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 #define _LGPL_SOURCE
  21 #include <assert.h>
  22 #include <poll.h>
  23 #include <pthread.h>
  24 #include <stdlib.h>
  25 #include <string.h>
  26 #include <sys/mman.h>
  27 #include <sys/socket.h>
  28 #include <sys/types.h>
  29 #include <inttypes.h>
  30 #include <unistd.h>
  31 #include <sys/stat.h>
  32
  33 #include <bin/lttng-consumerd/health-consumerd.h>
  34 #include <common/common.h>
  35 #include <common/kernel-ctl/kernel-ctl.h>
  36 #include <common/sessiond-comm/sessiond-comm.h>
  37 #include <common/sessiond-comm/relayd.h>
  38 #include <common/compat/fcntl.h>
  39 #include <common/compat/endian.h>
  40 #include <common/pipe.h>
  41 #include <common/relayd/relayd.h>
  42 #include <common/utils.h>
  43 #include <common/consumer-stream.h>
  44 #include <common/index/index.h>
  45 #include <common/consumer-timer.h>
  46
  47 #include "kernel-consumer.h"
  48
  49 extern struct lttng_consumer_global_data consumer_data;
  50 extern int consumer_poll_timeout;
  51 extern volatile int consumer_quit;
  52
  53 /*
  54  * Take a snapshot for a specific fd
  55  *
  56  * Returns 0 on success, < 0 on error
  57  */
  58 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  59 {
  60         int ret = 0;
  61         int infd = stream->wait_fd;
  62
  63         ret = kernctl_snapshot(infd);
  64         if (ret != 0) {
  65                 PERROR("Getting sub-buffer snapshot.");
  66                 ret = -errno;
  67         }
  68
  69         return ret;
  70 }
  71
  72 /*
  73  * Get the produced position
  74  *
  75  * Returns 0 on success, < 0 on error
  76  */
  77 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  78                 unsigned long *pos)
  79 {
  80         int ret;
  81         int infd = stream->wait_fd;
  82
  83         ret = kernctl_snapshot_get_produced(infd, pos);
  84         if (ret != 0) {
  85                 PERROR("kernctl_snapshot_get_produced");
  86                 ret = -errno;
  87         }
  88
  89         return ret;
  90 }
  91
  92 /*
  93  * Get the consumerd position
  94  *
  95  * Returns 0 on success, < 0 on error
  96  */
  97 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  98                 unsigned long *pos)
  99 {
 100         int ret;
 101         int infd = stream->wait_fd;
 102
 103         ret = kernctl_snapshot_get_consumed(infd, pos);
 104         if (ret != 0) {
 105                 PERROR("kernctl_snapshot_get_consumed");
 106                 ret = -errno;
 107         }
 108
 109         return ret;
 110 }
 111
 112 /*
 113  * Take a snapshot of all the stream of a channel
 114  *
 115  * Returns 0 on success, < 0 on error
 116  */
 117 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 118                 uint64_t relayd_id, uint64_t nb_packets_per_stream,
 119                 struct lttng_consumer_local_data *ctx)
 120 {
 121         int ret;
 122         unsigned long consumed_pos, produced_pos;
 123         struct lttng_consumer_channel *channel;
 124         struct lttng_consumer_stream *stream;
 125
 126         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 127
 128         rcu_read_lock();
 129
 130         channel = consumer_find_channel(key);
 131         if (!channel) {
 132                 ERR("No channel found for key %" PRIu64, key);
 133                 ret = -1;
 134                 goto end;
 135         }
 136
 137         /* Splice is not supported yet for channel snapshot. */
 138         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 139                 ERR("Unsupported output %d", channel->output);
 140                 ret = -1;
 141                 goto end;
 142         }
 143
 144         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 145
 146                 health_code_update();
 147
 148                 /*
 149                  * Lock stream because we are about to change its state.
 150                  */
 151                 pthread_mutex_lock(&stream->lock);
 152
 153                 /*
 154                  * Assign the received relayd ID so we can use it for streaming. The streams
 155                  * are not visible to anyone so this is OK to change it.
