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

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
   5  *
   6  * This program is free software; you can redistribute it and/or modify
   7  * it under the terms of the GNU General Public License, version 2 only,
   8  * as published by the Free Software Foundation.
   9  *
  10  * This program is distributed in the hope that it will be useful,
  11  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  13  * GNU General Public License for more details.
  14  *
  15  * You should have received a copy of the GNU General Public License along
  16  * with this program; if not, write to the Free Software Foundation, Inc.,
  17  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  18  */
  19
  20 #define _LGPL_SOURCE
  21 #include <assert.h>
  22 #include <poll.h>
  23 #include <pthread.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <sys/mman.h>
  27 #include <sys/socket.h>
  28 #include <sys/types.h>
  29 #include <inttypes.h>
  30 #include <unistd.h>
  31 #include <sys/stat.h>
  32
  33 #include <bin/lttng-consumerd/health-consumerd.h>
  34 #include <common/common.h>
  35 #include <common/kernel-ctl/kernel-ctl.h>
  36 #include <common/sessiond-comm/sessiond-comm.h>
  37 #include <common/sessiond-comm/relayd.h>
  38 #include <common/compat/fcntl.h>
  39 #include <common/compat/endian.h>
  40 #include <common/pipe.h>
  41 #include <common/relayd/relayd.h>
  42 #include <common/utils.h>
  43 #include <common/consumer/consumer-stream.h>
  44 #include <common/index/index.h>
  45 #include <common/consumer/consumer-timer.h>
  46
  47 #include "kernel-consumer.h"
  48
  49 extern struct lttng_consumer_global_data consumer_data;
  50 extern int consumer_poll_timeout;
  51
  52 /*
  53  * Take a snapshot for a specific fd
  54  *
  55  * Returns 0 on success, < 0 on error
  56  */
  57 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  58 {
  59         int ret = 0;
  60         int infd = stream->wait_fd;
  61
  62         ret = kernctl_snapshot(infd);
  63         if (ret != 0) {
  64                 PERROR("Getting sub-buffer snapshot.");
  65         }
  66
  67         return ret;
  68 }
  69
  70 /*
  71  * Sample consumed and produced positions for a specific fd.
  72  *
  73  * Returns 0 on success, < 0 on error.
  74  */
  75 int lttng_kconsumer_sample_snapshot_positions(
  76                 struct lttng_consumer_stream *stream)
  77 {
  78         assert(stream);
  79
  80         return kernctl_snapshot_sample_positions(stream->wait_fd);
  81 }
  82
  83 /*
  84  * Get the produced position
  85  *
  86  * Returns 0 on success, < 0 on error
  87  */
  88 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  89                 unsigned long *pos)
  90 {
  91         int ret;
  92         int infd = stream->wait_fd;
  93
  94         ret = kernctl_snapshot_get_produced(infd, pos);
  95         if (ret != 0) {
  96                 PERROR("kernctl_snapshot_get_produced");
  97         }
  98
  99         return ret;
 100 }
 101
 102 /*
 103  * Get the consumerd position
 104  *
 105  * Returns 0 on success, < 0 on error
 106  */
 107 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 108                 unsigned long *pos)
 109 {
 110         int ret;
 111         int infd = stream->wait_fd;
 112
 113         ret = kernctl_snapshot_get_consumed(infd, pos);
 114         if (ret != 0) {
 115                 PERROR("kernctl_snapshot_get_consumed");
 116         }
 117
 118         return ret;
 119 }
 120
 121 /*
 122  * Take a snapshot of all the stream of a channel
 123  *
 124  * Returns 0 on success, < 0 on error
 125  */
 126 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 127                 uint64_t relayd_id, uint64_t nb_packets_per_stream,
 128                 struct lttng_consumer_local_data *ctx)
 129 {
 130         int ret;
 131         struct lttng_consumer_channel *channel;
 132         struct lttng_consumer_stream *stream;
 133
 134         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 135
 136         rcu_read_lock();
 137
 138         channel = consumer_find_channel(key);
 139         if (!channel) {
 140                 ERR("No channel found for key %" PRIu64, key);
 141                 ret = -1;
 142                 goto end;
 143         }
 144
 145         /* Splice is not supported yet for channel snapshot. */
 146         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 147                 ERR("Unsupported output %d", channel->output);
 148                 ret = -1;
 149                 goto end;
 150         }
 151
 152         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 153                 unsigned long consumed_pos, produced_pos;
 154
 155                 health_code_update();
 156
 157                 /*
 158                  * Lock stream because we are about to change its state.
 159                  */
 160                 pthread_mutex_lock(&stream->lock);
 161
 162                 /*
 163                  * Assign the received relayd ID so we can use it for streaming. The streams
 164                  * are not visible to anyone so this is OK to change it.
 165                  */
 166                 stream->net_seq_idx = relayd_id;
 167                 channel->relayd_id = relayd_id;
 168                 if (relayd_id != (uint64_t) -1ULL) {
 169                         ret = consumer_send_relayd_stream(stream, path);
 170                         if (ret < 0) {
 171                                 ERR("sending stream to relayd");
 172                                 goto end_unlock;
 173                         }
 174                 } else {
 175                         ret = utils_create_stream_file(path, stream->name,
 176                                         stream->chan->tracefile_size,
 177                                         stream->tracefile_count_current,
 178                                         stream->uid, stream->gid, NULL);
 179                         if (ret < 0) {
 180                                 ERR("utils_create_stream_file");
 181                                 goto end_unlock;
 182                         }
 183
 184                         stream->out_fd = ret;
 185                         stream->tracefile_size_current = 0;
 186
 187                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 188                                         path, stream->name, stream->key);
 189                 }
 190                 if (relayd_id != -1ULL) {
 191                         ret = consumer_send_relayd_streams_sent(relayd_id);
 192                         if (ret < 0) {
 193                                 ERR("sending streams sent to relayd");
 194                                 goto end_unlock;
 195                         }
 196                         channel->streams_sent_to_relayd = true;
 197                 }
 198
 199                 ret = kernctl_buffer_flush_empty(stream->wait_fd);
 200                 if (ret < 0) {
 201                         /*
 202                          * Doing a buffer flush which does not take into
 203                          * account empty packets. This is not perfect
 204                          * for stream intersection, but required as a
 205                          * fall-back when "flush_empty" is not
 206                          * implemented by lttng-modules.
