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