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/compat/endian.h>
  39 #include <common/pipe.h>
  40 #include <common/relayd/relayd.h>
  41 #include <common/utils.h>
  42 #include <common/consumer-stream.h>
  43 #include <common/index/index.h>
  44 #include <common/consumer-timer.h>
  45
  46 #include "kernel-consumer.h"
  47
  48 extern struct lttng_consumer_global_data consumer_data;
  49 extern int consumer_poll_timeout;
  50 extern volatile int consumer_quit;
  51
  52 /*
  53  * Take a snapshot for a specific fd
  54  *
  55  * Returns 0 on success, < 0 on error
  56  */
  57 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  58 {
  59         int ret = 0;
  60         int infd = stream->wait_fd;
  61
  62         ret = kernctl_snapshot(infd);
  63         if (ret != 0) {
  64                 perror("Getting sub-buffer snapshot.");
  65                 ret = -errno;
  66         }
  67
  68         return ret;
  69 }
  70
  71 /*
  72  * Get the produced position
  73  *
  74  * Returns 0 on success, < 0 on error
  75  */
  76 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  77                 unsigned long *pos)
  78 {
  79         int ret;
  80         int infd = stream->wait_fd;
  81
  82         ret = kernctl_snapshot_get_produced(infd, pos);
  83         if (ret != 0) {
  84                 perror("kernctl_snapshot_get_produced");
  85                 ret = -errno;
  86         }
  87
  88         return ret;
  89 }
  90
  91 /*
  92  * Get the consumerd position
  93  *
  94  * Returns 0 on success, < 0 on error
  95  */
  96 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  97                 unsigned long *pos)
  98 {
  99         int ret;
 100         int infd = stream->wait_fd;
 101
 102         ret = kernctl_snapshot_get_consumed(infd, pos);
 103         if (ret != 0) {
 104                 perror("kernctl_snapshot_get_consumed");
 105                 ret = -errno;
 106         }
 107
 108         return ret;
 109 }
 110
 111 /*
 112  * Take a snapshot of all the stream of a channel
 113  *
 114  * Returns 0 on success, < 0 on error
 115  */
 116 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 117                 uint64_t relayd_id, uint64_t max_stream_size,
 118                 struct lttng_consumer_local_data *ctx)
 119 {
 120         int ret;
 121         unsigned long consumed_pos, produced_pos;
 122         struct lttng_consumer_channel *channel;
 123         struct lttng_consumer_stream *stream;
 124
 125         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 126
 127         rcu_read_lock();
 128
 129         channel = consumer_find_channel(key);
 130         if (!channel) {
 131                 ERR("No channel found for key %" PRIu64, key);
 132                 ret = -1;
 133                 goto end;
 134         }
 135
 136         /* Splice is not supported yet for channel snapshot. */
 137         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 138                 ERR("Unsupported output %d", channel->output);
 139                 ret = -1;
 140                 goto end;
 141         }
 142
 143         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 144
 145                 health_code_update();
 146
 147                 /*
 148                  * Lock stream because we are about to change its state.
 149                  */
 150                 pthread_mutex_lock(&stream->lock);
 151
 152                 /*
 153                  * Assign the received relayd ID so we can use it for streaming. The streams
 154                  * are not visible to anyone so this is OK to change it.
 155                  */
 156                 stream->net_seq_idx = relayd_id;
 157                 channel->relayd_id = relayd_id;
 158                 if (relayd_id != (uint64_t) -1ULL) {
 159                         ret = consumer_send_relayd_stream(stream, path);
 160                         if (ret < 0) {
 161                                 ERR("sending stream to relayd");
 162                                 goto end_unlock;
 163                         }
 164                 } else {
 165                         ret = utils_create_stream_file(path, stream->name,
 166                                         stream->chan->tracefile_size,
 167                                         stream->tracefile_count_current,
 168                                         stream->uid, stream->gid, NULL);
 169                         if (ret < 0) {
 170                                 ERR("utils_create_stream_file");
 171                                 goto end_unlock;
 172                         }
 173
 174                         stream->out_fd = ret;
 175                         stream->tracefile_size_current = 0;
 176
 177                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 178                                         path, stream->name, stream->key);
 179                 }
 180                 if (relayd_id != -1ULL) {
 181                         ret = consumer_send_relayd_streams_sent(relayd_id);
 182                         if (ret < 0) {
 183                                 ERR("sending streams sent to relayd");
 184                                 goto end_unlock;
 185                         }
 186                 }
 187
 188                 ret = kernctl_buffer_flush(stream->wait_fd);
 189                 if (ret < 0) {
 190                         ERR("Failed to flush kernel stream");
 191                         ret = -errno;
 192                         goto end_unlock;
 193                 }
 194
 195                 ret = lttng_kconsumer_take_snapshot(stream);
 196                 if (ret < 0) {
 197                         ERR("Taking kernel snapshot");
 198                         goto end_unlock;
 199                 }
 200
 201                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 202                 if (ret < 0) {
 203                         ERR("Produced kernel snapshot position");
 204                         goto end_unlock;
 205                 }
 206
 207                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 208                 if (ret < 0) {
 209                         ERR("Consumerd kernel snapshot position");
 210                         goto end_unlock;
 211                 }
 212
 213                 if (stream->max_sb_size == 0) {
 214                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 215                                         &stream->max_sb_size);
 216                         if (ret < 0) {
 217                                 ERR("Getting kernel max_sb_size");
 218                                 ret = -errno;
 219                                 goto end_unlock;
 220                         }
 221                 }
 222
 223                 /*
 224                  * The original value is sent back if max stream size is larger than
 225                  * the possible size of the snapshot. Also, we asume that the session
 226                  * daemon should never send a maximum stream size that is lower than
 227                  * subbuffer size.
