diff --git a/daemons/fenced/cts-fence-helper.c b/daemons/fenced/cts-fence-helper.c index a2488295ca..af006b5b41 100644 --- a/daemons/fenced/cts-fence-helper.c +++ b/daemons/fenced/cts-fence-helper.c @@ -1,673 +1,682 @@ /* * 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include +#include #include #include static GMainLoop *mainloop = NULL; static crm_trigger_t *trig = NULL; static int mainloop_iter = 0; static int callback_rc = 0; typedef void (*mainloop_test_iteration_cb) (int check_event); #define MAINLOOP_DEFAULT_TIMEOUT 2 #define mainloop_test_done(pass) \ if (pass) { \ crm_info("SUCCESS - %s", __func__); \ mainloop_iter++; \ mainloop_set_trigger(trig); \ } else { \ crm_err("FAILURE = %s async_callback %d", __func__, callback_rc); \ crm_exit(CRM_EX_ERROR); \ } \ callback_rc = 0; \ enum test_modes { test_standard = 0, // test using a specific developer environment test_passive, // watch notifications only test_api_sanity, // sanity-test stonith client API using fence_dummy test_api_mainloop, // sanity-test mainloop code with async responses }; static pcmk__cli_option_t long_options[] = { // long option, argument type, storage, short option, description, flags { "verbose", no_argument, NULL, 'V', NULL, pcmk__option_default }, { "version", no_argument, NULL, '$', NULL, pcmk__option_default }, { "help", no_argument, NULL, '?', NULL, pcmk__option_default }, { "passive", no_argument, NULL, 'p', NULL, pcmk__option_default }, { "api_test", no_argument, NULL, 't', NULL, pcmk__option_default }, { "mainloop_api_test", no_argument, NULL, 'm', NULL, pcmk__option_default }, { 0, 0, 0, 0 } }; static stonith_t *st = NULL; static struct pollfd pollfd; static const int st_opts = st_opt_sync_call; static int expected_notifications = 0; static int verbose = 0; static void dispatch_helper(int timeout) { int rc; crm_debug("Looking for notification"); pollfd.events = POLLIN; while (true) { rc = poll(&pollfd, 1, timeout); /* wait 10 minutes, -1 forever */ if (rc > 0) { if (!stonith_dispatch(st)) { break; } } else { break; } } } static void st_callback(stonith_t * st, stonith_event_t * e) { if (st->state == stonith_disconnected) { crm_exit(CRM_EX_DISCONNECT); } crm_notice("Operation %s requested by %s %s for peer %s. %s reported: %s (ref=%s)", e->operation, e->origin, e->result == pcmk_ok ? "completed" : "failed", e->target, e->executioner ? e->executioner : "", pcmk_strerror(e->result), e->id); if (expected_notifications) { expected_notifications--; } } static void st_global_callback(stonith_t * stonith, stonith_callback_data_t * data) { crm_notice("Call id %d completed with rc %d", data->call_id, data->rc); } static void passive_test(void) { int rc = 0; rc = st->cmds->connect(st, crm_system_name, &pollfd.fd); if (rc != pcmk_ok) { stonith_api_delete(st); crm_exit(CRM_EX_DISCONNECT); } st->cmds->register_notification(st, T_STONITH_NOTIFY_DISCONNECT, st_callback); st->cmds->register_notification(st, T_STONITH_NOTIFY_FENCE, st_callback); st->cmds->register_notification(st, STONITH_OP_DEVICE_ADD, st_callback); st->cmds->register_notification(st, STONITH_OP_DEVICE_DEL, st_callback); st->cmds->register_callback(st, 0, 120, st_opt_timeout_updates, NULL, "st_global_callback", st_global_callback); dispatch_helper(600 * 1000); } #define single_test(cmd, str, num_notifications, expected_rc) \ { \ int rc = 0; \ rc = cmd; \ expected_notifications = 0; \ if (num_notifications) { \ expected_notifications = num_notifications; \ dispatch_helper(500); \ } \ if (rc != expected_rc) { \ crm_err("FAILURE - expected rc %d != %d(%s) for cmd - %s", expected_rc, rc, pcmk_strerror(rc), str); \ crm_exit(CRM_EX_ERROR); \ } else if (expected_notifications) { \ crm_err("FAILURE - expected %d notifications, got only %d for cmd - %s", \ num_notifications, num_notifications - expected_notifications, str); \ crm_exit(CRM_EX_ERROR); \ } else { \ if (verbose) { \ crm_info("SUCCESS - %s: %d", str, rc); \ } else { \ crm_debug("SUCCESS - %s: %d", str, rc); \ } \ } \ }\ static void run_fence_failure_test(void) { stonith_key_value_t *params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "false_1_node1=1,2 false_1_node2=3,4"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "false_1_node1=1,2 false_1_node2=3,4"); params = stonith_key_value_add(params, "mode", "fail"); single_test(st-> cmds->register_device(st, st_opts, "test-id1", "stonith-ng", "fence_dummy", params), "Register device1 for failure test", 1, 0); single_test(st->cmds->fence(st, st_opts, "false_1_node2", "off", 3, 0), "Fence failure results off", 1, -pcmk_err_generic); single_test(st->cmds->fence(st, st_opts, "false_1_node2", "reboot", 3, 0), "Fence failure results reboot", 1, -pcmk_err_generic); single_test(st->cmds->remove_device(st, st_opts, "test-id1"), "Remove device1 for failure test", 1, 0); stonith_key_value_freeall(params, 1, 1); } static void run_fence_failure_rollover_test(void) { stonith_key_value_t *params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "false_1_node1=1,2 false_1_node2=3,4"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "false_1_node1=1,2 false_1_node2=3,4"); params = stonith_key_value_add(params, "mode", "fail"); single_test(st-> cmds->register_device(st, st_opts, "test-id1", "stonith-ng", "fence_dummy", params), "Register device1 for rollover test", 1, 0); stonith_key_value_freeall(params, 1, 1); params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "false_1_node1=1,2 false_1_node2=3,4"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "false_1_node1=1,2 false_1_node2=3,4"); params = stonith_key_value_add(params, "mode", "pass"); single_test(st-> cmds->register_device(st, st_opts, "test-id2", "stonith-ng", "fence_dummy", params), "Register device2 for rollover test", 1, 0); single_test(st->cmds->fence(st, st_opts, "false_1_node2", "off", 3, 0), "Fence rollover results off", 1, 0); /* Expect -ENODEV because fence_dummy requires 'on' to be executed on target */ single_test(st->cmds->fence(st, st_opts, "false_1_node2", "on", 3, 0), "Fence rollover results on", 1, -ENODEV); single_test(st->cmds->remove_device(st, st_opts, "test-id1"), "Remove device1 for rollover tests", 1, 0); single_test(st->cmds->remove_device(st, st_opts, "test-id2"), "Remove device2 for rollover tests", 1, 0); stonith_key_value_freeall(params, 1, 1); } static void run_standard_test(void) { stonith_key_value_t *params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "false_1_node1=1,2 false_1_node2=3,4"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "false_1_node1=1,2 false_1_node2=3,4"); params = stonith_key_value_add(params, "mode", "pass"); params = stonith_key_value_add(params, "mock_dynamic_hosts", "false_1_node1 false_1_node2"); single_test(st-> cmds->register_device(st, st_opts, "test-id", "stonith-ng", "fence_dummy", params), "Register", 1, 0); stonith_key_value_freeall(params, 1, 1); params = NULL; single_test(st->cmds->list(st, st_opts, "test-id", NULL, 1), "list", 1, 0); single_test(st->cmds->monitor(st, st_opts, "test-id", 1), "Monitor", 1, 0); single_test(st->cmds->status(st, st_opts, "test-id", "false_1_node2", 1), "Status false_1_node2", 1, 0); single_test(st->cmds->status(st, st_opts, "test-id", "false_1_node1", 1), "Status false_1_node1", 1, 0); single_test(st->cmds->fence(st, st_opts, "unknown-host", "off", 1, 0), "Fence unknown-host (expected failure)", 0, -ENODEV); single_test(st->cmds->fence(st, st_opts, "false_1_node1", "off", 1, 0), "Fence false_1_node1", 1, 0); /* Expect -ENODEV because fence_dummy requires 'on' to be executed on target */ single_test(st->cmds->fence(st, st_opts, "false_1_node1", "on", 1, 0), "Unfence false_1_node1", 1, -ENODEV); /* Confirm that an invalid level index is rejected */ single_test(st->cmds->register_level(st, st_opts, "node1", 999, params), "Attempt to register an invalid level index", 0, -EINVAL); single_test(st->cmds->remove_device(st, st_opts, "test-id"), "Remove test-id", 1, 0); stonith_key_value_freeall(params, 1, 1); } static void sanity_tests(void) { int rc = 0; rc = st->cmds->connect(st, crm_system_name, &pollfd.fd); if (rc != pcmk_ok) { stonith_api_delete(st); crm_exit(CRM_EX_DISCONNECT); } st->cmds->register_notification(st, T_STONITH_NOTIFY_DISCONNECT, st_callback); st->cmds->register_notification(st, T_STONITH_NOTIFY_FENCE, st_callback); st->cmds->register_notification(st, STONITH_OP_DEVICE_ADD, st_callback); st->cmds->register_notification(st, STONITH_OP_DEVICE_DEL, st_callback); st->cmds->register_callback(st, 0, 120, st_opt_timeout_updates, NULL, "st_global_callback", st_global_callback); crm_info("Starting API Sanity Tests"); run_standard_test(); run_fence_failure_test(); run_fence_failure_rollover_test(); crm_info("Sanity Tests Passed"); } static void standard_dev_test(void) { int rc = 0; char *tmp = NULL; stonith_key_value_t *params = NULL; rc = st->cmds->connect(st, crm_system_name, &pollfd.fd); if (rc != pcmk_ok) { stonith_api_delete(st); crm_exit(CRM_EX_DISCONNECT); } - params = stonith_key_value_add(params, "pcmk_host_map", "some-host=pcmk-7 true_1_node1=3,4"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "some-host=pcmk-7 true_1_node1=3,4"); rc = st->cmds->register_device(st, st_opts, "test-id", "stonith-ng", "fence_xvm", params); crm_debug("Register: %d", rc); rc = st->cmds->list(st, st_opts, "test-id", &tmp, 10); crm_debug("List: %d output: %s", rc, tmp ? tmp : ""); rc = st->cmds->monitor(st, st_opts, "test-id", 10); crm_debug("Monitor: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "false_1_node2", 10); crm_debug("Status false_1_node2: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "false_1_node1", 10); crm_debug("Status false_1_node1: %d", rc); rc = st->cmds->fence(st, st_opts, "unknown-host", "off", 60, 0); crm_debug("Fence unknown-host: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "false_1_node1", 10); crm_debug("Status false_1_node1: %d", rc); rc = st->cmds->fence(st, st_opts, "false_1_node1", "off", 60, 0); crm_debug("Fence false_1_node1: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "false_1_node1", 10); crm_debug("Status false_1_node1: %d", rc); rc = st->cmds->fence(st, st_opts, "false_1_node1", "on", 10, 0); crm_debug("Unfence false_1_node1: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "false_1_node1", 10); crm_debug("Status false_1_node1: %d", rc); rc = st->cmds->fence(st, st_opts, "some-host", "off", 10, 0); crm_debug("Fence alias: %d", rc); rc = st->cmds->status(st, st_opts, "test-id", "some-host", 10); crm_debug("Status alias: %d", rc); rc = st->cmds->fence(st, st_opts, "false_1_node1", "on", 10, 0); crm_debug("Unfence false_1_node1: %d", rc); rc = st->cmds->remove_device(st, st_opts, "test-id"); crm_debug("Remove test-id: %d", rc); stonith_key_value_freeall(params, 1, 1); } static void iterate_mainloop_tests(gboolean event_ready); static void mainloop_callback(stonith_t * stonith, stonith_callback_data_t * data) { callback_rc = data->rc; iterate_mainloop_tests(TRUE); } static int register_callback_helper(int callid) { return st->cmds->register_callback(st, callid, MAINLOOP_DEFAULT_TIMEOUT, st_opt_timeout_updates, NULL, "callback", mainloop_callback); } static void test_async_fence_pass(int check_event) { int rc = 0; if (check_event) { if (callback_rc != 0) { mainloop_test_done(FALSE); } else { mainloop_test_done(TRUE); } return; } rc = st->cmds->fence(st, 0, "true_1_node1", "off", MAINLOOP_DEFAULT_TIMEOUT, 0); if (rc < 0) { crm_err("fence failed with rc %d", rc); mainloop_test_done(FALSE); } register_callback_helper(rc); /* wait for event */ } #define CUSTOM_TIMEOUT_ADDITION 10 static void test_async_fence_custom_timeout(int check_event) { int rc = 0; static time_t begin = 0; if (check_event) { uint32_t diff = (time(NULL) - begin); if (callback_rc != -ETIME) { mainloop_test_done(FALSE); } else if (diff < CUSTOM_TIMEOUT_ADDITION + MAINLOOP_DEFAULT_TIMEOUT) { crm_err ("Custom timeout test failed, callback expiration should be updated to %d, actual timeout was %d", CUSTOM_TIMEOUT_ADDITION + MAINLOOP_DEFAULT_TIMEOUT, diff); mainloop_test_done(FALSE); } else { mainloop_test_done(TRUE); } return; } begin = time(NULL); rc = st->cmds->fence(st, 0, "custom_timeout_node1", "off", MAINLOOP_DEFAULT_TIMEOUT, 0); if (rc < 0) { crm_err("fence failed with rc %d", rc); mainloop_test_done(FALSE); } register_callback_helper(rc); /* wait for event */ } static void test_async_fence_timeout(int check_event) { int rc = 0; if (check_event) { if (callback_rc != -ENODEV) { mainloop_test_done(FALSE); } else { mainloop_test_done(TRUE); } return; } rc = st->cmds->fence(st, 0, "false_1_node2", "off", MAINLOOP_DEFAULT_TIMEOUT, 0); if (rc < 0) { crm_err("fence failed with rc %d", rc); mainloop_test_done(FALSE); } register_callback_helper(rc); /* wait for event */ } static void test_async_monitor(int check_event) { int rc = 0; if (check_event) { if (callback_rc) { mainloop_test_done(FALSE); } else { mainloop_test_done(TRUE); } return; } rc = st->cmds->monitor(st, 0, "false_1", MAINLOOP_DEFAULT_TIMEOUT); if (rc < 0) { crm_err("monitor failed with rc %d", rc); mainloop_test_done(FALSE); } register_callback_helper(rc); /* wait for event */ } static void test_register_async_devices(int check_event) { char buf[16] = { 0, }; stonith_key_value_t *params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "false_1_node1=1,2"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "false_1_node1=1,2"); params = stonith_key_value_add(params, "mode", "fail"); st->cmds->register_device(st, st_opts, "false_1", "stonith-ng", "fence_dummy", params); stonith_key_value_freeall(params, 1, 1); params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "true_1_node1=1,2"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "true_1_node1=1,2"); params = stonith_key_value_add(params, "mode", "pass"); st->cmds->register_device(st, st_opts, "true_1", "stonith-ng", "fence_dummy", params); stonith_key_value_freeall(params, 1, 1); params = NULL; - params = stonith_key_value_add(params, "pcmk_host_map", "custom_timeout_node1=1,2"); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_MAP, + "custom_timeout_node1=1,2"); params = stonith_key_value_add(params, "mode", "fail"); params = stonith_key_value_add(params, "delay", "1000"); snprintf(buf, sizeof(buf) - 1, "%d", MAINLOOP_DEFAULT_TIMEOUT + CUSTOM_TIMEOUT_ADDITION); params = stonith_key_value_add(params, "pcmk_off_timeout", buf); st->cmds->register_device(st, st_opts, "false_custom_timeout", "stonith-ng", "fence_dummy", params); stonith_key_value_freeall(params, 1, 1); mainloop_test_done(TRUE); } static void try_mainloop_connect(int check_event) { int rc = stonith_api_connect_retry(st, crm_system_name, 10); if (rc == pcmk_ok) { mainloop_test_done(TRUE); return; } crm_err("API CONNECTION FAILURE"); mainloop_test_done(FALSE); } static void iterate_mainloop_tests(gboolean event_ready) { static mainloop_test_iteration_cb callbacks[] = { try_mainloop_connect, test_register_async_devices, test_async_monitor, test_async_fence_pass, test_async_fence_timeout, test_async_fence_custom_timeout, }; if (mainloop_iter == (sizeof(callbacks) / sizeof(mainloop_test_iteration_cb))) { /* all tests ran, everything passed */ crm_info("ALL MAINLOOP TESTS PASSED!"); crm_exit(CRM_EX_OK); } callbacks[mainloop_iter] (event_ready); } static gboolean trigger_iterate_mainloop_tests(gpointer user_data) { iterate_mainloop_tests(FALSE); return TRUE; } static void test_shutdown(int nsig) { int rc = 0; if (st) { rc = st->cmds->disconnect(st); crm_info("Disconnect: %d", rc); crm_debug("Destroy"); stonith_api_delete(st); } if (rc) { crm_exit(CRM_EX_ERROR); } } static void mainloop_tests(void) { trig = mainloop_add_trigger(G_PRIORITY_HIGH, trigger_iterate_mainloop_tests, NULL); mainloop_set_trigger(trig); mainloop_add_signal(SIGTERM, test_shutdown); crm_info("Starting"); mainloop = g_main_loop_new(NULL, FALSE); g_main_loop_run(mainloop); } int main(int argc, char **argv) { int argerr = 0; int flag; int option_index = 0; enum test_modes mode = test_standard; crm_log_cli_init("cts-fence-helper"); pcmk__set_cli_options(NULL, " [options]", long_options, "inject commands into the Pacemaker fencer, " "and watch for events"); while (1) { flag = pcmk__next_cli_option(argc, argv, &option_index, NULL); if (flag == -1) { break; } switch (flag) { case 'V': verbose = 1; break; case '$': case '?': pcmk__cli_help(flag, CRM_EX_OK); break; case 'p': mode = test_passive; break; case 't': mode = test_api_sanity; break; case 'm': mode = test_api_mainloop; break; default: ++argerr; break; } } crm_log_init(NULL, LOG_INFO, TRUE, (verbose? TRUE : FALSE), argc, argv, FALSE); if (optind > argc) { ++argerr; } if (argerr) { pcmk__cli_help('?', CRM_EX_USAGE); } st = stonith_api_new(); if (st == NULL) { crm_err("Could not connect to fencer: API memory allocation failed"); crm_exit(CRM_EX_DISCONNECT); } switch (mode) { case test_standard: standard_dev_test(); break; case test_passive: passive_test(); break; case test_api_sanity: sanity_tests(); break; case test_api_mainloop: mainloop_tests(); break; } test_shutdown(0); return CRM_EX_OK; } diff --git a/daemons/fenced/fenced_commands.c b/daemons/fenced/fenced_commands.c index d9db985b1a..45fb4690fd 100644 --- a/daemons/fenced/fenced_commands.c +++ b/daemons/fenced/fenced_commands.c @@ -1,2863 +1,2864 @@ /* * Copyright 2009-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #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); 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, STONITH_ATTR_DELAY_MAX); + 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, STONITH_ATTR_DELAY_BASE); + 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, STONITH_ATTR_ACTION_LIMIT); + 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'%s%s on %s now running with pid=%d, timeout=%ds", cmd->action, cmd->victim ? " targeting " : "", cmd->victim ? cmd->victim : "", device->id, pid, cmd->timeout); cmd->active_on = device; cmd->activating_on = NULL; } 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, active_cmds > 1 ? "s" : ""); 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 on %s was asked to run too early, waiting for start_delay timeout of %ds", pending_op->action, pending_op->victim ? " targeting " : "", pending_op->victim ? 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("Nothing further to do for %s for now", 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 the stop operation for %s", device->id); } else { crm_err("failed to get secrets for %s, " "considering resource not configured", 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 on %s failed: %s (%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 on %s for remote peer %s with op id (%s) (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 on %s for %s (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("Base-delay (%ds) is larger than max-delay (%ds) " "for %s on %s - limiting to max-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 on %s for %ds (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); 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, STONITH_ATTR_HOSTLIST); + 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, STONITH_ATTR_HOSTMAP); + 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); } 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("The fencing device '%s' requires unfencing", device->id); } if (device->on_target_actions) { crm_info("The fencing device '%s' requires actions (%s) to be executed on the target node", 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, STONITH_ATTR_HOSTCHECK); + check_type = g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_CHECK); if (check_type == NULL) { - if (g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTLIST)) { + if (g_hash_table_lookup(dev->params, PCMK_STONITH_HOST_LIST)) { check_type = "static-list"; - } else if (g_hash_table_lookup(dev->params, STONITH_ATTR_HOSTMAP)) { + } 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 (%d): %s", dev->id, rc, output); /* Fall back to status */ - g_hash_table_replace(dev->params, strdup(STONITH_ATTR_HOSTCHECK), strdup("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); CRM_CHECK(device != NULL, return -ENOMEM); dup = device_has_duplicate(device); if (dup) { crm_debug("Device '%s' already existed in device list (%d active devices)", device->id, g_hash_table_size(device_list)); free_device(device); device = dup; } 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 an existing entry for %s from the 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); crm_notice("Added '%s' to the device list (%d active devices)", device->id, g_hash_table_size(device_list)); } 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); if (!device) { crm_info("Device '%s' not found (%d active devices)", id, g_hash_table_size(device_list)); 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); crm_info("Removed '%s' from the device list (%d active devices)", id, g_hash_table_size(device_list)); } else { crm_trace("Not removing '%s' from the device list (%d active devices) " "- still %s%s_registered", 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); crm_info("Target %s has %d active fencing levels", tp->target, count_active_levels(tp)); 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) { crm_info("Topology for %s not found (%d active entries)", target, g_hash_table_size(topology)); } else if (id == 0 && g_hash_table_remove(topology, target)) { crm_info("Removed all %s related entries from the topology (%d active entries)", target, g_hash_table_size(topology)); } else if (id > 0 && tp->levels[id] != NULL) { g_list_free_full(tp->levels[id], free); tp->levels[id] = NULL; crm_info("Removed level '%d' from topology for %s (%d active levels remaining)", id, target, count_active_levels(tp)); } 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) { crm_debug("Finished Search. %d devices can perform action (%s) on node %s", g_list_length(search->capable), search->action ? search->action : "", search->host ? search->host : ""); 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("'%s' operation with %s can only be executed for localhost 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, STONITH_ATTR_HOSTMAP) + } 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 " STONITH_ATTR_HOSTCHECK ": %s", check_type); - check_type = "Invalid " STONITH_ATTR_HOSTCHECK; + 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; if (!g_hash_table_size(device_list)) { 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("STONITH 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("STONITH 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 = g_hash_table_size(device_list); search->allow_suicide = suicide; search->callback = callback; search->user_data = user_data; /* kick off the search */ crm_debug("Searching through %d devices to see what is capable of action (%s) for target %s", search->replies_needed, search->action ? search->action : "", search->host ? search->host : ""); 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 on %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 on %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 on %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 on %s", action, delay_max, device->id); } else if ((delay_max == 0) && (delay_base > 0)) { crm_trace("Action '%s' has a static delay of %dms on %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 on %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' on %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 devices for '%s'", available_devices, query->target); } else { crm_debug("%d devices installed", 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] (call %d from %s) for host '%s' with device '%s' returned%s: %d (%s)%s%s", cmd->action, pid, cmd->id, cmd->client_name, cmd->victim, cmd->device, (op_merged? " (merged)" : ""), rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } else { do_crm_log_unlikely(rc == 0 ? LOG_DEBUG : LOG_NOTICE, "Operation '%s' [%d] for device '%s' returned%s: %d (%s)%s%s", cmd->action, pid, cmd->device, (op_merged? " (merged)" : ""), rc, pcmk_strerror(rc), (next? ", retrying with " : ""), (next ? next : "")); } if (output) { /* Logging the whole string confuses syslog when the string is xml */ 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 on %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' on '%s' completed with rc=%d (%d 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 stonith action '%s' targeting %s originating from client %s with identical stonith 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; crm_info("Found %d matching devices for '%s'", g_list_length(devices), 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 = crm_find_known_peer_full(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 complex self fencing request"); 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_info("Delete the relay op : %s - %s of %s for %s.(replaced by op : %s - %s of %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); bool allowed = false; #if ENABLE_ACL /* IPC commands related to fencing configuration may be done only by * privileged users (i.e. root or hacluster) when ACLs are supported, * because all other users should go through the CIB to have ACLs applied. */ if (client != NULL) { allowed = pcmk_is_set(client->flags, pcmk__client_privileged); } #else allowed = true; #endif 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 %s (%s)", flag_name, client->name, client->id); 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 %s (%s)", flag_name, client->name, client->id); 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("Peer %s has received a forwarded fencing request from %s to fence (%s) peer %s", stonith_our_uname, client ? client->name : remote_peer, 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.%.8s wants to fence (%s) '%s' with device '%s'", client->name, client->id, action, target, device ? device : "(any)"); 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 complex self fencing request to peer %s", 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", op, (client? client->name : remote_peer)); } 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", op, client ? client->name : remote_peer); 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 (%16x)", op, is_reply ? " reply" : "", id, client ? client->name : remote_peer, 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 (%d)", op, is_reply ? " reply" : "", client ? client->name : remote_peer, rc > 0 ? "" : pcmk_strerror(rc), rc); free(op); } diff --git a/daemons/fenced/pacemaker-fenced.c b/daemons/fenced/pacemaker-fenced.c index 092f604bf9..5c2cc3a099 100644 --- a/daemons/fenced/pacemaker-fenced.c +++ b/daemons/fenced/pacemaker-fenced.c @@ -1,1534 +1,1542 @@ /* * Copyright 2009-2020 the Pacemaker project contributors * * The version control history for this file may have further details. * * This source code is licensed under the GNU General Public License version 2 * or later (GPLv2+) WITHOUT ANY WARRANTY. */ #include #include #include #include #include #include #include #include #include #include #include /* U32T ~ PRIu32, X32T ~ PRIx32 */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include char *stonith_our_uname = NULL; char *stonith_our_uuid = NULL; long stonith_watchdog_timeout_ms = 0; static GMainLoop *mainloop = NULL; gboolean stand_alone = FALSE; static gboolean no_cib_connect = FALSE; static gboolean stonith_shutdown_flag = FALSE; static qb_ipcs_service_t *ipcs = NULL; static xmlNode *local_cib = NULL; static pe_working_set_t *fenced_data_set = NULL; static cib_t *cib_api = NULL; static void stonith_shutdown(int nsig); static void stonith_cleanup(void); static int32_t st_ipc_accept(qb_ipcs_connection_t * c, uid_t uid, gid_t gid) { if (stonith_shutdown_flag) { crm_info("Ignoring new client [%d] during shutdown", pcmk__client_pid(c)); return -EPERM; } if (pcmk__new_client(c, uid, gid) == NULL) { return -EIO; } return 0; } /* Exit code means? */ static int32_t st_ipc_dispatch(qb_ipcs_connection_t * qbc, void *data, size_t size) { uint32_t id = 0; uint32_t flags = 0; int call_options = 0; xmlNode *request = NULL; pcmk__client_t *c = pcmk__find_client(qbc); const char *op = NULL; if (c == NULL) { crm_info("Invalid client: %p", qbc); return 0; } request = pcmk__client_data2xml(c, data, &id, &flags); if (request == NULL) { pcmk__ipc_send_ack(c, id, flags, "nack", CRM_EX_PROTOCOL); return 0; } op = crm_element_value(request, F_CRM_TASK); if(pcmk__str_eq(op, CRM_OP_RM_NODE_CACHE, pcmk__str_casei)) { crm_xml_add(request, F_TYPE, T_STONITH_NG); crm_xml_add(request, F_STONITH_OPERATION, op); crm_xml_add(request, F_STONITH_CLIENTID, c->id); crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname); send_cluster_message(NULL, crm_msg_stonith_ng, request, FALSE); free_xml(request); return 0; } if (c->name == NULL) { const char *value = crm_element_value(request, F_STONITH_CLIENTNAME); if (value == NULL) { value = "unknown"; } c->name = crm_strdup_printf("%s.%u", value, c->pid); } crm_element_value_int(request, F_STONITH_CALLOPTS, &call_options); crm_trace("Flags %" X32T "/%u for command %" U32T " from %s", flags, call_options, id, pcmk__client_name(c)); if (pcmk_is_set(call_options, st_opt_sync_call)) { CRM_ASSERT(flags & crm_ipc_client_response); CRM_LOG_ASSERT(c->request_id == 0); /* This means the client has two synchronous events in-flight */ c->request_id = id; /* Reply only to the last one */ } crm_xml_add(request, F_STONITH_CLIENTID, c->id); crm_xml_add(request, F_STONITH_CLIENTNAME, pcmk__client_name(c)); crm_xml_add(request, F_STONITH_CLIENTNODE, stonith_our_uname); stonith_command(c, id, flags, request, NULL); free_xml(request); return 0; } /* Error code means? */ static int32_t st_ipc_closed(qb_ipcs_connection_t * c) { pcmk__client_t *client = pcmk__find_client(c); if (client == NULL) { return 0; } crm_trace("Connection %p closed", c); pcmk__free_client(client); /* 0 means: yes, go ahead and destroy the connection */ return 0; } static void st_ipc_destroy(qb_ipcs_connection_t * c) { crm_trace("Connection %p destroyed", c); st_ipc_closed(c); } static void stonith_peer_callback(xmlNode * msg, void *private_data) { const char *remote_peer = crm_element_value(msg, F_ORIG); const char *op = crm_element_value(msg, F_STONITH_OPERATION); if (pcmk__str_eq(op, "poke", pcmk__str_none)) { return; } crm_log_xml_trace(msg, "Peer[inbound]"); stonith_command(NULL, 0, 0, msg, remote_peer); } #if SUPPORT_COROSYNC static void stonith_peer_ais_callback(cpg_handle_t handle, const struct cpg_name *groupName, uint32_t nodeid, uint32_t pid, void *msg, size_t msg_len) { uint32_t kind = 0; xmlNode *xml = NULL; const char *from = NULL; char *data = pcmk_message_common_cs(handle, nodeid, pid, msg, &kind, &from); if(data == NULL) { return; } if (kind == crm_class_cluster) { xml = string2xml(data); if (xml == NULL) { crm_err("Invalid XML: '%.120s'", data); free(data); return; } crm_xml_add(xml, F_ORIG, from); /* crm_xml_add_int(xml, F_SEQ, wrapper->id); */ stonith_peer_callback(xml, NULL); } free_xml(xml); free(data); return; } static void stonith_peer_cs_destroy(gpointer user_data) { crm_crit("Lost connection to cluster layer, shutting down"); stonith_shutdown(0); } #endif void do_local_reply(xmlNode * notify_src, const char *client_id, gboolean sync_reply, gboolean from_peer) { /* send callback to originating child */ pcmk__client_t *client_obj = NULL; int local_rc = pcmk_rc_ok; crm_trace("Sending response"); client_obj = pcmk__find_client_by_id(client_id); crm_trace("Sending callback to request originator"); if (client_obj == NULL) { local_rc = EPROTO; crm_trace("No client to sent the response to. F_STONITH_CLIENTID not set."); } else { int rid = 0; if (sync_reply) { CRM_LOG_ASSERT(client_obj->request_id); rid = client_obj->request_id; client_obj->request_id = 0; crm_trace("Sending response %d to %s %s", rid, client_obj->name, from_peer ? "(originator of delegated request)" : ""); } else { crm_trace("Sending an event to %s %s", client_obj->name, from_peer ? "(originator of delegated request)" : ""); } local_rc = pcmk__ipc_send_xml(client_obj, rid, notify_src, (sync_reply? crm_ipc_flags_none : crm_ipc_server_event)); } if ((local_rc != pcmk_rc_ok) && (client_obj != NULL)) { crm_warn("%s reply to %s failed: %s", (sync_reply? "Synchronous" : "Asynchronous"), (client_obj? client_obj->name : "unknown client"), pcmk_rc_str(local_rc)); } } uint64_t get_stonith_flag(const char *name) { if (pcmk__str_eq(name, T_STONITH_NOTIFY_FENCE, pcmk__str_casei)) { return st_callback_notify_fence; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_ADD, pcmk__str_casei)) { return st_callback_device_add; } else if (pcmk__str_eq(name, STONITH_OP_DEVICE_DEL, pcmk__str_casei)) { return st_callback_device_del; } else if (pcmk__str_eq(name, T_STONITH_NOTIFY_HISTORY, pcmk__str_casei)) { return st_callback_notify_history; } else if (pcmk__str_eq(name, T_STONITH_NOTIFY_HISTORY_SYNCED, pcmk__str_casei)) { return st_callback_notify_history_synced; } return st_callback_unknown; } static void stonith_notify_client(gpointer key, gpointer value, gpointer user_data) { xmlNode *update_msg = user_data; pcmk__client_t *client = value; const char *type = NULL; CRM_CHECK(client != NULL, return); CRM_CHECK(update_msg != NULL, return); type = crm_element_value(update_msg, F_SUBTYPE); CRM_CHECK(type != NULL, crm_log_xml_err(update_msg, "notify"); return); if (client->ipcs == NULL) { crm_trace("Skipping client with NULL channel"); return; } if (pcmk_is_set(client->flags, get_stonith_flag(type))) { int rc = pcmk__ipc_send_xml(client, 0, update_msg, crm_ipc_server_event|crm_ipc_server_error); if (rc != pcmk_rc_ok) { crm_warn("%s notification of client %s failed: %s " CRM_XS " id=%.8s rc=%d", type, pcmk__client_name(client), pcmk_rc_str(rc), client->id, rc); } else { crm_trace("Sent %s notification to client %s.%.6s", type, pcmk__client_name(client), client->id); } } } void do_stonith_async_timeout_update(const char *client_id, const char *call_id, int timeout) { pcmk__client_t *client = NULL; xmlNode *notify_data = NULL; if (!timeout || !call_id || !client_id) { return; } client = pcmk__find_client_by_id(client_id); if (!client) { return; } notify_data = create_xml_node(NULL, T_STONITH_TIMEOUT_VALUE); crm_xml_add(notify_data, F_TYPE, T_STONITH_TIMEOUT_VALUE); crm_xml_add(notify_data, F_STONITH_CALLID, call_id); crm_xml_add_int(notify_data, F_STONITH_TIMEOUT, timeout); crm_trace("timeout update is %d for client %s and call id %s", timeout, client_id, call_id); if (client) { pcmk__ipc_send_xml(client, 0, notify_data, crm_ipc_server_event); } free_xml(notify_data); } void do_stonith_notify(int options, const char *type, int result, xmlNode * data) { /* TODO: Standardize the contents of data */ xmlNode *update_msg = create_xml_node(NULL, "notify"); CRM_CHECK(type != NULL,;); crm_xml_add(update_msg, F_TYPE, T_STONITH_NOTIFY); crm_xml_add(update_msg, F_SUBTYPE, type); crm_xml_add(update_msg, F_STONITH_OPERATION, type); crm_xml_add_int(update_msg, F_STONITH_RC, result); if (data != NULL) { add_message_xml(update_msg, F_STONITH_CALLDATA, data); } crm_trace("Notifying clients"); pcmk__foreach_ipc_client(stonith_notify_client, update_msg); free_xml(update_msg); crm_trace("Notify complete"); } static void do_stonith_notify_config(int options, const char *op, int rc, const char *desc, int active) { xmlNode *notify_data = create_xml_node(NULL, op); CRM_CHECK(notify_data != NULL, return); crm_xml_add(notify_data, F_STONITH_DEVICE, desc); crm_xml_add_int(notify_data, F_STONITH_ACTIVE, active); do_stonith_notify(options, op, rc, notify_data); free_xml(notify_data); } void do_stonith_notify_device(int options, const char *op, int rc, const char *desc) { do_stonith_notify_config(options, op, rc, desc, g_hash_table_size(device_list)); } void do_stonith_notify_level(int options, const char *op, int rc, const char *desc) { do_stonith_notify_config(options, op, rc, desc, g_hash_table_size(topology)); } static void topology_remove_helper(const char *node, int level) { int rc; char *desc = NULL; xmlNode *data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level); crm_xml_add(data, XML_ATTR_STONITH_TARGET, node); rc = stonith_level_remove(data, &desc); do_stonith_notify_level(0, STONITH_OP_LEVEL_DEL, rc, desc); free_xml(data); free(desc); } static void remove_cib_device(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { standard = crm_element_value(match, XML_AGENT_ATTR_CLASS); } if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } rsc_id = crm_element_value(match, XML_ATTR_ID); stonith_device_remove(rsc_id, TRUE); } } static void handle_topology_change(xmlNode *match, bool remove) { int rc; char *desc = NULL; CRM_CHECK(match != NULL, return); crm_trace("Updating %s", ID(match)); if(remove) { int index = 0; char *key = stonith_level_key(match, -1); crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index); topology_remove_helper(key, index); free(key); } rc = stonith_level_register(match, &desc); do_stonith_notify_level(0, STONITH_OP_LEVEL_ADD, rc, desc); free(desc); } static void remove_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if (match && crm_element_value(match, XML_DIFF_MARKER)) { /* Deletion */ int index = 0; char *target = stonith_level_key(match, -1); crm_element_value_int(match, XML_ATTR_STONITH_INDEX, &index); if (target == NULL) { crm_err("Invalid fencing target in element %s", ID(match)); } else if (index <= 0) { crm_err("Invalid level for %s in element %s", target, ID(match)); } else { topology_remove_helper(target, index); } /* } else { Deal with modifications during the 'addition' stage */ } } } static void register_fencing_topology(xmlXPathObjectPtr xpathObj) { int max = numXpathResults(xpathObj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); handle_topology_change(match, TRUE); } } /* Fencing */ static void fencing_topology_init(void) { xmlXPathObjectPtr xpathObj = NULL; const char *xpath = "//" XML_TAG_FENCING_LEVEL; crm_trace("Full topology refresh"); free_topology_list(); init_topology_list(); /* Grab everything */ xpathObj = xpath_search(local_cib, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } #define rsc_name(x) x->clone_name?x->clone_name:x->id /*! * \internal * \brief Check whether our uname is in a resource's allowed node list * * \param[in] rsc Resource to check * * \return Pointer to node object if found, NULL otherwise */ static pe_node_t * our_node_allowed_for(pe_resource_t *rsc) { GHashTableIter iter; pe_node_t *node = NULL; if (rsc && stonith_our_uname) { g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if (node && strcmp(node->details->uname, stonith_our_uname) == 0) { break; } node = NULL; } } return node; } /*! * \internal * \brief If a resource or any of its children are STONITH devices, update their * definitions given a cluster working set. * * \param[in] rsc Resource to check * \param[in] data_set Cluster working set with device information */ static void cib_device_update(pe_resource_t *rsc, pe_working_set_t *data_set) { pe_node_t *node = NULL; const char *value = NULL; const char *rclass = NULL; pe_node_t *parent = NULL; gboolean remove = TRUE; /* If this is a complex resource, check children rather than this resource itself. * TODO: Mark each installed device and remove if untouched when this process finishes. */ if(rsc->children) { GListPtr gIter = NULL; for (gIter = rsc->children; gIter != NULL; gIter = gIter->next) { cib_device_update(gIter->data, data_set); if(pe_rsc_is_clone(rsc)) { crm_trace("Only processing one copy of the clone %s", rsc->id); break; } } return; } /* We only care about STONITH resources. */ rclass = crm_element_value(rsc->xml, XML_AGENT_ATTR_CLASS); if (!pcmk__str_eq(rclass, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { return; } /* If this STONITH resource is disabled, just remove it. */ if (pe__resource_is_disabled(rsc)) { crm_info("Device %s has been disabled", rsc->id); goto update_done; } /* Check whether our node is allowed for this resource (and its parent if in a group) */ node = our_node_allowed_for(rsc); if (rsc->parent && (rsc->parent->variant == pe_group)) { parent = our_node_allowed_for(rsc->parent); } if(node == NULL) { /* Our node is disallowed, so remove the device */ GHashTableIter iter; crm_info("Device %s has been disabled on %s: unknown", rsc->id, stonith_our_uname); g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { crm_trace("Available: %s = %d", node->details->uname, node->weight); } goto update_done; } else if(node->weight < 0 || (parent && parent->weight < 0)) { /* Our node (or its group) is disallowed by score, so remove the device */ char *score = score2char((node->weight < 0) ? node->weight : parent->weight); crm_info("Device %s has been disabled on %s: score=%s", rsc->id, stonith_our_uname, score); free(score); goto update_done; } else { /* Our node is allowed, so update the device information */ int rc; xmlNode *data; GHashTableIter gIter; stonith_key_value_t *params = NULL; const char *name = NULL; const char *agent = crm_element_value(rsc->xml, XML_EXPR_ATTR_TYPE); const char *rsc_provides = NULL; crm_debug("Device %s is allowed on %s: score=%d", rsc->id, stonith_our_uname, node->weight); get_rsc_attributes(rsc->parameters, rsc, node, data_set); get_meta_attributes(rsc->meta, rsc, node, data_set); - rsc_provides = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_PROVIDES); + rsc_provides = g_hash_table_lookup(rsc->meta, PCMK_STONITH_PROVIDES); g_hash_table_iter_init(&gIter, rsc->parameters); while (g_hash_table_iter_next(&gIter, (gpointer *) & name, (gpointer *) & value)) { if (!name || !value) { continue; } params = stonith_key_value_add(params, name, value); crm_trace(" %s=%s", name, value); } remove = FALSE; data = create_device_registration_xml(rsc_name(rsc), st_namespace_any, agent, params, rsc_provides); stonith_key_value_freeall(params, 1, 1); rc = stonith_device_register(data, NULL, TRUE); CRM_ASSERT(rc == pcmk_ok); free_xml(data); } update_done: if(remove && g_hash_table_lookup(device_list, rsc_name(rsc))) { stonith_device_remove(rsc_name(rsc), TRUE); } } /*! * \internal * \brief Update all STONITH device definitions based on current CIB */ static void cib_devices_update(void) { GListPtr gIter = NULL; crm_info("Updating devices to version %s.%s.