| 1 | /* |
| 2 | * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca> |
| 3 | * Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 4 | * 2012 - David Goulet <dgoulet@efficios.com> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License, version 2 only, |
| 8 | * as published by the Free Software Foundation. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but WITHOUT |
| 11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 13 | * more details. |
| 14 | * |
| 15 | * You should have received a copy of the GNU General Public License along |
| 16 | * with this program; if not, write to the Free Software Foundation, Inc., |
| 17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 18 | */ |
| 19 | |
| 20 | #define _LGPL_SOURCE |
| 21 | #include <assert.h> |
| 22 | #include <poll.h> |
| 23 | #include <pthread.h> |
| 24 | #include <stdlib.h> |
| 25 | #include <string.h> |
| 26 | #include <sys/mman.h> |
| 27 | #include <sys/socket.h> |
| 28 | #include <sys/types.h> |
| 29 | #include <unistd.h> |
| 30 | #include <inttypes.h> |
| 31 | #include <signal.h> |
| 32 | |
| 33 | #include <bin/lttng-consumerd/health-consumerd.h> |
| 34 | #include <common/common.h> |
| 35 | #include <common/utils.h> |
| 36 | #include <common/time.h> |
| 37 | #include <common/compat/poll.h> |
| 38 | #include <common/compat/endian.h> |
| 39 | #include <common/index/index.h> |
| 40 | #include <common/kernel-ctl/kernel-ctl.h> |
| 41 | #include <common/sessiond-comm/relayd.h> |
| 42 | #include <common/sessiond-comm/sessiond-comm.h> |
| 43 | #include <common/kernel-consumer/kernel-consumer.h> |
| 44 | #include <common/relayd/relayd.h> |
| 45 | #include <common/ust-consumer/ust-consumer.h> |
| 46 | #include <common/consumer/consumer-timer.h> |
| 47 | #include <common/consumer/consumer.h> |
| 48 | #include <common/consumer/consumer-stream.h> |
| 49 | #include <common/consumer/consumer-testpoint.h> |
| 50 | #include <common/align.h> |
| 51 | #include <common/consumer/consumer-metadata-cache.h> |
| 52 | #include <common/trace-chunk.h> |
| 53 | #include <common/trace-chunk-registry.h> |
| 54 | #include <common/string-utils/format.h> |
| 55 | #include <common/dynamic-array.h> |
| 56 | |
| 57 | struct lttng_consumer_global_data consumer_data = { |
| 58 | .stream_count = 0, |
| 59 | .need_update = 1, |
| 60 | .type = LTTNG_CONSUMER_UNKNOWN, |
| 61 | }; |
| 62 | |
| 63 | enum consumer_channel_action { |
| 64 | CONSUMER_CHANNEL_ADD, |
| 65 | CONSUMER_CHANNEL_DEL, |
| 66 | CONSUMER_CHANNEL_QUIT, |
| 67 | }; |
| 68 | |
| 69 | struct consumer_channel_msg { |
| 70 | enum consumer_channel_action action; |
| 71 | struct lttng_consumer_channel *chan; /* add */ |
| 72 | uint64_t key; /* del */ |
| 73 | }; |
| 74 | |
| 75 | /* Flag used to temporarily pause data consumption from testpoints. */ |
| 76 | int data_consumption_paused; |
| 77 | |
| 78 | /* |
| 79 | * Flag to inform the polling thread to quit when all fd hung up. Updated by |
| 80 | * the consumer_thread_receive_fds when it notices that all fds has hung up. |
| 81 | * Also updated by the signal handler (consumer_should_exit()). Read by the |
| 82 | * polling threads. |
| 83 | */ |
| 84 | int consumer_quit; |
| 85 | |
| 86 | /* |
| 87 | * Global hash table containing respectively metadata and data streams. The |
| 88 | * stream element in this ht should only be updated by the metadata poll thread |
| 89 | * for the metadata and the data poll thread for the data. |
| 90 | */ |
| 91 | static struct lttng_ht *metadata_ht; |
| 92 | static struct lttng_ht *data_ht; |
| 93 | |
| 94 | /* |
| 95 | * Notify a thread lttng pipe to poll back again. This usually means that some |
| 96 | * global state has changed so we just send back the thread in a poll wait |
| 97 | * call. |
| 98 | */ |
| 99 | static void notify_thread_lttng_pipe(struct lttng_pipe *pipe) |
| 100 | { |
| 101 | struct lttng_consumer_stream *null_stream = NULL; |
| 102 | |
| 103 | assert(pipe); |
| 104 | |
| 105 | (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream)); |
| 106 | } |
| 107 | |
| 108 | static void notify_health_quit_pipe(int *pipe) |
| 109 | { |
| 110 | ssize_t ret; |
| 111 | |
| 112 | ret = lttng_write(pipe[1], "4", 1); |
| 113 | if (ret < 1) { |
| 114 | PERROR("write consumer health quit"); |
| 115 | } |
| 116 | } |
| 117 | |
| 118 | static void notify_channel_pipe(struct lttng_consumer_local_data *ctx, |
| 119 | struct lttng_consumer_channel *chan, |
| 120 | uint64_t key, |
| 121 | enum consumer_channel_action action) |
| 122 | { |
| 123 | struct consumer_channel_msg msg; |
| 124 | ssize_t ret; |
| 125 | |
| 126 | memset(&msg, 0, sizeof(msg)); |
| 127 | |
| 128 | msg.action = action; |
| 129 | msg.chan = chan; |
| 130 | msg.key = key; |
| 131 | ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg)); |
| 132 | if (ret < sizeof(msg)) { |
| 133 | PERROR("notify_channel_pipe write error"); |
| 134 | } |
| 135 | } |
| 136 | |
| 137 | void notify_thread_del_channel(struct lttng_consumer_local_data *ctx, |
| 138 | uint64_t key) |
| 139 | { |
| 140 | notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL); |
| 141 | } |
| 142 | |
| 143 | static int read_channel_pipe(struct lttng_consumer_local_data *ctx, |
| 144 | struct lttng_consumer_channel **chan, |
| 145 | uint64_t *key, |
| 146 | enum consumer_channel_action *action) |
| 147 | { |
| 148 | struct consumer_channel_msg msg; |
| 149 | ssize_t ret; |
| 150 | |
| 151 | ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg)); |
| 152 | if (ret < sizeof(msg)) { |
| 153 | ret = -1; |
| 154 | goto error; |
| 155 | } |
| 156 | *action = msg.action; |
| 157 | *chan = msg.chan; |
| 158 | *key = msg.key; |
| 159 | error: |
| 160 | return (int) ret; |
| 161 | } |
| 162 | |
| 163 | /* |
| 164 | * Cleanup the stream list of a channel. Those streams are not yet globally |
| 165 | * visible |
| 166 | */ |
| 167 | static void clean_channel_stream_list(struct lttng_consumer_channel *channel) |
| 168 | { |
| 169 | struct lttng_consumer_stream *stream, *stmp; |
| 170 | |
| 171 | assert(channel); |
| 172 | |
| 173 | /* Delete streams that might have been left in the stream list. */ |
| 174 | cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head, |
| 175 | send_node) { |
| 176 | cds_list_del(&stream->send_node); |
| 177 | /* |
| 178 | * Once a stream is added to this list, the buffers were created so we |
| 179 | * have a guarantee that this call will succeed. Setting the monitor |
| 180 | * mode to 0 so we don't lock nor try to delete the stream from the |
| 181 | * global hash table. |
| 182 | */ |
| 183 | stream->monitor = 0; |
| 184 | consumer_stream_destroy(stream, NULL); |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | /* |
| 189 | * Find a stream. The consumer_data.lock must be locked during this |
| 190 | * call. |
| 191 | */ |
| 192 | static struct lttng_consumer_stream *find_stream(uint64_t key, |
| 193 | struct lttng_ht *ht) |
| 194 | { |
| 195 | struct lttng_ht_iter iter; |
| 196 | struct lttng_ht_node_u64 *node; |
| 197 | struct lttng_consumer_stream *stream = NULL; |
| 198 | |
| 199 | assert(ht); |
| 200 | |
| 201 | /* -1ULL keys are lookup failures */ |
| 202 | if (key == (uint64_t) -1ULL) { |
| 203 | return NULL; |
| 204 | } |
| 205 | |
| 206 | rcu_read_lock(); |
| 207 | |
| 208 | lttng_ht_lookup(ht, &key, &iter); |
| 209 | node = lttng_ht_iter_get_node_u64(&iter); |
| 210 | if (node != NULL) { |
| 211 | stream = caa_container_of(node, struct lttng_consumer_stream, node); |
| 212 | } |
| 213 | |
| 214 | rcu_read_unlock(); |
| 215 | |
| 216 | return stream; |
| 217 | } |
| 218 | |
| 219 | static void steal_stream_key(uint64_t key, struct lttng_ht *ht) |
| 220 | { |
| 221 | struct lttng_consumer_stream *stream; |
| 222 | |
| 223 | rcu_read_lock(); |
| 224 | stream = find_stream(key, ht); |
| 225 | if (stream) { |
| 226 | stream->key = (uint64_t) -1ULL; |
| 227 | /* |
| 228 | * We don't want the lookup to match, but we still need |
| 229 | * to iterate on this stream when iterating over the hash table. Just |
| 230 | * change the node key. |
| 231 | */ |
| 232 | stream->node.key = (uint64_t) -1ULL; |
| 233 | } |
| 234 | rcu_read_unlock(); |
| 235 | } |
| 236 | |
| 237 | /* |
| 238 | * Return a channel object for the given key. |
| 239 | * |
| 240 | * RCU read side lock MUST be acquired before calling this function and |
| 241 | * protects the channel ptr. |
| 242 | */ |
| 243 | struct lttng_consumer_channel *consumer_find_channel(uint64_t key) |
| 244 | { |
| 245 | struct lttng_ht_iter iter; |
| 246 | struct lttng_ht_node_u64 *node; |
| 247 | struct lttng_consumer_channel *channel = NULL; |
| 248 | |
| 249 | /* -1ULL keys are lookup failures */ |
| 250 | if (key == (uint64_t) -1ULL) { |
| 251 | return NULL; |
| 252 | } |
| 253 | |
| 254 | lttng_ht_lookup(consumer_data.channel_ht, &key, &iter); |
| 255 | node = lttng_ht_iter_get_node_u64(&iter); |
| 256 | if (node != NULL) { |
| 257 | channel = caa_container_of(node, struct lttng_consumer_channel, node); |
| 258 | } |
| 259 | |
| 260 | return channel; |
| 261 | } |
| 262 | |
| 263 | /* |
| 264 | * There is a possibility that the consumer does not have enough time between |
| 265 | * the close of the channel on the session daemon and the cleanup in here thus |
| 266 | * once we have a channel add with an existing key, we know for sure that this |
| 267 | * channel will eventually get cleaned up by all streams being closed. |
| 268 | * |
| 269 | * This function just nullifies the already existing channel key. |
| 270 | */ |
| 271 | static void steal_channel_key(uint64_t key) |
| 272 | { |
| 273 | struct lttng_consumer_channel *channel; |
| 274 | |
| 275 | rcu_read_lock(); |
| 276 | channel = consumer_find_channel(key); |
| 277 | if (channel) { |
| 278 | channel->key = (uint64_t) -1ULL; |
| 279 | /* |
| 280 | * We don't want the lookup to match, but we still need to iterate on |
| 281 | * this channel when iterating over the hash table. Just change the |
| 282 | * node key. |
| 283 | */ |
| 284 | channel->node.key = (uint64_t) -1ULL; |
| 285 | } |
| 286 | rcu_read_unlock(); |
| 287 | } |
| 288 | |
| 289 | static void free_channel_rcu(struct rcu_head *head) |
| 290 | { |
| 291 | struct lttng_ht_node_u64 *node = |
| 292 | caa_container_of(head, struct lttng_ht_node_u64, head); |
| 293 | struct lttng_consumer_channel *channel = |
| 294 | caa_container_of(node, struct lttng_consumer_channel, node); |
| 295 | |
| 296 | switch (consumer_data.type) { |
| 297 | case LTTNG_CONSUMER_KERNEL: |
| 298 | break; |
| 299 | case LTTNG_CONSUMER32_UST: |
| 300 | case LTTNG_CONSUMER64_UST: |
| 301 | lttng_ustconsumer_free_channel(channel); |
| 302 | break; |
| 303 | default: |
| 304 | ERR("Unknown consumer_data type"); |
| 305 | abort(); |
| 306 | } |
| 307 | free(channel); |
| 308 | } |
| 309 | |
| 310 | /* |
| 311 | * RCU protected relayd socket pair free. |
| 312 | */ |
| 313 | static void free_relayd_rcu(struct rcu_head *head) |
| 314 | { |
| 315 | struct lttng_ht_node_u64 *node = |
| 316 | caa_container_of(head, struct lttng_ht_node_u64, head); |
| 317 | struct consumer_relayd_sock_pair *relayd = |
| 318 | caa_container_of(node, struct consumer_relayd_sock_pair, node); |
| 319 | |
| 320 | /* |
| 321 | * Close all sockets. This is done in the call RCU since we don't want the |
| 322 | * socket fds to be reassigned thus potentially creating bad state of the |
| 323 | * relayd object. |
| 324 | * |
| 325 | * We do not have to lock the control socket mutex here since at this stage |
| 326 | * there is no one referencing to this relayd object. |
| 327 | */ |
| 328 | (void) relayd_close(&relayd->control_sock); |
| 329 | (void) relayd_close(&relayd->data_sock); |
| 330 | |
| 331 | pthread_mutex_destroy(&relayd->ctrl_sock_mutex); |
| 332 | free(relayd); |
| 333 | } |
| 334 | |
| 335 | /* |
| 336 | * Destroy and free relayd socket pair object. |
| 337 | */ |
| 338 | void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd) |
| 339 | { |
| 340 | int ret; |
| 341 | struct lttng_ht_iter iter; |
| 342 | |
| 343 | if (relayd == NULL) { |
| 344 | return; |
| 345 | } |
| 346 | |
| 347 | DBG("Consumer destroy and close relayd socket pair"); |
| 348 | |
| 349 | iter.iter.node = &relayd->node.node; |
| 350 | ret = lttng_ht_del(consumer_data.relayd_ht, &iter); |
| 351 | if (ret != 0) { |
| 352 | /* We assume the relayd is being or is destroyed */ |
| 353 | return; |
| 354 | } |
| 355 | |
| 356 | /* RCU free() call */ |
| 357 | call_rcu(&relayd->node.head, free_relayd_rcu); |
| 358 | } |
| 359 | |
| 360 | /* |
| 361 | * Remove a channel from the global list protected by a mutex. This function is |
| 362 | * also responsible for freeing its data structures. |
| 363 | */ |
| 364 | void consumer_del_channel(struct lttng_consumer_channel *channel) |
| 365 | { |
| 366 | struct lttng_ht_iter iter; |
| 367 | |
| 368 | DBG("Consumer delete channel key %" PRIu64, channel->key); |
| 369 | |
| 370 | pthread_mutex_lock(&consumer_data.lock); |
| 371 | pthread_mutex_lock(&channel->lock); |
| 372 | |
| 373 | /* Destroy streams that might have been left in the stream list. */ |
| 374 | clean_channel_stream_list(channel); |
| 375 | |
| 376 | if (channel->live_timer_enabled == 1) { |
| 377 | consumer_timer_live_stop(channel); |
| 378 | } |
| 379 | if (channel->monitor_timer_enabled == 1) { |
| 380 | consumer_timer_monitor_stop(channel); |
| 381 | } |
| 382 | |
| 383 | switch (consumer_data.type) { |
| 384 | case LTTNG_CONSUMER_KERNEL: |
| 385 | break; |
| 386 | case LTTNG_CONSUMER32_UST: |
| 387 | case LTTNG_CONSUMER64_UST: |
| 388 | lttng_ustconsumer_del_channel(channel); |
| 389 | break; |
| 390 | default: |
| 391 | ERR("Unknown consumer_data type"); |
| 392 | assert(0); |
| 393 | goto end; |
| 394 | } |
| 395 | |
| 396 | lttng_trace_chunk_put(channel->trace_chunk); |
| 397 | channel->trace_chunk = NULL; |
| 398 | |
| 399 | if (channel->is_published) { |
| 400 | int ret; |
| 401 | |
| 402 | rcu_read_lock(); |
| 403 | iter.iter.node = &channel->node.node; |
| 404 | ret = lttng_ht_del(consumer_data.channel_ht, &iter); |
| 405 | assert(!ret); |
| 406 | |
| 407 | iter.iter.node = &channel->channels_by_session_id_ht_node.node; |
| 408 | ret = lttng_ht_del(consumer_data.channels_by_session_id_ht, |
| 409 | &iter); |
| 410 | assert(!ret); |
| 411 | rcu_read_unlock(); |
| 412 | } |
| 413 | |
| 414 | channel->is_deleted = true; |
| 415 | call_rcu(&channel->node.head, free_channel_rcu); |
| 416 | end: |
| 417 | pthread_mutex_unlock(&channel->lock); |
| 418 | pthread_mutex_unlock(&consumer_data.lock); |
| 419 | } |
| 420 | |
| 421 | /* |
| 422 | * Iterate over the relayd hash table and destroy each element. Finally, |
| 423 | * destroy the whole hash table. |
| 424 | */ |
| 425 | static void cleanup_relayd_ht(void) |
| 426 | { |
| 427 | struct lttng_ht_iter iter; |
| 428 | struct consumer_relayd_sock_pair *relayd; |
| 429 | |
| 430 | rcu_read_lock(); |
| 431 | |
| 432 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, |
| 433 | node.node) { |
| 434 | consumer_destroy_relayd(relayd); |
| 435 | } |
| 436 | |
| 437 | rcu_read_unlock(); |
| 438 | |
| 439 | lttng_ht_destroy(consumer_data.relayd_ht); |
| 440 | } |
| 441 | |
| 442 | /* |
| 443 | * Update the end point status of all streams having the given network sequence |
| 444 | * index (relayd index). |
| 445 | * |
| 446 | * It's atomically set without having the stream mutex locked which is fine |
| 447 | * because we handle the write/read race with a pipe wakeup for each thread. |
| 448 | */ |
| 449 | static void update_endpoint_status_by_netidx(uint64_t net_seq_idx, |
| 450 | enum consumer_endpoint_status status) |
| 451 | { |
| 452 | struct lttng_ht_iter iter; |
| 453 | struct lttng_consumer_stream *stream; |
| 454 | |
| 455 | DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx); |
| 456 | |
| 457 | rcu_read_lock(); |
| 458 | |
| 459 | /* Let's begin with metadata */ |
| 460 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { |
| 461 | if (stream->net_seq_idx == net_seq_idx) { |
| 462 | uatomic_set(&stream->endpoint_status, status); |
| 463 | DBG("Delete flag set to metadata stream %d", stream->wait_fd); |
| 464 | } |
| 465 | } |
| 466 | |
| 467 | /* Follow up by the data streams */ |
| 468 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { |
| 469 | if (stream->net_seq_idx == net_seq_idx) { |
| 470 | uatomic_set(&stream->endpoint_status, status); |
| 471 | DBG("Delete flag set to data stream %d", stream->wait_fd); |
| 472 | } |
| 473 | } |
| 474 | rcu_read_unlock(); |
| 475 | } |
| 476 | |
| 477 | /* |
| 478 | * Cleanup a relayd object by flagging every associated streams for deletion, |
| 479 | * destroying the object meaning removing it from the relayd hash table, |
| 480 | * closing the sockets and freeing the memory in a RCU call. |
| 481 | * |
| 482 | * If a local data context is available, notify the threads that the streams' |
| 483 | * state have changed. |
| 484 | */ |
| 485 | void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd) |
| 486 | { |
| 487 | uint64_t netidx; |
| 488 | |
| 489 | assert(relayd); |
| 490 | |
| 491 | DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx); |
| 492 | |
| 493 | /* Save the net sequence index before destroying the object */ |
| 494 | netidx = relayd->net_seq_idx; |
| 495 | |
| 496 | /* |
| 497 | * Delete the relayd from the relayd hash table, close the sockets and free |
| 498 | * the object in a RCU call. |
| 499 | */ |
| 500 | consumer_destroy_relayd(relayd); |
| 501 | |
| 502 | /* Set inactive endpoint to all streams */ |
| 503 | update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE); |
| 504 | |
| 505 | /* |
| 506 | * With a local data context, notify the threads that the streams' state |
| 507 | * have changed. The write() action on the pipe acts as an "implicit" |
| 508 | * memory barrier ordering the updates of the end point status from the |
| 509 | * read of this status which happens AFTER receiving this notify. |
| 510 | */ |
| 511 | notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe); |
| 512 | notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe); |
| 513 | } |
| 514 | |
| 515 | /* |
| 516 | * Flag a relayd socket pair for destruction. Destroy it if the refcount |
| 517 | * reaches zero. |
| 518 | * |
| 519 | * RCU read side lock MUST be aquired before calling this function. |
| 520 | */ |
| 521 | void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd) |
| 522 | { |
| 523 | assert(relayd); |
| 524 | |
| 525 | /* Set destroy flag for this object */ |
| 526 | uatomic_set(&relayd->destroy_flag, 1); |
| 527 | |
| 528 | /* Destroy the relayd if refcount is 0 */ |
| 529 | if (uatomic_read(&relayd->refcount) == 0) { |
| 530 | consumer_destroy_relayd(relayd); |
| 531 | } |
| 532 | } |
| 533 | |
| 534 | /* |
| 535 | * Completly destroy stream from every visiable data structure and the given |
| 536 | * hash table if one. |
| 537 | * |
| 538 | * One this call returns, the stream object is not longer usable nor visible. |
| 539 | */ |
| 540 | void consumer_del_stream(struct lttng_consumer_stream *stream, |
| 541 | struct lttng_ht *ht) |
| 542 | { |
| 543 | consumer_stream_destroy(stream, ht); |
| 544 | } |
| 545 | |
| 546 | /* |
| 547 | * XXX naming of del vs destroy is all mixed up. |
| 548 | */ |
| 549 | void consumer_del_stream_for_data(struct lttng_consumer_stream *stream) |
| 550 | { |
| 551 | consumer_stream_destroy(stream, data_ht); |
| 552 | } |
| 553 | |
| 554 | void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream) |
| 555 | { |
| 556 | consumer_stream_destroy(stream, metadata_ht); |
| 557 | } |
| 558 | |
| 559 | void consumer_stream_update_channel_attributes( |
| 560 | struct lttng_consumer_stream *stream, |
| 561 | struct lttng_consumer_channel *channel) |
| 562 | { |
| 563 | stream->channel_read_only_attributes.tracefile_size = |
| 564 | channel->tracefile_size; |
| 565 | } |
| 566 | |
| 567 | struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key, |
| 568 | uint64_t stream_key, |
| 569 | const char *channel_name, |
| 570 | uint64_t relayd_id, |
| 571 | uint64_t session_id, |
| 572 | struct lttng_trace_chunk *trace_chunk, |
| 573 | int cpu, |
| 574 | int *alloc_ret, |
| 575 | enum consumer_channel_type type, |
| 576 | unsigned int monitor) |
| 577 | { |
| 578 | int ret; |
| 579 | struct lttng_consumer_stream *stream; |
| 580 | |
| 581 | stream = zmalloc(sizeof(*stream)); |
| 582 | if (stream == NULL) { |
| 583 | PERROR("malloc struct lttng_consumer_stream"); |
| 584 | ret = -ENOMEM; |
| 585 | goto end; |
| 586 | } |
| 587 | |
| 588 | if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) { |
| 589 | ERR("Failed to acquire trace chunk reference during the creation of a stream"); |
| 590 | ret = -1; |
| 591 | goto error; |
| 592 | } |
| 593 | |
| 594 | rcu_read_lock(); |
| 595 | stream->key = stream_key; |
| 596 | stream->trace_chunk = trace_chunk; |
| 597 | stream->out_fd = -1; |
| 598 | stream->out_fd_offset = 0; |
| 599 | stream->output_written = 0; |
| 600 | stream->net_seq_idx = relayd_id; |
| 601 | stream->session_id = session_id; |
| 602 | stream->monitor = monitor; |
| 603 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
| 604 | stream->index_file = NULL; |
| 605 | stream->last_sequence_number = -1ULL; |
| 606 | pthread_mutex_init(&stream->lock, NULL); |
| 607 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
| 608 | |
| 609 | /* If channel is the metadata, flag this stream as metadata. */ |
| 610 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { |
| 611 | stream->metadata_flag = 1; |
| 612 | /* Metadata is flat out. */ |
| 613 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); |
| 614 | /* Live rendez-vous point. */ |
| 615 | pthread_cond_init(&stream->metadata_rdv, NULL); |