 228                  */
 229                 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
 230                                 produced_pos, max_stream_size);
 231
 232                 while (consumed_pos < produced_pos) {
 233                         ssize_t read_len;
 234                         unsigned long len, padded_len;
 235
 236                         health_code_update();
 237
 238                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 239
 240                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 241                         if (ret < 0) {
 242                                 if (errno != EAGAIN) {
 243                                         PERROR("kernctl_get_subbuf snapshot");
 244                                         ret = -errno;
 245                                         goto end_unlock;
 246                                 }
 247                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 248                                 consumed_pos += stream->max_sb_size;
 249                                 continue;
 250                         }
 251
 252                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 253                         if (ret < 0) {
 254                                 ERR("Snapshot kernctl_get_subbuf_size");
 255                                 ret = -errno;
 256                                 goto error_put_subbuf;
 257                         }
 258
 259                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 260                         if (ret < 0) {
 261                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 262                                 ret = -errno;
 263                                 goto error_put_subbuf;
 264                         }
 265
 266                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 267                                         padded_len - len, NULL);
 268                         /*
 269                          * We write the padded len in local tracefiles but the data len
 270                          * when using a relay. Display the error but continue processing
 271                          * to try to release the subbuffer.
 272                          */
 273                         if (relayd_id != (uint64_t) -1ULL) {
 274                                 if (read_len != len) {
 275                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 276                                                         read_len, len);
 277                                 }
 278                         } else {
 279                                 if (read_len != padded_len) {
 280                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 281                                                         read_len, padded_len);
 282                                 }
 283                         }
 284
 285                         ret = kernctl_put_subbuf(stream->wait_fd);
 286                         if (ret < 0) {
 287                                 ERR("Snapshot kernctl_put_subbuf");
 288                                 ret = -errno;
 289                                 goto end_unlock;
 290                         }
 291                         consumed_pos += stream->max_sb_size;
 292                 }
 293
 294                 if (relayd_id == (uint64_t) -1ULL) {
 295                         if (stream->out_fd >= 0) {
 296                                 ret = close(stream->out_fd);
 297                                 if (ret < 0) {
 298                                         PERROR("Kernel consumer snapshot close out_fd");
 299                                         goto end_unlock;
 300                                 }
 301                                 stream->out_fd = -1;
 302                         }
 303                 } else {
 304                         close_relayd_stream(stream);
 305                         stream->net_seq_idx = (uint64_t) -1ULL;
 306                 }
 307                 pthread_mutex_unlock(&stream->lock);
 308         }
 309
 310         /* All good! */
 311         ret = 0;
 312         goto end;
 313
 314 error_put_subbuf:
 315         ret = kernctl_put_subbuf(stream->wait_fd);
 316         if (ret < 0) {
 317                 ret = -errno;
 318                 ERR("Snapshot kernctl_put_subbuf error path");
 319         }
 320 end_unlock:
 321         pthread_mutex_unlock(&stream->lock);
 322 end:
 323         rcu_read_unlock();
 324         return ret;
 325 }
 326
 327 /*
 328  * Read the whole metadata available for a snapshot.
