src/common/kernel-consumer/kernel-consumer.c

   1 /*
   2  * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
   3  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  * Copyright (C) 2017 Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  *
   6  * SPDX-License-Identifier: GPL-2.0-only
   7  *
   8  */
   9
  10 #define _LGPL_SOURCE
  11 #include <assert.h>
  12 #include <poll.h>
  13 #include <pthread.h>
  14 #include <stdlib.h>
  15 #include <string.h>
  16 #include <sys/mman.h>
  17 #include <sys/socket.h>
  18 #include <sys/types.h>
  19 #include <inttypes.h>
  20 #include <unistd.h>
  21 #include <sys/stat.h>
  22
  23 #include <bin/lttng-consumerd/health-consumerd.h>
  24 #include <common/common.h>
  25 #include <common/kernel-ctl/kernel-ctl.h>
  26 #include <common/sessiond-comm/sessiond-comm.h>
  27 #include <common/sessiond-comm/relayd.h>
  28 #include <common/compat/fcntl.h>
  29 #include <common/compat/endian.h>
  30 #include <common/pipe.h>
  31 #include <common/relayd/relayd.h>
  32 #include <common/utils.h>
  33 #include <common/consumer/consumer-stream.h>
  34 #include <common/index/index.h>
  35 #include <common/consumer/consumer-timer.h>
  36 #include <common/optional.h>
  37 #include <common/buffer-view.h>
  38 #include <common/consumer/consumer.h>
  39 #include <stdint.h>
  40
  41 #include "kernel-consumer.h"
  42
  43 extern struct lttng_consumer_global_data consumer_data;
  44 extern int consumer_poll_timeout;
  45
  46 /*
  47  * Take a snapshot for a specific fd
  48  *
  49  * Returns 0 on success, < 0 on error
  50  */
  51 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  52 {
  53         int ret = 0;
  54         int infd = stream->wait_fd;
  55
  56         ret = kernctl_snapshot(infd);
  57         /*
  58          * -EAGAIN is not an error, it just means that there is no data to
  59          *  be read.
  60          */
  61         if (ret != 0 && ret != -EAGAIN) {
  62                 PERROR("Getting sub-buffer snapshot.");
  63         }
  64
  65         return ret;
  66 }
  67
  68 /*
  69  * Sample consumed and produced positions for a specific fd.
  70  *
  71  * Returns 0 on success, < 0 on error.
  72  */
  73 int lttng_kconsumer_sample_snapshot_positions(
  74                 struct lttng_consumer_stream *stream)
  75 {
  76         assert(stream);
  77
  78         return kernctl_snapshot_sample_positions(stream->wait_fd);
  79 }
  80
  81 /*
  82  * Get the produced position
  83  *
  84  * Returns 0 on success, < 0 on error
  85  */
  86 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  87                 unsigned long *pos)
  88 {
  89         int ret;
  90         int infd = stream->wait_fd;
  91
  92         ret = kernctl_snapshot_get_produced(infd, pos);
  93         if (ret != 0) {
  94                 PERROR("kernctl_snapshot_get_produced");
  95         }
  96
  97         return ret;
  98 }
  99
 100 /*
 101  * Get the consumerd position
 102  *
 103  * Returns 0 on success, < 0 on error
 104  */
 105 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 106                 unsigned long *pos)
 107 {
 108         int ret;
 109         int infd = stream->wait_fd;
 110
 111         ret = kernctl_snapshot_get_consumed(infd, pos);
 112         if (ret != 0) {
 113                 PERROR("kernctl_snapshot_get_consumed");
 114         }
 115
 116         return ret;
 117 }
 118
 119 static
 120 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
 121                 const char **addr)
 122 {
 123         int ret;
 124         unsigned long mmap_offset;
 125         const char *mmap_base = stream->mmap_base;
 126
 127         ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
 128         if (ret < 0) {
 129                 PERROR("Failed to get mmap read offset");
 130                 goto error;
 131         }
 132
 133         *addr = mmap_base + mmap_offset;
 134 error:
 135         return ret;
 136 }
 137
 138 /*
 139  * Take a snapshot of all the stream of a channel
 140  * RCU read-side lock must be held across this function to ensure existence of
 141  * channel. The channel lock must be held by the caller.
 142  *
 143  * Returns 0 on success, < 0 on error
 144  */
 145 static int lttng_kconsumer_snapshot_channel(
 146                 struct lttng_consumer_channel *channel,
 147                 uint64_t key, char *path, uint64_t relayd_id,
 148                 uint64_t nb_packets_per_stream,
 149                 struct lttng_consumer_local_data *ctx)
 150 {
 151         int ret;
 152         struct lttng_consumer_stream *stream;
 153
 154         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 155
 156         rcu_read_lock();
 157
 158         /* Splice is not supported yet for channel snapshot. */
 159         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 160                 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
 161                                 channel->name);
 162                 ret = -1;
 163                 goto end;
 164         }
 165
 166         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 167                 unsigned long consumed_pos, produced_pos;
 168
 169                 health_code_update();
 170
 171                 /*
 172                  * Lock stream because we are about to change its state.
 173                  */
 174                 pthread_mutex_lock(&stream->lock);
 175
 176                 assert(channel->trace_chunk);
 177                 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
 178                         /*
 179                          * Can't happen barring an internal error as the channel
 180                          * holds a reference to the trace chunk.
 181                          */
 182                         ERR("Failed to acquire reference to channel's trace chunk");
 183                         ret = -1;
 184                         goto end_unlock;
 185                 }
 186                 assert(!stream->trace_chunk);
 187                 stream->trace_chunk = channel->trace_chunk;
 188
 189                 /*
 190                  * Assign the received relayd ID so we can use it for streaming. The streams
 191                  * are not visible to anyone so this is OK to change it.
 192                  */
 193                 stream->net_seq_idx = relayd_id;
 194                 channel->relayd_id = relayd_id;
 195                 if (relayd_id != (uint64_t) -1ULL) {
 196                         ret = consumer_send_relayd_stream(stream, path);
 197                         if (ret < 0) {
 198                                 ERR("sending stream to relayd");
 199                                 goto end_unlock;
 200                         }
 201                 } else {
 202                         ret = consumer_stream_create_output_files(stream,
 203                                         false);
 204                         if (ret < 0) {
 205                                 goto end_unlock;
 206                         }
 207                         DBG("Kernel consumer snapshot stream (%" PRIu64 ")",
 208                                         stream->key);
 209                 }
 210
 211                 ret = kernctl_buffer_flush_empty(stream->wait_fd);
 212                 if (ret < 0) {
 213                         /*
 214                          * Doing a buffer flush which does not take into
 215                          * account empty packets. This is not perfect
 216                          * for stream intersection, but required as a
 217                          * fall-back when "flush_empty" is not
 218                          * implemented by lttng-modules.
