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  * This program is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License, version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
  12  * more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along with
  15  * this program; if not, write to the Free Software Foundation, Inc., 51
  16  * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _LGPL_SOURCE
  20 #include <lttng/trigger/trigger.h>
  21 #include <common/error.h>
  22 #include <common/config/session-config.h>
  23 #include <common/defaults.h>
  24 #include <common/utils.h>
  25 #include <common/futex.h>
  26 #include <common/align.h>
  27 #include <common/time.h>
  28 #include <common/hashtable/utils.h>
  29 #include <sys/eventfd.h>
  30 #include <sys/stat.h>
  31 #include <time.h>
  32 #include <signal.h>
  33 #include <inttypes.h>
  34
  35 #include <common/kernel-ctl/kernel-ctl.h>
  36 #include <lttng/notification/channel-internal.h>
  37 #include <lttng/rotate-internal.h>
  38
  39 #include "rotation-thread.h"
  40 #include "lttng-sessiond.h"
  41 #include "health-sessiond.h"
  42 #include "rotate.h"
  43 #include "cmd.h"
  44 #include "session.h"
  45 #include "timer.h"
  46 #include "notification-thread-commands.h"
  47 #include "utils.h"
  48 #include "thread.h"
  49
  50 #include <urcu.h>
  51 #include <urcu/list.h>
  52
  53 struct lttng_notification_channel *rotate_notification_channel = NULL;
  54
  55 struct rotation_thread {
  56         struct lttng_poll_event events;
  57 };
  58
  59 struct rotation_thread_job {
  60         enum rotation_thread_job_type type;
  61         struct ltt_session *session;
  62         /* List member in struct rotation_thread_timer_queue. */
  63         struct cds_list_head head;
  64 };
  65
  66 /*
  67  * The timer thread enqueues jobs and wakes up the rotation thread.
  68  * When the rotation thread wakes up, it empties the queue.
  69  */
  70 struct rotation_thread_timer_queue {
  71         struct lttng_pipe *event_pipe;
  72         struct cds_list_head list;
  73         pthread_mutex_t lock;
  74 };
  75
  76 struct rotation_thread_handle {
  77         struct rotation_thread_timer_queue *rotation_timer_queue;
  78         /* Access to the notification thread cmd_queue */
  79         struct notification_thread_handle *notification_thread_handle;
  80         /* Thread-specific quit pipe. */
  81         struct lttng_pipe *quit_pipe;
  82 };
  83
  84 static
  85 const char *get_job_type_str(enum rotation_thread_job_type job_type)
  86 {
  87         switch (job_type) {
  88         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
  89                 return "CHECK_PENDING_ROTATION";
  90         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
  91                 return "SCHEDULED_ROTATION";
  92         default:
  93                 abort();
  94         }
  95 }
  96
  97 struct rotation_thread_timer_queue *rotation_thread_timer_queue_create(void)
  98 {
  99         struct rotation_thread_timer_queue *queue = NULL;
 100
 101         queue = zmalloc(sizeof(*queue));
 102         if (!queue) {
 103                 PERROR("Failed to allocate timer rotate queue");
 104                 goto end;
 105         }
 106
 107         queue->event_pipe = lttng_pipe_open(FD_CLOEXEC | O_NONBLOCK);
 108         CDS_INIT_LIST_HEAD(&queue->list);
 109         pthread_mutex_init(&queue->lock, NULL);
 110 end:
 111         return queue;
 112 }
 113
 114 void rotation_thread_timer_queue_destroy(
 115                 struct rotation_thread_timer_queue *queue)
 116 {
 117         if (!queue) {
 118                 return;
 119         }
 120
 121         lttng_pipe_destroy(queue->event_pipe);
 122
 123         pthread_mutex_lock(&queue->lock);
 124         assert(cds_list_empty(&queue->list));
 125         pthread_mutex_unlock(&queue->lock);
 126         pthread_mutex_destroy(&queue->lock);
 127         free(queue);
 128 }
 129
 130 /*
 131  * Destroy the thread data previously created by the init function.
