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

   1 /*
   2  * Copyright (C) 2011 EfficiOS Inc.
   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 #include <stdint.h>
  23
  24 #include <bin/lttng-consumerd/health-consumerd.h>
  25 #include <common/common.h>
  26 #include <common/kernel-ctl/kernel-ctl.h>
  27 #include <common/sessiond-comm/sessiond-comm.h>
  28 #include <common/sessiond-comm/relayd.h>
  29 #include <common/compat/fcntl.h>
  30 #include <common/compat/endian.h>
  31 #include <common/pipe.h>
  32 #include <common/relayd/relayd.h>
  33 #include <common/utils.h>
  34 #include <common/consumer/consumer-stream.h>
  35 #include <common/index/index.h>
  36 #include <common/consumer/consumer-timer.h>
  37 #include <common/optional.h>
  38 #include <common/buffer-view.h>
  39 #include <common/consumer/consumer.h>
  40 #include <common/consumer/metadata-bucket.h>
  41
  42 #include "kernel-consumer.h"
  43
  44 extern struct lttng_consumer_global_data the_consumer_data;
  45 extern int consumer_poll_timeout;
  46
  47 /*
  48  * Take a snapshot for a specific fd
  49  *
  50  * Returns 0 on success, < 0 on error
  51  */
  52 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  53 {
  54         int ret = 0;
  55         int infd = stream->wait_fd;
  56
  57         ret = kernctl_snapshot(infd);
  58         /*
  59          * -EAGAIN is not an error, it just means that there is no data to
  60          *  be read.
  61          */
  62         if (ret != 0 && ret != -EAGAIN) {
  63                 PERROR("Getting sub-buffer snapshot.");
  64         }
  65
  66         return ret;
  67 }
  68
  69 /*
  70  * Sample consumed and produced positions for a specific fd.
  71  *
  72  * Returns 0 on success, < 0 on error.
  73  */
  74 int lttng_kconsumer_sample_snapshot_positions(
  75                 struct lttng_consumer_stream *stream)
  76 {
  77         assert(stream);
  78
  79         return kernctl_snapshot_sample_positions(stream->wait_fd);
  80 }
  81
  82 /*
  83  * Get the produced position
  84  *
  85  * Returns 0 on success, < 0 on error
  86  */
  87 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  88                 unsigned long *pos)
  89 {
  90         int ret;
  91         int infd = stream->wait_fd;
  92
  93         ret = kernctl_snapshot_get_produced(infd, pos);
  94         if (ret != 0) {
  95                 PERROR("kernctl_snapshot_get_produced");
  96         }
  97
  98         return ret;
  99 }
 100
 101 /*
 102  * Get the consumerd position
 103  *
 104  * Returns 0 on success, < 0 on error
 105  */
 106 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 107                 unsigned long *pos)
 108 {
 109         int ret;
 110         int infd = stream->wait_fd;
 111
 112         ret = kernctl_snapshot_get_consumed(infd, pos);
 113         if (ret != 0) {
 114                 PERROR("kernctl_snapshot_get_consumed");
 115         }
 116
 117         return ret;
 118 }
 119
 120 static
 121 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
 122                 const char **addr)
 123 {
 124         int ret;
 125         unsigned long mmap_offset;
 126         const char *mmap_base = stream->mmap_base;
 127
 128         ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
 129         if (ret < 0) {
 130                 PERROR("Failed to get mmap read offset");
 131                 goto error;
 132         }
 133
 134         *addr = mmap_base + mmap_offset;
 135 error:
 136         return ret;
 137 }
 138
 139 static void finalize_snapshot_stream(
 140                 struct lttng_consumer_stream *stream, uint64_t relayd_id)
 141 {
 142         ASSERT_LOCKED(stream->lock);
 143
 144         if (relayd_id == (uint64_t) -1ULL) {
 145                 if (stream->out_fd >= 0) {
 146                         const int ret = close(stream->out_fd);
 147
 148                         if (ret < 0) {
 149                                 PERROR("Failed to close stream snapshot output file descriptor");
 150                         }
 151
 152                         stream->out_fd = -1;
 153                 }
 154         } else {
 155                 close_relayd_stream(stream);
 156                 stream->net_seq_idx = (uint64_t) -1ULL;
 157         }
 158
 159         lttng_trace_chunk_put(stream->trace_chunk);
 160         stream->trace_chunk = NULL;
 161 }
 162
 163 /*
 164  * Take a snapshot of all the stream of a channel
 165  * RCU read-side lock must be held across this function to ensure existence of
 166  * channel.
 167  *
 168  * Returns 0 on success, < 0 on error
 169  */
 170 static int lttng_kconsumer_snapshot_channel(
 171                 struct lttng_consumer_channel *channel,
 172                 uint64_t key, char *path, uint64_t relayd_id,
 173                 uint64_t nb_packets_per_stream,
 174                 struct lttng_consumer_local_data *ctx)
 175 {
 176         int ret;
 177         struct lttng_consumer_stream *stream;
 178
 179         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 180
 181         /* Prevent channel modifications while we perform the snapshot.*/
 182         pthread_mutex_lock(&channel->lock);
 183
 184         rcu_read_lock();
 185
 186         /* Splice is not supported yet for channel snapshot. */
 187         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 188                 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
 189                                 channel->name);
 190                 ret = -1;
 191                 goto end;
 192         }
 193
 194         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 195                 unsigned long consumed_pos, produced_pos;
 196
 197                 health_code_update();
 198
 199                 /*
 200                  * Lock stream because we are about to change its state.
 201                  */
 202                 pthread_mutex_lock(&stream->lock);
 203
 204                 assert(channel->trace_chunk);
 205                 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
 206                         /*
 207                          * Can't happen barring an internal error as the channel
 208                          * holds a reference to the trace chunk.
 209                          */
 210                         ERR("Failed to acquire reference to channel's trace chunk");
 211                         ret = -1;
 212                         goto end_unlock;
 213                 }
 214                 assert(!stream->trace_chunk);
 215                 stream->trace_chunk = channel->trace_chunk;
 216
 217                 /*
 218                  * Assign the received relayd ID so we can use it for streaming. The streams
 219                  * are not visible to anyone so this is OK to change it.
 220                  */
 221                 stream->net_seq_idx = relayd_id;
 222                 channel->relayd_id = relayd_id;
 223                 if (relayd_id != (uint64_t) -1ULL) {
 224                         ret = consumer_send_relayd_stream(stream, path);
 225                         if (ret < 0) {
 226                                 ERR("sending stream to relayd");
 227                                 goto error_finalize_stream;
 228                         }
 229                 } else {
 230                         ret = consumer_stream_create_output_files(stream,
 231                                         false);
 232                         if (ret < 0) {
 233                                 goto error_finalize_stream;
 234                         }
 235                         DBG("Kernel consumer snapshot stream (%" PRIu64 ")",
 236                                         stream->key);
 237                 }
 238
 239                 ret = kernctl_buffer_flush_empty(stream->wait_fd);
 240                 if (ret < 0) {
 241                         /*
 242                          * Doing a buffer flush which does not take into
 243                          * account empty packets. This is not perfect
 244                          * for stream intersection, but required as a
 245                          * fall-back when "flush_empty" is not
 246                          * implemented by lttng-modules.
