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

   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.hpp>
  12 #include <common/config/session-config.hpp>
  13 #include <common/defaults.hpp>
  14 #include <common/utils.hpp>
  15 #include <common/futex.hpp>
  16 #include <common/align.hpp>
  17 #include <common/time.hpp>
  18 #include <common/hashtable/utils.hpp>
  19 #include <sys/stat.h>
  20 #include <time.h>
  21 #include <signal.h>
  22 #include <inttypes.h>
  23
  24 #include <common/kernel-ctl/kernel-ctl.hpp>
  25 #include <lttng/notification/channel-internal.hpp>
  26 #include <lttng/rotate-internal.hpp>
  27 #include <lttng/location-internal.hpp>
  28 #include <lttng/condition/condition-internal.hpp>
  29
  30 #include "rotation-thread.hpp"
  31 #include "lttng-sessiond.hpp"
  32 #include "health-sessiond.hpp"
  33 #include "rotate.hpp"
  34 #include "cmd.hpp"
  35 #include "session.hpp"
  36 #include "timer.hpp"
  37 #include "notification-thread-commands.hpp"
  38 #include "utils.hpp"
  39 #include "thread.hpp"
  40
  41 #include <urcu.h>
  42 #include <urcu/list.h>
  43
  44 struct lttng_notification_channel *rotate_notification_channel = NULL;
  45
  46 struct rotation_thread {
  47         struct lttng_poll_event events;
  48 };
  49
  50 struct rotation_thread_job {
  51         enum rotation_thread_job_type type;
  52         struct ltt_session *session;
  53         /* List member in struct rotation_thread_timer_queue. */
  54         struct cds_list_head head;
  55 };
  56
  57 /*
  58  * The timer thread enqueues jobs and wakes up the rotation thread.
  59  * When the rotation thread wakes up, it empties the queue.
  60  */
  61 struct rotation_thread_timer_queue {
  62         struct lttng_pipe *event_pipe;
  63         struct cds_list_head list;
  64         pthread_mutex_t lock;
  65 };
  66
  67 struct rotation_thread_handle {
  68         struct rotation_thread_timer_queue *rotation_timer_queue;
  69         /* Access to the notification thread cmd_queue */
  70         struct notification_thread_handle *notification_thread_handle;
  71         /* Thread-specific quit pipe. */
  72         struct lttng_pipe *quit_pipe;
  73 };
  74
  75 static
  76 const char *get_job_type_str(enum rotation_thread_job_type job_type)
  77 {
  78         switch (job_type) {
  79         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
  80                 return "CHECK_PENDING_ROTATION";
  81         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
  82                 return "SCHEDULED_ROTATION";
  83         default:
  84                 abort();
  85         }
  86 }
  87
  88 struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void)
  89 {
  90         struct rotation_thread_timer_queue *queue = NULL;
  91
  92         queue = zmalloc<rotation_thread_timer_queue>();
  93         if (!queue) {
  94                 PERROR("Failed to allocate timer rotate queue");
  95                 goto end;
  96         }
  97
  98         queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
  99         CDS_INIT_LIST_HEAD(&queue->list);
 100         pthread_mutex_init(&queue->lock, NULL);
 101 end:
 102         return queue;
 103 }
 104
 105 void rotation_thread_timer_queue_destroy(
 106                 struct rotation_thread_timer_queue *queue)
 107 {
 108         if (!queue) {
 109                 return;
 110         }
 111
 112         lttng_pipe_destroy(queue->event_pipe);
 113
 114         pthread_mutex_lock(&queue->lock);
 115         LTTNG_ASSERT(cds_list_empty(&queue->list));
 116         pthread_mutex_unlock(&queue->lock);
 117         pthread_mutex_destroy(&queue->lock);
 118         free(queue);
 119 }
 120
 121 /*
 122  * Destroy the thread data previously created by the init function.