 132  */
 133 void rotation_thread_handle_destroy(
 134                 struct rotation_thread_handle *handle)
 135 {
 136         lttng_pipe_destroy(handle->quit_pipe);
 137         free(handle);
 138 }
 139
 140 struct rotation_thread_handle *rotation_thread_handle_create(
 141                 struct rotation_thread_timer_queue *rotation_timer_queue,
 142                 struct notification_thread_handle *notification_thread_handle)
 143 {
 144         struct rotation_thread_handle *handle;
 145
 146         handle = zmalloc(sizeof(*handle));
 147         if (!handle) {
 148                 goto end;
 149         }
 150
 151         handle->rotation_timer_queue = rotation_timer_queue;
 152         handle->notification_thread_handle = notification_thread_handle;
 153         handle->quit_pipe = lttng_pipe_open(FD_CLOEXEC);
 154         if (!handle->quit_pipe) {
 155                 goto error;
 156         }
 157
 158 end:
 159         return handle;
 160 error:
 161         rotation_thread_handle_destroy(handle);
 162         return NULL;
 163 }
 164
 165 /*
 166  * Called with the rotation_thread_timer_queue lock held.
 167  * Return true if the same timer job already exists in the queue, false if not.
 168  */
 169 static
 170 bool timer_job_exists(const struct rotation_thread_timer_queue *queue,
 171                 enum rotation_thread_job_type job_type,
 172                 struct ltt_session *session)
 173 {
 174         bool exists = false;
 175         struct rotation_thread_job *job;
 176
 177         cds_list_for_each_entry(job, &queue->list, head) {
 178                 if (job->session == session && job->type == job_type) {
 179                         exists = true;
 180                         goto end;
 181                 }
 182         }
 183 end:
 184         return exists;
 185 }
 186
 187 void rotation_thread_enqueue_job(struct rotation_thread_timer_queue *queue,
 188                 enum rotation_thread_job_type job_type,
 189                 struct ltt_session *session)
 190 {
 191         int ret;
 192         const char dummy = '!';
 193         struct rotation_thread_job *job = NULL;
 194         const char *job_type_str = get_job_type_str(job_type);
 195
 196         pthread_mutex_lock(&queue->lock);
 197         if (timer_job_exists(queue, job_type, session)) {
 198                 /*
 199                  * This timer job is already pending, we don't need to add
 200                  * it.
 201                  */
 202                 goto end;
 203         }
 204
 205         job = zmalloc(sizeof(struct rotation_thread_job));
 206         if (!job) {
 207                 PERROR("Failed to allocate rotation thread job of type \"%s\" for session \"%s\"",
 208                                 job_type_str, session->name);
 209                 goto end;
 210         }
 211         /* No reason for this to fail as the caller must hold a reference. */
 212         (void) session_get(session);
 213
 214         job->session = session;
 215         job->type = job_type;
 216         cds_list_add_tail(&job->head, &queue->list);
 217
 218         ret = lttng_write(lttng_pipe_get_writefd(queue->event_pipe), &dummy,
 219                         sizeof(dummy));
 220         if (ret < 0) {
 221                 /*
 222                  * We do not want to block in the timer handler, the job has
 223                  * been enqueued in the list, the wakeup pipe is probably full,
 224                  * the job will be processed when the rotation_thread catches
 225                  * up.
 226                  */
 227                 if (errno == EAGAIN || errno == EWOULDBLOCK) {
 228                         /*
 229                          * Not an error, but would be surprising and indicate
 230                          * that the rotation thread can't keep up with the
 231                          * current load.