%s", crm_element_value(local_cib, XML_ATTR_GENERATION_ADMIN), crm_element_value(local_cib, XML_ATTR_GENERATION), crm_element_value(local_cib, XML_ATTR_NUMUPDATES)); CRM_ASSERT(fenced_data_set != NULL); fenced_data_set->input = local_cib; fenced_data_set->now = crm_time_new(NULL); fenced_data_set->localhost = stonith_our_uname; pe__set_working_set_flags(fenced_data_set, pe_flag_quick_location); cluster_status(fenced_data_set); pcmk__schedule_actions(fenced_data_set, NULL, NULL); for (gIter = fenced_data_set->resources; gIter != NULL; gIter = gIter->next) { cib_device_update(gIter->data, fenced_data_set); } fenced_data_set->input = NULL; // Wasn't a copy, so don't let API free it pe_reset_working_set(fenced_data_set); } static void update_cib_stonith_devices_v2(const char *event, xmlNode * msg) { xmlNode *change = NULL; char *reason = NULL; bool needs_update = FALSE; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, XML_DIFF_OP); const char *xpath = crm_element_value(change, XML_DIFF_PATH); const char *shortpath = NULL; if ((op == NULL) || (strcmp(op, "move") == 0) || strstr(xpath, "/"XML_CIB_TAG_STATUS)) { continue; } else if (pcmk__str_eq(op, "delete", pcmk__str_casei) && strstr(xpath, "/"XML_CIB_TAG_RESOURCE)) { const char *rsc_id = NULL; char *search = NULL; char *mutable = NULL; if (strstr(xpath, XML_TAG_ATTR_SETS) || strstr(xpath, XML_TAG_META_SETS)) { needs_update = TRUE; reason = strdup("(meta) attribute deleted from resource"); break; } mutable = strdup(xpath); rsc_id = strstr(mutable, "primitive[@id=\'"); if (rsc_id != NULL) { rsc_id += strlen("primitive[@id=\'"); search = strchr(rsc_id, '\''); } if (search != NULL) { *search = 0; stonith_device_remove(rsc_id, TRUE); } else { crm_warn("Ignoring malformed CIB update (resource deletion)"); } free(mutable); } else if (strstr(xpath, "/"XML_CIB_TAG_RESOURCES) || strstr(xpath, "/"XML_CIB_TAG_CONSTRAINTS) || strstr(xpath, "/"XML_CIB_TAG_RSCCONFIG)) { shortpath = strrchr(xpath, '/'); CRM_ASSERT(shortpath); reason = crm_strdup_printf("%s %s", op, shortpath+1); needs_update = TRUE; break; } } if(needs_update) { crm_info("Updating device list from the cib: %s", reason); cib_devices_update(); } else { crm_trace("No updates for device list found in cib"); } free(reason); } static void update_cib_stonith_devices_v1(const char *event, xmlNode * msg) { const char *reason = "none"; gboolean needs_update = FALSE; xmlXPathObjectPtr xpath_obj = NULL; /* process new constraints */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_CONS_TAG_RSC_LOCATION); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; /* Safest and simplest to always recompute */ needs_update = TRUE; reason = "new location constraint"; for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpath_obj, lpc); crm_log_xml_trace(match, "new constraint"); } } freeXpathObject(xpath_obj); /* process deletions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_CIB_TAG_RESOURCE); if (numXpathResults(xpath_obj) > 0) { remove_cib_device(xpath_obj); } freeXpathObject(xpath_obj); /* process additions */ xpath_obj = xpath_search(msg, "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_CIB_TAG_RESOURCE); if (numXpathResults(xpath_obj) > 0) { int max = numXpathResults(xpath_obj), lpc = 0; for (lpc = 0; lpc < max; lpc++) { const char *rsc_id = NULL; const char *standard = NULL; xmlNode *match = getXpathResult(xpath_obj, lpc); rsc_id = crm_element_value(match, XML_ATTR_ID); standard = crm_element_value(match, XML_AGENT_ATTR_CLASS); if (!pcmk__str_eq(standard, PCMK_RESOURCE_CLASS_STONITH, pcmk__str_casei)) { continue; } crm_trace("Fencing resource %s was added or modified", rsc_id); reason = "new resource"; needs_update = TRUE; } } freeXpathObject(xpath_obj); if(needs_update) { crm_info("Updating device list from the cib: %s", reason); cib_devices_update(); } } static void update_cib_stonith_devices(const char *event, xmlNode * msg) { int format = 1; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, "format", &format); switch(format) { case 1: update_cib_stonith_devices_v1(event, msg); break; case 2: update_cib_stonith_devices_v2(event, msg); break; default: crm_warn("Unknown patch format: %d", format); } } /* Needs to hold node name + attribute name + attribute value + 75 */ #define XPATH_MAX 512 /*! * \internal * \brief Check whether a node has a specific attribute name/value * * \param[in] node Name of node to check * \param[in] name Name of an attribute to look for * \param[in] value The value the named attribute needs to be set to in order to be considered a match * * \return TRUE if the locally cached CIB has the specified node attribute */ gboolean node_has_attr(const char *node, const char *name, const char *value) { char xpath[XPATH_MAX]; xmlNode *match; int n; CRM_CHECK(local_cib != NULL, return FALSE); /* Search for the node's attributes in the CIB. While the schema allows * multiple sets of instance attributes, and allows instance attributes to * use id-ref to reference values elsewhere, that is intended for resources, * so we ignore that here. */ n = snprintf(xpath, XPATH_MAX, "//" XML_CIB_TAG_NODES "/" XML_CIB_TAG_NODE "[@uname='%s']/" XML_TAG_ATTR_SETS "/" XML_CIB_TAG_NVPAIR "[@name='%s' and @value='%s']", node, name, value); match = get_xpath_object(xpath, local_cib, LOG_NEVER); CRM_CHECK(n < XPATH_MAX, return FALSE); return (match != NULL); } static void update_fencing_topology(const char *event, xmlNode * msg) { int format = 1; const char *xpath; xmlXPathObjectPtr xpathObj = NULL; xmlNode *patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); CRM_ASSERT(patchset); crm_element_value_int(patchset, "format", &format); if(format == 1) { /* Process deletions (only) */ xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_REMOVED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); remove_fencing_topology(xpathObj); freeXpathObject(xpathObj); /* Process additions and changes */ xpath = "//" F_CIB_UPDATE_RESULT "//" XML_TAG_DIFF_ADDED "//" XML_TAG_FENCING_LEVEL; xpathObj = xpath_search(msg, xpath); register_fencing_topology(xpathObj); freeXpathObject(xpathObj); } else if(format == 2) { xmlNode *change = NULL; int add[] = { 0, 0, 0 }; int del[] = { 0, 0, 0 }; xml_patch_versions(patchset, add, del); for (change = pcmk__xml_first_child(patchset); change != NULL; change = pcmk__xml_next(change)) { const char *op = crm_element_value(change, XML_DIFF_OP); const char *xpath = crm_element_value(change, XML_DIFF_PATH); if(op == NULL) { continue; } else if(strstr(xpath, "/" XML_TAG_FENCING_LEVEL) != NULL) { /* Change to a specific entry */ crm_trace("Handling %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); if(strcmp(op, "move") == 0) { continue; } else if(strcmp(op, "create") == 0) { handle_topology_change(change->children, FALSE); } else if(strcmp(op, "modify") == 0) { xmlNode *match = first_named_child(change, XML_DIFF_RESULT); if(match) { handle_topology_change(match->children, TRUE); } } else if(strcmp(op, "delete") == 0) { /* Nuclear option, all we have is the path and an id... not enough to remove a specific entry */ crm_info("Re-initializing fencing topology after %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else if (strstr(xpath, "/" XML_TAG_FENCING_TOPOLOGY) != NULL) { /* Change to the topology in general */ crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } else if (strstr(xpath, "/" XML_CIB_TAG_CONFIGURATION)) { /* Changes to the whole config section, possibly including the topology as a whild */ if(first_named_child(change, XML_TAG_FENCING_TOPOLOGY) == NULL) { crm_trace("Nothing for us in %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); } else if(strcmp(op, "delete") == 0 || strcmp(op, "create") == 0) { crm_info("Re-initializing fencing topology after top-level %s operation %d.%d.%d for %s.", op, add[0], add[1], add[2], xpath); fencing_topology_init(); return; } } else { crm_trace("Nothing for us in %s operation %d.%d.%d for %s", op, add[0], add[1], add[2], xpath); } } } else { crm_warn("Unknown patch format: %d", format); } } static bool have_cib_devices = FALSE; static void update_cib_cache_cb(const char *event, xmlNode * msg) { int rc = pcmk_ok; xmlNode *stonith_enabled_xml = NULL; xmlNode *stonith_watchdog_xml = NULL; const char *stonith_enabled_s = NULL; static gboolean stonith_enabled_saved = TRUE; if(!have_cib_devices) { crm_trace("Skipping updates until we get a full dump"); return; } else if(msg == NULL) { crm_trace("Missing %s update", event); return; } /* Maintain a local copy of the CIB so that we have full access * to device definitions, location constraints, and node attributes */ if (local_cib != NULL) { int rc = pcmk_ok; xmlNode *patchset = NULL; crm_element_value_int(msg, F_CIB_RC, &rc); if (rc != pcmk_ok) { return; } patchset = get_message_xml(msg, F_CIB_UPDATE_RESULT); xml_log_patchset(LOG_TRACE, "Config update", patchset); rc = xml_apply_patchset(local_cib, patchset, TRUE); switch (rc) { case pcmk_ok: case -pcmk_err_old_data: break; case -pcmk_err_diff_resync: case -pcmk_err_diff_failed: crm_notice("[%s] Patch aborted: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; break; default: crm_warn("[%s] ABORTED: %s (%d)", event, pcmk_strerror(rc), rc); free_xml(local_cib); local_cib = NULL; } } if (local_cib == NULL) { crm_trace("Re-requesting the full cib"); rc = cib_api->cmds->query(cib_api, NULL, &local_cib, cib_scope_local | cib_sync_call); if(rc != pcmk_ok) { crm_err("Couldn't retrieve the CIB: %s (%d)", pcmk_strerror(rc), rc); return; } CRM_ASSERT(local_cib != NULL); stonith_enabled_saved = FALSE; /* Trigger a full refresh below */ } crm_peer_caches_refresh(local_cib); stonith_enabled_xml = get_xpath_object("//nvpair[@name='stonith-enabled']", local_cib, LOG_NEVER); if (stonith_enabled_xml) { stonith_enabled_s = crm_element_value(stonith_enabled_xml, XML_NVPAIR_ATTR_VALUE); } if (stonith_enabled_s == NULL || crm_is_true(stonith_enabled_s)) { long timeout_ms = 0; const char *value = NULL; stonith_watchdog_xml = get_xpath_object("//nvpair[@name='stonith-watchdog-timeout']", local_cib, LOG_NEVER); if (stonith_watchdog_xml) { value = crm_element_value(stonith_watchdog_xml, XML_NVPAIR_ATTR_VALUE); } if(value) { timeout_ms = crm_get_msec(value); } if (timeout_ms < 0) { timeout_ms = pcmk__auto_watchdog_timeout(); } if(timeout_ms != stonith_watchdog_timeout_ms) { crm_notice("New watchdog timeout %lds (was %lds)", timeout_ms/1000, stonith_watchdog_timeout_ms/1000); stonith_watchdog_timeout_ms = timeout_ms; } } else { stonith_watchdog_timeout_ms = 0; } if (stonith_enabled_s && crm_is_true(stonith_enabled_s) == FALSE) { crm_trace("Ignoring cib updates while stonith is disabled"); stonith_enabled_saved = FALSE; return; } else if (stonith_enabled_saved == FALSE) { crm_info("Updating stonith device and topology lists now that stonith is enabled"); stonith_enabled_saved = TRUE; fencing_topology_init(); cib_devices_update(); } else { update_fencing_topology(event, msg); update_cib_stonith_devices(event, msg); } } static void init_cib_cache_cb(xmlNode * msg, int call_id, int rc, xmlNode * output, void *user_data) { crm_info("Updating device list from the cib: init"); have_cib_devices = TRUE; local_cib = copy_xml(output); crm_peer_caches_refresh(local_cib); fencing_topology_init(); cib_devices_update(); } static void stonith_shutdown(int nsig) { crm_info("Terminating with %d clients", pcmk__ipc_client_count()); stonith_shutdown_flag = TRUE; if (mainloop != NULL && g_main_loop_is_running(mainloop)) { g_main_loop_quit(mainloop); } else { stonith_cleanup(); crm_exit(CRM_EX_OK); } } static void cib_connection_destroy(gpointer user_data) { if (stonith_shutdown_flag) { crm_info("Connection to the CIB manager closed"); return; } else { crm_crit("Lost connection to the CIB manager, shutting down"); } if (cib_api) { cib_api->cmds->signoff(cib_api); } stonith_shutdown(0); } static void stonith_cleanup(void) { if (cib_api) { cib_api->cmds->del_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb); cib_api->cmds->signoff(cib_api); } if (ipcs) { qb_ipcs_destroy(ipcs); } crm_peer_destroy(); pcmk__client_cleanup(); free_stonith_remote_op_list(); free_topology_list(); free_device_list(); free_metadata_cache(); free(stonith_our_uname); stonith_our_uname = NULL; free_xml(local_cib); local_cib = NULL; } static pcmk__cli_option_t long_options[] = { // long option, argument type, storage, short option, description, flags { "stand-alone", no_argument, 0, 's', NULL, pcmk__option_default }, { "stand-alone-w-cpg", no_argument, 0, 'c', NULL, pcmk__option_default }, { "logfile", required_argument, 0, 'l', NULL, pcmk__option_default }, { "verbose", no_argument, 0, 'V', NULL, pcmk__option_default }, { "version", no_argument, 0, '$', NULL, pcmk__option_default }, { "help", no_argument, 0, '?', NULL, pcmk__option_default }, { 0, 0, 0, 0 } }; static void setup_cib(void) { int rc, retries = 0; cib_api = cib_new(); if (cib_api == NULL) { crm_err("No connection to the CIB manager"); return; } do { sleep(retries); rc = cib_api->cmds->signon(cib_api, CRM_SYSTEM_STONITHD, cib_command); } while (rc == -ENOTCONN && ++retries < 5); if (rc != pcmk_ok) { crm_err("Could not connect to the CIB manager: %s (%d)", pcmk_strerror(rc), rc); } else if (pcmk_ok != cib_api->cmds->add_notify_callback(cib_api, T_CIB_DIFF_NOTIFY, update_cib_cache_cb)) { crm_err("Could not set CIB notification callback"); } else { rc = cib_api->cmds->query(cib_api, NULL, NULL, cib_scope_local); cib_api->cmds->register_callback(cib_api, rc, 120, FALSE, NULL, "init_cib_cache_cb", init_cib_cache_cb); cib_api->cmds->set_connection_dnotify(cib_api, cib_connection_destroy); crm_info("Watching for stonith topology changes"); } } struct qb_ipcs_service_handlers ipc_callbacks = { .connection_accept = st_ipc_accept, .connection_created = NULL, .msg_process = st_ipc_dispatch, .connection_closed = st_ipc_closed, .connection_destroyed = st_ipc_destroy }; /*! * \internal * \brief Callback for peer status changes * * \param[in] type What changed * \param[in] node What peer had the change * \param[in] data Previous value of what changed */ static void st_peer_update_callback(enum crm_status_type type, crm_node_t * node, const void *data) { if ((type != crm_status_processes) && !pcmk_is_set(node->flags, crm_remote_node)) { /* * This is a hack until we can send to a nodeid and/or we fix node name lookups * These messages are ignored in stonith_peer_callback() */ xmlNode *query = create_xml_node(NULL, "stonith_command"); crm_xml_add(query, F_XML_TAGNAME, "stonith_command"); crm_xml_add(query, F_TYPE, T_STONITH_NG); crm_xml_add(query, F_STONITH_OPERATION, "poke"); crm_debug("Broadcasting our uname because of node %u", node->id); send_cluster_message(NULL, crm_msg_stonith_ng, query, FALSE); free_xml(query); } } int main(int argc, char **argv) { int flag; int lpc = 0; int argerr = 0; int option_index = 0; crm_cluster_t cluster; const char *actions[] = { "reboot", "off", "on", "list", "monitor", "status" }; crm_ipc_t *old_instance = NULL; crm_log_preinit(NULL, argc, argv); pcmk__set_cli_options(NULL, "[options]", long_options, "daemon for executing fencing devices in a " "Pacemaker cluster"); while (1) { flag = pcmk__next_cli_option(argc, argv, &option_index, NULL); if (flag == -1) { break; } switch (flag) { case 'V': crm_bump_log_level(argc, argv); break; case 'l': crm_add_logfile(optarg); break; case 's': stand_alone = TRUE; break; case 'c': stand_alone = FALSE; no_cib_connect = TRUE; break; case '$': case '?': pcmk__cli_help(flag, CRM_EX_OK); break; default: ++argerr; break; } } if (argc - optind == 1 && pcmk__str_eq("metadata", argv[optind], pcmk__str_casei)) { printf("\n"); printf("\n"); printf(" 1.0\n"); printf(" Instance attributes available for all \"stonith\"-class resources" " and used by Pacemaker's fence daemon, formerly known as stonithd\n"); printf(" Instance attributes available for all \"stonith\"-class resources\n"); printf(" \n"); #if 0 // priority is not implemented yet printf(" \n"); printf(" Devices that are not in a topology " "are tried in order of highest to lowest integer priority\n"); printf(" \n"); printf(" \n"); #endif - printf(" \n", STONITH_ATTR_HOSTARG); + printf(" \n", + PCMK_STONITH_HOST_ARGUMENT); printf (" Advanced use only: An alternate parameter to supply instead of 'port'\n"); printf (" Some devices do not support the standard 'port' parameter or may provide additional ones.\n" "Use this to specify an alternate, device-specific, parameter that should indicate the machine to be fenced.\n" "A value of 'none' can be used to tell the cluster not to supply any additional parameters.\n" " \n"); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_HOSTMAP); + printf(" \n", + PCMK_STONITH_HOST_MAP); printf (" A mapping of host names to ports numbers for devices that do not support host names.\n"); printf (" Eg. node1:1;node2:2,3 would tell the cluster to use port 1 for node1 and ports 2 and 3 for node2\n"); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_HOSTLIST); - printf - (" A list of machines controlled by this device (Optional unless %s=static-list).\n", - STONITH_ATTR_HOSTCHECK); + printf(" \n", + PCMK_STONITH_HOST_LIST); + printf(" A list of machines controlled by " + "this device (Optional unless %s=static-list).\n", + PCMK_STONITH_HOST_CHECK); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_HOSTCHECK); + printf(" \n", + PCMK_STONITH_HOST_CHECK); printf (" How to determine which machines are controlled by the device.\n"); printf(" Allowed values: dynamic-list " "(query the device via the 'list' command), static-list " - "(check the " STONITH_ATTR_HOSTLIST " attribute), status " + "(check the " PCMK_STONITH_HOST_LIST " attribute), status " "(query the device via the 'status' command), none (assume " "every device can fence every machine)\n"); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_DELAY_MAX); + printf(" \n", + PCMK_STONITH_DELAY_MAX); printf (" Enable a random delay for stonith actions and specify the maximum of random delay.\n"); printf (" This prevents double fencing when using slow devices such as sbd.\n" "Use this to enable a random delay for stonith actions.\n" "The overall delay is derived from this random delay value adding a static delay so that the sum is kept below the maximum delay.\n"); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_DELAY_BASE); + printf(" \n", + PCMK_STONITH_DELAY_BASE); printf (" Enable a base delay for stonith actions and specify base delay value.\n"); printf (" This prevents double fencing when different delays are configured on the nodes.\n" "Use this to enable a static delay for stonith actions.\n" "The overall delay is derived from a random delay value adding this static delay so that the sum is kept below the maximum delay.\n"); printf(" \n"); printf(" \n"); - printf(" \n", STONITH_ATTR_ACTION_LIMIT); + printf(" \n", + PCMK_STONITH_ACTION_LIMIT); printf (" The maximum number of actions can be performed in parallel on this device\n"); printf (" Cluster property concurrent-fencing=true needs to be configured first.\n" "Then use this to specify the maximum number of actions can be performed in parallel on this device. -1 is unlimited.\n"); printf(" \n"); printf(" \n"); for (lpc = 0; lpc < DIMOF(actions); lpc++) { printf(" \n", actions[lpc]); printf (" Advanced use only: An alternate command to run instead of '%s'\n", actions[lpc]); printf (" Some devices do not support the standard commands or may provide additional ones.\n" "Use this to specify an alternate, device-specific, command that implements the '%s' action.\n", actions[lpc]); printf(" \n", actions[lpc]); printf(" \n"); printf(" \n", actions[lpc]); printf (" Advanced use only: Specify an alternate timeout to use for %s actions instead of stonith-timeout\n", actions[lpc]); printf (" Some devices need much more/less time to complete than normal.\n" "Use this to specify an alternate, device-specific, timeout for '%s' actions.\n", actions[lpc]); printf(" \n"); printf(" \n"); printf(" \n", actions[lpc]); printf (" Advanced use only: The maximum number of times to retry the '%s' command within the timeout period\n", actions[lpc]); printf(" Some devices do not support multiple connections." " Operations may 'fail' if the device is busy with another task so Pacemaker will automatically retry the operation, if there is time remaining." " Use this option to alter the number of times Pacemaker retries '%s' actions before giving up." "\n", actions[lpc]); printf(" \n"); printf(" \n"); } printf(" \n"); printf("\n"); return CRM_EX_OK; } if (optind != argc) { ++argerr; } if (argerr) { pcmk__cli_help('?', CRM_EX_USAGE); } crm_log_init(NULL, LOG_INFO, TRUE, FALSE, argc, argv, FALSE); crm_notice("Starting Pacemaker fencer"); old_instance = crm_ipc_new("stonith-ng", 0); if (crm_ipc_connect(old_instance)) { /* IPC end-point already up */ crm_ipc_close(old_instance); crm_ipc_destroy(old_instance); crm_err("pacemaker-fenced is already active, aborting startup"); crm_exit(CRM_EX_OK); } else { /* not up or not authentic, we'll proceed either way */ crm_ipc_destroy(old_instance); old_instance = NULL; } mainloop_add_signal(SIGTERM, stonith_shutdown); crm_peer_init(); fenced_data_set = pe_new_working_set(); CRM_ASSERT(fenced_data_set != NULL); pe__set_working_set_flags(fenced_data_set, pe_flag_no_counts|pe_flag_no_compat); if (stand_alone == FALSE) { if (is_corosync_cluster()) { #if SUPPORT_COROSYNC cluster.destroy = stonith_peer_cs_destroy; cluster.cpg.cpg_deliver_fn = stonith_peer_ais_callback; cluster.cpg.cpg_confchg_fn = pcmk_cpg_membership; #endif } crm_set_status_callback(&st_peer_update_callback); if (crm_cluster_connect(&cluster) == FALSE) { crm_crit("Cannot sign in to the cluster... terminating"); crm_exit(CRM_EX_FATAL); } stonith_our_uname = cluster.uname; stonith_our_uuid = cluster.uuid; if (no_cib_connect == FALSE) { setup_cib(); } } else { stonith_our_uname = strdup("localhost"); } init_device_list(); init_topology_list(); if(stonith_watchdog_timeout_ms > 0) { int rc; xmlNode *xml; stonith_key_value_t *params = NULL; - params = stonith_key_value_add(params, STONITH_ATTR_HOSTLIST, stonith_our_uname); + params = stonith_key_value_add(params, PCMK_STONITH_HOST_LIST, + stonith_our_uname); xml = create_device_registration_xml("watchdog", st_namespace_internal, STONITH_WATCHDOG_AGENT, params, NULL); stonith_key_value_freeall(params, 1, 1); rc = stonith_device_register(xml, NULL, FALSE); free_xml(xml); if (rc != pcmk_ok) { crm_crit("Cannot register watchdog pseudo fence agent"); crm_exit(CRM_EX_FATAL); } } pcmk__serve_fenced_ipc(&ipcs, &ipc_callbacks); /* Create the mainloop and run it... */ mainloop = g_main_loop_new(NULL, FALSE); crm_notice("Pacemaker fencer successfully started and accepting connections"); g_main_loop_run(mainloop); stonith_cleanup(); pe_free_working_set(fenced_data_set); crm_exit(CRM_EX_OK); } diff --git a/include/crm/common/agents.h b/include/crm/common/agents.h index b585ada05f..e9089aeab9 100644 --- a/include/crm/common/agents.h +++ b/include/crm/common/agents.h @@ -1,58 +1,71 @@ /* * Copyright 2017-2020 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. */ #ifndef PCMK__AGENTS__H # define PCMK__AGENTS__H #ifdef __cplusplus extern "C" { #endif /** * \file * \brief API related to resource agents * \ingroup core */ #include // uint32_t #include +/* Special stonith-class agent parameters interpreted directly by Pacemaker + * (not including the pcmk_ACTION_{action,retries,timeout} parameters) + */ +#define PCMK_STONITH_ACTION_LIMIT "pcmk_action_limit" +#define PCMK_STONITH_DELAY_BASE "pcmk_delay_base" +#define PCMK_STONITH_DELAY_MAX "pcmk_delay_max" +#define PCMK_STONITH_HOST_ARGUMENT "pcmk_host_argument" +#define PCMK_STONITH_HOST_CHECK "pcmk_host_check" +#define PCMK_STONITH_HOST_LIST "pcmk_host_list" +#define PCMK_STONITH_HOST_MAP "pcmk_host_map" +#define PCMK_STONITH_PROVIDES "provides" +#define PCMK_STONITH_STONITH_TIMEOUT "stonith-timeout" + // Capabilities supported by a resource agent standard enum pcmk_ra_caps { pcmk_ra_cap_none = 0, pcmk_ra_cap_provider = (1 << 0), // Requires provider pcmk_ra_cap_status = (1 << 1), // Supports status instead of monitor pcmk_ra_cap_params = (1 << 2), // Supports parameters pcmk_ra_cap_unique = (1 << 3), // Supports unique clones pcmk_ra_cap_promotable = (1 << 4), // Supports promotable clones pcmk_ra_cap_stdin = (1 << 5), // Reads from standard input pcmk_ra_cap_fence_params = (1 << 6), // Supports pcmk_monitor_timeout, etc. }; uint32_t pcmk_get_ra_caps(const char *standard); char *crm_generate_ra_key(const char *standard, const char *provider, const char *type); int crm_parse_agent_spec(const char *spec, char **standard, char **provider, char **type); #ifndef PCMK__NO_COMPAT /* Everything here is deprecated and kept only for public API backward * compatibility. It will be moved to compatibility.h in a future release. */ //! \deprecated Use pcmk_get_ra_caps() instead bool crm_provider_required(const char *standard); #endif // PCMK__NO_COMPAT #ifdef __cplusplus } #endif #endif // PCMK__AGENTS__H diff --git a/include/crm/fencing/internal.h b/include/crm/fencing/internal.h index f33957f749..391ab72a68 100644 --- a/include/crm/fencing/internal.h +++ b/include/crm/fencing/internal.h @@ -1,218 +1,210 @@ /* * Copyright 2011-2020 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. */ #ifndef STONITH_NG_INTERNAL__H # define STONITH_NG_INTERNAL__H # include # include # include # include # include enum st_device_flags { st_device_supports_list = 0x0001, st_device_supports_status = 0x0002, st_device_supports_reboot = 0x0004, st_device_supports_parameter_plug = 0x0008, st_device_supports_parameter_port = 0x0010, }; #define stonith__set_device_flags(device_flags, device_id, flags_to_set) do { \ device_flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Fence device", device_id, \ (device_flags), (flags_to_set), \ #flags_to_set); \ } while (0) #define stonith__set_call_options(st_call_opts, call_for, flags_to_set) do { \ st_call_opts = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, \ "Fencer call", (call_for), \ (st_call_opts), (flags_to_set), \ #flags_to_set); \ } while (0) #define stonith__clear_call_options(st_call_opts, call_for, flags_to_clear) do { \ st_call_opts = pcmk__clear_flags_as(__func__, __LINE__, LOG_TRACE, \ "Fencer call", (call_for), \ (st_call_opts), (flags_to_clear), \ #flags_to_clear); \ } while (0) struct stonith_action_s; typedef struct stonith_action_s stonith_action_t; stonith_action_t *stonith_action_create(const char *agent, const char *_action, const char *victim, uint32_t victim_nodeid, int timeout, GHashTable * device_args, GHashTable * port_map, const char * host_arg); void stonith__destroy_action(stonith_action_t *action); void stonith__action_result(stonith_action_t *action, int *rc, char **output, char **error_output); int stonith_action_execute_async(stonith_action_t * action, void *userdata, void (*done) (GPid pid, int rc, const char *output, gpointer user_data), void (*fork_cb) (GPid pid, gpointer user_data)); int stonith__execute(stonith_action_t *action); xmlNode *create_level_registration_xml(const char *node, const char *pattern, const char *attr, const char *value, int level, stonith_key_value_t *device_list); xmlNode *create_device_registration_xml(const char *id, enum stonith_namespace namespace, const char *agent, stonith_key_value_t *params, const char *rsc_provides); void stonith__register_messages(pcmk__output_t *out); GList *stonith__parse_targets(const char *hosts); gboolean stonith__later_succeeded(stonith_history_t *event, stonith_history_t *top_history); stonith_history_t *stonith__sort_history(stonith_history_t *history); void stonith__device_parameter_flags(uint32_t *device_flags, const char *device_name, xmlNode *metadata); # define ST_LEVEL_MAX 10 # define F_STONITH_CLIENTID "st_clientid" # define F_STONITH_CALLOPTS "st_callopt" # define F_STONITH_CALLID "st_callid" # define F_STONITH_CALLDATA "st_calldata" # define F_STONITH_OPERATION "st_op" # define F_STONITH_TARGET "st_target" # define F_STONITH_REMOTE_OP_ID "st_remote_op" # define F_STONITH_REMOTE_OP_ID_RELAY "st_remote_op_relay" # define F_STONITH_RC "st_rc" /*! Timeout period per a device execution */ # define F_STONITH_TIMEOUT "st_timeout" # define F_STONITH_TOLERANCE "st_tolerance" # define F_STONITH_DELAY "st_delay" /*! Action specific timeout period returned in query of fencing devices. */ # define F_STONITH_ACTION_TIMEOUT "st_action_timeout" /*! Host in query result is not allowed to run this action */ # define F_STONITH_ACTION_DISALLOWED "st_action_disallowed" /*! Maximum of random fencing delay for a device */ # define F_STONITH_DELAY_MAX "st_delay_max" /*! Base delay used for a fencing delay */ # define F_STONITH_DELAY_BASE "st_delay_base" /*! Has this device been verified using a monitor type * operation (monitor, list, status) */ # define F_STONITH_DEVICE_VERIFIED "st_monitor_verified" /*! device is required for this action */ # define F_STONITH_DEVICE_REQUIRED "st_required" /*! number of available devices in query result */ # define F_STONITH_AVAILABLE_DEVICES "st-available-devices" # define F_STONITH_CALLBACK_TOKEN "st_async_id" # define F_STONITH_CLIENTNAME "st_clientname" # define F_STONITH_CLIENTNODE "st_clientnode" # define F_STONITH_NOTIFY_ACTIVATE "st_notify_activate" # define F_STONITH_NOTIFY_DEACTIVATE "st_notify_deactivate" # define F_STONITH_DELEGATE "st_delegate" /*! The node initiating the stonith operation. If an operation * is relayed, this is the last node the operation lands on. When * in standalone mode, origin is the client's id that originated the * operation. */ # define F_STONITH_ORIGIN "st_origin" # define F_STONITH_HISTORY_LIST "st_history" # define F_STONITH_DATE "st_date" # define F_STONITH_STATE "st_state" # define F_STONITH_ACTIVE "st_active" # define F_STONITH_DIFFERENTIAL "st_differential" # define F_STONITH_DEVICE "st_device_id" # define F_STONITH_ACTION "st_device_action" # define F_STONITH_MODE "st_mode" # define F_STONITH_MERGED "st_op_merged" # define T_STONITH_NG "stonith-ng" # define T_STONITH_REPLY "st-reply" /*! For async operations, an event from the server containing * the total amount of time the server is allowing for the operation * to take place is returned to the client. */ # define T_STONITH_TIMEOUT_VALUE "st-async-timeout-value" # define T_STONITH_NOTIFY "st_notify" -# define STONITH_ATTR_HOSTARG "pcmk_host_argument" -# define STONITH_ATTR_HOSTMAP "pcmk_host_map" -# define STONITH_ATTR_HOSTLIST "pcmk_host_list" -# define STONITH_ATTR_HOSTCHECK "pcmk_host_check" -# define STONITH_ATTR_DELAY_MAX "pcmk_delay_max" -# define STONITH_ATTR_DELAY_BASE "pcmk_delay_base" -# define STONITH_ATTR_ACTION_LIMIT "pcmk_action_limit" - # define STONITH_ATTR_ACTION_OP "action" # define STONITH_OP_EXEC "st_execute" # define STONITH_OP_TIMEOUT_UPDATE "st_timeout_update" # define STONITH_OP_QUERY "st_query" # define STONITH_OP_FENCE "st_fence" # define STONITH_OP_RELAY "st_relay" # define STONITH_OP_DEVICE_ADD "st_device_register" # define STONITH_OP_DEVICE_DEL "st_device_remove" # define STONITH_OP_FENCE_HISTORY "st_fence_history" # define STONITH_OP_LEVEL_ADD "st_level_add" # define STONITH_OP_LEVEL_DEL "st_level_remove" # define STONITH_WATCHDOG_AGENT "#watchdog" # ifdef HAVE_STONITH_STONITH_H // utilities from st_lha.c int stonith__list_lha_agents(stonith_key_value_t **devices); int stonith__lha_metadata(const char *agent, int timeout, char **output); bool stonith__agent_is_lha(const char *agent); int stonith__lha_validate(stonith_t *st, int call_options, const char *target, const char *agent, GHashTable *params, int timeout, char **output, char **error_output); # endif // utilities from st_rhcs.c int stonith__list_rhcs_agents(stonith_key_value_t **devices); int stonith__rhcs_metadata(const char *agent, int timeout, char **output); bool stonith__agent_is_rhcs(const char *agent); int stonith__rhcs_validate(stonith_t *st, int call_options, const char *target, const char *agent, GHashTable *params, const char *host_arg, int timeout, char **output, char **error_output); int stonith__failed_history(pcmk__output_t *out, va_list args); int stonith__history(pcmk__output_t *out, va_list args); int stonith__full_history(pcmk__output_t *out, va_list args); int stonith__full_history_xml(pcmk__output_t *out, va_list args); int stonith__last_fenced_html(pcmk__output_t *out, va_list args); int stonith__last_fenced_text(pcmk__output_t *out, va_list args); int stonith__last_fenced_xml(pcmk__output_t *out, va_list args); int stonith__pending_actions(pcmk__output_t *out, va_list args); int stonith__event_html(pcmk__output_t *out, va_list args); int stonith__event_text(pcmk__output_t *out, va_list args); int stonith__event_xml(pcmk__output_t *out, va_list args); int stonith__validate_agent_html(pcmk__output_t *out, va_list args); int stonith__validate_agent_text(pcmk__output_t *out, va_list args); int stonith__validate_agent_xml(pcmk__output_t *out, va_list args); stonith_history_t *stonith__first_matching_event(stonith_history_t *history, bool (*matching_fn)(stonith_history_t *, void *), void *user_data); bool stonith__event_state_pending(stonith_history_t *history, void *user_data); bool stonith__event_state_eq(stonith_history_t *history, void *user_data); bool stonith__event_state_neq(stonith_history_t *history, void *user_data); #endif diff --git a/include/crm/msg_xml.h b/include/crm/msg_xml.h index 1fcb72d090..c8b528b8db 100644 --- a/include/crm/msg_xml.h +++ b/include/crm/msg_xml.h @@ -1,432 +1,442 @@ /* * 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. */ #ifndef XML_TAGS__H # define XML_TAGS__H #ifdef __cplusplus extern "C" { #endif # ifndef F_ORIG # define F_ORIG "src" # endif # ifndef F_SEQ # define F_SEQ "seq" # endif # ifndef F_SUBTYPE # define F_SUBTYPE "subt" # endif # ifndef F_TYPE # define F_TYPE "t" # endif # ifndef F_CLIENTNAME # define F_CLIENTNAME "cn" # endif # ifndef F_XML_TAGNAME # define F_XML_TAGNAME "__name__" # endif # ifndef T_CRM # define T_CRM "crmd" # endif # ifndef T_ATTRD # define T_ATTRD "attrd" # endif # define CIB_OPTIONS_FIRST "cib-bootstrap-options" # define F_CRM_DATA "crm_xml" # define F_CRM_TASK "crm_task" # define F_CRM_HOST_TO "crm_host_to" # define F_CRM_MSG_TYPE F_SUBTYPE # define F_CRM_SYS_TO "crm_sys_to" # define F_CRM_SYS_FROM "crm_sys_from" # define F_CRM_HOST_FROM F_ORIG # define F_CRM_REFERENCE XML_ATTR_REFERENCE # define F_CRM_VERSION XML_ATTR_VERSION # define F_CRM_ORIGIN "origin" # define F_CRM_USER "crm_user" # define F_CRM_JOIN_ID "join_id" # define F_CRM_DC_LEAVING "dc-leaving" # define F_CRM_ELECTION_ID "election-id" # define F_CRM_ELECTION_AGE_S "election-age-sec" # define F_CRM_ELECTION_AGE_US "election-age-nano-sec" # define F_CRM_ELECTION_OWNER "election-owner" # define F_CRM_TGRAPH "crm-tgraph-file" # define F_CRM_TGRAPH_INPUT "crm-tgraph-in" # define F_CRM_THROTTLE_MODE "crm-limit-mode" # define F_CRM_THROTTLE_MAX "crm-limit-max" /*---- Common tags/attrs */ # define XML_DIFF_MARKER "__crm_diff_marker__" # define XML_TAG_CIB "cib" # define XML_TAG_FAILED "failed" # define XML_ATTR_CRM_VERSION "crm_feature_set" # define XML_ATTR_DIGEST "digest" # define XML_ATTR_VALIDATION "validate-with" # define XML_ATTR_RA_VERSION "ra-version" # define XML_ATTR_QUORUM_PANIC "no-quorum-panic" # define XML_ATTR_HAVE_QUORUM "have-quorum" # define XML_ATTR_HAVE_WATCHDOG "have-watchdog" # define XML_ATTR_GENERATION "epoch" # define XML_ATTR_GENERATION_ADMIN "admin_epoch" # define XML_ATTR_NUMUPDATES "num_updates" # define XML_ATTR_TIMEOUT "timeout" # define XML_ATTR_ORIGIN "crm-debug-origin" # define XML_ATTR_TSTAMP "crm-timestamp" # define XML_CIB_ATTR_WRITTEN "cib-last-written" # define XML_ATTR_VERSION "version" # define XML_ATTR_DESC "description" # define XML_ATTR_ID "id" # define XML_ATTR_IDREF "id-ref" # define XML_ATTR_ID_LONG "long-id" # define XML_ATTR_TYPE "type" # define XML_ATTR_VERBOSE "verbose" # define XML_ATTR_OP "op" # define XML_ATTR_DC_UUID "dc-uuid" # define XML_ATTR_UPDATE_ORIG "update-origin" # define XML_ATTR_UPDATE_CLIENT "update-client" # define XML_ATTR_UPDATE_USER "update-user" # define XML_BOOLEAN_TRUE "true" # define XML_BOOLEAN_FALSE "false" # define XML_BOOLEAN_YES XML_BOOLEAN_TRUE # define XML_BOOLEAN_NO XML_BOOLEAN_FALSE # define XML_TAG_OPTIONS "options" /*---- top level tags/attrs */ # define XML_ATTR_REQUEST "request" # define XML_ATTR_RESPONSE "response" # define XML_ATTR_UNAME "uname" # define XML_ATTR_UUID "id" # define XML_ATTR_REFERENCE "reference" # define XML_CRM_TAG_PING "ping_response" # define XML_PING_ATTR_STATUS "result" # define XML_PING_ATTR_SYSFROM "crm_subsystem" # define XML_PING_ATTR_CRMDSTATE "crmd_state" # define XML_PING_ATTR_PACEMAKERDSTATE "pacemakerd_state" # define XML_PING_ATTR_PACEMAKERDSTATE_INIT "init" # define XML_PING_ATTR_PACEMAKERDSTATE_STARTINGDAEMONS "starting_daemons" # define XML_PING_ATTR_PACEMAKERDSTATE_WAITPING "wait_for_ping" # define XML_PING_ATTR_PACEMAKERDSTATE_RUNNING "running" # define XML_PING_ATTR_PACEMAKERDSTATE_SHUTTINGDOWN "shutting_down" # define XML_PING_ATTR_PACEMAKERDSTATE_SHUTDOWNCOMPLETE "shutdown_complete" # define XML_TAG_FRAGMENT "cib_fragment" # define XML_FAIL_TAG_CIB "failed_update" # define XML_FAILCIB_ATTR_ID "id" # define XML_FAILCIB_ATTR_OBJTYPE "object_type" # define XML_FAILCIB_ATTR_OP "operation" # define XML_FAILCIB_ATTR_REASON "reason" /*---- CIB specific tags/attrs */ # define XML_CIB_TAG_SECTION_ALL "all" # define XML_CIB_TAG_CONFIGURATION "configuration" # define XML_CIB_TAG_STATUS "status" # define XML_CIB_TAG_RESOURCES "resources" # define XML_CIB_TAG_NODES "nodes" # define XML_CIB_TAG_DOMAINS "domains" # define XML_CIB_TAG_CONSTRAINTS "constraints" # define XML_CIB_TAG_CRMCONFIG "crm_config" # define XML_CIB_TAG_OPCONFIG "op_defaults" # define XML_CIB_TAG_RSCCONFIG "rsc_defaults" # define XML_CIB_TAG_ACLS "acls" # define XML_CIB_TAG_ALERTS "alerts" # define XML_CIB_TAG_ALERT "alert" # define XML_CIB_TAG_ALERT_RECIPIENT "recipient" # define XML_CIB_TAG_ALERT_SELECT "select" # define XML_CIB_TAG_ALERT_ATTRIBUTES "select_attributes" # define XML_CIB_TAG_ALERT_FENCING "select_fencing" # define XML_CIB_TAG_ALERT_NODES "select_nodes" # define XML_CIB_TAG_ALERT_RESOURCES "select_resources" # define XML_CIB_TAG_ALERT_ATTR "attribute" # define XML_CIB_TAG_STATE "node_state" # define XML_CIB_TAG_NODE "node" # define XML_CIB_TAG_NVPAIR "nvpair" # define XML_CIB_TAG_PROPSET "cluster_property_set" # define XML_TAG_ATTR_SETS "instance_attributes" # define XML_TAG_META_SETS "meta_attributes" # define XML_TAG_ATTRS "attributes" # define XML_TAG_RSC_VER_ATTRS "rsc_versioned_attrs" # define XML_TAG_OP_VER_ATTRS "op_versioned_attrs" # define XML_TAG_OP_VER_META "op_versioned_meta" # define XML_TAG_PARAMS "parameters" # define XML_TAG_PARAM "param" # define XML_TAG_UTILIZATION "utilization" # define XML_TAG_RESOURCE_REF "resource_ref" # define XML_CIB_TAG_RESOURCE "primitive" # define XML_CIB_TAG_GROUP "group" # define XML_CIB_TAG_INCARNATION "clone" # define XML_CIB_TAG_MASTER "master" // deprecated since 2.0.0 # define XML_CIB_TAG_CONTAINER "bundle" # define XML_CIB_TAG_RSC_TEMPLATE "template" # define XML_RSC_ATTR_TARGET "container-attribute-target" # define XML_RSC_ATTR_RESTART "restart-type" # define XML_RSC_ATTR_ORDERED "ordered" # define XML_RSC_ATTR_INTERLEAVE "interleave" # define XML_RSC_ATTR_INCARNATION "clone" # define XML_RSC_ATTR_INCARNATION_MAX "clone-max" # define XML_RSC_ATTR_INCARNATION_MIN "clone-min" # define XML_RSC_ATTR_INCARNATION_NODEMAX "clone-node-max" # define XML_RSC_ATTR_PROMOTABLE "promotable" # define XML_RSC_ATTR_PROMOTED_MAX "promoted-max" # define XML_RSC_ATTR_PROMOTED_NODEMAX "promoted-node-max" # define XML_RSC_ATTR_MASTER_MAX "master-max" // deprecated since 2.0.0 # define XML_RSC_ATTR_MASTER_NODEMAX "master-node-max" // deprecated since 2.0.0 # define XML_RSC_ATTR_MANAGED "is-managed" # define XML_RSC_ATTR_TARGET_ROLE "target-role" # define XML_RSC_ATTR_UNIQUE "globally-unique" # define XML_RSC_ATTR_NOTIFY "notify" # define XML_RSC_ATTR_STICKINESS "resource-stickiness" # define XML_RSC_ATTR_FAIL_STICKINESS "migration-threshold" # define XML_RSC_ATTR_FAIL_TIMEOUT "failure-timeout" # define XML_RSC_ATTR_MULTIPLE "multiple-active" # define XML_RSC_ATTR_REQUIRES "requires" -# define XML_RSC_ATTR_PROVIDES "provides" # define XML_RSC_ATTR_CONTAINER "container" # define XML_RSC_ATTR_INTERNAL_RSC "internal_rsc" # define XML_RSC_ATTR_MAINTENANCE "maintenance" # define XML_RSC_ATTR_REMOTE_NODE "remote-node" # define XML_RSC_ATTR_CLEAR_OP "clear_failure_op" # define XML_RSC_ATTR_CLEAR_INTERVAL "clear_failure_interval" # define XML_RSC_ATTR_REMOTE_RA_ADDR "addr" # define XML_RSC_ATTR_REMOTE_RA_SERVER "server" # define XML_RSC_ATTR_REMOTE_RA_PORT "port" # define XML_REMOTE_ATTR_RECONNECT_INTERVAL "reconnect_interval" # define XML_OP_ATTR_ON_FAIL "on-fail" # define XML_OP_ATTR_START_DELAY "start-delay" # define XML_OP_ATTR_ALLOW_MIGRATE "allow-migrate" # define XML_OP_ATTR_ORIGIN "interval-origin" # define XML_OP_ATTR_PENDING "record-pending" # define XML_OP_ATTR_DIGESTS_ALL "digests-all" # define XML_OP_ATTR_DIGESTS_SECURE "digests-secure" # define XML_CIB_TAG_LRM "lrm" # define XML_LRM_TAG_RESOURCES "lrm_resources" # define XML_LRM_TAG_RESOURCE "lrm_resource" # define XML_LRM_TAG_RSC_OP "lrm_rsc_op" # define XML_AGENT_ATTR_CLASS "class" # define XML_AGENT_ATTR_PROVIDER "provider" # define XML_CIB_ATTR_REPLACE "replace" # define XML_CIB_ATTR_SOURCE "source" # define XML_CIB_ATTR_PRIORITY "priority" # define XML_CIB_ATTR_SOURCE "source" # define XML_NODE_JOIN_STATE "join" # define XML_NODE_EXPECTED "expected" # define XML_NODE_IN_CLUSTER "in_ccm" # define XML_NODE_IS_PEER "crmd" # define XML_NODE_IS_REMOTE "remote_node" # define XML_NODE_IS_FENCED "node_fenced" # define XML_NODE_IS_MAINTENANCE "node_in_maintenance" # define XML_CIB_ATTR_SHUTDOWN "shutdown" /* Aside from being an old name for the executor, LRM is a misnomer here because * the controller and scheduler use these to track actions, which are not always * executor operations. */ // XML attribute that takes interval specification (user-facing configuration) # define XML_LRM_ATTR_INTERVAL "interval" // XML attribute that takes interval in milliseconds (daemon APIs) // (identical value as above, but different constant allows clearer code intent) # define XML_LRM_ATTR_INTERVAL_MS XML_LRM_ATTR_INTERVAL # define XML_LRM_ATTR_TASK "operation" # define XML_LRM_ATTR_TASK_KEY "operation_key" # define XML_LRM_ATTR_TARGET "on_node" # define XML_LRM_ATTR_TARGET_UUID "on_node_uuid" /*! Actions to be executed on Pacemaker Remote nodes are routed through the * controller on the cluster node hosting the remote connection. That cluster * node is considered the router node for the action. */ # define XML_LRM_ATTR_ROUTER_NODE "router_node" # define XML_LRM_ATTR_RSCID "rsc-id" # define XML_LRM_ATTR_OPSTATUS "op-status" # define XML_LRM_ATTR_RC "rc-code" # define XML_LRM_ATTR_CALLID "call-id" # define XML_LRM_ATTR_OP_DIGEST "op-digest" # define XML_LRM_ATTR_OP_RESTART "op-force-restart" # define XML_LRM_ATTR_OP_SECURE "op-secure-params" # define XML_LRM_ATTR_RESTART_DIGEST "op-restart-digest" # define XML_LRM_ATTR_SECURE_DIGEST "op-secure-digest" # define XML_LRM_ATTR_EXIT_REASON "exit-reason" # define XML_RSC_OP_LAST_CHANGE "last-rc-change" # define XML_RSC_OP_LAST_RUN "last-run" // deprecated since 2.0.3 # define XML_RSC_OP_T_EXEC "exec-time" # define XML_RSC_OP_T_QUEUE "queue-time" # define XML_LRM_ATTR_MIGRATE_SOURCE "migrate_source" # define XML_LRM_ATTR_MIGRATE_TARGET "migrate_target" # define XML_TAG_GRAPH "transition_graph" # define XML_GRAPH_TAG_RSC_OP "rsc_op" # define XML_GRAPH_TAG_PSEUDO_EVENT "pseudo_event" # define XML_GRAPH_TAG_CRM_EVENT "crm_event" # define XML_GRAPH_TAG_DOWNED "downed" # define XML_GRAPH_TAG_MAINTENANCE "maintenance" # define XML_TAG_RULE "rule" # define XML_RULE_ATTR_SCORE "score" # define XML_RULE_ATTR_SCORE_ATTRIBUTE "score-attribute" # define XML_RULE_ATTR_ROLE "role" # define XML_RULE_ATTR_BOOLEAN_OP "boolean-op" # define XML_TAG_EXPRESSION "expression" # define XML_EXPR_ATTR_ATTRIBUTE "attribute" # define XML_EXPR_ATTR_OPERATION "operation" # define XML_EXPR_ATTR_VALUE "value" # define XML_EXPR_ATTR_TYPE "type" # define XML_EXPR_ATTR_VALUE_SOURCE "value-source" # define XML_CONS_TAG_RSC_DEPEND "rsc_colocation" # define XML_CONS_TAG_RSC_ORDER "rsc_order" # define XML_CONS_TAG_RSC_LOCATION "rsc_location" # define XML_CONS_TAG_RSC_TICKET "rsc_ticket" # define XML_CONS_TAG_RSC_SET "resource_set" # define XML_CONS_ATTR_SYMMETRICAL "symmetrical" # define XML_LOCATION_ATTR_DISCOVERY "resource-discovery" # define XML_COLOC_ATTR_SOURCE "rsc" # define XML_COLOC_ATTR_SOURCE_ROLE "rsc-role" # define XML_COLOC_ATTR_TARGET "with-rsc" # define XML_COLOC_ATTR_TARGET_ROLE "with-rsc-role" # define XML_COLOC_ATTR_NODE_ATTR "node-attribute" # define XML_COLOC_ATTR_SOURCE_INSTANCE "rsc-instance" # define XML_COLOC_ATTR_TARGET_INSTANCE "with-rsc-instance" # define XML_LOC_ATTR_SOURCE "rsc" # define XML_LOC_ATTR_SOURCE_PATTERN "rsc-pattern" # define XML_ORDER_ATTR_FIRST "first" # define XML_ORDER_ATTR_THEN "then" # define XML_ORDER_ATTR_FIRST_ACTION "first-action" # define XML_ORDER_ATTR_THEN_ACTION "then-action" # define XML_ORDER_ATTR_FIRST_INSTANCE "first-instance" # define XML_ORDER_ATTR_THEN_INSTANCE "then-instance" # define XML_ORDER_ATTR_KIND "kind" # define XML_TICKET_ATTR_TICKET "ticket" # define XML_TICKET_ATTR_LOSS_POLICY "loss-policy" # define XML_NVPAIR_ATTR_NAME "name" # define XML_NVPAIR_ATTR_VALUE "value" # define XML_NODE_ATTR_RSC_DISCOVERY "resource-discovery-enabled" # define XML_CONFIG_ATTR_DC_DEADTIME "dc-deadtime" # define XML_CONFIG_ATTR_ELECTION_FAIL "election-timeout" # define XML_CONFIG_ATTR_FORCE_QUIT "shutdown-escalation" # define XML_CONFIG_ATTR_RECHECK "cluster-recheck-interval" # define XML_CONFIG_ATTR_FENCE_REACTION "fence-reaction" # define XML_CONFIG_ATTR_SHUTDOWN_LOCK "shutdown-lock" # define XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT "shutdown-lock-limit" # define XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY "priority-fencing-delay" # define XML_ALERT_ATTR_PATH "path" # define XML_ALERT_ATTR_TIMEOUT "timeout" # define XML_ALERT_ATTR_TSTAMP_FORMAT "timestamp-format" # define XML_ALERT_ATTR_REC_VALUE "value" # define XML_CIB_TAG_GENERATION_TUPPLE "generation_tuple" # define XML_ATTR_TRANSITION_MAGIC "transition-magic" # define XML_ATTR_TRANSITION_KEY "transition-key" # define XML_ATTR_TE_NOWAIT "op_no_wait" # define XML_ATTR_TE_TARGET_RC "op_target_rc" # define XML_TAG_TRANSIENT_NODEATTRS "transient_attributes" # define XML_TAG_DIFF_ADDED "diff-added" # define XML_TAG_DIFF_REMOVED "diff-removed" # define XML_ACL_TAG_USER "acl_target" # define XML_ACL_TAG_USERv1 "acl_user" # define XML_ACL_TAG_GROUP "acl_group" # define XML_ACL_TAG_ROLE "acl_role" # define XML_ACL_TAG_PERMISSION "acl_permission" # define XML_ACL_TAG_ROLE_REF "role" # define XML_ACL_TAG_ROLE_REFv1 "role_ref" # define XML_ACL_ATTR_KIND "kind" # define XML_ACL_TAG_READ "read" # define XML_ACL_TAG_WRITE "write" # define XML_ACL_TAG_DENY "deny" # define XML_ACL_ATTR_REF "reference" # define XML_ACL_ATTR_REFv1 "ref" # define XML_ACL_ATTR_TAG "object-type" # define XML_ACL_ATTR_TAGv1 "tag" # define XML_ACL_ATTR_XPATH "xpath" # define XML_ACL_ATTR_ATTRIBUTE "attribute" # define XML_CIB_TAG_TICKETS "tickets" # define XML_CIB_TAG_TICKET_STATE "ticket_state" # define XML_CIB_TAG_TAGS "tags" # define XML_CIB_TAG_TAG "tag" # define XML_CIB_TAG_OBJ_REF "obj_ref" # define XML_TAG_FENCING_TOPOLOGY "fencing-topology" # define XML_TAG_FENCING_LEVEL "fencing-level" # define XML_ATTR_STONITH_INDEX "index" # define XML_ATTR_STONITH_TARGET "target" # define XML_ATTR_STONITH_TARGET_VALUE "target-value" # define XML_ATTR_STONITH_TARGET_PATTERN "target-pattern" # define XML_ATTR_STONITH_TARGET_ATTRIBUTE "target-attribute" # define XML_ATTR_STONITH_DEVICES "devices" # define XML_TAG_DIFF "diff" # define XML_DIFF_VERSION "version" # define XML_DIFF_VSOURCE "source" # define XML_DIFF_VTARGET "target" # define XML_DIFF_CHANGE "change" # define XML_DIFF_LIST "change-list" # define XML_DIFF_ATTR "change-attr" # define XML_DIFF_RESULT "change-result" # define XML_DIFF_OP "operation" # define XML_DIFF_PATH "path" # define XML_DIFF_POSITION "position" # include # define ID(x) crm_element_value(x, XML_ATTR_ID) # define TYPE(x) crm_element_name(x) + +#ifndef PCMK__NO_COMPAT +/* Everything here is deprecated and kept only for public API backward + * compatibility. It will be moved to compatibility.h in a future release. + */ + +//! \deprecated Use PCMK_STONITH_PROVIDES instead +# define XML_RSC_ATTR_PROVIDES "provides" + +#endif + #ifdef __cplusplus } #endif #endif diff --git a/lib/fencing/st_client.c b/lib/fencing/st_client.c index 9d7d030f3d..d5784dce9e 100644 --- a/lib/fencing/st_client.c +++ b/lib/fencing/st_client.c @@ -1,2698 +1,2700 @@ /* * 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(stonith); struct stonith_action_s { /*! user defined data */ char *agent; char *action; char *victim; GHashTable *args; int timeout; int async; void *userdata; void (*done_cb) (GPid pid, gint status, const char *output, gpointer user_data); void (*fork_cb) (GPid pid, gpointer user_data); svc_action_t *svc_action; /*! internal timing information */ time_t initial_start_time; int tries; int remaining_timeout; int max_retries; /* device output data */ GPid pid; int rc; char *output; char *error; }; typedef struct stonith_private_s { char *token; crm_ipc_t *ipc; mainloop_io_t *source; GHashTable *stonith_op_callback_table; GList *notify_list; int notify_refcnt; bool notify_deletes; void (*op_callback) (stonith_t * st, stonith_callback_data_t * data); } stonith_private_t; typedef struct stonith_notify_client_s { const char *event; const char *obj_id; /* implement one day */ const char *obj_type; /* implement one day */ void (*notify) (stonith_t * st, stonith_event_t * e); bool delete; } stonith_notify_client_t; typedef struct stonith_callback_client_s { void (*callback) (stonith_t * st, stonith_callback_data_t * data); const char *id; void *user_data; gboolean only_success; gboolean allow_timeout_updates; struct timer_rec_s *timer; } stonith_callback_client_t; struct notify_blob_s { stonith_t *stonith; xmlNode *xml; }; struct timer_rec_s { int call_id; int timeout; guint ref; stonith_t *stonith; }; typedef int (*stonith_op_t) (const char *, int, const char *, xmlNode *, xmlNode *, xmlNode *, xmlNode **, xmlNode **); bool stonith_dispatch(stonith_t * st); xmlNode *stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options); static int stonith_send_command(stonith_t *stonith, const char *op, xmlNode *data, xmlNode **output_data, int call_options, int timeout); static void stonith_connection_destroy(gpointer user_data); static void stonith_send_notification(gpointer data, gpointer user_data); static int internal_stonith_action_execute(stonith_action_t * action); static void log_action(stonith_action_t *action, pid_t pid); /*! * \brief Get agent namespace by name * * \param[in] namespace_s Name of namespace as string * * \return Namespace as enum value */ enum stonith_namespace stonith_text2namespace(const char *namespace_s) { if (pcmk__str_eq(namespace_s, "any", pcmk__str_null_matches)) { return st_namespace_any; } else if (!strcmp(namespace_s, "redhat") || !strcmp(namespace_s, "stonith-ng")) { return st_namespace_rhcs; } else if (!strcmp(namespace_s, "internal")) { return st_namespace_internal; } else if (!strcmp(namespace_s, "heartbeat")) { return st_namespace_lha; } return st_namespace_invalid; } /*! * \brief Get agent namespace name * * \param[in] namespace Namespace as enum value * * \return Namespace name as string */ const char * stonith_namespace2text(enum stonith_namespace st_namespace) { switch (st_namespace) { case st_namespace_any: return "any"; case st_namespace_rhcs: return "stonith-ng"; case st_namespace_internal: return "internal"; case st_namespace_lha: return "heartbeat"; default: break; } return "unsupported"; } /*! * \brief Determine namespace of a fence agent * * \param[in] agent Fence agent type * \param[in] namespace_s Name of agent namespace as string, if known * * \return Namespace of specified agent, as enum value */ enum stonith_namespace stonith_get_namespace(const char *agent, const char *namespace_s) { if (pcmk__str_eq(namespace_s, "internal", pcmk__str_casei)) { return st_namespace_internal; } if (stonith__agent_is_rhcs(agent)) { return st_namespace_rhcs; } #if HAVE_STONITH_STONITH_H if (stonith__agent_is_lha(agent)) { return st_namespace_lha; } #endif crm_err("Unknown fence agent: %s", agent); return st_namespace_invalid; } static void log_action(stonith_action_t *action, pid_t pid) { if (action->output) { /* Logging the whole string confuses syslog when the string is xml */ char *prefix = crm_strdup_printf("%s[%d] stdout:", action->agent, pid); crm_log_output(LOG_TRACE, prefix, action->output); free(prefix); } if (action->error) { /* Logging the whole string confuses syslog when the string is xml */ char *prefix = crm_strdup_printf("%s[%d] stderr:", action->agent, pid); crm_log_output(LOG_WARNING, prefix, action->error); free(prefix); } } /* when cycling through the list we don't want to delete items so just mark them and when we know nobody is using the list loop over it to remove the marked items */ static void foreach_notify_entry (stonith_private_t *private, GFunc func, gpointer user_data) { private->notify_refcnt++; g_list_foreach(private->notify_list, func, user_data); private->notify_refcnt--; if ((private->notify_refcnt == 0) && private->notify_deletes) { GList *list_item = private->notify_list; private->notify_deletes = FALSE; while (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; GList *next = g_list_next(list_item); if (list_client->delete) { free(list_client); private->notify_list = g_list_delete_link(private->notify_list, list_item); } list_item = next; } } } static void stonith_connection_destroy(gpointer user_data) { stonith_t *stonith = user_data; stonith_private_t *native = NULL; struct notify_blob_s blob; crm_trace("Sending destroyed notification"); blob.stonith = stonith; blob.xml = create_xml_node(NULL, "notify"); native = stonith->st_private; native->ipc = NULL; native->source = NULL; free(native->token); native->token = NULL; stonith->state = stonith_disconnected; crm_xml_add(blob.xml, F_TYPE, T_STONITH_NOTIFY); crm_xml_add(blob.xml, F_SUBTYPE, T_STONITH_NOTIFY_DISCONNECT); foreach_notify_entry(native, stonith_send_notification, &blob); free_xml(blob.xml); } xmlNode * create_device_registration_xml(const char *id, enum stonith_namespace namespace, const char *agent, stonith_key_value_t *params, const char *rsc_provides) { xmlNode *data = create_xml_node(NULL, F_STONITH_DEVICE); xmlNode *args = create_xml_node(data, XML_TAG_ATTRS); #if HAVE_STONITH_STONITH_H if (namespace == st_namespace_any) { namespace = stonith_get_namespace(agent, NULL); } if (namespace == st_namespace_lha) { hash2field((gpointer) "plugin", (gpointer) agent, args); agent = "fence_legacy"; } #endif crm_xml_add(data, XML_ATTR_ID, id); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add(data, "agent", agent); if ((namespace != st_namespace_any) && (namespace != st_namespace_invalid)) { crm_xml_add(data, "namespace", stonith_namespace2text(namespace)); } if (rsc_provides) { crm_xml_add(data, "rsc_provides", rsc_provides); } for (; params; params = params->next) { hash2field((gpointer) params->key, (gpointer) params->value, args); } return data; } static int stonith_api_register_device(stonith_t * st, int call_options, const char *id, const char *namespace, const char *agent, stonith_key_value_t * params) { int rc = 0; xmlNode *data = NULL; data = create_device_registration_xml(id, stonith_text2namespace(namespace), agent, params, NULL); rc = stonith_send_command(st, STONITH_OP_DEVICE_ADD, data, NULL, call_options, 0); free_xml(data); return rc; } static int stonith_api_remove_device(stonith_t * st, int call_options, const char *name) { int rc = 0; xmlNode *data = NULL; data = create_xml_node(NULL, F_STONITH_DEVICE); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add(data, XML_ATTR_ID, name); rc = stonith_send_command(st, STONITH_OP_DEVICE_DEL, data, NULL, call_options, 0); free_xml(data); return rc; } static int stonith_api_remove_level_full(stonith_t *st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level) { int rc = 0; xmlNode *data = NULL; CRM_CHECK(node || pattern || (attr && value), return -EINVAL); data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL); crm_xml_add(data, F_STONITH_ORIGIN, __func__); if (node) { crm_xml_add(data, XML_ATTR_STONITH_TARGET, node); } else if (pattern) { crm_xml_add(data, XML_ATTR_STONITH_TARGET_PATTERN, pattern); } else { crm_xml_add(data, XML_ATTR_STONITH_TARGET_ATTRIBUTE, attr); crm_xml_add(data, XML_ATTR_STONITH_TARGET_VALUE, value); } crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level); rc = stonith_send_command(st, STONITH_OP_LEVEL_DEL, data, NULL, options, 0); free_xml(data); return rc; } static int stonith_api_remove_level(stonith_t * st, int options, const char *node, int level) { return stonith_api_remove_level_full(st, options, node, NULL, NULL, NULL, level); } /*! * \internal * \brief Create XML for fence topology level registration request * * \param[in] node If not NULL, target level by this node name * \param[in] pattern If not NULL, target by node name using this regex * \param[in] attr If not NULL, target by this node attribute * \param[in] value If not NULL, target by this node attribute value * \param[in] level Index number of level to register * \param[in] device_list List of devices in level * * \return Newly allocated XML tree on success, NULL otherwise * * \note The caller should set only one of node, pattern or attr/value. */ xmlNode * create_level_registration_xml(const char *node, const char *pattern, const char *attr, const char *value, int level, stonith_key_value_t *device_list) { size_t len = 0; char *list = NULL; xmlNode *data; CRM_CHECK(node || pattern || (attr && value), return NULL); data = create_xml_node(NULL, XML_TAG_FENCING_LEVEL); CRM_CHECK(data, return NULL); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add_int(data, XML_ATTR_ID, level); crm_xml_add_int(data, XML_ATTR_STONITH_INDEX, level); if (node) { crm_xml_add(data, XML_ATTR_STONITH_TARGET, node); } else if (pattern) { crm_xml_add(data, XML_ATTR_STONITH_TARGET_PATTERN, pattern); } else { crm_xml_add(data, XML_ATTR_STONITH_TARGET_ATTRIBUTE, attr); crm_xml_add(data, XML_ATTR_STONITH_TARGET_VALUE, value); } // cppcheck seems not to understand the abort logic behind pcmk__realloc // cppcheck-suppress memleak for (; device_list; device_list = device_list->next) { pcmk__add_separated_word(&list, &len, device_list->value, ","); } crm_xml_add(data, XML_ATTR_STONITH_DEVICES, list); free(list); return data; } static int stonith_api_register_level_full(stonith_t * st, int options, const char *node, const char *pattern, const char *attr, const char *value, int level, stonith_key_value_t *device_list) { int rc = 0; xmlNode *data = create_level_registration_xml(node, pattern, attr, value, level, device_list); CRM_CHECK(data != NULL, return -EINVAL); rc = stonith_send_command(st, STONITH_OP_LEVEL_ADD, data, NULL, options, 0); free_xml(data); return rc; } static int stonith_api_register_level(stonith_t * st, int options, const char *node, int level, stonith_key_value_t * device_list) { return stonith_api_register_level_full(st, options, node, NULL, NULL, NULL, level, device_list); } static void append_arg(const char *key, const char *value, GHashTable **args) { CRM_CHECK(key != NULL, return); CRM_CHECK(value != NULL, return); CRM_CHECK(args != NULL, return); if (strstr(key, "pcmk_")) { return; } else if (strstr(key, CRM_META)) { return; } else if (pcmk__str_eq(key, "crm_feature_set", pcmk__str_casei)) { return; } if (!*args) { *args = crm_str_table_new(); } CRM_CHECK(*args != NULL, return); crm_trace("Appending: %s=%s", key, value); g_hash_table_replace(*args, strdup(key), strdup(value)); } static void append_config_arg(gpointer key, gpointer value, gpointer user_data) { /* The fencer will filter action out when it registers the device, * but ignore it here just in case any other library callers * fail to do so. */ if (!pcmk__str_eq(key, STONITH_ATTR_ACTION_OP, pcmk__str_casei)) { append_arg(key, value, user_data); return; } } static GHashTable * make_args(const char *agent, const char *action, const char *victim, uint32_t victim_nodeid, GHashTable * device_args, GHashTable * port_map, const char *host_arg) { char buffer[512]; GHashTable *arg_list = NULL; const char *value = NULL; CRM_CHECK(action != NULL, return NULL); snprintf(buffer, sizeof(buffer), "pcmk_%s_action", action); if (device_args) { value = g_hash_table_lookup(device_args, buffer); } if (value) { crm_debug("Substituting action '%s' for requested operation '%s'", value, action); action = value; } append_arg(STONITH_ATTR_ACTION_OP, action, &arg_list); if (victim && device_args) { const char *alias = victim; - const char *param = g_hash_table_lookup(device_args, STONITH_ATTR_HOSTARG); + const char *param = g_hash_table_lookup(device_args, + PCMK_STONITH_HOST_ARGUMENT); if (port_map && g_hash_table_lookup(port_map, victim)) { alias = g_hash_table_lookup(port_map, victim); } /* Always supply the node's name, too: * https://github.com/ClusterLabs/fence-agents/blob/master/doc/FenceAgentAPI.md */ append_arg("nodename", victim, &arg_list); if (victim_nodeid) { char nodeid_str[33] = { 0, }; if (snprintf(nodeid_str, 33, "%u", (unsigned int)victim_nodeid)) { crm_info("For stonith action (%s) for victim %s, adding nodeid (%s) to parameters", action, victim, nodeid_str); append_arg("nodeid", nodeid_str, &arg_list); } } /* Check if we need to supply the victim in any other form */ if(pcmk__str_eq(agent, "fence_legacy", pcmk__str_casei)) { value = agent; } else if (param == NULL) { // By default, `port` is added if (host_arg == NULL) { param = "port"; } else { param = host_arg; } value = g_hash_table_lookup(device_args, param); } else if (pcmk__str_eq(param, "none", pcmk__str_casei)) { value = param; /* Nothing more to do */ } else { value = g_hash_table_lookup(device_args, param); } /* Don't overwrite explictly set values for $param */ if (pcmk__str_eq(value, "dynamic", pcmk__str_null_matches | pcmk__str_casei)) { crm_debug("Performing '%s' action targeting '%s' as '%s=%s'", action, victim, param, alias); append_arg(param, alias, &arg_list); } } if (device_args) { g_hash_table_foreach(device_args, append_config_arg, &arg_list); } return arg_list; } /*! * \internal * \brief Free all memory used by a stonith action * * \param[in,out] action Action to free */ void stonith__destroy_action(stonith_action_t *action) { if (action) { free(action->agent); if (action->args) { g_hash_table_destroy(action->args); } free(action->action); free(action->victim); if (action->svc_action) { services_action_free(action->svc_action); } free(action->output); free(action->error); free(action); } } /*! * \internal * \brief Get the result of an executed stonith action * * \param[in,out] action Executed action * \param[out] rc Where to store result code (or NULL) * \param[out] output Where to store standard output (or NULL) * \param[out] error_output Where to store standard error output (or NULL) * * \note If output or error_output is not NULL, the caller is responsible for * freeing the memory. */ void stonith__action_result(stonith_action_t *action, int *rc, char **output, char **error_output) { if (rc) { *rc = pcmk_ok; } if (output) { *output = NULL; } if (error_output) { *error_output = NULL; } if (action != NULL) { if (rc) { *rc = action->rc; } if (output && action->output) { *output = action->output; action->output = NULL; // hand off memory management to caller } if (error_output && action->error) { *error_output = action->error; action->error = NULL; // hand off memory management to caller } } } #define FAILURE_MAX_RETRIES 2 stonith_action_t * stonith_action_create(const char *agent, const char *_action, const char *victim, uint32_t victim_nodeid, int timeout, GHashTable * device_args, GHashTable * port_map, const char *host_arg) { stonith_action_t *action; action = calloc(1, sizeof(stonith_action_t)); action->args = make_args(agent, _action, victim, victim_nodeid, device_args, port_map, host_arg); crm_debug("Preparing '%s' action for %s using agent %s", _action, (victim? victim : "no target"), agent); action->agent = strdup(agent); action->action = strdup(_action); if (victim) { action->victim = strdup(victim); } action->timeout = action->remaining_timeout = timeout; action->max_retries = FAILURE_MAX_RETRIES; if (device_args) { char buffer[512]; const char *value = NULL; snprintf(buffer, sizeof(buffer), "pcmk_%s_retries", _action); value = g_hash_table_lookup(device_args, buffer); if (value) { action->max_retries = atoi(value); } } return action; } static gboolean update_remaining_timeout(stonith_action_t * action) { int diff = time(NULL) - action->initial_start_time; if (action->tries >= action->max_retries) { crm_info("Attempted to execute agent %s (%s) the maximum number of times (%d) allowed", action->agent, action->action, action->max_retries); action->remaining_timeout = 0; } else if ((action->rc != -ETIME) && diff < (action->timeout * 0.7)) { /* only set remaining timeout period if there is 30% * or greater of the original timeout period left */ action->remaining_timeout = action->timeout - diff; } else { action->remaining_timeout = 0; } return action->remaining_timeout ? TRUE : FALSE; } static int svc_action_to_errno(svc_action_t *svc_action) { int rv = pcmk_ok; if (svc_action->rc > 0) { /* Try to provide a useful error code based on the fence agent's * error output. */ if (svc_action->rc == PCMK_OCF_TIMEOUT) { rv = -ETIME; } else if (svc_action->stderr_data == NULL) { rv = -ENODATA; } else if (strstr(svc_action->stderr_data, "imed out")) { /* Some agents have their own internal timeouts */ rv = -ETIME; } else if (strstr(svc_action->stderr_data, "Unrecognised action")) { rv = -EOPNOTSUPP; } else { rv = -pcmk_err_generic; } } return rv; } static void stonith_action_async_done(svc_action_t *svc_action) { stonith_action_t *action = (stonith_action_t *) svc_action->cb_data; action->rc = svc_action_to_errno(svc_action); action->output = svc_action->stdout_data; svc_action->stdout_data = NULL; action->error = svc_action->stderr_data; svc_action->stderr_data = NULL; svc_action->params = NULL; crm_debug("Child process %d performing action '%s' exited with rc %d", action->pid, action->action, svc_action->rc); log_action(action, action->pid); if (action->rc != pcmk_ok && update_remaining_timeout(action)) { int rc = internal_stonith_action_execute(action); if (rc == pcmk_ok) { return; } } if (action->done_cb) { action->done_cb(action->pid, action->rc, action->output, action->userdata); } action->svc_action = NULL; // don't remove our caller stonith__destroy_action(action); } static void stonith_action_async_forked(svc_action_t *svc_action) { stonith_action_t *action = (stonith_action_t *) svc_action->cb_data; action->pid = svc_action->pid; action->svc_action = svc_action; if (action->fork_cb) { (action->fork_cb) (svc_action->pid, action->userdata); } crm_trace("Child process %d performing action '%s' successfully forked", action->pid, action->action); } static int internal_stonith_action_execute(stonith_action_t * action) { int rc = -EPROTO; int is_retry = 0; svc_action_t *svc_action = NULL; static int stonith_sequence = 0; char *buffer = NULL; if (!action->tries) { action->initial_start_time = time(NULL); } action->tries++; if (action->tries > 1) { crm_info("Attempt %d to execute %s (%s). remaining timeout is %d", action->tries, action->agent, action->action, action->remaining_timeout); is_retry = 1; } if (action->args == NULL || action->agent == NULL) goto fail; buffer = crm_strdup_printf(RH_STONITH_DIR "/%s", basename(action->agent)); svc_action = services_action_create_generic(buffer, NULL); free(buffer); svc_action->timeout = 1000 * action->remaining_timeout; svc_action->standard = strdup(PCMK_RESOURCE_CLASS_STONITH); svc_action->id = crm_strdup_printf("%s_%s_%d", basename(action->agent), action->action, action->tries); svc_action->agent = strdup(action->agent); svc_action->sequence = stonith_sequence++; svc_action->params = action->args; svc_action->cb_data = (void *) action; svc_action->flags = pcmk__set_flags_as(__func__, __LINE__, LOG_TRACE, "Action", svc_action->id, svc_action->flags, SVC_ACTION_NON_BLOCKED, "SVC_ACTION_NON_BLOCKED"); /* keep retries from executing out of control and free previous results */ if (is_retry) { free(action->output); action->output = NULL; free(action->error); action->error = NULL; sleep(1); } if (action->async) { /* async */ if(services_action_async_fork_notify(svc_action, &stonith_action_async_done, &stonith_action_async_forked) == FALSE) { services_action_free(svc_action); svc_action = NULL; } else { rc = 0; } } else { /* sync */ if (services_action_sync(svc_action)) { rc = 0; action->rc = svc_action_to_errno(svc_action); action->output = svc_action->stdout_data; svc_action->stdout_data = NULL; action->error = svc_action->stderr_data; svc_action->stderr_data = NULL; } else { action->rc = -ECONNABORTED; rc = action->rc; } svc_action->params = NULL; services_action_free(svc_action); } fail: return rc; } /*! * \internal * \brief Kick off execution of an async stonith action * * \param[in,out] action Action to be executed * \param[in,out] userdata Datapointer to be passed to callbacks * \param[in] done Callback to notify action has failed/succeeded * \param[in] fork_callback Callback to notify successful fork of child * * \return pcmk_ok if ownership of action has been taken, -errno otherwise */ int stonith_action_execute_async(stonith_action_t * action, void *userdata, void (*done) (GPid pid, int rc, const char *output, gpointer user_data), void (*fork_cb) (GPid pid, gpointer user_data)) { if (!action) { return -EINVAL; } action->userdata = userdata; action->done_cb = done; action->fork_cb = fork_cb; action->async = 1; return internal_stonith_action_execute(action); } /*! * \internal * \brief Execute a stonith action * * \param[in,out] action Action to execute * * \return pcmk_ok on success, -errno otherwise */ int stonith__execute(stonith_action_t *action) { int rc = pcmk_ok; CRM_CHECK(action != NULL, return -EINVAL); // Keep trying until success, max retries, or timeout do { rc = internal_stonith_action_execute(action); } while ((rc != pcmk_ok) && update_remaining_timeout(action)); return rc; } static int stonith_api_device_list(stonith_t * stonith, int call_options, const char *namespace, stonith_key_value_t ** devices, int timeout) { int count = 0; enum stonith_namespace ns = stonith_text2namespace(namespace); if (devices == NULL) { crm_err("Parameter error: stonith_api_device_list"); return -EFAULT; } #if HAVE_STONITH_STONITH_H // Include Linux-HA agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_lha)) { count += stonith__list_lha_agents(devices); } #endif // Include Red Hat agents if requested if ((ns == st_namespace_any) || (ns == st_namespace_rhcs)) { count += stonith__list_rhcs_agents(devices); } return count; } static int stonith_api_device_metadata(stonith_t * stonith, int call_options, const char *agent, const char *namespace, char **output, int timeout) { /* By executing meta-data directly, we can get it from stonith_admin when * the cluster is not running, which is important for higher-level tools. */ enum stonith_namespace ns = stonith_get_namespace(agent, namespace); crm_trace("Looking up metadata for %s agent %s", stonith_namespace2text(ns), agent); switch (ns) { case st_namespace_rhcs: return stonith__rhcs_metadata(agent, timeout, output); #if HAVE_STONITH_STONITH_H case st_namespace_lha: return stonith__lha_metadata(agent, timeout, output); #endif default: crm_err("Can't get fence agent '%s' meta-data: No such agent", agent); break; } return -ENODEV; } static int stonith_api_query(stonith_t * stonith, int call_options, const char *target, stonith_key_value_t ** devices, int timeout) { int rc = 0, lpc = 0, max = 0; xmlNode *data = NULL; xmlNode *output = NULL; xmlXPathObjectPtr xpathObj = NULL; CRM_CHECK(devices != NULL, return -EINVAL); data = create_xml_node(NULL, F_STONITH_DEVICE); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add(data, F_STONITH_TARGET, target); crm_xml_add(data, F_STONITH_ACTION, "off"); rc = stonith_send_command(stonith, STONITH_OP_QUERY, data, &output, call_options, timeout); if (rc < 0) { return rc; } xpathObj = xpath_search(output, "//@agent"); if (xpathObj) { max = numXpathResults(xpathObj); for (lpc = 0; lpc < max; lpc++) { xmlNode *match = getXpathResult(xpathObj, lpc); CRM_LOG_ASSERT(match != NULL); if(match != NULL) { xmlChar *match_path = xmlGetNodePath(match); crm_info("%s[%d] = %s", "//@agent", lpc, match_path); free(match_path); *devices = stonith_key_value_add(*devices, NULL, crm_element_value(match, XML_ATTR_ID)); } } freeXpathObject(xpathObj); } free_xml(output); free_xml(data); return max; } static int stonith_api_call(stonith_t * stonith, int call_options, const char *id, const char *action, const char *victim, int timeout, xmlNode ** output) { int rc = 0; xmlNode *data = NULL; data = create_xml_node(NULL, F_STONITH_DEVICE); crm_xml_add(data, F_STONITH_ORIGIN, __func__); crm_xml_add(data, F_STONITH_DEVICE, id); crm_xml_add(data, F_STONITH_ACTION, action); crm_xml_add(data, F_STONITH_TARGET, victim); rc = stonith_send_command(stonith, STONITH_OP_EXEC, data, output, call_options, timeout); free_xml(data); return rc; } static int stonith_api_list(stonith_t * stonith, int call_options, const char *id, char **list_info, int timeout) { int rc; xmlNode *output = NULL; rc = stonith_api_call(stonith, call_options, id, "list", NULL, timeout, &output); if (output && list_info) { const char *list_str; list_str = crm_element_value(output, "st_output"); if (list_str) { *list_info = strdup(list_str); } } if (output) { free_xml(output); } return rc; } static int stonith_api_monitor(stonith_t * stonith, int call_options, const char *id, int timeout) { return stonith_api_call(stonith, call_options, id, "monitor", NULL, timeout, NULL); } static int stonith_api_status(stonith_t * stonith, int call_options, const char *id, const char *port, int timeout) { return stonith_api_call(stonith, call_options, id, "status", port, timeout, NULL); } static int stonith_api_fence_with_delay(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance, int delay) { int rc = 0; xmlNode *data = NULL; data = create_xml_node(NULL, __func__); crm_xml_add(data, F_STONITH_TARGET, node); crm_xml_add(data, F_STONITH_ACTION, action); crm_xml_add_int(data, F_STONITH_TIMEOUT, timeout); crm_xml_add_int(data, F_STONITH_TOLERANCE, tolerance); crm_xml_add_int(data, F_STONITH_DELAY, delay); rc = stonith_send_command(stonith, STONITH_OP_FENCE, data, NULL, call_options, timeout); free_xml(data); return rc; } static int stonith_api_fence(stonith_t * stonith, int call_options, const char *node, const char *action, int timeout, int tolerance) { return stonith_api_fence_with_delay(stonith, call_options, node, action, timeout, tolerance, 0); } static int stonith_api_confirm(stonith_t * stonith, int call_options, const char *target) { stonith__set_call_options(call_options, target, st_opt_manual_ack); return stonith_api_fence(stonith, call_options, target, "off", 0, 0); } static int stonith_api_history(stonith_t * stonith, int call_options, const char *node, stonith_history_t ** history, int timeout) { int rc = 0; xmlNode *data = NULL; xmlNode *output = NULL; stonith_history_t *last = NULL; *history = NULL; if (node) { data = create_xml_node(NULL, __func__); crm_xml_add(data, F_STONITH_TARGET, node); } stonith__set_call_options(call_options, node, st_opt_sync_call); rc = stonith_send_command(stonith, STONITH_OP_FENCE_HISTORY, data, &output, call_options, timeout); free_xml(data); if (rc == 0) { xmlNode *op = NULL; xmlNode *reply = get_xpath_object("//" F_STONITH_HISTORY_LIST, output, LOG_NEVER); for (op = pcmk__xml_first_child(reply); op != NULL; op = pcmk__xml_next(op)) { stonith_history_t *kvp; long long completed; kvp = calloc(1, sizeof(stonith_history_t)); kvp->target = crm_element_value_copy(op, F_STONITH_TARGET); kvp->action = crm_element_value_copy(op, F_STONITH_ACTION); kvp->origin = crm_element_value_copy(op, F_STONITH_ORIGIN); kvp->delegate = crm_element_value_copy(op, F_STONITH_DELEGATE); kvp->client = crm_element_value_copy(op, F_STONITH_CLIENTNAME); crm_element_value_ll(op, F_STONITH_DATE, &completed); kvp->completed = (time_t) completed; crm_element_value_int(op, F_STONITH_STATE, &kvp->state); if (last) { last->next = kvp; } else { *history = kvp; } last = kvp; } } free_xml(output); return rc; } void stonith_history_free(stonith_history_t *history) { stonith_history_t *hp, *hp_old; for (hp = history; hp; hp_old = hp, hp = hp->next, free(hp_old)) { free(hp->target); free(hp->action); free(hp->origin); free(hp->delegate); free(hp->client); } } static gint stonithlib_GCompareFunc(gconstpointer a, gconstpointer b) { int rc = 0; const stonith_notify_client_t *a_client = a; const stonith_notify_client_t *b_client = b; if (a_client->delete || b_client->delete) { /* make entries marked for deletion not findable */ return -1; } CRM_CHECK(a_client->event != NULL && b_client->event != NULL, return 0); rc = strcmp(a_client->event, b_client->event); if (rc == 0) { if (a_client->notify == NULL || b_client->notify == NULL) { return 0; } else if (a_client->notify == b_client->notify) { return 0; } else if (((long)a_client->notify) < ((long)b_client->notify)) { crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return -1; } crm_err("callbacks for %s are not equal: %p vs. %p", a_client->event, a_client->notify, b_client->notify); return 1; } return rc; } xmlNode * stonith_create_op(int call_id, const char *token, const char *op, xmlNode * data, int call_options) { xmlNode *op_msg = create_xml_node(NULL, "stonith_command"); CRM_CHECK(op_msg != NULL, return NULL); CRM_CHECK(token != NULL, return NULL); crm_xml_add(op_msg, F_XML_TAGNAME, "stonith_command"); crm_xml_add(op_msg, F_TYPE, T_STONITH_NG); crm_xml_add(op_msg, F_STONITH_CALLBACK_TOKEN, token); crm_xml_add(op_msg, F_STONITH_OPERATION, op); crm_xml_add_int(op_msg, F_STONITH_CALLID, call_id); crm_trace("Sending call options: %.8lx, %d", (long)call_options, call_options); crm_xml_add_int(op_msg, F_STONITH_CALLOPTS, call_options); if (data != NULL) { add_message_xml(op_msg, F_STONITH_CALLDATA, data); } return op_msg; } static void stonith_destroy_op_callback(gpointer data) { stonith_callback_client_t *blob = data; if (blob->timer && blob->timer->ref > 0) { g_source_remove(blob->timer->ref); } free(blob->timer); free(blob); } static int stonith_api_signoff(stonith_t * stonith) { stonith_private_t *native = stonith->st_private; crm_debug("Disconnecting from the fencer"); if (native->source != NULL) { /* Attached to mainloop */ mainloop_del_ipc_client(native->source); native->source = NULL; native->ipc = NULL; } else if (native->ipc) { /* Not attached to mainloop */ crm_ipc_t *ipc = native->ipc; native->ipc = NULL; crm_ipc_close(ipc); crm_ipc_destroy(ipc); } free(native->token); native->token = NULL; stonith->state = stonith_disconnected; return pcmk_ok; } static int stonith_api_del_callback(stonith_t * stonith, int call_id, bool all_callbacks) { stonith_private_t *private = stonith->st_private; if (all_callbacks) { private->op_callback = NULL; g_hash_table_destroy(private->stonith_op_callback_table); private->stonith_op_callback_table = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL, stonith_destroy_op_callback); } else if (call_id == 0) { private->op_callback = NULL; } else { g_hash_table_remove(private->stonith_op_callback_table, GINT_TO_POINTER(call_id)); } return pcmk_ok; } static void invoke_callback(stonith_t * st, int call_id, int rc, void *userdata, void (*callback) (stonith_t * st, stonith_callback_data_t * data)) { stonith_callback_data_t data = { 0, }; data.call_id = call_id; data.rc = rc; data.userdata = userdata; callback(st, &data); } static void stonith_perform_callback(stonith_t * stonith, xmlNode * msg, int call_id, int rc) { stonith_private_t *private = NULL; stonith_callback_client_t *blob = NULL; stonith_callback_client_t local_blob; CRM_CHECK(stonith != NULL, return); CRM_CHECK(stonith->st_private != NULL, return); private = stonith->st_private; local_blob.id = NULL; local_blob.callback = NULL; local_blob.user_data = NULL; local_blob.only_success = FALSE; if (msg != NULL) { crm_element_value_int(msg, F_STONITH_RC, &rc); crm_element_value_int(msg, F_STONITH_CALLID, &call_id); } CRM_CHECK(call_id > 0, crm_log_xml_err(msg, "Bad result")); blob = g_hash_table_lookup(private->stonith_op_callback_table, GINT_TO_POINTER(call_id)); if (blob != NULL) { local_blob = *blob; blob = NULL; stonith_api_del_callback(stonith, call_id, FALSE); } else { crm_trace("No callback found for call %d", call_id); local_blob.callback = NULL; } if (local_blob.callback != NULL && (rc == pcmk_ok || local_blob.only_success == FALSE)) { crm_trace("Invoking callback %s for call %d", crm_str(local_blob.id), call_id); invoke_callback(stonith, call_id, rc, local_blob.user_data, local_blob.callback); } else if (private->op_callback == NULL && rc != pcmk_ok) { crm_warn("Fencing command failed: %s", pcmk_strerror(rc)); crm_log_xml_debug(msg, "Failed fence update"); } if (private->op_callback != NULL) { crm_trace("Invoking global callback for call %d", call_id); invoke_callback(stonith, call_id, rc, NULL, private->op_callback); } crm_trace("OP callback activated."); } static gboolean stonith_async_timeout_handler(gpointer data) { struct timer_rec_s *timer = data; crm_err("Async call %d timed out after %dms", timer->call_id, timer->timeout); stonith_perform_callback(timer->stonith, NULL, timer->call_id, -ETIME); /* Always return TRUE, never remove the handler * We do that in stonith_del_callback() */ return TRUE; } static void set_callback_timeout(stonith_callback_client_t * callback, stonith_t * stonith, int call_id, int timeout) { struct timer_rec_s *async_timer = callback->timer; if (timeout <= 0) { return; } if (!async_timer) { async_timer = calloc(1, sizeof(struct timer_rec_s)); callback->timer = async_timer; } async_timer->stonith = stonith; async_timer->call_id = call_id; /* Allow a fair bit of grace to allow the server to tell us of a timeout * This is only a fallback */ async_timer->timeout = (timeout + 60) * 1000; if (async_timer->ref) { g_source_remove(async_timer->ref); } async_timer->ref = g_timeout_add(async_timer->timeout, stonith_async_timeout_handler, async_timer); } static void update_callback_timeout(int call_id, int timeout, stonith_t * st) { stonith_callback_client_t *callback = NULL; stonith_private_t *private = st->st_private; callback = g_hash_table_lookup(private->stonith_op_callback_table, GINT_TO_POINTER(call_id)); if (!callback || !callback->allow_timeout_updates) { return; } set_callback_timeout(callback, st, call_id, timeout); } static int stonith_dispatch_internal(const char *buffer, ssize_t length, gpointer userdata) { const char *type = NULL; struct notify_blob_s blob; stonith_t *st = userdata; stonith_private_t *private = NULL; CRM_ASSERT(st != NULL); private = st->st_private; blob.stonith = st; blob.xml = string2xml(buffer); if (blob.xml == NULL) { crm_warn("Received malformed message from fencer: %s", buffer); return 0; } /* do callbacks */ type = crm_element_value(blob.xml, F_TYPE); crm_trace("Activating %s callbacks...", type); if (pcmk__str_eq(type, T_STONITH_NG, pcmk__str_casei)) { stonith_perform_callback(st, blob.xml, 0, 0); } else if (pcmk__str_eq(type, T_STONITH_NOTIFY, pcmk__str_casei)) { foreach_notify_entry(private, stonith_send_notification, &blob); } else if (pcmk__str_eq(type, T_STONITH_TIMEOUT_VALUE, pcmk__str_casei)) { int call_id = 0; int timeout = 0; crm_element_value_int(blob.xml, F_STONITH_TIMEOUT, &timeout); crm_element_value_int(blob.xml, F_STONITH_CALLID, &call_id); update_callback_timeout(call_id, timeout, st); } else { crm_err("Unknown message type: %s", type); crm_log_xml_warn(blob.xml, "BadReply"); } free_xml(blob.xml); return 1; } static int stonith_api_signon(stonith_t * stonith, const char *name, int *stonith_fd) { int rc = pcmk_ok; stonith_private_t *native = NULL; const char *display_name = name? name : "client"; struct ipc_client_callbacks st_callbacks = { .dispatch = stonith_dispatch_internal, .destroy = stonith_connection_destroy }; CRM_CHECK(stonith != NULL, return -EINVAL); native = stonith->st_private; CRM_ASSERT(native != NULL); crm_debug("Attempting fencer connection by %s with%s mainloop", display_name, (stonith_fd? "out" : "")); stonith->state = stonith_connected_command; if (stonith_fd) { /* No mainloop */ native->ipc = crm_ipc_new("stonith-ng", 0); if (native->ipc && crm_ipc_connect(native->ipc)) { *stonith_fd = crm_ipc_get_fd(native->ipc); } else if (native->ipc) { crm_ipc_close(native->ipc); crm_ipc_destroy(native->ipc); native->ipc = NULL; } } else { /* With mainloop */ native->source = mainloop_add_ipc_client("stonith-ng", G_PRIORITY_MEDIUM, 0, stonith, &st_callbacks); native->ipc = mainloop_get_ipc_client(native->source); } if (native->ipc == NULL) { rc = -ENOTCONN; } else { xmlNode *reply = NULL; xmlNode *hello = create_xml_node(NULL, "stonith_command"); crm_xml_add(hello, F_TYPE, T_STONITH_NG); crm_xml_add(hello, F_STONITH_OPERATION, CRM_OP_REGISTER); crm_xml_add(hello, F_STONITH_CLIENTNAME, name); rc = crm_ipc_send(native->ipc, hello, crm_ipc_client_response, -1, &reply); if (rc < 0) { crm_debug("Couldn't register with the fencer: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); rc = -ECOMM; } else if (reply == NULL) { crm_debug("Couldn't register with the fencer: no reply"); rc = -EPROTO; } else { const char *msg_type = crm_element_value(reply, F_STONITH_OPERATION); native->token = crm_element_value_copy(reply, F_STONITH_CLIENTID); if (!pcmk__str_eq(msg_type, CRM_OP_REGISTER, pcmk__str_casei)) { crm_debug("Couldn't register with the fencer: invalid reply type '%s'", (msg_type? msg_type : "(missing)")); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else if (native->token == NULL) { crm_debug("Couldn't register with the fencer: no token in reply"); crm_log_xml_debug(reply, "Invalid fencer reply"); rc = -EPROTO; } else { #if HAVE_MSGFROMIPC_TIMEOUT stonith->call_timeout = MAX_IPC_DELAY; #endif crm_debug("Connection to fencer by %s succeeded (registration token: %s)", display_name, native->token); rc = pcmk_ok; } } free_xml(reply); free_xml(hello); } if (rc != pcmk_ok) { crm_debug("Connection attempt to fencer by %s failed: %s " CRM_XS " rc=%d", display_name, pcmk_strerror(rc), rc); stonith->cmds->disconnect(stonith); } return rc; } static int stonith_set_notification(stonith_t * stonith, const char *callback, int enabled) { int rc = pcmk_ok; xmlNode *notify_msg = create_xml_node(NULL, __func__); stonith_private_t *native = stonith->st_private; if (stonith->state != stonith_disconnected) { crm_xml_add(notify_msg, F_STONITH_OPERATION, T_STONITH_NOTIFY); if (enabled) { crm_xml_add(notify_msg, F_STONITH_NOTIFY_ACTIVATE, callback); } else { crm_xml_add(notify_msg, F_STONITH_NOTIFY_DEACTIVATE, callback); } rc = crm_ipc_send(native->ipc, notify_msg, crm_ipc_client_response, -1, NULL); if (rc < 0) { crm_perror(LOG_DEBUG, "Couldn't register for fencing notifications: %d", rc); rc = -ECOMM; } else { rc = pcmk_ok; } } free_xml(notify_msg); return rc; } static int stonith_api_add_notification(stonith_t * stonith, const char *event, void (*callback) (stonith_t * stonith, stonith_event_t * e)) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = NULL; private = stonith->st_private; crm_trace("Adding callback for %s events (%d)", event, g_list_length(private->notify_list)); new_client = calloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = callback; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); if (list_item != NULL) { crm_warn("Callback already present"); free(new_client); return -ENOTUNIQ; } else { private->notify_list = g_list_append(private->notify_list, new_client); stonith_set_notification(stonith, event, 1); crm_trace("Callback added (%d)", g_list_length(private->notify_list)); } return pcmk_ok; } static int stonith_api_del_notification(stonith_t * stonith, const char *event) { GList *list_item = NULL; stonith_notify_client_t *new_client = NULL; stonith_private_t *private = NULL; crm_debug("Removing callback for %s events", event); private = stonith->st_private; new_client = calloc(1, sizeof(stonith_notify_client_t)); new_client->event = event; new_client->notify = NULL; list_item = g_list_find_custom(private->notify_list, new_client, stonithlib_GCompareFunc); stonith_set_notification(stonith, event, 0); if (list_item != NULL) { stonith_notify_client_t *list_client = list_item->data; if (private->notify_refcnt) { list_client->delete = TRUE; private->notify_deletes = TRUE; } else { private->notify_list = g_list_remove(private->notify_list, list_client); free(list_client); } crm_trace("Removed callback"); } else { crm_trace("Callback not present"); } free(new_client); return pcmk_ok; } static int stonith_api_add_callback(stonith_t * stonith, int call_id, int timeout, int options, void *user_data, const char *callback_name, void (*callback) (stonith_t * st, stonith_callback_data_t * data)) { stonith_callback_client_t *blob = NULL; stonith_private_t *private = NULL; CRM_CHECK(stonith != NULL, return -EINVAL); CRM_CHECK(stonith->st_private != NULL, return -EINVAL); private = stonith->st_private; if (call_id == 0) { private->op_callback = callback; } else if (call_id < 0) { if (!