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 extern volatile int consumer_quit;
  52
  53 /*
  54  * Take a snapshot for a specific fd
  55  *
  56  * Returns 0 on success, < 0 on error
  57  */
  58 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  59 {
  60         int ret = 0;
  61         int infd = stream->wait_fd;
  62
  63         ret = kernctl_snapshot(infd);
  64         if (ret != 0) {
  65                 PERROR("Getting sub-buffer snapshot.");
  66         }
  67
  68         return ret;
  69 }
  70
  71 /*
  72  * Sample consumed and produced positions for a specific fd.
  73  *
  74  * Returns 0 on success, < 0 on error.
  75  */
  76 int lttng_kconsumer_sample_snapshot_positions(
  77                 struct lttng_consumer_stream *stream)
  78 {
  79         assert(stream);
  80
  81         return kernctl_snapshot_sample_positions(stream->wait_fd);
  82 }
  83
  84 /*
  85  * Get the produced position
  86  *
  87  * Returns 0 on success, < 0 on error
  88  */
  89 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  90                 unsigned long *pos)
  91 {
  92         int ret;
  93         int infd = stream->wait_fd;
  94
  95         ret = kernctl_snapshot_get_produced(infd, pos);
  96         if (ret != 0) {
  97                 PERROR("kernctl_snapshot_get_produced");
  98         }
  99
 100         return ret;
 101 }
 102
 103 /*
 104  * Get the consumerd position
 105  *
 106  * Returns 0 on success, < 0 on error
 107  */
 108 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
 109                 unsigned long *pos)
 110 {
 111         int ret;
 112         int infd = stream->wait_fd;
 113
 114         ret = kernctl_snapshot_get_consumed(infd, pos);
 115         if (ret != 0) {
 116                 PERROR("kernctl_snapshot_get_consumed");
 117         }
 118
 119         return ret;
 120 }
 121
 122 /*
 123  * Take a snapshot of all the stream of a channel
 124  *
 125  * Returns 0 on success, < 0 on error
 126  */
 127 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 128                 uint64_t relayd_id, uint64_t nb_packets_per_stream,
 129                 struct lttng_consumer_local_data *ctx)
 130 {
 131         int ret;
 132         struct lttng_consumer_channel *channel;
 133         struct lttng_consumer_stream *stream;
 134
 135         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 136
 137         rcu_read_lock();
 138
 139         channel = consumer_find_channel(key);
 140         if (!channel) {
 141                 ERR("No channel found for key %" PRIu64, key);
 142                 ret = -1;
 143                 goto end;
 144         }
 145
 146         /* Splice is not supported yet for channel snapshot. */
 147         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 148                 ERR("Unsupported output %d", channel->output);
 149                 ret = -1;
 150                 goto end;
 151         }
 152
 153         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 154                 /* Are we at a position _before_ the first available packet ? */
 155                 bool before_first_packet = true;
 156                 unsigned long consumed_pos, produced_pos;
 157
 158                 health_code_update();
 159
 160                 /*
 161                  * Lock stream because we are about to change its state.
 162                  */
 163                 pthread_mutex_lock(&stream->lock);
 164
 165                 /*
 166                  * Assign the received relayd ID so we can use it for streaming. The streams
 167                  * are not visible to anyone so this is OK to change it.
 168                  */
 169                 stream->net_seq_idx = relayd_id;
 170                 channel->relayd_id = relayd_id;
 171                 if (relayd_id != (uint64_t) -1ULL) {
 172                         ret = consumer_send_relayd_stream(stream, path);
 173                         if (ret < 0) {
 174                                 ERR("sending stream to relayd");
 175                                 goto end_unlock;
 176                         }
 177                 } else {
 178                         ret = utils_create_stream_file(path, stream->name,
 179                                         stream->chan->tracefile_size,
 180                                         stream->tracefile_count_current,
 181                                         stream->uid, stream->gid, NULL);
 182                         if (ret < 0) {
 183                                 ERR("utils_create_stream_file");
 184                                 goto end_unlock;
 185                         }
 186
 187                         stream->out_fd = ret;
 188                         stream->tracefile_size_current = 0;
 189
 190                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 191                                         path, stream->name, stream->key);
 192                 }
 193                 if (relayd_id != -1ULL) {
 194                         ret = consumer_send_relayd_streams_sent(relayd_id);
 195                         if (ret < 0) {
 196                                 ERR("sending streams sent to relayd");
 197                                 goto end_unlock;
 198                         }
 199                         channel->streams_sent_to_relayd = true;
 200                 }
 201
 202                 ret = kernctl_buffer_flush_empty(stream->wait_fd);
 203                 if (ret < 0) {
 204                         /*
 205                          * Doing a buffer flush which does not take into
 206                          * account empty packets. This is not perfect
 207                          * for stream intersection, but required as a
 208                          * fall-back when "flush_empty" is not
 209                          * implemented by lttng-modules.
 210                          */
 211                         ret = kernctl_buffer_flush(stream->wait_fd);
 212                         if (ret < 0) {
 213                                 ERR("Failed to flush kernel stream");
 214                                 goto end_unlock;
 215                         }
 216                         goto end_unlock;
 217                 }
 218
 219                 ret = lttng_kconsumer_take_snapshot(stream);
 220                 if (ret < 0) {
 221                         ERR("Taking kernel snapshot");
 222                         goto end_unlock;
 223                 }
 224
 225                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 226                 if (ret < 0) {
 227                         ERR("Produced kernel snapshot position");
 228                         goto end_unlock;
 229                 }
 230
 231                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 232                 if (ret < 0) {
 233                         ERR("Consumerd kernel snapshot position");
 234                         goto end_unlock;
 235                 }
 236
 237                 if (stream->max_sb_size == 0) {
 238                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 239                                         &stream->max_sb_size);
 240                         if (ret < 0) {
 241                                 ERR("Getting kernel max_sb_size");
 242                                 goto end_unlock;
 243                         }
 244                 }
 245
 246                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 247                                 produced_pos, nb_packets_per_stream,
 248                                 stream->max_sb_size);
 249
 250                 while (consumed_pos < produced_pos) {
 251                         ssize_t read_len;
 252                         unsigned long len, padded_len;
 253                         int lost_packet = 0;
 254
 255                         health_code_update();
 256
 257                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 258
 259                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 260                         if (ret < 0) {
 261                                 if (ret != -EAGAIN) {
 262                                         PERROR("kernctl_get_subbuf snapshot");
 263                                         goto end_unlock;
 264                                 }
 265                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 266                                 consumed_pos += stream->max_sb_size;
 267
 268                                 /*
 269                                  * Start accounting lost packets only when we
 270                                  * already have extracted packets (to match the
 271                                  * content of the final snapshot).