 219                          */
 220                         ret = kernctl_buffer_flush(stream->wait_fd);
 221                         if (ret < 0) {
 222                                 ERR("Failed to flush kernel stream");
 223                                 goto end_unlock;
 224                         }
 225                         goto end_unlock;
 226                 }
 227
 228                 ret = lttng_kconsumer_take_snapshot(stream);
 229                 if (ret < 0) {
 230                         ERR("Taking kernel snapshot");
 231                         goto end_unlock;
 232                 }
 233
 234                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 235                 if (ret < 0) {
 236                         ERR("Produced kernel snapshot position");
 237                         goto end_unlock;
 238                 }
 239
 240                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 241                 if (ret < 0) {
 242                         ERR("Consumerd kernel snapshot position");
 243                         goto end_unlock;
 244                 }
 245
 246                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 247                                 produced_pos, nb_packets_per_stream,
 248                                 stream->max_sb_size);
 249
 250                 while ((long) (consumed_pos - produced_pos) < 0) {
 251                         ssize_t read_len;
 252                         unsigned long len, padded_len;
 253                         const char *subbuf_addr;
 254                         struct lttng_buffer_view subbuf_view;
 255
 256                         health_code_update();
 257                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 258
 259                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 260                         if (ret < 0) {
 261                                 if (ret != -EAGAIN) {
 262                                         PERROR("kernctl_get_subbuf snapshot");
 263                                         goto end_unlock;
 264                                 }
 265                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 266                                 consumed_pos += stream->max_sb_size;
 267                                 stream->chan->lost_packets++;
 268                                 continue;
 269                         }
 270
 271                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 272                         if (ret < 0) {
 273                                 ERR("Snapshot kernctl_get_subbuf_size");
 274                                 goto error_put_subbuf;
 275                         }
 276
 277                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 278                         if (ret < 0) {
 279                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 280                                 goto error_put_subbuf;
 281                         }
 282
 283                         ret = get_current_subbuf_addr(stream, &subbuf_addr);
 284                         if (ret) {
 285                                 goto error_put_subbuf;
 286                         }
 287
 288                         subbuf_view = lttng_buffer_view_init(
 289                                         subbuf_addr, 0, padded_len);
 290                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx,
 291                                         stream, &subbuf_view,
 292                                         padded_len - len);
 293                         /*
 294                          * We write the padded len in local tracefiles but the data len
 295                          * when using a relay. Display the error but continue processing
 296                          * to try to release the subbuffer.
 297                          */
 298                         if (relayd_id != (uint64_t) -1ULL) {
 299                                 if (read_len != len) {
 300                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 301                                                         read_len, len);
 302                                 }
 303                         } else {
 304                                 if (read_len != padded_len) {
 305                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 306                                                         read_len, padded_len);
 307                                 }
 308                         }
 309
 310                         ret = kernctl_put_subbuf(stream->wait_fd);
 311                         if (ret < 0) {
 312                                 ERR("Snapshot kernctl_put_subbuf");
 313                                 goto end_unlock;
 314                         }
 315                         consumed_pos += stream->max_sb_size;
 316                 }
 317
 318                 if (relayd_id == (uint64_t) -1ULL) {
 319                         if (stream->out_fd >= 0) {
 320                                 ret = close(stream->out_fd);
 321                                 if (ret < 0) {
 322                                         PERROR("Kernel consumer snapshot close out_fd");
 323                                         goto end_unlock;
 324                                 }
 325                                 stream->out_fd = -1;
 326                         }
 327                 } else {
 328                         close_relayd_stream(stream);
 329                         stream->net_seq_idx = (uint64_t) -1ULL;
 330                 }
 331                 lttng_trace_chunk_put(stream->trace_chunk);
 332                 stream->trace_chunk = NULL;
 333                 pthread_mutex_unlock(&stream->lock);
 334         }
 335
 336         /* All good! */
 337         ret = 0;
 338         goto end;
 339
 340 error_put_subbuf:
 341         ret = kernctl_put_subbuf(stream->wait_fd);
 342         if (ret < 0) {
 343                 ERR("Snapshot kernctl_put_subbuf error path");
 344         }
 345 end_unlock:
 346         pthread_mutex_unlock(&stream->lock);
 347 end:
 348         rcu_read_unlock();
 349         return ret;
 350 }
 351
 352 /*
 353  * Read the whole metadata available for a snapshot.
 354  * RCU read-side lock must be held across this function to ensure existence of
 355  * metadata_channel. The channel lock must be held by the caller.
 356  *
 357  * Returns 0 on success, < 0 on error
 358  */
 359 static int lttng_kconsumer_snapshot_metadata(
 360                 struct lttng_consumer_channel *metadata_channel,
 361                 uint64_t key, char *path, uint64_t relayd_id,
 362                 struct lttng_consumer_local_data *ctx)
 363 {
 364         int ret, use_relayd = 0;
 365         ssize_t ret_read;
 366         struct lttng_consumer_stream *metadata_stream;
 367
 368         assert(ctx);
 369
 370         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 371                         key, path);
 372
 373         rcu_read_lock();
 374
 375         metadata_stream = metadata_channel->metadata_stream;
 376         assert(metadata_stream);
 377
 378         pthread_mutex_lock(&metadata_stream->lock);
 379         assert(metadata_channel->trace_chunk);
 380         assert(metadata_stream->trace_chunk);
 381
 382         /* Flag once that we have a valid relayd for the stream. */
 383         if (relayd_id != (uint64_t) -1ULL) {
 384                 use_relayd = 1;
 385         }
 386
 387         if (use_relayd) {
 388                 ret = consumer_send_relayd_stream(metadata_stream, path);
 389                 if (ret < 0) {
 390                         goto error_snapshot;
 391                 }
 392         } else {
 393                 ret = consumer_stream_create_output_files(metadata_stream,
 394                                 false);
 395                 if (ret < 0) {
 396                         goto error_snapshot;
 397                 }
 398         }
 399
 400         do {
 401                 health_code_update();
 402
 403                 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
 404                 if (ret_read < 0) {
 405                         if (ret_read != -EAGAIN) {
 406                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 407                                                 ret_read);
 408                                 ret = ret_read;
 409                                 goto error_snapshot;
 410                         }
 411                         /* ret_read is negative at this point so we will exit the loop. */
 412                         continue;
 413                 }
 414         } while (ret_read >= 0);
 415
 416         if (use_relayd) {
 417                 close_relayd_stream(metadata_stream);
 418                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 419         } else {
 420                 if (metadata_stream->out_fd >= 0) {
 421                         ret = close(metadata_stream->out_fd);
 422                         if (ret < 0) {
 423                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 424                                 /*
 425                                  * Don't go on error here since the snapshot was successful at this
 426                                  * point but somehow the close failed.
