src/common/kernel-consumer/kernel-consumer.cpp

   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 <poll.h>
  12 #include <pthread.h>
  13 #include <stdlib.h>
  14 #include <string.h>
  15 #include <sys/mman.h>
  16 #include <sys/socket.h>
  17 #include <sys/types.h>
  18 #include <inttypes.h>
  19 #include <unistd.h>
  20 #include <sys/stat.h>
  21 #include <stdint.h>
  22
  23 #include <bin/lttng-consumerd/health-consumerd.hpp>
  24 #include <common/common.hpp>
  25 #include <common/kernel-ctl/kernel-ctl.hpp>
  26 #include <common/sessiond-comm/sessiond-comm.hpp>
  27 #include <common/sessiond-comm/relayd.hpp>
  28 #include <common/compat/fcntl.hpp>
  29 #include <common/compat/endian.hpp>
  30 #include <common/pipe.hpp>
  31 #include <common/relayd/relayd.hpp>
  32 #include <common/utils.hpp>
  33 #include <common/consumer/consumer-stream.hpp>
  34 #include <common/index/index.hpp>
  35 #include <common/consumer/consumer-timer.hpp>
  36 #include <common/optional.hpp>
  37 #include <common/buffer-view.hpp>
  38 #include <common/consumer/consumer.hpp>
  39 #include <common/consumer/metadata-bucket.hpp>
  40
  41 #include "kernel-consumer.hpp"
  42
  43 extern struct lttng_consumer_global_data the_consumer_data;
  44 extern int consumer_poll_timeout;
  45
  46 /*
  47  * Take a snapshot for a specific fd
  48  *
  49  * Returns 0 on success, < 0 on error
  50  */
  51 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  52 {
  53         int ret = 0;
  54         int infd = stream->wait_fd;
  55
  56         ret = kernctl_snapshot(infd);
  57         /*
  58          * -EAGAIN is not an error, it just means that there is no data to
  59          *  be read.
  60          */
  61         if (ret != 0 && ret != -EAGAIN) {
  62                 PERROR("Getting sub-buffer snapshot.");
  63         }
  64
  65         return ret;
  66 }
  67
  68 /*
  69  * Sample consumed and produced positions for a specific fd.
  70  *
  71  * Returns 0 on success, < 0 on error.
  72  */
  73 int lttng_kconsumer_sample_snapshot_positions(
  74                 struct lttng_consumer_stream *stream)
  75 {
  76         LTTNG_ASSERT(stream);
  77
  78         return kernctl_snapshot_sample_positions(stream->wait_fd);
  79 }
  80
  81 /*
  82  * Get the produced position
  83  *
  84  * Returns 0 on success, < 0 on error
  85  */
  86 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  87                 unsigned long *pos)
  88 {
  89         int ret;
  90         int infd = stream->wait_fd;
  91
  92         ret = kernctl_snapshot_get_produced(infd, pos);
  93         if (ret != 0) {
  94                 PERROR("kernctl_snapshot_get_produced");
  95         }
  96
  97         return ret;
  98 }
  99
 100 /*
 101  * Get the consumerd position
 102  *
 103  * Returns 0 on success, < 0 on error
 104  */
 105 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 106                 unsigned long *pos)
 107 {
 108         int ret;
 109         int infd = stream->wait_fd;
 110
 111         ret = kernctl_snapshot_get_consumed(infd, pos);
 112         if (ret != 0) {
 113                 PERROR("kernctl_snapshot_get_consumed");
 114         }
 115
 116         return ret;
 117 }
 118
 119 static
 120 int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
 121                 const char **addr)
 122 {
 123         int ret;
 124         unsigned long mmap_offset;
 125         const char *mmap_base = (const char *) stream->mmap_base;
 126
 127         ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
 128         if (ret < 0) {
 129                 PERROR("Failed to get mmap read offset");
 130                 goto error;
 131         }
 132
 133         *addr = mmap_base + mmap_offset;
 134 error:
 135         return ret;
 136 }
 137
 138 static void finalize_snapshot_stream(
 139                 struct lttng_consumer_stream *stream, uint64_t relayd_id)
 140 {
 141         ASSERT_LOCKED(stream->lock);
 142
 143         if (relayd_id == (uint64_t) -1ULL) {
 144                 if (stream->out_fd >= 0) {
 145                         const int ret = close(stream->out_fd);
 146
 147                         if (ret < 0) {
 148                                 PERROR("Failed to close stream snapshot output file descriptor");
 149                         }
 150
 151                         stream->out_fd = -1;
 152                 }
 153         } else {
 154                 close_relayd_stream(stream);
 155                 stream->net_seq_idx = (uint64_t) -1ULL;
 156         }
 157
 158         lttng_trace_chunk_put(stream->trace_chunk);
 159         stream->trace_chunk = NULL;
 160 }
 161
 162 /*
 163  * Take a snapshot of all the stream of a channel
 164  * RCU read-side lock must be held across this function to ensure existence of
 165  * channel.