 272                                  */
 273                                 if (!before_first_packet) {
 274                                         lost_packet = 1;
 275                                 }
 276                                 continue;
 277                         }
 278
 279                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 280                         if (ret < 0) {
 281                                 ERR("Snapshot kernctl_get_subbuf_size");
 282                                 goto error_put_subbuf;
 283                         }
 284
 285                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 286                         if (ret < 0) {
 287                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 288                                 goto error_put_subbuf;
 289                         }
 290
 291                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 292                                         padded_len - len, NULL);
 293                         /*
 294                          * We write the padded len in local tracefiles but the data len
 295                          * when using a relay. Display the error but continue processing
 296                          * to try to release the subbuffer.
 297                          */
 298                         if (relayd_id != (uint64_t) -1ULL) {
 299                                 if (read_len != len) {
 300                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 301                                                         read_len, len);
 302                                 }
 303                         } else {
 304                                 if (read_len != padded_len) {
 305                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 306                                                         read_len, padded_len);
 307                                 }
 308                         }
 309
 310                         ret = kernctl_put_subbuf(stream->wait_fd);
 311                         if (ret < 0) {
 312                                 ERR("Snapshot kernctl_put_subbuf");
 313                                 goto end_unlock;
 314                         }
 315                         consumed_pos += stream->max_sb_size;
 316
 317                         /*
 318                          * Only account lost packets located between
 319                          * succesfully extracted packets (do not account before
 320                          * and after since they are not visible in the
 321                          * resulting snapshot).
 322                          */
 323                         stream->chan->lost_packets += lost_packet;
 324                         lost_packet = 0;
 325                         before_first_packet = false;
 326                 }
 327
 328                 if (relayd_id == (uint64_t) -1ULL) {
 329                         if (stream->out_fd >= 0) {
 330                                 ret = close(stream->out_fd);
 331                                 if (ret < 0) {
 332                                         PERROR("Kernel consumer snapshot close out_fd");
 333                                         goto end_unlock;
 334                                 }
 335                                 stream->out_fd = -1;
 336                         }
 337                 } else {
 338                         close_relayd_stream(stream);
 339                         stream->net_seq_idx = (uint64_t) -1ULL;
 340                 }
 341                 pthread_mutex_unlock(&stream->lock);
 342         }
 343
 344         /* All good! */
 345         ret = 0;
 346         goto end;
 347
 348 error_put_subbuf:
 349         ret = kernctl_put_subbuf(stream->wait_fd);
 350         if (ret < 0) {
 351                 ERR("Snapshot kernctl_put_subbuf error path");
 352         }
 353 end_unlock:
 354         pthread_mutex_unlock(&stream->lock);
 355 end:
 356         rcu_read_unlock();
 357         return ret;
 358 }
 359
 360 /*
 361  * Read the whole metadata available for a snapshot.
 362  *
 363  * Returns 0 on success, < 0 on error
 364  */
 365 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 366                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 367 {
 368         int ret, use_relayd = 0;
 369         ssize_t ret_read;
 370         struct lttng_consumer_channel *metadata_channel;
 371         struct lttng_consumer_stream *metadata_stream;
 372
 373         assert(ctx);
 374
 375         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 376                         key, path);
 377
 378         rcu_read_lock();
 379
 380         metadata_channel = consumer_find_channel(key);
 381         if (!metadata_channel) {
 382                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 383                 ret = -1;
 384                 goto error;
 385         }
 386
 387         metadata_stream = metadata_channel->metadata_stream;
 388         assert(metadata_stream);
 389
 390         /* Flag once that we have a valid relayd for the stream. */
 391         if (relayd_id != (uint64_t) -1ULL) {
 392                 use_relayd = 1;
 393         }
 394
 395         if (use_relayd) {
 396                 ret = consumer_send_relayd_stream(metadata_stream, path);
 397                 if (ret < 0) {
 398                         goto error;
 399                 }
 400         } else {
 401                 ret = utils_create_stream_file(path, metadata_stream->name,
 402                                 metadata_stream->chan->tracefile_size,
 403                                 metadata_stream->tracefile_count_current,
 404                                 metadata_stream->uid, metadata_stream->gid, NULL);
 405                 if (ret < 0) {
 406                         goto error;
 407                 }
 408                 metadata_stream->out_fd = ret;
 409         }
 410
 411         do {
 412                 health_code_update();
 413
 414                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 415                 if (ret_read < 0) {
 416                         if (ret_read != -EAGAIN) {
 417                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 418                                                 ret_read);
 419                                 goto error;
 420                         }
 421                         /* ret_read is negative at this point so we will exit the loop. */
 422                         continue;
 423                 }
 424         } while (ret_read >= 0);
 425
 426         if (use_relayd) {
 427                 close_relayd_stream(metadata_stream);
 428                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 429         } else {
 430                 if (metadata_stream->out_fd >= 0) {
 431                         ret = close(metadata_stream->out_fd);
 432                         if (ret < 0) {
 433                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 434                                 /*
 435                                  * Don't go on error here since the snapshot was successful at this
 436                                  * point but somehow the close failed.
 437                                  */
 438                         }
 439                         metadata_stream->out_fd = -1;
 440                 }
 441         }
 442
 443         ret = 0;
 444
 445         cds_list_del(&metadata_stream->send_node);
 446         consumer_stream_destroy(metadata_stream, NULL);
 447         metadata_channel->metadata_stream = NULL;
 448 error:
 449         rcu_read_unlock();
 450         return ret;
 451 }
 452
 453 /*
 454  * Receive command from session daemon and process it.
 455  *
 456  * Return 1 on success else a negative value or 0.
