| 1 | /* |
| 2 | * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca> |
| 3 | * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com> |
| 4 | * Copyright (C) 2013 David Goulet <dgoulet@efficios.com> |
| 5 | * |
| 6 | * SPDX-License-Identifier: GPL-2.0-only |
| 7 | * |
| 8 | */ |
| 9 | |
| 10 | #define _LGPL_SOURCE |
| 11 | #include <assert.h> |
| 12 | #include <inttypes.h> |
| 13 | #include <sys/mman.h> |
| 14 | #include <unistd.h> |
| 15 | |
| 16 | #include <common/common.h> |
| 17 | #include <common/index/index.h> |
| 18 | #include <common/kernel-consumer/kernel-consumer.h> |
| 19 | #include <common/relayd/relayd.h> |
| 20 | #include <common/ust-consumer/ust-consumer.h> |
| 21 | #include <common/utils.h> |
| 22 | #include <common/consumer/consumer.h> |
| 23 | #include <common/consumer/consumer-timer.h> |
| 24 | #include <common/consumer/metadata-bucket.h> |
| 25 | |
| 26 | #include "consumer-stream.h" |
| 27 | |
| 28 | /* |
| 29 | * RCU call to free stream. MUST only be used with call_rcu(). |
| 30 | */ |
| 31 | static void free_stream_rcu(struct rcu_head *head) |
| 32 | { |
| 33 | struct lttng_ht_node_u64 *node = |
| 34 | caa_container_of(head, struct lttng_ht_node_u64, head); |
| 35 | struct lttng_consumer_stream *stream = |
| 36 | caa_container_of(node, struct lttng_consumer_stream, node); |
| 37 | |
| 38 | pthread_mutex_destroy(&stream->lock); |
| 39 | free(stream); |
| 40 | } |
| 41 | |
| 42 | static void consumer_stream_data_lock_all(struct lttng_consumer_stream *stream) |
| 43 | { |
| 44 | pthread_mutex_lock(&stream->chan->lock); |
| 45 | pthread_mutex_lock(&stream->lock); |
| 46 | } |
| 47 | |
| 48 | static void consumer_stream_data_unlock_all(struct lttng_consumer_stream *stream) |
| 49 | { |
| 50 | pthread_mutex_unlock(&stream->lock); |
| 51 | pthread_mutex_unlock(&stream->chan->lock); |
| 52 | } |
| 53 | |
| 54 | static void consumer_stream_metadata_lock_all(struct lttng_consumer_stream *stream) |
| 55 | { |
| 56 | consumer_stream_data_lock_all(stream); |
| 57 | pthread_mutex_lock(&stream->metadata_rdv_lock); |
| 58 | } |
| 59 | |
| 60 | static void consumer_stream_metadata_unlock_all(struct lttng_consumer_stream *stream) |
| 61 | { |
| 62 | pthread_mutex_unlock(&stream->metadata_rdv_lock); |
| 63 | consumer_stream_data_unlock_all(stream); |
| 64 | } |
| 65 | |
| 66 | /* Only used for data streams. */ |
| 67 | static int consumer_stream_update_stats(struct lttng_consumer_stream *stream, |
| 68 | const struct stream_subbuffer *subbuf) |
| 69 | { |
| 70 | int ret = 0; |
| 71 | uint64_t sequence_number; |
| 72 | const uint64_t discarded_events = subbuf->info.data.events_discarded; |
| 73 | |
| 74 | if (!subbuf->info.data.sequence_number.is_set) { |
| 75 | /* Command not supported by the tracer. */ |
| 76 | sequence_number = -1ULL; |
| 77 | stream->sequence_number_unavailable = true; |
| 78 | } else { |
| 79 | sequence_number = subbuf->info.data.sequence_number.value; |
| 80 | } |
| 81 | |
| 82 | /* |
| 83 | * Start the sequence when we extract the first packet in case we don't |
| 84 | * start at 0 (for example if a consumer is not connected to the |
| 85 | * session immediately after the beginning). |
| 86 | */ |
| 87 | if (stream->last_sequence_number == -1ULL) { |
| 88 | stream->last_sequence_number = sequence_number; |
| 89 | } else if (sequence_number > stream->last_sequence_number) { |
| 90 | stream->chan->lost_packets += sequence_number - |
| 91 | stream->last_sequence_number - 1; |
| 92 | } else { |
| 93 | /* seq <= last_sequence_number */ |
| 94 | ERR("Sequence number inconsistent : prev = %" PRIu64 |
| 95 | ", current = %" PRIu64, |
| 96 | stream->last_sequence_number, sequence_number); |
| 97 | ret = -1; |
| 98 | goto end; |
| 99 | } |
| 100 | stream->last_sequence_number = sequence_number; |
| 101 | |
| 102 | if (discarded_events < stream->last_discarded_events) { |
| 103 | /* |
| 104 | * Overflow has occurred. We assume only one wrap-around |
| 105 | * has occurred. |
| 106 | */ |
| 107 | stream->chan->discarded_events += |
| 108 | (1ULL << (CAA_BITS_PER_LONG - 1)) - |
| 109 | stream->last_discarded_events + |
| 110 | discarded_events; |
| 111 | } else { |
| 112 | stream->chan->discarded_events += discarded_events - |
| 113 | stream->last_discarded_events; |
| 114 | } |
| 115 | stream->last_discarded_events = discarded_events; |
| 116 | ret = 0; |
| 117 | |
| 118 | end: |
| 119 | return ret; |
| 120 | } |
| 121 | |
| 122 | static |
| 123 | void ctf_packet_index_populate(struct ctf_packet_index *index, |
| 124 | off_t offset, const struct stream_subbuffer *subbuffer) |
| 125 | { |
| 126 | *index = (typeof(*index)){ |
| 127 | .offset = htobe64(offset), |
| 128 | .packet_size = htobe64(subbuffer->info.data.packet_size), |
| 129 | .content_size = htobe64(subbuffer->info.data.content_size), |
| 130 | .timestamp_begin = htobe64( |
| 131 | subbuffer->info.data.timestamp_begin), |
| 132 | .timestamp_end = htobe64( |
| 133 | subbuffer->info.data.timestamp_end), |
