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 #include <lttng/location-internal.h>
  29
  30 #include "rotation-thread.h"
  31 #include "lttng-sessiond.h"
  32 #include "health-sessiond.h"
  33 #include "rotate.h"
  34 #include "cmd.h"
  35 #include "session.h"
  36 #include "timer.h"
  37 #include "notification-thread-commands.h"
  38 #include "utils.h"
  39 #include "thread.h"
  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(sizeof(*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         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(sizeof(*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(sizeof(struct 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                 if (errno == EAGAIN || errno == EWOULDBLOCK) {
 219                         /*
 220                          * Not an error, but would be surprising and indicate
 221                          * that the rotation thread can't keep up with the
 222                          * current load.
 223                          */
 224                         DBG("Wake-up pipe of rotation thread job queue is full");
 225                         goto end;
 226                 }
 227                 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
 228                                 job_type_str, session->name);
 229                 goto end;
 230         }
 231
 232 end:
 233         pthread_mutex_unlock(&queue->lock);
 234 }
 235
 236 static
 237 int init_poll_set(struct lttng_poll_event *poll_set,
 238                 struct rotation_thread_handle *handle)
 239 {
 240         int ret;
 241
 242         /*
 243          * Create pollset with size 3:
 244          *      - rotation thread quit pipe,
 245          *      - rotation thread timer queue pipe,
 246          *      - notification channel sock,
 247          */
 248         ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
 249         if (ret < 0) {
 250                 goto error;
 251         }
 252
 253         ret = lttng_poll_add(poll_set,
 254                         lttng_pipe_get_readfd(handle->quit_pipe),
 255                         LPOLLIN | LPOLLERR);
 256         if (ret < 0) {
 257                 ERR("[rotation-thread] Failed to add quit pipe read fd to poll set");
 258                 goto error;
 259         }
 260
 261         ret = lttng_poll_add(poll_set,
 262                         lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
 263                         LPOLLIN | LPOLLERR);
 264         if (ret < 0) {
 265                 ERR("[rotation-thread] Failed to add rotate_pending fd to poll set");
 266                 goto error;
 267         }
 268
 269         return ret;
 270 error:
 271         lttng_poll_clean(poll_set);
 272         return ret;
 273 }
 274
 275 static
 276 void fini_thread_state(struct rotation_thread *state)
 277 {
 278         lttng_poll_clean(&state->events);
 279         if (rotate_notification_channel) {
 280                 lttng_notification_channel_destroy(rotate_notification_channel);
 281         }
 282 }
 283
 284 static
 285 int init_thread_state(struct rotation_thread_handle *handle,
 286                 struct rotation_thread *state)
 287 {
 288         int ret;
 289
 290         memset(state, 0, sizeof(*state));
 291         lttng_poll_init(&state->events);
 292
 293         ret = init_poll_set(&state->events, handle);
 294         if (ret) {
 295                 ERR("[rotation-thread] Failed to initialize rotation thread poll set");
 296                 goto end;
 297         }
 298
 299         rotate_notification_channel = lttng_notification_channel_create(
 300                         lttng_session_daemon_notification_endpoint);
 301         if (!rotate_notification_channel) {
 302                 ERR("[rotation-thread] Could not create notification channel");
 303                 ret = -1;
 304                 goto end;
 305         }
 306         ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
 307                         LPOLLIN | LPOLLERR);
 308         if (ret < 0) {
 309                 ERR("[rotation-thread] Failed to add notification fd to pollset");
 310                 goto end;
 311         }
 312
 313 end:
 314         return ret;
 315 }
 316
 317 static
 318 void check_session_rotation_pending_on_consumers(struct ltt_session *session,
 319                 bool *_rotation_completed)
 320 {
 321         int ret = 0;
 322         struct consumer_socket *socket;
 323         struct cds_lfht_iter iter;
 324         enum consumer_trace_chunk_exists_status exists_status;
 325         uint64_t relayd_id;
 326         bool chunk_exists_on_peer = false;
 327         enum lttng_trace_chunk_status chunk_status;
 328
 329         assert(session->chunk_being_archived);
 330
 331         /*
 332          * Check for a local pending rotation on all consumers (32-bit
 333          * user space, 64-bit user space, and kernel).
 334          */
 335         rcu_read_lock();
 336         if (!session->ust_session) {
 337                 goto skip_ust;
 338         }
 339         cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
 340                         &iter, socket, node.node) {
 341                 relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ?
