| 1 | /* |
| 2 | * Copyright (C) 2020 Jérémie Galarneau <jeremie.galarneau@efficios.com> |
| 3 | * |
| 4 | * SPDX-License-Identifier: GPL-2.0-only |
| 5 | * |
| 6 | */ |
| 7 | |
| 8 | #include "action-executor.h" |
| 9 | #include "cmd.h" |
| 10 | #include "health-sessiond.h" |
| 11 | #include "lttng-sessiond.h" |
| 12 | #include "notification-thread-internal.h" |
| 13 | #include "session.h" |
| 14 | #include "thread.h" |
| 15 | #include <common/macros.h> |
| 16 | #include <common/optional.h> |
| 17 | #include <lttng/action/action-internal.h> |
| 18 | #include <lttng/action/group-internal.h> |
| 19 | #include <lttng/action/group.h> |
| 20 | #include <lttng/action/notify-internal.h> |
| 21 | #include <lttng/action/notify.h> |
| 22 | #include <lttng/action/rotate-session.h> |
| 23 | #include <lttng/action/snapshot-session.h> |
| 24 | #include <lttng/action/start-session.h> |
| 25 | #include <lttng/action/stop-session.h> |
| 26 | #include <lttng/condition/evaluation.h> |
| 27 | #include <lttng/condition/on-event-internal.h> |
| 28 | #include <lttng/lttng-error.h> |
| 29 | #include <lttng/trigger/trigger-internal.h> |
| 30 | #include <pthread.h> |
| 31 | #include <stdbool.h> |
| 32 | #include <stddef.h> |
| 33 | #include <urcu/list.h> |
| 34 | |
| 35 | #define THREAD_NAME "Action Executor" |
| 36 | #define MAX_QUEUED_WORK_COUNT 8192 |
| 37 | |
| 38 | struct action_work_item { |
| 39 | uint64_t id; |
| 40 | struct lttng_trigger *trigger; |
| 41 | struct lttng_evaluation *evaluation; |
| 42 | struct notification_client_list *client_list; |
| 43 | LTTNG_OPTIONAL(struct lttng_credentials) object_creds; |
| 44 | struct cds_list_head list_node; |
| 45 | }; |
| 46 | |
| 47 | struct action_executor { |
| 48 | struct lttng_thread *thread; |
| 49 | struct notification_thread_handle *notification_thread_handle; |
| 50 | struct { |
| 51 | uint64_t pending_count; |
| 52 | struct cds_list_head list; |
| 53 | pthread_cond_t cond; |
| 54 | pthread_mutex_t lock; |
| 55 | } work; |
| 56 | bool should_quit; |
| 57 | uint64_t next_work_item_id; |
| 58 | }; |
| 59 | |
| 60 | /* |
| 61 | * Only return non-zero on a fatal error that should shut down the action |
| 62 | * executor. |
| 63 | */ |
| 64 | typedef int (*action_executor_handler)(struct action_executor *executor, |
| 65 | const struct action_work_item *, |
| 66 | struct lttng_action *action); |
| 67 | |
| 68 | static int action_executor_notify_handler(struct action_executor *executor, |
| 69 | const struct action_work_item *, |
| 70 | struct lttng_action *); |
| 71 | static int action_executor_start_session_handler( |
| 72 | struct action_executor *executor, |
| 73 | const struct action_work_item *, |
| 74 | struct lttng_action *); |
| 75 | static int action_executor_stop_session_handler( |
| 76 | struct action_executor *executor, |
| 77 | const struct action_work_item *, |
| 78 | struct lttng_action *); |
| 79 | static int action_executor_rotate_session_handler( |
| 80 | struct action_executor *executor, |
| 81 | const struct action_work_item *, |
| 82 | struct lttng_action *); |
| 83 | static int action_executor_snapshot_session_handler( |
| 84 | struct action_executor *executor, |
| 85 | const struct action_work_item *, |
| 86 | struct lttng_action *); |
| 87 | static int action_executor_group_handler(struct action_executor *executor, |
| 88 | const struct action_work_item *, |
| 89 | struct lttng_action *); |
| 90 | static int action_executor_generic_handler(struct action_executor *executor, |
| 91 | const struct action_work_item *, |
| 92 | struct lttng_action *); |
| 93 | |
| 94 | static const action_executor_handler action_executors[] = { |
| 95 | [LTTNG_ACTION_TYPE_NOTIFY] = action_executor_notify_handler, |
| 96 | [LTTNG_ACTION_TYPE_START_SESSION] = action_executor_start_session_handler, |
