diff --git a/cts/fence_dummy.in b/cts/fence_dummy.in index 00552fea71..18d7699e74 100644 --- a/cts/fence_dummy.in +++ b/cts/fence_dummy.in @@ -1,496 +1,499 @@ #!@PYTHON@ """Dummy fence agent for testing """ __copyright__ = "Copyright 2012-2020 the Pacemaker project contributors" __license__ = "GNU General Public License version 2 or later (GPLv2+) WITHOUT ANY WARRANTY" import io import os import re import sys import time import random import atexit import getopt AGENT_VERSION = "4.0.1" OCF_VERSION = "1.0" SHORT_DESC = "Dummy fence agent" LONG_DESC = """fence_dummy is a fake fencing agent which reports success based on its mode (pass|fail|random) without doing anything.""" # Short options used: difhmnoqsvBDHMRUV ALL_OPT = { "quiet" : { "getopt" : "q", "help" : "", "order" : 50 }, "verbose" : { "getopt" : "v", "longopt" : "verbose", "help" : "-v, --verbose Verbose mode", "required" : "0", "shortdesc" : "Verbose mode", "order" : 51 }, "debug" : { "getopt" : "D:", "longopt" : "debug-file", "help" : "-D, --debug-file=[debugfile] Debugging to output file", "required" : "0", "shortdesc" : "Write debug information to given file", "order" : 52 }, "version" : { "getopt" : "V", "longopt" : "version", "help" : "-V, --version Display version information and exit", "required" : "0", "shortdesc" : "Display version information and exit", "order" : 53 }, "help" : { "getopt" : "h", "longopt" : "help", "help" : "-h, --help Display this help and exit", "required" : "0", "shortdesc" : "Display help and exit", "order" : 54 }, "action" : { "getopt" : "o:", "longopt" : "action", "help" : "-o, --action=[action] Action: status, list, reboot (default), off or on", "required" : "1", "shortdesc" : "Fencing Action", "default" : "reboot", "order" : 1 }, "nodename" : { "getopt" : "N:", "longopt" : "nodename", "help" : "-N, --nodename Node name of fence victim (ignored)", "required" : "0", "shortdesc" : "The node name of fence victim (ignored)", "order" : 2 }, "mode": { "getopt" : "M:", "longopt" : "mode", "required" : "0", "help" : "-M, --mode=(pass|fail|random) Exit status to return for non-monitor operations", "shortdesc" : "Whether fence operations should always pass, always fail, or fail at random", "order" : 3 }, "monitor_mode" : { "getopt" : "m:", "longopt" : "monitor_mode", "help" : "-m, --monitor_mode=(pass|fail|random) Exit status to return for monitor operations", "required" : "0", "shortdesc" : "Whether monitor operations should always pass, always fail, or fail at random", "order" : 3 }, "random_sleep_range": { "getopt" : "R:", "required" : "0", "longopt" : "random_sleep_range", "help" : "-R, --random_sleep_range=[seconds] Sleep between 1 and [seconds] before returning", "shortdesc" : "Wait randomly between 1 and [seconds]", "order" : 3 }, "mock_dynamic_hosts" : { "getopt" : "H:", "longopt" : "mock_dynamic_hosts", "help" : "-H, --mock_dynamic_hosts=[list] What to return when dynamically queried for possible targets", "required" : "0", "shortdesc" : "A list of hosts we can fence", "order" : 3 }, "delay" : { "getopt" : "f:", "longopt" : "delay", "help" : "-f, --delay [seconds] Wait X seconds before fencing is started", "required" : "0", "shortdesc" : "Wait X seconds before fencing is started", "default" : "0", "order" : 3 }, "monitor_delay" : { "getopt" : "d:", "longopt" : "monitor_delay", "help" : "-d, --monitor_delay [seconds] Wait X seconds before monitor completes", "required" : "0", "shortdesc" : "Wait X seconds before monitor completes", "default" : "0", "order" : 3 }, "plug" : { "getopt" : "n:", "longopt" : "plug", "help" : "-n, --plug=[id] Physical plug number on device (ignored)", "required" : "1", "shortdesc" : "Ignored", "order" : 4 }, "port" : { "getopt" : "n:", "longopt" : "plug", "help" : "-n, --plug=[id] Physical plug number on device (ignored)", "required" : "1", "shortdesc" : "Ignored", "order" : 4 }, "switch" : { "getopt" : "s:", "longopt" : "switch", "help" : "-s, --switch=[id] Physical switch number on device (ignored)", "required" : "0", "shortdesc" : "Ignored", "order" : 4 }, "nodeid" : { "getopt" : "i:", "longopt" : "nodeid", "help" : "-i, --nodeid Corosync id of fence victim (ignored)", "required" : "0", "shortdesc" : "Ignored", "order" : 4 }, "uuid" : { "getopt" : "U:", "longopt" : "uuid", "help" : "-U, --uuid UUID of the VM to fence (ignored)", "required" : "0", "shortdesc" : "Ignored", "order" : 4 } } auto_unfence = False no_reboot = False def agent(): """ Return name this file was run as. """ return os.path.basename(sys.argv[0]) def fail_usage(message): """ Print a usage message and exit. """ sys.exit("%s\nPlease use '-h' for usage" % message) def show_docs(options): """ Handle informational options (display info and exit). """ device_opt = options["device_opt"] if "-h" in options: usage(device_opt) sys.exit(0) if "-o" in options and options["-o"].lower() == "metadata": - metadata(device_opt, options) + if not os.path.exists(__file__ + ".fail"): + metadata(device_opt, options) + else: + os.remove(__file__ + ".fail") sys.exit(0) if "-V" in options: print(AGENT_VERSION) sys.exit(0) def sorted_options(avail_opt): """ Return a list of all options, in their internally specified order. """ sorted_list = [(key, ALL_OPT[key]) for key in avail_opt] sorted_list.sort(key=lambda x: x[1]["order"]) return sorted_list def usage(avail_opt): """ Print a usage message. """ print("Usage:") print("\t" + agent() + " [options]") print("Options:") for dummy, value in sorted_options(avail_opt): if len(value["help"]) != 0: print(" " + value["help"]) def metadata(avail_opt, options): """ Print agent metadata. """ # This log is just for testing handling of stderr output print("asked for fence_dummy metadata", file=sys.stderr) print(""" %s %s """ % (agent(), SHORT_DESC, AGENT_VERSION, OCF_VERSION, LONG_DESC)) for option, dummy in sorted_options(avail_opt): if "shortdesc" in ALL_OPT[option]: print(' ') default = "" default_name_arg = "-" + ALL_OPT[option]["getopt"][:-1] default_name_no_arg = "-" + ALL_OPT[option]["getopt"] if "default" in ALL_OPT[option]: default = 'default="%s"' % str(ALL_OPT[option]["default"]) elif default_name_arg in options: if options[default_name_arg]: try: default = 'default="%s"' % options[default_name_arg] except TypeError: ## @todo/@note: Currently there is no clean way how to handle lists ## we can create a string from it but we can't set it on command line default = 'default="%s"' % str(options[default_name_arg]) elif default_name_no_arg in options: default = 'default="true"' mixed = ALL_OPT[option]["help"] ## split it between option and help text res = re.compile(r"^(.*--\S+)\s+", re.IGNORECASE | re.S).search(mixed) if None != res: mixed = res.group(1) mixed = mixed.replace("<", "<").replace(">", ">") print(' ') if ALL_OPT[option]["getopt"].count(":") > 0: print(' ') else: print(' ') print(' ' + ALL_OPT[option]["shortdesc"] + '') print(' ') print(' \n ') if auto_unfence: attr_name = 'automatic' else: attr_name = 'on_target' print(' ') print(' ') if not no_reboot: print(' ') print(' ') print(' ') print(' ') print(' ') print(' ') print('') def option_longopt(option): """ Return the getopt-compatible long-option name of the given option. """ if ALL_OPT[option]["getopt"].endswith(":"): return ALL_OPT[option]["longopt"] + "=" else: return ALL_OPT[option]["longopt"] def opts_from_command_line(argv, avail_opt): """ Read options from command-line arguments. """ # Prepare list of options for getopt getopt_string = "" longopt_list = [] for k in avail_opt: if k in ALL_OPT: getopt_string += ALL_OPT[k]["getopt"] else: fail_usage("Parse error: unknown option '"+k+"'") if k in ALL_OPT and "longopt" in ALL_OPT[k]: longopt_list.append(option_longopt(k)) try: opt, dummy = getopt.gnu_getopt(argv, getopt_string, longopt_list) except getopt.GetoptError as error: fail_usage("Parse error: " + error.msg) # Transform longopt to short one which are used in fencing agents old_opt = opt opt = {} for old_option in dict(old_opt).keys(): if old_option.startswith("--"): for option in ALL_OPT.keys(): if "longopt" in ALL_OPT[option] and "--" + ALL_OPT[option]["longopt"] == old_option: opt["-" + ALL_OPT[option]["getopt"].rstrip(":")] = dict(old_opt)[old_option] else: opt[old_option] = dict(old_opt)[old_option] # Compatibility Layer (with what? probably not needed for fence_dummy) new_opt = dict(opt) if "-T" in new_opt: new_opt["-o"] = "status" if "-n" in new_opt: new_opt["-m"] = new_opt["-n"] opt = new_opt return opt def opts_from_stdin(avail_opt): """ Read options from standard input. """ opt = {} name = "" for line in sys.stdin.readlines(): line = line.strip() if line.startswith("#") or (len(line) == 0): continue (name, value) = (line + "=").split("=", 1) value = value[:-1] # Compatibility Layer (with what? probably not needed for fence_dummy) if name == "option": name = "action" if name not in avail_opt: print("Parse error: Ignoring unknown option '%s'" % line, file=sys.stderr) continue if ALL_OPT[name]["getopt"].endswith(":"): opt["-"+ALL_OPT[name]["getopt"].rstrip(":")] = value elif value.lower() in ["1", "yes", "on", "true"]: opt["-"+ALL_OPT[name]["getopt"]] = "1" return opt def process_input(avail_opt): """ Set standard environment variables, and parse all options. """ # Set standard environment os.putenv("LANG", "C") os.putenv("LC_ALL", "C") # Read options from command line or standard input if len(sys.argv) > 1: return opts_from_command_line(sys.argv[1:], avail_opt) else: return opts_from_stdin(avail_opt) def atexit_handler(): """ Close stdout on exit. """ try: sys.stdout.close() os.close(1) except IOError: sys.exit("%s failed to close standard output" % agent()) def success_mode(options, option, default_value): """ Return exit code specified by option. """ if option in options: test_value = options[option] else: test_value = default_value if test_value == "pass": exitcode = 0 elif test_value == "fail": exitcode = 1 else: exitcode = random.randint(0, 1) return exitcode def write_options(options): """ Write out all options to debug file. """ try: debugfile = io.open(options["-D"], 'at') debugfile.write("### %s ###\n" % (time.strftime("%Y-%m-%d %H:%M:%S"))) for option in sorted(options): debugfile.write("%s=%s\n" % (option, options[option])) debugfile.write("###\n") debugfile.close() except IOError: pass def main(): """ Make it so! """ global auto_unfence global no_reboot # Meta-data can't take parameters, so we simulate different meta-data # behavior based on the executable name (which can be a symbolic link). if (sys.argv[0].endswith("_auto_unfence")): auto_unfence = True elif (sys.argv[0].endswith("_no_reboot")): no_reboot = True device_opt = ALL_OPT.keys() ## Defaults for fence agent atexit.register(atexit_handler) options = process_input(device_opt) options["device_opt"] = device_opt show_docs(options) # dump input to file if "-D" in options: write_options(options) if "-f" in options: val = int(options["-f"]) print("delay sleep for %d seconds" % val, file=sys.stderr) time.sleep(val) # random sleep for testing if "-R" in options: val = int(options["-R"]) ran = random.randint(1, val) print("random sleep for %d seconds" % ran, file=sys.stderr) time.sleep(ran) if "-o" in options: action = options["-o"] else: action = "action" if action == "monitor": if "-d" in options: time.sleep(int(options["-d"])) exitcode = success_mode(options, "-m", "pass") elif action == "list": print("fence_dummy action (list) called", file=sys.stderr) if "-H" in options: print(options["-H"]) exitcode = 0 else: print("dynamic hostlist requires mock_dynamic_hosts to be set", file=sys.stderr) exitcode = 1 else: exitcode = success_mode(options, "-M", "random") # Ensure we generate some error output on failure exit. if exitcode == 1: print("simulated %s failure" % action, file=sys.stderr) sys.exit(exitcode) if __name__ == "__main__": main() diff --git a/daemons/fenced/fenced_commands.c b/daemons/fenced/fenced_commands.c index 41901e515b..65b41c54e4 100644 --- a/daemons/fenced/fenced_commands.c +++ b/daemons/fenced/fenced_commands.c @@ -1,2916 +1,2951 @@ /* * Copyright 2009-2021 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include GHashTable *device_list = NULL; GHashTable *topology = NULL; GList *cmd_list = NULL; struct device_search_s { /* target of fence action */ char *host; /* requested fence action */ char *action; /* timeout to use if a device is queried dynamically for possible targets */ int per_device_timeout; /* number of registered fencing devices at time of request */ int replies_needed; /* number of device replies received so far */ int replies_received; /* whether the target is eligible to perform requested action (or off) */ bool allow_suicide; /* private data to pass to search callback function */ void *user_data; /* function to call when all replies have been received */ void (*callback) (GList * devices, void *user_data); /* devices capable of performing requested action (or off if remapping) */ GListPtr capable; }; static gboolean stonith_device_dispatch(gpointer user_data); static void st_child_done(GPid pid, int rc, const char *output, gpointer user_data); static void stonith_send_reply(xmlNode * reply, int call_options, const char *remote_peer, const char *client_id); static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence); +static xmlNode * get_agent_metadata(const char *agent); +static void read_action_metadata(stonith_device_t *device); + typedef struct async_command_s { int id; int pid; int fd_stdout; int options; int default_timeout; /* seconds */ int timeout; /* seconds */ int start_delay; /* seconds */ int delay_id; char *op; char *origin; char *client; char *client_name; char *remote_op_id; char *victim; uint32_t victim_nodeid; char *action; char *device; char *mode; GListPtr device_list; GListPtr device_next; void *internal_user_data; void (*done_cb) (GPid pid, int rc, const char *output, gpointer user_data); guint timer_sigterm; guint timer_sigkill; /*! If the operation timed out, this is the last signal * we sent to the process to get it to terminate */ int last_timeout_signo; stonith_device_t *active_on; stonith_device_t *activating_on; } async_command_t; static xmlNode *stonith_construct_async_reply(async_command_t * cmd, const char *output, xmlNode * data, int rc); static gboolean is_action_required(const char *action, stonith_device_t *device) { return device && device->automatic_unfencing && pcmk__str_eq(action, "on", pcmk__str_casei); } static int get_action_delay_max(stonith_device_t * device, const char * action) { const char *value = NULL; int delay_max = 0; if (!pcmk__strcase_any_of(action, "off", "reboot", NULL)) { return 0; } value = g_hash_table_lookup(device->params, PCMK_STONITH_DELAY_MAX); if (value) { delay_max = crm_parse_interval_spec(value) / 1000; } return delay_max; } static int get_action_delay_base(stonith_device_t * device, const char * action) { const char *value = NULL; int delay_base = 0; if (!pcmk__strcase_any_of(action, "off", "reboot", NULL)) { return 0; } value = g_hash_table_lookup(device->params, PCMK_STONITH_DELAY_BASE); if (value) { delay_base = crm_parse_interval_spec(value) / 1000; } return delay_base; } /*! * \internal * \brief Override STONITH timeout with pcmk_*_timeout if available * * \param[in] device STONITH device to use * \param[in] action STONITH action name * \param[in] default_timeout Timeout to use if device does not have * a pcmk_*_timeout parameter for action * * \return Value of pcmk_(action)_timeout if available, otherwise default_timeout * \note For consistency, it would be nice if reboot/off/on timeouts could be * set the same way as start/stop/monitor timeouts, i.e. with an * entry in the fencing resource configuration. However that * is insufficient because fencing devices may be registered directly via * the fencer's register_device() API instead of going through the CIB * (e.g. stonith_admin uses it for its -R option, and the executor uses it * to ensure a device is registered when a command is issued). As device * properties, pcmk_*_timeout parameters can be grabbed by the fencer when * the device is registered, whether by CIB change or API call. */ static int get_action_timeout(stonith_device_t * device, const char *action, int default_timeout) { if (action && device && device->params) { char buffer[64] = { 0, }; const char *value = NULL; /* If "reboot" was requested but the device does not support it, * we will remap to "off", so check timeout for "off" instead */ if (pcmk__str_eq(action, "reboot", pcmk__str_casei) && !pcmk_is_set(device->flags, st_device_supports_reboot)) { crm_trace("%s doesn't support reboot, using timeout for off instead", device->id); action = "off"; } /* If the device config specified an action-specific timeout, use it */ snprintf(buffer, sizeof(buffer), "pcmk_%s_timeout", action); value = g_hash_table_lookup(device->params, buffer); if (value) { return atoi(value); } } return default_timeout; } static void free_async_command(async_command_t * cmd) { if (!cmd) { return; } if (cmd->delay_id) { g_source_remove(cmd->delay_id); } cmd_list = g_list_remove(cmd_list, cmd); g_list_free_full(cmd->device_list, free); free(cmd->device); free(cmd->action); free(cmd->victim); free(cmd->remote_op_id); free(cmd->client); free(cmd->client_name); free(cmd->origin); free(cmd->mode); free(cmd->op); free(cmd); } static async_command_t * create_async_command(xmlNode * msg) { async_command_t *cmd = NULL; xmlNode *op = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_ERR); const char *action = crm_element_value(op, F_STONITH_ACTION); CRM_CHECK(action != NULL, crm_log_xml_warn(msg, "NoAction"); return NULL); crm_log_xml_trace(msg, "Command"); cmd = calloc(1, sizeof(async_command_t)); crm_element_value_int(msg, F_STONITH_CALLID, &(cmd->id)); crm_element_value_int(msg, F_STONITH_CALLOPTS, &(cmd->options)); crm_element_value_int(msg, F_STONITH_TIMEOUT, &(cmd->default_timeout)); cmd->timeout = cmd->default_timeout; // Value -1 means disable any static/random