 427                                  */
 428                         }
 429                         metadata_stream->out_fd = -1;
 430                         lttng_trace_chunk_put(metadata_stream->trace_chunk);
 431                         metadata_stream->trace_chunk = NULL;
 432                 }
 433         }
 434
 435         ret = 0;
 436 error_snapshot:
 437         pthread_mutex_unlock(&metadata_stream->lock);
 438         cds_list_del(&metadata_stream->send_node);
 439         consumer_stream_destroy(metadata_stream, NULL);
 440         metadata_channel->metadata_stream = NULL;
 441         rcu_read_unlock();
 442         return ret;
 443 }
 444
 445 /*
 446  * Receive command from session daemon and process it.
 447  *
 448  * Return 1 on success else a negative value or 0.
 449  */
 450 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 451                 int sock, struct pollfd *consumer_sockpoll)
 452 {
 453         ssize_t ret;
 454         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 455         struct lttcomm_consumer_msg msg;
 456
 457         health_code_update();
 458
 459         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 460         if (ret != sizeof(msg)) {
 461                 if (ret > 0) {
 462                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 463                         ret = -1;
 464                 }
 465                 return ret;
 466         }
 467
 468         health_code_update();
 469
 470         /* Deprecated command */
 471         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 472
 473         health_code_update();
 474
 475         /* relayd needs RCU read-side protection */
 476         rcu_read_lock();
 477
 478         switch (msg.cmd_type) {
 479         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 480         {
 481                 /* Session daemon status message are handled in the following call. */
 482                 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 483                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 484                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 485                                 msg.u.relayd_sock.relayd_session_id);
 486                 goto end_nosignal;
 487         }
 488         case LTTNG_CONSUMER_ADD_CHANNEL:
 489         {
 490                 struct lttng_consumer_channel *new_channel;
 491                 int ret_recv;
 492                 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
 493
 494                 health_code_update();
 495
 496                 /* First send a status message before receiving the fds. */
 497                 ret = consumer_send_status_msg(sock, ret_code);
 498                 if (ret < 0) {
 499                         /* Somehow, the session daemon is not responding anymore. */
 500                         goto error_fatal;
 501                 }
 502
 503                 health_code_update();
 504
 505                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 506                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 507                                 msg.u.channel.session_id,
 508                                 msg.u.channel.chunk_id.is_set ?
 509                                                 &chunk_id : NULL,
 510                                 msg.u.channel.pathname,
 511                                 msg.u.channel.name,
 512                                 msg.u.channel.relayd_id, msg.u.channel.output,
 513                                 msg.u.channel.tracefile_size,
 514                                 msg.u.channel.tracefile_count, 0,
 515                                 msg.u.channel.monitor,
 516                                 msg.u.channel.live_timer_interval,
 517                                 msg.u.channel.is_live,
 518                                 NULL, NULL);
 519                 if (new_channel == NULL) {
 520                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 521                         goto end_nosignal;
 522                 }
 523                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 524                 switch (msg.u.channel.output) {
 525                 case LTTNG_EVENT_SPLICE:
 526                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 527                         break;
 528                 case LTTNG_EVENT_MMAP:
 529                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 530                         break;
 531                 default:
 532                         ERR("Channel output unknown %d", msg.u.channel.output);
 533                         goto end_nosignal;
 534                 }
 535
 536                 /* Translate and save channel type. */
 537                 switch (msg.u.channel.type) {
 538                 case CONSUMER_CHANNEL_TYPE_DATA:
 539                 case CONSUMER_CHANNEL_TYPE_METADATA:
 540                         new_channel->type = msg.u.channel.type;
 541                         break;
 542                 default:
 543                         assert(0);
 544                         goto end_nosignal;
 545                 };
 546
 547                 health_code_update();
 548
 549                 if (ctx->on_recv_channel != NULL) {
 550                         ret_recv = ctx->on_recv_channel(new_channel);
 551                         if (ret_recv == 0) {
 552                                 ret = consumer_add_channel(new_channel, ctx);
 553                         } else if (ret_recv < 0) {
 554                                 goto end_nosignal;
 555                         }
 556                 } else {
 557                         ret = consumer_add_channel(new_channel, ctx);
 558                 }
 559                 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
 560                         int monitor_start_ret;
 561
 562                         DBG("Consumer starting monitor timer");
 563                         consumer_timer_live_start(new_channel,
 564                                         msg.u.channel.live_timer_interval);
 565                         monitor_start_ret = consumer_timer_monitor_start(
 566                                         new_channel,
 567                                         msg.u.channel.monitor_timer_interval);
 568                         if (monitor_start_ret < 0) {
 569                                 ERR("Starting channel monitoring timer failed");
 570                                 goto end_nosignal;
 571                         }
 572
 573                 }
 574
 575                 health_code_update();
 576
 577                 /* If we received an error in add_channel, we need to report it. */
 578                 if (ret < 0) {
 579                         ret = consumer_send_status_msg(sock, ret);
 580                         if (ret < 0) {
 581                                 goto error_fatal;
 582                         }
 583                         goto end_nosignal;
 584                 }
 585
 586                 goto end_nosignal;
 587         }
 588         case LTTNG_CONSUMER_ADD_STREAM:
 589         {
 590                 int fd;
 591                 struct lttng_pipe *stream_pipe;
 592                 struct lttng_consumer_stream *new_stream;
 593                 struct lttng_consumer_channel *channel;
 594                 int alloc_ret = 0;
 595
 596                 /*
 597                  * Get stream's channel reference. Needed when adding the stream to the
 598                  * global hash table.
 599                  */
 600                 channel = consumer_find_channel(msg.u.stream.channel_key);
 601                 if (!channel) {
 602                         /*
 603                          * We could not find the channel. Can happen if cpu hotplug
 604                          * happens while tearing down.
 605                          */
 606                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 607                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 608                 }
 609
 610                 health_code_update();
 611
 612                 /* First send a status message before receiving the fds. */
 613                 ret = consumer_send_status_msg(sock, ret_code);
 614                 if (ret < 0) {
 615                         /* Somehow, the session daemon is not responding anymore. */
 616                         goto error_add_stream_fatal;
 617                 }
 618
 619                 health_code_update();
 620
 621                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 622                         /* Channel was not found. */
 623                         goto error_add_stream_nosignal;
 624                 }
 625
 626                 /* Blocking call */
 627                 health_poll_entry();
 628                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 629                 health_poll_exit();
 630                 if (ret) {
 631                         goto error_add_stream_fatal;
 632                 }
 633
 634                 health_code_update();
 635
 636                 /* Get stream file descriptor from socket */
 637                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 638                 if (ret != sizeof(fd)) {
 639                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 640                         goto end;
 641                 }
 642
 643                 health_code_update();
 644
 645                 /*
 646                  * Send status code to session daemon only if the recv works. If the
 647                  * above recv() failed, the session daemon is notified through the
 648                  * error socket and the teardown is eventually done.