| 97 | [LTTNG_ACTION_TYPE_STOP_SESSION] = action_executor_stop_session_handler, |
| 98 | [LTTNG_ACTION_TYPE_ROTATE_SESSION] = action_executor_rotate_session_handler, |
| 99 | [LTTNG_ACTION_TYPE_SNAPSHOT_SESSION] = action_executor_snapshot_session_handler, |
| 100 | [LTTNG_ACTION_TYPE_GROUP] = action_executor_group_handler, |
| 101 | }; |
| 102 | |
| 103 | static const char *get_action_name(const struct lttng_action *action) |
| 104 | { |
| 105 | const enum lttng_action_type action_type = lttng_action_get_type(action); |
| 106 | |
| 107 | assert(action_type != LTTNG_ACTION_TYPE_UNKNOWN); |
| 108 | |
| 109 | return lttng_action_type_string(action_type); |
| 110 | } |
| 111 | |
| 112 | /* Check if this trigger allowed to interect with a given session. */ |
| 113 | static bool is_trigger_allowed_for_session(const struct lttng_trigger *trigger, |
| 114 | struct ltt_session *session) |
| 115 | { |
| 116 | bool is_allowed = false; |
| 117 | const struct lttng_credentials session_creds = { |
| 118 | .uid = LTTNG_OPTIONAL_INIT_VALUE(session->uid), |
| 119 | .gid = LTTNG_OPTIONAL_INIT_VALUE(session->gid), |
| 120 | }; |
| 121 | /* Can never be NULL. */ |
| 122 | const struct lttng_credentials *trigger_creds = |
| 123 | lttng_trigger_get_credentials(trigger); |
| 124 | |
| 125 | is_allowed = (lttng_credentials_is_equal_uid(trigger_creds, &session_creds)) || |
| 126 | (lttng_credentials_get_uid(trigger_creds) == 0); |
| 127 | if (!is_allowed) { |
| 128 | WARN("Trigger is not allowed to interact with session `%s`: session uid = %ld, session gid = %ld, trigger uid = %ld", |
| 129 | session->name, |
| 130 | (long int) session->uid, |
| 131 | (long int) session->gid, |
| 132 | (long int) lttng_credentials_get_uid(trigger_creds)); |
| 133 | } |
| 134 | |
| 135 | return is_allowed; |
| 136 | } |
| 137 | |
| 138 | static const char *get_trigger_name(const struct lttng_trigger *trigger) |
| 139 | { |
| 140 | const char *trigger_name; |
| 141 | enum lttng_trigger_status trigger_status; |
| 142 | |
| 143 | trigger_status = lttng_trigger_get_name(trigger, &trigger_name); |
| 144 | assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); |
| 145 | |
| 146 | return trigger_name; |
| 147 | } |
| 148 | |
| 149 | static int client_handle_transmission_status( |
| 150 | struct notification_client *client, |
| 151 | enum client_transmission_status status, |
| 152 | void *user_data) |
| 153 | { |
| 154 | int ret = 0; |
| 155 | struct action_executor *executor = user_data; |
| 156 | bool update_communication = true; |
| 157 | |
| 158 | switch (status) { |
| 159 | case CLIENT_TRANSMISSION_STATUS_COMPLETE: |
| 160 | DBG("Successfully sent full notification to client, client_id = %" PRIu64, |
| 161 | client->id); |
| 162 | update_communication = false; |
| 163 | break; |
| 164 | case CLIENT_TRANSMISSION_STATUS_QUEUED: |
| 165 | DBG("Queued notification in client outgoing buffer, client_id = %" PRIu64, |
| 166 | client->id); |
| 167 | break; |
| 168 | case CLIENT_TRANSMISSION_STATUS_FAIL: |
| 169 | DBG("Communication error occurred while sending notification to client, client_id = %" PRIu64, |
| 170 | client->id); |
| 171 | break; |
| 172 | default: |
| 173 | ERR("Fatal error encoutered while sending notification to client, client_id = %" PRIu64, |
| 174 | client->id); |
| 175 | ret = -1; |
| 176 | goto end; |
| 177 | } |
| 178 | |
| 179 | if (!update_communication) { |
| 180 | goto end; |
| 181 | } |
| 182 | |
| 183 | /* Safe to read client's id without locking as it is immutable. */ |
| 184 | ret = notification_thread_client_communication_update( |
| 185 | executor->notification_thread_handle, client->id, |
| 186 | status); |
| 187 | end: |
| 188 | return ret; |
| 189 | } |
| 190 | |
