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 = LTTNG_ERR_KERN_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                         break;
 645                 case CONSUMER_CHANNEL_MMAP:
 646                         new_stream->output = LTTNG_EVENT_MMAP;
 647                         break;
 648                 default:
 649                         ERR("Stream output unknown %d", channel->output);
 650                         goto end_nosignal;
 651                 }
 652
 653                 /*
 654                  * We've just assigned the channel to the stream so increment the
 655                  * refcount right now. We don't need to increment the refcount for
 656                  * streams in no monitor because we handle manually the cleanup of
 657                  * those. It is very important to make sure there is NO prior
 658                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 659                  */
 660                 if (channel->monitor) {
 661                         uatomic_inc(&new_stream->chan->refcount);
 662                 }
 663
 664                 /*
 665                  * The buffer flush is done on the session daemon side for the kernel
 666                  * so no need for the stream "hangup_flush_done" variable to be
 667                  * tracked. This is important for a kernel stream since we don't rely
 668                  * on the flush state of the stream to read data. It's not the case for
 669                  * user space tracing.
 670                  */
 671                 new_stream->hangup_flush_done = 0;
 672
 673                 health_code_update();
 674
 675                 if (ctx->on_recv_stream) {
 676                         ret = ctx->on_recv_stream(new_stream);
 677                         if (ret < 0) {
 678                                 consumer_stream_free(new_stream);
 679                                 goto end_nosignal;
 680                         }
 681                 }
 682
 683                 health_code_update();
 684
 685                 if (new_stream->metadata_flag) {
 686                         channel->metadata_stream = new_stream;
 687                 }
 688
 689                 /* Do not monitor this stream. */
 690                 if (!channel->monitor) {
 691                         DBG("Kernel consumer add stream %s in no monitor mode with "
 692                                         "relayd id %" PRIu64, new_stream->name,
 693                                         new_stream->net_seq_idx);
 694                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 695                         break;
 696                 }
 697
 698                 /* Send stream to relayd if the stream has an ID. */
 699                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 700                         ret = consumer_send_relayd_stream(new_stream,
 701                                         new_stream->chan->pathname);
 702                         if (ret < 0) {
 703                                 consumer_stream_free(new_stream);
 704                                 goto end_nosignal;
 705                         }
 706                 }
 707
 708                 /* Get the right pipe where the stream will be sent. */
 709                 if (new_stream->metadata_flag) {
 710                         ret = consumer_add_metadata_stream(new_stream);
 711                         if (ret) {
 712                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 713                                                 new_stream->key);
 714                                 consumer_stream_free(new_stream);
 715                                 goto end_nosignal;
 716                         }
 717                         stream_pipe = ctx->consumer_metadata_pipe;
 718                 } else {
 719                         ret = consumer_add_data_stream(new_stream);
 720                         if (ret) {
 721                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 722                                                 new_stream->key);
 723                                 consumer_stream_free(new_stream);
 724                                 goto end_nosignal;
 725                         }
 726                         stream_pipe = ctx->consumer_data_pipe;
 727                 }
 728
 729                 /* Vitible to other threads */
 730                 new_stream->globally_visible = 1;
 731
 732                 health_code_update();
 733
 734                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 735                 if (ret < 0) {
 736                         ERR("Consumer write %s stream to pipe %d",
 737                                         new_stream->metadata_flag ? "metadata" : "data",
 738                                         lttng_pipe_get_writefd(stream_pipe));
 739                         if (new_stream->metadata_flag) {
 740                                 consumer_del_stream_for_metadata(new_stream);
 741                         } else {
 742                                 consumer_del_stream_for_data(new_stream);
 743                         }
 744                         goto end_nosignal;
 745                 }
 746
 747                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 748                                 new_stream->name, fd, new_stream->relayd_stream_id);
 749                 break;
 750         }
 751         case LTTNG_CONSUMER_STREAMS_SENT:
 752         {
 753                 struct lttng_consumer_channel *channel;
 754
 755                 /*
 756                  * Get stream's channel reference. Needed when adding the stream to the
 757                  * global hash table.
