action_t *
custom_action(resource_t * rsc, char *key, const char *task,
node_t * on_node, gboolean optional, gboolean save_action,
pe_working_set_t * data_set)
{
action_t *action = NULL;
GListPtr possible_matches = NULL;
CRM_CHECK(key != NULL, return NULL);
CRM_CHECK(task != NULL, return NULL);
if (save_action && rsc != NULL) {
possible_matches = find_actions(rsc->actions, key, on_node);
}
if (possible_matches != NULL) {
if (g_list_length(possible_matches) > 1) {
pe_warn("Action %s for %s on %s exists %d times",
task, rsc ? rsc->id : "<NULL>",
on_node ? on_node->details->uname : "<NULL>", g_list_length(possible_matches));
}
action = g_list_nth_data(possible_matches, 0);
crm_trace("Found existing action (%d) %s for %s on %s",
action->id, task, rsc ? rsc->id : "<NULL>",
on_node ? on_node->details->uname : "<NULL>");
g_list_free(possible_matches);
}
if (action == NULL) {
if (save_action) {
crm_trace("Creating%s action %d: %s for %s on %s",
optional ? "" : " manditory", data_set->action_id, key,
rsc ? rsc->id : "<NULL>", on_node ? on_node->details->uname : "<NULL>");
}
action = calloc(1, sizeof(action_t));
if (save_action) {
action->id = data_set->action_id++;
} else {
action->id = 0;
}
action->rsc = rsc;
CRM_ASSERT(task != NULL);
action->task = strdup(task);
if (on_node) {
action->node = node_copy(on_node);
}
action->uuid = strdup(key);
pe_set_action_bit(action, pe_action_failure_is_fatal);
pe_set_action_bit(action, pe_action_runnable);
if (optional) {
pe_set_action_bit(action, pe_action_optional);
} else {
pe_clear_action_bit(action, pe_action_optional);
}
/*
Implied by calloc()...
action->actions_before = NULL;
action->actions_after = NULL;
action->pseudo = FALSE;
action->dumped = FALSE;
action->processed = FALSE;
action->seen_count = 0;
*/
action->extra = g_hash_table_new_full(crm_str_hash, g_str_equal, free, free);
action->meta = g_hash_table_new_full(crm_str_hash, g_str_equal, free, free);
if (save_action) {
data_set->actions = g_list_prepend(data_set->actions, action);
}
if (rsc != NULL) {
action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE);
unpack_operation(action, action->op_entry, data_set);
if (save_action) {
rsc->actions = g_list_prepend(rsc->actions, action);
}
}
if (save_action) {
crm_trace("Action %d created", action->id);
}
}
if (optional == FALSE) {
crm_trace("Action %d (%s) marked manditory", action->id, action->uuid);
pe_clear_action_bit(action, pe_action_optional);
}
if (rsc != NULL) {
enum action_tasks a_task = text2task(action->task);
int warn_level = LOG_DEBUG_3;
if (save_action) {
warn_level = LOG_WARNING;
}
if (is_set(action->flags, pe_action_have_node_attrs) == FALSE
&& action->node != NULL && action->op_entry != NULL) {
pe_set_action_bit(action, pe_action_have_node_attrs);
unpack_instance_attributes(data_set->input, action->op_entry, XML_TAG_ATTR_SETS,
action->node->details->attrs,
action->extra, NULL, FALSE, data_set->now);
}
if (is_set(action->flags, pe_action_pseudo)) {
/* leave untouched */
} else if (action->node == NULL) {
pe_clear_action_bit(action, pe_action_runnable);
} else if (is_not_set(rsc->flags, pe_rsc_managed)
&& g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL) == NULL) {
crm_debug("Action %s (unmanaged)", action->uuid);
pe_set_action_bit(action, pe_action_optional);
/* action->runnable = FALSE; */
} else if (action->node->details->online == FALSE) {
pe_clear_action_bit(action, pe_action_runnable);
do_crm_log(warn_level, "Action %s on %s is unrunnable (offline)",
action->uuid, action->node->details->uname);
if (is_set(action->rsc->flags, pe_rsc_managed)
&& save_action && a_task == stop_rsc) {
do_crm_log(warn_level, "Marking node %s unclean", action->node->details->uname);
action->node->details->unclean = TRUE;
}
} else if (action->node->details->pending) {
pe_clear_action_bit(action, pe_action_runnable);
do_crm_log(warn_level, "Action %s on %s is unrunnable (pending)",
action->uuid, action->node->details->uname);
} else if (action->needs == rsc_req_nothing) {
crm_trace("Action %s doesnt require anything", action->uuid);
pe_set_action_bit(action, pe_action_runnable);
#if 0
/*
* No point checking this
* - if we dont have quorum we cant stonith anyway
*/
} else if (action->needs == rsc_req_stonith) {
crm_trace("Action %s requires only stonith", action->uuid);
action->runnable = TRUE;
#endif
} else if (is_set(data_set->flags, pe_flag_have_quorum) == FALSE
&& data_set->no_quorum_policy == no_quorum_stop) {
pe_clear_action_bit(action, pe_action_runnable);
crm_debug("%s\t%s (cancelled : quorum)", action->node->details->uname, action->uuid);
} else if (is_set(data_set->flags, pe_flag_have_quorum) == FALSE
&& data_set->no_quorum_policy == no_quorum_freeze) {
crm_trace("Check resource is already active");
if (rsc->fns->active(rsc, TRUE) == FALSE) {
pe_clear_action_bit(action, pe_action_runnable);
crm_debug("%s\t%s (cancelled : quorum freeze)",
action->node->details->uname, action->uuid);
}
} else {
crm_trace("Action %s is runnable", action->uuid);
pe_set_action_bit(action, pe_action_runnable);
}
if (save_action) {
switch (a_task) {
case stop_rsc:
set_bit(rsc->flags, pe_rsc_stopping);
break;
case start_rsc:
clear_bit(rsc->flags, pe_rsc_starting);
if (is_set(action->flags, pe_action_runnable)) {
set_bit(rsc->flags, pe_rsc_starting);
}
break;
default:
break;
}
}
}
free(key);
return action;
}
gboolean
process_graph_event(xmlNode * event, const char *event_node)
{
int rc = -1;
int status = -1;
int callid = -1;
int action = -1;
int target_rc = -1;
int transition_num = -1;
char *update_te_uuid = NULL;
gboolean stop_early = FALSE;
gboolean passed = FALSE;
const char *id = NULL;
const char *desc = NULL;
const char *magic = NULL;
CRM_ASSERT(event != NULL);
/*
<lrm_rsc_op id="rsc_east-05_last_0" operation_key="rsc_east-05_monitor_0" operation="monitor" crm-debug-origin="do_update_resource" crm_feature_set="3.0.6" transition-key="9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" transition-magic="0:7;9:2:7:be2e97d9-05e2-439d-863e-48f7aecab2aa" call-id="17" rc-code="7" op-status="0" interval="0" last-run="1355361636" last-rc-change="1355361636" exec-time="128" queue-time="0" op-digest="c81f5f40b1c9e859c992e800b1aa6972"/>
*/
id = crm_element_value(event, XML_LRM_ATTR_TASK_KEY);
crm_element_value_int(event, XML_LRM_ATTR_RC, &rc);
crm_element_value_int(event, XML_LRM_ATTR_OPSTATUS, &status);
crm_element_value_int(event, XML_LRM_ATTR_CALLID, &callid);
magic = crm_element_value(event, XML_ATTR_TRANSITION_KEY);
if (magic == NULL) {
/* non-change */
return FALSE;
}
if (decode_transition_key(magic, &update_te_uuid, &transition_num, &action, &target_rc) ==
FALSE) {
crm_err("Invalid event %s.%d detected: %s", id, callid, magic);
abort_transition(INFINITY, tg_restart, "Bad event", event);
return FALSE;
}
if (status == PCMK_LRM_OP_PENDING) {
goto bail;
}
if (transition_num == -1) {
desc = "initiated outside of the cluster";
abort_transition(INFINITY, tg_restart, "Unexpected event", event);
} else if (action < 0 || crm_str_eq(update_te_uuid, te_uuid, TRUE) == FALSE) {
desc = "initiated by a different node";
abort_transition(INFINITY, tg_restart, "Foreign event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->id != transition_num) {
desc = "arrived really late";
abort_transition(INFINITY, tg_restart, "Old event", event);
stop_early = TRUE; /* This could be an lrm status refresh */
} else if (transition_graph->complete) {
desc = "arrived late";
abort_transition(INFINITY, tg_restart, "Inactive graph", event);
} else if (match_graph_event(action, event, event_node, status, rc, target_rc) < 0) {
desc = "unknown";
abort_transition(INFINITY, tg_restart, "Unknown event", event);
} else if (rc == target_rc) {
passed = TRUE;
crm_trace("Processed update to %s: %s", id, magic);
}
if (passed == FALSE) {
if (update_failcount(event, event_node, rc, target_rc, transition_num == -1)) {
/* Turns out this wasn't an lrm status refresh update aferall */
stop_early = FALSE;
desc = "failed";
}
crm_info("Detected action (%d.%d) %s.%d=%s: %s", transition_num, action, id, callid,
services_ocf_exitcode_str(rc), desc);
}
bail:
free(update_te_uuid);
return stop_early;
}
static void
convert_ha_field(xmlNode * parent, void *msg_v, int lpc)
{
int type = 0;
const char *name = NULL;
const char *value = NULL;
xmlNode *xml = NULL;
HA_Message *msg = msg_v;
int rc = BZ_OK;
size_t orig_len = 0;
unsigned int used = 0;
char *uncompressed = NULL;
char *compressed = NULL;
int size = orig_len * 10;
CRM_CHECK(parent != NULL, return);
CRM_CHECK(msg != NULL, return);
name = msg->names[lpc];
type = cl_get_type(msg, name);
switch (type) {
case FT_STRUCT:
convert_ha_message(parent, msg->values[lpc], name);
break;
case FT_COMPRESS:
case FT_UNCOMPRESS:
convert_ha_message(parent, cl_get_struct(msg, name), name);
break;
case FT_STRING:
value = msg->values[lpc];
CRM_CHECK(value != NULL, return);
crm_trace("Converting %s/%d/%s", name, type, value[0] == '<' ? "xml" : "field");
if (value[0] != '<') {
crm_xml_add(parent, name, value);
break;
}
/* unpack xml string */
xml = string2xml(value);
if (xml == NULL) {
crm_err("Conversion of field '%s' failed", name);
return;
}
add_node_nocopy(parent, NULL, xml);
break;
case FT_BINARY:
value = cl_get_binary(msg, name, &orig_len);
size = orig_len * 10 + 1; /* +1 because an exact 10x compression factor happens occasionally */
if (orig_len < 3 || value[0] != 'B' || value[1] != 'Z' || value[2] != 'h') {
if (strstr(name, "uuid") == NULL) {
crm_err("Skipping non-bzip binary field: %s", name);
}
return;
}
compressed = calloc(1, orig_len);
memcpy(compressed, value, orig_len);
crm_trace("Trying to decompress %d bytes", (int)orig_len);
retry:
uncompressed = realloc_safe(uncompressed, size);
memset(uncompressed, 0, size);
used = size - 1; /* always leave room for a trailing '\0'
* BZ2_bzBuffToBuffDecompress won't say anything if
* the uncompressed data is exactly 'size' bytes
*/
rc = BZ2_bzBuffToBuffDecompress(uncompressed, &used, compressed, orig_len, 1, 0);
if (rc == BZ_OUTBUFF_FULL) {
size = size * 2;
/* don't try to allocate more memory than we have */
if (size > 0) {
goto retry;
}
}
if (rc != BZ_OK) {
crm_err("Decompression of %s (%d bytes) into %d failed: %d",
name, (int)orig_len, size, rc);
} else if (used >= size) {
CRM_ASSERT(used < size);
} else {
CRM_LOG_ASSERT(uncompressed[used] == 0);
uncompressed[used] = 0;
xml = string2xml(uncompressed);
}
if (xml != NULL) {
add_node_copy(parent, xml);
free_xml(xml);
}
free(uncompressed);
free(compressed);
break;
}
}
GList *
upstart_job_listall(void)
{
GList *units = NULL;
DBusMessageIter args;
DBusMessageIter unit;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
const char *method = "GetAllJobs";
DBusError error;
int lpc = 0;
if (upstart_init() == FALSE) {
return NULL;
}
/*
com.ubuntu.Upstart0_6.GetAllJobs (out <Array of ObjectPath> jobs)
*/
dbus_error_init(&error);
msg = dbus_message_new_method_call(BUS_NAME, // target for the method call
BUS_PATH, // object to call on
UPSTART_06_API, // interface to call on
method); // method name
CRM_ASSERT(msg != NULL);
reply = pcmk_dbus_send_recv(msg, upstart_proxy, &error, DBUS_TIMEOUT_USE_DEFAULT);
dbus_message_unref(msg);
if(error.name) {
crm_err("Call to %s failed: %s", method, error.name);
return NULL;
} else if (!dbus_message_iter_init(reply, &args)) {
crm_err("Call to %s failed: Message has no arguments", method);
dbus_message_unref(reply);
return NULL;
}
if(!pcmk_dbus_type_check(reply, &args, DBUS_TYPE_ARRAY, __FUNCTION__, __LINE__)) {
crm_err("Call to %s failed: Message has invalid arguments", method);
dbus_message_unref(reply);
return NULL;
}
dbus_message_iter_recurse(&args, &unit);
while (dbus_message_iter_get_arg_type (&unit) != DBUS_TYPE_INVALID) {
DBusBasicValue value;
const char *job = NULL;
char *path = NULL;
if(!pcmk_dbus_type_check(reply, &unit, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
continue;
}
dbus_message_iter_get_basic(&unit, &value);
if(value.str) {
int llpc = 0;
path = value.str;
job = value.str;
while (path[llpc] != 0) {
if (path[llpc] == '/') {
job = path + llpc + 1;
}
llpc++;
}
lpc++;
crm_trace("%s -> %s\n", path, job);
units = g_list_append(units, fix_upstart_name(job));
}
dbus_message_iter_next (&unit);
}
dbus_message_unref(reply);
crm_trace("Found %d upstart jobs", lpc);
return units;
}
/*!