| 191 | static int action_executor_notify_handler(struct action_executor *executor, |
| 192 | const struct action_work_item *work_item, |
| 193 | struct lttng_action *action) |
| 194 | { |
| 195 | return notification_client_list_send_evaluation(work_item->client_list, |
| 196 | work_item->trigger, |
| 197 | work_item->evaluation, |
| 198 | work_item->object_creds.is_set ? |
| 199 | &(work_item->object_creds.value) : |
| 200 | NULL, |
| 201 | client_handle_transmission_status, executor); |
| 202 | } |
| 203 | |
| 204 | static int action_executor_start_session_handler( |
| 205 | struct action_executor *executor, |
| 206 | const struct action_work_item *work_item, |
| 207 | struct lttng_action *action) |
| 208 | { |
| 209 | int ret = 0; |
| 210 | const char *session_name; |
| 211 | enum lttng_action_status action_status; |
| 212 | struct ltt_session *session; |
| 213 | enum lttng_error_code cmd_ret; |
| 214 | |
| 215 | action_status = lttng_action_start_session_get_session_name( |
| 216 | action, &session_name); |
| 217 | if (action_status != LTTNG_ACTION_STATUS_OK) { |
| 218 | ERR("Failed to get session name from `%s` action", |
| 219 | get_action_name(action)); |
| 220 | ret = -1; |
| 221 | goto end; |
| 222 | } |
| 223 | |
| 224 | session_lock_list(); |
| 225 | session = session_find_by_name(session_name); |
| 226 | if (!session) { |
| 227 | DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`", |
| 228 | session_name, get_action_name(action), |
| 229 | get_trigger_name(work_item->trigger)); |
| 230 | goto error_unlock_list; |
| 231 | } |
| 232 | |
| 233 | session_lock(session); |
| 234 | if (!is_trigger_allowed_for_session(work_item->trigger, session)) { |
| 235 | goto error_dispose_session; |
| 236 | } |
| 237 | |
| 238 | cmd_ret = cmd_start_trace(session); |
| 239 | switch (cmd_ret) { |
| 240 | case LTTNG_OK: |
| 241 | DBG("Successfully started session `%s` on behalf of trigger `%s`", |
| 242 | session_name, get_trigger_name(work_item->trigger)); |
| 243 | break; |
| 244 | case LTTNG_ERR_TRACE_ALREADY_STARTED: |
| 245 | DBG("Attempted to start session `%s` on behalf of trigger `%s` but it was already started", |
| 246 | session_name, get_trigger_name(work_item->trigger)); |
| 247 | break; |
| 248 | default: |
| 249 | WARN("Failed to start session `%s` on behalf of trigger `%s`: %s", |
| 250 | session_name, get_trigger_name(work_item->trigger), |
| 251 | lttng_strerror(-cmd_ret)); |
| 252 | lttng_action_increase_execution_failure_count(action); |
| 253 | break; |
| 254 | } |
| 255 | |
| 256 | error_dispose_session: |
| 257 | session_unlock(session); |
| 258 | session_put(session); |
| 259 | error_unlock_list: |
| 260 | session_unlock_list(); |
| 261 | end: |
| 262 | return ret; |
| 263 | } |
| 264 | |
| 265 | static int action_executor_stop_session_handler( |
| 266 | struct action_executor *executor, |
| 267 | const struct action_work_item *work_item, |
| 268 | struct lttng_action *action) |
| 269 | { |
| 270 | int ret = 0; |
| 271 | const char *session_name; |
| 272 | enum lttng_action_status action_status; |
| 273 | struct ltt_session *session; |
| 274 | enum lttng_error_code cmd_ret; |
| 275 | |
| 276 | action_status = lttng_action_stop_session_get_session_name( |
| 277 | action, &session_name); |
| 278 | if (action_status != LTTNG_ACTION_STATUS_OK) { |
| 279 | ERR("Failed to get session name from `%s` action", |
| 280 | get_action_name(action)); |
| 281 | ret = -1; |
| 282 | goto end; |
| 283 | } |
| 284 | |
| 285 | session_lock_list(); |
| 286 | session = session_find_by_name(session_name); |
| 287 | if (!session) { |
| 288 | DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`", |
| 289 | session_name, get_action_name(action), |
| 290 | get_trigger_name(work_item->trigger)); |
| 291 | lttng_action_increase_execution_failure_count(action); |