 342                                 -1ULL :
 343                                 session->ust_session->consumer->net_seq_index;
 344
 345                 pthread_mutex_lock(socket->lock);
 346                 ret = consumer_trace_chunk_exists(socket,
 347                                 relayd_id,
 348                                 session->id, session->chunk_being_archived,
 349                                 &exists_status);
 350                 if (ret) {
 351                         pthread_mutex_unlock(socket->lock);
 352                         ERR("Error occurred while checking rotation status on consumer daemon");
 353                         goto end;
 354                 }
 355
 356                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 357                         pthread_mutex_unlock(socket->lock);
 358                         chunk_exists_on_peer = true;
 359                         goto end;
 360                 }
 361                 pthread_mutex_unlock(socket->lock);
 362         }
 363
 364 skip_ust:
 365         if (!session->kernel_session) {
 366                 goto skip_kernel;
 367         }
 368         cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
 369                                 &iter, socket, node.node) {
 370                 pthread_mutex_lock(socket->lock);
 371                 relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ?
 372                                 -1ULL :
 373                                 session->kernel_session->consumer->net_seq_index;
 374
 375                 ret = consumer_trace_chunk_exists(socket,
 376                                 relayd_id,
 377                                 session->id, session->chunk_being_archived,
 378                                 &exists_status);
 379                 if (ret) {
 380                         pthread_mutex_unlock(socket->lock);
 381                         ERR("Error occurred while checking rotation status on consumer daemon");
 382                         goto end;
 383                 }
 384
 385                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 386                         pthread_mutex_unlock(socket->lock);
 387                         chunk_exists_on_peer = true;
 388                         goto end;
 389                 }
 390                 pthread_mutex_unlock(socket->lock);
 391         }
 392 skip_kernel:
 393 end:
 394         rcu_read_unlock();
 395
 396         if (!chunk_exists_on_peer) {
 397                 uint64_t chunk_being_archived_id;
 398
 399                 chunk_status = lttng_trace_chunk_get_id(
 400                                 session->chunk_being_archived,
 401                                 &chunk_being_archived_id);
 402                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 403                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
 404                                 chunk_being_archived_id,
 405                                 session->name);
 406         }
 407         *_rotation_completed = !chunk_exists_on_peer;
 408         if (ret) {
 409                 ret = session_reset_rotation_state(session,
 410                                 LTTNG_ROTATION_STATE_ERROR);
 411                 if (ret) {
 412                         ERR("Failed to reset rotation state of session \"%s\"",
 413                                         session->name);
 414                 }
 415         }
 416 }
 417
 418 /*
 419  * Check if the last rotation was completed, called with session lock held.
 420  * Should only return non-zero in the event of a fatal error. Doing so will
 421  * shutdown the thread.
 422  */
 423 static
 424 int check_session_rotation_pending(struct ltt_session *session,
 425                 struct notification_thread_handle *notification_thread_handle)
 426 {
 427         int ret;
 428         struct lttng_trace_archive_location *location;
 429         enum lttng_trace_chunk_status chunk_status;
 430         bool rotation_completed = false;
 431         const char *archived_chunk_name;
 432         uint64_t chunk_being_archived_id;
 433
 434         if (!session->chunk_being_archived) {
 435                 ret = 0;
 436                 goto end;
 437         }
 438
 439         chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived,
 440                         &chunk_being_archived_id);
 441         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 442
 443         DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
 444                         session->name, chunk_being_archived_id);
 445
 446         /*
 447          * The rotation-pending check timer of a session is launched in
 448          * one-shot mode. If the rotation is incomplete, the rotation
 449          * thread will re-enable the pending-check timer.
 450          *
 451          * The timer thread can't stop the timer itself since it is involved
 452          * in the check for the timer's quiescence.
 453          */
 454         ret = timer_session_rotation_pending_check_stop(session);
 455         if (ret) {
 456                 goto check_ongoing_rotation;
 457         }
 458
 459         check_session_rotation_pending_on_consumers(session,
 460                         &rotation_completed);
 461         if (!rotation_completed ||
 462                         session->rotation_state == LTTNG_ROTATION_STATE_ERROR) {
 463                 goto check_ongoing_rotation;
 464         }
 465
 466         /*
 467          * Now we can clear the "ONGOING" state in the session. New
 468          * rotations can start now.