fencing delays crm_element_value_int(msg, F_STONITH_DELAY, &(cmd->start_delay)); cmd->origin = crm_element_value_copy(msg, F_ORIG); cmd->remote_op_id = crm_element_value_copy(msg, F_STONITH_REMOTE_OP_ID); cmd->client = crm_element_value_copy(msg, F_STONITH_CLIENTID); cmd->client_name = crm_element_value_copy(msg, F_STONITH_CLIENTNAME); cmd->op = crm_element_value_copy(msg, F_STONITH_OPERATION); cmd->action = strdup(action); cmd->victim = crm_element_value_copy(op, F_STONITH_TARGET); cmd->mode = crm_element_value_copy(op, F_STONITH_MODE); cmd->device = crm_element_value_copy(op, F_STONITH_DEVICE); CRM_CHECK(cmd->op != NULL, crm_log_xml_warn(msg, "NoOp"); free_async_command(cmd); return NULL); CRM_CHECK(cmd->client != NULL, crm_log_xml_warn(msg, "NoClient")); cmd->done_cb = st_child_done; cmd_list = g_list_append(cmd_list, cmd); return cmd; } static int get_action_limit(stonith_device_t * device) { const char *value = NULL; int action_limit = 1; value = g_hash_table_lookup(device->params, PCMK_STONITH_ACTION_LIMIT); if (value) { action_limit = crm_parse_int(value, "1"); if (action_limit == 0) { /* pcmk_action_limit should not be 0. Enforce it to be 1. */ action_limit = 1; } } return action_limit; } static int get_active_cmds(stonith_device_t * device) { int counter = 0; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(device != NULL, return 0); for (gIter = cmd_list; gIter != NULL; gIter = gIterNext) { async_command_t *cmd = gIter->data; gIterNext = gIter->next; if (cmd->active_on == device) { counter++; } } return counter; } static void fork_cb(GPid pid, gpointer user_data) { async_command_t *cmd = (async_command_t *) user_data; stonith_device_t * device = /* in case of a retry we've done the move from activating_on to active_on already */ cmd->activating_on?cmd->activating_on:cmd->active_on; CRM_ASSERT(device); crm_debug("Operation '%s' [%d]%s%s using %s now running with %ds timeout", cmd->action, pid, ((cmd->victim == NULL)? "" : " targeting "), ((cmd->victim == NULL)? "" : cmd->victim), device->id, cmd->timeout); cmd->active_on = device; cmd->activating_on = NULL; } +static int +get_agent_metadata_cb(gpointer data) { + stonith_device_t *device = data; + + device->agent_metadata = get_agent_metadata(device->agent); + if (device->agent_metadata) { + read_action_metadata(device); + stonith__device_parameter_flags(&(device->flags), device->id, + device->agent_metadata); + return G_SOURCE_REMOVE; + } else { + guint period_ms = pcmk__mainloop_timer_get_period(device->timer); + if (period_ms < 160 * 1000) { + mainloop_timer_set_period(device->timer, 2 * period_ms); + } + return G_SOURCE_CONTINUE; + } +} + static gboolean stonith_device_execute(stonith_device_t * device) { int exec_rc = 0; const char *action_str = NULL; const char *host_arg = NULL; async_command_t *cmd = NULL; stonith_action_t *action = NULL; int active_cmds = 0; int action_limit = 0; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(device != NULL, return FALSE); active_cmds = get_active_cmds(device); action_limit = get_action_limit(device); if (action_limit > -1 && active_cmds >= action_limit) { crm_trace("%s is over its action limit of %d (%u active action%s)", device->id, action_limit, active_cmds, pcmk__plural_s(active_cmds)); return TRUE; } for (gIter = device->pending_ops; gIter != NULL; gIter = gIterNext) { async_command_t *pending_op = gIter->data; gIterNext = gIter->next; if (pending_op && pending_op->delay_id) { crm_trace("Operation '%s'%s%s using %s was asked to run too early, " "waiting for start delay of %ds", pending_op->action, ((pending_op->victim == NULL)? "" : " targeting "), ((pending_op->victim == NULL)? "" : pending_op->victim), device->id, pending_op->start_delay); continue; } device->pending_ops = g_list_remove_link(device->pending_ops, gIter); g_list_free_1(gIter); cmd = pending_op; break; } if (cmd == NULL) { crm_trace("No actions using %s are needed", device->id); return TRUE; } if(pcmk__str_eq(device->agent, STONITH_WATCHDOG_AGENT, pcmk__str_casei)) { if(pcmk__str_eq(cmd->action, "reboot", pcmk__str_casei)) { pcmk__panic(__func__); goto done; } else if(pcmk__str_eq(cmd->action, "off", pcmk__str_casei)) { pcmk__panic(__func__); goto done; } else { crm_info("Faking success for %s watchdog operation", cmd->action); cmd->done_cb(0, 0, NULL, cmd); goto done; } } #if SUPPORT_CIBSECRETS if (pcmk__substitute_secrets(device->id, device->params) != pcmk_rc_ok) { /* replacing secrets failed! */ if (pcmk__str_eq(cmd->action, "stop", pcmk__str_casei)) { /* don't fail on stop! */ crm_info("Proceeding with stop operation for %s", device->id); } else { crm_err("Considering %s unconfigured: Failed to get secrets", device->id); exec_rc = PCMK_OCF_NOT_CONFIGURED; cmd->done_cb(0, exec_rc, NULL, cmd); goto done; } } #endif action_str = cmd->action; if (pcmk__str_eq(cmd->action, "reboot", pcmk__str_casei) && !pcmk_is_set(device->flags, st_device_supports_reboot)) { crm_warn("Agent '%s' does not advertise support for 'reboot', performing 'off' action instead", device->agent); action_str = "off"; } if (pcmk_is_set(device->flags, st_device_supports_parameter_port)) { host_arg = "port"; } else if (pcmk_is_set(device->flags, st_device_supports_parameter_plug)) { host_arg = "plug"; } action = stonith_action_create(device->agent, action_str, cmd->victim, cmd->victim_nodeid, cmd->timeout, device->params, device->aliases, host_arg); /* for async exec, exec_rc is negative for early error exit otherwise handling of success/errors is done via callbacks */ cmd->activating_on = device; exec_rc = stonith_action_execute_async(action, (void *)cmd, cmd->done_cb, fork_cb); if (exec_rc < 0) { crm_warn("Operation '%s'%s%s using %s failed: %s " CRM_XS " rc=%d", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, pcmk_strerror(exec_rc), exec_rc); cmd->activating_on = NULL; cmd->done_cb(0, exec_rc, NULL, cmd); } done: /* Device might get triggered to work by multiple fencing commands * simultaneously. Trigger the device again to make sure any * remaining concurrent commands get executed. */ if (device->pending_ops) { mainloop_set_trigger(device->work); } return TRUE; } static gboolean stonith_device_dispatch(gpointer user_data) { return stonith_device_execute(user_data); } static gboolean start_delay_helper(gpointer data) { async_command_t *cmd = data; stonith_device_t *device = NULL; cmd->delay_id = 0; device = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; if (device) { mainloop_set_trigger(device->work); } return FALSE; } static void schedule_stonith_command(async_command_t * cmd, stonith_device_t * device) { int delay_max = 0; int delay_base = 0; int requested_delay = cmd->start_delay; CRM_CHECK(cmd != NULL, return); CRM_CHECK(device != NULL, return); if (cmd->device) { free(cmd->device); } if (device->include_nodeid && cmd->victim) { crm_node_t *node = crm_get_peer(0, cmd->victim); cmd->victim_nodeid = node->id; } cmd->device = strdup(device->id); cmd->timeout = get_action_timeout(device, cmd->action, cmd->default_timeout); if (cmd->remote_op_id) { crm_debug("Scheduling '%s' action%s%s using %s for remote peer %s " "with op id %.8s and timeout %ds", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->origin, cmd->remote_op_id, cmd->timeout); } else { crm_debug("Scheduling '%s' action%s%s using %s for %s with timeout %ds", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->client, cmd->timeout); } device->pending_ops = g_list_append(device->pending_ops, cmd); mainloop_set_trigger(device->work); // Value -1 means disable any static/random fencing delays if (requested_delay < 0) { return; } delay_max = get_action_delay_max(device, cmd->action); delay_base = get_action_delay_base(device, cmd->action); if (delay_max == 0) { delay_max = delay_base; } if (delay_max < delay_base) { crm_warn(PCMK_STONITH_DELAY_BASE " (%ds) is larger than " PCMK_STONITH_DELAY_MAX " (%ds) for %s using %s " "(limiting to maximum delay)", delay_base, delay_max, cmd->action, device->id); delay_base = delay_max; } if (delay_max > 0) { // coverity[dont_call] We're not using rand() for security cmd->start_delay += ((delay_max != delay_base)?(rand() % (delay_max - delay_base)):0) + delay_base; } if (cmd->start_delay > 0) { crm_notice("Delaying '%s' action%s%s using %s for %ds " CRM_XS " timeout=%ds requested_delay=%ds base=%ds max=%ds", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, cmd->start_delay, cmd->timeout, requested_delay, delay_base, delay_max); cmd->delay_id = g_timeout_add_seconds(cmd->start_delay, start_delay_helper, cmd); } } static void free_device(gpointer data) { GListPtr gIter = NULL; stonith_device_t *device = data; g_hash_table_destroy(device->params); g_hash_table_destroy(device->aliases); for (gIter = device->pending_ops; gIter != NULL; gIter = gIter->next) { async_command_t *cmd = gIter->data; crm_warn("Removal of device '%s' purged operation '%s'", device->id, cmd->action); cmd->done_cb(0, -ENODEV, NULL, cmd); } g_list_free(device->pending_ops); g_list_free_full(device->targets, free); + if (device->timer) { + mainloop_timer_stop(device->timer); + mainloop_timer_del(device->timer); + } + mainloop_destroy_trigger(device->work); free_xml(device->agent_metadata); free(device->namespace); free(device->on_target_actions); free(device->agent); free(device->id); free(device); } void free_device_list(void) { if (device_list != NULL) { g_hash_table_destroy(device_list); device_list = NULL; } } void init_device_list(void) { if (device_list == NULL) { device_list = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_device); } } static GHashTable * build_port_aliases(const char *hostmap, GListPtr * targets) { char *name = NULL; int last = 0, lpc = 0, max = 0, added = 0; GHashTable *aliases = crm_strcase_table_new(); if (hostmap == NULL) { return aliases; } max = strlen(hostmap); for (; lpc <= max; lpc++) { switch (hostmap[lpc]) { /* Assignment chars */ case '=': case ':': if (lpc > last) { free(name); name = calloc(1, 1 + lpc - last); memcpy(name, hostmap + last, lpc - last); } last = lpc + 1; break; /* Delimeter chars */ /* case ',': Potentially used to specify multiple ports */ case 0: case ';': case ' ': case '\t': if (name) { char *value = NULL; value = calloc(1, 1 + lpc - last); memcpy(value, hostmap + last, lpc - last); crm_debug("Adding alias '%s'='%s'", name, value); g_hash_table_replace(aliases, name, value); if (targets) { *targets = g_list_append(*targets, strdup(value)); } value = NULL; name = NULL; added++; } else if (lpc > last) { crm_debug("Parse error at offset %d near '%s'", lpc - last, hostmap + last); } last = lpc + 1; break; } if (hostmap[lpc] == 0) { break; } } if (added == 0) { crm_info("No host mappings detected in '%s'", hostmap); } free(name); return aliases; } GHashTable *metadata_cache = NULL; void free_metadata_cache(void) { if (metadata_cache != NULL) { g_hash_table_destroy(metadata_cache); metadata_cache = NULL; } } static void init_metadata_cache(void) { if (metadata_cache == NULL) { metadata_cache = crm_str_table_new(); } } static xmlNode * get_agent_metadata(const char *agent) { xmlNode *xml = NULL; char *buffer = NULL; init_metadata_cache(); buffer = g_hash_table_lookup(metadata_cache, agent); if(pcmk__str_eq(agent, STONITH_WATCHDOG_AGENT, pcmk__str_casei)) { return NULL; } else if(buffer == NULL) { stonith_t *st = stonith_api_new(); int rc; if (st == NULL) { crm_warn("Could not get agent meta-data: " "API memory allocation failed"); return NULL; } rc = st->cmds->metadata(st, st_opt_sync_call, agent, NULL, &buffer, 10); stonith_api_delete(st); if (rc || !buffer) { crm_err("Could not retrieve metadata for fencing agent %s", agent); return NULL; } g_hash_table_replace(metadata_cache, strdup(agent), buffer); } xml = string2xml(buffer); return xml; } static gboolean is_nodeid_required(xmlNode * xml) { xmlXPathObjectPtr xpath = NULL; if (stand_alone) { return FALSE; } if (!xml) { return FALSE; } xpath = xpath_search(xml, "//parameter[@name='nodeid']"); if (numXpathResults(xpath) <= 0) { freeXpathObject(xpath); return FALSE; } freeXpathObject(xpath); return TRUE; } #define MAX_ACTION_LEN 256 static char * add_action(char *actions, const char *action) { int offset = 0; if (actions == NULL) { actions = calloc(1, MAX_ACTION_LEN); } else { offset = strlen(actions); } if (offset > 0) { offset += snprintf(actions+offset, MAX_ACTION_LEN - offset, " "); } offset += snprintf(actions+offset, MAX_ACTION_LEN - offset, "%s", action); return actions; } static void read_action_metadata(stonith_device_t *device) { xmlXPathObjectPtr xpath = NULL; int max = 0; int lpc = 0; if (device->agent_metadata == NULL) { return; } xpath = xpath_search(device->agent_metadata, "//action"); max = numXpathResults(xpath); if (max <= 0) { freeXpathObject(xpath); return; } for (lpc = 0; lpc < max; lpc++) { const char *on_target = NULL; const char *action = NULL; xmlNode *match = getXpathResult(xpath, lpc); CRM_LOG_ASSERT(match != NULL); if(match == NULL) { continue; }; on_target = crm_element_value(match, "on_target"); action = crm_element_value(match, "name"); if(pcmk__str_eq(action, "list", pcmk__str_casei)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_list); } else if(pcmk__str_eq(action, "status", pcmk__str_casei)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_status); } else if(pcmk__str_eq(action, "reboot", pcmk__str_casei)) { stonith__set_device_flags(device->flags, device->id, st_device_supports_reboot); } else if (pcmk__str_eq(action, "on", pcmk__str_casei)) { /* "automatic" means the cluster will unfence node when it joins */ const char *automatic = crm_element_value(match, "automatic"); /* "required" is a deprecated synonym for "automatic" */ const char *required = crm_element_value(match, "required"); if (crm_is_true(automatic) || crm_is_true(required)) { device->automatic_unfencing = TRUE; } } if (action && crm_is_true(on_target)) { device->on_target_actions = add_action(device->on_target_actions, action); } } freeXpathObject(xpath); } /*! * \internal * \brief Set a pcmk_*_action parameter if not already set * * \param[in,out] params Device parameters * \param[in] action Name of action * \param[in] value Value to use if action is not already set */ static void map_action(GHashTable *params, const char *action, const char *value) { char *key = crm_strdup_printf("pcmk_%s_action", action); if (g_hash_table_lookup(params, key)) { crm_warn("Ignoring %s='%s', see %s instead", STONITH_ATTR_ACTION_OP, value, key); free(key); } else { crm_warn("Mapping %s='%s' to %s='%s'", STONITH_ATTR_ACTION_OP, value, key, value); g_hash_table_insert(params, key, strdup(value)); } } /*! * \internal * \brief Create device parameter table from XML * * \param[in] name Device name (used for logging only) * \param[in,out] params Device parameters */ static GHashTable * xml2device_params(const char *name, xmlNode *dev) { GHashTable *params = xml2list(dev); const char *value; /* Action should never be specified in the device configuration, * but we support it for users who are familiar with other software * that worked that way. */ value = g_hash_table_lookup(params, STONITH_ATTR_ACTION_OP); if (value != NULL) { crm_warn("%s has '%s' parameter, which should never be specified in configuration", name, STONITH_ATTR_ACTION_OP); if (*value == '\0') { crm_warn("Ignoring empty '%s' parameter", STONITH_ATTR_ACTION_OP); } else if (strcmp(value, "reboot") == 0) { crm_warn("Ignoring %s='reboot' (see stonith-action cluster property instead)", STONITH_ATTR_ACTION_OP); } else if (strcmp(value, "off") == 0) { map_action(params, "reboot", value); } else { map_action(params, "off", value); map_action(params, "reboot", value); } g_hash_table_remove(params, STONITH_ATTR_ACTION_OP); } return params; } static stonith_device_t * build_device_from_xml(xmlNode * msg) { const char *value; xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, msg, LOG_ERR); stonith_device_t *device = NULL; char *agent = crm_element_value_copy(dev, "agent"); CRM_CHECK(agent != NULL, return device); device = calloc(1, sizeof(stonith_device_t)); CRM_CHECK(device != NULL, {free(agent); return device;}); device->id = crm_element_value_copy(dev, XML_ATTR_ID); device->agent = agent; device->namespace = crm_element_value_copy(dev, "namespace"); device->params = xml2device_params(device->id, dev); value = g_hash_table_lookup(device->params, PCMK_STONITH_HOST_LIST); if (value) { device->targets = stonith__parse_targets(value); } value = g_hash_table_lookup(device->params, PCMK_STONITH_HOST_MAP); device->aliases = build_port_aliases(value, &(device->targets)); device->agent_metadata = get_agent_metadata(device->agent); if (device->agent_metadata) { read_action_metadata(device); stonith__device_parameter_flags(&(device->flags), device->id, device->agent_metadata); + } else { + if (device->timer == NULL) { + device->timer = mainloop_timer_add("get_agent_metadata", 10 * 1000, + TRUE, get_agent_metadata_cb, device); + } + if (!mainloop_timer_running(device->timer)) { + mainloop_timer_start(device->timer); + } } value = g_hash_table_lookup(device->params, "nodeid"); if (!value) { device->include_nodeid = is_nodeid_required(device->agent_metadata); } value = crm_element_value(dev, "rsc_provides"); if (pcmk__str_eq(value, "unfencing", pcmk__str_casei)) { device->automatic_unfencing = TRUE; } if (is_action_required("on", device)) { crm_info("Fencing device '%s' requires unfencing", device->id); } if (device->on_target_actions) { crm_info("Fencing device '%s' requires actions (%s) to be executed " "on target", device->id, device->on_target_actions); } device->work = mainloop_add_trigger(G_PRIORITY_HIGH, stonith_device_dispatch, device); /* TODO: Hook up priority */ return device; } static const char * target_list_type(stonith_device_t * dev) { const char *check_type = NULL; check_type = g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_CHECK); if (check_type == NULL) { if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_LIST)) { check_type = "static-list"; } else if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_MAP)) { check_type = "static-list"; } else if (pcmk_is_set(dev->flags, st_device_supports_list)) { check_type = "dynamic-list"; } else if (pcmk_is_set(dev->flags, st_device_supports_status)) { check_type = "status"; } else { check_type = "none"; } } return check_type; } static void schedule_internal_command(const char *origin, stonith_device_t * device, const char *action, const char *victim, int timeout, void *internal_user_data, void (*done_cb) (GPid pid, int rc, const char *output, gpointer user_data)) { async_command_t *cmd = NULL; cmd = calloc(1, sizeof(async_command_t)); cmd->id = -1; cmd->default_timeout = timeout ? timeout : 60; cmd->timeout = cmd->default_timeout; cmd->action = strdup(action); cmd->victim = victim ? strdup(victim) : NULL; cmd->device = strdup(device->id); cmd->origin = strdup(origin); cmd->client = strdup(crm_system_name); cmd->client_name = strdup(crm_system_name); cmd->internal_user_data = internal_user_data; cmd->done_cb = done_cb; /* cmd, not internal_user_data, is passed to 'done_cb' as the userdata */ schedule_stonith_command(cmd, device); } gboolean string_in_list(GListPtr list, const char *item) { int lpc = 0; int max = g_list_length(list); for (lpc = 0; lpc < max; lpc++) { const char *value = g_list_nth_data(list, lpc); if (pcmk__str_eq(item, value, pcmk__str_casei)) { return TRUE; } else { crm_trace("%d: '%s' != '%s'", lpc, item, value); } } return FALSE; } static void status_search_cb(GPid pid, int rc, const char *output, gpointer user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; gboolean can = FALSE; free_async_command(cmd); if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); if (rc == 1 /* unknown */ ) { crm_trace("Host %s is not known by %s", search->host, dev->id); } else if (rc == 0 /* active */ || rc == 2 /* inactive */ ) { crm_trace("Host %s is known by %s", search->host, dev->id); can = TRUE; } else { crm_notice("Unknown result when testing if %s can fence %s: rc=%d", dev->id, search->host, rc); } search_devices_record_result(search, dev->id, can); } static void dynamic_list_search_cb(GPid pid, int rc, const char *output, gpointer user_data) { async_command_t *cmd = user_data; struct device_search_s *search = cmd->internal_user_data; stonith_device_t *dev = cmd->device ? g_hash_table_lookup(device_list, cmd->device) : NULL; gboolean can_fence = FALSE; free_async_command(cmd); /* Host/alias must be in the list output to be eligible to be fenced * * Will cause problems if down'd nodes aren't listed or (for virtual nodes) * if the guest is still listed despite being moved to another machine */ if (!dev) { search_devices_record_result(search, NULL, FALSE); return; } mainloop_set_trigger(dev->work); /* If we successfully got the targets earlier, don't disable. */ if (rc != 0 && !dev->targets) { crm_notice("Disabling port list queries for %s: %s " CRM_XS " rc=%d", dev->id, output, rc); /* Fall back to status */ g_hash_table_replace(dev->params, strdup(PCMK_STONITH_HOST_CHECK), strdup("status")); g_list_free_full(dev->targets, free); dev->targets = NULL; } else if (!rc) { crm_info("Refreshing port list for %s", dev->id); g_list_free_full(dev->targets, free); dev->targets = stonith__parse_targets(output); dev->targets_age = time(NULL); } if (dev->targets) { const char *alias = g_hash_table_lookup(dev->aliases, search->host); if (!alias) { alias = search->host; } if (string_in_list(dev->targets, alias)) { can_fence = TRUE; } } search_devices_record_result(search, dev->id, can_fence); } /*! * \internal * \brief Returns true if any key in first is not in second or second has a different value for key */ static int device_params_diff(GHashTable *first, GHashTable *second) { char *key = NULL; char *value = NULL; GHashTableIter gIter; g_hash_table_iter_init(&gIter, first); while (g_hash_table_iter_next(&gIter, (void **)&key, (void **)&value)) { if(strstr(key, "CRM_meta") == key) { continue; } else if(strcmp(key, "crm_feature_set") == 0) { continue; } else { char *other_value = g_hash_table_lookup(second, key); if (!other_value || !pcmk__str_eq(other_value, value, pcmk__str_casei)) { crm_trace("Different value for %s: %s != %s", key, other_value, value); return 1; } } } return 0; } /*! * \internal * \brief Checks to see if an identical device already exists in the device_list */ static stonith_device_t * device_has_duplicate(stonith_device_t * device) { stonith_device_t *dup = g_hash_table_lookup(device_list, device->id); if (!dup) { crm_trace("No match for %s", device->id); return NULL; } else if (!pcmk__str_eq(dup->agent, device->agent, pcmk__str_casei)) { crm_trace("Different agent: %s != %s", dup->agent, device->agent); return NULL; } /* Use calculate_operation_digest() here? */ if (device_params_diff(device->params, dup->params) || device_params_diff(dup->params, device->params)) { return NULL; } crm_trace("Match"); return dup; } int stonith_device_register(xmlNode * msg, const char **desc, gboolean from_cib) { stonith_device_t *dup = NULL; stonith_device_t *device = build_device_from_xml(msg); guint ndevices = 0; CRM_CHECK(device != NULL, return -ENOMEM); dup = device_has_duplicate(device); if (dup) { ndevices = g_hash_table_size(device_list); crm_debug("Device '%s' already in device list (%d active device%s)", device->id, ndevices, pcmk__plural_s(ndevices)); free_device(device); device = dup; dup = g_hash_table_lookup(device_list, device->id); dup->dirty = FALSE; } else { stonith_device_t *old = g_hash_table_lookup(device_list, device->id); if (from_cib && old && old->api_registered) { /* If the cib is writing over an entry that is shared with a stonith client, * copy any pending ops that currently exist on the old entry to the new one. * Otherwise the pending ops will be reported as failures */ crm_info("Overwriting existing entry for %s from CIB", device->id); device->pending_ops = old->pending_ops; device->api_registered = TRUE; old->pending_ops = NULL; if (device->pending_ops) { mainloop_set_trigger(device->work); } } g_hash_table_replace(device_list, device->id, device); ndevices = g_hash_table_size(device_list); crm_notice("Added '%s' to device list (%d active device%s)", device->id, ndevices, pcmk__plural_s(ndevices)); } if (desc) { *desc = device->id; } if (from_cib) { device->cib_registered = TRUE; } else { device->api_registered = TRUE; } return pcmk_ok; } int stonith_device_remove(const char *id, gboolean from_cib) { stonith_device_t *device = g_hash_table_lookup(device_list, id); guint ndevices = 0; if (!device) { ndevices = g_hash_table_size(device_list); crm_info("Device '%s' not found (%d active device%s)", id, ndevices, pcmk__plural_s(ndevices)); return pcmk_ok; } if (from_cib) { device->cib_registered = FALSE; } else { device->verified = FALSE; device->api_registered = FALSE; } if (!device->cib_registered && !device->api_registered) { g_hash_table_remove(device_list, id); ndevices = g_hash_table_size(device_list); crm_info("Removed '%s' from device list (%d active device%s)", id, ndevices, pcmk__plural_s(ndevices)); } else { crm_trace("Not removing '%s' from device list (%d active) because " "still registered via:%s%s", id, g_hash_table_size(device_list), (device->cib_registered? " cib" : ""), (device->api_registered? " api" : "")); } return pcmk_ok; } /*! * \internal * \brief Return the number of stonith levels registered for a node * * \param[in] tp Node's topology table entry * * \return Number of non-NULL levels in topology entry * \note This function is used only for log messages. */ static int count_active_levels(stonith_topology_t * tp) { int lpc = 0; int count = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { count++; } } return count; } static void free_topology_entry(gpointer data) { stonith_topology_t *tp = data; int lpc = 0; for (lpc = 0; lpc < ST_LEVEL_MAX; lpc++) { if (tp->levels[lpc] != NULL) { g_list_free_full(tp->levels[lpc], free); } } free(tp->target); free(tp->target_value); free(tp->target_pattern); free(tp->target_attribute); free(tp); } void free_topology_list(void) { if (topology != NULL) { g_hash_table_destroy(topology); topology = NULL; } } void init_topology_list(void) { if (topology == NULL) { topology = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free_topology_entry); } } char *stonith_level_key(xmlNode *level, int mode) { if(mode == -1) { mode = stonith_level_kind(level); } switch(mode) { case 0: return crm_element_value_copy(level, XML_ATTR_STONITH_TARGET); case 1: return crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_PATTERN); case 2: { const char *name = crm_element_value(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE); const char *value = crm_element_value(level, XML_ATTR_STONITH_TARGET_VALUE); if(name && value) { return crm_strdup_printf("%s=%s", name, value); } } default: return crm_strdup_printf("Unknown-%d-%s", mode, ID(level)); } } int stonith_level_kind(xmlNode * level) { int mode = 0; const char *target = crm_element_value(level, XML_ATTR_STONITH_TARGET); if(target == NULL) { mode++; target = crm_element_value(level, XML_ATTR_STONITH_TARGET_PATTERN); } if(stand_alone == FALSE && target == NULL) { mode++; if(crm_element_value(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE) == NULL) { mode++; } else if(crm_element_value(level, XML_ATTR_STONITH_TARGET_VALUE) == NULL) { mode++; } } return mode; } static stonith_key_value_t * parse_device_list(const char *devices) { int lpc = 0; int max = 0; int last = 0; stonith_key_value_t *output = NULL; if (devices == NULL) { return output; } max = strlen(devices); for (lpc = 0; lpc <= max; lpc++) { if (devices[lpc] == ',' || devices[lpc] == 0) { char *line = strndup(devices + last, lpc - last); output = stonith_key_value_add(output, NULL, line); free(line); last = lpc + 1; } } return output; } /*! * \internal * \brief Register a STONITH level for a target * * Given an XML request specifying the target name, level index, and device IDs * for the level, this will create an entry for the target in the global topology * table if one does not already exist, then append the specified device IDs to * the entry's device list for the specified level. * * \param[in] msg XML request for STONITH level registration * \param[out] desc If not NULL, will be set to string representation ("TARGET[LEVEL]") * * \return pcmk_ok on success, -EINVAL if XML does not specify valid level index */ int stonith_level_register(xmlNode *msg, char **desc) { int id = 0; xmlNode *level; int mode; char *target; stonith_topology_t *tp; stonith_key_value_t *dIter = NULL; stonith_key_value_t *devices = NULL; /* Allow the XML here to point to the level tag directly, or wrapped in * another tag. If directly, don't search by xpath, because it might give * multiple hits (e.g. if the XML is the CIB). */ if (pcmk__str_eq(TYPE(msg), XML_TAG_FENCING_LEVEL, pcmk__str_casei)) { level = msg; } else { level = get_xpath_object("//" XML_TAG_FENCING_LEVEL, msg, LOG_ERR); } CRM_CHECK(level != NULL, return -EINVAL); mode = stonith_level_kind(level); target = stonith_level_key(level, mode); crm_element_value_int(level, XML_ATTR_STONITH_INDEX, &id); if (desc) { *desc = crm_strdup_printf("%s[%d]", target, id); } /* Sanity-check arguments */ if (mode >= 3 || (id <= 0) || (id >= ST_LEVEL_MAX)) { crm_trace("Could not add %s[%d] (%d) to the topology (%d active entries)", target, id, mode, g_hash_table_size(topology)); free(target); crm_log_xml_err(level, "Bad topology"); return -EINVAL; } /* Find or create topology table entry */ tp = g_hash_table_lookup(topology, target); if (tp == NULL) { tp = calloc(1, sizeof(stonith_topology_t)); tp->kind = mode; tp->target = target; tp->target_value = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_VALUE); tp->target_pattern = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_PATTERN); tp->target_attribute = crm_element_value_copy(level, XML_ATTR_STONITH_TARGET_ATTRIBUTE); g_hash_table_replace(topology, tp->target, tp); crm_trace("Added %s (%d) to the topology (%d active entries)", target, mode, g_hash_table_size(topology)); } else { free(target); } if (tp->levels[id] != NULL) { crm_info("Adding to the existing %s[%d] topology entry", tp->target, id); } devices = parse_device_list(crm_element_value(level, XML_ATTR_STONITH_DEVICES)); for (dIter = devices; dIter; dIter = dIter->next) { const char *device = dIter->value; crm_trace("Adding device '%s' for %s[%d]", device, tp->target, id); tp->levels[id] = g_list_append(tp->levels[id], strdup(device)); } stonith_key_value_freeall(devices, 1, 1); { int nlevels = count_active_levels(tp); crm_info("Target %s has %d active fencing level%s", tp->target, nlevels, pcmk__plural_s(nlevels)); } return pcmk_ok; } int stonith_level_remove(xmlNode *msg, char **desc) { int id = 0; stonith_topology_t *tp; char *target; /* Unlike additions, removal requests should always have one level tag */ xmlNode *level = get_xpath_object("//" XML_TAG_FENCING_LEVEL, msg, LOG_ERR); CRM_CHECK(level != NULL, return -EINVAL); target = stonith_level_key(level, -1); crm_element_value_int(level, XML_ATTR_STONITH_INDEX, &id); if (desc) { *desc = crm_strdup_printf("%s[%d]", target, id); } /* Sanity-check arguments */ if (id >= ST_LEVEL_MAX) { free(target); return -EINVAL; } tp = g_hash_table_lookup(topology, target); if (tp == NULL) { guint nentries = g_hash_table_size(topology); crm_info("No fencing topology found for %s (%d active %s)", target, nentries, pcmk__plural_alt(nentries, "entry", "entries")); } else if (id == 0 && g_hash_table_remove(topology, target)) { guint nentries = g_hash_table_size(topology); crm_info("Removed all fencing topology entries related to %s " "(%d active %s remaining)", target, nentries, pcmk__plural_alt(nentries, "entry", "entries")); } else if (id > 0 && tp->levels[id] != NULL) { guint nlevels; g_list_free_full(tp->levels[id], free); tp->levels[id] = NULL; nlevels = count_active_levels(tp); crm_info("Removed level %d from fencing topology for %s " "(%d active level%s remaining)", id, target, nlevels, pcmk__plural_s(nlevels)); } free(target); return pcmk_ok; } /*! * \internal * \brief Schedule an (asynchronous) action directly on a stonith device * * Handle a STONITH_OP_EXEC API message by scheduling a requested agent action * directly on a specified device. Only list, monitor, and status actions are * expected to use this call, though it should work with any agent command. * * \param[in] msg API message XML with desired action * \param[out] output Unused * * \return -EINPROGRESS on success, -errno otherwise * \note If the action is monitor, the device must be registered via the API * (CIB registration is not sufficient), because monitor should not be * possible unless the device is "started" (API registered). */ static int stonith_device_action(xmlNode * msg, char **output) { xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, msg, LOG_ERR); xmlNode *op = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_ERR); const char *id = crm_element_value(dev, F_STONITH_DEVICE); const char *action = crm_element_value(op, F_STONITH_ACTION); async_command_t *cmd = NULL; stonith_device_t *device = NULL; if ((id == NULL) || (action == NULL)) { crm_info("Malformed API action request: device %s, action %s", (id? id : "not specified"), (action? action : "not specified")); return -EPROTO; } device = g_hash_table_lookup(device_list, id); if ((device == NULL) || (!device->api_registered && !strcmp(action, "monitor"))) { // Monitors may run only on "started" (API-registered) devices crm_info("Ignoring API '%s' action request because device %s not found", action, id); return -ENODEV; } cmd = create_async_command(msg); if (cmd == NULL) { return -EPROTO; } schedule_stonith_command(cmd, device); return -EINPROGRESS; } static void search_devices_record_result(struct device_search_s *search, const char *device, gboolean can_fence) { search->replies_received++; if (can_fence && device) { search->capable = g_list_append(search->capable, strdup(device)); } if (search->replies_needed == search->replies_received) { guint ndevices = g_list_length(search->capable); crm_debug("Search found %d device%s that can perform '%s' targeting %s", ndevices, pcmk__plural_s(ndevices), (search->action? search->action : "unknown action"), (search->host? search->host : "any node")); search->callback(search->capable, search->user_data); free(search->host); free(search->action); free(search); } } /*! * \internal * \brief Check whether the local host is allowed to execute a fencing action * * \param[in] device Fence device to check * \param[in] action Fence action to check * \param[in] target Hostname of fence target * \param[in] allow_suicide Whether self-fencing is allowed for this operation * * \return TRUE if local host is allowed to execute action, FALSE otherwise */ static gboolean localhost_is_eligible(const stonith_device_t *device, const char *action, const char *target, gboolean allow_suicide) { gboolean localhost_is_target = pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei); if (device && action && device->on_target_actions && strstr(device->on_target_actions, action)) { if (!localhost_is_target) { crm_trace("Operation '%s' using %s can only be executed for " "local host, not %s", action, device->id, target); return FALSE; } } else if (localhost_is_target && !allow_suicide) { crm_trace("'%s' operation does not support self-fencing", action); return FALSE; } return TRUE; } static void can_fence_host_with_device(stonith_device_t * dev, struct device_search_s *search) { gboolean can = FALSE; const char *check_type = NULL; const char *host = search->host; const char *alias = NULL; CRM_LOG_ASSERT(dev != NULL); if (dev == NULL) { goto search_report_results; } else if (host == NULL) { can = TRUE; goto search_report_results; } /* Short-circuit query if this host is not allowed to perform the action */ if (pcmk__str_eq(search->action, "reboot", pcmk__str_casei)) { /* A "reboot" *might* get remapped to "off" then "on", so short-circuit * only if all three are disallowed. If only one or two are disallowed, * we'll report that with the results. We never allow suicide for * remapped "on" operations because the host is off at that point. */ if (!localhost_is_eligible(dev, "reboot", host, search->allow_suicide) && !localhost_is_eligible(dev, "off", host, search->allow_suicide) && !localhost_is_eligible(dev, "on", host, FALSE)) { goto search_report_results; } } else if (!localhost_is_eligible(dev, search->action, host, search->allow_suicide)) { goto search_report_results; } alias = g_hash_table_lookup(dev->aliases, host); if (alias == NULL) { alias = host; } check_type = target_list_type(dev); if (pcmk__str_eq(check_type, "none", pcmk__str_casei)) { can = TRUE; } else if (pcmk__str_eq(check_type, "static-list", pcmk__str_casei)) { /* Presence in the hostmap is sufficient * Only use if all hosts on which the device can be active can always fence all listed hosts */ if (string_in_list(dev->targets, host)) { can = TRUE; } else if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_MAP) && g_hash_table_lookup(dev->aliases, host)) { can = TRUE; } } else if (pcmk__str_eq(check_type, "dynamic-list", pcmk__str_casei)) { time_t now = time(NULL); if (dev->targets == NULL || dev->targets_age + 60 < now) { crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev->id, search->host, search->action); schedule_internal_command(__func__, dev, "list", NULL, search->per_device_timeout, search, dynamic_list_search_cb); /* we'll respond to this search request async in the cb */ return; } if (string_in_list(dev->targets, alias)) { can = TRUE; } } else if (pcmk__str_eq(check_type, "status", pcmk__str_casei)) { crm_trace("Running '%s' to check whether %s is eligible to fence %s (%s)", check_type, dev->id, search->host, search->action); schedule_internal_command(__func__, dev, "status", search->host, search->per_device_timeout, search, status_search_cb); /* we'll respond to this search request async in the cb */ return; } else { crm_err("Invalid value for " PCMK_STONITH_HOST_CHECK ": %s", check_type); check_type = "Invalid " PCMK_STONITH_HOST_CHECK; } if (pcmk__str_eq(host, alias, pcmk__str_casei)) { crm_notice("%s is%s eligible to fence (%s) %s: %s", dev->id, (can? "" : " not"), search->action, host, check_type); } else { crm_notice("%s is%s eligible to fence (%s) %s (aka. '%s'): %s", dev->id, (can? "" : " not"), search->action, host, alias, check_type); } search_report_results: search_devices_record_result(search, dev ? dev->id : NULL, can); } static void search_devices(gpointer key, gpointer value, gpointer user_data) { stonith_device_t *dev = value; struct device_search_s *search = user_data; can_fence_host_with_device(dev, search); } #define DEFAULT_QUERY_TIMEOUT 20 static void get_capable_devices(const char *host, const char *action, int timeout, bool suicide, void *user_data, void (*callback) (GList * devices, void *user_data)) { struct device_search_s *search; int per_device_timeout = DEFAULT_QUERY_TIMEOUT; int devices_needing_async_query = 0; char *key = NULL; const char *check_type = NULL; GHashTableIter gIter; stonith_device_t *device = NULL; guint ndevices = g_hash_table_size(device_list); if (ndevices == 0) { callback(NULL, user_data); return; } search = calloc(1, sizeof(struct device_search_s)); if (!search) { callback(NULL, user_data); return; } g_hash_table_iter_init(&gIter, device_list); while (g_hash_table_iter_next(&gIter, (void **)&key, (void **)&device)) { check_type = target_list_type(device); if (pcmk__strcase_any_of(check_type, "status", "dynamic-list", NULL)) { devices_needing_async_query++; } } /* If we have devices that require an async event in order to know what * nodes they can fence, we have to give the events a timeout. The total * query timeout is divided among those events. */ if (devices_needing_async_query) { per_device_timeout = timeout / devices_needing_async_query; if (!per_device_timeout) { crm_err("Fencing timeout %ds is too low; using %ds, " "but consider raising to at least %ds", timeout, DEFAULT_QUERY_TIMEOUT, DEFAULT_QUERY_TIMEOUT * devices_needing_async_query); per_device_timeout = DEFAULT_QUERY_TIMEOUT; } else if (per_device_timeout < DEFAULT_QUERY_TIMEOUT) { crm_notice("Fencing timeout %ds is low for the current " "configuration; consider raising to at least %ds", timeout, DEFAULT_QUERY_TIMEOUT * devices_needing_async_query); } } search->host = host ? strdup(host) : NULL; search->action = action ? strdup(action) : NULL; search->per_device_timeout = per_device_timeout; /* We are guaranteed this many replies. Even if a device gets * unregistered some how during the async search, we will get * the correct number of replies. */ search->replies_needed = ndevices; search->allow_suicide = suicide; search->callback = callback; search->user_data = user_data; /* kick off the search */ crm_debug("Searching %d device%s to see which can execute '%s' targeting %s", ndevices, pcmk__plural_s(ndevices), (search->action? search->action : "unknown action"), (search->host? search->host : "any node")); g_hash_table_foreach(device_list, search_devices, search); } struct st_query_data { xmlNode *reply; char *remote_peer; char *client_id; char *target; char *action; int call_options; }; /*! * \internal * \brief Add action-specific attributes to query reply XML * * \param[in,out] xml XML to add attributes to * \param[in] action Fence action * \param[in] device Fence device */ static void add_action_specific_attributes(xmlNode *xml, const char *action, stonith_device_t *device) { int action_specific_timeout; int delay_max; int delay_base; CRM_CHECK(xml && action && device, return); if (is_action_required(action, device)) { crm_trace("Action '%s' is required using %s", action, device->id); crm_xml_add_int(xml, F_STONITH_DEVICE_REQUIRED, 1); } action_specific_timeout = get_action_timeout(device, action, 0); if (action_specific_timeout) { crm_trace("Action '%s' has timeout %dms using %s", action, action_specific_timeout, device->id); crm_xml_add_int(xml, F_STONITH_ACTION_TIMEOUT, action_specific_timeout); } delay_max = get_action_delay_max(device, action); if (delay_max > 0) { crm_trace("Action '%s' has maximum random delay %dms using %s", action, delay_max, device->id); crm_xml_add_int(xml, F_STONITH_DELAY_MAX, delay_max / 1000); } delay_base = get_action_delay_base(device, action); if (delay_base > 0) { crm_xml_add_int(xml, F_STONITH_DELAY_BASE, delay_base / 1000); } if ((delay_max > 0) && (delay_base == 0)) { crm_trace("Action '%s' has maximum random delay %dms using %s", action, delay_max, device->id); } else if ((delay_max == 0) && (delay_base > 0)) { crm_trace("Action '%s' has a static delay of %dms using %s", action, delay_base, device->id); } else if ((delay_max > 0) && (delay_base > 0)) { crm_trace("Action '%s' has a minimum delay of %dms and a randomly chosen " "maximum delay of %dms using %s", action, delay_base, delay_max, device->id); } } /*! * \internal * \brief Add "disallowed" attribute to query reply XML if appropriate * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_disallowed(xmlNode *xml, const char *action, stonith_device_t *device, const char *target, gboolean allow_suicide) { if (!localhost_is_eligible(device, action, target, allow_suicide)) { crm_trace("Action '%s' using %s is disallowed for local host", action, device->id); crm_xml_add(xml, F_STONITH_ACTION_DISALLOWED, XML_BOOLEAN_TRUE); } } /*! * \internal * \brief Add child element with action-specific values to query reply XML * * \param[in,out] xml XML to add attribute to * \param[in] action Fence action * \param[in] device Fence device * \param[in] target Fence target * \param[in] allow_suicide Whether self-fencing is allowed */ static void add_action_reply(xmlNode *xml, const char *action, stonith_device_t *device, const char *target, gboolean allow_suicide) { xmlNode *child = create_xml_node(xml, F_STONITH_ACTION); crm_xml_add(child, XML_ATTR_ID, action); add_action_specific_attributes(child, action, device); add_disallowed(child, action, device, target, allow_suicide); } static void stonith_query_capable_device_cb(GList * devices, void *user_data) { struct st_query_data *query = user_data; int available_devices = 0; xmlNode *dev = NULL; xmlNode *list = NULL; GListPtr lpc = NULL; /* Pack the results into XML */ list = create_xml_node(NULL, __func__); crm_xml_add(list, F_STONITH_TARGET, query->target); for (lpc = devices; lpc != NULL; lpc = lpc->next) { stonith_device_t *device = g_hash_table_lookup(device_list, lpc->data); const char *action = query->action; if (!device) { /* It is possible the device got unregistered while * determining who can fence the target */ continue; } available_devices++; dev = create_xml_node(list, F_STONITH_DEVICE); crm_xml_add(dev, XML_ATTR_ID, device->id); crm_xml_add(dev, "namespace", device->namespace); crm_xml_add(dev, "agent", device->agent); crm_xml_add_int(dev, F_STONITH_DEVICE_VERIFIED, device->verified); /* If the originating fencer wants to reboot the node, and we have a * capable device that doesn't support "reboot", remap to "off" instead. */ if (!pcmk_is_set(device->flags, st_device_supports_reboot) && pcmk__str_eq(query->action, "reboot", pcmk__str_casei)) { crm_trace("%s doesn't support reboot, using values for off instead", device->id); action = "off"; } /* Add action-specific values if available */ add_action_specific_attributes(dev, action, device); if (pcmk__str_eq(query->action, "reboot", pcmk__str_casei)) { /* A "reboot" *might* get remapped to "off" then "on", so after * sending the "reboot"-specific values in the main element, we add * sub-elements for "off" and "on" values. * * We short-circuited earlier if "reboot", "off" and "on" are all * disallowed for the local host. However if only one or two are * disallowed, we send back the results and mark which ones are * disallowed. If "reboot" is disallowed, this might cause problems * with older fencer versions, which won't check for it. Older * versions will ignore "off" and "on", so they are not a problem. */ add_disallowed(dev, action, device, query->target, pcmk_is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, "off", device, query->target, pcmk_is_set(query->call_options, st_opt_allow_suicide)); add_action_reply(dev, "on", device, query->target, FALSE); } /* A query without a target wants device parameters */ if (query->target == NULL) { xmlNode *attrs = create_xml_node(dev, XML_TAG_ATTRS); g_hash_table_foreach(device->params, hash2field, attrs); } } crm_xml_add_int(list, F_STONITH_AVAILABLE_DEVICES, available_devices); if (query->target) { crm_debug("Found %d matching device%s for target '%s'", available_devices, pcmk__plural_s(available_devices), query->target); } else { crm_debug("%d device%s installed", available_devices, pcmk__plural_s(available_devices)); } if (list != NULL) { crm_log_xml_trace(list, "Add query results"); add_message_xml(query->reply, F_STONITH_CALLDATA, list); } stonith_send_reply(query->reply, query->call_options, query->remote_peer, query->client_id); free_xml(query->reply); free(query->remote_peer); free(query->client_id); free(query->target); free(query->action); free(query); free_xml(list); g_list_free_full(devices, free); } static void stonith_query(xmlNode * msg, const char *remote_peer, const char *client_id, int call_options) { struct st_query_data *query = NULL; const char *action = NULL; const char *target = NULL; int timeout = 0; xmlNode *dev = get_xpath_object("//@" F_STONITH_ACTION, msg, LOG_NEVER); crm_element_value_int(msg, F_STONITH_TIMEOUT, &timeout); if (dev) { const char *device = crm_element_value(dev, F_STONITH_DEVICE); target = crm_element_value(dev, F_STONITH_TARGET); action = crm_element_value(dev, F_STONITH_ACTION); if (device && pcmk__str_eq(device, "manual_ack", pcmk__str_casei)) { /* No query or reply necessary */ return; } } crm_log_xml_debug(msg, "Query"); query = calloc(1, sizeof(struct st_query_data)); query->reply = stonith_construct_reply(msg, NULL, NULL, pcmk_ok); query->remote_peer = remote_peer ? strdup(remote_peer) : NULL; query->client_id = client_id ? strdup(client_id) : NULL; query->target = target ? strdup(target) : NULL; query->action = action ? strdup(action) : NULL; query->call_options = call_options; get_capable_devices(target, action, timeout, pcmk_is_set(call_options, st_opt_allow_suicide), query, stonith_query_capable_device_cb); } #define ST_LOG_OUTPUT_MAX 512 static void log_operation(async_command_t * cmd, int rc, int pid, const char *next, const char *output, gboolean op_merged) { if (rc == 0) { next = NULL; } if (cmd->victim != NULL) { do_crm_log(((rc == 0)? LOG_NOTICE : LOG_ERR), "Operation '%s' [%d] (%scall %d from %s) targeting %s " "using %s returned %d (%s)%s%s", cmd->action, pid, (op_merged? "merged " : ""), cmd->id, cmd->client_name, cmd->victim, cmd->device, rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } else { do_crm_log_unlikely(((rc == 0)? LOG_DEBUG : LOG_NOTICE), "Operation '%s' [%d]%s using %s returned %d (%s)%s%s", cmd->action, pid, (op_merged? " (merged)" : ""), cmd->device, rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } if (output) { // Output may have multiple lines char *prefix = crm_strdup_printf("%s[%d]", cmd->device, pid); crm_log_output(rc == 0 ? LOG_DEBUG : LOG_WARNING, prefix, output); free(prefix); } } static void stonith_send_async_reply(async_command_t * cmd, const char *output, int rc, GPid pid, int options) { xmlNode *reply = NULL; gboolean bcast = FALSE; reply = stonith_construct_async_reply(cmd, output, NULL, rc); if (pcmk__str_eq(cmd->action, "metadata", pcmk__str_casei)) { /* Too verbose to log */ crm_trace("Metadata query for %s", cmd->device); output = NULL; } else if (pcmk__str_any_of(cmd->action, "monitor", "list", "status", NULL)) { crm_trace("Never broadcast '%s' replies", cmd->action); } else if (!stand_alone && pcmk__str_eq(cmd->origin, cmd->victim, pcmk__str_casei) && !pcmk__str_eq(cmd->action, "on", pcmk__str_casei)) { crm_trace("Broadcast '%s' reply for %s", cmd->action, cmd->victim); crm_xml_add(reply, F_SUBTYPE, "broadcast"); bcast = TRUE; } log_operation(cmd, rc, pid, NULL, output, (options & st_reply_opt_merged ? TRUE : FALSE)); crm_log_xml_trace(reply, "Reply"); if (options & st_reply_opt_merged) { crm_xml_add(reply, F_STONITH_MERGED, "true"); } if (bcast) { crm_xml_add(reply, F_STONITH_OPERATION, T_STONITH_NOTIFY); send_cluster_message(NULL, crm_msg_stonith_ng, reply, FALSE); } else if (cmd->origin) { crm_trace("Directed reply to %s", cmd->origin); send_cluster_message(crm_get_peer(0, cmd->origin), crm_msg_stonith_ng, reply, FALSE); } else { crm_trace("Directed local %ssync reply to %s", (cmd->options & st_opt_sync_call) ? "" : "a-", cmd->client_name); do_local_reply(reply, cmd->client, cmd->options & st_opt_sync_call, FALSE); } if (stand_alone) { /* Do notification with a clean data object */ xmlNode *notify_data = create_xml_node(NULL, T_STONITH_NOTIFY_FENCE); crm_xml_add_int(notify_data, F_STONITH_RC, rc); crm_xml_add(notify_data, F_STONITH_TARGET, cmd->victim); crm_xml_add(notify_data, F_STONITH_OPERATION, cmd->op); crm_xml_add(notify_data, F_STONITH_DELEGATE, "localhost"); crm_xml_add(notify_data, F_STONITH_DEVICE, cmd->device); crm_xml_add(notify_data, F_STONITH_REMOTE_OP_ID, cmd->remote_op_id); crm_xml_add(notify_data, F_STONITH_ORIGIN, cmd->client); do_stonith_notify(0, T_STONITH_NOTIFY_FENCE, rc, notify_data); do_stonith_notify(0, T_STONITH_NOTIFY_HISTORY, 0, NULL); } free_xml(reply); } static void cancel_stonith_command(async_command_t * cmd) { stonith_device_t *device; CRM_CHECK(cmd != NULL, return); if (!cmd->device) { return; } device = g_hash_table_lookup(device_list, cmd->device); if (device) { crm_trace("Cancel scheduled '%s' action using %s", cmd->action, device->id); device->pending_ops = g_list_remove(device->pending_ops, cmd); } } static void st_child_done(GPid pid, int rc, const char *output, gpointer user_data) { stonith_device_t *device = NULL; stonith_device_t *next_device = NULL; async_command_t *cmd = user_data; GListPtr gIter = NULL; GListPtr gIterNext = NULL; CRM_CHECK(cmd != NULL, return); cmd->active_on = NULL; /* The device is ready to do something else now */ device = g_hash_table_lookup(device_list, cmd->device); if (device) { if (!device->verified && (rc == pcmk_ok) && (pcmk__strcase_any_of(cmd->action, "list", "monitor", "status", NULL))) { device->verified = TRUE; } mainloop_set_trigger(device->work); } crm_debug("Operation '%s' using %s returned %d (%d devices remaining)", cmd->action, cmd->device, rc, g_list_length(cmd->device_next)); if (rc == 0) { GListPtr iter; /* see if there are any required devices left to execute for this op */ for (iter = cmd->device_next; iter != NULL; iter = iter->next) { next_device = g_hash_table_lookup(device_list, iter->data); if (next_device != NULL && is_action_required(cmd->action, next_device)) { cmd->device_next = iter->next; break; } next_device = NULL; } } else if (rc != 0 && cmd->device_next && (is_action_required(cmd->action, device) == FALSE)) { /* if this device didn't work out, see if there are any others we can try. * if the failed device was 'required', we can't pick another device. */ next_device = g_hash_table_lookup(device_list, cmd->device_next->data); cmd->device_next = cmd->device_next->next; } /* this operation requires more fencing, hooray! */ if (next_device) { log_operation(cmd, rc, pid, next_device->id, output, FALSE); schedule_stonith_command(cmd, next_device); /* Prevent cmd from being freed */ cmd = NULL; goto done; } stonith_send_async_reply(cmd, output, rc, pid, st_reply_opt_none); if (rc != 0) { goto done; } /* Check to see if any operations are scheduled to do the exact * same thing that just completed. If so, rather than * performing the same fencing operation twice, return the result * of this operation for all pending commands