src/bin/lttng-sessiond/rotation-thread.c

   1 /*
   2  * Copyright (C) 2017 Julien Desfossez <jdesfossez@efficios.com>
   3  * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   4  *
   5  * SPDX-License-Identifier: GPL-2.0-only
   6  *
   7  */
   8
   9 #define _LGPL_SOURCE
  10 #include <lttng/trigger/trigger.h>
  11 #include <common/error.h>
  12 #include <common/config/session-config.h>
  13 #include <common/defaults.h>
  14 #include <common/utils.h>
  15 #include <common/futex.h>
  16 #include <common/align.h>
  17 #include <common/time.h>
  18 #include <common/hashtable/utils.h>
  19 #include <sys/eventfd.h>
  20 #include <sys/stat.h>
  21 #include <time.h>
  22 #include <signal.h>
  23 #include <inttypes.h>
  24
  25 #include <common/kernel-ctl/kernel-ctl.h>
  26 #include <lttng/notification/channel-internal.h>
  27 #include <lttng/rotate-internal.h>
  28
  29 #include "rotation-thread.h"
  30 #include "lttng-sessiond.h"
  31 #include "health-sessiond.h"
  32 #include "rotate.h"
  33 #include "cmd.h"
  34 #include "session.h"
  35 #include "timer.h"
  36 #include "notification-thread-commands.h"
  37 #include "utils.h"
  38 #include "thread.h"
  39
  40 #include <urcu.h>
  41 #include <urcu/list.h>
  42
  43 struct lttng_notification_channel *rotate_notification_channel = NULL;
  44
  45 struct rotation_thread {
  46         struct lttng_poll_event events;
  47 };
  48
  49 struct rotation_thread_job {
  50         enum rotation_thread_job_type type;
  51         struct ltt_session *session;
  52         /* List member in struct rotation_thread_timer_queue. */
  53         struct cds_list_head head;
  54 };
  55
  56 /*
  57  * The timer thread enqueues jobs and wakes up the rotation thread.
  58  * When the rotation thread wakes up, it empties the queue.
  59  */
  60 struct rotation_thread_timer_queue {
  61         struct lttng_pipe *event_pipe;
  62         struct cds_list_head list;
  63         pthread_mutex_t lock;
  64 };
  65
  66 struct rotation_thread_handle {
  67         struct rotation_thread_timer_queue *rotation_timer_queue;
  68         /* Access to the notification thread cmd_queue */
  69         struct notification_thread_handle *notification_thread_handle;
  70         /* Thread-specific quit pipe. */
  71         struct lttng_pipe *quit_pipe;
  72 };
  73
  74 static
  75 const char *get_job_type_str(enum rotation_thread_job_type job_type)
  76 {
  77         switch (job_type) {
  78         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
  79                 return "CHECK_PENDING_ROTATION";
  80         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
  81                 return "SCHEDULED_ROTATION";
  82         default:
  83                 abort();
  84         }
  85 }
  86
  87 struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void)
  88 {
  89         struct rotation_thread_timer_queue *queue = NULL;
  90
  91         queue = zmalloc(sizeof(*queue));
  92         if (!queue) {
  93                 PERROR("Failed to allocate timer rotate queue");
  94                 goto end;
  95         }
  96
  97         queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
  98         CDS_INIT_LIST_HEAD(&queue->list);
  99         pthread_mutex_init(&queue->lock, NULL);
 100 end:
 101         return queue;
 102 }
 103
 104 void rotation_thread_timer_queue_destroy(
 105                 struct rotation_thread_timer_queue *queue)
 106 {
 107         if (!queue) {
 108                 return;
 109         }
 110
 111         lttng_pipe_destroy(queue->event_pipe);
 112
 113         pthread_mutex_lock(&queue->lock);
 114         assert(cds_list_empty(&queue->list));
 115         pthread_mutex_unlock(&queue->lock);
 116         pthread_mutex_destroy(&queue->lock);
 117         free(queue);
 118 }
 119
 120 /*
 121  * Destroy the thread data previously created by the init function.