 649                  */
 650                 ret = consumer_send_status_msg(sock, ret_code);
 651                 if (ret < 0) {
 652                         /* Somehow, the session daemon is not responding anymore. */
 653                         goto error_add_stream_nosignal;
 654                 }
 655
 656                 health_code_update();
 657
 658                 pthread_mutex_lock(&channel->lock);
 659                 new_stream = consumer_stream_create(
 660                                 channel,
 661                                 channel->key,
 662                                 fd,
 663                                 channel->name,
 664                                 channel->relayd_id,
 665                                 channel->session_id,
 666                                 channel->trace_chunk,
 667                                 msg.u.stream.cpu,
 668                                 &alloc_ret,
 669                                 channel->type,
 670                                 channel->monitor);
 671                 if (new_stream == NULL) {
 672                         switch (alloc_ret) {
 673                         case -ENOMEM:
 674                         case -EINVAL:
 675                         default:
 676                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 677                                 break;
 678                         }
 679                         pthread_mutex_unlock(&channel->lock);
 680                         goto error_add_stream_nosignal;
 681                 }
 682
 683                 new_stream->wait_fd = fd;
 684                 ret = kernctl_get_max_subbuf_size(new_stream->wait_fd,
 685                                 &new_stream->max_sb_size);
 686                 if (ret < 0) {
 687                         pthread_mutex_unlock(&channel->lock);
 688                         ERR("Failed to get kernel maximal subbuffer size");
 689                         goto error_add_stream_nosignal;
 690                 }
 691
 692                 consumer_stream_update_channel_attributes(new_stream,
 693                                 channel);
 694
 695                 /*
 696                  * We've just assigned the channel to the stream so increment the
 697                  * refcount right now. We don't need to increment the refcount for
 698                  * streams in no monitor because we handle manually the cleanup of
 699                  * those. It is very important to make sure there is NO prior
 700                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 701                  */
 702                 if (channel->monitor) {
 703                         uatomic_inc(&new_stream->chan->refcount);
 704                 }
 705
 706                 /*
 707                  * The buffer flush is done on the session daemon side for the kernel
 708                  * so no need for the stream "hangup_flush_done" variable to be
 709                  * tracked. This is important for a kernel stream since we don't rely
 710                  * on the flush state of the stream to read data. It's not the case for
 711                  * user space tracing.
 712                  */
 713                 new_stream->hangup_flush_done = 0;
 714
 715                 health_code_update();
 716
 717                 pthread_mutex_lock(&new_stream->lock);
 718                 if (ctx->on_recv_stream) {
 719                         ret = ctx->on_recv_stream(new_stream);
 720                         if (ret < 0) {
 721                                 pthread_mutex_unlock(&new_stream->lock);
 722                                 pthread_mutex_unlock(&channel->lock);
 723                                 consumer_stream_free(new_stream);
 724                                 goto error_add_stream_nosignal;
 725                         }
 726                 }
 727                 health_code_update();
 728
 729                 if (new_stream->metadata_flag) {
 730                         channel->metadata_stream = new_stream;
 731                 }
 732
 733                 /* Do not monitor this stream. */
 734                 if (!channel->monitor) {
 735                         DBG("Kernel consumer add stream %s in no monitor mode with "
 736                                         "relayd id %" PRIu64, new_stream->name,
 737                                         new_stream->net_seq_idx);
 738                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 739                         pthread_mutex_unlock(&new_stream->lock);
 740                         pthread_mutex_unlock(&channel->lock);
 741                         goto end_add_stream;
 742                 }
 743
 744                 /* Send stream to relayd if the stream has an ID. */
 745                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 746                         ret = consumer_send_relayd_stream(new_stream,
 747                                         new_stream->chan->pathname);
 748                         if (ret < 0) {
 749                                 pthread_mutex_unlock(&new_stream->lock);
 750                                 pthread_mutex_unlock(&channel->lock);
 751                                 consumer_stream_free(new_stream);
 752                                 goto error_add_stream_nosignal;
 753                         }
 754
 755                         /*
 756                          * If adding an extra stream to an already
 757                          * existing channel (e.g. cpu hotplug), we need
 758                          * to send the "streams_sent" command to relayd.
 759                          */
 760                         if (channel->streams_sent_to_relayd) {
 761                                 ret = consumer_send_relayd_streams_sent(
 762                                                 new_stream->net_seq_idx);
 763                                 if (ret < 0) {
 764                                         pthread_mutex_unlock(&new_stream->lock);
 765                                         pthread_mutex_unlock(&channel->lock);
 766                                         goto error_add_stream_nosignal;
 767                                 }
 768                         }
 769                 }
 770                 pthread_mutex_unlock(&new_stream->lock);
 771                 pthread_mutex_unlock(&channel->lock);
 772
 773                 /* Get the right pipe where the stream will be sent. */
 774                 if (new_stream->metadata_flag) {
 775                         consumer_add_metadata_stream(new_stream);
 776                         stream_pipe = ctx->consumer_metadata_pipe;
 777                 } else {
 778                         consumer_add_data_stream(new_stream);
 779                         stream_pipe = ctx->consumer_data_pipe;
 780                 }
 781
 782                 /* Visible to other threads */
 783                 new_stream->globally_visible = 1;
 784
 785                 health_code_update();
 786
 787                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 788                 if (ret < 0) {
 789                         ERR("Consumer write %s stream to pipe %d",
 790                                         new_stream->metadata_flag ? "metadata" : "data",
 791                                         lttng_pipe_get_writefd(stream_pipe));
 792                         if (new_stream->metadata_flag) {
 793                                 consumer_del_stream_for_metadata(new_stream);
 794                         } else {
 795                                 consumer_del_stream_for_data(new_stream);
 796                         }
 797                         goto error_add_stream_nosignal;
 798                 }
 799
 800                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
 801                                 new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
 802 end_add_stream:
 803                 break;
 804 error_add_stream_nosignal:
 805                 goto end_nosignal;
 806 error_add_stream_fatal:
 807                 goto error_fatal;
 808         }
 809         case LTTNG_CONSUMER_STREAMS_SENT:
 810         {
 811                 struct lttng_consumer_channel *channel;
 812
 813                 /*
 814                  * Get stream's channel reference. Needed when adding the stream to the
 815                  * global hash table.
 816                  */
 817                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 818                 if (!channel) {
 819                         /*
 820                          * We could not find the channel. Can happen if cpu hotplug
 821                          * happens while tearing down.
 822                          */
 823                         ERR("Unable to find channel key %" PRIu64,
 824                                         msg.u.sent_streams.channel_key);
 825                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 826                 }
 827
 828                 health_code_update();
 829
 830                 /*
 831                  * Send status code to session daemon.
 832                  */
 833                 ret = consumer_send_status_msg(sock, ret_code);
 834                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 835                         /* Somehow, the session daemon is not responding anymore. */
 836                         goto error_streams_sent_nosignal;
 837                 }
 838
 839                 health_code_update();
 840
 841                 /*
 842                  * We should not send this message if we don't monitor the
 843                  * streams in this channel.
