| 1 | /* |
| 2 | * Copyright (C) 2012 - Julien Desfossez <julien.desfossez@efficios.com> |
| 3 | * David Goulet <dgoulet@efficios.com> |
| 4 | * |
| 5 | * This program is free software; you can redistribute it and/or modify it |
| 6 | * under the terms of the GNU General Public License, version 2 only, as |
| 7 | * published by the Free Software Foundation. |
| 8 | * |
| 9 | * This program is distributed in the hope that it will be useful, but WITHOUT |
| 10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 12 | * more details. |
| 13 | * |
| 14 | * You should have received a copy of the GNU General Public License along with |
| 15 | * this program; if not, write to the Free Software Foundation, Inc., 51 |
| 16 | * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 17 | */ |
| 18 | |
| 19 | #define _LGPL_SOURCE |
| 20 | #include <assert.h> |
| 21 | #include <inttypes.h> |
| 22 | #include <signal.h> |
| 23 | |
| 24 | #include <bin/lttng-sessiond/ust-ctl.h> |
| 25 | #include <bin/lttng-consumerd/health-consumerd.h> |
| 26 | #include <common/common.h> |
| 27 | #include <common/compat/endian.h> |
| 28 | #include <common/kernel-ctl/kernel-ctl.h> |
| 29 | #include <common/kernel-consumer/kernel-consumer.h> |
| 30 | #include <common/consumer/consumer-stream.h> |
| 31 | #include <common/consumer/consumer-timer.h> |
| 32 | #include <common/consumer/consumer-testpoint.h> |
| 33 | #include <common/ust-consumer/ust-consumer.h> |
| 34 | |
| 35 | typedef int (*sample_positions_cb)(struct lttng_consumer_stream *stream); |
| 36 | typedef int (*get_consumed_cb)(struct lttng_consumer_stream *stream, |
| 37 | unsigned long *consumed); |
| 38 | typedef int (*get_produced_cb)(struct lttng_consumer_stream *stream, |
| 39 | unsigned long *produced); |
| 40 | |
| 41 | static struct timer_signal_data timer_signal = { |
| 42 | .tid = 0, |
| 43 | .setup_done = 0, |
| 44 | .qs_done = 0, |
| 45 | .lock = PTHREAD_MUTEX_INITIALIZER, |
| 46 | }; |
| 47 | |
| 48 | /* |
| 49 | * Set custom signal mask to current thread. |
| 50 | */ |
| 51 | static void setmask(sigset_t *mask) |
| 52 | { |
| 53 | int ret; |
| 54 | |
| 55 | ret = sigemptyset(mask); |
| 56 | if (ret) { |
| 57 | PERROR("sigemptyset"); |
| 58 | } |
| 59 | ret = sigaddset(mask, LTTNG_CONSUMER_SIG_SWITCH); |
| 60 | if (ret) { |
| 61 | PERROR("sigaddset switch"); |
| 62 | } |
| 63 | ret = sigaddset(mask, LTTNG_CONSUMER_SIG_TEARDOWN); |
| 64 | if (ret) { |
| 65 | PERROR("sigaddset teardown"); |
| 66 | } |
| 67 | ret = sigaddset(mask, LTTNG_CONSUMER_SIG_LIVE); |
| 68 | if (ret) { |
| 69 | PERROR("sigaddset live"); |
| 70 | } |
| 71 | ret = sigaddset(mask, LTTNG_CONSUMER_SIG_MONITOR); |
| 72 | if (ret) { |
| 73 | PERROR("sigaddset monitor"); |
| 74 | } |
| 75 | ret = sigaddset(mask, LTTNG_CONSUMER_SIG_EXIT); |
| 76 | if (ret) { |
| 77 | PERROR("sigaddset exit"); |
| 78 | } |
| 79 | } |
| 80 | |
| 81 | static int channel_monitor_pipe = -1; |
| 82 | |
| 83 | /* |
| 84 | * Execute action on a timer switch. |
| 85 | * |
| 86 | * Beware: metadata_switch_timer() should *never* take a mutex also held |
| 87 | * while consumer_timer_switch_stop() is called. It would result in |
| 88 | * deadlocks. |
| 89 | */ |
| 90 | static void metadata_switch_timer(struct lttng_consumer_local_data *ctx, |
| 91 | int sig, siginfo_t *si) |
| 92 | { |
| 93 | int ret; |
| 94 | struct lttng_consumer_channel *channel; |
| 95 | |
| 96 | channel = si->si_value.sival_ptr; |
| 97 | assert(channel); |
| 98 | |
| 99 | if (channel->switch_timer_error) { |
| 100 | return; |
| 101 | } |
| 102 | |