| 616 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); |
| 617 | } else { |
| 618 | /* Format stream name to <channel_name>_<cpu_number> */ |
| 619 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", |
| 620 | channel_name, cpu); |
| 621 | if (ret < 0) { |
| 622 | PERROR("snprintf stream name"); |
| 623 | goto error; |
| 624 | } |
| 625 | } |
| 626 | |
| 627 | /* Key is always the wait_fd for streams. */ |
| 628 | lttng_ht_node_init_u64(&stream->node, stream->key); |
| 629 | |
| 630 | /* Init node per channel id key */ |
| 631 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); |
| 632 | |
| 633 | /* Init session id node with the stream session id */ |
| 634 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
| 635 | |
| 636 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
| 637 | " relayd_id %" PRIu64 ", session_id %" PRIu64, |
| 638 | stream->name, stream->key, channel_key, |
| 639 | stream->net_seq_idx, stream->session_id); |
| 640 | |
| 641 | rcu_read_unlock(); |
| 642 | return stream; |
| 643 | |
| 644 | error: |
| 645 | rcu_read_unlock(); |
| 646 | lttng_trace_chunk_put(stream->trace_chunk); |
| 647 | free(stream); |
| 648 | end: |
| 649 | if (alloc_ret) { |
| 650 | *alloc_ret = ret; |
| 651 | } |
| 652 | return NULL; |
| 653 | } |
| 654 | |
| 655 | /* |
| 656 | * Add a stream to the global list protected by a mutex. |
| 657 | */ |
| 658 | void consumer_add_data_stream(struct lttng_consumer_stream *stream) |
| 659 | { |
| 660 | struct lttng_ht *ht = data_ht; |
| 661 | |
| 662 | assert(stream); |
| 663 | assert(ht); |
| 664 | |
| 665 | DBG3("Adding consumer stream %" PRIu64, stream->key); |
| 666 | |
| 667 | pthread_mutex_lock(&consumer_data.lock); |
| 668 | pthread_mutex_lock(&stream->chan->lock); |
| 669 | pthread_mutex_lock(&stream->chan->timer_lock); |
| 670 | pthread_mutex_lock(&stream->lock); |
| 671 | rcu_read_lock(); |
| 672 | |
| 673 | /* Steal stream identifier to avoid having streams with the same key */ |
| 674 | steal_stream_key(stream->key, ht); |
| 675 | |
| 676 | lttng_ht_add_unique_u64(ht, &stream->node); |
| 677 | |
| 678 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
| 679 | &stream->node_channel_id); |
| 680 | |
| 681 | /* |
| 682 | * Add stream to the stream_list_ht of the consumer data. No need to steal |
| 683 | * the key since the HT does not use it and we allow to add redundant keys |
| 684 | * into this table. |
| 685 | */ |
| 686 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
| 687 | |
| 688 | /* |
| 689 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
| 690 | * in terms of destroying the associated channel, because the action that |
| 691 | * causes the count to become 0 also causes a stream to be added. The |
| 692 | * channel deletion will thus be triggered by the following removal of this |
| 693 | * stream. |
| 694 | */ |
| 695 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
| 696 | /* Increment refcount before decrementing nb_init_stream_left */ |
| 697 | cmm_smp_wmb(); |
| 698 | uatomic_dec(&stream->chan->nb_init_stream_left); |
| 699 | } |
| 700 | |
| 701 | /* Update consumer data once the node is inserted. */ |
| 702 | consumer_data.stream_count++; |
| 703 | consumer_data.need_update = 1; |
| 704 | |
| 705 | rcu_read_unlock(); |
| 706 | pthread_mutex_unlock(&stream->lock); |
| 707 | pthread_mutex_unlock(&stream->chan->timer_lock); |
| 708 | pthread_mutex_unlock(&stream->chan->lock); |
| 709 | pthread_mutex_unlock(&consumer_data.lock); |
| 710 | } |
| 711 | |
| 712 | void consumer_del_data_stream(struct lttng_consumer_stream *stream) |
| 713 | { |
| 714 | consumer_del_stream(stream, data_ht); |
| 715 | } |
| 716 | |
| 717 | /* |
| 718 | * Add relayd socket to global consumer data hashtable. RCU read side lock MUST |
| 719 | * be acquired before calling this. |
| 720 | */ |
| 721 | static int add_relayd(struct consumer_relayd_sock_pair *relayd) |
| 722 | { |
| 723 | int ret = 0; |
| 724 | struct lttng_ht_node_u64 *node; |
| 725 | struct lttng_ht_iter iter; |
| 726 | |
| 727 | assert(relayd); |
| 728 | |
| 729 | lttng_ht_lookup(consumer_data.relayd_ht, |
| 730 | &relayd->net_seq_idx, &iter); |
| 731 | node = lttng_ht_iter_get_node_u64(&iter); |
| 732 | if (node != NULL) { |
| 733 | goto end; |
| 734 | } |
| 735 | lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node); |
| 736 | |
| 737 | end: |
| 738 | return ret; |
| 739 | } |
| 740 | |
| 741 | /* |
| 742 | * Allocate and return a consumer relayd socket. |
| 743 | */ |
| 744 | static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair( |
| 745 | uint64_t net_seq_idx) |
| 746 | { |
| 747 | struct consumer_relayd_sock_pair *obj = NULL; |
| 748 | |
| 749 | /* net sequence index of -1 is a failure */ |
| 750 | if (net_seq_idx == (uint64_t) -1ULL) { |
| 751 | goto error; |
| 752 | } |
| 753 | |
| 754 | obj = zmalloc(sizeof(struct consumer_relayd_sock_pair)); |
| 755 | if (obj == NULL) { |
| 756 | PERROR("zmalloc relayd sock"); |
| 757 | goto error; |
| 758 | } |
| 759 | |
| 760 | obj->net_seq_idx = net_seq_idx; |
| 761 | obj->refcount = 0; |
| 762 | obj->destroy_flag = 0; |
| 763 | obj->control_sock.sock.fd = -1; |
| 764 | obj->data_sock.sock.fd = -1; |
| 765 | lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx); |
| 766 | pthread_mutex_init(&obj->ctrl_sock_mutex, NULL); |
| 767 | |
| 768 | error: |
| 769 | return obj; |
| 770 | } |
| 771 | |
| 772 | /* |
| 773 | * Find a relayd socket pair in the global consumer data. |
| 774 | * |
| 775 | * Return the object if found else NULL. |
| 776 | * RCU read-side lock must be held across this call and while using the |
| 777 | * returned object. |
| 778 | */ |
| 779 | struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key) |
| 780 | { |
| 781 | struct lttng_ht_iter iter; |
| 782 | struct lttng_ht_node_u64 *node; |
| 783 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 784 | |
| 785 | /* Negative keys are lookup failures */ |
| 786 | if (key == (uint64_t) -1ULL) { |
| 787 | goto error; |
| 788 | } |
| 789 | |
| 790 | lttng_ht_lookup(consumer_data.relayd_ht, &key, |
| 791 | &iter); |
| 792 | node = lttng_ht_iter_get_node_u64(&iter); |
| 793 | if (node != NULL) { |
| 794 | relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node); |
| 795 | } |
| 796 | |
| 797 | error: |
| 798 | return relayd; |
| 799 | } |
| 800 | |
| 801 | /* |
| 802 | * Find a relayd and send the stream |
| 803 | * |
| 804 | * Returns 0 on success, < 0 on error |
| 805 | */ |
| 806 | int consumer_send_relayd_stream(struct lttng_consumer_stream *stream, |
| 807 | char *path) |
| 808 | { |
| 809 | int ret = 0; |
| 810 | struct consumer_relayd_sock_pair *relayd; |
| 811 | |
| 812 | assert(stream); |
| 813 | assert(stream->net_seq_idx != -1ULL); |
| 814 | assert(path); |
| 815 | |
| 816 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 817 | rcu_read_lock(); |
| 818 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 819 | if (relayd != NULL) { |
| 820 | /* Add stream on the relayd */ |
| 821 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 822 | ret = relayd_add_stream(&relayd->control_sock, stream->name, |
| 823 | path, &stream->relayd_stream_id, |
| 824 | stream->chan->tracefile_size, |
| 825 | stream->chan->tracefile_count, |
| 826 | stream->trace_chunk); |
| 827 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 828 | if (ret < 0) { |
| 829 | ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 830 | lttng_consumer_cleanup_relayd(relayd); |
| 831 | goto end; |
| 832 | } |
| 833 | |
| 834 | uatomic_inc(&relayd->refcount); |
| 835 | stream->sent_to_relayd = 1; |
| 836 | } else { |
| 837 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.", |
| 838 | stream->key, stream->net_seq_idx); |
| 839 | ret = -1; |
| 840 | goto end; |
| 841 | } |
| 842 | |
| 843 | DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64, |
| 844 | stream->name, stream->key, stream->net_seq_idx); |
| 845 | |
| 846 | end: |
| 847 | rcu_read_unlock(); |
| 848 | return ret; |
| 849 | } |
| 850 | |
| 851 | /* |
| 852 | * Find a relayd and send the streams sent message |
| 853 | * |
| 854 | * Returns 0 on success, < 0 on error |
| 855 | */ |
| 856 | int consumer_send_relayd_streams_sent(uint64_t net_seq_idx) |
| 857 | { |
| 858 | int ret = 0; |
| 859 | struct consumer_relayd_sock_pair *relayd; |
| 860 | |
| 861 | assert(net_seq_idx != -1ULL); |
| 862 | |
| 863 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 864 | rcu_read_lock(); |
| 865 | relayd = consumer_find_relayd(net_seq_idx); |
| 866 | if (relayd != NULL) { |
| 867 | /* Add stream on the relayd */ |
| 868 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 869 | ret = relayd_streams_sent(&relayd->control_sock); |
| 870 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 871 | if (ret < 0) { |
| 872 | ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 873 | lttng_consumer_cleanup_relayd(relayd); |
| 874 | goto end; |
| 875 | } |
| 876 | } else { |
| 877 | ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.", |
| 878 | net_seq_idx); |
| 879 | ret = -1; |
| 880 | goto end; |
| 881 | } |
| 882 | |
| 883 | ret = 0; |
| 884 | DBG("All streams sent relayd id %" PRIu64, net_seq_idx); |
| 885 | |
| 886 | end: |
| 887 | rcu_read_unlock(); |
| 888 | return ret; |
| 889 | } |
| 890 | |
| 891 | /* |
| 892 | * Find a relayd and close the stream |
| 893 | */ |
| 894 | void close_relayd_stream(struct lttng_consumer_stream *stream) |
| 895 | { |
| 896 | struct consumer_relayd_sock_pair *relayd; |
| 897 | |
| 898 | /* The stream is not metadata. Get relayd reference if exists. */ |
| 899 | rcu_read_lock(); |
| 900 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 901 | if (relayd) { |
| 902 | consumer_stream_relayd_close(stream, relayd); |
| 903 | } |
| 904 | rcu_read_unlock(); |
| 905 | } |
| 906 | |
| 907 | /* |
| 908 | * Handle stream for relayd transmission if the stream applies for network |
| 909 | * streaming where the net sequence index is set. |
| 910 | * |
| 911 | * Return destination file descriptor or negative value on error. |
| 912 | */ |
| 913 | static int write_relayd_stream_header(struct lttng_consumer_stream *stream, |
| 914 | size_t data_size, unsigned long padding, |
| 915 | struct consumer_relayd_sock_pair *relayd) |
| 916 | { |
| 917 | int outfd = -1, ret; |
| 918 | struct lttcomm_relayd_data_hdr data_hdr; |
| 919 | |
| 920 | /* Safety net */ |
| 921 | assert(stream); |
| 922 | assert(relayd); |
| 923 | |
| 924 | /* Reset data header */ |
| 925 | memset(&data_hdr, 0, sizeof(data_hdr)); |
| 926 | |
| 927 | if (stream->metadata_flag) { |
| 928 | /* Caller MUST acquire the relayd control socket lock */ |
| 929 | ret = relayd_send_metadata(&relayd->control_sock, data_size); |
| 930 | if (ret < 0) { |
| 931 | goto error; |
| 932 | } |
| 933 | |
| 934 | /* Metadata are always sent on the control socket. */ |
| 935 | outfd = relayd->control_sock.sock.fd; |
| 936 | } else { |
| 937 | /* Set header with stream information */ |
| 938 | data_hdr.stream_id = htobe64(stream->relayd_stream_id); |
| 939 | data_hdr.data_size = htobe32(data_size); |
| 940 | data_hdr.padding_size = htobe32(padding); |
| 941 | |
| 942 | /* |
| 943 | * Note that net_seq_num below is assigned with the *current* value of |
| 944 | * next_net_seq_num and only after that the next_net_seq_num will be |
| 945 | * increment. This is why when issuing a command on the relayd using |
| 946 | * this next value, 1 should always be substracted in order to compare |
| 947 | * the last seen sequence number on the relayd side to the last sent. |
| 948 | */ |
| 949 | data_hdr.net_seq_num = htobe64(stream->next_net_seq_num); |
| 950 | /* Other fields are zeroed previously */ |
| 951 | |
| 952 | ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr, |
| 953 | sizeof(data_hdr)); |
| 954 | if (ret < 0) { |
| 955 | goto error; |
| 956 | } |
| 957 | |
| 958 | ++stream->next_net_seq_num; |
| 959 | |
| 960 | /* Set to go on data socket */ |
| 961 | outfd = relayd->data_sock.sock.fd; |
| 962 | } |
| 963 | |
| 964 | error: |
| 965 | return outfd; |
| 966 | } |
| 967 | |
| 968 | /* |
| 969 | * Trigger a dump of the metadata content. Following/during the succesful |
| 970 | * completion of this call, the metadata poll thread will start receiving |
| 971 | * metadata packets to consume. |
| 972 | * |
| 973 | * The caller must hold the channel and stream locks. |
| 974 | */ |
| 975 | static |
| 976 | int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream) |
| 977 | { |
| 978 | int ret; |
| 979 | |
| 980 | ASSERT_LOCKED(stream->chan->lock); |
| 981 | ASSERT_LOCKED(stream->lock); |
| 982 | assert(stream->metadata_flag); |
| 983 | assert(stream->chan->trace_chunk); |
| 984 | |
| 985 | switch (consumer_data.type) { |
| 986 | case LTTNG_CONSUMER_KERNEL: |
| 987 | /* |
| 988 | * Reset the position of what has been read from the |
| 989 | * metadata cache to 0 so we can dump it again. |
| 990 | */ |
| 991 | ret = kernctl_metadata_cache_dump(stream->wait_fd); |
| 992 | break; |
| 993 | case LTTNG_CONSUMER32_UST: |
| 994 | case LTTNG_CONSUMER64_UST: |
| 995 | /* |
| 996 | * Reset the position pushed from the metadata cache so it |
| 997 | * will write from the beginning on the next push. |
| 998 | */ |
| 999 | stream->ust_metadata_pushed = 0; |
| 1000 | ret = consumer_metadata_wakeup_pipe(stream->chan); |
| 1001 | break; |
| 1002 | default: |
| 1003 | ERR("Unknown consumer_data type"); |
| 1004 | abort(); |
| 1005 | } |
| 1006 | if (ret < 0) { |
| 1007 | ERR("Failed to dump the metadata cache"); |
| 1008 | } |
| 1009 | return ret; |
| 1010 | } |
| 1011 | |
| 1012 | static |
| 1013 | int lttng_consumer_channel_set_trace_chunk( |
| 1014 | struct lttng_consumer_channel *channel, |
| 1015 | struct lttng_trace_chunk *new_trace_chunk) |
| 1016 | { |
| 1017 | int ret = 0; |
| 1018 | const bool is_local_trace = channel->relayd_id == -1ULL; |
| 1019 | bool update_stream_trace_chunk; |
| 1020 | struct cds_lfht_iter iter; |
| 1021 | struct lttng_consumer_stream *stream; |
| 1022 | unsigned long channel_hash; |
| 1023 | |
| 1024 | pthread_mutex_lock(&channel->lock); |
| 1025 | if (channel->is_deleted) { |
| 1026 | /* |
| 1027 | * The channel has been logically deleted and should no longer |
| 1028 | * be used. It has released its reference to its current trace |
| 1029 | * chunk and should not acquire a new one. |
| 1030 | * |
| 1031 | * Return success as there is nothing for the caller to do. |
| 1032 | */ |
| 1033 | goto end; |
| 1034 | } |
| 1035 | /* |
| 1036 | * A stream can transition to a state where it and its channel |
| 1037 | * no longer belong to a trace chunk. For instance, this happens when |
| 1038 | * a session is rotated while it is inactive. After the rotation |
| 1039 | * of an inactive session completes, the channel and its streams no |
| 1040 | * longer belong to a trace chunk. |
| 1041 | * |
| 1042 | * However, if a session is stopped, rotated, and started again, |
| 1043 | * the session daemon will create a new chunk and send it to its peers. |
| 1044 | * In that case, the streams' transition to a new chunk can be performed |
| 1045 | * immediately. |
| 1046 | * |
| 1047 | * This trace chunk transition could also be performed lazily when |
| 1048 | * a buffer is consumed. However, creating the files here allows the |
| 1049 | * consumer daemon to report any creation error to the session daemon |
| 1050 | * and cause the start of the tracing session to fail. |
| 1051 | */ |
| 1052 | update_stream_trace_chunk = !channel->trace_chunk && new_trace_chunk; |
| 1053 | |
| 1054 | /* |
| 1055 | * The acquisition of the reference cannot fail (barring |
| 1056 | * a severe internal error) since a reference to the published |
| 1057 | * chunk is already held by the caller. |
| 1058 | */ |
| 1059 | if (new_trace_chunk) { |
| 1060 | const bool acquired_reference = lttng_trace_chunk_get( |
| 1061 | new_trace_chunk); |
| 1062 | |
| 1063 | assert(acquired_reference); |
| 1064 | } |
| 1065 | |
| 1066 | lttng_trace_chunk_put(channel->trace_chunk); |
| 1067 | channel->trace_chunk = new_trace_chunk; |
| 1068 | if (!is_local_trace || !new_trace_chunk) { |
| 1069 | /* Not an error. */ |
| 1070 | goto end; |
| 1071 | } |
| 1072 | |
| 1073 | if (!update_stream_trace_chunk) { |
| 1074 | goto end; |
| 1075 | } |
| 1076 | |
| 1077 | channel_hash = consumer_data.stream_per_chan_id_ht->hash_fct( |
| 1078 | &channel->key, lttng_ht_seed); |
| 1079 | rcu_read_lock(); |
| 1080 | cds_lfht_for_each_entry_duplicate(consumer_data.stream_per_chan_id_ht->ht, |
| 1081 | channel_hash, |
| 1082 | consumer_data.stream_per_chan_id_ht->match_fct, |
| 1083 | &channel->key, &iter, stream, node_channel_id.node) { |
| 1084 | bool acquired_reference, should_regenerate_metadata = false; |
| 1085 | |
| 1086 | acquired_reference = lttng_trace_chunk_get(channel->trace_chunk); |
| 1087 | assert(acquired_reference); |
| 1088 | |
| 1089 | pthread_mutex_lock(&stream->lock); |
| 1090 | |
| 1091 | /* |
| 1092 | * On a transition from "no-chunk" to a new chunk, a metadata |
| 1093 | * stream's content must be entirely dumped. This must occcur |
| 1094 | * _after_ the creation of the metadata stream's output files |
| 1095 | * as the consumption thread (not necessarily the one executing |
| 1096 | * this) may start to consume during the call to |
| 1097 | * consumer_metadata_stream_dump(). |
| 1098 | */ |
| 1099 | should_regenerate_metadata = |
| 1100 | stream->metadata_flag && |
| 1101 | !stream->trace_chunk && channel->trace_chunk; |
| 1102 | stream->trace_chunk = channel->trace_chunk; |
| 1103 | ret = consumer_stream_create_output_files(stream, true); |
| 1104 | if (ret) { |
| 1105 | pthread_mutex_unlock(&stream->lock); |
| 1106 | goto end_rcu_unlock; |
| 1107 | } |
| 1108 | if (should_regenerate_metadata) { |
| 1109 | ret = consumer_metadata_stream_dump(stream); |
| 1110 | } |
| 1111 | pthread_mutex_unlock(&stream->lock); |
| 1112 | if (ret) { |
| 1113 | goto end_rcu_unlock; |
| 1114 | } |
| 1115 | } |
| 1116 | end_rcu_unlock: |
| 1117 | rcu_read_unlock(); |
| 1118 | end: |
| 1119 | pthread_mutex_unlock(&channel->lock); |
| 1120 | return ret; |
| 1121 | } |
| 1122 | |
| 1123 | /* |
| 1124 | * Allocate and return a new lttng_consumer_channel object using the given key |
| 1125 | * to initialize the hash table node. |
| 1126 | * |
| 1127 | * On error, return NULL. |
| 1128 | */ |
| 1129 | struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key, |
| 1130 | uint64_t session_id, |
| 1131 | const uint64_t *chunk_id, |
| 1132 | const char *pathname, |
| 1133 | const char *name, |
| 1134 | uint64_t relayd_id, |
| 1135 | enum lttng_event_output output, |
| 1136 | uint64_t tracefile_size, |
| 1137 | uint64_t tracefile_count, |
| 1138 | uint64_t session_id_per_pid, |
| 1139 | unsigned int monitor, |
| 1140 | unsigned int live_timer_interval, |
| 1141 | const char *root_shm_path, |
| 1142 | const char *shm_path) |
| 1143 | { |
| 1144 | struct lttng_consumer_channel *channel = NULL; |
| 1145 | struct lttng_trace_chunk *trace_chunk = NULL; |
| 1146 | |
| 1147 | if (chunk_id) { |
| 1148 | trace_chunk = lttng_trace_chunk_registry_find_chunk( |
| 1149 | consumer_data.chunk_registry, session_id, |
| 1150 | *chunk_id); |
| 1151 | if (!trace_chunk) { |
| 1152 | ERR("Failed to find trace chunk reference during creation of channel"); |
| 1153 | goto end; |
| 1154 | } |
| 1155 | } |
| 1156 | |
| 1157 | channel = zmalloc(sizeof(*channel)); |
| 1158 | if (channel == NULL) { |
| 1159 | PERROR("malloc struct lttng_consumer_channel"); |
| 1160 | goto end; |
| 1161 | } |
| 1162 | |
| 1163 | channel->key = key; |
| 1164 | channel->refcount = 0; |
| 1165 | channel->session_id = session_id; |
| 1166 | channel->session_id_per_pid = session_id_per_pid; |
| 1167 | channel->relayd_id = relayd_id; |
| 1168 | channel->tracefile_size = tracefile_size; |
| 1169 | channel->tracefile_count = tracefile_count; |
| 1170 | channel->monitor = monitor; |
| 1171 | channel->live_timer_interval = live_timer_interval; |
| 1172 | pthread_mutex_init(&channel->lock, NULL); |
| 1173 | pthread_mutex_init(&channel->timer_lock, NULL); |
| 1174 | |
| 1175 | switch (output) { |
| 1176 | case LTTNG_EVENT_SPLICE: |
| 1177 | channel->output = CONSUMER_CHANNEL_SPLICE; |
| 1178 | break; |
| 1179 | case LTTNG_EVENT_MMAP: |
| 1180 | channel->output = CONSUMER_CHANNEL_MMAP; |
| 1181 | break; |
| 1182 | default: |
| 1183 | assert(0); |
| 1184 | free(channel); |
| 1185 | channel = NULL; |
| 1186 | goto end; |
| 1187 | } |
| 1188 | |
| 1189 | /* |
| 1190 | * In monitor mode, the streams associated with the channel will be put in |
| 1191 | * a special list ONLY owned by this channel. So, the refcount is set to 1 |
| 1192 | * here meaning that the channel itself has streams that are referenced. |
| 1193 | * |
| 1194 | * On a channel deletion, once the channel is no longer visible, the |
| 1195 | * refcount is decremented and checked for a zero value to delete it. With |
| 1196 | * streams in no monitor mode, it will now be safe to destroy the channel. |
| 1197 | */ |
| 1198 | if (!channel->monitor) { |
| 1199 | channel->refcount = 1; |
| 1200 | } |
| 1201 | |
| 1202 | strncpy(channel->pathname, pathname, sizeof(channel->pathname)); |
| 1203 | channel->pathname[sizeof(channel->pathname) - 1] = '\0'; |
| 1204 | |
| 1205 | strncpy(channel->name, name, sizeof(channel->name)); |
| 1206 | channel->name[sizeof(channel->name) - 1] = '\0'; |
| 1207 | |
| 1208 | if (root_shm_path) { |
| 1209 | strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path)); |
| 1210 | channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0'; |
| 1211 | } |
| 1212 | if (shm_path) { |
| 1213 | strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path)); |
