| 1 | /* |
| 2 | * Copyright (C) 2011 EfficiOS Inc. |
| 3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 4 | * Copyright (C) 2013 Jérémie Galarneau <jeremie.galarneau@efficios.com> |
| 5 | * |
| 6 | * SPDX-License-Identifier: GPL-2.0-only |
| 7 | * |
| 8 | */ |
| 9 | |
| 10 | #include "health-sessiond.hpp" |
| 11 | #include "manage-consumer.hpp" |
| 12 | #include "testpoint.hpp" |
| 13 | #include "thread.hpp" |
| 14 | #include "ust-consumer.hpp" |
| 15 | #include "utils.hpp" |
| 16 | |
| 17 | #include <common/pipe.hpp> |
| 18 | #include <common/utils.hpp> |
| 19 | |
| 20 | #include <fcntl.h> |
| 21 | #include <signal.h> |
| 22 | |
| 23 | namespace { |
| 24 | struct thread_notifiers { |
| 25 | struct lttng_pipe *quit_pipe; |
| 26 | struct consumer_data *consumer_data; |
| 27 | sem_t ready; |
| 28 | int initialization_result; |
| 29 | }; |
| 30 | } /* namespace */ |
| 31 | |
| 32 | static void mark_thread_as_ready(struct thread_notifiers *notifiers) |
| 33 | { |
| 34 | DBG("Marking consumer management thread as ready"); |
| 35 | notifiers->initialization_result = 0; |
| 36 | sem_post(¬ifiers->ready); |
| 37 | } |
| 38 | |
| 39 | static void mark_thread_intialization_as_failed(struct thread_notifiers *notifiers) |
| 40 | { |
| 41 | ERR("Consumer management thread entering error state"); |
| 42 | notifiers->initialization_result = -1; |
| 43 | sem_post(¬ifiers->ready); |
| 44 | } |
| 45 | |
| 46 | static void wait_until_thread_is_ready(struct thread_notifiers *notifiers) |
| 47 | { |
| 48 | DBG("Waiting for consumer management thread to be ready"); |
| 49 | sem_wait(¬ifiers->ready); |
| 50 | DBG("Consumer management thread is ready"); |
| 51 | } |
| 52 | |
| 53 | /* |
| 54 | * This thread manage the consumer error sent back to the session daemon. |
| 55 | */ |
| 56 | static void *thread_consumer_management(void *data) |
| 57 | { |
| 58 | int sock = -1, i, ret, err = -1, should_quit = 0; |
| 59 | uint32_t nb_fd; |
| 60 | enum lttcomm_return_code code; |
| 61 | struct lttng_poll_event events; |
| 62 | struct thread_notifiers *notifiers = (thread_notifiers *) data; |
| 63 | struct consumer_data *consumer_data = notifiers->consumer_data; |
| 64 | const auto thread_quit_pipe_fd = lttng_pipe_get_readfd(notifiers->quit_pipe); |
| 65 | struct consumer_socket *cmd_socket_wrapper = nullptr; |
| 66 | |
| 67 | DBG("[thread] Manage consumer started"); |
| 68 | |
| 69 | rcu_register_thread(); |
| 70 | rcu_thread_online(); |
| 71 | |
| 72 | health_register(the_health_sessiond, HEALTH_SESSIOND_TYPE_CONSUMER); |
| 73 | |
| 74 | health_code_update(); |
| 75 | |
| 76 | /* |
| 77 | * Pass 3 as size here for the thread quit pipe, consumerd_err_sock and the |
| 78 | * metadata_sock. Nothing more will be added to this poll set. |
| 79 | */ |
| 80 | ret = lttng_poll_create(&events, 3, LTTNG_CLOEXEC); |
| 81 | if (ret < 0) { |
| 82 | mark_thread_intialization_as_failed(notifiers); |
| 83 | goto error_poll; |
| 84 | } |
| 85 | |
| 86 | ret = lttng_poll_add(&events, thread_quit_pipe_fd, LPOLLIN); |
| 87 | if (ret < 0) { |
| 88 | mark_thread_intialization_as_failed(notifiers); |
| 89 | goto error; |
| 90 | } |
| 91 | |
| 92 | /* |
| 93 | * The error socket here is already in a listening state which was done |
| 94 | * just before spawning this thread to avoid a race between the consumer |