 329  *
 330  * Returns 0 on success, < 0 on error
 331  */
 332 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 333                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 334 {
 335         int ret, use_relayd = 0;
 336         ssize_t ret_read;
 337         struct lttng_consumer_channel *metadata_channel;
 338         struct lttng_consumer_stream *metadata_stream;
 339
 340         assert(ctx);
 341
 342         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 343                         key, path);
 344
 345         rcu_read_lock();
 346
 347         metadata_channel = consumer_find_channel(key);
 348         if (!metadata_channel) {
 349                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 350                 ret = -1;
 351                 goto error;
 352         }
 353
 354         metadata_stream = metadata_channel->metadata_stream;
 355         assert(metadata_stream);
 356
 357         /* Flag once that we have a valid relayd for the stream. */
 358         if (relayd_id != (uint64_t) -1ULL) {
 359                 use_relayd = 1;
 360         }
 361
 362         if (use_relayd) {
 363                 ret = consumer_send_relayd_stream(metadata_stream, path);
 364                 if (ret < 0) {
 365                         goto error;
 366                 }
 367         } else {
 368                 ret = utils_create_stream_file(path, metadata_stream->name,
 369                                 metadata_stream->chan->tracefile_size,
 370                                 metadata_stream->tracefile_count_current,
 371                                 metadata_stream->uid, metadata_stream->gid, NULL);
 372                 if (ret < 0) {
 373                         goto error;
 374                 }
 375                 metadata_stream->out_fd = ret;
 376         }
 377
 378         do {
 379                 health_code_update();
 380
 381                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 382                 if (ret_read < 0) {
 383                         if (ret_read != -EAGAIN) {
 384                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 385                                                 ret_read);
 386                                 goto error;
 387                         }
 388                         /* ret_read is negative at this point so we will exit the loop. */
 389                         continue;
 390                 }
 391         } while (ret_read >= 0);
 392
 393         if (use_relayd) {
 394                 close_relayd_stream(metadata_stream);
 395                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 396         } else {
 397                 if (metadata_stream->out_fd >= 0) {
 398                         ret = close(metadata_stream->out_fd);
 399                         if (ret < 0) {
 400                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 401                                 /*
 402                                  * Don't go on error here since the snapshot was successful at this
 403                                  * point but somehow the close failed.
 404                                  */
 405                         }
 406                         metadata_stream->out_fd = -1;
 407                 }
 408         }
 409
 410         ret = 0;
 411
 412         cds_list_del(&metadata_stream->send_node);
 413         consumer_stream_destroy(metadata_stream, NULL);
 414         metadata_channel->metadata_stream = NULL;
 415 error:
 416         rcu_read_unlock();
 417         return ret;
 418 }
 419
 420 /*
 421  * Receive command from session daemon and process it.
 422  *
 423  * Return 1 on success else a negative value or 0.
 424  */
 425 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 426                 int sock, struct pollfd *consumer_sockpoll)
 427 {
 428         ssize_t ret;
 429         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 430         struct lttcomm_consumer_msg msg;
 431
 432         health_code_update();
 433
 434         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 435         if (ret != sizeof(msg)) {
 436                 if (ret > 0) {
 437                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 438                         ret = -1;
 439                 }
 440                 return ret;
 441         }
 442
 443         health_code_update();
 444
 445         /* Deprecated command */
 446         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 447
 448         health_code_update();
 449
 450         /* relayd needs RCU read-side protection */
 451         rcu_read_lock();
 452
 453         switch (msg.cmd_type) {
 454         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 455         {
 456                 /* Session daemon status message are handled in the following call. */
 457                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 458                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 459                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 460                                  msg.u.relayd_sock.relayd_session_id);
 461                 goto end_nosignal;
 462         }
 463         case LTTNG_CONSUMER_ADD_CHANNEL:
 464         {
 465                 struct lttng_consumer_channel *new_channel;
 466                 int ret_recv;
 467
 468                 health_code_update();
 469
 470                 /* First send a status message before receiving the fds. */
 471                 ret = consumer_send_status_msg(sock, ret_code);
 472                 if (ret < 0) {
 473                         /* Somehow, the session daemon is not responding anymore. */
 474                         goto error_fatal;
 475                 }
 476
 477                 health_code_update();
 478
 479                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 480                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 481                                 msg.u.channel.session_id, msg.u.channel.pathname,
 482                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 483                                 msg.u.channel.relayd_id, msg.u.channel.output,
 484                                 msg.u.channel.tracefile_size,
 485                                 msg.u.channel.tracefile_count, 0,
 486                                 msg.u.channel.monitor,
 487                                 msg.u.channel.live_timer_interval);
 488                 if (new_channel == NULL) {
 489                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 490                         goto end_nosignal;
 491                 }
 492                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 493                 switch (msg.u.channel.output) {
 494                 case LTTNG_EVENT_SPLICE:
 495                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 496                         break;
 497                 case LTTNG_EVENT_MMAP:
 498                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 499                         break;
 500                 default:
 501                         ERR("Channel output unknown %d", msg.u.channel.output);
 502                         goto end_nosignal;
 503                 }
 504
 505                 /* Translate and save channel type. */
 506                 switch (msg.u.channel.type) {
 507                 case CONSUMER_CHANNEL_TYPE_DATA:
 508                 case CONSUMER_CHANNEL_TYPE_METADATA:
 509                         new_channel->type = msg.u.channel.type;
 510                         break;
 511                 default:
 512                         assert(0);
 513                         goto end_nosignal;
 514                 };
 515
 516                 health_code_update();
 517
 518                 if (ctx->on_recv_channel != NULL) {
 519                         ret_recv = ctx->on_recv_channel(new_channel);
 520                         if (ret_recv == 0) {
 521                                 ret = consumer_add_channel(new_channel, ctx);
 522                         } else if (ret_recv < 0) {
 523                                 goto end_nosignal;
 524                         }
 525                 } else {
 526                         ret = consumer_add_channel(new_channel, ctx);
 527                 }
 528                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 529                         consumer_timer_live_start(new_channel,
 530                                         msg.u.channel.live_timer_interval);
 531                 }
 532
 533                 health_code_update();
 534
 535                 /* If we received an error in add_channel, we need to report it. */
 536                 if (ret < 0) {
 537                         ret = consumer_send_status_msg(sock, ret);
 538                         if (ret < 0) {
 539                                 goto error_fatal;
 540                         }
 541                         goto end_nosignal;
 542                 }
 543
 544                 goto end_nosignal;
 545         }
 546         case LTTNG_CONSUMER_ADD_STREAM:
 547         {
 548                 int fd;
 549                 struct lttng_pipe *stream_pipe;
 550                 struct lttng_consumer_stream *new_stream;
 551                 struct lttng_consumer_channel *channel;
 552                 int alloc_ret = 0;
 553
 554                 /*
 555                  * Get stream's channel reference. Needed when adding the stream to the
 556                  * global hash table.