 122  */
 123 void rotation_thread_handle_destroy(
 124                 struct rotation_thread_handle *handle)
 125 {
 126         lttng_pipe_destroy(handle->quit_pipe);
 127         free(handle);
 128 }
 129
 130 struct rotation_thread_handle *rotation_thread_handle_create(
 131                 struct rotation_thread_timer_queue *rotation_timer_queue,
 132                 struct notification_thread_handle *notification_thread_handle)
 133 {
 134         struct rotation_thread_handle *handle;
 135
 136         handle = zmalloc(sizeof(*handle));
 137         if (!handle) {
 138                 goto end;
 139         }
 140
 141         handle->rotation_timer_queue = rotation_timer_queue;
 142         handle->notification_thread_handle = notification_thread_handle;
 143         handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
 144         if (!handle->quit_pipe) {
 145                 goto error;
 146         }
 147
 148 end:
 149         return handle;
 150 error:
 151         rotation_thread_handle_destroy(handle);
 152         return NULL;
 153 }
 154
 155 /*
 156  * Called with the rotation_thread_timer_queue lock held.
 157  * Return true if the same timer job already exists in the queue, false if not.
 158  */
 159 static
 160 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
 161                 enum rotation_thread_job_type job_type,
 162                 struct ltt_session *session)
 163 {
 164         bool exists = false;
 165         struct rotation_thread_job *job;
 166
 167         cds_list_for_each_entry(job, &queue->list, head) {
 168                 if (job->session == session && job->type == job_type) {
 169                         exists = true;
 170                         goto end;
 171                 }
 172         }
 173 end:
 174         return exists;
 175 }
 176
 177 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 178                 enum rotation_thread_job_type job_type,
 179                 struct ltt_session *session)
 180 {
 181         int ret;
 182         const char dummy = '!';
 183         struct rotation_thread_job *job = NULL;
 184         const char *job_type_str = get_job_type_str(job_type);
 185
 186         pthread_mutex_lock(&queue->lock);
 187         if (timer_job_exists(queue, job_type, session)) {
 188                 /*
 189                  * This timer job is already pending, we don't need to add
 190                  * it.
 191                  */
 192                 goto end;
 193         }
 194
 195         job = zmalloc(sizeof(struct rotation_thread_job));
 196         if (!job) {
 197                 PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"",
 198                                 job_type_str, session->name);
 199                 goto end;
 200         }
 201         /* No reason for this to fail as the caller must hold a reference. */
 202         (void) session_get(session);
 203
 204         job->session = session;
 205         job->type = job_type;
 206         cds_list_add_tail(&job->head, &queue->list);
 207
 208         ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy,
 209                         sizeof(dummy));
 210         if (ret < 0) {
 211                 /*
 212                  * We do not want to block in the timer handler, the job has
 213                  * been enqueued in the list, the wakeup pipe is probably full,
 214                  * the job will be processed when the rotation_thread catches
 215                  * up.
 216                  */
 217                 if (errno == EAGAIN || errno == EWOULDBLOCK) {
 218                         /*
 219                          * Not an error, but would be surprising and indicate
 220                          * that the rotation thread can't keep up with the
 221                          * current load.