 469          */
 470         chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived,
 471                         &archived_chunk_name, NULL);
 472         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 473         free(session->last_archived_chunk_name);
 474         session->last_archived_chunk_name = strdup(archived_chunk_name);
 475         if (!session->last_archived_chunk_name) {
 476                 PERROR("Failed to duplicate archived chunk name");
 477         }
 478         session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED);
 479
 480         if (!session->quiet_rotation) {
 481                 location = session_get_trace_archive_location(session);
 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                 lttng_trace_archive_location_put(location);
 490                 if (ret != LTTNG_OK) {
 491                         ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
 492                                         session->name);
 493                 }
 494         }
 495
 496         ret = 0;
 497 check_ongoing_rotation:
 498         if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
 499                 uint64_t chunk_being_archived_id;
 500
 501                 chunk_status = lttng_trace_chunk_get_id(
 502                                 session->chunk_being_archived,
 503                                 &chunk_being_archived_id);
 504                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 505
 506                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s",
 507                                 chunk_being_archived_id, session->name);
 508                 ret = timer_session_rotation_pending_check_start(session,
 509                                 DEFAULT_ROTATE_PENDING_TIMER);
 510                 if (ret) {
 511                         ERR("Failed to re-enable rotation pending timer");
 512                         ret = -1;
 513                         goto end;
 514                 }
 515         }
 516
 517 end:
 518         return ret;
 519 }
 520
 521 /* Call with the session and session_list locks held. */
 522 static
 523 int launch_session_rotation(struct ltt_session *session)
 524 {
 525         int ret;
 526         struct lttng_rotate_session_return rotation_return;
 527
 528         DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"",
 529                         session->name);
 530
 531         ret = cmd_rotate_session(session, &rotation_return, false,
 532                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 533         if (ret == LTTNG_OK) {
 534                 DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"",
 535                                 session->name);
 536         } else {
 537                 /* Don't consider errors as fatal. */
 538                 DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s",
 539                                 session->name, lttng_strerror(ret));
 540         }
 541         return 0;
 542 }
 543
 544 static
 545 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
 546                 struct notification_thread_handle *notification_thread_handle)
 547 {
 548         int ret;
 549
 550         switch (job->type) {
 551         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
 552                 ret = launch_session_rotation(session);
 553                 break;
 554         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
 555                 ret = check_session_rotation_pending(session,
 556                                 notification_thread_handle);
 557                 break;
 558         default:
 559                 abort();
 560         }
 561         return ret;
 562 }
 563
 564 static
 565 int handle_job_queue(struct rotation_thread_handle *handle,
 566                 struct rotation_thread *state,
 567                 struct rotation_thread_timer_queue *queue)
 568 {
 569         int ret = 0;
 570
 571         for (;;) {
 572                 struct ltt_session *session;
 573                 struct rotation_thread_job *job;
 574
 575                 /* Take the queue lock only to pop an element from the list. */
 576                 pthread_mutex_lock(&queue->lock);
 577                 if (cds_list_empty(&queue->list)) {
 578                         pthread_mutex_unlock(&queue->lock);
 579                         break;
 580                 }
 581                 job = cds_list_first_entry(&queue->list,
 582                                 typeof(*job), head);
 583                 cds_list_del(&job->head);
 584                 pthread_mutex_unlock(&queue->lock);
 585
 586                 session_lock_list();
 587                 session = job->session;
 588                 if (!session) {
 589                         DBG("[rotation-thread] Session \"%s\" not found",
 590                                         session->name);
 591                         /*
 592                          * This is a non-fatal error, and we cannot report it to
 593                          * the user (timer), so just print the error and
 594                          * continue the processing.
 595                          *
 596                          * While the timer thread will purge pending signals for
 597                          * a session on the session's destruction, it is
 598                          * possible for a job targeting that session to have
 599                          * already been queued before it was destroyed.