 232                          */
 233                         DBG("Wake-up pipe of rotation thread job queue is full");
 234                         goto end;
 235                 }
 236                 PERROR("Failed to wake-up the rotation thread after pushing a job of type \"%s\" for session \"%s\"",
 237                                 job_type_str, session->name);
 238                 goto end;
 239         }
 240
 241 end:
 242         pthread_mutex_unlock(&queue->lock);
 243 }
 244
 245 static
 246 int init_poll_set(struct lttng_poll_event *poll_set,
 247                 struct rotation_thread_handle *handle)
 248 {
 249         int ret;
 250
 251         /*
 252          * Create pollset with size 3:
 253          *      - rotation thread quit pipe,
 254          *      - rotation thread timer queue pipe,
 255          *      - notification channel sock,
 256          */
 257         ret = lttng_poll_create(poll_set, 5, LTTNG_CLOEXEC);
 258         if (ret < 0) {
 259                 goto error;
 260         }
 261
 262         ret = lttng_poll_add(poll_set,
 263                         lttng_pipe_get_readfd(handle->quit_pipe),
 264                         LPOLLIN | LPOLLERR);
 265         if (ret < 0) {
 266                 ERR("[rotation-thread] Failed to add quit pipe read fd to poll set");
 267                 goto error;
 268         }
 269
 270         ret = lttng_poll_add(poll_set,
 271                         lttng_pipe_get_readfd(handle->rotation_timer_queue->event_pipe),
 272                         LPOLLIN | LPOLLERR);
 273         if (ret < 0) {
 274                 ERR("[rotation-thread] Failed to add rotate_pending fd to poll set");
 275                 goto error;
 276         }
 277
 278         return ret;
 279 error:
 280         lttng_poll_clean(poll_set);
 281         return ret;
 282 }
 283
 284 static
 285 void fini_thread_state(struct rotation_thread *state)
 286 {
 287         lttng_poll_clean(&state->events);
 288         if (rotate_notification_channel) {
 289                 lttng_notification_channel_destroy(rotate_notification_channel);
 290         }
 291 }
 292
 293 static
 294 int init_thread_state(struct rotation_thread_handle *handle,
 295                 struct rotation_thread *state)
 296 {
 297         int ret;
 298
 299         memset(state, 0, sizeof(*state));
 300         lttng_poll_init(&state->events);
 301
 302         ret = init_poll_set(&state->events, handle);
 303         if (ret) {
 304                 ERR("[rotation-thread] Failed to initialize rotation thread poll set");
 305                 goto end;
 306         }
 307
 308         rotate_notification_channel = lttng_notification_channel_create(
 309                         lttng_session_daemon_notification_endpoint);
 310         if (!rotate_notification_channel) {
 311                 ERR("[rotation-thread] Could not create notification channel");
 312                 ret = -1;
 313                 goto end;
 314         }
 315         ret = lttng_poll_add(&state->events, rotate_notification_channel->socket,
 316                         LPOLLIN | LPOLLERR);
 317         if (ret < 0) {
 318                 ERR("[rotation-thread] Failed to add notification fd to pollset");
 319                 goto end;
 320         }
 321
 322 end:
 323         return ret;
 324 }
 325
 326 static
 327 void check_session_rotation_pending_on_consumers(struct ltt_session *session,
 328                 bool *_rotation_completed)
 329 {
 330         int ret = 0;
 331         struct consumer_socket *socket;
 332         struct cds_lfht_iter iter;
 333         enum consumer_trace_chunk_exists_status exists_status;
 334         uint64_t relayd_id;
 335         bool chunk_exists_on_peer = false;
 336         enum lttng_trace_chunk_status chunk_status;
 337
 338         assert(session->chunk_being_archived);
 339
 340         /*
 341          * Check for a local pending rotation on all consumers (32-bit
 342          * user space, 64-bit user space, and kernel).
 343          */
 344         rcu_read_lock();
 345         if (!session->ust_session) {
 346                 goto skip_ust;
 347         }
 348         cds_lfht_for_each_entry(session->ust_session->consumer->socks->ht,
 349                         &iter, socket, node.node) {
 350                 relayd_id = session->ust_session->consumer->type == CONSUMER_DST_LOCAL ?
