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 _GNU_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/pipe.h>
  39 #include <common/relayd/relayd.h>
  40 #include <common/utils.h>
  41 #include <common/consumer-stream.h>
  42 #include <common/index/index.h>
  43 #include <common/consumer-timer.h>
  44
  45 #include "kernel-consumer.h"
  46
  47 extern struct lttng_consumer_global_data consumer_data;
  48 extern int consumer_poll_timeout;
  49 extern volatile int consumer_quit;
  50
  51 /*
  52  * Take a snapshot for a specific fd
  53  *
  54  * Returns 0 on success, < 0 on error
  55  */
  56 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  57 {
  58         int ret = 0;
  59         int infd = stream->wait_fd;
  60
  61         ret = kernctl_snapshot(infd);
  62         if (ret != 0) {
  63                 perror("Getting sub-buffer snapshot.");
  64                 ret = -errno;
  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                 ret = -errno;
  85         }
  86
  87         return ret;
  88 }
  89
  90 /*
  91  * Get the consumerd position
  92  *
  93  * Returns 0 on success, < 0 on error
  94  */
  95 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  96                 unsigned long *pos)
  97 {
  98         int ret;
  99         int infd = stream->wait_fd;
 100
 101         ret = kernctl_snapshot_get_consumed(infd, pos);
 102         if (ret != 0) {
 103                 perror("kernctl_snapshot_get_consumed");
 104                 ret = -errno;
 105         }
 106
 107         return ret;
 108 }
 109
 110 /*
 111  * Take a snapshot of all the stream of a channel
 112  *
 113  * Returns 0 on success, < 0 on error
 114  */
 115 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 116                 uint64_t relayd_id, uint64_t max_stream_size,
 117                 struct lttng_consumer_local_data *ctx)
 118 {
 119         int ret;
 120         unsigned long consumed_pos, produced_pos;
 121         struct lttng_consumer_channel *channel;
 122         struct lttng_consumer_stream *stream;
 123
 124         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 125
 126         rcu_read_lock();
 127
 128         channel = consumer_find_channel(key);
 129         if (!channel) {
 130                 ERR("No channel found for key %" PRIu64, key);
 131                 ret = -1;
 132                 goto end;
 133         }
 134
 135         /* Splice is not supported yet for channel snapshot. */
 136         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 137                 ERR("Unsupported output %d", channel->output);
 138                 ret = -1;
 139                 goto end;
 140         }
 141
 142         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 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                         ret = -errno;
 191                         goto end_unlock;
 192                 }
 193
 194                 ret = lttng_kconsumer_take_snapshot(stream);
 195                 if (ret < 0) {
 196                         ERR("Taking kernel snapshot");
 197                         goto end_unlock;
 198                 }
 199
 200                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 201                 if (ret < 0) {
 202                         ERR("Produced kernel snapshot position");
 203                         goto end_unlock;
 204                 }
 205
 206                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 207                 if (ret < 0) {
 208                         ERR("Consumerd kernel snapshot position");
 209                         goto end_unlock;
 210                 }
 211
 212                 if (stream->max_sb_size == 0) {
 213                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 214                                         &stream->max_sb_size);
 215                         if (ret < 0) {
 216                                 ERR("Getting kernel max_sb_size");
 217                                 ret = -errno;
 218                                 goto end_unlock;
 219                         }
 220                 }
 221
 222                 /*
 223                  * The original value is sent back if max stream size is larger than
 224                  * the possible size of the snapshot. Also, we asume that the session
 225                  * daemon should never send a maximum stream size that is lower than
 226                  * subbuffer size.