(options & st_opt_report_only_success)) { crm_trace("Call failed, calling %s: %s", callback_name, pcmk_strerror(call_id)); invoke_callback(stonith, call_id, call_id, user_data, callback); } else { crm_warn("Fencer call failed: %s", pcmk_strerror(call_id)); } return FALSE; } blob = calloc(1, sizeof(stonith_callback_client_t)); blob->id = callback_name; blob->only_success = (options & st_opt_report_only_success) ? TRUE : FALSE; blob->user_data = user_data; blob->callback = callback; blob->allow_timeout_updates = (options & st_opt_timeout_updates) ? TRUE : FALSE; if (timeout > 0) { set_callback_timeout(blob, stonith, call_id, timeout); } g_hash_table_insert(private->stonith_op_callback_table, GINT_TO_POINTER(call_id), blob); crm_trace("Added callback to %s for call %d", callback_name, call_id); return TRUE; } static void stonith_dump_pending_op(gpointer key, gpointer value, gpointer user_data) { int call = GPOINTER_TO_INT(key); stonith_callback_client_t *blob = value; crm_debug("Call %d (%s): pending", call, crm_str(blob->id)); } void stonith_dump_pending_callbacks(stonith_t * stonith) { stonith_private_t *private = stonith->st_private; if (private->stonith_op_callback_table == NULL) { return; } return g_hash_table_foreach(private->stonith_op_callback_table, stonith_dump_pending_op, NULL); } /* */ static stonith_event_t * xml_to_event(xmlNode * msg) { stonith_event_t *event = calloc(1, sizeof(stonith_event_t)); const char *ntype = crm_element_value(msg, F_SUBTYPE); char *data_addr = crm_strdup_printf("//%s", ntype); xmlNode *data = get_xpath_object(data_addr, msg, LOG_DEBUG); crm_log_xml_trace(msg, "stonith_notify"); crm_element_value_int(msg, F_STONITH_RC, &(event->result)); if (pcmk__str_eq(ntype, T_STONITH_NOTIFY_FENCE, pcmk__str_casei)) { event->operation = crm_element_value_copy(msg, F_STONITH_OPERATION); if (data) { event->origin = crm_element_value_copy(data, F_STONITH_ORIGIN); event->action = crm_element_value_copy(data, F_STONITH_ACTION); event->target = crm_element_value_copy(data, F_STONITH_TARGET); event->executioner = crm_element_value_copy(data, F_STONITH_DELEGATE); event->id = crm_element_value_copy(data, F_STONITH_REMOTE_OP_ID); event->client_origin = crm_element_value_copy(data, F_STONITH_CLIENTNAME); event->device = crm_element_value_copy(data, F_STONITH_DEVICE); } else { crm_err("No data for %s event", ntype); crm_log_xml_notice(msg, "BadEvent"); } } free(data_addr); return event; } static void event_free(stonith_event_t * event) { free(event->id); free(event->type); free(event->message); free(event->operation); free(event->origin); free(event->action); free(event->target); free(event->executioner); free(event->device); free(event->client_origin); free(event); } static void stonith_send_notification(gpointer data, gpointer user_data) { struct notify_blob_s *blob = user_data; stonith_notify_client_t *entry = data; stonith_event_t *st_event = NULL; const char *event = NULL; if (blob->xml == NULL) { crm_warn("Skipping callback - NULL message"); return; } event = crm_element_value(blob->xml, F_SUBTYPE); if (entry == NULL) { crm_warn("Skipping callback - NULL callback client"); return; } else if (entry->delete) { crm_trace("Skipping callback - marked for deletion"); return; } else if (entry->notify == NULL) { crm_warn("Skipping callback - NULL callback"); return; } else if (!pcmk__str_eq(entry->event, event, pcmk__str_casei)) { crm_trace("Skipping callback - event mismatch %p/%s vs. %s", entry, entry->event, event); return; } st_event = xml_to_event(blob->xml); crm_trace("Invoking callback for %p/%s event...", entry, event); entry->notify(blob->stonith, st_event); crm_trace("Callback invoked..."); event_free(st_event); } /*! * \internal * \brief Create and send an API request * * \param[in] stonith Stonith connection * \param[in] op API operation to request * \param[in] data Data to attach to request * \param[out] output_data If not NULL, will be set to reply if synchronous * \param[in] call_options Bitmask of stonith_call_options to use * \param[in] timeout Error if not completed within this many seconds * * \return pcmk_ok (for synchronous requests) or positive call ID * (for asynchronous requests) on success, -errno otherwise */ static int stonith_send_command(stonith_t * stonith, const char *op, xmlNode * data, xmlNode ** output_data, int call_options, int timeout) { int rc = 0; int reply_id = -1; xmlNode *op_msg = NULL; xmlNode *op_reply = NULL; stonith_private_t *native = NULL; CRM_ASSERT(stonith && stonith->st_private && op); native = stonith->st_private; if (output_data != NULL) { *output_data = NULL; } if ((stonith->state == stonith_disconnected) || (native->token == NULL)) { return -ENOTCONN; } /* Increment the call ID, which must be positive to avoid conflicting with * error codes. This shouldn't be a problem unless the client mucked with * it or the counter wrapped around. */ stonith->call_id++; if (stonith->call_id < 1) { stonith->call_id = 1; } op_msg = stonith_create_op(stonith->call_id, native->token, op, data, call_options); if (op_msg == NULL) { return -EINVAL; } crm_xml_add_int(op_msg, F_STONITH_TIMEOUT, timeout); crm_trace("Sending %s message to fencer with timeout %ds", op, timeout); if (data) { const char *delay_s = crm_element_value(data, F_STONITH_DELAY); if (delay_s) { crm_xml_add(op_msg, F_STONITH_DELAY, delay_s); } } { enum crm_ipc_flags ipc_flags = crm_ipc_flags_none; if (call_options & st_opt_sync_call) { pcmk__set_ipc_flags(ipc_flags, "stonith command", crm_ipc_client_response); } rc = crm_ipc_send(native->ipc, op_msg, ipc_flags, 1000 * (timeout + 60), &op_reply); } free_xml(op_msg); if (rc < 0) { crm_perror(LOG_ERR, "Couldn't perform %s operation (timeout=%ds): %d", op, timeout, rc); rc = -ECOMM; goto done; } crm_log_xml_trace(op_reply, "Reply"); if (!(call_options & st_opt_sync_call)) { crm_trace("Async call %d, returning", stonith->call_id); free_xml(op_reply); return stonith->call_id; } rc = pcmk_ok; crm_element_value_int(op_reply, F_STONITH_CALLID, &reply_id); if (reply_id == stonith->call_id) { crm_trace("Synchronous reply %d received", reply_id); if (crm_element_value_int(op_reply, F_STONITH_RC, &rc) != 0) { rc = -ENOMSG; } if ((call_options & st_opt_discard_reply) || output_data == NULL) { crm_trace("Discarding reply"); } else { *output_data = op_reply; op_reply = NULL; /* Prevent subsequent free */ } } else if (reply_id <= 0) { crm_err("Received bad reply: No id set"); crm_log_xml_err(op_reply, "Bad reply"); free_xml(op_reply); rc = -ENOMSG; } else { crm_err("Received bad reply: %d (wanted %d)", reply_id, stonith->call_id); crm_log_xml_err(op_reply, "Old reply"); free_xml(op_reply); rc = -ENOMSG; } done: if (crm_ipc_connected(native->ipc) == FALSE) { crm_err("Fencer disconnected"); free(native->token); native->token = NULL; stonith->state = stonith_disconnected; } free_xml(op_reply); return rc; } /* Not used with mainloop */ bool stonith_dispatch(stonith_t * st) { gboolean stay_connected = TRUE; stonith_private_t *private = NULL; CRM_ASSERT(st != NULL); private = st->st_private; while (crm_ipc_ready(private->ipc)) { if (crm_ipc_read(private->ipc) > 0) { const char *msg = crm_ipc_buffer(private->ipc); stonith_dispatch_internal(msg, strlen(msg), st); } if (crm_ipc_connected(private->ipc) == FALSE) { crm_err("Connection closed"); stay_connected = FALSE; } } return stay_connected; } static int stonith_api_free(stonith_t * stonith) { int rc = pcmk_ok; crm_trace("Destroying %p", stonith); if (stonith->state != stonith_disconnected) { crm_trace("Disconnecting %p first", stonith); rc = stonith->cmds->disconnect(stonith); } if (stonith->state == stonith_disconnected) { stonith_private_t *private = stonith->st_private; crm_trace("Removing %d callbacks", g_hash_table_size(private->stonith_op_callback_table)); g_hash_table_destroy(private->stonith_op_callback_table); crm_trace("Destroying %d notification clients", g_list_length(private->notify_list)); g_list_free_full(private->notify_list, free); free(stonith->st_private); free(stonith->cmds); free(stonith); } else { crm_err("Not free'ing active connection: %s (%d)", pcmk_strerror(rc), rc); } return rc; } void stonith_api_delete(stonith_t * stonith) { crm_trace("Destroying %p", stonith); if(stonith) { stonith->cmds->free(stonith); } } static int stonith_api_validate(stonith_t *st, int call_options, const char *rsc_id, const char *namespace_s, const char *agent, stonith_key_value_t *params, int timeout, char **output, char **error_output) { /* Validation should be done directly via the agent, so we can get it from * stonith_admin when the cluster is not running, which is important for * higher-level tools. */ int rc = pcmk_ok; /* Use a dummy node name in case the agent requires a target. We assume the * actual target doesn't matter for validation purposes (if in practice, * that is incorrect, we will need to allow the caller to pass the target). */ const char *target = "node1"; const char *host_arg = NULL; GHashTable *params_table = crm_str_table_new(); // Convert parameter list to a hash table for (; params; params = params->next) { - if (pcmk__str_eq(params->key, STONITH_ATTR_HOSTARG, pcmk__str_casei)) { + if (pcmk__str_eq(params->key, PCMK_STONITH_HOST_ARGUMENT, + pcmk__str_casei)) { host_arg = params->value; } // Strip out Pacemaker-implemented parameters if (!pcmk__starts_with(params->key, "pcmk_") - && strcmp(params->key, "provides") - && strcmp(params->key, "stonith-timeout")) { + && strcmp(params->key, PCMK_STONITH_PROVIDES) + && strcmp(params->key, PCMK_STONITH_STONITH_TIMEOUT)) { g_hash_table_insert(params_table, strdup(params->key), strdup(params->value)); } } #if SUPPORT_CIBSECRETS rc = pcmk__substitute_secrets(rsc_id, params_table); if (rc != pcmk_rc_ok) { crm_warn("Could not replace secret parameters for validation of %s: %s", agent, pcmk_rc_str(rc)); // rc is standard return value, don't return it in this function } #endif if (output) { *output = NULL; } if (error_output) { *error_output = NULL; } switch (stonith_get_namespace(agent, namespace_s)) { case st_namespace_rhcs: rc = stonith__rhcs_validate(st, call_options, target, agent, params_table, host_arg, timeout, output, error_output); break; #if HAVE_STONITH_STONITH_H case st_namespace_lha: rc = stonith__lha_validate(st, call_options, target, agent, params_table, timeout, output, error_output); break; #endif default: rc = -EINVAL; errno = EINVAL; crm_perror(LOG_ERR, "Agent %s not found or does not support validation", agent); break; } g_hash_table_destroy(params_table); return rc; } stonith_t * stonith_api_new(void) { stonith_t *new_stonith = NULL; stonith_private_t *private = NULL; new_stonith = calloc(1, sizeof(stonith_t)); if (new_stonith == NULL) { return NULL; } private = calloc(1, sizeof(stonith_private_t)); if (private == NULL) { free(new_stonith); return NULL; } new_stonith->st_private = private; private->stonith_op_callback_table = g_hash_table_new_full(g_direct_hash, g_direct_equal, NULL, stonith_destroy_op_callback); private->notify_list = NULL; private->notify_refcnt = 0; private->notify_deletes = FALSE; new_stonith->call_id = 1; new_stonith->state = stonith_disconnected; new_stonith->cmds = calloc(1, sizeof(stonith_api_operations_t)); if (new_stonith->cmds == NULL) { free(new_stonith->st_private); free(new_stonith); return NULL; } /* *INDENT-OFF* */ new_stonith->cmds->free = stonith_api_free; new_stonith->cmds->connect = stonith_api_signon; new_stonith->cmds->disconnect = stonith_api_signoff; new_stonith->cmds->list = stonith_api_list; new_stonith->cmds->monitor = stonith_api_monitor; new_stonith->cmds->status = stonith_api_status; new_stonith->cmds->fence = stonith_api_fence; new_stonith->cmds->fence_with_delay = stonith_api_fence_with_delay; new_stonith->cmds->confirm = stonith_api_confirm; new_stonith->cmds->history = stonith_api_history; new_stonith->cmds->list_agents = stonith_api_device_list; new_stonith->cmds->metadata = stonith_api_device_metadata; new_stonith->cmds->query = stonith_api_query; new_stonith->cmds->remove_device = stonith_api_remove_device; new_stonith->cmds->register_device = stonith_api_register_device; new_stonith->cmds->remove_level = stonith_api_remove_level; new_stonith->cmds->remove_level_full = stonith_api_remove_level_full; new_stonith->cmds->register_level = stonith_api_register_level; new_stonith->cmds->register_level_full = stonith_api_register_level_full; new_stonith->cmds->remove_callback = stonith_api_del_callback; new_stonith->cmds->register_callback = stonith_api_add_callback; new_stonith->cmds->remove_notification = stonith_api_del_notification; new_stonith->cmds->register_notification = stonith_api_add_notification; new_stonith->cmds->validate = stonith_api_validate; /* *INDENT-ON* */ return new_stonith; } /*! * \brief Make a blocking connection attempt to the fencer * * \param[in,out] st Fencer API object * \param[in] name Client name to use with fencer * \param[in] max_attempts Return error if this many attempts fail * * \return pcmk_ok on success, result of last attempt otherwise */ int stonith_api_connect_retry(stonith_t *st, const char *name, int max_attempts) { int rc = -EINVAL; // if max_attempts is not positive for (int attempt = 1; attempt <= max_attempts; attempt++) { rc = st->cmds->connect(st, name, NULL); if (rc == pcmk_ok) { return pcmk_ok; } else if (attempt < max_attempts) { crm_notice("Fencer connection attempt %d of %d failed (retrying in 2s): %s " CRM_XS " rc=%d", attempt, max_attempts, pcmk_strerror(rc), rc); sleep(2); } } crm_notice("Could not connect to fencer: %s " CRM_XS " rc=%d", pcmk_strerror(rc), rc); return rc; } stonith_key_value_t * stonith_key_value_add(stonith_key_value_t * head, const char *key, const char *value) { stonith_key_value_t *p, *end; p = calloc(1, sizeof(stonith_key_value_t)); if (key) { p->key = strdup(key); } if (value) { p->value = strdup(value); } end = head; while (end && end->next) { end = end->next; } if (end) { end->next = p; } else { head = p; } return head; } void stonith_key_value_freeall(stonith_key_value_t * head, int keys, int values) { stonith_key_value_t *p; while (head) { p = head->next; if (keys) { free(head->key); } if (values) { free(head->value); } free(head); head = p; } } #define api_log_open() openlog("stonith-api", LOG_CONS | LOG_NDELAY | LOG_PID, LOG_DAEMON) #define api_log(level, fmt, args...) syslog(level, "%s: "fmt, __func__, args) int stonith_api_kick(uint32_t nodeid, const char *uname, int timeout, bool off) { int rc = pcmk_ok; stonith_t *st = stonith_api_new(); const char *action = off? "off" : "reboot"; api_log_open(); if (st == NULL) { api_log(LOG_ERR, "API initialization failed, could not kick (%s) node %u/%s", action, nodeid, uname); return -EPROTO; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_ERR, "Connection failed, could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { char *name = (uname == NULL)? crm_itoa(nodeid) : strdup(uname); int opts = 0; stonith__set_call_options(opts, name, st_opt_sync_call|st_opt_allow_suicide); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->fence(st, opts, name, action, timeout, 0); free(name); if (rc != pcmk_ok) { api_log(LOG_ERR, "Could not kick (%s) node %u/%s : %s (%d)", action, nodeid, uname, pcmk_strerror(rc), rc); } else { api_log(LOG_NOTICE, "Node %u/%s kicked: %s", nodeid, uname, action); } } stonith_api_delete(st); return rc; } time_t stonith_api_time(uint32_t nodeid, const char *uname, bool in_progress) { int rc = pcmk_ok; time_t when = 0; stonith_t *st = stonith_api_new(); stonith_history_t *history = NULL, *hp = NULL; if (st == NULL) { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: " "API initialization failed", nodeid, uname); return when; } rc = st->cmds->connect(st, "stonith-api", NULL); if (rc != pcmk_ok) { api_log(LOG_NOTICE, "Connection failed: %s (%d)", pcmk_strerror(rc), rc); } else { int entries = 0; int progress = 0; int completed = 0; int opts = 0; char *name = (uname == NULL)? crm_itoa(nodeid) : strdup(uname); stonith__set_call_options(opts, name, st_opt_sync_call); if ((uname == NULL) && (nodeid > 0)) { stonith__set_call_options(opts, name, st_opt_cs_nodeid); } rc = st->cmds->history(st, opts, name, &history, 120); free(name); for (hp = history; hp; hp = hp->next) { entries++; if (in_progress) { progress++; if (hp->state != st_done && hp->state != st_failed) { when = time(NULL); } } else if (hp->state == st_done) { completed++; if (hp->completed > when) { when = hp->completed; } } } stonith_history_free(history); if(rc == pcmk_ok) { api_log(LOG_INFO, "Found %d entries for %u/%s: %d in progress, %d completed", entries, nodeid, uname, progress, completed); } else { api_log(LOG_ERR, "Could not retrieve fence history for %u/%s: %s (%d)", nodeid, uname, pcmk_strerror(rc), rc); } } stonith_api_delete(st); if(when) { api_log(LOG_INFO, "Node %u/%s last kicked at: %ld", nodeid, uname, (long int)when); } return when; } bool stonith_agent_exists(const char *agent, int timeout) { stonith_t *st = NULL; stonith_key_value_t *devices = NULL; stonith_key_value_t *dIter = NULL; bool rc = FALSE; if (agent == NULL) { return rc; } st = stonith_api_new(); if (st == NULL) { crm_err("Could not list fence agents: API memory allocation failed"); return FALSE; } st->cmds->list_agents(st, st_opt_sync_call, NULL, &devices, timeout == 0 ? 120 : timeout); for (dIter = devices; dIter != NULL; dIter = dIter->next) { if (pcmk__str_eq(dIter->value, agent, pcmk__str_none)) { rc = TRUE; break; } } stonith_key_value_freeall(devices, 1, 1); stonith_api_delete(st); return rc; } const char * stonith_action_str(const char *action) { if (action == NULL) { return "fencing"; } else if (!strcmp(action, "on")) { return "unfencing"; } else if (!strcmp(action, "off")) { return "turning off"; } else { return action; } } /*! * \internal * \brief Parse a target name from one line of a target list string * * \param[in] line One line of a target list string * \parma[in] len String length of line * \param[in,out] output List to add newly allocated target name to */ static void parse_list_line(const char *line, int len, GList **output) { size_t i = 0; size_t entry_start = 0; /* Skip complaints about additional parameters device doesn't understand * * @TODO Document or eliminate the implied restriction of target names */ if (strstr(line, "invalid") || strstr(line, "variable")) { crm_debug("Skipping list output line: %s", line); return; } // Process line content, character by character for (i = 0; i <= len; i++) { if (isspace(line[i]) || (line[i] == ',') || (line[i] == ';') || (line[i] == '\0')) { // We've found a separator (i.e. the end of an entry) int rc = 0; char *entry = NULL; if (i == entry_start) { // Skip leading and sequential separators entry_start = i + 1; continue; } entry = calloc(i - entry_start + 1, sizeof(char)); CRM_ASSERT(entry != NULL); /* Read entry, stopping at first separator * * @TODO Document or eliminate these character restrictions */ rc = sscanf(line + entry_start, "%[a-zA-Z0-9_-.]", entry); if (rc != 1) { crm_warn("Could not parse list output entry: %s " CRM_XS " entry_start=%d position=%d", line + entry_start, entry_start, i); free(entry); } else if (pcmk__strcase_any_of(entry, "on", "off", NULL)) { /* Some agents print the target status in the list output, * though none are known now (the separate list-status command * is used for this, but it can also print "UNKNOWN"). To handle * this possibility, skip such entries. * * @TODO Document or eliminate the implied restriction of target * names. */ free(entry); } else { // We have a valid entry *output = g_list_append(*output, entry); } entry_start = i + 1; } } } /*! * \internal * \brief Parse a list of targets from a string * * \param[in] list_output Target list as a string * * \return List of target names * \note The target list string format is flexible, to allow for user-specified * lists such pcmk_host_list and the output of an agent's list action * (whether direct or via the API, which escapes newlines). There may be * multiple lines, separated by either a newline or an escaped newline * (backslash n). Each line may have one or more target names, separated * by any combination of whitespace, commas, and semi-colons. Lines * containing "invalid" or "variable" will be ignored entirely. Target * names "on" or "off" (case-insensitive) will be ignored. Target names * may contain only alphanumeric characters, underbars (_), dashes (-), * and dots (.) (if any other character occurs in the name, it and all * subsequent characters in the name will be ignored). * \note The caller is responsible for freeing the result with * g_list_free_full(result, free). */ GList * stonith__parse_targets(const char *target_spec) { GList *targets = NULL; if (target_spec != NULL) { size_t out_len = strlen(target_spec); size_t line_start = 0; // Starting index of line being processed for (size_t i = 0; i <= out_len; ++i) { if ((target_spec[i] == '\n') || (target_spec[i] == '\0') || ((target_spec[i] == '\\') && (target_spec[i + 1] == 'n'))) { // We've reached the end of one line of output int len = i - line_start; if (len > 0) { char *line = strndup(target_spec + line_start, len); line[len] = '\0'; // Because it might be a newline parse_list_line(line, len, &targets); free(line); } if (target_spec[i] == '\\') { ++i; // backslash-n takes up two positions } line_start = i + 1; } } } return targets; } /*! * \internal * \brief Determine if a later stonith event succeeded. * * \note Before calling this function, use stonith__sort_history() to sort the * top_history argument. */ gboolean stonith__later_succeeded(stonith_history_t *event, stonith_history_t *top_history) { gboolean ret = FALSE; for (stonith_history_t *prev_hp = top_history; prev_hp; prev_hp = prev_hp->next) { if (prev_hp == event) { break; } if ((prev_hp->state == st_done) && pcmk__str_eq(event->target, prev_hp->target, pcmk__str_casei) && pcmk__str_eq(event->action, prev_hp->action, pcmk__str_casei) && pcmk__str_eq(event->delegate, prev_hp->delegate, pcmk__str_casei) && (event->completed < prev_hp->completed)) { ret = TRUE; break; } } return ret; } /*! * \internal * \brief Sort the stonith-history * sort by competed most current on the top * pending actions lacking a completed-stamp are gathered at the top * * \param[in] history List of stonith actions * */ stonith_history_t * stonith__sort_history(stonith_history_t *history) { stonith_history_t *new = NULL, *pending = NULL, *hp, *np, *tmp; for (hp = history; hp; ) { tmp = hp->next; if ((hp->state == st_done) || (hp->state == st_failed)) { /* sort into new */ if ((!new) || (hp->completed > new->completed)) { hp->next = new; new = hp; } else { np = new; do { if ((!np->next) || (hp->completed > np->next->completed)) { hp->next = np->next; np->next = hp; break; } np = np->next; } while (1); } } else { /* put into pending */ hp->next = pending; pending = hp; } hp = tmp; } /* pending actions don't have a completed-stamp so make them go front */ if (pending) { stonith_history_t *last_pending = pending; while (last_pending->next) { last_pending = last_pending->next; } last_pending->next = new; new = pending; } return new; } stonith_history_t * stonith__first_matching_event(stonith_history_t *history, bool (*matching_fn)(stonith_history_t *, void *), void *user_data) { for (stonith_history_t *hp = history; hp; hp = hp->next) { if (matching_fn(hp, user_data)) { return hp; } } return NULL; } bool stonith__event_state_pending(stonith_history_t *history, void *user_data) { return history->state != st_failed && history->state != st_done; } bool stonith__event_state_eq(stonith_history_t *history, void *user_data) { return history->state == GPOINTER_TO_INT(user_data); } bool stonith__event_state_neq(stonith_history_t *history, void *user_data) { return history->state != GPOINTER_TO_INT(user_data); } // Deprecated functions kept only for backward API compatibility const char *get_stonith_provider(const char *agent, const char *provider); /*! * \brief Deprecated (use stonith_get_namespace() instead) */ const char * get_stonith_provider(const char *agent, const char *provider) { return stonith_namespace2text(stonith_get_namespace(agent, provider)); } void stonith__device_parameter_flags(uint32_t *device_flags, const char *device_name, xmlNode *metadata) { xmlXPathObjectPtr xpath = NULL; int max = 0; int lpc = 0; CRM_CHECK((device_flags != NULL) && (metadata != NULL), return); xpath = xpath_search(metadata, "//parameter"); max = numXpathResults(xpath); if (max <= 0) { freeXpathObject(xpath); return; } for (lpc = 0; lpc < max; lpc++) { const char *parameter = NULL; xmlNode *match = getXpathResult(xpath, lpc); CRM_LOG_ASSERT(match != NULL); if (match == NULL) { continue; } parameter = crm_element_value(match, "name"); if (pcmk__str_eq(parameter, "plug", pcmk__str_casei)) { stonith__set_device_flags(*device_flags, device_name, st_device_supports_parameter_plug); } else if (pcmk__str_eq(parameter, "port", pcmk__str_casei)) { stonith__set_device_flags(*device_flags, device_name, st_device_supports_parameter_port); } } freeXpathObject(xpath); } diff --git a/lib/pengine/unpack.c b/lib/pengine/unpack.c index a9bbf4b6c0..44dba47e74 100644 --- a/lib/pengine/unpack.c +++ b/lib/pengine/unpack.c @@ -1,4024 +1,4024 @@ /* * 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 #include #include #include #include #include #include #include #include #include #include #include #include #include #include CRM_TRACE_INIT_DATA(pe_status); /* This uses pcmk__set_flags_as()/pcmk__clear_flags_as() directly rather than * use pe__set_working_set_flags()/pe__clear_working_set_flags() so that the * flag is stringified more readably in log messages. */ #define set_config_flag(data_set, option, flag) do { \ const char *scf_value = pe_pref((data_set)->config_hash, (option)); \ if (scf_value != NULL) { \ if (crm_is_true(scf_value)) { \ (data_set)->flags = pcmk__set_flags_as(__func__, __LINE__, \ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } else { \ (data_set)->flags = pcmk__clear_flags_as(__func__, __LINE__,\ LOG_TRACE, "Working set", \ crm_system_name, (data_set)->flags, \ (flag), #flag); \ } \ } \ } while(0) static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *failed, pe_working_set_t *data_set); static void determine_remote_online_status(pe_working_set_t *data_set, pe_node_t *this_node); static void add_node_attrs(xmlNode *attrs, pe_node_t *node, bool overwrite, pe_working_set_t *data_set); static void determine_online_status(xmlNode *node_state, pe_node_t *this_node, pe_working_set_t *data_set); static void unpack_lrm_resources(pe_node_t *node, xmlNode *lrm_state, pe_working_set_t *data_set); // Bitmask for warnings we only want to print once uint32_t pe_wo = 0; static gboolean is_dangling_guest_node(pe_node_t *node) { /* we are looking for a remote-node that was supposed to be mapped to a * container resource, but all traces of that container have disappeared * from both the config and the status section. */ if (pe__is_guest_or_remote_node(node) && node->details->remote_rsc && node->details->remote_rsc->container == NULL && pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_orphan_container_filler)) { return TRUE; } return FALSE; } /*! * \brief Schedule a fence action for a node * * \param[in,out] data_set Current working set of cluster * \param[in,out] node Node to fence * \param[in] reason Text description of why fencing is needed * \param[in] priority_delay Whether to consider `priority-fencing-delay` */ void pe_fence_node(pe_working_set_t * data_set, pe_node_t * node, const char *reason, bool priority_delay) { CRM_CHECK(node, return); /* A guest node is fenced by marking its container as failed */ if (pe__is_guest_node(node)) { pe_resource_t *rsc = node->details->remote_rsc->container; if (!pcmk_is_set(rsc->flags, pe_rsc_failed)) { if (!pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing guest node %s " "(otherwise would because %s): " "its guest resource %s is unmanaged", node->details->uname, reason, rsc->id); } else { crm_warn("Guest node %s will be fenced " "(by recovering its guest resource %s): %s", node->details->uname, rsc->id, reason); /* We don't mark the node as unclean because that would prevent the * node from running resources. We want to allow it to run resources * in this transition if the recovery succeeds. */ node->details->remote_requires_reset = TRUE; pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); } } } else if (is_dangling_guest_node(node)) { crm_info("Cleaning up dangling connection for guest node %s: " "fencing was already done because %s, " "and guest resource no longer exists", node->details->uname, reason); pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } else if (pe__is_remote_node(node)) { pe_resource_t *rsc = node->details->remote_rsc; if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed)) { crm_notice("Not fencing remote node %s " "(otherwise would because %s): connection is unmanaged", node->details->uname, reason); } else if(node->details->remote_requires_reset == FALSE) { node->details->remote_requires_reset = TRUE; crm_warn("Remote node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); } node->details->unclean = TRUE; // No need to apply `priority-fencing-delay` for remote nodes pe_fence_op(node, NULL, TRUE, reason, FALSE, data_set); } else if (node->details->unclean) { crm_trace("Cluster node %s %s because %s", node->details->uname, pe_can_fence(data_set, node)? "would also be fenced" : "also is unclean", reason); } else { crm_warn("Cluster node %s %s: %s", node->details->uname, pe_can_fence(data_set, node)? "will be fenced" : "is unclean", reason); node->details->unclean = TRUE; pe_fence_op(node, NULL, TRUE, reason, priority_delay, data_set); } } // @TODO xpaths can't handle templates, rules, or id-refs // nvpair with provides or requires set to unfencing #define XPATH_UNFENCING_NVPAIR XML_CIB_TAG_NVPAIR \ - "[(@" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_PROVIDES "'" \ + "[(@" XML_NVPAIR_ATTR_NAME "='" PCMK_STONITH_PROVIDES "'" \ "or @" XML_NVPAIR_ATTR_NAME "='" XML_RSC_ATTR_REQUIRES "') " \ "and @" XML_NVPAIR_ATTR_VALUE "='unfencing']" // unfencing in rsc_defaults or any resource #define XPATH_ENABLE_UNFENCING \ "/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RESOURCES \ "//" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR \ "|/" XML_TAG_CIB "/" XML_CIB_TAG_CONFIGURATION "/" XML_CIB_TAG_RSCCONFIG \ "/" XML_TAG_META_SETS "/" XPATH_UNFENCING_NVPAIR static void set_if_xpath(uint64_t flag, const char *xpath, pe_working_set_t *data_set) { xmlXPathObjectPtr result = NULL; if (!pcmk_is_set(data_set->flags, flag)) { result = xpath_search(data_set->input, xpath); if (result && (numXpathResults(result) > 0)) { pe__set_working_set_flags(data_set, flag); } freeXpathObject(result); } } gboolean unpack_config(xmlNode * config, pe_working_set_t * data_set) { const char *value = NULL; GHashTable *config_hash = crm_str_table_new(); pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; data_set->config_hash = config_hash; pe__unpack_dataset_nvpairs(config, XML_CIB_TAG_PROPSET, &rule_data, config_hash, CIB_OPTIONS_FIRST, FALSE, data_set); verify_pe_options(data_set->config_hash); set_config_flag(data_set, "enable-startup-probes", pe_flag_startup_probes); if (!pcmk_is_set(data_set->flags, pe_flag_startup_probes)) { crm_info("Startup probes: disabled (dangerous)"); } value = pe_pref(data_set->config_hash, XML_ATTR_HAVE_WATCHDOG); if (value && crm_is_true(value)) { crm_info("Watchdog-based self-fencing will be performed via SBD if " "fencing is required and stonith-watchdog-timeout is nonzero"); pe__set_working_set_flags(data_set, pe_flag_have_stonith_resource); } /* Set certain flags via xpath here, so they can be used before the relevant * configuration sections are unpacked. */ set_if_xpath(pe_flag_enable_unfencing, XPATH_ENABLE_UNFENCING, data_set); value = pe_pref(data_set->config_hash, "stonith-timeout"); data_set->stonith_timeout = (int) crm_parse_interval_spec(value); crm_debug("STONITH timeout: %d", data_set->stonith_timeout); set_config_flag(data_set, "stonith-enabled", pe_flag_stonith_enabled); crm_debug("STONITH of failed nodes is %s", pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)? "enabled" : "disabled"); data_set->stonith_action = pe_pref(data_set->config_hash, "stonith-action"); if (!strcmp(data_set->stonith_action, "poweroff")) { pe_warn_once(pe_wo_poweroff, "Support for stonith-action of 'poweroff' is deprecated " "and will be removed in a future release (use 'off' instead)"); data_set->stonith_action = "off"; } crm_trace("STONITH will %s nodes", data_set->stonith_action); set_config_flag(data_set, "concurrent-fencing", pe_flag_concurrent_fencing); crm_debug("Concurrent fencing is %s", pcmk_is_set(data_set->flags, pe_flag_concurrent_fencing)? "enabled" : "disabled"); value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY); if (value) { data_set->priority_fencing_delay = crm_parse_interval_spec(value) / 1000; crm_trace("Priority fencing delay is %ds", data_set->priority_fencing_delay); } set_config_flag(data_set, "stop-all-resources", pe_flag_stop_everything); crm_debug("Stop all active resources: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_stop_everything))); set_config_flag(data_set, "symmetric-cluster", pe_flag_symmetric_cluster); if (pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster)) { crm_debug("Cluster is symmetric" " - resources can run anywhere by default"); } value = pe_pref(data_set->config_hash, "no-quorum-policy"); if (pcmk__str_eq(value, "ignore", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_ignore; } else if (pcmk__str_eq(value, "freeze", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_freeze; } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { data_set->no_quorum_policy = no_quorum_demote; } else if (pcmk__str_eq(value, "suicide", pcmk__str_casei)) { if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { int do_panic = 0; crm_element_value_int(data_set->input, XML_ATTR_QUORUM_PANIC, &do_panic); if (do_panic || pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { data_set->no_quorum_policy = no_quorum_suicide; } else { crm_notice("Resetting no-quorum-policy to 'stop': cluster has never had quorum"); data_set->no_quorum_policy = no_quorum_stop; } } else { pcmk__config_err("Resetting no-quorum-policy to 'stop' because " "fencing is disabled"); data_set->no_quorum_policy = no_quorum_stop; } } else { data_set->no_quorum_policy = no_quorum_stop; } switch (data_set->no_quorum_policy) { case no_quorum_freeze: crm_debug("On loss of quorum: Freeze resources"); break; case no_quorum_stop: crm_debug("On loss of quorum: Stop ALL resources"); break; case no_quorum_demote: crm_debug("On loss of quorum: " "Demote promotable resources and stop other resources"); break; case no_quorum_suicide: crm_notice("On loss of quorum: Fence all remaining nodes"); break; case no_quorum_ignore: crm_notice("On loss of quorum: Ignore"); break; } set_config_flag(data_set, "stop-orphan-resources", pe_flag_stop_rsc_orphans); crm_trace("Orphan resources are %s", pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)? "stopped" : "ignored"); set_config_flag(data_set, "stop-orphan-actions", pe_flag_stop_action_orphans); crm_trace("Orphan resource actions are %s", pcmk_is_set(data_set->flags, pe_flag_stop_action_orphans)? "stopped" : "ignored"); set_config_flag(data_set, "remove-after-stop", pe_flag_remove_after_stop); crm_trace("Stopped resources are removed from the status section: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_remove_after_stop))); set_config_flag(data_set, "maintenance-mode", pe_flag_maintenance_mode); crm_trace("Maintenance mode: %s", pcmk__btoa(pcmk_is_set(data_set->flags, pe_flag_maintenance_mode))); set_config_flag(data_set, "start-failure-is-fatal", pe_flag_start_failure_fatal); crm_trace("Start failures are %s", pcmk_is_set(data_set->flags, pe_flag_start_failure_fatal)? "always fatal" : "handled by failcount"); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { set_config_flag(data_set, "startup-fencing", pe_flag_startup_fencing); } if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) { crm_trace("Unseen nodes will be fenced"); } else { pe_warn_once(pe_wo_blind, "Blind faith: not fencing unseen nodes"); } pcmk__score_red = char2score(pe_pref(data_set->config_hash, "node-health-red")); pcmk__score_green = char2score(pe_pref(data_set->config_hash, "node-health-green")); pcmk__score_yellow = char2score(pe_pref(data_set->config_hash, "node-health-yellow")); crm_debug("Node scores: 'red' = %s, 'yellow' = %s, 'green' = %s", pe_pref(data_set->config_hash, "node-health-red"), pe_pref(data_set->config_hash, "node-health-yellow"), pe_pref(data_set->config_hash, "node-health-green")); data_set->placement_strategy = pe_pref(data_set->config_hash, "placement-strategy"); crm_trace("Placement strategy: %s", data_set->placement_strategy); set_config_flag(data_set, "shutdown-lock", pe_flag_shutdown_lock); crm_trace("Resources will%s be locked to cleanly shut down nodes", (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)? "" : " not")); if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { value = pe_pref(data_set->config_hash, XML_CONFIG_ATTR_SHUTDOWN_LOCK_LIMIT); data_set->shutdown_lock = crm_parse_interval_spec(value) / 1000; crm_trace("Shutdown locks expire after %us", data_set->shutdown_lock); } return TRUE; } static void destroy_digest_cache(gpointer ptr) { op_digest_cache_t *data = ptr; free_xml(data->params_all); free_xml(data->params_secure); free_xml(data->params_restart); free(data->digest_all_calc); free(data->digest_restart_calc); free(data->digest_secure_calc); free(data); } pe_node_t * pe_create_node(const char *id, const char *uname, const char *type, const char *score, pe_working_set_t * data_set) { pe_node_t *new_node = NULL; if (pe_find_node(data_set->nodes, uname) != NULL) { pcmk__config_warn("More than one node entry has name '%s'", uname); } new_node = calloc(1, sizeof(pe_node_t)); if (new_node == NULL) { return NULL; } new_node->weight = char2score(score); new_node->fixed = FALSE; new_node->details = calloc(1, sizeof(struct pe_node_shared_s)); if (new_node->details == NULL) { free(new_node); return NULL; } crm_trace("Creating node for entry %s/%s", uname, id); new_node->details->id = id; new_node->details->uname = uname; new_node->details->online = FALSE; new_node->details->shutdown = FALSE; new_node->details->rsc_discovery_enabled = TRUE; new_node->details->running_rsc = NULL; new_node->details->type = node_ping; if (pcmk__str_eq(type, "remote", pcmk__str_casei)) { new_node->details->type = node_remote; pe__set_working_set_flags(data_set, pe_flag_have_remote_nodes); } else if (pcmk__str_eq(type, "member", pcmk__str_null_matches | pcmk__str_casei)) { new_node->details->type = node_member; } new_node->details->attrs = crm_str_table_new(); if (pe__is_guest_or_remote_node(new_node)) { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("remote")); } else { g_hash_table_insert(new_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("cluster")); } new_node->details->utilization = crm_str_table_new(); new_node->details->digest_cache = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_digest_cache); data_set->nodes = g_list_insert_sorted(data_set->nodes, new_node, sort_node_uname); return new_node; } static const char * expand_remote_rsc_meta(xmlNode *xml_obj, xmlNode *parent, pe_working_set_t *data) { xmlNode *attr_set = NULL; xmlNode *attr = NULL; const char *container_id = ID(xml_obj); const char *remote_name = NULL; const char *remote_server = NULL; const char *remote_port = NULL; const char *connect_timeout = "60s"; const char *remote_allow_migrate=NULL; const char *is_managed = NULL; for (attr_set = pcmk__xe_first_child(xml_obj); attr_set != NULL; attr_set = pcmk__xe_next(attr_set)) { if (!pcmk__str_eq((const char *)attr_set->name, XML_TAG_META_SETS, pcmk__str_casei)) { continue; } for (attr = pcmk__xe_first_child(attr_set); attr != NULL; attr = pcmk__xe_next(attr)) { const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); if (pcmk__str_eq(name, XML_RSC_ATTR_REMOTE_NODE, pcmk__str_casei)) { remote_name = value; } else if (pcmk__str_eq(name, "remote-addr", pcmk__str_casei)) { remote_server = value; } else if (pcmk__str_eq(name, "remote-port", pcmk__str_casei)) { remote_port = value; } else if (pcmk__str_eq(name, "remote-connect-timeout", pcmk__str_casei)) { connect_timeout = value; } else if (pcmk__str_eq(name, "remote-allow-migrate", pcmk__str_casei)) { remote_allow_migrate=value; } else if (pcmk__str_eq(name, XML_RSC_ATTR_MANAGED, pcmk__str_casei)) { is_managed = value; } } } if (remote_name == NULL) { return NULL; } if (pe_find_resource(data->resources, remote_name) != NULL) { return NULL; } pe_create_remote_xml(parent, remote_name, container_id, remote_allow_migrate, is_managed, connect_timeout, remote_server, remote_port); return remote_name; } static void handle_startup_fencing(pe_working_set_t *data_set, pe_node_t *new_node) { if ((new_node->details->type == node_remote) && (new_node->details->remote_rsc == NULL)) { /* Ignore fencing for remote nodes that don't have a connection resource * associated with them. This happens when remote node entries get left * in the nodes section after the connection resource is removed. */ return; } if (pcmk_is_set(data_set->flags, pe_flag_startup_fencing)) { // All nodes are unclean until we've seen their status entry new_node->details->unclean = TRUE; } else { // Blind faith ... new_node->details->unclean = FALSE; } /* We need to be able to determine if a node's status section * exists or not separate from whether the node is unclean. */ new_node->details->unseen = TRUE; } gboolean unpack_nodes(xmlNode * xml_nodes, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; pe_node_t *new_node = NULL; const char *id = NULL; const char *uname = NULL; const char *type = NULL; const char *score = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (xml_obj = pcmk__xe_first_child(xml_nodes); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_NODE, pcmk__str_none)) { new_node = NULL; id = crm_element_value(xml_obj, XML_ATTR_ID); uname = crm_element_value(xml_obj, XML_ATTR_UNAME); type = crm_element_value(xml_obj, XML_ATTR_TYPE); score = crm_element_value(xml_obj, XML_RULE_ATTR_SCORE); crm_trace("Processing node %s/%s", uname, id); if (id == NULL) { pcmk__config_err("Ignoring <" XML_CIB_TAG_NODE "> entry in configuration without id"); continue; } new_node = pe_create_node(id, uname, type, score, data_set); if (new_node == NULL) { return FALSE; } /* if(data_set->have_quorum == FALSE */ /* && data_set->no_quorum_policy == no_quorum_stop) { */ /* /\* start shutting resources down *\/ */ /* new_node->weight = -INFINITY; */ /* } */ handle_startup_fencing(data_set, new_node); add_node_attrs(xml_obj, new_node, FALSE, data_set); pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_UTILIZATION, &rule_data, new_node->details->utilization, NULL, FALSE, data_set); crm_trace("Done with node %s", crm_element_value(xml_obj, XML_ATTR_UNAME)); } } if (data_set->localhost && pe_find_node(data_set->nodes, data_set->localhost) == NULL) { crm_info("Creating a fake local node"); pe_create_node(data_set->localhost, data_set->localhost, NULL, 0, data_set); } return TRUE; } static void setup_container(pe_resource_t * rsc, pe_working_set_t * data_set) { const char *container_id = NULL; if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; setup_container(child_rsc, data_set); } return; } container_id = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_CONTAINER); if (container_id && !pcmk__str_eq(container_id, rsc->id, pcmk__str_casei)) { pe_resource_t *container = pe_find_resource(data_set->resources, container_id); if (container) { rsc->container = container; pe__set_resource_flags(container, pe_rsc_is_container); container->fillers = g_list_append(container->fillers, rsc); pe_rsc_trace(rsc, "Resource %s's container is %s", rsc->id, container_id); } else { pe_err("Resource %s: Unknown resource container (%s)", rsc->id, container_id); } } } gboolean unpack_remote_nodes(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; /* Create remote nodes and guest nodes from the resource configuration * before unpacking resources. */ for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { const char *new_node_id = NULL; /* Check for remote nodes, which are defined by ocf:pacemaker:remote * primitives. */ if (xml_contains_remote_node(xml_obj)) { new_node_id = ID(xml_obj); /* The "pe_find_node" check is here to make sure we don't iterate over * an expanded node that has already been added to the node list. */ if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found remote node %s defined by resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Check for guest nodes, which are defined by special meta-attributes * of a primitive of any type (for example, VirtualDomain or Xen). */ if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RESOURCE, pcmk__str_none)) { /* This will add an ocf:pacemaker:remote primitive to the * configuration for the guest node's connection, to be unpacked * later. */ new_node_id = expand_remote_rsc_meta(xml_obj, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s", new_node_id, ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } continue; } /* Check for guest nodes inside a group. Clones are currently not * supported as guest nodes. */ if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_GROUP, pcmk__str_none)) { xmlNode *xml_obj2 = NULL; for (xml_obj2 = pcmk__xe_first_child(xml_obj); xml_obj2 != NULL; xml_obj2 = pcmk__xe_next(xml_obj2)) { new_node_id = expand_remote_rsc_meta(xml_obj2, xml_resources, data_set); if (new_node_id && pe_find_node(data_set->nodes, new_node_id) == NULL) { crm_trace("Found guest node %s in resource %s inside group %s", new_node_id, ID(xml_obj2), ID(xml_obj)); pe_create_node(new_node_id, new_node_id, "remote", NULL, data_set); } } } } return TRUE; } /* Call this after all the nodes and resources have been * unpacked, but before the status section is read. * * A remote node's online status is reflected by the state * of the remote node's connection resource. We need to link * the remote node to this connection resource so we can have * easy access to the connection resource during the scheduler calculations. */ static void link_rsc2remotenode(pe_working_set_t *data_set, pe_resource_t *new_rsc) { pe_node_t *remote_node = NULL; if (new_rsc->is_remote_node == FALSE) { return; } if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* remote_nodes and remote_resources are not linked in quick location calculations */ return; } remote_node = pe_find_node(data_set->nodes, new_rsc->id); CRM_CHECK(remote_node != NULL, return;); pe_rsc_trace(new_rsc, "Linking remote connection resource %s to node %s", new_rsc->id, remote_node->details->uname); remote_node->details->remote_rsc = new_rsc; if (new_rsc->container == NULL) { /* Handle start-up fencing for remote nodes (as opposed to guest nodes) * the same as is done for cluster nodes. */ handle_startup_fencing(data_set, remote_node); } else { /* pe_create_node() marks the new node as "remote" or "cluster"; now * that we know the node is a guest node, update it correctly. */ g_hash_table_replace(remote_node->details->attrs, strdup(CRM_ATTR_KIND), strdup("container")); } } static void destroy_tag(gpointer data) { pe_tag_t *tag = data; if (tag) { free(tag->id); g_list_free_full(tag->refs, free); free(tag); } } /*! * \internal * \brief Parse configuration XML for resource information * * \param[in] xml_resources Top of resource configuration XML * \param[in,out] data_set Where to put resource information * * \return TRUE * * \note unpack_remote_nodes() MUST be called before this, so that the nodes can * be used when common_unpack() calls resource_location() */ gboolean unpack_resources(xmlNode * xml_resources, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; GListPtr gIter = NULL; data_set->template_rsc_sets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_tag); for (xml_obj = pcmk__xe_first_child(xml_resources); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { pe_resource_t *new_rsc = NULL; if (pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_RSC_TEMPLATE, pcmk__str_none)) { const char *template_id = ID(xml_obj); if (template_id && g_hash_table_lookup_extended(data_set->template_rsc_sets, template_id, NULL, NULL) == FALSE) { /* Record the template's ID for the knowledge of its existence anyway. */ g_hash_table_insert(data_set->template_rsc_sets, strdup(template_id), NULL); } continue; } crm_trace("Beginning unpack... <%s id=%s... >", crm_element_name(xml_obj), ID(xml_obj)); if (common_unpack(xml_obj, &new_rsc, NULL, data_set) && (new_rsc != NULL)) { data_set->resources = g_list_append(data_set->resources, new_rsc); pe_rsc_trace(new_rsc, "Added resource %s", new_rsc->id); } else { pcmk__config_err("Ignoring <%s> resource '%s' " "because configuration is invalid", crm_element_name(xml_obj), crm_str(ID(xml_obj))); if (new_rsc != NULL && new_rsc->fns != NULL) { new_rsc->fns->free(new_rsc); } } } for (gIter = data_set->resources; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; setup_container(rsc, data_set); link_rsc2remotenode(data_set, rsc); } data_set->resources = g_list_sort(data_set->resources, sort_rsc_priority); if (pcmk_is_set(data_set->flags, pe_flag_quick_location)) { /* Ignore */ } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && !pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) { pcmk__config_err("Resource start-up disabled since no STONITH resources have been defined"); pcmk__config_err("Either configure some or disable STONITH with the stonith-enabled option"); pcmk__config_err("NOTE: Clusters with shared data need STONITH to ensure data integrity"); } return TRUE; } gboolean unpack_tags(xmlNode * xml_tags, pe_working_set_t * data_set) { xmlNode *xml_tag = NULL; data_set->tags = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_tag); for (xml_tag = pcmk__xe_first_child(xml_tags); xml_tag != NULL; xml_tag = pcmk__xe_next(xml_tag)) { xmlNode *xml_obj_ref = NULL; const char *tag_id = ID(xml_tag); if (!pcmk__str_eq((const char *)xml_tag->name, XML_CIB_TAG_TAG, pcmk__str_none)) { continue; } if (tag_id == NULL) { pcmk__config_err("Ignoring <%s> without " XML_ATTR_ID, crm_element_name(xml_tag)); continue; } for (xml_obj_ref = pcmk__xe_first_child(xml_tag); xml_obj_ref != NULL; xml_obj_ref = pcmk__xe_next(xml_obj_ref)) { const char *obj_ref = ID(xml_obj_ref); if (!pcmk__str_eq((const char *)xml_obj_ref->name, XML_CIB_TAG_OBJ_REF, pcmk__str_none)) { continue; } if (obj_ref == NULL) { pcmk__config_err("Ignoring <%s> for tag '%s' without " XML_ATTR_ID, crm_element_name(xml_obj_ref), tag_id); continue; } if (add_tag_ref(data_set->tags, tag_id, obj_ref) == FALSE) { return FALSE; } } } return TRUE; } /* The ticket state section: * "/cib/status/tickets/ticket_state" */ static gboolean unpack_ticket_state(xmlNode * xml_ticket, pe_working_set_t * data_set) { const char *ticket_id = NULL; const char *granted = NULL; const char *last_granted = NULL; const char *standby = NULL; xmlAttrPtr xIter = NULL; pe_ticket_t *ticket = NULL; ticket_id = ID(xml_ticket); if (pcmk__str_empty(ticket_id)) { return FALSE; } crm_trace("Processing ticket state for %s", ticket_id); ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = ticket_new(ticket_id, data_set); if (ticket == NULL) { return FALSE; } } for (xIter = xml_ticket->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_ticket, prop_name); if (pcmk__str_eq(prop_name, XML_ATTR_ID, pcmk__str_none)) { continue; } g_hash_table_replace(ticket->state, strdup(prop_name), strdup(prop_value)); } granted = g_hash_table_lookup(ticket->state, "granted"); if (granted && crm_is_true(granted)) { ticket->granted = TRUE; crm_info("We have ticket '%s'", ticket->id); } else { ticket->granted = FALSE; crm_info("We do not have ticket '%s'", ticket->id); } last_granted = g_hash_table_lookup(ticket->state, "last-granted"); if (last_granted) { ticket->last_granted = crm_parse_int(last_granted, 0); } standby = g_hash_table_lookup(ticket->state, "standby"); if (standby && crm_is_true(standby)) { ticket->standby = TRUE; if (ticket->granted) { crm_info("Granted ticket '%s' is in standby-mode", ticket->id); } } else { ticket->standby = FALSE; } crm_trace("Done with ticket state for %s", ticket_id); return TRUE; } static gboolean unpack_tickets_state(xmlNode * xml_tickets, pe_working_set_t * data_set) { xmlNode *xml_obj = NULL; for (xml_obj = pcmk__xe_first_child(xml_tickets); xml_obj != NULL; xml_obj = pcmk__xe_next(xml_obj)) { if (!pcmk__str_eq((const char *)xml_obj->name, XML_CIB_TAG_TICKET_STATE, pcmk__str_none)) { continue; } unpack_ticket_state(xml_obj, data_set); } return TRUE; } static void unpack_handle_remote_attrs(pe_node_t *this_node, xmlNode *state, pe_working_set_t * data_set) { const char *resource_discovery_enabled = NULL; xmlNode *attrs = NULL; pe_resource_t *rsc = NULL; if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { return; } if ((this_node == NULL) || !pe__is_guest_or_remote_node(this_node)) { return; } crm_trace("Processing remote node id=%s, uname=%s", this_node->details->id, this_node->details->uname); this_node->details->remote_maintenance = crm_atoi(crm_element_value(state, XML_NODE_IS_MAINTENANCE), "0"); rsc = this_node->details->remote_rsc; if (this_node->details->remote_requires_reset == FALSE) { this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; } attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); if (pe__shutdown_requested(this_node)) { crm_info("Node %s is shutting down", this_node->details->uname); this_node->details->shutdown = TRUE; if (rsc) { rsc->next_role = RSC_ROLE_STOPPED; } } if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance")) || ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_managed))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { if (pe__is_remote_node(this_node) && !pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { crm_warn("Ignoring %s attribute on remote node %s because stonith is disabled", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } else { /* This is either a remote node with fencing enabled, or a guest * node. We don't care whether fencing is enabled when fencing guest * nodes, because they are "fenced" by recovering their containing * resource. */ crm_info("Node %s has resource discovery disabled", this_node->details->uname); this_node->details->rsc_discovery_enabled = FALSE; } } } static bool unpack_node_loop(xmlNode * status, bool fence, pe_working_set_t * data_set) { bool changed = false; xmlNode *lrm_rsc = NULL; for (xmlNode *state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { const char *id = NULL; const char *uname = NULL; pe_node_t *this_node = NULL; bool process = FALSE; if (!pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { continue; } id = crm_element_value(state, XML_ATTR_ID); uname = crm_element_value(state, XML_ATTR_UNAME); this_node = pe_find_node_any(data_set->nodes, id, uname); if (this_node == NULL) { crm_info("Node %s is unknown", id); continue; } else if (this_node->details->unpacked) { crm_trace("Node %s was already processed", id); continue; } else if (!pe__is_guest_or_remote_node(this_node) && pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { // A redundant test, but preserves the order for regression tests process = TRUE; } else if (pe__is_guest_or_remote_node(this_node)) { bool check = FALSE; pe_resource_t *rsc = this_node->details->remote_rsc; if(fence) { check = TRUE; } else if(rsc == NULL) { /* Not ready yet */ } else if (pe__is_guest_node(this_node) && rsc->role == RSC_ROLE_STARTED && rsc->container->role == RSC_ROLE_STARTED) { /* Both the connection and its containing resource need to be * known to be up before we process resources running in it. */ check = TRUE; crm_trace("Checking node %s/%s/%s status %d/%d/%d", id, rsc->id, rsc->container->id, fence, rsc->role, RSC_ROLE_STARTED); } else if (!pe__is_guest_node(this_node) && ((rsc->role == RSC_ROLE_STARTED) || pcmk_is_set(data_set->flags, pe_flag_shutdown_lock))) { check = TRUE; crm_trace("Checking node %s/%s status %d/%d/%d", id, rsc->id, fence, rsc->role, RSC_ROLE_STARTED); } if (check) { determine_remote_online_status(data_set, this_node); unpack_handle_remote_attrs(this_node, state, data_set); process = TRUE; } } else if (this_node->details->online) { process = TRUE; } else if (fence) { process = TRUE; } else if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { process = TRUE; } if(process) { crm_trace("Processing lrm resource entries on %shealthy%s node: %s", fence?"un":"", (pe__is_guest_or_remote_node(this_node)? " remote" : ""), this_node->details->uname); changed = TRUE; this_node->details->unpacked = TRUE; lrm_rsc = find_xml_node(state, XML_CIB_TAG_LRM, FALSE); lrm_rsc = find_xml_node(lrm_rsc, XML_LRM_TAG_RESOURCES, FALSE); unpack_lrm_resources(this_node, lrm_rsc, data_set); } } return changed; } /* remove nodes that are down, stopping */ /* create positive rsc_to_node constraints between resources and the nodes they are running on */ /* anything else? */ gboolean unpack_status(xmlNode * status, pe_working_set_t * data_set) { const char *id = NULL; const char *uname = NULL; xmlNode *state = NULL; pe_node_t *this_node = NULL; crm_trace("Beginning unpack"); if (data_set->tickets == NULL) { data_set->tickets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_ticket); } for (state = pcmk__xe_first_child(status); state != NULL; state = pcmk__xe_next(state)) { if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_TICKETS, pcmk__str_none)) { unpack_tickets_state((xmlNode *) state, data_set); } else if (pcmk__str_eq((const char *)state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { xmlNode *attrs = NULL; const char *resource_discovery_enabled = NULL; id = crm_element_value(state, XML_ATTR_ID); uname = crm_element_value(state, XML_ATTR_UNAME); this_node = pe_find_node_any(data_set->nodes, id, uname); if (uname == NULL) { /* error */ continue; } else if (this_node == NULL) { pcmk__config_warn("Ignoring recorded node status for '%s' " "because no longer in configuration", uname); continue; } else if (pe__is_guest_or_remote_node(this_node)) { /* online state for remote nodes is determined by the * rsc state after all the unpacking is done. we do however * need to mark whether or not the node has been fenced as this plays * a role during unpacking cluster node resource state */ this_node->details->remote_was_fenced = crm_atoi(crm_element_value(state, XML_NODE_IS_FENCED), "0"); continue; } crm_trace("Processing node id=%s, uname=%s", id, uname); /* Mark the node as provisionally clean * - at least we have seen it in the current cluster's lifetime */ this_node->details->unclean = FALSE; this_node->details->unseen = FALSE; attrs = find_xml_node(state, XML_TAG_TRANSIENT_NODEATTRS, FALSE); add_node_attrs(attrs, this_node, TRUE, data_set); if (crm_is_true(pe_node_attribute_raw(this_node, "standby"))) { crm_info("Node %s is in standby-mode", this_node->details->uname); this_node->details->standby = TRUE; } if (crm_is_true(pe_node_attribute_raw(this_node, "maintenance"))) { crm_info("Node %s is in maintenance-mode", this_node->details->uname); this_node->details->maintenance = TRUE; } resource_discovery_enabled = pe_node_attribute_raw(this_node, XML_NODE_ATTR_RSC_DISCOVERY); if (resource_discovery_enabled && !crm_is_true(resource_discovery_enabled)) { crm_warn("ignoring %s attribute on node %s, disabling resource discovery is not allowed on cluster nodes", XML_NODE_ATTR_RSC_DISCOVERY, this_node->details->uname); } crm_trace("determining node state"); determine_online_status(state, this_node, data_set); if (!pcmk_is_set(data_set->flags, pe_flag_have_quorum) && this_node->details->online && (data_set->no_quorum_policy == no_quorum_suicide)) { /* Everything else should flow from this automatically * (at least until the scheduler becomes able to migrate off * healthy resources) */ pe_fence_node(data_set, this_node, "cluster does not have quorum", FALSE); } } } while(unpack_node_loop(status, FALSE, data_set)) { crm_trace("Start another loop"); } // Now catch any nodes we didn't see unpack_node_loop(status, pcmk_is_set(data_set->flags, pe_flag_stonith_enabled), data_set); /* Now that we know where resources are, we can schedule stops of containers * with failed bundle connections */ if (data_set->stop_needed != NULL) { for (GList *item = data_set->stop_needed; item; item = item->next) { pe_resource_t *container = item->data; pe_node_t *node = pe__current_node(container); if (node) { stop_action(container, node, FALSE); } } g_list_free(data_set->stop_needed); data_set->stop_needed = NULL; } for (GListPtr gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *this_node = gIter->data; if (this_node == NULL) { continue; } else if (!pe__is_guest_or_remote_node(this_node)) { continue; } else if(this_node->details->unpacked) { continue; } determine_remote_online_status(data_set, this_node); } return TRUE; } static gboolean determine_online_status_no_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); if (!crm_is_true(in_cluster)) { crm_trace("Node is down: in_cluster=%s", crm_str(in_cluster)); } else if (pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei)) { if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { online = TRUE; } else { crm_debug("Node is not ready to run resources: %s", join); } } else if (this_node->details->expected_up == FALSE) { crm_trace("Controller is down: in_cluster=%s", crm_str(in_cluster)); crm_trace("\tis_peer=%s, join=%s, expected=%s", crm_str(is_peer), crm_str(join), crm_str(exp_state)); } else { /* mark it unclean */ pe_fence_node(data_set, this_node, "peer is unexpectedly down", FALSE); crm_info("\tin_cluster=%s, is_peer=%s, join=%s, expected=%s", crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state)); } return online; } static gboolean determine_online_status_fencing(pe_working_set_t * data_set, xmlNode * node_state, pe_node_t * this_node) { gboolean online = FALSE; gboolean do_terminate = FALSE; bool crmd_online = FALSE; const char *join = crm_element_value(node_state, XML_NODE_JOIN_STATE); const char *is_peer = crm_element_value(node_state, XML_NODE_IS_PEER); const char *in_cluster = crm_element_value(node_state, XML_NODE_IN_CLUSTER); const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); const char *terminate = pe_node_attribute_raw(this_node, "terminate"); /* - XML_NODE_IN_CLUSTER ::= true|false - XML_NODE_IS_PEER ::= online|offline - XML_NODE_JOIN_STATE ::= member|down|pending|banned - XML_NODE_EXPECTED ::= member|down */ if (crm_is_true(terminate)) { do_terminate = TRUE; } else if (terminate != NULL && strlen(terminate) > 0) { /* could be a time() value */ char t = terminate[0]; if (t != '0' && isdigit(t)) { do_terminate = TRUE; } } crm_trace("%s: in_cluster=%s, is_peer=%s, join=%s, expected=%s, term=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate); online = crm_is_true(in_cluster); crmd_online = pcmk__str_eq(is_peer, ONLINESTATUS, pcmk__str_casei); if (exp_state == NULL) { exp_state = CRMD_JOINSTATE_DOWN; } if (this_node->details->shutdown) { crm_debug("%s is shutting down", this_node->details->uname); /* Slightly different criteria since we can't shut down a dead peer */ online = crmd_online; } else if (in_cluster == NULL) { pe_fence_node(data_set, this_node, "peer has not been seen by the cluster", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_NACK, pcmk__str_casei)) { pe_fence_node(data_set, this_node, "peer failed the pacemaker membership criteria", FALSE); } else if (do_terminate == FALSE && pcmk__str_eq(exp_state, CRMD_JOINSTATE_DOWN, pcmk__str_casei)) { if (crm_is_true(in_cluster) || crmd_online) { crm_info("- Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { crm_trace("%s is down or still coming up", this_node->details->uname); } } else if (do_terminate && pcmk__str_eq(join, CRMD_JOINSTATE_DOWN, pcmk__str_casei) && crm_is_true(in_cluster) == FALSE && !crmd_online) { crm_info("Node %s was just shot", this_node->details->uname); online = FALSE; } else if (crm_is_true(in_cluster) == FALSE) { // Consider `priority-fencing-delay` for lost nodes pe_fence_node(data_set, this_node, "peer is no longer part of the cluster", TRUE); } else if (!crmd_online) { pe_fence_node(data_set, this_node, "peer process is no longer available", FALSE); /* Everything is running at this point, now check join state */ } else if (do_terminate) { pe_fence_node(data_set, this_node, "termination was requested", FALSE); } else if (pcmk__str_eq(join, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { crm_info("Node %s is active", this_node->details->uname); } else if (pcmk__strcase_any_of(join, CRMD_JOINSTATE_PENDING, CRMD_JOINSTATE_DOWN, NULL)) { crm_info("Node %s is not ready to run resources", this_node->details->uname); this_node->details->standby = TRUE; this_node->details->pending = TRUE; } else { pe_fence_node(data_set, this_node, "peer was in an unknown state", FALSE); crm_warn("%s: in-cluster=%s, is-peer=%s, join=%s, expected=%s, term=%d, shutdown=%d", this_node->details->uname, crm_str(in_cluster), crm_str(is_peer), crm_str(join), crm_str(exp_state), do_terminate, this_node->details->shutdown); } return online; } static void determine_remote_online_status(pe_working_set_t * data_set, pe_node_t * this_node) { pe_resource_t *rsc = this_node->details->remote_rsc; pe_resource_t *container = NULL; pe_node_t *host = NULL; /* If there is a node state entry for a (former) Pacemaker Remote node * but no resource creating that node, the node's connection resource will * be NULL. Consider it an offline remote node in that case. */ if (rsc == NULL) { this_node->details->online = FALSE; goto remote_online_done; } container = rsc->container; if (container && pcmk__list_of_1(rsc->running_on)) { host = rsc->running_on->data; } /* If the resource is currently started, mark it online. */ if (rsc->role == RSC_ROLE_STARTED) { crm_trace("%s node %s presumed ONLINE because connection resource is started", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = TRUE; } /* consider this node shutting down if transitioning start->stop */ if (rsc->role == RSC_ROLE_STARTED && rsc->next_role == RSC_ROLE_STOPPED) { crm_trace("%s node %s shutting down because connection resource is stopping", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->shutdown = TRUE; } /* Now check all the failure conditions. */ if(container && pcmk_is_set(container->flags, pe_rsc_failed)) { crm_trace("Guest node %s UNCLEAN because guest resource failed", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } else if (pcmk_is_set(rsc->flags, pe_rsc_failed)) { crm_trace("%s node %s OFFLINE because connection resource failed", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; } else if (rsc->role == RSC_ROLE_STOPPED || (container && container->role == RSC_ROLE_STOPPED)) { crm_trace("%s node %s OFFLINE because its resource is stopped", (container? "Guest" : "Remote"), this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = FALSE; } else if (host && (host->details->online == FALSE) && host->details->unclean) { crm_trace("Guest node %s UNCLEAN because host is unclean", this_node->details->id); this_node->details->online = FALSE; this_node->details->remote_requires_reset = TRUE; } remote_online_done: crm_trace("Remote node %s online=%s", this_node->details->id, this_node->details->online ? "TRUE" : "FALSE"); } static void determine_online_status(xmlNode * node_state, pe_node_t * this_node, pe_working_set_t * data_set) { gboolean online = FALSE; const char *exp_state = crm_element_value(node_state, XML_NODE_EXPECTED); CRM_CHECK(this_node != NULL, return); this_node->details->shutdown = FALSE; this_node->details->expected_up = FALSE; if (pe__shutdown_requested(this_node)) { this_node->details->shutdown = TRUE; } else if (pcmk__str_eq(exp_state, CRMD_JOINSTATE_MEMBER, pcmk__str_casei)) { this_node->details->expected_up = TRUE; } if (this_node->details->type == node_ping) { this_node->details->unclean = FALSE; online = FALSE; /* As far as resource management is concerned, * the node is safely offline. * Anyone caught abusing this logic will be shot */ } else if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { online = determine_online_status_no_fencing(data_set, node_state, this_node); } else { online = determine_online_status_fencing(data_set, node_state, this_node); } if (online) { this_node->details->online = TRUE; } else { /* remove node from contention */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (online && this_node->details->shutdown) { /* don't run resources here */ this_node->fixed = TRUE; this_node->weight = -INFINITY; } if (this_node->details->type == node_ping) { crm_info("Node %s is not a pacemaker node", this_node->details->uname); } else if (this_node->details->unclean) { pe_proc_warn("Node %s is unclean", this_node->details->uname); } else if (this_node->details->online) { crm_info("Node %s is %s", this_node->details->uname, this_node->details->shutdown ? "shutting down" : this_node->details->pending ? "pending" : this_node->details->standby ? "standby" : this_node->details->maintenance ? "maintenance" : "online"); } else { crm_trace("Node %s is offline", this_node->details->uname); } } /*! * \internal * \brief Find the end of a resource's name, excluding any clone suffix * * \param[in] id Resource ID to check * * \return Pointer to last character of resource's base name */ const char * pe_base_name_end(const char *id) { if (!pcmk__str_empty(id)) { const char *end = id + strlen(id) - 1; for (const char *s = end; s > id; --s) { switch (*s) { case '0': case '1': case '2': case '3': case '4': case '5': case '6': case '7': case '8': case '9': break; case ':': return (s == end)? s : (s - 1); default: return end; } } return end; } return NULL; } /*! * \internal * \brief Get a resource name excluding any clone suffix * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_strip(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); char *basename = NULL; CRM_ASSERT(end); basename = strndup(last_rsc_id, end - last_rsc_id + 1); CRM_ASSERT(basename); return basename; } /*! * \internal * \brief Get the name of the first instance of a cloned resource * * \param[in] last_rsc_id Resource ID to check * * \return Pointer to newly allocated string with resource's base name plus :0 * \note It is the caller's responsibility to free() the result. * This asserts on error, so callers can assume result is not NULL. */ char * clone_zero(const char *last_rsc_id) { const char *end = pe_base_name_end(last_rsc_id); size_t base_name_len = end - last_rsc_id + 1; char *zero = NULL; CRM_ASSERT(end); zero = calloc(base_name_len + 3, sizeof(char)); CRM_ASSERT(zero); memcpy(zero, last_rsc_id, base_name_len); zero[base_name_len] = ':'; zero[base_name_len + 1] = '0'; return zero; } static pe_resource_t * create_fake_resource(const char *rsc_id, xmlNode * rsc_entry, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; xmlNode *xml_rsc = create_xml_node(NULL, XML_CIB_TAG_RESOURCE); copy_in_properties(xml_rsc, rsc_entry); crm_xml_add(xml_rsc, XML_ATTR_ID, rsc_id); crm_log_xml_debug(xml_rsc, "Orphan resource"); if (!common_unpack(xml_rsc, &rsc, NULL, data_set)) { return NULL; } if (xml_contains_remote_node(xml_rsc)) { pe_node_t *node; crm_debug("Detected orphaned remote node %s", rsc_id); node = pe_find_node(data_set->nodes, rsc_id); if (node == NULL) { node = pe_create_node(rsc_id, rsc_id, "remote", NULL, data_set); } link_rsc2remotenode(data_set, rsc); if (node) { crm_trace("Setting node %s as shutting down due to orphaned connection resource", rsc_id); node->details->shutdown = TRUE; } } if (crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER)) { /* This orphaned rsc needs to be mapped to a container. */ crm_trace("Detected orphaned container filler %s", rsc_id); pe__set_resource_flags(rsc, pe_rsc_orphan_container_filler); } pe__set_resource_flags(rsc, pe_rsc_orphan); data_set->resources = g_list_append(data_set->resources, rsc); return rsc; } /*! * \internal * \brief Create orphan instance for anonymous clone resource history */ static pe_resource_t * create_anonymous_orphan(pe_resource_t *parent, const char *rsc_id, pe_node_t *node, pe_working_set_t *data_set) { pe_resource_t *top = pe__create_clone_child(parent, data_set); // find_rsc() because we might be a cloned group pe_resource_t *orphan = top->fns->find_rsc(top, rsc_id, NULL, pe_find_clone); pe_rsc_debug(parent, "Created orphan %s for %s: %s on %s", top->id, parent->id, rsc_id, node->details->uname); return orphan; } /*! * \internal * \brief Check a node for an instance of an anonymous clone * * Return a child instance of the specified anonymous clone, in order of * preference: (1) the instance running on the specified node, if any; * (2) an inactive instance (i.e. within the total of clone-max instances); * (3) a newly created orphan (i.e. clone-max instances are already active). * * \param[in] data_set Cluster information * \param[in] node Node on which to check for instance * \param[in] parent Clone to check * \param[in] rsc_id Name of cloned resource in history (without instance) */ static pe_resource_t * find_anonymous_clone(pe_working_set_t * data_set, pe_node_t * node, pe_resource_t * parent, const char *rsc_id) { GListPtr rIter = NULL; pe_resource_t *rsc = NULL; pe_resource_t *inactive_instance = NULL; gboolean skip_inactive = FALSE; CRM_ASSERT(parent != NULL); CRM_ASSERT(pe_rsc_is_clone(parent)); CRM_ASSERT(!pcmk_is_set(parent->flags, pe_rsc_unique)); // Check for active (or partially active, for cloned groups) instance pe_rsc_trace(parent, "Looking for %s on %s in %s", rsc_id, node->details->uname, parent->id); for (rIter = parent->children; rsc == NULL && rIter; rIter = rIter->next) { GListPtr locations = NULL; pe_resource_t *child = rIter->data; /* Check whether this instance is already known to be active or pending * anywhere, at this stage of unpacking. Because this function is called * for a resource before the resource's individual operation history * entries are unpacked, locations will generally not contain the * desired node. * * However, there are three exceptions: * (1) when child is a cloned group and we have already unpacked the * history of another member of the group on the same node; * (2) when we've already unpacked the history of another numbered * instance on the same node (which can happen if globally-unique * was flipped from true to false); and * (3) when we re-run calculations on the same data set as part of a * simulation. */ child->fns->location(child, &locations, 2); if (locations) { /* We should never associate the same numbered anonymous clone * instance with multiple nodes, and clone instances can't migrate, * so there must be only one location, regardless of history. */ CRM_LOG_ASSERT(locations->next == NULL); if (((pe_node_t *)locations->data)->details == node->details) { /* This child instance is active on the requested node, so check * for a corresponding configured resource. We use find_rsc() * instead of child because child may be a cloned group, and we * need the particular member corresponding to rsc_id. * * If the history entry is orphaned, rsc will be NULL. */ rsc = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); if (rsc) { /* If there are multiple instance history entries for an * anonymous clone in a single node's history (which can * happen if globally-unique is switched from true to * false), we want to consider the instances beyond the * first as orphans, even if there are inactive instance * numbers available. */ if (rsc->running_on) { crm_notice("Active (now-)anonymous clone %s has " "multiple (orphan) instance histories on %s", parent->id, node->details->uname); skip_inactive = TRUE; rsc = NULL; } else { pe_rsc_trace(parent, "Resource %s, active", rsc->id); } } } g_list_free(locations); } else { pe_rsc_trace(parent, "Resource %s, skip inactive", child->id); if (!skip_inactive && !inactive_instance && !pcmk_is_set(child->flags, pe_rsc_block)) { // Remember one inactive instance in case we don't find active inactive_instance = parent->fns->find_rsc(child, rsc_id, NULL, pe_find_clone); /* ... but don't use it if it was already associated with a * pending action on another node */ if (inactive_instance && inactive_instance->pending_node && (inactive_instance->pending_node->details != node->details)) { inactive_instance = NULL; } } } } if ((rsc == NULL) && !skip_inactive && (inactive_instance != NULL)) { pe_rsc_trace(parent, "Resource %s, empty slot", inactive_instance->id); rsc = inactive_instance; } /* If the resource has "requires" set to "quorum" or "nothing", and we don't * have a clone instance for every node, we don't want to consume a valid * instance number for unclean nodes. Such instances may appear to be active * according to the history, but should be considered inactive, so we can * start an instance elsewhere. Treat such instances as orphans. * * An exception is instances running on guest nodes -- since guest node * "fencing" is actually just a resource stop, requires shouldn't apply. * * @TODO Ideally, we'd use an inactive instance number if it is not needed * for any clean instances. However, we don't know that at this point. */ if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_needs_fencing) && (!node->details->online || node->details->unclean) && !pe__is_guest_node(node) && !pe__is_universal_clone(parent, data_set)) { rsc = NULL; } if (rsc == NULL) { rsc = create_anonymous_orphan(parent, rsc_id, node, data_set); pe_rsc_trace(parent, "Resource %s, orphan", rsc->id); } return rsc; } static pe_resource_t * unpack_find_resource(pe_working_set_t * data_set, pe_node_t * node, const char *rsc_id, xmlNode * rsc_entry) { pe_resource_t *rsc = NULL; pe_resource_t *parent = NULL; crm_trace("looking for %s", rsc_id); rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL) { /* If we didn't find the resource by its name in the operation history, * check it again as a clone instance. Even when clone-max=0, we create * a single :0 orphan to match against here. */ char *clone0_id = clone_zero(rsc_id); pe_resource_t *clone0 = pe_find_resource(data_set->resources, clone0_id); if (clone0 && !pcmk_is_set(clone0->flags, pe_rsc_unique)) { rsc = clone0; parent = uber_parent(clone0); crm_trace("%s found as %s (%s)", rsc_id, clone0_id, parent->id); } else { crm_trace("%s is not known as %s either (orphan)", rsc_id, clone0_id); } free(clone0_id); } else if (rsc->variant > pe_native) { crm_trace("Resource history for %s is orphaned because it is no longer primitive", rsc_id); return NULL; } else { parent = uber_parent(rsc); } if (pe_rsc_is_anon_clone(parent)) { if (pe_rsc_is_bundled(parent)) { rsc = pe__find_bundle_replica(parent->parent, node); } else { char *base = clone_strip(rsc_id); rsc = find_anonymous_clone(data_set, node, parent, base); free(base); CRM_ASSERT(rsc != NULL); } } if (rsc && !pcmk__str_eq(rsc_id, rsc->id, pcmk__str_casei) && !pcmk__str_eq(rsc_id, rsc->clone_name, pcmk__str_casei)) { free(rsc->clone_name); rsc->clone_name = strdup(rsc_id); pe_rsc_debug(rsc, "Internally renamed %s on %s to %s%s", rsc_id, node->details->uname, rsc->id, (pcmk_is_set(rsc->flags, pe_rsc_orphan)? " (ORPHAN)" : "")); } return rsc; } static pe_resource_t * process_orphan_resource(xmlNode * rsc_entry, pe_node_t * node, pe_working_set_t * data_set) { pe_resource_t *rsc = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); crm_debug("Detected orphan resource %s on %s", rsc_id, node->details->uname); rsc = create_fake_resource(rsc_id, rsc_entry, data_set); if (!pcmk_is_set(data_set->flags, pe_flag_stop_rsc_orphans)) { pe__clear_resource_flags(rsc, pe_rsc_managed); } else { CRM_CHECK(rsc != NULL, return NULL); pe_rsc_trace(rsc, "Added orphan %s", rsc->id); resource_location(rsc, NULL, -INFINITY, "__orphan_do_not_run__", data_set); } return rsc; } static void process_rsc_state(pe_resource_t * rsc, pe_node_t * node, enum action_fail_response on_fail, xmlNode * migrate_op, pe_working_set_t * data_set) { pe_node_t *tmpnode = NULL; char *reason = NULL; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "Resource %s is %s on %s: on_fail=%s", rsc->id, role2text(rsc->role), node->details->uname, fail2text(on_fail)); /* process current state */ if (rsc->role != RSC_ROLE_UNKNOWN) { pe_resource_t *iter = rsc; while (iter) { if (g_hash_table_lookup(iter->known_on, node->details->id) == NULL) { pe_node_t *n = pe__copy_node(node); pe_rsc_trace(rsc, "%s (aka. %s) known on %s", rsc->id, rsc->clone_name, n->details->uname); g_hash_table_insert(iter->known_on, (gpointer) n->details->id, n); } if (pcmk_is_set(iter->flags, pe_rsc_unique)) { break; } iter = iter->parent; } } /* If a managed resource is believed to be running, but node is down ... */ if (rsc->role > RSC_ROLE_STOPPED && node->details->online == FALSE && node->details->maintenance == FALSE && pcmk_is_set(rsc->flags, pe_rsc_managed)) { gboolean should_fence = FALSE; /* If this is a guest node, fence it (regardless of whether fencing is * enabled, because guest node fencing is done by recovery of the * container resource rather than by the fencer). Mark the resource * we're processing as failed. When the guest comes back up, its * operation history in the CIB will be cleared, freeing the affected * resource to run again once we are sure we know its state. */ if (pe__is_guest_node(node)) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); should_fence = TRUE; } else if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { if (pe__is_remote_node(node) && node->details->remote_rsc && !pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_failed)) { /* Setting unseen means that fencing of the remote node will * occur only if the connection resource is not going to start * somewhere. This allows connection resources on a failed * cluster node to move to another node without requiring the * remote nodes to be fenced as well. */ node->details->unseen = TRUE; reason = crm_strdup_printf("%s is active there (fencing will be" " revoked if remote connection can " "be re-established elsewhere)", rsc->id); } should_fence = TRUE; } if (should_fence) { if (reason == NULL) { reason = crm_strdup_printf("%s is thought to be active there", rsc->id); } pe_fence_node(data_set, node, reason, FALSE); } free(reason); } if (node->details->unclean) { /* No extra processing needed * Also allows resources to be started again after a node is shot */ on_fail = action_fail_ignore; } switch (on_fail) { case action_fail_ignore: /* nothing to do */ break; case action_fail_demote: pe__set_resource_flags(rsc, pe_rsc_failed); demote_action(rsc, node, FALSE); break; case action_fail_fence: /* treat it as if it is still running * but also mark the node as unclean */ reason = crm_strdup_printf("%s failed there", rsc->id); pe_fence_node(data_set, node, reason, FALSE); free(reason); break; case action_fail_standby: node->details->standby = TRUE; node->details->standby_onfail = TRUE; break; case action_fail_block: /* is_managed == FALSE will prevent any * actions being sent for the resource */ pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); break; case action_fail_migrate: /* make sure it comes up somewhere else * or not at all */ resource_location(rsc, node, -INFINITY, "__action_migration_auto__", data_set); break; case action_fail_stop: rsc->next_role = RSC_ROLE_STOPPED; break; case action_fail_recover: if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); stop_action(rsc, node, FALSE); } break; case action_fail_restart_container: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (rsc->container && pe_rsc_is_bundled(rsc)) { /* A bundle's remote connection can run on a different node than * the bundle's container. We don't necessarily know where the * container is running yet, so remember it and add a stop * action for it later. */ data_set->stop_needed = g_list_prepend(data_set->stop_needed, rsc->container); } else if (rsc->container) { stop_action(rsc->container, node, FALSE); } else if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { stop_action(rsc, node, FALSE); } break; case action_fail_reset_remote: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { tmpnode = NULL; if (rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); } if (tmpnode && pe__is_remote_node(tmpnode) && tmpnode->details->remote_was_fenced == 0) { /* The remote connection resource failed in a way that * should result in fencing the remote node. */ pe_fence_node(data_set, tmpnode, "remote connection is unrecoverable", FALSE); } } /* require the stop action regardless if fencing is occurring or not. */ if (rsc->role > RSC_ROLE_STOPPED) { stop_action(rsc, node, FALSE); } /* if reconnect delay is in use, prevent the connection from exiting the * "STOPPED" role until the failure is cleared by the delay timeout. */ if (rsc->remote_reconnect_ms) { rsc->next_role = RSC_ROLE_STOPPED; } break; } /* ensure a remote-node connection failure forces an unclean remote-node * to be fenced. By setting unseen = FALSE, the remote-node failure will * result in a fencing operation regardless if we're going to attempt to * reconnect to the remote-node in this transition or not. */ if (pcmk_is_set(rsc->flags, pe_rsc_failed) && rsc->is_remote_node) { tmpnode = pe_find_node(data_set->nodes, rsc->id); if (tmpnode && tmpnode->details->unclean) { tmpnode->details->unseen = FALSE; } } if (rsc->role != RSC_ROLE_STOPPED && rsc->role != RSC_ROLE_UNKNOWN) { if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { if (pcmk_is_set(rsc->flags, pe_rsc_managed)) { pcmk__config_warn("Detected active orphan %s running on %s", rsc->id, node->details->uname); } else { pcmk__config_warn("Resource '%s' must be stopped manually on " "%s because cluster is configured not to " "stop active orphans", rsc->id, node->details->uname); } } native_add_running(rsc, node, data_set); switch (on_fail) { case action_fail_ignore: break; case action_fail_demote: case action_fail_block: pe__set_resource_flags(rsc, pe_rsc_failed); break; default: pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); break; } } else if (rsc->clone_name && strchr(rsc->clone_name, ':') != NULL) { /* Only do this for older status sections that included instance numbers * Otherwise stopped instances will appear as orphans */ pe_rsc_trace(rsc, "Resetting clone_name %s for %s (stopped)", rsc->clone_name, rsc->id); free(rsc->clone_name); rsc->clone_name = NULL; } else { GList *possible_matches = pe__resource_actions(rsc, node, RSC_STOP, FALSE); GListPtr gIter = possible_matches; for (; gIter != NULL; gIter = gIter->next) { pe_action_t *stop = (pe_action_t *) gIter->data; pe__set_action_flags(stop, pe_action_optional); } g_list_free(possible_matches); } } /* create active recurring operations as optional */ static void process_recurring(pe_node_t * node, pe_resource_t * rsc, int start_index, int stop_index, GListPtr sorted_op_list, pe_working_set_t * data_set) { int counter = -1; const char *task = NULL; const char *status = NULL; GListPtr gIter = sorted_op_list; CRM_ASSERT(rsc); pe_rsc_trace(rsc, "%s: Start index %d, stop index = %d", rsc->id, start_index, stop_index); for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; guint interval_ms = 0; char *key = NULL; const char *id = ID(rsc_op); counter++; if (node->details->online == FALSE) { pe_rsc_trace(rsc, "Skipping %s/%s: node is offline", rsc->id, node->details->uname); break; /* Need to check if there's a monitor for role="Stopped" */ } else if (start_index < stop_index && counter <= stop_index) { pe_rsc_trace(rsc, "Skipping %s/%s: resource is not active", id, node->details->uname); continue; } else if (counter < start_index) { pe_rsc_trace(rsc, "Skipping %s/%s: old %d", id, node->details->uname, counter); continue; } crm_element_value_ms(rsc_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if (interval_ms == 0) { pe_rsc_trace(rsc, "Skipping %s/%s: non-recurring", id, node->details->uname); continue; } status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(status, "-1", pcmk__str_casei)) { pe_rsc_trace(rsc, "Skipping %s/%s: status", id, node->details->uname); continue; } task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); /* create the action */ key = pcmk__op_key(rsc->id, task, interval_ms); pe_rsc_trace(rsc, "Creating %s/%s", key, node->details->uname); custom_action(rsc, key, task, node, TRUE, TRUE, data_set); } } void calculate_active_ops(GListPtr sorted_op_list, int *start_index, int *stop_index) { int counter = -1; int implied_monitor_start = -1; int implied_clone_start = -1; const char *task = NULL; const char *status = NULL; GListPtr gIter = sorted_op_list; *stop_index = -1; *start_index = -1; for (; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); status = crm_element_value(rsc_op, XML_LRM_ATTR_OPSTATUS); if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei) && pcmk__str_eq(status, "0", pcmk__str_casei)) { *stop_index = counter; } else if (pcmk__strcase_any_of(task, CRMD_ACTION_START, CRMD_ACTION_MIGRATED, NULL)) { *start_index = counter; } else if ((implied_monitor_start <= *stop_index) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { const char *rc = crm_element_value(rsc_op, XML_LRM_ATTR_RC); if (pcmk__strcase_any_of(rc, "0", "8", NULL)) { implied_monitor_start = counter; } } else if (pcmk__strcase_any_of(task, CRMD_ACTION_PROMOTE, CRMD_ACTION_DEMOTE, NULL)) { implied_clone_start = counter; } } if (*start_index == -1) { if (implied_clone_start != -1) { *start_index = implied_clone_start; } else if (implied_monitor_start != -1) { *start_index = implied_monitor_start; } } } // If resource history entry has shutdown lock, remember lock node and time static void unpack_shutdown_lock(xmlNode *rsc_entry, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { time_t lock_time = 0; // When lock started (i.e. node shutdown time) if ((crm_element_value_epoch(rsc_entry, XML_CONFIG_ATTR_SHUTDOWN_LOCK, &lock_time) == pcmk_ok) && (lock_time != 0)) { if ((data_set->shutdown_lock > 0) && (get_effective_time(data_set) > (lock_time + data_set->shutdown_lock))) { pe_rsc_info(rsc, "Shutdown lock for %s on %s expired", rsc->id, node->details->uname); pe__clear_resource_history(rsc, node, data_set); } else { rsc->lock_node = node; rsc->lock_time = lock_time; } } } static pe_resource_t * unpack_lrm_rsc_state(pe_node_t * node, xmlNode * rsc_entry, pe_working_set_t * data_set) { GListPtr gIter = NULL; int stop_index = -1; int start_index = -1; enum rsc_role_e req_role = RSC_ROLE_UNKNOWN; const char *task = NULL; const char *rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); pe_resource_t *rsc = NULL; GListPtr op_list = NULL; GListPtr sorted_op_list = NULL; xmlNode *migrate_op = NULL; xmlNode *rsc_op = NULL; xmlNode *last_failure = NULL; enum action_fail_response on_fail = action_fail_ignore; enum rsc_role_e saved_role = RSC_ROLE_UNKNOWN; crm_trace("[%s] Processing %s on %s", crm_element_name(rsc_entry), rsc_id, node->details->uname); /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) { op_list = g_list_prepend(op_list, rsc_op); } } if (!pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } } /* find the resource */ rsc = unpack_find_resource(data_set, node, rsc_id, rsc_entry); if (rsc == NULL) { if (op_list == NULL) { // If there are no operations, there is nothing to do return NULL; } else { rsc = process_orphan_resource(rsc_entry, node, data_set); } } CRM_ASSERT(rsc != NULL); // Check whether the resource is "shutdown-locked" to this node if (pcmk_is_set(data_set->flags, pe_flag_shutdown_lock)) { unpack_shutdown_lock(rsc_entry, rsc, node, data_set); } /* process operations */ saved_role = rsc->role; rsc->role = RSC_ROLE_UNKNOWN; sorted_op_list = g_list_sort(op_list, sort_op_by_callid); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; task = crm_element_value(rsc_op, XML_LRM_ATTR_TASK); if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { migrate_op = rsc_op; } unpack_rsc_op(rsc, node, rsc_op, &last_failure, &on_fail, data_set); } /* create active recurring operations as optional */ calculate_active_ops(sorted_op_list, &start_index, &stop_index); process_recurring(node, rsc, start_index, stop_index, sorted_op_list, data_set); /* no need to free the contents */ g_list_free(sorted_op_list); process_rsc_state(rsc, node, on_fail, migrate_op, data_set); if (get_target_role(rsc, &req_role)) { if (rsc->next_role == RSC_ROLE_UNKNOWN || req_role < rsc->next_role) { pe_rsc_debug(rsc, "%s: Overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); rsc->next_role = req_role; } else if (req_role > rsc->next_role) { pe_rsc_info(rsc, "%s: Not overwriting calculated next role %s" " with requested next role %s", rsc->id, role2text(rsc->next_role), role2text(req_role)); } } if (saved_role > rsc->role) { rsc->role = saved_role; } return rsc; } static void handle_orphaned_container_fillers(xmlNode * lrm_rsc_list, pe_working_set_t * data_set) { xmlNode *rsc_entry = NULL; for (rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { pe_resource_t *rsc; pe_resource_t *container; const char *rsc_id; const char *container_id; if (!pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_casei)) { continue; } container_id = crm_element_value(rsc_entry, XML_RSC_ATTR_CONTAINER); rsc_id = crm_element_value(rsc_entry, XML_ATTR_ID); if (container_id == NULL || rsc_id == NULL) { continue; } container = pe_find_resource(data_set->resources, container_id); if (container == NULL) { continue; } rsc = pe_find_resource(data_set->resources, rsc_id); if (rsc == NULL || !pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler) || rsc->container != NULL) { continue; } pe_rsc_trace(rsc, "Mapped container of orphaned resource %s to %s", rsc->id, container_id); rsc->container = container; container->fillers = g_list_append(container->fillers, rsc); } } static void unpack_lrm_resources(pe_node_t *node, xmlNode *lrm_rsc_list, pe_working_set_t *data_set) { xmlNode *rsc_entry = NULL; gboolean found_orphaned_container_filler = FALSE; for (rsc_entry = pcmk__xe_first_child(lrm_rsc_list); rsc_entry != NULL; rsc_entry = pcmk__xe_next(rsc_entry)) { if (pcmk__str_eq((const char *)rsc_entry->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) { pe_resource_t *rsc = unpack_lrm_rsc_state(node, rsc_entry, data_set); if (!rsc) { continue; } if (pcmk_is_set(rsc->flags, pe_rsc_orphan_container_filler)) { found_orphaned_container_filler = TRUE; } } } /* now that all the resource state has been unpacked for this node * we have to go back and map any orphaned container fillers to their * container resource */ if (found_orphaned_container_filler) { handle_orphaned_container_fillers(lrm_rsc_list, data_set); } } static void set_active(pe_resource_t * rsc) { pe_resource_t *top = uber_parent(rsc); if (top && pcmk_is_set(top->flags, pe_rsc_promotable)) { rsc->role = RSC_ROLE_SLAVE; } else { rsc->role = RSC_ROLE_STARTED; } } static void set_node_score(gpointer key, gpointer value, gpointer user_data) { pe_node_t *node = value; int *score = user_data; node->weight = *score; } #define STATUS_PATH_MAX 1024 static xmlNode * find_lrm_op(const char *resource, const char *op, const char *node, const char *source, bool success_only, pe_working_set_t *data_set) { int offset = 0; char xpath[STATUS_PATH_MAX]; xmlNode *xml = NULL; offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//node_state[@uname='%s']", node); offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "//" XML_LRM_TAG_RESOURCE "[@id='%s']", resource); /* Need to check against transition_magic too? */ if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_target='%s']", op, source); } else if (source && pcmk__str_eq(op, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s' and @migrate_source='%s']", op, source); } else { offset += snprintf(xpath + offset, STATUS_PATH_MAX - offset, "/" XML_LRM_TAG_RSC_OP "[@operation='%s']", op); } CRM_LOG_ASSERT(offset > 0); xml = get_xpath_object(xpath, data_set->input, LOG_DEBUG); if (xml && success_only) { int rc = PCMK_OCF_UNKNOWN_ERROR; int status = PCMK_LRM_OP_ERROR; crm_element_value_int(xml, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml, XML_LRM_ATTR_OPSTATUS, &status); if ((rc != PCMK_OCF_OK) || (status != PCMK_LRM_OP_DONE)) { return NULL; } } return xml; } static int pe__call_id(xmlNode *op_xml) { int id = 0; if (op_xml) { crm_element_value_int(op_xml, XML_LRM_ATTR_CALLID, &id); } return id; } /*! * \brief Check whether a stop happened on the same node after some event * * \param[in] rsc Resource being checked * \param[in] node Node being checked * \param[in] xml_op Event that stop is being compared to * \param[in] data_set Cluster working set * * \return TRUE if stop happened after event, FALSE otherwise * * \note This is really unnecessary, but kept as a safety mechanism. We * currently don't save more than one successful event in history, so this * only matters when processing really old CIB files that we don't * technically support anymore, or as preparation for logging an extended * history in the future. */ static bool stop_happened_after(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *stop_op = find_lrm_op(rsc->id, CRMD_ACTION_STOP, node->details->uname, NULL, TRUE, data_set); return (stop_op && (pe__call_id(stop_op) > pe__call_id(xml_op))); } static void unpack_migrate_to_success(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { /* A successful migration sequence is: * migrate_to on source node * migrate_from on target node * stop on source node * * If a migrate_to is followed by a stop, the entire migration (successful * or failed) is complete, and we don't care what happened on the target. * * If no migrate_from has happened, the migration is considered to be * "partial". If the migrate_from failed, make sure the resource gets * stopped on both source and target (if up). * * If the migrate_to and migrate_from both succeeded (which also implies the * resource is no longer running on the source), but there is no stop, the * migration is considered to be "dangling". Schedule a stop on the source * in this case. */ int from_rc = 0; int from_status = 0; pe_node_t *target_node = NULL; pe_node_t *source_node = NULL; xmlNode *migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), return); if (stop_happened_after(rsc, node, xml_op, data_set)) { return; } // Clones are not allowed to migrate, so role can't be master rsc->role = RSC_ROLE_STARTED; target_node = pe_find_node(data_set->nodes, target); source_node = pe_find_node(data_set->nodes, source); // Check whether there was a migrate_from action on the target migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, FALSE, data_set); if (migrate_from) { crm_element_value_int(migrate_from, XML_LRM_ATTR_RC, &from_rc); crm_element_value_int(migrate_from, XML_LRM_ATTR_OPSTATUS, &from_status); pe_rsc_trace(rsc, "%s op on %s exited with status=%d, rc=%d", ID(migrate_from), target, from_status, from_rc); } if (migrate_from && from_rc == PCMK_OCF_OK && from_status == PCMK_LRM_OP_DONE) { /* The migrate_to and migrate_from both succeeded, so mark the migration * as "dangling". This will be used to schedule a stop action on the * source without affecting the target. */ pe_rsc_trace(rsc, "Detected dangling migration op: %s on %s", ID(xml_op), source); rsc->role = RSC_ROLE_STOPPED; rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } else if (migrate_from && (from_status != PCMK_LRM_OP_PENDING)) { // Failed if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set); } } else { // Pending, or complete but erased if (target_node && target_node->details->online) { pe_rsc_trace(rsc, "Marking active on %s %p %d", target, target_node, target_node->details->online); native_add_running(rsc, target_node, data_set); if (source_node && source_node->details->online) { /* This is a partial migration: the migrate_to completed * successfully on the source, but the migrate_from has not * completed. Remember the source and target; if the newly * chosen target remains the same when we schedule actions * later, we may continue with the migration. */ rsc->partial_migration_target = target_node; rsc->partial_migration_source = source_node; } } else { /* Consider it failed here - forces a restart, prevents migration */ pe__set_resource_flags(rsc, pe_rsc_failed|pe_rsc_stop); pe__clear_resource_flags(rsc, pe_rsc_allow_migrate); } } } // Is there an action_name in node_name's rsc history newer than call_id? static bool newer_op(pe_resource_t *rsc, const char *action_name, const char *node_name, int call_id, pe_working_set_t *data_set) { xmlNode *action = find_lrm_op(rsc->id, action_name, node_name, NULL, TRUE, data_set); return pe__call_id(action) > call_id; } static void unpack_migrate_to_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { int target_stop_id = 0; int target_migrate_from_id = 0; xmlNode *target_stop = NULL; xmlNode *target_migrate_from = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(source, node->details->uname), return); /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be master. */ rsc->role = RSC_ROLE_STARTED; // Check for stop on the target target_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, target, NULL, TRUE, data_set); target_stop_id = pe__call_id(target_stop); // Check for migrate_from on the target target_migrate_from = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATED, target, source, TRUE, data_set); target_migrate_from_id = pe__call_id(target_migrate_from); if ((target_stop == NULL) || (target_stop_id < target_migrate_from_id)) { /* There was no stop on the target, or a stop that happened before a * migrate_from, so assume the resource is still active on the target * (if it is up). */ pe_node_t *target_node = pe_find_node(data_set->nodes, target); pe_rsc_trace(rsc, "stop (%d) + migrate_from (%d)", target_stop_id, target_migrate_from_id); if (target_node && target_node->details->online) { native_add_running(rsc, target_node, data_set); } } else if (target_migrate_from == NULL) { /* We know there was a stop on the target, but there may not have been a * migrate_from (the stop could have happened before migrate_from was * scheduled or attempted). * * That means this could be a "dangling" migration. But first, check * whether there is a newer successful stop, start, or migrate_from on * the source node -- it's possible the failed migration was followed by * a successful stop, full restart, or migration in the reverse * direction, in which case we don't want to force a stop. */ int source_migrate_to_id = pe__call_id(xml_op); if (newer_op(rsc, CRMD_ACTION_MIGRATED, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_START, source, source_migrate_to_id, data_set) || newer_op(rsc, CRMD_ACTION_STOP, source, source_migrate_to_id, data_set)) { return; } // Mark node as having dangling migration so we can force a stop later rsc->dangling_migrations = g_list_prepend(rsc->dangling_migrations, node); } } static void unpack_migrate_from_failure(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, pe_working_set_t *data_set) { xmlNode *source_stop = NULL; xmlNode *source_migrate_to = NULL; const char *source = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_SOURCE); const char *target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); // Sanity check CRM_CHECK(source && target && !strcmp(target, node->details->uname), return); /* If a migration failed, we have to assume the resource is active. Clones * are not allowed to migrate, so role can't be master. */ rsc->role = RSC_ROLE_STARTED; // Check for a stop on the source source_stop = find_lrm_op(rsc->id, CRMD_ACTION_STOP, source, NULL, TRUE, data_set); // Check for a migrate_to on the source source_migrate_to = find_lrm_op(rsc->id, CRMD_ACTION_MIGRATE, source, target, TRUE, data_set); if ((source_stop == NULL) || (pe__call_id(source_stop) < pe__call_id(source_migrate_to))) { /* There was no stop on the source, or a stop that happened before * migrate_to, so assume the resource is still active on the source (if * it is up). */ pe_node_t *source_node = pe_find_node(data_set->nodes, source); if (source_node && source_node->details->online) { native_add_running(rsc, source_node, data_set); } } } static void record_failed_op(xmlNode *op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { xmlNode *xIter = NULL; const char *op_key = crm_element_value(op, XML_LRM_ATTR_TASK_KEY); if (node->details->online == FALSE) { return; } for (xIter = data_set->failed->children; xIter; xIter = xIter->next) { const char *key = crm_element_value(xIter, XML_LRM_ATTR_TASK_KEY); const char *uname = crm_element_value(xIter, XML_ATTR_UNAME); if(pcmk__str_eq(op_key, key, pcmk__str_casei) && pcmk__str_eq(uname, node->details->uname, pcmk__str_casei)) { crm_trace("Skipping duplicate entry %s on %s", op_key, node->details->uname); return; } } crm_trace("Adding entry %s on %s", op_key, node->details->uname); crm_xml_add(op, XML_ATTR_UNAME, node->details->uname); crm_xml_add(op, XML_LRM_ATTR_RSCID, rsc->id); add_node_copy(data_set->failed, op); } static const char *get_op_key(xmlNode *xml_op) { const char *key = crm_element_value(xml_op, XML_LRM_ATTR_TASK_KEY); if(key == NULL) { key = ID(xml_op); } return key; } static const char * last_change_str(xmlNode *xml_op) { time_t when; const char *when_s = NULL; if (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &when) == pcmk_ok) { when_s = pcmk__epoch2str(&when); if (when_s) { // Skip day of week to make message shorter when_s = strchr(when_s, ' '); if (when_s) { ++when_s; } } } return ((when_s && *when_s)? when_s : "unknown time"); } /*! * \internal * \brief Compare two on-fail values * * \param[in] first One on-fail value to compare * \param[in] second The other on-fail value to compare * * \return A negative number if second is more severe than first, zero if they * are equal, or a positive number if first is more severe than second. * \note This is only needed until the action_fail_response values can be * renumbered at the next API compatibility break. */ static int cmp_on_fail(enum action_fail_response first, enum action_fail_response second) { switch (first) { case action_fail_demote: switch (second) { case action_fail_ignore: return 1; case action_fail_demote: return 0; default: return -1; } break; case action_fail_reset_remote: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return 1; case action_fail_reset_remote: return 0; default: return -1; } break; case action_fail_restart_container: switch (second) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return 1; case action_fail_restart_container: return 0; default: return -1; } break; default: break; } switch (second) { case action_fail_demote: return (first == action_fail_ignore)? -1 : 1; case action_fail_reset_remote: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: return -1; default: return 1; } break; case action_fail_restart_container: switch (first) { case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_reset_remote: return -1; default: return 1; } break; default: break; } return first - second; } static void unpack_rsc_op_failure(pe_resource_t * rsc, pe_node_t * node, int rc, xmlNode * xml_op, xmlNode ** last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; pe_action_t *action = NULL; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return); *last_failure = xml_op; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; } if (exit_reason == NULL) { exit_reason = ""; } if (!pcmk_is_set(data_set->flags, pe_flag_symmetric_cluster) && (rc == PCMK_OCF_NOT_INSTALLED)) { crm_trace("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); } else { crm_warn("Unexpected result (%s%s%s) was recorded for " "%s of %s on %s at %s " CRM_XS " rc=%d id=%s", services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, (is_probe? "probe" : task), rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); if (is_probe && (rc != PCMK_OCF_OK) && (rc != PCMK_OCF_NOT_RUNNING) && (rc != PCMK_OCF_RUNNING_MASTER)) { /* A failed (not just unexpected) probe result could mean the user * didn't know resources will be probed even where they can't run. */ crm_notice("If it is not possible for %s to run on %s, see " "the resource-discovery option for location constraints", rsc->id, node->details->uname); } record_failed_op(xml_op, node, rsc, data_set); } action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); if (cmp_on_fail(*on_fail, action->on_fail) < 0) { pe_rsc_trace(rsc, "on-fail %s -> %s for %s (%s)", fail2text(*on_fail), fail2text(action->on_fail), action->uuid, key); *on_fail = action->on_fail; } if (!strcmp(task, CRMD_ACTION_STOP)) { resource_location(rsc, node, -INFINITY, "__stop_fail__", data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATE)) { unpack_migrate_to_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_MIGRATED)) { unpack_migrate_from_failure(rsc, node, xml_op, data_set); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_MASTER; } else if (!strcmp(task, CRMD_ACTION_DEMOTE)) { if (action->on_fail == action_fail_block) { rsc->role = RSC_ROLE_MASTER; rsc->next_role = RSC_ROLE_STOPPED; } else if(rc == PCMK_OCF_NOT_RUNNING) { rsc->role = RSC_ROLE_STOPPED; } else { /* Staying in master role would put the scheduler and controller * into a loop. Setting slave role is not dangerous because the * resource will be stopped as part of recovery, and any master * promotion will be ordered after that stop. */ rsc->role = RSC_ROLE_SLAVE; } } if(is_probe && rc == PCMK_OCF_NOT_INSTALLED) { /* leave stopped */ pe_rsc_trace(rsc, "Leaving %s stopped", rsc->id); rsc->role = RSC_ROLE_STOPPED; } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "Setting %s active", rsc->id); set_active(rsc); } pe_rsc_trace(rsc, "Resource %s: role=%s, unclean=%s, on_fail=%s, fail_role=%s", rsc->id, role2text(rsc->role), pcmk__btoa(node->details->unclean), fail2text(action->on_fail), role2text(action->fail_role)); if (action->fail_role != RSC_ROLE_STARTED && rsc->next_role < action->fail_role) { rsc->next_role = action->fail_role; } if (action->fail_role == RSC_ROLE_STOPPED) { int score = -INFINITY; pe_resource_t *fail_rsc = rsc; if (fail_rsc->parent) { pe_resource_t *parent = uber_parent(fail_rsc); if (pe_rsc_is_clone(parent) && !pcmk_is_set(parent->flags, pe_rsc_unique)) { /* For clone resources, if a child fails on an operation * with on-fail = stop, all the resources fail. Do this by preventing * the parent from coming up again. */ fail_rsc = parent; } } crm_notice("%s will not be started under current conditions", fail_rsc->id); /* make sure it doesn't come up again */ if (fail_rsc->allowed_nodes != NULL) { g_hash_table_destroy(fail_rsc->allowed_nodes); } fail_rsc->allowed_nodes = pe__node_list2table(data_set->nodes); g_hash_table_foreach(fail_rsc->allowed_nodes, set_node_score, &score); } pe_free_action(action); } /*! * \internal * \brief Remap operation status based on action result * * Given an action result, determine an appropriate operation status for the * purposes of responding to the action (the status provided by the executor is * not directly usable since the executor does not know what was expected). * * \param[in,out] rsc Resource that operation history entry is for * \param[in] rc Actual return code of operation * \param[in] target_rc Expected return code of operation * \param[in] node Node where operation was executed * \param[in] xml_op Operation history entry XML from CIB status * \param[in,out] on_fail What should be done about the result * \param[in] data_set Current cluster working set * * \return Operation status based on return code and action info * \note This may update the resource's current and next role. */ static int determine_op_status( pe_resource_t *rsc, int rc, int target_rc, pe_node_t * node, xmlNode * xml_op, enum action_fail_response * on_fail, pe_working_set_t * data_set) { guint interval_ms = 0; bool is_probe = false; int result = PCMK_LRM_OP_DONE; const char *key = get_op_key(xml_op); const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); CRM_ASSERT(rsc); CRM_CHECK(task != NULL, return PCMK_LRM_OP_ERROR); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((interval_ms == 0) && !strcmp(task, CRMD_ACTION_STATUS)) { is_probe = true; task = "probe"; } if (target_rc < 0) { /* Pre-1.0 Pacemaker versions, and Pacemaker 1.1.6 or earlier with * Heartbeat 2.0.7 or earlier as the cluster layer, did not include the * target_rc in the transition key, which (along with the similar case * of a corrupted transition key in the CIB) will be reported to this * function as -1. Pacemaker 2.0+ does not support rolling upgrades from * those versions or processing of saved CIB files from those versions, * so we do not need to care much about this case. */ result = PCMK_LRM_OP_ERROR; crm_warn("Expected result not found for %s on %s (corrupt or obsolete CIB?)", key, node->details->uname); } else if (target_rc != rc) { result = PCMK_LRM_OP_ERROR; pe_rsc_debug(rsc, "%s on %s: expected %d (%s), got %d (%s%s%s)", key, node->details->uname, target_rc, services_ocf_exitcode_str(target_rc), rc, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason); } switch (rc) { case PCMK_OCF_OK: if (is_probe && (target_rc == PCMK_OCF_NOT_RUNNING)) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Probe found %s active on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } break; case PCMK_OCF_NOT_RUNNING: if (is_probe || (target_rc == rc) || !pcmk_is_set(rsc->flags, pe_rsc_managed)) { result = PCMK_LRM_OP_DONE; rsc->role = RSC_ROLE_STOPPED; /* clear any previous failure actions */ *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; } break; case PCMK_OCF_RUNNING_MASTER: if (is_probe && (rc != target_rc)) { result = PCMK_LRM_OP_DONE; pe_rsc_info(rsc, "Probe found %s active and promoted on %s at %s", rsc->id, node->details->uname, last_change_str(xml_op)); } rsc->role = RSC_ROLE_MASTER; break; case PCMK_OCF_DEGRADED_MASTER: case PCMK_OCF_FAILED_MASTER: rsc->role = RSC_ROLE_MASTER; result = PCMK_LRM_OP_ERROR; break; case PCMK_OCF_NOT_CONFIGURED: result = PCMK_LRM_OP_ERROR_FATAL; break; case PCMK_OCF_UNIMPLEMENT_FEATURE: if (interval_ms > 0) { result = PCMK_LRM_OP_NOTSUPPORTED; break; } // fall through case PCMK_OCF_NOT_INSTALLED: case PCMK_OCF_INVALID_PARAM: case PCMK_OCF_INSUFFICIENT_PRIV: if (!pe_can_fence(data_set, node) && !strcmp(task, CRMD_ACTION_STOP)) { /* If a stop fails and we can't fence, there's nothing else we can do */ pe_proc_err("No further recovery can be attempted for %s " "because %s on %s failed (%s%s%s) at %s " CRM_XS " rc=%d id=%s", rsc->id, task, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, last_change_str(xml_op), rc, ID(xml_op)); pe__clear_resource_flags(rsc, pe_rsc_managed); pe__set_resource_flags(rsc, pe_rsc_block); } result = PCMK_LRM_OP_ERROR_HARD; break; default: if (result == PCMK_LRM_OP_DONE) { crm_info("Treating unknown exit status %d from %s of %s " "on %s at %s as failure", rc, task, rsc->id, node->details->uname, last_change_str(xml_op)); result = PCMK_LRM_OP_ERROR; } break; } return result; } // return TRUE if start or monitor last failure but parameters changed static bool should_clear_for_param_change(xmlNode *xml_op, const char *task, pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { if (!strcmp(task, "start") || !strcmp(task, "monitor")) { if (pe__bundle_needs_remote_name(rsc)) { /* We haven't allocated resources yet, so we can't reliably * substitute addr parameters for the REMOTE_CONTAINER_HACK. * When that's needed, defer the check until later. */ pe__add_param_check(xml_op, rsc, node, pe_check_last_failure, data_set); } else { op_digest_cache_t *digest_data = NULL; digest_data = rsc_action_digest_cmp(rsc, xml_op, node, data_set); switch (digest_data->rc) { case RSC_DIGEST_UNKNOWN: crm_trace("Resource %s history entry %s on %s" " has no digest to compare", rsc->id, get_op_key(xml_op), node->details->id); break; case RSC_DIGEST_MATCH: break; default: return TRUE; } } } return FALSE; } // Order action after fencing of remote node, given connection rsc static void order_after_remote_fencing(pe_action_t *action, pe_resource_t *remote_conn, pe_working_set_t *data_set) { pe_node_t *remote_node = pe_find_node(data_set->nodes, remote_conn->id); if (remote_node) { pe_action_t *fence = pe_fence_op(remote_node, NULL, TRUE, NULL, FALSE, data_set); order_actions(fence, action, pe_order_implies_then); } } static bool should_ignore_failure_timeout(pe_resource_t *rsc, xmlNode *xml_op, const char *task, guint interval_ms, bool is_last_failure, pe_working_set_t *data_set) { /* Clearing failures of recurring monitors has special concerns. The * executor reports only changes in the monitor result, so if the * monitor is still active and still getting the same failure result, * that will go undetected after the failure is cleared. * * Also, the operation history will have the time when the recurring * monitor result changed to the given code, not the time when the * result last happened. * * @TODO We probably should clear such failures only when the failure * timeout has passed since the last occurrence of the failed result. * However we don't record that information. We could maybe approximate * that by clearing only if there is a more recent successful monitor or * stop result, but we don't even have that information at this point * since we are still unpacking the resource's operation history. * * This is especially important for remote connection resources with a * reconnect interval, so in that case, we skip clearing failures * if the remote node hasn't been fenced. */ if (rsc->remote_reconnect_ms && pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && (interval_ms != 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { pe_node_t *remote_node = pe_find_node(data_set->nodes, rsc->id); if (remote_node && !remote_node->details->remote_was_fenced) { if (is_last_failure) { crm_info("Waiting to clear monitor failure for remote node %s" " until fencing has occurred", rsc->id); } return TRUE; } } return FALSE; } /*! * \internal * \brief Check operation age and schedule failure clearing when appropriate * * This function has two distinct purposes. The first is to check whether an * operation history entry is expired (i.e. the resource has a failure timeout, * the entry is older than the timeout, and the resource either has no fail * count or its fail count is entirely older than the timeout). The second is to * schedule fail count clearing when appropriate (i.e. the operation is expired * and either the resource has an expired fail count or the operation is a * last_failure for a remote connection resource with a reconnect interval, * or the operation is a last_failure for a start or monitor operation and the * resource's parameters have changed since the operation). * * \param[in] rsc Resource that operation happened to * \param[in] node Node that operation happened on * \param[in] rc Actual result of operation * \param[in] xml_op Operation history entry XML * \param[in] data_set Current working set * * \return TRUE if operation history entry is expired, FALSE otherwise */ static bool check_operation_expiry(pe_resource_t *rsc, pe_node_t *node, int rc, xmlNode *xml_op, pe_working_set_t *data_set) { bool expired = FALSE; bool is_last_failure = pcmk__ends_with(ID(xml_op), "_last_failure_0"); time_t last_run = 0; guint interval_ms = 0; int unexpired_fail_count = 0; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *clear_reason = NULL; crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); if ((rsc->failure_timeout > 0) && (crm_element_value_epoch(xml_op, XML_RSC_OP_LAST_CHANGE, &last_run) == 0)) { // Resource has a failure-timeout, and history entry has a timestamp time_t now = get_effective_time(data_set); time_t last_failure = 0; // Is this particular operation history older than the failure timeout? if ((now >= (last_run + rsc->failure_timeout)) && !should_ignore_failure_timeout(rsc, xml_op, task, interval_ms, is_last_failure, data_set)) { expired = TRUE; } // Does the resource as a whole have an unexpired fail count? unexpired_fail_count = pe_get_failcount(node, rsc, &last_failure, pe_fc_effective, xml_op, data_set); // Update scheduler recheck time according to *last* failure crm_trace("%s@%lld is %sexpired @%lld with unexpired_failures=%d timeout=%ds" " last-failure@%lld", ID(xml_op), (long long) last_run, (expired? "" : "not "), (long long) now, unexpired_fail_count, rsc->failure_timeout, (long long) last_failure); last_failure += rsc->failure_timeout + 1; if (unexpired_fail_count && (now < last_failure)) { pe__update_recheck_time(last_failure, data_set); } } if (expired) { if (pe_get_failcount(node, rsc, NULL, pe_fc_default, xml_op, data_set)) { // There is a fail count ignoring timeout if (unexpired_fail_count == 0) { // There is no fail count considering timeout clear_reason = "it expired"; } else { /* This operation is old, but there is an unexpired fail count. * In a properly functioning cluster, this should only be * possible if this operation is not a failure (otherwise the * fail count should be expired too), so this is really just a * failsafe. */ expired = FALSE; } } else if (is_last_failure && rsc->remote_reconnect_ms) { /* Clear any expired last failure when reconnect interval is set, * even if there is no fail count. */ clear_reason = "reconnect interval is set"; } } if (!expired && is_last_failure && should_clear_for_param_change(xml_op, task, rsc, node, data_set)) { clear_reason = "resource parameters have changed"; } if (clear_reason != NULL) { // Schedule clearing of the fail count pe_action_t *clear_op = pe__clear_failcount(rsc, node, clear_reason, data_set); if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled) && rsc->remote_reconnect_ms) { /* If we're clearing a remote connection due to a reconnect * interval, we want to wait until any scheduled fencing * completes. * * We could limit this to remote_node->details->unclean, but at * this point, that's always true (it won't be reliable until * after unpack_node_loop() is done). */ crm_info("Clearing %s failure will wait until any scheduled " "fencing of %s completes", task, rsc->id); order_after_remote_fencing(clear_op, rsc, data_set); } } if (expired && (interval_ms == 0) && pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { switch(rc) { case PCMK_OCF_OK: case PCMK_OCF_NOT_RUNNING: case PCMK_OCF_RUNNING_MASTER: case PCMK_OCF_DEGRADED: case PCMK_OCF_DEGRADED_MASTER: // Don't expire probes that return these values expired = FALSE; break; } } return expired; } int pe__target_rc_from_xml(xmlNode *xml_op) { int target_rc = 0; const char *key = crm_element_value(xml_op, XML_ATTR_TRANSITION_KEY); if (key == NULL) { return -1; } decode_transition_key(key, NULL, NULL, NULL, &target_rc); return target_rc; } static enum action_fail_response get_action_on_fail(pe_resource_t *rsc, const char *key, const char *task, pe_working_set_t * data_set) { enum action_fail_response result = action_fail_recover; pe_action_t *action = custom_action(rsc, strdup(key), task, NULL, TRUE, FALSE, data_set); result = action->on_fail; pe_free_action(action); return result; } static void update_resource_state(pe_resource_t * rsc, pe_node_t * node, xmlNode * xml_op, const char * task, int rc, xmlNode * last_failure, enum action_fail_response * on_fail, pe_working_set_t * data_set) { gboolean clear_past_failure = FALSE; CRM_ASSERT(rsc); CRM_ASSERT(xml_op); if (rc == PCMK_OCF_NOT_RUNNING) { clear_past_failure = TRUE; } else if (rc == PCMK_OCF_NOT_INSTALLED) { rsc->role = RSC_ROLE_STOPPED; } else if (pcmk__str_eq(task, CRMD_ACTION_STATUS, pcmk__str_casei)) { if (last_failure) { const char *op_key = get_op_key(xml_op); const char *last_failure_key = get_op_key(last_failure); if (pcmk__str_eq(op_key, last_failure_key, pcmk__str_casei)) { clear_past_failure = TRUE; } } if (rsc->role < RSC_ROLE_STARTED) { set_active(rsc); } } else if (pcmk__str_eq(task, CRMD_ACTION_START, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_STOP, pcmk__str_casei)) { rsc->role = RSC_ROLE_STOPPED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) { rsc->role = RSC_ROLE_MASTER; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_DEMOTE, pcmk__str_casei)) { if (*on_fail == action_fail_demote) { // Demote clears an error only if on-fail=demote clear_past_failure = TRUE; } rsc->role = RSC_ROLE_SLAVE; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATED, pcmk__str_casei)) { rsc->role = RSC_ROLE_STARTED; clear_past_failure = TRUE; } else if (pcmk__str_eq(task, CRMD_ACTION_MIGRATE, pcmk__str_casei)) { unpack_migrate_to_success(rsc, node, xml_op, data_set); } else if (rsc->role < RSC_ROLE_STARTED) { pe_rsc_trace(rsc, "%s active on %s", rsc->id, node->details->uname); set_active(rsc); } /* clear any previous failure actions */ if (clear_past_failure) { switch (*on_fail) { case action_fail_stop: case action_fail_fence: case action_fail_migrate: case action_fail_standby: pe_rsc_trace(rsc, "%s.