 351                                 -1ULL :
 352                                 session->ust_session->consumer->net_seq_index;
 353
 354                 pthread_mutex_lock(socket->lock);
 355                 ret = consumer_trace_chunk_exists(socket,
 356                                 relayd_id,
 357                                 session->id, session->chunk_being_archived,
 358                                 &exists_status);
 359                 if (ret) {
 360                         pthread_mutex_unlock(socket->lock);
 361                         ERR("Error occurred while checking rotation status on consumer daemon");
 362                         goto end;
 363                 }
 364
 365                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 366                         pthread_mutex_unlock(socket->lock);
 367                         chunk_exists_on_peer = true;
 368                         goto end;
 369                 }
 370                 pthread_mutex_unlock(socket->lock);
 371         }
 372
 373 skip_ust:
 374         if (!session->kernel_session) {
 375                 goto skip_kernel;
 376         }
 377         cds_lfht_for_each_entry(session->kernel_session->consumer->socks->ht,
 378                                 &iter, socket, node.node) {
 379                 pthread_mutex_lock(socket->lock);
 380                 relayd_id = session->kernel_session->consumer->type == CONSUMER_DST_LOCAL ?
 381                                 -1ULL :
 382                                 session->kernel_session->consumer->net_seq_index;
 383
 384                 ret = consumer_trace_chunk_exists(socket,
 385                                 relayd_id,
 386                                 session->id, session->chunk_being_archived,
 387                                 &exists_status);
 388                 if (ret) {
 389                         pthread_mutex_unlock(socket->lock);
 390                         ERR("Error occurred while checking rotation status on consumer daemon");
 391                         goto end;
 392                 }
 393
 394                 if (exists_status != CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK) {
 395                         pthread_mutex_unlock(socket->lock);
 396                         chunk_exists_on_peer = true;
 397                         goto end;
 398                 }
 399                 pthread_mutex_unlock(socket->lock);
 400         }
 401 skip_kernel:
 402 end:
 403         rcu_read_unlock();
 404
 405         if (!chunk_exists_on_peer) {
 406                 uint64_t chunk_being_archived_id;
 407
 408                 chunk_status = lttng_trace_chunk_get_id(
 409                                 session->chunk_being_archived,
 410                                 &chunk_being_archived_id);
 411                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 412                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " of session \"%s\" is complete on all consumers",
 413                                 chunk_being_archived_id,
 414                                 session->name);
 415         }
 416         *_rotation_completed = !chunk_exists_on_peer;
 417         if (ret) {
 418                 ret = session_reset_rotation_state(session,
 419                                 LTTNG_ROTATION_STATE_ERROR);
 420                 if (ret) {
 421                         ERR("Failed to reset rotation state of session \"%s\"",
 422                                         session->name);
 423                 }
 424         }
 425 }
 426
 427 /*
 428  * Check if the last rotation was completed, called with session lock held.
 429  * Should only return non-zero in the event of a fatal error. Doing so will
 430  * shutdown the thread.
 431  */
 432 static
 433 int check_session_rotation_pending(struct ltt_session *session,
 434                 struct notification_thread_handle *notification_thread_handle)
 435 {
 436         int ret;
 437         struct lttng_trace_archive_location *location;
 438         enum lttng_trace_chunk_status chunk_status;
 439         bool rotation_completed = false;
 440         const char *archived_chunk_name;
 441         uint64_t chunk_being_archived_id;
 442
 443         if (!session->chunk_being_archived) {
 444                 ret = 0;
 445                 goto end;
 446         }
 447
 448         chunk_status = lttng_trace_chunk_get_id(session->chunk_being_archived,
 449                         &chunk_being_archived_id);
 450         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 451
 452         DBG("[rotation-thread] Checking for pending rotation on session \"%s\", trace archive %" PRIu64,
 453                         session->name, chunk_being_archived_id);
 454
 455         /*
 456          * The rotation-pending check timer of a session is launched in
 457          * one-shot mode. If the rotation is incomplete, the rotation
 458          * thread will re-enable the pending-check timer.
 459          *
 460          * The timer thread can't stop the timer itself since it is involved
 461          * in the check for the timer's quiescence.