| 103 | DBG("Switch timer for channel %" PRIu64, channel->key); |
| 104 | switch (ctx->type) { |
| 105 | case LTTNG_CONSUMER32_UST: |
| 106 | case LTTNG_CONSUMER64_UST: |
| 107 | /* |
| 108 | * Locks taken by lttng_ustconsumer_request_metadata(): |
| 109 | * - metadata_socket_lock |
| 110 | * - Calling lttng_ustconsumer_recv_metadata(): |
| 111 | * - channel->metadata_cache->lock |
| 112 | * - Calling consumer_metadata_cache_flushed(): |
| 113 | * - channel->timer_lock |
| 114 | * - channel->metadata_cache->lock |
| 115 | * |
| 116 | * Ensure that neither consumer_data.lock nor |
| 117 | * channel->lock are taken within this function, since |
| 118 | * they are held while consumer_timer_switch_stop() is |
| 119 | * called. |
| 120 | */ |
| 121 | ret = lttng_ustconsumer_request_metadata(ctx, channel, 1, 1); |
| 122 | if (ret < 0) { |
| 123 | channel->switch_timer_error = 1; |
| 124 | } |
| 125 | break; |
| 126 | case LTTNG_CONSUMER_KERNEL: |
| 127 | case LTTNG_CONSUMER_UNKNOWN: |
| 128 | assert(0); |
| 129 | break; |
| 130 | } |
| 131 | } |
| 132 | |
| 133 | static int send_empty_index(struct lttng_consumer_stream *stream, uint64_t ts, |
| 134 | uint64_t stream_id) |
| 135 | { |
| 136 | int ret; |
| 137 | struct ctf_packet_index index; |
| 138 | |
| 139 | memset(&index, 0, sizeof(index)); |
| 140 | index.stream_id = htobe64(stream_id); |
| 141 | index.timestamp_end = htobe64(ts); |
| 142 | ret = consumer_stream_write_index(stream, &index); |
| 143 | if (ret < 0) { |
| 144 | goto error; |
| 145 | } |
| 146 | |
| 147 | error: |
| 148 | return ret; |
| 149 | } |
| 150 | |
| 151 | int consumer_flush_kernel_index(struct lttng_consumer_stream *stream) |
| 152 | { |
| 153 | uint64_t ts, stream_id; |
| 154 | int ret; |
| 155 | |
| 156 | ret = kernctl_get_current_timestamp(stream->wait_fd, &ts); |
| 157 | if (ret < 0) { |
| 158 | ERR("Failed to get the current timestamp"); |
| 159 | goto end; |
| 160 | } |
| 161 | ret = kernctl_buffer_flush(stream->wait_fd); |
| 162 | if (ret < 0) { |
| 163 | ERR("Failed to flush kernel stream"); |
| 164 | goto end; |
| 165 | } |
| 166 | ret = kernctl_snapshot(stream->wait_fd); |
| 167 | if (ret < 0) { |
| 168 | if (ret != -EAGAIN && ret != -ENODATA) { |
| 169 | PERROR("live timer kernel snapshot"); |
| 170 | ret = -1; |
| 171 | goto end; |
| 172 | } |
| 173 | ret = kernctl_get_stream_id(stream->wait_fd, &stream_id); |
| 174 | if (ret < 0) { |
| 175 | PERROR("kernctl_get_stream_id"); |
| 176 | goto end; |
| 177 | } |
| 178 | DBG("Stream %" PRIu64 " empty, sending beacon", stream->key); |
| 179 | ret = send_empty_index(stream, ts, stream_id); |
| 180 | if (ret < 0) { |
| 181 | goto end; |
| 182 | } |
| 183 | } |
| 184 | ret = 0; |
| 185 | end: |
| 186 | return ret; |
| 187 | } |
| 188 | |
| 189 | static int check_kernel_stream(struct lttng_consumer_stream *stream) |
| 190 | { |
| 191 | int ret; |
| 192 | |
| 193 | /* |
| 194 | * While holding the stream mutex, try to take a snapshot, if it |
| 195 | * succeeds, it means that data is ready to be sent, just let the data |
| 196 | * thread handle that. Otherwise, if the snapshot returns EAGAIN, it |
| 197 | * means that there is no data to read after the flush, so we can |
| 198 | * safely send the empty index. |
| 199 | * |
| 200 | * Doing a trylock and checking if waiting on metadata if |
| 201 | * trylock fails. Bail out of the stream is indeed waiting for |
| 202 | * metadata to be pushed. Busy wait on trylock otherwise. |
| 203 | */ |
| 204 | for (;;) { |
| 205 | ret = pthread_mutex_trylock(&stream->lock); |