 123  */
 124 void rotation_thread_handle_destroy(
 125                 struct rotation_thread_handle *handle)
 126 {
 127         lttng_pipe_destroy(handle->quit_pipe);
 128         free(handle);
 129 }
 130
 131 struct rotation_thread_handle *rotation_thread_handle_create(
 132                 struct rotation_thread_timer_queue *rotation_timer_queue,
 133                 struct notification_thread_handle *notification_thread_handle)
 134 {
 135         struct rotation_thread_handle *handle;
 136
 137         handle = zmalloc<rotation_thread_handle>();
 138         if (!handle) {
 139                 goto end;
 140         }
 141
 142         handle->rotation_timer_queue = rotation_timer_queue;
 143         handle->notification_thread_handle = notification_thread_handle;
 144         handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
 145         if (!handle->quit_pipe) {
 146                 goto error;
 147         }
 148
 149 end:
 150         return handle;
 151 error:
 152         rotation_thread_handle_destroy(handle);
 153         return NULL;
 154 }
 155
 156 /*
 157  * Called with the rotation_thread_timer_queue lock held.
 158  * Return true if the same timer job already exists in the queue, false if not.
 159  */
 160 static
 161 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
 162                 enum rotation_thread_job_type job_type,
 163                 struct ltt_session *session)
 164 {
 165         bool exists = false;
 166         struct rotation_thread_job *job;
 167
 168         cds_list_for_each_entry(job, &queue->list, head) {
 169                 if (job->session == session && job->type == job_type) {
 170                         exists = true;
 171                         goto end;
 172                 }
 173         }
 174 end:
 175         return exists;
 176 }
 177
 178 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 179                 enum rotation_thread_job_type job_type,
 180                 struct ltt_session *session)
 181 {
 182         int ret;
 183         const char dummy = '!';
 184         struct rotation_thread_job *job = NULL;
 185         const char *job_type_str = get_job_type_str(job_type);
 186
 187         pthread_mutex_lock(&queue->lock);
 188         if (timer_job_exists(queue, job_type, session)) {
 189                 /*
 190                  * This timer job is already pending, we don't need to add
 191                  * it.
 192                  */
 193                 goto end;
 194         }
 195
 196         job = zmalloc<rotation_thread_job>();
 197         if (!job) {
 198                 PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"",
 199                                 job_type_str, session->name);
 200                 goto end;
 201         }
 202         /* No reason for this to fail as the caller must hold a reference. */
 203         (void) session_get(session);
 204
 205         job->session = session;
 206         job->type = job_type;
 207         cds_list_add_tail(&job->head, &queue->list);
 208
 209         ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy,
 210                         sizeof(dummy));
 211         if (ret < 0) {
 212                 /*
 213                  * We do not want to block in the timer handler, the job has
 214                  * been enqueued in the list, the wakeup pipe is probably full,
 215                  * the job will be processed when the rotation_thread catches
 216                  * up.
 217                  */
 218                 DIAGNOSTIC_PUSH
 219                 DIAGNOSTIC_IGNORE_LOGICAL_OP
 220                 if (errno == EAGAIN || errno == EWOULDBLOCK) {
 221                 DIAGNOSTIC_POP
 222                         /*
 223                          * Not an error, but would be surprising and indicate
 224                          * that the rotation thread can't keep up with the
 225                          * current load.