| 1214 | channel->shm_path[sizeof(channel->shm_path) - 1] = '\0'; |
| 1215 | } |
| 1216 | |
| 1217 | lttng_ht_node_init_u64(&channel->node, channel->key); |
| 1218 | lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node, |
| 1219 | channel->session_id); |
| 1220 | |
| 1221 | channel->wait_fd = -1; |
| 1222 | CDS_INIT_LIST_HEAD(&channel->streams.head); |
| 1223 | |
| 1224 | if (trace_chunk) { |
| 1225 | int ret = lttng_consumer_channel_set_trace_chunk(channel, |
| 1226 | trace_chunk); |
| 1227 | if (ret) { |
| 1228 | goto error; |
| 1229 | } |
| 1230 | } |
| 1231 | |
| 1232 | DBG("Allocated channel (key %" PRIu64 ")", channel->key); |
| 1233 | |
| 1234 | end: |
| 1235 | lttng_trace_chunk_put(trace_chunk); |
| 1236 | return channel; |
| 1237 | error: |
| 1238 | consumer_del_channel(channel); |
| 1239 | channel = NULL; |
| 1240 | goto end; |
| 1241 | } |
| 1242 | |
| 1243 | /* |
| 1244 | * Add a channel to the global list protected by a mutex. |
| 1245 | * |
| 1246 | * Always return 0 indicating success. |
| 1247 | */ |
| 1248 | int consumer_add_channel(struct lttng_consumer_channel *channel, |
| 1249 | struct lttng_consumer_local_data *ctx) |
| 1250 | { |
| 1251 | pthread_mutex_lock(&consumer_data.lock); |
| 1252 | pthread_mutex_lock(&channel->lock); |
| 1253 | pthread_mutex_lock(&channel->timer_lock); |
| 1254 | |
| 1255 | /* |
| 1256 | * This gives us a guarantee that the channel we are about to add to the |
| 1257 | * channel hash table will be unique. See this function comment on the why |
| 1258 | * we need to steel the channel key at this stage. |
| 1259 | */ |
| 1260 | steal_channel_key(channel->key); |
| 1261 | |
| 1262 | rcu_read_lock(); |
| 1263 | lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node); |
| 1264 | lttng_ht_add_u64(consumer_data.channels_by_session_id_ht, |
| 1265 | &channel->channels_by_session_id_ht_node); |
| 1266 | rcu_read_unlock(); |
| 1267 | channel->is_published = true; |
| 1268 | |
| 1269 | pthread_mutex_unlock(&channel->timer_lock); |
| 1270 | pthread_mutex_unlock(&channel->lock); |
| 1271 | pthread_mutex_unlock(&consumer_data.lock); |
| 1272 | |
| 1273 | if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) { |
| 1274 | notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD); |
| 1275 | } |
| 1276 | |
| 1277 | return 0; |
| 1278 | } |
| 1279 | |
| 1280 | /* |
| 1281 | * Allocate the pollfd structure and the local view of the out fds to avoid |
| 1282 | * doing a lookup in the linked list and concurrency issues when writing is |
| 1283 | * needed. Called with consumer_data.lock held. |
| 1284 | * |
| 1285 | * Returns the number of fds in the structures. |
| 1286 | */ |
| 1287 | static int update_poll_array(struct lttng_consumer_local_data *ctx, |
| 1288 | struct pollfd **pollfd, struct lttng_consumer_stream **local_stream, |
| 1289 | struct lttng_ht *ht, int *nb_inactive_fd) |
| 1290 | { |
| 1291 | int i = 0; |
| 1292 | struct lttng_ht_iter iter; |
| 1293 | struct lttng_consumer_stream *stream; |
| 1294 | |
| 1295 | assert(ctx); |
| 1296 | assert(ht); |
| 1297 | assert(pollfd); |
| 1298 | assert(local_stream); |
| 1299 | |
| 1300 | DBG("Updating poll fd array"); |
| 1301 | *nb_inactive_fd = 0; |
| 1302 | rcu_read_lock(); |
| 1303 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
| 1304 | /* |
| 1305 | * Only active streams with an active end point can be added to the |
| 1306 | * poll set and local stream storage of the thread. |
| 1307 | * |
| 1308 | * There is a potential race here for endpoint_status to be updated |
| 1309 | * just after the check. However, this is OK since the stream(s) will |
| 1310 | * be deleted once the thread is notified that the end point state has |
| 1311 | * changed where this function will be called back again. |
| 1312 | * |
| 1313 | * We track the number of inactive FDs because they still need to be |
| 1314 | * closed by the polling thread after a wakeup on the data_pipe or |
| 1315 | * metadata_pipe. |
| 1316 | */ |
| 1317 | if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) { |
| 1318 | (*nb_inactive_fd)++; |
| 1319 | continue; |
| 1320 | } |
| 1321 | /* |
| 1322 | * This clobbers way too much the debug output. Uncomment that if you |
| 1323 | * need it for debugging purposes. |
| 1324 | */ |
| 1325 | (*pollfd)[i].fd = stream->wait_fd; |
| 1326 | (*pollfd)[i].events = POLLIN | POLLPRI; |
| 1327 | local_stream[i] = stream; |
| 1328 | i++; |
| 1329 | } |
| 1330 | rcu_read_unlock(); |
| 1331 | |
| 1332 | /* |
| 1333 | * Insert the consumer_data_pipe at the end of the array and don't |
| 1334 | * increment i so nb_fd is the number of real FD. |
| 1335 | */ |
| 1336 | (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe); |
| 1337 | (*pollfd)[i].events = POLLIN | POLLPRI; |
| 1338 | |
| 1339 | (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe); |
| 1340 | (*pollfd)[i + 1].events = POLLIN | POLLPRI; |
| 1341 | return i; |
| 1342 | } |
| 1343 | |
| 1344 | /* |
| 1345 | * Poll on the should_quit pipe and the command socket return -1 on |
| 1346 | * error, 1 if should exit, 0 if data is available on the command socket |
| 1347 | */ |
| 1348 | int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll) |
| 1349 | { |
| 1350 | int num_rdy; |
| 1351 | |
| 1352 | restart: |
| 1353 | num_rdy = poll(consumer_sockpoll, 2, -1); |
| 1354 | if (num_rdy == -1) { |
| 1355 | /* |
| 1356 | * Restart interrupted system call. |
| 1357 | */ |
| 1358 | if (errno == EINTR) { |
| 1359 | goto restart; |
| 1360 | } |
| 1361 | PERROR("Poll error"); |
| 1362 | return -1; |
| 1363 | } |
| 1364 | if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) { |
| 1365 | DBG("consumer_should_quit wake up"); |
| 1366 | return 1; |
| 1367 | } |
| 1368 | return 0; |
| 1369 | } |
| 1370 | |
| 1371 | /* |
| 1372 | * Set the error socket. |
| 1373 | */ |
| 1374 | void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx, |
| 1375 | int sock) |
| 1376 | { |
| 1377 | ctx->consumer_error_socket = sock; |
| 1378 | } |
| 1379 | |
| 1380 | /* |
| 1381 | * Set the command socket path. |
| 1382 | */ |
| 1383 | void lttng_consumer_set_command_sock_path( |
| 1384 | struct lttng_consumer_local_data *ctx, char *sock) |
| 1385 | { |
| 1386 | ctx->consumer_command_sock_path = sock; |
| 1387 | } |
| 1388 | |
| 1389 | /* |
| 1390 | * Send return code to the session daemon. |
| 1391 | * If the socket is not defined, we return 0, it is not a fatal error |
| 1392 | */ |
| 1393 | int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd) |
| 1394 | { |
| 1395 | if (ctx->consumer_error_socket > 0) { |
| 1396 | return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd, |
| 1397 | sizeof(enum lttcomm_sessiond_command)); |
| 1398 | } |
| 1399 | |
| 1400 | return 0; |
| 1401 | } |
| 1402 | |
| 1403 | /* |
| 1404 | * Close all the tracefiles and stream fds and MUST be called when all |
| 1405 | * instances are destroyed i.e. when all threads were joined and are ended. |
| 1406 | */ |
| 1407 | void lttng_consumer_cleanup(void) |
| 1408 | { |
| 1409 | struct lttng_ht_iter iter; |
| 1410 | struct lttng_consumer_channel *channel; |
| 1411 | unsigned int trace_chunks_left; |
| 1412 | |
| 1413 | rcu_read_lock(); |
| 1414 | |
| 1415 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel, |
| 1416 | node.node) { |
| 1417 | consumer_del_channel(channel); |
| 1418 | } |
| 1419 | |
| 1420 | rcu_read_unlock(); |
| 1421 | |
| 1422 | lttng_ht_destroy(consumer_data.channel_ht); |
| 1423 | lttng_ht_destroy(consumer_data.channels_by_session_id_ht); |
| 1424 | |
| 1425 | cleanup_relayd_ht(); |
| 1426 | |
| 1427 | lttng_ht_destroy(consumer_data.stream_per_chan_id_ht); |
| 1428 | |
| 1429 | /* |
| 1430 | * This HT contains streams that are freed by either the metadata thread or |
| 1431 | * the data thread so we do *nothing* on the hash table and simply destroy |
| 1432 | * it. |
| 1433 | */ |
| 1434 | lttng_ht_destroy(consumer_data.stream_list_ht); |
| 1435 | |
| 1436 | /* |
| 1437 | * Trace chunks in the registry may still exist if the session |
| 1438 | * daemon has encountered an internal error and could not |
| 1439 | * tear down its sessions and/or trace chunks properly. |
| 1440 | * |
| 1441 | * Release the session daemon's implicit reference to any remaining |
| 1442 | * trace chunk and print an error if any trace chunk was found. Note |
| 1443 | * that there are _no_ legitimate cases for trace chunks to be left, |
| 1444 | * it is a leak. However, it can happen following a crash of the |
| 1445 | * session daemon and not emptying the registry would cause an assertion |
| 1446 | * to hit. |
| 1447 | */ |
| 1448 | trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk( |
| 1449 | consumer_data.chunk_registry); |
| 1450 | if (trace_chunks_left) { |
| 1451 | ERR("%u trace chunks are leaked by lttng-consumerd. " |
| 1452 | "This can be caused by an internal error of the session daemon.", |
| 1453 | trace_chunks_left); |
| 1454 | } |
| 1455 | /* Run all callbacks freeing each chunk. */ |
| 1456 | rcu_barrier(); |
| 1457 | lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry); |
| 1458 | } |
| 1459 | |
| 1460 | /* |
| 1461 | * Called from signal handler. |
| 1462 | */ |
| 1463 | void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx) |
| 1464 | { |
| 1465 | ssize_t ret; |
| 1466 | |
| 1467 | CMM_STORE_SHARED(consumer_quit, 1); |
| 1468 | ret = lttng_write(ctx->consumer_should_quit[1], "4", 1); |
| 1469 | if (ret < 1) { |
| 1470 | PERROR("write consumer quit"); |
| 1471 | } |
| 1472 | |
| 1473 | DBG("Consumer flag that it should quit"); |
| 1474 | } |
| 1475 | |
| 1476 | |
| 1477 | /* |
| 1478 | * Flush pending writes to trace output disk file. |
| 1479 | */ |
| 1480 | static |
| 1481 | void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream, |
| 1482 | off_t orig_offset) |
| 1483 | { |
| 1484 | int ret; |
| 1485 | int outfd = stream->out_fd; |
| 1486 | |
| 1487 | /* |
| 1488 | * This does a blocking write-and-wait on any page that belongs to the |
| 1489 | * subbuffer prior to the one we just wrote. |
| 1490 | * Don't care about error values, as these are just hints and ways to |
| 1491 | * limit the amount of page cache used. |
| 1492 | */ |
| 1493 | if (orig_offset < stream->max_sb_size) { |
| 1494 | return; |
| 1495 | } |
| 1496 | lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size, |
| 1497 | stream->max_sb_size, |
| 1498 | SYNC_FILE_RANGE_WAIT_BEFORE |
| 1499 | | SYNC_FILE_RANGE_WRITE |
| 1500 | | SYNC_FILE_RANGE_WAIT_AFTER); |
| 1501 | /* |
| 1502 | * Give hints to the kernel about how we access the file: |
| 1503 | * POSIX_FADV_DONTNEED : we won't re-access data in a near future after |
| 1504 | * we write it. |
| 1505 | * |
| 1506 | * We need to call fadvise again after the file grows because the |
| 1507 | * kernel does not seem to apply fadvise to non-existing parts of the |
| 1508 | * file. |
| 1509 | * |
| 1510 | * Call fadvise _after_ having waited for the page writeback to |
| 1511 | * complete because the dirty page writeback semantic is not well |
| 1512 | * defined. So it can be expected to lead to lower throughput in |
| 1513 | * streaming. |
| 1514 | */ |
| 1515 | ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size, |
| 1516 | stream->max_sb_size, POSIX_FADV_DONTNEED); |
| 1517 | if (ret && ret != -ENOSYS) { |
| 1518 | errno = ret; |
| 1519 | PERROR("posix_fadvise on fd %i", outfd); |
| 1520 | } |
| 1521 | } |
| 1522 | |
| 1523 | /* |
| 1524 | * Initialise the necessary environnement : |
| 1525 | * - create a new context |
| 1526 | * - create the poll_pipe |
| 1527 | * - create the should_quit pipe (for signal handler) |
| 1528 | * - create the thread pipe (for splice) |
| 1529 | * |
| 1530 | * Takes a function pointer as argument, this function is called when data is |
| 1531 | * available on a buffer. This function is responsible to do the |
| 1532 | * kernctl_get_next_subbuf, read the data with mmap or splice depending on the |
| 1533 | * buffer configuration and then kernctl_put_next_subbuf at the end. |
| 1534 | * |
| 1535 | * Returns a pointer to the new context or NULL on error. |
| 1536 | */ |
| 1537 | struct lttng_consumer_local_data *lttng_consumer_create( |
| 1538 | enum lttng_consumer_type type, |
| 1539 | ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream, |
| 1540 | struct lttng_consumer_local_data *ctx), |
| 1541 | int (*recv_channel)(struct lttng_consumer_channel *channel), |
| 1542 | int (*recv_stream)(struct lttng_consumer_stream *stream), |
| 1543 | int (*update_stream)(uint64_t stream_key, uint32_t state)) |
| 1544 | { |
| 1545 | int ret; |
| 1546 | struct lttng_consumer_local_data *ctx; |
| 1547 | |
| 1548 | assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN || |
| 1549 | consumer_data.type == type); |
| 1550 | consumer_data.type = type; |
| 1551 | |
| 1552 | ctx = zmalloc(sizeof(struct lttng_consumer_local_data)); |
| 1553 | if (ctx == NULL) { |
| 1554 | PERROR("allocating context"); |
| 1555 | goto error; |
| 1556 | } |
| 1557 | |
| 1558 | ctx->consumer_error_socket = -1; |
| 1559 | ctx->consumer_metadata_socket = -1; |
| 1560 | pthread_mutex_init(&ctx->metadata_socket_lock, NULL); |
| 1561 | /* assign the callbacks */ |
| 1562 | ctx->on_buffer_ready = buffer_ready; |
| 1563 | ctx->on_recv_channel = recv_channel; |
| 1564 | ctx->on_recv_stream = recv_stream; |
| 1565 | ctx->on_update_stream = update_stream; |
| 1566 | |
| 1567 | ctx->consumer_data_pipe = lttng_pipe_open(0); |
| 1568 | if (!ctx->consumer_data_pipe) { |
| 1569 | goto error_poll_pipe; |
| 1570 | } |
| 1571 | |
| 1572 | ctx->consumer_wakeup_pipe = lttng_pipe_open(0); |
| 1573 | if (!ctx->consumer_wakeup_pipe) { |
| 1574 | goto error_wakeup_pipe; |
| 1575 | } |
| 1576 | |
| 1577 | ret = pipe(ctx->consumer_should_quit); |
| 1578 | if (ret < 0) { |
| 1579 | PERROR("Error creating recv pipe"); |
| 1580 | goto error_quit_pipe; |
| 1581 | } |
| 1582 | |
| 1583 | ret = pipe(ctx->consumer_channel_pipe); |
| 1584 | if (ret < 0) { |
| 1585 | PERROR("Error creating channel pipe"); |
| 1586 | goto error_channel_pipe; |
| 1587 | } |
| 1588 | |
| 1589 | ctx->consumer_metadata_pipe = lttng_pipe_open(0); |
| 1590 | if (!ctx->consumer_metadata_pipe) { |
| 1591 | goto error_metadata_pipe; |
| 1592 | } |
| 1593 | |
| 1594 | ctx->channel_monitor_pipe = -1; |
| 1595 | |
| 1596 | return ctx; |
| 1597 | |
| 1598 | error_metadata_pipe: |
| 1599 | utils_close_pipe(ctx->consumer_channel_pipe); |
| 1600 | error_channel_pipe: |
| 1601 | utils_close_pipe(ctx->consumer_should_quit); |
| 1602 | error_quit_pipe: |
| 1603 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
| 1604 | error_wakeup_pipe: |
| 1605 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
| 1606 | error_poll_pipe: |
| 1607 | free(ctx); |
| 1608 | error: |
| 1609 | return NULL; |
| 1610 | } |
| 1611 | |
| 1612 | /* |
| 1613 | * Iterate over all streams of the hashtable and free them properly. |
| 1614 | */ |
| 1615 | static void destroy_data_stream_ht(struct lttng_ht *ht) |
| 1616 | { |
| 1617 | struct lttng_ht_iter iter; |
| 1618 | struct lttng_consumer_stream *stream; |
| 1619 | |
| 1620 | if (ht == NULL) { |
| 1621 | return; |
| 1622 | } |
| 1623 | |
| 1624 | rcu_read_lock(); |
| 1625 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
| 1626 | /* |
| 1627 | * Ignore return value since we are currently cleaning up so any error |
| 1628 | * can't be handled. |
| 1629 | */ |
| 1630 | (void) consumer_del_stream(stream, ht); |
| 1631 | } |
| 1632 | rcu_read_unlock(); |
| 1633 | |
| 1634 | lttng_ht_destroy(ht); |
| 1635 | } |
| 1636 | |
| 1637 | /* |
| 1638 | * Iterate over all streams of the metadata hashtable and free them |
| 1639 | * properly. |
| 1640 | */ |
| 1641 | static void destroy_metadata_stream_ht(struct lttng_ht *ht) |
| 1642 | { |
| 1643 | struct lttng_ht_iter iter; |
| 1644 | struct lttng_consumer_stream *stream; |
| 1645 | |
| 1646 | if (ht == NULL) { |
| 1647 | return; |
| 1648 | } |
| 1649 | |
| 1650 | rcu_read_lock(); |
| 1651 | cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) { |
| 1652 | /* |
| 1653 | * Ignore return value since we are currently cleaning up so any error |
| 1654 | * can't be handled. |
| 1655 | */ |
| 1656 | (void) consumer_del_metadata_stream(stream, ht); |
| 1657 | } |
| 1658 | rcu_read_unlock(); |
| 1659 | |
| 1660 | lttng_ht_destroy(ht); |
| 1661 | } |
| 1662 | |
| 1663 | /* |
| 1664 | * Close all fds associated with the instance and free the context. |
| 1665 | */ |
| 1666 | void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx) |
| 1667 | { |
| 1668 | int ret; |
| 1669 | |
| 1670 | DBG("Consumer destroying it. Closing everything."); |
| 1671 | |
| 1672 | if (!ctx) { |
| 1673 | return; |
| 1674 | } |
| 1675 | |
| 1676 | destroy_data_stream_ht(data_ht); |
| 1677 | destroy_metadata_stream_ht(metadata_ht); |
| 1678 | |
| 1679 | ret = close(ctx->consumer_error_socket); |
| 1680 | if (ret) { |
| 1681 | PERROR("close"); |
| 1682 | } |
| 1683 | ret = close(ctx->consumer_metadata_socket); |
| 1684 | if (ret) { |
| 1685 | PERROR("close"); |
| 1686 | } |
| 1687 | utils_close_pipe(ctx->consumer_channel_pipe); |
| 1688 | lttng_pipe_destroy(ctx->consumer_data_pipe); |
| 1689 | lttng_pipe_destroy(ctx->consumer_metadata_pipe); |
| 1690 | lttng_pipe_destroy(ctx->consumer_wakeup_pipe); |
| 1691 | utils_close_pipe(ctx->consumer_should_quit); |
| 1692 | |
| 1693 | unlink(ctx->consumer_command_sock_path); |
| 1694 | free(ctx); |
| 1695 | } |
| 1696 | |
| 1697 | /* |
| 1698 | * Write the metadata stream id on the specified file descriptor. |
| 1699 | */ |
| 1700 | static int write_relayd_metadata_id(int fd, |
| 1701 | struct lttng_consumer_stream *stream, |
| 1702 | unsigned long padding) |
| 1703 | { |
| 1704 | ssize_t ret; |
| 1705 | struct lttcomm_relayd_metadata_payload hdr; |
| 1706 | |
| 1707 | hdr.stream_id = htobe64(stream->relayd_stream_id); |
| 1708 | hdr.padding_size = htobe32(padding); |
| 1709 | ret = lttng_write(fd, (void *) &hdr, sizeof(hdr)); |
| 1710 | if (ret < sizeof(hdr)) { |
| 1711 | /* |
| 1712 | * This error means that the fd's end is closed so ignore the PERROR |
| 1713 | * not to clubber the error output since this can happen in a normal |
| 1714 | * code path. |
| 1715 | */ |
| 1716 | if (errno != EPIPE) { |
| 1717 | PERROR("write metadata stream id"); |
| 1718 | } |
| 1719 | DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno); |
| 1720 | /* |
| 1721 | * Set ret to a negative value because if ret != sizeof(hdr), we don't |
| 1722 | * handle writting the missing part so report that as an error and |
| 1723 | * don't lie to the caller. |
| 1724 | */ |
| 1725 | ret = -1; |
| 1726 | goto end; |
| 1727 | } |
| 1728 | DBG("Metadata stream id %" PRIu64 " with padding %lu written before data", |
| 1729 | stream->relayd_stream_id, padding); |
| 1730 | |
| 1731 | end: |
| 1732 | return (int) ret; |
| 1733 | } |
| 1734 | |
| 1735 | /* |
| 1736 | * Mmap the ring buffer, read it and write the data to the tracefile. This is a |
| 1737 | * core function for writing trace buffers to either the local filesystem or |
| 1738 | * the network. |
| 1739 | * |
| 1740 | * It must be called with the stream and the channel lock held. |
| 1741 | * |
| 1742 | * Careful review MUST be put if any changes occur! |
| 1743 | * |
| 1744 | * Returns the number of bytes written |
| 1745 | */ |
| 1746 | ssize_t lttng_consumer_on_read_subbuffer_mmap( |
| 1747 | struct lttng_consumer_local_data *ctx, |
| 1748 | struct lttng_consumer_stream *stream, unsigned long len, |
| 1749 | unsigned long padding, |
| 1750 | struct ctf_packet_index *index) |
| 1751 | { |
| 1752 | unsigned long mmap_offset; |
| 1753 | void *mmap_base; |
| 1754 | ssize_t ret = 0; |
| 1755 | off_t orig_offset = stream->out_fd_offset; |
| 1756 | /* Default is on the disk */ |
| 1757 | int outfd = stream->out_fd; |
| 1758 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 1759 | unsigned int relayd_hang_up = 0; |
| 1760 | |
| 1761 | /* RCU lock for the relayd pointer */ |
| 1762 | rcu_read_lock(); |
| 1763 | assert(stream->net_seq_idx != (uint64_t) -1ULL || |
| 1764 | stream->trace_chunk); |
| 1765 | |
| 1766 | /* Flag that the current stream if set for network streaming. */ |
| 1767 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
| 1768 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 1769 | if (relayd == NULL) { |
| 1770 | ret = -EPIPE; |
| 1771 | goto end; |
| 1772 | } |
| 1773 | } |
| 1774 | |
| 1775 | /* get the offset inside the fd to mmap */ |
| 1776 | switch (consumer_data.type) { |
| 1777 | case LTTNG_CONSUMER_KERNEL: |
| 1778 | mmap_base = stream->mmap_base; |
| 1779 | ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset); |
| 1780 | if (ret < 0) { |
| 1781 | PERROR("tracer ctl get_mmap_read_offset"); |
| 1782 | goto end; |
| 1783 | } |
| 1784 | break; |
| 1785 | case LTTNG_CONSUMER32_UST: |
| 1786 | case LTTNG_CONSUMER64_UST: |
| 1787 | mmap_base = lttng_ustctl_get_mmap_base(stream); |
| 1788 | if (!mmap_base) { |
| 1789 | ERR("read mmap get mmap base for stream %s", stream->name); |
| 1790 | ret = -EPERM; |
| 1791 | goto end; |
| 1792 | } |
| 1793 | ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset); |
| 1794 | if (ret != 0) { |
| 1795 | PERROR("tracer ctl get_mmap_read_offset"); |
| 1796 | ret = -EINVAL; |
| 1797 | goto end; |
| 1798 | } |
| 1799 | break; |
| 1800 | default: |
| 1801 | ERR("Unknown consumer_data type"); |
| 1802 | assert(0); |
| 1803 | } |
| 1804 | |
| 1805 | /* Handle stream on the relayd if the output is on the network */ |
| 1806 | if (relayd) { |
| 1807 | unsigned long netlen = len; |
| 1808 | |
| 1809 | /* |
| 1810 | * Lock the control socket for the complete duration of the function |