| 95 | * daemon exec trying to connect and the listen() call. |
| 96 | */ |
| 97 | ret = lttng_poll_add(&events, consumer_data->err_sock, LPOLLIN | LPOLLRDHUP); |
| 98 | if (ret < 0) { |
| 99 | mark_thread_intialization_as_failed(notifiers); |
| 100 | goto error; |
| 101 | } |
| 102 | |
| 103 | health_code_update(); |
| 104 | |
| 105 | /* Infinite blocking call, waiting for transmission */ |
| 106 | health_poll_entry(); |
| 107 | |
| 108 | if (testpoint(sessiond_thread_manage_consumer)) { |
| 109 | mark_thread_intialization_as_failed(notifiers); |
| 110 | goto error; |
| 111 | } |
| 112 | |
| 113 | ret = lttng_poll_wait(&events, -1); |
| 114 | health_poll_exit(); |
| 115 | if (ret < 0) { |
| 116 | mark_thread_intialization_as_failed(notifiers); |
| 117 | goto error; |
| 118 | } |
| 119 | |
| 120 | nb_fd = ret; |
| 121 | |
| 122 | for (i = 0; i < nb_fd; i++) { |
| 123 | /* Fetch once the poll data */ |
| 124 | const auto revents = LTTNG_POLL_GETEV(&events, i); |
| 125 | const auto pollfd = LTTNG_POLL_GETFD(&events, i); |
| 126 | |
| 127 | health_code_update(); |
| 128 | |
| 129 | /* Activity on thread quit pipe, exiting. */ |
| 130 | if (pollfd == thread_quit_pipe_fd) { |
| 131 | DBG("Activity on thread quit pipe"); |
| 132 | err = 0; |
| 133 | mark_thread_intialization_as_failed(notifiers); |
| 134 | goto exit; |
| 135 | } else if (pollfd == consumer_data->err_sock) { |
| 136 | /* Event on the registration socket */ |
| 137 | if (revents & LPOLLIN) { |
| 138 | continue; |
| 139 | } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { |
| 140 | ERR("consumer err socket poll error"); |
| 141 | mark_thread_intialization_as_failed(notifiers); |
| 142 | goto error; |
| 143 | } else { |
| 144 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); |
| 145 | mark_thread_intialization_as_failed(notifiers); |
| 146 | goto error; |
| 147 | } |
| 148 | } |
| 149 | } |
| 150 | |
| 151 | sock = lttcomm_accept_unix_sock(consumer_data->err_sock); |
| 152 | if (sock < 0) { |
| 153 | mark_thread_intialization_as_failed(notifiers); |
| 154 | goto error; |
| 155 | } |
| 156 | |
| 157 | /* |
| 158 | * Set the CLOEXEC flag. Return code is useless because either way, the |
| 159 | * show must go on. |
| 160 | */ |
| 161 | (void) utils_set_fd_cloexec(sock); |
| 162 | |
| 163 | health_code_update(); |
| 164 | |
| 165 | DBG2("Receiving code from consumer err_sock"); |
| 166 | |
| 167 | /* Getting status code from consumerd */ |
| 168 | { |
| 169 | std::int32_t comm_code = 0; |
| 170 | |
| 171 | ret = lttcomm_recv_unix_sock(sock, &comm_code, sizeof(comm_code)); |
| 172 | code = static_cast<decltype(code)>(comm_code); |
| 173 | } |
| 174 | if (ret <= 0) { |
| 175 | mark_thread_intialization_as_failed(notifiers); |
| 176 | goto error; |
| 177 | } |
| 178 | |
| 179 | health_code_update(); |
| 180 | if (code != LTTCOMM_CONSUMERD_COMMAND_SOCK_READY) { |
| 181 | ERR("consumer error when waiting for SOCK_READY : %s", |
| 182 | lttcomm_get_readable_code((lttcomm_return_code) -code)); |
| 183 | mark_thread_intialization_as_failed(notifiers); |
| 184 | goto error; |
| 185 | } |
| 186 | |
| 187 | /* Connect both command and metadata sockets. */ |
| 188 | consumer_data->cmd_sock = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); |
| 189 | consumer_data->metadata_fd = lttcomm_connect_unix_sock(consumer_data->cmd_unix_sock_path); |