 226                          */
 227                         DBG("Wake-up pipe of rotation thread job queue is full");
 228                         goto end;
 229                 }
 230                 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
 231                                 job_type_str, session->name);
 232                 goto end;
 233         }
 234
 235 end:
 236         pthread_mutex_unlock(&queue->lock);
 237 }
 238
 239 static
 240 int init_poll_set(struct lttng_poll_event *poll_set,
 241                 struct rotation_thread_handle *handle)
 242 {
 243         int ret;
 244
 245         /*
 246          * Create pollset with size 3:
 247          *      - rotation thread quit pipe,
 248          *      - rotation thread timer queue pipe,
 249          *      - notification channel sock,
 250          */
 251         ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
 252         if (ret < 0) {
 253                 goto error;
 254         }
 255
 256         ret = lttng_poll_add(poll_set,
 257                         lttng_pipe_get_readfd(handle->quit_pipe),
 258                         LPOLLIN | LPOLLERR);
 259         if (ret < 0) {
 260                 ERR("Failed to add quit pipe read fd to poll set");
 261                 goto error;
 262         }
 263
 264         ret = lttng_poll_add(poll_set,
 265                         lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
 266                         LPOLLIN | LPOLLERR);
 267         if (ret < 0) {
 268                 ERR("Failed to add rotate_pending fd to poll set");
 269                 goto error;
 270         }
 271
 272         return ret;
 273 error:
 274         lttng_poll_clean(poll_set);
 275         return ret;
 276 }
 277
 278 static
 279 void fini_thread_state(struct rotation_thread *state)
 280 {
 281         lttng_poll_clean(&state->events);
 282         if (rotate_notification_channel) {
 283                 lttng_notification_channel_destroy(rotate_notification_channel);
 284         }
 285 }
 286
 287 static
 288 int init_thread_state(struct rotation_thread_handle *handle,
 289                 struct rotation_thread *state)
 290 {
 291         int ret;
 292
 293         memset(state, 0, sizeof(*state));
 294         lttng_poll_init(&state->events);
 295
 296         ret = init_poll_set(&state->events, handle);
 297         if (ret) {
 298                 ERR("Failed to initialize rotation thread poll set");
 299                 goto end;
 300         }
 301
 302         rotate_notification_channel = lttng_notification_channel_create(
 303                         lttng_session_daemon_notification_endpoint);
 304         if (!rotate_notification_channel) {
 305                 ERR("Could not create notification channel");
 306                 ret = -1;
 307                 goto end;
 308         }
 309         ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
 310                         LPOLLIN | LPOLLERR);
 311         if (ret < 0) {
 312                 ERR("Failed to add notification fd to pollset");
 313                 goto end;
 314         }
 315
 316 end:
 317         return ret;
 318 }
 319
 320 static
 321 void check_session_rotation_pending_on_consumers(struct ltt_session *session,
 322                 bool *_rotation_completed)
 323 {
 324         int ret = 0;
 325         struct consumer_socket *socket;
 326         struct cds_lfht_iter iter;
 327         enum consumer_trace_chunk_exists_status exists_status;
 328         uint64_t relayd_id;
 329         bool chunk_exists_on_peer = false;
 330         enum lttng_trace_chunk_status chunk_status;
 331
 332         LTTNG_ASSERT(session->chunk_being_archived);
 333
 334         /*
 335          * Check for a local pending rotation on all consumers (32-bit
 336          * user space, 64-bit user space, and kernel).
 337          */
 338         rcu_read_lock();
 339         if (!session->ust_session) {
 340                 goto skip_ust;
 341         }
 342         cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
 343                         &iter, socket, node.node) {
 344                 relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ?
 345                                 -1ULL :
 346                                 session->ust_session->consumer->net_seq_index;
 347
 348                 pthread_mutex_lock(socket->lock);
 349                 ret = consumer_trace_chunk_exists(socket,
 350                                 relayd_id,
 351                                 session->id, session->chunk_being_archived,
 352                                 &exists_status);
 353                 if (ret) {
 354                         pthread_mutex_unlock(socket->lock);
 355                         ERR("Error occurred while checking rotation status on consumer daemon");
 356                         goto end;
 357                 }
 358
 359                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 360                         pthread_mutex_unlock(socket->lock);
 361                         chunk_exists_on_peer = true;
 362                         goto end;
 363                 }
 364                 pthread_mutex_unlock(socket->lock);
 365         }
 366
 367 skip_ust:
 368         if (!session->kernel_session) {
 369                 goto skip_kernel;
 370         }
 371         cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
 372                                 &iter, socket, node.node) {
 373                 pthread_mutex_lock(socket->lock);
 374                 relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ?