| 134 | .events_discarded = htobe64( |
| 135 | subbuffer->info.data.events_discarded), |
| 136 | .stream_id = htobe64(subbuffer->info.data.stream_id), |
| 137 | .stream_instance_id = htobe64( |
| 138 | subbuffer->info.data.stream_instance_id.is_set ? |
| 139 | subbuffer->info.data.stream_instance_id.value : -1ULL), |
| 140 | .packet_seq_num = htobe64( |
| 141 | subbuffer->info.data.sequence_number.is_set ? |
| 142 | subbuffer->info.data.sequence_number.value : -1ULL), |
| 143 | }; |
| 144 | } |
| 145 | |
| 146 | static ssize_t consumer_stream_consume_mmap( |
| 147 | struct lttng_consumer_local_data *ctx, |
| 148 | struct lttng_consumer_stream *stream, |
| 149 | const struct stream_subbuffer *subbuffer) |
| 150 | { |
| 151 | const unsigned long padding_size = |
| 152 | subbuffer->info.data.padded_subbuf_size - |
| 153 | subbuffer->info.data.subbuf_size; |
| 154 | const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_mmap( |
| 155 | stream, &subbuffer->buffer.buffer, padding_size); |
| 156 | |
| 157 | if (stream->net_seq_idx == -1ULL) { |
| 158 | /* |
| 159 | * When writing on disk, check that only the subbuffer (no |
| 160 | * padding) was written to disk. |
| 161 | */ |
| 162 | if (written_bytes != subbuffer->info.data.padded_subbuf_size) { |
| 163 | DBG("Failed to write the entire padded subbuffer on disk (written_bytes: %zd, padded subbuffer size %lu)", |
| 164 | written_bytes, |
| 165 | subbuffer->info.data.padded_subbuf_size); |
| 166 | } |
| 167 | } else { |
| 168 | /* |
| 169 | * When streaming over the network, check that the entire |
| 170 | * subbuffer including padding was successfully written. |
| 171 | */ |
| 172 | if (written_bytes != subbuffer->info.data.subbuf_size) { |
| 173 | DBG("Failed to write only the subbuffer over the network (written_bytes: %zd, subbuffer size %lu)", |
| 174 | written_bytes, |
| 175 | subbuffer->info.data.subbuf_size); |
| 176 | } |
| 177 | } |
| 178 | |
| 179 | /* |
| 180 | * If `lttng_consumer_on_read_subbuffer_mmap()` returned an error, pass |
| 181 | * it along to the caller, else return zero. |
| 182 | */ |
| 183 | if (written_bytes < 0) { |
| 184 | ERR("Error reading mmap subbuffer: %zd", written_bytes); |
| 185 | } |
| 186 | |
| 187 | return written_bytes; |
| 188 | } |
| 189 | |
| 190 | static ssize_t consumer_stream_consume_splice( |
| 191 | struct lttng_consumer_local_data *ctx, |
| 192 | struct lttng_consumer_stream *stream, |
| 193 | const struct stream_subbuffer *subbuffer) |
| 194 | { |
| 195 | const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_splice( |
| 196 | ctx, stream, subbuffer->info.data.padded_subbuf_size, 0); |
| 197 | |
| 198 | if (written_bytes != subbuffer->info.data.padded_subbuf_size) { |
| 199 | DBG("Failed to write the entire padded subbuffer (written_bytes: %zd, padded subbuffer size %lu)", |
| 200 | written_bytes, |
| 201 | subbuffer->info.data.padded_subbuf_size); |
| 202 | } |
| 203 | |
| 204 | /* |
| 205 | * If `lttng_consumer_on_read_subbuffer_splice()` returned an error, |
| 206 | * pass it along to the caller, else return zero. |
| 207 | */ |
| 208 | if (written_bytes < 0) { |
| 209 | ERR("Error reading splice subbuffer: %zd", written_bytes); |
| 210 | } |
| 211 | |
| 212 | return written_bytes; |
| 213 | } |
| 214 | |
| 215 | static int consumer_stream_send_index( |
| 216 | struct lttng_consumer_stream *stream, |
| 217 | const struct stream_subbuffer *subbuffer, |
| 218 | struct lttng_consumer_local_data *ctx) |
| 219 | { |
| 220 | off_t packet_offset = 0; |
| 221 | struct ctf_packet_index index = {}; |
| 222 | |
| 223 | /* |
| 224 | * This is called after consuming the sub-buffer; substract the |
| 225 | * effect this sub-buffer from the offset. |
| 226 | */ |
| 227 | if (stream->net_seq_idx == (uint64_t) -1ULL) { |
| 228 | packet_offset = stream->out_fd_offset - |
| 229 | subbuffer->info.data.padded_subbuf_size; |
| 230 | } |
| 231 | |
| 232 | ctf_packet_index_populate(&index, packet_offset, subbuffer); |
| 233 | return consumer_stream_write_index(stream, &index); |
| 234 | } |
| 235 | |
| 236 | /* |
| 237 | * Actually do the metadata sync using the given metadata stream. |
| 238 | * |
| 239 | * Return 0 on success else a negative value. ENODATA can be returned also |
| 240 | * indicating that there is no metadata available for that stream. |
| 241 | */ |
| 242 | static int do_sync_metadata(struct lttng_consumer_stream *metadata, |
| 243 | struct lttng_consumer_local_data *ctx) |
| 244 | { |
| 245 | int ret; |
| 246 | enum sync_metadata_status status; |
| 247 | |
| 248 | assert(metadata); |
| 249 | assert(metadata->metadata_flag); |
| 250 | assert(ctx); |
| 251 | |
| 252 | /* |
| 253 | * In UST, since we have to write the metadata from the cache packet |
| 254 | * by packet, we might need to start this procedure multiple times |
| 255 | * until all the metadata from the cache has been extracted. |
| 256 | */ |
| 257 | do { |
| 258 | /* |
| 259 | * Steps : |
| 260 | * - Lock the metadata stream |