 557                  */
 558                 channel = consumer_find_channel(msg.u.stream.channel_key);
 559                 if (!channel) {
 560                         /*
 561                          * We could not find the channel. Can happen if cpu hotplug
 562                          * happens while tearing down.
 563                          */
 564                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 565                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 566                 }
 567
 568                 health_code_update();
 569
 570                 /* First send a status message before receiving the fds. */
 571                 ret = consumer_send_status_msg(sock, ret_code);
 572                 if (ret < 0) {
 573                         /* Somehow, the session daemon is not responding anymore. */
 574                         goto error_fatal;
 575                 }
 576
 577                 health_code_update();
 578
 579                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 580                         /* Channel was not found. */
 581                         goto end_nosignal;
 582                 }
 583
 584                 /* Blocking call */
 585                 health_poll_entry();
 586                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 587                 health_poll_exit();
 588                 if (ret) {
 589                         goto error_fatal;
 590                 }
 591
 592                 health_code_update();
 593
 594                 /* Get stream file descriptor from socket */
 595                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 596                 if (ret != sizeof(fd)) {
 597                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 598                         rcu_read_unlock();
 599                         return ret;
 600                 }
 601
 602                 health_code_update();
 603
 604                 /*
 605                  * Send status code to session daemon only if the recv works. If the
 606                  * above recv() failed, the session daemon is notified through the
 607                  * error socket and the teardown is eventually done.
 608                  */
 609                 ret = consumer_send_status_msg(sock, ret_code);
 610                 if (ret < 0) {
 611                         /* Somehow, the session daemon is not responding anymore. */
 612                         goto end_nosignal;
 613                 }
 614
 615                 health_code_update();
 616
 617                 new_stream = consumer_allocate_stream(channel->key,
 618                                 fd,
 619                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 620                                 channel->name,
 621                                 channel->uid,
 622                                 channel->gid,
 623                                 channel->relayd_id,
 624                                 channel->session_id,
 625                                 msg.u.stream.cpu,
 626                                 &alloc_ret,
 627                                 channel->type,
 628                                 channel->monitor);
 629                 if (new_stream == NULL) {
 630                         switch (alloc_ret) {
 631                         case -ENOMEM:
 632                         case -EINVAL:
 633                         default:
 634                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 635                                 break;
 636                         }
 637                         goto end_nosignal;
 638                 }
 639
 640                 new_stream->chan = channel;
 641                 new_stream->wait_fd = fd;
 642                 switch (channel->output) {
 643                 case CONSUMER_CHANNEL_SPLICE:
 644                         new_stream->output = LTTNG_EVENT_SPLICE;
 645                         ret = utils_create_pipe(new_stream->splice_pipe);
 646                         if (ret < 0) {
 647                                 goto end_nosignal;
 648                         }
 649                         break;
 650                 case CONSUMER_CHANNEL_MMAP:
 651                         new_stream->output = LTTNG_EVENT_MMAP;
 652                         break;
 653                 default:
 654                         ERR("Stream output unknown %d", channel->output);
 655                         goto end_nosignal;
 656                 }
 657
 658                 /*
 659                  * We've just assigned the channel to the stream so increment the
 660                  * refcount right now. We don't need to increment the refcount for
 661                  * streams in no monitor because we handle manually the cleanup of
 662                  * those. It is very important to make sure there is NO prior
 663                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 664                  */
 665                 if (channel->monitor) {
 666                         uatomic_inc(&new_stream->chan->refcount);
 667                 }
 668
 669                 /*
 670                  * The buffer flush is done on the session daemon side for the kernel
 671                  * so no need for the stream "hangup_flush_done" variable to be
 672                  * tracked. This is important for a kernel stream since we don't rely
 673                  * on the flush state of the stream to read data. It's not the case for
 674                  * user space tracing.