 156                  */
 157                 stream->net_seq_idx = relayd_id;
 158                 channel->relayd_id = relayd_id;
 159                 if (relayd_id != (uint64_t) -1ULL) {
 160                         ret = consumer_send_relayd_stream(stream, path);
 161                         if (ret < 0) {
 162                                 ERR("sending stream to relayd");
 163                                 goto end_unlock;
 164                         }
 165                 } else {
 166                         ret = utils_create_stream_file(path, stream->name,
 167                                         stream->chan->tracefile_size,
 168                                         stream->tracefile_count_current,
 169                                         stream->uid, stream->gid, NULL);
 170                         if (ret < 0) {
 171                                 ERR("utils_create_stream_file");
 172                                 goto end_unlock;
 173                         }
 174
 175                         stream->out_fd = ret;
 176                         stream->tracefile_size_current = 0;
 177
 178                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 179                                         path, stream->name, stream->key);
 180                 }
 181                 if (relayd_id != -1ULL) {
 182                         ret = consumer_send_relayd_streams_sent(relayd_id);
 183                         if (ret < 0) {
 184                                 ERR("sending streams sent to relayd");
 185                                 goto end_unlock;
 186                         }
 187                 }
 188
 189                 ret = kernctl_buffer_flush(stream->wait_fd);
 190                 if (ret < 0) {
 191                         ERR("Failed to flush kernel stream");
 192                         ret = -errno;
 193                         goto end_unlock;
 194                 }
 195
 196                 ret = lttng_kconsumer_take_snapshot(stream);
 197                 if (ret < 0) {
 198                         ERR("Taking kernel snapshot");
 199                         goto end_unlock;
 200                 }
 201
 202                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 203                 if (ret < 0) {
 204                         ERR("Produced kernel snapshot position");
 205                         goto end_unlock;
 206                 }
 207
 208                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 209                 if (ret < 0) {
 210                         ERR("Consumerd kernel snapshot position");
 211                         goto end_unlock;
 212                 }
 213
 214                 if (stream->max_sb_size == 0) {
 215                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 216                                         &stream->max_sb_size);
 217                         if (ret < 0) {
 218                                 ERR("Getting kernel max_sb_size");
 219                                 ret = -errno;
 220                                 goto end_unlock;
 221                         }
 222                 }
 223
 224                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 225                                 produced_pos, nb_packets_per_stream,
 226                                 stream->max_sb_size);
 227
 228                 while (consumed_pos < produced_pos) {
 229                         ssize_t read_len;
 230                         unsigned long len, padded_len;
 231
 232                         health_code_update();
 233
 234                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 235
 236                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 237                         if (ret < 0) {
 238                                 if (errno != EAGAIN) {
 239                                         PERROR("kernctl_get_subbuf snapshot");
 240                                         ret = -errno;
 241                                         goto end_unlock;
 242                                 }
 243                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 244                                 consumed_pos += stream->max_sb_size;
 245                                 continue;
 246                         }
 247
 248                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 249                         if (ret < 0) {
 250                                 ERR("Snapshot kernctl_get_subbuf_size");
 251                                 ret = -errno;
 252                                 goto error_put_subbuf;
 253                         }
 254
 255                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 256                         if (ret < 0) {
 257                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 258                                 ret = -errno;
 259                                 goto error_put_subbuf;
 260                         }
 261
 262                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 263                                         padded_len - len, NULL);
 264                         /*
 265                          * We write the padded len in local tracefiles but the data len
 266                          * when using a relay. Display the error but continue processing
 267                          * to try to release the subbuffer.
 268                          */
 269                         if (relayd_id != (uint64_t) -1ULL) {
 270                                 if (read_len != len) {
 271                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 272                                                         read_len, len);
 273                                 }
 274                         } else {
 275                                 if (read_len != padded_len) {
 276                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 277                                                         read_len, padded_len);
 278                                 }
 279                         }
 280
 281                         ret = kernctl_put_subbuf(stream->wait_fd);
 282                         if (ret < 0) {
 283                                 ERR("Snapshot kernctl_put_subbuf");
 284                                 ret = -errno;
 285                                 goto end_unlock;
 286                         }
 287                         consumed_pos += stream->max_sb_size;
 288                 }
 289
 290                 if (relayd_id == (uint64_t) -1ULL) {
 291                         if (stream->out_fd >= 0) {
 292                                 ret = close(stream->out_fd);
 293                                 if (ret < 0) {
 294                                         PERROR("Kernel consumer snapshot close out_fd");
 295                                         goto end_unlock;
 296                                 }
 297                                 stream->out_fd = -1;
 298                         }
 299                 } else {
 300                         close_relayd_stream(stream);
 301                         stream->net_seq_idx = (uint64_t) -1ULL;
 302                 }
 303                 pthread_mutex_unlock(&stream->lock);
 304         }
 305
 306         /* All good! */
 307         ret = 0;
 308         goto end;
 309
 310 error_put_subbuf:
 311         ret = kernctl_put_subbuf(stream->wait_fd);
 312         if (ret < 0) {
 313                 ret = -errno;
 314                 ERR("Snapshot kernctl_put_subbuf error path");
 315         }
 316 end_unlock:
 317         pthread_mutex_unlock(&stream->lock);
 318 end:
 319         rcu_read_unlock();
 320         return ret;
 321 }
 322
 323 /*
 324  * Read the whole metadata available for a snapshot.