| 1811 | * since from this point on we will use the socket. |
| 1812 | */ |
| 1813 | if (stream->metadata_flag) { |
| 1814 | /* Metadata requires the control socket. */ |
| 1815 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 1816 | if (stream->reset_metadata_flag) { |
| 1817 | ret = relayd_reset_metadata(&relayd->control_sock, |
| 1818 | stream->relayd_stream_id, |
| 1819 | stream->metadata_version); |
| 1820 | if (ret < 0) { |
| 1821 | relayd_hang_up = 1; |
| 1822 | goto write_error; |
| 1823 | } |
| 1824 | stream->reset_metadata_flag = 0; |
| 1825 | } |
| 1826 | netlen += sizeof(struct lttcomm_relayd_metadata_payload); |
| 1827 | } |
| 1828 | |
| 1829 | ret = write_relayd_stream_header(stream, netlen, padding, relayd); |
| 1830 | if (ret < 0) { |
| 1831 | relayd_hang_up = 1; |
| 1832 | goto write_error; |
| 1833 | } |
| 1834 | /* Use the returned socket. */ |
| 1835 | outfd = ret; |
| 1836 | |
| 1837 | /* Write metadata stream id before payload */ |
| 1838 | if (stream->metadata_flag) { |
| 1839 | ret = write_relayd_metadata_id(outfd, stream, padding); |
| 1840 | if (ret < 0) { |
| 1841 | relayd_hang_up = 1; |
| 1842 | goto write_error; |
| 1843 | } |
| 1844 | } |
| 1845 | } else { |
| 1846 | /* No streaming, we have to set the len with the full padding */ |
| 1847 | len += padding; |
| 1848 | |
| 1849 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
| 1850 | ret = utils_truncate_stream_file(stream->out_fd, 0); |
| 1851 | if (ret < 0) { |
| 1852 | ERR("Reset metadata file"); |
| 1853 | goto end; |
| 1854 | } |
| 1855 | stream->reset_metadata_flag = 0; |
| 1856 | } |
| 1857 | |
| 1858 | /* |
| 1859 | * Check if we need to change the tracefile before writing the packet. |
| 1860 | */ |
| 1861 | if (stream->chan->tracefile_size > 0 && |
| 1862 | (stream->tracefile_size_current + len) > |
| 1863 | stream->chan->tracefile_size) { |
| 1864 | ret = consumer_stream_rotate_output_files(stream); |
| 1865 | if (ret) { |
| 1866 | goto end; |
| 1867 | } |
| 1868 | outfd = stream->out_fd; |
| 1869 | orig_offset = 0; |
| 1870 | } |
| 1871 | stream->tracefile_size_current += len; |
| 1872 | if (index) { |
| 1873 | index->offset = htobe64(stream->out_fd_offset); |
| 1874 | } |
| 1875 | } |
| 1876 | |
| 1877 | /* |
| 1878 | * This call guarantee that len or less is returned. It's impossible to |
| 1879 | * receive a ret value that is bigger than len. |
| 1880 | */ |
| 1881 | ret = lttng_write(outfd, mmap_base + mmap_offset, len); |
| 1882 | DBG("Consumer mmap write() ret %zd (len %lu)", ret, len); |
| 1883 | if (ret < 0 || ((size_t) ret != len)) { |
| 1884 | /* |
| 1885 | * Report error to caller if nothing was written else at least send the |
| 1886 | * amount written. |
| 1887 | */ |
| 1888 | if (ret < 0) { |
| 1889 | ret = -errno; |
| 1890 | } |
| 1891 | relayd_hang_up = 1; |
| 1892 | |
| 1893 | /* Socket operation failed. We consider the relayd dead */ |
| 1894 | if (errno == EPIPE) { |
| 1895 | /* |
| 1896 | * This is possible if the fd is closed on the other side |
| 1897 | * (outfd) or any write problem. It can be verbose a bit for a |
| 1898 | * normal execution if for instance the relayd is stopped |
| 1899 | * abruptly. This can happen so set this to a DBG statement. |
| 1900 | */ |
| 1901 | DBG("Consumer mmap write detected relayd hang up"); |
| 1902 | } else { |
| 1903 | /* Unhandled error, print it and stop function right now. */ |
| 1904 | PERROR("Error in write mmap (ret %zd != len %lu)", ret, len); |
| 1905 | } |
| 1906 | goto write_error; |
| 1907 | } |
| 1908 | stream->output_written += ret; |
| 1909 | |
| 1910 | /* This call is useless on a socket so better save a syscall. */ |
| 1911 | if (!relayd) { |
| 1912 | /* This won't block, but will start writeout asynchronously */ |
| 1913 | lttng_sync_file_range(outfd, stream->out_fd_offset, len, |
| 1914 | SYNC_FILE_RANGE_WRITE); |
| 1915 | stream->out_fd_offset += len; |
| 1916 | lttng_consumer_sync_trace_file(stream, orig_offset); |
| 1917 | } |
| 1918 | |
| 1919 | write_error: |
| 1920 | /* |
| 1921 | * This is a special case that the relayd has closed its socket. Let's |
| 1922 | * cleanup the relayd object and all associated streams. |
| 1923 | */ |
| 1924 | if (relayd && relayd_hang_up) { |
| 1925 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 1926 | lttng_consumer_cleanup_relayd(relayd); |
| 1927 | } |
| 1928 | |
| 1929 | end: |
| 1930 | /* Unlock only if ctrl socket used */ |
| 1931 | if (relayd && stream->metadata_flag) { |
| 1932 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 1933 | } |
| 1934 | |
| 1935 | rcu_read_unlock(); |
| 1936 | return ret; |
| 1937 | } |
| 1938 | |
| 1939 | /* |
| 1940 | * Splice the data from the ring buffer to the tracefile. |
| 1941 | * |
| 1942 | * It must be called with the stream lock held. |
| 1943 | * |
| 1944 | * Returns the number of bytes spliced. |
| 1945 | */ |
| 1946 | ssize_t lttng_consumer_on_read_subbuffer_splice( |
| 1947 | struct lttng_consumer_local_data *ctx, |
| 1948 | struct lttng_consumer_stream *stream, unsigned long len, |
| 1949 | unsigned long padding, |
| 1950 | struct ctf_packet_index *index) |
| 1951 | { |
| 1952 | ssize_t ret = 0, written = 0, ret_splice = 0; |
| 1953 | loff_t offset = 0; |
| 1954 | off_t orig_offset = stream->out_fd_offset; |
| 1955 | int fd = stream->wait_fd; |
| 1956 | /* Default is on the disk */ |
| 1957 | int outfd = stream->out_fd; |
| 1958 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 1959 | int *splice_pipe; |
| 1960 | unsigned int relayd_hang_up = 0; |
| 1961 | |
| 1962 | switch (consumer_data.type) { |
| 1963 | case LTTNG_CONSUMER_KERNEL: |
| 1964 | break; |
| 1965 | case LTTNG_CONSUMER32_UST: |
| 1966 | case LTTNG_CONSUMER64_UST: |
| 1967 | /* Not supported for user space tracing */ |
| 1968 | return -ENOSYS; |
| 1969 | default: |
| 1970 | ERR("Unknown consumer_data type"); |
| 1971 | assert(0); |
| 1972 | } |
| 1973 | |
| 1974 | /* RCU lock for the relayd pointer */ |
| 1975 | rcu_read_lock(); |
| 1976 | |
| 1977 | /* Flag that the current stream if set for network streaming. */ |
| 1978 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
| 1979 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 1980 | if (relayd == NULL) { |
| 1981 | written = -ret; |
| 1982 | goto end; |
| 1983 | } |
| 1984 | } |
| 1985 | splice_pipe = stream->splice_pipe; |
| 1986 | |
| 1987 | /* Write metadata stream id before payload */ |
| 1988 | if (relayd) { |
| 1989 | unsigned long total_len = len; |
| 1990 | |
| 1991 | if (stream->metadata_flag) { |
| 1992 | /* |
| 1993 | * Lock the control socket for the complete duration of the function |
| 1994 | * since from this point on we will use the socket. |
| 1995 | */ |
| 1996 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 1997 | |
| 1998 | if (stream->reset_metadata_flag) { |
| 1999 | ret = relayd_reset_metadata(&relayd->control_sock, |
| 2000 | stream->relayd_stream_id, |
| 2001 | stream->metadata_version); |
| 2002 | if (ret < 0) { |
| 2003 | relayd_hang_up = 1; |
| 2004 | goto write_error; |
| 2005 | } |
| 2006 | stream->reset_metadata_flag = 0; |
| 2007 | } |
| 2008 | ret = write_relayd_metadata_id(splice_pipe[1], stream, |
| 2009 | padding); |
| 2010 | if (ret < 0) { |
| 2011 | written = ret; |
| 2012 | relayd_hang_up = 1; |
| 2013 | goto write_error; |
| 2014 | } |
| 2015 | |
| 2016 | total_len += sizeof(struct lttcomm_relayd_metadata_payload); |
| 2017 | } |
| 2018 | |
| 2019 | ret = write_relayd_stream_header(stream, total_len, padding, relayd); |
| 2020 | if (ret < 0) { |
| 2021 | written = ret; |
| 2022 | relayd_hang_up = 1; |
| 2023 | goto write_error; |
| 2024 | } |
| 2025 | /* Use the returned socket. */ |
| 2026 | outfd = ret; |
| 2027 | } else { |
| 2028 | /* No streaming, we have to set the len with the full padding */ |
| 2029 | len += padding; |
| 2030 | |
| 2031 | if (stream->metadata_flag && stream->reset_metadata_flag) { |
| 2032 | ret = utils_truncate_stream_file(stream->out_fd, 0); |
| 2033 | if (ret < 0) { |
| 2034 | ERR("Reset metadata file"); |
| 2035 | goto end; |
| 2036 | } |
| 2037 | stream->reset_metadata_flag = 0; |
| 2038 | } |
| 2039 | /* |
| 2040 | * Check if we need to change the tracefile before writing the packet. |
| 2041 | */ |
| 2042 | if (stream->chan->tracefile_size > 0 && |
| 2043 | (stream->tracefile_size_current + len) > |
| 2044 | stream->chan->tracefile_size) { |
| 2045 | ret = consumer_stream_rotate_output_files(stream); |
| 2046 | if (ret < 0) { |
| 2047 | written = ret; |
| 2048 | goto end; |
| 2049 | } |
| 2050 | outfd = stream->out_fd; |
| 2051 | orig_offset = 0; |
| 2052 | } |
| 2053 | stream->tracefile_size_current += len; |
| 2054 | index->offset = htobe64(stream->out_fd_offset); |
| 2055 | } |
| 2056 | |
| 2057 | while (len > 0) { |
| 2058 | DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)", |
| 2059 | (unsigned long)offset, len, fd, splice_pipe[1]); |
| 2060 | ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len, |
| 2061 | SPLICE_F_MOVE | SPLICE_F_MORE); |
| 2062 | DBG("splice chan to pipe, ret %zd", ret_splice); |
| 2063 | if (ret_splice < 0) { |
| 2064 | ret = errno; |
| 2065 | written = -ret; |
| 2066 | PERROR("Error in relay splice"); |
| 2067 | goto splice_error; |
| 2068 | } |
| 2069 | |
| 2070 | /* Handle stream on the relayd if the output is on the network */ |
| 2071 | if (relayd && stream->metadata_flag) { |
| 2072 | size_t metadata_payload_size = |
| 2073 | sizeof(struct lttcomm_relayd_metadata_payload); |
| 2074 | |
| 2075 | /* Update counter to fit the spliced data */ |
| 2076 | ret_splice += metadata_payload_size; |
| 2077 | len += metadata_payload_size; |
| 2078 | /* |
| 2079 | * We do this so the return value can match the len passed as |
| 2080 | * argument to this function. |
| 2081 | */ |
| 2082 | written -= metadata_payload_size; |
| 2083 | } |
| 2084 | |
| 2085 | /* Splice data out */ |
| 2086 | ret_splice = splice(splice_pipe[0], NULL, outfd, NULL, |
| 2087 | ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE); |
| 2088 | DBG("Consumer splice pipe to file (out_fd: %d), ret %zd", |
| 2089 | outfd, ret_splice); |
| 2090 | if (ret_splice < 0) { |
| 2091 | ret = errno; |
| 2092 | written = -ret; |
| 2093 | relayd_hang_up = 1; |
| 2094 | goto write_error; |
| 2095 | } else if (ret_splice > len) { |
| 2096 | /* |
| 2097 | * We don't expect this code path to be executed but you never know |
| 2098 | * so this is an extra protection agains a buggy splice(). |
| 2099 | */ |
| 2100 | ret = errno; |
| 2101 | written += ret_splice; |
| 2102 | PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice, |
| 2103 | len); |
| 2104 | goto splice_error; |
| 2105 | } else { |
| 2106 | /* All good, update current len and continue. */ |
| 2107 | len -= ret_splice; |
| 2108 | } |
| 2109 | |
| 2110 | /* This call is useless on a socket so better save a syscall. */ |
| 2111 | if (!relayd) { |
| 2112 | /* This won't block, but will start writeout asynchronously */ |
| 2113 | lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice, |
| 2114 | SYNC_FILE_RANGE_WRITE); |
| 2115 | stream->out_fd_offset += ret_splice; |
| 2116 | } |
| 2117 | stream->output_written += ret_splice; |
| 2118 | written += ret_splice; |
| 2119 | } |
| 2120 | if (!relayd) { |
| 2121 | lttng_consumer_sync_trace_file(stream, orig_offset); |
| 2122 | } |
| 2123 | goto end; |
| 2124 | |
| 2125 | write_error: |
| 2126 | /* |
| 2127 | * This is a special case that the relayd has closed its socket. Let's |
| 2128 | * cleanup the relayd object and all associated streams. |
| 2129 | */ |
| 2130 | if (relayd && relayd_hang_up) { |
| 2131 | ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 2132 | lttng_consumer_cleanup_relayd(relayd); |
| 2133 | /* Skip splice error so the consumer does not fail */ |
| 2134 | goto end; |
| 2135 | } |
| 2136 | |
| 2137 | splice_error: |
| 2138 | /* send the appropriate error description to sessiond */ |
| 2139 | switch (ret) { |
| 2140 | case EINVAL: |
| 2141 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL); |
| 2142 | break; |
| 2143 | case ENOMEM: |
| 2144 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM); |
| 2145 | break; |
| 2146 | case ESPIPE: |
| 2147 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE); |
| 2148 | break; |
| 2149 | } |
| 2150 | |
| 2151 | end: |
| 2152 | if (relayd && stream->metadata_flag) { |
| 2153 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 2154 | } |
| 2155 | |
| 2156 | rcu_read_unlock(); |
| 2157 | return written; |
| 2158 | } |
| 2159 | |
| 2160 | /* |
| 2161 | * Sample the snapshot positions for a specific fd |
| 2162 | * |
| 2163 | * Returns 0 on success, < 0 on error |
| 2164 | */ |
| 2165 | int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream) |
| 2166 | { |
| 2167 | switch (consumer_data.type) { |
| 2168 | case LTTNG_CONSUMER_KERNEL: |
| 2169 | return lttng_kconsumer_sample_snapshot_positions(stream); |
| 2170 | case LTTNG_CONSUMER32_UST: |
| 2171 | case LTTNG_CONSUMER64_UST: |
| 2172 | return lttng_ustconsumer_sample_snapshot_positions(stream); |
| 2173 | default: |
| 2174 | ERR("Unknown consumer_data type"); |
| 2175 | assert(0); |
| 2176 | return -ENOSYS; |
| 2177 | } |
| 2178 | } |
| 2179 | /* |
| 2180 | * Take a snapshot for a specific fd |
| 2181 | * |
| 2182 | * Returns 0 on success, < 0 on error |
| 2183 | */ |
| 2184 | int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream) |
| 2185 | { |
| 2186 | switch (consumer_data.type) { |
| 2187 | case LTTNG_CONSUMER_KERNEL: |
| 2188 | return lttng_kconsumer_take_snapshot(stream); |
| 2189 | case LTTNG_CONSUMER32_UST: |
| 2190 | case LTTNG_CONSUMER64_UST: |
| 2191 | return lttng_ustconsumer_take_snapshot(stream); |
| 2192 | default: |
| 2193 | ERR("Unknown consumer_data type"); |
| 2194 | assert(0); |
| 2195 | return -ENOSYS; |
| 2196 | } |
| 2197 | } |
| 2198 | |
| 2199 | /* |
| 2200 | * Get the produced position |
| 2201 | * |
| 2202 | * Returns 0 on success, < 0 on error |
| 2203 | */ |
| 2204 | int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream, |
| 2205 | unsigned long *pos) |
| 2206 | { |
| 2207 | switch (consumer_data.type) { |
| 2208 | case LTTNG_CONSUMER_KERNEL: |
| 2209 | return lttng_kconsumer_get_produced_snapshot(stream, pos); |
| 2210 | case LTTNG_CONSUMER32_UST: |
| 2211 | case LTTNG_CONSUMER64_UST: |
| 2212 | return lttng_ustconsumer_get_produced_snapshot(stream, pos); |
| 2213 | default: |
| 2214 | ERR("Unknown consumer_data type"); |
| 2215 | assert(0); |
| 2216 | return -ENOSYS; |
| 2217 | } |
| 2218 | } |
| 2219 | |
| 2220 | /* |
| 2221 | * Get the consumed position (free-running counter position in bytes). |
| 2222 | * |
| 2223 | * Returns 0 on success, < 0 on error |
| 2224 | */ |
| 2225 | int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream, |
| 2226 | unsigned long *pos) |
| 2227 | { |
| 2228 | switch (consumer_data.type) { |
| 2229 | case LTTNG_CONSUMER_KERNEL: |
| 2230 | return lttng_kconsumer_get_consumed_snapshot(stream, pos); |
| 2231 | case LTTNG_CONSUMER32_UST: |
| 2232 | case LTTNG_CONSUMER64_UST: |
| 2233 | return lttng_ustconsumer_get_consumed_snapshot(stream, pos); |
| 2234 | default: |
| 2235 | ERR("Unknown consumer_data type"); |
| 2236 | assert(0); |
| 2237 | return -ENOSYS; |
| 2238 | } |
| 2239 | } |
| 2240 | |
| 2241 | int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx, |
| 2242 | int sock, struct pollfd *consumer_sockpoll) |
| 2243 | { |
| 2244 | switch (consumer_data.type) { |
| 2245 | case LTTNG_CONSUMER_KERNEL: |
| 2246 | return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
| 2247 | case LTTNG_CONSUMER32_UST: |
| 2248 | case LTTNG_CONSUMER64_UST: |
| 2249 | return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll); |
| 2250 | default: |
| 2251 | ERR("Unknown consumer_data type"); |
| 2252 | assert(0); |
| 2253 | return -ENOSYS; |
| 2254 | } |
| 2255 | } |
| 2256 | |
| 2257 | void lttng_consumer_close_all_metadata(void) |
| 2258 | { |
| 2259 | switch (consumer_data.type) { |
| 2260 | case LTTNG_CONSUMER_KERNEL: |
| 2261 | /* |
| 2262 | * The Kernel consumer has a different metadata scheme so we don't |
| 2263 | * close anything because the stream will be closed by the session |
| 2264 | * daemon. |
| 2265 | */ |
| 2266 | break; |
| 2267 | case LTTNG_CONSUMER32_UST: |
| 2268 | case LTTNG_CONSUMER64_UST: |
| 2269 | /* |
| 2270 | * Close all metadata streams. The metadata hash table is passed and |
| 2271 | * this call iterates over it by closing all wakeup fd. This is safe |
| 2272 | * because at this point we are sure that the metadata producer is |
| 2273 | * either dead or blocked. |
| 2274 | */ |
| 2275 | lttng_ustconsumer_close_all_metadata(metadata_ht); |
| 2276 | break; |
| 2277 | default: |
| 2278 | ERR("Unknown consumer_data type"); |
| 2279 | assert(0); |
| 2280 | } |
| 2281 | } |
| 2282 | |
| 2283 | /* |
| 2284 | * Clean up a metadata stream and free its memory. |
| 2285 | */ |
| 2286 | void consumer_del_metadata_stream(struct lttng_consumer_stream *stream, |
| 2287 | struct lttng_ht *ht) |
| 2288 | { |
| 2289 | struct lttng_consumer_channel *channel = NULL; |
| 2290 | bool free_channel = false; |
| 2291 | |
| 2292 | assert(stream); |
| 2293 | /* |
| 2294 | * This call should NEVER receive regular stream. It must always be |
| 2295 | * metadata stream and this is crucial for data structure synchronization. |
| 2296 | */ |
| 2297 | assert(stream->metadata_flag); |
| 2298 | |
| 2299 | DBG3("Consumer delete metadata stream %d", stream->wait_fd); |
| 2300 | |
| 2301 | pthread_mutex_lock(&consumer_data.lock); |
| 2302 | /* |
| 2303 | * Note that this assumes that a stream's channel is never changed and |
| 2304 | * that the stream's lock doesn't need to be taken to sample its |
| 2305 | * channel. |
| 2306 | */ |
| 2307 | channel = stream->chan; |
| 2308 | pthread_mutex_lock(&channel->lock); |
| 2309 | pthread_mutex_lock(&stream->lock); |
| 2310 | if (channel->metadata_cache) { |
| 2311 | /* Only applicable to userspace consumers. */ |
| 2312 | pthread_mutex_lock(&channel->metadata_cache->lock); |
| 2313 | } |
| 2314 | |
| 2315 | /* Remove any reference to that stream. */ |
| 2316 | consumer_stream_delete(stream, ht); |
| 2317 | |
| 2318 | /* Close down everything including the relayd if one. */ |
| 2319 | consumer_stream_close(stream); |
| 2320 | /* Destroy tracer buffers of the stream. */ |
| 2321 | consumer_stream_destroy_buffers(stream); |
| 2322 | |
| 2323 | /* Atomically decrement channel refcount since other threads can use it. */ |
| 2324 | if (!uatomic_sub_return(&channel->refcount, 1) |
| 2325 | && !uatomic_read(&channel->nb_init_stream_left)) { |
| 2326 | /* Go for channel deletion! */ |
| 2327 | free_channel = true; |
| 2328 | } |
| 2329 | stream->chan = NULL; |
| 2330 | |
| 2331 | /* |
| 2332 | * Nullify the stream reference so it is not used after deletion. The |
| 2333 | * channel lock MUST be acquired before being able to check for a NULL |
| 2334 | * pointer value. |
| 2335 | */ |
| 2336 | channel->metadata_stream = NULL; |
| 2337 | |
| 2338 | if (channel->metadata_cache) { |
| 2339 | pthread_mutex_unlock(&channel->metadata_cache->lock); |
| 2340 | } |
| 2341 | pthread_mutex_unlock(&stream->lock); |
| 2342 | pthread_mutex_unlock(&channel->lock); |
| 2343 | pthread_mutex_unlock(&consumer_data.lock); |
| 2344 | |
| 2345 | if (free_channel) { |
| 2346 | consumer_del_channel(channel); |
| 2347 | } |
| 2348 | |
| 2349 | lttng_trace_chunk_put(stream->trace_chunk); |
| 2350 | stream->trace_chunk = NULL; |
| 2351 | consumer_stream_free(stream); |
| 2352 | } |
| 2353 | |
| 2354 | /* |
| 2355 | * Action done with the metadata stream when adding it to the consumer internal |
| 2356 | * data structures to handle it. |
| 2357 | */ |
| 2358 | void consumer_add_metadata_stream(struct lttng_consumer_stream *stream) |
| 2359 | { |
| 2360 | struct lttng_ht *ht = metadata_ht; |
| 2361 | struct lttng_ht_iter iter; |
| 2362 | struct lttng_ht_node_u64 *node; |
| 2363 | |
| 2364 | assert(stream); |
| 2365 | assert(ht); |
| 2366 | |
| 2367 | DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key); |
| 2368 | |
| 2369 | pthread_mutex_lock(&consumer_data.lock); |
| 2370 | pthread_mutex_lock(&stream->chan->lock); |
| 2371 | pthread_mutex_lock(&stream->chan->timer_lock); |
| 2372 | pthread_mutex_lock(&stream->lock); |
| 2373 | |
| 2374 | /* |
| 2375 | * From here, refcounts are updated so be _careful_ when returning an error |
| 2376 | * after this point. |
| 2377 | */ |
| 2378 | |
| 2379 | rcu_read_lock(); |
| 2380 | |
| 2381 | /* |
| 2382 | * Lookup the stream just to make sure it does not exist in our internal |
| 2383 | * state. This should NEVER happen. |
| 2384 | */ |
| 2385 | lttng_ht_lookup(ht, &stream->key, &iter); |
| 2386 | node = lttng_ht_iter_get_node_u64(&iter); |
| 2387 | assert(!node); |
| 2388 | |
| 2389 | /* |
| 2390 | * When nb_init_stream_left reaches 0, we don't need to trigger any action |
| 2391 | * in terms of destroying the associated channel, because the action that |
| 2392 | * causes the count to become 0 also causes a stream to be added. The |
| 2393 | * channel deletion will thus be triggered by the following removal of this |
| 2394 | * stream. |
| 2395 | */ |
| 2396 | if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) { |
| 2397 | /* Increment refcount before decrementing nb_init_stream_left */ |
| 2398 | cmm_smp_wmb(); |
| 2399 | uatomic_dec(&stream->chan->nb_init_stream_left); |
| 2400 | } |
| 2401 | |
| 2402 | lttng_ht_add_unique_u64(ht, &stream->node); |
| 2403 | |
| 2404 | lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht, |
| 2405 | &stream->node_channel_id); |
| 2406 | |
| 2407 | /* |
| 2408 | * Add stream to the stream_list_ht of the consumer data. No need to steal |