 247                          */
 248                         ret = kernctl_buffer_flush(stream->wait_fd);
 249                         if (ret < 0) {
 250                                 ERR("Failed to flush kernel stream");
 251                                 goto error_finalize_stream;
 252                         }
 253                         goto end_unlock;
 254                 }
 255
 256                 ret = lttng_kconsumer_take_snapshot(stream);
 257                 if (ret < 0) {
 258                         ERR("Taking kernel snapshot");
 259                         goto error_finalize_stream;
 260                 }
 261
 262                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 263                 if (ret < 0) {
 264                         ERR("Produced kernel snapshot position");
 265                         goto error_finalize_stream;
 266                 }
 267
 268                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 269                 if (ret < 0) {
 270                         ERR("Consumerd kernel snapshot position");
 271                         goto error_finalize_stream;
 272                 }
 273
 274                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 275                                 produced_pos, nb_packets_per_stream,
 276                                 stream->max_sb_size);
 277
 278                 while ((long) (consumed_pos - produced_pos) < 0) {
 279                         ssize_t read_len;
 280                         unsigned long len, padded_len;
 281                         const char *subbuf_addr;
 282                         struct lttng_buffer_view subbuf_view;
 283
 284                         health_code_update();
 285                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 286
 287                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 288                         if (ret < 0) {
 289                                 if (ret != -EAGAIN) {
 290                                         PERROR("kernctl_get_subbuf snapshot");
 291                                         goto error_finalize_stream;
 292                                 }
 293                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 294                                 consumed_pos += stream->max_sb_size;
 295                                 stream->chan->lost_packets++;
 296                                 continue;
 297                         }
 298
 299                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 300                         if (ret < 0) {
 301                                 ERR("Snapshot kernctl_get_subbuf_size");
 302                                 goto error_put_subbuf;
 303                         }
 304
 305                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 306                         if (ret < 0) {
 307                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 308                                 goto error_put_subbuf;
 309                         }
 310
 311                         ret = get_current_subbuf_addr(stream, &subbuf_addr);
 312                         if (ret) {
 313                                 goto error_put_subbuf;
 314                         }
 315
 316                         subbuf_view = lttng_buffer_view_init(
 317                                         subbuf_addr, 0, padded_len);
 318                         read_len = lttng_consumer_on_read_subbuffer_mmap(
 319                                         stream, &subbuf_view,
 320                                         padded_len - len);
 321                         /*
 322                          * We write the padded len in local tracefiles but the data len
 323                          * when using a relay. Display the error but continue processing
 324                          * to try to release the subbuffer.
 325                          */
 326                         if (relayd_id != (uint64_t) -1ULL) {
 327                                 if (read_len != len) {
 328                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 329                                                         read_len, len);
 330                                 }
 331                         } else {
 332                                 if (read_len != padded_len) {
 333                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 334                                                         read_len, padded_len);
 335                                 }
 336                         }
 337
 338                         ret = kernctl_put_subbuf(stream->wait_fd);
 339                         if (ret < 0) {
 340                                 ERR("Snapshot kernctl_put_subbuf");
 341                                 goto error_finalize_stream;
 342                         }
 343                         consumed_pos += stream->max_sb_size;
 344                 }
 345
 346                 finalize_snapshot_stream(stream, relayd_id);
 347                 pthread_mutex_unlock(&stream->lock);
 348         }
 349
 350         /* All good! */
 351         ret = 0;
 352         goto end;
 353
 354 error_put_subbuf:
 355         ret = kernctl_put_subbuf(stream->wait_fd);
 356         if (ret < 0) {
 357                 ERR("Snapshot kernctl_put_subbuf error path");
 358         }
 359 error_finalize_stream:
 360         finalize_snapshot_stream(stream, relayd_id);
 361 end_unlock:
 362         pthread_mutex_unlock(&stream->lock);
 363 end:
 364         rcu_read_unlock();
 365         pthread_mutex_unlock(&channel->lock);
 366         return ret;
 367 }
 368
 369 /*
 370  * Read the whole metadata available for a snapshot.
 371  * RCU read-side lock must be held across this function to ensure existence of
 372  * metadata_channel.
 373  *
 374  * Returns 0 on success, < 0 on error
 375  */
 376 static int lttng_kconsumer_snapshot_metadata(
 377                 struct lttng_consumer_channel *metadata_channel,
 378                 uint64_t key, char *path, uint64_t relayd_id,
 379                 struct lttng_consumer_local_data *ctx)
 380 {
 381         int ret, use_relayd = 0;
 382         ssize_t ret_read;
 383         struct lttng_consumer_stream *metadata_stream;
 384
 385         assert(ctx);
 386
 387         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 388                         key, path);
 389
 390         rcu_read_lock();
 391
 392         metadata_stream = metadata_channel->metadata_stream;
 393         assert(metadata_stream);
 394
 395         metadata_stream->read_subbuffer_ops.lock(metadata_stream);
 396         assert(metadata_channel->trace_chunk);
 397         assert(metadata_stream->trace_chunk);
 398
 399         /* Flag once that we have a valid relayd for the stream. */
 400         if (relayd_id != (uint64_t) -1ULL) {
 401                 use_relayd = 1;
 402         }
 403
 404         if (use_relayd) {
 405                 ret = consumer_send_relayd_stream(metadata_stream, path);
 406                 if (ret < 0) {
 407                         goto error_snapshot;
 408                 }
 409         } else {
 410                 ret = consumer_stream_create_output_files(metadata_stream,
 411                                 false);
 412                 if (ret < 0) {
 413                         goto error_snapshot;
 414                 }
 415         }
 416
 417         do {
 418                 health_code_update();
 419
 420                 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
 421                 if (ret_read < 0) {
 422                         ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 423                                         ret_read);
 424                         ret = ret_read;
 425                         goto error_snapshot;
 426                 }
 427         } while (ret_read > 0);
 428
 429         if (use_relayd) {
 430                 close_relayd_stream(metadata_stream);
 431                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 432         } else {
 433                 if (metadata_stream->out_fd >= 0) {
 434                         ret = close(metadata_stream->out_fd);
 435                         if (ret < 0) {
 436                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 437                                 /*
 438                                  * Don't go on error here since the snapshot was successful at this
 439                                  * point but somehow the close failed.
 440                                  */
 441                         }
 442                         metadata_stream->out_fd = -1;
 443                         lttng_trace_chunk_put(metadata_stream->trace_chunk);
 444                         metadata_stream->trace_chunk = NULL;
 445                 }
 446         }
 447
 448         ret = 0;
 449 error_snapshot:
 450         metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
 451         consumer_stream_destroy(metadata_stream, NULL);
 452         metadata_channel->metadata_stream = NULL;
 453         rcu_read_unlock();
 454         return ret;
 455 }
 456
 457 /*
 458  * Receive command from session daemon and process it.
 459  *
 460  * Return 1 on success else a negative value or 0.
 461  */
 462 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 463                 int sock, struct pollfd *consumer_sockpoll)
 464 {
 465         int ret_func;
 466         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 467         struct lttcomm_consumer_msg msg;
 468
 469         health_code_update();
 470
 471         {
 472                 ssize_t ret_recv;
 473
 474                 ret_recv = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 475                 if (ret_recv != sizeof(msg)) {
 476                         if (ret_recv > 0) {
 477                                 lttng_consumer_send_error(ctx,
 478                                                 LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 479                                 ret_recv = -1;
 480                         }
 481                         return ret_recv;
 482                 }
 483         }
 484
 485         health_code_update();
 486
 487         /* Deprecated command */
 488         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 489
 490         health_code_update();
 491
 492         /* relayd needs RCU read-side protection */
 493         rcu_read_lock();
 494
 495         switch (msg.cmd_type) {
 496         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 497         {
 498                 uint32_t major = msg.u.relayd_sock.major;
 499                 uint32_t minor = msg.u.relayd_sock.minor;
 500                 enum lttcomm_sock_proto protocol = (enum lttcomm_sock_proto)
 501                                 msg.u.relayd_sock.relayd_socket_protocol;
 502
 503                 /* Session daemon status message are handled in the following call. */
 504                 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 505                                 msg.u.relayd_sock.type, ctx, sock,
 506                                 consumer_sockpoll, msg.u.relayd_sock.session_id,
 507                                 msg.u.relayd_sock.relayd_session_id, major,
 508                                 minor, protocol);
 509                 goto end_nosignal;
 510         }
 511         case LTTNG_CONSUMER_ADD_CHANNEL:
 512         {
 513                 struct lttng_consumer_channel *new_channel;
 514                 int ret_send_status, ret_add_channel = 0;
 515                 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
 516
 517                 health_code_update();
 518
 519                 /* First send a status message before receiving the fds. */
 520                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 521                 if (ret_send_status < 0) {
 522                         /* Somehow, the session daemon is not responding anymore. */
 523                         goto error_fatal;
 524                 }
 525
 526                 health_code_update();
 527
 528                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 529                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 530                                 msg.u.channel.session_id,
 531                                 msg.u.channel.chunk_id.is_set ?