 207                          */
 208                         ret = kernctl_buffer_flush(stream->wait_fd);
 209                         if (ret < 0) {
 210                                 ERR("Failed to flush kernel stream");
 211                                 goto end_unlock;
 212                         }
 213                         goto end_unlock;
 214                 }
 215
 216                 ret = lttng_kconsumer_take_snapshot(stream);
 217                 if (ret < 0) {
 218                         ERR("Taking kernel snapshot");
 219                         goto end_unlock;
 220                 }
 221
 222                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 223                 if (ret < 0) {
 224                         ERR("Produced kernel snapshot position");
 225                         goto end_unlock;
 226                 }
 227
 228                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 229                 if (ret < 0) {
 230                         ERR("Consumerd kernel snapshot position");
 231                         goto end_unlock;
 232                 }
 233
 234                 if (stream->max_sb_size == 0) {
 235                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 236                                         &stream->max_sb_size);
 237                         if (ret < 0) {
 238                                 ERR("Getting kernel max_sb_size");
 239                                 goto end_unlock;
 240                         }
 241                 }
 242
 243                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 244                                 produced_pos, nb_packets_per_stream,
 245                                 stream->max_sb_size);
 246
 247                 while (consumed_pos < produced_pos) {
 248                         ssize_t read_len;
 249                         unsigned long len, padded_len;
 250
 251                         health_code_update();
 252
 253                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 254
 255                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 256                         if (ret < 0) {
 257                                 if (ret != -EAGAIN) {
 258                                         PERROR("kernctl_get_subbuf snapshot");
 259                                         goto end_unlock;
 260                                 }
 261                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 262                                 consumed_pos += stream->max_sb_size;
 263                                 stream->chan->lost_packets++;
 264                                 continue;
 265                         }
 266
 267                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 268                         if (ret < 0) {
 269                                 ERR("Snapshot kernctl_get_subbuf_size");
 270                                 goto error_put_subbuf;
 271                         }
 272
 273                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 274                         if (ret < 0) {
 275                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 276                                 goto error_put_subbuf;
 277                         }
 278
 279                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 280                                         padded_len - len, NULL);
 281                         /*
 282                          * We write the padded len in local tracefiles but the data len
 283                          * when using a relay. Display the error but continue processing
 284                          * to try to release the subbuffer.
 285                          */
 286                         if (relayd_id != (uint64_t) -1ULL) {
 287                                 if (read_len != len) {
 288                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 289                                                         read_len, len);
 290                                 }
 291                         } else {
 292                                 if (read_len != padded_len) {
 293                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 294                                                         read_len, padded_len);
 295                                 }
 296                         }
 297
 298                         ret = kernctl_put_subbuf(stream->wait_fd);
 299                         if (ret < 0) {
 300                                 ERR("Snapshot kernctl_put_subbuf");
 301                                 goto end_unlock;
 302                         }
 303                         consumed_pos += stream->max_sb_size;
 304                 }
 305
 306                 if (relayd_id == (uint64_t) -1ULL) {
 307                         if (stream->out_fd >= 0) {
 308                                 ret = close(stream->out_fd);
 309                                 if (ret < 0) {
 310                                         PERROR("Kernel consumer snapshot close out_fd");
 311                                         goto end_unlock;
 312                                 }
 313                                 stream->out_fd = -1;
 314                         }
 315                 } else {
 316                         close_relayd_stream(stream);
 317                         stream->net_seq_idx = (uint64_t) -1ULL;
 318                 }
 319                 pthread_mutex_unlock(&stream->lock);
 320         }
 321
 322         /* All good! */
 323         ret = 0;
 324         goto end;
 325
 326 error_put_subbuf:
 327         ret = kernctl_put_subbuf(stream->wait_fd);
 328         if (ret < 0) {
 329                 ERR("Snapshot kernctl_put_subbuf error path");
 330         }
 331 end_unlock:
 332         pthread_mutex_unlock(&stream->lock);
 333 end:
 334         rcu_read_unlock();
 335         return ret;
 336 }
 337
 338 /*
 339  * Read the whole metadata available for a snapshot.
 340  *
 341  * Returns 0 on success, < 0 on error
 342  */
 343 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 344                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 345 {
 346         int ret, use_relayd = 0;
 347         ssize_t ret_read;
 348         struct lttng_consumer_channel *metadata_channel;
 349         struct lttng_consumer_stream *metadata_stream;
 350
 351         assert(ctx);
 352
 353         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 354                         key, path);
 355
 356         rcu_read_lock();
 357
 358         metadata_channel = consumer_find_channel(key);
 359         if (!metadata_channel) {
 360                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 361                 ret = -1;
 362                 goto error;
 363         }
 364
 365         metadata_stream = metadata_channel->metadata_stream;
 366         assert(metadata_stream);
 367
 368         /* Flag once that we have a valid relayd for the stream. */
 369         if (relayd_id != (uint64_t) -1ULL) {
 370                 use_relayd = 1;
 371         }
 372
 373         if (use_relayd) {
 374                 ret = consumer_send_relayd_stream(metadata_stream, path);
 375                 if (ret < 0) {
 376                         goto error;
 377                 }
 378         } else {
 379                 ret = utils_create_stream_file(path, metadata_stream->name,
 380                                 metadata_stream->chan->tracefile_size,
 381                                 metadata_stream->tracefile_count_current,
 382                                 metadata_stream->uid, metadata_stream->gid, NULL);
 383                 if (ret < 0) {
 384                         goto error;
 385                 }
 386                 metadata_stream->out_fd = ret;
 387         }
 388
 389         do {
 390                 health_code_update();
 391
 392                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 393                 if (ret_read < 0) {
 394                         if (ret_read != -EAGAIN) {
 395                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 396                                                 ret_read);
 397                                 goto error;
 398                         }
 399                         /* ret_read is negative at this point so we will exit the loop. */
 400                         continue;
 401                 }
 402         } while (ret_read >= 0);
 403
 404         if (use_relayd) {
 405                 close_relayd_stream(metadata_stream);
 406                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 407         } else {
 408                 if (metadata_stream->out_fd >= 0) {
 409                         ret = close(metadata_stream->out_fd);
 410                         if (ret < 0) {
 411                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 412                                 /*
 413                                  * Don't go on error here since the snapshot was successful at this
 414                                  * point but somehow the close failed.