* \internal
* \brief handles the recv buffer and parsing out msgs.
* \note new_data is owned by this function once it is passed in.
*/
xmlNode *
crm_remote_parse_buffer(crm_remote_t * remote)
{
xmlNode *xml = NULL;
struct crm_remote_header_v0 *header = crm_remote_header(remote);
if (remote->buffer == NULL || header == NULL) {
return NULL;
}
/* take ownership of the buffer */
remote->buffer_offset = 0;
/* Support compression on the receiving end now, in case we ever want to add it later */
if (header->payload_compressed) {
int rc = 0;
unsigned int size_u = 1 + header->payload_uncompressed;
char *uncompressed = calloc(1, header->payload_offset + size_u);
crm_trace("Decompressing message data %d bytes into %d bytes",
header->payload_compressed, size_u);
rc = BZ2_bzBuffToBuffDecompress(uncompressed + header->payload_offset, &size_u,
remote->buffer + header->payload_offset,
header->payload_compressed, 1, 0);
if (rc != BZ_OK && header->version > REMOTE_MSG_VERSION) {
crm_warn("Couldn't decompress v%d message, we only understand v%d",
header->version, REMOTE_MSG_VERSION);
free(uncompressed);
return NULL;
} else if (rc != BZ_OK) {
crm_err("Decompression failed: %s (%d)", bz2_strerror(rc), rc);
free(uncompressed);
return NULL;
}
CRM_ASSERT(size_u == header->payload_uncompressed);
memcpy(uncompressed, remote->buffer, header->payload_offset); /* Preserve the header */
remote->buffer_size = header->payload_offset + size_u;
free(remote->buffer);
remote->buffer = uncompressed;
header = crm_remote_header(remote);
}
CRM_LOG_ASSERT(remote->buffer[sizeof(struct crm_remote_header_v0) + header->payload_uncompressed - 1] == 0);
xml = string2xml(remote->buffer + header->payload_offset);
if (xml == NULL && header->version > REMOTE_MSG_VERSION) {
crm_warn("Couldn't parse v%d message, we only understand v%d",
header->version, REMOTE_MSG_VERSION);
} else if (xml == NULL) {
crm_err("Couldn't parse: '%.120s'", remote->buffer + header->payload_offset);
}
return xml;
}
static void
operation_finished(mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)
{
svc_action_t *op = mainloop_child_userdata(p);
char *prefix = crm_strdup_printf("%s:%d", op->id, op->pid);
mainloop_clear_child_userdata(p);
op->status = PCMK_LRM_OP_DONE;
CRM_ASSERT(op->pid == pid);
crm_trace("%s %p %p", prefix, op->opaque->stderr_gsource, op->opaque->stdout_gsource);
if (op->opaque->stderr_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
crm_trace("%s dispatching stderr", prefix);
dispatch_stderr(op);
crm_trace("%s: %p", op->id, op->stderr_data);
mainloop_del_fd(op->opaque->stderr_gsource);
op->opaque->stderr_gsource = NULL;
}
if (op->opaque->stdout_gsource) {
/* Make sure we have read everything from the buffer.
* Depending on the priority mainloop gives the fd, operation_finished
* could occur before all the reads are done. Force the read now.*/
crm_trace("%s dispatching stdout", prefix);
dispatch_stdout(op);
crm_trace("%s: %p", op->id, op->stdout_data);
mainloop_del_fd(op->opaque->stdout_gsource);
op->opaque->stdout_gsource = NULL;
}
if (signo) {
if (mainloop_child_timeout(p)) {
crm_warn("%s - timed out after %dms", prefix, op->timeout);
op->status = PCMK_LRM_OP_TIMEOUT;
op->rc = PCMK_OCF_TIMEOUT;
} else {
do_crm_log_unlikely((op->cancel) ? LOG_INFO : LOG_WARNING,
"%s - terminated with signal %d", prefix, signo);
op->status = PCMK_LRM_OP_ERROR;
op->rc = PCMK_OCF_SIGNAL;
}
} else {
op->rc = exitcode;
crm_debug("%s - exited with rc=%d", prefix, exitcode);
}
free(prefix);
prefix = crm_strdup_printf("%s:%d:stderr", op->id, op->pid);
crm_log_output(LOG_NOTICE, prefix, op->stderr_data);
free(prefix);
prefix = crm_strdup_printf("%s:%d:stdout", op->id, op->pid);
crm_log_output(LOG_DEBUG, prefix, op->stdout_data);
free(prefix);
operation_finalize(op);
}
char *
pcmk_message_common_cs(cpg_handle_t handle, uint32_t nodeid, uint32_t pid, void *content,
uint32_t *kind, const char **from)
{
char *data = NULL;
AIS_Message *msg = (AIS_Message *) content;
if(handle) {
/* 'msg' came from CPG not the plugin
* Do filtering and field massaging
*/
uint32_t local_nodeid = get_local_nodeid(handle);
const char *local_name = get_local_node_name();
if (msg->sender.id > 0 && msg->sender.id != nodeid) {
crm_err("Nodeid mismatch from %d.%d: claimed nodeid=%u", nodeid, pid, msg->sender.id);
return NULL;
} else if (msg->host.id != 0 && (local_nodeid != msg->host.id)) {
/* Not for us */
crm_trace("Not for us: %u != %u", msg->host.id, local_nodeid);
return NULL;
} else if (msg->host.size != 0 && safe_str_neq(msg->host.uname, local_name)) {
/* Not for us */
crm_trace("Not for us: %s != %s", msg->host.uname, local_name);
return NULL;
}
msg->sender.id = nodeid;
if (msg->sender.size == 0) {
crm_node_t *peer = crm_get_peer(nodeid, NULL);
if (peer == NULL) {
crm_err("Peer with nodeid=%u is unknown", nodeid);
} else if (peer->uname == NULL) {
crm_err("No uname for peer with nodeid=%u", nodeid);
} else {
crm_notice("Fixing uname for peer with nodeid=%u", nodeid);
msg->sender.size = strlen(peer->uname);
memset(msg->sender.uname, 0, MAX_NAME);
memcpy(msg->sender.uname, peer->uname, msg->sender.size);
}
}
}
crm_trace("Got new%s message (size=%d, %d, %d)",
msg->is_compressed ? " compressed" : "",
ais_data_len(msg), msg->size, msg->compressed_size);
if (kind != NULL) {
*kind = msg->header.id;
}
if (from != NULL) {
*from = msg->sender.uname;
}
if (msg->is_compressed && msg->size > 0) {
int rc = BZ_OK;
char *uncompressed = NULL;
unsigned int new_size = msg->size + 1;
if (check_message_sanity(msg, NULL) == FALSE) {
goto badmsg;
}
crm_trace("Decompressing message data");
uncompressed = calloc(1, new_size);
rc = BZ2_bzBuffToBuffDecompress(uncompressed, &new_size, msg->data, msg->compressed_size, 1, 0);
if (rc != BZ_OK) {
crm_err("Decompression failed: %d", rc);
free(uncompressed);
goto badmsg;
}
CRM_ASSERT(rc == BZ_OK);
CRM_ASSERT(new_size == msg->size);
data = uncompressed;
} else if (check_message_sanity(msg, data) == FALSE) {
goto badmsg;
} else if (safe_str_eq("identify", data)) {
int pid = getpid();
char *pid_s = crm_itoa(pid);
send_cluster_text(crm_class_cluster, pid_s, TRUE, NULL, crm_msg_ais);
free(pid_s);
return NULL;
} else {
data = strdup(msg->data);
}
if (msg->header.id != crm_class_members) {
/* Is this even needed anymore? */
crm_get_peer(msg->sender.id, msg->sender.uname);
}
if (msg->header.id == crm_class_rmpeer) {
uint32_t id = crm_int_helper(data, NULL);
crm_info("Removing peer %s/%u", data, id);
reap_crm_member(id, NULL);
free(data);
return NULL;
#if SUPPORT_PLUGIN
} else if (is_classic_ais_cluster()) {
plugin_handle_membership(msg);
#endif
}
crm_trace("Payload: %.200s", data);
return data;
badmsg:
crm_err("Invalid message (id=%d, dest=%s:%s, from=%s:%s.%d):"
" min=%d, total=%d, size=%d, bz2_size=%d",
msg->id, ais_dest(&(msg->host)), msg_type2text(msg->host.type),
ais_dest(&(msg->sender)), msg_type2text(msg->sender.type),
msg->sender.pid, (int)sizeof(AIS_Message),
msg->header.size, msg->size, msg->compressed_size);
free(data);
return NULL;
}
static char *
convert_non_atomic_uuid(char *old_uuid, resource_t * rsc, gboolean allow_notify,
gboolean free_original)
{
int interval = 0;
char *uuid = NULL;
char *rid = NULL;
char *raw_task = NULL;
int task = no_action;
CRM_ASSERT(rsc);
pe_rsc_trace(rsc, "Processing %s", old_uuid);
if (old_uuid == NULL) {
return NULL;
} else if (strstr(old_uuid, "notify") != NULL) {
goto done; /* no conversion */
} else if (rsc->variant < pe_group) {
goto done; /* no conversion */
}
CRM_ASSERT(parse_op_key(old_uuid, &rid, &raw_task, &interval));
if (interval > 0) {
goto done; /* no conversion */
}
task = text2task(raw_task);
switch (task) {
case stop_rsc:
case start_rsc:
case action_notify:
case action_promote:
case action_demote:
break;
case stopped_rsc:
case started_rsc:
case action_notified:
case action_promoted:
case action_demoted:
task--;
break;
case monitor_rsc:
case shutdown_crm:
case stonith_node:
task = no_action;
break;
default:
crm_err("Unknown action: %s", raw_task);
task = no_action;
break;
}
if (task != no_action) {
if (is_set(rsc->flags, pe_rsc_notify) && allow_notify) {
uuid = generate_notify_key(rid, "confirmed-post", task2text(task + 1));
} else {
uuid = generate_op_key(rid, task2text(task + 1), 0);
}
pe_rsc_trace(rsc, "Converted %s -> %s", old_uuid, uuid);
}
done:
if (uuid == NULL) {
uuid = strdup(old_uuid);
}
if (free_original) {
free(old_uuid);
}
free(raw_task);
free(rid);
return uuid;
}
gboolean
cib_config_changed(xmlNode * last, xmlNode * next, xmlNode ** diff)
{
int lpc = 0, max = 0;
gboolean config_changes = FALSE;
xmlXPathObject *xpathObj = NULL;
CRM_ASSERT(diff != NULL);
if (*diff == NULL && last != NULL && next != NULL) {
*diff = diff_xml_object(last, next, FALSE);
}
if (*diff == NULL) {
goto done;
}
xpathObj = xpath_search(*diff, "//" XML_CIB_TAG_CONFIGURATION);
if (numXpathResults(xpathObj) > 0) {
config_changes = TRUE;
goto done;
}
freeXpathObject(xpathObj);
/*
* Do not check XML_TAG_DIFF_ADDED "//" XML_TAG_CIB
* This always contains every field and would produce a false positive
* every time if the checked value existed
*/
xpathObj = xpath_search(*diff, "//" XML_TAG_DIFF_REMOVED "//" XML_TAG_CIB);
max = numXpathResults(xpathObj);