 222                          */
 223                         DBG("Wake-up pipe of rotation thread job queue is full");
 224                         goto end;
 225                 }
 226                 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
 227                                 job_type_str, session->name);
 228                 goto end;
 229         }
 230
 231 end:
 232         pthread_mutex_unlock(&queue->lock);
 233 }
 234
 235 static
 236 int init_poll_set(struct lttng_poll_event *poll_set,
 237                 struct rotation_thread_handle *handle)
 238 {
 239         int ret;
 240
 241         /*
 242          * Create pollset with size 3:
 243          *      - rotation thread quit pipe,
 244          *      - rotation thread timer queue pipe,
 245          *      - notification channel sock,
 246          */
 247         ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
 248         if (ret < 0) {
 249                 goto error;
 250         }
 251
 252         ret = lttng_poll_add(poll_set,
 253                         lttng_pipe_get_readfd(handle->quit_pipe),
 254                         LPOLLIN | LPOLLERR);
 255         if (ret < 0) {
 256                 ERR("[rotation-thread] Failed to add quit pipe read fd to poll set");
 257                 goto error;
 258         }
 259
 260         ret = lttng_poll_add(poll_set,
 261                         lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
 262                         LPOLLIN | LPOLLERR);
 263         if (ret < 0) {
 264                 ERR("[rotation-thread] Failed to add rotate_pending fd to poll set");
 265                 goto error;
 266         }
 267
 268         return ret;
 269 error:
 270         lttng_poll_clean(poll_set);
 271         return ret;
 272 }
 273
 274 static
 275 void fini_thread_state(struct rotation_thread *state)
 276 {
 277         lttng_poll_clean(&state->events);
 278         if (rotate_notification_channel) {
 279                 lttng_notification_channel_destroy(rotate_notification_channel);
 280         }
 281 }
 282
 283 static
 284 int init_thread_state(struct rotation_thread_handle *handle,
 285                 struct rotation_thread *state)
 286 {
 287         int ret;
 288
 289         memset(state, 0, sizeof(*state));
 290         lttng_poll_init(&state->events);
 291
 292         ret = init_poll_set(&state->events, handle);
 293         if (ret) {
 294                 ERR("[rotation-thread] Failed to initialize rotation thread poll set");
 295                 goto end;
 296         }
 297
 298         rotate_notification_channel = lttng_notification_channel_create(
 299                         lttng_session_daemon_notification_endpoint);
 300         if (!rotate_notification_channel) {
 301                 ERR("[rotation-thread] Could not create notification channel");
 302                 ret = -1;
 303                 goto end;
 304         }
 305         ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
 306                         LPOLLIN | LPOLLERR);
 307         if (ret < 0) {
 308                 ERR("[rotation-thread] Failed to add notification fd to pollset");
 309                 goto end;
 310         }
 311
 312 end:
 313         return ret;
 314 }
 315
 316 static
 317 void check_session_rotation_pending_on_consumers(struct ltt_session *session,
 318                 bool *_rotation_completed)
 319 {
 320         int ret = 0;
 321         struct consumer_socket *socket;
 322         struct cds_lfht_iter iter;
 323         enum consumer_trace_chunk_exists_status exists_status;
 324         uint64_t relayd_id;
 325         bool chunk_exists_on_peer = false;
 326         enum lttng_trace_chunk_status chunk_status;
 327
 328         assert(session->chunk_being_archived);
 329
 330         /*
 331          * Check for a local pending rotation on all consumers (32-bit
 332          * user space, 64-bit user space, and kernel).
 333          */
 334         rcu_read_lock();
 335         if (!session->ust_session) {
 336                 goto skip_ust;
 337         }
 338         cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
 339                         &iter, socket, node.node) {
 340                 relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ?
 341                                 -1ULL :
 342                                 session->ust_session->consumer->net_seq_index;
 343
 344                 pthread_mutex_lock(socket->lock);
 345                 ret = consumer_trace_chunk_exists(socket,
 346                                 relayd_id,
 347                                 session->id, session->chunk_being_archived,
 348                                 &exists_status);
 349                 if (ret) {
 350                         pthread_mutex_unlock(socket->lock);
 351                         ERR("Error occurred while checking rotation status on consumer daemon");
 352                         goto end;
 353                 }
 354
 355                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 356                         pthread_mutex_unlock(socket->lock);
 357                         chunk_exists_on_peer = true;
 358                         goto end;
 359                 }
 360                 pthread_mutex_unlock(socket->lock);
 361         }
 362
 363 skip_ust:
 364         if (!session->kernel_session) {
 365                 goto skip_kernel;
 366         }
 367         cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
 368                                 &iter, socket, node.node) {
 369                 pthread_mutex_lock(socket->lock);
 370                 relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ?