 675                  */
 676                 new_stream->hangup_flush_done = 0;
 677
 678                 health_code_update();
 679
 680                 if (ctx->on_recv_stream) {
 681                         ret = ctx->on_recv_stream(new_stream);
 682                         if (ret < 0) {
 683                                 consumer_stream_free(new_stream);
 684                                 goto end_nosignal;
 685                         }
 686                 }
 687
 688                 health_code_update();
 689
 690                 if (new_stream->metadata_flag) {
 691                         channel->metadata_stream = new_stream;
 692                 }
 693
 694                 /* Do not monitor this stream. */
 695                 if (!channel->monitor) {
 696                         DBG("Kernel consumer add stream %s in no monitor mode with "
 697                                         "relayd id %" PRIu64, new_stream->name,
 698                                         new_stream->net_seq_idx);
 699                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 700                         break;
 701                 }
 702
 703                 /* Send stream to relayd if the stream has an ID. */
 704                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 705                         ret = consumer_send_relayd_stream(new_stream,
 706                                         new_stream->chan->pathname);
 707                         if (ret < 0) {
 708                                 consumer_stream_free(new_stream);
 709                                 goto end_nosignal;
 710                         }
 711                 }
 712
 713                 /* Get the right pipe where the stream will be sent. */
 714                 if (new_stream->metadata_flag) {
 715                         ret = consumer_add_metadata_stream(new_stream);
 716                         if (ret) {
 717                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 718                                                 new_stream->key);
 719                                 consumer_stream_free(new_stream);
 720                                 goto end_nosignal;
 721                         }
 722                         stream_pipe = ctx->consumer_metadata_pipe;
 723                 } else {
 724                         ret = consumer_add_data_stream(new_stream);
 725                         if (ret) {
 726                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 727                                                 new_stream->key);
 728                                 consumer_stream_free(new_stream);
 729                                 goto end_nosignal;
 730                         }
 731                         stream_pipe = ctx->consumer_data_pipe;
 732                 }
 733
 734                 /* Vitible to other threads */
 735                 new_stream->globally_visible = 1;
 736
 737                 health_code_update();
 738
 739                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 740                 if (ret < 0) {
 741                         ERR("Consumer write %s stream to pipe %d",
 742                                         new_stream->metadata_flag ? "metadata" : "data",
 743                                         lttng_pipe_get_writefd(stream_pipe));
 744                         if (new_stream->metadata_flag) {
 745                                 consumer_del_stream_for_metadata(new_stream);
 746                         } else {
 747                                 consumer_del_stream_for_data(new_stream);
 748                         }
 749                         goto end_nosignal;
 750                 }
 751
 752                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 753                                 new_stream->name, fd, new_stream->relayd_stream_id);
 754                 break;
 755         }
 756         case LTTNG_CONSUMER_STREAMS_SENT:
 757         {
 758                 struct lttng_consumer_channel *channel;
 759
 760                 /*
 761                  * Get stream's channel reference. Needed when adding the stream to the
 762                  * global hash table.
 763                  */
 764                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 765                 if (!channel) {
 766                         /*
 767                          * We could not find the channel. Can happen if cpu hotplug
 768                          * happens while tearing down.
 769                          */
 770                         ERR("Unable to find channel key %" PRIu64,
 771                                         msg.u.sent_streams.channel_key);
 772                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 773                 }
 774
 775                 health_code_update();
 776
 777                 /*
 778                  * Send status code to session daemon.
 779                  */
 780                 ret = consumer_send_status_msg(sock, ret_code);
 781                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 782                         /* Somehow, the session daemon is not responding anymore. */
 783                         goto end_nosignal;
 784                 }
 785
 786                 health_code_update();
 787
 788                 /*
 789                  * We should not send this message if we don't monitor the
 790                  * streams in this channel.
 791                  */
 792                 if (!channel->monitor) {
 793                         break;
 794                 }
 795
 796                 health_code_update();
 797                 /* Send stream to relayd if the stream has an ID. */
 798                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 799                         ret = consumer_send_relayd_streams_sent(
 800                                         msg.u.sent_streams.net_seq_idx);
 801                         if (ret < 0) {
 802                                 goto end_nosignal;
 803                         }
 804                 }
 805                 break;
 806         }
 807         case LTTNG_CONSUMER_UPDATE_STREAM:
 808         {
 809                 rcu_read_unlock();
 810                 return -ENOSYS;
 811         }
 812         case LTTNG_CONSUMER_DESTROY_RELAYD:
 813         {
 814                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 815                 struct consumer_relayd_sock_pair *relayd;
 816
 817                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 818
 819                 /* Get relayd reference if exists. */
 820                 relayd = consumer_find_relayd(index);
 821                 if (relayd == NULL) {
 822                         DBG("Unable to find relayd %" PRIu64, index);
 823                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 824                 }
 825
 826                 /*
 827                  * Each relayd socket pair has a refcount of stream attached to it
 828                  * which tells if the relayd is still active or not depending on the
 829                  * refcount value.