 375                                 -1ULL :
 376                                 session->kernel_session->consumer->net_seq_index;
 377
 378                 ret = consumer_trace_chunk_exists(socket,
 379                                 relayd_id,
 380                                 session->id, session->chunk_being_archived,
 381                                 &exists_status);
 382                 if (ret) {
 383                         pthread_mutex_unlock(socket->lock);
 384                         ERR("Error occurred while checking rotation status on consumer daemon");
 385                         goto end;
 386                 }
 387
 388                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 389                         pthread_mutex_unlock(socket->lock);
 390                         chunk_exists_on_peer = true;
 391                         goto end;
 392                 }
 393                 pthread_mutex_unlock(socket->lock);
 394         }
 395 skip_kernel:
 396 end:
 397         rcu_read_unlock();
 398
 399         if (!chunk_exists_on_peer) {
 400                 uint64_t chunk_being_archived_id;
 401
 402                 chunk_status = lttng_trace_chunk_get_id(
 403                                 session->chunk_being_archived,
 404                                 &chunk_being_archived_id);
 405                 LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 406                 DBG("Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
 407                                 chunk_being_archived_id,
 408                                 session->name);
 409         }
 410         *_rotation_completed = !chunk_exists_on_peer;
 411         if (ret) {
 412                 ret = session_reset_rotation_state(session,
 413                                 LTTNG_ROTATION_STATE_ERROR);
 414                 if (ret) {
 415                         ERR("Failed to reset rotation state of session \"%s\"",
 416                                         session->name);
 417                 }
 418         }
 419 }
 420
 421 /*
 422  * Check if the last rotation was completed, called with session lock held.
 423  * Should only return non-zero in the event of a fatal error. Doing so will
 424  * shutdown the thread.
 425  */
 426 static
 427 int check_session_rotation_pending(struct ltt_session *session,
 428                 struct notification_thread_handle *notification_thread_handle)
 429 {
 430         int ret;
 431         struct lttng_trace_archive_location *location;
 432         enum lttng_trace_chunk_status chunk_status;
 433         bool rotation_completed = false;
 434         const char *archived_chunk_name;
 435         uint64_t chunk_being_archived_id;
 436
 437         if (!session->chunk_being_archived) {
 438                 ret = 0;
 439                 goto end;
 440         }
 441
 442         chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived,
 443                         &chunk_being_archived_id);
 444         LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 445
 446         DBG("Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
 447                         session->name, chunk_being_archived_id);
 448
 449         /*
 450          * The rotation-pending check timer of a session is launched in
 451          * one-shot mode. If the rotation is incomplete, the rotation
 452          * thread will re-enable the pending-check timer.
 453          *
 454          * The timer thread can't stop the timer itself since it is involved
 455          * in the check for the timer's quiescence.
 456          */
 457         ret = timer_session_rotation_pending_check_stop(session);
 458         if (ret) {
 459                 goto check_ongoing_rotation;
 460         }
 461
 462         check_session_rotation_pending_on_consumers(session,
 463                         &rotation_completed);
 464         if (!rotation_completed ||
 465                         session->rotation_state == LTTNG_ROTATION_STATE_ERROR) {
 466                 goto check_ongoing_rotation;
 467         }
 468
 469         /*
 470          * Now we can clear the "ONGOING" state in the session. New
 471          * rotations can start now.
 472          */
 473         chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived,
 474                         &archived_chunk_name, NULL);
 475         LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 476         free(session->last_archived_chunk_name);
 477         session->last_archived_chunk_name = strdup(archived_chunk_name);
 478         if (!session->last_archived_chunk_name) {
 479                 PERROR("Failed to duplicate archived chunk name");
 480         }
 481         session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED);
 482
 483         if (!session->quiet_rotation) {
 484                 location = session_get_trace_archive_location(session);
 485                 ret = notification_thread_command_session_rotation_completed(
 486                                 notification_thread_handle,
 487                                 session->name,
 488                                 session->uid,
 489                                 session->gid,
 490                                 session->last_archived_chunk_id.value,
 491                                 location);
 492                 lttng_trace_archive_location_put(location);
 493                 if (ret != LTTNG_OK) {
 494                         ERR("Failed to notify notification thread of completed rotation for session %s",
 495                                         session->name);
 496                 }
 497         }
 498
 499         ret = 0;
 500 check_ongoing_rotation:
 501         if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
 502                 chunk_status = lttng_trace_chunk_get_id(
 503                                 session->chunk_being_archived,
 504                                 &chunk_being_archived_id);
 505                 LTTNG_ASSERT(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 506
 507                 DBG("Rotation of trace archive %" PRIu64 " is still pending for session %s",
 508                                 chunk_being_archived_id, session->name);
 509                 ret = timer_session_rotation_pending_check_start(session,
 510                                 DEFAULT_ROTATE_PENDING_TIMER);
 511                 if (ret) {
 512                         ERR("Failed to re-enable rotation pending timer");
 513                         ret = -1;
 514                         goto end;