 227                  */
 228                 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
 229                                 produced_pos, max_stream_size);
 230
 231                 while (consumed_pos < produced_pos) {
 232                         ssize_t read_len;
 233                         unsigned long len, padded_len;
 234
 235                         health_code_update();
 236
 237                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 238
 239                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 240                         if (ret < 0) {
 241                                 if (errno != EAGAIN) {
 242                                         PERROR("kernctl_get_subbuf snapshot");
 243                                         ret = -errno;
 244                                         goto end_unlock;
 245                                 }
 246                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 247                                 consumed_pos += stream->max_sb_size;
 248                                 continue;
 249                         }
 250
 251                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 252                         if (ret < 0) {
 253                                 ERR("Snapshot kernctl_get_subbuf_size");
 254                                 ret = -errno;
 255                                 goto error_put_subbuf;
 256                         }
 257
 258                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 259                         if (ret < 0) {
 260                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 261                                 ret = -errno;
 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                                 ret = -errno;
 288                                 goto end_unlock;
 289                         }
 290                         consumed_pos += stream->max_sb_size;
 291                 }
 292
 293                 if (relayd_id == (uint64_t) -1ULL) {
 294                         if (stream->out_fd >= 0) {
 295                                 ret = close(stream->out_fd);
 296                                 if (ret < 0) {
 297                                         PERROR("Kernel consumer snapshot close out_fd");
 298                                         goto end_unlock;
 299                                 }
 300                                 stream->out_fd = -1;
 301                         }
 302                 } else {
 303                         close_relayd_stream(stream);
 304                         stream->net_seq_idx = (uint64_t) -1ULL;
 305                 }
 306                 pthread_mutex_unlock(&stream->lock);
 307         }
 308
 309         /* All good! */
 310         ret = 0;
 311         goto end;
 312
 313 error_put_subbuf:
 314         ret = kernctl_put_subbuf(stream->wait_fd);
 315         if (ret < 0) {
 316                 ret = -errno;
 317                 ERR("Snapshot kernctl_put_subbuf error path");
 318         }
 319 end_unlock:
 320         pthread_mutex_unlock(&stream->lock);
 321 end:
 322         rcu_read_unlock();
 323         return ret;
 324 }
 325
 326 /*
 327  * Read the whole metadata available for a snapshot.
 328  *
 329  * Returns 0 on success, < 0 on error
 330  */
 331 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 332                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 333 {
 334         int ret, use_relayd = 0;
 335         ssize_t ret_read;
 336         struct lttng_consumer_channel *metadata_channel;
 337         struct lttng_consumer_stream *metadata_stream;
 338
 339         assert(ctx);
 340
 341         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 342                         key, path);
 343
 344         rcu_read_lock();
 345
 346         metadata_channel = consumer_find_channel(key);
 347         if (!metadata_channel) {
 348                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 349                 ret = -1;
 350                 goto error;
 351         }
 352
 353         metadata_stream = metadata_channel->metadata_stream;
 354         assert(metadata_stream);
 355
 356         /* Flag once that we have a valid relayd for the stream. */
 357         if (relayd_id != (uint64_t) -1ULL) {
 358                 use_relayd = 1;
 359         }
 360
 361         if (use_relayd) {
 362                 ret = consumer_send_relayd_stream(metadata_stream, path);
 363                 if (ret < 0) {
 364                         goto error;
 365                 }
 366         } else {
 367                 ret = utils_create_stream_file(path, metadata_stream->name,
 368                                 metadata_stream->chan->tracefile_size,
 369                                 metadata_stream->tracefile_count_current,
 370                                 metadata_stream->uid, metadata_stream->gid, NULL);
 371                 if (ret < 0) {
 372                         goto error;
 373                 }
 374                 metadata_stream->out_fd = ret;
 375         }
 376
 377         do {
 378                 health_code_update();
 379
 380                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 381                 if (ret_read < 0) {
 382                         if (ret_read != -EAGAIN) {
 383                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 384                                                 ret_read);
 385                                 goto error;
 386                         }
 387                         /* ret_read is negative at this point so we will exit the loop. */
 388                         continue;
 389                 }
 390         } while (ret_read >= 0);
 391
 392         if (use_relayd) {
 393                 close_relayd_stream(metadata_stream);
 394                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 395         } else {
 396                 if (metadata_stream->out_fd >= 0) {
 397                         ret = close(metadata_stream->out_fd);
 398                         if (ret < 0) {
 399                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 400                                 /*
 401                                  * Don't go on error here since the snapshot was successful at this
 402                                  * point but somehow the close failed.
 403                                  */
 404                         }
 405                         metadata_stream->out_fd = -1;
 406                 }
 407         }
 408
 409         ret = 0;
 410
 411         cds_list_del(&metadata_stream->send_node);
 412         consumer_stream_destroy(metadata_stream, NULL);
 413         metadata_channel->metadata_stream = NULL;
 414 error:
 415         rcu_read_unlock();
 416         return ret;
 417 }
 418
 419 /*
 420  * Receive command from session daemon and process it.
 421  *
 422  * Return 1 on success else a negative value or 0.