 830                  *
 831                  * This will set the destroy flag of the relayd object and destroy it
 832                  * if the refcount reaches zero when called.
 833                  *
 834                  * The destroy can happen either here or when a stream fd hangs up.
 835                  */
 836                 if (relayd) {
 837                         consumer_flag_relayd_for_destroy(relayd);
 838                 }
 839
 840                 health_code_update();
 841
 842                 ret = consumer_send_status_msg(sock, ret_code);
 843                 if (ret < 0) {
 844                         /* Somehow, the session daemon is not responding anymore. */
 845                         goto error_fatal;
 846                 }
 847
 848                 goto end_nosignal;
 849         }
 850         case LTTNG_CONSUMER_DATA_PENDING:
 851         {
 852                 int32_t ret;
 853                 uint64_t id = msg.u.data_pending.session_id;
 854
 855                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 856
 857                 ret = consumer_data_pending(id);
 858
 859                 health_code_update();
 860
 861                 /* Send back returned value to session daemon */
 862                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 863                 if (ret < 0) {
 864                         PERROR("send data pending ret code");
 865                         goto error_fatal;
 866                 }
 867
 868                 /*
 869                  * No need to send back a status message since the data pending
 870                  * returned value is the response.
 871                  */
 872                 break;
 873         }
 874         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 875         {
 876                 if (msg.u.snapshot_channel.metadata == 1) {
 877                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 878                                         msg.u.snapshot_channel.pathname,
 879                                         msg.u.snapshot_channel.relayd_id, ctx);
 880                         if (ret < 0) {
 881                                 ERR("Snapshot metadata failed");
 882                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 883                         }
 884                 } else {
 885                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 886                                         msg.u.snapshot_channel.pathname,
 887                                         msg.u.snapshot_channel.relayd_id,
 888                                         msg.u.snapshot_channel.max_stream_size,
 889                                         ctx);
 890                         if (ret < 0) {
 891                                 ERR("Snapshot channel failed");
 892                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 893                         }
 894                 }
 895
 896                 health_code_update();
 897
 898                 ret = consumer_send_status_msg(sock, ret_code);
 899                 if (ret < 0) {
 900                         /* Somehow, the session daemon is not responding anymore. */
 901                         goto end_nosignal;
 902                 }
 903                 break;
 904         }
 905         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 906         {
 907                 uint64_t key = msg.u.destroy_channel.key;
 908                 struct lttng_consumer_channel *channel;
 909
 910                 channel = consumer_find_channel(key);
 911                 if (!channel) {
 912                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 913                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 914                 }
 915
 916                 health_code_update();
 917
 918                 ret = consumer_send_status_msg(sock, ret_code);
 919                 if (ret < 0) {
 920                         /* Somehow, the session daemon is not responding anymore. */
 921                         goto end_nosignal;
 922                 }
 923
 924                 health_code_update();
 925
 926                 /* Stop right now if no channel was found. */
 927                 if (!channel) {
 928                         goto end_nosignal;
 929                 }
 930
 931                 /*
 932                  * This command should ONLY be issued for channel with streams set in
 933                  * no monitor mode.
 934                  */
 935                 assert(!channel->monitor);
 936
 937                 /*
 938                  * The refcount should ALWAYS be 0 in the case of a channel in no
 939                  * monitor mode.
 940                  */
 941                 assert(!uatomic_sub_return(&channel->refcount, 1));
 942
 943                 consumer_del_channel(channel);
 944
 945                 goto end_nosignal;
 946         }
 947         default:
 948                 goto end_nosignal;
 949         }
 950
 951 end_nosignal:
 952         rcu_read_unlock();
 953
 954         /*
 955          * Return 1 to indicate success since the 0 value can be a socket
 956          * shutdown during the recv() or send() call.
 957          */
 958         health_code_update();
 959         return 1;
 960
 961 error_fatal:
 962         rcu_read_unlock();
 963         /* This will issue a consumer stop. */
 964         return -1;
 965 }
 966
 967 /*
 968  * Populate index values of a kernel stream. Values are set in big endian order.
 969  *
 970  * Return 0 on success or else a negative value.
