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 nb_packets_per_stream,
 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                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 224                                 produced_pos, nb_packets_per_stream,
 225                                 stream->max_sb_size);
 226
 227                 while (consumed_pos < produced_pos) {
 228                         ssize_t read_len;
 229                         unsigned long len, padded_len;
 230
 231                         health_code_update();
 232
 233                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 234
 235                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 236                         if (ret < 0) {
 237                                 if (errno != EAGAIN) {
 238                                         PERROR("kernctl_get_subbuf snapshot");
 239                                         ret = -errno;
 240                                         goto end_unlock;
 241                                 }
 242                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 243                                 consumed_pos += stream->max_sb_size;
 244                                 continue;
 245                         }
 246
 247                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 248                         if (ret < 0) {
 249                                 ERR("Snapshot kernctl_get_subbuf_size");
 250                                 ret = -errno;
 251                                 goto error_put_subbuf;
 252                         }
 253
 254                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 255                         if (ret < 0) {
 256                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 257                                 ret = -errno;
 258                                 goto error_put_subbuf;
 259                         }
 260
 261                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 262                                         padded_len - len, NULL);
 263                         /*
 264                          * We write the padded len in local tracefiles but the data len
 265                          * when using a relay. Display the error but continue processing
 266                          * to try to release the subbuffer.
 267                          */
 268                         if (relayd_id != (uint64_t) -1ULL) {
 269                                 if (read_len != len) {
 270                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 271                                                         read_len, len);
 272                                 }
 273                         } else {
 274                                 if (read_len != padded_len) {
 275                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 276                                                         read_len, padded_len);
 277                                 }
 278                         }
 279
 280                         ret = kernctl_put_subbuf(stream->wait_fd);
 281                         if (ret < 0) {
 282                                 ERR("Snapshot kernctl_put_subbuf");
 283                                 ret = -errno;
 284                                 goto end_unlock;
 285                         }
 286                         consumed_pos += stream->max_sb_size;
 287                 }
 288
 289                 if (relayd_id == (uint64_t) -1ULL) {
 290                         if (stream->out_fd >= 0) {
 291                                 ret = close(stream->out_fd);
 292                                 if (ret < 0) {
 293                                         PERROR("Kernel consumer snapshot close out_fd");
 294                                         goto end_unlock;
 295                                 }
 296                                 stream->out_fd = -1;
 297                         }
 298                 } else {
 299                         close_relayd_stream(stream);
 300                         stream->net_seq_idx = (uint64_t) -1ULL;
 301                 }
 302                 pthread_mutex_unlock(&stream->lock);
 303         }
 304
 305         /* All good! */
 306         ret = 0;
 307         goto end;
 308
 309 error_put_subbuf:
 310         ret = kernctl_put_subbuf(stream->wait_fd);
 311         if (ret < 0) {
 312                 ret = -errno;
 313                 ERR("Snapshot kernctl_put_subbuf error path");
 314         }
 315 end_unlock:
 316         pthread_mutex_unlock(&stream->lock);
 317 end:
 318         rcu_read_unlock();
 319         return ret;
 320 }
 321
 322 /*
 323  * Read the whole metadata available for a snapshot.