 457  */
 458 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 459                 int sock, struct pollfd *consumer_sockpoll)
 460 {
 461         ssize_t ret;
 462         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 463         struct lttcomm_consumer_msg msg;
 464
 465         health_code_update();
 466
 467         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 468         if (ret != sizeof(msg)) {
 469                 if (ret > 0) {
 470                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 471                         ret = -1;
 472                 }
 473                 return ret;
 474         }
 475
 476         health_code_update();
 477
 478         /* Deprecated command */
 479         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 480
 481         health_code_update();
 482
 483         /* relayd needs RCU read-side protection */
 484         rcu_read_lock();
 485
 486         switch (msg.cmd_type) {
 487         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 488         {
 489                 /* Session daemon status message are handled in the following call. */
 490                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 491                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 492                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 493                                  msg.u.relayd_sock.relayd_session_id);
 494                 goto end_nosignal;
 495         }
 496         case LTTNG_CONSUMER_ADD_CHANNEL:
 497         {
 498                 struct lttng_consumer_channel *new_channel;
 499                 int ret_recv;
 500
 501                 health_code_update();
 502
 503                 /* First send a status message before receiving the fds. */
 504                 ret = consumer_send_status_msg(sock, ret_code);
 505                 if (ret < 0) {
 506                         /* Somehow, the session daemon is not responding anymore. */
 507                         goto error_fatal;
 508                 }
 509
 510                 health_code_update();
 511
 512                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 513                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 514                                 msg.u.channel.session_id, msg.u.channel.pathname,
 515                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 516                                 msg.u.channel.relayd_id, msg.u.channel.output,
 517                                 msg.u.channel.tracefile_size,
 518                                 msg.u.channel.tracefile_count, 0,
 519                                 msg.u.channel.monitor,
 520                                 msg.u.channel.live_timer_interval,
 521                                 NULL, NULL);
 522                 if (new_channel == NULL) {
 523                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 524                         goto end_nosignal;
 525                 }
 526                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 527                 switch (msg.u.channel.output) {
 528                 case LTTNG_EVENT_SPLICE:
 529                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 530                         break;
 531                 case LTTNG_EVENT_MMAP:
 532                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 533                         break;
 534                 default:
 535                         ERR("Channel output unknown %d", msg.u.channel.output);
 536                         goto end_nosignal;
 537                 }
 538
 539                 /* Translate and save channel type. */
 540                 switch (msg.u.channel.type) {
 541                 case CONSUMER_CHANNEL_TYPE_DATA:
 542                 case CONSUMER_CHANNEL_TYPE_METADATA:
 543                         new_channel->type = msg.u.channel.type;
 544                         break;
 545                 default:
 546                         assert(0);
 547                         goto end_nosignal;
 548                 };
 549
 550                 health_code_update();
 551
 552                 if (ctx->on_recv_channel != NULL) {
 553                         ret_recv = ctx->on_recv_channel(new_channel);
 554                         if (ret_recv == 0) {
 555                                 ret = consumer_add_channel(new_channel, ctx);
 556                         } else if (ret_recv < 0) {
 557                                 goto end_nosignal;
 558                         }
 559                 } else {
 560                         ret = consumer_add_channel(new_channel, ctx);
 561                 }
 562                 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
 563                         int monitor_start_ret;
 564
 565                         DBG("Consumer starting monitor timer");
 566                         consumer_timer_live_start(new_channel,
 567                                         msg.u.channel.live_timer_interval);
 568                         monitor_start_ret = consumer_timer_monitor_start(
 569                                         new_channel,
 570                                         msg.u.channel.monitor_timer_interval);
 571                         if (monitor_start_ret < 0) {
 572                                 ERR("Starting channel monitoring timer failed");
 573                                 goto end_nosignal;
 574                         }
 575
 576                 }
 577
 578                 health_code_update();
 579
 580                 /* If we received an error in add_channel, we need to report it. */
 581                 if (ret < 0) {
 582                         ret = consumer_send_status_msg(sock, ret);
 583                         if (ret < 0) {
 584                                 goto error_fatal;
 585                         }
 586                         goto end_nosignal;
 587                 }
 588
 589                 goto end_nosignal;
 590         }
 591         case LTTNG_CONSUMER_ADD_STREAM:
 592         {
 593                 int fd;
 594                 struct lttng_pipe *stream_pipe;
 595                 struct lttng_consumer_stream *new_stream;
 596                 struct lttng_consumer_channel *channel;
 597                 int alloc_ret = 0;
 598
 599                 /*
 600                  * Get stream's channel reference. Needed when adding the stream to the
 601                  * global hash table.
 602                  */
 603                 channel = consumer_find_channel(msg.u.stream.channel_key);
 604                 if (!channel) {
 605                         /*
 606                          * We could not find the channel. Can happen if cpu hotplug
 607                          * happens while tearing down.
 608                          */
 609                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 610                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 611                 }
 612
 613                 health_code_update();
 614
 615                 /* First send a status message before receiving the fds. */
 616                 ret = consumer_send_status_msg(sock, ret_code);
 617                 if (ret < 0) {
 618                         /* Somehow, the session daemon is not responding anymore. */
 619                         goto error_fatal;
 620                 }
 621
 622                 health_code_update();
 623
 624                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 625                         /* Channel was not found. */
 626                         goto end_nosignal;
 627                 }
 628
 629                 /* Blocking call */
 630                 health_poll_entry();
 631                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 632                 health_poll_exit();
 633                 if (ret) {
 634                         goto error_fatal;
 635                 }
 636
 637                 health_code_update();
 638
 639                 /* Get stream file descriptor from socket */
 640                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 641                 if (ret != sizeof(fd)) {
 642                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 643                         rcu_read_unlock();
 644                         return ret;
 645                 }
 646
 647                 health_code_update();
 648
 649                 /*
 650                  * Send status code to session daemon only if the recv works. If the
 651                  * above recv() failed, the session daemon is notified through the
 652                  * error socket and the teardown is eventually done.