| 206 | switch (ret) { |
| 207 | case 0: |
| 208 | break; /* We have the lock. */ |
| 209 | case EBUSY: |
| 210 | pthread_mutex_lock(&stream->metadata_timer_lock); |
| 211 | if (stream->waiting_on_metadata) { |
| 212 | ret = 0; |
| 213 | stream->missed_metadata_flush = true; |
| 214 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 215 | goto end; /* Bail out. */ |
| 216 | } |
| 217 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 218 | /* Try again. */ |
| 219 | caa_cpu_relax(); |
| 220 | continue; |
| 221 | default: |
| 222 | ERR("Unexpected pthread_mutex_trylock error %d", ret); |
| 223 | ret = -1; |
| 224 | goto end; |
| 225 | } |
| 226 | break; |
| 227 | } |
| 228 | ret = consumer_flush_kernel_index(stream); |
| 229 | pthread_mutex_unlock(&stream->lock); |
| 230 | end: |
| 231 | return ret; |
| 232 | } |
| 233 | |
| 234 | int consumer_flush_ust_index(struct lttng_consumer_stream *stream) |
| 235 | { |
| 236 | uint64_t ts, stream_id; |
| 237 | int ret; |
| 238 | |
| 239 | ret = cds_lfht_is_node_deleted(&stream->node.node); |
| 240 | if (ret) { |
| 241 | goto end; |
| 242 | } |
| 243 | |
| 244 | ret = lttng_ustconsumer_get_current_timestamp(stream, &ts); |
| 245 | if (ret < 0) { |
| 246 | ERR("Failed to get the current timestamp"); |
| 247 | goto end; |
| 248 | } |
| 249 | lttng_ustconsumer_flush_buffer(stream, 1); |
| 250 | ret = lttng_ustconsumer_take_snapshot(stream); |
| 251 | if (ret < 0) { |
| 252 | if (ret != -EAGAIN) { |
| 253 | ERR("Taking UST snapshot"); |
| 254 | ret = -1; |
| 255 | goto end; |
| 256 | } |
| 257 | ret = lttng_ustconsumer_get_stream_id(stream, &stream_id); |
| 258 | if (ret < 0) { |
| 259 | PERROR("ustctl_get_stream_id"); |
| 260 | goto end; |
| 261 | } |
| 262 | DBG("Stream %" PRIu64 " empty, sending beacon", stream->key); |
| 263 | ret = send_empty_index(stream, ts, stream_id); |
| 264 | if (ret < 0) { |
| 265 | goto end; |
| 266 | } |
| 267 | } |
| 268 | ret = 0; |
| 269 | end: |
| 270 | return ret; |
| 271 | } |
| 272 | |
| 273 | static int check_ust_stream(struct lttng_consumer_stream *stream) |
| 274 | { |
| 275 | int ret; |
| 276 | |
| 277 | assert(stream); |
| 278 | assert(stream->ustream); |
| 279 | /* |
| 280 | * While holding the stream mutex, try to take a snapshot, if it |
| 281 | * succeeds, it means that data is ready to be sent, just let the data |
| 282 | * thread handle that. Otherwise, if the snapshot returns EAGAIN, it |
| 283 | * means that there is no data to read after the flush, so we can |
| 284 | * safely send the empty index. |
| 285 | * |
| 286 | * Doing a trylock and checking if waiting on metadata if |
| 287 | * trylock fails. Bail out of the stream is indeed waiting for |
| 288 | * metadata to be pushed. Busy wait on trylock otherwise. |
| 289 | */ |
| 290 | for (;;) { |
| 291 | ret = pthread_mutex_trylock(&stream->lock); |
| 292 | switch (ret) { |
| 293 | case 0: |
| 294 | break; /* We have the lock. */ |
| 295 | case EBUSY: |
| 296 | pthread_mutex_lock(&stream->metadata_timer_lock); |
| 297 | if (stream->waiting_on_metadata) { |
| 298 | ret = 0; |
| 299 | stream->missed_metadata_flush = true; |
| 300 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 301 | goto end; /* Bail out. */ |
| 302 | } |
| 303 | pthread_mutex_unlock(&stream->metadata_timer_lock); |
| 304 | /* Try again. */ |
| 305 | caa_cpu_relax(); |
| 306 | continue; |
| 307 | default: |
| 308 | ERR("Unexpected pthread_mutex_trylock error %d", ret); |
| 309 | ret = -1; |
| 310 | goto end; |
| 311 | } |
| 312 | break; |
| 313 | } |
| 314 | ret = consumer_flush_ust_index(stream); |