| 261 | * - Check if metadata stream node was deleted before locking. |
| 262 | * - if yes, release and return success |
| 263 | * - Check if new metadata is ready (flush + snapshot pos) |
| 264 | * - If nothing : release and return. |
| 265 | * - Lock the metadata_rdv_lock |
| 266 | * - Unlock the metadata stream |
| 267 | * - cond_wait on metadata_rdv to wait the wakeup from the |
| 268 | * metadata thread |
| 269 | * - Unlock the metadata_rdv_lock |
| 270 | */ |
| 271 | pthread_mutex_lock(&metadata->lock); |
| 272 | |
| 273 | /* |
| 274 | * There is a possibility that we were able to acquire a reference on the |
| 275 | * stream from the RCU hash table but between then and now, the node might |
| 276 | * have been deleted just before the lock is acquired. Thus, after locking, |
| 277 | * we make sure the metadata node has not been deleted which means that the |
| 278 | * buffers are closed. |
| 279 | * |
| 280 | * In that case, there is no need to sync the metadata hence returning a |
| 281 | * success return code. |
| 282 | */ |
| 283 | ret = cds_lfht_is_node_deleted(&metadata->node.node); |
| 284 | if (ret) { |
| 285 | ret = 0; |
| 286 | goto end_unlock_mutex; |
| 287 | } |
| 288 | |
| 289 | switch (ctx->type) { |
| 290 | case LTTNG_CONSUMER_KERNEL: |
| 291 | /* |
| 292 | * Empty the metadata cache and flush the current stream. |
| 293 | */ |
| 294 | status = lttng_kconsumer_sync_metadata(metadata); |
| 295 | break; |
| 296 | case LTTNG_CONSUMER32_UST: |
| 297 | case LTTNG_CONSUMER64_UST: |
| 298 | /* |
| 299 | * Ask the sessiond if we have new metadata waiting and update the |
| 300 | * consumer metadata cache. |
| 301 | */ |
| 302 | status = lttng_ustconsumer_sync_metadata(ctx, metadata); |
| 303 | break; |
| 304 | default: |
| 305 | abort(); |
| 306 | } |
| 307 | |
| 308 | switch (status) { |
| 309 | case SYNC_METADATA_STATUS_NEW_DATA: |
| 310 | break; |
| 311 | case SYNC_METADATA_STATUS_NO_DATA: |
| 312 | ret = 0; |
| 313 | goto end_unlock_mutex; |
| 314 | case SYNC_METADATA_STATUS_ERROR: |
| 315 | ret = -1; |
| 316 | goto end_unlock_mutex; |
| 317 | default: |
| 318 | abort(); |
| 319 | } |
| 320 | |
| 321 | /* |
| 322 | * At this point, new metadata have been flushed, so we wait on the |
| 323 | * rendez-vous point for the metadata thread to wake us up when it |
| 324 | * finishes consuming the metadata and continue execution. |
| 325 | */ |
| 326 | |
| 327 | pthread_mutex_lock(&metadata->metadata_rdv_lock); |
| 328 | |
| 329 | /* |
| 330 | * Release metadata stream lock so the metadata thread can process it. |
| 331 | */ |
| 332 | pthread_mutex_unlock(&metadata->lock); |
| 333 | |
| 334 | /* |
| 335 | * Wait on the rendez-vous point. Once woken up, it means the metadata was |
| 336 | * consumed and thus synchronization is achieved. |
| 337 | */ |
| 338 | pthread_cond_wait(&metadata->metadata_rdv, &metadata->metadata_rdv_lock); |
| 339 | pthread_mutex_unlock(&metadata->metadata_rdv_lock); |
| 340 | } while (status == SYNC_METADATA_STATUS_NEW_DATA); |
| 341 | |
| 342 | /* Success */ |
| 343 | return 0; |
| 344 | |
| 345 | end_unlock_mutex: |
| 346 | pthread_mutex_unlock(&metadata->lock); |
| 347 | return ret; |
| 348 | } |
| 349 | |
| 350 | /* |
| 351 | * Synchronize the metadata using a given session ID. A successful acquisition |
| 352 | * of a metadata stream will trigger a request to the session daemon and a |
| 353 | * snapshot so the metadata thread can consume it. |
| 354 | * |
| 355 | * This function call is a rendez-vous point between the metadata thread and |
| 356 | * the data thread. |
| 357 | * |
| 358 | * Return 0 on success or else a negative value. |
| 359 | */ |
| 360 | int consumer_stream_sync_metadata(struct lttng_consumer_local_data *ctx, |
| 361 | uint64_t session_id) |
| 362 | { |
| 363 | int ret; |
| 364 | struct lttng_consumer_stream *stream = NULL; |
| 365 | struct lttng_ht_iter iter; |
| 366 | struct lttng_ht *ht; |
| 367 | |
| 368 | assert(ctx); |
| 369 | |
| 370 | /* Ease our life a bit. */ |
| 371 | ht = consumer_data.stream_list_ht; |
| 372 | |
| 373 | rcu_read_lock(); |
| 374 | |
| 375 | /* Search the metadata associated with the session id of the given stream. */ |
| 376 | |
| 377 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 378 | ht->hash_fct(&session_id, lttng_ht_seed), ht->match_fct, |
| 379 | &session_id, &iter.iter, stream, node_session_id.node) { |
| 380 | if (!stream->metadata_flag) { |
| 381 | continue; |
| 382 | } |
| 383 | |
| 384 | ret = do_sync_metadata(stream, ctx); |
| 385 | if (ret < 0) { |
| 386 | goto end; |
| 387 | } |
| 388 | } |
| 389 | |
| 390 | /* |
| 391 | * Force return code to 0 (success) since ret might be ENODATA for instance |
| 392 | * which is not an error but rather that we should come back. |
| 393 | */ |
| 394 | ret = 0; |
| 395 | |
| 396 | end: |
| 397 | rcu_read_unlock(); |
| 398 | return ret; |
| 399 | } |
| 400 | |
| 401 | static int consumer_stream_sync_metadata_index( |