 415                                  */
 416                         }
 417                         metadata_stream->out_fd = -1;
 418                 }
 419         }
 420
 421         ret = 0;
 422
 423         cds_list_del(&metadata_stream->send_node);
 424         consumer_stream_destroy(metadata_stream, NULL);
 425         metadata_channel->metadata_stream = NULL;
 426 error:
 427         rcu_read_unlock();
 428         return ret;
 429 }
 430
 431 /*
 432  * Receive command from session daemon and process it.
 433  *
 434  * Return 1 on success else a negative value or 0.
 435  */
 436 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 437                 int sock, struct pollfd *consumer_sockpoll)
 438 {
 439         ssize_t ret;
 440         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 441         struct lttcomm_consumer_msg msg;
 442
 443         health_code_update();
 444
 445         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 446         if (ret != sizeof(msg)) {
 447                 if (ret > 0) {
 448                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 449                         ret = -1;
 450                 }
 451                 return ret;
 452         }
 453
 454         health_code_update();
 455
 456         /* Deprecated command */
 457         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 458
 459         health_code_update();
 460
 461         /* relayd needs RCU read-side protection */
 462         rcu_read_lock();
 463
 464         switch (msg.cmd_type) {
 465         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 466         {
 467                 /* Session daemon status message are handled in the following call. */
 468                 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 469                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 470                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 471                                 msg.u.relayd_sock.relayd_session_id);
 472                 goto end_nosignal;
 473         }
 474         case LTTNG_CONSUMER_ADD_CHANNEL:
 475         {
 476                 struct lttng_consumer_channel *new_channel;
 477                 int ret_recv;
 478
 479                 health_code_update();
 480
 481                 /* First send a status message before receiving the fds. */
 482                 ret = consumer_send_status_msg(sock, ret_code);
 483                 if (ret < 0) {
 484                         /* Somehow, the session daemon is not responding anymore. */
 485                         goto error_fatal;
 486                 }
 487
 488                 health_code_update();
 489
 490                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 491                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 492                                 msg.u.channel.session_id, msg.u.channel.pathname,
 493                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 494                                 msg.u.channel.relayd_id, msg.u.channel.output,
 495                                 msg.u.channel.tracefile_size,
 496                                 msg.u.channel.tracefile_count, 0,
 497                                 msg.u.channel.monitor,
 498                                 msg.u.channel.live_timer_interval,
 499                                 NULL, NULL);
 500                 if (new_channel == NULL) {
 501                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 502                         goto end_nosignal;
 503                 }
 504                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 505                 switch (msg.u.channel.output) {
 506                 case LTTNG_EVENT_SPLICE:
 507                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 508                         break;
 509                 case LTTNG_EVENT_MMAP:
 510                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 511                         break;
 512                 default:
 513                         ERR("Channel output unknown %d", msg.u.channel.output);
 514                         goto end_nosignal;
 515                 }
 516
 517                 /* Translate and save channel type. */
 518                 switch (msg.u.channel.type) {
 519                 case CONSUMER_CHANNEL_TYPE_DATA:
 520                 case CONSUMER_CHANNEL_TYPE_METADATA:
 521                         new_channel->type = msg.u.channel.type;
 522                         break;
 523                 default:
 524                         assert(0);
 525                         goto end_nosignal;
 526                 };
 527
 528                 health_code_update();
 529
 530                 if (ctx->on_recv_channel != NULL) {
 531                         ret_recv = ctx->on_recv_channel(new_channel);
 532                         if (ret_recv == 0) {
 533                                 ret = consumer_add_channel(new_channel, ctx);
 534                         } else if (ret_recv < 0) {
 535                                 goto end_nosignal;
 536                         }
 537                 } else {
 538                         ret = consumer_add_channel(new_channel, ctx);
 539                 }
 540                 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
 541                         int monitor_start_ret;
 542
 543                         DBG("Consumer starting monitor timer");
 544                         consumer_timer_live_start(new_channel,
 545                                         msg.u.channel.live_timer_interval);
 546                         monitor_start_ret = consumer_timer_monitor_start(
 547                                         new_channel,
 548                                         msg.u.channel.monitor_timer_interval);
 549                         if (monitor_start_ret < 0) {
 550                                 ERR("Starting channel monitoring timer failed");
 551                                 goto end_nosignal;
 552                         }
 553
 554                 }
 555
 556                 health_code_update();
 557
 558                 /* If we received an error in add_channel, we need to report it. */
 559                 if (ret < 0) {
 560                         ret = consumer_send_status_msg(sock, ret);
 561                         if (ret < 0) {
 562                                 goto error_fatal;
 563                         }
 564                         goto end_nosignal;
 565                 }
 566
 567                 goto end_nosignal;
 568         }
 569         case LTTNG_CONSUMER_ADD_STREAM:
 570         {
 571                 int fd;
 572                 struct lttng_pipe *stream_pipe;
 573                 struct lttng_consumer_stream *new_stream;
 574                 struct lttng_consumer_channel *channel;
 575                 int alloc_ret = 0;
 576
 577                 /*
 578                  * Get stream's channel reference. Needed when adding the stream to the
 579                  * global hash table.
 580                  */
 581                 channel = consumer_find_channel(msg.u.stream.channel_key);
 582                 if (!channel) {
 583                         /*
 584                          * We could not find the channel. Can happen if cpu hotplug
 585                          * happens while tearing down.
 586                          */
 587                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 588                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 589                 }
 590
 591                 health_code_update();
 592
 593                 /* First send a status message before receiving the fds. */
 594                 ret = consumer_send_status_msg(sock, ret_code);
 595                 if (ret < 0) {
 596                         /* Somehow, the session daemon is not responding anymore. */
 597                         goto error_fatal;
 598                 }
 599
 600                 health_code_update();
 601
 602                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 603                         /* Channel was not found. */
 604                         goto end_nosignal;
 605                 }
 606
 607                 /* Blocking call */
 608                 health_poll_entry();
 609                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 610                 health_poll_exit();
 611                 if (ret) {
 612                         goto error_fatal;
 613                 }
 614
 615                 health_code_update();
 616
 617                 /* Get stream file descriptor from socket */
 618                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 619                 if (ret != sizeof(fd)) {
 620                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 621                         rcu_read_unlock();
 622                         return ret;
 623                 }
 624
 625                 health_code_update();
 626
 627                 /*
 628                  * Send status code to session daemon only if the recv works. If the
 629                  * above recv() failed, the session daemon is notified through the
 630                  * error socket and the teardown is eventually done.