 325  *
 326  * Returns 0 on success, < 0 on error
 327  */
 328 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 329                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 330 {
 331         int ret, use_relayd = 0;
 332         ssize_t ret_read;
 333         struct lttng_consumer_channel *metadata_channel;
 334         struct lttng_consumer_stream *metadata_stream;
 335
 336         assert(ctx);
 337
 338         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 339                         key, path);
 340
 341         rcu_read_lock();
 342
 343         metadata_channel = consumer_find_channel(key);
 344         if (!metadata_channel) {
 345                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 346                 ret = -1;
 347                 goto error;
 348         }
 349
 350         metadata_stream = metadata_channel->metadata_stream;
 351         assert(metadata_stream);
 352
 353         /* Flag once that we have a valid relayd for the stream. */
 354         if (relayd_id != (uint64_t) -1ULL) {
 355                 use_relayd = 1;
 356         }
 357
 358         if (use_relayd) {
 359                 ret = consumer_send_relayd_stream(metadata_stream, path);
 360                 if (ret < 0) {
 361                         goto error;
 362                 }
 363         } else {
 364                 ret = utils_create_stream_file(path, metadata_stream->name,
 365                                 metadata_stream->chan->tracefile_size,
 366                                 metadata_stream->tracefile_count_current,
 367                                 metadata_stream->uid, metadata_stream->gid, NULL);
 368                 if (ret < 0) {
 369                         goto error;
 370                 }
 371                 metadata_stream->out_fd = ret;
 372         }
 373
 374         do {
 375                 health_code_update();
 376
 377                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 378                 if (ret_read < 0) {
 379                         if (ret_read != -EAGAIN) {
 380                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 381                                                 ret_read);
 382                                 goto error;
 383                         }
 384                         /* ret_read is negative at this point so we will exit the loop. */
 385                         continue;
 386                 }
 387         } while (ret_read >= 0);
 388
 389         if (use_relayd) {
 390                 close_relayd_stream(metadata_stream);
 391                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 392         } else {
 393                 if (metadata_stream->out_fd >= 0) {
 394                         ret = close(metadata_stream->out_fd);
 395                         if (ret < 0) {
 396                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 397                                 /*
 398                                  * Don't go on error here since the snapshot was successful at this
 399                                  * point but somehow the close failed.
 400                                  */
 401                         }
 402                         metadata_stream->out_fd = -1;
 403                 }
 404         }
 405
 406         ret = 0;
 407
 408         cds_list_del(&metadata_stream->send_node);
 409         consumer_stream_destroy(metadata_stream, NULL);
 410         metadata_channel->metadata_stream = NULL;
 411 error:
 412         rcu_read_unlock();
 413         return ret;
 414 }
 415
 416 /*
 417  * Receive command from session daemon and process it.
 418  *
 419  * Return 1 on success else a negative value or 0.