| 402 | struct lttng_consumer_stream *stream, |
| 403 | const struct stream_subbuffer *subbuffer, |
| 404 | struct lttng_consumer_local_data *ctx) |
| 405 | { |
| 406 | int ret; |
| 407 | |
| 408 | /* Block until all the metadata is sent. */ |
| 409 | pthread_mutex_lock(&stream->metadata_timer_lock); |
| 410 | assert(!stream->missed_metadata_flush); |
| 411 | stream->waiting_on_metadata = true; |
| 412 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 413 | |
| 414 | ret = consumer_stream_sync_metadata(ctx, stream->session_id); |
| 415 | |
| 416 | pthread_mutex_lock(&stream->metadata_timer_lock); |
| 417 | stream->waiting_on_metadata = false; |
| 418 | if (stream->missed_metadata_flush) { |
| 419 | stream->missed_metadata_flush = false; |
| 420 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 421 | (void) stream->read_subbuffer_ops.send_live_beacon(stream); |
| 422 | } else { |
| 423 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 424 | } |
| 425 | if (ret < 0) { |
| 426 | goto end; |
| 427 | } |
| 428 | |
| 429 | ret = consumer_stream_send_index(stream, subbuffer, ctx); |
| 430 | end: |
| 431 | return ret; |
| 432 | } |
| 433 | |
| 434 | /* |
| 435 | * Check if the local version of the metadata stream matches with the version |
| 436 | * of the metadata stream in the kernel. If it was updated, set the reset flag |
| 437 | * on the stream. |
| 438 | */ |
| 439 | static |
| 440 | int metadata_stream_check_version(struct lttng_consumer_stream *stream, |
| 441 | const struct stream_subbuffer *subbuffer) |
| 442 | { |
| 443 | if (stream->metadata_version == subbuffer->info.metadata.version) { |
| 444 | goto end; |
| 445 | } |
| 446 | |
| 447 | DBG("New metadata version detected"); |
| 448 | consumer_stream_metadata_set_version(stream, |
| 449 | subbuffer->info.metadata.version); |
| 450 | |
| 451 | if (stream->read_subbuffer_ops.reset_metadata) { |
| 452 | stream->read_subbuffer_ops.reset_metadata(stream); |
| 453 | } |
| 454 | |
| 455 | end: |
| 456 | return 0; |
| 457 | } |
| 458 | |
| 459 | struct lttng_consumer_stream *consumer_stream_create( |
| 460 | struct lttng_consumer_channel *channel, |
| 461 | uint64_t channel_key, |
| 462 | uint64_t stream_key, |
| 463 | const char *channel_name, |
| 464 | uint64_t relayd_id, |
| 465 | uint64_t session_id, |
| 466 | struct lttng_trace_chunk *trace_chunk, |
| 467 | int cpu, |
| 468 | int *alloc_ret, |
| 469 | enum consumer_channel_type type, |
| 470 | unsigned int monitor) |
| 471 | { |
| 472 | int ret; |
| 473 | struct lttng_consumer_stream *stream; |
| 474 | |
| 475 | stream = zmalloc(sizeof(*stream)); |
| 476 | if (stream == NULL) { |
| 477 | PERROR("malloc struct lttng_consumer_stream"); |
| 478 | ret = -ENOMEM; |
| 479 | goto end; |
| 480 | } |
| 481 | |
| 482 | rcu_read_lock(); |
| 483 | |
| 484 | if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) { |
| 485 | ERR("Failed to acquire trace chunk reference during the creation of a stream"); |
| 486 | ret = -1; |
| 487 | goto error; |
| 488 | } |
| 489 | |
| 490 | stream->chan = channel; |
| 491 | stream->key = stream_key; |
| 492 | stream->trace_chunk = trace_chunk; |
| 493 | stream->out_fd = -1; |
| 494 | stream->out_fd_offset = 0; |
| 495 | stream->output_written = 0; |
| 496 | stream->net_seq_idx = relayd_id; |
| 497 | stream->session_id = session_id; |
| 498 | stream->monitor = monitor; |
| 499 | stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE; |
| 500 | stream->index_file = NULL; |
| 501 | stream->last_sequence_number = -1ULL; |
| 502 | stream->rotate_position = -1ULL; |
| 503 | /* Buffer is created with an open packet. */ |
| 504 | stream->opened_packet_in_current_trace_chunk = true; |
| 505 | pthread_mutex_init(&stream->lock, NULL); |
| 506 | pthread_mutex_init(&stream->metadata_timer_lock, NULL); |
| 507 | |
| 508 | /* If channel is the metadata, flag this stream as metadata. */ |
| 509 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { |
| 510 | stream->metadata_flag = 1; |
| 511 | /* Metadata is flat out. */ |
| 512 | strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name)); |
| 513 | /* Live rendez-vous point. */ |
| 514 | pthread_cond_init(&stream->metadata_rdv, NULL); |
| 515 | pthread_mutex_init(&stream->metadata_rdv_lock, NULL); |
| 516 | } else { |
| 517 | /* Format stream name to <channel_name>_<cpu_number> */ |
| 518 | ret = snprintf(stream->name, sizeof(stream->name), "%s_%d", |
| 519 | channel_name, cpu); |
| 520 | if (ret < 0) { |
| 521 | PERROR("snprintf stream name"); |
| 522 | goto error; |
| 523 | } |
| 524 | } |
| 525 | |
| 526 | switch (channel->output) { |
| 527 | case CONSUMER_CHANNEL_SPLICE: |
| 528 | stream->output = LTTNG_EVENT_SPLICE; |
| 529 | ret = utils_create_pipe(stream->splice_pipe); |
| 530 | if (ret < 0) { |
| 531 | goto error; |
| 532 | } |
| 533 | break; |
| 534 | case CONSUMER_CHANNEL_MMAP: |
| 535 | stream->output = LTTNG_EVENT_MMAP; |
| 536 | break; |
| 537 | default: |
| 538 | abort(); |
| 539 | } |
| 540 | |