 532                                                 &chunk_id : NULL,
 533                                 msg.u.channel.pathname,
 534                                 msg.u.channel.name,
 535                                 msg.u.channel.relayd_id, msg.u.channel.output,
 536                                 msg.u.channel.tracefile_size,
 537                                 msg.u.channel.tracefile_count, 0,
 538                                 msg.u.channel.monitor,
 539                                 msg.u.channel.live_timer_interval,
 540                                 msg.u.channel.is_live,
 541                                 NULL, NULL);
 542                 if (new_channel == NULL) {
 543                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 544                         goto end_nosignal;
 545                 }
 546                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 547                 switch (msg.u.channel.output) {
 548                 case LTTNG_EVENT_SPLICE:
 549                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 550                         break;
 551                 case LTTNG_EVENT_MMAP:
 552                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 553                         break;
 554                 default:
 555                         ERR("Channel output unknown %d", msg.u.channel.output);
 556                         goto end_nosignal;
 557                 }
 558
 559                 /* Translate and save channel type. */
 560                 switch (msg.u.channel.type) {
 561                 case CONSUMER_CHANNEL_TYPE_DATA:
 562                 case CONSUMER_CHANNEL_TYPE_METADATA:
 563                         new_channel->type = msg.u.channel.type;
 564                         break;
 565                 default:
 566                         assert(0);
 567                         goto end_nosignal;
 568                 };
 569
 570                 health_code_update();
 571
 572                 if (ctx->on_recv_channel != NULL) {
 573                         int ret_recv_channel =
 574                                         ctx->on_recv_channel(new_channel);
 575                         if (ret_recv_channel == 0) {
 576                                 ret_add_channel = consumer_add_channel(
 577                                                 new_channel, ctx);
 578                         } else if (ret_recv_channel < 0) {
 579                                 goto end_nosignal;
 580                         }
 581                 } else {
 582                         ret_add_channel =
 583                                         consumer_add_channel(new_channel, ctx);
 584                 }
 585                 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA &&
 586                                 !ret_add_channel) {
 587                         int monitor_start_ret;
 588
 589                         DBG("Consumer starting monitor timer");
 590                         consumer_timer_live_start(new_channel,
 591                                         msg.u.channel.live_timer_interval);
 592                         monitor_start_ret = consumer_timer_monitor_start(
 593                                         new_channel,
 594                                         msg.u.channel.monitor_timer_interval);
 595                         if (monitor_start_ret < 0) {
 596                                 ERR("Starting channel monitoring timer failed");
 597                                 goto end_nosignal;
 598                         }
 599                 }
 600
 601                 health_code_update();
 602
 603                 /* If we received an error in add_channel, we need to report it. */
 604                 if (ret_add_channel < 0) {
 605                         ret_send_status = consumer_send_status_msg(
 606                                         sock, ret_add_channel);
 607                         if (ret_send_status < 0) {
 608                                 goto error_fatal;
 609                         }
 610                         goto end_nosignal;
 611                 }
 612
 613                 goto end_nosignal;
 614         }
 615         case LTTNG_CONSUMER_ADD_STREAM:
 616         {
 617                 int fd;
 618                 struct lttng_pipe *stream_pipe;
 619                 struct lttng_consumer_stream *new_stream;
 620                 struct lttng_consumer_channel *channel;
 621                 int alloc_ret = 0;
 622                 int ret_send_status, ret_poll, ret_get_max_subbuf_size;
 623                 ssize_t ret_pipe_write, ret_recv;
 624
 625                 /*
 626                  * Get stream's channel reference. Needed when adding the stream to the
 627                  * global hash table.
 628                  */
 629                 channel = consumer_find_channel(msg.u.stream.channel_key);
 630                 if (!channel) {
 631                         /*
 632                          * We could not find the channel. Can happen if cpu hotplug
 633                          * happens while tearing down.
 634                          */
 635                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 636                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 637                 }
 638
 639                 health_code_update();
 640
 641                 /* First send a status message before receiving the fds. */
 642                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 643                 if (ret_send_status < 0) {
 644                         /* Somehow, the session daemon is not responding anymore. */
 645                         goto error_add_stream_fatal;
 646                 }
 647
 648                 health_code_update();
 649
 650                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 651                         /* Channel was not found. */
 652                         goto error_add_stream_nosignal;
 653                 }
 654
 655                 /* Blocking call */
 656                 health_poll_entry();
 657                 ret_poll = lttng_consumer_poll_socket(consumer_sockpoll);
 658                 health_poll_exit();
 659                 if (ret_poll) {
 660                         goto error_add_stream_fatal;
 661                 }
 662
 663                 health_code_update();
 664
 665                 /* Get stream file descriptor from socket */
 666                 ret_recv = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 667                 if (ret_recv != sizeof(fd)) {
 668                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 669                         ret_func = ret_recv;
 670                         goto end;
 671                 }
 672
 673                 health_code_update();
 674
 675                 /*
 676                  * Send status code to session daemon only if the recv works. If the
 677                  * above recv() failed, the session daemon is notified through the
 678                  * error socket and the teardown is eventually done.
 679                  */
 680                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 681                 if (ret_send_status < 0) {
 682                         /* Somehow, the session daemon is not responding anymore. */
 683                         goto error_add_stream_nosignal;
 684                 }
 685
 686                 health_code_update();
 687
 688                 pthread_mutex_lock(&channel->lock);
 689                 new_stream = consumer_stream_create(
 690                                 channel,
 691                                 channel->key,
 692                                 fd,
 693                                 channel->name,
 694                                 channel->relayd_id,
 695                                 channel->session_id,
 696                                 channel->trace_chunk,
 697                                 msg.u.stream.cpu,
 698                                 &alloc_ret,
 699                                 channel->type,
 700                                 channel->monitor);
 701                 if (new_stream == NULL) {
 702                         switch (alloc_ret) {
 703                         case -ENOMEM:
 704                         case -EINVAL:
 705                         default:
 706                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 707                                 break;
 708                         }
 709                         pthread_mutex_unlock(&channel->lock);
 710                         goto error_add_stream_nosignal;
 711                 }
 712
 713                 new_stream->wait_fd = fd;
 714                 ret_get_max_subbuf_size = kernctl_get_max_subbuf_size(
 715                                 new_stream->wait_fd, &new_stream->max_sb_size);
 716                 if (ret_get_max_subbuf_size < 0) {
 717                         pthread_mutex_unlock(&channel->lock);
 718                         ERR("Failed to get kernel maximal subbuffer size");
 719                         goto error_add_stream_nosignal;
 720                 }
 721
 722                 consumer_stream_update_channel_attributes(new_stream,
 723                                 channel);
 724
 725                 /*
 726                  * We've just assigned the channel to the stream so increment the
 727                  * refcount right now. We don't need to increment the refcount for
 728                  * streams in no monitor because we handle manually the cleanup of
 729                  * those. It is very important to make sure there is NO prior
 730                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 731                  */
 732                 if (channel->monitor) {
 733                         uatomic_inc(&new_stream->chan->refcount);
 734                 }
 735
 736                 /*
 737                  * The buffer flush is done on the session daemon side for the kernel
 738                  * so no need for the stream "hangup_flush_done" variable to be
 739                  * tracked. This is important for a kernel stream since we don't rely
 740                  * on the flush state of the stream to read data. It's not the case for
 741                  * user space tracing.