| 292 | goto error_unlock_list; |
| 293 | } |
| 294 | |
| 295 | session_lock(session); |
| 296 | if (!is_trigger_allowed_for_session(work_item->trigger, session)) { |
| 297 | goto error_dispose_session; |
| 298 | } |
| 299 | |
| 300 | cmd_ret = cmd_stop_trace(session); |
| 301 | switch (cmd_ret) { |
| 302 | case LTTNG_OK: |
| 303 | DBG("Successfully stopped session `%s` on behalf of trigger `%s`", |
| 304 | session_name, get_trigger_name(work_item->trigger)); |
| 305 | break; |
| 306 | case LTTNG_ERR_TRACE_ALREADY_STOPPED: |
| 307 | DBG("Attempted to stop session `%s` on behalf of trigger `%s` but it was already stopped", |
| 308 | session_name, get_trigger_name(work_item->trigger)); |
| 309 | break; |
| 310 | default: |
| 311 | WARN("Failed to stop session `%s` on behalf of trigger `%s`: %s", |
| 312 | session_name, get_trigger_name(work_item->trigger), |
| 313 | lttng_strerror(-cmd_ret)); |
| 314 | lttng_action_increase_execution_failure_count(action); |
| 315 | break; |
| 316 | } |
| 317 | |
| 318 | error_dispose_session: |
| 319 | session_unlock(session); |
| 320 | session_put(session); |
| 321 | error_unlock_list: |
| 322 | session_unlock_list(); |
| 323 | end: |
| 324 | return ret; |
| 325 | } |
| 326 | |
| 327 | static int action_executor_rotate_session_handler( |
| 328 | struct action_executor *executor, |
| 329 | const struct action_work_item *work_item, |
| 330 | struct lttng_action *action) |
| 331 | { |
| 332 | int ret = 0; |
| 333 | const char *session_name; |
| 334 | enum lttng_action_status action_status; |
| 335 | struct ltt_session *session; |
| 336 | enum lttng_error_code cmd_ret; |
| 337 | |
| 338 | action_status = lttng_action_rotate_session_get_session_name( |
| 339 | action, &session_name); |
| 340 | if (action_status != LTTNG_ACTION_STATUS_OK) { |
| 341 | ERR("Failed to get session name from `%s` action", |
| 342 | get_action_name(action)); |
| 343 | ret = -1; |
| 344 | goto end; |
| 345 | } |
| 346 | |
| 347 | session_lock_list(); |
| 348 | session = session_find_by_name(session_name); |
| 349 | if (!session) { |
| 350 | DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`", |
| 351 | session_name, get_action_name(action), |
| 352 | get_trigger_name(work_item->trigger)); |
| 353 | lttng_action_increase_execution_failure_count(action); |
| 354 | goto error_unlock_list; |
| 355 | } |
| 356 | |
| 357 | session_lock(session); |
| 358 | if (!is_trigger_allowed_for_session(work_item->trigger, session)) { |
| 359 | goto error_dispose_session; |
| 360 | } |
| 361 | |
| 362 | cmd_ret = cmd_rotate_session(session, NULL, false, |
| 363 | LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED); |
| 364 | switch (cmd_ret) { |
| 365 | case LTTNG_OK: |
| 366 | DBG("Successfully started rotation of session `%s` on behalf of trigger `%s`", |
| 367 | session_name, get_trigger_name(work_item->trigger)); |
| 368 | break; |
| 369 | case LTTNG_ERR_ROTATION_PENDING: |
| 370 | DBG("Attempted to start a rotation of session `%s` on behalf of trigger `%s` but a rotation is already ongoing", |
| 371 | session_name, get_trigger_name(work_item->trigger)); |
| 372 | lttng_action_increase_execution_failure_count(action); |
| 373 | break; |
| 374 | case LTTNG_ERR_ROTATION_MULTIPLE_AFTER_STOP: |
| 375 | case LTTNG_ERR_ROTATION_AFTER_STOP_CLEAR: |
| 376 | DBG("Attempted to start a rotation of session `%s` on behalf of trigger `%s` but a rotation has already been completed since the last stop or clear", |
| 377 | session_name, get_trigger_name(work_item->trigger)); |
| 378 | break; |
| 379 | default: |
| 380 | WARN("Failed to start a rotation of session `%s` on behalf of trigger `%s`: %s", |
| 381 | session_name, get_trigger_name(work_item->trigger), |
| 382 | lttng_strerror(-cmd_ret)); |