 324  *
 325  * Returns 0 on success, < 0 on error
 326  */
 327 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 328                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 329 {
 330         int ret, use_relayd = 0;
 331         ssize_t ret_read;
 332         struct lttng_consumer_channel *metadata_channel;
 333         struct lttng_consumer_stream *metadata_stream;
 334
 335         assert(ctx);
 336
 337         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 338                         key, path);
 339
 340         rcu_read_lock();
 341
 342         metadata_channel = consumer_find_channel(key);
 343         if (!metadata_channel) {
 344                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 345                 ret = -1;
 346                 goto error;
 347         }
 348
 349         metadata_stream = metadata_channel->metadata_stream;
 350         assert(metadata_stream);
 351
 352         /* Flag once that we have a valid relayd for the stream. */
 353         if (relayd_id != (uint64_t) -1ULL) {
 354                 use_relayd = 1;
 355         }
 356
 357         if (use_relayd) {
 358                 ret = consumer_send_relayd_stream(metadata_stream, path);
 359                 if (ret < 0) {
 360                         goto error;
 361                 }
 362         } else {
 363                 ret = utils_create_stream_file(path, metadata_stream->name,
 364                                 metadata_stream->chan->tracefile_size,
 365                                 metadata_stream->tracefile_count_current,
 366                                 metadata_stream->uid, metadata_stream->gid, NULL);
 367                 if (ret < 0) {
 368                         goto error;
 369                 }
 370                 metadata_stream->out_fd = ret;
 371         }
 372
 373         do {
 374                 health_code_update();
 375
 376                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 377                 if (ret_read < 0) {
 378                         if (ret_read != -EAGAIN) {
 379                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 380                                                 ret_read);
 381                                 goto error;
 382                         }
 383                         /* ret_read is negative at this point so we will exit the loop. */
 384                         continue;
 385                 }
 386         } while (ret_read >= 0);
 387
 388         if (use_relayd) {
 389                 close_relayd_stream(metadata_stream);
 390                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 391         } else {
 392                 if (metadata_stream->out_fd >= 0) {
 393                         ret = close(metadata_stream->out_fd);
 394                         if (ret < 0) {
 395                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 396                                 /*
 397                                  * Don't go on error here since the snapshot was successful at this
 398                                  * point but somehow the close failed.
 399                                  */
 400                         }
 401                         metadata_stream->out_fd = -1;
 402                 }
 403         }
 404
 405         ret = 0;
 406
 407         cds_list_del(&metadata_stream->send_node);
 408         consumer_stream_destroy(metadata_stream, NULL);
 409         metadata_channel->metadata_stream = NULL;
 410 error:
 411         rcu_read_unlock();
 412         return ret;
 413 }
 414
 415 /*
 416  * Receive command from session daemon and process it.
 417  *
 418  * Return 1 on success else a negative value or 0.
 419  */
 420 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 421                 int sock, struct pollfd *consumer_sockpoll)
 422 {
 423         ssize_t ret;
 424         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 425         struct lttcomm_consumer_msg msg;
 426
 427         health_code_update();
 428
 429         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 430         if (ret != sizeof(msg)) {
 431                 if (ret > 0) {
 432                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 433                         ret = -1;
 434                 }
 435                 return ret;
 436         }
 437
 438         health_code_update();
 439
 440         /* Deprecated command */
 441         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 442
 443         health_code_update();
 444
 445         /* relayd needs RCU read-side protection */
 446         rcu_read_lock();
 447
 448         switch (msg.cmd_type) {
 449         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 450         {
 451                 /* Session daemon status message are handled in the following call. */
 452                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 453                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 454                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 455                                  msg.u.relayd_sock.relayd_session_id);
 456                 goto end_nosignal;
 457         }
 458         case LTTNG_CONSUMER_ADD_CHANNEL:
 459         {
 460                 struct lttng_consumer_channel *new_channel;
 461                 int ret_recv;
 462
 463                 health_code_update();
 464
 465                 /* First send a status message before receiving the fds. */
 466                 ret = consumer_send_status_msg(sock, ret_code);
 467                 if (ret < 0) {
 468                         /* Somehow, the session daemon is not responding anymore. */
 469                         goto error_fatal;
 470                 }
 471
 472                 health_code_update();
 473