| 315 | pthread_mutex_unlock(&stream->lock); |
| 316 | end: |
| 317 | return ret; |
| 318 | } |
| 319 | |
| 320 | /* |
| 321 | * Execute action on a live timer |
| 322 | */ |
| 323 | static void live_timer(struct lttng_consumer_local_data *ctx, |
| 324 | int sig, siginfo_t *si) |
| 325 | { |
| 326 | int ret; |
| 327 | struct lttng_consumer_channel *channel; |
| 328 | struct lttng_consumer_stream *stream; |
| 329 | struct lttng_ht *ht; |
| 330 | struct lttng_ht_iter iter; |
| 331 | |
| 332 | channel = si->si_value.sival_ptr; |
| 333 | assert(channel); |
| 334 | |
| 335 | if (channel->switch_timer_error) { |
| 336 | goto error; |
| 337 | } |
| 338 | ht = consumer_data.stream_per_chan_id_ht; |
| 339 | |
| 340 | DBG("Live timer for channel %" PRIu64, channel->key); |
| 341 | |
| 342 | rcu_read_lock(); |
| 343 | switch (ctx->type) { |
| 344 | case LTTNG_CONSUMER32_UST: |
| 345 | case LTTNG_CONSUMER64_UST: |
| 346 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 347 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 348 | ht->match_fct, &channel->key, &iter.iter, |
| 349 | stream, node_channel_id.node) { |
| 350 | ret = check_ust_stream(stream); |
| 351 | if (ret < 0) { |
| 352 | goto error_unlock; |
| 353 | } |
| 354 | } |
| 355 | break; |
| 356 | case LTTNG_CONSUMER_KERNEL: |
| 357 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 358 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 359 | ht->match_fct, &channel->key, &iter.iter, |
| 360 | stream, node_channel_id.node) { |
| 361 | ret = check_kernel_stream(stream); |
| 362 | if (ret < 0) { |
| 363 | goto error_unlock; |
| 364 | } |
| 365 | } |
| 366 | break; |
| 367 | case LTTNG_CONSUMER_UNKNOWN: |
| 368 | assert(0); |
| 369 | break; |
| 370 | } |
| 371 | |
| 372 | error_unlock: |
| 373 | rcu_read_unlock(); |
| 374 | |
| 375 | error: |
| 376 | return; |
| 377 | } |
| 378 | |
| 379 | static |
| 380 | void consumer_timer_signal_thread_qs(unsigned int signr) |
| 381 | { |
| 382 | sigset_t pending_set; |
| 383 | int ret; |
| 384 | |
| 385 | /* |
| 386 | * We need to be the only thread interacting with the thread |
| 387 | * that manages signals for teardown synchronization. |
| 388 | */ |
| 389 | pthread_mutex_lock(&timer_signal.lock); |
| 390 | |
| 391 | /* Ensure we don't have any signal queued for this channel. */ |
| 392 | for (;;) { |
| 393 | ret = sigemptyset(&pending_set); |
| 394 | if (ret == -1) { |
| 395 | PERROR("sigemptyset"); |
| 396 | } |
| 397 | ret = sigpending(&pending_set); |
| 398 | if (ret == -1) { |
| 399 | PERROR("sigpending"); |
| 400 | } |
| 401 | if (!sigismember(&pending_set, signr)) { |
| 402 | break; |
| 403 | } |
| 404 | caa_cpu_relax(); |
| 405 | } |
| 406 | |
| 407 | /* |
| 408 | * From this point, no new signal handler will be fired that would try to |
| 409 | * access "chan". However, we still need to wait for any currently |
| 410 | * executing handler to complete. |
| 411 | */ |
| 412 | cmm_smp_mb(); |
| 413 | CMM_STORE_SHARED(timer_signal.qs_done, 0); |
| 414 | cmm_smp_mb(); |
| 415 | |
| 416 | /* |
| 417 | * Kill with LTTNG_CONSUMER_SIG_TEARDOWN, so signal management thread wakes |
| 418 | * up. |
| 419 | */ |
| 420 | kill(getpid(), LTTNG_CONSUMER_SIG_TEARDOWN); |
| 421 | |
| 422 | while (!CMM_LOAD_SHARED(timer_signal.qs_done)) { |
| 423 | caa_cpu_relax(); |
| 424 | } |
| 425 | cmm_smp_mb(); |
| 426 | |
| 427 | pthread_mutex_unlock(&timer_signal.lock); |
| 428 | } |
| 429 | |
| 430 | /* |
| 431 | * Start a timer channel timer which will fire at a given interval |