 653                  */
 654                 ret = consumer_send_status_msg(sock, ret_code);
 655                 if (ret < 0) {
 656                         /* Somehow, the session daemon is not responding anymore. */
 657                         goto end_nosignal;
 658                 }
 659
 660                 health_code_update();
 661
 662                 new_stream = consumer_allocate_stream(channel->key,
 663                                 fd,
 664                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 665                                 channel->name,
 666                                 channel->uid,
 667                                 channel->gid,
 668                                 channel->relayd_id,
 669                                 channel->session_id,
 670                                 msg.u.stream.cpu,
 671                                 &alloc_ret,
 672                                 channel->type,
 673                                 channel->monitor);
 674                 if (new_stream == NULL) {
 675                         switch (alloc_ret) {
 676                         case -ENOMEM:
 677                         case -EINVAL:
 678                         default:
 679                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 680                                 break;
 681                         }
 682                         goto end_nosignal;
 683                 }
 684
 685                 new_stream->chan = channel;
 686                 new_stream->wait_fd = fd;
 687                 switch (channel->output) {
 688                 case CONSUMER_CHANNEL_SPLICE:
 689                         new_stream->output = LTTNG_EVENT_SPLICE;
 690                         ret = utils_create_pipe(new_stream->splice_pipe);
 691                         if (ret < 0) {
 692                                 goto end_nosignal;
 693                         }
 694                         break;
 695                 case CONSUMER_CHANNEL_MMAP:
 696                         new_stream->output = LTTNG_EVENT_MMAP;
 697                         break;
 698                 default:
 699                         ERR("Stream output unknown %d", channel->output);
 700                         goto end_nosignal;
 701                 }
 702
 703                 /*
 704                  * We've just assigned the channel to the stream so increment the
 705                  * refcount right now. We don't need to increment the refcount for
 706                  * streams in no monitor because we handle manually the cleanup of
 707                  * those. It is very important to make sure there is NO prior
 708                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 709                  */
 710                 if (channel->monitor) {
 711                         uatomic_inc(&new_stream->chan->refcount);
 712                 }
 713
 714                 /*
 715                  * The buffer flush is done on the session daemon side for the kernel
 716                  * so no need for the stream "hangup_flush_done" variable to be
 717                  * tracked. This is important for a kernel stream since we don't rely
 718                  * on the flush state of the stream to read data. It's not the case for
 719                  * user space tracing.
 720                  */
 721                 new_stream->hangup_flush_done = 0;
 722
 723                 health_code_update();
 724
 725                 if (ctx->on_recv_stream) {
 726                         ret = ctx->on_recv_stream(new_stream);
 727                         if (ret < 0) {
 728                                 consumer_stream_free(new_stream);
 729                                 goto end_nosignal;
 730                         }
 731                 }
 732
 733                 health_code_update();
 734
 735                 if (new_stream->metadata_flag) {
 736                         channel->metadata_stream = new_stream;
 737                 }
 738
 739                 /* Do not monitor this stream. */
 740                 if (!channel->monitor) {
 741                         DBG("Kernel consumer add stream %s in no monitor mode with "
 742                                         "relayd id %" PRIu64, new_stream->name,
 743                                         new_stream->net_seq_idx);
 744                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 745                         break;
 746                 }
 747
 748                 /* Send stream to relayd if the stream has an ID. */
 749                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 750                         ret = consumer_send_relayd_stream(new_stream,
 751                                         new_stream->chan->pathname);
 752                         if (ret < 0) {
 753                                 consumer_stream_free(new_stream);
 754                                 goto end_nosignal;
 755                         }
 756
 757                         /*
 758                          * If adding an extra stream to an already
 759                          * existing channel (e.g. cpu hotplug), we need
 760                          * to send the "streams_sent" command to relayd.
 761                          */
 762                         if (channel->streams_sent_to_relayd) {
 763                                 ret = consumer_send_relayd_streams_sent(
 764                                                 new_stream->net_seq_idx);
 765                                 if (ret < 0) {
 766                                         goto end_nosignal;
 767                                 }
 768                         }
 769                 }
 770
 771                 /* Get the right pipe where the stream will be sent. */
 772                 if (new_stream->metadata_flag) {
 773                         ret = consumer_add_metadata_stream(new_stream);
 774                         if (ret) {
 775                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 776                                                 new_stream->key);
 777                                 consumer_stream_free(new_stream);
 778                                 goto end_nosignal;
 779                         }
 780                         stream_pipe = ctx->consumer_metadata_pipe;
 781                 } else {
 782                         ret = consumer_add_data_stream(new_stream);
 783                         if (ret) {
 784                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 785                                                 new_stream->key);
 786                                 consumer_stream_free(new_stream);
 787                                 goto end_nosignal;
 788                         }
 789                         stream_pipe = ctx->consumer_data_pipe;
 790                 }
 791
 792                 /* Vitible to other threads */
 793                 new_stream->globally_visible = 1;
 794
 795                 health_code_update();
 796
 797                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 798                 if (ret < 0) {
 799                         ERR("Consumer write %s stream to pipe %d",
 800                                         new_stream->metadata_flag ? "metadata" : "data",
 801                                         lttng_pipe_get_writefd(stream_pipe));
 802                         if (new_stream->metadata_flag) {
 803                                 consumer_del_stream_for_metadata(new_stream);
 804                         } else {
 805                                 consumer_del_stream_for_data(new_stream);
 806                         }
 807                         goto end_nosignal;
 808                 }
 809
 810                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 811                                 new_stream->name, fd, new_stream->relayd_stream_id);
 812                 break;
 813         }
 814         case LTTNG_CONSUMER_STREAMS_SENT:
 815         {
 816                 struct lttng_consumer_channel *channel;
 817
 818                 /*
 819                  * Get stream's channel reference. Needed when adding the stream to the
 820                  * global hash table.
 821                  */
 822                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 823                 if (!channel) {
 824                         /*
 825                          * We could not find the channel. Can happen if cpu hotplug
 826                          * happens while tearing down.
 827                          */
 828                         ERR("Unable to find channel key %" PRIu64,
 829                                         msg.u.sent_streams.channel_key);
 830                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 831                 }
 832
 833                 health_code_update();
 834
 835                 /*
 836                  * Send status code to session daemon.
 837                  */
 838                 ret = consumer_send_status_msg(sock, ret_code);
 839                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 840                         /* Somehow, the session daemon is not responding anymore. */
 841                         goto end_nosignal;
 842                 }
 843
 844                 health_code_update();
 845
 846                 /*
 847                  * We should not send this message if we don't monitor the
 848                  * streams in this channel.