 420  */
 421 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 422                 int sock, struct pollfd *consumer_sockpoll)
 423 {
 424         ssize_t ret;
 425         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 426         struct lttcomm_consumer_msg msg;
 427
 428         health_code_update();
 429
 430         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 431         if (ret != sizeof(msg)) {
 432                 if (ret > 0) {
 433                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 434                         ret = -1;
 435                 }
 436                 return ret;
 437         }
 438
 439         health_code_update();
 440
 441         /* Deprecated command */
 442         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 443
 444         health_code_update();
 445
 446         /* relayd needs RCU read-side protection */
 447         rcu_read_lock();
 448
 449         switch (msg.cmd_type) {
 450         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 451         {
 452                 /* Session daemon status message are handled in the following call. */
 453                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 454                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 455                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 456                                  msg.u.relayd_sock.relayd_session_id);
 457                 goto end_nosignal;
 458         }
 459         case LTTNG_CONSUMER_ADD_CHANNEL:
 460         {
 461                 struct lttng_consumer_channel *new_channel;
 462                 int ret_recv;
 463
 464                 health_code_update();
 465
 466                 /* First send a status message before receiving the fds. */
 467                 ret = consumer_send_status_msg(sock, ret_code);
 468                 if (ret < 0) {
 469                         /* Somehow, the session daemon is not responding anymore. */
 470                         goto error_fatal;
 471                 }
 472
 473                 health_code_update();
 474
 475                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 476                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 477                                 msg.u.channel.session_id, msg.u.channel.pathname,
 478                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 479                                 msg.u.channel.relayd_id, msg.u.channel.output,
 480                                 msg.u.channel.tracefile_size,
 481                                 msg.u.channel.tracefile_count, 0,
 482                                 msg.u.channel.monitor,
 483                                 msg.u.channel.live_timer_interval,
 484                                 NULL, NULL);
 485                 if (new_channel == NULL) {
 486                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 487                         goto end_nosignal;
 488                 }
 489                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 490                 switch (msg.u.channel.output) {
 491                 case LTTNG_EVENT_SPLICE:
 492                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 493                         break;
 494                 case LTTNG_EVENT_MMAP:
 495                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 496                         break;
 497                 default:
 498                         ERR("Channel output unknown %d", msg.u.channel.output);
 499                         goto end_nosignal;
 500                 }
 501
 502                 /* Translate and save channel type. */
 503                 switch (msg.u.channel.type) {
 504                 case CONSUMER_CHANNEL_TYPE_DATA:
 505                 case CONSUMER_CHANNEL_TYPE_METADATA:
 506                         new_channel->type = msg.u.channel.type;
 507                         break;
 508                 default:
 509                         assert(0);
 510                         goto end_nosignal;
 511                 };
 512
 513                 health_code_update();
 514
 515                 if (ctx->on_recv_channel != NULL) {
 516                         ret_recv = ctx->on_recv_channel(new_channel);
 517                         if (ret_recv == 0) {
 518                                 ret = consumer_add_channel(new_channel, ctx);
 519                         } else if (ret_recv < 0) {
 520                                 goto end_nosignal;
 521                         }
 522                 } else {
 523                         ret = consumer_add_channel(new_channel, ctx);
 524                 }
 525                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 526                         consumer_timer_live_start(new_channel,
 527                                         msg.u.channel.live_timer_interval);
 528                 }
 529
 530                 health_code_update();
 531
 532                 /* If we received an error in add_channel, we need to report it. */
 533                 if (ret < 0) {
 534                         ret = consumer_send_status_msg(sock, ret);
 535                         if (ret < 0) {
 536                                 goto error_fatal;
 537                         }
 538                         goto end_nosignal;
 539                 }
 540
 541                 goto end_nosignal;
 542         }
 543         case LTTNG_CONSUMER_ADD_STREAM:
 544         {
 545                 int fd;
 546                 struct lttng_pipe *stream_pipe;
 547                 struct lttng_consumer_stream *new_stream;
 548                 struct lttng_consumer_channel *channel;
 549                 int alloc_ret = 0;
 550
 551                 /*
 552                  * Get stream's channel reference. Needed when adding the stream to the
 553                  * global hash table.
 554                  */
 555                 channel = consumer_find_channel(msg.u.stream.channel_key);
 556                 if (!channel) {
 557                         /*
 558                          * We could not find the channel. Can happen if cpu hotplug
 559                          * happens while tearing down.