| 541 | /* Key is always the wait_fd for streams. */ |
| 542 | lttng_ht_node_init_u64(&stream->node, stream->key); |
| 543 | |
| 544 | /* Init node per channel id key */ |
| 545 | lttng_ht_node_init_u64(&stream->node_channel_id, channel_key); |
| 546 | |
| 547 | /* Init session id node with the stream session id */ |
| 548 | lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id); |
| 549 | |
| 550 | DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64 |
| 551 | " relayd_id %" PRIu64 ", session_id %" PRIu64, |
| 552 | stream->name, stream->key, channel_key, |
| 553 | stream->net_seq_idx, stream->session_id); |
| 554 | |
| 555 | rcu_read_unlock(); |
| 556 | |
| 557 | if (type == CONSUMER_CHANNEL_TYPE_METADATA) { |
| 558 | stream->read_subbuffer_ops.lock = |
| 559 | consumer_stream_metadata_lock_all; |
| 560 | stream->read_subbuffer_ops.unlock = |
| 561 | consumer_stream_metadata_unlock_all; |
| 562 | stream->read_subbuffer_ops.pre_consume_subbuffer = |
| 563 | metadata_stream_check_version; |
| 564 | } else { |
| 565 | stream->read_subbuffer_ops.lock = consumer_stream_data_lock_all; |
| 566 | stream->read_subbuffer_ops.unlock = |
| 567 | consumer_stream_data_unlock_all; |
| 568 | stream->read_subbuffer_ops.pre_consume_subbuffer = |
| 569 | consumer_stream_update_stats; |
| 570 | if (channel->is_live) { |
| 571 | stream->read_subbuffer_ops.post_consume = |
| 572 | consumer_stream_sync_metadata_index; |
| 573 | } else { |
| 574 | stream->read_subbuffer_ops.post_consume = |
| 575 | consumer_stream_send_index; |
| 576 | } |
| 577 | } |
| 578 | |
| 579 | if (channel->output == CONSUMER_CHANNEL_MMAP) { |
| 580 | stream->read_subbuffer_ops.consume_subbuffer = |
| 581 | consumer_stream_consume_mmap; |
| 582 | } else { |
| 583 | stream->read_subbuffer_ops.consume_subbuffer = |
| 584 | consumer_stream_consume_splice; |
| 585 | } |
| 586 | |
| 587 | return stream; |
| 588 | |
| 589 | error: |
| 590 | rcu_read_unlock(); |
| 591 | lttng_trace_chunk_put(stream->trace_chunk); |
| 592 | free(stream); |
| 593 | end: |
| 594 | if (alloc_ret) { |
| 595 | *alloc_ret = ret; |
| 596 | } |
| 597 | return NULL; |
| 598 | } |
| 599 | |
| 600 | /* |
| 601 | * Close stream on the relayd side. This call can destroy a relayd if the |
| 602 | * conditions are met. |
| 603 | * |
| 604 | * A RCU read side lock MUST be acquired if the relayd object was looked up in |
| 605 | * a hash table before calling this. |
| 606 | */ |
| 607 | void consumer_stream_relayd_close(struct lttng_consumer_stream *stream, |
| 608 | struct consumer_relayd_sock_pair *relayd) |
| 609 | { |
| 610 | int ret; |
| 611 | |
| 612 | assert(stream); |
| 613 | assert(relayd); |
| 614 | |
| 615 | if (stream->sent_to_relayd) { |
| 616 | uatomic_dec(&relayd->refcount); |
| 617 | assert(uatomic_read(&relayd->refcount) >= 0); |
| 618 | } |
| 619 | |
| 620 | /* Closing streams requires to lock the control socket. */ |
| 621 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 622 | ret = relayd_send_close_stream(&relayd->control_sock, |
| 623 | stream->relayd_stream_id, |
| 624 | stream->next_net_seq_num - 1); |
| 625 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 626 | if (ret < 0) { |
| 627 | ERR("Relayd send close stream failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx); |
| 628 | lttng_consumer_cleanup_relayd(relayd); |
| 629 | } |
| 630 | |
| 631 | /* Both conditions are met, we destroy the relayd. */ |
| 632 | if (uatomic_read(&relayd->refcount) == 0 && |
| 633 | uatomic_read(&relayd->destroy_flag)) { |
| 634 | consumer_destroy_relayd(relayd); |
| 635 | } |
| 636 | stream->net_seq_idx = (uint64_t) -1ULL; |
| 637 | stream->sent_to_relayd = 0; |
| 638 | } |
| 639 | |
| 640 | /* |
| 641 | * Close stream's file descriptors and, if needed, close stream also on the |
| 642 | * relayd side. |
| 643 | * |
| 644 | * The consumer data lock MUST be acquired. |
| 645 | * The stream lock MUST be acquired. |
| 646 | */ |
| 647 | void consumer_stream_close(struct lttng_consumer_stream *stream) |
| 648 | { |
| 649 | int ret; |
| 650 | struct consumer_relayd_sock_pair *relayd; |
| 651 | |
| 652 | assert(stream); |
| 653 | |
| 654 | switch (consumer_data.type) { |
| 655 | case LTTNG_CONSUMER_KERNEL: |
| 656 | if (stream->mmap_base != NULL) { |
| 657 | ret = munmap(stream->mmap_base, stream->mmap_len); |
| 658 | if (ret != 0) { |
| 659 | PERROR("munmap"); |
| 660 | } |
| 661 | } |
| 662 | |
| 663 | if (stream->wait_fd >= 0) { |
| 664 | ret = close(stream->wait_fd); |
| 665 | if (ret) { |
| 666 | PERROR("close"); |
| 667 | } |
| 668 | stream->wait_fd = -1; |
| 669 | } |
| 670 | if (stream->chan->output == CONSUMER_CHANNEL_SPLICE) { |
| 671 | utils_close_pipe(stream->splice_pipe); |
| 672 | } |
| 673 | break; |
| 674 | case LTTNG_CONSUMER32_UST: |
| 675 | case LTTNG_CONSUMER64_UST: |
| 676 | { |
| 677 | /* |
| 678 | * Special case for the metadata since the wait fd is an internal pipe |
| 679 | * polled in the metadata thread. |
| 680 | */ |