 631                  */
 632                 ret = consumer_send_status_msg(sock, ret_code);
 633                 if (ret < 0) {
 634                         /* Somehow, the session daemon is not responding anymore. */
 635                         goto end_nosignal;
 636                 }
 637
 638                 health_code_update();
 639
 640                 new_stream = consumer_allocate_stream(channel->key,
 641                                 fd,
 642                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 643                                 channel->name,
 644                                 channel->uid,
 645                                 channel->gid,
 646                                 channel->relayd_id,
 647                                 channel->session_id,
 648                                 msg.u.stream.cpu,
 649                                 &alloc_ret,
 650                                 channel->type,
 651                                 channel->monitor);
 652                 if (new_stream == NULL) {
 653                         switch (alloc_ret) {
 654                         case -ENOMEM:
 655                         case -EINVAL:
 656                         default:
 657                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 658                                 break;
 659                         }
 660                         goto end_nosignal;
 661                 }
 662
 663                 new_stream->chan = channel;
 664                 new_stream->wait_fd = fd;
 665                 switch (channel->output) {
 666                 case CONSUMER_CHANNEL_SPLICE:
 667                         new_stream->output = LTTNG_EVENT_SPLICE;
 668                         ret = utils_create_pipe(new_stream->splice_pipe);
 669                         if (ret < 0) {
 670                                 goto end_nosignal;
 671                         }
 672                         break;
 673                 case CONSUMER_CHANNEL_MMAP:
 674                         new_stream->output = LTTNG_EVENT_MMAP;
 675                         break;
 676                 default:
 677                         ERR("Stream output unknown %d", channel->output);
 678                         goto end_nosignal;
 679                 }
 680
 681                 /*
 682                  * We've just assigned the channel to the stream so increment the
 683                  * refcount right now. We don't need to increment the refcount for
 684                  * streams in no monitor because we handle manually the cleanup of
 685                  * those. It is very important to make sure there is NO prior
 686                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 687                  */
 688                 if (channel->monitor) {
 689                         uatomic_inc(&new_stream->chan->refcount);
 690                 }
 691
 692                 /*
 693                  * The buffer flush is done on the session daemon side for the kernel
 694                  * so no need for the stream "hangup_flush_done" variable to be
 695                  * tracked. This is important for a kernel stream since we don't rely
 696                  * on the flush state of the stream to read data. It's not the case for
 697                  * user space tracing.
 698                  */
 699                 new_stream->hangup_flush_done = 0;
 700
 701                 health_code_update();
 702
 703                 if (ctx->on_recv_stream) {
 704                         ret = ctx->on_recv_stream(new_stream);
 705                         if (ret < 0) {
 706                                 consumer_stream_free(new_stream);
 707                                 goto end_nosignal;
 708                         }
 709                 }
 710
 711                 health_code_update();
 712
 713                 if (new_stream->metadata_flag) {
 714                         channel->metadata_stream = new_stream;
 715                 }
 716
 717                 /* Do not monitor this stream. */
 718                 if (!channel->monitor) {
 719                         DBG("Kernel consumer add stream %s in no monitor mode with "
 720                                         "relayd id %" PRIu64, new_stream->name,
 721                                         new_stream->net_seq_idx);
 722                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 723                         break;
 724                 }
 725
 726                 /* Send stream to relayd if the stream has an ID. */
 727                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 728                         ret = consumer_send_relayd_stream(new_stream,
 729                                         new_stream->chan->pathname);
 730                         if (ret < 0) {
 731                                 consumer_stream_free(new_stream);
 732                                 goto end_nosignal;
 733                         }
 734
 735                         /*
 736                          * If adding an extra stream to an already
 737                          * existing channel (e.g. cpu hotplug), we need
 738                          * to send the "streams_sent" command to relayd.
 739                          */
 740                         if (channel->streams_sent_to_relayd) {
 741                                 ret = consumer_send_relayd_streams_sent(
 742                                                 new_stream->net_seq_idx);
 743                                 if (ret < 0) {
 744                                         goto end_nosignal;
 745                                 }
 746                         }
 747                 }
 748
 749                 /* Get the right pipe where the stream will be sent. */
 750                 if (new_stream->metadata_flag) {
 751                         ret = consumer_add_metadata_stream(new_stream);
 752                         if (ret) {
 753                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 754                                                 new_stream->key);
 755                                 consumer_stream_free(new_stream);
 756                                 goto end_nosignal;
 757                         }
 758                         stream_pipe = ctx->consumer_metadata_pipe;
 759                 } else {
 760                         ret = consumer_add_data_stream(new_stream);
 761                         if (ret) {
 762                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 763                                                 new_stream->key);
 764                                 consumer_stream_free(new_stream);
 765                                 goto end_nosignal;
 766                         }
 767                         stream_pipe = ctx->consumer_data_pipe;
 768                 }
 769
 770                 /* Vitible to other threads */
 771                 new_stream->globally_visible = 1;
 772
 773                 health_code_update();
 774
 775                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 776                 if (ret < 0) {
 777                         ERR("Consumer write %s stream to pipe %d",
 778                                         new_stream->metadata_flag ? "metadata" : "data",
 779                                         lttng_pipe_get_writefd(stream_pipe));
 780                         if (new_stream->metadata_flag) {
 781                                 consumer_del_stream_for_metadata(new_stream);
 782                         } else {
 783                                 consumer_del_stream_for_data(new_stream);
 784                         }
 785                         goto end_nosignal;
 786                 }
 787
 788                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 789                                 new_stream->name, fd, new_stream->relayd_stream_id);
 790                 break;
 791         }
 792         case LTTNG_CONSUMER_STREAMS_SENT:
 793         {
 794                 struct lttng_consumer_channel *channel;
 795
 796                 /*
 797                  * Get stream's channel reference. Needed when adding the stream to the
 798                  * global hash table.
 799                  */
 800                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 801                 if (!channel) {
 802                         /*
 803                          * We could not find the channel. Can happen if cpu hotplug
 804                          * happens while tearing down.
 805                          */
 806                         ERR("Unable to find channel key %" PRIu64,
 807                                         msg.u.sent_streams.channel_key);
 808                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 809                 }
 810
 811                 health_code_update();
 812
 813                 /*
 814                  * Send status code to session daemon.
 815                  */
 816                 ret = consumer_send_status_msg(sock, ret_code);
 817                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 818                         /* Somehow, the session daemon is not responding anymore. */
 819                         goto end_nosignal;
 820                 }
 821
 822                 health_code_update();
 823
 824                 /*
 825                  * We should not send this message if we don't monitor the
 826                  * streams in this channel.