 844                  */
 845                 if (!channel->monitor) {
 846                         goto end_error_streams_sent;
 847                 }
 848
 849                 health_code_update();
 850                 /* Send stream to relayd if the stream has an ID. */
 851                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 852                         ret = consumer_send_relayd_streams_sent(
 853                                         msg.u.sent_streams.net_seq_idx);
 854                         if (ret < 0) {
 855                                 goto error_streams_sent_nosignal;
 856                         }
 857                         channel->streams_sent_to_relayd = true;
 858                 }
 859 end_error_streams_sent:
 860                 break;
 861 error_streams_sent_nosignal:
 862                 goto end_nosignal;
 863         }
 864         case LTTNG_CONSUMER_UPDATE_STREAM:
 865         {
 866                 rcu_read_unlock();
 867                 return -ENOSYS;
 868         }
 869         case LTTNG_CONSUMER_DESTROY_RELAYD:
 870         {
 871                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 872                 struct consumer_relayd_sock_pair *relayd;
 873
 874                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 875
 876                 /* Get relayd reference if exists. */
 877                 relayd = consumer_find_relayd(index);
 878                 if (relayd == NULL) {
 879                         DBG("Unable to find relayd %" PRIu64, index);
 880                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 881                 }
 882
 883                 /*
 884                  * Each relayd socket pair has a refcount of stream attached to it
 885                  * which tells if the relayd is still active or not depending on the
 886                  * refcount value.
 887                  *
 888                  * This will set the destroy flag of the relayd object and destroy it
 889                  * if the refcount reaches zero when called.
 890                  *
 891                  * The destroy can happen either here or when a stream fd hangs up.
 892                  */
 893                 if (relayd) {
 894                         consumer_flag_relayd_for_destroy(relayd);
 895                 }
 896
 897                 health_code_update();
 898
 899                 ret = consumer_send_status_msg(sock, ret_code);
 900                 if (ret < 0) {
 901                         /* Somehow, the session daemon is not responding anymore. */
 902                         goto error_fatal;
 903                 }
 904
 905                 goto end_nosignal;
 906         }
 907         case LTTNG_CONSUMER_DATA_PENDING:
 908         {
 909                 int32_t ret;
 910                 uint64_t id = msg.u.data_pending.session_id;
 911
 912                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 913
 914                 ret = consumer_data_pending(id);
 915
 916                 health_code_update();
 917
 918                 /* Send back returned value to session daemon */
 919                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 920                 if (ret < 0) {
 921                         PERROR("send data pending ret code");
 922                         goto error_fatal;
 923                 }
 924
 925                 /*
 926                  * No need to send back a status message since the data pending
 927                  * returned value is the response.
 928                  */
 929                 break;
 930         }
 931         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 932         {
 933                 struct lttng_consumer_channel *channel;
 934                 uint64_t key = msg.u.snapshot_channel.key;
 935
 936                 channel = consumer_find_channel(key);
 937                 if (!channel) {
 938                         ERR("Channel %" PRIu64 " not found", key);
 939                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 940                 } else {
 941                         pthread_mutex_lock(&channel->lock);
 942                         if (msg.u.snapshot_channel.metadata == 1) {
 943                                 ret = lttng_kconsumer_snapshot_metadata(channel, key,
 944                                                 msg.u.snapshot_channel.pathname,
 945                                                 msg.u.snapshot_channel.relayd_id, ctx);
 946                                 if (ret < 0) {
 947                                         ERR("Snapshot metadata failed");
 948                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
 949                                 }
 950                         } else {
 951                                 ret = lttng_kconsumer_snapshot_channel(channel, key,
 952                                                 msg.u.snapshot_channel.pathname,
 953                                                 msg.u.snapshot_channel.relayd_id,
 954                                                 msg.u.snapshot_channel.nb_packets_per_stream,
 955                                                 ctx);
 956                                 if (ret < 0) {
 957                                         ERR("Snapshot channel failed");
 958                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
 959                                 }
 960                         }
 961                         pthread_mutex_unlock(&channel->lock);
 962                 }
 963                 health_code_update();
 964
 965                 ret = consumer_send_status_msg(sock, ret_code);
 966                 if (ret < 0) {
 967                         /* Somehow, the session daemon is not responding anymore. */
 968                         goto end_nosignal;
 969                 }
 970                 break;
 971         }
 972         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 973         {
 974                 uint64_t key = msg.u.destroy_channel.key;
 975                 struct lttng_consumer_channel *channel;
 976
 977                 channel = consumer_find_channel(key);
 978                 if (!channel) {
 979                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 980                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 981                 }
 982
 983                 health_code_update();
 984
 985                 ret = consumer_send_status_msg(sock, ret_code);
 986                 if (ret < 0) {
 987                         /* Somehow, the session daemon is not responding anymore. */
 988                         goto end_destroy_channel;
 989                 }
 990
 991                 health_code_update();
 992
 993                 /* Stop right now if no channel was found. */
 994                 if (!channel) {
 995                         goto end_destroy_channel;
 996                 }
 997
 998                 /*
 999                  * This command should ONLY be issued for channel with streams set in
1000                  * no monitor mode.
1001                  */
1002                 assert(!channel->monitor);
1003
1004                 /*
1005                  * The refcount should ALWAYS be 0 in the case of a channel in no
1006                  * monitor mode.