| 681 | if (stream->metadata_flag && stream->chan->monitor) { |
| 682 | int rpipe = stream->ust_metadata_poll_pipe[0]; |
| 683 | |
| 684 | /* |
| 685 | * This will stop the channel timer if one and close the write side |
| 686 | * of the metadata poll pipe. |
| 687 | */ |
| 688 | lttng_ustconsumer_close_metadata(stream->chan); |
| 689 | if (rpipe >= 0) { |
| 690 | ret = close(rpipe); |
| 691 | if (ret < 0) { |
| 692 | PERROR("closing metadata pipe read side"); |
| 693 | } |
| 694 | stream->ust_metadata_poll_pipe[0] = -1; |
| 695 | } |
| 696 | } |
| 697 | break; |
| 698 | } |
| 699 | default: |
| 700 | ERR("Unknown consumer_data type"); |
| 701 | assert(0); |
| 702 | } |
| 703 | |
| 704 | /* Close output fd. Could be a socket or local file at this point. */ |
| 705 | if (stream->out_fd >= 0) { |
| 706 | ret = close(stream->out_fd); |
| 707 | if (ret) { |
| 708 | PERROR("close"); |
| 709 | } |
| 710 | stream->out_fd = -1; |
| 711 | } |
| 712 | |
| 713 | if (stream->index_file) { |
| 714 | lttng_index_file_put(stream->index_file); |
| 715 | stream->index_file = NULL; |
| 716 | } |
| 717 | |
| 718 | lttng_trace_chunk_put(stream->trace_chunk); |
| 719 | stream->trace_chunk = NULL; |
| 720 | |
| 721 | /* Check and cleanup relayd if needed. */ |
| 722 | rcu_read_lock(); |
| 723 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 724 | if (relayd != NULL) { |
| 725 | consumer_stream_relayd_close(stream, relayd); |
| 726 | } |
| 727 | rcu_read_unlock(); |
| 728 | } |
| 729 | |
| 730 | /* |
| 731 | * Delete the stream from all possible hash tables. |
| 732 | * |
| 733 | * The consumer data lock MUST be acquired. |
| 734 | * The stream lock MUST be acquired. |
| 735 | */ |
| 736 | void consumer_stream_delete(struct lttng_consumer_stream *stream, |
| 737 | struct lttng_ht *ht) |
| 738 | { |
| 739 | int ret; |
| 740 | struct lttng_ht_iter iter; |
| 741 | |
| 742 | assert(stream); |
| 743 | /* Should NEVER be called not in monitor mode. */ |
| 744 | assert(stream->chan->monitor); |
| 745 | |
| 746 | rcu_read_lock(); |
| 747 | |
| 748 | if (ht) { |
| 749 | iter.iter.node = &stream->node.node; |
| 750 | ret = lttng_ht_del(ht, &iter); |
| 751 | assert(!ret); |
| 752 | } |
| 753 | |
| 754 | /* Delete from stream per channel ID hash table. */ |
| 755 | iter.iter.node = &stream->node_channel_id.node; |
| 756 | /* |
| 757 | * The returned value is of no importance. Even if the node is NOT in the |
| 758 | * hash table, we continue since we may have been called by a code path |
| 759 | * that did not add the stream to a (all) hash table. Same goes for the |
| 760 | * next call ht del call. |
| 761 | */ |
| 762 | (void) lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter); |
| 763 | |
| 764 | /* Delete from the global stream list. */ |
| 765 | iter.iter.node = &stream->node_session_id.node; |
| 766 | /* See the previous ht del on why we ignore the returned value. */ |
| 767 | (void) lttng_ht_del(consumer_data.stream_list_ht, &iter); |
| 768 | |
| 769 | rcu_read_unlock(); |
| 770 | |
| 771 | if (!stream->metadata_flag) { |
| 772 | /* Decrement the stream count of the global consumer data. */ |
| 773 | assert(consumer_data.stream_count > 0); |
| 774 | consumer_data.stream_count--; |
| 775 | } |
| 776 | } |
| 777 | |
| 778 | /* |
| 779 | * Free the given stream within a RCU call. |
| 780 | */ |
| 781 | void consumer_stream_free(struct lttng_consumer_stream *stream) |
| 782 | { |
| 783 | assert(stream); |
| 784 | |
| 785 | metadata_bucket_destroy(stream->metadata_bucket); |
| 786 | call_rcu(&stream->node.head, free_stream_rcu); |
| 787 | } |
| 788 | |
| 789 | /* |
| 790 | * Destroy the stream's buffers of the tracer. |
| 791 | */ |
| 792 | void consumer_stream_destroy_buffers(struct lttng_consumer_stream *stream) |
| 793 | { |
| 794 | assert(stream); |
| 795 | |
| 796 | switch (consumer_data.type) { |
| 797 | case LTTNG_CONSUMER_KERNEL: |
| 798 | break; |
| 799 | case LTTNG_CONSUMER32_UST: |
| 800 | case LTTNG_CONSUMER64_UST: |
| 801 | lttng_ustconsumer_del_stream(stream); |
| 802 | break; |
| 803 | default: |
| 804 | ERR("Unknown consumer_data type"); |
| 805 | assert(0); |
| 806 | } |
| 807 | } |
| 808 | |
| 809 | /* |
| 810 | * Destroy and close a already created stream. |
| 811 | */ |
| 812 | static void destroy_close_stream(struct lttng_consumer_stream *stream) |
| 813 | { |
| 814 | assert(stream); |
| 815 | |
| 816 | DBG("Consumer stream destroy monitored key: %" PRIu64, stream->key); |
| 817 | |
| 818 | /* Destroy tracer buffers of the stream. */ |
| 819 | consumer_stream_destroy_buffers(stream); |
| 820 | /* Close down everything including the relayd if one. */ |
| 821 | consumer_stream_close(stream); |
| 822 | } |
| 823 | |
| 824 | /* |
| 825 | * Decrement the stream's channel refcount and if down to 0, return the channel |
| 826 | * pointer so it can be destroyed by the caller or NULL if not. |
| 827 | */ |
| 828 | static struct lttng_consumer_channel *unref_channel( |
| 829 | struct lttng_consumer_stream *stream) |