 474                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 475                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 476                                 msg.u.channel.session_id, msg.u.channel.pathname,
 477                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 478                                 msg.u.channel.relayd_id, msg.u.channel.output,
 479                                 msg.u.channel.tracefile_size,
 480                                 msg.u.channel.tracefile_count, 0,
 481                                 msg.u.channel.monitor,
 482                                 msg.u.channel.live_timer_interval);
 483                 if (new_channel == NULL) {
 484                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 485                         goto end_nosignal;
 486                 }
 487                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 488                 switch (msg.u.channel.output) {
 489                 case LTTNG_EVENT_SPLICE:
 490                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 491                         break;
 492                 case LTTNG_EVENT_MMAP:
 493                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 494                         break;
 495                 default:
 496                         ERR("Channel output unknown %d", msg.u.channel.output);
 497                         goto end_nosignal;
 498                 }
 499
 500                 /* Translate and save channel type. */
 501                 switch (msg.u.channel.type) {
 502                 case CONSUMER_CHANNEL_TYPE_DATA:
 503                 case CONSUMER_CHANNEL_TYPE_METADATA:
 504                         new_channel->type = msg.u.channel.type;
 505                         break;
 506                 default:
 507                         assert(0);
 508                         goto end_nosignal;
 509                 };
 510
 511                 health_code_update();
 512
 513                 if (ctx->on_recv_channel != NULL) {
 514                         ret_recv = ctx->on_recv_channel(new_channel);
 515                         if (ret_recv == 0) {
 516                                 ret = consumer_add_channel(new_channel, ctx);
 517                         } else if (ret_recv < 0) {
 518                                 goto end_nosignal;
 519                         }
 520                 } else {
 521                         ret = consumer_add_channel(new_channel, ctx);
 522                 }
 523                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 524                         consumer_timer_live_start(new_channel,
 525                                         msg.u.channel.live_timer_interval);
 526                 }
 527
 528                 health_code_update();
 529
 530                 /* If we received an error in add_channel, we need to report it. */
 531                 if (ret < 0) {
 532                         ret = consumer_send_status_msg(sock, ret);
 533                         if (ret < 0) {
 534                                 goto error_fatal;
 535                         }
 536                         goto end_nosignal;
 537                 }
 538
 539                 goto end_nosignal;
 540         }
 541         case LTTNG_CONSUMER_ADD_STREAM:
 542         {
 543                 int fd;
 544                 struct lttng_pipe *stream_pipe;
 545                 struct lttng_consumer_stream *new_stream;
 546                 struct lttng_consumer_channel *channel;
 547                 int alloc_ret = 0;
 548
 549                 /*
 550                  * Get stream's channel reference. Needed when adding the stream to the
 551                  * global hash table.
 552                  */
 553                 channel = consumer_find_channel(msg.u.stream.channel_key);
 554                 if (!channel) {
 555                         /*
 556                          * We could not find the channel. Can happen if cpu hotplug
 557                          * happens while tearing down.
 558                          */
 559                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 560                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 561                 }
 562
 563                 health_code_update();
 564
 565                 /* First send a status message before receiving the fds. */
 566                 ret = consumer_send_status_msg(sock, ret_code);
 567                 if (ret < 0) {
 568                         /* Somehow, the session daemon is not responding anymore. */
 569                         goto error_fatal;
 570                 }
 571
 572                 health_code_update();
 573
 574                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 575                         /* Channel was not found. */
 576                         goto end_nosignal;
 577                 }
 578
 579                 /* Blocking call */
 580                 health_poll_entry();
 581                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 582                 health_poll_exit();
 583                 if (ret) {
 584                         goto error_fatal;
 585                 }
 586
 587                 health_code_update();
 588
 589                 /* Get stream file descriptor from socket */
 590                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 591                 if (ret != sizeof(fd)) {
 592                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 593                         rcu_read_unlock();
 594                         return ret;
 595                 }
 596
 597                 health_code_update();
 598
 599                 /*
 600                  * Send status code to session daemon only if the recv works. If the
 601                  * above recv() failed, the session daemon is notified through the
 602                  * error socket and the teardown is eventually done.