 560                          */
 561                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 562                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 563                 }
 564
 565                 health_code_update();
 566
 567                 /* First send a status message before receiving the fds. */
 568                 ret = consumer_send_status_msg(sock, ret_code);
 569                 if (ret < 0) {
 570                         /* Somehow, the session daemon is not responding anymore. */
 571                         goto error_fatal;
 572                 }
 573
 574                 health_code_update();
 575
 576                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 577                         /* Channel was not found. */
 578                         goto end_nosignal;
 579                 }
 580
 581                 /* Blocking call */
 582                 health_poll_entry();
 583                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 584                 health_poll_exit();
 585                 if (ret) {
 586                         goto error_fatal;
 587                 }
 588
 589                 health_code_update();
 590
 591                 /* Get stream file descriptor from socket */
 592                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 593                 if (ret != sizeof(fd)) {
 594                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 595                         rcu_read_unlock();
 596                         return ret;
 597                 }
 598
 599                 health_code_update();
 600
 601                 /*
 602                  * Send status code to session daemon only if the recv works. If the
 603                  * above recv() failed, the session daemon is notified through the
 604                  * error socket and the teardown is eventually done.
 605                  */
 606                 ret = consumer_send_status_msg(sock, ret_code);
 607                 if (ret < 0) {
 608                         /* Somehow, the session daemon is not responding anymore. */
 609                         goto end_nosignal;
 610                 }
 611
 612                 health_code_update();
 613
 614                 new_stream = consumer_allocate_stream(channel->key,
 615                                 fd,
 616                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 617                                 channel->name,
 618                                 channel->uid,
 619                                 channel->gid,
 620                                 channel->relayd_id,
 621                                 channel->session_id,
 622                                 msg.u.stream.cpu,
 623                                 &alloc_ret,
 624                                 channel->type,
 625                                 channel->monitor);
 626                 if (new_stream == NULL) {
 627                         switch (alloc_ret) {
 628                         case -ENOMEM:
 629                         case -EINVAL:
 630                         default:
 631                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 632                                 break;
 633                         }
 634                         goto end_nosignal;
 635                 }
 636
 637                 new_stream->chan = channel;
 638                 new_stream->wait_fd = fd;
 639                 switch (channel->output) {
 640                 case CONSUMER_CHANNEL_SPLICE:
 641                         new_stream->output = LTTNG_EVENT_SPLICE;
 642                         ret = utils_create_pipe(new_stream->splice_pipe);
 643                         if (ret < 0) {
 644                                 goto end_nosignal;
 645                         }
 646                         break;
 647                 case CONSUMER_CHANNEL_MMAP:
 648                         new_stream->output = LTTNG_EVENT_MMAP;
 649                         break;
 650                 default:
 651                         ERR("Stream output unknown %d", channel->output);
 652                         goto end_nosignal;
 653                 }
 654
 655                 /*
 656                  * We've just assigned the channel to the stream so increment the
 657                  * refcount right now. We don't need to increment the refcount for
 658                  * streams in no monitor because we handle manually the cleanup of
 659                  * those. It is very important to make sure there is NO prior
 660                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 661                  */
 662                 if (channel->monitor) {
 663                         uatomic_inc(&new_stream->chan->refcount);
 664                 }
 665
 666                 /*
 667                  * The buffer flush is done on the session daemon side for the kernel
 668                  * so no need for the stream "hangup_flush_done" variable to be
 669                  * tracked. This is important for a kernel stream since we don't rely
 670                  * on the flush state of the stream to read data. It's not the case for
 671                  * user space tracing.