 166  *
 167  * Returns 0 on success, < 0 on error
 168  */
 169 static int lttng_kconsumer_snapshot_channel(
 170                 struct lttng_consumer_channel *channel,
 171                 uint64_t key, char *path, uint64_t relayd_id,
 172                 uint64_t nb_packets_per_stream)
 173 {
 174         int ret;
 175         struct lttng_consumer_stream *stream;
 176
 177         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 178
 179         /* Prevent channel modifications while we perform the snapshot.*/
 180         pthread_mutex_lock(&channel->lock);
 181
 182         rcu_read_lock();
 183
 184         /* Splice is not supported yet for channel snapshot. */
 185         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 186                 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
 187                                 channel->name);
 188                 ret = -1;
 189                 goto end;
 190         }
 191
 192         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 193                 unsigned long consumed_pos, produced_pos;
 194
 195                 health_code_update();
 196
 197                 /*
 198                  * Lock stream because we are about to change its state.
 199                  */
 200                 pthread_mutex_lock(&stream->lock);
 201
 202                 LTTNG_ASSERT(channel->trace_chunk);
 203                 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
 204                         /*
 205                          * Can't happen barring an internal error as the channel
 206                          * holds a reference to the trace chunk.
 207                          */
 208                         ERR("Failed to acquire reference to channel's trace chunk");
 209                         ret = -1;
 210                         goto end_unlock;
 211                 }
 212                 LTTNG_ASSERT(!stream->trace_chunk);
 213                 stream->trace_chunk = channel->trace_chunk;
 214
 215                 /*
 216                  * Assign the received relayd ID so we can use it for streaming. The streams
 217                  * are not visible to anyone so this is OK to change it.
 218                  */
 219                 stream->net_seq_idx = relayd_id;
 220                 channel->relayd_id = relayd_id;
 221                 if (relayd_id != (uint64_t) -1ULL) {
 222                         ret = consumer_send_relayd_stream(stream, path);
 223                         if (ret < 0) {
 224                                 ERR("sending stream to relayd");
 225                                 goto error_finalize_stream;
 226                         }
 227                 } else {
 228                         ret = consumer_stream_create_output_files(stream,
 229                                         false);
 230                         if (ret < 0) {
 231                                 goto error_finalize_stream;
 232                         }
 233                         DBG("Kernel consumer snapshot stream (%" PRIu64 ")",
 234                                         stream->key);
 235                 }
 236
 237                 ret = kernctl_buffer_flush_empty(stream->wait_fd);
 238                 if (ret < 0) {
 239                         /*
 240                          * Doing a buffer flush which does not take into
 241                          * account empty packets. This is not perfect
 242                          * for stream intersection, but required as a
 243                          * fall-back when "flush_empty" is not
 244                          * implemented by lttng-modules.
 245                          */
 246                         ret = kernctl_buffer_flush(stream->wait_fd);
 247                         if (ret < 0) {
 248                                 ERR("Failed to flush kernel stream");
 249                                 goto error_finalize_stream;
 250                         }
 251                         goto end_unlock;
 252                 }
 253
 254                 ret = lttng_kconsumer_take_snapshot(stream);
 255                 if (ret < 0) {
 256                         ERR("Taking kernel snapshot");
 257                         goto error_finalize_stream;
 258                 }
 259
 260                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 261                 if (ret < 0) {
 262                         ERR("Produced kernel snapshot position");
 263                         goto error_finalize_stream;
 264                 }
 265
 266                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 267                 if (ret < 0) {
 268                         ERR("Consumerd kernel snapshot position");
 269                         goto error_finalize_stream;
 270                 }
 271
 272                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 273                                 produced_pos, nb_packets_per_stream,
 274                                 stream->max_sb_size);
 275
 276                 while ((long) (consumed_pos - produced_pos) < 0) {
 277                         ssize_t read_len;
 278                         unsigned long len, padded_len;
 279                         const char *subbuf_addr;
 280                         struct lttng_buffer_view subbuf_view;
 281
 282                         health_code_update();
 283                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 284
 285                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 286                         if (ret < 0) {
 287                                 if (ret != -EAGAIN) {
 288                                         PERROR("kernctl_get_subbuf snapshot");
 289                                         goto error_finalize_stream;
 290                                 }
 291                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 292                                 consumed_pos += stream->max_sb_size;
 293                                 stream->chan->lost_packets++;
 294                                 continue;
 295                         }
 296
 297                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 298                         if (ret < 0) {
 299                                 ERR("Snapshot kernctl_get_subbuf_size");
 300                                 goto error_put_subbuf;
 301                         }
 302
 303                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 304                         if (ret < 0) {
 305                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 306                                 goto error_put_subbuf;
 307                         }
 308
 309                         ret = get_current_subbuf_addr(stream, &subbuf_addr);
 310                         if (ret) {
 311                                 goto error_put_subbuf;
 312                         }
 313
 314                         subbuf_view = lttng_buffer_view_init(
 315                                         subbuf_addr, 0, padded_len);
 316                         read_len = lttng_consumer_on_read_subbuffer_mmap(
 317                                         stream, &subbuf_view,
 318                                         padded_len - len);
 319                         /*
 320                          * We write the padded len in local tracefiles but the data len
 321                          * when using a relay. Display the error but continue processing
 322                          * to try to release the subbuffer.