 827                  */
 828                 if (!channel->monitor) {
 829                         break;
 830                 }
 831
 832                 health_code_update();
 833                 /* Send stream to relayd if the stream has an ID. */
 834                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 835                         ret = consumer_send_relayd_streams_sent(
 836                                         msg.u.sent_streams.net_seq_idx);
 837                         if (ret < 0) {
 838                                 goto end_nosignal;
 839                         }
 840                         channel->streams_sent_to_relayd = true;
 841                 }
 842                 break;
 843         }
 844         case LTTNG_CONSUMER_UPDATE_STREAM:
 845         {
 846                 rcu_read_unlock();
 847                 return -ENOSYS;
 848         }
 849         case LTTNG_CONSUMER_DESTROY_RELAYD:
 850         {
 851                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 852                 struct consumer_relayd_sock_pair *relayd;
 853
 854                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 855
 856                 /* Get relayd reference if exists. */
 857                 relayd = consumer_find_relayd(index);
 858                 if (relayd == NULL) {
 859                         DBG("Unable to find relayd %" PRIu64, index);
 860                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 861                 }
 862
 863                 /*
 864                  * Each relayd socket pair has a refcount of stream attached to it
 865                  * which tells if the relayd is still active or not depending on the
 866                  * refcount value.
 867                  *
 868                  * This will set the destroy flag of the relayd object and destroy it
 869                  * if the refcount reaches zero when called.
 870                  *
 871                  * The destroy can happen either here or when a stream fd hangs up.
 872                  */
 873                 if (relayd) {
 874                         consumer_flag_relayd_for_destroy(relayd);
 875                 }
 876
 877                 health_code_update();
 878
 879                 ret = consumer_send_status_msg(sock, ret_code);
 880                 if (ret < 0) {
 881                         /* Somehow, the session daemon is not responding anymore. */
 882                         goto error_fatal;
 883                 }
 884
 885                 goto end_nosignal;
 886         }
 887         case LTTNG_CONSUMER_DATA_PENDING:
 888         {
 889                 int32_t ret;
 890                 uint64_t id = msg.u.data_pending.session_id;
 891
 892                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 893
 894                 ret = consumer_data_pending(id);
 895
 896                 health_code_update();
 897
 898                 /* Send back returned value to session daemon */
 899                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 900                 if (ret < 0) {
 901                         PERROR("send data pending ret code");
 902                         goto error_fatal;
 903                 }
 904
 905                 /*
 906                  * No need to send back a status message since the data pending
 907                  * returned value is the response.
 908                  */
 909                 break;
 910         }
 911         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 912         {
 913                 if (msg.u.snapshot_channel.metadata == 1) {
 914                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 915                                         msg.u.snapshot_channel.pathname,
 916                                         msg.u.snapshot_channel.relayd_id, ctx);
 917                         if (ret < 0) {
 918                                 ERR("Snapshot metadata failed");
 919                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 920                         }
 921                 } else {
 922                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 923                                         msg.u.snapshot_channel.pathname,
 924                                         msg.u.snapshot_channel.relayd_id,
 925                                         msg.u.snapshot_channel.nb_packets_per_stream,
 926                                         ctx);
 927                         if (ret < 0) {
 928                                 ERR("Snapshot channel failed");
 929                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 930                         }
 931                 }
 932
 933                 health_code_update();
 934
 935                 ret = consumer_send_status_msg(sock, ret_code);
 936                 if (ret < 0) {
 937                         /* Somehow, the session daemon is not responding anymore. */
 938                         goto end_nosignal;
 939                 }
 940                 break;
 941         }
 942         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 943         {
 944                 uint64_t key = msg.u.destroy_channel.key;
 945                 struct lttng_consumer_channel *channel;
 946
 947                 channel = consumer_find_channel(key);
 948                 if (!channel) {
 949                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 950                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 951                 }
 952
 953                 health_code_update();
 954
 955                 ret = consumer_send_status_msg(sock, ret_code);
 956                 if (ret < 0) {
 957                         /* Somehow, the session daemon is not responding anymore. */
 958                         goto end_nosignal;
 959                 }
 960
 961                 health_code_update();
 962
 963                 /* Stop right now if no channel was found. */
 964                 if (!channel) {
 965                         goto end_nosignal;
 966                 }
 967
 968                 /*
 969                  * This command should ONLY be issued for channel with streams set in
 970                  * no monitor mode.
 971                  */
 972                 assert(!channel->monitor);
 973
 974                 /*
 975                  * The refcount should ALWAYS be 0 in the case of a channel in no
 976                  * monitor mode.