 515                 }
 516         }
 517
 518 end:
 519         return ret;
 520 }
 521
 522 /* Call with the session and session_list locks held. */
 523 static
 524 int launch_session_rotation(struct ltt_session *session)
 525 {
 526         int ret;
 527         struct lttng_rotate_session_return rotation_return;
 528
 529         DBG("Launching scheduled time-based rotation on session \"%s\"",
 530                         session->name);
 531
 532         ret = cmd_rotate_session(session, &rotation_return, false,
 533                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 534         if (ret == LTTNG_OK) {
 535                 DBG("Scheduled time-based rotation successfully launched on session \"%s\"",
 536                                 session->name);
 537         } else {
 538                 /* Don't consider errors as fatal. */
 539                 DBG("Scheduled time-based rotation aborted for session %s: %s",
 540                                 session->name, lttng_strerror(ret));
 541         }
 542         return 0;
 543 }
 544
 545 static
 546 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
 547                 struct notification_thread_handle *notification_thread_handle)
 548 {
 549         int ret;
 550
 551         switch (job->type) {
 552         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
 553                 ret = launch_session_rotation(session);
 554                 break;
 555         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
 556                 ret = check_session_rotation_pending(session,
 557                                 notification_thread_handle);
 558                 break;
 559         default:
 560                 abort();
 561         }
 562         return ret;
 563 }
 564
 565 static
 566 int handle_job_queue(struct rotation_thread_handle *handle,
 567                 struct rotation_thread *state __attribute__((unused)),
 568                 struct rotation_thread_timer_queue *queue)
 569 {
 570         int ret = 0;
 571
 572         for (;;) {
 573                 struct ltt_session *session;
 574                 struct rotation_thread_job *job;
 575
 576                 /* Take the queue lock only to pop an element from the list. */
 577                 pthread_mutex_lock(&queue->lock);
 578                 if (cds_list_empty(&queue->list)) {
 579                         pthread_mutex_unlock(&queue->lock);
 580                         break;
 581                 }
 582                 job = cds_list_first_entry(&queue->list,
 583                                 typeof(*job), head);
 584                 cds_list_del(&job->head);
 585                 pthread_mutex_unlock(&queue->lock);
 586
 587                 session_lock_list();
 588                 session = job->session;
 589                 if (!session) {
 590                         DBG("Session \"%s\" not found",
 591                                         session->name != NULL ? session->name : "");
 592                         /*
 593                          * This is a non-fatal error, and we cannot report it to
 594                          * the user (timer), so just print the error and
 595                          * continue the processing.
 596                          *
 597                          * While the timer thread will purge pending signals for
 598                          * a session on the session's destruction, it is
 599                          * possible for a job targeting that session to have
 600                          * already been queued before it was destroyed.
 601                          */
 602                         free(job);
 603                         session_put(session);
 604                         session_unlock_list();
 605                         continue;
 606                 }
 607
 608                 session_lock(session);
 609                 ret = run_job(job, session, handle->notification_thread_handle);
 610                 session_unlock(session);
 611                 /* Release reference held by the job. */
 612                 session_put(session);
 613                 session_unlock_list();
 614                 free(job);
 615                 if (ret) {
 616                         goto end;
 617                 }
 618         }
 619
 620         ret = 0;
 621
 622 end:
 623         return ret;
 624 }
 625
 626 static
 627 int handle_condition(const struct lttng_condition *condition,
 628                 const struct lttng_evaluation *evaluation,
 629                 struct notification_thread_handle *notification_thread_handle)
 630 {
 631         int ret = 0;
 632         const char *condition_session_name = NULL;
 633         enum lttng_condition_type condition_type;
 634         enum lttng_condition_status condition_status;
 635         enum lttng_evaluation_status evaluation_status;
 636         uint64_t consumed;
 637         struct ltt_session *session;
 638
 639         condition_type = lttng_condition_get_type(condition);
 640
 641         if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
 642                 ret = -1;
 643                 ERR("Condition type and session usage type are not the same");
 644                 goto end;
 645         }
 646
 647         /* Fetch info to test */
 648         condition_status = lttng_condition_session_consumed_size_get_session_name(
 649                         condition, &condition_session_name);
 650         if (condition_status != LTTNG_CONDITION_STATUS_OK) {
 651                 ERR("Session name could not be fetched");
 652                 ret = -1;
 653                 goto end;
 654         }
 655         evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
 656                         &consumed);
 657         if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
 658                 ERR("Failed to get evaluation");
 659                 ret = -1;
 660                 goto end;
 661         }
 662
 663         session_lock_list();
 664         session = session_find_by_name(condition_session_name);
 665         if (!session) {
 666                 DBG("Failed to find session while handling notification: notification type = %s, session name = `%s`",
 667                                 lttng_condition_type_str(condition_type),
 668                                 condition_session_name);
 669                 /*
 670                  * Not a fatal error: a session can be destroyed before we get
 671                  * the chance to handle the notification.