 323                          */
 324                         if (relayd_id != (uint64_t) -1ULL) {
 325                                 if (read_len != len) {
 326                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 327                                                         read_len, len);
 328                                 }
 329                         } else {
 330                                 if (read_len != padded_len) {
 331                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 332                                                         read_len, padded_len);
 333                                 }
 334                         }
 335
 336                         ret = kernctl_put_subbuf(stream->wait_fd);
 337                         if (ret < 0) {
 338                                 ERR("Snapshot kernctl_put_subbuf");
 339                                 goto error_finalize_stream;
 340                         }
 341                         consumed_pos += stream->max_sb_size;
 342                 }
 343
 344                 finalize_snapshot_stream(stream, relayd_id);
 345                 pthread_mutex_unlock(&stream->lock);
 346         }
 347
 348         /* All good! */
 349         ret = 0;
 350         goto end;
 351
 352 error_put_subbuf:
 353         ret = kernctl_put_subbuf(stream->wait_fd);
 354         if (ret < 0) {
 355                 ERR("Snapshot kernctl_put_subbuf error path");
 356         }
 357 error_finalize_stream:
 358         finalize_snapshot_stream(stream, relayd_id);
 359 end_unlock:
 360         pthread_mutex_unlock(&stream->lock);
 361 end:
 362         rcu_read_unlock();
 363         pthread_mutex_unlock(&channel->lock);
 364         return ret;
 365 }
 366
 367 /*
 368  * Read the whole metadata available for a snapshot.
 369  * RCU read-side lock must be held across this function to ensure existence of
 370  * metadata_channel.
 371  *
 372  * Returns 0 on success, < 0 on error
 373  */
 374 static int lttng_kconsumer_snapshot_metadata(
 375                 struct lttng_consumer_channel *metadata_channel,
 376                 uint64_t key, char *path, uint64_t relayd_id,
 377                 struct lttng_consumer_local_data *ctx)
 378 {
 379         int ret, use_relayd = 0;
 380         ssize_t ret_read;
 381         struct lttng_consumer_stream *metadata_stream;
 382
 383         LTTNG_ASSERT(ctx);
 384
 385         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 386                         key, path);
 387
 388         rcu_read_lock();
 389
 390         metadata_stream = metadata_channel->metadata_stream;
 391         LTTNG_ASSERT(metadata_stream);
 392
 393         metadata_stream->read_subbuffer_ops.lock(metadata_stream);
 394         LTTNG_ASSERT(metadata_channel->trace_chunk);
 395         LTTNG_ASSERT(metadata_stream->trace_chunk);
 396
 397         /* Flag once that we have a valid relayd for the stream. */
 398         if (relayd_id != (uint64_t) -1ULL) {
 399                 use_relayd = 1;
 400         }
 401
 402         if (use_relayd) {
 403                 ret = consumer_send_relayd_stream(metadata_stream, path);
 404                 if (ret < 0) {
 405                         goto error_snapshot;
 406                 }
 407         } else {
 408                 ret = consumer_stream_create_output_files(metadata_stream,
 409                                 false);
 410                 if (ret < 0) {
 411                         goto error_snapshot;
 412                 }
 413         }
 414
 415         do {
 416                 health_code_update();
 417
 418                 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
 419                 if (ret_read < 0) {
 420                         ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 421                                         ret_read);
 422                         ret = ret_read;
 423                         goto error_snapshot;
 424                 }
 425         } while (ret_read > 0);
 426
 427         if (use_relayd) {
 428                 close_relayd_stream(metadata_stream);
 429                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 430         } else {
 431                 if (metadata_stream->out_fd >= 0) {
 432                         ret = close(metadata_stream->out_fd);
 433                         if (ret < 0) {
 434                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 435                                 /*
 436                                  * Don't go on error here since the snapshot was successful at this
 437                                  * point but somehow the close failed.
 438                                  */
 439                         }
 440                         metadata_stream->out_fd = -1;
 441                         lttng_trace_chunk_put(metadata_stream->trace_chunk);
 442                         metadata_stream->trace_chunk = NULL;
 443                 }
 444         }
 445
 446         ret = 0;
 447 error_snapshot:
 448         metadata_stream->read_subbuffer_ops.unlock(metadata_stream);
 449         consumer_stream_destroy(metadata_stream, NULL);
 450         metadata_channel->metadata_stream = NULL;
 451         rcu_read_unlock();
 452         return ret;
 453 }
 454
 455 /*
 456  * Receive command from session daemon and process it.
 457  *
 458  * Return 1 on success else a negative value or 0.
 459  */
 460 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 461                 int sock, struct pollfd *consumer_sockpoll)
 462 {
 463         int ret_func;
 464         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 465         struct lttcomm_consumer_msg msg;
 466
 467         health_code_update();
 468
 469         {
 470                 ssize_t ret_recv;
 471
 472                 ret_recv = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 473                 if (ret_recv != sizeof(msg)) {
 474                         if (ret_recv > 0) {
 475                                 lttng_consumer_send_error(ctx,
 476                                                 LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 477                                 ret_recv = -1;
 478                         }
 479                         return ret_recv;
 480                 }
 481         }
 482
 483         health_code_update();
 484
 485         /* Deprecated command */
 486         LTTNG_ASSERT(msg.cmd_type != LTTNG_CONSUMER_STOP);
 487
 488         health_code_update();
 489
 490         /* relayd needs RCU read-side protection */
 491         rcu_read_lock();
 492
 493         switch (msg.cmd_type) {
 494         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 495         {
 496                 uint32_t major = msg.u.relayd_sock.major;
 497                 uint32_t minor = msg.u.relayd_sock.minor;
 498                 enum lttcomm_sock_proto protocol = (enum lttcomm_sock_proto)
 499                                 msg.u.relayd_sock.relayd_socket_protocol;
 500
 501                 /* Session daemon status message are handled in the following call. */
 502                 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 503                                 msg.u.relayd_sock.type, ctx, sock,
 504                                 consumer_sockpoll, msg.u.relayd_sock.session_id,
 505                                 msg.u.relayd_sock.relayd_session_id, major,
 506                                 minor, protocol);
 507                 goto end_nosignal;
 508         }
 509         case LTTNG_CONSUMER_ADD_CHANNEL:
 510         {
 511                 struct lttng_consumer_channel *new_channel;
 512                 int ret_send_status, ret_add_channel = 0;
 513                 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
 514
 515                 health_code_update();
 516
 517                 /* First send a status message before receiving the fds. */
 518                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 519                 if (ret_send_status < 0) {
 520                         /* Somehow, the session daemon is not responding anymore. */
 521                         goto error_fatal;
 522                 }
 523
 524                 health_code_update();
 525
 526                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 527                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 528                                 msg.u.channel.session_id,
 529                                 msg.u.channel.chunk_id.is_set ?