 371                                 -1ULL :
 372                                 session->kernel_session->consumer->net_seq_index;
 373
 374                 ret = consumer_trace_chunk_exists(socket,
 375                                 relayd_id,
 376                                 session->id, session->chunk_being_archived,
 377                                 &exists_status);
 378                 if (ret) {
 379                         pthread_mutex_unlock(socket->lock);
 380                         ERR("Error occurred while checking rotation status on consumer daemon");
 381                         goto end;
 382                 }
 383
 384                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 385                         pthread_mutex_unlock(socket->lock);
 386                         chunk_exists_on_peer = true;
 387                         goto end;
 388                 }
 389                 pthread_mutex_unlock(socket->lock);
 390         }
 391 skip_kernel:
 392 end:
 393         rcu_read_unlock();
 394
 395         if (!chunk_exists_on_peer) {
 396                 uint64_t chunk_being_archived_id;
 397
 398                 chunk_status = lttng_trace_chunk_get_id(
 399                                 session->chunk_being_archived,
 400                                 &chunk_being_archived_id);
 401                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 402                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
 403                                 chunk_being_archived_id,
 404                                 session->name);
 405         }
 406         *_rotation_completed = !chunk_exists_on_peer;
 407         if (ret) {
 408                 ret = session_reset_rotation_state(session,
 409                                 LTTNG_ROTATION_STATE_ERROR);
 410                 if (ret) {
 411                         ERR("Failed to reset rotation state of session \"%s\"",
 412                                         session->name);
 413                 }
 414         }
 415 }
 416
 417 /*
 418  * Check if the last rotation was completed, called with session lock held.
 419  * Should only return non-zero in the event of a fatal error. Doing so will
 420  * shutdown the thread.
 421  */
 422 static
 423 int check_session_rotation_pending(struct ltt_session *session,
 424                 struct notification_thread_handle *notification_thread_handle)
 425 {
 426         int ret;
 427         struct lttng_trace_archive_location *location;
 428         enum lttng_trace_chunk_status chunk_status;
 429         bool rotation_completed = false;
 430         const char *archived_chunk_name;
 431         uint64_t chunk_being_archived_id;
 432
 433         if (!session->chunk_being_archived) {
 434                 ret = 0;
 435                 goto end;
 436         }
 437
 438         chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived,
 439                         &chunk_being_archived_id);
 440         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 441
 442         DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
 443                         session->name, chunk_being_archived_id);
 444
 445         /*
 446          * The rotation-pending check timer of a session is launched in
 447          * one-shot mode. If the rotation is incomplete, the rotation
 448          * thread will re-enable the pending-check timer.
 449          *
 450          * The timer thread can't stop the timer itself since it is involved
 451          * in the check for the timer's quiescence.
 452          */
 453         ret = timer_session_rotation_pending_check_stop(session);
 454         if (ret) {
 455                 goto check_ongoing_rotation;
 456         }
 457
 458         check_session_rotation_pending_on_consumers(session,
 459                         &rotation_completed);
 460         if (!rotation_completed ||
 461                         session->rotation_state == LTTNG_ROTATION_STATE_ERROR) {
 462                 goto check_ongoing_rotation;
 463         }
 464
 465         /*
 466          * Now we can clear the "ONGOING" state in the session. New
 467          * rotations can start now.