| 2409 | * the key since the HT does not use it and we allow to add redundant keys |
| 2410 | * into this table. |
| 2411 | */ |
| 2412 | lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id); |
| 2413 | |
| 2414 | rcu_read_unlock(); |
| 2415 | |
| 2416 | pthread_mutex_unlock(&stream->lock); |
| 2417 | pthread_mutex_unlock(&stream->chan->lock); |
| 2418 | pthread_mutex_unlock(&stream->chan->timer_lock); |
| 2419 | pthread_mutex_unlock(&consumer_data.lock); |
| 2420 | } |
| 2421 | |
| 2422 | /* |
| 2423 | * Delete data stream that are flagged for deletion (endpoint_status). |
| 2424 | */ |
| 2425 | static void validate_endpoint_status_data_stream(void) |
| 2426 | { |
| 2427 | struct lttng_ht_iter iter; |
| 2428 | struct lttng_consumer_stream *stream; |
| 2429 | |
| 2430 | DBG("Consumer delete flagged data stream"); |
| 2431 | |
| 2432 | rcu_read_lock(); |
| 2433 | cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) { |
| 2434 | /* Validate delete flag of the stream */ |
| 2435 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
| 2436 | continue; |
| 2437 | } |
| 2438 | /* Delete it right now */ |
| 2439 | consumer_del_stream(stream, data_ht); |
| 2440 | } |
| 2441 | rcu_read_unlock(); |
| 2442 | } |
| 2443 | |
| 2444 | /* |
| 2445 | * Delete metadata stream that are flagged for deletion (endpoint_status). |
| 2446 | */ |
| 2447 | static void validate_endpoint_status_metadata_stream( |
| 2448 | struct lttng_poll_event *pollset) |
| 2449 | { |
| 2450 | struct lttng_ht_iter iter; |
| 2451 | struct lttng_consumer_stream *stream; |
| 2452 | |
| 2453 | DBG("Consumer delete flagged metadata stream"); |
| 2454 | |
| 2455 | assert(pollset); |
| 2456 | |
| 2457 | rcu_read_lock(); |
| 2458 | cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) { |
| 2459 | /* Validate delete flag of the stream */ |
| 2460 | if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) { |
| 2461 | continue; |
| 2462 | } |
| 2463 | /* |
| 2464 | * Remove from pollset so the metadata thread can continue without |
| 2465 | * blocking on a deleted stream. |
| 2466 | */ |
| 2467 | lttng_poll_del(pollset, stream->wait_fd); |
| 2468 | |
| 2469 | /* Delete it right now */ |
| 2470 | consumer_del_metadata_stream(stream, metadata_ht); |
| 2471 | } |
| 2472 | rcu_read_unlock(); |
| 2473 | } |
| 2474 | |
| 2475 | /* |
| 2476 | * Thread polls on metadata file descriptor and write them on disk or on the |
| 2477 | * network. |
| 2478 | */ |
| 2479 | void *consumer_thread_metadata_poll(void *data) |
| 2480 | { |
| 2481 | int ret, i, pollfd, err = -1; |
| 2482 | uint32_t revents, nb_fd; |
| 2483 | struct lttng_consumer_stream *stream = NULL; |
| 2484 | struct lttng_ht_iter iter; |
| 2485 | struct lttng_ht_node_u64 *node; |
| 2486 | struct lttng_poll_event events; |
| 2487 | struct lttng_consumer_local_data *ctx = data; |
| 2488 | ssize_t len; |
| 2489 | |
| 2490 | rcu_register_thread(); |
| 2491 | |
| 2492 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA); |
| 2493 | |
| 2494 | if (testpoint(consumerd_thread_metadata)) { |
| 2495 | goto error_testpoint; |
| 2496 | } |
| 2497 | |
| 2498 | health_code_update(); |
| 2499 | |
| 2500 | DBG("Thread metadata poll started"); |
| 2501 | |
| 2502 | /* Size is set to 1 for the consumer_metadata pipe */ |
| 2503 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); |
| 2504 | if (ret < 0) { |
| 2505 | ERR("Poll set creation failed"); |
| 2506 | goto end_poll; |
| 2507 | } |
| 2508 | |
| 2509 | ret = lttng_poll_add(&events, |
| 2510 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN); |
| 2511 | if (ret < 0) { |
| 2512 | goto end; |
| 2513 | } |
| 2514 | |
| 2515 | /* Main loop */ |
| 2516 | DBG("Metadata main loop started"); |
| 2517 | |
| 2518 | while (1) { |
| 2519 | restart: |
| 2520 | health_code_update(); |
| 2521 | health_poll_entry(); |
| 2522 | DBG("Metadata poll wait"); |
| 2523 | ret = lttng_poll_wait(&events, -1); |
| 2524 | DBG("Metadata poll return from wait with %d fd(s)", |
| 2525 | LTTNG_POLL_GETNB(&events)); |
| 2526 | health_poll_exit(); |
| 2527 | DBG("Metadata event caught in thread"); |
| 2528 | if (ret < 0) { |
| 2529 | if (errno == EINTR) { |
| 2530 | ERR("Poll EINTR caught"); |
| 2531 | goto restart; |
| 2532 | } |
| 2533 | if (LTTNG_POLL_GETNB(&events) == 0) { |
| 2534 | err = 0; /* All is OK */ |
| 2535 | } |
| 2536 | goto end; |
| 2537 | } |
| 2538 | |
| 2539 | nb_fd = ret; |
| 2540 | |
| 2541 | /* From here, the event is a metadata wait fd */ |
| 2542 | for (i = 0; i < nb_fd; i++) { |
| 2543 | health_code_update(); |
| 2544 | |
| 2545 | revents = LTTNG_POLL_GETEV(&events, i); |
| 2546 | pollfd = LTTNG_POLL_GETFD(&events, i); |
| 2547 | |
| 2548 | if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) { |
| 2549 | if (revents & LPOLLIN) { |
| 2550 | ssize_t pipe_len; |
| 2551 | |
| 2552 | pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe, |
| 2553 | &stream, sizeof(stream)); |
| 2554 | if (pipe_len < sizeof(stream)) { |
| 2555 | if (pipe_len < 0) { |
| 2556 | PERROR("read metadata stream"); |
| 2557 | } |
| 2558 | /* |
| 2559 | * Remove the pipe from the poll set and continue the loop |
| 2560 | * since their might be data to consume. |
| 2561 | */ |
| 2562 | lttng_poll_del(&events, |
| 2563 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); |
| 2564 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); |
| 2565 | continue; |
| 2566 | } |
| 2567 | |
| 2568 | /* A NULL stream means that the state has changed. */ |
| 2569 | if (stream == NULL) { |
| 2570 | /* Check for deleted streams. */ |
| 2571 | validate_endpoint_status_metadata_stream(&events); |
| 2572 | goto restart; |
| 2573 | } |
| 2574 | |
| 2575 | DBG("Adding metadata stream %d to poll set", |
| 2576 | stream->wait_fd); |
| 2577 | |
| 2578 | /* Add metadata stream to the global poll events list */ |
| 2579 | lttng_poll_add(&events, stream->wait_fd, |
| 2580 | LPOLLIN | LPOLLPRI | LPOLLHUP); |
| 2581 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
| 2582 | DBG("Metadata thread pipe hung up"); |
| 2583 | /* |
| 2584 | * Remove the pipe from the poll set and continue the loop |
| 2585 | * since their might be data to consume. |
| 2586 | */ |
| 2587 | lttng_poll_del(&events, |
| 2588 | lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)); |
| 2589 | lttng_pipe_read_close(ctx->consumer_metadata_pipe); |
| 2590 | continue; |
| 2591 | } else { |
| 2592 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); |
| 2593 | goto end; |
| 2594 | } |
| 2595 | |
| 2596 | /* Handle other stream */ |
| 2597 | continue; |
| 2598 | } |
| 2599 | |
| 2600 | rcu_read_lock(); |
| 2601 | { |
| 2602 | uint64_t tmp_id = (uint64_t) pollfd; |
| 2603 | |
| 2604 | lttng_ht_lookup(metadata_ht, &tmp_id, &iter); |
| 2605 | } |
| 2606 | node = lttng_ht_iter_get_node_u64(&iter); |
| 2607 | assert(node); |
| 2608 | |
| 2609 | stream = caa_container_of(node, struct lttng_consumer_stream, |
| 2610 | node); |
| 2611 | |
| 2612 | if (revents & (LPOLLIN | LPOLLPRI)) { |
| 2613 | /* Get the data out of the metadata file descriptor */ |
| 2614 | DBG("Metadata available on fd %d", pollfd); |
| 2615 | assert(stream->wait_fd == pollfd); |
| 2616 | |
| 2617 | do { |
| 2618 | health_code_update(); |
| 2619 | |
| 2620 | len = ctx->on_buffer_ready(stream, ctx); |
| 2621 | /* |
| 2622 | * We don't check the return value here since if we get |
| 2623 | * a negative len, it means an error occurred thus we |
| 2624 | * simply remove it from the poll set and free the |
| 2625 | * stream. |
| 2626 | */ |
| 2627 | } while (len > 0); |
| 2628 | |
| 2629 | /* It's ok to have an unavailable sub-buffer */ |
| 2630 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
| 2631 | /* Clean up stream from consumer and free it. */ |
| 2632 | lttng_poll_del(&events, stream->wait_fd); |
| 2633 | consumer_del_metadata_stream(stream, metadata_ht); |
| 2634 | } |
| 2635 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
| 2636 | DBG("Metadata fd %d is hup|err.", pollfd); |
| 2637 | if (!stream->hangup_flush_done |
| 2638 | && (consumer_data.type == LTTNG_CONSUMER32_UST |
| 2639 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { |
| 2640 | DBG("Attempting to flush and consume the UST buffers"); |
| 2641 | lttng_ustconsumer_on_stream_hangup(stream); |
| 2642 | |
| 2643 | /* We just flushed the stream now read it. */ |
| 2644 | do { |
| 2645 | health_code_update(); |
| 2646 | |
| 2647 | len = ctx->on_buffer_ready(stream, ctx); |
| 2648 | /* |
| 2649 | * We don't check the return value here since if we get |
| 2650 | * a negative len, it means an error occurred thus we |
| 2651 | * simply remove it from the poll set and free the |
| 2652 | * stream. |
| 2653 | */ |
| 2654 | } while (len > 0); |
| 2655 | } |
| 2656 | |
| 2657 | lttng_poll_del(&events, stream->wait_fd); |
| 2658 | /* |
| 2659 | * This call update the channel states, closes file descriptors |
| 2660 | * and securely free the stream. |
| 2661 | */ |
| 2662 | consumer_del_metadata_stream(stream, metadata_ht); |
| 2663 | } else { |
| 2664 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); |
| 2665 | rcu_read_unlock(); |
| 2666 | goto end; |
| 2667 | } |
| 2668 | /* Release RCU lock for the stream looked up */ |
| 2669 | rcu_read_unlock(); |
| 2670 | } |
| 2671 | } |
| 2672 | |
| 2673 | /* All is OK */ |
| 2674 | err = 0; |
| 2675 | end: |
| 2676 | DBG("Metadata poll thread exiting"); |
| 2677 | |
| 2678 | lttng_poll_clean(&events); |
| 2679 | end_poll: |
| 2680 | error_testpoint: |
| 2681 | if (err) { |
| 2682 | health_error(); |
| 2683 | ERR("Health error occurred in %s", __func__); |
| 2684 | } |
| 2685 | health_unregister(health_consumerd); |
| 2686 | rcu_unregister_thread(); |
| 2687 | return NULL; |
| 2688 | } |
| 2689 | |
| 2690 | /* |
| 2691 | * This thread polls the fds in the set to consume the data and write |
| 2692 | * it to tracefile if necessary. |
| 2693 | */ |
| 2694 | void *consumer_thread_data_poll(void *data) |
| 2695 | { |
| 2696 | int num_rdy, num_hup, high_prio, ret, i, err = -1; |
| 2697 | struct pollfd *pollfd = NULL; |
| 2698 | /* local view of the streams */ |
| 2699 | struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL; |
| 2700 | /* local view of consumer_data.fds_count */ |
| 2701 | int nb_fd = 0; |
| 2702 | /* 2 for the consumer_data_pipe and wake up pipe */ |
| 2703 | const int nb_pipes_fd = 2; |
| 2704 | /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */ |
| 2705 | int nb_inactive_fd = 0; |
| 2706 | struct lttng_consumer_local_data *ctx = data; |
| 2707 | ssize_t len; |
| 2708 | |
| 2709 | rcu_register_thread(); |
| 2710 | |
| 2711 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA); |
| 2712 | |
| 2713 | if (testpoint(consumerd_thread_data)) { |
| 2714 | goto error_testpoint; |
| 2715 | } |
| 2716 | |
| 2717 | health_code_update(); |
| 2718 | |
| 2719 | local_stream = zmalloc(sizeof(struct lttng_consumer_stream *)); |
| 2720 | if (local_stream == NULL) { |
| 2721 | PERROR("local_stream malloc"); |
| 2722 | goto end; |
| 2723 | } |
| 2724 | |
| 2725 | while (1) { |
| 2726 | health_code_update(); |
| 2727 | |
| 2728 | high_prio = 0; |
| 2729 | num_hup = 0; |
| 2730 | |
| 2731 | /* |
| 2732 | * the fds set has been updated, we need to update our |
| 2733 | * local array as well |
| 2734 | */ |
| 2735 | pthread_mutex_lock(&consumer_data.lock); |
| 2736 | if (consumer_data.need_update) { |
| 2737 | free(pollfd); |
| 2738 | pollfd = NULL; |
| 2739 | |
| 2740 | free(local_stream); |
| 2741 | local_stream = NULL; |
| 2742 | |
| 2743 | /* Allocate for all fds */ |
| 2744 | pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd)); |
| 2745 | if (pollfd == NULL) { |
| 2746 | PERROR("pollfd malloc"); |
| 2747 | pthread_mutex_unlock(&consumer_data.lock); |
| 2748 | goto end; |
| 2749 | } |
| 2750 | |
| 2751 | local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) * |
| 2752 | sizeof(struct lttng_consumer_stream *)); |
| 2753 | if (local_stream == NULL) { |
| 2754 | PERROR("local_stream malloc"); |
| 2755 | pthread_mutex_unlock(&consumer_data.lock); |
| 2756 | goto end; |
| 2757 | } |
| 2758 | ret = update_poll_array(ctx, &pollfd, local_stream, |
| 2759 | data_ht, &nb_inactive_fd); |
| 2760 | if (ret < 0) { |
| 2761 | ERR("Error in allocating pollfd or local_outfds"); |
| 2762 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
| 2763 | pthread_mutex_unlock(&consumer_data.lock); |
| 2764 | goto end; |
| 2765 | } |
| 2766 | nb_fd = ret; |
| 2767 | consumer_data.need_update = 0; |
| 2768 | } |
| 2769 | pthread_mutex_unlock(&consumer_data.lock); |
| 2770 | |
| 2771 | /* No FDs and consumer_quit, consumer_cleanup the thread */ |
| 2772 | if (nb_fd == 0 && nb_inactive_fd == 0 && |
| 2773 | CMM_LOAD_SHARED(consumer_quit) == 1) { |
| 2774 | err = 0; /* All is OK */ |
| 2775 | goto end; |
| 2776 | } |
| 2777 | /* poll on the array of fds */ |
| 2778 | restart: |
| 2779 | DBG("polling on %d fd", nb_fd + nb_pipes_fd); |
| 2780 | if (testpoint(consumerd_thread_data_poll)) { |
| 2781 | goto end; |
| 2782 | } |
| 2783 | health_poll_entry(); |
| 2784 | num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1); |
| 2785 | health_poll_exit(); |
| 2786 | DBG("poll num_rdy : %d", num_rdy); |
| 2787 | if (num_rdy == -1) { |
| 2788 | /* |
| 2789 | * Restart interrupted system call. |
| 2790 | */ |
| 2791 | if (errno == EINTR) { |
| 2792 | goto restart; |
| 2793 | } |
| 2794 | PERROR("Poll error"); |
| 2795 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
| 2796 | goto end; |
| 2797 | } else if (num_rdy == 0) { |
| 2798 | DBG("Polling thread timed out"); |
| 2799 | goto end; |
| 2800 | } |
| 2801 | |
| 2802 | if (caa_unlikely(data_consumption_paused)) { |
| 2803 | DBG("Data consumption paused, sleeping..."); |
| 2804 | sleep(1); |
| 2805 | goto restart; |
| 2806 | } |
| 2807 | |
| 2808 | /* |
| 2809 | * If the consumer_data_pipe triggered poll go directly to the |
| 2810 | * beginning of the loop to update the array. We want to prioritize |
| 2811 | * array update over low-priority reads. |
| 2812 | */ |
| 2813 | if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) { |
| 2814 | ssize_t pipe_readlen; |
| 2815 | |
| 2816 | DBG("consumer_data_pipe wake up"); |
| 2817 | pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe, |
| 2818 | &new_stream, sizeof(new_stream)); |
| 2819 | if (pipe_readlen < sizeof(new_stream)) { |
| 2820 | PERROR("Consumer data pipe"); |
| 2821 | /* Continue so we can at least handle the current stream(s). */ |
| 2822 | continue; |
| 2823 | } |
| 2824 | |
| 2825 | /* |
| 2826 | * If the stream is NULL, just ignore it. It's also possible that |
| 2827 | * the sessiond poll thread changed the consumer_quit state and is |
| 2828 | * waking us up to test it. |
| 2829 | */ |
| 2830 | if (new_stream == NULL) { |
| 2831 | validate_endpoint_status_data_stream(); |
| 2832 | continue; |
| 2833 | } |
| 2834 | |
| 2835 | /* Continue to update the local streams and handle prio ones */ |
| 2836 | continue; |
| 2837 | } |
| 2838 | |
| 2839 | /* Handle wakeup pipe. */ |
| 2840 | if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) { |
| 2841 | char dummy; |
| 2842 | ssize_t pipe_readlen; |
| 2843 | |
| 2844 | pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy, |
| 2845 | sizeof(dummy)); |
| 2846 | if (pipe_readlen < 0) { |
| 2847 | PERROR("Consumer data wakeup pipe"); |
| 2848 | } |
| 2849 | /* We've been awakened to handle stream(s). */ |
| 2850 | ctx->has_wakeup = 0; |
| 2851 | } |
| 2852 | |
| 2853 | /* Take care of high priority channels first. */ |
| 2854 | for (i = 0; i < nb_fd; i++) { |
| 2855 | health_code_update(); |
| 2856 | |
| 2857 | if (local_stream[i] == NULL) { |
| 2858 | continue; |
| 2859 | } |
| 2860 | if (pollfd[i].revents & POLLPRI) { |
| 2861 | DBG("Urgent read on fd %d", pollfd[i].fd); |
| 2862 | high_prio = 1; |
| 2863 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
| 2864 | /* it's ok to have an unavailable sub-buffer */ |
| 2865 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
| 2866 | /* Clean the stream and free it. */ |
| 2867 | consumer_del_stream(local_stream[i], data_ht); |
| 2868 | local_stream[i] = NULL; |
| 2869 | } else if (len > 0) { |
| 2870 | local_stream[i]->data_read = 1; |
| 2871 | } |
| 2872 | } |
| 2873 | } |
| 2874 | |
| 2875 | /* |
| 2876 | * If we read high prio channel in this loop, try again |
| 2877 | * for more high prio data. |
| 2878 | */ |
| 2879 | if (high_prio) { |
| 2880 | continue; |
| 2881 | } |
| 2882 | |
| 2883 | /* Take care of low priority channels. */ |
| 2884 | for (i = 0; i < nb_fd; i++) { |
| 2885 | health_code_update(); |
| 2886 | |
| 2887 | if (local_stream[i] == NULL) { |
| 2888 | continue; |
| 2889 | } |
| 2890 | if ((pollfd[i].revents & POLLIN) || |
| 2891 | local_stream[i]->hangup_flush_done || |
| 2892 | local_stream[i]->has_data) { |
| 2893 | DBG("Normal read on fd %d", pollfd[i].fd); |
| 2894 | len = ctx->on_buffer_ready(local_stream[i], ctx); |
| 2895 | /* it's ok to have an unavailable sub-buffer */ |
| 2896 | if (len < 0 && len != -EAGAIN && len != -ENODATA) { |
| 2897 | /* Clean the stream and free it. */ |
| 2898 | consumer_del_stream(local_stream[i], data_ht); |
| 2899 | local_stream[i] = NULL; |
| 2900 | } else if (len > 0) { |
| 2901 | local_stream[i]->data_read = 1; |
| 2902 | } |
| 2903 | } |
| 2904 | } |
| 2905 | |
| 2906 | /* Handle hangup and errors */ |
| 2907 | for (i = 0; i < nb_fd; i++) { |
| 2908 | health_code_update(); |
| 2909 | |
| 2910 | if (local_stream[i] == NULL) { |
| 2911 | continue; |
| 2912 | } |
| 2913 | if (!local_stream[i]->hangup_flush_done |
| 2914 | && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL)) |
| 2915 | && (consumer_data.type == LTTNG_CONSUMER32_UST |
| 2916 | || consumer_data.type == LTTNG_CONSUMER64_UST)) { |
| 2917 | DBG("fd %d is hup|err|nval. Attempting flush and read.", |
| 2918 | pollfd[i].fd); |
| 2919 | lttng_ustconsumer_on_stream_hangup(local_stream[i]); |
| 2920 | /* Attempt read again, for the data we just flushed. */ |
| 2921 | local_stream[i]->data_read = 1; |
| 2922 | } |
| 2923 | /* |
| 2924 | * If the poll flag is HUP/ERR/NVAL and we have |
| 2925 | * read no data in this pass, we can remove the |
| 2926 | * stream from its hash table. |
| 2927 | */ |
| 2928 | if ((pollfd[i].revents & POLLHUP)) { |
| 2929 | DBG("Polling fd %d tells it has hung up.", pollfd[i].fd); |
| 2930 | if (!local_stream[i]->data_read) { |
| 2931 | consumer_del_stream(local_stream[i], data_ht); |
| 2932 | local_stream[i] = NULL; |
| 2933 | num_hup++; |
| 2934 | } |
| 2935 | } else if (pollfd[i].revents & POLLERR) { |
| 2936 | ERR("Error returned in polling fd %d.", pollfd[i].fd); |
| 2937 | if (!local_stream[i]->data_read) { |
| 2938 | consumer_del_stream(local_stream[i], data_ht); |
| 2939 | local_stream[i] = NULL; |
| 2940 | num_hup++; |
| 2941 | } |
| 2942 | } else if (pollfd[i].revents & POLLNVAL) { |
| 2943 | ERR("Polling fd %d tells fd is not open.", pollfd[i].fd); |
| 2944 | if (!local_stream[i]->data_read) { |
| 2945 | consumer_del_stream(local_stream[i], data_ht); |
| 2946 | local_stream[i] = NULL; |
| 2947 | num_hup++; |
| 2948 | } |
| 2949 | } |
| 2950 | if (local_stream[i] != NULL) { |
| 2951 | local_stream[i]->data_read = 0; |
| 2952 | } |
| 2953 | } |
| 2954 | } |
| 2955 | /* All is OK */ |
| 2956 | err = 0; |
| 2957 | end: |
| 2958 | DBG("polling thread exiting"); |
| 2959 | free(pollfd); |
| 2960 | free(local_stream); |
| 2961 | |
| 2962 | /* |
| 2963 | * Close the write side of the pipe so epoll_wait() in |
| 2964 | * consumer_thread_metadata_poll can catch it. The thread is monitoring the |
| 2965 | * read side of the pipe. If we close them both, epoll_wait strangely does |
| 2966 | * not return and could create a endless wait period if the pipe is the |
| 2967 | * only tracked fd in the poll set. The thread will take care of closing |
| 2968 | * the read side. |
| 2969 | */ |
| 2970 | (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe); |
| 2971 | |
| 2972 | error_testpoint: |
| 2973 | if (err) { |
| 2974 | health_error(); |
| 2975 | ERR("Health error occurred in %s", __func__); |
| 2976 | } |
| 2977 | health_unregister(health_consumerd); |
| 2978 | |
| 2979 | rcu_unregister_thread(); |
| 2980 | return NULL; |
| 2981 | } |
| 2982 | |
| 2983 | /* |
| 2984 | * Close wake-up end of each stream belonging to the channel. This will |
| 2985 | * allow the poll() on the stream read-side to detect when the |
| 2986 | * write-side (application) finally closes them. |
| 2987 | */ |
| 2988 | static |
| 2989 | void consumer_close_channel_streams(struct lttng_consumer_channel *channel) |
| 2990 | { |
| 2991 | struct lttng_ht *ht; |
| 2992 | struct lttng_consumer_stream *stream; |
| 2993 | struct lttng_ht_iter iter; |
| 2994 | |
| 2995 | ht = consumer_data.stream_per_chan_id_ht; |
| 2996 | |
| 2997 | rcu_read_lock(); |
| 2998 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 2999 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 3000 | ht->match_fct, &channel->key, |
| 3001 | &iter.iter, stream, node_channel_id.node) { |
| 3002 | /* |
| 3003 | * Protect against teardown with mutex. |
| 3004 | */ |
| 3005 | pthread_mutex_lock(&stream->lock); |
| 3006 | if (cds_lfht_is_node_deleted(&stream->node.node)) { |
| 3007 | goto next; |
| 3008 | } |
| 3009 | switch (consumer_data.type) { |
| 3010 | case LTTNG_CONSUMER_KERNEL: |
| 3011 | break; |
| 3012 | case LTTNG_CONSUMER32_UST: |
| 3013 | case LTTNG_CONSUMER64_UST: |
| 3014 | if (stream->metadata_flag) { |
| 3015 | /* Safe and protected by the stream lock. */ |
| 3016 | lttng_ustconsumer_close_metadata(stream->chan); |
| 3017 | } else { |
| 3018 | /* |
| 3019 | * Note: a mutex is taken internally within |