 672                  */
 673                 ret = 0;
 674                 session_unlock_list();
 675                 goto end;
 676         }
 677         session_lock(session);
 678
 679         ret = unsubscribe_session_consumed_size_rotation(session,
 680                         notification_thread_handle);
 681         if (ret) {
 682                 goto end_unlock;
 683         }
 684
 685         ret = cmd_rotate_session(session, NULL, false,
 686                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 687         if (ret == -LTTNG_ERR_ROTATION_PENDING) {
 688                 DBG("Rotate already pending, subscribe to the next threshold value");
 689         } else if (ret != LTTNG_OK) {
 690                 ERR("Failed to rotate on size notification with error: %s",
 691                                 lttng_strerror(ret));
 692                 ret = -1;
 693                 goto end_unlock;
 694         }
 695         ret = subscribe_session_consumed_size_rotation(session,
 696                         consumed + session->rotate_size,
 697                         notification_thread_handle);
 698         if (ret) {
 699                 ERR("Failed to subscribe to session consumed size condition");
 700                 goto end_unlock;
 701         }
 702         ret = 0;
 703
 704 end_unlock:
 705         session_unlock(session);
 706         session_put(session);
 707         session_unlock_list();
 708 end:
 709         return ret;
 710 }
 711
 712 static
 713 int handle_notification_channel(int fd __attribute__((unused)),
 714                 struct rotation_thread_handle *handle,
 715                 struct rotation_thread *state __attribute__((unused)))
 716 {
 717         int ret;
 718         bool notification_pending;
 719         struct lttng_notification *notification = NULL;
 720         enum lttng_notification_channel_status status;
 721         const struct lttng_evaluation *notification_evaluation;
 722         const struct lttng_condition *notification_condition;
 723
 724         status = lttng_notification_channel_has_pending_notification(
 725                         rotate_notification_channel, &notification_pending);
 726         if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
 727                 ERR("Error occurred while checking for pending notification");
 728                 ret = -1;
 729                 goto end;
 730         }
 731
 732         if (!notification_pending) {
 733                 ret = 0;
 734                 goto end;
 735         }
 736
 737         /* Receive the next notification. */
 738         status = lttng_notification_channel_get_next_notification(
 739                         rotate_notification_channel,
 740                         &notification);
 741
 742         switch (status) {
 743         case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
 744                 break;
 745         case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
 746                 /* Not an error, we will wait for the next one */
 747                 ret = 0;
 748                 goto end;;
 749         case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
 750                 ERR("Notification channel was closed");
 751                 ret = -1;
 752                 goto end;
 753         default:
 754                 /* Unhandled conditions / errors. */
 755                 ERR("Unknown notification channel status");
 756                 ret = -1;
 757                 goto end;
 758         }
 759
 760         notification_condition = lttng_notification_get_condition(notification);
 761         notification_evaluation = lttng_notification_get_evaluation(notification);
 762
 763         ret = handle_condition(notification_condition, notification_evaluation,
 764                         handle->notification_thread_handle);
 765
 766 end:
 767         lttng_notification_destroy(notification);
 768         return ret;
 769 }
 770
 771 static
 772 void *thread_rotation(void *data)
 773 {
 774         int ret;
 775         struct rotation_thread_handle *handle = (rotation_thread_handle *) data;
 776         struct rotation_thread thread;
 777         int queue_pipe_fd;
 778
 779         DBG("Started rotation thread");
 780         rcu_register_thread();
 781         rcu_thread_online();
 782         health_register(the_health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
 783         health_code_update();
 784
 785         if (!handle) {
 786                 ERR("Invalid thread context provided");
 787                 goto end;
 788         }
 789
 790         queue_pipe_fd = lttng_pipe_get_readfd(
 791                         handle->rotation_timer_queue->event_pipe);
 792
 793
 794         ret = init_thread_state(handle, &thread);
 795         if (ret) {
 796                 goto error;
 797         }
 798
 799         while (true) {
 800                 int fd_count, i;
 801
 802                 health_poll_entry();
 803                 DBG("Entering poll wait");
 804                 ret = lttng_poll_wait(&thread.events, -1);
 805                 DBG("Poll wait returned (%i)", ret);
 806                 health_poll_exit();
 807                 if (ret < 0) {
 808                         /*
 809                          * Restart interrupted system call.