 600                          */
 601                         free(job);
 602                         session_put(session);
 603                         session_unlock_list();
 604                         continue;
 605                 }
 606
 607                 session_lock(session);
 608                 ret = run_job(job, session, handle->notification_thread_handle);
 609                 session_unlock(session);
 610                 /* Release reference held by the job. */
 611                 session_put(session);
 612                 session_unlock_list();
 613                 free(job);
 614                 if (ret) {
 615                         goto end;
 616                 }
 617         }
 618
 619         ret = 0;
 620
 621 end:
 622         return ret;
 623 }
 624
 625 static
 626 int handle_condition(const struct lttng_condition *condition,
 627                 const struct lttng_evaluation *evaluation,
 628                 struct notification_thread_handle *notification_thread_handle)
 629 {
 630         int ret = 0;
 631         const char *condition_session_name = NULL;
 632         enum lttng_condition_type condition_type;
 633         enum lttng_condition_status condition_status;
 634         enum lttng_evaluation_status evaluation_status;
 635         uint64_t consumed;
 636         struct ltt_session *session;
 637
 638         condition_type = lttng_condition_get_type(condition);
 639
 640         if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
 641                 ret = -1;
 642                 ERR("[rotation-thread] Condition type and session usage type are not the same");
 643                 goto end;
 644         }
 645
 646         /* Fetch info to test */
 647         condition_status = lttng_condition_session_consumed_size_get_session_name(
 648                         condition, &condition_session_name);
 649         if (condition_status != LTTNG_CONDITION_STATUS_OK) {
 650                 ERR("[rotation-thread] Session name could not be fetched");
 651                 ret = -1;
 652                 goto end;
 653         }
 654         evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
 655                         &consumed);
 656         if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
 657                 ERR("[rotation-thread] Failed to get evaluation");
 658                 ret = -1;
 659                 goto end;
 660         }
 661
 662         session_lock_list();
 663         session = session_find_by_name(condition_session_name);
 664         if (!session) {
 665                 ret = -1;
 666                 session_unlock_list();
 667                 ERR("[rotation-thread] Session \"%s\" not found",
 668                                 condition_session_name);
 669                 goto end;
 670         }
 671         session_lock(session);
 672
 673         ret = unsubscribe_session_consumed_size_rotation(session,
 674                         notification_thread_handle);
 675         if (ret) {
 676                 goto end_unlock;
 677         }
 678
 679         ret = cmd_rotate_session(session, NULL, false,
 680                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 681         if (ret == -LTTNG_ERR_ROTATION_PENDING) {
 682                 DBG("Rotate already pending, subscribe to the next threshold value");
 683         } else if (ret != LTTNG_OK) {
 684                 ERR("[rotation-thread] Failed to rotate on size notification with error: %s",
 685                                 lttng_strerror(ret));
 686                 ret = -1;
 687                 goto end_unlock;
 688         }
 689         ret = subscribe_session_consumed_size_rotation(session,
 690                         consumed + session->rotate_size,
 691                         notification_thread_handle);
 692         if (ret) {
 693                 ERR("[rotation-thread] Failed to subscribe to session consumed size condition");
 694                 goto end_unlock;
 695         }
 696         ret = 0;
 697
 698 end_unlock:
 699         session_unlock(session);
 700         session_put(session);
 701         session_unlock_list();
 702 end:
 703         return ret;
 704 }
 705
 706 static
 707 int handle_notification_channel(int fd,
 708                 struct rotation_thread_handle *handle,
 709                 struct rotation_thread *state)
 710 {
 711         int ret;
 712         bool notification_pending;
 713         struct lttng_notification *notification = NULL;
 714         enum lttng_notification_channel_status status;
 715         const struct lttng_evaluation *notification_evaluation;
 716         const struct lttng_condition *notification_condition;
 717
 718         status = lttng_notification_channel_has_pending_notification(
 719                         rotate_notification_channel, &notification_pending);
 720         if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
 721                 ERR("[rotation-thread ]Error occurred while checking for pending notification");
 722                 ret = -1;
 723                 goto end;
 724         }
 725
 726         if (!notification_pending) {
 727                 ret = 0;
 728                 goto end;
 729         }
 730
 731         /* Receive the next notification. */
 732         status = lttng_notification_channel_get_next_notification(
 733                         rotate_notification_channel,
 734                         &notification);
 735
 736         switch (status) {
 737         case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
 738                 break;
 739         case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
 740                 /* Not an error, we will wait for the next one */
 741                 ret = 0;
 742                 goto end;;
 743         case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
 744                 ERR("Notification channel was closed");
 745                 ret = -1;
 746                 goto end;
 747         default:
 748                 /* Unhandled conditions / errors. */
 749                 ERR("Unknown notification channel status");