| 432 | * (timer_interval_us)and fire a given signal (signal). |
| 433 | * |
| 434 | * Returns a negative value on error, 0 if a timer was created, and |
| 435 | * a positive value if no timer was created (not an error). |
| 436 | */ |
| 437 | static |
| 438 | int consumer_channel_timer_start(timer_t *timer_id, |
| 439 | struct lttng_consumer_channel *channel, |
| 440 | unsigned int timer_interval_us, int signal) |
| 441 | { |
| 442 | int ret = 0, delete_ret; |
| 443 | struct sigevent sev; |
| 444 | struct itimerspec its; |
| 445 | |
| 446 | assert(channel); |
| 447 | assert(channel->key); |
| 448 | |
| 449 | if (timer_interval_us == 0) { |
| 450 | /* No creation needed; not an error. */ |
| 451 | ret = 1; |
| 452 | goto end; |
| 453 | } |
| 454 | |
| 455 | sev.sigev_notify = SIGEV_SIGNAL; |
| 456 | sev.sigev_signo = signal; |
| 457 | sev.sigev_value.sival_ptr = channel; |
| 458 | ret = timer_create(CLOCKID, &sev, timer_id); |
| 459 | if (ret == -1) { |
| 460 | PERROR("timer_create"); |
| 461 | goto end; |
| 462 | } |
| 463 | |
| 464 | its.it_value.tv_sec = timer_interval_us / 1000000; |
| 465 | its.it_value.tv_nsec = (timer_interval_us % 1000000) * 1000; |
| 466 | its.it_interval.tv_sec = its.it_value.tv_sec; |
| 467 | its.it_interval.tv_nsec = its.it_value.tv_nsec; |
| 468 | |
| 469 | ret = timer_settime(*timer_id, 0, &its, NULL); |
| 470 | if (ret == -1) { |
| 471 | PERROR("timer_settime"); |
| 472 | goto error_destroy_timer; |
| 473 | } |
| 474 | end: |
| 475 | return ret; |
| 476 | error_destroy_timer: |
| 477 | delete_ret = timer_delete(*timer_id); |
| 478 | if (delete_ret == -1) { |
| 479 | PERROR("timer_delete"); |
| 480 | } |
| 481 | goto end; |
| 482 | } |
| 483 | |
| 484 | static |
| 485 | int consumer_channel_timer_stop(timer_t *timer_id, int signal) |
| 486 | { |
| 487 | int ret = 0; |
| 488 | |
| 489 | ret = timer_delete(*timer_id); |
| 490 | if (ret == -1) { |
| 491 | PERROR("timer_delete"); |
| 492 | goto end; |
| 493 | } |
| 494 | |
| 495 | consumer_timer_signal_thread_qs(signal); |
| 496 | *timer_id = 0; |
| 497 | end: |
| 498 | return ret; |
| 499 | } |
| 500 | |
| 501 | /* |
| 502 | * Set the channel's switch timer. |
| 503 | */ |
| 504 | void consumer_timer_switch_start(struct lttng_consumer_channel *channel, |
| 505 | unsigned int switch_timer_interval_us) |
| 506 | { |
| 507 | int ret; |
| 508 | |
| 509 | assert(channel); |
| 510 | assert(channel->key); |
| 511 | |
| 512 | ret = consumer_channel_timer_start(&channel->switch_timer, channel, |
| 513 | switch_timer_interval_us, LTTNG_CONSUMER_SIG_SWITCH); |
| 514 | |
| 515 | channel->switch_timer_enabled = !!(ret == 0); |
| 516 | } |
| 517 | |
| 518 | /* |
| 519 | * Stop and delete the channel's switch timer. |
| 520 | */ |
| 521 | void consumer_timer_switch_stop(struct lttng_consumer_channel *channel) |
| 522 | { |
| 523 | int ret; |
| 524 | |
| 525 | assert(channel); |
| 526 | |
| 527 | ret = consumer_channel_timer_stop(&channel->switch_timer, |
| 528 | LTTNG_CONSUMER_SIG_SWITCH); |
| 529 | if (ret == -1) { |
| 530 | ERR("Failed to stop switch timer"); |
| 531 | } |
| 532 | |
| 533 | channel->switch_timer_enabled = 0; |
| 534 | } |
| 535 | |
| 536 | /* |
| 537 | * Set the channel's live timer. |
| 538 | */ |
| 539 | void consumer_timer_live_start(struct lttng_consumer_channel *channel, |
| 540 | unsigned int live_timer_interval_us) |
| 541 | { |
| 542 | int ret; |
| 543 | |
| 544 | assert(channel); |
| 545 | assert(channel->key); |
| 546 | |
| 547 | ret = consumer_channel_timer_start(&channel->live_timer, channel, |