 810                          */
 811                         if (errno == EINTR) {
 812                                 continue;
 813                         }
 814                         ERR("Error encountered during lttng_poll_wait (%i)", ret);
 815                         goto error;
 816                 }
 817
 818                 fd_count = ret;
 819                 for (i = 0; i < fd_count; i++) {
 820                         int fd = LTTNG_POLL_GETFD(&thread.events, i);
 821                         uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
 822
 823                         DBG("Handling fd (%i) activity (%u)",
 824                                         fd, revents);
 825
 826                         if (revents & LPOLLERR) {
 827                                 ERR("Polling returned an error on fd %i", fd);
 828                                 goto error;
 829                         }
 830
 831                         if (fd == rotate_notification_channel->socket) {
 832                                 ret = handle_notification_channel(fd, handle,
 833                                                 &thread);
 834                                 if (ret) {
 835                                         ERR("Error occurred while handling activity on notification channel socket");
 836                                         goto error;
 837                                 }
 838                         } else {
 839                                 /* Job queue or quit pipe activity. */
 840
 841                                 /*
 842                                  * The job queue is serviced if there is
 843                                  * activity on the quit pipe to ensure it is
 844                                  * flushed and all references held in the queue
 845                                  * are released.
 846                                  */
 847                                 ret = handle_job_queue(handle, &thread,
 848                                                 handle->rotation_timer_queue);
 849                                 if (ret) {
 850                                         ERR("Failed to handle rotation timer pipe event");
 851                                         goto error;
 852                                 }
 853
 854                                 if (fd == queue_pipe_fd) {
 855                                         char buf;
 856
 857                                         ret = lttng_read(fd, &buf, 1);
 858                                         if (ret != 1) {
 859                                                 ERR("Failed to read from wakeup pipe (fd = %i)", fd);
 860                                                 goto error;
 861                                         }
 862                                 } else {
 863                                         DBG("Quit pipe activity");
 864                                         goto exit;
 865                                 }
 866                         }
 867                 }
 868         }
 869 exit:
 870 error:
 871         DBG("Thread exit");
 872         fini_thread_state(&thread);
 873 end:
 874         health_unregister(the_health_sessiond);
 875         rcu_thread_offline();
 876         rcu_unregister_thread();
 877         return NULL;
 878 }
 879
 880 static
 881 bool shutdown_rotation_thread(void *thread_data)
 882 {
 883         struct rotation_thread_handle *handle = (rotation_thread_handle *) thread_data;
 884         const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
 885
 886         return notify_thread_pipe(write_fd) == 1;
 887 }
 888
 889 bool launch_rotation_thread(struct rotation_thread_handle *handle)
 890 {
 891         struct lttng_thread *thread;
 892
 893         thread = lttng_thread_create("Rotation",
 894                         thread_rotation,
 895                         shutdown_rotation_thread,
 896                         NULL,
 897                         handle);
 898         if (!thread) {
 899                 goto error;
 900         }
 901         lttng_thread_put(thread);
 902         return true;
 903 error:
 904         return false;
 905 }