_ipcs_connection_t *c)
{
crm_trace("Connection %p", c);
mock_based_ipc_closed(c);
}
/* see based/based_callbacks.c:cib_process_command (and more) */
static void
mock_based_handle_query(pcmk__client_t *cib_client, uint32_t flags,
const xmlNode *op_request)
{
xmlNode *reply, *cib;
const char cib_str[] =
#if 0
"<cib/>";
#else
"<cib validate-with='pacemaker-1.2' admin_epoch='0' epoch='0' num_updates='0'>"\
" <configuration>"\
" <crm_config/>"\
" <nodes/>"\
" <resources/>"\
" <constraints/>"\
" </configuration>"\
" <status/>"\
"</cib>";
#endif
cib = xmlReadMemory(cib_str, sizeof(cib_str), "file:///tmp/foo", NULL, 0)->children;
reply = create_xml_node(NULL, "cib-reply");
crm_xml_add(reply, F_TYPE, T_CIB);
crm_xml_add(reply, F_CIB_OPERATION,
crm_element_value(op_request, F_CIB_OPERATION));
crm_xml_add(reply, F_CIB_CALLID,
crm_element_value(op_request, F_CIB_CALLID));
crm_xml_add(reply, F_CIB_CLIENTID,
crm_element_value(op_request, F_CIB_CLIENTID));
crm_xml_add_int(reply, F_CIB_CALLOPTS, flags);
crm_xml_add_int(reply, F_CIB_RC, pcmk_ok);
if (cib != NULL) {
crm_trace("Attaching reply output");
add_message_xml(reply, F_CIB_CALLDATA, cib);
}
pcmk__ipc_send_xml(cib_client, cib_client->request_id, reply,
((flags & cib_sync_call)? crm_ipc_flags_none
: crm_ipc_server_event));
free_xml(reply);
free_xml(cib);
}
/* see based/based_callbacks.c:cib_common_callback_worker */
static void
mock_based_common_callback_worker(uint32_t id, uint32_t flags,
xmlNode *op_request,
pcmk__client_t *cib_client)
{
const char *op = crm_element_value(op_request, F_CIB_OPERATION);
mock_based_context_t *ctxt;
if (!strcmp(op, CRM_OP_REGISTER)) {
if (flags & crm_ipc_client_response) {
xmlNode *ack = create_xml_node(NULL, __FUNCTION__);
crm_xml_add(ack, F_CIB_OPERATION, CRM_OP_REGISTER);
crm_xml_add(ack, F_CIB_CLIENTID, cib_client->id);
pcmk__ipc_send_xml(cib_client, id, ack, flags);
cib_client->request_id = 0;
free_xml(ack);
}
} else if (!strcmp(op, T_CIB_NOTIFY)) {
int on_off = 0;
const char *type = crm_element_value(op_request, F_CIB_NOTIFY_TYPE);
crm_element_value_int(op_request, F_CIB_NOTIFY_ACTIVATE, &on_off);
crm_debug("Setting %s callbacks for %s (%s): %s",
type, cib_client->name, cib_client->id, on_off ? "on" : "off");
if (!strcmp(type, T_CIB_DIFF_NOTIFY) && on_off) {
cib_client->options |= cib_notify_diff;
}
ctxt = (mock_based_context_t *) cib_client->userdata;
for (size_t c = ctxt->modules_cnt; c > 0; c--) {
if (ctxt->modules[c - 1]->hooks.cib_notify != NULL) {
ctxt->modules[c - 1]->hooks.cib_notify(cib_client);
}
}
if (flags & crm_ipc_client_response) {
pcmk__ipc_send_ack(cib_client, id, flags, "ack");
}
} else if (!strcmp(op, CIB_OP_QUERY)) {
mock_based_handle_query(cib_client, flags, op_request);
} else {
crm_notice("Discarded request %s", op);
}
}
/* see based/based_callbacks.c:cib_ipc_dispatch_rw */
static int32_t
mock_based_dispatch_command(qb_ipcs_connection_t *c, void *data, size_t size)
{
uint32_t id = 0, flags = 0;
int call_options = 0;
pcmk__client_t *cib_client = pcmk__find_client(c);
xmlNode *op_request = pcmk__client_data2xml(cib_client, data, size, &id,
&flags);
crm_notice("Got connection %p", c);
assert(op_request != NULL);
if (cib_client == NULL || op_request == NULL) {
if (op_request == NULL) {
crm_trace("Invalid message from %p", c);
pcmk__ipc_send_ack(cib_client, id, flags, "nack");
}
return 0;
}
crm_element_value_int(op_request, F_CIB_CALLOPTS, &call_options);
if (call_options & cib_sync_call) {
assert(flags & crm_ipc_client_response);
cib_client->request_id = id; /* reply only to last in-flight request */