 603                  */
 604                 ret = consumer_send_status_msg(sock, ret_code);
 605                 if (ret < 0) {
 606                         /* Somehow, the session daemon is not responding anymore. */
 607                         goto end_nosignal;
 608                 }
 609
 610                 health_code_update();
 611
 612                 new_stream = consumer_allocate_stream(channel->key,
 613                                 fd,
 614                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 615                                 channel->name,
 616                                 channel->uid,
 617                                 channel->gid,
 618                                 channel->relayd_id,
 619                                 channel->session_id,
 620                                 msg.u.stream.cpu,
 621                                 &alloc_ret,
 622                                 channel->type,
 623                                 channel->monitor);
 624                 if (new_stream == NULL) {
 625                         switch (alloc_ret) {
 626                         case -ENOMEM:
 627                         case -EINVAL:
 628                         default:
 629                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 630                                 break;
 631                         }
 632                         goto end_nosignal;
 633                 }
 634
 635                 new_stream->chan = channel;
 636                 new_stream->wait_fd = fd;
 637                 switch (channel->output) {
 638                 case CONSUMER_CHANNEL_SPLICE:
 639                         new_stream->output = LTTNG_EVENT_SPLICE;
 640                         ret = utils_create_pipe(new_stream->splice_pipe);
 641                         if (ret < 0) {
 642                                 goto end_nosignal;
 643                         }
 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 = LTTCOMM_CONSUMERD_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 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 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 = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 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 = LTTCOMM_CONSUMERD_ERROR_METADATA;
 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.nb_packets_per_stream,
 884                                         ctx);
 885                         if (ret < 0) {
 886                                 ERR("Snapshot channel failed");
 887                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 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 = LTTCOMM_CONSUMERD_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                 err = kernctl_put_subbuf(infd);
1087                 if (err != 0) {
1088                         if (errno == EFAULT) {
1089                                 PERROR("Error in unreserving sub buffer\n");
1090                         } else if (errno == EIO) {
1091                                 /* Should never happen with newer LTTng versions */
1092                                 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1093                         }
1094                         ret = -errno;
1095                         goto end;
1096                 }
1097                 ret = -errno;
1098                 goto end;
1099         }
1100
1101         if (!stream->metadata_flag) {
1102                 ret = get_index_values(&index, infd);
1103                 if (ret < 0) {
1104                         err = kernctl_put_subbuf(infd);
1105                         if (err != 0) {
1106                                 if (errno == EFAULT) {
1107                                         PERROR("Error in unreserving sub buffer\n");
1108                                 } else if (errno == EIO) {
1109                                         /* Should never happen with newer LTTng versions */
1110                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1111                                 }
1112                                 ret = -errno;
1113                                 goto end;
1114                         }
1115                         goto end;
1116                 }
1117         } else {
1118                 write_index = 0;
1119         }
1120
1121         switch (stream->chan->output) {
1122         case CONSUMER_CHANNEL_SPLICE:
1123                 /*
1124                  * XXX: The lttng-modules splice "actor" does not handle copying
1125                  * partial pages hence only using the subbuffer size without the
1126                  * padding makes the splice fail.
1127                  */
1128                 subbuf_size = len;
1129                 padding = 0;
1130
1131                 /* splice the subbuffer to the tracefile */
1132                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1133                                 padding, &index);
1134                 /*
1135                  * XXX: Splice does not support network streaming so the return value
1136                  * is simply checked against subbuf_size and not like the mmap() op.
1137                  */
1138                 if (ret != subbuf_size) {
1139                         /*
1140                          * display the error but continue processing to try
1141                          * to release the subbuffer
1142                          */
1143                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1144                                         ret, subbuf_size);
1145                         write_index = 0;
1146                 }
1147                 break;
1148         case CONSUMER_CHANNEL_MMAP:
1149                 /* Get subbuffer size without padding */
1150                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1151                 if (err != 0) {
1152                         PERROR("Getting sub-buffer len failed.");
1153                         err = kernctl_put_subbuf(infd);
1154                         if (err != 0) {
1155                                 if (errno == EFAULT) {
1156                                         PERROR("Error in unreserving sub buffer\n");
1157                                 } else if (errno == EIO) {
1158                                         /* Should never happen with newer LTTng versions */
1159                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1160                                 }
1161                                 ret = -errno;
1162                                 goto end;
1163                         }
1164                         ret = -errno;
1165                         goto end;
1166                 }
1167
1168                 /* Make sure the tracer is not gone mad on us! */
1169                 assert(len >= subbuf_size);
1170
1171                 padding = len - subbuf_size;
1172
1173                 /* write the subbuffer to the tracefile */
1174                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1175                                 padding, &index);
1176                 /*
1177                  * The mmap operation should write subbuf_size amount of data when
1178                  * network streaming or the full padding (len) size when we are _not_
1179                  * streaming.
1180                  */
1181                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1182                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1183                         /*
1184                          * Display the error but continue processing to try to release the
1185                          * subbuffer. This is a DBG statement since this is possible to
1186                          * happen without being a critical error.
1187                          */
1188                         DBG("Error writing to tracefile "
1189                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1190                                         ret, len, subbuf_size);
1191                         write_index = 0;
1192                 }
1193                 break;
1194         default:
1195                 ERR("Unknown output method");
1196                 ret = -EPERM;
1197         }
1198
1199         err = kernctl_put_next_subbuf(infd);
1200         if (err != 0) {
1201                 if (errno == EFAULT) {
1202                         PERROR("Error in unreserving sub buffer\n");
1203                 } else if (errno == EIO) {
1204                         /* Should never happen with newer LTTng versions */
1205                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1206                 }
1207                 ret = -errno;
1208                 goto end;
1209         }
1210
1211         /* Write index if needed. */
1212         if (!write_index) {
1213                 goto end;
1214         }
1215
1216         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1217                 /*
1218                  * In live, block until all the metadata is sent.