 742                  */
 743                 new_stream->hangup_flush_done = 0;
 744
 745                 health_code_update();
 746
 747                 pthread_mutex_lock(&new_stream->lock);
 748                 if (ctx->on_recv_stream) {
 749                         int ret_recv_stream = ctx->on_recv_stream(new_stream);
 750                         if (ret_recv_stream < 0) {
 751                                 pthread_mutex_unlock(&new_stream->lock);
 752                                 pthread_mutex_unlock(&channel->lock);
 753                                 consumer_stream_free(new_stream);
 754                                 goto error_add_stream_nosignal;
 755                         }
 756                 }
 757                 health_code_update();
 758
 759                 if (new_stream->metadata_flag) {
 760                         channel->metadata_stream = new_stream;
 761                 }
 762
 763                 /* Do not monitor this stream. */
 764                 if (!channel->monitor) {
 765                         DBG("Kernel consumer add stream %s in no monitor mode with "
 766                                         "relayd id %" PRIu64, new_stream->name,
 767                                         new_stream->net_seq_idx);
 768                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 769                         pthread_mutex_unlock(&new_stream->lock);
 770                         pthread_mutex_unlock(&channel->lock);
 771                         goto end_add_stream;
 772                 }
 773
 774                 /* Send stream to relayd if the stream has an ID. */
 775                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 776                         int ret_send_relayd_stream;
 777
 778                         ret_send_relayd_stream = consumer_send_relayd_stream(
 779                                         new_stream, new_stream->chan->pathname);
 780                         if (ret_send_relayd_stream < 0) {
 781                                 pthread_mutex_unlock(&new_stream->lock);
 782                                 pthread_mutex_unlock(&channel->lock);
 783                                 consumer_stream_free(new_stream);
 784                                 goto error_add_stream_nosignal;
 785                         }
 786
 787                         /*
 788                          * If adding an extra stream to an already
 789                          * existing channel (e.g. cpu hotplug), we need
 790                          * to send the "streams_sent" command to relayd.
 791                          */
 792                         if (channel->streams_sent_to_relayd) {
 793                                 int ret_send_relayd_streams_sent;
 794
 795                                 ret_send_relayd_streams_sent =
 796                                                 consumer_send_relayd_streams_sent(
 797                                                                 new_stream->net_seq_idx);
 798                                 if (ret_send_relayd_streams_sent < 0) {
 799                                         pthread_mutex_unlock(&new_stream->lock);
 800                                         pthread_mutex_unlock(&channel->lock);
 801                                         goto error_add_stream_nosignal;
 802                                 }
 803                         }
 804                 }
 805                 pthread_mutex_unlock(&new_stream->lock);
 806                 pthread_mutex_unlock(&channel->lock);
 807
 808                 /* Get the right pipe where the stream will be sent. */
 809                 if (new_stream->metadata_flag) {
 810                         consumer_add_metadata_stream(new_stream);
 811                         stream_pipe = ctx->consumer_metadata_pipe;
 812                 } else {
 813                         consumer_add_data_stream(new_stream);
 814                         stream_pipe = ctx->consumer_data_pipe;
 815                 }
 816
 817                 /* Visible to other threads */
 818                 new_stream->globally_visible = 1;
 819
 820                 health_code_update();
 821
 822                 ret_pipe_write = lttng_pipe_write(
 823                                 stream_pipe, &new_stream, sizeof(new_stream));
 824                 if (ret_pipe_write < 0) {
 825                         ERR("Consumer write %s stream to pipe %d",
 826                                         new_stream->metadata_flag ? "metadata" : "data",
 827                                         lttng_pipe_get_writefd(stream_pipe));
 828                         if (new_stream->metadata_flag) {
 829                                 consumer_del_stream_for_metadata(new_stream);
 830                         } else {
 831                                 consumer_del_stream_for_data(new_stream);
 832                         }
 833                         goto error_add_stream_nosignal;
 834                 }
 835
 836                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
 837                                 new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
 838 end_add_stream:
 839                 break;
 840 error_add_stream_nosignal:
 841                 goto end_nosignal;
 842 error_add_stream_fatal:
 843                 goto error_fatal;
 844         }
 845         case LTTNG_CONSUMER_STREAMS_SENT:
 846         {
 847                 struct lttng_consumer_channel *channel;
 848                 int ret_send_status;
 849
 850                 /*
 851                  * Get stream's channel reference. Needed when adding the stream to the
 852                  * global hash table.
 853                  */
 854                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 855                 if (!channel) {
 856                         /*
 857                          * We could not find the channel. Can happen if cpu hotplug
 858                          * happens while tearing down.
 859                          */
 860                         ERR("Unable to find channel key %" PRIu64,
 861                                         msg.u.sent_streams.channel_key);
 862                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 863                 }
 864
 865                 health_code_update();
 866
 867                 /*
 868                  * Send status code to session daemon.
 869                  */
 870                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 871                 if (ret_send_status < 0 ||
 872                                 ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 873                         /* Somehow, the session daemon is not responding anymore. */
 874                         goto error_streams_sent_nosignal;
 875                 }
 876
 877                 health_code_update();
 878
 879                 /*
 880                  * We should not send this message if we don't monitor the
 881                  * streams in this channel.
 882                  */
 883                 if (!channel->monitor) {
 884                         goto end_error_streams_sent;
 885                 }
 886
 887                 health_code_update();
 888                 /* Send stream to relayd if the stream has an ID. */
 889                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 890                         int ret_send_relay_streams;
 891
 892                         ret_send_relay_streams = consumer_send_relayd_streams_sent(
 893                                         msg.u.sent_streams.net_seq_idx);
 894                         if (ret_send_relay_streams < 0) {
 895                                 goto error_streams_sent_nosignal;
 896                         }
 897                         channel->streams_sent_to_relayd = true;
 898                 }
 899 end_error_streams_sent:
 900                 break;
 901 error_streams_sent_nosignal:
 902                 goto end_nosignal;
 903         }
 904         case LTTNG_CONSUMER_UPDATE_STREAM:
 905         {
 906                 rcu_read_unlock();
 907                 return -ENOSYS;
 908         }
 909         case LTTNG_CONSUMER_DESTROY_RELAYD:
 910         {
 911                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 912                 struct consumer_relayd_sock_pair *relayd;
 913                 int ret_send_status;
 914
 915                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 916
 917                 /* Get relayd reference if exists. */
 918                 relayd = consumer_find_relayd(index);
 919                 if (relayd == NULL) {
 920                         DBG("Unable to find relayd %" PRIu64, index);
 921                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 922                 }
 923
 924                 /*
 925                  * Each relayd socket pair has a refcount of stream attached to it
 926                  * which tells if the relayd is still active or not depending on the
 927                  * refcount value.
 928                  *
 929                  * This will set the destroy flag of the relayd object and destroy it
 930                  * if the refcount reaches zero when called.
 931                  *
 932                  * The destroy can happen either here or when a stream fd hangs up.
 933                  */
 934                 if (relayd) {
 935                         consumer_flag_relayd_for_destroy(relayd);
 936                 }
 937
 938                 health_code_update();
 939
 940                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 941                 if (ret_send_status < 0) {
 942                         /* Somehow, the session daemon is not responding anymore. */
 943                         goto error_fatal;
 944                 }
 945
 946                 goto end_nosignal;
 947         }
 948         case LTTNG_CONSUMER_DATA_PENDING:
 949         {
 950                 int32_t ret_data_pending;
 951                 uint64_t id = msg.u.data_pending.session_id;
 952                 ssize_t ret_send;
 953
 954                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 955
 956                 ret_data_pending = consumer_data_pending(id);
 957
 958                 health_code_update();
 959
 960                 /* Send back returned value to session daemon */
 961                 ret_send = lttcomm_send_unix_sock(sock, &ret_data_pending,
 962                                 sizeof(ret_data_pending));
 963                 if (ret_send < 0) {
 964                         PERROR("send data pending ret code");
 965                         goto error_fatal;
 966                 }
 967
 968                 /*
 969                  * No need to send back a status message since the data pending
 970                  * returned value is the response.