 672                  */
 673                 new_stream->hangup_flush_done = 0;
 674
 675                 health_code_update();
 676
 677                 if (ctx->on_recv_stream) {
 678                         ret = ctx->on_recv_stream(new_stream);
 679                         if (ret < 0) {
 680                                 consumer_stream_free(new_stream);
 681                                 goto end_nosignal;
 682                         }
 683                 }
 684
 685                 health_code_update();
 686
 687                 if (new_stream->metadata_flag) {
 688                         channel->metadata_stream = new_stream;
 689                 }
 690
 691                 /* Do not monitor this stream. */
 692                 if (!channel->monitor) {
 693                         DBG("Kernel consumer add stream %s in no monitor mode with "
 694                                         "relayd id %" PRIu64, new_stream->name,
 695                                         new_stream->net_seq_idx);
 696                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 697                         break;
 698                 }
 699
 700                 /* Send stream to relayd if the stream has an ID. */
 701                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 702                         ret = consumer_send_relayd_stream(new_stream,
 703                                         new_stream->chan->pathname);
 704                         if (ret < 0) {
 705                                 consumer_stream_free(new_stream);
 706                                 goto end_nosignal;
 707                         }
 708                 }
 709
 710                 /* Get the right pipe where the stream will be sent. */
 711                 if (new_stream->metadata_flag) {
 712                         ret = consumer_add_metadata_stream(new_stream);
 713                         if (ret) {
 714                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 715                                                 new_stream->key);
 716                                 consumer_stream_free(new_stream);
 717                                 goto end_nosignal;
 718                         }
 719                         stream_pipe = ctx->consumer_metadata_pipe;
 720                 } else {
 721                         ret = consumer_add_data_stream(new_stream);
 722                         if (ret) {
 723                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 724                                                 new_stream->key);
 725                                 consumer_stream_free(new_stream);
 726                                 goto end_nosignal;
 727                         }
 728                         stream_pipe = ctx->consumer_data_pipe;
 729                 }
 730
 731                 /* Vitible to other threads */
 732                 new_stream->globally_visible = 1;
 733
 734                 health_code_update();
 735
 736                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 737                 if (ret < 0) {
 738                         ERR("Consumer write %s stream to pipe %d",
 739                                         new_stream->metadata_flag ? "metadata" : "data",
 740                                         lttng_pipe_get_writefd(stream_pipe));
 741                         if (new_stream->metadata_flag) {
 742                                 consumer_del_stream_for_metadata(new_stream);
 743                         } else {
 744                                 consumer_del_stream_for_data(new_stream);
 745                         }
 746                         goto end_nosignal;
 747                 }
 748
 749                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 750                                 new_stream->name, fd, new_stream->relayd_stream_id);
 751                 break;
 752         }
 753         case LTTNG_CONSUMER_STREAMS_SENT:
 754         {
 755                 struct lttng_consumer_channel *channel;
 756
 757                 /*
 758                  * Get stream's channel reference. Needed when adding the stream to the
 759                  * global hash table.
 760                  */
 761                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 762                 if (!channel) {
 763                         /*
 764                          * We could not find the channel. Can happen if cpu hotplug
 765                          * happens while tearing down.
 766                          */
 767                         ERR("Unable to find channel key %" PRIu64,
 768                                         msg.u.sent_streams.channel_key);
 769                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 770                 }
 771
 772                 health_code_update();
 773
 774                 /*
 775                  * Send status code to session daemon.
 776                  */
 777                 ret = consumer_send_status_msg(sock, ret_code);
 778                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 779                         /* Somehow, the session daemon is not responding anymore. */
 780                         goto end_nosignal;
 781                 }
 782
 783                 health_code_update();
 784
 785                 /*
 786                  * We should not send this message if we don't monitor the
 787                  * streams in this channel.
 788                  */
 789                 if (!channel->monitor) {
 790                         break;
 791                 }
 792
 793                 health_code_update();
 794                 /* Send stream to relayd if the stream has an ID. */
 795                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 796                         ret = consumer_send_relayd_streams_sent(
 797                                         msg.u.sent_streams.net_seq_idx);
 798                         if (ret < 0) {
 799                                 goto end_nosignal;
 800                         }
 801                 }
 802                 break;
 803         }
 804         case LTTNG_CONSUMER_UPDATE_STREAM:
 805         {
 806                 rcu_read_unlock();
 807                 return -ENOSYS;
 808         }
 809         case LTTNG_CONSUMER_DESTROY_RELAYD:
 810         {
 811                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 812                 struct consumer_relayd_sock_pair *relayd;
 813
 814                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 815
 816                 /* Get relayd reference if exists. */
 817                 relayd = consumer_find_relayd(index);
 818                 if (relayd == NULL) {
 819                         DBG("Unable to find relayd %" PRIu64, index);
 820                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 821                 }
 822
 823                 /*
 824                  * Each relayd socket pair has a refcount of stream attached to it
 825                  * which tells if the relayd is still active or not depending on the
 826                  * refcount value.
 827                  *
 828                  * This will set the destroy flag of the relayd object and destroy it
 829                  * if the refcount reaches zero when called.
 830                  *
 831                  * The destroy can happen either here or when a stream fd hangs up.