| 3020 | * liblttng-ust-ctl to protect timer wakeup_fd |
| 3021 | * use from concurrent close. |
| 3022 | */ |
| 3023 | lttng_ustconsumer_close_stream_wakeup(stream); |
| 3024 | } |
| 3025 | break; |
| 3026 | default: |
| 3027 | ERR("Unknown consumer_data type"); |
| 3028 | assert(0); |
| 3029 | } |
| 3030 | next: |
| 3031 | pthread_mutex_unlock(&stream->lock); |
| 3032 | } |
| 3033 | rcu_read_unlock(); |
| 3034 | } |
| 3035 | |
| 3036 | static void destroy_channel_ht(struct lttng_ht *ht) |
| 3037 | { |
| 3038 | struct lttng_ht_iter iter; |
| 3039 | struct lttng_consumer_channel *channel; |
| 3040 | int ret; |
| 3041 | |
| 3042 | if (ht == NULL) { |
| 3043 | return; |
| 3044 | } |
| 3045 | |
| 3046 | rcu_read_lock(); |
| 3047 | cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) { |
| 3048 | ret = lttng_ht_del(ht, &iter); |
| 3049 | assert(ret != 0); |
| 3050 | } |
| 3051 | rcu_read_unlock(); |
| 3052 | |
| 3053 | lttng_ht_destroy(ht); |
| 3054 | } |
| 3055 | |
| 3056 | /* |
| 3057 | * This thread polls the channel fds to detect when they are being |
| 3058 | * closed. It closes all related streams if the channel is detected as |
| 3059 | * closed. It is currently only used as a shim layer for UST because the |
| 3060 | * consumerd needs to keep the per-stream wakeup end of pipes open for |
| 3061 | * periodical flush. |
| 3062 | */ |
| 3063 | void *consumer_thread_channel_poll(void *data) |
| 3064 | { |
| 3065 | int ret, i, pollfd, err = -1; |
| 3066 | uint32_t revents, nb_fd; |
| 3067 | struct lttng_consumer_channel *chan = NULL; |
| 3068 | struct lttng_ht_iter iter; |
| 3069 | struct lttng_ht_node_u64 *node; |
| 3070 | struct lttng_poll_event events; |
| 3071 | struct lttng_consumer_local_data *ctx = data; |
| 3072 | struct lttng_ht *channel_ht; |
| 3073 | |
| 3074 | rcu_register_thread(); |
| 3075 | |
| 3076 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL); |
| 3077 | |
| 3078 | if (testpoint(consumerd_thread_channel)) { |
| 3079 | goto error_testpoint; |
| 3080 | } |
| 3081 | |
| 3082 | health_code_update(); |
| 3083 | |
| 3084 | channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3085 | if (!channel_ht) { |
| 3086 | /* ENOMEM at this point. Better to bail out. */ |
| 3087 | goto end_ht; |
| 3088 | } |
| 3089 | |
| 3090 | DBG("Thread channel poll started"); |
| 3091 | |
| 3092 | /* Size is set to 1 for the consumer_channel pipe */ |
| 3093 | ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC); |
| 3094 | if (ret < 0) { |
| 3095 | ERR("Poll set creation failed"); |
| 3096 | goto end_poll; |
| 3097 | } |
| 3098 | |
| 3099 | ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN); |
| 3100 | if (ret < 0) { |
| 3101 | goto end; |
| 3102 | } |
| 3103 | |
| 3104 | /* Main loop */ |
| 3105 | DBG("Channel main loop started"); |
| 3106 | |
| 3107 | while (1) { |
| 3108 | restart: |
| 3109 | health_code_update(); |
| 3110 | DBG("Channel poll wait"); |
| 3111 | health_poll_entry(); |
| 3112 | ret = lttng_poll_wait(&events, -1); |
| 3113 | DBG("Channel poll return from wait with %d fd(s)", |
| 3114 | LTTNG_POLL_GETNB(&events)); |
| 3115 | health_poll_exit(); |
| 3116 | DBG("Channel event caught in thread"); |
| 3117 | if (ret < 0) { |
| 3118 | if (errno == EINTR) { |
| 3119 | ERR("Poll EINTR caught"); |
| 3120 | goto restart; |
| 3121 | } |
| 3122 | if (LTTNG_POLL_GETNB(&events) == 0) { |
| 3123 | err = 0; /* All is OK */ |
| 3124 | } |
| 3125 | goto end; |
| 3126 | } |
| 3127 | |
| 3128 | nb_fd = ret; |
| 3129 | |
| 3130 | /* From here, the event is a channel wait fd */ |
| 3131 | for (i = 0; i < nb_fd; i++) { |
| 3132 | health_code_update(); |
| 3133 | |
| 3134 | revents = LTTNG_POLL_GETEV(&events, i); |
| 3135 | pollfd = LTTNG_POLL_GETFD(&events, i); |
| 3136 | |
| 3137 | if (pollfd == ctx->consumer_channel_pipe[0]) { |
| 3138 | if (revents & LPOLLIN) { |
| 3139 | enum consumer_channel_action action; |
| 3140 | uint64_t key; |
| 3141 | |
| 3142 | ret = read_channel_pipe(ctx, &chan, &key, &action); |
| 3143 | if (ret <= 0) { |
| 3144 | if (ret < 0) { |
| 3145 | ERR("Error reading channel pipe"); |
| 3146 | } |
| 3147 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); |
| 3148 | continue; |
| 3149 | } |
| 3150 | |
| 3151 | switch (action) { |
| 3152 | case CONSUMER_CHANNEL_ADD: |
| 3153 | DBG("Adding channel %d to poll set", |
| 3154 | chan->wait_fd); |
| 3155 | |
| 3156 | lttng_ht_node_init_u64(&chan->wait_fd_node, |
| 3157 | chan->wait_fd); |
| 3158 | rcu_read_lock(); |
| 3159 | lttng_ht_add_unique_u64(channel_ht, |
| 3160 | &chan->wait_fd_node); |
| 3161 | rcu_read_unlock(); |
| 3162 | /* Add channel to the global poll events list */ |
| 3163 | lttng_poll_add(&events, chan->wait_fd, |
| 3164 | LPOLLERR | LPOLLHUP); |
| 3165 | break; |
| 3166 | case CONSUMER_CHANNEL_DEL: |
| 3167 | { |
| 3168 | /* |
| 3169 | * This command should never be called if the channel |
| 3170 | * has streams monitored by either the data or metadata |
| 3171 | * thread. The consumer only notify this thread with a |
| 3172 | * channel del. command if it receives a destroy |
| 3173 | * channel command from the session daemon that send it |
| 3174 | * if a command prior to the GET_CHANNEL failed. |
| 3175 | */ |
| 3176 | |
| 3177 | rcu_read_lock(); |
| 3178 | chan = consumer_find_channel(key); |
| 3179 | if (!chan) { |
| 3180 | rcu_read_unlock(); |
| 3181 | ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key); |
| 3182 | break; |
| 3183 | } |
| 3184 | lttng_poll_del(&events, chan->wait_fd); |
| 3185 | iter.iter.node = &chan->wait_fd_node.node; |
| 3186 | ret = lttng_ht_del(channel_ht, &iter); |
| 3187 | assert(ret == 0); |
| 3188 | |
| 3189 | switch (consumer_data.type) { |
| 3190 | case LTTNG_CONSUMER_KERNEL: |
| 3191 | break; |
| 3192 | case LTTNG_CONSUMER32_UST: |
| 3193 | case LTTNG_CONSUMER64_UST: |
| 3194 | health_code_update(); |
| 3195 | /* Destroy streams that might have been left in the stream list. */ |
| 3196 | clean_channel_stream_list(chan); |
| 3197 | break; |
| 3198 | default: |
| 3199 | ERR("Unknown consumer_data type"); |
| 3200 | assert(0); |
| 3201 | } |
| 3202 | |
| 3203 | /* |
| 3204 | * Release our own refcount. Force channel deletion even if |
| 3205 | * streams were not initialized. |
| 3206 | */ |
| 3207 | if (!uatomic_sub_return(&chan->refcount, 1)) { |
| 3208 | consumer_del_channel(chan); |
| 3209 | } |
| 3210 | rcu_read_unlock(); |
| 3211 | goto restart; |
| 3212 | } |
| 3213 | case CONSUMER_CHANNEL_QUIT: |
| 3214 | /* |
| 3215 | * Remove the pipe from the poll set and continue the loop |
| 3216 | * since their might be data to consume. |
| 3217 | */ |
| 3218 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); |
| 3219 | continue; |
| 3220 | default: |
| 3221 | ERR("Unknown action"); |
| 3222 | break; |
| 3223 | } |
| 3224 | } else if (revents & (LPOLLERR | LPOLLHUP)) { |
| 3225 | DBG("Channel thread pipe hung up"); |
| 3226 | /* |
| 3227 | * Remove the pipe from the poll set and continue the loop |
| 3228 | * since their might be data to consume. |
| 3229 | */ |
| 3230 | lttng_poll_del(&events, ctx->consumer_channel_pipe[0]); |
| 3231 | continue; |
| 3232 | } else { |
| 3233 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); |
| 3234 | goto end; |
| 3235 | } |
| 3236 | |
| 3237 | /* Handle other stream */ |
| 3238 | continue; |
| 3239 | } |
| 3240 | |
| 3241 | rcu_read_lock(); |
| 3242 | { |
| 3243 | uint64_t tmp_id = (uint64_t) pollfd; |
| 3244 | |
| 3245 | lttng_ht_lookup(channel_ht, &tmp_id, &iter); |
| 3246 | } |
| 3247 | node = lttng_ht_iter_get_node_u64(&iter); |
| 3248 | assert(node); |
| 3249 | |
| 3250 | chan = caa_container_of(node, struct lttng_consumer_channel, |
| 3251 | wait_fd_node); |
| 3252 | |
| 3253 | /* Check for error event */ |
| 3254 | if (revents & (LPOLLERR | LPOLLHUP)) { |
| 3255 | DBG("Channel fd %d is hup|err.", pollfd); |
| 3256 | |
| 3257 | lttng_poll_del(&events, chan->wait_fd); |
| 3258 | ret = lttng_ht_del(channel_ht, &iter); |
| 3259 | assert(ret == 0); |
| 3260 | |
| 3261 | /* |
| 3262 | * This will close the wait fd for each stream associated to |
| 3263 | * this channel AND monitored by the data/metadata thread thus |
| 3264 | * will be clean by the right thread. |
| 3265 | */ |
| 3266 | consumer_close_channel_streams(chan); |
| 3267 | |
| 3268 | /* Release our own refcount */ |
| 3269 | if (!uatomic_sub_return(&chan->refcount, 1) |
| 3270 | && !uatomic_read(&chan->nb_init_stream_left)) { |
| 3271 | consumer_del_channel(chan); |
| 3272 | } |
| 3273 | } else { |
| 3274 | ERR("Unexpected poll events %u for sock %d", revents, pollfd); |
| 3275 | rcu_read_unlock(); |
| 3276 | goto end; |
| 3277 | } |
| 3278 | |
| 3279 | /* Release RCU lock for the channel looked up */ |
| 3280 | rcu_read_unlock(); |
| 3281 | } |
| 3282 | } |
| 3283 | |
| 3284 | /* All is OK */ |
| 3285 | err = 0; |
| 3286 | end: |
| 3287 | lttng_poll_clean(&events); |
| 3288 | end_poll: |
| 3289 | destroy_channel_ht(channel_ht); |
| 3290 | end_ht: |
| 3291 | error_testpoint: |
| 3292 | DBG("Channel poll thread exiting"); |
| 3293 | if (err) { |
| 3294 | health_error(); |
| 3295 | ERR("Health error occurred in %s", __func__); |
| 3296 | } |
| 3297 | health_unregister(health_consumerd); |
| 3298 | rcu_unregister_thread(); |
| 3299 | return NULL; |
| 3300 | } |
| 3301 | |
| 3302 | static int set_metadata_socket(struct lttng_consumer_local_data *ctx, |
| 3303 | struct pollfd *sockpoll, int client_socket) |
| 3304 | { |
| 3305 | int ret; |
| 3306 | |
| 3307 | assert(ctx); |
| 3308 | assert(sockpoll); |
| 3309 | |
| 3310 | ret = lttng_consumer_poll_socket(sockpoll); |
| 3311 | if (ret) { |
| 3312 | goto error; |
| 3313 | } |
| 3314 | DBG("Metadata connection on client_socket"); |
| 3315 | |
| 3316 | /* Blocking call, waiting for transmission */ |
| 3317 | ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket); |
| 3318 | if (ctx->consumer_metadata_socket < 0) { |
| 3319 | WARN("On accept metadata"); |
| 3320 | ret = -1; |
| 3321 | goto error; |
| 3322 | } |
| 3323 | ret = 0; |
| 3324 | |
| 3325 | error: |
| 3326 | return ret; |
| 3327 | } |
| 3328 | |
| 3329 | /* |
| 3330 | * This thread listens on the consumerd socket and receives the file |
| 3331 | * descriptors from the session daemon. |
| 3332 | */ |
| 3333 | void *consumer_thread_sessiond_poll(void *data) |
| 3334 | { |
| 3335 | int sock = -1, client_socket, ret, err = -1; |
| 3336 | /* |
| 3337 | * structure to poll for incoming data on communication socket avoids |
| 3338 | * making blocking sockets. |
| 3339 | */ |
| 3340 | struct pollfd consumer_sockpoll[2]; |
| 3341 | struct lttng_consumer_local_data *ctx = data; |
| 3342 | |
| 3343 | rcu_register_thread(); |
| 3344 | |
| 3345 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND); |
| 3346 | |
| 3347 | if (testpoint(consumerd_thread_sessiond)) { |
| 3348 | goto error_testpoint; |
| 3349 | } |
| 3350 | |
| 3351 | health_code_update(); |
| 3352 | |
| 3353 | DBG("Creating command socket %s", ctx->consumer_command_sock_path); |
| 3354 | unlink(ctx->consumer_command_sock_path); |
| 3355 | client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path); |
| 3356 | if (client_socket < 0) { |
| 3357 | ERR("Cannot create command socket"); |
| 3358 | goto end; |
| 3359 | } |
| 3360 | |
| 3361 | ret = lttcomm_listen_unix_sock(client_socket); |
| 3362 | if (ret < 0) { |
| 3363 | goto end; |
| 3364 | } |
| 3365 | |
| 3366 | DBG("Sending ready command to lttng-sessiond"); |
| 3367 | ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY); |
| 3368 | /* return < 0 on error, but == 0 is not fatal */ |
| 3369 | if (ret < 0) { |
| 3370 | ERR("Error sending ready command to lttng-sessiond"); |
| 3371 | goto end; |
| 3372 | } |
| 3373 | |
| 3374 | /* prepare the FDs to poll : to client socket and the should_quit pipe */ |
| 3375 | consumer_sockpoll[0].fd = ctx->consumer_should_quit[0]; |
| 3376 | consumer_sockpoll[0].events = POLLIN | POLLPRI; |
| 3377 | consumer_sockpoll[1].fd = client_socket; |
| 3378 | consumer_sockpoll[1].events = POLLIN | POLLPRI; |
| 3379 | |
| 3380 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
| 3381 | if (ret) { |
| 3382 | if (ret > 0) { |
| 3383 | /* should exit */ |
| 3384 | err = 0; |
| 3385 | } |
| 3386 | goto end; |
| 3387 | } |
| 3388 | DBG("Connection on client_socket"); |
| 3389 | |
| 3390 | /* Blocking call, waiting for transmission */ |
| 3391 | sock = lttcomm_accept_unix_sock(client_socket); |
| 3392 | if (sock < 0) { |
| 3393 | WARN("On accept"); |
| 3394 | goto end; |
| 3395 | } |
| 3396 | |
| 3397 | /* |
| 3398 | * Setup metadata socket which is the second socket connection on the |
| 3399 | * command unix socket. |
| 3400 | */ |
| 3401 | ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket); |
| 3402 | if (ret) { |
| 3403 | if (ret > 0) { |
| 3404 | /* should exit */ |
| 3405 | err = 0; |
| 3406 | } |
| 3407 | goto end; |
| 3408 | } |
| 3409 | |
| 3410 | /* This socket is not useful anymore. */ |
| 3411 | ret = close(client_socket); |
| 3412 | if (ret < 0) { |
| 3413 | PERROR("close client_socket"); |
| 3414 | } |
| 3415 | client_socket = -1; |
| 3416 | |
| 3417 | /* update the polling structure to poll on the established socket */ |
| 3418 | consumer_sockpoll[1].fd = sock; |
| 3419 | consumer_sockpoll[1].events = POLLIN | POLLPRI; |
| 3420 | |
| 3421 | while (1) { |
| 3422 | health_code_update(); |
| 3423 | |
| 3424 | health_poll_entry(); |
| 3425 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
| 3426 | health_poll_exit(); |
| 3427 | if (ret) { |
| 3428 | if (ret > 0) { |
| 3429 | /* should exit */ |
| 3430 | err = 0; |
| 3431 | } |
| 3432 | goto end; |
| 3433 | } |
| 3434 | DBG("Incoming command on sock"); |
| 3435 | ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll); |
| 3436 | if (ret <= 0) { |
| 3437 | /* |
| 3438 | * This could simply be a session daemon quitting. Don't output |
| 3439 | * ERR() here. |
| 3440 | */ |
| 3441 | DBG("Communication interrupted on command socket"); |
| 3442 | err = 0; |
| 3443 | goto end; |
| 3444 | } |
| 3445 | if (CMM_LOAD_SHARED(consumer_quit)) { |
| 3446 | DBG("consumer_thread_receive_fds received quit from signal"); |
| 3447 | err = 0; /* All is OK */ |
| 3448 | goto end; |
| 3449 | } |
| 3450 | DBG("received command on sock"); |
| 3451 | } |
| 3452 | /* All is OK */ |
| 3453 | err = 0; |
| 3454 | |
| 3455 | end: |
| 3456 | DBG("Consumer thread sessiond poll exiting"); |
| 3457 | |
| 3458 | /* |
| 3459 | * Close metadata streams since the producer is the session daemon which |
| 3460 | * just died. |
| 3461 | * |
| 3462 | * NOTE: for now, this only applies to the UST tracer. |
| 3463 | */ |
| 3464 | lttng_consumer_close_all_metadata(); |
| 3465 | |
| 3466 | /* |
| 3467 | * when all fds have hung up, the polling thread |
| 3468 | * can exit cleanly |
| 3469 | */ |
| 3470 | CMM_STORE_SHARED(consumer_quit, 1); |
| 3471 | |
| 3472 | /* |
| 3473 | * Notify the data poll thread to poll back again and test the |
| 3474 | * consumer_quit state that we just set so to quit gracefully. |
| 3475 | */ |
| 3476 | notify_thread_lttng_pipe(ctx->consumer_data_pipe); |
| 3477 | |
| 3478 | notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT); |
| 3479 | |
| 3480 | notify_health_quit_pipe(health_quit_pipe); |
| 3481 | |
| 3482 | /* Cleaning up possibly open sockets. */ |
| 3483 | if (sock >= 0) { |
| 3484 | ret = close(sock); |
| 3485 | if (ret < 0) { |
| 3486 | PERROR("close sock sessiond poll"); |
| 3487 | } |
| 3488 | } |
| 3489 | if (client_socket >= 0) { |
| 3490 | ret = close(client_socket); |
| 3491 | if (ret < 0) { |
| 3492 | PERROR("close client_socket sessiond poll"); |
| 3493 | } |
| 3494 | } |
| 3495 | |
| 3496 | error_testpoint: |
| 3497 | if (err) { |
| 3498 | health_error(); |
| 3499 | ERR("Health error occurred in %s", __func__); |
| 3500 | } |
| 3501 | health_unregister(health_consumerd); |
| 3502 | |
| 3503 | rcu_unregister_thread(); |
| 3504 | return NULL; |
| 3505 | } |
| 3506 | |
| 3507 | ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream, |
| 3508 | struct lttng_consumer_local_data *ctx) |
| 3509 | { |
| 3510 | ssize_t ret; |
| 3511 | |
| 3512 | pthread_mutex_lock(&stream->chan->lock); |
| 3513 | pthread_mutex_lock(&stream->lock); |
| 3514 | if (stream->metadata_flag) { |
| 3515 | pthread_mutex_lock(&stream->metadata_rdv_lock); |
| 3516 | } |
| 3517 | |
| 3518 | switch (consumer_data.type) { |
| 3519 | case LTTNG_CONSUMER_KERNEL: |
| 3520 | ret = lttng_kconsumer_read_subbuffer(stream, ctx); |
| 3521 | break; |
| 3522 | case LTTNG_CONSUMER32_UST: |
| 3523 | case LTTNG_CONSUMER64_UST: |
| 3524 | ret = lttng_ustconsumer_read_subbuffer(stream, ctx); |
| 3525 | break; |
| 3526 | default: |
| 3527 | ERR("Unknown consumer_data type"); |
| 3528 | assert(0); |
| 3529 | ret = -ENOSYS; |
| 3530 | break; |
| 3531 | } |
| 3532 | |
| 3533 | if (stream->metadata_flag) { |
| 3534 | pthread_cond_broadcast(&stream->metadata_rdv); |
| 3535 | pthread_mutex_unlock(&stream->metadata_rdv_lock); |
| 3536 | } |
| 3537 | pthread_mutex_unlock(&stream->lock); |
| 3538 | pthread_mutex_unlock(&stream->chan->lock); |
| 3539 | |
| 3540 | return ret; |
| 3541 | } |
| 3542 | |
| 3543 | int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream) |
| 3544 | { |
| 3545 | switch (consumer_data.type) { |
| 3546 | case LTTNG_CONSUMER_KERNEL: |
| 3547 | return lttng_kconsumer_on_recv_stream(stream); |
| 3548 | case LTTNG_CONSUMER32_UST: |
| 3549 | case LTTNG_CONSUMER64_UST: |
| 3550 | return lttng_ustconsumer_on_recv_stream(stream); |
| 3551 | default: |
| 3552 | ERR("Unknown consumer_data type"); |
| 3553 | assert(0); |
| 3554 | return -ENOSYS; |
| 3555 | } |
| 3556 | } |
| 3557 | |
| 3558 | /* |
| 3559 | * Allocate and set consumer data hash tables. |
| 3560 | */ |
| 3561 | int lttng_consumer_init(void) |
| 3562 | { |
| 3563 | consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3564 | if (!consumer_data.channel_ht) { |
| 3565 | goto error; |
| 3566 | } |
| 3567 | |
| 3568 | consumer_data.channels_by_session_id_ht = |
| 3569 | lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3570 | if (!consumer_data.channels_by_session_id_ht) { |
| 3571 | goto error; |
| 3572 | } |
| 3573 | |
| 3574 | consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3575 | if (!consumer_data.relayd_ht) { |
| 3576 | goto error; |
| 3577 | } |
| 3578 | |
| 3579 | consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3580 | if (!consumer_data.stream_list_ht) { |
| 3581 | goto error; |
| 3582 | } |
| 3583 | |
| 3584 | consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3585 | if (!consumer_data.stream_per_chan_id_ht) { |
| 3586 | goto error; |
| 3587 | } |
| 3588 | |
| 3589 | data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3590 | if (!data_ht) { |
| 3591 | goto error; |
| 3592 | } |
| 3593 | |
| 3594 | metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
| 3595 | if (!metadata_ht) { |
| 3596 | goto error; |
| 3597 | } |
| 3598 | |
| 3599 | consumer_data.chunk_registry = lttng_trace_chunk_registry_create(); |
| 3600 | if (!consumer_data.chunk_registry) { |
| 3601 | goto error; |
| 3602 | } |
| 3603 | |
| 3604 | return 0; |
| 3605 | |
| 3606 | error: |
| 3607 | return -1; |
| 3608 | } |
| 3609 | |
| 3610 | /* |
| 3611 | * Process the ADD_RELAYD command receive by a consumer. |
| 3612 | * |
| 3613 | * This will create a relayd socket pair and add it to the relayd hash table. |
| 3614 | * The caller MUST acquire a RCU read side lock before calling it. |
| 3615 | */ |
| 3616 | void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type, |
| 3617 | struct lttng_consumer_local_data *ctx, int sock, |
| 3618 | struct pollfd *consumer_sockpoll, |
| 3619 | struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id, |
| 3620 | uint64_t relayd_session_id) |
| 3621 | { |
| 3622 | int fd = -1, ret = -1, relayd_created = 0; |
| 3623 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
| 3624 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 3625 | |
| 3626 | assert(ctx); |
| 3627 | assert(relayd_sock); |
| 3628 | |
| 3629 | DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx); |
| 3630 | |
| 3631 | /* Get relayd reference if exists. */ |
| 3632 | relayd = consumer_find_relayd(net_seq_idx); |
| 3633 | if (relayd == NULL) { |
| 3634 | assert(sock_type == LTTNG_STREAM_CONTROL); |
| 3635 | /* Not found. Allocate one. */ |
| 3636 | relayd = consumer_allocate_relayd_sock_pair(net_seq_idx); |
| 3637 | if (relayd == NULL) { |
| 3638 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
| 3639 | goto error; |
| 3640 | } else { |
| 3641 | relayd->sessiond_session_id = sessiond_id; |
| 3642 | relayd_created = 1; |
| 3643 | } |
| 3644 | |
| 3645 | /* |
| 3646 | * This code path MUST continue to the consumer send status message to |
| 3647 | * we can notify the session daemon and continue our work without |
| 3648 | * killing everything. |
| 3649 | */ |
| 3650 | } else { |
| 3651 | /* |
| 3652 | * relayd key should never be found for control socket. |
| 3653 | */ |
| 3654 | assert(sock_type != LTTNG_STREAM_CONTROL); |
| 3655 | } |
| 3656 | |
| 3657 | /* First send a status message before receiving the fds. */ |
| 3658 | ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS); |
| 3659 | if (ret < 0) { |
| 3660 | /* Somehow, the session daemon is not responding anymore. */ |
| 3661 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
| 3662 | goto error_nosignal; |
| 3663 | } |
| 3664 | |
| 3665 | /* Poll on consumer socket. */ |
| 3666 | ret = lttng_consumer_poll_socket(consumer_sockpoll); |
| 3667 | if (ret) { |
| 3668 | /* Needing to exit in the middle of a command: error. */ |
| 3669 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR); |
| 3670 | goto error_nosignal; |
| 3671 | } |
| 3672 | |