 849                  */
 850                 if (!channel->monitor) {
 851                         break;
 852                 }
 853
 854                 health_code_update();
 855                 /* Send stream to relayd if the stream has an ID. */
 856                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 857                         ret = consumer_send_relayd_streams_sent(
 858                                         msg.u.sent_streams.net_seq_idx);
 859                         if (ret < 0) {
 860                                 goto end_nosignal;
 861                         }
 862                         channel->streams_sent_to_relayd = true;
 863                 }
 864                 break;
 865         }
 866         case LTTNG_CONSUMER_UPDATE_STREAM:
 867         {
 868                 rcu_read_unlock();
 869                 return -ENOSYS;
 870         }
 871         case LTTNG_CONSUMER_DESTROY_RELAYD:
 872         {
 873                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 874                 struct consumer_relayd_sock_pair *relayd;
 875
 876                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 877
 878                 /* Get relayd reference if exists. */
 879                 relayd = consumer_find_relayd(index);
 880                 if (relayd == NULL) {
 881                         DBG("Unable to find relayd %" PRIu64, index);
 882                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 883                 }
 884
 885                 /*
 886                  * Each relayd socket pair has a refcount of stream attached to it
 887                  * which tells if the relayd is still active or not depending on the
 888                  * refcount value.
 889                  *
 890                  * This will set the destroy flag of the relayd object and destroy it
 891                  * if the refcount reaches zero when called.
 892                  *
 893                  * The destroy can happen either here or when a stream fd hangs up.
 894                  */
 895                 if (relayd) {
 896                         consumer_flag_relayd_for_destroy(relayd);
 897                 }
 898
 899                 health_code_update();
 900
 901                 ret = consumer_send_status_msg(sock, ret_code);
 902                 if (ret < 0) {
 903                         /* Somehow, the session daemon is not responding anymore. */
 904                         goto error_fatal;
 905                 }
 906
 907                 goto end_nosignal;
 908         }
 909         case LTTNG_CONSUMER_DATA_PENDING:
 910         {
 911                 int32_t ret;
 912                 uint64_t id = msg.u.data_pending.session_id;
 913
 914                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 915
 916                 ret = consumer_data_pending(id);
 917
 918                 health_code_update();
 919
 920                 /* Send back returned value to session daemon */
 921                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 922                 if (ret < 0) {
 923                         PERROR("send data pending ret code");
 924                         goto error_fatal;
 925                 }
 926
 927                 /*
 928                  * No need to send back a status message since the data pending
 929                  * returned value is the response.
 930                  */
 931                 break;
 932         }
 933         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 934         {
 935                 if (msg.u.snapshot_channel.metadata == 1) {
 936                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 937                                         msg.u.snapshot_channel.pathname,
 938                                         msg.u.snapshot_channel.relayd_id, ctx);
 939                         if (ret < 0) {
 940                                 ERR("Snapshot metadata failed");
 941                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 942                         }
 943                 } else {
 944                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 945                                         msg.u.snapshot_channel.pathname,
 946                                         msg.u.snapshot_channel.relayd_id,
 947                                         msg.u.snapshot_channel.nb_packets_per_stream,
 948                                         ctx);
 949                         if (ret < 0) {
 950                                 ERR("Snapshot channel failed");
 951                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 952                         }
 953                 }
 954
 955                 health_code_update();
 956
 957                 ret = consumer_send_status_msg(sock, ret_code);
 958                 if (ret < 0) {
 959                         /* Somehow, the session daemon is not responding anymore. */
 960                         goto end_nosignal;
 961                 }
 962                 break;
 963         }
 964         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 965         {
 966                 uint64_t key = msg.u.destroy_channel.key;
 967                 struct lttng_consumer_channel *channel;
 968
 969                 channel = consumer_find_channel(key);
 970                 if (!channel) {
 971                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 972                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 973                 }
 974
 975                 health_code_update();
 976
 977                 ret = consumer_send_status_msg(sock, ret_code);
 978                 if (ret < 0) {
 979                         /* Somehow, the session daemon is not responding anymore. */
 980                         goto end_nosignal;
 981                 }
 982
 983                 health_code_update();
 984
 985                 /* Stop right now if no channel was found. */
 986                 if (!channel) {
 987                         goto end_nosignal;
 988                 }
 989
 990                 /*
 991                  * This command should ONLY be issued for channel with streams set in
 992                  * no monitor mode.
 993                  */
 994                 assert(!channel->monitor);
 995
 996                 /*
 997                  * The refcount should ALWAYS be 0 in the case of a channel in no
 998                  * monitor mode.
 999                  */
1000                 assert(!uatomic_sub_return(&channel->refcount, 1));
1001
1002                 consumer_del_channel(channel);
1003
1004                 goto end_nosignal;
1005         }
1006         case LTTNG_CONSUMER_DISCARDED_EVENTS:
1007         {
1008                 uint64_t ret;
1009                 struct lttng_consumer_channel *channel;
1010                 uint64_t id = msg.u.discarded_events.session_id;
1011                 uint64_t key = msg.u.discarded_events.channel_key;
1012
1013                 DBG("Kernel consumer discarded events command for session id %"
1014                                 PRIu64 ", channel key %" PRIu64, id, key);
1015
1016                 channel = consumer_find_channel(key);
1017                 if (!channel) {
1018                         ERR("Kernel consumer discarded events channel %"
1019                                         PRIu64 " not found", key);
1020                         ret = 0;
1021                 } else {
1022                         ret = channel->discarded_events;
1023                 }
1024
1025                 health_code_update();
1026
1027                 /* Send back returned value to session daemon */
1028                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
1029                 if (ret < 0) {
1030                         PERROR("send discarded events");
1031                         goto error_fatal;
1032                 }
1033
1034                 break;
1035         }
1036         case LTTNG_CONSUMER_LOST_PACKETS:
1037         {
1038                 uint64_t ret;
1039                 struct lttng_consumer_channel *channel;
1040                 uint64_t id = msg.u.lost_packets.session_id;
1041                 uint64_t key = msg.u.lost_packets.channel_key;
1042
1043                 DBG("Kernel consumer lost packets command for session id %"
1044                                 PRIu64 ", channel key %" PRIu64, id, key);
1045
1046                 channel = consumer_find_channel(key);
1047                 if (!channel) {
1048                         ERR("Kernel consumer lost packets channel %"
1049                                         PRIu64 " not found", key);
1050                         ret = 0;
1051                 } else {
1052                         ret = channel->lost_packets;
1053                 }
1054
1055                 health_code_update();
1056
1057                 /* Send back returned value to session daemon */
1058                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
1059                 if (ret < 0) {
1060                         PERROR("send lost packets");
1061                         goto error_fatal;
1062                 }
1063
1064                 break;
1065         }
1066         case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1067         {
1068                 int channel_monitor_pipe;
1069
1070                 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1071                 /* Successfully received the command's type. */
1072                 ret = consumer_send_status_msg(sock, ret_code);
1073                 if (ret < 0) {
1074                         goto error_fatal;
1075                 }
1076
1077                 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1078                                 1);
1079                 if (ret != sizeof(channel_monitor_pipe)) {
1080                         ERR("Failed to receive channel monitor pipe");
1081                         goto error_fatal;
1082                 }
1083
1084                 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1085                 ret = consumer_timer_thread_set_channel_monitor_pipe(
1086                                 channel_monitor_pipe);
1087                 if (!ret) {
1088                         int flags;
1089
1090                         ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1091                         /* Set the pipe as non-blocking. */
1092                         ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1093                         if (ret == -1) {
1094                                 PERROR("fcntl get flags of the channel monitoring pipe");
1095                                 goto error_fatal;
1096                         }
1097                         flags = ret;
1098
1099                         ret = fcntl(channel_monitor_pipe, F_SETFL,
1100                                         flags | O_NONBLOCK);
1101                         if (ret == -1) {
1102                                 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1103                                 goto error_fatal;
1104                         }
1105                         DBG("Channel monitor pipe set as non-blocking");
1106                 } else {
1107                         ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1108                 }
1109                 ret = consumer_send_status_msg(sock, ret_code);
1110                 if (ret < 0) {
1111                         goto error_fatal;
1112                 }
1113                 break;
1114         }
1115         default:
1116                 goto end_nosignal;
1117         }
1118
1119 end_nosignal:
1120         rcu_read_unlock();
1121
1122         /*
1123          * Return 1 to indicate success since the 0 value can be a socket
1124          * shutdown during the recv() or send() call.