 832                  */
 833                 if (relayd) {
 834                         consumer_flag_relayd_for_destroy(relayd);
 835                 }
 836
 837                 health_code_update();
 838
 839                 ret = consumer_send_status_msg(sock, ret_code);
 840                 if (ret < 0) {
 841                         /* Somehow, the session daemon is not responding anymore. */
 842                         goto error_fatal;
 843                 }
 844
 845                 goto end_nosignal;
 846         }
 847         case LTTNG_CONSUMER_DATA_PENDING:
 848         {
 849                 int32_t ret;
 850                 uint64_t id = msg.u.data_pending.session_id;
 851
 852                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 853
 854                 ret = consumer_data_pending(id);
 855
 856                 health_code_update();
 857
 858                 /* Send back returned value to session daemon */
 859                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 860                 if (ret < 0) {
 861                         PERROR("send data pending ret code");
 862                         goto error_fatal;
 863                 }
 864
 865                 /*
 866                  * No need to send back a status message since the data pending
 867                  * returned value is the response.
 868                  */
 869                 break;
 870         }
 871         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 872         {
 873                 if (msg.u.snapshot_channel.metadata == 1) {
 874                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 875                                         msg.u.snapshot_channel.pathname,
 876                                         msg.u.snapshot_channel.relayd_id, ctx);
 877                         if (ret < 0) {
 878                                 ERR("Snapshot metadata failed");
 879                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 880                         }
 881                 } else {
 882                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 883                                         msg.u.snapshot_channel.pathname,
 884                                         msg.u.snapshot_channel.relayd_id,
 885                                         msg.u.snapshot_channel.nb_packets_per_stream,
 886                                         ctx);
 887                         if (ret < 0) {
 888                                 ERR("Snapshot channel failed");
 889                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 890                         }
 891                 }
 892
 893                 health_code_update();
 894
 895                 ret = consumer_send_status_msg(sock, ret_code);
 896                 if (ret < 0) {
 897                         /* Somehow, the session daemon is not responding anymore. */
 898                         goto end_nosignal;
 899                 }
 900                 break;
 901         }
 902         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 903         {
 904                 uint64_t key = msg.u.destroy_channel.key;
 905                 struct lttng_consumer_channel *channel;
 906
 907                 channel = consumer_find_channel(key);
 908                 if (!channel) {
 909                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 910                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 911                 }
 912
 913                 health_code_update();
 914
 915                 ret = consumer_send_status_msg(sock, ret_code);
 916                 if (ret < 0) {
 917                         /* Somehow, the session daemon is not responding anymore. */
 918                         goto end_nosignal;
 919                 }
 920
 921                 health_code_update();
 922
 923                 /* Stop right now if no channel was found. */
 924                 if (!channel) {
 925                         goto end_nosignal;
 926                 }
 927
 928                 /*
 929                  * This command should ONLY be issued for channel with streams set in
 930                  * no monitor mode.
 931                  */
 932                 assert(!channel->monitor);
 933
 934                 /*
 935                  * The refcount should ALWAYS be 0 in the case of a channel in no
 936                  * monitor mode.
 937                  */
 938                 assert(!uatomic_sub_return(&channel->refcount, 1));
 939
 940                 consumer_del_channel(channel);
 941
 942                 goto end_nosignal;
 943         }
 944         default:
 945                 goto end_nosignal;
 946         }
 947
 948 end_nosignal:
 949         rcu_read_unlock();
 950
 951         /*
 952          * Return 1 to indicate success since the 0 value can be a socket
 953          * shutdown during the recv() or send() call.
 954          */
 955         health_code_update();
 956         return 1;
 957
 958 error_fatal:
 959         rcu_read_unlock();
 960         /* This will issue a consumer stop. */
 961         return -1;
 962 }
 963
 964 /*
 965  * Populate index values of a kernel stream. Values are set in big endian order.
 966  *
 967  * Return 0 on success or else a negative value.