 977                  */
 978                 assert(!uatomic_sub_return(&channel->refcount, 1));
 979
 980                 consumer_del_channel(channel);
 981
 982                 goto end_nosignal;
 983         }
 984         case LTTNG_CONSUMER_DISCARDED_EVENTS:
 985         {
 986                 ssize_t ret;
 987                 uint64_t count;
 988                 struct lttng_consumer_channel *channel;
 989                 uint64_t id = msg.u.discarded_events.session_id;
 990                 uint64_t key = msg.u.discarded_events.channel_key;
 991
 992                 DBG("Kernel consumer discarded events command for session id %"
 993                                 PRIu64 ", channel key %" PRIu64, id, key);
 994
 995                 channel = consumer_find_channel(key);
 996                 if (!channel) {
 997                         ERR("Kernel consumer discarded events channel %"
 998                                         PRIu64 " not found", key);
 999                         count = 0;
1000                 } else {
1001                         count = channel->discarded_events;
1002                 }
1003
1004                 health_code_update();
1005
1006                 /* Send back returned value to session daemon */
1007                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1008                 if (ret < 0) {
1009                         PERROR("send discarded events");
1010                         goto error_fatal;
1011                 }
1012
1013                 break;
1014         }
1015         case LTTNG_CONSUMER_LOST_PACKETS:
1016         {
1017                 ssize_t ret;
1018                 uint64_t count;
1019                 struct lttng_consumer_channel *channel;
1020                 uint64_t id = msg.u.lost_packets.session_id;
1021                 uint64_t key = msg.u.lost_packets.channel_key;
1022
1023                 DBG("Kernel consumer lost packets command for session id %"
1024                                 PRIu64 ", channel key %" PRIu64, id, key);
1025
1026                 channel = consumer_find_channel(key);
1027                 if (!channel) {
1028                         ERR("Kernel consumer lost packets channel %"
1029                                         PRIu64 " not found", key);
1030                         count = 0;
1031                 } else {
1032                         count = channel->lost_packets;
1033                 }
1034
1035                 health_code_update();
1036
1037                 /* Send back returned value to session daemon */
1038                 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
1039                 if (ret < 0) {
1040                         PERROR("send lost packets");
1041                         goto error_fatal;
1042                 }
1043
1044                 break;
1045         }
1046         case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1047         {
1048                 int channel_monitor_pipe;
1049
1050                 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1051                 /* Successfully received the command's type. */
1052                 ret = consumer_send_status_msg(sock, ret_code);
1053                 if (ret < 0) {
1054                         goto error_fatal;
1055                 }
1056
1057                 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1058                                 1);
1059                 if (ret != sizeof(channel_monitor_pipe)) {
1060                         ERR("Failed to receive channel monitor pipe");
1061                         goto error_fatal;
1062                 }
1063
1064                 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1065                 ret = consumer_timer_thread_set_channel_monitor_pipe(
1066                                 channel_monitor_pipe);
1067                 if (!ret) {
1068                         int flags;
1069
1070                         ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1071                         /* Set the pipe as non-blocking. */
1072                         ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1073                         if (ret == -1) {
1074                                 PERROR("fcntl get flags of the channel monitoring pipe");
1075                                 goto error_fatal;
1076                         }
1077                         flags = ret;
1078
1079                         ret = fcntl(channel_monitor_pipe, F_SETFL,
1080                                         flags | O_NONBLOCK);
1081                         if (ret == -1) {
1082                                 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1083                                 goto error_fatal;
1084                         }
1085                         DBG("Channel monitor pipe set as non-blocking");
1086                 } else {
1087                         ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1088                 }
1089                 ret = consumer_send_status_msg(sock, ret_code);
1090                 if (ret < 0) {
1091                         goto error_fatal;
1092                 }
1093                 break;
1094         }
1095         default:
1096                 goto end_nosignal;
1097         }
1098
1099 end_nosignal:
1100         rcu_read_unlock();
1101
1102         /*
1103          * Return 1 to indicate success since the 0 value can be a socket
1104          * shutdown during the recv() or send() call.
1105          */
1106         health_code_update();
1107         return 1;
1108
1109 error_fatal:
1110         rcu_read_unlock();
1111         /* This will issue a consumer stop. */
1112         return -1;
1113 }
1114
1115 /*
1116  * Populate index values of a kernel stream. Values are set in big endian order.
1117  *
1118  * Return 0 on success or else a negative value.
1119  */
1120 static int get_index_values(struct ctf_packet_index *index, int infd)
1121 {
1122         int ret;
1123
1124         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
1125         if (ret < 0) {
1126                 PERROR("kernctl_get_timestamp_begin");
1127                 goto error;
1128         }
1129         index->timestamp_begin = htobe64(index->timestamp_begin);
1130
1131         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
1132         if (ret < 0) {
1133                 PERROR("kernctl_get_timestamp_end");
1134                 goto error;
1135         }
1136         index->timestamp_end = htobe64(index->timestamp_end);
1137
1138         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
1139         if (ret < 0) {
1140                 PERROR("kernctl_get_events_discarded");
1141                 goto error;
1142         }
1143         index->events_discarded = htobe64(index->events_discarded);
1144
1145         ret = kernctl_get_content_size(infd, &index->content_size);
1146         if (ret < 0) {
1147                 PERROR("kernctl_get_content_size");
1148                 goto error;
1149         }
1150         index->content_size = htobe64(index->content_size);
1151
1152         ret = kernctl_get_packet_size(infd, &index->packet_size);
1153         if (ret < 0) {
1154                 PERROR("kernctl_get_packet_size");
1155                 goto error;
1156         }
1157         index->packet_size = htobe64(index->packet_size);
1158
1159         ret = kernctl_get_stream_id(infd, &index->stream_id);
1160         if (ret < 0) {
1161                 PERROR("kernctl_get_stream_id");
1162                 goto error;
1163         }
1164         index->stream_id = htobe64(index->stream_id);
1165
1166         ret = kernctl_get_instance_id(infd, &index->stream_instance_id);
1167         if (ret < 0) {
1168                 if (ret == -ENOTTY) {
1169                         /* Command not implemented by lttng-modules. */
1170                         index->stream_instance_id = -1ULL;
1171                 } else {
1172                         PERROR("kernctl_get_instance_id");
1173                         goto error;
1174                 }
1175         }
1176         index->stream_instance_id = htobe64(index->stream_instance_id);
1177
1178         ret = kernctl_get_sequence_number(infd, &index->packet_seq_num);
1179         if (ret < 0) {
1180                 if (ret == -ENOTTY) {
1181                         /* Command not implemented by lttng-modules. */
1182                         index->packet_seq_num = -1ULL;
1183                         ret = 0;
1184                 } else {
1185                         PERROR("kernctl_get_sequence_number");
1186                         goto error;
1187                 }
1188         }
1189         index->packet_seq_num = htobe64(index->packet_seq_num);
1190
1191 error:
1192         return ret;
1193 }
1194 /*
1195  * Sync metadata meaning request them to the session daemon and snapshot to the
1196  * metadata thread can consumer them.
1197  *
1198  * Metadata stream lock MUST be acquired.
1199  *
1200  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1201  * is empty or a negative value on error.
1202  */
1203 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1204 {
1205         int ret;
1206
1207         assert(metadata);
1208
1209         ret = kernctl_buffer_flush(metadata->wait_fd);
1210         if (ret < 0) {
1211                 ERR("Failed to flush kernel stream");
1212                 goto end;
1213         }
1214
1215         ret = kernctl_snapshot(metadata->wait_fd);
1216         if (ret < 0) {
1217                 if (ret != -EAGAIN) {
1218                         ERR("Sync metadata, taking kernel snapshot failed.");
1219                         goto end;
1220                 }
1221                 DBG("Sync metadata, no new kernel metadata");
1222                 /* No new metadata, exit. */
1223                 ret = ENODATA;
1224                 goto end;
1225         }
1226
1227 end:
1228         return ret;
1229 }
1230
1231 static
1232 int update_stream_stats(struct lttng_consumer_stream *stream)
1233 {
1234         int ret;
1235         uint64_t seq, discarded;
1236
1237         ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
1238         if (ret < 0) {
1239                 if (ret == -ENOTTY) {
1240                         /* Command not implemented by lttng-modules. */
1241                         seq = -1ULL;
1242                 } else {
1243                         PERROR("kernctl_get_sequence_number");
1244                         goto end;
1245                 }
1246         }
1247
1248         /*
1249          * Start the sequence when we extract the first packet in case we don't
1250          * start at 0 (for example if a consumer is not connected to the
1251          * session immediately after the beginning).