| 383 | lttng_action_increase_execution_failure_count(action); |
| 384 | break; |
| 385 | } |
| 386 | |
| 387 | error_dispose_session: |
| 388 | session_unlock(session); |
| 389 | session_put(session); |
| 390 | error_unlock_list: |
| 391 | session_unlock_list(); |
| 392 | end: |
| 393 | return ret; |
| 394 | } |
| 395 | |
| 396 | static int action_executor_snapshot_session_handler( |
| 397 | struct action_executor *executor, |
| 398 | const struct action_work_item *work_item, |
| 399 | struct lttng_action *action) |
| 400 | { |
| 401 | int ret = 0; |
| 402 | const char *session_name; |
| 403 | enum lttng_action_status action_status; |
| 404 | struct ltt_session *session; |
| 405 | const struct lttng_snapshot_output default_snapshot_output = { |
| 406 | .max_size = UINT64_MAX, |
| 407 | }; |
| 408 | const struct lttng_snapshot_output *snapshot_output = |
| 409 | &default_snapshot_output; |
| 410 | enum lttng_error_code cmd_ret; |
| 411 | |
| 412 | action_status = lttng_action_snapshot_session_get_session_name( |
| 413 | action, &session_name); |
| 414 | if (action_status != LTTNG_ACTION_STATUS_OK) { |
| 415 | ERR("Failed to get session name from `%s` action", |
| 416 | get_action_name(action)); |
| 417 | ret = -1; |
| 418 | goto end; |
| 419 | } |
| 420 | |
| 421 | action_status = lttng_action_snapshot_session_get_output( |
| 422 | action, &snapshot_output); |
| 423 | if (action_status != LTTNG_ACTION_STATUS_OK && |
| 424 | action_status != LTTNG_ACTION_STATUS_UNSET) { |
| 425 | ERR("Failed to get output from `%s` action", |
| 426 | get_action_name(action)); |
| 427 | ret = -1; |
| 428 | goto end; |
| 429 | } |
| 430 | |
| 431 | session_lock_list(); |
| 432 | session = session_find_by_name(session_name); |
| 433 | if (!session) { |
| 434 | DBG("Failed to find session `%s` by name while executing `%s` action of trigger `%s`", |
| 435 | session_name, get_action_name(action), |
| 436 | get_trigger_name(work_item->trigger)); |
| 437 | lttng_action_increase_execution_failure_count(action); |
| 438 | goto error_unlock_list; |
| 439 | } |
| 440 | |
| 441 | |
| 442 | session_lock(session); |
| 443 | if (!is_trigger_allowed_for_session(work_item->trigger, session)) { |
| 444 | goto error_dispose_session; |
| 445 | } |
| 446 | |
| 447 | cmd_ret = cmd_snapshot_record(session, snapshot_output, 0); |
| 448 | switch (cmd_ret) { |
| 449 | case LTTNG_OK: |
| 450 | DBG("Successfully recorded snapshot of session `%s` on behalf of trigger `%s`", |
| 451 | session_name, get_trigger_name(work_item->trigger)); |
| 452 | break; |
| 453 | default: |
| 454 | WARN("Failed to record snapshot of session `%s` on behalf of trigger `%s`: %s", |
| 455 | session_name, get_trigger_name(work_item->trigger), |
| 456 | lttng_strerror(-cmd_ret)); |
| 457 | lttng_action_increase_execution_failure_count(action); |
| 458 | break; |
| 459 | } |
| 460 | |
| 461 | error_dispose_session: |
| 462 | session_unlock(session); |
| 463 | session_put(session); |
| 464 | error_unlock_list: |
| 465 | session_unlock_list(); |
| 466 | end: |
| 467 | return ret; |
| 468 | } |
| 469 | |
| 470 | static int action_executor_group_handler(struct action_executor *executor, |
| 471 | const struct action_work_item *work_item, |
| 472 | struct lttng_action *action_group) |
| 473 | { |
| 474 | int ret = 0; |
| 475 | unsigned int i, count; |
| 476 | enum lttng_action_status action_status; |
| 477 | |
| 478 | action_status = lttng_action_group_get_count(action_group, &count); |
| 479 | if (action_status != LTTNG_ACTION_STATUS_OK) { |
| 480 | /* Fatal error. */ |
| 481 | ERR("Failed to get count of action in action group"); |
| 482 | ret = -1; |
| 483 | goto end; |
| 484 | } |
| 485 | |
| 486 | DBG("Action group has %u action%s", count, count != 1 ? "s" : ""); |