%s is not cleared by a completed stop", rsc->id, fail2text(*on_fail)); break; case action_fail_block: case action_fail_ignore: case action_fail_demote: case action_fail_recover: case action_fail_restart_container: *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; break; case action_fail_reset_remote: if (rsc->remote_reconnect_ms == 0) { /* With no reconnect interval, the connection is allowed to * start again after the remote node is fenced and * completely stopped. (With a reconnect interval, we wait * for the failure to be cleared entirely before attempting * to reconnect.) */ *on_fail = action_fail_ignore; rsc->next_role = RSC_ROLE_UNKNOWN; } break; } } } /*! * \internal * \brief Remap informational monitor results to usual values * * Certain OCF result codes are for providing extended information to the * user about services that aren't yet failed but not entirely healthy either. * These must be treated as the "normal" result by pacemaker. * * \param[in] rc Actual result of a monitor action * \param[in] xml_op Operation history XML * \param[in] node Node that operation happened on * \param[in] rsc Resource that operation happened to * \param[in] data_set Cluster working set * * \return Result code that pacemaker should use * * \note If the result is remapped, and the node is not shutting down or failed, * the operation will be recorded in the data set's list of failed * operations, to highlight it for the user. */ static int remap_monitor_rc(int rc, xmlNode *xml_op, const pe_node_t *node, const pe_resource_t *rsc, pe_working_set_t *data_set) { int remapped_rc = pcmk__effective_rc(rc); if (rc != remapped_rc) { crm_trace("Remapping monitor result %d to %d", rc, remapped_rc); if (!node->details->shutdown || node->details->online) { record_failed_op(xml_op, node, rsc, data_set); } } return remapped_rc; } static void unpack_rsc_op(pe_resource_t *rsc, pe_node_t *node, xmlNode *xml_op, xmlNode **last_failure, enum action_fail_response *on_fail, pe_working_set_t *data_set) { int rc = 0; int task_id = 0; int target_rc = 0; int status = PCMK_LRM_OP_UNKNOWN; guint interval_ms = 0; const char *task = NULL; const char *task_key = NULL; const char *exit_reason = NULL; bool expired = FALSE; pe_resource_t *parent = rsc; enum action_fail_response failure_strategy = action_fail_recover; CRM_CHECK(rsc && node && xml_op, return); target_rc = pe__target_rc_from_xml(xml_op); task_key = get_op_key(xml_op); task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); exit_reason = crm_element_value(xml_op, XML_LRM_ATTR_EXIT_REASON); if (exit_reason == NULL) { exit_reason = ""; } crm_element_value_int(xml_op, XML_LRM_ATTR_RC, &rc); crm_element_value_int(xml_op, XML_LRM_ATTR_CALLID, &task_id); crm_element_value_int(xml_op, XML_LRM_ATTR_OPSTATUS, &status); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); CRM_CHECK(task != NULL, return); CRM_CHECK(status <= PCMK_LRM_OP_INVALID, return); CRM_CHECK(status >= PCMK_LRM_OP_PENDING, return); if (!strcmp(task, CRMD_ACTION_NOTIFY) || !strcmp(task, CRMD_ACTION_METADATA)) { /* safe to ignore these */ return; } if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { parent = uber_parent(rsc); } pe_rsc_trace(rsc, "Unpacking task %s/%s (call_id=%d, status=%d, rc=%d) on %s (role=%s)", task_key, task, task_id, status, rc, node->details->uname, role2text(rsc->role)); if (node->details->unclean) { pe_rsc_trace(rsc, "Node %s (where %s is running) is unclean." " Further action depends on the value of the stop's on-fail attribute", node->details->uname, rsc->id); } /* It should be possible to call remap_monitor_rc() first then call * check_operation_expiry() only if rc != target_rc, because there should * never be a fail count without at least one unexpected result in the * resource history. That would be more efficient by avoiding having to call * check_operation_expiry() for expected results. * * However, we do have such configurations in the scheduler regression * tests, even if it shouldn't be possible with the current code. It's * probably a good idea anyway, but that would require updating the test * inputs to something currently possible. */ if ((status != PCMK_LRM_OP_NOT_INSTALLED) && check_operation_expiry(rsc, node, rc, xml_op, data_set)) { expired = TRUE; } if (!strcmp(task, CRMD_ACTION_STATUS)) { rc = remap_monitor_rc(rc, xml_op, node, rsc, data_set); } if (expired && (rc != target_rc)) { const char *magic = crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC); if (interval_ms == 0) { crm_notice("Ignoring expired %s failure on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); goto done; } else if(node->details->online && node->details->unclean == FALSE) { /* Reschedule the recurring monitor. CancelXmlOp() won't work at * this stage, so as a hacky workaround, forcibly change the restart * digest so check_action_definition() does what we want later. * * @TODO We should skip this if there is a newer successful monitor. * Also, this causes rescheduling only if the history entry * has an op-digest (which the expire-non-blocked-failure * scheduler regression test doesn't, but that may not be a * realistic scenario in production). */ crm_notice("Rescheduling %s after failure expired on %s " CRM_XS " actual=%d expected=%d magic=%s", task_key, node->details->uname, rc, target_rc, magic); crm_xml_add(xml_op, XML_LRM_ATTR_RESTART_DIGEST, "calculated-failure-timeout"); goto done; } } /* If the executor reported an operation status of anything but done or * error, consider that final. But for done or error, we know better whether * it should be treated as a failure or not, because we know the expected * result. */ if(status == PCMK_LRM_OP_DONE || status == PCMK_LRM_OP_ERROR) { status = determine_op_status(rsc, rc, target_rc, node, xml_op, on_fail, data_set); pe_rsc_trace(rsc, "Remapped %s status to %d", task_key, status); } switch (status) { case PCMK_LRM_OP_CANCELLED: // Should never happen pe_err("Resource history contains cancellation '%s' " "(%s of %s on %s at %s)", ID(xml_op), task, rsc->id, node->details->uname, last_change_str(xml_op)); break; case PCMK_LRM_OP_PENDING: if (!strcmp(task, CRMD_ACTION_START)) { pe__set_resource_flags(rsc, pe_rsc_start_pending); set_active(rsc); } else if (!strcmp(task, CRMD_ACTION_PROMOTE)) { rsc->role = RSC_ROLE_MASTER; } else if (!strcmp(task, CRMD_ACTION_MIGRATE) && node->details->unclean) { /* If a pending migrate_to action is out on a unclean node, * we have to force the stop action on the target. */ const char *migrate_target = crm_element_value(xml_op, XML_LRM_ATTR_MIGRATE_TARGET); pe_node_t *target = pe_find_node(data_set->nodes, migrate_target); if (target) { stop_action(rsc, target, FALSE); } } if (rsc->pending_task == NULL) { if ((interval_ms != 0) || strcmp(task, CRMD_ACTION_STATUS)) { rsc->pending_task = strdup(task); rsc->pending_node = node; } else { /* Pending probes are not printed, even if pending * operations are requested. If someone ever requests that * behavior, enable the below and the corresponding part of * native.c:native_pending_task(). */ #if 0 rsc->pending_task = strdup("probe"); rsc->pending_node = node; #endif } } break; case PCMK_LRM_OP_DONE: pe_rsc_trace(rsc, "%s of %s on %s completed at %s " CRM_XS " id=%s", task, rsc->id, node->details->uname, last_change_str(xml_op), ID(xml_op)); update_resource_state(rsc, node, xml_op, task, rc, *last_failure, on_fail, data_set); break; case PCMK_LRM_OP_NOT_INSTALLED: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if (failure_strategy == action_fail_ignore) { crm_warn("Cannot ignore failed %s of %s on %s: " "Resource agent doesn't exist " CRM_XS " status=%d rc=%d id=%s", task, rsc->id, node->details->uname, status, rc, ID(xml_op)); /* Also for printing it as "FAILED" by marking it as pe_rsc_failed later */ *on_fail = action_fail_migrate; } resource_location(parent, node, -INFINITY, "hard-error", data_set); unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); break; case PCMK_LRM_OP_NOT_CONNECTED: if (pe__is_guest_or_remote_node(node) && pcmk_is_set(node->details->remote_rsc->flags, pe_rsc_managed)) { /* We should never get into a situation where a managed remote * connection resource is considered OK but a resource action * behind the connection gets a "not connected" status. But as a * fail-safe in case a bug or unusual circumstances do lead to * that, ensure the remote connection is considered failed. */ pe__set_resource_flags(node->details->remote_rsc, pe_rsc_failed|pe_rsc_stop); } // fall through case PCMK_LRM_OP_ERROR: case PCMK_LRM_OP_ERROR_HARD: case PCMK_LRM_OP_ERROR_FATAL: case PCMK_LRM_OP_TIMEOUT: case PCMK_LRM_OP_NOTSUPPORTED: case PCMK_LRM_OP_INVALID: failure_strategy = get_action_on_fail(rsc, task_key, task, data_set); if ((failure_strategy == action_fail_ignore) || (failure_strategy == action_fail_restart_container && !strcmp(task, CRMD_ACTION_STOP))) { crm_warn("Pretending failed %s (%s%s%s) of %s on %s at %s " "succeeded " CRM_XS " rc=%d id=%s", task, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rsc->id, node->details->uname, last_change_str(xml_op), rc, ID(xml_op)); update_resource_state(rsc, node, xml_op, task, target_rc, *last_failure, on_fail, data_set); crm_xml_add(xml_op, XML_ATTR_UNAME, node->details->uname); pe__set_resource_flags(rsc, pe_rsc_failure_ignored); record_failed_op(xml_op, node, rsc, data_set); if ((failure_strategy == action_fail_restart_container) && cmp_on_fail(*on_fail, action_fail_recover) <= 0) { *on_fail = failure_strategy; } } else { unpack_rsc_op_failure(rsc, node, rc, xml_op, last_failure, on_fail, data_set); if(status == PCMK_LRM_OP_ERROR_HARD) { do_crm_log(rc != PCMK_OCF_NOT_INSTALLED?LOG_ERR:LOG_NOTICE, "Preventing %s from restarting on %s because " "of hard failure (%s%s%s)" CRM_XS " rc=%d id=%s", parent->id, node->details->uname, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, node, -INFINITY, "hard-error", data_set); } else if(status == PCMK_LRM_OP_ERROR_FATAL) { crm_err("Preventing %s from restarting anywhere because " "of fatal failure (%s%s%s) " CRM_XS " rc=%d id=%s", parent->id, services_ocf_exitcode_str(rc), (*exit_reason? ": " : ""), exit_reason, rc, ID(xml_op)); resource_location(parent, NULL, -INFINITY, "fatal-error", data_set); } } break; } done: pe_rsc_trace(rsc, "Resource %s after %s: role=%s, next=%s", rsc->id, task, role2text(rsc->role), role2text(rsc->next_role)); } static void add_node_attrs(xmlNode *xml_obj, pe_node_t *node, bool overwrite, pe_working_set_t *data_set) { const char *cluster_name = NULL; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_UNAME), strdup(node->details->uname)); g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_ID), strdup(node->details->id)); if (pcmk__str_eq(node->details->id, data_set->dc_uuid, pcmk__str_casei)) { data_set->dc_node = node; node->details->is_dc = TRUE; g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_TRUE)); } else { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_IS_DC), strdup(XML_BOOLEAN_FALSE)); } cluster_name = g_hash_table_lookup(data_set->config_hash, "cluster-name"); if (cluster_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_CLUSTER_NAME), strdup(cluster_name)); } pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_ATTR_SETS, &rule_data, node->details->attrs, NULL, overwrite, data_set); if (pe_node_attribute_raw(node, CRM_ATTR_SITE_NAME) == NULL) { const char *site_name = pe_node_attribute_raw(node, "site-name"); if (site_name) { g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(site_name)); } else if (cluster_name) { /* Default to cluster-name if unset */ g_hash_table_insert(node->details->attrs, strdup(CRM_ATTR_SITE_NAME), strdup(cluster_name)); } } } static GListPtr extract_operations(const char *node, const char *rsc, xmlNode * rsc_entry, gboolean active_filter) { int counter = -1; int stop_index = -1; int start_index = -1; xmlNode *rsc_op = NULL; GListPtr gIter = NULL; GListPtr op_list = NULL; GListPtr sorted_op_list = NULL; /* extract operations */ op_list = NULL; sorted_op_list = NULL; for (rsc_op = pcmk__xe_first_child(rsc_entry); rsc_op != NULL; rsc_op = pcmk__xe_next(rsc_op)) { if (pcmk__str_eq((const char *)rsc_op->name, XML_LRM_TAG_RSC_OP, pcmk__str_none)) { crm_xml_add(rsc_op, "resource", rsc); crm_xml_add(rsc_op, XML_ATTR_UNAME, node); op_list = g_list_prepend(op_list, rsc_op); } } if (op_list == NULL) { /* if there are no operations, there is nothing to do */ return NULL; } sorted_op_list = g_list_sort(op_list, sort_op_by_callid); /* create active recurring operations as optional */ if (active_filter == FALSE) { return sorted_op_list; } op_list = NULL; calculate_active_ops(sorted_op_list, &start_index, &stop_index); for (gIter = sorted_op_list; gIter != NULL; gIter = gIter->next) { xmlNode *rsc_op = (xmlNode *) gIter->data; counter++; if (start_index < stop_index) { crm_trace("Skipping %s: not active", ID(rsc_entry)); break; } else if (counter < start_index) { crm_trace("Skipping %s: old", ID(rsc_op)); continue; } op_list = g_list_append(op_list, rsc_op); } g_list_free(sorted_op_list); return op_list; } GListPtr find_operations(const char *rsc, const char *node, gboolean active_filter, pe_working_set_t * data_set) { GListPtr output = NULL; GListPtr intermediate = NULL; xmlNode *tmp = NULL; xmlNode *status = find_xml_node(data_set->input, XML_CIB_TAG_STATUS, TRUE); pe_node_t *this_node = NULL; xmlNode *node_state = NULL; for (node_state = pcmk__xe_first_child(status); node_state != NULL; node_state = pcmk__xe_next(node_state)) { if (pcmk__str_eq((const char *)node_state->name, XML_CIB_TAG_STATE, pcmk__str_none)) { const char *uname = crm_element_value(node_state, XML_ATTR_UNAME); if (node != NULL && !pcmk__str_eq(uname, node, pcmk__str_casei)) { continue; } this_node = pe_find_node(data_set->nodes, uname); if(this_node == NULL) { CRM_LOG_ASSERT(this_node != NULL); continue; } else if (pe__is_guest_or_remote_node(this_node)) { determine_remote_online_status(data_set, this_node); } else { determine_online_status(node_state, this_node, data_set); } if (this_node->details->online || pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { /* offline nodes run no resources... * unless stonith is enabled in which case we need to * make sure rsc start events happen after the stonith */ xmlNode *lrm_rsc = NULL; tmp = find_xml_node(node_state, XML_CIB_TAG_LRM, FALSE); tmp = find_xml_node(tmp, XML_LRM_TAG_RESOURCES, FALSE); for (lrm_rsc = pcmk__xe_first_child(tmp); lrm_rsc != NULL; lrm_rsc = pcmk__xe_next(lrm_rsc)) { if (pcmk__str_eq((const char *)lrm_rsc->name, XML_LRM_TAG_RESOURCE, pcmk__str_none)) { const char *rsc_id = crm_element_value(lrm_rsc, XML_ATTR_ID); if (rsc != NULL && !pcmk__str_eq(rsc_id, rsc, pcmk__str_casei)) { continue; } intermediate = extract_operations(uname, rsc_id, lrm_rsc, active_filter); output = g_list_concat(output, intermediate); } } } } } return output; } diff --git a/lib/pengine/utils.c b/lib/pengine/utils.c index 2ad87800cf..fd238dfca9 100644 --- a/lib/pengine/utils.c +++ b/lib/pengine/utils.c @@ -1,2840 +1,2841 @@ /* * 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 #include #include #include #include #include #include #include #include #include extern xmlNode *get_object_root(const char *object_type, xmlNode * the_root); void print_str_str(gpointer key, gpointer value, gpointer user_data); gboolean ghash_free_str_str(gpointer key, gpointer value, gpointer user_data); static void unpack_operation(pe_action_t * action, xmlNode * xml_obj, pe_resource_t * container, pe_working_set_t * data_set, guint interval_ms); static xmlNode *find_rsc_op_entry_helper(pe_resource_t * rsc, const char *key, gboolean include_disabled); #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t * pe_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; CRM_CHECK(action != NULL, return NULL); if (action->action_details == NULL) { action->action_details = calloc(1, sizeof(pe_rsc_action_details_t)); CRM_CHECK(action->action_details != NULL, return NULL); } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters == NULL) { details->versioned_parameters = create_xml_node(NULL, XML_TAG_OP_VER_ATTRS); } if (details->versioned_meta == NULL) { details->versioned_meta = create_xml_node(NULL, XML_TAG_OP_VER_META); } return details; } static void pe_free_rsc_action_details(pe_action_t *action) { pe_rsc_action_details_t *details; if ((action == NULL) || (action->action_details == NULL)) { return; } details = (pe_rsc_action_details_t *) action->action_details; if (details->versioned_parameters) { free_xml(details->versioned_parameters); } if (details->versioned_meta) { free_xml(details->versioned_meta); } action->action_details = NULL; } #endif /*! * \internal * \brief Check whether we can fence a particular node * * \param[in] data_set Working set for cluster * \param[in] node Name of node to check * * \return true if node can be fenced, false otherwise */ bool pe_can_fence(pe_working_set_t *data_set, pe_node_t *node) { if (pe__is_guest_node(node)) { /* Guest nodes are fenced by stopping their container resource. We can * do that if the container's host is either online or fenceable. */ pe_resource_t *rsc = node->details->remote_rsc->container; for (GList *n = rsc->running_on; n != NULL; n = n->next) { pe_node_t *container_node = n->data; if (!container_node->details->online && !pe_can_fence(data_set, container_node)) { return false; } } return true; } else if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { return false; /* Turned off */ } else if (!pcmk_is_set(data_set->flags, pe_flag_have_stonith_resource)) { return false; /* No devices */ } else if (pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { return true; } else if (data_set->no_quorum_policy == no_quorum_ignore) { return true; } else if(node == NULL) { return false; } else if(node->details->online) { crm_notice("We can fence %s without quorum because they're in our membership", node->details->uname); return true; } crm_trace("Cannot fence %s", node->details->uname); return false; } /*! * \internal * \brief Copy a node object * * \param[in] this_node Node object to copy * * \return Newly allocated shallow copy of this_node * \note This function asserts on errors and is guaranteed to return non-NULL. */ pe_node_t * pe__copy_node(const pe_node_t *this_node) { pe_node_t *new_node = NULL; CRM_ASSERT(this_node != NULL); new_node = calloc(1, sizeof(pe_node_t)); CRM_ASSERT(new_node != NULL); new_node->rsc_discover_mode = this_node->rsc_discover_mode; new_node->weight = this_node->weight; new_node->fixed = this_node->fixed; new_node->details = this_node->details; return new_node; } /* any node in list1 or list2 and not in the other gets a score of -INFINITY */ void node_list_exclude(GHashTable * hash, GListPtr list, gboolean merge_scores) { GHashTable *result = hash; pe_node_t *other_node = NULL; GListPtr gIter = list; GHashTableIter iter; pe_node_t *node = NULL; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { other_node = pe_find_node_id(list, node->details->id); if (other_node == NULL) { node->weight = -INFINITY; } else if (merge_scores) { node->weight = pe__add_scores(node->weight, other_node->weight); } } for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; other_node = pe_hash_table_lookup(result, node->details->id); if (other_node == NULL) { pe_node_t *new_node = pe__copy_node(node); new_node->weight = -INFINITY; g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } } } /*! * \internal * \brief Create a node hash table from a node list * * \param[in] list Node list * * \return Hash table equivalent of node list */ GHashTable * pe__node_list2table(GList *list) { GHashTable *result = NULL; result = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, free); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { pe_node_t *new_node = pe__copy_node((pe_node_t *) gIter->data); g_hash_table_insert(result, (gpointer) new_node->details->id, new_node); } return result; } gint sort_node_uname(gconstpointer a, gconstpointer b) { return pcmk_numeric_strcasecmp(((const pe_node_t *) a)->details->uname, ((const pe_node_t *) b)->details->uname); } /*! * \internal * \brief Output node weights to stdout * * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes */ static void pe__output_node_weights(pe_resource_t *rsc, const char *comment, GHashTable *nodes) { char score[128]; // Stack-allocated since this is called frequently // Sort the nodes so the output is consistent for regression tests GList *list = g_list_sort(g_hash_table_get_values(nodes), sort_node_uname); for (GList *gIter = list; gIter != NULL; gIter = gIter->next) { pe_node_t *node = (pe_node_t *) gIter->data; score2char_stack(node->weight, score, sizeof(score)); if (rsc) { printf("%s: %s allocation score on %s: %s\n", comment, rsc->id, node->details->uname, score); } else { printf("%s: %s = %s\n", comment, node->details->uname, score); } } g_list_free(list); } /*! * \internal * \brief Log node weights at trace level * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes */ static void pe__log_node_weights(const char *file, const char *function, int line, pe_resource_t *rsc, const char *comment, GHashTable *nodes) { GHashTableIter iter; pe_node_t *node = NULL; char score[128]; // Stack-allocated since this is called frequently // Don't waste time if we're not tracing at this point pcmk__log_else(LOG_TRACE, return); g_hash_table_iter_init(&iter, nodes); while (g_hash_table_iter_next(&iter, NULL, (void **) &node)) { score2char_stack(node->weight, score, sizeof(score)); if (rsc) { qb_log_from_external_source(function, file, "%s: %s allocation score on %s: %s", LOG_TRACE, line, 0, comment, rsc->id, node->details->uname, score); } else { qb_log_from_external_source(function, file, "%s: %s = %s", LOG_TRACE, line, 0, comment, node->details->uname, score); } } } /*! * \internal * \brief Log or output node weights * * \param[in] file Caller's filename * \param[in] function Caller's function name * \param[in] line Caller's line number * \param[in] to_log Log if true, otherwise output * \param[in] rsc Use allowed nodes for this resource * \param[in] comment Text description to prefix lines with * \param[in] nodes If rsc is not specified, use these nodes */ void pe__show_node_weights_as(const char *file, const char *function, int line, bool to_log, pe_resource_t *rsc, const char *comment, GHashTable *nodes) { if (rsc != NULL) { if (pcmk_is_set(rsc->flags, pe_rsc_orphan)) { // Don't show allocation scores for orphans return; } nodes = rsc->allowed_nodes; } if (nodes == NULL) { // Nothing to show return; } if (to_log) { pe__log_node_weights(file, function, line, rsc, comment, nodes); } else { pe__output_node_weights(rsc, comment, nodes); } // If this resource has children, repeat recursively for each if (rsc && rsc->children) { for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe_resource_t *child = (pe_resource_t *) gIter->data; pe__show_node_weights_as(file, function, line, to_log, child, comment, nodes); } } } static void append_dump_text(gpointer key, gpointer value, gpointer user_data) { char **dump_text = user_data; char *new_text = crm_strdup_printf("%s %s=%s", *dump_text, (char *)key, (char *)value); free(*dump_text); *dump_text = new_text; } void dump_node_capacity(int level, const char *comment, pe_node_t * node) { char *dump_text = crm_strdup_printf("%s: %s capacity:", comment, node->details->uname); g_hash_table_foreach(node->details->utilization, append_dump_text, &dump_text); if (level == LOG_STDOUT) { fprintf(stdout, "%s\n", dump_text); } else { crm_trace("%s", dump_text); } free(dump_text); } void dump_rsc_utilization(int level, const char *comment, pe_resource_t * rsc, pe_node_t * node) { char *dump_text = crm_strdup_printf("%s: %s utilization on %s:", comment, rsc->id, node->details->uname); g_hash_table_foreach(rsc->utilization, append_dump_text, &dump_text); switch (level) { case LOG_STDOUT: fprintf(stdout, "%s\n", dump_text); break; case LOG_NEVER: break; default: crm_trace("%s", dump_text); } free(dump_text); } gint sort_rsc_index(gconstpointer a, gconstpointer b) { const pe_resource_t *resource1 = (const pe_resource_t *)a; const pe_resource_t *resource2 = (const pe_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->sort_index > resource2->sort_index) { return -1; } if (resource1->sort_index < resource2->sort_index) { return 1; } return 0; } gint sort_rsc_priority(gconstpointer a, gconstpointer b) { const pe_resource_t *resource1 = (const pe_resource_t *)a; const pe_resource_t *resource2 = (const pe_resource_t *)b; if (a == NULL && b == NULL) { return 0; } if (a == NULL) { return 1; } if (b == NULL) { return -1; } if (resource1->priority > resource2->priority) { return -1; } if (resource1->priority < resource2->priority) { return 1; } return 0; } static enum pe_quorum_policy effective_quorum_policy(pe_resource_t *rsc, pe_working_set_t *data_set) { enum pe_quorum_policy policy = data_set->no_quorum_policy; if (pcmk_is_set(data_set->flags, pe_flag_have_quorum)) { policy = no_quorum_ignore; } else if (data_set->no_quorum_policy == no_quorum_demote) { switch (rsc->role) { case RSC_ROLE_MASTER: case RSC_ROLE_SLAVE: if (rsc->next_role > RSC_ROLE_SLAVE) { rsc->next_role = RSC_ROLE_SLAVE; } policy = no_quorum_ignore; break; default: policy = no_quorum_stop; break; } } return policy; } pe_action_t * custom_action(pe_resource_t * rsc, char *key, const char *task, pe_node_t * on_node, gboolean optional, gboolean save_action, pe_working_set_t * data_set) { pe_action_t *action = NULL; GListPtr possible_matches = NULL; CRM_CHECK(key != NULL, return NULL); CRM_CHECK(task != NULL, free(key); return NULL); if (save_action && rsc != NULL) { possible_matches = find_actions(rsc->actions, key, on_node); } else if(save_action) { #if 0 action = g_hash_table_lookup(data_set->singletons, key); #else /* More expensive but takes 'node' into account */ possible_matches = find_actions(data_set->actions, key, on_node); #endif } if(data_set->singletons == NULL) { data_set->singletons = g_hash_table_new_full(crm_str_hash, g_str_equal, NULL, NULL); } if (possible_matches != NULL) { if (pcmk__list_of_multiple(possible_matches)) { pe_warn("Action %s for %s on %s exists %d times", task, rsc ? rsc->id : "", on_node ? on_node->details->uname : "", g_list_length(possible_matches)); } action = g_list_nth_data(possible_matches, 0); pe_rsc_trace(rsc, "Found action %d: %s for %s (%s) on %s", action->id, task, (rsc? rsc->id : "no resource"), action->uuid, (on_node? on_node->details->uname : "no node")); g_list_free(possible_matches); } if (action == NULL) { if (save_action) { pe_rsc_trace(rsc, "Creating action %d (%s): %s for %s (%s) on %s", data_set->action_id, (optional? "optional" : "required"), task, (rsc? rsc->id : "no resource"), key, (on_node? on_node->details->uname : "no node")); } action = calloc(1, sizeof(pe_action_t)); if (save_action) { action->id = data_set->action_id++; } else { action->id = 0; } action->rsc = rsc; action->task = strdup(task); if (on_node) { action->node = pe__copy_node(on_node); } action->uuid = strdup(key); if (pcmk__str_eq(task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { // Resource history deletion for a node can be done on the DC pe__set_action_flags(action, pe_action_dc); } pe__set_action_flags(action, pe_action_runnable); if (optional) { pe__set_action_flags(action, pe_action_optional); } else { pe__clear_action_flags(action, pe_action_optional); } action->extra = crm_str_table_new(); action->meta = crm_str_table_new(); if (save_action) { data_set->actions = g_list_prepend(data_set->actions, action); if(rsc == NULL) { g_hash_table_insert(data_set->singletons, action->uuid, action); } } if (rsc != NULL) { guint interval_ms = 0; action->op_entry = find_rsc_op_entry_helper(rsc, key, TRUE); parse_op_key(key, NULL, NULL, &interval_ms); unpack_operation(action, action->op_entry, rsc->container, data_set, interval_ms); if (save_action) { rsc->actions = g_list_prepend(rsc->actions, action); } } } if (!optional && pcmk_is_set(action->flags, pe_action_optional)) { pe__clear_action_flags(action, pe_action_optional); } if (rsc != NULL) { enum action_tasks a_task = text2task(action->task); enum pe_quorum_policy quorum_policy = effective_quorum_policy(rsc, data_set); int warn_level = LOG_TRACE; if (save_action) { warn_level = LOG_WARNING; } if (!pcmk_is_set(action->flags, pe_action_have_node_attrs) && action->node != NULL && action->op_entry != NULL) { pe_rule_eval_data_t rule_data = { .node_hash = action->node->details->attrs, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; pe__set_action_flags(action, pe_action_have_node_attrs); pe__unpack_dataset_nvpairs(action->op_entry, XML_TAG_ATTR_SETS, &rule_data, action->extra, NULL, FALSE, data_set); } if (pcmk_is_set(action->flags, pe_action_pseudo)) { /* leave untouched */ } else if (action->node == NULL) { pe_rsc_trace(rsc, "%s is unrunnable (unallocated)", action->uuid); pe__clear_action_flags(action, pe_action_runnable); } else if (!pcmk_is_set(rsc->flags, pe_rsc_managed) && g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS) == NULL) { pe_rsc_debug(rsc, "%s on %s is optional (%s is unmanaged)", action->uuid, action->node->details->uname, rsc->id); pe__set_action_flags(action, pe_action_optional); //pe__clear_action_flags(action, pe_action_runnable); } else if (!pcmk_is_set(action->flags, pe_action_dc) && !(action->node->details->online) && (!pe__is_guest_node(action->node) || action->node->details->remote_requires_reset)) { pe__clear_action_flags(action, pe_action_runnable); do_crm_log(warn_level, "%s on %s is unrunnable (node is offline)", action->uuid, action->node->details->uname); if (pcmk_is_set(action->rsc->flags, pe_rsc_managed) && save_action && a_task == stop_rsc && action->node->details->unclean == FALSE) { pe_fence_node(data_set, action->node, "resource actions are unrunnable", FALSE); } } else if (!pcmk_is_set(action->flags, pe_action_dc) && action->node->details->pending) { pe__clear_action_flags(action, pe_action_runnable); do_crm_log(warn_level, "Action %s on %s is unrunnable (node is pending)", action->uuid, action->node->details->uname); } else if (action->needs == rsc_req_nothing) { pe_action_set_reason(action, NULL, TRUE); if (pe__is_guest_node(action->node) && !pe_can_fence(data_set, action->node)) { /* An action that requires nothing usually does not require any * fencing in order to be runnable. However, there is an * exception: an action cannot be completed if it is on a guest * node whose host is unclean and cannot be fenced. */ pe_rsc_debug(rsc, "%s on %s is unrunnable " "(node's host cannot be fenced)", action->uuid, action->node->details->uname); pe__clear_action_flags(action, pe_action_runnable); } else { pe_rsc_trace(rsc, "%s on %s does not require fencing or quorum", action->uuid, action->node->details->uname); pe__set_action_flags(action, pe_action_runnable); } #if 0 /* * No point checking this * - if we don't have quorum we can't stonith anyway */ } else if (action->needs == rsc_req_stonith) { crm_trace("Action %s requires only stonith", action->uuid); action->runnable = TRUE; #endif } else if (quorum_policy == no_quorum_stop) { pe_rsc_debug(rsc, "%s on %s is unrunnable (no quorum)", action->uuid, action->node->details->uname); pe_action_set_flag_reason(__func__, __LINE__, action, NULL, "no quorum", pe_action_runnable, TRUE); } else if (quorum_policy == no_quorum_freeze) { if (rsc->fns->active(rsc, TRUE) == FALSE || rsc->next_role > rsc->role) { pe_rsc_debug(rsc, "%s on %s is unrunnable (no quorum)", action->uuid, action->node->details->uname); pe_action_set_flag_reason(__func__, __LINE__, action, NULL, "quorum freeze", pe_action_runnable, TRUE); } } else { //pe_action_set_reason(action, NULL, TRUE); pe__set_action_flags(action, pe_action_runnable); } if (save_action) { switch (a_task) { case stop_rsc: pe__set_resource_flags(rsc, pe_rsc_stopping); break; case start_rsc: pe__clear_resource_flags(rsc, pe_rsc_starting); if (pcmk_is_set(action->flags, pe_action_runnable)) { pe__set_resource_flags(rsc, pe_rsc_starting); } break; default: break; } } } free(key); return action; } static bool valid_stop_on_fail(const char *value) { return !pcmk__strcase_any_of(value, "standby", "demote", "stop", NULL); } static const char * unpack_operation_on_fail(pe_action_t * action) { const char *name = NULL; const char *role = NULL; const char *on_fail = NULL; const char *interval_spec = NULL; const char *enabled = NULL; const char *value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ON_FAIL); if (pcmk__str_eq(action->task, CRMD_ACTION_STOP, pcmk__str_casei) && !valid_stop_on_fail(value)) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s stop " "action to default value because '%s' is not " "allowed for stop", action->rsc->id, value); return NULL; } else if (pcmk__str_eq(action->task, CRMD_ACTION_DEMOTE, pcmk__str_casei) && !value) { /* demote on_fail defaults to master monitor value if present */ xmlNode *operation = NULL; CRM_CHECK(action->rsc != NULL, return NULL); for (operation = pcmk__xe_first_child(action->rsc->ops_xml); (operation != NULL) && (value == NULL); operation = pcmk__xe_next(operation)) { if (!pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { continue; } name = crm_element_value(operation, "name"); role = crm_element_value(operation, "role"); on_fail = crm_element_value(operation, XML_OP_ATTR_ON_FAIL); enabled = crm_element_value(operation, "enabled"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); if (!on_fail) { continue; } else if (enabled && !crm_is_true(enabled)) { continue; } else if (!pcmk__str_eq(name, "monitor", pcmk__str_casei) || !pcmk__str_eq(role, "Master", pcmk__str_casei)) { continue; } else if (crm_parse_interval_spec(interval_spec) == 0) { continue; } else if (pcmk__str_eq(on_fail, "demote", pcmk__str_casei)) { continue; } value = on_fail; } } else if (pcmk__str_eq(action->task, CRM_OP_LRM_DELETE, pcmk__str_casei)) { value = "ignore"; } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { name = crm_element_value(action->op_entry, "name"); role = crm_element_value(action->op_entry, "role"); interval_spec = crm_element_value(action->op_entry, XML_LRM_ATTR_INTERVAL); if (!pcmk__str_eq(name, CRMD_ACTION_PROMOTE, pcmk__str_casei) && (!pcmk__str_eq(name, CRMD_ACTION_STATUS, pcmk__str_casei) || !pcmk__str_eq(role, "Master", pcmk__str_casei) || (crm_parse_interval_spec(interval_spec) == 0))) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for %s %s " "action to default value because 'demote' is not " "allowed for it", action->rsc->id, name); return NULL; } } return value; } static xmlNode * find_min_interval_mon(pe_resource_t * rsc, gboolean include_disabled) { guint interval_ms = 0; guint min_interval_ms = G_MAXUINT; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; xmlNode *op = NULL; xmlNode *operation = NULL; for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL; operation = pcmk__xe_next(operation)) { if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } if (!pcmk__str_eq(name, RSC_STATUS, pcmk__str_casei)) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); if (interval_ms && (interval_ms < min_interval_ms)) { min_interval_ms = interval_ms; op = operation; } } } return op; } static int unpack_start_delay(const char *value, GHashTable *meta) { int start_delay = 0; if (value != NULL) { start_delay = crm_get_msec(value); if (start_delay < 0) { start_delay = 0; } if (meta) { g_hash_table_replace(meta, strdup(XML_OP_ATTR_START_DELAY), crm_itoa(start_delay)); } } return start_delay; } // true if value contains valid, non-NULL interval origin for recurring op static bool unpack_interval_origin(const char *value, xmlNode *xml_obj, guint interval_ms, crm_time_t *now, long long *start_delay) { long long result = 0; guint interval_sec = interval_ms / 1000; crm_time_t *origin = NULL; // Ignore unspecified values and non-recurring operations if ((value == NULL) || (interval_ms == 0) || (now == NULL)) { return false; } // Parse interval origin from text origin = crm_time_new(value); if (origin == NULL) { pcmk__config_err("Ignoring '" XML_OP_ATTR_ORIGIN "' for operation " "'%s' because '%s' is not valid", (ID(xml_obj)? ID(xml_obj) : "(missing ID)"), value); return false; } // Get seconds since origin (negative if origin is in the future) result = crm_time_get_seconds(now) - crm_time_get_seconds(origin); crm_time_free(origin); // Calculate seconds from closest interval to now result = result % interval_sec; // Calculate seconds remaining until next interval result = ((result <= 0)? 0 : interval_sec) - result; crm_info("Calculated a start delay of %llds for operation '%s'", result, (ID(xml_obj)? ID(xml_obj) : "(unspecified)")); if (start_delay != NULL) { *start_delay = result * 1000; // milliseconds } return true; } static int unpack_timeout(const char *value) { int timeout_ms = crm_get_msec(value); if (timeout_ms < 0) { timeout_ms = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } return timeout_ms; } int pe_get_configured_timeout(pe_resource_t *rsc, const char *action, pe_working_set_t *data_set) { xmlNode *child = NULL; GHashTable *action_meta = NULL; const char *timeout_spec = NULL; int timeout_ms = 0; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = NULL, .op_data = NULL }; for (child = first_named_child(rsc->ops_xml, XML_ATTR_OP); child != NULL; child = crm_next_same_xml(child)) { if (pcmk__str_eq(action, crm_element_value(child, XML_NVPAIR_ATTR_NAME), pcmk__str_casei)) { timeout_spec = crm_element_value(child, XML_ATTR_TIMEOUT); break; } } if (timeout_spec == NULL && data_set->op_defaults) { action_meta = crm_str_table_new(); pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, &rule_data, action_meta, NULL, FALSE, data_set); timeout_spec = g_hash_table_lookup(action_meta, XML_ATTR_TIMEOUT); } // @TODO check meta-attributes (including versioned meta-attributes) // @TODO maybe use min-interval monitor timeout as default for monitors timeout_ms = crm_get_msec(timeout_spec); if (timeout_ms < 0) { timeout_ms = crm_get_msec(CRM_DEFAULT_OP_TIMEOUT_S); } if (action_meta != NULL) { g_hash_table_destroy(action_meta); } return timeout_ms; } #if ENABLE_VERSIONED_ATTRS static void unpack_versioned_meta(xmlNode *versioned_meta, xmlNode *xml_obj, guint interval_ms, crm_time_t *now) { xmlNode *attrs = NULL; xmlNode *attr = NULL; for (attrs = pcmk__xe_first_child(versioned_meta); attrs != NULL; attrs = pcmk__xe_next(attrs)) { for (attr = pcmk__xe_first_child(attrs); attr != NULL; attr = pcmk__xe_next(attr)) { const char *name = crm_element_value(attr, XML_NVPAIR_ATTR_NAME); const char *value = crm_element_value(attr, XML_NVPAIR_ATTR_VALUE); if (pcmk__str_eq(name, XML_OP_ATTR_START_DELAY, pcmk__str_casei)) { int start_delay = unpack_start_delay(value, NULL); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } else if (pcmk__str_eq(name, XML_OP_ATTR_ORIGIN, pcmk__str_casei)) { long long start_delay = 0; if (unpack_interval_origin(value, xml_obj, interval_ms, now, &start_delay)) { crm_xml_add(attr, XML_NVPAIR_ATTR_NAME, XML_OP_ATTR_START_DELAY); crm_xml_add_ll(attr, XML_NVPAIR_ATTR_VALUE, start_delay); } } else if (pcmk__str_eq(name, XML_ATTR_TIMEOUT, pcmk__str_casei)) { int timeout_ms = unpack_timeout(value); crm_xml_add_int(attr, XML_NVPAIR_ATTR_VALUE, timeout_ms); } } } } #endif /*! * \brief Unpack operation XML into an action structure * * Unpack an operation's meta-attributes (normalizing the interval, timeout, * and start delay values as integer milliseconds), requirements, and * failure policy. * * \param[in,out] action Action to unpack into * \param[in] xml_obj Operation XML (or NULL if all defaults) * \param[in] container Resource that contains affected resource, if any * \param[in] data_set Cluster state * \param[in] interval_ms How frequently to perform the operation */ static void unpack_operation(pe_action_t * action, xmlNode * xml_obj, pe_resource_t * container, pe_working_set_t * data_set, guint interval_ms) { int timeout_ms = 0; const char *value = NULL; #if ENABLE_VERSIONED_ATTRS pe_rsc_action_details_t *rsc_details = NULL; #endif pe_rsc_eval_data_t rsc_rule_data = { .standard = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_CLASS), .provider = crm_element_value(action->rsc->xml, XML_AGENT_ATTR_PROVIDER), .agent = crm_element_value(action->rsc->xml, XML_EXPR_ATTR_TYPE) }; pe_op_eval_data_t op_rule_data = { .op_name = action->task, .interval = interval_ms }; pe_rule_eval_data_t rule_data = { .node_hash = NULL, .role = RSC_ROLE_UNKNOWN, .now = data_set->now, .match_data = NULL, .rsc_data = &rsc_rule_data, .op_data = &op_rule_data }; CRM_CHECK(action && action->rsc, return); // Cluster-wide pe__unpack_dataset_nvpairs(data_set->op_defaults, XML_TAG_META_SETS, &rule_data, action->meta, NULL, FALSE, data_set); // Determine probe default timeout differently if (pcmk__str_eq(action->task, RSC_STATUS, pcmk__str_casei) && (interval_ms == 0)) { xmlNode *min_interval_mon = find_min_interval_mon(action->rsc, FALSE); if (min_interval_mon) { value = crm_element_value(min_interval_mon, XML_ATTR_TIMEOUT); if (value) { crm_trace("\t%s: Setting default timeout to minimum-interval " "monitor's timeout '%s'", action->uuid, value); g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT), strdup(value)); } } } if (xml_obj) { xmlAttrPtr xIter = NULL; // take precedence over defaults pe__unpack_dataset_nvpairs(xml_obj, XML_TAG_META_SETS, &rule_data, action->meta, NULL, TRUE, data_set); #if ENABLE_VERSIONED_ATTRS rsc_details = pe_rsc_action_details(action); pe_eval_versioned_attributes(data_set->input, xml_obj, XML_TAG_ATTR_SETS, &rule_data, rsc_details->versioned_parameters, NULL); pe_eval_versioned_attributes(data_set->input, xml_obj, XML_TAG_META_SETS, &rule_data, rsc_details->versioned_meta, NULL); #endif /* Anything set as an XML property has highest precedence. * This ensures we use the name and interval from the tag. */ for (xIter = xml_obj->properties; xIter; xIter = xIter->next) { const char *prop_name = (const char *)xIter->name; const char *prop_value = crm_element_value(xml_obj, prop_name); g_hash_table_replace(action->meta, strdup(prop_name), strdup(prop_value)); } } g_hash_table_remove(action->meta, "id"); // Normalize interval to milliseconds if (interval_ms > 0) { g_hash_table_replace(action->meta, strdup(XML_LRM_ATTR_INTERVAL), crm_strdup_printf("%u", interval_ms)); } else { g_hash_table_remove(action->meta, XML_LRM_ATTR_INTERVAL); } /* * Timeout order of precedence: * 1. pcmk_monitor_timeout (if rsc has pcmk_ra_cap_fence_params * and task is start or a probe; pcmk_monitor_timeout works * by default for a recurring monitor) * 2. explicit op timeout on the primitive * 3. default op timeout * a. if probe, then min-interval monitor's timeout * b. else, in XML_CIB_TAG_OPCONFIG * 4. CRM_DEFAULT_OP_TIMEOUT_S * * #1 overrides general rule of XML property having highest * precedence. */ if (pcmk_is_set(pcmk_get_ra_caps(rsc_rule_data.standard), pcmk_ra_cap_fence_params) && (pcmk__str_eq(action->task, RSC_START, pcmk__str_casei) || (pcmk__str_eq(action->task, RSC_STATUS, pcmk__str_casei) && (interval_ms == 0))) && action->rsc->parameters) { value = g_hash_table_lookup(action->rsc->parameters, "pcmk_monitor_timeout"); if (value) { crm_trace("\t%s: Setting timeout to pcmk_monitor_timeout '%s', " "overriding default", action->uuid, value); g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT), strdup(value)); } } // Normalize timeout to positive milliseconds value = g_hash_table_lookup(action->meta, XML_ATTR_TIMEOUT); timeout_ms = unpack_timeout(value); g_hash_table_replace(action->meta, strdup(XML_ATTR_TIMEOUT), crm_itoa(timeout_ms)); if (!pcmk__strcase_any_of(action->task, RSC_START, RSC_PROMOTE, NULL)) { action->needs = rsc_req_nothing; value = "nothing (not start or promote)"; } else if (pcmk_is_set(action->rsc->flags, pe_rsc_needs_fencing)) { action->needs = rsc_req_stonith; value = "fencing"; } else if (pcmk_is_set(action->rsc->flags, pe_rsc_needs_quorum)) { action->needs = rsc_req_quorum; value = "quorum"; } else { action->needs = rsc_req_nothing; value = "nothing"; } pe_rsc_trace(action->rsc, "%s requires %s", action->uuid, value); value = unpack_operation_on_fail(action); if (value == NULL) { } else if (pcmk__str_eq(value, "block", pcmk__str_casei)) { action->on_fail = action_fail_block; g_hash_table_insert(action->meta, strdup(XML_OP_ATTR_ON_FAIL), strdup("block")); value = "block"; // The above could destroy the original string } else if (pcmk__str_eq(value, "fence", pcmk__str_casei)) { action->on_fail = action_fail_fence; value = "node fencing"; if (!pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { pcmk__config_err("Resetting '" XML_OP_ATTR_ON_FAIL "' for " "operation '%s' to 'stop' because 'fence' is not " "valid when fencing is disabled", action->uuid); action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } } else if (pcmk__str_eq(value, "standby", pcmk__str_casei)) { action->on_fail = action_fail_standby; value = "node standby"; } else if (pcmk__strcase_any_of(value, "ignore", "nothing", NULL)) { action->on_fail = action_fail_ignore; value = "ignore"; } else if (pcmk__str_eq(value, "migrate", pcmk__str_casei)) { action->on_fail = action_fail_migrate; value = "force migration"; } else if (pcmk__str_eq(value, "stop", pcmk__str_casei)) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop resource"; } else if (pcmk__str_eq(value, "restart", pcmk__str_casei)) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate)"; } else if (pcmk__str_eq(value, "restart-container", pcmk__str_casei)) { if (container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate)"; } else { value = NULL; } } else if (pcmk__str_eq(value, "demote", pcmk__str_casei)) { action->on_fail = action_fail_demote; value = "demote instance"; } else { pe_err("Resource %s: Unknown failure type (%s)", action->rsc->id, value); value = NULL; } /* defaults */ if (value == NULL && container) { action->on_fail = action_fail_restart_container; value = "restart container (and possibly migrate) (default)"; /* For remote nodes, ensure that any failure that results in dropping an * active connection to the node results in fencing of the node. * * There are only two action failures that don't result in fencing. * 1. probes - probe failures are expected. * 2. start - a start failure indicates that an active connection does not already * exist. The user can set op on-fail=fence if they really want to fence start * failures. */ } else if (((value == NULL) || !pcmk_is_set(action->rsc->flags, pe_rsc_managed)) && pe__resource_is_remote_conn(action->rsc, data_set) && !