1252          */
1253         if (stream->last_sequence_number == -1ULL) {
1254                 stream->last_sequence_number = seq;
1255         } else if (seq > stream->last_sequence_number) {
1256                 stream->chan->lost_packets += seq -
1257                                 stream->last_sequence_number - 1;
1258         } else {
1259                 /* seq <= last_sequence_number */
1260                 ERR("Sequence number inconsistent : prev = %" PRIu64
1261                                 ", current = %" PRIu64,
1262                                 stream->last_sequence_number, seq);
1263                 ret = -1;
1264                 goto end;
1265         }
1266         stream->last_sequence_number = seq;
1267
1268         ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
1269         if (ret < 0) {
1270                 PERROR("kernctl_get_events_discarded");
1271                 goto end;
1272         }
1273         if (discarded < stream->last_discarded_events) {
1274                 /*
1275                  * Overflow has occurred. We assume only one wrap-around
1276                  * has occurred.
1277                  */
1278                 stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
1279                         stream->last_discarded_events + discarded;
1280         } else {
1281                 stream->chan->discarded_events += discarded -
1282                         stream->last_discarded_events;
1283         }
1284         stream->last_discarded_events = discarded;
1285         ret = 0;
1286
1287 end:
1288         return ret;
1289 }
1290
1291 /*
1292  * Check if the local version of the metadata stream matches with the version
1293  * of the metadata stream in the kernel. If it was updated, set the reset flag
1294  * on the stream.
1295  */
1296 static
1297 int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
1298 {
1299         int ret;
1300         uint64_t cur_version;
1301
1302         ret = kernctl_get_metadata_version(infd, &cur_version);
1303         if (ret < 0) {
1304                 if (ret == -ENOTTY) {
1305                         /*
1306                          * LTTng-modules does not implement this
1307                          * command.
1308                          */
1309                         ret = 0;
1310                         goto end;
1311                 }
1312                 ERR("Failed to get the metadata version");
1313                 goto end;
1314         }
1315
1316         if (stream->metadata_version == cur_version) {
1317                 ret = 0;
1318                 goto end;
1319         }
1320
1321         DBG("New metadata version detected");
1322         stream->metadata_version = cur_version;
1323         stream->reset_metadata_flag = 1;
1324         ret = 0;
1325
1326 end:
1327         return ret;
1328 }
1329
1330 /*
1331  * Consume data on a file descriptor and write it on a trace file.
1332  */
1333 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1334                 struct lttng_consumer_local_data *ctx)
1335 {
1336         unsigned long len, subbuf_size, padding;
1337         int err, write_index = 1;
1338         ssize_t ret = 0;
1339         int infd = stream->wait_fd;
1340         struct ctf_packet_index index;
1341
1342         DBG("In read_subbuffer (infd : %d)", infd);
1343
1344         /* Get the next subbuffer */
1345         err = kernctl_get_next_subbuf(infd);
1346         if (err != 0) {
1347                 /*
1348                  * This is a debug message even for single-threaded consumer,
1349                  * because poll() have more relaxed criterions than get subbuf,
1350                  * so get_subbuf may fail for short race windows where poll()
1351                  * would issue wakeups.
1352                  */
1353                 DBG("Reserving sub buffer failed (everything is normal, "
1354                                 "it is due to concurrency)");
1355                 ret = err;
1356                 goto end;
1357         }
1358
1359         /* Get the full subbuffer size including padding */
1360         err = kernctl_get_padded_subbuf_size(infd, &len);
1361         if (err != 0) {
1362                 PERROR("Getting sub-buffer len failed.");
1363                 err = kernctl_put_subbuf(infd);
1364                 if (err != 0) {
1365                         if (err == -EFAULT) {
1366                                 PERROR("Error in unreserving sub buffer\n");
1367                         } else if (err == -EIO) {
1368                                 /* Should never happen with newer LTTng versions */
1369                                 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1370                         }
1371                         ret = err;
1372                         goto end;
1373                 }
1374                 ret = err;
1375                 goto end;
1376         }
1377
1378         if (!stream->metadata_flag) {
1379                 ret = get_index_values(&index, infd);
1380                 if (ret < 0) {
1381                         err = kernctl_put_subbuf(infd);
1382                         if (err != 0) {
1383                                 if (err == -EFAULT) {
1384                                         PERROR("Error in unreserving sub buffer\n");
1385                                 } else if (err == -EIO) {
1386                                         /* Should never happen with newer LTTng versions */
1387                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1388                                 }
1389                                 ret = err;
1390                                 goto end;
1391                         }
1392                         goto end;
1393                 }
1394                 ret = update_stream_stats(stream);
1395                 if (ret < 0) {
1396                         err = kernctl_put_subbuf(infd);
1397                         if (err != 0) {
1398                                 if (err == -EFAULT) {
1399                                         PERROR("Error in unreserving sub buffer\n");
1400                                 } else if (err == -EIO) {
1401                                         /* Should never happen with newer LTTng versions */
1402                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1403                                 }
1404                                 ret = err;
1405                                 goto end;
1406                         }
1407                         goto end;
1408                 }
1409         } else {
1410                 write_index = 0;
1411                 ret = metadata_stream_check_version(infd, stream);
1412                 if (ret < 0) {
1413                         err = kernctl_put_subbuf(infd);
1414                         if (err != 0) {
1415                                 if (err == -EFAULT) {
1416                                         PERROR("Error in unreserving sub buffer\n");
1417                                 } else if (err == -EIO) {
1418                                         /* Should never happen with newer LTTng versions */
1419                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1420                                 }
1421                                 ret = err;
1422                                 goto end;
1423                         }
1424                         goto end;
1425                 }
1426         }
1427
1428         switch (stream->chan->output) {
1429         case CONSUMER_CHANNEL_SPLICE:
1430                 /*
1431                  * XXX: The lttng-modules splice "actor" does not handle copying
1432                  * partial pages hence only using the subbuffer size without the
1433                  * padding makes the splice fail.
1434                  */
1435                 subbuf_size = len;
1436                 padding = 0;
1437
1438                 /* splice the subbuffer to the tracefile */
1439                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1440                                 padding, &index);
1441                 /*
1442                  * XXX: Splice does not support network streaming so the return value
1443                  * is simply checked against subbuf_size and not like the mmap() op.