| 190 | if (consumer_data->cmd_sock < 0 || consumer_data->metadata_fd < 0) { |
| 191 | PERROR("consumer connect cmd socket"); |
| 192 | mark_thread_intialization_as_failed(notifiers); |
| 193 | goto error; |
| 194 | } |
| 195 | |
| 196 | consumer_data->metadata_sock.fd_ptr = &consumer_data->metadata_fd; |
| 197 | |
| 198 | /* Create metadata socket lock. */ |
| 199 | consumer_data->metadata_sock.lock = zmalloc<pthread_mutex_t>(); |
| 200 | if (consumer_data->metadata_sock.lock == nullptr) { |
| 201 | PERROR("zmalloc pthread mutex"); |
| 202 | mark_thread_intialization_as_failed(notifiers); |
| 203 | goto error; |
| 204 | } |
| 205 | pthread_mutex_init(consumer_data->metadata_sock.lock, nullptr); |
| 206 | |
| 207 | DBG("Consumer command socket ready (fd: %d)", consumer_data->cmd_sock); |
| 208 | DBG("Consumer metadata socket ready (fd: %d)", consumer_data->metadata_fd); |
| 209 | |
| 210 | /* |
| 211 | * Remove the consumerd error sock since we've established a connection. |
| 212 | */ |
| 213 | ret = lttng_poll_del(&events, consumer_data->err_sock); |
| 214 | if (ret < 0) { |
| 215 | mark_thread_intialization_as_failed(notifiers); |
| 216 | goto error; |
| 217 | } |
| 218 | |
| 219 | /* Add new accepted error socket. */ |
| 220 | ret = lttng_poll_add(&events, sock, LPOLLIN | LPOLLRDHUP); |
| 221 | if (ret < 0) { |
| 222 | mark_thread_intialization_as_failed(notifiers); |
| 223 | goto error; |
| 224 | } |
| 225 | |
| 226 | /* Add metadata socket that is successfully connected. */ |
| 227 | ret = lttng_poll_add(&events, consumer_data->metadata_fd, LPOLLIN | LPOLLRDHUP); |
| 228 | if (ret < 0) { |
| 229 | mark_thread_intialization_as_failed(notifiers); |
| 230 | goto error; |
| 231 | } |
| 232 | |
| 233 | health_code_update(); |
| 234 | |
| 235 | /* |
| 236 | * Transfer the write-end of the channel monitoring pipe to the consumer |
| 237 | * by issuing a SET_CHANNEL_MONITOR_PIPE command. |
| 238 | */ |
| 239 | cmd_socket_wrapper = consumer_allocate_socket(&consumer_data->cmd_sock); |
| 240 | if (!cmd_socket_wrapper) { |
| 241 | mark_thread_intialization_as_failed(notifiers); |
| 242 | goto error; |
| 243 | } |
| 244 | cmd_socket_wrapper->lock = &consumer_data->lock; |
| 245 | |
| 246 | pthread_mutex_lock(cmd_socket_wrapper->lock); |
| 247 | ret = consumer_init(cmd_socket_wrapper, the_sessiond_uuid); |
| 248 | if (ret) { |
| 249 | ERR("Failed to send sessiond uuid to consumer daemon"); |
| 250 | mark_thread_intialization_as_failed(notifiers); |
| 251 | pthread_mutex_unlock(cmd_socket_wrapper->lock); |
| 252 | goto error; |
| 253 | } |
| 254 | pthread_mutex_unlock(cmd_socket_wrapper->lock); |
| 255 | |
| 256 | ret = consumer_send_channel_monitor_pipe(cmd_socket_wrapper, |
| 257 | consumer_data->channel_monitor_pipe); |
| 258 | if (ret) { |
| 259 | mark_thread_intialization_as_failed(notifiers); |
| 260 | goto error; |
| 261 | } |
| 262 | |
| 263 | /* Discard the socket wrapper as it is no longer needed. */ |
| 264 | consumer_destroy_socket(cmd_socket_wrapper); |
| 265 | cmd_socket_wrapper = nullptr; |
| 266 | |
| 267 | /* The thread is completely initialized, signal that it is ready. */ |
| 268 | mark_thread_as_ready(notifiers); |
| 269 | |
| 270 | /* Infinite blocking call, waiting for transmission */ |
| 271 | while (true) { |
| 272 | health_code_update(); |
| 273 | |