| 3673 | /* Get relayd socket from session daemon */ |
| 3674 | ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1); |
| 3675 | if (ret != sizeof(fd)) { |
| 3676 | fd = -1; /* Just in case it gets set with an invalid value. */ |
| 3677 | |
| 3678 | /* |
| 3679 | * Failing to receive FDs might indicate a major problem such as |
| 3680 | * reaching a fd limit during the receive where the kernel returns a |
| 3681 | * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we |
| 3682 | * don't take any chances and stop everything. |
| 3683 | * |
| 3684 | * XXX: Feature request #558 will fix that and avoid this possible |
| 3685 | * issue when reaching the fd limit. |
| 3686 | */ |
| 3687 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD); |
| 3688 | ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD; |
| 3689 | goto error; |
| 3690 | } |
| 3691 | |
| 3692 | /* Copy socket information and received FD */ |
| 3693 | switch (sock_type) { |
| 3694 | case LTTNG_STREAM_CONTROL: |
| 3695 | /* Copy received lttcomm socket */ |
| 3696 | lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock); |
| 3697 | ret = lttcomm_create_sock(&relayd->control_sock.sock); |
| 3698 | /* Handle create_sock error. */ |
| 3699 | if (ret < 0) { |
| 3700 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
| 3701 | goto error; |
| 3702 | } |
| 3703 | /* |
| 3704 | * Close the socket created internally by |
| 3705 | * lttcomm_create_sock, so we can replace it by the one |
| 3706 | * received from sessiond. |
| 3707 | */ |
| 3708 | if (close(relayd->control_sock.sock.fd)) { |
| 3709 | PERROR("close"); |
| 3710 | } |
| 3711 | |
| 3712 | /* Assign new file descriptor */ |
| 3713 | relayd->control_sock.sock.fd = fd; |
| 3714 | /* Assign version values. */ |
| 3715 | relayd->control_sock.major = relayd_sock->major; |
| 3716 | relayd->control_sock.minor = relayd_sock->minor; |
| 3717 | |
| 3718 | relayd->relayd_session_id = relayd_session_id; |
| 3719 | |
| 3720 | break; |
| 3721 | case LTTNG_STREAM_DATA: |
| 3722 | /* Copy received lttcomm socket */ |
| 3723 | lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock); |
| 3724 | ret = lttcomm_create_sock(&relayd->data_sock.sock); |
| 3725 | /* Handle create_sock error. */ |
| 3726 | if (ret < 0) { |
| 3727 | ret_code = LTTCOMM_CONSUMERD_ENOMEM; |
| 3728 | goto error; |
| 3729 | } |
| 3730 | /* |
| 3731 | * Close the socket created internally by |
| 3732 | * lttcomm_create_sock, so we can replace it by the one |
| 3733 | * received from sessiond. |
| 3734 | */ |
| 3735 | if (close(relayd->data_sock.sock.fd)) { |
| 3736 | PERROR("close"); |
| 3737 | } |
| 3738 | |
| 3739 | /* Assign new file descriptor */ |
| 3740 | relayd->data_sock.sock.fd = fd; |
| 3741 | /* Assign version values. */ |
| 3742 | relayd->data_sock.major = relayd_sock->major; |
| 3743 | relayd->data_sock.minor = relayd_sock->minor; |
| 3744 | break; |
| 3745 | default: |
| 3746 | ERR("Unknown relayd socket type (%d)", sock_type); |
| 3747 | ret_code = LTTCOMM_CONSUMERD_FATAL; |
| 3748 | goto error; |
| 3749 | } |
| 3750 | |
| 3751 | DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)", |
| 3752 | sock_type == LTTNG_STREAM_CONTROL ? "control" : "data", |
| 3753 | relayd->net_seq_idx, fd); |
| 3754 | /* |
| 3755 | * We gave the ownership of the fd to the relayd structure. Set the |
| 3756 | * fd to -1 so we don't call close() on it in the error path below. |
| 3757 | */ |
| 3758 | fd = -1; |
| 3759 | |
| 3760 | /* We successfully added the socket. Send status back. */ |
| 3761 | ret = consumer_send_status_msg(sock, ret_code); |
| 3762 | if (ret < 0) { |
| 3763 | /* Somehow, the session daemon is not responding anymore. */ |
| 3764 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
| 3765 | goto error_nosignal; |
| 3766 | } |
| 3767 | |
| 3768 | /* |
| 3769 | * Add relayd socket pair to consumer data hashtable. If object already |
| 3770 | * exists or on error, the function gracefully returns. |
| 3771 | */ |
| 3772 | relayd->ctx = ctx; |
| 3773 | add_relayd(relayd); |
| 3774 | |
| 3775 | /* All good! */ |
| 3776 | return; |
| 3777 | |
| 3778 | error: |
| 3779 | if (consumer_send_status_msg(sock, ret_code) < 0) { |
| 3780 | lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL); |
| 3781 | } |
| 3782 | |
| 3783 | error_nosignal: |
| 3784 | /* Close received socket if valid. */ |
| 3785 | if (fd >= 0) { |
| 3786 | if (close(fd)) { |
| 3787 | PERROR("close received socket"); |
| 3788 | } |
| 3789 | } |
| 3790 | |
| 3791 | if (relayd_created) { |
| 3792 | free(relayd); |
| 3793 | } |
| 3794 | } |
| 3795 | |
| 3796 | /* |
| 3797 | * Search for a relayd associated to the session id and return the reference. |
| 3798 | * |
| 3799 | * A rcu read side lock MUST be acquire before calling this function and locked |
| 3800 | * until the relayd object is no longer necessary. |
| 3801 | */ |
| 3802 | static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id) |
| 3803 | { |
| 3804 | struct lttng_ht_iter iter; |
| 3805 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 3806 | |
| 3807 | /* Iterate over all relayd since they are indexed by net_seq_idx. */ |
| 3808 | cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd, |
| 3809 | node.node) { |
| 3810 | /* |
| 3811 | * Check by sessiond id which is unique here where the relayd session |
| 3812 | * id might not be when having multiple relayd. |
| 3813 | */ |
| 3814 | if (relayd->sessiond_session_id == id) { |
| 3815 | /* Found the relayd. There can be only one per id. */ |
| 3816 | goto found; |
| 3817 | } |
| 3818 | } |
| 3819 | |
| 3820 | return NULL; |
| 3821 | |
| 3822 | found: |
| 3823 | return relayd; |
| 3824 | } |
| 3825 | |
| 3826 | /* |
| 3827 | * Check if for a given session id there is still data needed to be extract |
| 3828 | * from the buffers. |
| 3829 | * |
| 3830 | * Return 1 if data is pending or else 0 meaning ready to be read. |
| 3831 | */ |
| 3832 | int consumer_data_pending(uint64_t id) |
| 3833 | { |
| 3834 | int ret; |
| 3835 | struct lttng_ht_iter iter; |
| 3836 | struct lttng_ht *ht; |
| 3837 | struct lttng_consumer_stream *stream; |
| 3838 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 3839 | int (*data_pending)(struct lttng_consumer_stream *); |
| 3840 | |
| 3841 | DBG("Consumer data pending command on session id %" PRIu64, id); |
| 3842 | |
| 3843 | rcu_read_lock(); |
| 3844 | pthread_mutex_lock(&consumer_data.lock); |
| 3845 | |
| 3846 | switch (consumer_data.type) { |
| 3847 | case LTTNG_CONSUMER_KERNEL: |
| 3848 | data_pending = lttng_kconsumer_data_pending; |
| 3849 | break; |
| 3850 | case LTTNG_CONSUMER32_UST: |
| 3851 | case LTTNG_CONSUMER64_UST: |
| 3852 | data_pending = lttng_ustconsumer_data_pending; |
| 3853 | break; |
| 3854 | default: |
| 3855 | ERR("Unknown consumer data type"); |
| 3856 | assert(0); |
| 3857 | } |
| 3858 | |
| 3859 | /* Ease our life a bit */ |
| 3860 | ht = consumer_data.stream_list_ht; |
| 3861 | |
| 3862 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 3863 | ht->hash_fct(&id, lttng_ht_seed), |
| 3864 | ht->match_fct, &id, |
| 3865 | &iter.iter, stream, node_session_id.node) { |
| 3866 | pthread_mutex_lock(&stream->lock); |
| 3867 | |
| 3868 | /* |
| 3869 | * A removed node from the hash table indicates that the stream has |
| 3870 | * been deleted thus having a guarantee that the buffers are closed |
| 3871 | * on the consumer side. However, data can still be transmitted |
| 3872 | * over the network so don't skip the relayd check. |
| 3873 | */ |
| 3874 | ret = cds_lfht_is_node_deleted(&stream->node.node); |
| 3875 | if (!ret) { |
| 3876 | /* Check the stream if there is data in the buffers. */ |
| 3877 | ret = data_pending(stream); |
| 3878 | if (ret == 1) { |
| 3879 | pthread_mutex_unlock(&stream->lock); |
| 3880 | goto data_pending; |
| 3881 | } |
| 3882 | } |
| 3883 | |
| 3884 | pthread_mutex_unlock(&stream->lock); |
| 3885 | } |
| 3886 | |
| 3887 | relayd = find_relayd_by_session_id(id); |
| 3888 | if (relayd) { |
| 3889 | unsigned int is_data_inflight = 0; |
| 3890 | |
| 3891 | /* Send init command for data pending. */ |
| 3892 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 3893 | ret = relayd_begin_data_pending(&relayd->control_sock, |
| 3894 | relayd->relayd_session_id); |
| 3895 | if (ret < 0) { |
| 3896 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 3897 | /* Communication error thus the relayd so no data pending. */ |
| 3898 | goto data_not_pending; |
| 3899 | } |
| 3900 | |
| 3901 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 3902 | ht->hash_fct(&id, lttng_ht_seed), |
| 3903 | ht->match_fct, &id, |
| 3904 | &iter.iter, stream, node_session_id.node) { |
| 3905 | if (stream->metadata_flag) { |
| 3906 | ret = relayd_quiescent_control(&relayd->control_sock, |
| 3907 | stream->relayd_stream_id); |
| 3908 | } else { |
| 3909 | ret = relayd_data_pending(&relayd->control_sock, |
| 3910 | stream->relayd_stream_id, |
| 3911 | stream->next_net_seq_num - 1); |
| 3912 | } |
| 3913 | |
| 3914 | if (ret == 1) { |
| 3915 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 3916 | goto data_pending; |
| 3917 | } else if (ret < 0) { |
| 3918 | ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 3919 | lttng_consumer_cleanup_relayd(relayd); |
| 3920 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 3921 | goto data_not_pending; |
| 3922 | } |
| 3923 | } |
| 3924 | |
| 3925 | /* Send end command for data pending. */ |
| 3926 | ret = relayd_end_data_pending(&relayd->control_sock, |
| 3927 | relayd->relayd_session_id, &is_data_inflight); |
| 3928 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 3929 | if (ret < 0) { |
| 3930 | ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx); |
| 3931 | lttng_consumer_cleanup_relayd(relayd); |
| 3932 | goto data_not_pending; |
| 3933 | } |
| 3934 | if (is_data_inflight) { |
| 3935 | goto data_pending; |
| 3936 | } |
| 3937 | } |
| 3938 | |
| 3939 | /* |
| 3940 | * Finding _no_ node in the hash table and no inflight data means that the |
| 3941 | * stream(s) have been removed thus data is guaranteed to be available for |
| 3942 | * analysis from the trace files. |
| 3943 | */ |
| 3944 | |
| 3945 | data_not_pending: |
| 3946 | /* Data is available to be read by a viewer. */ |
| 3947 | pthread_mutex_unlock(&consumer_data.lock); |
| 3948 | rcu_read_unlock(); |
| 3949 | return 0; |
| 3950 | |
| 3951 | data_pending: |
| 3952 | /* Data is still being extracted from buffers. */ |
| 3953 | pthread_mutex_unlock(&consumer_data.lock); |
| 3954 | rcu_read_unlock(); |
| 3955 | return 1; |
| 3956 | } |
| 3957 | |
| 3958 | /* |
| 3959 | * Send a ret code status message to the sessiond daemon. |
| 3960 | * |
| 3961 | * Return the sendmsg() return value. |
| 3962 | */ |
| 3963 | int consumer_send_status_msg(int sock, int ret_code) |
| 3964 | { |
| 3965 | struct lttcomm_consumer_status_msg msg; |
| 3966 | |
| 3967 | memset(&msg, 0, sizeof(msg)); |
| 3968 | msg.ret_code = ret_code; |
| 3969 | |
| 3970 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); |
| 3971 | } |
| 3972 | |
| 3973 | /* |
| 3974 | * Send a channel status message to the sessiond daemon. |
| 3975 | * |
| 3976 | * Return the sendmsg() return value. |
| 3977 | */ |
| 3978 | int consumer_send_status_channel(int sock, |
| 3979 | struct lttng_consumer_channel *channel) |
| 3980 | { |
| 3981 | struct lttcomm_consumer_status_channel msg; |
| 3982 | |
| 3983 | assert(sock >= 0); |
| 3984 | |
| 3985 | memset(&msg, 0, sizeof(msg)); |
| 3986 | if (!channel) { |
| 3987 | msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL; |
| 3988 | } else { |
| 3989 | msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
| 3990 | msg.key = channel->key; |
| 3991 | msg.stream_count = channel->streams.count; |
| 3992 | } |
| 3993 | |
| 3994 | return lttcomm_send_unix_sock(sock, &msg, sizeof(msg)); |
| 3995 | } |
| 3996 | |
| 3997 | unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos, |
| 3998 | unsigned long produced_pos, uint64_t nb_packets_per_stream, |
| 3999 | uint64_t max_sb_size) |
| 4000 | { |
| 4001 | unsigned long start_pos; |
| 4002 | |
| 4003 | if (!nb_packets_per_stream) { |
| 4004 | return consumed_pos; /* Grab everything */ |
| 4005 | } |
| 4006 | start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size); |
| 4007 | start_pos -= max_sb_size * nb_packets_per_stream; |
| 4008 | if ((long) (start_pos - consumed_pos) < 0) { |
| 4009 | return consumed_pos; /* Grab everything */ |
| 4010 | } |
| 4011 | return start_pos; |
| 4012 | } |
| 4013 | |
| 4014 | static |
| 4015 | int consumer_flush_buffer(struct lttng_consumer_stream *stream, int producer_active) |
| 4016 | { |
| 4017 | int ret = 0; |
| 4018 | |
| 4019 | switch (consumer_data.type) { |
| 4020 | case LTTNG_CONSUMER_KERNEL: |
| 4021 | ret = kernctl_buffer_flush(stream->wait_fd); |
| 4022 | if (ret < 0) { |
| 4023 | ERR("Failed to flush kernel stream"); |
| 4024 | goto end; |
| 4025 | } |
| 4026 | break; |
| 4027 | case LTTNG_CONSUMER32_UST: |
| 4028 | case LTTNG_CONSUMER64_UST: |
| 4029 | lttng_ustctl_flush_buffer(stream, producer_active); |
| 4030 | break; |
| 4031 | default: |
| 4032 | ERR("Unknown consumer_data type"); |
| 4033 | abort(); |
| 4034 | } |
| 4035 | |
| 4036 | end: |
| 4037 | return ret; |
| 4038 | } |
| 4039 | |
| 4040 | /* |
| 4041 | * Sample the rotate position for all the streams of a channel. If a stream |
| 4042 | * is already at the rotate position (produced == consumed), we flag it as |
| 4043 | * ready for rotation. The rotation of ready streams occurs after we have |
| 4044 | * replied to the session daemon that we have finished sampling the positions. |
| 4045 | * Must be called with RCU read-side lock held to ensure existence of channel. |
| 4046 | * |
| 4047 | * Returns 0 on success, < 0 on error |
| 4048 | */ |
| 4049 | int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel, |
| 4050 | uint64_t key, uint64_t relayd_id, uint32_t metadata, |
| 4051 | struct lttng_consumer_local_data *ctx) |
| 4052 | { |
| 4053 | int ret; |
| 4054 | struct lttng_consumer_stream *stream; |
| 4055 | struct lttng_ht_iter iter; |
| 4056 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; |
| 4057 | struct lttng_dynamic_array stream_rotation_positions; |
| 4058 | uint64_t next_chunk_id, stream_count = 0; |
| 4059 | enum lttng_trace_chunk_status chunk_status; |
| 4060 | const bool is_local_trace = relayd_id == -1ULL; |
| 4061 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 4062 | bool rotating_to_new_chunk = true; |
| 4063 | |
| 4064 | DBG("Consumer sample rotate position for channel %" PRIu64, key); |
| 4065 | |
| 4066 | lttng_dynamic_array_init(&stream_rotation_positions, |
| 4067 | sizeof(struct relayd_stream_rotation_position), NULL); |
| 4068 | |
| 4069 | rcu_read_lock(); |
| 4070 | |
| 4071 | pthread_mutex_lock(&channel->lock); |
| 4072 | assert(channel->trace_chunk); |
| 4073 | chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk, |
| 4074 | &next_chunk_id); |
| 4075 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4076 | ret = -1; |
| 4077 | goto end_unlock_channel; |
| 4078 | } |
| 4079 | |
| 4080 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 4081 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 4082 | ht->match_fct, &channel->key, &iter.iter, |
| 4083 | stream, node_channel_id.node) { |
| 4084 | unsigned long consumed_pos; |
| 4085 | |
| 4086 | health_code_update(); |
| 4087 | |
| 4088 | /* |
| 4089 | * Lock stream because we are about to change its state. |
| 4090 | */ |
| 4091 | pthread_mutex_lock(&stream->lock); |
| 4092 | |
| 4093 | if (stream->trace_chunk == stream->chan->trace_chunk) { |
| 4094 | rotating_to_new_chunk = false; |
| 4095 | } |
| 4096 | |
| 4097 | ret = lttng_consumer_sample_snapshot_positions(stream); |
| 4098 | if (ret < 0) { |
| 4099 | ERR("Failed to sample snapshot position during channel rotation"); |
| 4100 | goto end_unlock_stream; |
| 4101 | } |
| 4102 | |
| 4103 | ret = lttng_consumer_get_produced_snapshot(stream, |
| 4104 | &stream->rotate_position); |
| 4105 | if (ret < 0) { |
| 4106 | ERR("Failed to sample produced position during channel rotation"); |
| 4107 | goto end_unlock_stream; |
| 4108 | } |
| 4109 | |
| 4110 | lttng_consumer_get_consumed_snapshot(stream, |
| 4111 | &consumed_pos); |
| 4112 | if (consumed_pos == stream->rotate_position) { |
| 4113 | stream->rotate_ready = true; |
| 4114 | } |
| 4115 | |
| 4116 | /* |
| 4117 | * Active flush; has no effect if the production position |
| 4118 | * is at a packet boundary. |
| 4119 | */ |
| 4120 | ret = consumer_flush_buffer(stream, 1); |
| 4121 | if (ret < 0) { |
| 4122 | ERR("Failed to flush stream %" PRIu64 " during channel rotation", |
| 4123 | stream->key); |
| 4124 | goto end_unlock_stream; |
| 4125 | } |
| 4126 | |
| 4127 | if (!is_local_trace) { |
| 4128 | /* |
| 4129 | * The relay daemon control protocol expects a rotation |
| 4130 | * position as "the sequence number of the first packet |
| 4131 | * _after_ the current trace chunk. |
| 4132 | * |
| 4133 | * At the moment when the positions of the buffers are |
| 4134 | * sampled, the production position does not necessarily |
| 4135 | * sit at a packet boundary. The 'active' flush |
| 4136 | * operation above will push the production position to |
| 4137 | * the next packet boundary _if_ it is not already |
| 4138 | * sitting at such a boundary. |
| 4139 | * |
| 4140 | * Assuming a current production position that is not |
| 4141 | * on the bound of a packet, the 'target' sequence |
| 4142 | * number is |
| 4143 | * (consumed_pos / subbuffer_size) + 1 |
| 4144 | * Note the '+ 1' to ensure the current packet is |
| 4145 | * part of the current trace chunk. |
| 4146 | * |
| 4147 | * However, if the production position is already at |
| 4148 | * a packet boundary, the '+ 1' is not necessary as the |
| 4149 | * last packet of the current chunk is already |
| 4150 | * 'complete'. |
| 4151 | */ |
| 4152 | const struct relayd_stream_rotation_position position = { |
| 4153 | .stream_id = stream->relayd_stream_id, |
| 4154 | .rotate_at_seq_num = (stream->rotate_position / stream->max_sb_size) + |
| 4155 | !!(stream->rotate_position % stream->max_sb_size), |
| 4156 | }; |
| 4157 | |
| 4158 | ret = lttng_dynamic_array_add_element( |
| 4159 | &stream_rotation_positions, |
| 4160 | &position); |
| 4161 | if (ret) { |
| 4162 | ERR("Failed to allocate stream rotation position"); |
| 4163 | goto end_unlock_stream; |
| 4164 | } |
| 4165 | stream_count++; |
| 4166 | } |
| 4167 | pthread_mutex_unlock(&stream->lock); |
| 4168 | } |
| 4169 | stream = NULL; |
| 4170 | pthread_mutex_unlock(&channel->lock); |
| 4171 | |
| 4172 | if (is_local_trace) { |
| 4173 | ret = 0; |
| 4174 | goto end; |
| 4175 | } |
| 4176 | |
| 4177 | relayd = consumer_find_relayd(relayd_id); |
| 4178 | if (!relayd) { |
| 4179 | ERR("Failed to find relayd %" PRIu64, relayd_id); |
| 4180 | ret = -1; |
| 4181 | goto end; |
| 4182 | } |
| 4183 | |
| 4184 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 4185 | ret = relayd_rotate_streams(&relayd->control_sock, stream_count, |
| 4186 | rotating_to_new_chunk ? &next_chunk_id : NULL, |
| 4187 | (const struct relayd_stream_rotation_position *) |
| 4188 | stream_rotation_positions.buffer.data); |
| 4189 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 4190 | if (ret < 0) { |
| 4191 | ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64, |
| 4192 | relayd->net_seq_idx); |
| 4193 | lttng_consumer_cleanup_relayd(relayd); |
| 4194 | goto end; |
| 4195 | } |
| 4196 | |
| 4197 | ret = 0; |
| 4198 | goto end; |
| 4199 | |
| 4200 | end_unlock_stream: |
| 4201 | pthread_mutex_unlock(&stream->lock); |
| 4202 | end_unlock_channel: |
| 4203 | pthread_mutex_unlock(&channel->lock); |
| 4204 | end: |
| 4205 | rcu_read_unlock(); |
| 4206 | lttng_dynamic_array_reset(&stream_rotation_positions); |
| 4207 | return ret; |
| 4208 | } |
| 4209 | |
| 4210 | /* |
| 4211 | * Check if a stream is ready to be rotated after extracting it. |
| 4212 | * |
| 4213 | * Return 1 if it is ready for rotation, 0 if it is not, a negative value on |
| 4214 | * error. Stream lock must be held. |
| 4215 | */ |
| 4216 | int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream) |
| 4217 | { |
| 4218 | int ret; |
| 4219 | unsigned long consumed_pos; |
| 4220 | |
| 4221 | if (!stream->rotate_position && !stream->rotate_ready) { |
| 4222 | ret = 0; |
| 4223 | goto end; |
| 4224 | } |
| 4225 | |
| 4226 | if (stream->rotate_ready) { |
| 4227 | ret = 1; |
| 4228 | goto end; |
| 4229 | } |
| 4230 | |
| 4231 | /* |
| 4232 | * If we don't have the rotate_ready flag, check the consumed position |
| 4233 | * to determine if we need to rotate. |
| 4234 | */ |
| 4235 | ret = lttng_consumer_sample_snapshot_positions(stream); |
| 4236 | if (ret < 0) { |
| 4237 | ERR("Taking snapshot positions"); |
| 4238 | goto end; |
| 4239 | } |
| 4240 | |
| 4241 | ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos); |
| 4242 | if (ret < 0) { |
| 4243 | ERR("Consumed snapshot position"); |
| 4244 | goto end; |
| 4245 | } |
| 4246 | |
| 4247 | /* Rotate position not reached yet (with check for overflow). */ |
| 4248 | if ((long) (consumed_pos - stream->rotate_position) < 0) { |
| 4249 | ret = 0; |
| 4250 | goto end; |
| 4251 | } |
| 4252 | ret = 1; |
| 4253 | |
| 4254 | end: |
| 4255 | return ret; |
| 4256 | } |
| 4257 | |
| 4258 | /* |
| 4259 | * Reset the state for a stream after a rotation occurred. |
| 4260 | */ |
| 4261 | void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream) |
| 4262 | { |
| 4263 | stream->rotate_position = 0; |
| 4264 | stream->rotate_ready = false; |
| 4265 | } |
| 4266 | |
| 4267 | /* |
| 4268 | * Perform the rotation a local stream file. |