1125          */
1126         health_code_update();
1127         return 1;
1128
1129 error_fatal:
1130         rcu_read_unlock();
1131         /* This will issue a consumer stop. */
1132         return -1;
1133 }
1134
1135 /*
1136  * Populate index values of a kernel stream. Values are set in big endian order.
1137  *
1138  * Return 0 on success or else a negative value.
1139  */
1140 static int get_index_values(struct ctf_packet_index *index, int infd)
1141 {
1142         int ret;
1143
1144         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
1145         if (ret < 0) {
1146                 PERROR("kernctl_get_timestamp_begin");
1147                 goto error;
1148         }
1149         index->timestamp_begin = htobe64(index->timestamp_begin);
1150
1151         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
1152         if (ret < 0) {
1153                 PERROR("kernctl_get_timestamp_end");
1154                 goto error;
1155         }
1156         index->timestamp_end = htobe64(index->timestamp_end);
1157
1158         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
1159         if (ret < 0) {
1160                 PERROR("kernctl_get_events_discarded");
1161                 goto error;
1162         }
1163         index->events_discarded = htobe64(index->events_discarded);
1164
1165         ret = kernctl_get_content_size(infd, &index->content_size);
1166         if (ret < 0) {
1167                 PERROR("kernctl_get_content_size");
1168                 goto error;
1169         }
1170         index->content_size = htobe64(index->content_size);
1171
1172         ret = kernctl_get_packet_size(infd, &index->packet_size);
1173         if (ret < 0) {
1174                 PERROR("kernctl_get_packet_size");
1175                 goto error;
1176         }
1177         index->packet_size = htobe64(index->packet_size);
1178
1179         ret = kernctl_get_stream_id(infd, &index->stream_id);
1180         if (ret < 0) {
1181                 PERROR("kernctl_get_stream_id");
1182                 goto error;
1183         }
1184         index->stream_id = htobe64(index->stream_id);
1185
1186         ret = kernctl_get_instance_id(infd, &index->stream_instance_id);
1187         if (ret < 0) {
1188                 if (ret == -ENOTTY) {
1189                         /* Command not implemented by lttng-modules. */
1190                         index->stream_instance_id = -1ULL;
1191                         ret = 0;
1192                 } else {
1193                         PERROR("kernctl_get_instance_id");
1194                         goto error;
1195                 }
1196         }
1197         index->stream_instance_id = htobe64(index->stream_instance_id);
1198
1199         ret = kernctl_get_sequence_number(infd, &index->packet_seq_num);
1200         if (ret < 0) {
1201                 if (ret == -ENOTTY) {
1202                         /* Command not implemented by lttng-modules. */
1203                         index->packet_seq_num = -1ULL;
1204                         ret = 0;
1205                 } else {
1206                         PERROR("kernctl_get_sequence_number");
1207                         goto error;
1208                 }
1209         }
1210         index->packet_seq_num = htobe64(index->packet_seq_num);
1211
1212 error:
1213         return ret;
1214 }
1215 /*
1216  * Sync metadata meaning request them to the session daemon and snapshot to the
1217  * metadata thread can consumer them.
1218  *
1219  * Metadata stream lock MUST be acquired.
1220  *
1221  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1222  * is empty or a negative value on error.
1223  */
1224 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1225 {
1226         int ret;
1227
1228         assert(metadata);
1229
1230         ret = kernctl_buffer_flush(metadata->wait_fd);
1231         if (ret < 0) {
1232                 ERR("Failed to flush kernel stream");
1233                 goto end;
1234         }
1235
1236         ret = kernctl_snapshot(metadata->wait_fd);
1237         if (ret < 0) {
1238                 if (ret != -EAGAIN) {
1239                         ERR("Sync metadata, taking kernel snapshot failed.");
1240                         goto end;
1241                 }
1242                 DBG("Sync metadata, no new kernel metadata");
1243                 /* No new metadata, exit. */
1244                 ret = ENODATA;
1245                 goto end;
1246         }
1247
1248 end:
1249         return ret;
1250 }
1251
1252 static
1253 int update_stream_stats(struct lttng_consumer_stream *stream)
1254 {
1255         int ret;
1256         uint64_t seq, discarded;
1257
1258         ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
1259         if (ret < 0) {
1260                 if (ret == -ENOTTY) {
1261                         /* Command not implemented by lttng-modules. */
1262                         seq = -1ULL;
1263                         ret = 0;
1264                 } else {
1265                         PERROR("kernctl_get_sequence_number");
1266                         goto end;
1267                 }
1268         }
1269
1270         /*
1271          * Start the sequence when we extract the first packet in case we don't
1272          * start at 0 (for example if a consumer is not connected to the
1273          * session immediately after the beginning).