for (lpc = 0; lpc < max; lpc++) {
xmlNode *top = getXpathResult(xpathObj, lpc);
if (crm_element_value(top, XML_ATTR_GENERATION) != NULL) {
config_changes = TRUE;
goto done;
}
if (crm_element_value(top, XML_ATTR_GENERATION_ADMIN) != NULL) {
config_changes = TRUE;
goto done;
}
if (crm_element_value(top, XML_ATTR_VALIDATION) != NULL) {
config_changes = TRUE;
goto done;
}
if (crm_element_value(top, XML_ATTR_CRM_VERSION) != NULL) {
config_changes = TRUE;
goto done;
}
if (crm_element_value(top, "remote-clear-port") != NULL) {
config_changes = TRUE;
goto done;
}
if (crm_element_value(top, "remote-tls-port") != NULL) {
config_changes = TRUE;
goto done;
}
}
done:
freeXpathObject(xpathObj);
return config_changes;
}
/* A_CIB_STOP, A_CIB_START, A_CIB_RESTART, */
void
do_cib_control(long long action,
enum crmd_fsa_cause cause,
enum crmd_fsa_state cur_state,
enum crmd_fsa_input current_input,
fsa_data_t *msg_data)
{
struct crm_subsystem_s *this_subsys = cib_subsystem;
long long stop_actions = A_CIB_STOP;
long long start_actions = A_CIB_START;
if(action & stop_actions) {
if (fsa_cib_conn->state != cib_disconnected && last_resource_update != 0) {
crm_info("Waiting for resource update %d to complete", last_resource_update);
crmd_fsa_stall(NULL);
return;
}
crm_info("Disconnecting CIB");
clear_bit_inplace(fsa_input_register, R_CIB_CONNECTED);
CRM_ASSERT(fsa_cib_conn != NULL);
fsa_cib_conn->cmds->del_notify_callback(
fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated);
if(fsa_cib_conn->state != cib_disconnected) {
fsa_cib_conn->cmds->set_slave(
fsa_cib_conn, cib_scope_local);
fsa_cib_conn->cmds->signoff(fsa_cib_conn);
}
}
if(action & start_actions) {
int rc = cib_ok;
CRM_ASSERT(fsa_cib_conn != NULL);
if(cur_state == S_STOPPING) {
crm_err("Ignoring request to start %s after shutdown",
this_subsys->name);
return;
}
rc = fsa_cib_conn->cmds->signon(
fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command);
if(rc != cib_ok) {
/* a short wait that usually avoids stalling the FSA */
sleep(1);
rc = fsa_cib_conn->cmds->signon(
fsa_cib_conn, CRM_SYSTEM_CRMD, cib_command);
}
if(rc != cib_ok){
crm_info("Could not connect to the CIB service: %s", cib_error2string(rc));
} else if(cib_ok != fsa_cib_conn->cmds->set_connection_dnotify(
fsa_cib_conn, crmd_cib_connection_destroy)) {
crm_err("Could not set dnotify callback");
} else if(cib_ok != fsa_cib_conn->cmds->add_notify_callback(
fsa_cib_conn, T_CIB_REPLACE_NOTIFY,
do_cib_replaced)) {
crm_err("Could not set CIB notification callback");
} else if(cib_ok != fsa_cib_conn->cmds->add_notify_callback(
fsa_cib_conn, T_CIB_DIFF_NOTIFY, do_cib_updated)) {
crm_err("Could not set CIB notification callback");
} else {
set_bit_inplace(
fsa_input_register, R_CIB_CONNECTED);
}
if(is_set(fsa_input_register, R_CIB_CONNECTED) == FALSE) {
cib_retries++;
crm_warn("Couldn't complete CIB registration %d"
" times... pause and retry",
cib_retries);
if(cib_retries < 30) {
crm_timer_start(wait_timer);
crmd_fsa_stall(NULL);
} else {
crm_err("Could not complete CIB"
" registration %d times..."
" hard error", cib_retries);
register_fsa_error(
C_FSA_INTERNAL, I_ERROR, NULL);
}
} else {
int call_id = 0;
crm_info("CIB connection established");
call_id = fsa_cib_conn->cmds->query(
fsa_cib_conn, NULL, NULL, cib_scope_local);
add_cib_op_callback(fsa_cib_conn, call_id, FALSE, NULL,
revision_check_callback);
cib_retries = 0;
}
}
}
static void
master_promotion_order(resource_t * rsc, pe_working_set_t * data_set)
{
GListPtr gIter = NULL;
node_t *node = NULL;
node_t *chosen = NULL;
clone_variant_data_t *clone_data = NULL;
get_clone_variant_data(clone_data, rsc);
if (clone_data->merged_master_weights) {
return;
}
clone_data->merged_master_weights = TRUE;
crm_trace("Merging weights for %s", rsc->id);
set_bit(rsc->flags, pe_rsc_merging);
gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
crm_trace("Sort index: %s = %d", child->id, child->sort_index);
}
dump_node_scores(LOG_DEBUG_3, rsc, "Before", rsc->allowed_nodes);
gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
chosen = child->fns->location(child, NULL, FALSE);
if (chosen == NULL || child->sort_index < 0) {
crm_trace("Skipping %s", child->id);
continue;
}
node = (node_t *) pe_hash_table_lookup(rsc->allowed_nodes, chosen->details->id);
CRM_ASSERT(node != NULL);
/* adds in master preferences and rsc_location.role=Master */
crm_trace("Adding %s to %s from %s", score2char(child->sort_index), node->details->uname,
child->id);
node->weight = merge_weights(child->sort_index, node->weight);
}
dump_node_scores(LOG_DEBUG_3, rsc, "Middle", rsc->allowed_nodes);
gIter = rsc->rsc_cons;
for (; gIter != NULL; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
/* (re-)adds location preferences of resources that the
* master instance should/must be colocated with
*/
if (constraint->role_lh == RSC_ROLE_MASTER) {
crm_trace("RHS: %s with %s: %d", constraint->rsc_lh->id, constraint->rsc_rh->id,
constraint->score);
rsc->allowed_nodes =
constraint->rsc_rh->cmds->merge_weights(constraint->rsc_rh, rsc->id,
rsc->allowed_nodes,
constraint->node_attribute,
constraint->score / INFINITY,
constraint->score ==
INFINITY ? FALSE : TRUE, FALSE);
}
}
gIter = rsc->rsc_cons_lhs;
for (; gIter != NULL; gIter = gIter->next) {
rsc_colocation_t *constraint = (rsc_colocation_t *) gIter->data;
/* (re-)adds location preferences of resource that wish to be
* colocated with the master instance
*/
if (constraint->role_rh == RSC_ROLE_MASTER) {
crm_trace("LHS: %s with %s: %d", constraint->rsc_lh->id, constraint->rsc_rh->id,
constraint->score);
rsc->allowed_nodes =
constraint->rsc_lh->cmds->merge_weights(constraint->rsc_lh, rsc->id,
rsc->allowed_nodes,
constraint->node_attribute,
constraint->score / INFINITY, TRUE, TRUE);
}
}
gIter = rsc->rsc_tickets;
for (; gIter != NULL; gIter = gIter->next) {
rsc_ticket_t *rsc_ticket = (rsc_ticket_t *) gIter->data;
if (rsc_ticket->role_lh == RSC_ROLE_MASTER && rsc_ticket->ticket->granted == FALSE) {
resource_location(rsc, NULL, -INFINITY, "__stateful_without_ticket__", data_set);
}
}
dump_node_scores(LOG_DEBUG_3, rsc, "After", rsc->allowed_nodes);
/* write them back and sort */
gIter = rsc->children;
for (; gIter != NULL; gIter = gIter->next) {
resource_t *child = (resource_t *) gIter->data;
chosen = child->fns->location(child, NULL, FALSE);
if (is_not_set(child->flags, pe_rsc_managed) && child->next_role == RSC_ROLE_MASTER) {
child->sort_index = INFINITY;
} else if (chosen == NULL || child->sort_index < 0) {
crm_trace("%s: %d", child->id, child->sort_index);
} else {
node = (node_t *) pe_hash_table_lookup(rsc->allowed_nodes, chosen->details->id);
CRM_ASSERT(node != NULL);
child->sort_index = node->weight;
}
crm_trace("Set sort index: %s = %d", child->id, child->sort_index);
}
rsc->children = g_list_sort_with_data(rsc->children, sort_master_instance, data_set);
clear_bit(rsc->flags, pe_rsc_merging);
}
static gboolean
te_fence_node(crm_graph_t *graph, crm_action_t *action)
{
const char *id = NULL;
const char *uuid = NULL;
const char *target = NULL;
const char *type = NULL;
stonith_ops_t * st_op = NULL;
id = ID(action->xml);
target = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
uuid = crm_element_value(action->xml, XML_LRM_ATTR_TARGET_UUID);
type = crm_meta_value(action->params, "stonith_action");
CRM_CHECK(id != NULL,
crm_log_xml_warn(action->xml, "BadAction");
return FALSE);
CRM_CHECK(uuid != NULL,
crm_log_xml_warn(action->xml, "BadAction");
return FALSE);
CRM_CHECK(type != NULL,
crm_log_xml_warn(action->xml, "BadAction");
return FALSE);
CRM_CHECK(target != NULL,
crm_log_xml_warn(action->xml, "BadAction");
return FALSE);
te_log_action(LOG_INFO,
"Executing %s fencing operation (%s) on %s (timeout=%d)",
type, id, target, transition_graph->stonith_timeout);
/* Passing NULL means block until we can connect... */
te_connect_stonith(NULL);
crm_malloc0(st_op, sizeof(stonith_ops_t));
if(safe_str_eq(type, "poweroff")) {
st_op->optype = POWEROFF;
} else {
st_op->optype = RESET;
}
st_op->timeout = transition_graph->stonith_timeout;
st_op->node_name = crm_strdup(target);
st_op->node_uuid = crm_strdup(uuid);
st_op->private_data = generate_transition_key(
transition_graph->id, action->id, 0, te_uuid);
CRM_ASSERT(stonithd_input_IPC_channel() != NULL);
if (ST_OK != stonithd_node_fence( st_op )) {
crm_err("Cannot fence %s: stonithd_node_fence() call failed ",
target);
}
crm_free(st_op->node_name);
crm_free(st_op->node_uuid);
crm_free(st_op->private_data);
crm_free(st_op);
return TRUE;
}
/* A_ELECTION_COUNT */
void
do_election_count_vote(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)
{
struct timeval your_age;
int age;
int election_id = -1;
int log_level = LOG_INFO;
gboolean use_born_on = FALSE;
gboolean done = FALSE;
gboolean we_loose = FALSE;
const char *op = NULL;
const char *vote_from = NULL;
const char *your_version = NULL;
const char *election_owner = NULL;
const char *reason = "unknown";
crm_node_t *our_node = NULL, *your_node = NULL;
ha_msg_input_t *vote = fsa_typed_data(fsa_dt_ha_msg);
static time_t last_election_loss = 0;
/* if the membership copy is NULL we REALLY shouldnt be voting
* the question is how we managed to get here.