| 4269 | */ |
| 4270 | static |
| 4271 | int rotate_local_stream(struct lttng_consumer_local_data *ctx, |
| 4272 | struct lttng_consumer_stream *stream) |
| 4273 | { |
| 4274 | int ret = 0; |
| 4275 | |
| 4276 | DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64, |
| 4277 | stream->key, |
| 4278 | stream->chan->key); |
| 4279 | stream->tracefile_size_current = 0; |
| 4280 | stream->tracefile_count_current = 0; |
| 4281 | |
| 4282 | if (stream->out_fd >= 0) { |
| 4283 | ret = close(stream->out_fd); |
| 4284 | if (ret) { |
| 4285 | PERROR("Failed to close stream out_fd of channel \"%s\"", |
| 4286 | stream->chan->name); |
| 4287 | } |
| 4288 | stream->out_fd = -1; |
| 4289 | } |
| 4290 | |
| 4291 | if (stream->index_file) { |
| 4292 | lttng_index_file_put(stream->index_file); |
| 4293 | stream->index_file = NULL; |
| 4294 | } |
| 4295 | |
| 4296 | if (!stream->trace_chunk) { |
| 4297 | goto end; |
| 4298 | } |
| 4299 | |
| 4300 | ret = consumer_stream_create_output_files(stream, true); |
| 4301 | end: |
| 4302 | return ret; |
| 4303 | } |
| 4304 | |
| 4305 | /* |
| 4306 | * Performs the stream rotation for the rotate session feature if needed. |
| 4307 | * It must be called with the channel and stream locks held. |
| 4308 | * |
| 4309 | * Return 0 on success, a negative number of error. |
| 4310 | */ |
| 4311 | int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx, |
| 4312 | struct lttng_consumer_stream *stream) |
| 4313 | { |
| 4314 | int ret; |
| 4315 | |
| 4316 | DBG("Consumer rotate stream %" PRIu64, stream->key); |
| 4317 | |
| 4318 | /* |
| 4319 | * Update the stream's 'current' chunk to the session's (channel) |
| 4320 | * now-current chunk. |
| 4321 | */ |
| 4322 | lttng_trace_chunk_put(stream->trace_chunk); |
| 4323 | if (stream->chan->trace_chunk == stream->trace_chunk) { |
| 4324 | /* |
| 4325 | * A channel can be rotated and not have a "next" chunk |
| 4326 | * to transition to. In that case, the channel's "current chunk" |
| 4327 | * has not been closed yet, but it has not been updated to |
| 4328 | * a "next" trace chunk either. Hence, the stream, like its |
| 4329 | * parent channel, becomes part of no chunk and can't output |
| 4330 | * anything until a new trace chunk is created. |
| 4331 | */ |
| 4332 | stream->trace_chunk = NULL; |
| 4333 | } else if (stream->chan->trace_chunk && |
| 4334 | !lttng_trace_chunk_get(stream->chan->trace_chunk)) { |
| 4335 | ERR("Failed to acquire a reference to channel's trace chunk during stream rotation"); |
| 4336 | ret = -1; |
| 4337 | goto error; |
| 4338 | } else { |
| 4339 | /* |
| 4340 | * Update the stream's trace chunk to its parent channel's |
| 4341 | * current trace chunk. |
| 4342 | */ |
| 4343 | stream->trace_chunk = stream->chan->trace_chunk; |
| 4344 | } |
| 4345 | |
| 4346 | if (stream->net_seq_idx == (uint64_t) -1ULL) { |
| 4347 | ret = rotate_local_stream(ctx, stream); |
| 4348 | if (ret < 0) { |
| 4349 | ERR("Failed to rotate stream, ret = %i", ret); |
| 4350 | goto error; |
| 4351 | } |
| 4352 | } |
| 4353 | |
| 4354 | if (stream->metadata_flag && stream->trace_chunk) { |
| 4355 | /* |
| 4356 | * If the stream has transitioned to a new trace |
| 4357 | * chunk, the metadata should be re-dumped to the |
| 4358 | * newest chunk. |
| 4359 | * |
| 4360 | * However, it is possible for a stream to transition to |
| 4361 | * a "no-chunk" state. This can happen if a rotation |
| 4362 | * occurs on an inactive session. In such cases, the metadata |
| 4363 | * regeneration will happen when the next trace chunk is |
| 4364 | * created. |
| 4365 | */ |
| 4366 | ret = consumer_metadata_stream_dump(stream); |
| 4367 | if (ret) { |
| 4368 | goto error; |
| 4369 | } |
| 4370 | } |
| 4371 | lttng_consumer_reset_stream_rotate_state(stream); |
| 4372 | |
| 4373 | ret = 0; |
| 4374 | |
| 4375 | error: |
| 4376 | return ret; |
| 4377 | } |
| 4378 | |
| 4379 | /* |
| 4380 | * Rotate all the ready streams now. |
| 4381 | * |
| 4382 | * This is especially important for low throughput streams that have already |
| 4383 | * been consumed, we cannot wait for their next packet to perform the |
| 4384 | * rotation. |
| 4385 | * Need to be called with RCU read-side lock held to ensure existence of |
| 4386 | * channel. |
| 4387 | * |
| 4388 | * Returns 0 on success, < 0 on error |
| 4389 | */ |
| 4390 | int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel, |
| 4391 | uint64_t key, struct lttng_consumer_local_data *ctx) |
| 4392 | { |
| 4393 | int ret; |
| 4394 | struct lttng_consumer_stream *stream; |
| 4395 | struct lttng_ht_iter iter; |
| 4396 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; |
| 4397 | |
| 4398 | rcu_read_lock(); |
| 4399 | |
| 4400 | DBG("Consumer rotate ready streams in channel %" PRIu64, key); |
| 4401 | |
| 4402 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 4403 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 4404 | ht->match_fct, &channel->key, &iter.iter, |
| 4405 | stream, node_channel_id.node) { |
| 4406 | health_code_update(); |
| 4407 | |
| 4408 | pthread_mutex_lock(&stream->chan->lock); |
| 4409 | pthread_mutex_lock(&stream->lock); |
| 4410 | |
| 4411 | if (!stream->rotate_ready) { |
| 4412 | pthread_mutex_unlock(&stream->lock); |
| 4413 | pthread_mutex_unlock(&stream->chan->lock); |
| 4414 | continue; |
| 4415 | } |
| 4416 | DBG("Consumer rotate ready stream %" PRIu64, stream->key); |
| 4417 | |
| 4418 | ret = lttng_consumer_rotate_stream(ctx, stream); |
| 4419 | pthread_mutex_unlock(&stream->lock); |
| 4420 | pthread_mutex_unlock(&stream->chan->lock); |
| 4421 | if (ret) { |
| 4422 | goto end; |
| 4423 | } |
| 4424 | } |
| 4425 | |
| 4426 | ret = 0; |
| 4427 | |
| 4428 | end: |
| 4429 | rcu_read_unlock(); |
| 4430 | return ret; |
| 4431 | } |
| 4432 | |
| 4433 | enum lttcomm_return_code lttng_consumer_init_command( |
| 4434 | struct lttng_consumer_local_data *ctx, |
| 4435 | const lttng_uuid sessiond_uuid) |
| 4436 | { |
| 4437 | enum lttcomm_return_code ret; |
| 4438 | char uuid_str[UUID_STR_LEN]; |
| 4439 | |
| 4440 | if (ctx->sessiond_uuid.is_set) { |
| 4441 | ret = LTTCOMM_CONSUMERD_ALREADY_SET; |
| 4442 | goto end; |
| 4443 | } |
| 4444 | |
| 4445 | ctx->sessiond_uuid.is_set = true; |
| 4446 | memcpy(ctx->sessiond_uuid.value, sessiond_uuid, sizeof(lttng_uuid)); |
| 4447 | ret = LTTCOMM_CONSUMERD_SUCCESS; |
| 4448 | lttng_uuid_to_str(sessiond_uuid, uuid_str); |
| 4449 | DBG("Received session daemon UUID: %s", uuid_str); |
| 4450 | end: |
| 4451 | return ret; |
| 4452 | } |
| 4453 | |
| 4454 | enum lttcomm_return_code lttng_consumer_create_trace_chunk( |
| 4455 | const uint64_t *relayd_id, uint64_t session_id, |
| 4456 | uint64_t chunk_id, |
| 4457 | time_t chunk_creation_timestamp, |
| 4458 | const char *chunk_override_name, |
| 4459 | const struct lttng_credentials *credentials, |
| 4460 | struct lttng_directory_handle *chunk_directory_handle) |
| 4461 | { |
| 4462 | int ret; |
| 4463 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
| 4464 | struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL; |
| 4465 | enum lttng_trace_chunk_status chunk_status; |
| 4466 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; |
| 4467 | char creation_timestamp_buffer[ISO8601_STR_LEN]; |
| 4468 | const char *relayd_id_str = "(none)"; |
| 4469 | const char *creation_timestamp_str; |
| 4470 | struct lttng_ht_iter iter; |
| 4471 | struct lttng_consumer_channel *channel; |
| 4472 | |
| 4473 | if (relayd_id) { |
| 4474 | /* Only used for logging purposes. */ |
| 4475 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), |
| 4476 | "%" PRIu64, *relayd_id); |
| 4477 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { |
| 4478 | relayd_id_str = relayd_id_buffer; |
| 4479 | } else { |
| 4480 | relayd_id_str = "(formatting error)"; |
| 4481 | } |
| 4482 | } |
| 4483 | |
| 4484 | /* Local protocol error. */ |
| 4485 | assert(chunk_creation_timestamp); |
| 4486 | ret = time_to_iso8601_str(chunk_creation_timestamp, |
| 4487 | creation_timestamp_buffer, |
| 4488 | sizeof(creation_timestamp_buffer)); |
| 4489 | creation_timestamp_str = !ret ? creation_timestamp_buffer : |
| 4490 | "(formatting error)"; |
| 4491 | |
| 4492 | DBG("Consumer create trace chunk command: relay_id = %s" |
| 4493 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 |
| 4494 | ", chunk_override_name = %s" |
| 4495 | ", chunk_creation_timestamp = %s", |
| 4496 | relayd_id_str, session_id, chunk_id, |
| 4497 | chunk_override_name ? : "(none)", |
| 4498 | creation_timestamp_str); |
| 4499 | |
| 4500 | /* |
| 4501 | * The trace chunk registry, as used by the consumer daemon, implicitly |
| 4502 | * owns the trace chunks. This is only needed in the consumer since |
| 4503 | * the consumer has no notion of a session beyond session IDs being |
| 4504 | * used to identify other objects. |
| 4505 | * |
| 4506 | * The lttng_trace_chunk_registry_publish() call below provides a |
| 4507 | * reference which is not released; it implicitly becomes the session |
| 4508 | * daemon's reference to the chunk in the consumer daemon. |
| 4509 | * |
| 4510 | * The lifetime of trace chunks in the consumer daemon is managed by |
| 4511 | * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK |
| 4512 | * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands. |
| 4513 | */ |
| 4514 | created_chunk = lttng_trace_chunk_create(chunk_id, |
| 4515 | chunk_creation_timestamp); |
| 4516 | if (!created_chunk) { |
| 4517 | ERR("Failed to create trace chunk"); |
| 4518 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4519 | goto error; |
| 4520 | } |
| 4521 | |
| 4522 | if (chunk_override_name) { |
| 4523 | chunk_status = lttng_trace_chunk_override_name(created_chunk, |
| 4524 | chunk_override_name); |
| 4525 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4526 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4527 | goto error; |
| 4528 | } |
| 4529 | } |
| 4530 | |
| 4531 | if (chunk_directory_handle) { |
| 4532 | chunk_status = lttng_trace_chunk_set_credentials(created_chunk, |
| 4533 | credentials); |
| 4534 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4535 | ERR("Failed to set trace chunk credentials"); |
| 4536 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4537 | goto error; |
| 4538 | } |
| 4539 | /* |
| 4540 | * The consumer daemon has no ownership of the chunk output |
| 4541 | * directory. |
| 4542 | */ |
| 4543 | chunk_status = lttng_trace_chunk_set_as_user(created_chunk, |
| 4544 | chunk_directory_handle); |
| 4545 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4546 | ERR("Failed to set trace chunk's directory handle"); |
| 4547 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4548 | goto error; |
| 4549 | } |
| 4550 | } |
| 4551 | |
| 4552 | published_chunk = lttng_trace_chunk_registry_publish_chunk( |
| 4553 | consumer_data.chunk_registry, session_id, |
| 4554 | created_chunk); |
| 4555 | lttng_trace_chunk_put(created_chunk); |
| 4556 | created_chunk = NULL; |
| 4557 | if (!published_chunk) { |
| 4558 | ERR("Failed to publish trace chunk"); |
| 4559 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4560 | goto error; |
| 4561 | } |
| 4562 | |
| 4563 | rcu_read_lock(); |
| 4564 | cds_lfht_for_each_entry_duplicate(consumer_data.channels_by_session_id_ht->ht, |
| 4565 | consumer_data.channels_by_session_id_ht->hash_fct( |
| 4566 | &session_id, lttng_ht_seed), |
| 4567 | consumer_data.channels_by_session_id_ht->match_fct, |
| 4568 | &session_id, &iter.iter, channel, |
| 4569 | channels_by_session_id_ht_node.node) { |
| 4570 | ret = lttng_consumer_channel_set_trace_chunk(channel, |
| 4571 | published_chunk); |
| 4572 | if (ret) { |
| 4573 | /* |
| 4574 | * Roll-back the creation of this chunk. |
| 4575 | * |
| 4576 | * This is important since the session daemon will |
| 4577 | * assume that the creation of this chunk failed and |
| 4578 | * will never ask for it to be closed, resulting |
| 4579 | * in a leak and an inconsistent state for some |
| 4580 | * channels. |
| 4581 | */ |
| 4582 | enum lttcomm_return_code close_ret; |
| 4583 | char path[LTTNG_PATH_MAX]; |
| 4584 | |
| 4585 | DBG("Failed to set new trace chunk on existing channels, rolling back"); |
| 4586 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, |
| 4587 | session_id, chunk_id, |
| 4588 | chunk_creation_timestamp, NULL, |
| 4589 | path); |
| 4590 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
| 4591 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, |
| 4592 | session_id, chunk_id); |
| 4593 | } |
| 4594 | |
| 4595 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4596 | break; |
| 4597 | } |
| 4598 | } |
| 4599 | |
| 4600 | if (relayd_id) { |
| 4601 | struct consumer_relayd_sock_pair *relayd; |
| 4602 | |
| 4603 | relayd = consumer_find_relayd(*relayd_id); |
| 4604 | if (relayd) { |
| 4605 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 4606 | ret = relayd_create_trace_chunk( |
| 4607 | &relayd->control_sock, published_chunk); |
| 4608 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 4609 | } else { |
| 4610 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, *relayd_id); |
| 4611 | } |
| 4612 | |
| 4613 | if (!relayd || ret) { |
| 4614 | enum lttcomm_return_code close_ret; |
| 4615 | char path[LTTNG_PATH_MAX]; |
| 4616 | |
| 4617 | close_ret = lttng_consumer_close_trace_chunk(relayd_id, |
| 4618 | session_id, |
| 4619 | chunk_id, |
| 4620 | chunk_creation_timestamp, |
| 4621 | NULL, path); |
| 4622 | if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) { |
| 4623 | ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64, |
| 4624 | session_id, |
| 4625 | chunk_id); |
| 4626 | } |
| 4627 | |
| 4628 | ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED; |
| 4629 | goto error_unlock; |
| 4630 | } |
| 4631 | } |
| 4632 | error_unlock: |
| 4633 | rcu_read_unlock(); |
| 4634 | error: |
| 4635 | /* Release the reference returned by the "publish" operation. */ |
| 4636 | lttng_trace_chunk_put(published_chunk); |
| 4637 | lttng_trace_chunk_put(created_chunk); |
| 4638 | return ret_code; |
| 4639 | } |
| 4640 | |
| 4641 | enum lttcomm_return_code lttng_consumer_close_trace_chunk( |
| 4642 | const uint64_t *relayd_id, uint64_t session_id, |
| 4643 | uint64_t chunk_id, time_t chunk_close_timestamp, |
| 4644 | const enum lttng_trace_chunk_command_type *close_command, |
| 4645 | char *path) |
| 4646 | { |
| 4647 | enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS; |
| 4648 | struct lttng_trace_chunk *chunk; |
| 4649 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; |
| 4650 | const char *relayd_id_str = "(none)"; |
| 4651 | const char *close_command_name = "none"; |
| 4652 | struct lttng_ht_iter iter; |
| 4653 | struct lttng_consumer_channel *channel; |
| 4654 | enum lttng_trace_chunk_status chunk_status; |
| 4655 | |
| 4656 | if (relayd_id) { |
| 4657 | int ret; |
| 4658 | |
| 4659 | /* Only used for logging purposes. */ |
| 4660 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), |
| 4661 | "%" PRIu64, *relayd_id); |
| 4662 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { |
| 4663 | relayd_id_str = relayd_id_buffer; |
| 4664 | } else { |
| 4665 | relayd_id_str = "(formatting error)"; |
| 4666 | } |
| 4667 | } |
| 4668 | if (close_command) { |
| 4669 | close_command_name = lttng_trace_chunk_command_type_get_name( |
| 4670 | *close_command); |
| 4671 | } |
| 4672 | |
| 4673 | DBG("Consumer close trace chunk command: relayd_id = %s" |
| 4674 | ", session_id = %" PRIu64 ", chunk_id = %" PRIu64 |
| 4675 | ", close command = %s", |
| 4676 | relayd_id_str, session_id, chunk_id, |
| 4677 | close_command_name); |
| 4678 | |
| 4679 | chunk = lttng_trace_chunk_registry_find_chunk( |
| 4680 | consumer_data.chunk_registry, session_id, chunk_id); |
| 4681 | if (!chunk) { |
| 4682 | ERR("Failed to find chunk: session_id = %" PRIu64 |
| 4683 | ", chunk_id = %" PRIu64, |
| 4684 | session_id, chunk_id); |
| 4685 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; |
| 4686 | goto end; |
| 4687 | } |
| 4688 | |
| 4689 | chunk_status = lttng_trace_chunk_set_close_timestamp(chunk, |
| 4690 | chunk_close_timestamp); |
| 4691 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4692 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; |
| 4693 | goto end; |
| 4694 | } |
| 4695 | |
| 4696 | if (close_command) { |
| 4697 | chunk_status = lttng_trace_chunk_set_close_command( |
| 4698 | chunk, *close_command); |
| 4699 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 4700 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; |
| 4701 | goto end; |
| 4702 | } |
| 4703 | } |
| 4704 | |
| 4705 | /* |
| 4706 | * chunk is now invalid to access as we no longer hold a reference to |
| 4707 | * it; it is only kept around to compare it (by address) to the |
| 4708 | * current chunk found in the session's channels. |
| 4709 | */ |
| 4710 | rcu_read_lock(); |
| 4711 | cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, |
| 4712 | channel, node.node) { |
| 4713 | int ret; |
| 4714 | |
| 4715 | /* |
| 4716 | * Only change the channel's chunk to NULL if it still |
| 4717 | * references the chunk being closed. The channel may |
| 4718 | * reference a newer channel in the case of a session |
| 4719 | * rotation. When a session rotation occurs, the "next" |
| 4720 | * chunk is created before the "current" chunk is closed. |
| 4721 | */ |
| 4722 | if (channel->trace_chunk != chunk) { |
| 4723 | continue; |
| 4724 | } |
| 4725 | ret = lttng_consumer_channel_set_trace_chunk(channel, NULL); |
| 4726 | if (ret) { |
| 4727 | /* |
| 4728 | * Attempt to close the chunk on as many channels as |
| 4729 | * possible. |
| 4730 | */ |
| 4731 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; |
| 4732 | } |
| 4733 | } |
| 4734 | |
| 4735 | if (relayd_id) { |
| 4736 | int ret; |
| 4737 | struct consumer_relayd_sock_pair *relayd; |
| 4738 | |
| 4739 | relayd = consumer_find_relayd(*relayd_id); |
| 4740 | if (relayd) { |
| 4741 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 4742 | ret = relayd_close_trace_chunk( |
| 4743 | &relayd->control_sock, chunk, |
| 4744 | path); |
| 4745 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 4746 | } else { |
| 4747 | ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, |
| 4748 | *relayd_id); |
| 4749 | } |
| 4750 | |
| 4751 | if (!relayd || ret) { |
| 4752 | ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED; |
| 4753 | goto error_unlock; |
| 4754 | } |
| 4755 | } |
| 4756 | error_unlock: |
| 4757 | rcu_read_unlock(); |
| 4758 | end: |
| 4759 | /* |
| 4760 | * Release the reference returned by the "find" operation and |
| 4761 | * the session daemon's implicit reference to the chunk. |
| 4762 | */ |
| 4763 | lttng_trace_chunk_put(chunk); |
| 4764 | lttng_trace_chunk_put(chunk); |
| 4765 | |
| 4766 | return ret_code; |
| 4767 | } |
| 4768 | |
| 4769 | enum lttcomm_return_code lttng_consumer_trace_chunk_exists( |
| 4770 | const uint64_t *relayd_id, uint64_t session_id, |
| 4771 | uint64_t chunk_id) |
| 4772 | { |
| 4773 | int ret; |
| 4774 | enum lttcomm_return_code ret_code; |
| 4775 | struct lttng_trace_chunk *chunk; |
| 4776 | char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)]; |
| 4777 | const char *relayd_id_str = "(none)"; |
| 4778 | const bool is_local_trace = !relayd_id; |
| 4779 | struct consumer_relayd_sock_pair *relayd = NULL; |
| 4780 | bool chunk_exists_remote; |
| 4781 | |
| 4782 | if (relayd_id) { |
| 4783 | int ret; |
| 4784 | |
| 4785 | /* Only used for logging purposes. */ |
| 4786 | ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer), |
| 4787 | "%" PRIu64, *relayd_id); |
| 4788 | if (ret > 0 && ret < sizeof(relayd_id_buffer)) { |
| 4789 | relayd_id_str = relayd_id_buffer; |
| 4790 | } else { |
| 4791 | relayd_id_str = "(formatting error)"; |
| 4792 | } |
| 4793 | } |
| 4794 | |
| 4795 | DBG("Consumer trace chunk exists command: relayd_id = %s" |
| 4796 | ", chunk_id = %" PRIu64, relayd_id_str, |
| 4797 | chunk_id); |
| 4798 | chunk = lttng_trace_chunk_registry_find_chunk( |
| 4799 | consumer_data.chunk_registry, session_id, |
| 4800 | chunk_id); |
| 4801 | DBG("Trace chunk %s locally", chunk ? "exists" : "does not exist"); |
| 4802 | if (chunk) { |
| 4803 | ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL; |
| 4804 | lttng_trace_chunk_put(chunk); |
| 4805 | goto end; |
| 4806 | } else if (is_local_trace) { |
| 4807 | ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; |
| 4808 | goto end; |
| 4809 | } |
| 4810 | |
| 4811 | rcu_read_lock(); |
| 4812 | relayd = consumer_find_relayd(*relayd_id); |
| 4813 | if (!relayd) { |
| 4814 | ERR("Failed to find relayd %" PRIu64, *relayd_id); |
| 4815 | ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS; |
| 4816 | goto end_rcu_unlock; |
| 4817 | } |
| 4818 | DBG("Looking up existence of trace chunk on relay daemon"); |
| 4819 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 4820 | ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id, |
| 4821 | &chunk_exists_remote); |
| 4822 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 4823 | if (ret < 0) { |
| 4824 | ERR("Failed to look-up the existence of trace chunk on relay daemon"); |
| 4825 | ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL; |
| 4826 | goto end_rcu_unlock; |
| 4827 | } |
| 4828 | |
| 4829 | ret_code = chunk_exists_remote ? |
| 4830 | LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE : |
| 4831 | LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK; |
| 4832 | DBG("Trace chunk %s on relay daemon", |
| 4833 | chunk_exists_remote ? "exists" : "does not exist"); |
| 4834 | |
| 4835 | end_rcu_unlock: |
| 4836 | rcu_read_unlock(); |
| 4837 | end: |
| 4838 | return ret_code; |
| 4839 | } |