 423  */
 424 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 425                 int sock, struct pollfd *consumer_sockpoll)
 426 {
 427         ssize_t ret;
 428         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 429         struct lttcomm_consumer_msg msg;
 430
 431         health_code_update();
 432
 433         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 434         if (ret != sizeof(msg)) {
 435                 if (ret > 0) {
 436                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 437                         ret = -1;
 438                 }
 439                 return ret;
 440         }
 441
 442         health_code_update();
 443
 444         /* Deprecated command */
 445         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 446
 447         health_code_update();
 448
 449         /* relayd needs RCU read-side protection */
 450         rcu_read_lock();
 451
 452         switch (msg.cmd_type) {
 453         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 454         {
 455                 /* Session daemon status message are handled in the following call. */
 456                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 457                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 458                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 459                                  msg.u.relayd_sock.relayd_session_id);
 460                 goto end_nosignal;
 461         }
 462         case LTTNG_CONSUMER_ADD_CHANNEL:
 463         {
 464                 struct lttng_consumer_channel *new_channel;
 465                 int ret_recv;
 466
 467                 health_code_update();
 468
 469                 /* First send a status message before receiving the fds. */
 470                 ret = consumer_send_status_msg(sock, ret_code);
 471                 if (ret < 0) {
 472                         /* Somehow, the session daemon is not responding anymore. */
 473                         goto error_fatal;
 474                 }
 475
 476                 health_code_update();
 477
 478                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 479                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 480                                 msg.u.channel.session_id, msg.u.channel.pathname,
 481                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 482                                 msg.u.channel.relayd_id, msg.u.channel.output,
 483                                 msg.u.channel.tracefile_size,
 484                                 msg.u.channel.tracefile_count, 0,
 485                                 msg.u.channel.monitor,
 486                                 msg.u.channel.live_timer_interval);
 487                 if (new_channel == NULL) {
 488                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 489                         goto end_nosignal;
 490                 }
 491                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 492                 switch (msg.u.channel.output) {
 493                 case LTTNG_EVENT_SPLICE:
 494                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 495                         break;
 496                 case LTTNG_EVENT_MMAP:
 497                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 498                         break;
 499                 default:
 500                         ERR("Channel output unknown %d", msg.u.channel.output);
 501                         goto end_nosignal;
 502                 }
 503
 504                 /* Translate and save channel type. */
 505                 switch (msg.u.channel.type) {
 506                 case CONSUMER_CHANNEL_TYPE_DATA:
 507                 case CONSUMER_CHANNEL_TYPE_METADATA:
 508                         new_channel->type = msg.u.channel.type;
 509                         break;
 510                 default:
 511                         assert(0);
 512                         goto end_nosignal;
 513                 };
 514
 515                 health_code_update();
 516
 517                 if (ctx->on_recv_channel != NULL) {
 518                         ret_recv = ctx->on_recv_channel(new_channel);
 519                         if (ret_recv == 0) {
 520                                 ret = consumer_add_channel(new_channel, ctx);
 521                         } else if (ret_recv < 0) {
 522                                 goto end_nosignal;
 523                         }
 524                 } else {
 525                         ret = consumer_add_channel(new_channel, ctx);
 526                 }
 527                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 528                         consumer_timer_live_start(new_channel,
 529                                         msg.u.channel.live_timer_interval);
 530                 }
 531
 532                 health_code_update();
 533
 534                 /* If we received an error in add_channel, we need to report it. */
 535                 if (ret < 0) {
 536                         ret = consumer_send_status_msg(sock, ret);
 537                         if (ret < 0) {
 538                                 goto error_fatal;
 539                         }
 540                         goto end_nosignal;
 541                 }
 542
 543                 goto end_nosignal;
 544         }
 545         case LTTNG_CONSUMER_ADD_STREAM:
 546         {
 547                 int fd;
 548                 struct lttng_pipe *stream_pipe;
 549                 struct lttng_consumer_stream *new_stream;
 550                 struct lttng_consumer_channel *channel;
 551                 int alloc_ret = 0;
 552
 553                 /*
 554                  * Get stream's channel reference. Needed when adding the stream to the
 555                  * global hash table.
 556                  */
 557                 channel = consumer_find_channel(msg.u.stream.channel_key);
 558                 if (!channel) {
 559                         /*
 560                          * We could not find the channel. Can happen if cpu hotplug
 561                          * happens while tearing down.