1007                  */
1008                 assert(!uatomic_sub_return(&channel->refcount, 1));
1009
1010                 consumer_del_channel(channel);
1011 end_destroy_channel:
1012                 goto end_nosignal;
1013         }
1014         case LTTNG_CONSUMER_DISCARDED_EVENTS:
1015         {
1016                 ssize_t ret;
1017                 uint64_t count;
1018                 struct lttng_consumer_channel *channel;
1019                 uint64_t id = msg.u.discarded_events.session_id;
1020                 uint64_t key = msg.u.discarded_events.channel_key;
1021
1022                 DBG("Kernel consumer discarded events command for session id %"
1023                                 PRIu64 ", channel key %" PRIu64, id, key);
1024
1025                 channel = consumer_find_channel(key);
1026                 if (!channel) {
1027                         ERR("Kernel consumer discarded events channel %"
1028                                         PRIu64 " not found", key);
1029                         count = 0;
1030                 } else {
1031                         count = channel->discarded_events;
1032                 }
1033
1034                 health_code_update();
1035
1036                 /* Send back returned value to session daemon */
1037                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1038                 if (ret < 0) {
1039                         PERROR("send discarded events");
1040                         goto error_fatal;
1041                 }
1042
1043                 break;
1044         }
1045         case LTTNG_CONSUMER_LOST_PACKETS:
1046         {
1047                 ssize_t ret;
1048                 uint64_t count;
1049                 struct lttng_consumer_channel *channel;
1050                 uint64_t id = msg.u.lost_packets.session_id;
1051                 uint64_t key = msg.u.lost_packets.channel_key;
1052
1053                 DBG("Kernel consumer lost packets command for session id %"
1054                                 PRIu64 ", channel key %" PRIu64, id, key);
1055
1056                 channel = consumer_find_channel(key);
1057                 if (!channel) {
1058                         ERR("Kernel consumer lost packets channel %"
1059                                         PRIu64 " not found", key);
1060                         count = 0;
1061                 } else {
1062                         count = channel->lost_packets;
1063                 }
1064
1065                 health_code_update();
1066
1067                 /* Send back returned value to session daemon */
1068                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1069                 if (ret < 0) {
1070                         PERROR("send lost packets");
1071                         goto error_fatal;
1072                 }
1073
1074                 break;
1075         }
1076         case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1077         {
1078                 int channel_monitor_pipe;
1079
1080                 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1081                 /* Successfully received the command's type. */
1082                 ret = consumer_send_status_msg(sock, ret_code);
1083                 if (ret < 0) {
1084                         goto error_fatal;
1085                 }
1086
1087                 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1088                                 1);
1089                 if (ret != sizeof(channel_monitor_pipe)) {
1090                         ERR("Failed to receive channel monitor pipe");
1091                         goto error_fatal;
1092                 }
1093
1094                 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1095                 ret = consumer_timer_thread_set_channel_monitor_pipe(
1096                                 channel_monitor_pipe);
1097                 if (!ret) {
1098                         int flags;
1099
1100                         ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1101                         /* Set the pipe as non-blocking. */
1102                         ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1103                         if (ret == -1) {
1104                                 PERROR("fcntl get flags of the channel monitoring pipe");
1105                                 goto error_fatal;
1106                         }
1107                         flags = ret;
1108
1109                         ret = fcntl(channel_monitor_pipe, F_SETFL,
1110                                         flags | O_NONBLOCK);
1111                         if (ret == -1) {
1112                                 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1113                                 goto error_fatal;
1114                         }
1115                         DBG("Channel monitor pipe set as non-blocking");
1116                 } else {
1117                         ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1118                 }
1119                 ret = consumer_send_status_msg(sock, ret_code);
1120                 if (ret < 0) {
1121                         goto error_fatal;
1122                 }
1123                 break;
1124         }
1125         case LTTNG_CONSUMER_ROTATE_CHANNEL:
1126         {
1127                 struct lttng_consumer_channel *channel;
1128                 uint64_t key = msg.u.rotate_channel.key;
1129
1130                 DBG("Consumer rotate channel %" PRIu64, key);
1131
1132                 channel = consumer_find_channel(key);
1133                 if (!channel) {
1134                         ERR("Channel %" PRIu64 " not found", key);
1135                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1136                 } else {
1137                         /*
1138                          * Sample the rotate position of all the streams in this channel.
1139                          */
1140                         ret = lttng_consumer_rotate_channel(channel, key,
1141                                         msg.u.rotate_channel.relayd_id,
1142                                         msg.u.rotate_channel.metadata,
1143                                         ctx);
1144                         if (ret < 0) {
1145                                 ERR("Rotate channel failed");
1146                                 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1147                         }
1148
1149                         health_code_update();
1150                 }
1151                 ret = consumer_send_status_msg(sock, ret_code);
1152                 if (ret < 0) {
1153                         /* Somehow, the session daemon is not responding anymore. */
1154                         goto error_rotate_channel;
1155                 }
1156                 if (channel) {
1157                         /* Rotate the streams that are ready right now. */
1158                         ret = lttng_consumer_rotate_ready_streams(
1159                                         channel, key, ctx);
1160                         if (ret < 0) {
1161                                 ERR("Rotate ready streams failed");
1162                         }
1163                 }
1164                 break;
1165 error_rotate_channel:
1166                 goto end_nosignal;
1167         }
1168         case LTTNG_CONSUMER_CLEAR_CHANNEL:
1169         {
1170                 struct lttng_consumer_channel *channel;
1171                 uint64_t key = msg.u.clear_channel.key;
1172
1173                 channel = consumer_find_channel(key);
1174                 if (!channel) {
1175                         DBG("Channel %" PRIu64 " not found", key);
1176                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1177                 } else {
1178                         ret = lttng_consumer_clear_channel(channel);
1179                         if (ret) {
1180                                 ERR("Clear channel failed");
1181                                 ret_code = ret;
1182                         }
1183
1184                         health_code_update();
1185                 }
1186                 ret = consumer_send_status_msg(sock, ret_code);
1187                 if (ret < 0) {
1188                         /* Somehow, the session daemon is not responding anymore. */
1189                         goto end_nosignal;
1190                 }
1191
1192                 break;
1193         }
1194         case LTTNG_CONSUMER_INIT:
1195         {
1196                 ret_code = lttng_consumer_init_command(ctx,
1197                                 msg.u.init.sessiond_uuid);
1198                 health_code_update();
1199                 ret = consumer_send_status_msg(sock, ret_code);
1200                 if (ret < 0) {
1201                         /* Somehow, the session daemon is not responding anymore. */
1202                         goto end_nosignal;
1203                 }
1204                 break;
1205         }
1206         case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
1207         {
1208                 const struct lttng_credentials credentials = {
1209                         .uid = msg.u.create_trace_chunk.credentials.value.uid,
1210                         .gid = msg.u.create_trace_chunk.credentials.value.gid,
1211                 };
1212                 const bool is_local_trace =
1213                                 !msg.u.create_trace_chunk.relayd_id.is_set;
1214                 const uint64_t relayd_id =
1215                                 msg.u.create_trace_chunk.relayd_id.value;
1216                 const char *chunk_override_name =
1217                                 *msg.u.create_trace_chunk.override_name ?
1218                                         msg.u.create_trace_chunk.override_name :
1219                                         NULL;
1220                 struct lttng_directory_handle *chunk_directory_handle = NULL;
1221
1222                 /*
1223                  * The session daemon will only provide a chunk directory file
1224                  * descriptor for local traces.
1225                  */
1226                 if (is_local_trace) {
1227                         int chunk_dirfd;
1228
1229                         /* Acnowledge the reception of the command. */
1230                         ret = consumer_send_status_msg(sock,
1231                                         LTTCOMM_CONSUMERD_SUCCESS);
1232                         if (ret < 0) {
1233                                 /* Somehow, the session daemon is not responding anymore. */
1234                                 goto end_nosignal;
1235                         }
1236
1237                         ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
1238                         if (ret != sizeof(chunk_dirfd)) {
1239                                 ERR("Failed to receive trace chunk directory file descriptor");
1240                                 goto error_fatal;
1241                         }
1242
1243                         DBG("Received trace chunk directory fd (%d)",
1244                                         chunk_dirfd);
1245                         chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
1246                                         chunk_dirfd);
1247                         if (!chunk_directory_handle) {
1248                                 ERR("Failed to initialize chunk directory handle from directory file descriptor");
1249                                 if (close(chunk_dirfd)) {
1250                                         PERROR("Failed to close chunk directory file descriptor");
1251                                 }
1252                                 goto error_fatal;
1253                         }
1254                 }
1255
1256                 ret_code = lttng_consumer_create_trace_chunk(
1257                                 !is_local_trace ? &relayd_id : NULL,
1258                                 msg.u.create_trace_chunk.session_id,
1259                                 msg.u.create_trace_chunk.chunk_id,
1260                                 (time_t) msg.u.create_trace_chunk
1261                                                 .creation_timestamp,
1262                                 chunk_override_name,
1263                                 msg.u.create_trace_chunk.credentials.is_set ?