 758                  */
 759                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 760                 if (!channel) {
 761                         /*
 762                          * We could not find the channel. Can happen if cpu hotplug
 763                          * happens while tearing down.
 764                          */
 765                         ERR("Unable to find channel key %" PRIu64,
 766                                         msg.u.sent_streams.channel_key);
 767                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 768                 }
 769
 770                 health_code_update();
 771
 772                 /*
 773                  * Send status code to session daemon.
 774                  */
 775                 ret = consumer_send_status_msg(sock, ret_code);
 776                 if (ret < 0) {
 777                         /* Somehow, the session daemon is not responding anymore. */
 778                         goto end_nosignal;
 779                 }
 780
 781                 health_code_update();
 782
 783                 /*
 784                  * We should not send this message if we don't monitor the
 785                  * streams in this channel.
 786                  */
 787                 if (!channel->monitor) {
 788                         break;
 789                 }
 790
 791                 health_code_update();
 792                 /* Send stream to relayd if the stream has an ID. */
 793                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 794                         ret = consumer_send_relayd_streams_sent(
 795                                         msg.u.sent_streams.net_seq_idx);
 796                         if (ret < 0) {
 797                                 goto end_nosignal;
 798                         }
 799                 }
 800                 break;
 801         }
 802         case LTTNG_CONSUMER_UPDATE_STREAM:
 803         {
 804                 rcu_read_unlock();
 805                 return -ENOSYS;
 806         }
 807         case LTTNG_CONSUMER_DESTROY_RELAYD:
 808         {
 809                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 810                 struct consumer_relayd_sock_pair *relayd;
 811
 812                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 813
 814                 /* Get relayd reference if exists. */
 815                 relayd = consumer_find_relayd(index);
 816                 if (relayd == NULL) {
 817                         DBG("Unable to find relayd %" PRIu64, index);
 818                         ret_code = LTTNG_ERR_NO_CONSUMER;
 819                 }
 820
 821                 /*
 822                  * Each relayd socket pair has a refcount of stream attached to it
 823                  * which tells if the relayd is still active or not depending on the
 824                  * refcount value.
 825                  *
 826                  * This will set the destroy flag of the relayd object and destroy it
 827                  * if the refcount reaches zero when called.
 828                  *
 829                  * The destroy can happen either here or when a stream fd hangs up.
 830                  */
 831                 if (relayd) {
 832                         consumer_flag_relayd_for_destroy(relayd);
 833                 }
 834
 835                 health_code_update();
 836
 837                 ret = consumer_send_status_msg(sock, ret_code);
 838                 if (ret < 0) {
 839                         /* Somehow, the session daemon is not responding anymore. */
 840                         goto error_fatal;
 841                 }
 842
 843                 goto end_nosignal;
 844         }
 845         case LTTNG_CONSUMER_DATA_PENDING:
 846         {
 847                 int32_t ret;
 848                 uint64_t id = msg.u.data_pending.session_id;
 849
 850                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 851
 852                 ret = consumer_data_pending(id);
 853
 854                 health_code_update();
 855
 856                 /* Send back returned value to session daemon */
 857                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 858                 if (ret < 0) {
 859                         PERROR("send data pending ret code");
 860                         goto error_fatal;
 861                 }
 862
 863                 /*
 864                  * No need to send back a status message since the data pending
 865                  * returned value is the response.