(pcmk__str_eq(action->task, CRMD_ACTION_STATUS, pcmk__str_casei) && (interval_ms == 0)) && !pcmk__str_eq(action->task, CRMD_ACTION_START, pcmk__str_casei)) { if (!pcmk_is_set(action->rsc->flags, pe_rsc_managed)) { action->on_fail = action_fail_stop; action->fail_role = RSC_ROLE_STOPPED; value = "stop unmanaged remote node (enforcing default)"; } else { if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { value = "fence remote node (default)"; } else { value = "recover remote node connection (default)"; } if (action->rsc->remote_reconnect_ms) { action->fail_role = RSC_ROLE_STOPPED; } action->on_fail = action_fail_reset_remote; } } else if (value == NULL && pcmk__str_eq(action->task, CRMD_ACTION_STOP, pcmk__str_casei)) { if (pcmk_is_set(data_set->flags, pe_flag_stonith_enabled)) { action->on_fail = action_fail_fence; value = "resource fence (default)"; } else { action->on_fail = action_fail_block; value = "resource block (default)"; } } else if (value == NULL) { action->on_fail = action_fail_recover; value = "restart (and possibly migrate) (default)"; } pe_rsc_trace(action->rsc, "%s failure handling: %s", action->uuid, value); value = NULL; if (xml_obj != NULL) { value = g_hash_table_lookup(action->meta, "role_after_failure"); if (value) { pe_warn_once(pe_wo_role_after, "Support for role_after_failure is deprecated and will be removed in a future release"); } } if (value != NULL && action->fail_role == RSC_ROLE_UNKNOWN) { action->fail_role = text2role(value); } /* defaults */ if (action->fail_role == RSC_ROLE_UNKNOWN) { if (pcmk__str_eq(action->task, CRMD_ACTION_PROMOTE, pcmk__str_casei)) { action->fail_role = RSC_ROLE_SLAVE; } else { action->fail_role = RSC_ROLE_STARTED; } } pe_rsc_trace(action->rsc, "%s failure results in: %s", action->uuid, role2text(action->fail_role)); value = g_hash_table_lookup(action->meta, XML_OP_ATTR_START_DELAY); if (value) { unpack_start_delay(value, action->meta); } else { long long start_delay = 0; value = g_hash_table_lookup(action->meta, XML_OP_ATTR_ORIGIN); if (unpack_interval_origin(value, xml_obj, interval_ms, data_set->now, &start_delay)) { g_hash_table_replace(action->meta, strdup(XML_OP_ATTR_START_DELAY), crm_strdup_printf("%lld", start_delay)); } } #if ENABLE_VERSIONED_ATTRS unpack_versioned_meta(rsc_details->versioned_meta, xml_obj, interval_ms, data_set->now); #endif } static xmlNode * find_rsc_op_entry_helper(pe_resource_t * rsc, const char *key, gboolean include_disabled) { guint interval_ms = 0; gboolean do_retry = TRUE; char *local_key = NULL; const char *name = NULL; const char *value = NULL; const char *interval_spec = NULL; char *match_key = NULL; xmlNode *op = NULL; xmlNode *operation = NULL; retry: for (operation = pcmk__xe_first_child(rsc->ops_xml); operation != NULL; operation = pcmk__xe_next(operation)) { if (pcmk__str_eq((const char *)operation->name, "op", pcmk__str_none)) { name = crm_element_value(operation, "name"); interval_spec = crm_element_value(operation, XML_LRM_ATTR_INTERVAL); value = crm_element_value(operation, "enabled"); if (!include_disabled && value && crm_is_true(value) == FALSE) { continue; } interval_ms = crm_parse_interval_spec(interval_spec); match_key = pcmk__op_key(rsc->id, name, interval_ms); if (pcmk__str_eq(key, match_key, pcmk__str_casei)) { op = operation; } free(match_key); if (rsc->clone_name) { match_key = pcmk__op_key(rsc->clone_name, name, interval_ms); if (pcmk__str_eq(key, match_key, pcmk__str_casei)) { op = operation; } free(match_key); } if (op != NULL) { free(local_key); return op; } } } free(local_key); if (do_retry == FALSE) { return NULL; } do_retry = FALSE; if (strstr(key, CRMD_ACTION_MIGRATE) || strstr(key, CRMD_ACTION_MIGRATED)) { local_key = pcmk__op_key(rsc->id, "migrate", 0); key = local_key; goto retry; } else if (strstr(key, "_notify_")) { local_key = pcmk__op_key(rsc->id, "notify", 0); key = local_key; goto retry; } return NULL; } xmlNode * find_rsc_op_entry(pe_resource_t * rsc, const char *key) { return find_rsc_op_entry_helper(rsc, key, FALSE); } void print_node(const char *pre_text, pe_node_t * node, gboolean details) { if (node == NULL) { crm_trace("%s%s: ", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": "); return; } CRM_ASSERT(node->details); crm_trace("%s%s%sNode %s: (weight=%d, fixed=%s)", pre_text == NULL ? "" : pre_text, pre_text == NULL ? "" : ": ", node->details->online ? "" : "Unavailable/Unclean ", node->details->uname, node->weight, node->fixed ? "True" : "False"); if (details) { int log_level = LOG_TRACE; char *pe_mutable = strdup("\t\t"); GListPtr gIter = node->details->running_rsc; crm_trace("\t\t===Node Attributes"); g_hash_table_foreach(node->details->attrs, print_str_str, pe_mutable); free(pe_mutable); crm_trace("\t\t=== Resources"); for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; rsc->fns->print(rsc, "\t\t", pe_print_log|pe_print_pending, &log_level); } } } /* * Used by the HashTable for-loop */ void print_str_str(gpointer key, gpointer value, gpointer user_data) { crm_trace("%s%s %s ==> %s", user_data == NULL ? "" : (char *)user_data, user_data == NULL ? "" : ": ", (char *)key, (char *)value); } void pe_free_action(pe_action_t * action) { if (action == NULL) { return; } g_list_free_full(action->actions_before, free); /* pe_action_wrapper_t* */ g_list_free_full(action->actions_after, free); /* pe_action_wrapper_t* */ if (action->extra) { g_hash_table_destroy(action->extra); } if (action->meta) { g_hash_table_destroy(action->meta); } #if ENABLE_VERSIONED_ATTRS if (action->rsc) { pe_free_rsc_action_details(action); } #endif free(action->cancel_task); free(action->reason); free(action->task); free(action->uuid); free(action->node); free(action); } GListPtr find_recurring_actions(GListPtr input, pe_node_t * not_on_node) { const char *value = NULL; GListPtr result = NULL; GListPtr gIter = input; CRM_CHECK(input != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; value = g_hash_table_lookup(action->meta, XML_LRM_ATTR_INTERVAL_MS); if (value == NULL) { /* skip */ } else if (pcmk__str_eq(value, "0", pcmk__str_casei)) { /* skip */ } else if (pcmk__str_eq(CRMD_ACTION_CANCEL, action->task, pcmk__str_casei)) { /* skip */ } else if (not_on_node == NULL) { crm_trace("(null) Found: %s", action->uuid); result = g_list_prepend(result, action); } else if (action->node == NULL) { /* skip */ } else if (action->node->details != not_on_node->details) { crm_trace("Found: %s", action->uuid); result = g_list_prepend(result, action); } } return result; } enum action_tasks get_complex_task(pe_resource_t * rsc, const char *name, gboolean allow_non_atomic) { enum action_tasks task = text2task(name); if (rsc == NULL) { return task; } else if (allow_non_atomic == FALSE || rsc->variant == pe_native) { switch (task) { case stopped_rsc: case started_rsc: case action_demoted: case action_promoted: crm_trace("Folding %s back into its atomic counterpart for %s", name, rsc->id); return task - 1; default: break; } } return task; } pe_action_t * find_first_action(GListPtr input, const char *uuid, const char *task, pe_node_t * on_node) { GListPtr gIter = NULL; CRM_CHECK(uuid || task, return NULL); for (gIter = input; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (uuid != NULL && !pcmk__str_eq(uuid, action->uuid, pcmk__str_casei)) { continue; } else if (task != NULL && !pcmk__str_eq(task, action->task, pcmk__str_casei)) { continue; } else if (on_node == NULL) { return action; } else if (action->node == NULL) { continue; } else if (on_node->details == action->node->details) { return action; } } return NULL; } GListPtr find_actions(GListPtr input, const char *key, const pe_node_t *on_node) { GListPtr gIter = input; GListPtr result = NULL; CRM_CHECK(key != NULL, return NULL); for (; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) { crm_trace("%s does not match action %s", key, action->uuid); continue; } else if (on_node == NULL) { crm_trace("Action %s matches (ignoring node)", key); result = g_list_prepend(result, action); } else if (action->node == NULL) { crm_trace("Action %s matches (unallocated, assigning to %s)", key, on_node->details->uname); action->node = pe__copy_node(on_node); result = g_list_prepend(result, action); } else if (on_node->details == action->node->details) { crm_trace("Action %s on %s matches", key, on_node->details->uname); result = g_list_prepend(result, action); } else { crm_trace("Action %s on node %s does not match requested node %s", key, action->node->details->uname, on_node->details->uname); } } return result; } GList * find_actions_exact(GList *input, const char *key, const pe_node_t *on_node) { GList *result = NULL; CRM_CHECK(key != NULL, return NULL); if (on_node == NULL) { crm_trace("Not searching for action %s because node not specified", key); return NULL; } for (GList *gIter = input; gIter != NULL; gIter = gIter->next) { pe_action_t *action = (pe_action_t *) gIter->data; if (action->node == NULL) { crm_trace("Skipping comparison of %s vs action %s without node", key, action->uuid); } else if (!pcmk__str_eq(key, action->uuid, pcmk__str_casei)) { crm_trace("Desired action %s doesn't match %s", key, action->uuid); } else if (!pcmk__str_eq(on_node->details->id, action->node->details->id, pcmk__str_casei)) { crm_trace("Action %s desired node ID %s doesn't match %s", key, on_node->details->id, action->node->details->id); } else { crm_trace("Action %s matches", key); result = g_list_prepend(result, action); } } return result; } /*! * \brief Find all actions of given type for a resource * * \param[in] rsc Resource to search * \param[in] node Find only actions scheduled on this node * \param[in] task Action name to search for * \param[in] require_node If TRUE, NULL node or action node will not match * * \return List of actions found (or NULL if none) * \note If node is not NULL and require_node is FALSE, matching actions * without a node will be assigned to node. */ GList * pe__resource_actions(const pe_resource_t *rsc, const pe_node_t *node, const char *task, bool require_node) { GList *result = NULL; char *key = pcmk__op_key(rsc->id, task, 0); if (require_node) { result = find_actions_exact(rsc->actions, key, node); } else { result = find_actions(rsc->actions, key, node); } free(key); return result; } static void resource_node_score(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag) { pe_node_t *match = NULL; if ((rsc->exclusive_discover || (node->rsc_discover_mode == pe_discover_never)) && pcmk__str_eq(tag, "symmetric_default", pcmk__str_casei)) { /* This string comparision may be fragile, but exclusive resources and * exclusive nodes should not have the symmetric_default constraint * applied to them. */ return; } else if (rsc->children) { GListPtr gIter = rsc->children; for (; gIter != NULL; gIter = gIter->next) { pe_resource_t *child_rsc = (pe_resource_t *) gIter->data; resource_node_score(child_rsc, node, score, tag); } } pe_rsc_trace(rsc, "Setting %s for %s on %s: %d", tag, rsc->id, node->details->uname, score); match = pe_hash_table_lookup(rsc->allowed_nodes, node->details->id); if (match == NULL) { match = pe__copy_node(node); g_hash_table_insert(rsc->allowed_nodes, (gpointer) match->details->id, match); } match->weight = pe__add_scores(match->weight, score); } void resource_location(pe_resource_t * rsc, pe_node_t * node, int score, const char *tag, pe_working_set_t * data_set) { if (node != NULL) { resource_node_score(rsc, node, score, tag); } else if (data_set != NULL) { GListPtr gIter = data_set->nodes; for (; gIter != NULL; gIter = gIter->next) { pe_node_t *node_iter = (pe_node_t *) gIter->data; resource_node_score(rsc, node_iter, score, tag); } } else { GHashTableIter iter; pe_node_t *node_iter = NULL; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node_iter)) { resource_node_score(rsc, node_iter, score, tag); } } if (node == NULL && score == -INFINITY) { if (rsc->allocated_to) { crm_info("Deallocating %s from %s", rsc->id, rsc->allocated_to->details->uname); free(rsc->allocated_to); rsc->allocated_to = NULL; } } } #define sort_return(an_int, why) do { \ free(a_uuid); \ free(b_uuid); \ crm_trace("%s (%d) %c %s (%d) : %s", \ a_xml_id, a_call_id, an_int>0?'>':an_int<0?'<':'=', \ b_xml_id, b_call_id, why); \ return an_int; \ } while(0) gint sort_op_by_callid(gconstpointer a, gconstpointer b) { int a_call_id = -1; int b_call_id = -1; char *a_uuid = NULL; char *b_uuid = NULL; const xmlNode *xml_a = a; const xmlNode *xml_b = b; const char *a_xml_id = crm_element_value(xml_a, XML_ATTR_ID); const char *b_xml_id = crm_element_value(xml_b, XML_ATTR_ID); if (pcmk__str_eq(a_xml_id, b_xml_id, pcmk__str_casei)) { /* We have duplicate lrm_rsc_op entries in the status * section which is unlikely to be a good thing * - we can handle it easily enough, but we need to get * to the bottom of why it's happening. */ pe_err("Duplicate lrm_rsc_op entries named %s", a_xml_id); sort_return(0, "duplicate"); } crm_element_value_int(xml_a, XML_LRM_ATTR_CALLID, &a_call_id); crm_element_value_int(xml_b, XML_LRM_ATTR_CALLID, &b_call_id); if (a_call_id == -1 && b_call_id == -1) { /* both are pending ops so it doesn't matter since * stops are never pending */ sort_return(0, "pending"); } else if (a_call_id >= 0 && a_call_id < b_call_id) { sort_return(-1, "call id"); } else if (b_call_id >= 0 && a_call_id > b_call_id) { sort_return(1, "call id"); } else if (b_call_id >= 0 && a_call_id == b_call_id) { /* * The op and last_failed_op are the same * Order on last-rc-change */ time_t last_a = -1; time_t last_b = -1; crm_element_value_epoch(xml_a, XML_RSC_OP_LAST_CHANGE, &last_a); crm_element_value_epoch(xml_b, XML_RSC_OP_LAST_CHANGE, &last_b); crm_trace("rc-change: %lld vs %lld", (long long) last_a, (long long) last_b); if (last_a >= 0 && last_a < last_b) { sort_return(-1, "rc-change"); } else if (last_b >= 0 && last_a > last_b) { sort_return(1, "rc-change"); } sort_return(0, "rc-change"); } else { /* One of the inputs is a pending operation * Attempt to use XML_ATTR_TRANSITION_MAGIC to determine its age relative to the other */ int a_id = -1; int b_id = -1; const char *a_magic = crm_element_value(xml_a, XML_ATTR_TRANSITION_MAGIC); const char *b_magic = crm_element_value(xml_b, XML_ATTR_TRANSITION_MAGIC); CRM_CHECK(a_magic != NULL && b_magic != NULL, sort_return(0, "No magic")); if (!decode_transition_magic(a_magic, &a_uuid, &a_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic a"); } if (!decode_transition_magic(b_magic, &b_uuid, &b_id, NULL, NULL, NULL, NULL)) { sort_return(0, "bad magic b"); } /* try to determine the relative age of the operation... * some pending operations (e.g. a start) may have been superseded * by a subsequent stop * * [a|b]_id == -1 means it's a shutdown operation and _always_ comes last */ if (!pcmk__str_eq(a_uuid, b_uuid, pcmk__str_casei) || a_id == b_id) { /* * some of the logic in here may be redundant... * * if the UUID from the TE doesn't match then one better * be a pending operation. * pending operations don't survive between elections and joins * because we query the LRM directly */ if (b_call_id == -1) { sort_return(-1, "transition + call"); } else if (a_call_id == -1) { sort_return(1, "transition + call"); } } else if ((a_id >= 0 && a_id < b_id) || b_id == -1) { sort_return(-1, "transition"); } else if ((b_id >= 0 && a_id > b_id) || a_id == -1) { sort_return(1, "transition"); } } /* we should never end up here */ CRM_CHECK(FALSE, sort_return(0, "default")); } time_t get_effective_time(pe_working_set_t * data_set) { if(data_set) { if (data_set->now == NULL) { crm_trace("Recording a new 'now'"); data_set->now = crm_time_new(NULL); } return crm_time_get_seconds_since_epoch(data_set->now); } crm_trace("Defaulting to 'now'"); return time(NULL); } gboolean get_target_role(pe_resource_t * rsc, enum rsc_role_e * role) { enum rsc_role_e local_role = RSC_ROLE_UNKNOWN; const char *value = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); CRM_CHECK(role != NULL, return FALSE); if (pcmk__str_eq(value, "started", pcmk__str_null_matches | pcmk__str_casei) || pcmk__str_eq("default", value, pcmk__str_casei)) { return FALSE; } local_role = text2role(value); if (local_role == RSC_ROLE_UNKNOWN) { pcmk__config_err("Ignoring '" XML_RSC_ATTR_TARGET_ROLE "' for %s " "because '%s' is not valid", rsc->id, value); return FALSE; } else if (local_role > RSC_ROLE_STARTED) { if (pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable)) { if (local_role > RSC_ROLE_SLAVE) { /* This is what we'd do anyway, just leave the default to avoid messing up the placement algorithm */ return FALSE; } } else { pcmk__config_err("Ignoring '" XML_RSC_ATTR_TARGET_ROLE "' for %s " "because '%s' only makes sense for promotable " "clones", rsc->id, value); return FALSE; } } *role = local_role; return TRUE; } gboolean order_actions(pe_action_t * lh_action, pe_action_t * rh_action, enum pe_ordering order) { GListPtr gIter = NULL; pe_action_wrapper_t *wrapper = NULL; GListPtr list = NULL; if (order == pe_order_none) { return FALSE; } if (lh_action == NULL || rh_action == NULL) { return FALSE; } crm_trace("Ordering Action %s before %s", lh_action->uuid, rh_action->uuid); /* Ensure we never create a dependency on ourselves... it's happened */ CRM_ASSERT(lh_action != rh_action); /* Filter dups, otherwise update_action_states() has too much work to do */ gIter = lh_action->actions_after; for (; gIter != NULL; gIter = gIter->next) { pe_action_wrapper_t *after = (pe_action_wrapper_t *) gIter->data; if (after->action == rh_action && (after->type & order)) { return FALSE; } } wrapper = calloc(1, sizeof(pe_action_wrapper_t)); wrapper->action = rh_action; wrapper->type = order; list = lh_action->actions_after; list = g_list_prepend(list, wrapper); lh_action->actions_after = list; wrapper = calloc(1, sizeof(pe_action_wrapper_t)); wrapper->action = lh_action; wrapper->type = order; list = rh_action->actions_before; list = g_list_prepend(list, wrapper); rh_action->actions_before = list; return TRUE; } pe_action_t * get_pseudo_op(const char *name, pe_working_set_t * data_set) { pe_action_t *op = NULL; if(data_set->singletons) { op = g_hash_table_lookup(data_set->singletons, name); } if (op == NULL) { op = custom_action(NULL, strdup(name), name, NULL, TRUE, TRUE, data_set); pe__set_action_flags(op, pe_action_pseudo|pe_action_runnable); } return op; } void destroy_ticket(gpointer data) { pe_ticket_t *ticket = data; if (ticket->state) { g_hash_table_destroy(ticket->state); } free(ticket->id); free(ticket); } pe_ticket_t * ticket_new(const char *ticket_id, pe_working_set_t * data_set) { pe_ticket_t *ticket = NULL; if (pcmk__str_empty(ticket_id)) { return NULL; } if (data_set->tickets == NULL) { data_set->tickets = g_hash_table_new_full(crm_str_hash, g_str_equal, free, destroy_ticket); } ticket = g_hash_table_lookup(data_set->tickets, ticket_id); if (ticket == NULL) { ticket = calloc(1, sizeof(pe_ticket_t)); if (ticket == NULL) { crm_err("Cannot allocate ticket '%s'", ticket_id); return NULL; } crm_trace("Creaing ticket entry for %s", ticket_id); ticket->id = strdup(ticket_id); ticket->granted = FALSE; ticket->last_granted = -1; ticket->standby = FALSE; ticket->state = crm_str_table_new(); g_hash_table_insert(data_set->tickets, strdup(ticket->id), ticket); } return ticket; } static void filter_parameters(xmlNode * param_set, const char *param_string, bool need_present) { if (param_set && param_string) { xmlAttrPtr xIter = param_set->properties; while (xIter) { const char *prop_name = (const char *)xIter->name; char *name = crm_strdup_printf(" %s ", prop_name); char *match = strstr(param_string, name); free(name); // Do now, because current entry might get removed below xIter = xIter->next; if (need_present && match == NULL) { crm_trace("%s not found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } else if (need_present == FALSE && match) { crm_trace("%s found in %s", prop_name, param_string); xml_remove_prop(param_set, prop_name); } } } } #if ENABLE_VERSIONED_ATTRS static void append_versioned_params(xmlNode *versioned_params, const char *ra_version, xmlNode *params) { GHashTable *hash = pe_unpack_versioned_parameters(versioned_params, ra_version); char *key = NULL; char *value = NULL; GHashTableIter iter; g_hash_table_iter_init(&iter, hash); while (g_hash_table_iter_next(&iter, (gpointer *) &key, (gpointer *) &value)) { crm_xml_add(params, key, value); } g_hash_table_destroy(hash); } #endif /*! * \internal * \brief Calculate action digests and store in node's digest cache * * \param[in] rsc Resource that action was for * \param[in] task Name of action performed * \param[in] key Action's task key * \param[in] node Node action was performed on * \param[in] xml_op XML of operation in CIB status (if available) * \param[in] calc_secure Whether to calculate secure digest * \param[in] data_set Cluster working set * * \return Pointer to node's digest cache entry */ static op_digest_cache_t * rsc_action_digest(pe_resource_t *rsc, const char *task, const char *key, pe_node_t *node, xmlNode *xml_op, bool calc_secure, pe_working_set_t *data_set) { op_digest_cache_t *data = NULL; data = g_hash_table_lookup(node->details->digest_cache, key); if (data == NULL) { GHashTable *local_rsc_params = crm_str_table_new(); pe_action_t *action = custom_action(rsc, strdup(key), task, node, TRUE, FALSE, data_set); #if ENABLE_VERSIONED_ATTRS xmlNode *local_versioned_params = create_xml_node(NULL, XML_TAG_RSC_VER_ATTRS); const char *ra_version = NULL; #endif const char *op_version; const char *restart_list = NULL; const char *secure_list = " passwd password "; data = calloc(1, sizeof(op_digest_cache_t)); CRM_ASSERT(data != NULL); get_rsc_attributes(local_rsc_params, rsc, node, data_set); #if ENABLE_VERSIONED_ATTRS pe_get_versioned_attributes(local_versioned_params, rsc, node, data_set); #endif data->params_all = create_xml_node(NULL, XML_TAG_PARAMS); // REMOTE_CONTAINER_HACK: Allow remote nodes that start containers with pacemaker remote inside if (pe__add_bundle_remote_name(rsc, data->params_all, XML_RSC_ATTR_REMOTE_RA_ADDR)) { crm_trace("Set address for bundle connection %s (on %s)", rsc->id, node->details->uname); } g_hash_table_foreach(local_rsc_params, hash2field, data->params_all); g_hash_table_foreach(action->extra, hash2field, data->params_all); g_hash_table_foreach(rsc->parameters, hash2field, data->params_all); g_hash_table_foreach(action->meta, hash2metafield, data->params_all); if(xml_op) { secure_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_SECURE); restart_list = crm_element_value(xml_op, XML_LRM_ATTR_OP_RESTART); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); #if ENABLE_VERSIONED_ATTRS ra_version = crm_element_value(xml_op, XML_ATTR_RA_VERSION); #endif } else { op_version = CRM_FEATURE_SET; } #if ENABLE_VERSIONED_ATTRS append_versioned_params(local_versioned_params, ra_version, data->params_all); append_versioned_params(rsc->versioned_parameters, ra_version, data->params_all); { pe_rsc_action_details_t *details = pe_rsc_action_details(action); append_versioned_params(details->versioned_parameters, ra_version, data->params_all); } #endif pcmk__filter_op_for_digest(data->params_all); g_hash_table_destroy(local_rsc_params); pe_free_action(action); data->digest_all_calc = calculate_operation_digest(data->params_all, op_version); if (calc_secure) { /* The controller doesn't create a digest of *all* non-sensitive * parameters, only those listed in resource agent meta-data. The * equivalent here is rsc->parameters. */ data->params_secure = create_xml_node(NULL, XML_TAG_PARAMS); g_hash_table_foreach(rsc->parameters, hash2field, data->params_secure); if(secure_list) { filter_parameters(data->params_secure, secure_list, FALSE); } data->digest_secure_calc = calculate_operation_digest(data->params_secure, op_version); } if(xml_op && crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST) != NULL) { data->params_restart = copy_xml(data->params_all); if (restart_list) { filter_parameters(data->params_restart, restart_list, TRUE); } data->digest_restart_calc = calculate_operation_digest(data->params_restart, op_version); } g_hash_table_insert(node->details->digest_cache, strdup(key), data); } return data; } op_digest_cache_t * rsc_action_digest_cmp(pe_resource_t * rsc, xmlNode * xml_op, pe_node_t * node, pe_working_set_t * data_set) { op_digest_cache_t *data = NULL; char *key = NULL; guint interval_ms = 0; const char *op_version; const char *task = crm_element_value(xml_op, XML_LRM_ATTR_TASK); const char *digest_all; const char *digest_restart; CRM_ASSERT(node != NULL); op_version = crm_element_value(xml_op, XML_ATTR_CRM_VERSION); digest_all = crm_element_value(xml_op, XML_LRM_ATTR_OP_DIGEST); digest_restart = crm_element_value(xml_op, XML_LRM_ATTR_RESTART_DIGEST); crm_element_value_ms(xml_op, XML_LRM_ATTR_INTERVAL_MS, &interval_ms); key = pcmk__op_key(rsc->id, task, interval_ms); data = rsc_action_digest(rsc, task, key, node, xml_op, pcmk_is_set(data_set->flags, pe_flag_sanitized), data_set); data->rc = RSC_DIGEST_MATCH; if (digest_restart && data->digest_restart_calc && strcmp(data->digest_restart_calc, digest_restart) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (restart:%s) %s", key, node->details->uname, crm_str(digest_restart), data->digest_restart_calc, op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_RESTART; } else if (digest_all == NULL) { /* it is unknown what the previous op digest was */ data->rc = RSC_DIGEST_UNKNOWN; } else if (strcmp(digest_all, data->digest_all_calc) != 0) { pe_rsc_info(rsc, "Parameters to %s on %s changed: was %s vs. now %s (%s:%s) %s", key, node->details->uname, crm_str(digest_all), data->digest_all_calc, (interval_ms > 0)? "reschedule" : "reload", op_version, crm_element_value(xml_op, XML_ATTR_TRANSITION_MAGIC)); data->rc = RSC_DIGEST_ALL; } free(key); return data; } /*! * \internal * \brief Create an unfencing summary for use in special node attribute * * Create a string combining a fence device's resource ID, agent type, and * parameter digest (whether for all parameters or just non-private parameters). * This can be stored in a special node attribute, allowing us to detect changes * in either the agent type or parameters, to know whether unfencing must be * redone or can be safely skipped when the device's history is cleaned. * * \param[in] rsc_id Fence device resource ID * \param[in] agent_type Fence device agent * \param[in] param_digest Fence device parameter digest * * \return Newly allocated string with unfencing digest * \note The caller is responsible for freeing the result. */ static inline char * create_unfencing_summary(const char *rsc_id, const char *agent_type, const char *param_digest) { return crm_strdup_printf("%s:%s:%s", rsc_id, agent_type, param_digest); } /*! * \internal * \brief Check whether a node can skip unfencing * * Check whether a fence device's current definition matches a node's * stored summary of when it was last unfenced by the device. * * \param[in] rsc_id Fence device's resource ID * \param[in] agent Fence device's agent type * \param[in] digest_calc Fence device's current parameter digest * \param[in] node_summary Value of node's special unfencing node attribute * (a comma-separated list of unfencing summaries for * all devices that have unfenced this node) * * \return TRUE if digest matches, FALSE otherwise */ static bool unfencing_digest_matches(const char *rsc_id, const char *agent, const char *digest_calc, const char *node_summary) { bool matches = FALSE; if (rsc_id && agent && digest_calc && node_summary) { char *search_secure = create_unfencing_summary(rsc_id, agent, digest_calc); /* The digest was calculated including the device ID and agent, * so there is no risk of collision using strstr(). */ matches = (strstr(node_summary, search_secure) != NULL); crm_trace("Calculated unfencing digest '%s' %sfound in '%s'", search_secure, matches? "" : "not ", node_summary); free(search_secure); } return matches; } /* Magic string to use as action name for digest cache entries used for * unfencing checks. This is not a real action name (i.e. "on"), so * check_action_definition() won't confuse these entries with real actions. */ #define STONITH_DIGEST_TASK "stonith-on" /*! * \internal * \brief Calculate fence device digests and digest comparison result * * \param[in] rsc Fence device resource * \param[in] agent Fence device's agent type * \param[in] node Node with digest cache to use * \param[in] data_set Cluster working set * * \return Node's digest cache entry */ static op_digest_cache_t * fencing_action_digest_cmp(pe_resource_t *rsc, const char *agent, pe_node_t *node, pe_working_set_t *data_set) { const char *node_summary = NULL; // Calculate device's current parameter digests char *key = pcmk__op_key(rsc->id, STONITH_DIGEST_TASK, 0); op_digest_cache_t *data = rsc_action_digest(rsc, STONITH_DIGEST_TASK, key, node, NULL, TRUE, data_set); free(key); // Check whether node has special unfencing summary node attribute node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_ALL); if (node_summary == NULL) { data->rc = RSC_DIGEST_UNKNOWN; return data; } // Check whether full parameter digest matches if (unfencing_digest_matches(rsc->id, agent, data->digest_all_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; return data; } // Check whether secure parameter digest matches node_summary = pe_node_attribute_raw(node, CRM_ATTR_DIGESTS_SECURE); if (unfencing_digest_matches(rsc->id, agent, data->digest_secure_calc, node_summary)) { data->rc = RSC_DIGEST_MATCH; if (pcmk_is_set(data_set->flags, pe_flag_stdout)) { printf("Only 'private' parameters to %s for unfencing %s changed\n", rsc->id, node->details->uname); } return data; } // Parameters don't match data->rc = RSC_DIGEST_ALL; if (pcmk_is_set(data_set->flags, (pe_flag_sanitized|pe_flag_stdout)) && data->digest_secure_calc) { char *digest = create_unfencing_summary(rsc->id, agent, data->digest_secure_calc); printf("Parameters to %s for unfencing %s changed, try '%s'\n", rsc->id, node->details->uname, digest); free(digest); } return data; } const char *rsc_printable_id(pe_resource_t *rsc) { if (!pcmk_is_set(rsc->flags, pe_rsc_unique)) { return ID(rsc->xml); } return rsc->id; } void pe__clear_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags) { pe__clear_resource_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__clear_resource_flags_recursive((pe_resource_t *) gIter->data, flags); } } void pe__set_resource_flags_recursive(pe_resource_t *rsc, uint64_t flags) { pe__set_resource_flags(rsc, flags); for (GList *gIter = rsc->children; gIter != NULL; gIter = gIter->next) { pe__set_resource_flags_recursive((pe_resource_t *) gIter->data, flags); } } static GListPtr find_unfencing_devices(GListPtr candidates, GListPtr matches) { for (GListPtr gIter = candidates; gIter != NULL; gIter = gIter->next) { pe_resource_t *candidate = gIter->data; - const char *provides = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_PROVIDES); + const char *provides = g_hash_table_lookup(candidate->meta, + PCMK_STONITH_PROVIDES); const char *requires = g_hash_table_lookup(candidate->meta, XML_RSC_ATTR_REQUIRES); if(candidate->children) { matches = find_unfencing_devices(candidate->children, matches); } else if (!pcmk_is_set(candidate->flags, pe_rsc_fence_device)) { continue; } else if (pcmk__str_eq(provides, "unfencing", pcmk__str_casei) || pcmk__str_eq(requires, "unfencing", pcmk__str_casei)) { matches = g_list_prepend(matches, candidate); } } return matches; } static int node_priority_fencing_delay(pe_node_t * node, pe_working_set_t * data_set) { int member_count = 0; int online_count = 0; int top_priority = 0; int lowest_priority = 0; GListPtr gIter = NULL; // `priority-fencing-delay` is disabled if (data_set->priority_fencing_delay <= 0) { return 0; } /* No need to request a delay if the fencing target is not a normal cluster * member, for example if it's a remote node or a guest node. */ if (node->details->type != node_member) { return 0; } // No need to request a delay if the fencing target is in our partition if (node->details->online) { return 0; } for (gIter = data_set->nodes; gIter != NULL; gIter = gIter->next) { pe_node_t *n = gIter->data; if (n->details->type != node_member) { continue; } member_count ++; if (n->details->online) { online_count++; } if (member_count == 1 || n->details->priority > top_priority) { top_priority = n->details->priority; } if (member_count == 1 || n->details->priority < lowest_priority) { lowest_priority = n->details->priority; } } // No need to delay if we have more than half of the cluster members if (online_count > member_count / 2) { return 0; } /* All the nodes have equal priority. * Any configured corresponding `pcmk_delay_base/max` will be applied. */ if (lowest_priority == top_priority) { return 0; } if (node->details->priority < top_priority) { return 0; } return data_set->priority_fencing_delay; } pe_action_t * pe_fence_op(pe_node_t * node, const char *op, bool optional, const char *reason, bool priority_delay, pe_working_set_t * data_set) { char *op_key = NULL; pe_action_t *stonith_op = NULL; if(op == NULL) { op = data_set->stonith_action; } op_key = crm_strdup_printf("%s-%s-%s", CRM_OP_FENCE, node->details->uname, op); if(data_set->singletons) { stonith_op = g_hash_table_lookup(data_set->singletons, op_key); } if(stonith_op == NULL) { stonith_op = custom_action(NULL, op_key, CRM_OP_FENCE, node, TRUE, TRUE, data_set); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET, node->details->uname); add_hash_param(stonith_op->meta, XML_LRM_ATTR_TARGET_UUID, node->details->id); add_hash_param(stonith_op->meta, "stonith_action", op); if (pe__is_guest_or_remote_node(node) && pcmk_is_set(data_set->flags, pe_flag_enable_unfencing)) { /* Extra work to detect device changes on remotes * * We may do this for all nodes in the future, but for now * the check_action_definition() based stuff works fine. */ long max = 1024; long digests_all_offset = 0; long digests_secure_offset = 0; char *digests_all = calloc(max, sizeof(char)); char *digests_secure = calloc(max, sizeof(char)); GListPtr matches = find_unfencing_devices(data_set->resources, NULL); for (GListPtr gIter = matches; gIter != NULL; gIter = gIter->next) { pe_resource_t *match = gIter->data; const char *agent = g_hash_table_lookup(match->meta, XML_ATTR_TYPE); op_digest_cache_t *data = NULL; data = fencing_action_digest_cmp(match, agent, node, data_set); if(data->rc == RSC_DIGEST_ALL) { optional = FALSE; crm_notice("Unfencing %s (remote): because the definition of %s changed", node->details->uname, match->id); if (pcmk_is_set(data_set->flags, pe_flag_stdout)) { fprintf(stdout, " notice: Unfencing %s (remote): because the definition of %s changed\n", node->details->uname, match->id); } } digests_all_offset += snprintf( digests_all+digests_all_offset, max-digests_all_offset, "%s:%s:%s,", match->id, agent, data->digest_all_calc); digests_secure_offset += snprintf( digests_secure+digests_secure_offset, max-digests_secure_offset, "%s:%s:%s,", match->id, agent, data->digest_secure_calc); } g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_ALL), digests_all); g_hash_table_insert(stonith_op->meta, strdup(XML_OP_ATTR_DIGESTS_SECURE), digests_secure); } } else { free(op_key); } if (data_set->priority_fencing_delay > 0 /* It's a suitable case where `priority-fencing-delay` applies. * At least add `priority-fencing-delay` field as an indicator. */ && (priority_delay /* Re-calculate priority delay for the suitable case when * pe_fence_op() is called again by stage6() after node priority has * been actually calculated with native_add_running() */ || g_hash_table_lookup(stonith_op->meta, XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY) != NULL)) { /* Add `priority-fencing-delay` to the fencing op even if it's 0 for * the targeting node. So that it takes precedence over any possible * `pcmk_delay_base/max`. */ char *delay_s = crm_itoa(node_priority_fencing_delay(node, data_set)); g_hash_table_insert(stonith_op->meta, strdup(XML_CONFIG_ATTR_PRIORITY_FENCING_DELAY), delay_s); } if(optional == FALSE && pe_can_fence(data_set, node)) { pe_action_required(stonith_op, NULL, reason); } else if(reason && stonith_op->reason == NULL) { stonith_op->reason = strdup(reason); } return stonith_op; } void trigger_unfencing( pe_resource_t * rsc, pe_node_t *node, const char *reason, pe_action_t *dependency, pe_working_set_t * data_set) { if (!pcmk_is_set(data_set->flags, pe_flag_enable_unfencing)) { /* No resources require it */ return; } else if ((rsc != NULL) && !pcmk_is_set(rsc->flags, pe_rsc_fence_device)) { /* Wasn't a stonith device */ return; } else if(node && node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { pe_action_t *unfence = pe_fence_op(node, "on", FALSE, reason, FALSE, data_set); if(dependency) { order_actions(unfence, dependency, pe_order_optional); } } else if(rsc) { GHashTableIter iter; g_hash_table_iter_init(&iter, rsc->allowed_nodes); while (g_hash_table_iter_next(&iter, NULL, (void **)&node)) { if(node->details->online && node->details->unclean == FALSE && node->details->shutdown == FALSE) { trigger_unfencing(rsc, node, reason, dependency, data_set); } } } } gboolean add_tag_ref(GHashTable * tags, const char * tag_name, const char * obj_ref) { pe_tag_t *tag = NULL; GListPtr gIter = NULL; gboolean is_existing = FALSE; CRM_CHECK(tags && tag_name && obj_ref, return FALSE); tag = g_hash_table_lookup(tags, tag_name); if (tag == NULL) { tag = calloc(1, sizeof(pe_tag_t)); if (tag == NULL) { return FALSE; } tag->id = strdup(tag_name); tag->refs = NULL; g_hash_table_insert(tags, strdup(tag_name), tag); } for (gIter = tag->refs; gIter != NULL; gIter = gIter->next) { const char *existing_ref = (const char *) gIter->data; if (pcmk__str_eq(existing_ref, obj_ref, pcmk__str_none)){ is_existing = TRUE; break; } } if (is_existing == FALSE) { tag->refs = g_list_append(tag->refs, strdup(obj_ref)); crm_trace("Added: tag=%s ref=%s", tag->id, obj_ref); } return TRUE; } void pe_action_set_flag_reason(const char *function, long line, pe_action_t *action, pe_action_t *reason, const char *text, enum pe_action_flags flags, bool overwrite) { bool unset = FALSE; bool update = FALSE; const char *change = NULL; if (pcmk_is_set(flags, pe_action_runnable)) { unset = TRUE; change = "unrunnable"; } else if (pcmk_is_set(flags, pe_action_optional)) { unset = TRUE; change = "required"; } else if (pcmk_is_set(flags, pe_action_migrate_runnable)) { unset = TRUE; overwrite = TRUE; change = "unrunnable"; } else if (pcmk_is_set(flags, pe_action_dangle)) { change = "dangling"; } else if (pcmk_is_set(flags, pe_action_requires_any)) { change = "required"; } else { crm_err("Unknown flag change to %x by %s: 0x%s", flags, action->uuid, (reason? reason->uuid : "0")); } if(unset) { if (pcmk_is_set(action->flags, flags)) { pe__clear_action_flags_as(function, line, action, flags); update = TRUE; } } else { if (!pcmk_is_set(action->flags, flags)) { pe__set_action_flags_as(function, line, action, flags); update = TRUE; } } if((change && update) || text) { char *reason_text = NULL; if(reason == NULL) { pe_action_set_reason(action, text, overwrite); } else if(reason->rsc == NULL) { reason_text = crm_strdup_printf("%s %s%c %s", change, reason->task, text?':':0, text?text:""); } else { reason_text = crm_strdup_printf("%s %s %s%c %s", change, reason->rsc->id, reason->task, text?':':0, text?text:"NA"); } if(reason_text && action->rsc != reason->rsc) { pe_action_set_reason(action, reason_text, overwrite); } free(reason_text); } } void pe_action_set_reason(pe_action_t *action, const char *reason, bool overwrite) { if (action->reason != NULL && overwrite) { pe_rsc_trace(action->rsc, "Changing %s reason from '%s' to '%s'", action->uuid, action->reason, crm_str(reason)); free(action->reason); } else if (action->reason == NULL) { pe_rsc_trace(action->rsc, "Set %s reason to '%s'", action->uuid, crm_str(reason)); } else { // crm_assert(action->reason != NULL && !overwrite); return; } if (reason != NULL) { action->reason = strdup(reason); } else { action->reason = NULL; } } /*! * \internal * \brief Check whether shutdown has been requested for a node * * \param[in] node Node to check * * \return TRUE if node has shutdown attribute set and nonzero, FALSE otherwise * \note This differs from simply using node->details->shutdown in that it can * be used before that has been determined (and in fact to determine it), * and it can also be used to distinguish requested shutdown from implicit * shutdown of remote nodes by virtue of their connection stopping. */ bool pe__shutdown_requested(pe_node_t *node) { const char *shutdown = pe_node_attribute_raw(node, XML_CIB_ATTR_SHUTDOWN); return !pcmk__str_eq(shutdown, "0", pcmk__str_null_matches); } /*! * \internal * \brief Update a data set's "recheck by" time * * \param[in] recheck Epoch time when recheck should happen * \param[in,out] data_set Current working set */ void pe__update_recheck_time(time_t recheck, pe_working_set_t *data_set) { if ((recheck > get_effective_time(data_set)) && ((data_set->recheck_by == 0) || (data_set->recheck_by > recheck))) { data_set->recheck_by = recheck; } } /*! * \internal * \brief Wrapper for pe_unpack_nvpairs() using a cluster working set */ void pe__unpack_dataset_nvpairs(xmlNode *xml_obj, const char *set_name, pe_rule_eval_data_t *rule_data, GHashTable *hash, const char *always_first, gboolean overwrite, pe_working_set_t *data_set) { crm_time_t *next_change = crm_time_new_undefined(); pe_eval_nvpairs(data_set->input, xml_obj, set_name, rule_data, hash, always_first, overwrite, next_change); if (crm_time_is_defined(next_change)) { time_t recheck = (time_t) crm_time_get_seconds_since_epoch(next_change); pe__update_recheck_time(recheck, data_set); } crm_time_free(next_change); } bool pe__resource_is_disabled(pe_resource_t *rsc) { const char *target_role = NULL; CRM_CHECK(rsc != NULL, return false); target_role = g_hash_table_lookup(rsc->meta, XML_RSC_ATTR_TARGET_ROLE); if (target_role) { enum rsc_role_e target_role_e = text2role(target_role); if ((target_role_e == RSC_ROLE_STOPPED) || ((target_role_e == RSC_ROLE_SLAVE) && pcmk_is_set(uber_parent(rsc)->flags, pe_rsc_promotable))) { return true; } } return false; } /*! * \internal * \brief Create an action to clear a resource's history from CIB * * \param[in] rsc Resource to clear * \param[in] node Node to clear history on * * \return New action to clear resource history */ pe_action_t * pe__clear_resource_history(pe_resource_t *rsc, pe_node_t *node, pe_working_set_t *data_set) { char *key = NULL; CRM_ASSERT(rsc && node); key = pcmk__op_key(rsc->id, CRM_OP_LRM_DELETE, 0); return custom_action(rsc, key, CRM_OP_LRM_DELETE, node, FALSE, TRUE, data_set); } bool pe__rsc_running_on_any_node_in_list(pe_resource_t *rsc, GListPtr node_list) { for (GListPtr ele = rsc->running_on; ele; ele = ele->next) { pe_node_t *node = (pe_node_t *) ele->data; if (pcmk__str_in_list(node_list, node->details->uname)) { return true; } } return false; } bool pcmk__rsc_filtered_by_node(pe_resource_t *rsc, GListPtr only_node) { return (rsc->fns->active(rsc, FALSE) && !pe__rsc_running_on_any_node_in_list(rsc, only_node)); } GListPtr pe__filter_rsc_list(GListPtr rscs, GListPtr filter) { GListPtr retval = NULL; for (GListPtr gIter = rscs; gIter; gIter = gIter->next) { pe_resource_t *rsc = (pe_resource_t *) gIter->data; /* I think the second condition is safe here for all callers of this * function. If not, it needs to move into pe__node_text. */ if (pcmk__str_in_list(filter, rsc_printable_id(rsc)) || (rsc->parent && pcmk__str_in_list(filter, rsc_printable_id(rsc->parent)))) { retval = g_list_prepend(retval, rsc); } } return retval; }