| 487 | for (i = 0; i < count; i++) { |
| 488 | struct lttng_action *action = |
| 489 | lttng_action_group_borrow_mutable_at_index( |
| 490 | action_group, i); |
| 491 | |
| 492 | ret = action_executor_generic_handler( |
| 493 | executor, work_item, action); |
| 494 | if (ret) { |
| 495 | ERR("Stopping the execution of the action group of trigger `%s` following a fatal error", |
| 496 | get_trigger_name(work_item->trigger)); |
| 497 | goto end; |
| 498 | } |
| 499 | } |
| 500 | end: |
| 501 | return ret; |
| 502 | } |
| 503 | |
| 504 | static int action_executor_generic_handler(struct action_executor *executor, |
| 505 | const struct action_work_item *work_item, |
| 506 | struct lttng_action *action) |
| 507 | { |
| 508 | int ret; |
| 509 | const enum lttng_action_type action_type = lttng_action_get_type(action); |
| 510 | |
| 511 | assert(action_type != LTTNG_ACTION_TYPE_UNKNOWN); |
| 512 | |
| 513 | lttng_action_increase_execution_request_count(action); |
| 514 | if (!lttng_action_should_execute(action)) { |
| 515 | DBG("Policy prevented execution of action `%s` of trigger `%s` action work item %" PRIu64, |
| 516 | get_action_name(action), |
| 517 | get_trigger_name(work_item->trigger), |
| 518 | work_item->id); |
| 519 | ret = 0; |
| 520 | goto end; |
| 521 | } |
| 522 | |
| 523 | lttng_action_increase_execution_count(action); |
| 524 | DBG("Executing action `%s` of trigger `%s` action work item %" PRIu64, |
| 525 | get_action_name(action), |
| 526 | get_trigger_name(work_item->trigger), |
| 527 | work_item->id); |
| 528 | ret = action_executors[action_type](executor, work_item, action); |
| 529 | end: |
| 530 | return ret; |
| 531 | } |
| 532 | |
| 533 | static int action_work_item_execute(struct action_executor *executor, |
| 534 | struct action_work_item *work_item) |
| 535 | { |
| 536 | int ret; |
| 537 | struct lttng_action *action = |
| 538 | lttng_trigger_get_action(work_item->trigger); |
| 539 | |
| 540 | DBG("Starting execution of action work item %" PRIu64 " of trigger `%s`", |
| 541 | work_item->id, get_trigger_name(work_item->trigger)); |
| 542 | ret = action_executor_generic_handler(executor, work_item, action); |
| 543 | DBG("Completed execution of action work item %" PRIu64 " of trigger `%s`", |
| 544 | work_item->id, get_trigger_name(work_item->trigger)); |
| 545 | return ret; |
| 546 | } |
| 547 | |
| 548 | static void action_work_item_destroy(struct action_work_item *work_item) |
| 549 | { |
| 550 | lttng_trigger_put(work_item->trigger); |
| 551 | lttng_evaluation_destroy(work_item->evaluation); |
| 552 | notification_client_list_put(work_item->client_list); |
| 553 | free(work_item); |
| 554 | } |
| 555 | |
| 556 | static void *action_executor_thread(void *_data) |
| 557 | { |
| 558 | struct action_executor *executor = _data; |
| 559 | |
| 560 | assert(executor); |
| 561 | |
| 562 | health_register(the_health_sessiond, |
| 563 | HEALTH_SESSIOND_TYPE_ACTION_EXECUTOR); |
| 564 | |
| 565 | rcu_register_thread(); |
| 566 | rcu_thread_online(); |
| 567 | |
| 568 | DBG("Entering work execution loop"); |
| 569 | pthread_mutex_lock(&executor->work.lock); |
| 570 | while (!executor->should_quit) { |
| 571 | int ret; |
| 572 | struct action_work_item *work_item; |
| 573 | |
| 574 | health_code_update(); |
| 575 | if (executor->work.pending_count == 0) { |
| 576 | health_poll_entry(); |
| 577 | DBG("No work items enqueued, entering wait"); |
| 578 | pthread_cond_wait(&executor->work.cond, |
| 579 | &executor->work.lock); |
| 580 | DBG("Woke-up from wait"); |
| 581 | health_poll_exit(); |
| 582 | continue; |
| 583 | } |
| 584 | |
| 585 | /* Pop item from front of the list with work lock held. */ |
| 586 | work_item = cds_list_first_entry(&executor->work.list, |
| 587 | struct action_work_item, list_node); |