 462          */
 463         ret = timer_session_rotation_pending_check_stop(session);
 464         if (ret) {
 465                 goto check_ongoing_rotation;
 466         }
 467
 468         check_session_rotation_pending_on_consumers(session,
 469                         &rotation_completed);
 470         if (!rotation_completed ||
 471                         session->rotation_state == LTTNG_ROTATION_STATE_ERROR) {
 472                 goto check_ongoing_rotation;
 473         }
 474
 475         /*
 476          * Now we can clear the "ONGOING" state in the session. New
 477          * rotations can start now.
 478          */
 479         chunk_status = lttng_trace_chunk_get_name(session->chunk_being_archived,
 480                         &archived_chunk_name, NULL);
 481         assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 482         free(session->last_archived_chunk_name);
 483         session->last_archived_chunk_name = strdup(archived_chunk_name);
 484         if (!session->last_archived_chunk_name) {
 485                 PERROR("Failed to duplicate archived chunk name");
 486         }
 487         session_reset_rotation_state(session, LTTNG_ROTATION_STATE_COMPLETED);
 488
 489         if (!session->quiet_rotation) {
 490                 location = session_get_trace_archive_location(session);
 491                 /* Ownership of location is transferred. */
 492                 ret = notification_thread_command_session_rotation_completed(
 493                                 notification_thread_handle,
 494                                 session->name,
 495                                 session->uid,
 496                                 session->gid,
 497                                 session->last_archived_chunk_id.value,
 498                                 location);
 499                 if (ret != LTTNG_OK) {
 500                         ERR("[rotation-thread] Failed to notify notification thread of completed rotation for session %s",
 501                                         session->name);
 502                 }
 503         }
 504
 505         ret = 0;
 506 check_ongoing_rotation:
 507         if (session->rotation_state == LTTNG_ROTATION_STATE_ONGOING) {
 508                 uint64_t chunk_being_archived_id;
 509
 510                 chunk_status = lttng_trace_chunk_get_id(
 511                                 session->chunk_being_archived,
 512                                 &chunk_being_archived_id);
 513                 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
 514
 515                 DBG("[rotation-thread] Rotation of trace archive %" PRIu64 " is still pending for session %s",
 516                                 chunk_being_archived_id, session->name);
 517                 ret = timer_session_rotation_pending_check_start(session,
 518                                 DEFAULT_ROTATE_PENDING_TIMER);
 519                 if (ret) {
 520                         ERR("Failed to re-enable rotation pending timer");
 521                         ret = -1;
 522                         goto end;
 523                 }
 524         }
 525
 526 end:
 527         return ret;
 528 }
 529
 530 /* Call with the session and session_list locks held. */
 531 static
 532 int launch_session_rotation(struct ltt_session *session)
 533 {
 534         int ret;
 535         struct lttng_rotate_session_return rotation_return;
 536
 537         DBG("[rotation-thread] Launching scheduled time-based rotation on session \"%s\"",
 538                         session->name);
 539
 540         ret = cmd_rotate_session(session, &rotation_return, false,
 541                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 542         if (ret == LTTNG_OK) {
 543                 DBG("[rotation-thread] Scheduled time-based rotation successfully launched on session \"%s\"",
 544                                 session->name);
 545         } else {
 546                 /* Don't consider errors as fatal. */
 547                 DBG("[rotation-thread] Scheduled time-based rotation aborted for session %s: %s",
 548                                 session->name, lttng_strerror(ret));
 549         }
 550         return 0;
 551 }
 552
 553 static
 554 int run_job(struct rotation_thread_job *job, struct ltt_session *session,
 555                 struct notification_thread_handle *notification_thread_handle)
 556 {
 557         int ret;
 558
 559         switch (job->type) {
 560         case ROTATION_THREAD_JOB_TYPE_SCHEDULED_ROTATION:
 561                 ret = launch_session_rotation(session);
 562                 break;
 563         case ROTATION_THREAD_JOB_TYPE_CHECK_PENDING_ROTATION:
 564                 ret = check_session_rotation_pending(session,
 565                                 notification_thread_handle);
 566                 break;
 567         default:
 568                 abort();
 569         }
 570         return ret;