*/
CRM_CHECK(msg_data != NULL, return);
CRM_CHECK(crm_peer_cache != NULL, return);
CRM_CHECK(vote != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
CRM_CHECK(vote->msg != NULL, crm_err("Bogus data from %s", msg_data->origin); return);
your_age.tv_sec = 0;
your_age.tv_usec = 0;
op = crm_element_value(vote->msg, F_CRM_TASK);
vote_from = crm_element_value(vote->msg, F_CRM_HOST_FROM);
your_version = crm_element_value(vote->msg, F_CRM_VERSION);
election_owner = crm_element_value(vote->msg, F_CRM_ELECTION_OWNER);
crm_element_value_int(vote->msg, F_CRM_ELECTION_ID, &election_id);
crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_S, (int *)&(your_age.tv_sec));
crm_element_value_int(vote->msg, F_CRM_ELECTION_AGE_US, (int *)&(your_age.tv_usec));
CRM_CHECK(vote_from != NULL, vote_from = fsa_our_uname);
your_node = crm_get_peer(0, vote_from);
our_node = crm_get_peer(0, fsa_our_uname);
if (voted == NULL) {
crm_debug("Created voted hash");
voted = g_hash_table_new_full(crm_str_hash, g_str_equal,
g_hash_destroy_str, g_hash_destroy_str);
}
if (is_heartbeat_cluster()) {
use_born_on = TRUE;
} else if (is_classic_ais_cluster()) {
use_born_on = TRUE;
}
age = crm_compare_age(your_age);
if (cur_state == S_STARTING) {
reason = "Still starting";
we_loose = TRUE;
} else if (our_node == NULL || crm_is_peer_active(our_node) == FALSE) {
reason = "We are not part of the cluster";
log_level = LOG_ERR;
we_loose = TRUE;
} else if (election_id != current_election_id && crm_str_eq(fsa_our_uuid, election_owner, TRUE)) {
log_level = LOG_DEBUG_2;
reason = "Superceeded";
done = TRUE;
} else if (your_node == NULL || crm_is_peer_active(your_node) == FALSE) {
/* Possibly we cached the message in the FSA queue at a point that it wasn't */
reason = "Peer is not part of our cluster";
log_level = LOG_WARNING;
done = TRUE;
} else if (crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update the list of nodes that have voted */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if (crm_str_eq(vote_from, fsa_our_uname, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(vote_from);
CRM_ASSERT(crm_str_eq(fsa_our_uuid, election_owner, TRUE));
/* update ourselves in the list of nodes that have voted */
g_hash_table_replace(voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) < 0) {
reason = "Version";
we_loose = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) > 0) {
reason = "Version";
} else if (age < 0) {
reason = "Uptime";
we_loose = TRUE;
} else if (age > 0) {
reason = "Uptime";
/* TODO: Check for y(our) born < 0 */
} else if (use_born_on && your_node->born < our_node->born) {
reason = "Born";
we_loose = TRUE;
} else if (use_born_on && your_node->born > our_node->born) {
reason = "Born";
} else if (fsa_our_uname == NULL) {
reason = "Unknown host name";
we_loose = TRUE;
} else if (strcasecmp(fsa_our_uname, vote_from) > 0) {
reason = "Host name";
we_loose = TRUE;
} else {
reason = "Host name";
CRM_ASSERT(strcmp(fsa_our_uname, vote_from) != 0);
/* cant happen...
* } else if(strcasecmp(fsa_our_uname, vote_from) == 0) {
*
* default...
* } else { // strcasecmp(fsa_our_uname, vote_from) < 0
* we win
*/
}
if (done) {
do_crm_log(log_level + 1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)",
election_id, current_election_id, election_owner, op, vote_from, reason);
} else if (we_loose) {
xmlNode *novote = create_request(CRM_OP_NOVOTE, NULL, vote_from,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
do_crm_log(log_level, "Election %d (owner: %s) lost: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
update_dc(NULL);
crm_timer_stop(election_timeout);
if (fsa_input_register & R_THE_DC) {
crm_trace("Give up the DC to %s", vote_from);
register_fsa_input(C_FSA_INTERNAL, I_RELEASE_DC, NULL);
} else if (cur_state != S_STARTING) {
crm_trace("We werent the DC anyway");
register_fsa_input(C_FSA_INTERNAL, I_PENDING, NULL);
}
crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner);
crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id);
send_cluster_message(crm_get_peer(0, vote_from), crm_msg_crmd, novote, TRUE);
free_xml(novote);
fsa_cib_conn->cmds->set_slave(fsa_cib_conn, cib_scope_local);
last_election_loss = time(NULL);
} else {
do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)",
election_id, election_owner, op, vote_from, reason);
if (last_election_loss) {
time_t tm_now = time(NULL);
if (tm_now - last_election_loss < (time_t) loss_dampen) {
crm_info("Election %d ignore: We already lost an election less than %ds ago (%s)",
election_id, loss_dampen, ctime(&last_election_loss));
update_dc(NULL);
return;
}
last_election_loss = 0;
}
register_fsa_input(C_FSA_INTERNAL, I_ELECTION, NULL);
g_hash_table_destroy(voted);
voted = NULL;
}
}
/* A_ELECTION_COUNT */
enum election_result
election_count_vote(election_t *e, xmlNode *vote, bool can_win)
{
int age = 0;
int election_id = -1;
int log_level = LOG_INFO;
gboolean use_born_on = FALSE;
gboolean done = FALSE;
gboolean we_loose = FALSE;
const char *op = NULL;
const char *from = NULL;
const char *reason = "unknown";
const char *election_owner = NULL;
crm_node_t *our_node = NULL, *your_node = NULL;
static int election_wins = 0;
xmlNode *novote = NULL;
time_t tm_now = time(NULL);
static time_t expires = 0;
static time_t last_election_loss = 0;
/* if the membership copy is NULL we REALLY shouldn't be voting
* the question is how we managed to get here.
*/
CRM_CHECK(vote != NULL, return election_error);
if(e == NULL) {
crm_info("Not voting in election: not initialized");
return election_lost;
} else if(crm_peer_cache == NULL) {
crm_info("Not voting in election: no peer cache");
return election_lost;
}
op = crm_element_value(vote, F_CRM_TASK);
from = crm_element_value(vote, F_CRM_HOST_FROM);
election_owner = crm_element_value(vote, F_CRM_ELECTION_OWNER);
crm_element_value_int(vote, F_CRM_ELECTION_ID, &election_id);
your_node = crm_get_peer(0, from);
our_node = crm_get_peer(0, e->uname);
if (e->voted == NULL) {
crm_debug("Created voted hash");
e->voted = g_hash_table_new_full(crm_str_hash, g_str_equal,
g_hash_destroy_str, g_hash_destroy_str);
}
if (is_heartbeat_cluster()) {
use_born_on = TRUE;
} else if (is_classic_ais_cluster()) {
use_born_on = TRUE;
}
if(can_win == FALSE) {
reason = "Not eligible";
we_loose = TRUE;
} else if (our_node == NULL || crm_is_peer_active(our_node) == FALSE) {
reason = "We are not part of the cluster";
log_level = LOG_ERR;
we_loose = TRUE;
} else if (election_id != e->count && crm_str_eq(our_node->uuid, election_owner, TRUE)) {
log_level = LOG_TRACE;
reason = "Superseded";
done = TRUE;
} else if (your_node == NULL || crm_is_peer_active(your_node) == FALSE) {
/* Possibly we cached the message in the FSA queue at a point that it wasn't */
reason = "Peer is not part of our cluster";
log_level = LOG_WARNING;
done = TRUE;
} else if (crm_str_eq(op, CRM_OP_NOVOTE, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(from);
CRM_ASSERT(crm_str_eq(our_node->uuid, election_owner, TRUE));
/* update the list of nodes that have voted */
g_hash_table_replace(e->voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else {
struct timeval your_age;
const char *your_version = crm_element_value(vote, F_CRM_VERSION);
int tv_sec = 0;
int tv_usec = 0;
crm_element_value_int(vote, F_CRM_ELECTION_AGE_S, &tv_sec);
crm_element_value_int(vote, F_CRM_ELECTION_AGE_US, &tv_usec);
your_age.tv_sec = tv_sec;
your_age.tv_usec = tv_usec;
age = crm_compare_age(your_age);
if (crm_str_eq(from, e->uname, TRUE)) {
char *op_copy = strdup(op);
char *uname_copy = strdup(from);
CRM_ASSERT(crm_str_eq(our_node->uuid, election_owner, TRUE));
/* update ourselves in the list of nodes that have voted */
g_hash_table_replace(e->voted, uname_copy, op_copy);
reason = "Recorded";
done = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) < 0) {
reason = "Version";
we_loose = TRUE;
} else if (compare_version(your_version, CRM_FEATURE_SET) > 0) {
reason = "Version";
} else if (age < 0) {
reason = "Uptime";
we_loose = TRUE;
} else if (age > 0) {
reason = "Uptime";
/* TODO: Check for y(our) born < 0 */
} else if (use_born_on && your_node->born < our_node->born) {
reason = "Born";
we_loose = TRUE;
} else if (use_born_on && your_node->born > our_node->born) {
reason = "Born";
} else if (e->uname == NULL) {
reason = "Unknown host name";
we_loose = TRUE;
} else if (strcasecmp(e->uname, from) > 0) {
reason = "Host name";
we_loose = TRUE;
} else {
reason = "Host name";
CRM_ASSERT(strcasecmp(e->uname, from) < 0);
/* can't happen...