it matches. */ for (gIter = cmd_list; gIter != NULL; gIter = gIterNext) { async_command_t *cmd_other = gIter->data; gIterNext = gIter->next; if (cmd == cmd_other) { continue; } /* A pending scheduled command matches the command that just finished if. * 1. The client connections are different. * 2. The node victim is the same. * 3. The fencing action is the same. * 4. The device scheduled to execute the action is the same. */ if (pcmk__str_eq(cmd->client, cmd_other->client, pcmk__str_casei) || !pcmk__str_eq(cmd->victim, cmd_other->victim, pcmk__str_casei) || !pcmk__str_eq(cmd->action, cmd_other->action, pcmk__str_casei) || !pcmk__str_eq(cmd->device, cmd_other->device, pcmk__str_casei)) { continue; } /* Duplicate merging will do the right thing for either type of remapped * reboot. If the executing fencer remapped an unsupported reboot to * off, then cmd->action will be reboot and will be merged with any * other reboot requests. If the originating fencer remapped a * topology reboot to off then on, we will get here once with * cmd->action "off" and once with "on", and they will be merged * separately with similar requests. */ crm_notice("Merging fencing action '%s' targeting %s originating from " "client %s with identical fencing request from client %s", cmd_other->action, cmd_other->victim, cmd_other->client_name, cmd->client_name); cmd_list = g_list_remove_link(cmd_list, gIter); stonith_send_async_reply(cmd_other, output, rc, pid, st_reply_opt_merged); cancel_stonith_command(cmd_other); free_async_command(cmd_other); g_list_free_1(gIter); } done: free_async_command(cmd); } static gint sort_device_priority(gconstpointer a, gconstpointer b) { const stonith_device_t *dev_a = a; const stonith_device_t *dev_b = b; if (dev_a->priority > dev_b->priority) { return -1; } else if (dev_a->priority < dev_b->priority) { return 1; } return 0; } static void stonith_fence_get_devices_cb(GList * devices, void *user_data) { async_command_t *cmd = user_data; stonith_device_t *device = NULL; guint ndevices = g_list_length(devices); crm_info("Found %d matching device%s for target '%s'", ndevices, pcmk__plural_s(ndevices), cmd->victim); if (devices != NULL) { /* Order based on priority */ devices = g_list_sort(devices, sort_device_priority); device = g_hash_table_lookup(device_list, devices->data); if (device) { cmd->device_list = devices; cmd->device_next = devices->next; devices = NULL; /* list owned by cmd now */ } } /* we have a device, schedule it for fencing. */ if (device) { schedule_stonith_command(cmd, device); /* in progress */ return; } /* no device found! */ stonith_send_async_reply(cmd, NULL, -ENODEV, 0, st_reply_opt_none); free_async_command(cmd); g_list_free_full(devices, free); } static int stonith_fence(xmlNode * msg) { const char *device_id = NULL; stonith_device_t *device = NULL; async_command_t *cmd = create_async_command(msg); xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, msg, LOG_ERR); if (cmd == NULL) { return -EPROTO; } device_id = crm_element_value(dev, F_STONITH_DEVICE); if (device_id) { device = g_hash_table_lookup(device_list, device_id); if (device == NULL) { crm_err("Requested device '%s' is not available", device_id); return -ENODEV; } schedule_stonith_command(cmd, device); } else { const char *host = crm_element_value(dev, F_STONITH_TARGET); if (cmd->options & st_opt_cs_nodeid) { int nodeid = crm_atoi(host, NULL); crm_node_t *node = pcmk__search_known_node_cache(nodeid, NULL, CRM_GET_PEER_ANY); if (node) { host = node->uname; } } /* If we get to here, then self-fencing is implicitly allowed */ get_capable_devices(host, cmd->action, cmd->default_timeout, TRUE, cmd, stonith_fence_get_devices_cb); } return -EINPROGRESS; } xmlNode * stonith_construct_reply(xmlNode * request, const char *output, xmlNode * data, int rc) { xmlNode *reply = NULL; reply = create_xml_node(NULL, T_STONITH_REPLY); crm_xml_add(reply, "st_origin", __func__); crm_xml_add(reply, F_TYPE, T_STONITH_NG); crm_xml_add(reply, "st_output", output); crm_xml_add_int(reply, F_STONITH_RC, rc); if (request == NULL) { /* Most likely, this is the result of a stonith operation that was * initiated before we came up. Unfortunately that means we lack enough * information to provide clients with a full result. * * @TODO Maybe synchronize this information at start-up? */ crm_warn("Missing request information for client notifications for " "operation with result %d (initiated before we came up?)", rc); } else { const char *name = NULL; const char *value = NULL; const char *names[] = { F_STONITH_OPERATION, F_STONITH_CALLID, F_STONITH_CLIENTID, F_STONITH_CLIENTNAME, F_STONITH_REMOTE_OP_ID, F_STONITH_CALLOPTS }; crm_trace("Creating a result reply with%s reply output (rc=%d)", (data? "" : "out"), rc); for (int lpc = 0; lpc < DIMOF(names); lpc++) { name = names[lpc]; value = crm_element_value(request, name); crm_xml_add(reply, name, value); } if (data != NULL) { add_message_xml(reply, F_STONITH_CALLDATA, data); } } return reply; } static xmlNode * stonith_construct_async_reply(async_command_t * cmd, const char *output, xmlNode * data, int rc) { xmlNode *reply = NULL; crm_trace("Creating a basic reply"); reply = create_xml_node(NULL, T_STONITH_REPLY); crm_xml_add(reply, "st_origin", __func__); crm_xml_add(reply, F_TYPE, T_STONITH_NG); crm_xml_add(reply, F_STONITH_OPERATION, cmd->op); crm_xml_add(reply, F_STONITH_DEVICE, cmd->device); crm_xml_add(reply, F_STONITH_REMOTE_OP_ID, cmd->remote_op_id); crm_xml_add(reply, F_STONITH_CLIENTID, cmd->client); crm_xml_add(reply, F_STONITH_CLIENTNAME, cmd->client_name); crm_xml_add(reply, F_STONITH_TARGET, cmd->victim); crm_xml_add(reply, F_STONITH_ACTION, cmd->op); crm_xml_add(reply, F_STONITH_ORIGIN, cmd->origin); crm_xml_add_int(reply, F_STONITH_CALLID, cmd->id); crm_xml_add_int(reply, F_STONITH_CALLOPTS, cmd->options); crm_xml_add_int(reply, F_STONITH_RC, rc); crm_xml_add(reply, "st_output", output); if (data != NULL) { crm_info("Attaching reply output"); add_message_xml(reply, F_STONITH_CALLDATA, data); } return reply; } bool fencing_peer_active(crm_node_t *peer) { if (peer == NULL) { return FALSE; } else if (peer->uname == NULL) { return FALSE; } else if (pcmk_is_set(peer->processes, crm_get_cluster_proc())) { return TRUE; } return FALSE; } /*! * \internal * \brief Determine if we need to use an alternate node to * fence the target. If so return that node's uname * * \retval NULL, no alternate host * \retval uname, uname of alternate host to use */ static const char * check_alternate_host(const char *target) { const char *alternate_host = NULL; crm_trace("Checking if we (%s) can fence %s", stonith_our_uname, target); if (find_topology_for_host(target) && pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei)) { GHashTableIter gIter; crm_node_t *entry = NULL; g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { crm_trace("Checking for %s.%d != %s", entry->uname, entry->id, target); if (fencing_peer_active(entry) && !pcmk__str_eq(entry->uname, target, pcmk__str_casei)) { alternate_host = entry->uname; break; } } if (alternate_host == NULL) { crm_err("No alternate host available to handle request " "for self-fencing with topology"); g_hash_table_iter_init(&gIter, crm_peer_cache); while (g_hash_table_iter_next(&gIter, NULL, (void **)&entry)) { crm_notice("Peer[%d] %s", entry->id, entry->uname); } } } return alternate_host; } static void stonith_send_reply(xmlNode * reply, int call_options, const char *remote_peer, const char *client_id) { if (remote_peer) { send_cluster_message(crm_get_peer(0, remote_peer), crm_msg_stonith_ng, reply, FALSE); } else { do_local_reply(reply, client_id, pcmk_is_set(call_options, st_opt_sync_call), (remote_peer != NULL)); } } static void remove_relay_op(xmlNode * request) { xmlNode *dev = get_xpath_object("//@" F_STONITH_ACTION, request, LOG_TRACE); const char *relay_op_id = NULL; const char *op_id = NULL; const char *client_name = NULL; const char *target = NULL; remote_fencing_op_t *relay_op = NULL; if (dev) { target = crm_element_value(dev, F_STONITH_TARGET); } relay_op_id = crm_element_value(request, F_STONITH_REMOTE_OP_ID_RELAY); op_id = crm_element_value(request, F_STONITH_REMOTE_OP_ID); client_name = crm_element_value(request, F_STONITH_CLIENTNAME); /* Delete RELAY operation. */ if (relay_op_id && target && pcmk__str_eq(target, stonith_our_uname, pcmk__str_casei)) { relay_op = g_hash_table_lookup(stonith_remote_op_list, relay_op_id); if (relay_op) { GHashTableIter iter; remote_fencing_op_t *list_op = NULL; g_hash_table_iter_init(&iter, stonith_remote_op_list); /* If the operation to be deleted is registered as a duplicate, delete the registration. */ while (g_hash_table_iter_next(&iter, NULL, (void **)&list_op)) { GListPtr dup_iter = NULL; if (list_op != relay_op) { for (dup_iter = list_op->duplicates; dup_iter != NULL; dup_iter = dup_iter->next) { remote_fencing_op_t *other = dup_iter->data; if (other == relay_op) { other->duplicates = g_list_remove(other->duplicates, relay_op); break; } } } } crm_debug("Deleting relay op %s ('%s' targeting %s for %s), " "replaced by op %s ('%s' targeting %s for %s)", relay_op->id, relay_op->action, relay_op->target, relay_op->client_name, op_id, relay_op->action, target, client_name); g_hash_table_remove(stonith_remote_op_list, relay_op_id); } } } static int handle_request(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *request, const char *remote_peer) { int call_options = 0; int rc = -EOPNOTSUPP; xmlNode *data = NULL; xmlNode *reply = NULL; char *output = NULL; const char *op = crm_element_value(request, F_STONITH_OPERATION); const char *client_id = crm_element_value(request, F_STONITH_CLIENTID); /* IPC commands related to fencing configuration may be done only by * privileged users (i.e. root or hacluster), because all other users should * go through the CIB to have ACLs applied. * * If no client was given, this is a peer request, which is always allowed. */ bool allowed = (client == NULL) || pcmk_is_set(client->flags, pcmk__client_privileged); crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } if (pcmk__str_eq(op, CRM_OP_REGISTER, pcmk__str_none)) { xmlNode *reply = create_xml_node(NULL, "reply"); CRM_ASSERT(client); crm_xml_add(reply, F_STONITH_OPERATION, CRM_OP_REGISTER); crm_xml_add(reply, F_STONITH_CLIENTID, client->id); pcmk__ipc_send_xml(client, id, reply, flags); client->request_id = 0; free_xml(reply); return 0; } else if (pcmk__str_eq(op, STONITH_OP_EXEC, pcmk__str_none)) { rc = stonith_device_action(request, &output); } else if (pcmk__str_eq(op, STONITH_OP_TIMEOUT_UPDATE, pcmk__str_none)) { const char *call_id = crm_element_value(request, F_STONITH_CALLID); const char *client_id = crm_element_value(request, F_STONITH_CLIENTID); int op_timeout = 0; crm_element_value_int(request, F_STONITH_TIMEOUT, &op_timeout); do_stonith_async_timeout_update(client_id, call_id, op_timeout); return 0; } else if (pcmk__str_eq(op, STONITH_OP_QUERY, pcmk__str_none)) { if (remote_peer) { create_remote_stonith_op(client_id, request, TRUE); /* Record it for the future notification */ } /* Delete the DC node RELAY operation. */ remove_relay_op(request); stonith_query(request, remote_peer, client_id, call_options); return 0; } else if (pcmk__str_eq(op, T_STONITH_NOTIFY, pcmk__str_none)) { const char *flag_name = NULL; CRM_ASSERT(client); flag_name = crm_element_value(request, F_STONITH_NOTIFY_ACTIVATE); if (flag_name) { crm_debug("Enabling %s callbacks for client %s", flag_name, pcmk__client_name(client)); pcmk__set_client_flags(client, get_stonith_flag(flag_name)); } flag_name = crm_element_value(request, F_STONITH_NOTIFY_DEACTIVATE); if (flag_name) { crm_debug("Disabling %s callbacks for client %s", flag_name, pcmk__client_name(client)); pcmk__clear_client_flags(client, get_stonith_flag(flag_name)); } pcmk__ipc_send_ack(client, id, flags, "ack", CRM_EX_OK); return 0; } else if (pcmk__str_eq(op, STONITH_OP_RELAY, pcmk__str_none)) { xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); crm_notice("Received forwarded fencing request from " "%s%s to fence (%s) peer %s", ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client)), crm_element_value(dev, F_STONITH_ACTION), crm_element_value(dev, F_STONITH_TARGET)); if (initiate_remote_stonith_op(NULL, request, FALSE) != NULL) { rc = -EINPROGRESS; } } else if (pcmk__str_eq(op, STONITH_OP_FENCE, pcmk__str_none)) { if (remote_peer || stand_alone) { rc = stonith_fence(request); } else if (call_options & st_opt_manual_ack) { remote_fencing_op_t *rop = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); const char *target = crm_element_value(dev, F_STONITH_TARGET); crm_notice("Received manual confirmation that %s is fenced", target); rop = initiate_remote_stonith_op(client, request, TRUE); rc = stonith_manual_ack(request, rop); } else { const char *alternate_host = NULL; xmlNode *dev = get_xpath_object("//@" F_STONITH_TARGET, request, LOG_TRACE); const char *target = crm_element_value(dev, F_STONITH_TARGET); const char *action = crm_element_value(dev, F_STONITH_ACTION); const char *device = crm_element_value(dev, F_STONITH_DEVICE); if (client) { int tolerance = 0; crm_notice("Client %s wants to fence (%s) %s using %s", pcmk__client_name(client), action, target, (device? device : "any device")); crm_element_value_int(dev, F_STONITH_TOLERANCE, &tolerance); if (stonith_check_fence_tolerance(tolerance, target, action)) { rc = 0; goto done; } } else { crm_notice("Peer %s wants to fence (%s) '%s' with device '%s'", remote_peer, action, target, device ? device : "(any)"); } alternate_host = check_alternate_host(target); if (alternate_host && client) { const char *client_id = NULL; remote_fencing_op_t *op = NULL; crm_notice("Forwarding self-fencing request to peer %s" "due to topology", alternate_host); if (client->id) { client_id = client->id; } else { client_id = crm_element_value(request, F_STONITH_CLIENTID); } /* Create an operation for RELAY and send the ID in the RELAY message. */ /* When a QUERY response is received, delete the RELAY operation to avoid the existence of duplicate operations. */ op = create_remote_stonith_op(client_id, request, FALSE); crm_xml_add(request, F_STONITH_OPERATION, STONITH_OP_RELAY); crm_xml_add(request, F_STONITH_CLIENTID, client->id); crm_xml_add(request, F_STONITH_REMOTE_OP_ID, op->id); send_cluster_message(crm_get_peer(0, alternate_host), crm_msg_stonith_ng, request, FALSE); rc = -EINPROGRESS; } else if (initiate_remote_stonith_op(client, request, FALSE) != NULL) { rc = -EINPROGRESS; } } } else if (pcmk__str_eq(op, STONITH_OP_FENCE_HISTORY, pcmk__str_none)) { rc = stonith_fence_history(request, &data, remote_peer, call_options); if (call_options & st_opt_discard_reply) { /* we don't expect answers to the broadcast * we might have sent out */ free_xml(data); return pcmk_ok; } } else if (pcmk__str_eq(op, STONITH_OP_DEVICE_ADD, pcmk__str_none)) { const char *device_id = NULL; if (allowed) { rc = stonith_device_register(request, &device_id, FALSE); } else { rc = -EACCES; } do_stonith_notify_device(call_options, op, rc, device_id); } else if (pcmk__str_eq(op, STONITH_OP_DEVICE_DEL, pcmk__str_none)) { xmlNode *dev = get_xpath_object("//" F_STONITH_DEVICE, request, LOG_ERR); const char *device_id = crm_element_value(dev, XML_ATTR_ID); if (allowed) { rc = stonith_device_remove(device_id, FALSE); } else { rc = -EACCES; } do_stonith_notify_device(call_options, op, rc, device_id); } else if (pcmk__str_eq(op, STONITH_OP_LEVEL_ADD, pcmk__str_none)) { char *device_id = NULL; if (allowed) { rc = stonith_level_register(request, &device_id); } else { rc = -EACCES; } do_stonith_notify_level(call_options, op, rc, device_id); free(device_id); } else if (pcmk__str_eq(op, STONITH_OP_LEVEL_DEL, pcmk__str_none)) { char *device_id = NULL; if (allowed) { rc = stonith_level_remove(request, &device_id); } else { rc = -EACCES; } do_stonith_notify_level(call_options, op, rc, device_id); } else if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) { int node_id = 0; const char *name = NULL; crm_element_value_int(request, XML_ATTR_ID, &node_id); name = crm_element_value(request, XML_ATTR_UNAME); reap_crm_member(node_id, name); return pcmk_ok; } else { crm_err("Unknown IPC request %s from %s %s", op, ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client))); } done: if (rc == -EACCES) { crm_warn("Rejecting IPC request '%s' from unprivileged client %s", crm_str(op), pcmk__client_name(client)); } /* Always reply unless the request is in process still. * If in progress, a reply will happen async after the request * processing is finished */ if (rc != -EINPROGRESS) { crm_trace("Reply handling: %p %u %u %d %d %s", client, client?client->request_id:0, id, pcmk_is_set(call_options, st_opt_sync_call), call_options, crm_element_value(request, F_STONITH_CALLOPTS)); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } reply = stonith_construct_reply(request, output, data, rc); stonith_send_reply(reply, call_options, remote_peer, client_id); } free(output); free_xml(data); free_xml(reply); return rc; } static void handle_reply(pcmk__client_t *client, xmlNode *request, const char *remote_peer) { const char *op = crm_element_value(request, F_STONITH_OPERATION); if (pcmk__str_eq(op, STONITH_OP_QUERY, pcmk__str_none)) { process_remote_stonith_query(request); } else if (pcmk__str_eq(op, T_STONITH_NOTIFY, pcmk__str_none)) { process_remote_stonith_exec(request); } else if (pcmk__str_eq(op, STONITH_OP_FENCE, pcmk__str_none)) { /* Reply to a complex fencing op */ process_remote_stonith_exec(request); } else { crm_err("Unknown %s reply from %s %s", op, ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client))); crm_log_xml_warn(request, "UnknownOp"); } } void stonith_command(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *request, const char *remote_peer) { int call_options = 0; int rc = 0; gboolean is_reply = FALSE; /* Copy op for reporting. The original might get freed by handle_reply() * before