| 588 | cds_list_del(&work_item->list_node); |
| 589 | executor->work.pending_count--; |
| 590 | |
| 591 | /* |
| 592 | * Work can be performed without holding the work lock, |
| 593 | * allowing new items to be queued. |
| 594 | */ |
| 595 | pthread_mutex_unlock(&executor->work.lock); |
| 596 | |
| 597 | /* Execute item only if a trigger is registered. */ |
| 598 | lttng_trigger_lock(work_item->trigger); |
| 599 | if (!lttng_trigger_is_registered(work_item->trigger)) { |
| 600 | const char *trigger_name = NULL; |
| 601 | uid_t trigger_owner_uid; |
| 602 | enum lttng_trigger_status trigger_status; |
| 603 | |
| 604 | trigger_status = lttng_trigger_get_name( |
| 605 | work_item->trigger, &trigger_name); |
| 606 | switch (trigger_status) { |
| 607 | case LTTNG_TRIGGER_STATUS_OK: |
| 608 | break; |
| 609 | case LTTNG_TRIGGER_STATUS_UNSET: |
| 610 | trigger_name = "(unset)"; |
| 611 | break; |
| 612 | default: |
| 613 | abort(); |
| 614 | } |
| 615 | |
| 616 | trigger_status = lttng_trigger_get_owner_uid( |
| 617 | work_item->trigger, &trigger_owner_uid); |
| 618 | assert(trigger_status == LTTNG_TRIGGER_STATUS_OK); |
| 619 | |
| 620 | DBG("Work item skipped since the associated trigger is no longer registered: work item id = %" PRIu64 ", trigger name = '%s', trigger owner uid = %d", |
| 621 | work_item->id, trigger_name, |
| 622 | (int) trigger_owner_uid); |
| 623 | ret = 0; |
| 624 | goto skip_execute; |
| 625 | } |
| 626 | |
| 627 | ret = action_work_item_execute(executor, work_item); |
| 628 | |
| 629 | skip_execute: |
| 630 | lttng_trigger_unlock(work_item->trigger); |
| 631 | action_work_item_destroy(work_item); |
| 632 | if (ret) { |
| 633 | /* Fatal error. */ |
| 634 | break; |
| 635 | } |
| 636 | |
| 637 | health_code_update(); |
| 638 | pthread_mutex_lock(&executor->work.lock); |
| 639 | } |
| 640 | |
| 641 | if (executor->should_quit) { |
| 642 | pthread_mutex_unlock(&executor->work.lock); |
| 643 | } |
| 644 | DBG("Left work execution loop"); |
| 645 | |
| 646 | health_code_update(); |
| 647 | |
| 648 | rcu_thread_offline(); |
| 649 | rcu_unregister_thread(); |
| 650 | health_unregister(the_health_sessiond); |
| 651 | |
| 652 | return NULL; |
| 653 | } |
| 654 | |
| 655 | static bool shutdown_action_executor_thread(void *_data) |
| 656 | { |
| 657 | struct action_executor *executor = _data; |
| 658 | |
| 659 | pthread_mutex_lock(&executor->work.lock); |
| 660 | executor->should_quit = true; |
| 661 | pthread_cond_signal(&executor->work.cond); |
| 662 | pthread_mutex_unlock(&executor->work.lock); |
| 663 | return true; |
| 664 | } |
| 665 | |
| 666 | static void clean_up_action_executor_thread(void *_data) |
| 667 | { |
| 668 | struct action_executor *executor = _data; |
| 669 | |
| 670 | assert(cds_list_empty(&executor->work.list)); |
| 671 | |
| 672 | pthread_mutex_destroy(&executor->work.lock); |
| 673 | pthread_cond_destroy(&executor->work.cond); |
| 674 | free(executor); |
| 675 | } |
| 676 | |
| 677 | struct action_executor *action_executor_create( |
| 678 | struct notification_thread_handle *handle) |
| 679 | { |
| 680 | struct action_executor *executor = zmalloc(sizeof(*executor)); |
| 681 | |
| 682 | if (!executor) { |
| 683 | goto end; |
| 684 | } |
| 685 | |
| 686 | CDS_INIT_LIST_HEAD(&executor->work.list); |
| 687 | pthread_cond_init(&executor->work.cond, NULL); |
| 688 | pthread_mutex_init(&executor->work.lock, NULL); |
| 689 | executor->notification_thread_handle = handle; |
| 690 | |
| 691 | executor->thread = lttng_thread_create(THREAD_NAME, |
| 692 | action_executor_thread, shutdown_action_executor_thread, |
| 693 | clean_up_action_executor_thread, executor); |
| 694 | end: |
| 695 | return executor; |
| 696 | } |
| 697 | |
| 698 | void action_executor_destroy(struct action_executor *executor) |