 530                                                 &chunk_id : NULL,
 531                                 msg.u.channel.pathname,
 532                                 msg.u.channel.name,
 533                                 msg.u.channel.relayd_id, msg.u.channel.output,
 534                                 msg.u.channel.tracefile_size,
 535                                 msg.u.channel.tracefile_count, 0,
 536                                 msg.u.channel.monitor,
 537                                 msg.u.channel.live_timer_interval,
 538                                 msg.u.channel.is_live,
 539                                 NULL, NULL);
 540                 if (new_channel == NULL) {
 541                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 542                         goto end_nosignal;
 543                 }
 544                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 545                 switch (msg.u.channel.output) {
 546                 case LTTNG_EVENT_SPLICE:
 547                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 548                         break;
 549                 case LTTNG_EVENT_MMAP:
 550                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 551                         break;
 552                 default:
 553                         ERR("Channel output unknown %d", msg.u.channel.output);
 554                         goto end_nosignal;
 555                 }
 556
 557                 /* Translate and save channel type. */
 558                 switch (msg.u.channel.type) {
 559                 case CONSUMER_CHANNEL_TYPE_DATA:
 560                 case CONSUMER_CHANNEL_TYPE_METADATA:
 561                         new_channel->type = (consumer_channel_type) msg.u.channel.type;
 562                         break;
 563                 default:
 564                         abort();
 565                         goto end_nosignal;
 566                 };
 567
 568                 health_code_update();
 569
 570                 if (ctx->on_recv_channel != NULL) {
 571                         int ret_recv_channel =
 572                                         ctx->on_recv_channel(new_channel);
 573                         if (ret_recv_channel == 0) {
 574                                 ret_add_channel = consumer_add_channel(
 575                                                 new_channel, ctx);
 576                         } else if (ret_recv_channel < 0) {
 577                                 goto end_nosignal;
 578                         }
 579                 } else {
 580                         ret_add_channel =
 581                                         consumer_add_channel(new_channel, ctx);
 582                 }
 583                 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA &&
 584                                 !ret_add_channel) {
 585                         int monitor_start_ret;
 586
 587                         DBG("Consumer starting monitor timer");
 588                         consumer_timer_live_start(new_channel,
 589                                         msg.u.channel.live_timer_interval);
 590                         monitor_start_ret = consumer_timer_monitor_start(
 591                                         new_channel,
 592                                         msg.u.channel.monitor_timer_interval);
 593                         if (monitor_start_ret < 0) {
 594                                 ERR("Starting channel monitoring timer failed");
 595                                 goto end_nosignal;
 596                         }
 597                 }
 598
 599                 health_code_update();
 600
 601                 /* If we received an error in add_channel, we need to report it. */
 602                 if (ret_add_channel < 0) {
 603                         ret_send_status = consumer_send_status_msg(
 604                                         sock, ret_add_channel);
 605                         if (ret_send_status < 0) {
 606                                 goto error_fatal;
 607                         }
 608                         goto end_nosignal;
 609                 }
 610
 611                 goto end_nosignal;
 612         }
 613         case LTTNG_CONSUMER_ADD_STREAM:
 614         {
 615                 int fd;
 616                 struct lttng_pipe *stream_pipe;
 617                 struct lttng_consumer_stream *new_stream;
 618                 struct lttng_consumer_channel *channel;
 619                 int alloc_ret = 0;
 620                 int ret_send_status, ret_poll, ret_get_max_subbuf_size;
 621                 ssize_t ret_pipe_write, ret_recv;
 622
 623                 /*
 624                  * Get stream's channel reference. Needed when adding the stream to the
 625                  * global hash table.
 626                  */
 627                 channel = consumer_find_channel(msg.u.stream.channel_key);
 628                 if (!channel) {
 629                         /*
 630                          * We could not find the channel. Can happen if cpu hotplug
 631                          * happens while tearing down.