1264                                                 &credentials :
1265                                                 NULL,
1266                                 chunk_directory_handle);
1267                 lttng_directory_handle_put(chunk_directory_handle);
1268                 goto end_msg_sessiond;
1269         }
1270         case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
1271         {
1272                 enum lttng_trace_chunk_command_type close_command =
1273                                 msg.u.close_trace_chunk.close_command.value;
1274                 const uint64_t relayd_id =
1275                                 msg.u.close_trace_chunk.relayd_id.value;
1276                 struct lttcomm_consumer_close_trace_chunk_reply reply;
1277                 char path[LTTNG_PATH_MAX];
1278
1279                 ret_code = lttng_consumer_close_trace_chunk(
1280                                 msg.u.close_trace_chunk.relayd_id.is_set ?
1281                                                 &relayd_id :
1282                                                 NULL,
1283                                 msg.u.close_trace_chunk.session_id,
1284                                 msg.u.close_trace_chunk.chunk_id,
1285                                 (time_t) msg.u.close_trace_chunk.close_timestamp,
1286                                 msg.u.close_trace_chunk.close_command.is_set ?
1287                                                 &close_command :
1288                                                 NULL, path);
1289                 reply.ret_code = ret_code;
1290                 reply.path_length = strlen(path) + 1;
1291                 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
1292                 if (ret != sizeof(reply)) {
1293                         goto error_fatal;
1294                 }
1295                 ret = lttcomm_send_unix_sock(sock, path, reply.path_length);
1296                 if (ret != reply.path_length) {
1297                         goto error_fatal;
1298                 }
1299                 goto end_nosignal;
1300         }
1301         case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
1302         {
1303                 const uint64_t relayd_id =
1304                                 msg.u.trace_chunk_exists.relayd_id.value;
1305
1306                 ret_code = lttng_consumer_trace_chunk_exists(
1307                                 msg.u.trace_chunk_exists.relayd_id.is_set ?
1308                                                 &relayd_id : NULL,
1309                                 msg.u.trace_chunk_exists.session_id,
1310                                 msg.u.trace_chunk_exists.chunk_id);
1311                 goto end_msg_sessiond;
1312         }
1313         default:
1314                 goto end_nosignal;
1315         }
1316
1317 end_nosignal:
1318         /*
1319          * Return 1 to indicate success since the 0 value can be a socket
1320          * shutdown during the recv() or send() call.
1321          */
1322         ret = 1;
1323         goto end;
1324 error_fatal:
1325         /* This will issue a consumer stop. */
1326         ret = -1;
1327         goto end;
1328 end_msg_sessiond:
1329         /*
1330          * The returned value here is not useful since either way we'll return 1 to
1331          * the caller because the session daemon socket management is done
1332          * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1333          */
1334         ret = consumer_send_status_msg(sock, ret_code);
1335         if (ret < 0) {
1336                 goto error_fatal;
1337         }
1338         ret = 1;
1339 end:
1340         health_code_update();
1341         rcu_read_unlock();
1342         return ret;
1343 }
1344
1345 /*
1346  * Sync metadata meaning request them to the session daemon and snapshot to the
1347  * metadata thread can consumer them.
1348  *
1349  * Metadata stream lock MUST be acquired.
1350  *
1351  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1352  * is empty or a negative value on error.
1353  */
1354 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1355 {
1356         int ret;
1357
1358         assert(metadata);
1359
1360         ret = kernctl_buffer_flush(metadata->wait_fd);
1361         if (ret < 0) {
1362                 ERR("Failed to flush kernel stream");
1363                 goto end;
1364         }
1365
1366         ret = kernctl_snapshot(metadata->wait_fd);
1367         if (ret < 0) {
1368                 if (ret != -EAGAIN) {
1369                         ERR("Sync metadata, taking kernel snapshot failed.");
1370                         goto end;
1371                 }
1372                 DBG("Sync metadata, no new kernel metadata");
1373                 /* No new metadata, exit. */
1374                 ret = ENODATA;
1375                 goto end;
1376         }
1377
1378 end:
1379         return ret;
1380 }
1381
1382 static
1383 int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
1384                 struct stream_subbuffer *subbuf)
1385 {
1386         int ret;
1387
1388         ret = kernctl_get_subbuf_size(
1389                         stream->wait_fd, &subbuf->info.data.subbuf_size);
1390         if (ret) {
1391                 goto end;
1392         }
1393
1394         ret = kernctl_get_padded_subbuf_size(
1395                         stream->wait_fd, &subbuf->info.data.padded_subbuf_size);
1396         if (ret) {
1397                 goto end;
1398         }
1399
1400 end:
1401         return ret;
1402 }
1403
1404 static
1405 int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
1406                 struct stream_subbuffer *subbuf)
1407 {
1408         int ret;
1409
1410         ret = extract_common_subbuffer_info(stream, subbuf);
1411         if (ret) {
1412                 goto end;
1413         }
1414
1415         ret = kernctl_get_metadata_version(
1416                         stream->wait_fd, &subbuf->info.metadata.version);
1417         if (ret) {
1418                 goto end;
1419         }
1420
1421 end:
1422         return ret;
1423 }
1424
1425 static
1426 int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
1427                 struct stream_subbuffer *subbuf)
1428 {
1429         int ret;
1430
1431         ret = extract_common_subbuffer_info(stream, subbuf);
1432         if (ret) {
1433                 goto end;
1434         }
1435
1436         ret = kernctl_get_packet_size(
1437                         stream->wait_fd, &subbuf->info.data.packet_size);
1438         if (ret < 0) {
1439                 PERROR("Failed to get sub-buffer packet size");
1440                 goto end;
1441         }
1442
1443         ret = kernctl_get_content_size(
1444                         stream->wait_fd, &subbuf->info.data.content_size);
1445         if (ret < 0) {
1446                 PERROR("Failed to get sub-buffer content size");
1447                 goto end;
1448         }
1449
1450         ret = kernctl_get_timestamp_begin(
1451                         stream->wait_fd, &subbuf->info.data.timestamp_begin);
1452         if (ret < 0) {
1453                 PERROR("Failed to get sub-buffer begin timestamp");
1454                 goto end;
1455         }
1456
1457         ret = kernctl_get_timestamp_end(
1458                         stream->wait_fd, &subbuf->info.data.timestamp_end);
1459         if (ret < 0) {
1460                 PERROR("Failed to get sub-buffer end timestamp");
1461                 goto end;
1462         }
1463
1464         ret = kernctl_get_events_discarded(
1465                         stream->wait_fd, &subbuf->info.data.events_discarded);
1466         if (ret) {
1467                 PERROR("Failed to get sub-buffer events discarded count");
1468                 goto end;
1469         }