 971                  */
 972                 break;
 973         }
 974         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 975         {
 976                 struct lttng_consumer_channel *channel;
 977                 uint64_t key = msg.u.snapshot_channel.key;
 978                 int ret_send_status;
 979
 980                 channel = consumer_find_channel(key);
 981                 if (!channel) {
 982                         ERR("Channel %" PRIu64 " not found", key);
 983                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 984                 } else {
 985                         if (msg.u.snapshot_channel.metadata == 1) {
 986                                 int ret_snapshot;
 987
 988                                 ret_snapshot = lttng_kconsumer_snapshot_metadata(
 989                                                 channel, key,
 990                                                 msg.u.snapshot_channel.pathname,
 991                                                 msg.u.snapshot_channel.relayd_id,
 992                                                 ctx);
 993                                 if (ret_snapshot < 0) {
 994                                         ERR("Snapshot metadata failed");
 995                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
 996                                 }
 997                         } else {
 998                                 int ret_snapshot;
 999
1000                                 ret_snapshot = lttng_kconsumer_snapshot_channel(
1001                                                 channel, key,
1002                                                 msg.u.snapshot_channel.pathname,
1003                                                 msg.u.snapshot_channel.relayd_id,
1004                                                 msg.u.snapshot_channel
1005                                                                 .nb_packets_per_stream,
1006                                                 ctx);
1007                                 if (ret_snapshot < 0) {
1008                                         ERR("Snapshot channel failed");
1009                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1010                                 }
1011                         }
1012                 }
1013                 health_code_update();
1014
1015                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1016                 if (ret_send_status < 0) {
1017                         /* Somehow, the session daemon is not responding anymore. */
1018                         goto end_nosignal;
1019                 }
1020                 break;
1021         }
1022         case LTTNG_CONSUMER_DESTROY_CHANNEL:
1023         {
1024                 uint64_t key = msg.u.destroy_channel.key;
1025                 struct lttng_consumer_channel *channel;
1026                 int ret_send_status;
1027
1028                 channel = consumer_find_channel(key);
1029                 if (!channel) {
1030                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
1031                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1032                 }
1033
1034                 health_code_update();
1035
1036                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1037                 if (ret_send_status < 0) {
1038                         /* Somehow, the session daemon is not responding anymore. */
1039                         goto end_destroy_channel;
1040                 }
1041
1042                 health_code_update();
1043
1044                 /* Stop right now if no channel was found. */
1045                 if (!channel) {
1046                         goto end_destroy_channel;
1047                 }
1048
1049                 /*
1050                  * This command should ONLY be issued for channel with streams set in
1051                  * no monitor mode.
1052                  */
1053                 assert(!channel->monitor);
1054
1055                 /*
1056                  * The refcount should ALWAYS be 0 in the case of a channel in no
1057                  * monitor mode.
1058                  */
1059                 assert(!uatomic_sub_return(&channel->refcount, 1));
1060
1061                 consumer_del_channel(channel);
1062 end_destroy_channel:
1063                 goto end_nosignal;
1064         }
1065         case LTTNG_CONSUMER_DISCARDED_EVENTS:
1066         {
1067                 ssize_t ret;
1068                 uint64_t count;
1069                 struct lttng_consumer_channel *channel;
1070                 uint64_t id = msg.u.discarded_events.session_id;
1071                 uint64_t key = msg.u.discarded_events.channel_key;
1072
1073                 DBG("Kernel consumer discarded events command for session id %"
1074                                 PRIu64 ", channel key %" PRIu64, id, key);
1075
1076                 channel = consumer_find_channel(key);
1077                 if (!channel) {
1078                         ERR("Kernel consumer discarded events channel %"
1079                                         PRIu64 " not found", key);
1080                         count = 0;
1081                 } else {
1082                         count = channel->discarded_events;
1083                 }
1084
1085                 health_code_update();
1086
1087                 /* Send back returned value to session daemon */
1088                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1089                 if (ret < 0) {
1090                         PERROR("send discarded events");
1091                         goto error_fatal;
1092                 }
1093
1094                 break;
1095         }
1096         case LTTNG_CONSUMER_LOST_PACKETS:
1097         {
1098                 ssize_t ret;
1099                 uint64_t count;
1100                 struct lttng_consumer_channel *channel;
1101                 uint64_t id = msg.u.lost_packets.session_id;
1102                 uint64_t key = msg.u.lost_packets.channel_key;
1103
1104                 DBG("Kernel consumer lost packets command for session id %"
1105                                 PRIu64 ", channel key %" PRIu64, id, key);
1106
1107                 channel = consumer_find_channel(key);
1108                 if (!channel) {
1109                         ERR("Kernel consumer lost packets channel %"
1110                                         PRIu64 " not found", key);
1111                         count = 0;
1112                 } else {
1113                         count = channel->lost_packets;
1114                 }
1115
1116                 health_code_update();
1117
1118                 /* Send back returned value to session daemon */
1119                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1120                 if (ret < 0) {
1121                         PERROR("send lost packets");
1122                         goto error_fatal;
1123                 }
1124
1125                 break;
1126         }
1127         case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1128         {
1129                 int channel_monitor_pipe;
1130                 int ret_send_status, ret_set_channel_monitor_pipe;
1131                 ssize_t ret_recv;
1132
1133                 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1134                 /* Successfully received the command's type. */
1135                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1136                 if (ret_send_status < 0) {
1137                         goto error_fatal;
1138                 }
1139
1140                 ret_recv = lttcomm_recv_fds_unix_sock(
1141                                 sock, &channel_monitor_pipe, 1);
1142                 if (ret_recv != sizeof(channel_monitor_pipe)) {
1143                         ERR("Failed to receive channel monitor pipe");
1144                         goto error_fatal;
1145                 }
1146
1147                 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1148                 ret_set_channel_monitor_pipe =
1149                                 consumer_timer_thread_set_channel_monitor_pipe(
1150                                                 channel_monitor_pipe);
1151                 if (!ret_set_channel_monitor_pipe) {
1152                         int flags;
1153                         int ret_fcntl;
1154
1155                         ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1156                         /* Set the pipe as non-blocking. */
1157                         ret_fcntl = fcntl(channel_monitor_pipe, F_GETFL, 0);
1158                         if (ret_fcntl == -1) {
1159                                 PERROR("fcntl get flags of the channel monitoring pipe");
1160                                 goto error_fatal;
1161                         }
1162                         flags = ret_fcntl;
1163
1164                         ret_fcntl = fcntl(channel_monitor_pipe, F_SETFL,
1165                                         flags | O_NONBLOCK);
1166                         if (ret_fcntl == -1) {
1167                                 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1168                                 goto error_fatal;
1169                         }
1170                         DBG("Channel monitor pipe set as non-blocking");
1171                 } else {
1172                         ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1173                 }
1174                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1175                 if (ret_send_status < 0) {
1176                         goto error_fatal;
1177                 }
1178                 break;
1179         }
1180         case LTTNG_CONSUMER_ROTATE_CHANNEL:
1181         {
1182                 struct lttng_consumer_channel *channel;
1183                 uint64_t key = msg.u.rotate_channel.key;
1184                 int ret_send_status;
1185
1186                 DBG("Consumer rotate channel %" PRIu64, key);
1187
1188                 channel = consumer_find_channel(key);
1189                 if (!channel) {
1190                         ERR("Channel %" PRIu64 " not found", key);
1191                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1192                 } else {
1193                         /*
1194                          * Sample the rotate position of all the streams in this channel.