 632                          */
 633                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 634                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 635                 }
 636
 637                 health_code_update();
 638
 639                 /* First send a status message before receiving the fds. */
 640                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 641                 if (ret_send_status < 0) {
 642                         /* Somehow, the session daemon is not responding anymore. */
 643                         goto error_add_stream_fatal;
 644                 }
 645
 646                 health_code_update();
 647
 648                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 649                         /* Channel was not found. */
 650                         goto error_add_stream_nosignal;
 651                 }
 652
 653                 /* Blocking call */
 654                 health_poll_entry();
 655                 ret_poll = lttng_consumer_poll_socket(consumer_sockpoll);
 656                 health_poll_exit();
 657                 if (ret_poll) {
 658                         goto error_add_stream_fatal;
 659                 }
 660
 661                 health_code_update();
 662
 663                 /* Get stream file descriptor from socket */
 664                 ret_recv = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 665                 if (ret_recv != sizeof(fd)) {
 666                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 667                         ret_func = ret_recv;
 668                         goto end;
 669                 }
 670
 671                 health_code_update();
 672
 673                 /*
 674                  * Send status code to session daemon only if the recv works. If the
 675                  * above recv() failed, the session daemon is notified through the
 676                  * error socket and the teardown is eventually done.
 677                  */
 678                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 679                 if (ret_send_status < 0) {
 680                         /* Somehow, the session daemon is not responding anymore. */
 681                         goto error_add_stream_nosignal;
 682                 }
 683
 684                 health_code_update();
 685
 686                 pthread_mutex_lock(&channel->lock);
 687                 new_stream = consumer_stream_create(
 688                                 channel,
 689                                 channel->key,
 690                                 fd,
 691                                 channel->name,
 692                                 channel->relayd_id,
 693                                 channel->session_id,
 694                                 channel->trace_chunk,
 695                                 msg.u.stream.cpu,
 696                                 &alloc_ret,
 697                                 channel->type,
 698                                 channel->monitor);
 699                 if (new_stream == NULL) {
 700                         switch (alloc_ret) {
 701                         case -ENOMEM:
 702                         case -EINVAL:
 703                         default:
 704                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 705                                 break;
 706                         }
 707                         pthread_mutex_unlock(&channel->lock);
 708                         goto error_add_stream_nosignal;
 709                 }
 710
 711                 new_stream->wait_fd = fd;
 712                 ret_get_max_subbuf_size = kernctl_get_max_subbuf_size(
 713                                 new_stream->wait_fd, &new_stream->max_sb_size);
 714                 if (ret_get_max_subbuf_size < 0) {
 715                         pthread_mutex_unlock(&channel->lock);
 716                         ERR("Failed to get kernel maximal subbuffer size");
 717                         goto error_add_stream_nosignal;
 718                 }
 719
 720                 consumer_stream_update_channel_attributes(new_stream,
 721                                 channel);
 722
 723                 /*
 724                  * We've just assigned the channel to the stream so increment the
 725                  * refcount right now. We don't need to increment the refcount for
 726                  * streams in no monitor because we handle manually the cleanup of
 727                  * those. It is very important to make sure there is NO prior
 728                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 729                  */
 730                 if (channel->monitor) {
 731                         uatomic_inc(&new_stream->chan->refcount);
 732                 }
 733
 734                 /*
 735                  * The buffer flush is done on the session daemon side for the kernel
 736                  * so no need for the stream "hangup_flush_done" variable to be
 737                  * tracked. This is important for a kernel stream since we don't rely
 738                  * on the flush state of the stream to read data. It's not the case for
 739                  * user space tracing.
 740                  */
 741                 new_stream->hangup_flush_done = 0;
 742
 743                 health_code_update();
 744
 745                 pthread_mutex_lock(&new_stream->lock);
 746                 if (ctx->on_recv_stream) {
 747                         int ret_recv_stream = ctx->on_recv_stream(new_stream);
 748                         if (ret_recv_stream < 0) {
 749                                 pthread_mutex_unlock(&new_stream->lock);
 750                                 pthread_mutex_unlock(&channel->lock);
 751                                 consumer_stream_free(new_stream);
 752                                 goto error_add_stream_nosignal;
 753                         }
 754                 }
 755                 health_code_update();
 756
 757                 if (new_stream->metadata_flag) {
 758                         channel->metadata_stream = new_stream;
 759                 }
 760
 761                 /* Do not monitor this stream. */
 762                 if (!channel->monitor) {
 763                         DBG("Kernel consumer add stream %s in no monitor mode with "
 764                                         "relayd id %" PRIu64, new_stream->name,
 765                                         new_stream->net_seq_idx);
 766                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 767                         pthread_mutex_unlock(&new_stream->lock);
 768                         pthread_mutex_unlock(&channel->lock);
 769                         goto end_add_stream;
 770                 }
 771
 772                 /* Send stream to relayd if the stream has an ID. */
 773                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 774                         int ret_send_relayd_stream;
 775
 776                         ret_send_relayd_stream = consumer_send_relayd_stream(
 777                                         new_stream, new_stream->chan->pathname);
 778                         if (ret_send_relayd_stream < 0) {
 779                                 pthread_mutex_unlock(&new_stream->lock);
 780                                 pthread_mutex_unlock(&channel->lock);
 781                                 consumer_stream_free(new_stream);
 782                                 goto error_add_stream_nosignal;
 783                         }
 784
 785                         /*
 786                          * If adding an extra stream to an already
 787                          * existing channel (e.g. cpu hotplug), we need
 788                          * to send the "streams_sent" command to relayd.
