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