1195                          */
1196                         int ret_rotate_channel;
1197
1198                         ret_rotate_channel = lttng_consumer_rotate_channel(
1199                                         channel, key,
1200                                         msg.u.rotate_channel.relayd_id,
1201                                         msg.u.rotate_channel.metadata, ctx);
1202                         if (ret_rotate_channel < 0) {
1203                                 ERR("Rotate channel failed");
1204                                 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1205                         }
1206
1207                         health_code_update();
1208                 }
1209
1210                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1211                 if (ret_send_status < 0) {
1212                         /* Somehow, the session daemon is not responding anymore. */
1213                         goto error_rotate_channel;
1214                 }
1215                 if (channel) {
1216                         /* Rotate the streams that are ready right now. */
1217                         int ret_rotate;
1218
1219                         ret_rotate = lttng_consumer_rotate_ready_streams(
1220                                         channel, key, ctx);
1221                         if (ret_rotate < 0) {
1222                                 ERR("Rotate ready streams failed");
1223                         }
1224                 }
1225                 break;
1226 error_rotate_channel:
1227                 goto end_nosignal;
1228         }
1229         case LTTNG_CONSUMER_CLEAR_CHANNEL:
1230         {
1231                 struct lttng_consumer_channel *channel;
1232                 uint64_t key = msg.u.clear_channel.key;
1233                 int ret_send_status;
1234
1235                 channel = consumer_find_channel(key);
1236                 if (!channel) {
1237                         DBG("Channel %" PRIu64 " not found", key);
1238                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1239                 } else {
1240                         int ret_clear_channel;
1241
1242                         ret_clear_channel =
1243                                         lttng_consumer_clear_channel(channel);
1244                         if (ret_clear_channel) {
1245                                 ERR("Clear channel failed");
1246                                 ret_code = ret_clear_channel;
1247                         }
1248
1249                         health_code_update();
1250                 }
1251
1252                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1253                 if (ret_send_status < 0) {
1254                         /* Somehow, the session daemon is not responding anymore. */
1255                         goto end_nosignal;
1256                 }
1257
1258                 break;
1259         }
1260         case LTTNG_CONSUMER_INIT:
1261         {
1262                 int ret_send_status;
1263
1264                 ret_code = lttng_consumer_init_command(ctx,
1265                                 msg.u.init.sessiond_uuid);
1266                 health_code_update();
1267                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1268                 if (ret_send_status < 0) {
1269                         /* Somehow, the session daemon is not responding anymore. */
1270                         goto end_nosignal;
1271                 }
1272                 break;
1273         }
1274         case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
1275         {
1276                 const struct lttng_credentials credentials = {
1277                         .uid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.uid),
1278                         .gid = LTTNG_OPTIONAL_INIT_VALUE(msg.u.create_trace_chunk.credentials.value.gid),
1279                 };
1280                 const bool is_local_trace =
1281                                 !msg.u.create_trace_chunk.relayd_id.is_set;
1282                 const uint64_t relayd_id =
1283                                 msg.u.create_trace_chunk.relayd_id.value;
1284                 const char *chunk_override_name =
1285                                 *msg.u.create_trace_chunk.override_name ?
1286                                         msg.u.create_trace_chunk.override_name :
1287                                         NULL;
1288                 struct lttng_directory_handle *chunk_directory_handle = NULL;
1289
1290                 /*
1291                  * The session daemon will only provide a chunk directory file
1292                  * descriptor for local traces.
1293                  */
1294                 if (is_local_trace) {
1295                         int chunk_dirfd;
1296                         int ret_send_status;
1297                         ssize_t ret_recv;
1298
1299                         /* Acnowledge the reception of the command. */
1300                         ret_send_status = consumer_send_status_msg(
1301                                         sock, LTTCOMM_CONSUMERD_SUCCESS);
1302                         if (ret_send_status < 0) {
1303                                 /* Somehow, the session daemon is not responding anymore. */
1304                                 goto end_nosignal;
1305                         }
1306
1307                         ret_recv = lttcomm_recv_fds_unix_sock(
1308                                         sock, &chunk_dirfd, 1);
1309                         if (ret_recv != sizeof(chunk_dirfd)) {
1310                                 ERR("Failed to receive trace chunk directory file descriptor");
1311                                 goto error_fatal;
1312                         }
1313
1314                         DBG("Received trace chunk directory fd (%d)",
1315                                         chunk_dirfd);
1316                         chunk_directory_handle = lttng_directory_handle_create_from_dirfd(
1317                                         chunk_dirfd);
1318                         if (!chunk_directory_handle) {
1319                                 ERR("Failed to initialize chunk directory handle from directory file descriptor");
1320                                 if (close(chunk_dirfd)) {
1321                                         PERROR("Failed to close chunk directory file descriptor");
1322                                 }
1323                                 goto error_fatal;
1324                         }
1325                 }
1326
1327                 ret_code = lttng_consumer_create_trace_chunk(
1328                                 !is_local_trace ? &relayd_id : NULL,
1329                                 msg.u.create_trace_chunk.session_id,
1330                                 msg.u.create_trace_chunk.chunk_id,
1331                                 (time_t) msg.u.create_trace_chunk
1332                                                 .creation_timestamp,
1333                                 chunk_override_name,
1334                                 msg.u.create_trace_chunk.credentials.is_set ?
1335                                                 &credentials :
1336                                                 NULL,
1337                                 chunk_directory_handle);
1338                 lttng_directory_handle_put(chunk_directory_handle);
1339                 goto end_msg_sessiond;
1340         }
1341         case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
1342         {
1343                 enum lttng_trace_chunk_command_type close_command =
1344                                 msg.u.close_trace_chunk.close_command.value;
1345                 const uint64_t relayd_id =
1346                                 msg.u.close_trace_chunk.relayd_id.value;
1347                 struct lttcomm_consumer_close_trace_chunk_reply reply;
1348                 char path[LTTNG_PATH_MAX];
1349                 ssize_t ret_send;
1350
1351                 ret_code = lttng_consumer_close_trace_chunk(
1352                                 msg.u.close_trace_chunk.relayd_id.is_set ?
1353                                                 &relayd_id :
1354                                                 NULL,
1355                                 msg.u.close_trace_chunk.session_id,
1356                                 msg.u.close_trace_chunk.chunk_id,
1357                                 (time_t) msg.u.close_trace_chunk.close_timestamp,
1358                                 msg.u.close_trace_chunk.close_command.is_set ?
1359                                                 &close_command :
1360                                                 NULL, path);
1361                 reply.ret_code = ret_code;
1362                 reply.path_length = strlen(path) + 1;
1363                 ret_send = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
1364                 if (ret_send != sizeof(reply)) {
1365                         goto error_fatal;
1366                 }
1367                 ret_send = lttcomm_send_unix_sock(
1368                                 sock, path, reply.path_length);
1369                 if (ret_send != reply.path_length) {
1370                         goto error_fatal;
1371                 }
1372                 goto end_nosignal;
1373         }
1374         case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
1375         {
1376                 const uint64_t relayd_id =
1377                                 msg.u.trace_chunk_exists.relayd_id.value;
1378
1379                 ret_code = lttng_consumer_trace_chunk_exists(
1380                                 msg.u.trace_chunk_exists.relayd_id.is_set ?
1381                                                 &relayd_id : NULL,
1382                                 msg.u.trace_chunk_exists.session_id,
1383                                 msg.u.trace_chunk_exists.chunk_id);
1384                 goto end_msg_sessiond;
1385         }
1386         case LTTNG_CONSUMER_OPEN_CHANNEL_PACKETS:
1387         {
1388                 const uint64_t key = msg.u.open_channel_packets.key;
1389                 struct lttng_consumer_channel *channel =
1390                                 consumer_find_channel(key);
1391
1392                 if (channel) {
1393                         pthread_mutex_lock(&channel->lock);
1394                         ret_code = lttng_consumer_open_channel_packets(channel);
1395                         pthread_mutex_unlock(&channel->lock);
1396                 } else {
1397                         WARN("Channel %" PRIu64 " not found", key);
1398                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1399                 }
1400
1401                 health_code_update();
1402                 goto end_msg_sessiond;
1403         }
1404         default:
1405                 goto end_nosignal;
1406         }
1407
1408 end_nosignal:
1409         /*
1410          * Return 1 to indicate success since the 0 value can be a socket
1411          * shutdown during the recv() or send() call.
1412          */
1413         ret_func = 1;
1414         goto end;
1415 error_fatal:
1416         /* This will issue a consumer stop. */
1417         ret_func = -1;
1418         goto end;
1419 end_msg_sessiond:
1420         /*
1421          * The returned value here is not useful since either way we'll return 1 to
1422          * the caller because the session daemon socket management is done
1423          * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1424          */
1425         {
1426                 int ret_send_status;
1427
1428                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1429                 if (ret_send_status < 0) {
1430                         goto error_fatal;
1431                 }
1432         }
1433
1434         ret_func = 1;
1435
1436 end:
1437         health_code_update();
1438         rcu_read_unlock();
1439         return ret_func;
1440 }
1441
1442 /*
1443  * Sync metadata meaning request them to the session daemon and snapshot to the
1444  * metadata thread can consumer them.
1445  *
1446  * Metadata stream lock MUST be acquired.