 866                  */
 867                 break;
 868         }
 869         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 870         {
 871                 if (msg.u.snapshot_channel.metadata == 1) {
 872                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 873                                         msg.u.snapshot_channel.pathname,
 874                                         msg.u.snapshot_channel.relayd_id, ctx);
 875                         if (ret < 0) {
 876                                 ERR("Snapshot metadata failed");
 877                                 ret_code = LTTNG_ERR_KERN_META_FAIL;
 878                         }
 879                 } else {
 880                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 881                                         msg.u.snapshot_channel.pathname,
 882                                         msg.u.snapshot_channel.relayd_id,
 883                                         msg.u.snapshot_channel.max_stream_size,
 884                                         ctx);
 885                         if (ret < 0) {
 886                                 ERR("Snapshot channel failed");
 887                                 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
 888                         }
 889                 }
 890
 891                 health_code_update();
 892
 893                 ret = consumer_send_status_msg(sock, ret_code);
 894                 if (ret < 0) {
 895                         /* Somehow, the session daemon is not responding anymore. */
 896                         goto end_nosignal;
 897                 }
 898                 break;
 899         }
 900         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 901         {
 902                 uint64_t key = msg.u.destroy_channel.key;
 903                 struct lttng_consumer_channel *channel;
 904
 905                 channel = consumer_find_channel(key);
 906                 if (!channel) {
 907                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 908                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 909                 }
 910
 911                 health_code_update();
 912
 913                 ret = consumer_send_status_msg(sock, ret_code);
 914                 if (ret < 0) {
 915                         /* Somehow, the session daemon is not responding anymore. */
 916                         goto end_nosignal;
 917                 }
 918
 919                 health_code_update();
 920
 921                 /* Stop right now if no channel was found. */
 922                 if (!channel) {
 923                         goto end_nosignal;
 924                 }
 925
 926                 /*
 927                  * This command should ONLY be issued for channel with streams set in
 928                  * no monitor mode.
 929                  */
 930                 assert(!channel->monitor);
 931
 932                 /*
 933                  * The refcount should ALWAYS be 0 in the case of a channel in no
 934                  * monitor mode.
 935                  */
 936                 assert(!uatomic_sub_return(&channel->refcount, 1));
 937
 938                 consumer_del_channel(channel);
 939
 940                 goto end_nosignal;
 941         }
 942         default:
 943                 goto end_nosignal;
 944         }
 945
 946 end_nosignal:
 947         rcu_read_unlock();
 948
 949         /*
 950          * Return 1 to indicate success since the 0 value can be a socket
 951          * shutdown during the recv() or send() call.
 952          */
 953         health_code_update();
 954         return 1;
 955
 956 error_fatal:
 957         rcu_read_unlock();
 958         /* This will issue a consumer stop. */
 959         return -1;
 960 }
 961
 962 /*
 963  * Populate index values of a kernel stream. Values are set in big endian order.
 964  *
 965  * Return 0 on success or else a negative value.
 966  */
 967 static int get_index_values(struct ctf_packet_index *index, int infd)
 968 {
 969         int ret;
 970
 971         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 972         if (ret < 0) {
 973                 PERROR("kernctl_get_timestamp_begin");
 974                 goto error;
 975         }
 976         index->timestamp_begin = htobe64(index->timestamp_begin);
 977
 978         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 979         if (ret < 0) {
 980                 PERROR("kernctl_get_timestamp_end");
 981                 goto error;
 982         }
 983         index->timestamp_end = htobe64(index->timestamp_end);
 984
 985         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 986         if (ret < 0) {
 987                 PERROR("kernctl_get_events_discarded");
 988                 goto error;
 989         }
 990         index->events_discarded = htobe64(index->events_discarded);
 991
 992         ret = kernctl_get_content_size(infd, &index->content_size);
 993         if (ret < 0) {
 994                 PERROR("kernctl_get_content_size");
 995                 goto error;
 996         }
 997         index->content_size = htobe64(index->content_size);
 998
 999         ret = kernctl_get_packet_size(infd, &index->packet_size);
1000         if (ret < 0) {
1001                 PERROR("kernctl_get_packet_size");
1002                 goto error;
1003         }
1004         index->packet_size = htobe64(index->packet_size);
1005
1006         ret = kernctl_get_stream_id(infd, &index->stream_id);
1007         if (ret < 0) {
1008                 PERROR("kernctl_get_stream_id");
1009                 goto error;
1010         }
1011         index->stream_id = htobe64(index->stream_id);
1012
1013 error:
1014         return ret;
1015 }
1016 /*
1017  * Sync metadata meaning request them to the session daemon and snapshot to the
1018  * metadata thread can consumer them.
1019  *
1020  * Metadata stream lock MUST be acquired.
1021  *
1022  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1023  * is empty or a negative value on error.