 562                          */
 563                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 564                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 565                 }
 566
 567                 health_code_update();
 568
 569                 /* First send a status message before receiving the fds. */
 570                 ret = consumer_send_status_msg(sock, ret_code);
 571                 if (ret < 0) {
 572                         /* Somehow, the session daemon is not responding anymore. */
 573                         goto error_fatal;
 574                 }
 575
 576                 health_code_update();
 577
 578                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 579                         /* Channel was not found. */
 580                         goto end_nosignal;
 581                 }
 582
 583                 /* Blocking call */
 584                 health_poll_entry();
 585                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 586                 health_poll_exit();
 587                 if (ret) {
 588                         goto error_fatal;
 589                 }
 590
 591                 health_code_update();
 592
 593                 /* Get stream file descriptor from socket */
 594                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 595                 if (ret != sizeof(fd)) {
 596                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 597                         rcu_read_unlock();
 598                         return ret;
 599                 }
 600
 601                 health_code_update();
 602
 603                 /*
 604                  * Send status code to session daemon only if the recv works. If the
 605                  * above recv() failed, the session daemon is notified through the
 606                  * error socket and the teardown is eventually done.
 607                  */
 608                 ret = consumer_send_status_msg(sock, ret_code);
 609                 if (ret < 0) {
 610                         /* Somehow, the session daemon is not responding anymore. */
 611                         goto end_nosignal;
 612                 }
 613
 614                 health_code_update();
 615
 616                 new_stream = consumer_allocate_stream(channel->key,
 617                                 fd,
 618                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 619                                 channel->name,
 620                                 channel->uid,
 621                                 channel->gid,
 622                                 channel->relayd_id,
 623                                 channel->session_id,
 624                                 msg.u.stream.cpu,
 625                                 &alloc_ret,
 626                                 channel->type,
 627                                 channel->monitor);
 628                 if (new_stream == NULL) {
 629                         switch (alloc_ret) {
 630                         case -ENOMEM:
 631                         case -EINVAL:
 632                         default:
 633                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 634                                 break;
 635                         }
 636                         goto end_nosignal;
 637                 }
 638
 639                 new_stream->chan = channel;
 640                 new_stream->wait_fd = fd;
 641                 switch (channel->output) {
 642                 case CONSUMER_CHANNEL_SPLICE:
 643                         new_stream->output = LTTNG_EVENT_SPLICE;
 644                         ret = utils_create_pipe(new_stream->splice_pipe);
 645                         if (ret < 0) {
 646                                 goto end_nosignal;
 647                         }
 648                         break;
 649                 case CONSUMER_CHANNEL_MMAP:
 650                         new_stream->output = LTTNG_EVENT_MMAP;
 651                         break;
 652                 default:
 653                         ERR("Stream output unknown %d", channel->output);
 654                         goto end_nosignal;
 655                 }
 656
 657                 /*
 658                  * We've just assigned the channel to the stream so increment the
 659                  * refcount right now. We don't need to increment the refcount for
 660                  * streams in no monitor because we handle manually the cleanup of
 661                  * those. It is very important to make sure there is NO prior
 662                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 663                  */
 664                 if (channel->monitor) {
 665                         uatomic_inc(&new_stream->chan->refcount);
 666                 }
 667
 668                 /*
 669                  * The buffer flush is done on the session daemon side for the kernel
 670                  * so no need for the stream "hangup_flush_done" variable to be
 671                  * tracked. This is important for a kernel stream since we don't rely
 672                  * on the flush state of the stream to read data. It's not the case for
 673                  * user space tracing.