we use it in crm_debug(): * handle_reply() * |- process_remote_stonith_exec() * |-- remote_op_done() * |--- handle_local_reply_and_notify() * |---- crm_xml_add(...F_STONITH_OPERATION...) * |--- free_xml(op->request) */ char *op = crm_element_value_copy(request, F_STONITH_OPERATION); if (get_xpath_object("//" T_STONITH_REPLY, request, LOG_NEVER)) { is_reply = TRUE; } crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); crm_debug("Processing %s%s %u from %s %s with call options 0x%08x", op, (is_reply? " reply" : ""), id, ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client)), call_options); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(client == NULL || client->request_id == id); } if (is_reply) { handle_reply(client, request, remote_peer); } else { rc = handle_request(client, id, flags, request, remote_peer); } crm_debug("Processed %s%s from %s %s: %s (rc=%d)", op, (is_reply? " reply" : ""), ((client == NULL)? "peer" : "client"), ((client == NULL)? remote_peer : pcmk__client_name(client)), ((rc > 0)? "" : pcmk_strerror(rc)), rc); free(op); } diff --git a/daemons/fenced/pacemaker-fenced.h b/daemons/fenced/pacemaker-fenced.h index 13cf6dccfb..e342692240 100644 --- a/daemons/fenced/pacemaker-fenced.h +++ b/daemons/fenced/pacemaker-fenced.h @@ -1,272 +1,273 @@ /* * Copyright 2009-2020 the Pacemaker project contributors * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include // uint32_t, uint64_t #include /*! * \internal * \brief Check to see if target was fenced in the last few seconds. * \param tolerance, The number of seconds to look back in time * \param target, The node to search for * \param action, The action we want to match. * * \retval FALSE, not match * \retval TRUE, fencing operation took place in the last 'tolerance' number of seconds. */ gboolean stonith_check_fence_tolerance(int tolerance, const char *target, const char *action); typedef struct stonith_device_s { char *id; char *agent; char *namespace; /*! list of actions that must execute on the target node. Used for unfencing */ char *on_target_actions; GListPtr targets; time_t targets_age; gboolean has_attr_map; /* should nodeid parameter for victim be included in agent arguments */ gboolean include_nodeid; /* whether the cluster should automatically unfence nodes with the device */ gboolean automatic_unfencing; guint priority; uint32_t flags; // Group of enum st_device_flags GHashTable *params; GHashTable *aliases; GList *pending_ops; + mainloop_timer_t *timer; crm_trigger_t *work; xmlNode *agent_metadata; /*! A verified device is one that has contacted the * agent successfully to perform a monitor operation */ gboolean verified; gboolean cib_registered; gboolean api_registered; gboolean dirty; } stonith_device_t; /* These values are used to index certain arrays by "phase". Usually an * operation has only one "phase", so phase is always zero. However, some * reboots are remapped to "off" then "on", in which case "reboot" will be * phase 0, "off" will be phase 1 and "on" will be phase 2. */ enum st_remap_phase { st_phase_requested = 0, st_phase_off = 1, st_phase_on = 2, st_phase_max = 3 }; /* These values provide additional information for STONITH's asynchronous reply response. * The st_reply_opt_merged value indicates an operation that has been merged and completed without being executed. */ enum st_replay_option { st_reply_opt_none = 0x00000000, st_reply_opt_merged = 0x00000001, }; typedef struct remote_fencing_op_s { /* The unique id associated with this operation */ char *id; /*! The node this operation will fence */ char *target; /*! The fencing action to perform on the target. (reboot, on, off) */ char *action; /*! When was the fencing action recorded (seconds since epoch) */ time_t created; /*! Marks if the final notifications have been sent to local stonith clients. */ gboolean notify_sent; /*! The number of query replies received */ guint replies; /*! The number of query replies expected */ guint replies_expected; /*! Does this node own control of this operation */ gboolean owner; /*! After query is complete, This the high level timer that expires the entire operation */ guint op_timer_total; /*! This timer expires the current fencing request. Many fencing * requests may exist in a single operation */ guint op_timer_one; /*! This timer expires the query request sent out to determine * what nodes are contain what devices, and who those devices can fence */ guint query_timer; /*! This is the default timeout to use for each fencing device if no * custom timeout is received in the query. */ gint base_timeout; /*! This is the calculated total timeout an operation can take before * expiring. This is calculated by adding together all the timeout * values associated with the devices this fencing operation may call */ gint total_timeout; /*! Requested fencing delay. * Value -1 means disable any static/random fencing delays. */ int delay; /*! Delegate is the node being asked to perform a fencing action * on behalf of the node that owns the remote operation. Some operations * will involve multiple delegates. This value represents the final delegate * that is used. */ char *delegate; /*! The point at which the remote operation completed */ time_t completed; //! Group of enum stonith_call_options associated with this operation uint32_t call_options; /*! The current state of the remote operation. This indicates * what stage the op is in, query, exec, done, duplicate, failed. */ enum op_state state; /*! The node that owns the remote operation */ char *originator; /*! The local client id that initiated the fencing request */ char *client_id; /*! The client's call_id that initiated the fencing request */ int client_callid; /*! The name of client that initiated the fencing request */ char *client_name; /*! List of the received query results for all the nodes in the cpg group */ GListPtr query_results; /*! The original request that initiated the remote stonith operation */ xmlNode *request; /*! The current topology level being executed */ guint level; /*! The current operation phase being executed */ enum st_remap_phase phase; /*! Devices with automatic unfencing (always run if "on" requested, never if remapped) */ GListPtr automatic_list; /*! List of all devices at the currently executing topology level */ GListPtr devices_list; /*! Current entry in the topology device list */ GListPtr devices; /*! List of duplicate operations attached to this operation. Once this operation * completes, the duplicate operations will be closed out as well. */ GListPtr duplicates; } remote_fencing_op_t; /*! * \internal * \brief Broadcast the result of an operation to the peers. * \param op, Operation whose result should be broadcast * \param rc, Result of the operation */ void stonith_bcast_result_to_peers(remote_fencing_op_t * op, int rc, gboolean op_merged); // Fencer-specific client flags enum st_client_flags { st_callback_unknown = UINT64_C(0), st_callback_notify_fence = (UINT64_C(1) << 0), st_callback_device_add = (UINT64_C(1) << 2), st_callback_device_del = (UINT64_C(1) << 4), st_callback_notify_history = (UINT64_C(1) << 5), st_callback_notify_history_synced = (UINT64_C(1) << 6) }; /* * Complex fencing requirements are specified via fencing topologies. * A topology consists of levels; each level is a list of fencing devices. * Topologies are stored in a hash table by node name. When a node needs to be * fenced, if it has an entry in the topology table, the levels are tried * sequentially, and the devices in each level are tried sequentially. * Fencing is considered successful as soon as any level succeeds; * a level is considered successful if all its devices succeed. * Essentially, all devices at a given level are "and-ed" and the * levels are "or-ed". * * This structure is used for the topology table entries. * Topology levels start from 1, so levels[0] is unused and always NULL. */ typedef struct stonith_topology_s { int kind; /*! Node name regex or attribute name=value for which topology applies */ char *target; char *target_value; char *target_pattern; char *target_attribute; /*! Names of fencing devices at each topology level */ GListPtr levels[ST_LEVEL_MAX]; } stonith_topology_t; void init_device_list(void); void free_device_list(void); void init_topology_list(void); void free_topology_list(void); void free_stonith_remote_op_list(void); void init_stonith_remote_op_hash_table(GHashTable **table); void free_metadata_cache(void); uint64_t get_stonith_flag(const char *name); void stonith_command(pcmk__client_t *client, uint32_t id, uint32_t flags, xmlNode *op_request, const char *remote_peer); int stonith_device_register(xmlNode * msg, const char **desc, gboolean from_cib); int stonith_device_remove(const char *id, gboolean from_cib); char *stonith_level_key(xmlNode * msg, int mode); int stonith_level_kind(xmlNode * msg); int stonith_level_register(xmlNode * msg, char **desc); int stonith_level_remove(xmlNode * msg, char **desc); stonith_topology_t *find_topology_for_host(const char *host); void do_local_reply(xmlNode * notify_src, const char *client_id, gboolean sync_reply, gboolean from_peer); xmlNode *stonith_construct_reply(xmlNode * request, const char *output, xmlNode * data, int rc); void do_stonith_async_timeout_update(const char *client, const char *call_id, int timeout); void do_stonith_notify(int options, const char *type, int result, xmlNode * data); void do_stonith_notify_device(int options, const char *op, int rc, const char *desc); void do_stonith_notify_level(int options, const char *op, int rc, const char *desc); remote_fencing_op_t *initiate_remote_stonith_op(pcmk__client_t *client, xmlNode *request, gboolean manual_ack); int process_remote_stonith_exec(xmlNode * msg); int process_remote_stonith_query(xmlNode * msg); void *create_remote_stonith_op(const char *client, xmlNode * request, gboolean peer); int stonith_fence_history(xmlNode *msg, xmlNode **output, const char *remote_peer, int options); void stonith_fence_history_trim(void); bool fencing_peer_active(crm_node_t *peer); int stonith_manual_ack(xmlNode * msg, remote_fencing_op_t * op); gboolean string_in_list(GListPtr list, const char *item); gboolean node_has_attr(const char *node, const char *name, const char *value); extern char *stonith_our_uname; extern gboolean stand_alone; extern GHashTable *device_list; extern GHashTable *topology; extern long stonith_watchdog_timeout_ms; extern GHashTable *stonith_remote_op_list; diff --git a/include/crm/common/internal.h b/include/crm/common/internal.h index f69abe8c0c..63bfd2cddf 100644 --- a/include/crm/common/internal.h +++ b/include/crm/common/internal.h @@ -1,342 +1,343 @@ /* * Copyright 2015-2021 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #ifndef CRM_COMMON_INTERNAL__H #define CRM_COMMON_INTERNAL__H #include // getpid() #include // bool #include // uint8_t, uint64_t #include // strcmp() #include // open() #include // uid_t, gid_t, pid_t #include // guint, GList, GHashTable #include // xmlNode #include // crm_strdup_printf() #include // do_crm_log_unlikely(), etc. #include // mainloop_io_t, struct ipc_client_callbacks #include // Internal ACL-related utilities (from acl.c) char *pcmk__uid2username(uid_t uid); const char *pcmk__update_acl_user(xmlNode *request, const char *field, const char *peer_user); static inline bool pcmk__is_privileged(const char *user) { return user && (!strcmp(user, CRM_DAEMON_USER) || !strcmp(user, "root")); } #if SUPPORT_CIBSECRETS // Internal CIB utilities (from cib_secrets.c) */ int pcmk__substitute_secrets(const char *rsc_id, GHashTable *params); #endif /* internal digest-related utilities (from digest.c) */ bool pcmk__verify_digest(xmlNode *input, const char *expected); /* internal I/O utilities (from io.c) */ int pcmk__real_path(const char *path, char **resolved_path); char *pcmk__series_filename(const char *directory, const char *series, int sequence, bool bzip); int pcmk__read_series_sequence(const char *directory, const char *series, unsigned int *seq); void pcmk__write_series_sequence(const char *directory, const char *series, unsigned int sequence, int max); int pcmk__chown_series_sequence(const char *directory, const char *series, uid_t uid, gid_t gid); int pcmk__build_path(const char *path_c, mode_t mode); bool pcmk__daemon_can_write(const char *dir, const char *file); void pcmk__sync_directory(const char *name); int pcmk__file_contents(const char *filename, char **contents); int pcmk__write_sync(int fd, const char *contents); int pcmk__set_nonblocking(int fd); const char *pcmk__get_tmpdir(void); void pcmk__close_fds_in_child(bool); /*! * \internal * \brief Open /dev/null to consume next available file descriptor * * Open /dev/null, disregarding the result. This is intended when daemonizing to * be able to null stdin, stdout, and stderr. * * \param[in] flags O_RDONLY (stdin) or O_WRONLY (stdout and stderr) */ static inline void pcmk__open_devnull(int flags) { // Static analysis clutter // cppcheck-suppress leakReturnValNotUsed (void) open("/dev/null", flags); } /* internal main loop utilities (from mainloop.c) */ int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata, struct ipc_client_callbacks *callbacks, mainloop_io_t **source); +guint pcmk__mainloop_timer_get_period(mainloop_timer_t *timer); /* internal messaging utilities (from messages.c) */ const char *pcmk__message_name(const char *name); /* internal procfs utilities (from procfs.c) */ pid_t pcmk__procfs_pid_of(const char *name); unsigned int pcmk__procfs_num_cores(void); /* internal XML schema functions (from xml.c) */ void crm_schema_init(void); void crm_schema_cleanup(void); /* internal functions related to process IDs (from pid.c) */ /*! * \internal * \brief Check whether process exists (by PID and optionally executable path) * * \param[in] pid PID of process to check * \param[in] daemon If not NULL, path component to match with procfs entry * * \return Standard Pacemaker return code * \note Particular return codes of interest include pcmk_rc_ok for alive, * ESRCH for process is not alive (verified by kill and/or executable path * match), EACCES for caller unable or not allowed to check. A result of * "alive" is less reliable when \p daemon is not provided or procfs is * not available, since there is no guarantee that the PID has not been * recycled for another process. * \note This function cannot be used to verify \e authenticity of the process. */ int pcmk__pid_active(pid_t pid, const char *daemon); int pcmk__read_pidfile(const char *filename, pid_t *pid); int pcmk__pidfile_matches(const char *filename, pid_t expected_pid, const char *expected_name, pid_t *pid); int pcmk__lock_pidfile(const char *filename, const char *name); /* internal functions related to resource operations (from operations.c) */ // printf-style format to create operation ID from resource, action, interval #define PCMK__OP_FMT "%s_%s_%u" char *pcmk__op_key(const char *rsc_id, const char *op_type, guint interval_ms); char *pcmk__notify_key(const char *rsc_id, const char *notify_type, const char *op_type); char *pcmk__transition_key(int transition_id, int action_id, int target_rc, const char *node); void pcmk__filter_op_for_digest(xmlNode *param_set); // bitwise arithmetic utilities /*! * \internal * \brief Set specified flags in a flag group * * \param[in] function Function name of caller * \param[in] line Line number of caller * \param[in] log_level Log a message at this level * \param[in] flag_type Label describing this flag group (for logging) * \param[in] target Name of object whose flags these are (for logging) * \param[in] flag_group Flag group being manipulated * \param[in] flags Which flags in the group should be set * \param[in] flags_str Readable equivalent of \p flags (for logging) * * \return Possibly modified flag group */ static inline uint64_t pcmk__set_flags_as(const char *function, int line, uint8_t log_level, const char *flag_type, const char *target, uint64_t flag_group, uint64_t flags, const char *flags_str) { uint64_t result = flag_group | flags; if (result != flag_group) { do_crm_log_unlikely(log_level, "%s flags 0x%.8llx (%s) for %s set by %s:%d", ((flag_type == NULL)? "Group of" : flag_type), (unsigned long long) flags, ((flags_str == NULL)? "flags" : flags_str), ((target == NULL)? "target" : target), function, line); } return result; } /*! * \internal * \brief Clear specified flags in a flag group * * \param[in] function Function name of caller * \param[in] line Line number of caller * \param[in] log_level Log a message at this level * \param[in] flag_type Label describing this flag group (for logging) * \param[in] target Name of object whose flags these are (for logging) * \param[in] flag_group Flag group being manipulated * \param[in] flags Which flags in the group should be cleared * \param[in] flags_str Readable equivalent of \p flags (for logging) * * \return Possibly modified flag group */ static inline uint64_t pcmk__clear_flags_as(const char *function, int line, uint8_t log_level, const char *flag_type, const char *target, uint64_t flag_group, uint64_t flags, const char *flags_str) { uint64_t result = flag_group & ~flags; if (result != flag_group) { do_crm_log_unlikely(log_level, "%s flags 0x%.8llx (%s) for %s cleared by %s:%d", ((flag_type == NULL)? "Group of" : flag_type), (unsigned long long) flags, ((flags_str == NULL)? "flags" : flags_str), ((target == NULL)? "target" : target), function, line); } return result; } // miscellaneous utilities (from utils.c) void pcmk__daemonize(const char *name, const char *pidfile); void pcmk__panic(const char *origin); pid_t pcmk__locate_sbd(void); void pcmk__sleep_ms(unsigned int ms); extern int pcmk__score_red; extern int pcmk__score_green; extern int pcmk__score_yellow; /*! * \internal * \brief Resize a dynamically allocated memory block * * \param[in] ptr Memory block to resize (or NULL to allocate new memory) * \param[in] size New size of memory block in bytes (must be > 0) * * \return Pointer to resized memory block * * \note This asserts on error, so the result is guaranteed to be non-NULL * (which is the main advantage of this over directly using realloc()). */ static inline void * pcmk__realloc(void *ptr, size_t size) { void *new_ptr; // realloc(p, 0) can replace free(p) but this wrapper can't CRM_ASSERT(size > 0); new_ptr = realloc(ptr, size); if (new_ptr == NULL) { free(ptr); abort(); } return new_ptr; } /* Error domains for use with g_set_error (from results.c) */ GQuark pcmk__rc_error_quark(void); GQuark pcmk__exitc_error_quark(void); #define PCMK__RC_ERROR pcmk__rc_error_quark() #define PCMK__EXITC_ERROR pcmk__exitc_error_quark() static inline char * pcmk__getpid_s(void) { return crm_strdup_printf("%lu", (unsigned long) getpid()); } // More efficient than g_list_length(list) == 1 static inline bool pcmk__list_of_1(GList *list) { return list && (list->next == NULL); } // More efficient than g_list_length(list) > 1 static inline bool pcmk__list_of_multiple(GList *list) { return list && (list->next != NULL); } /* convenience functions for failure-related node attributes */ #define PCMK__FAIL_COUNT_PREFIX "fail-count" #define PCMK__LAST_FAILURE_PREFIX "last-failure" /*! * \internal * \brief Generate a failure-related node attribute name for a resource * * \param[in] prefix Start of attribute name * \param[in] rsc_id Resource name * \param[in] op Operation name * \param[in] interval_ms Operation interval * * \return Newly allocated string with attribute name * * \note Failure attributes are named like PREFIX-RSC#OP_INTERVAL (for example, * "fail-count-myrsc#monitor_30000"). The '#' is used because it is not * a valid character in a resource ID, to reliably distinguish where the * operation name begins. The '_' is used simply to be more comparable to * action labels like "myrsc_monitor_30000". */ static inline char * pcmk__fail_attr_name(const char *prefix, const char *rsc_id, const char *op, guint interval_ms) { CRM_CHECK(prefix && rsc_id && op, return NULL); return crm_strdup_printf("%s-%s#%s_%u", prefix, rsc_id, op, interval_ms); } static inline char * pcmk__failcount_name(const char *rsc_id, const char *op, guint interval_ms) { return pcmk__fail_attr_name(PCMK__FAIL_COUNT_PREFIX, rsc_id, op, interval_ms); } static inline char * pcmk__lastfailure_name(const char *rsc_id, const char *op, guint interval_ms) { return pcmk__fail_attr_name(PCMK__LAST_FAILURE_PREFIX, rsc_id, op, interval_ms); } // internal resource agent functions (from agents.c) int pcmk__effective_rc(int rc); #endif /* CRM_COMMON_INTERNAL__H */ diff --git a/lib/common/mainloop.c b/lib/common/mainloop.c index 2f00e314cd..75f24e2323 100644 --- a/lib/common/mainloop.c +++ b/lib/common/mainloop.c @@ -1,1455 +1,1471 @@ /* * 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 Lesser General Public License * version 2.1 or later (LGPLv2.1+) WITHOUT ANY WARRANTY. */ #include #ifndef _GNU_SOURCE # define _GNU_SOURCE #endif #include #include #include #include #include #include #include #include #include #include struct mainloop_child_s { pid_t pid; char *desc; unsigned timerid; gboolean timeout; void *privatedata; enum mainloop_child_flags flags; /* Called when a process dies */ void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode); }; struct trigger_s { GSource source; gboolean running; gboolean trigger; void *user_data; guint id; }; +struct mainloop_timer_s { + guint id; + guint period_ms; + bool repeat; + char *name; + GSourceFunc cb; + void *userdata; +}; + static gboolean crm_trigger_prepare(GSource * source, gint * timeout) { crm_trigger_t *trig = (crm_trigger_t *) source; /* cluster-glue's FD and IPC related sources make use of * g_source_add_poll() but do not set a timeout in their prepare * functions * * This means mainloop's poll() will block until an event for one * of these sources occurs - any /other/ type of source, such as * this one or g_idle_*, that doesn't use g_source_add_poll() is * S-O-L and won't be processed until there is something fd-based * happens. * * Luckily the timeout we can set here affects all sources and * puts an upper limit on how long poll() can take. * * So unconditionally set a small-ish timeout, not too small that * we're in constant motion, which will act as an upper bound on * how long the signal handling might be delayed for. */ *timeout = 500; /* Timeout in ms */ return trig->trigger; } static gboolean crm_trigger_check(GSource * source) { crm_trigger_t *trig = (crm_trigger_t *) source; return trig->trigger; } static gboolean crm_trigger_dispatch(GSource * source, GSourceFunc callback, gpointer userdata) { int rc = TRUE; crm_trigger_t *trig = (crm_trigger_t *) source; if (trig->running) { /* Wait until the existing job is complete before starting the next one */ return TRUE; } trig->trigger = FALSE; if (callback) { rc = callback(trig->user_data); if (rc < 0) { crm_trace("Trigger handler %p not yet complete", trig); trig->running = TRUE; rc = TRUE; } } return rc; } static void crm_trigger_finalize(GSource * source) { crm_trace("Trigger %p destroyed", source); } static GSourceFuncs crm_trigger_funcs = { crm_trigger_prepare, crm_trigger_check, crm_trigger_dispatch, crm_trigger_finalize, }; static crm_trigger_t * mainloop_setup_trigger(GSource * source, int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { crm_trigger_t *trigger = NULL; trigger = (crm_trigger_t *) source; trigger->id = 0; trigger->trigger = FALSE; trigger->user_data = userdata; if (dispatch) { g_source_set_callback(source, dispatch, trigger, NULL); } g_source_set_priority(source, priority); g_source_set_can_recurse(source, FALSE); trigger->id = g_source_attach(source, NULL); return trigger; } void mainloop_trigger_complete(crm_trigger_t * trig) { crm_trace("Trigger handler %p complete", trig); trig->running = FALSE; } /* If dispatch returns: * -1: Job running but not complete * 0: Remove the trigger from mainloop * 1: Leave the trigger in mainloop */ crm_trigger_t * mainloop_add_trigger(int priority, int (*dispatch) (gpointer user_data), gpointer userdata) { GSource *source = NULL; CRM_ASSERT(sizeof(crm_trigger_t) > sizeof(GSource)); source = g_source_new(&crm_trigger_funcs, sizeof(crm_trigger_t)); CRM_ASSERT(source != NULL); return mainloop_setup_trigger(source, priority, dispatch, userdata); } void mainloop_set_trigger(crm_trigger_t * source) { if(source) { source->trigger = TRUE; } } gboolean mainloop_destroy_trigger(crm_trigger_t * source) { GSource *gs = NULL; if(source == NULL) { return TRUE; } gs = (GSource *)source; g_source_destroy(gs); /* Remove from mainloop, ref_count-- */ g_source_unref(gs); /* The caller no longer carries a reference to source * * At this point the source should be free'd, * unless we're currently processing said * source, in which case mainloop holds an * additional reference and it will be free'd * once our processing completes */ return TRUE; } // Define a custom glib source for signal handling // Data structure for custom glib source typedef struct signal_s { crm_trigger_t trigger; // trigger that invoked source (must be first) void (*handler) (int sig); // signal handler int signal; // signal that was received } crm_signal_t; // Table to associate signal handlers with signal numbers static crm_signal_t *crm_signals[NSIG]; /*! * \internal * \brief Dispatch an event from custom glib source for signals * * Given an signal event, clear the event trigger and call any registered * signal handler. * * \param[in] source glib source that triggered this dispatch * \param[in] callback (ignored) * \param[in] userdata (ignored) */ static gboolean crm_signal_dispatch(GSource * source, GSourceFunc callback, gpointer userdata) { crm_signal_t *sig = (crm_signal_t *) source; if(sig->signal != SIGCHLD) { crm_notice("Caught '%s' signal "CRM_XS" %d (%s handler)", strsignal(sig->signal), sig->signal, (sig->handler? "invoking" : "no")); } sig->trigger.trigger = FALSE; if (sig->handler) { sig->handler(sig->signal); } return TRUE; } /*! * \internal * \brief Handle a signal by setting a trigger for signal source * * \param[in] sig Signal number that was received * * \note This is the true signal handler for the mainloop signal source, and * must be async-safe. */ static void mainloop_signal_handler(int sig) { if (sig > 0 && sig < NSIG && crm_signals[sig] != NULL) { mainloop_set_trigger((crm_trigger_t *) crm_signals[sig]); } } // Functions implementing our custom glib source for signal handling static GSourceFuncs crm_signal_funcs = { crm_trigger_prepare, crm_trigger_check, crm_signal_dispatch, crm_trigger_finalize, }; /*! * \internal * \brief Set a true signal handler * * signal()-like interface to sigaction() * * \param[in] sig Signal number to register handler for * \param[in] dispatch Signal handler * * \return The previous value of the signal handler, or SIG_ERR on error * \note The dispatch function must be async-safe. */ sighandler_t crm_signal_handler(int sig, sighandler_t dispatch) { sigset_t mask; struct sigaction sa; struct sigaction old; if (sigemptyset(&mask) < 0) { crm_err("Could not set handler for signal %d: %s", sig, pcmk_strerror(errno)); return SIG_ERR; } memset(&sa, 0, sizeof(struct sigaction)); sa.sa_handler = dispatch; sa.sa_flags = SA_RESTART; sa.sa_mask = mask; if (sigaction(sig, &sa, &old) < 0) { crm_err("Could not set handler for signal %d: %s", sig, pcmk_strerror(errno)); return SIG_ERR; } return old.sa_handler; } static void mainloop_destroy_signal_entry(int sig) { crm_signal_t *tmp = crm_signals[sig]; crm_signals[sig] = NULL; crm_trace("Destroying signal %d", sig); mainloop_destroy_trigger((crm_trigger_t *) tmp); } /*! * \internal * \brief Add a signal handler to a mainloop * * \param[in] sig Signal number to handle * \param[in] dispatch Signal handler function * * \note The true signal handler merely sets a mainloop trigger to call this * dispatch function via the mainloop. Therefore, the dispatch function * does not need to be async-safe. */ gboolean mainloop_add_signal(int sig, void (*dispatch) (int sig)) { GSource *source = NULL; int priority = G_PRIORITY_HIGH - 1; if (sig == SIGTERM) { /* TERM is higher priority than other signals, * signals are higher priority than other ipc. * Yes, minus: smaller is "higher" */ priority--; } if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signals[sig] != NULL && crm_signals[sig]->handler == dispatch) { crm_trace("Signal handler for %d is already installed", sig); return TRUE; } else if (crm_signals[sig] != NULL) { crm_err("Different signal handler for %d is already installed", sig); return FALSE; } CRM_ASSERT(sizeof(crm_signal_t) > sizeof(GSource)); source = g_source_new(&crm_signal_funcs, sizeof(crm_signal_t)); crm_signals[sig] = (crm_signal_t *) mainloop_setup_trigger(source, priority, NULL, NULL); CRM_ASSERT(crm_signals[sig] != NULL); crm_signals[sig]->handler = dispatch; crm_signals[sig]->signal = sig; if (crm_signal_handler(sig, mainloop_signal_handler) == SIG_ERR) { mainloop_destroy_signal_entry(sig); return FALSE; } #if 0 /* If we want signals to interrupt mainloop's poll(), instead of waiting for * the timeout, then we should call siginterrupt() below * * For now, just enforce a low timeout */ if (siginterrupt(sig, 1) < 0) { crm_perror(LOG_INFO, "Could not enable system call interruptions for signal %d", sig); } #endif return TRUE; } gboolean mainloop_destroy_signal(int sig) { if (sig >= NSIG || sig < 0) { crm_err("Signal %d is out of range", sig); return FALSE; } else if (crm_signal_handler(sig, NULL) == SIG_ERR) { crm_perror(LOG_ERR, "Could not uninstall signal handler for signal %d", sig); return FALSE; } else if (crm_signals[sig] == NULL) { return TRUE; } mainloop_destroy_signal_entry(sig); return TRUE; } static qb_array_t *gio_map = NULL; void mainloop_cleanup(void) { if (gio_map) { qb_array_free(gio_map); } for (int sig = 0; sig < NSIG; ++sig) { mainloop_destroy_signal_entry(sig); } } /* * libqb... */ struct gio_to_qb_poll { int32_t is_used; guint source; int32_t events; void *data; qb_ipcs_dispatch_fn_t fn; enum qb_loop_priority p; }; static gboolean gio_read_socket(GIOChannel * gio, GIOCondition condition, gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; gint fd = g_io_channel_unix_get_fd(gio); crm_trace("%p.%d %d", data, fd, condition); /* if this assert get's hit, then there is a race condition between * when we destroy a fd and when mainloop actually gives it up */ CRM_ASSERT(adaptor->is_used > 0); return (adaptor->fn(fd, condition, adaptor->data) == 0); } static void gio_poll_destroy(gpointer data) { struct gio_to_qb_poll *adaptor = (struct gio_to_qb_poll *)data; adaptor->is_used--; CRM_ASSERT(adaptor->is_used >= 0); if (adaptor->is_used == 0) { crm_trace("Marking adaptor %p unused", adaptor); adaptor->source = 0; } } /*! * \internal * \brief Convert libqb's poll priority into GLib's one * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching GLib's priority */ static gint conv_prio_libqb2glib(enum qb_loop_priority prio) { gint ret = G_PRIORITY_DEFAULT; switch (prio) { case QB_LOOP_LOW: ret = G_PRIORITY_LOW; break; case QB_LOOP_HIGH: ret = G_PRIORITY_HIGH; break; default: crm_trace("Invalid libqb's loop priority %d, assuming QB_LOOP_MED", prio); /* fall-through */ case QB_LOOP_MED: break; } return ret; } /*! * \internal * \brief Convert libqb's poll priority to rate limiting spec * * \param[in] prio libqb's poll priority (#QB_LOOP_MED assumed as fallback) * * \return best matching rate limiting spec */ static enum qb_ipcs_rate_limit conv_libqb_prio2ratelimit(enum qb_loop_priority prio) { /* this is an inversion of what libqb's qb_ipcs_request_rate_limit does */ enum qb_ipcs_rate_limit ret = QB_IPCS_RATE_NORMAL; switch (prio) { case QB_LOOP_LOW: ret = QB_IPCS_RATE_SLOW; break; case QB_LOOP_HIGH: ret = QB_IPCS_RATE_FAST; break; default: crm_trace("Invalid libqb's loop priority %d, assuming QB_LOOP_MED", prio); /* fall-through */ case QB_LOOP_MED: break; } return ret; } static int32_t gio_poll_dispatch_update(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn, int32_t add) { struct gio_to_qb_poll *adaptor; GIOChannel *channel; int32_t res = 0; res = qb_array_index(gio_map, fd, (void **)&adaptor); if (res < 0) { crm_err("Array lookup failed for fd=%d: %d", fd, res); return res; } crm_trace("Adding fd=%d to mainloop as adaptor %p", fd, adaptor); if (add && adaptor->source) { crm_err("Adaptor for descriptor %d is still in-use", fd); return -EEXIST; } if (!add && !adaptor->is_used) { crm_err("Adaptor for descriptor %d is not in-use", fd); return -ENOENT; } /* channel is created with ref_count = 1 */ channel = g_io_channel_unix_new(fd); if (!channel) { crm_err("No memory left to add fd=%d", fd); return -ENOMEM; } if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } /* Because unlike the poll() API, glib doesn't tell us about HUPs by default */ evts |= (G_IO_HUP | G_IO_NVAL | G_IO_ERR); adaptor->fn = fn; adaptor->events = evts; adaptor->data = data; adaptor->p = p; adaptor->is_used++; adaptor->source = g_io_add_watch_full(channel, conv_prio_libqb2glib(p), evts, gio_read_socket, adaptor, gio_poll_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after gio_poll_destroy() completes */ g_io_channel_unref(channel); crm_trace("Added to mainloop with gsource id=%d", adaptor->source); if (adaptor->source > 0) { return 0; } return -EINVAL; } static int32_t gio_poll_dispatch_add(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_TRUE); } static int32_t gio_poll_dispatch_mod(enum qb_loop_priority p, int32_t fd, int32_t evts, void *data, qb_ipcs_dispatch_fn_t fn) { return gio_poll_dispatch_update(p, fd, evts, data, fn, QB_FALSE); } static int32_t gio_poll_dispatch_del(int32_t fd) { struct gio_to_qb_poll *adaptor; crm_trace("Looking for fd=%d", fd); if (qb_array_index(gio_map, fd, (void **)&adaptor) == 0) { if (adaptor->source) { g_source_remove(adaptor->source); adaptor->source = 0; } } return 0; } struct qb_ipcs_poll_handlers gio_poll_funcs = { .job_add = NULL, .dispatch_add = gio_poll_dispatch_add, .dispatch_mod = gio_poll_dispatch_mod, .dispatch_del = gio_poll_dispatch_del, }; static enum qb_ipc_type pick_ipc_type(enum qb_ipc_type requested) { const char *env = getenv("PCMK_ipc_type"); if (env && strcmp("shared-mem", env) == 0) { return QB_IPC_SHM; } else if (env && strcmp("socket", env) == 0) { return QB_IPC_SOCKET; } else if (env && strcmp("posix", env) == 0) { return QB_IPC_POSIX_MQ; } else if (env && strcmp("sysv", env) == 0) { return QB_IPC_SYSV_MQ; } else if (requested == QB_IPC_NATIVE) { /* We prefer shared memory because the server never blocks on * send. If part of a message fits into the socket, libqb * needs to block until the remainder can be sent also. * Otherwise the client will wait forever for the remaining * bytes. */ return QB_IPC_SHM; } return requested; } qb_ipcs_service_t * mainloop_add_ipc_server(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks) { return mainloop_add_ipc_server_with_prio(name, type, callbacks, QB_LOOP_MED); } qb_ipcs_service_t * mainloop_add_ipc_server_with_prio(const char *name, enum qb_ipc_type type, struct qb_ipcs_service_handlers *callbacks, enum qb_loop_priority prio) { int rc = 0; qb_ipcs_service_t *server = NULL; if (gio_map == NULL) { gio_map = qb_array_create_2(64, sizeof(struct gio_to_qb_poll), 1); } server = qb_ipcs_create(name, 0, pick_ipc_type(type), callbacks); if (server == NULL) { crm_err("Could not create %s IPC server: %s (%d)", name, pcmk_strerror(rc), rc); return NULL; } if (prio != QB_LOOP_MED) { qb_ipcs_request_rate_limit(server, conv_libqb_prio2ratelimit(prio)); } /* All clients should use at least ipc_buffer_max as their buffer size */ qb_ipcs_enforce_buffer_size(server, crm_ipc_default_buffer_size()); qb_ipcs_poll_handlers_set(server, &gio_poll_funcs); rc = qb_ipcs_run(server); if (rc < 0) { crm_err("Could not start %s IPC server: %s (%d)", name, pcmk_strerror(rc), rc); return NULL; } return server; } void