| 274 | /* Exit the thread because the thread quit pipe has been triggered. */ |
| 275 | if (should_quit) { |
| 276 | /* Not a health error. */ |
| 277 | err = 0; |
| 278 | goto exit; |
| 279 | } |
| 280 | |
| 281 | health_poll_entry(); |
| 282 | ret = lttng_poll_wait(&events, -1); |
| 283 | health_poll_exit(); |
| 284 | if (ret < 0) { |
| 285 | goto error; |
| 286 | } |
| 287 | |
| 288 | nb_fd = ret; |
| 289 | |
| 290 | for (i = 0; i < nb_fd; i++) { |
| 291 | /* Fetch once the poll data */ |
| 292 | const auto revents = LTTNG_POLL_GETEV(&events, i); |
| 293 | const auto pollfd = LTTNG_POLL_GETFD(&events, i); |
| 294 | |
| 295 | health_code_update(); |
| 296 | |
| 297 | /* |
| 298 | * Thread quit pipe has been triggered, flag that we should stop |
| 299 | * but continue the current loop to handle potential data from |
| 300 | * consumer. |
| 301 | */ |
| 302 | if (pollfd == thread_quit_pipe_fd) { |
| 303 | should_quit = 1; |
| 304 | } else if (pollfd == sock) { |
| 305 | /* Event on the consumerd socket */ |
| 306 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) && |
| 307 | !(revents & LPOLLIN)) { |
| 308 | ERR("consumer err socket second poll error"); |
| 309 | goto error; |
| 310 | } |
| 311 | health_code_update(); |
| 312 | /* Wait for any consumerd error */ |
| 313 | { |
| 314 | std::int32_t comm_code = 0; |
| 315 | |
| 316 | ret = lttcomm_recv_unix_sock( |
| 317 | sock, &comm_code, sizeof(comm_code)); |
| 318 | code = static_cast<decltype(code)>(comm_code); |
| 319 | } |
| 320 | if (ret <= 0) { |
| 321 | ERR("consumer closed the command socket"); |
| 322 | goto error; |
| 323 | } |
| 324 | |
| 325 | ERR("consumer return code : %s", |
| 326 | lttcomm_get_readable_code((lttcomm_return_code) -code)); |
| 327 | |
| 328 | goto exit; |
| 329 | } else if (pollfd == consumer_data->metadata_fd) { |
| 330 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP) && |
| 331 | !(revents & LPOLLIN)) { |
| 332 | ERR("consumer err metadata socket second poll error"); |
| 333 | goto error; |
| 334 | } |
| 335 | /* UST metadata requests */ |
| 336 | ret = ust_consumer_metadata_request(&consumer_data->metadata_sock); |
| 337 | if (ret < 0) { |
| 338 | ERR("Handling metadata request"); |
| 339 | goto error; |
| 340 | } |
| 341 | } |
| 342 | /* No need for an else branch all FDs are tested prior. */ |
| 343 | } |
| 344 | health_code_update(); |
| 345 | } |
| 346 | |
| 347 | exit: |
| 348 | error: |
| 349 | /* |
| 350 | * We lock here because we are about to close the sockets and some other |
| 351 | * thread might be using them so get exclusive access which will abort all |
| 352 | * other consumer command by other threads. |
| 353 | */ |
| 354 | pthread_mutex_lock(&consumer_data->lock); |
| 355 | |
| 356 | /* Immediately set the consumerd state to stopped */ |
| 357 | if (consumer_data->type == LTTNG_CONSUMER_KERNEL) { |
| 358 | uatomic_set(&the_kernel_consumerd_state, CONSUMER_ERROR); |
| 359 | } else if (consumer_data->type == LTTNG_CONSUMER64_UST || |
| 360 | consumer_data->type == LTTNG_CONSUMER32_UST) { |
| 361 | uatomic_set(&the_ust_consumerd_state, CONSUMER_ERROR); |
| 362 | } else { |
| 363 | /* Code flow error... */ |
| 364 | abort(); |
| 365 | } |
| 366 | |
| 367 | if (consumer_data->err_sock >= 0) { |
| 368 | ret = close(consumer_data->err_sock); |
| 369 | if (ret) { |