 971  */
 972 static int get_index_values(struct ctf_packet_index *index, int infd)
 973 {
 974         int ret;
 975
 976         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 977         if (ret < 0) {
 978                 PERROR("kernctl_get_timestamp_begin");
 979                 goto error;
 980         }
 981         index->timestamp_begin = htobe64(index->timestamp_begin);
 982
 983         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 984         if (ret < 0) {
 985                 PERROR("kernctl_get_timestamp_end");
 986                 goto error;
 987         }
 988         index->timestamp_end = htobe64(index->timestamp_end);
 989
 990         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 991         if (ret < 0) {
 992                 PERROR("kernctl_get_events_discarded");
 993                 goto error;
 994         }
 995         index->events_discarded = htobe64(index->events_discarded);
 996
 997         ret = kernctl_get_content_size(infd, &index->content_size);
 998         if (ret < 0) {
 999                 PERROR("kernctl_get_content_size");
1000                 goto error;
1001         }
1002         index->content_size = htobe64(index->content_size);
1003
1004         ret = kernctl_get_packet_size(infd, &index->packet_size);
1005         if (ret < 0) {
1006                 PERROR("kernctl_get_packet_size");
1007                 goto error;
1008         }
1009         index->packet_size = htobe64(index->packet_size);
1010
1011         ret = kernctl_get_stream_id(infd, &index->stream_id);
1012         if (ret < 0) {
1013                 PERROR("kernctl_get_stream_id");
1014                 goto error;
1015         }
1016         index->stream_id = htobe64(index->stream_id);
1017
1018 error:
1019         return ret;
1020 }
1021 /*
1022  * Sync metadata meaning request them to the session daemon and snapshot to the
1023  * metadata thread can consumer them.
1024  *
1025  * Metadata stream lock MUST be acquired.
1026  *
1027  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1028  * is empty or a negative value on error.
1029  */
1030 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1031 {
1032         int ret;
1033
1034         assert(metadata);
1035
1036         ret = kernctl_buffer_flush(metadata->wait_fd);
1037         if (ret < 0) {
1038                 ERR("Failed to flush kernel stream");
1039                 goto end;
1040         }
1041
1042         ret = kernctl_snapshot(metadata->wait_fd);
1043         if (ret < 0) {
1044                 if (errno != EAGAIN) {
1045                         ERR("Sync metadata, taking kernel snapshot failed.");
1046                         goto end;
1047                 }
1048                 DBG("Sync metadata, no new kernel metadata");
1049                 /* No new metadata, exit. */
1050                 ret = ENODATA;
1051                 goto end;
1052         }
1053
1054 end:
1055         return ret;
1056 }
1057
1058 /*
1059  * Consume data on a file descriptor and write it on a trace file.
1060  */
1061 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1062                 struct lttng_consumer_local_data *ctx)
1063 {
1064         unsigned long len, subbuf_size, padding;
1065         int err, write_index = 1;
1066         ssize_t ret = 0;
1067         int infd = stream->wait_fd;
1068         struct ctf_packet_index index;
1069
1070         DBG("In read_subbuffer (infd : %d)", infd);
1071
1072         /* Get the next subbuffer */
1073         err = kernctl_get_next_subbuf(infd);
1074         if (err != 0) {
1075                 /*
1076                  * This is a debug message even for single-threaded consumer,
1077                  * because poll() have more relaxed criterions than get subbuf,
1078                  * so get_subbuf may fail for short race windows where poll()
1079                  * would issue wakeups.
1080                  */
1081                 DBG("Reserving sub buffer failed (everything is normal, "
1082                                 "it is due to concurrency)");
1083                 ret = -errno;
1084                 goto end;
1085         }
1086
1087         /* Get the full subbuffer size including padding */
1088         err = kernctl_get_padded_subbuf_size(infd, &len);
1089         if (err != 0) {
1090                 perror("Getting sub-buffer len failed.");
1091                 ret = -errno;
1092                 goto end;
1093         }
1094
1095         if (!stream->metadata_flag) {
1096                 ret = get_index_values(&index, infd);
1097                 if (ret < 0) {
1098                         goto end;
1099                 }
1100         } else {
1101                 write_index = 0;
1102         }
1103
1104         switch (stream->chan->output) {
1105         case CONSUMER_CHANNEL_SPLICE:
1106                 /*
1107                  * XXX: The lttng-modules splice "actor" does not handle copying
1108                  * partial pages hence only using the subbuffer size without the
1109                  * padding makes the splice fail.
1110                  */
1111                 subbuf_size = len;
1112                 padding = 0;
1113
1114                 /* splice the subbuffer to the tracefile */
1115                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1116                                 padding, &index);
1117                 /*
1118                  * XXX: Splice does not support network streaming so the return value
1119                  * is simply checked against subbuf_size and not like the mmap() op.