| 830 | { |
| 831 | struct lttng_consumer_channel *free_chan = NULL; |
| 832 | |
| 833 | assert(stream); |
| 834 | assert(stream->chan); |
| 835 | |
| 836 | /* Update refcount of channel and see if we need to destroy it. */ |
| 837 | if (!uatomic_sub_return(&stream->chan->refcount, 1) |
| 838 | && !uatomic_read(&stream->chan->nb_init_stream_left)) { |
| 839 | free_chan = stream->chan; |
| 840 | } |
| 841 | |
| 842 | return free_chan; |
| 843 | } |
| 844 | |
| 845 | /* |
| 846 | * Destroy a stream completely. This will delete, close and free the stream. |
| 847 | * Once return, the stream is NO longer usable. Its channel may get destroyed |
| 848 | * if conditions are met for a monitored stream. |
| 849 | * |
| 850 | * This MUST be called WITHOUT the consumer data and stream lock acquired if |
| 851 | * the stream is in _monitor_ mode else it does not matter. |
| 852 | */ |
| 853 | void consumer_stream_destroy(struct lttng_consumer_stream *stream, |
| 854 | struct lttng_ht *ht) |
| 855 | { |
| 856 | assert(stream); |
| 857 | |
| 858 | /* Stream is in monitor mode. */ |
| 859 | if (stream->monitor) { |
| 860 | struct lttng_consumer_channel *free_chan = NULL; |
| 861 | |
| 862 | /* |
| 863 | * This means that the stream was successfully removed from the streams |
| 864 | * list of the channel and sent to the right thread managing this |
| 865 | * stream thus being globally visible. |
| 866 | */ |
| 867 | if (stream->globally_visible) { |
| 868 | pthread_mutex_lock(&consumer_data.lock); |
| 869 | pthread_mutex_lock(&stream->chan->lock); |
| 870 | pthread_mutex_lock(&stream->lock); |
| 871 | /* Remove every reference of the stream in the consumer. */ |
| 872 | consumer_stream_delete(stream, ht); |
| 873 | |
| 874 | destroy_close_stream(stream); |
| 875 | |
| 876 | /* Update channel's refcount of the stream. */ |
| 877 | free_chan = unref_channel(stream); |
| 878 | |
| 879 | /* Indicates that the consumer data state MUST be updated after this. */ |
| 880 | consumer_data.need_update = 1; |
| 881 | |
| 882 | pthread_mutex_unlock(&stream->lock); |
| 883 | pthread_mutex_unlock(&stream->chan->lock); |
| 884 | pthread_mutex_unlock(&consumer_data.lock); |
| 885 | } else { |
| 886 | /* |
| 887 | * If the stream is not visible globally, this needs to be done |
| 888 | * outside of the consumer data lock section. |
| 889 | */ |
| 890 | free_chan = unref_channel(stream); |
| 891 | } |
| 892 | |
| 893 | if (free_chan) { |
| 894 | consumer_del_channel(free_chan); |
| 895 | } |
| 896 | } else { |
| 897 | destroy_close_stream(stream); |
| 898 | } |
| 899 | |
| 900 | /* Free stream within a RCU call. */ |
| 901 | lttng_trace_chunk_put(stream->trace_chunk); |
| 902 | stream->trace_chunk = NULL; |
| 903 | consumer_stream_free(stream); |
| 904 | } |
| 905 | |
| 906 | /* |
| 907 | * Write index of a specific stream either on the relayd or local disk. |
| 908 | * |
| 909 | * Return 0 on success or else a negative value. |
| 910 | */ |
| 911 | int consumer_stream_write_index(struct lttng_consumer_stream *stream, |
| 912 | struct ctf_packet_index *element) |
| 913 | { |
| 914 | int ret; |
| 915 | |
| 916 | assert(stream); |
| 917 | assert(element); |
| 918 | |
| 919 | rcu_read_lock(); |
| 920 | if (stream->net_seq_idx != (uint64_t) -1ULL) { |
| 921 | struct consumer_relayd_sock_pair *relayd; |
| 922 | relayd = consumer_find_relayd(stream->net_seq_idx); |
| 923 | if (relayd) { |
| 924 | pthread_mutex_lock(&relayd->ctrl_sock_mutex); |
| 925 | ret = relayd_send_index(&relayd->control_sock, element, |
| 926 | stream->relayd_stream_id, stream->next_net_seq_num - 1); |
| 927 | if (ret < 0) { |
| 928 | /* |
| 929 | * Communication error with lttng-relayd, |
| 930 | * perform cleanup now |
| 931 | */ |
| 932 | ERR("Relayd send index failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx); |
| 933 | lttng_consumer_cleanup_relayd(relayd); |
| 934 | ret = -1; |
| 935 | } |
| 936 | pthread_mutex_unlock(&relayd->ctrl_sock_mutex); |
| 937 | } else { |
| 938 | ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't write index.", |
| 939 | stream->key, stream->net_seq_idx); |
| 940 | ret = -1; |
| 941 | } |
| 942 | } else { |
| 943 | if (lttng_index_file_write(stream->index_file, element)) { |
| 944 | ret = -1; |
| 945 | } else { |
| 946 | ret = 0; |
| 947 | } |
| 948 | } |
| 949 | if (ret < 0) { |
| 950 | goto error; |
| 951 | } |
| 952 | |
| 953 | error: |
| 954 | rcu_read_unlock(); |
| 955 | return ret; |
| 956 | } |
| 957 | |
| 958 | int consumer_stream_create_output_files(struct lttng_consumer_stream *stream, |
| 959 | bool create_index) |
| 960 | { |
| 961 | int ret; |
| 962 | enum lttng_trace_chunk_status chunk_status; |
| 963 | const int flags = O_WRONLY | O_CREAT | O_TRUNC; |
| 964 | const mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP; |
| 965 | char stream_path[LTTNG_PATH_MAX]; |
| 966 | |
| 967 | ASSERT_LOCKED(stream->lock); |
| 968 | assert(stream->trace_chunk); |
| 969 | |