* } else if(strcasecmp(e->uname, from) == 0) {
*
*/
}
}
if (expires < tm_now) {
election_wins = 0;
expires = tm_now + STORM_INTERVAL;
} else if (done == FALSE && we_loose == FALSE) {
int peers = 1 + g_hash_table_size(crm_peer_cache);
/* If every node has to vote down every other node, thats N*(N-1) total elections
* Allow some leway before _really_ complaining
*/
election_wins++;
if (election_wins > (peers * peers)) {
crm_warn("Election storm detected: %d elections in %d seconds", election_wins,
STORM_INTERVAL);
election_wins = 0;
expires = tm_now + STORM_INTERVAL;
crm_write_blackbox(0, NULL);
}
}
if (done) {
do_crm_log(log_level + 1, "Election %d (current: %d, owner: %s): Processed %s from %s (%s)",
election_id, e->count, election_owner, op, from, reason);
return e->state;
} else if(we_loose == FALSE) {
do_crm_log(log_level, "Election %d (owner: %s) pass: %s from %s (%s)",
election_id, election_owner, op, from, reason);
if (last_election_loss == 0
|| tm_now - last_election_loss > (time_t) loss_dampen) {
last_election_loss = 0;
election_timeout_stop(e);
/* Start a new election by voting down this, and other, peers */
e->state = election_start;
return e->state;
}
crm_info("Election %d ignore: We already lost an election less than %ds ago (%s)",
election_id, loss_dampen, ctime(&last_election_loss));
}
novote = create_request(CRM_OP_NOVOTE, NULL, from,
CRM_SYSTEM_CRMD, CRM_SYSTEM_CRMD, NULL);
do_crm_log(log_level, "Election %d (owner: %s) lost: %s from %s (%s)",
election_id, election_owner, op, from, reason);
election_timeout_stop(e);
crm_xml_add(novote, F_CRM_ELECTION_OWNER, election_owner);
crm_xml_add_int(novote, F_CRM_ELECTION_ID, election_id);
send_cluster_message(your_node, crm_msg_crmd, novote, TRUE);
free_xml(novote);
last_election_loss = tm_now;
e->state = election_lost;
return e->state;
}
const char *
crm_ipc_name(crm_ipc_t * client)
{
CRM_ASSERT(client != NULL);
return client->name;
}
/* 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;
crm_client_t *c = crm_client_get(qbc);
const char *op = NULL;
if (c == NULL) {
crm_info("Invalid client: %p", qbc);
return 0;
}
request = crm_ipcs_recv(c, data, size, &id, &flags);
if (request == NULL) {
crm_ipcs_send_ack(c, id, flags, "nack", __FUNCTION__, __LINE__);
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, crm_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 %u/%u for command %u from %s", flags, call_options, id, crm_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, crm_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;
}
ssize_t
crm_ipc_prepare(uint32_t request, xmlNode * message, struct iovec ** result, uint32_t max_send_size)
{
static unsigned int biggest = 0;
struct iovec *iov;
unsigned int total = 0;
char *compressed = NULL;
char *buffer = dump_xml_unformatted(message);
struct crm_ipc_response_header *header = calloc(1, sizeof(struct crm_ipc_response_header));
CRM_ASSERT(result != NULL);
crm_ipc_init();
if (max_send_size == 0) {
max_send_size = ipc_buffer_max;
}
CRM_LOG_ASSERT(max_send_size != 0);
*result = NULL;
iov = calloc(2, sizeof(struct iovec));
iov[0].iov_len = hdr_offset;
iov[0].iov_base = header;
header->version = PCMK_IPC_VERSION;
header->size_uncompressed = 1 + strlen(buffer);
total = iov[0].iov_len + header->size_uncompressed;
if (total < max_send_size) {
iov[1].iov_base = buffer;
iov[1].iov_len = header->size_uncompressed;
} else {
unsigned int new_size = 0;
if (crm_compress_string
(buffer, header->size_uncompressed, max_send_size, &compressed, &new_size)) {
header->flags |= crm_ipc_compressed;
header->size_compressed = new_size;
iov[1].iov_len = header->size_compressed;
iov[1].iov_base = compressed;
free(buffer);
biggest = QB_MAX(header->size_compressed, biggest);
} else {
ssize_t rc = -EMSGSIZE;
crm_log_xml_trace(message, "EMSGSIZE");
biggest = QB_MAX(header->size_uncompressed, biggest);
crm_err
("Could not compress the message (%u bytes) into less than the configured ipc limit (%u bytes). "
"Set PCMK_ipc_buffer to a higher value (%u bytes suggested)",
header->size_uncompressed, max_send_size, 4 * biggest);
free(compressed);
free(buffer);
free(header);
free(iov);
return rc;
}
}
header->qb.size = iov[0].iov_len + iov[1].iov_len;
header->qb.id = (int32_t)request; /* Replying to a specific request */
*result = iov;
CRM_ASSERT(header->qb.size > 0);
return header->qb.size;
}
/* coverity[-alloc] Memory is referenced in one or both hashtables */
crm_node_t *
crm_get_peer(unsigned int id, const char *uname)
{
crm_node_t *node = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
crm_peer_init();
if (node == NULL && uname != NULL) {
node = g_hash_table_lookup(crm_peer_cache, uname);
}
if (node == NULL && id > 0) {
node = g_hash_table_lookup(crm_peer_id_cache, GUINT_TO_POINTER(id));
if (node && node->uname && uname) {
crm_crit("Node %s and %s share the same cluster node id '%u'!", node->uname, uname, id);
/* NOTE: Calling crm_new_peer() means the entry in
* crm_peer_id_cache will point to the new entity
*
* TO-DO: Replace the old uname instead?
*/
node = NULL;
}
}
if (node == NULL) {
crm_debug("Creating entry for node %s/%u", uname, id);
node = calloc(1, sizeof(crm_node_t));
CRM_ASSERT(node);
}
if (id > 0 && node->id != id) {
node->id = id;
crm_info("Node %s now has id: %u", crm_str(uname), id);
g_hash_table_replace(crm_peer_id_cache, GUINT_TO_POINTER(node->id), node);
}
if (uname && node->uname == NULL) {
node->uname = strdup(uname);
crm_info("Node %u is now known as %s", id, uname);
g_hash_table_replace(crm_peer_cache, node->uname, node);
if (crm_status_callback) {
crm_status_callback(crm_status_uname, node, NULL);
}
}
if (node && node->uname && node->uuid == NULL) {
const char *uuid = get_node_uuid(id, node->uname);
if(uuid) {
node->uuid = strdup(uuid);
crm_info("Node %u has uuid %s", id, node->uuid);
} else {
crm_warn("Cannot obtain a UUID for node %d/%s", id, node->uname);
}
}
return node;
}
static void
action_synced_wait(svc_action_t * op, sigset_t mask)
{
#ifndef HAVE_SYS_SIGNALFD_H
CRM_ASSERT(FALSE);
#else
int status = 0;
int timeout = op->timeout;
int sfd = -1;
time_t start = -1;
struct pollfd fds[3];
int wait_rc = 0;
sfd = signalfd(-1, &mask, SFD_NONBLOCK);
if (sfd < 0) {
crm_perror(LOG_ERR, "signalfd() failed");
}
fds[0].fd = op->opaque->stdout_fd;
fds[0].events = POLLIN;
fds[0].revents = 0;
fds[1].fd = op->opaque->stderr_fd;
fds[1].events = POLLIN;
fds[1].revents = 0;
fds[2].fd = sfd;
fds[2].events = POLLIN;
fds[2].revents = 0;
crm_trace("Waiting for %d", op->pid);
start = time(NULL);
do {
int poll_rc = poll(fds, 3, timeout);
if (poll_rc > 0) {
if (fds[0].revents & POLLIN) {
svc_read_output(op->opaque->stdout_fd, op, FALSE);
}
if (fds[1].revents & POLLIN) {
svc_read_output(op->opaque->stderr_fd, op, TRUE);
}
if (fds[2].revents & POLLIN) {
struct signalfd_siginfo fdsi;
ssize_t s;
s = read(sfd, &fdsi, sizeof(struct signalfd_siginfo));
if (s != sizeof(struct signalfd_siginfo)) {
crm_perror(LOG_ERR, "Read from signal fd %d failed", sfd);
} else if (fdsi.ssi_signo == SIGCHLD) {
wait_rc = waitpid(op->pid, &status, WNOHANG);
if (wait_rc < 0){
crm_perror(LOG_ERR, "waitpid() for %d failed", op->pid);
} else if (wait_rc > 0) {
break;
}
}
}
} else if (poll_rc == 0) {
timeout = 0;
break;
} else if (poll_rc < 0) {
if (errno != EINTR) {
crm_perror(LOG_ERR, "poll() failed");
break;
}
}
timeout = op->timeout - (time(NULL) - start) * 1000;
} while ((op->timeout < 0 || timeout > 0));
crm_trace("Child done: %d", op->pid);
if (wait_rc <= 0) {
int killrc = kill(op->pid, SIGKILL);
op->rc = PCMK_OCF_UNKNOWN_ERROR;
if (op->timeout > 0 && timeout <= 0) {
op->status = PCMK_LRM_OP_TIMEOUT;
crm_warn("%s:%d - timed out after %dms", op->id, op->pid, op->timeout);
} else {
op->status = PCMK_LRM_OP_ERROR;
}
if (killrc && errno != ESRCH) {
crm_err("kill(%d, KILL) failed: %d", op->pid, errno);
}
/*
* From sigprocmask(2):
* It is not possible to block SIGKILL or SIGSTOP. Attempts to do so are silently ignored.
*
* This makes it safe to skip WNOHANG here
*/
waitpid(op->pid, &status, 0);
} else if (WIFEXITED(status)) {
op->status = PCMK_LRM_OP_DONE;
op->rc = WEXITSTATUS(status);
crm_info("Managed %s process %d exited with rc=%d", op->id, op->pid, op->rc);
} else if (WIFSIGNALED(status)) {
int signo = WTERMSIG(status);
op->status = PCMK_LRM_OP_ERROR;
crm_err("Managed %s process %d exited with signal=%d", op->id, op->pid, signo);
}
#ifdef WCOREDUMP
if (WCOREDUMP(status)) {
crm_err("Managed %s process %d dumped core", op->id, op->pid);
}
#endif
svc_read_output(op->opaque->stdout_fd, op, FALSE);
svc_read_output(op->opaque->stderr_fd, op, TRUE);
close(op->opaque->stdout_fd);
close(op->opaque->stderr_fd);
close(sfd);
#endif
}
/*
* Useful Examples:
* http://www.kernel.org/pub/linux/libs/pam/Linux-PAM-html
* http://developer.apple.com/samplecode/CryptNoMore/index.html
*/
static int
construct_pam_passwd(int num_msg, const struct pam_message **msg,
struct pam_response **response, void *data)
{
int count = 0;
struct pam_response *reply;
char *string = (char *)data;
CRM_CHECK(data, return PAM_CONV_ERR);
CRM_CHECK(num_msg == 1, return PAM_CONV_ERR); /* We only want to handle one message */
reply = calloc(1, sizeof(struct pam_response));
CRM_ASSERT(reply != NULL);
for (count = 0; count < num_msg; ++count) {
switch (msg[count]->msg_style) {
case PAM_TEXT_INFO:
crm_info("PAM: %s\n", msg[count]->msg);
break;
case PAM_PROMPT_ECHO_OFF:
case PAM_PROMPT_ECHO_ON:
reply[count].resp_retcode = 0;
reply[count].resp = string; /* We already made a copy */
case PAM_ERROR_MSG:
/* In theory we'd want to print this, but then
* we see the password prompt in the logs
*/
/* crm_err("PAM error: %s\n", msg[count]->msg); */
break;
default:
crm_err("Unhandled conversation type: %d", msg[count]->msg_style);
goto bail;
}
}
*response = reply;
reply = NULL;
return PAM_SUCCESS;
bail:
for (count = 0; count < num_msg; ++count) {
if (reply[count].resp != NULL) {
switch (msg[count]->msg_style) {
case PAM_PROMPT_ECHO_ON:
case PAM_PROMPT_ECHO_OFF:
/* Erase the data - it contained a password */
while (*(reply[count].resp)) {
*(reply[count].resp)++ = '\0';
}
free(reply[count].resp);
break;
}
reply[count].resp = NULL;
}
}
free(reply);
reply = NULL;
return PAM_CONV_ERR;
}
/*!