 571 }
 572
 573 static
 574 int handle_job_queue(struct rotation_thread_handle *handle,
 575                 struct rotation_thread *state,
 576                 struct rotation_thread_timer_queue *queue)
 577 {
 578         int ret = 0;
 579
 580         for (;;) {
 581                 struct ltt_session *session;
 582                 struct rotation_thread_job *job;
 583
 584                 /* Take the queue lock only to pop an element from the list. */
 585                 pthread_mutex_lock(&queue->lock);
 586                 if (cds_list_empty(&queue->list)) {
 587                         pthread_mutex_unlock(&queue->lock);
 588                         break;
 589                 }
 590                 job = cds_list_first_entry(&queue->list,
 591                                 typeof(*job), head);
 592                 cds_list_del(&job->head);
 593                 pthread_mutex_unlock(&queue->lock);
 594
 595                 session_lock_list();
 596                 session = job->session;
 597                 if (!session) {
 598                         DBG("[rotation-thread] Session \"%s\" not found",
 599                                         session->name);
 600                         /*
 601                          * This is a non-fatal error, and we cannot report it to
 602                          * the user (timer), so just print the error and
 603                          * continue the processing.
 604                          *
 605                          * While the timer thread will purge pending signals for
 606                          * a session on the session's destruction, it is
 607                          * possible for a job targeting that session to have
 608                          * already been queued before it was destroyed.
 609                          */
 610                         free(job);
 611                         session_put(session);
 612                         session_unlock_list();
 613                         continue;
 614                 }
 615
 616                 session_lock(session);
 617                 ret = run_job(job, session, handle->notification_thread_handle);
 618                 session_unlock(session);
 619                 /* Release reference held by the job. */
 620                 session_put(session);
 621                 session_unlock_list();
 622                 free(job);
 623                 if (ret) {
 624                         goto end;
 625                 }
 626         }
 627
 628         ret = 0;
 629
 630 end:
 631         return ret;
 632 }
 633
 634 static
 635 int handle_condition(const struct lttng_condition *condition,
 636                 const struct lttng_evaluation *evaluation,
 637                 struct notification_thread_handle *notification_thread_handle)
 638 {
 639         int ret = 0;
 640         const char *condition_session_name = NULL;
 641         enum lttng_condition_type condition_type;
 642         enum lttng_condition_status condition_status;
 643         enum lttng_evaluation_status evaluation_status;
 644         uint64_t consumed;
 645         struct ltt_session *session;
 646
 647         condition_type = lttng_condition_get_type(condition);
 648
 649         if (condition_type != LTTNG_CONDITION_TYPE_SESSION_CONSUMED_SIZE) {
 650                 ret = -1;
 651                 ERR("[rotation-thread] Condition type and session usage type are not the same");
 652                 goto end;
 653         }
 654
 655         /* Fetch info to test */
 656         condition_status = lttng_condition_session_consumed_size_get_session_name(
 657                         condition, &condition_session_name);
 658         if (condition_status != LTTNG_CONDITION_STATUS_OK) {
 659                 ERR("[rotation-thread] Session name could not be fetched");
 660                 ret = -1;
 661                 goto end;
 662         }
 663         evaluation_status = lttng_evaluation_session_consumed_size_get_consumed_size(evaluation,
 664                         &consumed);
 665         if (evaluation_status != LTTNG_EVALUATION_STATUS_OK) {
 666                 ERR("[rotation-thread] Failed to get evaluation");
 667                 ret = -1;
 668                 goto end;
 669         }
 670
 671         session_lock_list();
 672         session = session_find_by_name(condition_session_name);
 673         if (!session) {
 674                 ret = -1;
 675                 session_unlock_list();
 676                 ERR("[rotation-thread] Session \"%s\" not found",
 677                                 condition_session_name);
 678                 goto end;
 679         }
 680         session_lock(session);
 681
 682         ret = unsubscribe_session_consumed_size_rotation(session,
 683                         notification_thread_handle);
 684         if (ret) {
 685                 goto end_unlock;
 686         }
 687