 468          */
 469         chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived,
 470                         &archived_chunk_name, NULL);
 471         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 472         free(session->last_archived_chunk_name);
 473         session->last_archived_chunk_name = strdup(archived_chunk_name);
 474         if (!session->last_archived_chunk_name) {
 475                 PERROR("Failed to duplicate archived chunk name");
 476         }
 477         session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED);
 478
 479         if (!session->quiet_rotation) {
 480                 location = session_get_trace_archive_location(session);
 481                 /* Ownership of location is transferred. */
 482                 ret = notification_thread_command_session_rotation_completed(
 483                                 notification_thread_handle,
 484                                 session->name,
 485                                 session->uid,
 486                                 session->gid,
 487                                 session->last_archived_chunk_id.value,
 488                                 location);
 489                 if (ret != LTTNG_OK) {
 490                         ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
 491                                         session->name);
 492                 }
 493         }
 494
 495         ret = 0;
 496 check_ongoing_rotation:
 497         if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
 498                 uint64_t chunk_being_archived_id;
 499
 500                 chunk_status = lttng_trace_chunk_get_id(
 501                                 session->chunk_being_archived,
 502                                 &chunk_being_archived_id);
 503                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 504
 505                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s",
 506                                 chunk_being_archived_id, session->name);
 507                 ret = timer_session_rotation_pending_check_start(session,
 508                                 DEFAULT_ROTATE_PENDING_TIMER);
 509                 if (ret) {
 510                         ERR("Failed to re-enable rotation pending timer");
 511                         ret = -1;
 512                         goto end;
 513                 }
 514         }
 515
 516 end:
 517         return ret;
 518 }
 519
 520 /* Call with the session and session_list locks held. */
 521 static
 522 int launch_session_rotation(struct ltt_session *session)
 523 {
 524         int ret;
 525         struct lttng_rotate_session_return rotation_return;
 526
 527         DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"",
 528                         session->name);
 529
 530         ret = cmd_rotate_session(session, &rotation_return, false,
 531                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 532         if (ret == LTTNG_OK) {
 533                 DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"",
 534                                 session->name);
 535         } else {
 536                 /* Don't consider errors as fatal. */
 537                 DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s",
 538                                 session->name, lttng_strerror(ret));
 539         }
 540         return 0;
 541 }
 542
 543 static
 544 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
 545                 struct notification_thread_handle *notification_thread_handle)
 546 {
 547         int ret;
 548
 549         switch (job->type) {
 550         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
 551                 ret = launch_session_rotation(session);
 552                 break;
 553         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
 554                 ret = check_session_rotation_pending(session,
 555                                 notification_thread_handle);
 556                 break;
 557         default:
 558                 abort();
 559         }
 560         return ret;
 561 }
 562
 563 static
 564 int handle_job_queue(struct rotation_thread_handle *handle,
 565                 struct rotation_thread *state,
 566                 struct rotation_thread_timer_queue *queue)
 567 {
 568         int ret = 0;
 569
 570         for (;;) {
 571                 struct ltt_session *session;
 572                 struct rotation_thread_job *job;
 573
 574                 /* Take the queue lock only to pop an element from the list. */
 575                 pthread_mutex_lock(&queue->lock);
 576                 if (cds_list_empty(&queue->list)) {
 577                         pthread_mutex_unlock(&queue->lock);
 578                         break;
 579                 }
 580                 job = cds_list_first_entry(&queue->list,
 581                                 typeof(*job), head);
 582                 cds_list_del(&job->head);
 583                 pthread_mutex_unlock(&queue->lock);
 584
 585                 session_lock_list();
 586                 session = job->session;
 587                 if (!session) {
 588                         DBG("[rotation-thread] Session \"%s\" not found",
 589                                         session->name);
 590                         /*
 591                          * This is a non-fatal error, and we cannot report it to
 592                          * the user (timer), so just print the error and
 593                          * continue the processing.
 594                          *
 595                          * While the timer thread will purge pending signals for
 596                          * a session on the session's destruction, it is
 597                          * possible for a job targeting that session to have
 598                          * already been queued before it was destroyed.