1447  */
1448 enum sync_metadata_status lttng_kconsumer_sync_metadata(
1449                 struct lttng_consumer_stream *metadata)
1450 {
1451         int ret;
1452         enum sync_metadata_status status;
1453
1454         assert(metadata);
1455
1456         ret = kernctl_buffer_flush(metadata->wait_fd);
1457         if (ret < 0) {
1458                 ERR("Failed to flush kernel stream");
1459                 status = SYNC_METADATA_STATUS_ERROR;
1460                 goto end;
1461         }
1462
1463         ret = kernctl_snapshot(metadata->wait_fd);
1464         if (ret < 0) {
1465                 if (errno == EAGAIN) {
1466                         /* No new metadata, exit. */
1467                         DBG("Sync metadata, no new kernel metadata");
1468                         status = SYNC_METADATA_STATUS_NO_DATA;
1469                 } else {
1470                         ERR("Sync metadata, taking kernel snapshot failed.");
1471                         status = SYNC_METADATA_STATUS_ERROR;
1472                 }
1473         } else {
1474                 status = SYNC_METADATA_STATUS_NEW_DATA;
1475         }
1476
1477 end:
1478         return status;
1479 }
1480
1481 static
1482 int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
1483                 struct stream_subbuffer *subbuf)
1484 {
1485         int ret;
1486
1487         ret = kernctl_get_subbuf_size(
1488                         stream->wait_fd, &subbuf->info.data.subbuf_size);
1489         if (ret) {
1490                 goto end;
1491         }
1492
1493         ret = kernctl_get_padded_subbuf_size(
1494                         stream->wait_fd, &subbuf->info.data.padded_subbuf_size);
1495         if (ret) {
1496                 goto end;
1497         }
1498
1499 end:
1500         return ret;
1501 }
1502
1503 static
1504 int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
1505                 struct stream_subbuffer *subbuf)
1506 {
1507         int ret;
1508
1509         ret = extract_common_subbuffer_info(stream, subbuf);
1510         if (ret) {
1511                 goto end;
1512         }
1513
1514         ret = kernctl_get_metadata_version(
1515                         stream->wait_fd, &subbuf->info.metadata.version);
1516         if (ret) {
1517                 goto end;
1518         }
1519
1520 end:
1521         return ret;
1522 }
1523
1524 static
1525 int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
1526                 struct stream_subbuffer *subbuf)
1527 {
1528         int ret;
1529
1530         ret = extract_common_subbuffer_info(stream, subbuf);
1531         if (ret) {
1532                 goto end;
1533         }
1534
1535         ret = kernctl_get_packet_size(
1536                         stream->wait_fd, &subbuf->info.data.packet_size);
1537         if (ret < 0) {
1538                 PERROR("Failed to get sub-buffer packet size");
1539                 goto end;
1540         }
1541
1542         ret = kernctl_get_content_size(
1543                         stream->wait_fd, &subbuf->info.data.content_size);
1544         if (ret < 0) {
1545                 PERROR("Failed to get sub-buffer content size");
1546                 goto end;
1547         }
1548
1549         ret = kernctl_get_timestamp_begin(
1550                         stream->wait_fd, &subbuf->info.data.timestamp_begin);
1551         if (ret < 0) {
1552                 PERROR("Failed to get sub-buffer begin timestamp");
1553                 goto end;
1554         }
1555
1556         ret = kernctl_get_timestamp_end(
1557                         stream->wait_fd, &subbuf->info.data.timestamp_end);
1558         if (ret < 0) {
1559                 PERROR("Failed to get sub-buffer end timestamp");
1560                 goto end;
1561         }
1562
1563         ret = kernctl_get_events_discarded(
1564                         stream->wait_fd, &subbuf->info.data.events_discarded);
1565         if (ret) {
1566                 PERROR("Failed to get sub-buffer events discarded count");
1567                 goto end;
1568         }
1569
1570         ret = kernctl_get_sequence_number(stream->wait_fd,
1571                         &subbuf->info.data.sequence_number.value);
1572         if (ret) {
1573                 /* May not be supported by older LTTng-modules. */
1574                 if (ret != -ENOTTY) {
1575                         PERROR("Failed to get sub-buffer sequence number");
1576                         goto end;
1577                 }
1578         } else {
1579                 subbuf->info.data.sequence_number.is_set = true;
1580         }
1581
1582         ret = kernctl_get_stream_id(
1583                         stream->wait_fd, &subbuf->info.data.stream_id);
1584         if (ret < 0) {
1585                 PERROR("Failed to get stream id");
1586                 goto end;
1587         }
1588
1589         ret = kernctl_get_instance_id(stream->wait_fd,
1590                         &subbuf->info.data.stream_instance_id.value);
1591         if (ret) {
1592                 /* May not be supported by older LTTng-modules. */
1593                 if (ret != -ENOTTY) {
1594                         PERROR("Failed to get stream instance id");
1595                         goto end;
1596                 }
1597         } else {
1598                 subbuf->info.data.stream_instance_id.is_set = true;
1599         }
1600 end:
1601         return ret;
1602 }
1603
1604 static
1605 enum get_next_subbuffer_status get_subbuffer_common(
1606                 struct lttng_consumer_stream *stream,
1607                 struct stream_subbuffer *subbuffer)
1608 {
1609         int ret;
1610         enum get_next_subbuffer_status status;
1611
1612         ret = kernctl_get_next_subbuf(stream->wait_fd);
1613         switch (ret) {
1614         case 0:
1615                 status = GET_NEXT_SUBBUFFER_STATUS_OK;
1616                 break;
1617         case -ENODATA:
1618                 case -EAGAIN:
1619                 /*
1620                  * The caller only expects -ENODATA when there is no data to
1621                  * read, but the kernel tracer returns -EAGAIN when there is
1622                  * currently no data for a non-finalized stream, and -ENODATA
1623                  * when there is no data for a finalized stream. Those can be
1624                  * combined into a -ENODATA return value.
1625                  */
1626                 status = GET_NEXT_SUBBUFFER_STATUS_NO_DATA;
1627                 goto end;
1628         default:
1629                 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1630                 goto end;
1631         }
1632
1633         ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1634                 stream, subbuffer);
1635         if (ret) {
1636                 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1637         }
1638 end:
1639         return status;
1640 }
1641
1642 static
1643 enum get_next_subbuffer_status get_next_subbuffer_splice(
1644                 struct lttng_consumer_stream *stream,
1645                 struct stream_subbuffer *subbuffer)
1646 {
1647         const enum get_next_subbuffer_status status =
1648                         get_subbuffer_common(stream, subbuffer);
1649
1650         if (status != GET_NEXT_SUBBUFFER_STATUS_OK) {
1651                 goto end;
1652         }
1653
1654         subbuffer->buffer.fd = stream->wait_fd;
1655 end:
1656         return status;
1657 }
1658
1659 static
1660 enum get_next_subbuffer_status get_next_subbuffer_mmap(
1661                 struct lttng_consumer_stream *stream,
1662                 struct stream_subbuffer *subbuffer)
1663 {
1664         int ret;
1665         enum get_next_subbuffer_status status;
1666         const char *addr;
1667
1668         status = get_subbuffer_common(stream, subbuffer);
1669         if (status != GET_NEXT_SUBBUFFER_STATUS_OK) {
1670                 goto end;
1671         }
1672
1673         ret = get_current_subbuf_addr(stream, &addr);
1674         if (ret) {
1675                 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1676                 goto end;
1677         }
1678
1679         subbuffer->buffer.buffer = lttng_buffer_view_init(
1680                         addr, 0, subbuffer->info.data.padded_subbuf_size);
1681 end:
1682         return status;
1683 }
1684
1685 static
1686 enum get_next_subbuffer_status get_next_subbuffer_metadata_check(struct lttng_consumer_stream *stream,
1687                 struct stream_subbuffer *subbuffer)
1688 {
1689         int ret;
1690         const char *addr;
1691         bool coherent;
1692         enum get_next_subbuffer_status status;
1693
1694         ret = kernctl_get_next_subbuf_metadata_check(stream->wait_fd,
1695                         &coherent);
1696         if (ret) {
1697                 goto end;
1698         }
1699
1700         ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1701                         stream, subbuffer);
1702         if (ret) {
1703                 goto end;
1704         }
1705
1706         LTTNG_OPTIONAL_SET(&subbuffer->info.metadata.coherent, coherent);
1707
1708         ret = get_current_subbuf_addr(stream, &addr);
1709         if (ret) {
1710                 goto end;
1711         }
1712
1713         subbuffer->buffer.buffer = lttng_buffer_view_init(
1714                         addr, 0, subbuffer->info.data.padded_subbuf_size);
1715         DBG("Got metadata packet with padded_subbuf_size = %lu, coherent = %s",
1716                         subbuffer->info.metadata.padded_subbuf_size,
1717                         coherent ? "true" : "false");
1718 end:
1719         /*
1720          * The caller only expects -ENODATA when there is no data to read, but
1721          * the kernel tracer returns -EAGAIN when there is currently no data
1722          * for a non-finalized stream, and -ENODATA when there is no data for a
1723          * finalized stream. Those can be combined into a -ENODATA return value.