| 548 | live_timer_interval_us, LTTNG_CONSUMER_SIG_LIVE); |
| 549 | |
| 550 | channel->live_timer_enabled = !!(ret == 0); |
| 551 | } |
| 552 | |
| 553 | /* |
| 554 | * Stop and delete the channel's live timer. |
| 555 | */ |
| 556 | void consumer_timer_live_stop(struct lttng_consumer_channel *channel) |
| 557 | { |
| 558 | int ret; |
| 559 | |
| 560 | assert(channel); |
| 561 | |
| 562 | ret = consumer_channel_timer_stop(&channel->live_timer, |
| 563 | LTTNG_CONSUMER_SIG_LIVE); |
| 564 | if (ret == -1) { |
| 565 | ERR("Failed to stop live timer"); |
| 566 | } |
| 567 | |
| 568 | channel->live_timer_enabled = 0; |
| 569 | } |
| 570 | |
| 571 | /* |
| 572 | * Set the channel's monitoring timer. |
| 573 | * |
| 574 | * Returns a negative value on error, 0 if a timer was created, and |
| 575 | * a positive value if no timer was created (not an error). |
| 576 | */ |
| 577 | int consumer_timer_monitor_start(struct lttng_consumer_channel *channel, |
| 578 | unsigned int monitor_timer_interval_us) |
| 579 | { |
| 580 | int ret; |
| 581 | |
| 582 | assert(channel); |
| 583 | assert(channel->key); |
| 584 | assert(!channel->monitor_timer_enabled); |
| 585 | |
| 586 | ret = consumer_channel_timer_start(&channel->monitor_timer, channel, |
| 587 | monitor_timer_interval_us, LTTNG_CONSUMER_SIG_MONITOR); |
| 588 | channel->monitor_timer_enabled = !!(ret == 0); |
| 589 | return ret; |
| 590 | } |
| 591 | |
| 592 | /* |
| 593 | * Stop and delete the channel's monitoring timer. |
| 594 | */ |
| 595 | int consumer_timer_monitor_stop(struct lttng_consumer_channel *channel) |
| 596 | { |
| 597 | int ret; |
| 598 | |
| 599 | assert(channel); |
| 600 | assert(channel->monitor_timer_enabled); |
| 601 | |
| 602 | ret = consumer_channel_timer_stop(&channel->monitor_timer, |
| 603 | LTTNG_CONSUMER_SIG_MONITOR); |
| 604 | if (ret == -1) { |
| 605 | ERR("Failed to stop live timer"); |
| 606 | goto end; |
| 607 | } |
| 608 | |
| 609 | channel->monitor_timer_enabled = 0; |
| 610 | end: |
| 611 | return ret; |
| 612 | } |
| 613 | |
| 614 | /* |
| 615 | * Block the RT signals for the entire process. It must be called from the |
| 616 | * consumer main before creating the threads |
| 617 | */ |
| 618 | int consumer_signal_init(void) |
| 619 | { |
| 620 | int ret; |
| 621 | sigset_t mask; |
| 622 | |
| 623 | /* Block signal for entire process, so only our thread processes it. */ |
| 624 | setmask(&mask); |
| 625 | ret = pthread_sigmask(SIG_BLOCK, &mask, NULL); |
| 626 | if (ret) { |
| 627 | errno = ret; |
| 628 | PERROR("pthread_sigmask"); |
| 629 | return -1; |
| 630 | } |
| 631 | return 0; |
| 632 | } |
| 633 | |
| 634 | static |
| 635 | int sample_channel_positions(struct lttng_consumer_channel *channel, |
| 636 | uint64_t *_highest_use, uint64_t *_lowest_use, uint64_t *_total_consumed, |
| 637 | sample_positions_cb sample, get_consumed_cb get_consumed, |
| 638 | get_produced_cb get_produced) |
| 639 | { |
| 640 | int ret = 0; |
| 641 | struct lttng_ht_iter iter; |
| 642 | struct lttng_consumer_stream *stream; |
| 643 | bool empty_channel = true; |
| 644 | uint64_t high = 0, low = UINT64_MAX; |
| 645 | struct lttng_ht *ht = consumer_data.stream_per_chan_id_ht; |
| 646 | |
| 647 | *_total_consumed = 0; |
| 648 | |
| 649 | rcu_read_lock(); |
| 650 | |
| 651 | cds_lfht_for_each_entry_duplicate(ht->ht, |
| 652 | ht->hash_fct(&channel->key, lttng_ht_seed), |
| 653 | ht->match_fct, &channel->key, |
| 654 | &iter.iter, stream, node_channel_id.node) { |
| 655 | unsigned long produced, consumed, usage; |
| 656 | |
| 657 | empty_channel = false; |
| 658 | |