 688         ret = cmd_rotate_session(session, NULL, false,
 689                 LTTNG_TRACE_CHUNK_COMMAND_TYPE_MOVE_TO_COMPLETED);
 690         if (ret == -LTTNG_ERR_ROTATION_PENDING) {
 691                 DBG("Rotate already pending, subscribe to the next threshold value");
 692         } else if (ret != LTTNG_OK) {
 693                 ERR("[rotation-thread] Failed to rotate on size notification with error: %s",
 694                                 lttng_strerror(ret));
 695                 ret = -1;
 696                 goto end_unlock;
 697         }
 698         ret = subscribe_session_consumed_size_rotation(session,
 699                         consumed + session->rotate_size,
 700                         notification_thread_handle);
 701         if (ret) {
 702                 ERR("[rotation-thread] Failed to subscribe to session consumed size condition");
 703                 goto end_unlock;
 704         }
 705         ret = 0;
 706
 707 end_unlock:
 708         session_unlock(session);
 709         session_put(session);
 710         session_unlock_list();
 711 end:
 712         return ret;
 713 }
 714
 715 static
 716 int handle_notification_channel(int fd,
 717                 struct rotation_thread_handle *handle,
 718                 struct rotation_thread *state)
 719 {
 720         int ret;
 721         bool notification_pending;
 722         struct lttng_notification *notification = NULL;
 723         enum lttng_notification_channel_status status;
 724         const struct lttng_evaluation *notification_evaluation;
 725         const struct lttng_condition *notification_condition;
 726
 727         status = lttng_notification_channel_has_pending_notification(
 728                         rotate_notification_channel, &notification_pending);
 729         if (status != LTTNG_NOTIFICATION_CHANNEL_STATUS_OK) {
 730                 ERR("[rotation-thread ]Error occurred while checking for pending notification");
 731                 ret = -1;
 732                 goto end;
 733         }
 734
 735         if (!notification_pending) {
 736                 ret = 0;
 737                 goto end;
 738         }
 739
 740         /* Receive the next notification. */
 741         status = lttng_notification_channel_get_next_notification(
 742                         rotate_notification_channel,
 743                         &notification);
 744
 745         switch (status) {
 746         case LTTNG_NOTIFICATION_CHANNEL_STATUS_OK:
 747                 break;
 748         case LTTNG_NOTIFICATION_CHANNEL_STATUS_NOTIFICATIONS_DROPPED:
 749                 /* Not an error, we will wait for the next one */
 750                 ret = 0;
 751                 goto end;;
 752         case LTTNG_NOTIFICATION_CHANNEL_STATUS_CLOSED:
 753                 ERR("Notification channel was closed");
 754                 ret = -1;
 755                 goto end;
 756         default:
 757                 /* Unhandled conditions / errors. */
 758                 ERR("Unknown notification channel status");
 759                 ret = -1;
 760                 goto end;
 761         }
 762
 763         notification_condition = lttng_notification_get_condition(notification);
 764         notification_evaluation = lttng_notification_get_evaluation(notification);
 765
 766         ret = handle_condition(notification_condition, notification_evaluation,
 767                         handle->notification_thread_handle);
 768
 769 end:
 770         lttng_notification_destroy(notification);
 771         return ret;
 772 }
 773
 774 static
 775 void *thread_rotation(void *data)
 776 {
 777         int ret;
 778         struct rotation_thread_handle *handle = data;
 779         struct rotation_thread thread;
 780         int queue_pipe_fd;
 781
 782         DBG("[rotation-thread] Started rotation thread");
 783         rcu_register_thread();
 784         rcu_thread_online();
 785         health_register(health_sessiond, HEALTH_SESSIOND_TYPE_ROTATION);
 786         health_code_update();
 787
 788         if (!handle) {
 789                 ERR("[rotation-thread] Invalid thread context provided");
 790                 goto end;
 791         }
 792
 793         queue_pipe_fd = lttng_pipe_get_readfd(
 794                         handle->rotation_timer_queue->event_pipe);
 795
 796
 797         ret = init_thread_state(handle, &thread);
 798         if (ret) {
 799                 goto error;
 800         }
 801
 802         while (true) {
 803                 int fd_count, i;
 804
 805                 health_poll_entry();
 806                 DBG("[rotation-thread] Entering poll wait");
 807                 ret = lttng_poll_wait(&thread.events, -1);
 808                 DBG("[rotation-thread] Poll wait returned (%i)", ret);
 809                 health_poll_exit();
 810                 if (ret < 0) {
 811                         /*
 812                          * Restart interrupted system call.