 789                          */
 790                         if (channel->streams_sent_to_relayd) {
 791                                 int ret_send_relayd_streams_sent;
 792
 793                                 ret_send_relayd_streams_sent =
 794                                                 consumer_send_relayd_streams_sent(
 795                                                                 new_stream->net_seq_idx);
 796                                 if (ret_send_relayd_streams_sent < 0) {
 797                                         pthread_mutex_unlock(&new_stream->lock);
 798                                         pthread_mutex_unlock(&channel->lock);
 799                                         goto error_add_stream_nosignal;
 800                                 }
 801                         }
 802                 }
 803                 pthread_mutex_unlock(&new_stream->lock);
 804                 pthread_mutex_unlock(&channel->lock);
 805
 806                 /* Get the right pipe where the stream will be sent. */
 807                 if (new_stream->metadata_flag) {
 808                         consumer_add_metadata_stream(new_stream);
 809                         stream_pipe = ctx->consumer_metadata_pipe;
 810                 } else {
 811                         consumer_add_data_stream(new_stream);
 812                         stream_pipe = ctx->consumer_data_pipe;
 813                 }
 814
 815                 /* Visible to other threads */
 816                 new_stream->globally_visible = 1;
 817
 818                 health_code_update();
 819
 820                 ret_pipe_write = lttng_pipe_write(
 821                                 stream_pipe, &new_stream, sizeof(new_stream));
 822                 if (ret_pipe_write < 0) {
 823                         ERR("Consumer write %s stream to pipe %d",
 824                                         new_stream->metadata_flag ? "metadata" : "data",
 825                                         lttng_pipe_get_writefd(stream_pipe));
 826                         if (new_stream->metadata_flag) {
 827                                 consumer_del_stream_for_metadata(new_stream);
 828                         } else {
 829                                 consumer_del_stream_for_data(new_stream);
 830                         }
 831                         goto error_add_stream_nosignal;
 832                 }
 833
 834                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
 835                                 new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
 836 end_add_stream:
 837                 break;
 838 error_add_stream_nosignal:
 839                 goto end_nosignal;
 840 error_add_stream_fatal:
 841                 goto error_fatal;
 842         }
 843         case LTTNG_CONSUMER_STREAMS_SENT:
 844         {
 845                 struct lttng_consumer_channel *channel;
 846                 int ret_send_status;
 847
 848                 /*
 849                  * Get stream's channel reference. Needed when adding the stream to the
 850                  * global hash table.
 851                  */
 852                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 853                 if (!channel) {
 854                         /*
 855                          * We could not find the channel. Can happen if cpu hotplug
 856                          * happens while tearing down.
 857                          */
 858                         ERR("Unable to find channel key %" PRIu64,
 859                                         msg.u.sent_streams.channel_key);
 860                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 861                 }
 862
 863                 health_code_update();
 864
 865                 /*
 866                  * Send status code to session daemon.
 867                  */
 868                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 869                 if (ret_send_status < 0 ||
 870                                 ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 871                         /* Somehow, the session daemon is not responding anymore. */
 872                         goto error_streams_sent_nosignal;
 873                 }
 874
 875                 health_code_update();
 876
 877                 /*
 878                  * We should not send this message if we don't monitor the
 879                  * streams in this channel.
 880                  */
 881                 if (!channel->monitor) {
 882                         goto end_error_streams_sent;
 883                 }
 884
 885                 health_code_update();
 886                 /* Send stream to relayd if the stream has an ID. */
 887                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 888                         int ret_send_relay_streams;
 889
 890                         ret_send_relay_streams = consumer_send_relayd_streams_sent(
 891                                         msg.u.sent_streams.net_seq_idx);
 892                         if (ret_send_relay_streams < 0) {
 893                                 goto error_streams_sent_nosignal;
 894                         }
 895                         channel->streams_sent_to_relayd = true;
 896                 }
 897 end_error_streams_sent:
 898                 break;
 899 error_streams_sent_nosignal:
 900                 goto end_nosignal;
 901         }
 902         case LTTNG_CONSUMER_UPDATE_STREAM:
 903         {
 904                 rcu_read_unlock();
 905                 return -ENOSYS;
 906         }
 907         case LTTNG_CONSUMER_DESTROY_RELAYD:
 908         {
 909                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 910                 struct consumer_relayd_sock_pair *relayd;
 911                 int ret_send_status;
 912
 913                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 914
 915                 /* Get relayd reference if exists. */
 916                 relayd = consumer_find_relayd(index);
 917                 if (relayd == NULL) {
 918                         DBG("Unable to find relayd %" PRIu64, index);
 919                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 920                 }
 921
 922                 /*
 923                  * Each relayd socket pair has a refcount of stream attached to it
 924                  * which tells if the relayd is still active or not depending on the
 925                  * refcount value.
 926                  *
 927                  * This will set the destroy flag of the relayd object and destroy it
 928                  * if the refcount reaches zero when called.
 929                  *
 930                  * The destroy can happen either here or when a stream fd hangs up.
 931                  */
 932                 if (relayd) {
 933                         consumer_flag_relayd_for_destroy(relayd);
 934                 }
 935
 936                 health_code_update();
 937
 938                 ret_send_status = consumer_send_status_msg(sock, ret_code);
 939                 if (ret_send_status < 0) {
 940                         /* Somehow, the session daemon is not responding anymore. */
 941                         goto error_fatal;
 942                 }
 943
 944                 goto end_nosignal;
 945         }
 946         case LTTNG_CONSUMER_DATA_PENDING:
 947         {
 948                 int32_t ret_data_pending;
 949                 uint64_t id = msg.u.data_pending.session_id;
 950                 ssize_t ret_send;
 951
 952                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 953
 954                 ret_data_pending = consumer_data_pending(id);
 955
 956                 health_code_update();
 957
 958                 /* Send back returned value to session daemon */
 959                 ret_send = lttcomm_send_unix_sock(sock, &ret_data_pending,
 960                                 sizeof(ret_data_pending));
 961                 if (ret_send < 0) {
 962                         PERROR("send data pending ret code");
 963                         goto error_fatal;
 964                 }
 965
 966                 /*
 967                  * No need to send back a status message since the data pending
 968                  * returned value is the response.