 674                  */
 675                 new_stream->hangup_flush_done = 0;
 676
 677                 health_code_update();
 678
 679                 if (ctx->on_recv_stream) {
 680                         ret = ctx->on_recv_stream(new_stream);
 681                         if (ret < 0) {
 682                                 consumer_stream_free(new_stream);
 683                                 goto end_nosignal;
 684                         }
 685                 }
 686
 687                 health_code_update();
 688
 689                 if (new_stream->metadata_flag) {
 690                         channel->metadata_stream = new_stream;
 691                 }
 692
 693                 /* Do not monitor this stream. */
 694                 if (!channel->monitor) {
 695                         DBG("Kernel consumer add stream %s in no monitor mode with "
 696                                         "relayd id %" PRIu64, new_stream->name,
 697                                         new_stream->net_seq_idx);
 698                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 699                         break;
 700                 }
 701
 702                 /* Send stream to relayd if the stream has an ID. */
 703                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 704                         ret = consumer_send_relayd_stream(new_stream,
 705                                         new_stream->chan->pathname);
 706                         if (ret < 0) {
 707                                 consumer_stream_free(new_stream);
 708                                 goto end_nosignal;
 709                         }
 710                 }
 711
 712                 /* Get the right pipe where the stream will be sent. */
 713                 if (new_stream->metadata_flag) {
 714                         ret = consumer_add_metadata_stream(new_stream);
 715                         if (ret) {
 716                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 717                                                 new_stream->key);
 718                                 consumer_stream_free(new_stream);
 719                                 goto end_nosignal;
 720                         }
 721                         stream_pipe = ctx->consumer_metadata_pipe;
 722                 } else {
 723                         ret = consumer_add_data_stream(new_stream);
 724                         if (ret) {
 725                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 726                                                 new_stream->key);
 727                                 consumer_stream_free(new_stream);
 728                                 goto end_nosignal;
 729                         }
 730                         stream_pipe = ctx->consumer_data_pipe;
 731                 }
 732
 733                 /* Vitible to other threads */
 734                 new_stream->globally_visible = 1;
 735
 736                 health_code_update();
 737
 738                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 739                 if (ret < 0) {
 740                         ERR("Consumer write %s stream to pipe %d",
 741                                         new_stream->metadata_flag ? "metadata" : "data",
 742                                         lttng_pipe_get_writefd(stream_pipe));
 743                         if (new_stream->metadata_flag) {
 744                                 consumer_del_stream_for_metadata(new_stream);
 745                         } else {
 746                                 consumer_del_stream_for_data(new_stream);
 747                         }
 748                         goto end_nosignal;
 749                 }
 750
 751                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 752                                 new_stream->name, fd, new_stream->relayd_stream_id);
 753                 break;
 754         }
 755         case LTTNG_CONSUMER_STREAMS_SENT:
 756         {
 757                 struct lttng_consumer_channel *channel;
 758
 759                 /*
 760                  * Get stream's channel reference. Needed when adding the stream to the
 761                  * global hash table.
 762                  */
 763                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 764                 if (!channel) {
 765                         /*
 766                          * We could not find the channel. Can happen if cpu hotplug
 767                          * happens while tearing down.
 768                          */
 769                         ERR("Unable to find channel key %" PRIu64,
 770                                         msg.u.sent_streams.channel_key);
 771                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 772                 }
 773
 774                 health_code_update();
 775
 776                 /*
 777                  * Send status code to session daemon.
 778                  */
 779                 ret = consumer_send_status_msg(sock, ret_code);
 780                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 781                         /* Somehow, the session daemon is not responding anymore. */
 782                         goto end_nosignal;
 783                 }
 784
 785                 health_code_update();
 786
 787                 /*
 788                  * We should not send this message if we don't monitor the
 789                  * streams in this channel.
 790                  */
 791                 if (!channel->monitor) {
 792                         break;
 793                 }
 794
 795                 health_code_update();
 796                 /* Send stream to relayd if the stream has an ID. */
 797                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 798                         ret = consumer_send_relayd_streams_sent(
 799                                         msg.u.sent_streams.net_seq_idx);
 800                         if (ret < 0) {
 801                                 goto end_nosignal;
 802                         }
 803                 }
 804                 break;
 805         }
 806         case LTTNG_CONSUMER_UPDATE_STREAM:
 807         {
 808                 rcu_read_unlock();
 809                 return -ENOSYS;
 810         }
 811         case LTTNG_CONSUMER_DESTROY_RELAYD:
 812         {
 813                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 814                 struct consumer_relayd_sock_pair *relayd;
 815
 816                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 817
 818                 /* Get relayd reference if exists. */
 819                 relayd = consumer_find_relayd(index);
 820                 if (relayd == NULL) {
 821                         DBG("Unable to find relayd %" PRIu64, index);
 822                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 823                 }
 824
 825                 /*
 826                  * Each relayd socket pair has a refcount of stream attached to it
 827                  * which tells if the relayd is still active or not depending on the
 828                  * refcount value.
 829                  *
 830                  * This will set the destroy flag of the relayd object and destroy it
 831                  * if the refcount reaches zero when called.
 832                  *
 833                  * The destroy can happen either here or when a stream fd hangs up.
 834                  */
 835                 if (relayd) {
 836                         consumer_flag_relayd_for_destroy(relayd);
 837                 }
 838
 839                 health_code_update();
 840
 841                 ret = consumer_send_status_msg(sock, ret_code);
 842                 if (ret < 0) {
 843                         /* Somehow, the session daemon is not responding anymore. */
 844                         goto error_fatal;
 845                 }
 846
 847                 goto end_nosignal;
 848         }
 849         case LTTNG_CONSUMER_DATA_PENDING:
 850         {
 851                 int32_t ret;
 852                 uint64_t id = msg.u.data_pending.session_id;
 853
 854                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 855
 856                 ret = consumer_data_pending(id);
 857
 858                 health_code_update();
 859
 860                 /* Send back returned value to session daemon */
 861                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 862                 if (ret < 0) {
 863                         PERROR("send data pending ret code");
 864                         goto error_fatal;
 865                 }
 866
 867                 /*
 868                  * No need to send back a status message since the data pending
 869                  * returned value is the response.