 750                 ret = -1;
 751                 goto end;
 752         }
 753
 754         notification_condition = lttng_notification_get_condition(notification);
 755         notification_evaluation = lttng_notification_get_evaluation(notification);
 756
 757         ret = handle_condition(notification_condition, notification_evaluation,
 758                         handle->notification_thread_handle);
 759
 760 end:
 761         lttng_notification_destroy(notification);
 762         return ret;
 763 }
 764
 765 static
 766 void *thread_rotation(void *data)
 767 {
 768         int ret;
 769         struct rotation_thread_handle *handle = data;
 770         struct rotation_thread thread;
 771         int queue_pipe_fd;
 772
 773         DBG("[rotation-thread] Started rotation thread");
 774         rcu_register_thread();
 775         rcu_thread_online();
 776         health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
 777         health_code_update();
 778
 779         if (!handle) {
 780                 ERR("[rotation-thread] Invalid thread context provided");
 781                 goto end;
 782         }
 783
 784         queue_pipe_fd = lttng_pipe_get_readfd(
 785                         handle->rotation_timer_queue->event_pipe);
 786
 787
 788         ret = init_thread_state(handle, &thread);
 789         if (ret) {
 790                 goto error;
 791         }
 792
 793         while (true) {
 794                 int fd_count, i;
 795
 796                 health_poll_entry();
 797                 DBG("[rotation-thread] Entering poll wait");
 798                 ret = lttng_poll_wait(&thread.events, -1);
 799                 DBG("[rotation-thread] Poll wait returned (%i)", ret);
 800                 health_poll_exit();
 801                 if (ret < 0) {
 802                         /*
 803                          * Restart interrupted system call.
 804                          */
 805                         if (errno == EINTR) {
 806                                 continue;
 807                         }
 808                         ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret);
 809                         goto error;
 810                 }
 811
 812                 fd_count = ret;
 813                 for (i = 0; i < fd_count; i++) {
 814                         int fd = LTTNG_POLL_GETFD(&thread.events, i);
 815                         uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
 816
 817                         DBG("[rotation-thread] Handling fd (%i) activity (%u)",
 818                                         fd, revents);
 819
 820                         if (revents & LPOLLERR) {
 821                                 ERR("[rotation-thread] Polling returned an error on fd %i", fd);
 822                                 goto error;
 823                         }
 824
 825                         if (fd == rotate_notification_channel->socket) {
 826                                 ret = handle_notification_channel(fd, handle,
 827                                                 &thread);
 828                                 if (ret) {
 829                                         ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
 830                                         goto error;
 831                                 }
 832                         } else {
 833                                 /* Job queue or quit pipe activity. */
 834
 835                                 /*
 836                                  * The job queue is serviced if there is
 837                                  * activity on the quit pipe to ensure it is
 838                                  * flushed and all references held in the queue
 839                                  * are released.
 840                                  */
 841                                 ret = handle_job_queue(handle, &thread,
 842                                                 handle->rotation_timer_queue);
 843                                 if (ret) {
 844                                         ERR("[rotation-thread] Failed to handle rotation timer pipe event");
 845                                         goto error;
 846                                 }
 847
 848                                 if (fd == queue_pipe_fd) {
 849                                         char buf;
 850
 851                                         ret = lttng_read(fd, &buf, 1);
 852                                         if (ret != 1) {
 853                                                 ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
 854                                                 goto error;
 855                                         }
 856                                 } else {
 857                                         DBG("[rotation-thread] Quit pipe activity");
 858                                         goto exit;
 859                                 }
 860                         }
 861                 }
 862         }
 863 exit:
 864 error:
 865         DBG("[rotation-thread] Exit");
 866         fini_thread_state(&thread);
 867 end:
 868         health_unregister(health_sessiond);
 869         rcu_thread_offline();
 870         rcu_unregister_thread();
 871         return NULL;
 872 }
 873
 874 static
 875 bool shutdown_rotation_thread(void *thread_data)
 876 {
 877         struct rotation_thread_handle *handle = thread_data;
 878         const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
 879
 880         return notify_thread_pipe(write_fd) == 1;
 881 }
 882
 883 bool launch_rotation_thread(struct rotation_thread_handle *handle)
 884 {
 885         struct lttng_thread *thread;
 886
 887         thread = lttng_thread_create("Rotation",
 888                         thread_rotation,
 889                         shutdown_rotation_thread,
 890                         NULL,
 891                         handle);
 892         if (!thread) {
 893                 goto error;
 894         }
 895         lttng_thread_put(thread);
 896         return true;
 897 error:
 898         return false;
 899 }