 969                  */
 970                 break;
 971         }
 972         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 973         {
 974                 struct lttng_consumer_channel *channel;
 975                 uint64_t key = msg.u.snapshot_channel.key;
 976                 int ret_send_status;
 977
 978                 channel = consumer_find_channel(key);
 979                 if (!channel) {
 980                         ERR("Channel %" PRIu64 " not found", key);
 981                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 982                 } else {
 983                         if (msg.u.snapshot_channel.metadata == 1) {
 984                                 int ret_snapshot;
 985
 986                                 ret_snapshot = lttng_kconsumer_snapshot_metadata(
 987                                                 channel, key,
 988                                                 msg.u.snapshot_channel.pathname,
 989                                                 msg.u.snapshot_channel.relayd_id,
 990                                                 ctx);
 991                                 if (ret_snapshot < 0) {
 992                                         ERR("Snapshot metadata failed");
 993                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
 994                                 }
 995                         } else {
 996                                 int ret_snapshot;
 997
 998                                 ret_snapshot = lttng_kconsumer_snapshot_channel(
 999                                                 channel, key,
1000                                                 msg.u.snapshot_channel.pathname,
1001                                                 msg.u.snapshot_channel.relayd_id,
1002                                                 msg.u.snapshot_channel
1003                                                                 .nb_packets_per_stream);
1004                                 if (ret_snapshot < 0) {
1005                                         ERR("Snapshot channel failed");
1006                                         ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
1007                                 }
1008                         }
1009                 }
1010                 health_code_update();
1011
1012                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1013                 if (ret_send_status < 0) {
1014                         /* Somehow, the session daemon is not responding anymore. */
1015                         goto end_nosignal;
1016                 }
1017                 break;
1018         }
1019         case LTTNG_CONSUMER_DESTROY_CHANNEL:
1020         {
1021                 uint64_t key = msg.u.destroy_channel.key;
1022                 struct lttng_consumer_channel *channel;
1023                 int ret_send_status;
1024
1025                 channel = consumer_find_channel(key);
1026                 if (!channel) {
1027                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
1028                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1029                 }
1030
1031                 health_code_update();
1032
1033                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1034                 if (ret_send_status < 0) {
1035                         /* Somehow, the session daemon is not responding anymore. */
1036                         goto end_destroy_channel;
1037                 }
1038
1039                 health_code_update();
1040
1041                 /* Stop right now if no channel was found. */
1042                 if (!channel) {
1043                         goto end_destroy_channel;
1044                 }
1045
1046                 /*
1047                  * This command should ONLY be issued for channel with streams set in
1048                  * no monitor mode.
1049                  */
1050                 LTTNG_ASSERT(!channel->monitor);
1051
1052                 /*
1053                  * The refcount should ALWAYS be 0 in the case of a channel in no
1054                  * monitor mode.
1055                  */
1056                 LTTNG_ASSERT(!uatomic_sub_return(&channel->refcount, 1));
1057
1058                 consumer_del_channel(channel);
1059 end_destroy_channel:
1060                 goto end_nosignal;
1061         }
1062         case LTTNG_CONSUMER_DISCARDED_EVENTS:
1063         {
1064                 ssize_t ret;
1065                 uint64_t count;
1066                 struct lttng_consumer_channel *channel;
1067                 uint64_t id = msg.u.discarded_events.session_id;
1068                 uint64_t key = msg.u.discarded_events.channel_key;
1069
1070                 DBG("Kernel consumer discarded events command for session id %"
1071                                 PRIu64 ", channel key %" PRIu64, id, key);
1072
1073                 channel = consumer_find_channel(key);
1074                 if (!channel) {
1075                         ERR("Kernel consumer discarded events channel %"
1076                                         PRIu64 " not found", key);
1077                         count = 0;
1078                 } else {
1079                         count = channel->discarded_events;
1080                 }
1081
1082                 health_code_update();
1083
1084                 /* Send back returned value to session daemon */
1085                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1086                 if (ret < 0) {
1087                         PERROR("send discarded events");
1088                         goto error_fatal;
1089                 }
1090
1091                 break;
1092         }
1093         case LTTNG_CONSUMER_LOST_PACKETS:
1094         {
1095                 ssize_t ret;
1096                 uint64_t count;
1097                 struct lttng_consumer_channel *channel;
1098                 uint64_t id = msg.u.lost_packets.session_id;
1099                 uint64_t key = msg.u.lost_packets.channel_key;
1100
1101                 DBG("Kernel consumer lost packets command for session id %"
1102                                 PRIu64 ", channel key %" PRIu64, id, key);
1103
1104                 channel = consumer_find_channel(key);
1105                 if (!channel) {
1106                         ERR("Kernel consumer lost packets channel %"
1107                                         PRIu64 " not found", key);
1108                         count = 0;
1109                 } else {
1110                         count = channel->lost_packets;
1111                 }
1112
1113                 health_code_update();
1114
1115                 /* Send back returned value to session daemon */
1116                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1117                 if (ret < 0) {
1118                         PERROR("send lost packets");
1119                         goto error_fatal;
1120                 }
1121
1122                 break;
1123         }
1124         case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1125         {
1126                 int channel_monitor_pipe;
1127                 int ret_send_status, ret_set_channel_monitor_pipe;