 870                  */
 871                 break;
 872         }
 873         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 874         {
 875                 if (msg.u.snapshot_channel.metadata == 1) {
 876                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 877                                         msg.u.snapshot_channel.pathname,
 878                                         msg.u.snapshot_channel.relayd_id, ctx);
 879                         if (ret < 0) {
 880                                 ERR("Snapshot metadata failed");
 881                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 882                         }
 883                 } else {
 884                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 885                                         msg.u.snapshot_channel.pathname,
 886                                         msg.u.snapshot_channel.relayd_id,
 887                                         msg.u.snapshot_channel.max_stream_size,
 888                                         ctx);
 889                         if (ret < 0) {
 890                                 ERR("Snapshot channel failed");
 891                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 892                         }
 893                 }
 894
 895                 health_code_update();
 896
 897                 ret = consumer_send_status_msg(sock, ret_code);
 898                 if (ret < 0) {
 899                         /* Somehow, the session daemon is not responding anymore. */
 900                         goto end_nosignal;
 901                 }
 902                 break;
 903         }
 904         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 905         {
 906                 uint64_t key = msg.u.destroy_channel.key;
 907                 struct lttng_consumer_channel *channel;
 908
 909                 channel = consumer_find_channel(key);
 910                 if (!channel) {
 911                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 912                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 913                 }
 914
 915                 health_code_update();
 916
 917                 ret = consumer_send_status_msg(sock, ret_code);
 918                 if (ret < 0) {
 919                         /* Somehow, the session daemon is not responding anymore. */
 920                         goto end_nosignal;
 921                 }
 922
 923                 health_code_update();
 924
 925                 /* Stop right now if no channel was found. */
 926                 if (!channel) {
 927                         goto end_nosignal;
 928                 }
 929
 930                 /*
 931                  * This command should ONLY be issued for channel with streams set in
 932                  * no monitor mode.
 933                  */
 934                 assert(!channel->monitor);
 935
 936                 /*
 937                  * The refcount should ALWAYS be 0 in the case of a channel in no
 938                  * monitor mode.
 939                  */
 940                 assert(!uatomic_sub_return(&channel->refcount, 1));
 941
 942                 consumer_del_channel(channel);
 943
 944                 goto end_nosignal;
 945         }
 946         default:
 947                 goto end_nosignal;
 948         }
 949
 950 end_nosignal:
 951         rcu_read_unlock();
 952
 953         /*
 954          * Return 1 to indicate success since the 0 value can be a socket
 955          * shutdown during the recv() or send() call.
 956          */
 957         health_code_update();
 958         return 1;
 959
 960 error_fatal:
 961         rcu_read_unlock();
 962         /* This will issue a consumer stop. */
 963         return -1;
 964 }
 965
 966 /*
 967  * Populate index values of a kernel stream. Values are set in big endian order.
 968  *
 969  * Return 0 on success or else a negative value.
 970  */
 971 static int get_index_values(struct ctf_packet_index *index, int infd)
 972 {
 973         int ret;
 974
 975         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 976         if (ret < 0) {
 977                 PERROR("kernctl_get_timestamp_begin");
 978                 goto error;
 979         }
 980         index->timestamp_begin = htobe64(index->timestamp_begin);
 981
 982         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 983         if (ret < 0) {
 984                 PERROR("kernctl_get_timestamp_end");
 985                 goto error;
 986         }
 987         index->timestamp_end = htobe64(index->timestamp_end);
 988
 989         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 990         if (ret < 0) {
 991                 PERROR("kernctl_get_events_discarded");
 992                 goto error;
 993         }
 994         index->events_discarded = htobe64(index->events_discarded);
 995
 996         ret = kernctl_get_content_size(infd, &index->content_size);
 997         if (ret < 0) {
 998                 PERROR("kernctl_get_content_size");
 999                 goto error;
1000         }
1001         index->content_size = htobe64(index->content_size);
1002
1003         ret = kernctl_get_packet_size(infd, &index->packet_size);
1004         if (ret < 0) {
1005                 PERROR("kernctl_get_packet_size");
1006                 goto error;
1007         }
1008         index->packet_size = htobe64(index->packet_size);
1009
1010         ret = kernctl_get_stream_id(infd, &index->stream_id);
1011         if (ret < 0) {
1012                 PERROR("kernctl_get_stream_id");
1013                 goto error;
1014         }
1015         index->stream_id = htobe64(index->stream_id);
1016
1017 error:
1018         return ret;
1019 }
1020 /*
1021  * Sync metadata meaning request them to the session daemon and snapshot to the
1022  * metadata thread can consumer them.