| 370 | PERROR("close"); |
| 371 | } |
| 372 | consumer_data->err_sock = -1; |
| 373 | } |
| 374 | if (consumer_data->cmd_sock >= 0) { |
| 375 | ret = close(consumer_data->cmd_sock); |
| 376 | if (ret) { |
| 377 | PERROR("close"); |
| 378 | } |
| 379 | consumer_data->cmd_sock = -1; |
| 380 | } |
| 381 | if (consumer_data->metadata_sock.fd_ptr && *consumer_data->metadata_sock.fd_ptr >= 0) { |
| 382 | ret = close(*consumer_data->metadata_sock.fd_ptr); |
| 383 | if (ret) { |
| 384 | PERROR("close"); |
| 385 | } |
| 386 | } |
| 387 | if (sock >= 0) { |
| 388 | ret = close(sock); |
| 389 | if (ret) { |
| 390 | PERROR("close"); |
| 391 | } |
| 392 | } |
| 393 | |
| 394 | unlink(consumer_data->err_unix_sock_path); |
| 395 | unlink(consumer_data->cmd_unix_sock_path); |
| 396 | pthread_mutex_unlock(&consumer_data->lock); |
| 397 | |
| 398 | /* Cleanup metadata socket mutex. */ |
| 399 | if (consumer_data->metadata_sock.lock) { |
| 400 | pthread_mutex_destroy(consumer_data->metadata_sock.lock); |
| 401 | free(consumer_data->metadata_sock.lock); |
| 402 | } |
| 403 | lttng_poll_clean(&events); |
| 404 | |
| 405 | if (cmd_socket_wrapper) { |
| 406 | consumer_destroy_socket(cmd_socket_wrapper); |
| 407 | } |
| 408 | error_poll: |
| 409 | if (err) { |
| 410 | health_error(); |
| 411 | ERR("Health error occurred in %s", __func__); |
| 412 | } |
| 413 | health_unregister(the_health_sessiond); |
| 414 | DBG("consumer thread cleanup completed"); |
| 415 | |
| 416 | rcu_thread_offline(); |
| 417 | rcu_unregister_thread(); |
| 418 | |
| 419 | return nullptr; |
| 420 | } |
| 421 | |
| 422 | static bool shutdown_consumer_management_thread(void *data) |
| 423 | { |
| 424 | struct thread_notifiers *notifiers = (thread_notifiers *) data; |
| 425 | const int write_fd = lttng_pipe_get_writefd(notifiers->quit_pipe); |
| 426 | |
| 427 | return notify_thread_pipe(write_fd) == 1; |
| 428 | } |
| 429 | |
| 430 | static void cleanup_consumer_management_thread(void *data) |
| 431 | { |
| 432 | struct thread_notifiers *notifiers = (thread_notifiers *) data; |
| 433 | |
| 434 | lttng_pipe_destroy(notifiers->quit_pipe); |
| 435 | free(notifiers); |
| 436 | } |
| 437 | |
| 438 | bool launch_consumer_management_thread(struct consumer_data *consumer_data) |
| 439 | { |
| 440 | struct lttng_pipe *quit_pipe; |
| 441 | struct thread_notifiers *notifiers = nullptr; |
| 442 | struct lttng_thread *thread; |
| 443 | |
| 444 | notifiers = zmalloc<thread_notifiers>(); |
| 445 | if (!notifiers) { |
| 446 | goto error_alloc; |
| 447 | } |
| 448 | |
| 449 | quit_pipe = lttng_pipe_open(FD_CLOEXEC); |
| 450 | if (!quit_pipe) { |
| 451 | goto error; |
| 452 | } |
| 453 | notifiers->quit_pipe = quit_pipe; |
| 454 | notifiers->consumer_data = consumer_data; |
| 455 | sem_init(¬ifiers->ready, 0, 0); |
| 456 | |
| 457 | thread = lttng_thread_create("Consumer management", |
| 458 | thread_consumer_management, |
| 459 | shutdown_consumer_management_thread, |
| 460 | cleanup_consumer_management_thread, |
| 461 | notifiers); |
| 462 | if (!thread) { |
| 463 | goto error; |
| 464 | } |
| 465 | wait_until_thread_is_ready(notifiers); |
| 466 | lttng_thread_put(thread); |
| 467 | return notifiers->initialization_result == 0; |
| 468 | error: |
| 469 | cleanup_consumer_management_thread(notifiers); |
| 470 | error_alloc: |
| 471 | return false; |
| 472 | } |