| 970 | ret = utils_stream_file_path(stream->chan->pathname, stream->name, |
| 971 | stream->chan->tracefile_size, |
| 972 | stream->tracefile_count_current, NULL, |
| 973 | stream_path, sizeof(stream_path)); |
| 974 | if (ret < 0) { |
| 975 | goto end; |
| 976 | } |
| 977 | |
| 978 | if (stream->out_fd >= 0) { |
| 979 | ret = close(stream->out_fd); |
| 980 | if (ret < 0) { |
| 981 | PERROR("Failed to close stream file \"%s\"", |
| 982 | stream->name); |
| 983 | goto end; |
| 984 | } |
| 985 | stream->out_fd = -1; |
| 986 | } |
| 987 | |
| 988 | DBG("Opening stream output file \"%s\"", stream_path); |
| 989 | chunk_status = lttng_trace_chunk_open_file(stream->trace_chunk, stream_path, |
| 990 | flags, mode, &stream->out_fd, false); |
| 991 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 992 | ERR("Failed to open stream file \"%s\"", stream->name); |
| 993 | ret = -1; |
| 994 | goto end; |
| 995 | } |
| 996 | |
| 997 | if (!stream->metadata_flag && (create_index || stream->index_file)) { |
| 998 | if (stream->index_file) { |
| 999 | lttng_index_file_put(stream->index_file); |
| 1000 | } |
| 1001 | chunk_status = lttng_index_file_create_from_trace_chunk( |
| 1002 | stream->trace_chunk, |
| 1003 | stream->chan->pathname, |
| 1004 | stream->name, |
| 1005 | stream->chan->tracefile_size, |
| 1006 | stream->tracefile_count_current, |
| 1007 | CTF_INDEX_MAJOR, CTF_INDEX_MINOR, |
| 1008 | false, &stream->index_file); |
| 1009 | if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) { |
| 1010 | ret = -1; |
| 1011 | goto end; |
| 1012 | } |
| 1013 | } |
| 1014 | |
| 1015 | /* Reset current size because we just perform a rotation. */ |
| 1016 | stream->tracefile_size_current = 0; |
| 1017 | stream->out_fd_offset = 0; |
| 1018 | end: |
| 1019 | return ret; |
| 1020 | } |
| 1021 | |
| 1022 | int consumer_stream_rotate_output_files(struct lttng_consumer_stream *stream) |
| 1023 | { |
| 1024 | int ret; |
| 1025 | |
| 1026 | stream->tracefile_count_current++; |
| 1027 | if (stream->chan->tracefile_count > 0) { |
| 1028 | stream->tracefile_count_current %= |
| 1029 | stream->chan->tracefile_count; |
| 1030 | } |
| 1031 | |
| 1032 | DBG("Rotating output files of stream \"%s\"", stream->name); |
| 1033 | ret = consumer_stream_create_output_files(stream, true); |
| 1034 | if (ret) { |
| 1035 | goto end; |
| 1036 | } |
| 1037 | |
| 1038 | end: |
| 1039 | return ret; |
| 1040 | } |
| 1041 | |
| 1042 | bool consumer_stream_is_deleted(struct lttng_consumer_stream *stream) |
| 1043 | { |
| 1044 | /* |
| 1045 | * This function does not take a const stream since |
| 1046 | * cds_lfht_is_node_deleted was not const before liburcu 0.12. |
| 1047 | */ |
| 1048 | assert(stream); |
| 1049 | return cds_lfht_is_node_deleted(&stream->node.node); |
| 1050 | } |
| 1051 | |
| 1052 | static ssize_t metadata_bucket_flush( |
| 1053 | const struct stream_subbuffer *buffer, void *data) |
| 1054 | { |
| 1055 | ssize_t ret; |
| 1056 | struct lttng_consumer_stream *stream = data; |
| 1057 | |
| 1058 | ret = consumer_stream_consume_mmap(NULL, stream, buffer); |
| 1059 | if (ret < 0) { |
| 1060 | goto end; |
| 1061 | } |
| 1062 | end: |
| 1063 | return ret; |
| 1064 | } |
| 1065 | |
| 1066 | static ssize_t metadata_bucket_consume( |
| 1067 | struct lttng_consumer_local_data *unused, |
| 1068 | struct lttng_consumer_stream *stream, |
| 1069 | const struct stream_subbuffer *subbuffer) |
| 1070 | { |
| 1071 | ssize_t ret; |
| 1072 | enum metadata_bucket_status status; |
| 1073 | |
| 1074 | status = metadata_bucket_fill(stream->metadata_bucket, subbuffer); |
| 1075 | switch (status) { |
| 1076 | case METADATA_BUCKET_STATUS_OK: |
| 1077 | /* Return consumed size. */ |
| 1078 | ret = subbuffer->buffer.buffer.size; |
| 1079 | break; |
| 1080 | default: |
| 1081 | ret = -1; |
| 1082 | } |
| 1083 | |
| 1084 | return ret; |
| 1085 | } |
| 1086 | |
| 1087 | int consumer_stream_enable_metadata_bucketization( |
| 1088 | struct lttng_consumer_stream *stream) |
| 1089 | { |
| 1090 | int ret = 0; |
| 1091 | |
| 1092 | assert(stream->metadata_flag); |
| 1093 | assert(!stream->metadata_bucket); |
| 1094 | assert(stream->chan->output == CONSUMER_CHANNEL_MMAP); |
| 1095 | |
| 1096 | stream->metadata_bucket = metadata_bucket_create( |
| 1097 | metadata_bucket_flush, stream); |
| 1098 | if (!stream->metadata_bucket) { |
| 1099 | ret = -1; |
| 1100 | goto end; |
| 1101 | } |
| 1102 | |
| 1103 | stream->read_subbuffer_ops.consume_subbuffer = metadata_bucket_consume; |
| 1104 | end: |
| 1105 | return ret; |
| 1106 | } |
| 1107 | |
| 1108 | void consumer_stream_metadata_set_version( |
| 1109 | struct lttng_consumer_stream *stream, uint64_t new_version) |
| 1110 | { |
| 1111 | assert(new_version > stream->metadata_version); |
| 1112 | stream->metadata_version = new_version; |
| 1113 | stream->reset_metadata_flag = 1; |
| 1114 | |
| 1115 | if (stream->metadata_bucket) { |
| 1116 | metadata_bucket_reset(stream->metadata_bucket); |
| 1117 | } |
| 1118 | } |