1274          */
1275         if (stream->last_sequence_number == -1ULL) {
1276                 stream->last_sequence_number = seq;
1277         } else if (seq > stream->last_sequence_number) {
1278                 stream->chan->lost_packets += seq -
1279                                 stream->last_sequence_number - 1;
1280         } else {
1281                 /* seq <= last_sequence_number */
1282                 ERR("Sequence number inconsistent : prev = %" PRIu64
1283                                 ", current = %" PRIu64,
1284                                 stream->last_sequence_number, seq);
1285                 ret = -1;
1286                 goto end;
1287         }
1288         stream->last_sequence_number = seq;
1289
1290         ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
1291         if (ret < 0) {
1292                 PERROR("kernctl_get_events_discarded");
1293                 goto end;
1294         }
1295         if (discarded < stream->last_discarded_events) {
1296                 /*
1297                  * Overflow has occurred. We assume only one wrap-around
1298                  * has occurred.
1299                  */
1300                 stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
1301                         stream->last_discarded_events + discarded;
1302         } else {
1303                 stream->chan->discarded_events += discarded -
1304                         stream->last_discarded_events;
1305         }
1306         stream->last_discarded_events = discarded;
1307         ret = 0;
1308
1309 end:
1310         return ret;
1311 }
1312
1313 /*
1314  * Check if the local version of the metadata stream matches with the version
1315  * of the metadata stream in the kernel. If it was updated, set the reset flag
1316  * on the stream.
1317  */
1318 static
1319 int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
1320 {
1321         int ret;
1322         uint64_t cur_version;
1323
1324         ret = kernctl_get_metadata_version(infd, &cur_version);
1325         if (ret < 0) {
1326                 if (ret == -ENOTTY) {
1327                         /*
1328                          * LTTng-modules does not implement this
1329                          * command.
1330                          */
1331                         ret = 0;
1332                         goto end;
1333                 }
1334                 ERR("Failed to get the metadata version");
1335                 goto end;
1336         }
1337
1338         if (stream->metadata_version == cur_version) {
1339                 ret = 0;
1340                 goto end;
1341         }
1342
1343         DBG("New metadata version detected");
1344         stream->metadata_version = cur_version;
1345         stream->reset_metadata_flag = 1;
1346         ret = 0;
1347
1348 end:
1349         return ret;
1350 }
1351
1352 /*
1353  * Consume data on a file descriptor and write it on a trace file.
1354  */
1355 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1356                 struct lttng_consumer_local_data *ctx)
1357 {
1358         unsigned long len, subbuf_size, padding;
1359         int err, write_index = 1;
1360         ssize_t ret = 0;
1361         int infd = stream->wait_fd;
1362         struct ctf_packet_index index;
1363
1364         DBG("In read_subbuffer (infd : %d)", infd);
1365
1366         /* Get the next subbuffer */
1367         err = kernctl_get_next_subbuf(infd);
1368         if (err != 0) {
1369                 /*
1370                  * This is a debug message even for single-threaded consumer,
1371                  * because poll() have more relaxed criterions than get subbuf,
1372                  * so get_subbuf may fail for short race windows where poll()
1373                  * would issue wakeups.
1374                  */
1375                 DBG("Reserving sub buffer failed (everything is normal, "
1376                                 "it is due to concurrency)");
1377                 ret = err;
1378                 goto end;
1379         }
1380
1381         /* Get the full subbuffer size including padding */
1382         err = kernctl_get_padded_subbuf_size(infd, &len);
1383         if (err != 0) {
1384                 PERROR("Getting sub-buffer len failed.");
1385                 err = kernctl_put_subbuf(infd);
1386                 if (err != 0) {
1387                         if (err == -EFAULT) {
1388                                 PERROR("Error in unreserving sub buffer\n");
1389                         } else if (err == -EIO) {
1390                                 /* Should never happen with newer LTTng versions */
1391                                 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1392                         }
1393                         ret = err;
1394                         goto end;
1395                 }
1396                 ret = err;
1397                 goto end;
1398         }
1399
1400         if (!stream->metadata_flag) {
1401                 ret = get_index_values(&index, infd);
1402                 if (ret < 0) {
1403                         err = kernctl_put_subbuf(infd);
1404                         if (err != 0) {
1405                                 if (err == -EFAULT) {
1406                                         PERROR("Error in unreserving sub buffer\n");
1407                                 } else if (err == -EIO) {
1408                                         /* Should never happen with newer LTTng versions */
1409                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1410                                 }
1411                                 ret = err;
1412                                 goto end;
1413                         }
1414                         goto end;
1415                 }
1416                 ret = update_stream_stats(stream);
1417                 if (ret < 0) {
1418                         err = kernctl_put_subbuf(infd);
1419                         if (err != 0) {
1420                                 if (err == -EFAULT) {
1421                                         PERROR("Error in unreserving sub buffer\n");
1422                                 } else if (err == -EIO) {
1423                                         /* Should never happen with newer LTTng versions */
1424                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1425                                 }
1426                                 ret = err;
1427                                 goto end;
1428                         }
1429                         goto end;
1430                 }
1431         } else {
1432                 write_index = 0;
1433                 ret = metadata_stream_check_version(infd, stream);
1434                 if (ret < 0) {
1435                         err = kernctl_put_subbuf(infd);
1436                         if (err != 0) {
1437                                 if (err == -EFAULT) {
1438                                         PERROR("Error in unreserving sub buffer\n");
1439                                 } else if (err == -EIO) {
1440                                         /* Should never happen with newer LTTng versions */
1441                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1442                                 }
1443                                 ret = err;
1444                                 goto end;
1445                         }
1446                         goto end;
1447                 }
1448         }
1449
1450         switch (stream->chan->output) {
1451         case CONSUMER_CHANNEL_SPLICE:
1452                 /*
1453                  * XXX: The lttng-modules splice "actor" does not handle copying
1454                  * partial pages hence only using the subbuffer size without the
1455                  * padding makes the splice fail.
1456                  */
1457                 subbuf_size = len;
1458                 padding = 0;
1459
1460                 /* splice the subbuffer to the tracefile */
1461                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1462                                 padding, &index);
1463                 /*
1464                  * XXX: Splice does not support network streaming so the return value
1465                  * is simply checked against subbuf_size and not like the mmap() op.