1024  */
1025 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1026 {
1027         int ret;
1028
1029         assert(metadata);
1030
1031         ret = kernctl_buffer_flush(metadata->wait_fd);
1032         if (ret < 0) {
1033                 ERR("Failed to flush kernel stream");
1034                 goto end;
1035         }
1036
1037         ret = kernctl_snapshot(metadata->wait_fd);
1038         if (ret < 0) {
1039                 if (errno != EAGAIN) {
1040                         ERR("Sync metadata, taking kernel snapshot failed.");
1041                         goto end;
1042                 }
1043                 DBG("Sync metadata, no new kernel metadata");
1044                 /* No new metadata, exit. */
1045                 ret = ENODATA;
1046                 goto end;
1047         }
1048
1049 end:
1050         return ret;
1051 }
1052
1053 /*
1054  * Consume data on a file descriptor and write it on a trace file.
1055  */
1056 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1057                 struct lttng_consumer_local_data *ctx)
1058 {
1059         unsigned long len, subbuf_size, padding;
1060         int err, write_index = 1;
1061         ssize_t ret = 0;
1062         int infd = stream->wait_fd;
1063         struct ctf_packet_index index;
1064
1065         DBG("In read_subbuffer (infd : %d)", infd);
1066
1067         /* Get the next subbuffer */
1068         err = kernctl_get_next_subbuf(infd);
1069         if (err != 0) {
1070                 /*
1071                  * This is a debug message even for single-threaded consumer,
1072                  * because poll() have more relaxed criterions than get subbuf,
1073                  * so get_subbuf may fail for short race windows where poll()
1074                  * would issue wakeups.
1075                  */
1076                 DBG("Reserving sub buffer failed (everything is normal, "
1077                                 "it is due to concurrency)");
1078                 ret = -errno;
1079                 goto end;
1080         }
1081
1082         /* Get the full subbuffer size including padding */
1083         err = kernctl_get_padded_subbuf_size(infd, &len);
1084         if (err != 0) {
1085                 perror("Getting sub-buffer len failed.");
1086                 ret = -errno;
1087                 goto end;
1088         }
1089
1090         if (!stream->metadata_flag) {
1091                 ret = get_index_values(&index, infd);
1092                 if (ret < 0) {
1093                         goto end;
1094                 }
1095         } else {
1096                 write_index = 0;
1097         }
1098
1099         switch (stream->chan->output) {
1100         case CONSUMER_CHANNEL_SPLICE:
1101                 /*
1102                  * XXX: The lttng-modules splice "actor" does not handle copying
1103                  * partial pages hence only using the subbuffer size without the
1104                  * padding makes the splice fail.
1105                  */
1106                 subbuf_size = len;
1107                 padding = 0;
1108
1109                 /* splice the subbuffer to the tracefile */
1110                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1111                                 padding, &index);
1112                 /*
1113                  * XXX: Splice does not support network streaming so the return value
1114                  * is simply checked against subbuf_size and not like the mmap() op.
1115                  */
1116                 if (ret != subbuf_size) {
1117                         /*
1118                          * display the error but continue processing to try
1119                          * to release the subbuffer
1120                          */
1121                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1122                                         ret, subbuf_size);
1123                         write_index = 0;
1124                 }
1125                 break;
1126         case CONSUMER_CHANNEL_MMAP:
1127                 /* Get subbuffer size without padding */
1128                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1129                 if (err != 0) {
1130                         perror("Getting sub-buffer len failed.");
1131                         ret = -errno;
1132                         goto end;
1133                 }
1134
1135                 /* Make sure the tracer is not gone mad on us! */
1136                 assert(len >= subbuf_size);
1137
1138                 padding = len - subbuf_size;
1139
1140                 /* write the subbuffer to the tracefile */
1141                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1142                                 padding, &index);
1143                 /*
1144                  * The mmap operation should write subbuf_size amount of data when
1145                  * network streaming or the full padding (len) size when we are _not_
1146                  * streaming.
1147                  */
1148                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1149                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1150                         /*
1151                          * Display the error but continue processing to try to release the
1152                          * subbuffer. This is a DBG statement since this is possible to
1153                          * happen without being a critical error.