 599                          */
 600                         free(job);
 601                         session_put(session);
 602                         session_unlock_list();
 603                         continue;
 604                 }
 605
 606                 session_lock(session);
 607                 ret = run_job(job, session, handle->notification_thread_handle);
 608                 session_unlock(session);
 609                 /* Release reference held by the job. */
 610                 session_put(session);
 611                 session_unlock_list();
 612                 free(job);
 613                 if (ret) {
 614                         goto end;
 615                 }
 616         }
 617
 618         ret = 0;
 619
 620 end:
 621         return ret;
 622 }
 623
 624 static
 625 int handle_condition(const struct lttng_condition *condition,
 626                 const struct lttng_evaluation *evaluation,
 627                 struct notification_thread_handle *notification_thread_handle)
 628 {
 629         int ret = 0;
 630         const char *condition_session_name = NULL;
 631         enum lttng_condition_type condition_type;
 632         enum lttng_condition_status condition_status;
 633         enum lttng_evaluation_status evaluation_status;
 634         uint64_t consumed;
 635         struct ltt_session *session;
 636
 637         condition_type = lttng_condition_get_type(condition);
 638
 639         if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
 640                 ret = -1;
 641                 ERR("[rotation-thread] Condition type and session usage type are not the same");
 642                 goto end;
 643         }
 644
 645         /* Fetch info to test */
 646         condition_status = lttng_condition_session_consumed_size_get_session_name(
 647                         condition, &condition_session_name);
 648         if (condition_status != LTTNG_CONDITION_STATUS_OK) {
 649                 ERR("[rotation-thread] Session name could not be fetched");
 650                 ret = -1;
 651                 goto end;
 652         }
 653         evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
 654                         &consumed);
 655         if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
 656                 ERR("[rotation-thread] Failed to get evaluation");
 657                 ret = -1;
 658                 goto end;
 659         }
 660
 661         session_lock_list();
 662         session = session_find_by_name(condition_session_name);
 663         if (!session) {
 664                 ret = -1;
 665                 session_unlock_list();
 666                 ERR("[rotation-thread] Session \"%s\" not found",
 667                                 condition_session_name);
 668                 goto end;
 669         }
 670         session_lock(session);
 671
 672         ret = unsubscribe_session_consumed_size_rotation(session,
 673                         notification_thread_handle);
 674         if (ret) {
 675                 goto end_unlock;
 676         }
 677
 678         ret = cmd_rotate_session(session, NULL, false,
 679                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 680         if (ret == -LTTNG_ERR_ROTATION_PENDING) {
 681                 DBG("Rotate already pending, subscribe to the next threshold value");
 682         } else if (ret != LTTNG_OK) {
 683                 ERR("[rotation-thread] Failed to rotate on size notification with error: %s",
 684                                 lttng_strerror(ret));
 685                 ret = -1;
 686                 goto end_unlock;
 687         }
 688         ret = subscribe_session_consumed_size_rotation(session,
 689                         consumed + session->rotate_size,
 690                         notification_thread_handle);
 691         if (ret) {
 692                 ERR("[rotation-thread] Failed to subscribe to session consumed size condition");
 693                 goto end_unlock;
 694         }
 695         ret = 0;
 696
 697 end_unlock:
 698         session_unlock(session);
 699         session_put(session);
 700         session_unlock_list();
 701 end:
 702         return ret;
 703 }
 704
 705 static
 706 int handle_notification_channel(int fd,
 707                 struct rotation_thread_handle *handle,
 708                 struct rotation_thread *state)
 709 {
 710         int ret;
 711         bool notification_pending;
 712         struct lttng_notification *notification = NULL;
 713         enum lttng_notification_channel_status status;
 714         const struct lttng_evaluation *notification_evaluation;
 715         const struct lttng_condition *notification_condition;
 716
 717         status = lttng_notification_channel_has_pending_notification(
 718                         rotate_notification_channel, &notification_pending);
 719         if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
 720                 ERR("[rotation-thread ]Error occurred while checking for pending notification");
 721                 ret = -1;
 722                 goto end;
 723         }
 724
 725         if (!notification_pending) {
 726                 ret = 0;
 727                 goto end;
 728         }
 729
 730         /* Receive the next notification. */
 731         status = lttng_notification_channel_get_next_notification(
 732                         rotate_notification_channel,
 733                         &notification);
 734
 735         switch (status) {
 736         case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
 737                 break;
 738         case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
 739                 /* Not an error, we will wait for the next one */
 740                 ret = 0;
 741                 goto end;;
 742         case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
 743                 ERR("Notification channel was closed");
 744                 ret = -1;
 745                 goto end;
 746         default:
 747                 /* Unhandled conditions / errors. */
 748                 ERR("Unknown notification channel status");
 749                 ret = -1;
 750                 goto end;
 751         }
 752
 753         notification_condition = lttng_notification_get_condition(notification);
 754         notification_evaluation = lttng_notification_get_evaluation(notification);
 755
 756         ret = handle_condition(notification_condition, notification_evaluation,
 757                         handle->notification_thread_handle);
 758
 759 end:
 760         lttng_notification_destroy(notification);
 761         return ret;
 762 }
 763
 764 static
 765 void *thread_rotation(void *data)
 766 {
 767         int ret;
 768         struct rotation_thread_handle *handle = data;
 769         struct rotation_thread thread;
 770         int queue_pipe_fd;
 771
 772         DBG("[rotation-thread] Started rotation thread");
 773         rcu_register_thread();
 774         rcu_thread_online();
 775         health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
 776         health_code_update();
 777
 778         if (!handle) {
 779                 ERR("[rotation-thread] Invalid thread context provided");
 780                 goto end;
 781         }
 782
 783         queue_pipe_fd = lttng_pipe_get_readfd(
 784                         handle->rotation_timer_queue->event_pipe);
 785
 786
 787         ret = init_thread_state(handle, &thread);
 788         if (ret) {
 789                 goto error;
 790         }
 791
 792         while (true) {
 793                 int fd_count, i;
 794
 795                 health_poll_entry();
 796                 DBG("[rotation-thread] Entering poll wait");
 797                 ret = lttng_poll_wait(&thread.events, -1);
 798                 DBG("[rotation-thread] Poll wait returned (%i)", ret);
 799                 health_poll_exit();
 800                 if (ret < 0) {
 801                         /*
 802                          * Restart interrupted system call.