1219                  */
1220                 pthread_mutex_lock(&stream->metadata_timer_lock);
1221                 assert(!stream->missed_metadata_flush);
1222                 stream->waiting_on_metadata = true;
1223                 pthread_mutex_unlock(&stream->metadata_timer_lock);
1224
1225                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1226
1227                 pthread_mutex_lock(&stream->metadata_timer_lock);
1228                 stream->waiting_on_metadata = false;
1229                 if (stream->missed_metadata_flush) {
1230                         stream->missed_metadata_flush = false;
1231                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1232                         (void) consumer_flush_kernel_index(stream);
1233                 } else {
1234                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1235                 }
1236                 if (err < 0) {
1237                         goto end;
1238                 }
1239         }
1240
1241         err = consumer_stream_write_index(stream, &index);
1242         if (err < 0) {
1243                 goto end;
1244         }
1245
1246 end:
1247         return ret;
1248 }
1249
1250 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1251 {
1252         int ret;
1253
1254         assert(stream);
1255
1256         /*
1257          * Don't create anything if this is set for streaming or should not be
1258          * monitored.
1259          */
1260         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1261                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1262                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1263                                 stream->uid, stream->gid, NULL);
1264                 if (ret < 0) {
1265                         goto error;
1266                 }
1267                 stream->out_fd = ret;
1268                 stream->tracefile_size_current = 0;
1269
1270                 if (!stream->metadata_flag) {
1271                         ret = index_create_file(stream->chan->pathname,
1272                                         stream->name, stream->uid, stream->gid,
1273                                         stream->chan->tracefile_size,
1274                                         stream->tracefile_count_current);
1275                         if (ret < 0) {
1276                                 goto error;
1277                         }
1278                         stream->index_fd = ret;
1279                 }
1280         }
1281
1282         if (stream->output == LTTNG_EVENT_MMAP) {
1283                 /* get the len of the mmap region */
1284                 unsigned long mmap_len;
1285
1286                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1287                 if (ret != 0) {
1288                         PERROR("kernctl_get_mmap_len");
1289                         ret = -errno;
1290                         goto error_close_fd;
1291                 }
1292                 stream->mmap_len = (size_t) mmap_len;
1293
1294                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1295                                 MAP_PRIVATE, stream->wait_fd, 0);
1296                 if (stream->mmap_base == MAP_FAILED) {
1297                         PERROR("Error mmaping");
1298                         ret = -1;
1299                         goto error_close_fd;
1300                 }
1301         }
1302
1303         /* we return 0 to let the library handle the FD internally */
1304         return 0;
1305
1306 error_close_fd:
1307         if (stream->out_fd >= 0) {
1308                 int err;
1309
1310                 err = close(stream->out_fd);
1311                 assert(!err);
1312                 stream->out_fd = -1;
1313         }
1314 error:
1315         return ret;
1316 }
1317
1318 /*
1319  * Check if data is still being extracted from the buffers for a specific
1320  * stream. Consumer data lock MUST be acquired before calling this function
1321  * and the stream lock.
1322  *
1323  * Return 1 if the traced data are still getting read else 0 meaning that the
1324  * data is available for trace viewer reading.
1325  */
1326 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1327 {
1328         int ret;
1329
1330         assert(stream);
1331
1332         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1333                 ret = 0;
1334                 goto end;
1335         }
1336
1337         ret = kernctl_get_next_subbuf(stream->wait_fd);
1338         if (ret == 0) {
1339                 /* There is still data so let's put back this subbuffer. */
1340                 ret = kernctl_put_subbuf(stream->wait_fd);
1341                 assert(ret == 0);
1342                 ret = 1;   /* Data is pending */
1343                 goto end;
1344         }
1345
1346         /* Data is NOT pending and ready to be read. */
1347         ret = 0;
1348
1349 end:
1350         return ret;
1351 }