| 659 | pthread_mutex_lock(&stream->lock); |
| 660 | if (cds_lfht_is_node_deleted(&stream->node.node)) { |
| 661 | goto next; |
| 662 | } |
| 663 | |
| 664 | ret = sample(stream); |
| 665 | if (ret) { |
| 666 | ERR("Failed to take buffer position snapshot in monitor timer (ret = %d)", ret); |
| 667 | pthread_mutex_unlock(&stream->lock); |
| 668 | goto end; |
| 669 | } |
| 670 | ret = get_consumed(stream, &consumed); |
| 671 | if (ret) { |
| 672 | ERR("Failed to get buffer consumed position in monitor timer"); |
| 673 | pthread_mutex_unlock(&stream->lock); |
| 674 | goto end; |
| 675 | } |
| 676 | ret = get_produced(stream, &produced); |
| 677 | if (ret) { |
| 678 | ERR("Failed to get buffer produced position in monitor timer"); |
| 679 | pthread_mutex_unlock(&stream->lock); |
| 680 | goto end; |
| 681 | } |
| 682 | |
| 683 | usage = produced - consumed; |
| 684 | high = (usage > high) ? usage : high; |
| 685 | low = (usage < low) ? usage : low; |
| 686 | |
| 687 | /* |
| 688 | * We don't use consumed here for 2 reasons: |
| 689 | * - output_written takes into account the padding written in the |
| 690 | * tracefiles when we stop the session; |
| 691 | * - the consumed position is not the accurate representation of what |
| 692 | * was extracted from a buffer in overwrite mode. |
| 693 | */ |
| 694 | *_total_consumed += stream->output_written; |
| 695 | next: |
| 696 | pthread_mutex_unlock(&stream->lock); |
| 697 | } |
| 698 | |
| 699 | *_highest_use = high; |
| 700 | *_lowest_use = low; |
| 701 | end: |
| 702 | rcu_read_unlock(); |
| 703 | if (empty_channel) { |
| 704 | ret = -1; |
| 705 | } |
| 706 | return ret; |
| 707 | } |
| 708 | |
| 709 | /* |
| 710 | * Execute action on a monitor timer. |
| 711 | */ |
| 712 | static |
| 713 | void monitor_timer(struct lttng_consumer_local_data *ctx, |
| 714 | struct lttng_consumer_channel *channel) |
| 715 | { |
| 716 | int ret; |
| 717 | int channel_monitor_pipe = |
| 718 | consumer_timer_thread_get_channel_monitor_pipe(); |
| 719 | struct lttcomm_consumer_channel_monitor_msg msg = { |
| 720 | .key = channel->key, |
| 721 | }; |
| 722 | sample_positions_cb sample; |
| 723 | get_consumed_cb get_consumed; |
| 724 | get_produced_cb get_produced; |
| 725 | |
| 726 | assert(channel); |
| 727 | |
| 728 | if (channel_monitor_pipe < 0) { |
| 729 | return; |
| 730 | } |
| 731 | |
| 732 | switch (consumer_data.type) { |
| 733 | case LTTNG_CONSUMER_KERNEL: |
| 734 | sample = lttng_kconsumer_sample_snapshot_positions; |
| 735 | get_consumed = lttng_kconsumer_get_consumed_snapshot; |
| 736 | get_produced = lttng_kconsumer_get_produced_snapshot; |
| 737 | break; |
| 738 | case LTTNG_CONSUMER32_UST: |
| 739 | case LTTNG_CONSUMER64_UST: |
| 740 | sample = lttng_ustconsumer_sample_snapshot_positions; |
| 741 | get_consumed = lttng_ustconsumer_get_consumed_snapshot; |
| 742 | get_produced = lttng_ustconsumer_get_produced_snapshot; |
| 743 | break; |
| 744 | default: |
| 745 | abort(); |
| 746 | } |
| 747 | |
| 748 | ret = sample_channel_positions(channel, &msg.highest, &msg.lowest, |
| 749 | &msg.total_consumed, sample, get_consumed, get_produced); |
| 750 | if (ret) { |
| 751 | return; |
| 752 | } |
| 753 | |
| 754 | /* |
| 755 | * Writes performed here are assumed to be atomic which is only |
| 756 | * guaranteed for sizes < than PIPE_BUF. |
| 757 | */ |
| 758 | assert(sizeof(msg) <= PIPE_BUF); |
| 759 | |
| 760 | do { |
| 761 | ret = write(channel_monitor_pipe, &msg, sizeof(msg)); |
| 762 | } while (ret == -1 && errno == EINTR); |
| 763 | if (ret == -1) { |