1128                 ssize_t ret_recv;
1129
1130                 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1131                 /* Successfully received the command's type. */
1132                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1133                 if (ret_send_status < 0) {
1134                         goto error_fatal;
1135                 }
1136
1137                 ret_recv = lttcomm_recv_fds_unix_sock(
1138                                 sock, &channel_monitor_pipe, 1);
1139                 if (ret_recv != sizeof(channel_monitor_pipe)) {
1140                         ERR("Failed to receive channel monitor pipe");
1141                         goto error_fatal;
1142                 }
1143
1144                 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1145                 ret_set_channel_monitor_pipe =
1146                                 consumer_timer_thread_set_channel_monitor_pipe(
1147                                                 channel_monitor_pipe);
1148                 if (!ret_set_channel_monitor_pipe) {
1149                         int flags;
1150                         int ret_fcntl;
1151
1152                         ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1153                         /* Set the pipe as non-blocking. */
1154                         ret_fcntl = fcntl(channel_monitor_pipe, F_GETFL, 0);
1155                         if (ret_fcntl == -1) {
1156                                 PERROR("fcntl get flags of the channel monitoring pipe");
1157                                 goto error_fatal;
1158                         }
1159                         flags = ret_fcntl;
1160
1161                         ret_fcntl = fcntl(channel_monitor_pipe, F_SETFL,
1162                                         flags | O_NONBLOCK);
1163                         if (ret_fcntl == -1) {
1164                                 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1165                                 goto error_fatal;
1166                         }
1167                         DBG("Channel monitor pipe set as non-blocking");
1168                 } else {
1169                         ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1170                 }
1171                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1172                 if (ret_send_status < 0) {
1173                         goto error_fatal;
1174                 }
1175                 break;
1176         }
1177         case LTTNG_CONSUMER_ROTATE_CHANNEL:
1178         {
1179                 struct lttng_consumer_channel *channel;
1180                 uint64_t key = msg.u.rotate_channel.key;
1181                 int ret_send_status;
1182
1183                 DBG("Consumer rotate channel %" PRIu64, key);
1184
1185                 channel = consumer_find_channel(key);
1186                 if (!channel) {
1187                         ERR("Channel %" PRIu64 " not found", key);
1188                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1189                 } else {
1190                         /*
1191                          * Sample the rotate position of all the streams in this channel.
1192                          */
1193                         int ret_rotate_channel;
1194
1195                         ret_rotate_channel = lttng_consumer_rotate_channel(
1196                                         channel, key,
1197                                         msg.u.rotate_channel.relayd_id);
1198                         if (ret_rotate_channel < 0) {
1199                                 ERR("Rotate channel failed");
1200                                 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1201                         }
1202
1203                         health_code_update();
1204                 }
1205
1206                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1207                 if (ret_send_status < 0) {
1208                         /* Somehow, the session daemon is not responding anymore. */
1209                         goto error_rotate_channel;
1210                 }
1211                 if (channel) {
1212                         /* Rotate the streams that are ready right now. */
1213                         int ret_rotate;
1214
1215                         ret_rotate = lttng_consumer_rotate_ready_streams(
1216                                         channel, key);
1217                         if (ret_rotate < 0) {
1218                                 ERR("Rotate ready streams failed");
1219                         }
1220                 }
1221                 break;
1222 error_rotate_channel:
1223                 goto end_nosignal;
1224         }
1225         case LTTNG_CONSUMER_CLEAR_CHANNEL:
1226         {
1227                 struct lttng_consumer_channel *channel;
1228                 uint64_t key = msg.u.clear_channel.key;
1229                 int ret_send_status;
1230
1231                 channel = consumer_find_channel(key);
1232                 if (!channel) {
1233                         DBG("Channel %" PRIu64 " not found", key);
1234                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1235                 } else {
1236                         int ret_clear_channel;
1237
1238                         ret_clear_channel =
1239                                         lttng_consumer_clear_channel(channel);
1240                         if (ret_clear_channel) {
1241                                 ERR("Clear channel failed");
1242                                 ret_code = (lttcomm_return_code) ret_clear_channel;
1243                         }
1244
1245                         health_code_update();
1246                 }
1247
1248                 ret_send_status = consumer_send_status_msg(sock, ret_code);
1249                 if (ret_send_status < 0) {
1250                         /* Somehow, the session daemon is not responding anymore. */
1251                         goto end_nosignal;
1252                 }
1253
1254                 break;
1255         }
1256         case LTTNG_CONSUMER_INIT:
1257         {
1258                 int ret_send_status;
1259                 lttng_uuid sessiond_uuid;
1260
1261                 std::copy(std::begin(msg.u.init.sessiond_uuid), std::end(msg.u.init.sessiond_uuid),
1262                                 sessiond_uuid.begin());
1263
1264                 ret_code = lttng_consumer_init_command(ctx,
1265                                 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                                 (lttng_trace_chunk_command_type) 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         LTTNG_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 __attribute__((unused)))
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 __attribute__((unused)))
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         LTTNG_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                 LTTNG_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         LTTNG_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                 LTTNG_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 }