1470
1471         ret = kernctl_get_sequence_number(stream->wait_fd,
1472                         &subbuf->info.data.sequence_number.value);
1473         if (ret) {
1474                 /* May not be supported by older LTTng-modules. */
1475                 if (ret != -ENOTTY) {
1476                         PERROR("Failed to get sub-buffer sequence number");
1477                         goto end;
1478                 }
1479         } else {
1480                 subbuf->info.data.sequence_number.is_set = true;
1481         }
1482
1483         ret = kernctl_get_stream_id(
1484                         stream->wait_fd, &subbuf->info.data.stream_id);
1485         if (ret < 0) {
1486                 PERROR("Failed to get stream id");
1487                 goto end;
1488         }
1489
1490         ret = kernctl_get_instance_id(stream->wait_fd,
1491                         &subbuf->info.data.stream_instance_id.value);
1492         if (ret) {
1493                 /* May not be supported by older LTTng-modules. */
1494                 if (ret != -ENOTTY) {
1495                         PERROR("Failed to get stream instance id");
1496                         goto end;
1497                 }
1498         } else {
1499                 subbuf->info.data.stream_instance_id.is_set = true;
1500         }
1501 end:
1502         return ret;
1503 }
1504
1505 static
1506 int get_subbuffer_common(struct lttng_consumer_stream *stream,
1507                 struct stream_subbuffer *subbuffer)
1508 {
1509         int ret;
1510
1511         ret = kernctl_get_next_subbuf(stream->wait_fd);
1512         if (ret) {
1513                 goto end;
1514         }
1515
1516         ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1517                         stream, subbuffer);
1518 end:
1519         return ret;
1520 }
1521
1522 static
1523 int get_next_subbuffer_splice(struct lttng_consumer_stream *stream,
1524                 struct stream_subbuffer *subbuffer)
1525 {
1526         int ret;
1527
1528         ret = get_subbuffer_common(stream, subbuffer);
1529         if (ret) {
1530                 goto end;
1531         }
1532
1533         subbuffer->buffer.fd = stream->wait_fd;
1534 end:
1535         return ret;
1536 }
1537
1538 static
1539 int get_next_subbuffer_mmap(struct lttng_consumer_stream *stream,
1540                 struct stream_subbuffer *subbuffer)
1541 {
1542         int ret;
1543         const char *addr;
1544
1545         ret = get_subbuffer_common(stream, subbuffer);
1546         if (ret) {
1547                 goto end;
1548         }
1549
1550         ret = get_current_subbuf_addr(stream, &addr);
1551         if (ret) {
1552                 goto end;
1553         }
1554
1555         subbuffer->buffer.buffer = lttng_buffer_view_init(
1556                         addr, 0, subbuffer->info.data.padded_subbuf_size);
1557 end:
1558         return ret;
1559 }
1560
1561 static
1562 int put_next_subbuffer(struct lttng_consumer_stream *stream,
1563                 struct stream_subbuffer *subbuffer)
1564 {
1565         const int ret = kernctl_put_next_subbuf(stream->wait_fd);
1566
1567         if (ret) {
1568                 if (ret == -EFAULT) {
1569                         PERROR("Error in unreserving sub buffer");
1570                 } else if (ret == -EIO) {
1571                         /* Should never happen with newer LTTng versions */
1572                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted");
1573                 }
1574         }
1575
1576         return ret;
1577 }
1578
1579 static void lttng_kconsumer_set_stream_ops(
1580                 struct lttng_consumer_stream *stream)
1581 {
1582         if (stream->chan->output == CONSUMER_CHANNEL_MMAP) {
1583                 stream->read_subbuffer_ops.get_next_subbuffer =
1584                                 get_next_subbuffer_mmap;
1585         } else {
1586                 stream->read_subbuffer_ops.get_next_subbuffer =
1587                                 get_next_subbuffer_splice;
1588         }
1589
1590         if (stream->metadata_flag) {
1591                 stream->read_subbuffer_ops.extract_subbuffer_info =
1592                                 extract_metadata_subbuffer_info;
1593         } else {
1594                 stream->read_subbuffer_ops.extract_subbuffer_info =
1595                                 extract_data_subbuffer_info;
1596                 if (stream->chan->is_live) {
1597                         stream->read_subbuffer_ops.send_live_beacon =
1598                                         consumer_flush_kernel_index;
1599                 }
1600         }
1601
1602         stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
1603 }
1604
1605 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1606 {
1607         int ret;
1608
1609         assert(stream);
1610
1611         /*
1612          * Don't create anything if this is set for streaming or if there is
1613          * no current trace chunk on the parent channel.
1614          */
1615         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
1616                         stream->chan->trace_chunk) {
1617                 ret = consumer_stream_create_output_files(stream, true);
1618                 if (ret) {
1619                         goto error;
1620                 }
1621         }
1622
1623         if (stream->output == LTTNG_EVENT_MMAP) {
1624                 /* get the len of the mmap region */
1625                 unsigned long mmap_len;
1626
1627                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1628                 if (ret != 0) {
1629                         PERROR("kernctl_get_mmap_len");
1630                         goto error_close_fd;
1631                 }
1632                 stream->mmap_len = (size_t) mmap_len;
1633
1634                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1635                                 MAP_PRIVATE, stream->wait_fd, 0);
1636                 if (stream->mmap_base == MAP_FAILED) {
1637                         PERROR("Error mmaping");
1638                         ret = -1;
1639                         goto error_close_fd;
1640                 }
1641         }
1642
1643         lttng_kconsumer_set_stream_ops(stream);
1644
1645         /* we return 0 to let the library handle the FD internally */
1646         return 0;
1647
1648 error_close_fd:
1649         if (stream->out_fd >= 0) {
1650                 int err;
1651
1652                 err = close(stream->out_fd);
1653                 assert(!err);
1654                 stream->out_fd = -1;
1655         }
1656 error:
1657         return ret;
1658 }
1659
1660 /*
1661  * Check if data is still being extracted from the buffers for a specific
1662  * stream. Consumer data lock MUST be acquired before calling this function
1663  * and the stream lock.
1664  *
1665  * Return 1 if the traced data are still getting read else 0 meaning that the
1666  * data is available for trace viewer reading.
1667  */
1668 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1669 {
1670         int ret;
1671
1672         assert(stream);
1673
1674         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1675                 ret = 0;
1676                 goto end;
1677         }
1678
1679         ret = kernctl_get_next_subbuf(stream->wait_fd);
1680         if (ret == 0) {
1681                 /* There is still data so let's put back this subbuffer. */
1682                 ret = kernctl_put_subbuf(stream->wait_fd);
1683                 assert(ret == 0);
1684                 ret = 1;   /* Data is pending */
1685                 goto end;
1686         }
1687
1688         /* Data is NOT pending and ready to be read. */
1689         ret = 0;
1690
1691 end:
1692         return ret;
1693 }