 813                          */
 814                         if (errno == EINTR) {
 815                                 continue;
 816                         }
 817                         ERR("[rotation-thread] Error encountered during lttng_poll_wait (%i)", ret);
 818                         goto error;
 819                 }
 820
 821                 fd_count = ret;
 822                 for (i = 0; i < fd_count; i++) {
 823                         int fd = LTTNG_POLL_GETFD(&thread.events, i);
 824                         uint32_t revents = LTTNG_POLL_GETEV(&thread.events, i);
 825
 826                         DBG("[rotation-thread] Handling fd (%i) activity (%u)",
 827                                         fd, revents);
 828
 829                         if (revents & LPOLLERR) {
 830                                 ERR("[rotation-thread] Polling returned an error on fd %i", fd);
 831                                 goto error;
 832                         }
 833
 834                         if (fd == rotate_notification_channel->socket) {
 835                                 ret = handle_notification_channel(fd, handle,
 836                                                 &thread);
 837                                 if (ret) {
 838                                         ERR("[rotation-thread] Error occurred while handling activity on notification channel socket");
 839                                         goto error;
 840                                 }
 841                         } else {
 842                                 /* Job queue or quit pipe activity. */
 843
 844                                 /*
 845                                  * The job queue is serviced if there is
 846                                  * activity on the quit pipe to ensure it is
 847                                  * flushed and all references held in the queue
 848                                  * are released.
 849                                  */
 850                                 ret = handle_job_queue(handle, &thread,
 851                                                 handle->rotation_timer_queue);
 852                                 if (ret) {
 853                                         ERR("[rotation-thread] Failed to handle rotation timer pipe event");
 854                                         goto error;
 855                                 }
 856
 857                                 if (fd == queue_pipe_fd) {
 858                                         char buf;
 859
 860                                         ret = lttng_read(fd, &buf, 1);
 861                                         if (ret != 1) {
 862                                                 ERR("[rotation-thread] Failed to read from wakeup pipe (fd = %i)", fd);
 863                                                 goto error;
 864                                         }
 865                                 } else {
 866                                         DBG("[rotation-thread] Quit pipe activity");
 867                                         goto exit;
 868                                 }
 869                         }
 870                 }
 871         }
 872 exit:
 873 error:
 874         DBG("[rotation-thread] Exit");
 875         fini_thread_state(&thread);
 876 end:
 877         health_unregister(health_sessiond);
 878         rcu_thread_offline();
 879         rcu_unregister_thread();
 880         return NULL;
 881 }
 882
 883 static
 884 bool shutdown_rotation_thread(void *thread_data)
 885 {
 886         struct rotation_thread_handle *handle = thread_data;
 887         const int write_fd = lttng_pipe_get_writefd(handle->quit_pipe);
 888
 889         return notify_thread_pipe(write_fd) == 1;
 890 }
 891
 892 bool launch_rotation_thread(struct rotation_thread_handle *handle)
 893 {
 894         struct lttng_thread *thread;
 895
 896         thread = lttng_thread_create("Rotation",
 897                         thread_rotation,
 898                         shutdown_rotation_thread,
 899                         NULL,
 900                         handle);
 901         if (!thread) {
 902                 goto error;
 903         }
 904         lttng_thread_put(thread);
 905         return true;
 906 error:
 907         return false;
 908 }