1120                  */
1121                 if (ret != subbuf_size) {
1122                         /*
1123                          * display the error but continue processing to try
1124                          * to release the subbuffer
1125                          */
1126                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1127                                         ret, subbuf_size);
1128                         write_index = 0;
1129                 }
1130                 break;
1131         case CONSUMER_CHANNEL_MMAP:
1132                 /* Get subbuffer size without padding */
1133                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1134                 if (err != 0) {
1135                         perror("Getting sub-buffer len failed.");
1136                         ret = -errno;
1137                         goto end;
1138                 }
1139
1140                 /* Make sure the tracer is not gone mad on us! */
1141                 assert(len >= subbuf_size);
1142
1143                 padding = len - subbuf_size;
1144
1145                 /* write the subbuffer to the tracefile */
1146                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1147                                 padding, &index);
1148                 /*
1149                  * The mmap operation should write subbuf_size amount of data when
1150                  * network streaming or the full padding (len) size when we are _not_
1151                  * streaming.
1152                  */
1153                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1154                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1155                         /*
1156                          * Display the error but continue processing to try to release the
1157                          * subbuffer. This is a DBG statement since this is possible to
1158                          * happen without being a critical error.
1159                          */
1160                         DBG("Error writing to tracefile "
1161                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1162                                         ret, len, subbuf_size);
1163                         write_index = 0;
1164                 }
1165                 break;
1166         default:
1167                 ERR("Unknown output method");
1168                 ret = -EPERM;
1169         }
1170
1171         err = kernctl_put_next_subbuf(infd);
1172         if (err != 0) {
1173                 if (errno == EFAULT) {
1174                         perror("Error in unreserving sub buffer\n");
1175                 } else if (errno == EIO) {
1176                         /* Should never happen with newer LTTng versions */
1177                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1178                 }
1179                 ret = -errno;
1180                 goto end;
1181         }
1182
1183         /* Write index if needed. */
1184         if (!write_index) {
1185                 goto end;
1186         }
1187
1188         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1189                 /*
1190                  * In live, block until all the metadata is sent.
1191                  */
1192                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1193                 if (err < 0) {
1194                         goto end;
1195                 }
1196         }
1197
1198         err = consumer_stream_write_index(stream, &index);
1199         if (err < 0) {
1200                 goto end;
1201         }
1202
1203 end:
1204         return ret;
1205 }
1206
1207 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1208 {
1209         int ret;
1210
1211         assert(stream);
1212
1213         /*
1214          * Don't create anything if this is set for streaming or should not be
1215          * monitored.
1216          */
1217         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1218                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1219                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1220                                 stream->uid, stream->gid, NULL);
1221                 if (ret < 0) {
1222                         goto error;
1223                 }
1224                 stream->out_fd = ret;
1225                 stream->tracefile_size_current = 0;
1226
1227                 if (!stream->metadata_flag) {
1228                         ret = index_create_file(stream->chan->pathname,
1229                                         stream->name, stream->uid, stream->gid,
1230                                         stream->chan->tracefile_size,
1231                                         stream->tracefile_count_current);
1232                         if (ret < 0) {
1233                                 goto error;
1234                         }
1235                         stream->index_fd = ret;
1236                 }
1237         }
1238
1239         if (stream->output == LTTNG_EVENT_MMAP) {
1240                 /* get the len of the mmap region */
1241                 unsigned long mmap_len;
1242
1243                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1244                 if (ret != 0) {
1245                         PERROR("kernctl_get_mmap_len");
1246                         ret = -errno;
1247                         goto error_close_fd;
1248                 }
1249                 stream->mmap_len = (size_t) mmap_len;
1250
1251                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1252                                 MAP_PRIVATE, stream->wait_fd, 0);
1253                 if (stream->mmap_base == MAP_FAILED) {
1254                         PERROR("Error mmaping");
1255                         ret = -1;
1256                         goto error_close_fd;
1257                 }
1258         }
1259
1260         /* we return 0 to let the library handle the FD internally */
1261         return 0;
1262
1263 error_close_fd:
1264         if (stream->out_fd >= 0) {
1265                 int err;
1266
1267                 err = close(stream->out_fd);
1268                 assert(!err);
1269                 stream->out_fd = -1;
1270         }
1271 error:
1272         return ret;
1273 }
1274
1275 /*
1276  * Check if data is still being extracted from the buffers for a specific
1277  * stream. Consumer data lock MUST be acquired before calling this function
1278  * and the stream lock.
1279  *
1280  * Return 1 if the traced data are still getting read else 0 meaning that the
1281  * data is available for trace viewer reading.
1282  */
1283 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1284 {
1285         int ret;
1286
1287         assert(stream);
1288
1289         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1290                 ret = 0;
1291                 goto end;
1292         }
1293
1294         ret = kernctl_get_next_subbuf(stream->wait_fd);
1295         if (ret == 0) {
1296                 /* There is still data so let's put back this subbuffer. */
1297                 ret = kernctl_put_subbuf(stream->wait_fd);
1298                 assert(ret == 0);
1299                 ret = 1;   /* Data is pending */
1300                 goto end;
1301         }
1302
1303         /* Data is NOT pending and ready to be read. */
1304         ret = 0;
1305
1306 end:
1307         return ret;
1308 }