}
assert(cib_client->name == NULL);
crm_element_value_int(op_request, F_CIB_CALLOPTS, &call_options);
crm_xml_add(op_request, F_CIB_CLIENTID, cib_client->id);
crm_xml_add(op_request, F_CIB_CLIENTNAME, cib_client->name);
mock_based_common_callback_worker(id, flags, op_request, cib_client);
free_xml(op_request);
return 0;
}
/* * */
size_t mock_based_register_module(module_t mod) {
module_t *module;
size_t ret = mock_based_context.modules_cnt++;
mock_based_context.modules = realloc(mock_based_context.modules,
sizeof(*mock_based_context.modules)
* mock_based_context.modules_cnt);
if (mock_based_context.modules == NULL
|| (module = malloc(sizeof(module_t))) == NULL) {
abort();
}
memcpy(module, &mod, sizeof(mod));
mock_based_context.modules[mock_based_context.modules_cnt - 1] = module;
return ret;
}
static int
mock_based_options(mock_based_context_t *ctxt,
bool usage, int argc, const char *argv[])
{
const char **args2argv;
char *s;
int ret = 0;
if (argc <= 1) {
const char *help_argv[] = {argv[0], "-h"};
return mock_based_options(ctxt, false, 2, (const char **) &help_argv);
}
for (size_t i = 1; i < argc; i++) {
if (argv[i][0] == '-' && argv[i][1] != '-' && argv[i][1] != '\0') {
if (usage) {
printf("\t-%c\t", argv[i][1]);
}
switch(argv[i][1]) {
case 'h':
if (usage) {
printf("show this help message\n");
ret = 1;
} else {
if ((args2argv
= malloc((ctxt->modules_cnt + 2) * sizeof(*args2argv))) == NULL
|| (s
= malloc((ctxt->modules_cnt * 2 + 2) * sizeof(*s))) == NULL) {
return -1;
}
s[0] = 'h';
args2argv[ctxt->modules_cnt + 1] = (char[]){'-', 'h', '\0'};
for (size_t c = ctxt->modules_cnt; c > 0; c--) {
args2argv[c] = (char[]){'-', ctxt->modules[c - 1]->shortopt, '\0'};
s[(ctxt->modules_cnt - i) + 1] = '|';
s[(ctxt->modules_cnt - i) + 2] = ctxt->modules[c - 1]->shortopt;
}
s[ctxt->modules_cnt * 2 + 1] = '\0';
printf("Usage: %s [-{%s}]\n", argv[0], s);
(void) mock_based_options(ctxt, true, 2 + ctxt->modules_cnt, args2argv);
free(args2argv);
free(s);
}
return ret;
default:
for (size_t c = ctxt->modules_cnt; c > 0; c--) {
if (ctxt->modules[c - 1]->shortopt == argv[i][1]) {
ret = ctxt->modules[c - 1]->hooks.argparse(ctxt, usage, argc - i, &argv[i]);
if (ret < 0) {
break;
} else if (ret > 1) {
i += (ret - 1);
}
}
}
if (ret == 0) {
printf("uknown option \"%s\"\n", argv[i]);
}
break;
}
}
}
return ret;
}
int main(int argc, char *argv[])
{
mock_based_context_t *ctxt = &mock_based_context;
if (mock_based_options(ctxt, false, argc, (const char **) argv) > 0) {
struct qb_ipcs_service_handlers cib_ipc_callbacks = {
.connection_accept = mock_based_ipc_accept,
- .connection_created = mock_based_ipc_created,
+ .connection_created = NULL,
.msg_process = mock_based_dispatch_command,
.connection_closed = mock_based_ipc_closed,
.connection_destroyed = mock_based_ipc_destroy,
};
crm_log_preinit(NULL, argc, argv);
crm_log_init(NULL, LOG_DEBUG, false, true, argc, argv, false);
qb_ipcs_service_t *ipcs_command =
mainloop_add_ipc_server(CIB_CHANNEL_RW, QB_IPC_NATIVE,
&cib_ipc_callbacks);
g_main_loop_run(g_main_loop_new(NULL, false));
qb_ipcs_destroy(ipcs_command);
}
for (size_t c = ctxt->modules_cnt; c > 0; c--) {
if (ctxt->modules[c - 1]->hooks.destroy != NULL) {
ctxt->modules[c - 1]->hooks.destroy(ctxt->modules[c - 1]);
}
free(mock_based_context.modules[c - 1]);
}
free(mock_based_context.modules);
}

File Metadata

Mime Type
text/x-diff
Expires
Sat, Jan 25, 11:57 AM (1 d, 19 h)
Storage Engine
blob
Storage Format
Raw Data
Storage Handle
1322470
Default Alt Text
(223 KB)

Event Timeline