mainloop_del_ipc_server(qb_ipcs_service_t * server) { if (server) { qb_ipcs_destroy(server); } } struct mainloop_io_s { char *name; void *userdata; int fd; guint source; crm_ipc_t *ipc; GIOChannel *channel; int (*dispatch_fn_ipc) (const char *buffer, ssize_t length, gpointer userdata); int (*dispatch_fn_io) (gpointer userdata); void (*destroy_fn) (gpointer userdata); }; static gboolean mainloop_gio_callback(GIOChannel * gio, GIOCondition condition, gpointer data) { gboolean keep = TRUE; mainloop_io_t *client = data; CRM_ASSERT(client->fd == g_io_channel_unix_get_fd(gio)); if (condition & G_IO_IN) { if (client->ipc) { long rc = 0; int max = 10; do { rc = crm_ipc_read(client->ipc); if (rc <= 0) { crm_trace("Message acquisition from %s[%p] failed: %s (%ld)", client->name, client, pcmk_strerror(rc), rc); } else if (client->dispatch_fn_ipc) { const char *buffer = crm_ipc_buffer(client->ipc); crm_trace("New message from %s[%p] = %ld (I/O condition=%d)", client->name, client, rc, condition); if (client->dispatch_fn_ipc(buffer, rc, client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); keep = FALSE; } } } while (keep && rc > 0 && --max > 0); } else { crm_trace("New message from %s[%p] %u", client->name, client, condition); if (client->dispatch_fn_io) { if (client->dispatch_fn_io(client->userdata) < 0) { crm_trace("Connection to %s no longer required", client->name); keep = FALSE; } } } } if (client->ipc && crm_ipc_connected(client->ipc) == FALSE) { crm_err("Connection to %s closed " CRM_XS "client=%p condition=%d", client->name, client, condition); keep = FALSE; } else if (condition & (G_IO_HUP | G_IO_NVAL | G_IO_ERR)) { crm_trace("The connection %s[%p] has been closed (I/O condition=%d)", client->name, client, condition); keep = FALSE; } else if ((condition & G_IO_IN) == 0) { /* #define GLIB_SYSDEF_POLLIN =1 #define GLIB_SYSDEF_POLLPRI =2 #define GLIB_SYSDEF_POLLOUT =4 #define GLIB_SYSDEF_POLLERR =8 #define GLIB_SYSDEF_POLLHUP =16 #define GLIB_SYSDEF_POLLNVAL =32 typedef enum { G_IO_IN GLIB_SYSDEF_POLLIN, G_IO_OUT GLIB_SYSDEF_POLLOUT, G_IO_PRI GLIB_SYSDEF_POLLPRI, G_IO_ERR GLIB_SYSDEF_POLLERR, G_IO_HUP GLIB_SYSDEF_POLLHUP, G_IO_NVAL GLIB_SYSDEF_POLLNVAL } GIOCondition; A bitwise combination representing a condition to watch for on an event source. G_IO_IN There is data to read. G_IO_OUT Data can be written (without blocking). G_IO_PRI There is urgent data to read. G_IO_ERR Error condition. G_IO_HUP Hung up (the connection has been broken, usually for pipes and sockets). G_IO_NVAL Invalid request. The file descriptor is not open. */ crm_err("Strange condition: %d", condition); } /* keep == FALSE results in mainloop_gio_destroy() being called * just before the source is removed from mainloop */ return keep; } static void mainloop_gio_destroy(gpointer c) { mainloop_io_t *client = c; char *c_name = strdup(client->name); /* client->source is valid but about to be destroyed (ref_count == 0) in gmain.c * client->channel will still have ref_count > 0... should be == 1 */ crm_trace("Destroying client %s[%p]", c_name, c); if (client->ipc) { crm_ipc_close(client->ipc); } if (client->destroy_fn) { void (*destroy_fn) (gpointer userdata) = client->destroy_fn; client->destroy_fn = NULL; destroy_fn(client->userdata); } if (client->ipc) { crm_ipc_t *ipc = client->ipc; client->ipc = NULL; crm_ipc_destroy(ipc); } crm_trace("Destroyed client %s[%p]", c_name, c); free(client->name); client->name = NULL; free(client); free(c_name); } /*! * \brief Connect to IPC and add it as a main loop source * * \param[in] ipc IPC connection to add * \param[in] priority Event source priority to use for connection * \param[in] userdata Data to register with callbacks * \param[in] callbacks Dispatch and destroy callbacks for connection * \param[out] source Newly allocated event source * * \return Standard Pacemaker return code * * \note On failure, the caller is still responsible for ipc. On success, the * caller should call mainloop_del_ipc_client() when source is no longer * needed, which will lead to the disconnection of the IPC later in the * main loop if it is connected. However the IPC disconnects, * mainloop_gio_destroy() will free ipc and source after calling the * destroy callback. */ int pcmk__add_mainloop_ipc(crm_ipc_t *ipc, int priority, void *userdata, struct ipc_client_callbacks *callbacks, mainloop_io_t **source) { CRM_CHECK((ipc != NULL) && (callbacks != NULL), return EINVAL); if (!crm_ipc_connect(ipc)) { int rc = errno; crm_debug("Connection to %s failed: %d", crm_ipc_name(ipc), errno); return rc; } *source = mainloop_add_fd(crm_ipc_name(ipc), priority, crm_ipc_get_fd(ipc), userdata, NULL); if (*source == NULL) { int rc = errno; crm_ipc_close(ipc); return rc; } (*source)->ipc = ipc; (*source)->destroy_fn = callbacks->destroy; (*source)->dispatch_fn_ipc = callbacks->dispatch; return pcmk_rc_ok; } +/*! + * \brief Get period for mainloop timer + * + * \param[in] timer Timer + * + * \return Period in ms + */ +guint +pcmk__mainloop_timer_get_period(mainloop_timer_t *timer) +{ + if (timer) { + return timer->period_ms; + } + return 0; +} + mainloop_io_t * mainloop_add_ipc_client(const char *name, int priority, size_t max_size, void *userdata, struct ipc_client_callbacks *callbacks) { crm_ipc_t *ipc = crm_ipc_new(name, max_size); mainloop_io_t *source = NULL; int rc = pcmk__add_mainloop_ipc(ipc, priority, userdata, callbacks, &source); if (rc != pcmk_rc_ok) { if (crm_log_level == LOG_STDOUT) { fprintf(stderr, "Connection to %s failed: %s", name, pcmk_rc_str(rc)); } crm_ipc_destroy(ipc); if (rc > 0) { errno = rc; } else { errno = ENOTCONN; } return NULL; } return source; } void mainloop_del_ipc_client(mainloop_io_t * client) { mainloop_del_fd(client); } crm_ipc_t * mainloop_get_ipc_client(mainloop_io_t * client) { if (client) { return client->ipc; } return NULL; } mainloop_io_t * mainloop_add_fd(const char *name, int priority, int fd, void *userdata, struct mainloop_fd_callbacks * callbacks) { mainloop_io_t *client = NULL; if (fd >= 0) { client = calloc(1, sizeof(mainloop_io_t)); if (client == NULL) { return NULL; } client->name = strdup(name); client->userdata = userdata; if (callbacks) { client->destroy_fn = callbacks->destroy; client->dispatch_fn_io = callbacks->dispatch; } client->fd = fd; client->channel = g_io_channel_unix_new(fd); client->source = g_io_add_watch_full(client->channel, priority, (G_IO_IN | G_IO_HUP | G_IO_NVAL | G_IO_ERR), mainloop_gio_callback, client, mainloop_gio_destroy); /* Now that mainloop now holds a reference to channel, * thanks to g_io_add_watch_full(), drop ours from g_io_channel_unix_new(). * * This means that channel will be free'd by: * g_main_context_dispatch() or g_source_remove() * -> g_source_destroy_internal() * -> g_source_callback_unref() * shortly after mainloop_gio_destroy() completes */ g_io_channel_unref(client->channel); crm_trace("Added connection %d for %s[%p].%d", client->source, client->name, client, fd); } else { errno = EINVAL; } return client; } void mainloop_del_fd(mainloop_io_t * client) { if (client != NULL) { crm_trace("Removing client %s[%p]", client->name, client); if (client->source) { /* Results in mainloop_gio_destroy() being called just * before the source is removed from mainloop */ g_source_remove(client->source); } } } static GListPtr child_list = NULL; pid_t mainloop_child_pid(mainloop_child_t * child) { return child->pid; } const char * mainloop_child_name(mainloop_child_t * child) { return child->desc; } int mainloop_child_timeout(mainloop_child_t * child) { return child->timeout; } void * mainloop_child_userdata(mainloop_child_t * child) { return child->privatedata; } void mainloop_clear_child_userdata(mainloop_child_t * child) { child->privatedata = NULL; } /* good function name */ static void child_free(mainloop_child_t *child) { if (child->timerid != 0) { crm_trace("Removing timer %d", child->timerid); g_source_remove(child->timerid); child->timerid = 0; } free(child->desc); free(child); } /* terrible function name */ static int child_kill_helper(mainloop_child_t *child) { int rc; if (child->flags & mainloop_leave_pid_group) { crm_debug("Kill pid %d only. leave group intact.", child->pid); rc = kill(child->pid, SIGKILL); } else { crm_debug("Kill pid %d's group", child->pid); rc = kill(-child->pid, SIGKILL); } if (rc < 0) { if (errno != ESRCH) { crm_perror(LOG_ERR, "kill(%d, KILL) failed", child->pid); } return -errno; } return 0; } static gboolean child_timeout_callback(gpointer p) { mainloop_child_t *child = p; int rc = 0; child->timerid = 0; if (child->timeout) { crm_crit("%s process (PID %d) will not die!", child->desc, (int)child->pid); return FALSE; } rc = child_kill_helper(child); if (rc == -ESRCH) { /* Nothing left to do. pid doesn't exist */ return FALSE; } child->timeout = TRUE; crm_warn("%s process (PID %d) timed out", child->desc, (int)child->pid); child->timerid = g_timeout_add(5000, child_timeout_callback, child); return FALSE; } static bool child_waitpid(mainloop_child_t *child, int flags) { int rc = 0; int core = 0; int signo = 0; int status = 0; int exitcode = 0; bool callback_needed = true; rc = waitpid(child->pid, &status, flags); if (rc == 0) { // WNOHANG in flags, and child status is not available crm_trace("Child process %d (%s) still active", child->pid, child->desc); callback_needed = false; } else if (rc != child->pid) { /* According to POSIX, possible conditions: * - child->pid was non-positive (process group or any child), * and rc is specific child * - errno ECHILD (pid does not exist or is not child) * - errno EINVAL (invalid flags) * - errno EINTR (caller interrupted by signal) * * @TODO Handle these cases more specifically. */ signo = SIGCHLD; exitcode = 1; crm_notice("Wait for child process %d (%s) interrupted: %s", child->pid, child->desc, pcmk_strerror(errno)); } else if (WIFEXITED(status)) { exitcode = WEXITSTATUS(status); crm_trace("Child process %d (%s) exited with status %d", child->pid, child->desc, exitcode); } else if (WIFSIGNALED(status)) { signo = WTERMSIG(status); crm_trace("Child process %d (%s) exited with signal %d (%s)", child->pid, child->desc, signo, strsignal(signo)); #ifdef WCOREDUMP // AIX, SunOS, maybe others } else if (WCOREDUMP(status)) { core = 1; crm_err("Child process %d (%s) dumped core", child->pid, child->desc); #endif } else { // flags must contain WUNTRACED and/or WCONTINUED to reach this crm_trace("Child process %d (%s) stopped or continued", child->pid, child->desc); callback_needed = false; } if (callback_needed && child->callback) { child->callback(child, child->pid, core, signo, exitcode); } return callback_needed; } static void child_death_dispatch(int signal) { for (GList *iter = child_list; iter; ) { GList *saved = iter; mainloop_child_t *child = iter->data; iter = iter->next; if (child_waitpid(child, WNOHANG)) { crm_trace("Removing completed process %d from child list", child->pid); child_list = g_list_remove_link(child_list, saved); g_list_free(saved); child_free(child); } } } static gboolean child_signal_init(gpointer p) { crm_trace("Installed SIGCHLD handler"); /* Do NOT use g_child_watch_add() and friends, they rely on pthreads */ mainloop_add_signal(SIGCHLD, child_death_dispatch); /* In case they terminated before the signal handler was installed */ child_death_dispatch(SIGCHLD); return FALSE; } gboolean mainloop_child_kill(pid_t pid) { GListPtr iter; mainloop_child_t *child = NULL; mainloop_child_t *match = NULL; /* It is impossible to block SIGKILL, this allows us to * call waitpid without WNOHANG flag.*/ int waitflags = 0, rc = 0; for (iter = child_list; iter != NULL && match == NULL; iter = iter->next) { child = iter->data; if (pid == child->pid) { match = child; } } if (match == NULL) { return FALSE; } rc = child_kill_helper(match); if(rc == -ESRCH) { /* It's gone, but hasn't shown up in waitpid() yet. Wait until we get * SIGCHLD and let handler clean it up as normal (so we get the correct * return code/status). The blocking alternative would be to call * child_waitpid(match, 0). */ crm_trace("Waiting for signal that child process %d completed", match->pid); return TRUE; } else if(rc != 0) { /* If KILL for some other reason set the WNOHANG flag since we * can't be certain what happened. */ waitflags = WNOHANG; } if (!child_waitpid(match, waitflags)) { /* not much we can do if this occurs */ return FALSE; } child_list = g_list_remove(child_list, match); child_free(match); return TRUE; } /* Create/Log a new tracked process * To track a process group, use -pid * * @TODO Using a non-positive pid (i.e. any child, or process group) would * likely not be useful since we will free the child after the first * completed process. */ void mainloop_child_add_with_flags(pid_t pid, int timeout, const char *desc, void *privatedata, enum mainloop_child_flags flags, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { static bool need_init = TRUE; mainloop_child_t *child = g_new(mainloop_child_t, 1); child->pid = pid; child->timerid = 0; child->timeout = FALSE; child->privatedata = privatedata; child->callback = callback; child->flags = flags; if(desc) { child->desc = strdup(desc); } if (timeout) { child->timerid = g_timeout_add(timeout, child_timeout_callback, child); } child_list = g_list_append(child_list, child); if(need_init) { need_init = FALSE; /* SIGCHLD processing has to be invoked from mainloop. * We do not want it to be possible to both add a child pid * to mainloop, and have the pid's exit callback invoked within * the same callstack. */ g_timeout_add(1, child_signal_init, NULL); } } void mainloop_child_add(pid_t pid, int timeout, const char *desc, void *privatedata, void (*callback) (mainloop_child_t * p, pid_t pid, int core, int signo, int exitcode)) { mainloop_child_add_with_flags(pid, timeout, desc, privatedata, 0, callback); } -struct mainloop_timer_s { - guint id; - guint period_ms; - bool repeat; - char *name; - GSourceFunc cb; - void *userdata; -}; - static gboolean mainloop_timer_cb(gpointer user_data) { int id = 0; bool repeat = FALSE; struct mainloop_timer_s *t = user_data; CRM_ASSERT(t != NULL); id = t->id; t->id = 0; /* Ensure it's unset during callbacks so that * mainloop_timer_running() works as expected */ if(t->cb) { crm_trace("Invoking callbacks for timer %s", t->name); repeat = t->repeat; if(t->cb(t->userdata) == FALSE) { crm_trace("Timer %s complete", t->name); repeat = FALSE; } } if(repeat) { /* Restore if repeating */ t->id = id; } return repeat; } bool mainloop_timer_running(mainloop_timer_t *t) { if(t && t->id != 0) { return TRUE; } return FALSE; } void mainloop_timer_start(mainloop_timer_t *t) { mainloop_timer_stop(t); if(t && t->period_ms > 0) { crm_trace("Starting timer %s", t->name); t->id = g_timeout_add(t->period_ms, mainloop_timer_cb, t); } } void mainloop_timer_stop(mainloop_timer_t *t) { if(t && t->id != 0) { crm_trace("Stopping timer %s", t->name); g_source_remove(t->id); t->id = 0; } } guint mainloop_timer_set_period(mainloop_timer_t *t, guint period_ms) { guint last = 0; if(t) { last = t->period_ms; t->period_ms = period_ms; } if(t && t->id != 0 && last != t->period_ms) { mainloop_timer_start(t); } return last; } mainloop_timer_t * mainloop_timer_add(const char *name, guint period_ms, bool repeat, GSourceFunc cb, void *userdata) { mainloop_timer_t *t = calloc(1, sizeof(mainloop_timer_t)); if(t) { if(name) { t->name = crm_strdup_printf("%s-%u-%d", name, period_ms, repeat); } else { t->name = crm_strdup_printf("%p-%u-%d", t, period_ms, repeat); } t->id = 0; t->period_ms = period_ms; t->repeat = repeat; t->cb = cb; t->userdata = userdata; crm_trace("Created timer %s with %p %p", t->name, userdata, t->userdata); } return t; } void mainloop_timer_del(mainloop_timer_t *t) { if(t) { crm_trace("Destroying timer %s", t->name); mainloop_timer_stop(t); free(t->name); free(t); } } /* * Helpers to make sure certain events aren't lost at shutdown */ static gboolean drain_timeout_cb(gpointer user_data) { bool *timeout_popped = (bool*) user_data; *timeout_popped = TRUE; return FALSE; } /*! * \brief Drain some remaining main loop events then quit it * * \param[in] mloop Main loop to drain and quit * \param[in] n Drain up to this many pending events */ void pcmk_quit_main_loop(GMainLoop *mloop, unsigned int n) { if ((mloop != NULL) && g_main_loop_is_running(mloop)) { GMainContext *ctx = g_main_loop_get_context(mloop); /* Drain up to n events in case some memory clean-up is pending * (helpful to reduce noise in valgrind output). */ for (int i = 0; (i < n) && g_main_context_pending(ctx); ++i) { g_main_context_dispatch(ctx); } g_main_loop_quit(mloop); } } /*! * \brief Process main loop events while a certain condition is met * * \param[in] mloop Main loop to process * \param[in] timer_ms Don't process longer than this amount of time * \param[in] check Function that returns TRUE if events should be processed * * \note This function is intended to be called at shutdown if certain important * events should not be missed. The caller would likely quit the main loop * or exit after calling this function. The check() function will be * passed the remaining timeout in milliseconds. */ void pcmk_drain_main_loop(GMainLoop *mloop, guint timer_ms, bool (*check)(guint)) { bool timeout_popped = FALSE; guint timer = 0; GMainContext *ctx = NULL; CRM_CHECK(mloop && check, return); ctx = g_main_loop_get_context(mloop); if (ctx) { time_t start_time = time(NULL); timer = g_timeout_add(timer_ms, drain_timeout_cb, &timeout_popped); while (!timeout_popped && check(timer_ms - (time(NULL) - start_time) * 1000)) { g_main_context_iteration(ctx, TRUE); } } if (!timeout_popped && (timer > 0)) { g_source_remove(timer); } } // Deprecated functions kept only for backward API compatibility gboolean crm_signal(int sig, void (*dispatch) (int sig)); /* * \brief Use crm_signal_handler() instead * \deprecated */ gboolean crm_signal(int sig, void (*dispatch) (int sig)) { return crm_signal_handler(sig, dispatch) != SIG_ERR; }