* \internal
* \brief Return whether a directory or file is writable by a user/group
*
* \param[in] dir Directory to check or that contains file
* \param[in] file File name to check (or NULL to check directory)
* \param[in] user Name of user that should have write permission
* \param[in] group Name of group that should have write permission
* \param[in] need_both Whether both user and group must be able to write
*
* \return TRUE if permissions match, FALSE if they don't or on error
*/
gboolean
crm_is_writable(const char *dir, const char *file,
const char *user, const char *group, gboolean need_both)
{
int s_res = -1;
struct stat buf;
char *full_file = NULL;
const char *target = NULL;
gboolean pass = TRUE;
gboolean readwritable = FALSE;
CRM_ASSERT(dir != NULL);
if (file != NULL) {
full_file = crm_concat(dir, file, '/');
target = full_file;
s_res = stat(full_file, &buf);
if (s_res == 0 && S_ISREG(buf.st_mode) == FALSE) {
crm_err("%s must be a regular file", target);
pass = FALSE;
goto out;
}
}
if (s_res != 0) {
target = dir;
s_res = stat(dir, &buf);
if (s_res != 0) {
crm_err("%s must exist and be a directory", dir);
pass = FALSE;
goto out;
} else if (S_ISDIR(buf.st_mode) == FALSE) {
crm_err("%s must be a directory", dir);
pass = FALSE;
}
}
if (user) {
struct passwd *sys_user = NULL;
sys_user = getpwnam(user);
readwritable = (sys_user != NULL
&& buf.st_uid == sys_user->pw_uid && (buf.st_mode & (S_IRUSR | S_IWUSR)));
if (readwritable == FALSE) {
crm_err("%s must be owned and r/w by user %s", target, user);
if (need_both) {
pass = FALSE;
}
}
}
if (group) {
struct group *sys_grp = getgrnam(group);
readwritable = (sys_grp != NULL
&& buf.st_gid == sys_grp->gr_gid && (buf.st_mode & (S_IRGRP | S_IWGRP)));
if (readwritable == FALSE) {
if (need_both || user == NULL) {
pass = FALSE;
crm_err("%s must be owned and r/w by group %s", target, group);
} else {
crm_warn("%s should be owned and r/w by group %s", target, group);
}
}
}
out:
free(full_file);
return pass;
}
/* coverity[-alloc] Memory is referenced in one or both hashtables */
crm_node_t *
crm_get_peer(unsigned int id, const char *uname)
{
GHashTableIter iter;
crm_node_t *node = NULL;
crm_node_t *by_id = NULL;
crm_node_t *by_name = NULL;
CRM_ASSERT(id > 0 || uname != NULL);
crm_peer_init();
if (uname != NULL) {
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if(node->uname && strcasecmp(node->uname, uname) == 0) {
crm_trace("Name match: %s = %p", node->uname, node);
by_name = node;
break;
}
}
}
if (id > 0) {
g_hash_table_iter_init(&iter, crm_peer_cache);
while (g_hash_table_iter_next(&iter, NULL, (gpointer *) &node)) {
if(node->id == id) {
crm_trace("ID match: %u = %p", node->id, node);
by_id = node;
break;
}
}
}
node = by_id; /* Good default */
if(by_id == by_name) {
/* Nothing to do if they match (both NULL counts) */
crm_trace("Consistent: %p for %u/%s", by_id, id, uname);
} else if(by_id == NULL && by_name) {
crm_trace("Only one: %p for %u/%s", by_name, id, uname);
if(id && by_name->id) {
crm_dump_peer_hash(LOG_WARNING, __FUNCTION__);
crm_crit("Node %u and %u share the same name '%s'",
id, by_name->id, uname);
node = NULL; /* Create a new one */
} else {
node = by_name;
}
} else if(by_name == NULL && by_id) {
crm_trace("Only one: %p for %u/%s", by_id, id, uname);
if(uname && by_id->uname) {
crm_dump_peer_hash(LOG_WARNING, __FUNCTION__);
crm_crit("Node '%s' and '%s' share the same cluster nodeid %u: assuming '%s' is correct",
uname, by_id->uname, id, uname);
}
} else if(uname && by_id->uname) {
crm_warn("Node '%s' and '%s' share the same cluster nodeid: %u", by_id->uname, by_name->uname, id);
} else if(id && by_name->id) {
crm_warn("Node %u and %u share the same name: '%s'", by_id->id, by_name->id, uname);
} else {
/* Simple merge */
/* Only corosync based clusters use nodeid's
*
* The functions that call crm_update_peer_state() only know nodeid
* so 'by_id' is authorative when merging
*
* Same for crm_update_peer_proc()
*/
crm_dump_peer_hash(LOG_DEBUG, __FUNCTION__);
crm_info("Merging %p into %p", by_name, by_id);
g_hash_table_foreach_remove(crm_peer_cache, crm_hash_find_by_data, by_name);
}
if (node == NULL) {
char *uniqueid = crm_generate_uuid();
node = calloc(1, sizeof(crm_node_t));
CRM_ASSERT(node);
crm_info("Created entry %s/%p for node %s/%u (%d total)",
uniqueid, node, uname, id, 1 + g_hash_table_size(crm_peer_cache));
g_hash_table_replace(crm_peer_cache, uniqueid, node);
}
if(id > 0 && uname && (node->id == 0 || node->uname == NULL)) {
crm_info("Node %u is now known as %s", id, uname);
}
if(id > 0 && node->id == 0) {
node->id = id;
}
if(uname && node->uname == NULL) {
int lpc, len = strlen(uname);
for (lpc = 0; lpc < len; lpc++) {
if (uname[lpc] >= 'A' && uname[lpc] <= 'Z') {
crm_warn("Node names with capitals are discouraged, consider changing '%s' to something else",
uname);
break;
}
}
node->uname = strdup(uname);
if (crm_status_callback) {
crm_status_callback(crm_status_uname, node, NULL);
}
}
if(node->uuid == NULL) {
const char *uuid = crm_peer_uuid(node);
if (uuid) {
crm_info("Node %u has uuid %s", id, uuid);
} else {
crm_info("Cannot obtain a UUID for node %d/%s", id, node->uname);
}
}
return node;
}
/* For a synchronous 'op', returns FALSE if 'op' fails */
gboolean
upstart_job_exec(svc_action_t * op, gboolean synchronous)
{
char *job = NULL;
int arg_wait = TRUE;
const char *arg_env = "pacemaker=1";
const char *action = op->action;
DBusError error;
DBusMessage *msg = NULL;
DBusMessage *reply = NULL;
DBusMessageIter iter, array_iter;
op->rc = PCMK_OCF_UNKNOWN_ERROR;
CRM_ASSERT(upstart_init());
if (safe_str_eq(op->action, "meta-data")) {
op->stdout_data = upstart_job_metadata(op->agent);
op->rc = PCMK_OCF_OK;
goto cleanup;
}
if(!upstart_job_by_name(op->agent, &job, op->timeout)) {
crm_debug("Could not obtain job named '%s' to %s", op->agent, action);
if (!g_strcmp0(action, "stop")) {
op->rc = PCMK_OCF_OK;
} else {
op->rc = PCMK_OCF_NOT_INSTALLED;
op->status = PCMK_LRM_OP_NOT_INSTALLED;
}
goto cleanup;
}
if (safe_str_eq(op->action, "monitor") || safe_str_eq(action, "status")) {
char *path = get_first_instance(job, op->timeout);
op->rc = PCMK_OCF_NOT_RUNNING;
if(path) {
DBusPendingCall *pending = NULL;
char *state = pcmk_dbus_get_property(
upstart_proxy, BUS_NAME, path, UPSTART_06_API ".Instance", "state",
op->synchronous?NULL:upstart_job_check, op,
op->synchronous?NULL:&pending, op->timeout);
free(job);
free(path);
if(op->synchronous) {
upstart_job_check("state", state, op);
free(state);
return op->rc == PCMK_OCF_OK;
} else if (pending) {
services_set_op_pending(op, pending);
services_add_inflight_op(op);
return TRUE;
}
return FALSE;
}
goto cleanup;
} else if (!g_strcmp0(action, "start")) {
action = "Start";
} else if (!g_strcmp0(action, "stop")) {
action = "Stop";
} else if (!g_strcmp0(action, "restart")) {
action = "Restart";
} else {
op->rc = PCMK_OCF_UNIMPLEMENT_FEATURE;
goto cleanup;
}
crm_debug("Calling %s for %s on %s", action, op->rsc, job);
msg = dbus_message_new_method_call(BUS_NAME, // target for the method call
job, // object to call on
UPSTART_JOB_IFACE, // interface to call on
action); // method name
CRM_ASSERT(msg != NULL);
dbus_message_iter_init_append (msg, &iter);
CRM_LOG_ASSERT(dbus_message_iter_open_container (&iter,
DBUS_TYPE_ARRAY,
DBUS_TYPE_STRING_AS_STRING,
&array_iter));
CRM_LOG_ASSERT(dbus_message_iter_append_basic (&array_iter, DBUS_TYPE_STRING, &arg_env));
CRM_LOG_ASSERT(dbus_message_iter_close_container (&iter, &array_iter));
CRM_LOG_ASSERT(dbus_message_append_args(msg, DBUS_TYPE_BOOLEAN, &arg_wait, DBUS_TYPE_INVALID));
if (op->synchronous == FALSE) {
DBusPendingCall* pending = pcmk_dbus_send(msg, upstart_proxy, upstart_async_dispatch, op, op->timeout);
free(job);
if(pending) {
services_set_op_pending(op, pending);
services_add_inflight_op(op);
return TRUE;
}
return FALSE;
}
dbus_error_init(&error);
reply = pcmk_dbus_send_recv(msg, upstart_proxy, &error, op->timeout);
if(error.name) {
if(!upstart_mask_error(op, error.name)) {
crm_err("Could not issue %s for %s: %s (%s)", action, op->rsc, error.name, job);
}
} else if (!g_strcmp0(op->action, "stop")) {
/* No return vaue */
op->rc = PCMK_OCF_OK;
} else if(!pcmk_dbus_type_check(reply, NULL, DBUS_TYPE_OBJECT_PATH, __FUNCTION__, __LINE__)) {
crm_warn("Call to %s passed but return type was unexpected", op->action);
op->rc = PCMK_OCF_OK;
} else {
const char *path = NULL;
dbus_message_get_args (reply, NULL,
DBUS_TYPE_OBJECT_PATH, &path,
DBUS_TYPE_INVALID);
crm_info("Call to %s passed: %s", op->action, path);
op->rc = PCMK_OCF_OK;
}
cleanup:
free(job);
if(msg) {
dbus_message_unref(msg);
}
if(reply) {
dbus_message_unref(reply);
}
if (op->synchronous == FALSE) {
return operation_finalize(op);
}
return op->rc == PCMK_OCF_OK;
}
crm_node_t *
crm_update_peer(const char *source, unsigned int id, uint64_t born, uint64_t seen, int32_t votes,
uint32_t children, const char *uuid, const char *uname, const char *addr,
const char *state)
{
#if SUPPORT_PLUGIN
gboolean addr_changed = FALSE;
gboolean votes_changed = FALSE;
#endif
crm_node_t *node = NULL;
id = get_corosync_id(id, uuid);
node = crm_get_peer(id, uname);
CRM_ASSERT(node != NULL);
if (node->uuid == NULL) {
if (is_openais_cluster()) {
/* Yes, overrule whatever was passed in */
crm_peer_uuid(node);
} else if (uuid != NULL) {
node->uuid = strdup(uuid);
}
}
if (children > 0) {
crm_update_peer_proc(source, node, children, state);
}
if (state != NULL) {
crm_update_peer_state(source, node, state, seen);
}
#if SUPPORT_HEARTBEAT
if (born != 0) {
node->born = born;
}
#endif
#if SUPPORT_PLUGIN
/* These were only used by the plugin */
if (born != 0) {
node->born = born;
}
if (votes > 0 && node->votes != votes) {
votes_changed = TRUE;
node->votes = votes;
}
if (addr != NULL) {
if (node->addr == NULL || crm_str_eq(node->addr, addr, FALSE) == FALSE) {
addr_changed = TRUE;
free(node->addr);
node->addr = strdup(addr);
}
}
if (addr_changed || votes_changed) {
crm_info("%s: Node %s: id=%u state=%s addr=%s%s votes=%d%s born=" U64T " seen=" U64T
" proc=%.32x", source, node->uname, node->id, node->state,
node->addr, addr_changed ? " (new)" : "", node->votes,
votes_changed ? " (new)" : "", node->born, node->last_seen, node->processes);
}
#endif
return node;
}
/*!