1154                          */
1155                         DBG("Error writing to tracefile "
1156                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1157                                         ret, len, subbuf_size);
1158                         write_index = 0;
1159                 }
1160                 break;
1161         default:
1162                 ERR("Unknown output method");
1163                 ret = -EPERM;
1164         }
1165
1166         err = kernctl_put_next_subbuf(infd);
1167         if (err != 0) {
1168                 if (errno == EFAULT) {
1169                         perror("Error in unreserving sub buffer\n");
1170                 } else if (errno == EIO) {
1171                         /* Should never happen with newer LTTng versions */
1172                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1173                 }
1174                 ret = -errno;
1175                 goto end;
1176         }
1177
1178         /* Write index if needed. */
1179         if (!write_index) {
1180                 goto end;
1181         }
1182
1183         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1184                 /*
1185                  * In live, block until all the metadata is sent.
1186                  */
1187                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1188                 if (err < 0) {
1189                         goto end;
1190                 }
1191         }
1192
1193         err = consumer_stream_write_index(stream, &index);
1194         if (err < 0) {
1195                 goto end;
1196         }
1197
1198 end:
1199         return ret;
1200 }
1201
1202 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1203 {
1204         int ret;
1205
1206         assert(stream);
1207
1208         /*
1209          * Don't create anything if this is set for streaming or should not be
1210          * monitored.
1211          */
1212         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1213                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1214                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1215                                 stream->uid, stream->gid, NULL);
1216                 if (ret < 0) {
1217                         goto error;
1218                 }
1219                 stream->out_fd = ret;
1220                 stream->tracefile_size_current = 0;
1221
1222                 if (!stream->metadata_flag) {
1223                         ret = index_create_file(stream->chan->pathname,
1224                                         stream->name, stream->uid, stream->gid,
1225                                         stream->chan->tracefile_size,
1226                                         stream->tracefile_count_current);
1227                         if (ret < 0) {
1228                                 goto error;
1229                         }
1230                         stream->index_fd = ret;
1231                 }
1232         }
1233
1234         if (stream->output == LTTNG_EVENT_MMAP) {
1235                 /* get the len of the mmap region */
1236                 unsigned long mmap_len;
1237
1238                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1239                 if (ret != 0) {
1240                         PERROR("kernctl_get_mmap_len");
1241                         ret = -errno;
1242                         goto error_close_fd;
1243                 }
1244                 stream->mmap_len = (size_t) mmap_len;
1245
1246                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1247                                 MAP_PRIVATE, stream->wait_fd, 0);
1248                 if (stream->mmap_base == MAP_FAILED) {
1249                         PERROR("Error mmaping");
1250                         ret = -1;
1251                         goto error_close_fd;
1252                 }
1253         }
1254
1255         /* we return 0 to let the library handle the FD internally */
1256         return 0;
1257
1258 error_close_fd:
1259         if (stream->out_fd >= 0) {
1260                 int err;
1261
1262                 err = close(stream->out_fd);
1263                 assert(!err);
1264                 stream->out_fd = -1;
1265         }
1266 error:
1267         return ret;
1268 }
1269
1270 /*
1271  * Check if data is still being extracted from the buffers for a specific
1272  * stream. Consumer data lock MUST be acquired before calling this function
1273  * and the stream lock.
1274  *
1275  * Return 1 if the traced data are still getting read else 0 meaning that the
1276  * data is available for trace viewer reading.
1277  */
1278 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1279 {
1280         int ret;
1281
1282         assert(stream);
1283
1284         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1285                 ret = 0;
1286                 goto end;
1287         }
1288
1289         ret = kernctl_get_next_subbuf(stream->wait_fd);
1290         if (ret == 0) {
1291                 /* There is still data so let's put back this subbuffer. */
1292                 ret = kernctl_put_subbuf(stream->wait_fd);
1293                 assert(ret == 0);
1294                 ret = 1;   /* Data is pending */
1295                 goto end;
1296         }
1297
1298         /* Data is NOT pending and ready to be read. */
1299         ret = 0;
1300
1301 end:
1302         return ret;
1303 }