1023  *
1024  * Metadata stream lock MUST be acquired.
1025  *
1026  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1027  * is empty or a negative value on error.
1028  */
1029 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1030 {
1031         int ret;
1032
1033         assert(metadata);
1034
1035         ret = kernctl_buffer_flush(metadata->wait_fd);
1036         if (ret < 0) {
1037                 ERR("Failed to flush kernel stream");
1038                 goto end;
1039         }
1040
1041         ret = kernctl_snapshot(metadata->wait_fd);
1042         if (ret < 0) {
1043                 if (errno != EAGAIN) {
1044                         ERR("Sync metadata, taking kernel snapshot failed.");
1045                         goto end;
1046                 }
1047                 DBG("Sync metadata, no new kernel metadata");
1048                 /* No new metadata, exit. */
1049                 ret = ENODATA;
1050                 goto end;
1051         }
1052
1053 end:
1054         return ret;
1055 }
1056
1057 /*
1058  * Consume data on a file descriptor and write it on a trace file.
1059  */
1060 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1061                 struct lttng_consumer_local_data *ctx)
1062 {
1063         unsigned long len, subbuf_size, padding;
1064         int err, write_index = 1;
1065         ssize_t ret = 0;
1066         int infd = stream->wait_fd;
1067         struct ctf_packet_index index;
1068
1069         DBG("In read_subbuffer (infd : %d)", infd);
1070
1071         /* Get the next subbuffer */
1072         err = kernctl_get_next_subbuf(infd);
1073         if (err != 0) {
1074                 /*
1075                  * This is a debug message even for single-threaded consumer,
1076                  * because poll() have more relaxed criterions than get subbuf,
1077                  * so get_subbuf may fail for short race windows where poll()
1078                  * would issue wakeups.
1079                  */
1080                 DBG("Reserving sub buffer failed (everything is normal, "
1081                                 "it is due to concurrency)");
1082                 ret = -errno;
1083                 goto end;
1084         }
1085
1086         /* Get the full subbuffer size including padding */
1087         err = kernctl_get_padded_subbuf_size(infd, &len);
1088         if (err != 0) {
1089                 perror("Getting sub-buffer len failed.");
1090                 ret = -errno;
1091                 goto end;
1092         }
1093
1094         if (!stream->metadata_flag) {
1095                 ret = get_index_values(&index, infd);
1096                 if (ret < 0) {
1097                         goto end;
1098                 }
1099         } else {
1100                 write_index = 0;
1101         }
1102
1103         switch (stream->chan->output) {
1104         case CONSUMER_CHANNEL_SPLICE:
1105                 /*
1106                  * XXX: The lttng-modules splice "actor" does not handle copying
1107                  * partial pages hence only using the subbuffer size without the
1108                  * padding makes the splice fail.
1109                  */
1110                 subbuf_size = len;
1111                 padding = 0;
1112
1113                 /* splice the subbuffer to the tracefile */
1114                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1115                                 padding, &index);
1116                 /*
1117                  * XXX: Splice does not support network streaming so the return value
1118                  * is simply checked against subbuf_size and not like the mmap() op.
1119                  */
1120                 if (ret != subbuf_size) {
1121                         /*
1122                          * display the error but continue processing to try
1123                          * to release the subbuffer
1124                          */
1125                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1126                                         ret, subbuf_size);
1127                         write_index = 0;
1128                 }
1129                 break;
1130         case CONSUMER_CHANNEL_MMAP:
1131                 /* Get subbuffer size without padding */
1132                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1133                 if (err != 0) {
1134                         perror("Getting sub-buffer len failed.");
1135                         ret = -errno;
1136                         goto end;
1137                 }
1138
1139                 /* Make sure the tracer is not gone mad on us! */
1140                 assert(len >= subbuf_size);
1141
1142                 padding = len - subbuf_size;
1143
1144                 /* write the subbuffer to the tracefile */
1145                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1146                                 padding, &index);
1147                 /*
1148                  * The mmap operation should write subbuf_size amount of data when
1149                  * network streaming or the full padding (len) size when we are _not_
1150                  * streaming.