1444                  */
1445                 if (ret != subbuf_size) {
1446                         /*
1447                          * display the error but continue processing to try
1448                          * to release the subbuffer
1449                          */
1450                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1451                                         ret, subbuf_size);
1452                         write_index = 0;
1453                 }
1454                 break;
1455         case CONSUMER_CHANNEL_MMAP:
1456                 /* Get subbuffer size without padding */
1457                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1458                 if (err != 0) {
1459                         PERROR("Getting sub-buffer len failed.");
1460                         err = kernctl_put_subbuf(infd);
1461                         if (err != 0) {
1462                                 if (err == -EFAULT) {
1463                                         PERROR("Error in unreserving sub buffer\n");
1464                                 } else if (err == -EIO) {
1465                                         /* Should never happen with newer LTTng versions */
1466                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1467                                 }
1468                                 ret = err;
1469                                 goto end;
1470                         }
1471                         ret = err;
1472                         goto end;
1473                 }
1474
1475                 /* Make sure the tracer is not gone mad on us! */
1476                 assert(len >= subbuf_size);
1477
1478                 padding = len - subbuf_size;
1479
1480                 /* write the subbuffer to the tracefile */
1481                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1482                                 padding, &index);
1483                 /*
1484                  * The mmap operation should write subbuf_size amount of data when
1485                  * network streaming or the full padding (len) size when we are _not_
1486                  * streaming.
1487                  */
1488                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1489                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1490                         /*
1491                          * Display the error but continue processing to try to release the
1492                          * subbuffer. This is a DBG statement since this is possible to
1493                          * happen without being a critical error.
1494                          */
1495                         DBG("Error writing to tracefile "
1496                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1497                                         ret, len, subbuf_size);
1498                         write_index = 0;
1499                 }
1500                 break;
1501         default:
1502                 ERR("Unknown output method");
1503                 ret = -EPERM;
1504         }
1505
1506         err = kernctl_put_next_subbuf(infd);
1507         if (err != 0) {
1508                 if (err == -EFAULT) {
1509                         PERROR("Error in unreserving sub buffer\n");
1510                 } else if (err == -EIO) {
1511                         /* Should never happen with newer LTTng versions */
1512                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1513                 }
1514                 ret = err;
1515                 goto end;
1516         }
1517
1518         /* Write index if needed. */
1519         if (!write_index) {
1520                 goto end;
1521         }
1522
1523         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1524                 /*
1525                  * In live, block until all the metadata is sent.
1526                  */
1527                 pthread_mutex_lock(&stream->metadata_timer_lock);
1528                 assert(!stream->missed_metadata_flush);
1529                 stream->waiting_on_metadata = true;
1530                 pthread_mutex_unlock(&stream->metadata_timer_lock);
1531
1532                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1533
1534                 pthread_mutex_lock(&stream->metadata_timer_lock);
1535                 stream->waiting_on_metadata = false;
1536                 if (stream->missed_metadata_flush) {
1537                         stream->missed_metadata_flush = false;
1538                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1539                         (void) consumer_flush_kernel_index(stream);
1540                 } else {
1541                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1542                 }
1543                 if (err < 0) {
1544                         goto end;
1545                 }
1546         }
1547
1548         err = consumer_stream_write_index(stream, &index);
1549         if (err < 0) {
1550                 goto end;
1551         }
1552
1553 end:
1554         return ret;
1555 }
1556
1557 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1558 {
1559         int ret;
1560
1561         assert(stream);
1562
1563         /*
1564          * Don't create anything if this is set for streaming or should not be
1565          * monitored.
1566          */
1567         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1568                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1569                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1570                                 stream->uid, stream->gid, NULL);
1571                 if (ret < 0) {
1572                         goto error;
1573                 }
1574                 stream->out_fd = ret;
1575                 stream->tracefile_size_current = 0;
1576
1577                 if (!stream->metadata_flag) {
1578                         struct lttng_index_file *index_file;
1579
1580                         index_file = lttng_index_file_create(stream->chan->pathname,
1581                                         stream->name, stream->uid, stream->gid,
1582                                         stream->chan->tracefile_size,
1583                                         stream->tracefile_count_current,
1584                                         CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1585                         if (!index_file) {
1586                                 goto error;
1587                         }
1588                         assert(!stream->index_file);
1589                         stream->index_file = index_file;
1590                 }
1591         }
1592
1593         if (stream->output == LTTNG_EVENT_MMAP) {
1594                 /* get the len of the mmap region */
1595                 unsigned long mmap_len;
1596
1597                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1598                 if (ret != 0) {
1599                         PERROR("kernctl_get_mmap_len");
1600                         goto error_close_fd;
1601                 }
1602                 stream->mmap_len = (size_t) mmap_len;
1603
1604                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1605                                 MAP_PRIVATE, stream->wait_fd, 0);
1606                 if (stream->mmap_base == MAP_FAILED) {
1607                         PERROR("Error mmaping");
1608                         ret = -1;
1609                         goto error_close_fd;
1610                 }
1611         }
1612
1613         /* we return 0 to let the library handle the FD internally */
1614         return 0;
1615
1616 error_close_fd:
1617         if (stream->out_fd >= 0) {
1618                 int err;
1619
1620                 err = close(stream->out_fd);
1621                 assert(!err);
1622                 stream->out_fd = -1;
1623         }
1624 error:
1625         return ret;
1626 }
1627
1628 /*
1629  * Check if data is still being extracted from the buffers for a specific
1630  * stream. Consumer data lock MUST be acquired before calling this function
1631  * and the stream lock.
1632  *
1633  * Return 1 if the traced data are still getting read else 0 meaning that the
1634  * data is available for trace viewer reading.
1635  */
1636 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1637 {
1638         int ret;
1639
1640         assert(stream);
1641
1642         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1643                 ret = 0;
1644                 goto end;
1645         }
1646
1647         ret = kernctl_get_next_subbuf(stream->wait_fd);
1648         if (ret == 0) {
1649                 /* There is still data so let's put back this subbuffer. */
1650                 ret = kernctl_put_subbuf(stream->wait_fd);
1651                 assert(ret == 0);
1652                 ret = 1;   /* Data is pending */
1653                 goto end;
1654         }
1655
1656         /* Data is NOT pending and ready to be read. */
1657         ret = 0;
1658
1659 end:
1660         return ret;
1661 }