 803                          */
 804                         if (errno == EINTR) {
 805                                 continue;
 806                         }
 807                         ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret);
 808                         goto error;
 809                 }
 810
 811                 fd_count = ret;
 812                 for (i = 0; i < fd_count; i++) {
 813                         int fd = LTTNG_POLL_GETFD(&thread.events, i);
 814                         uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
 815
 816                         DBG("[rotation-thread] Handling fd (%i) activity (%u)",
 817                                         fd, revents);
 818
 819                         if (revents & LPOLLERR) {
 820                                 ERR("[rotation-thread] Polling returned an error on fd %i", fd);
 821                                 goto error;
 822                         }
 823
 824                         if (fd == rotate_notification_channel->socket) {
 825                                 ret = handle_notification_channel(fd, handle,
 826                                                 &thread);
 827                                 if (ret) {
 828                                         ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
 829                                         goto error;
 830                                 }
 831                         } else {
 832                                 /* Job queue or quit pipe activity. */
 833
 834                                 /*
 835                                  * The job queue is serviced if there is
 836                                  * activity on the quit pipe to ensure it is
 837                                  * flushed and all references held in the queue
 838                                  * are released.
 839                                  */
 840                                 ret = handle_job_queue(handle, &thread,
 841                                                 handle->rotation_timer_queue);
 842                                 if (ret) {
 843                                         ERR("[rotation-thread] Failed to handle rotation timer pipe event");
 844                                         goto error;
 845                                 }
 846
 847                                 if (fd == queue_pipe_fd) {
 848                                         char buf;
 849
 850                                         ret = lttng_read(fd, &buf, 1);
 851                                         if (ret != 1) {
 852                                                 ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
 853                                                 goto error;
 854                                         }
 855                                 } else {
 856                                         DBG("[rotation-thread] Quit pipe activity");
 857                                         goto exit;
 858                                 }
 859                         }
 860                 }
 861         }
 862 exit:
 863 error:
 864         DBG("[rotation-thread] Exit");
 865         fini_thread_state(&thread);
 866 end:
 867         health_unregister(health_sessiond);
 868         rcu_thread_offline();
 869         rcu_unregister_thread();
 870         return NULL;
 871 }
 872
 873 static
 874 bool shutdown_rotation_thread(void *thread_data)
 875 {
 876         struct rotation_thread_handle *handle = thread_data;
 877         const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
 878
 879         return notify_thread_pipe(write_fd) == 1;
 880 }
 881
 882 bool launch_rotation_thread(struct rotation_thread_handle *handle)
 883 {
 884         struct lttng_thread *thread;
 885
 886         thread = lttng_thread_create("Rotation",
 887                         thread_rotation,
 888                         shutdown_rotation_thread,
 889                         NULL,
 890                         handle);
 891         if (!thread) {
 892                 goto error;
 893         }
 894         lttng_thread_put(thread);
 895         return true;
 896 error:
 897         return false;
 898 }