* \internal
* \brief Determine if a remote session has data to read
*
* \retval 0, timeout occured.
* \retval positive, data is ready to be read
* \retval negative, session has ended
*/
int
crm_remote_ready(crm_remote_t * remote, int timeout /* ms */ )
{
struct pollfd fds = { 0, };
int sock = 0;
int rc = 0;
time_t start;
#ifdef HAVE_GNUTLS_GNUTLS_H
if (remote->tls_session) {
void *sock_ptr = gnutls_transport_get_ptr(*remote->tls_session);
sock = GPOINTER_TO_INT(sock_ptr);
} else if (remote->tcp_socket) {
#else
if (remote->tcp_socket) {
#endif
sock = remote->tcp_socket;
} else {
crm_err("Unsupported connection type");
}
if (sock <= 0) {
crm_trace("No longer connected");
return -ENOTCONN;
}
start = time(NULL);
errno = 0;
do {
fds.fd = sock;
fds.events = POLLIN;
/* If we got an EINTR while polling, and we have a
* specific timeout we are trying to honor, attempt
* to adjust the timeout to the closest second. */
if (errno == EINTR && (timeout > 0)) {
timeout = timeout - ((time(NULL) - start) * 1000);
if (timeout < 1000) {
timeout = 1000;
}
}
rc = poll(&fds, 1, timeout);
} while (rc < 0 && errno == EINTR);
return rc;
}
/*!
* \internal
* \brief Read bytes off non blocking remote connection.
*
* \note only use with NON-Blocking sockets. Should only be used after polling socket.
* This function will return once max_size is met, the socket read buffer
* is empty, or an error is encountered.
*
* \retval number of bytes received
*/
static size_t
crm_remote_recv_once(crm_remote_t * remote)
{
int rc = 0;
size_t read_len = sizeof(struct crm_remote_header_v0);
struct crm_remote_header_v0 *header = crm_remote_header(remote);
if(header) {
/* Stop at the end of the current message */
read_len = header->size_total;
}
/* automatically grow the buffer when needed */
if(remote->buffer_size < read_len) {
remote->buffer_size = 2 * read_len;
crm_trace("Expanding buffer to %u bytes", remote->buffer_size);
remote->buffer = realloc_safe(remote->buffer, remote->buffer_size + 1);
CRM_ASSERT(remote->buffer != NULL);
}
#ifdef HAVE_GNUTLS_GNUTLS_H
if (remote->tls_session) {
rc = gnutls_record_recv(*(remote->tls_session),
remote->buffer + remote->buffer_offset,
remote->buffer_size - remote->buffer_offset);
if (rc == GNUTLS_E_INTERRUPTED) {
rc = -EINTR;
} else if (rc == GNUTLS_E_AGAIN) {
rc = -EAGAIN;
} else if (rc < 0) {
crm_debug("TLS receive failed: %s (%d)", gnutls_strerror(rc), rc);
rc = -pcmk_err_generic;
}
} else if (remote->tcp_socket) {
#else
if (remote->tcp_socket) {
#endif
errno = 0;
rc = read(remote->tcp_socket,
remote->buffer + remote->buffer_offset,
remote->buffer_size - remote->buffer_offset);
if(rc < 0) {
rc = -errno;
}
} else {
crm_err("Unsupported connection type");
return -ESOCKTNOSUPPORT;
}
/* process any errors. */
if (rc > 0) {
remote->buffer_offset += rc;
/* always null terminate buffer, the +1 to alloc always allows for this. */
remote->buffer[remote->buffer_offset] = '\0';
crm_trace("Received %u more bytes, %u total", rc, remote->buffer_offset);
} else if (rc == -EINTR || rc == -EAGAIN) {
crm_trace("non-blocking, exiting read: %s (%d)", pcmk_strerror(rc), rc);
} else if (rc == 0) {
crm_debug("EOF encoutered after %u bytes", remote->buffer_offset);
return -ENOTCONN;
} else {
crm_debug("Error receiving message after %u bytes: %s (%d)",
remote->buffer_offset, pcmk_strerror(rc), rc);
return -ENOTCONN;
}
header = crm_remote_header(remote);
if(header) {
if(remote->buffer_offset < header->size_total) {
crm_trace("Read less than the advertised length: %u < %u bytes",
remote->buffer_offset, header->size_total);
} else {
crm_trace("Read full message of %u bytes", remote->buffer_offset);
return remote->buffer_offset;
}
}
return -EAGAIN;
}
/*!
* \internal
* \brief Read data off the socket until at least one full message is present or timeout occures.
* \retval TRUE message read
* \retval FALSE full message not read
*/
gboolean
crm_remote_recv(crm_remote_t * remote, int total_timeout /*ms */ , int *disconnected)
{
int rc;
time_t start = time(NULL);
int remaining_timeout = 0;
if (total_timeout == 0) {
total_timeout = 10000;
} else if (total_timeout < 0) {
total_timeout = 60000;
}
*disconnected = 0;
remaining_timeout = total_timeout;
while ((remaining_timeout > 0) && !(*disconnected)) {
/* read some more off the tls buffer if we still have time left. */
crm_trace("waiting to receive remote msg, starting timeout %d, remaining_timeout %d",
total_timeout, remaining_timeout);
rc = crm_remote_ready(remote, remaining_timeout);
if (rc == 0) {
crm_err("poll timed out (%d ms) while waiting to receive msg", remaining_timeout);
return FALSE;
} else if(rc < 0) {
crm_debug("poll() failed: %s (%d)", pcmk_strerror(rc), rc);
} else {
rc = crm_remote_recv_once(remote);
if(rc > 0) {
return TRUE;
} else if (rc < 0) {
crm_debug("recv() failed: %s (%d)", pcmk_strerror(rc), rc);
}
}
if(rc == -ENOTCONN) {
*disconnected = 1;
return FALSE;
}
remaining_timeout = remaining_timeout - ((time(NULL) - start) * 1000);
}
return FALSE;
}
static void
expand_list(GListPtr list, char **rsc_list, char **node_list)
{
GListPtr gIter = NULL;
const char *uname = NULL;
const char *rsc_id = NULL;
const char *last_rsc_id = NULL;
if (rsc_list) {
*rsc_list = NULL;
}
if (list == NULL) {
if (rsc_list) {
*rsc_list = strdup(" ");
}
if (node_list) {
*node_list = strdup(" ");
}
return;
}
if (node_list) {
*node_list = NULL;
}
for (gIter = list; gIter != NULL; gIter = gIter->next) {
notify_entry_t *entry = (notify_entry_t *) gIter->data;
CRM_LOG_ASSERT(entry != NULL);
CRM_LOG_ASSERT(entry && entry->rsc != NULL);
if(entry == NULL || entry->rsc == NULL) {
continue;
}
/* Uh, why? */
CRM_LOG_ASSERT(node_list == NULL || entry->node != NULL);
if(node_list != NULL && entry->node == NULL) {
continue;
}
uname = NULL;
rsc_id = entry->rsc->id;
CRM_ASSERT(rsc_id != NULL);
/* filter dups */
if (safe_str_eq(rsc_id, last_rsc_id)) {
continue;
}
last_rsc_id = rsc_id;
if (rsc_list != NULL) {
int existing_len = 0;
int len = 2 + strlen(rsc_id); /* +1 space, +1 EOS */
if (rsc_list && *rsc_list) {
existing_len = strlen(*rsc_list);
}
crm_trace("Adding %s (%dc) at offset %d", rsc_id, len - 2, existing_len);
*rsc_list = realloc_safe(*rsc_list, len + existing_len);
sprintf(*rsc_list + existing_len, "%s%s", existing_len == 0 ? "":" ", rsc_id);
}
if (entry->node != NULL) {
uname = entry->node->details->uname;
}
if (node_list != NULL && uname) {
int existing_len = 0;
int len = 2 + strlen(uname);
if (node_list && *node_list) {
existing_len = strlen(*node_list);
}
crm_trace("Adding %s (%dc) at offset %d", uname, len - 2, existing_len);
*node_list = realloc_safe(*node_list, len + existing_len);
sprintf(*node_list + existing_len, "%s%s", existing_len == 0 ? "":" ", uname);
}
}
}
extern int
find_nvpair_attr_delegate(cib_t * the_cib, const char *attr, const char *section,
const char *node_uuid, const char *attr_set_type, const char *set_name,
const char *attr_id, const char *attr_name, gboolean to_console,
char **value, const char *user_name)
{
int offset = 0;
static int xpath_max = 1024;
int rc = pcmk_ok;
char *xpath_string = NULL;
xmlNode *xml_search = NULL;
const char *set_type = NULL;
const char *node_type = NULL;
if (attr_set_type) {
set_type = attr_set_type;
} else {
set_type = XML_TAG_ATTR_SETS;
}
CRM_ASSERT(value != NULL);
*value = NULL;
if (safe_str_eq(section, XML_CIB_TAG_CRMCONFIG)) {
node_uuid = NULL;
set_type = XML_CIB_TAG_PROPSET;
} else if (safe_str_eq(section, XML_CIB_TAG_OPCONFIG)
|| safe_str_eq(section, XML_CIB_TAG_RSCCONFIG)) {
node_uuid = NULL;
set_type = XML_TAG_META_SETS;
} else if (safe_str_eq(section, XML_CIB_TAG_TICKETS)) {
node_uuid = NULL;
section = XML_CIB_TAG_STATUS;
node_type = XML_CIB_TAG_TICKETS;
} else if (node_uuid == NULL) {
return -EINVAL;
}
xpath_string = calloc(1, xpath_max);
offset += snprintf(xpath_string + offset, xpath_max - offset, "%.128s", get_object_path(section));
if (safe_str_eq(node_type, XML_CIB_TAG_TICKETS)) {
offset += snprintf(xpath_string + offset, xpath_max - offset, "//%s", node_type);
} else if (node_uuid) {
const char *node_type = XML_CIB_TAG_NODE;
if (safe_str_eq(section, XML_CIB_TAG_STATUS)) {
node_type = XML_CIB_TAG_STATE;
set_type = XML_TAG_TRANSIENT_NODEATTRS;
}
offset +=
snprintf(xpath_string + offset, xpath_max - offset, "//%s[@id='%s']", node_type,
node_uuid);
}
if (set_name) {
offset +=
snprintf(xpath_string + offset, xpath_max - offset, "//%s[@id='%.128s']", set_type,
set_name);
} else {
offset += snprintf(xpath_string + offset, xpath_max - offset, "//%s", set_type);
}
offset += snprintf(xpath_string + offset, xpath_max - offset, "//nvpair[");
if (attr_id) {
offset += snprintf(xpath_string + offset, xpath_max - offset, "@id='%s'", attr_id);
}
if (attr_name) {
if (attr_id) {
offset += snprintf(xpath_string + offset, xpath_max - offset, " and ");
}
