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