1724          */
1725         switch (ret) {
1726         case 0:
1727                 status = GET_NEXT_SUBBUFFER_STATUS_OK;
1728                 break;
1729         case -ENODATA:
1730         case -EAGAIN:
1731                 /*
1732                  * The caller only expects -ENODATA when there is no data to
1733                  * read, but the kernel tracer returns -EAGAIN when there is
1734                  * currently no data for a non-finalized stream, and -ENODATA
1735                  * when there is no data for a finalized stream. Those can be
1736                  * combined into a -ENODATA return value.
1737                  */
1738                 status = GET_NEXT_SUBBUFFER_STATUS_NO_DATA;
1739                 break;
1740         default:
1741                 status = GET_NEXT_SUBBUFFER_STATUS_ERROR;
1742                 break;
1743         }
1744
1745         return status;
1746 }
1747
1748 static
1749 int put_next_subbuffer(struct lttng_consumer_stream *stream,
1750                 struct stream_subbuffer *subbuffer)
1751 {
1752         const int ret = kernctl_put_next_subbuf(stream->wait_fd);
1753
1754         if (ret) {
1755                 if (ret == -EFAULT) {
1756                         PERROR("Error in unreserving sub buffer");
1757                 } else if (ret == -EIO) {
1758                         /* Should never happen with newer LTTng versions */
1759                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted");
1760                 }
1761         }
1762
1763         return ret;
1764 }
1765
1766 static
1767 bool is_get_next_check_metadata_available(int tracer_fd)
1768 {
1769         const int ret = kernctl_get_next_subbuf_metadata_check(tracer_fd, NULL);
1770         const bool available = ret != -ENOTTY;
1771
1772         if (ret == 0) {
1773                 /* get succeeded, make sure to put the subbuffer. */
1774                 kernctl_put_subbuf(tracer_fd);
1775         }
1776
1777         return available;
1778 }
1779
1780 static
1781 int signal_metadata(struct lttng_consumer_stream *stream,
1782                 struct lttng_consumer_local_data *ctx)
1783 {
1784         ASSERT_LOCKED(stream->metadata_rdv_lock);
1785         return pthread_cond_broadcast(&stream->metadata_rdv) ? -errno : 0;
1786 }
1787
1788 static
1789 int lttng_kconsumer_set_stream_ops(
1790                 struct lttng_consumer_stream *stream)
1791 {
1792         int ret = 0;
1793
1794         if (stream->metadata_flag && stream->chan->is_live) {
1795                 DBG("Attempting to enable metadata bucketization for live consumers");
1796                 if (is_get_next_check_metadata_available(stream->wait_fd)) {
1797                         DBG("Kernel tracer supports get_next_subbuffer_metadata_check, metadata will be accumulated until a coherent state is reached");
1798                         stream->read_subbuffer_ops.get_next_subbuffer =
1799                                         get_next_subbuffer_metadata_check;
1800                         ret = consumer_stream_enable_metadata_bucketization(
1801                                         stream);
1802                         if (ret) {
1803                                 goto end;
1804                         }
1805                 } else {
1806                         /*
1807                          * The kernel tracer version is too old to indicate
1808                          * when the metadata stream has reached a "coherent"
1809                          * (parseable) point.
1810                          *
1811                          * This means that a live viewer may see an incoherent
1812                          * sequence of metadata and fail to parse it.
1813                          */
1814                         WARN("Kernel tracer does not support get_next_subbuffer_metadata_check which may cause live clients to fail to parse the metadata stream");
1815                         metadata_bucket_destroy(stream->metadata_bucket);
1816                         stream->metadata_bucket = NULL;
1817                 }
1818
1819                 stream->read_subbuffer_ops.on_sleep = signal_metadata;
1820         }
1821
1822         if (!stream->read_subbuffer_ops.get_next_subbuffer) {
1823                 if (stream->chan->output == CONSUMER_CHANNEL_MMAP) {
1824                         stream->read_subbuffer_ops.get_next_subbuffer =
1825                                         get_next_subbuffer_mmap;
1826                 } else {
1827                         stream->read_subbuffer_ops.get_next_subbuffer =
1828                                         get_next_subbuffer_splice;
1829                 }
1830         }
1831
1832         if (stream->metadata_flag) {
1833                 stream->read_subbuffer_ops.extract_subbuffer_info =
1834                                 extract_metadata_subbuffer_info;
1835         } else {
1836                 stream->read_subbuffer_ops.extract_subbuffer_info =
1837                                 extract_data_subbuffer_info;
1838                 if (stream->chan->is_live) {
1839                         stream->read_subbuffer_ops.send_live_beacon =
1840                                         consumer_flush_kernel_index;
1841                 }
1842         }
1843
1844         stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
1845 end:
1846         return ret;
1847 }
1848
1849 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1850 {
1851         int ret;
1852
1853         assert(stream);
1854
1855         /*
1856          * Don't create anything if this is set for streaming or if there is
1857          * no current trace chunk on the parent channel.
1858          */
1859         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
1860                         stream->chan->trace_chunk) {
1861                 ret = consumer_stream_create_output_files(stream, true);
1862                 if (ret) {
1863                         goto error;
1864                 }
1865         }
1866
1867         if (stream->output == LTTNG_EVENT_MMAP) {
1868                 /* get the len of the mmap region */
1869                 unsigned long mmap_len;
1870
1871                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1872                 if (ret != 0) {
1873                         PERROR("kernctl_get_mmap_len");
1874                         goto error_close_fd;
1875                 }
1876                 stream->mmap_len = (size_t) mmap_len;
1877
1878                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1879                                 MAP_PRIVATE, stream->wait_fd, 0);
1880                 if (stream->mmap_base == MAP_FAILED) {
1881                         PERROR("Error mmaping");
1882                         ret = -1;
1883                         goto error_close_fd;
1884                 }
1885         }
1886
1887         ret = lttng_kconsumer_set_stream_ops(stream);
1888         if (ret) {
1889                 goto error_close_fd;
1890         }
1891
1892         /* we return 0 to let the library handle the FD internally */
1893         return 0;
1894
1895 error_close_fd:
1896         if (stream->out_fd >= 0) {
1897                 int err;
1898
1899                 err = close(stream->out_fd);
1900                 assert(!err);
1901                 stream->out_fd = -1;
1902         }
1903 error:
1904         return ret;
1905 }
1906
1907 /*
1908  * Check if data is still being extracted from the buffers for a specific
1909  * stream. Consumer data lock MUST be acquired before calling this function
1910  * and the stream lock.
1911  *
1912  * Return 1 if the traced data are still getting read else 0 meaning that the
1913  * data is available for trace viewer reading.
1914  */
1915 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1916 {
1917         int ret;
1918
1919         assert(stream);
1920
1921         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1922                 ret = 0;
1923                 goto end;
1924         }
1925
1926         ret = kernctl_get_next_subbuf(stream->wait_fd);
1927         if (ret == 0) {
1928                 /* There is still data so let's put back this subbuffer. */
1929                 ret = kernctl_put_subbuf(stream->wait_fd);
1930                 assert(ret == 0);
1931                 ret = 1;   /* Data is pending */
1932                 goto end;
1933         }
1934
1935         /* Data is NOT pending and ready to be read. */
1936         ret = 0;
1937
1938 end:
1939         return ret;
1940 }