| 764 | if (errno == EAGAIN) { |
| 765 | /* Not an error, the sample is merely dropped. */ |
| 766 | DBG("Channel monitor pipe is full; dropping sample for channel key = %"PRIu64, |
| 767 | channel->key); |
| 768 | } else { |
| 769 | PERROR("write to the channel monitor pipe"); |
| 770 | } |
| 771 | } else { |
| 772 | DBG("Sent channel monitoring sample for channel key %" PRIu64 |
| 773 | ", (highest = %" PRIu64 ", lowest = %"PRIu64")", |
| 774 | channel->key, msg.highest, msg.lowest); |
| 775 | } |
| 776 | } |
| 777 | |
| 778 | int consumer_timer_thread_get_channel_monitor_pipe(void) |
| 779 | { |
| 780 | return uatomic_read(&channel_monitor_pipe); |
| 781 | } |
| 782 | |
| 783 | int consumer_timer_thread_set_channel_monitor_pipe(int fd) |
| 784 | { |
| 785 | int ret; |
| 786 | |
| 787 | ret = uatomic_cmpxchg(&channel_monitor_pipe, -1, fd); |
| 788 | if (ret != -1) { |
| 789 | ret = -1; |
| 790 | goto end; |
| 791 | } |
| 792 | ret = 0; |
| 793 | end: |
| 794 | return ret; |
| 795 | } |
| 796 | |
| 797 | /* |
| 798 | * This thread is the sighandler for signals LTTNG_CONSUMER_SIG_SWITCH, |
| 799 | * LTTNG_CONSUMER_SIG_TEARDOWN, LTTNG_CONSUMER_SIG_LIVE, and |
| 800 | * LTTNG_CONSUMER_SIG_MONITOR, LTTNG_CONSUMER_SIG_EXIT. |
| 801 | */ |
| 802 | void *consumer_timer_thread(void *data) |
| 803 | { |
| 804 | int signr; |
| 805 | sigset_t mask; |
| 806 | siginfo_t info; |
| 807 | struct lttng_consumer_local_data *ctx = data; |
| 808 | |
| 809 | rcu_register_thread(); |
| 810 | |
| 811 | health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA_TIMER); |
| 812 | |
| 813 | if (testpoint(consumerd_thread_metadata_timer)) { |
| 814 | goto error_testpoint; |
| 815 | } |
| 816 | |
| 817 | health_code_update(); |
| 818 | |
| 819 | /* Only self thread will receive signal mask. */ |
| 820 | setmask(&mask); |
| 821 | CMM_STORE_SHARED(timer_signal.tid, pthread_self()); |
| 822 | |
| 823 | while (1) { |
| 824 | health_code_update(); |
| 825 | |
| 826 | health_poll_entry(); |
| 827 | signr = sigwaitinfo(&mask, &info); |
| 828 | health_poll_exit(); |
| 829 | |
| 830 | /* |
| 831 | * NOTE: cascading conditions are used instead of a switch case |
| 832 | * since the use of SIGRTMIN in the definition of the signals' |
| 833 | * values prevents the reduction to an integer constant. |
| 834 | */ |
| 835 | if (signr == -1) { |
| 836 | if (errno != EINTR) { |
| 837 | PERROR("sigwaitinfo"); |
| 838 | } |
| 839 | continue; |
| 840 | } else if (signr == LTTNG_CONSUMER_SIG_SWITCH) { |
| 841 | metadata_switch_timer(ctx, info.si_signo, &info); |
| 842 | } else if (signr == LTTNG_CONSUMER_SIG_TEARDOWN) { |
| 843 | cmm_smp_mb(); |
| 844 | CMM_STORE_SHARED(timer_signal.qs_done, 1); |
| 845 | cmm_smp_mb(); |
| 846 | DBG("Signal timer metadata thread teardown"); |
| 847 | } else if (signr == LTTNG_CONSUMER_SIG_LIVE) { |
| 848 | live_timer(ctx, info.si_signo, &info); |
| 849 | } else if (signr == LTTNG_CONSUMER_SIG_MONITOR) { |
| 850 | struct lttng_consumer_channel *channel; |
| 851 | |
| 852 | channel = info.si_value.sival_ptr; |
| 853 | monitor_timer(ctx, channel); |
| 854 | } else if (signr == LTTNG_CONSUMER_SIG_EXIT) { |
| 855 | assert(CMM_LOAD_SHARED(consumer_quit)); |
| 856 | goto end; |
| 857 | } else { |
| 858 | ERR("Unexpected signal %d\n", info.si_signo); |
| 859 | } |
| 860 | } |
| 861 | |
| 862 | error_testpoint: |
| 863 | /* Only reached in testpoint error */ |
| 864 | health_error(); |
| 865 | end: |
| 866 | health_unregister(health_consumerd); |
| 867 | rcu_unregister_thread(); |
| 868 | return NULL; |
| 869 | } |