| 699 | { |
| 700 | struct action_work_item *work_item, *tmp; |
| 701 | |
| 702 | /* TODO Wait for work list to drain? */ |
| 703 | lttng_thread_shutdown(executor->thread); |
| 704 | pthread_mutex_lock(&executor->work.lock); |
| 705 | if (executor->work.pending_count != 0) { |
| 706 | WARN("%" PRIu64 |
| 707 | " trigger action%s still queued for execution and will be discarded", |
| 708 | executor->work.pending_count, |
| 709 | executor->work.pending_count == 1 ? " is" : |
| 710 | "s are"); |
| 711 | } |
| 712 | |
| 713 | cds_list_for_each_entry_safe ( |
| 714 | work_item, tmp, &executor->work.list, list_node) { |
| 715 | WARN("Discarding action work item %" PRIu64 |
| 716 | " associated to trigger `%s`", |
| 717 | work_item->id, get_trigger_name(work_item->trigger)); |
| 718 | cds_list_del(&work_item->list_node); |
| 719 | action_work_item_destroy(work_item); |
| 720 | } |
| 721 | pthread_mutex_unlock(&executor->work.lock); |
| 722 | lttng_thread_put(executor->thread); |
| 723 | } |
| 724 | |
| 725 | /* RCU read-lock must be held by the caller. */ |
| 726 | enum action_executor_status action_executor_enqueue( |
| 727 | struct action_executor *executor, |
| 728 | struct lttng_trigger *trigger, |
| 729 | struct lttng_evaluation *evaluation, |
| 730 | const struct lttng_credentials *object_creds, |
| 731 | struct notification_client_list *client_list) |
| 732 | { |
| 733 | enum action_executor_status executor_status = ACTION_EXECUTOR_STATUS_OK; |
| 734 | const uint64_t work_item_id = executor->next_work_item_id++; |
| 735 | struct action_work_item *work_item; |
| 736 | bool signal = false; |
| 737 | |
| 738 | pthread_mutex_lock(&executor->work.lock); |
| 739 | /* Check for queue overflow. */ |
| 740 | if (executor->work.pending_count >= MAX_QUEUED_WORK_COUNT) { |
| 741 | /* Most likely spammy, remove if it is the case. */ |
| 742 | DBG("Refusing to enqueue action for trigger `%s` as work item %" PRIu64 |
| 743 | " (overflow)", get_trigger_name(trigger), work_item_id); |
| 744 | executor_status = ACTION_EXECUTOR_STATUS_OVERFLOW; |
| 745 | goto error_unlock; |
| 746 | } |
| 747 | |
| 748 | work_item = zmalloc(sizeof(*work_item)); |
| 749 | if (!work_item) { |
| 750 | PERROR("Failed to allocate action executor work item on behalf of trigger `%s`", |
| 751 | get_trigger_name(trigger)); |
| 752 | executor_status = ACTION_EXECUTOR_STATUS_ERROR; |
| 753 | goto error_unlock; |
| 754 | } |
| 755 | |
| 756 | lttng_trigger_get(trigger); |
| 757 | if (client_list) { |
| 758 | const bool reference_acquired = |
| 759 | notification_client_list_get(client_list); |
| 760 | |
| 761 | assert(reference_acquired); |
| 762 | } |
| 763 | |
| 764 | *work_item = (typeof(*work_item)){ |
| 765 | .id = work_item_id, |
| 766 | .trigger = trigger, |
| 767 | /* Ownership transferred to the work item. */ |
| 768 | .evaluation = evaluation, |
| 769 | .object_creds = { |
| 770 | .is_set = !!object_creds, |
| 771 | .value = object_creds ? *object_creds : |
| 772 | (typeof(work_item->object_creds.value)) {}, |
| 773 | }, |
| 774 | .client_list = client_list, |
| 775 | .list_node = CDS_LIST_HEAD_INIT(work_item->list_node), |
| 776 | }; |
| 777 | |
| 778 | evaluation = NULL; |
| 779 | cds_list_add_tail(&work_item->list_node, &executor->work.list); |
| 780 | executor->work.pending_count++; |
| 781 | DBG("Enqueued action for trigger `%s` as work item #%" PRIu64, |
| 782 | get_trigger_name(trigger), work_item_id); |
| 783 | signal = true; |
| 784 | |
| 785 | error_unlock: |
| 786 | if (signal) { |
| 787 | pthread_cond_signal(&executor->work.cond); |
| 788 | } |
| 789 | pthread_mutex_unlock(&executor->work.lock); |
| 790 | |
| 791 | lttng_evaluation_destroy(evaluation); |
| 792 | return executor_status; |
| 793 | } |