1151                  */
1152                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1153                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1154                         /*
1155                          * Display the error but continue processing to try to release the
1156                          * subbuffer. This is a DBG statement since this is possible to
1157                          * happen without being a critical error.
1158                          */
1159                         DBG("Error writing to tracefile "
1160                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1161                                         ret, len, subbuf_size);
1162                         write_index = 0;
1163                 }
1164                 break;
1165         default:
1166                 ERR("Unknown output method");
1167                 ret = -EPERM;
1168         }
1169
1170         err = kernctl_put_next_subbuf(infd);
1171         if (err != 0) {
1172                 if (errno == EFAULT) {
1173                         perror("Error in unreserving sub buffer\n");
1174                 } else if (errno == EIO) {
1175                         /* Should never happen with newer LTTng versions */
1176                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1177                 }
1178                 ret = -errno;
1179                 goto end;
1180         }
1181
1182         /* Write index if needed. */
1183         if (!write_index) {
1184                 goto end;
1185         }
1186
1187         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1188                 /*
1189                  * In live, block until all the metadata is sent.
1190                  */
1191                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1192                 if (err < 0) {
1193                         goto end;
1194                 }
1195         }
1196
1197         err = consumer_stream_write_index(stream, &index);
1198         if (err < 0) {
1199                 goto end;
1200         }
1201
1202 end:
1203         return ret;
1204 }
1205
1206 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1207 {
1208         int ret;
1209
1210         assert(stream);
1211
1212         /*
1213          * Don't create anything if this is set for streaming or should not be
1214          * monitored.
1215          */
1216         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1217                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1218                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1219                                 stream->uid, stream->gid, NULL);
1220                 if (ret < 0) {
1221                         goto error;
1222                 }
1223                 stream->out_fd = ret;
1224                 stream->tracefile_size_current = 0;
1225
1226                 if (!stream->metadata_flag) {
1227                         ret = index_create_file(stream->chan->pathname,
1228                                         stream->name, stream->uid, stream->gid,
1229                                         stream->chan->tracefile_size,
1230                                         stream->tracefile_count_current);
1231                         if (ret < 0) {
1232                                 goto error;
1233                         }
1234                         stream->index_fd = ret;
1235                 }
1236         }
1237
1238         if (stream->output == LTTNG_EVENT_MMAP) {
1239                 /* get the len of the mmap region */
1240                 unsigned long mmap_len;
1241
1242                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1243                 if (ret != 0) {
1244                         PERROR("kernctl_get_mmap_len");
1245                         ret = -errno;
1246                         goto error_close_fd;
1247                 }
1248                 stream->mmap_len = (size_t) mmap_len;
1249
1250                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1251                                 MAP_PRIVATE, stream->wait_fd, 0);
1252                 if (stream->mmap_base == MAP_FAILED) {
1253                         PERROR("Error mmaping");
1254                         ret = -1;
1255                         goto error_close_fd;
1256                 }
1257         }
1258
1259         /* we return 0 to let the library handle the FD internally */
1260         return 0;
1261
1262 error_close_fd:
1263         if (stream->out_fd >= 0) {
1264                 int err;
1265
1266                 err = close(stream->out_fd);
1267                 assert(!err);
1268                 stream->out_fd = -1;
1269         }
1270 error:
1271         return ret;
1272 }
1273
1274 /*
1275  * Check if data is still being extracted from the buffers for a specific
1276  * stream. Consumer data lock MUST be acquired before calling this function
1277  * and the stream lock.
1278  *
1279  * Return 1 if the traced data are still getting read else 0 meaning that the
1280  * data is available for trace viewer reading.
1281  */
1282 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1283 {
1284         int ret;
1285
1286         assert(stream);
1287
1288         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1289                 ret = 0;
1290                 goto end;
1291         }
1292
1293         ret = kernctl_get_next_subbuf(stream->wait_fd);
1294         if (ret == 0) {
1295                 /* There is still data so let's put back this subbuffer. */
1296                 ret = kernctl_put_subbuf(stream->wait_fd);
1297                 assert(ret == 0);
1298                 ret = 1;   /* Data is pending */
1299                 goto end;
1300         }
1301
1302         /* Data is NOT pending and ready to be read. */
1303         ret = 0;
1304
1305 end:
1306         return ret;
1307 }