1466                  */
1467                 if (ret != subbuf_size) {
1468                         /*
1469                          * display the error but continue processing to try
1470                          * to release the subbuffer
1471                          */
1472                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1473                                         ret, subbuf_size);
1474                         write_index = 0;
1475                 }
1476                 break;
1477         case CONSUMER_CHANNEL_MMAP:
1478                 /* Get subbuffer size without padding */
1479                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1480                 if (err != 0) {
1481                         PERROR("Getting sub-buffer len failed.");
1482                         err = kernctl_put_subbuf(infd);
1483                         if (err != 0) {
1484                                 if (err == -EFAULT) {
1485                                         PERROR("Error in unreserving sub buffer\n");
1486                                 } else if (err == -EIO) {
1487                                         /* Should never happen with newer LTTng versions */
1488                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1489                                 }
1490                                 ret = err;
1491                                 goto end;
1492                         }
1493                         ret = err;
1494                         goto end;
1495                 }
1496
1497                 /* Make sure the tracer is not gone mad on us! */
1498                 assert(len >= subbuf_size);
1499
1500                 padding = len - subbuf_size;
1501
1502                 /* write the subbuffer to the tracefile */
1503                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1504                                 padding, &index);
1505                 /*
1506                  * The mmap operation should write subbuf_size amount of data when
1507                  * network streaming or the full padding (len) size when we are _not_
1508                  * streaming.
1509                  */
1510                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1511                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1512                         /*
1513                          * Display the error but continue processing to try to release the
1514                          * subbuffer. This is a DBG statement since this is possible to
1515                          * happen without being a critical error.
1516                          */
1517                         DBG("Error writing to tracefile "
1518                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1519                                         ret, len, subbuf_size);
1520                         write_index = 0;
1521                 }
1522                 break;
1523         default:
1524                 ERR("Unknown output method");
1525                 ret = -EPERM;
1526         }
1527
1528         err = kernctl_put_next_subbuf(infd);
1529         if (err != 0) {
1530                 if (err == -EFAULT) {
1531                         PERROR("Error in unreserving sub buffer\n");
1532                 } else if (err == -EIO) {
1533                         /* Should never happen with newer LTTng versions */
1534                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1535                 }
1536                 ret = err;
1537                 goto end;
1538         }
1539
1540         /* Write index if needed. */
1541         if (!write_index) {
1542                 goto end;
1543         }
1544
1545         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1546                 /*
1547                  * In live, block until all the metadata is sent.
1548                  */
1549                 pthread_mutex_lock(&stream->metadata_timer_lock);
1550                 assert(!stream->missed_metadata_flush);
1551                 stream->waiting_on_metadata = true;
1552                 pthread_mutex_unlock(&stream->metadata_timer_lock);
1553
1554                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1555
1556                 pthread_mutex_lock(&stream->metadata_timer_lock);
1557                 stream->waiting_on_metadata = false;
1558                 if (stream->missed_metadata_flush) {
1559                         stream->missed_metadata_flush = false;
1560                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1561                         (void) consumer_flush_kernel_index(stream);
1562                 } else {
1563                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1564                 }
1565                 if (err < 0) {
1566                         goto end;
1567                 }
1568         }
1569
1570         err = consumer_stream_write_index(stream, &index);
1571         if (err < 0) {
1572                 goto end;
1573         }
1574
1575 end:
1576         return ret;
1577 }
1578
1579 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1580 {
1581         int ret;
1582
1583         assert(stream);
1584
1585         /*
1586          * Don't create anything if this is set for streaming or should not be
1587          * monitored.
1588          */
1589         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1590                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1591                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1592                                 stream->uid, stream->gid, NULL);
1593                 if (ret < 0) {
1594                         goto error;
1595                 }
1596                 stream->out_fd = ret;
1597                 stream->tracefile_size_current = 0;
1598
1599                 if (!stream->metadata_flag) {
1600                         struct lttng_index_file *index_file;
1601
1602                         index_file = lttng_index_file_create(stream->chan->pathname,
1603                                         stream->name, stream->uid, stream->gid,
1604                                         stream->chan->tracefile_size,
1605                                         stream->tracefile_count_current,
1606                                         CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1607                         if (!index_file) {
1608                                 goto error;
1609                         }
1610                         assert(!stream->index_file);
1611                         stream->index_file = index_file;
1612                 }
1613         }
1614
1615         if (stream->output == LTTNG_EVENT_MMAP) {
1616                 /* get the len of the mmap region */
1617                 unsigned long mmap_len;
1618
1619                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1620                 if (ret != 0) {
1621                         PERROR("kernctl_get_mmap_len");
1622                         goto error_close_fd;
1623                 }
1624                 stream->mmap_len = (size_t) mmap_len;
1625
1626                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1627                                 MAP_PRIVATE, stream->wait_fd, 0);
1628                 if (stream->mmap_base == MAP_FAILED) {
1629                         PERROR("Error mmaping");
1630                         ret = -1;
1631                         goto error_close_fd;
1632                 }
1633         }
1634
1635         /* we return 0 to let the library handle the FD internally */
1636         return 0;
1637
1638 error_close_fd:
1639         if (stream->out_fd >= 0) {
1640                 int err;
1641
1642                 err = close(stream->out_fd);
1643                 assert(!err);
1644                 stream->out_fd = -1;
1645         }
1646 error:
1647         return ret;
1648 }
1649
1650 /*
1651  * Check if data is still being extracted from the buffers for a specific
1652  * stream. Consumer data lock MUST be acquired before calling this function
1653  * and the stream lock.
1654  *
1655  * Return 1 if the traced data are still getting read else 0 meaning that the
1656  * data is available for trace viewer reading.
1657  */
1658 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1659 {
1660         int ret;
1661
1662         assert(stream);
1663
1664         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1665                 ret = 0;
1666                 goto end;
1667         }
1668
1669         ret = kernctl_get_next_subbuf(stream->wait_fd);
1670         if (ret == 0) {
1671                 /* There is still data so let's put back this subbuffer. */
1672                 ret = kernctl_put_subbuf(stream->wait_fd);
1673                 assert(ret == 0);
1674                 ret = 1;   /* Data is pending */
1675                 goto end;
1676         }
1677
1678         /* Data is NOT pending and ready to be read. */
1679         ret = 0;
1680
1681 end:
1682         return ret;
1683 }