offset += snprintf(xpath_string + offset, xpath_max - offset, "@name='%.128s'", attr_name);
}
offset += snprintf(xpath_string + offset, xpath_max - offset, "]");
CRM_LOG_ASSERT(offset > 0);
rc = cib_internal_op(the_cib, CIB_OP_QUERY, NULL, xpath_string, NULL, &xml_search,
cib_sync_call | cib_scope_local | cib_xpath, user_name);
if (rc != pcmk_ok) {
crm_trace("Query failed for attribute %s (section=%s, node=%s, set=%s, xpath=%s): %s",
attr_name, section, crm_str(node_uuid), crm_str(set_name), xpath_string,
pcmk_strerror(rc));
goto done;
}
crm_log_xml_debug(xml_search, "Match");
if (xml_has_children(xml_search)) {
xmlNode *child = NULL;
rc = -ENOTUNIQ;
attr_msg(LOG_WARNING, "Multiple attributes match name=%s", attr_name);
for (child = __xml_first_child(xml_search); child != NULL; child = __xml_next(child)) {
attr_msg(LOG_INFO, " Value: %s \t(id=%s)",
crm_element_value(child, XML_NVPAIR_ATTR_VALUE), ID(child));
}
} else {
const char *tmp = crm_element_value(xml_search, attr);
if (tmp) {
*value = strdup(tmp);
}
}
done:
free(xpath_string);
free_xml(xml_search);
return rc;
}
const char *
crm_ipc_buffer(crm_ipc_t * client)
{
CRM_ASSERT(client != NULL);
return client->buffer + sizeof(struct crm_ipc_response_header);
}
static gboolean
te_rsc_command(crm_graph_t * graph, crm_action_t * action)
{
/* never overwrite stop actions in the CIB with
* anything other than completed results
*
* Writing pending stops makes it look like the
* resource is running again
*/
xmlNode *cmd = NULL;
xmlNode *rsc_op = NULL;
gboolean rc = TRUE;
gboolean no_wait = FALSE;
gboolean is_local = FALSE;
char *counter = NULL;
const char *task = NULL;
const char *value = NULL;
const char *on_node = NULL;
const char *task_uuid = NULL;
CRM_ASSERT(action != NULL);
CRM_ASSERT(action->xml != NULL);
action->executed = FALSE;
on_node = crm_element_value(action->xml, XML_LRM_ATTR_TARGET);
CRM_CHECK(on_node != NULL && strlen(on_node) != 0,
te_log_action(LOG_ERR, "Corrupted command(id=%s) %s: no node",
ID(action->xml), crm_str(task));
return FALSE);
rsc_op = action->xml;
task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK);
task_uuid = crm_element_value(action->xml, XML_LRM_ATTR_TASK_KEY);
on_node = crm_element_value(rsc_op, XML_LRM_ATTR_TARGET);
counter =
generate_transition_key(transition_graph->id, action->id, get_target_rc(action), te_uuid);
crm_xml_add(rsc_op, XML_ATTR_TRANSITION_KEY, counter);
if (safe_str_eq(on_node, fsa_our_uname)) {
is_local = TRUE;
}
value = crm_meta_value(action->params, XML_ATTR_TE_NOWAIT);
if (crm_is_true(value)) {
no_wait = TRUE;
}
crm_info("Initiating action %d: %s %s on %s%s%s",
action->id, task, task_uuid, on_node,
is_local ? " (local)" : "", no_wait ? " - no waiting" : "");
cmd = create_request(CRM_OP_INVOKE_LRM, rsc_op, on_node,
CRM_SYSTEM_LRMD, CRM_SYSTEM_TENGINE, NULL);
if (is_local) {
/* shortcut local resource commands */
ha_msg_input_t data = {
.msg = cmd,
.xml = rsc_op,
};
fsa_data_t msg = {
.id = 0,
.data = &data,
.data_type = fsa_dt_ha_msg,
.fsa_input = I_NULL,
.fsa_cause = C_FSA_INTERNAL,
.actions = A_LRM_INVOKE,
.origin = __FUNCTION__,
};
do_lrm_invoke(A_LRM_INVOKE, C_FSA_INTERNAL, fsa_state, I_NULL, &msg);
} else {
rc = send_cluster_message(on_node, crm_msg_lrmd, cmd, TRUE);
}
crm_free(counter);
free_xml(cmd);
action->executed = TRUE;
if (rc == FALSE) {
crm_err("Action %d failed: send", action->id);
return FALSE;
} else if (no_wait) {
action->confirmed = TRUE;
update_graph(transition_graph, action);
trigger_graph();
} else {
if (action->timeout <= 0) {
crm_err("Action %d: %s %s on %s had an invalid timeout (%dms). Using %dms instead",
action->id, task, task_uuid, on_node, action->timeout, graph->network_delay);
action->timeout = graph->network_delay;
}
te_start_action_timer(graph, action);
}
value = crm_meta_value(action->params, XML_OP_ATTR_PENDING);
if (crm_is_true(value)) {
/* write a "pending" entry to the CIB, inhibit notification */
crm_info("Recording pending op %s in the CIB", task_uuid);
cib_action_update(action, LRM_OP_PENDING, EXECRA_STATUS_UNKNOWN);
}
return TRUE;
}
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;
#if ENABLE_ACL
xmlNode *filtered_current_cib = NULL;
#endif
int call_type = 0;
int call_options = 0;
int log_level = LOG_DEBUG_4;
const char *op = NULL;
const char *section = NULL;
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;
CRM_ASSERT(cib_status == pcmk_ok);
*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, §ion);
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) {
#if ENABLE_ACL
if (acl_enabled(config_hash) == FALSE
|| acl_filter_cib(request, current_cib, current_cib, &filtered_current_cib) == 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);
} else if (filtered_current_cib == NULL) {
crm_debug("Pre-filtered the entire cib");
rc = -EACCES;
} else {
crm_debug("Pre-filtered the queried cib according to the ACLs");
rc = cib_perform_op(op, call_options, cib_op_func(call_type), TRUE,
section, request, input, FALSE, &config_changed,
filtered_current_cib, &result_cib, NULL, &output);
}
#else
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);
#endif
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) {
manage_counters = FALSE;
call_options |= cib_force_diff;
CRM_CHECK(call_type == 3 || call_type == 4, crm_err("Call type: %d", call_type);
crm_log_xml_err(request, "bad op"));
}
#ifdef SUPPORT_PRENOTIFY
if ((call_options & cib_inhibit_notify) == 0) {
cib_pre_notify(call_options, op, the_cib, input);
}
#endif
if (rc == pcmk_ok) {
if (call_options & cib_inhibit_bcast) {
/* skip */
crm_trace("Skipping update: inhibit broadcast");
manage_counters = FALSE;
}
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 ENABLE_ACL
if (acl_enabled(config_hash) == TRUE
&& acl_check_diff(request, current_cib, result_cib, *cib_diff) == FALSE) {
rc = -EACCES;
}
#endif
if (rc == pcmk_ok && config_changed) {
time_t now;
char *now_str = NULL;
const char *validation = crm_element_value(result_cib, XML_ATTR_VALIDATION);
if (validation) {
int current_version = get_schema_version(validation);
int support_version = get_schema_version("pacemaker-1.1");
/* Once the later schemas support the "update-*" attributes, change "==" to ">=" -- Changed */
if (current_version >= support_version) {
const char *origin = crm_element_value(request, F_ORIG);
crm_xml_replace(result_cib, XML_ATTR_UPDATE_ORIG,
origin ? origin : cib_our_uname);
crm_xml_replace(result_cib, XML_ATTR_UPDATE_CLIENT,
crm_element_value(request, F_CIB_CLIENTNAME));
#if ENABLE_ACL
crm_xml_replace(result_cib, XML_ATTR_UPDATE_USER,
crm_element_value(request, F_CIB_USER));
#endif
}
}
now = time(NULL);
now_str = ctime(&now);
now_str[24] = EOS; /* replace the newline */
crm_xml_replace(result_cib, XML_CIB_ATTR_WRITTEN, now_str);
}
if (manage_counters == FALSE) {
config_changed = cib_config_changed(current_cib, result_cib, cib_diff);
}
/* Always write to disk for replace ops,
* this negates the need to detect ordering changes
*/
if (config_changed == FALSE && crm_str_eq(CIB_OP_REPLACE, op, TRUE)) {
config_changed = TRUE;
}
}
if (rc == pcmk_ok && (call_options & cib_dryrun) == 0) {
rc = activateCibXml(result_cib, config_changed, op);
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;
}
} else if (rc == -pcmk_err_dtd_validation) {
if (output != NULL) {
crm_log_xml_info(output, "cib:output");
free_xml(output);
}
#if ENABLE_ACL
{
xmlNode *filtered_result_cib = NULL;
if (acl_enabled(config_hash) == FALSE
|| acl_filter_cib(request, current_cib, result_cib,
&filtered_result_cib) == FALSE) {
output = result_cib;
} else {
crm_debug("Filtered the result cib for output according to the ACLs");
output = filtered_result_cib;
if (result_cib != NULL) {
free_xml(result_cib);
}
}
}
#else
output = result_cib;
#endif
} else {
free_xml(result_cib);
}
if ((call_options & cib_inhibit_notify) == 0) {
const char *call_id = crm_element_value(request, F_CIB_CALLID);
const char *client = crm_element_value(request, F_CIB_CLIENTNAME);
#ifdef SUPPORT_POSTNOTIFY
cib_post_notify(call_options, op, input, rc, the_cib);
#endif
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);
}
if (rc != pcmk_ok) {
log_level = LOG_DEBUG_4;
if (rc == -pcmk_err_dtd_validation && global_update) {
log_level = LOG_WARNING;
crm_log_xml_info(input, "cib:global_update");
}
} else if (config_changed) {
log_level = LOG_DEBUG_3;
if (cib_is_master) {
log_level = LOG_NOTICE;
}
} else if (cib_is_master) {
log_level = LOG_DEBUG_2;
}
log_xml_diff(log_level, *cib_diff, "cib:diff");
done:
if ((call_options & cib_discard_reply) == 0) {
*reply = cib_construct_reply(request, output, rc);
crm_log_xml_trace(*reply, "cib:reply");
}
#if ENABLE_ACL
if (filtered_current_cib != NULL) {
free_xml(filtered_current_cib);
}
#endif
if (call_type >= 0) {
cib_op_cleanup(call_type, call_options, &input, &output);
}
return rc;
}