 968  */
 969 static int get_index_values(struct ctf_packet_index *index, int infd)
 970 {
 971         int ret;
 972
 973         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 974         if (ret < 0) {
 975                 PERROR("kernctl_get_timestamp_begin");
 976                 goto error;
 977         }
 978         index->timestamp_begin = htobe64(index->timestamp_begin);
 979
 980         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 981         if (ret < 0) {
 982                 PERROR("kernctl_get_timestamp_end");
 983                 goto error;
 984         }
 985         index->timestamp_end = htobe64(index->timestamp_end);
 986
 987         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 988         if (ret < 0) {
 989                 PERROR("kernctl_get_events_discarded");
 990                 goto error;
 991         }
 992         index->events_discarded = htobe64(index->events_discarded);
 993
 994         ret = kernctl_get_content_size(infd, &index->content_size);
 995         if (ret < 0) {
 996                 PERROR("kernctl_get_content_size");
 997                 goto error;
 998         }
 999         index->content_size = htobe64(index->content_size);
1000
1001         ret = kernctl_get_packet_size(infd, &index->packet_size);
1002         if (ret < 0) {
1003                 PERROR("kernctl_get_packet_size");
1004                 goto error;
1005         }
1006         index->packet_size = htobe64(index->packet_size);
1007
1008         ret = kernctl_get_stream_id(infd, &index->stream_id);
1009         if (ret < 0) {
1010                 PERROR("kernctl_get_stream_id");
1011                 goto error;
1012         }
1013         index->stream_id = htobe64(index->stream_id);
1014
1015 error:
1016         return ret;
1017 }
1018 /*
1019  * Sync metadata meaning request them to the session daemon and snapshot to the
1020  * metadata thread can consumer them.
1021  *
1022  * Metadata stream lock MUST be acquired.
1023  *
1024  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1025  * is empty or a negative value on error.
1026  */
1027 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1028 {
1029         int ret;
1030
1031         assert(metadata);
1032
1033         ret = kernctl_buffer_flush(metadata->wait_fd);
1034         if (ret < 0) {
1035                 ERR("Failed to flush kernel stream");
1036                 goto end;
1037         }
1038
1039         ret = kernctl_snapshot(metadata->wait_fd);
1040         if (ret < 0) {
1041                 if (errno != EAGAIN) {
1042                         ERR("Sync metadata, taking kernel snapshot failed.");
1043                         goto end;
1044                 }
1045                 DBG("Sync metadata, no new kernel metadata");
1046                 /* No new metadata, exit. */
1047                 ret = ENODATA;
1048                 goto end;
1049         }
1050
1051 end:
1052         return ret;
1053 }
1054
1055 /*
1056  * Consume data on a file descriptor and write it on a trace file.
1057  */
1058 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1059                 struct lttng_consumer_local_data *ctx)
1060 {
1061         unsigned long len, subbuf_size, padding;
1062         int err, write_index = 1;
1063         ssize_t ret = 0;
1064         int infd = stream->wait_fd;
1065         struct ctf_packet_index index;
1066
1067         DBG("In read_subbuffer (infd : %d)", infd);
1068
1069         /* Get the next subbuffer */
1070         err = kernctl_get_next_subbuf(infd);
1071         if (err != 0) {
1072                 /*
1073                  * This is a debug message even for single-threaded consumer,
1074                  * because poll() have more relaxed criterions than get subbuf,
1075                  * so get_subbuf may fail for short race windows where poll()
1076                  * would issue wakeups.
1077                  */
1078                 DBG("Reserving sub buffer failed (everything is normal, "
1079                                 "it is due to concurrency)");
1080                 ret = -errno;
1081                 goto end;
1082         }
1083
1084         /* Get the full subbuffer size including padding */
1085         err = kernctl_get_padded_subbuf_size(infd, &len);
1086         if (err != 0) {
1087                 PERROR("Getting sub-buffer len failed.");
1088                 err = kernctl_put_subbuf(infd);
1089                 if (err != 0) {
1090                         if (errno == EFAULT) {
1091                                 PERROR("Error in unreserving sub buffer\n");
1092                         } else if (errno == EIO) {
1093                                 /* Should never happen with newer LTTng versions */
1094                                 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1095                         }
1096                         ret = -errno;
1097                         goto end;
1098                 }
1099                 ret = -errno;
1100                 goto end;
1101         }
1102
1103         if (!stream->metadata_flag) {
1104                 ret = get_index_values(&index, infd);
1105                 if (ret < 0) {
1106                         err = kernctl_put_subbuf(infd);
1107                         if (err != 0) {
1108                                 if (errno == EFAULT) {
1109                                         PERROR("Error in unreserving sub buffer\n");
1110                                 } else if (errno == EIO) {
1111                                         /* Should never happen with newer LTTng versions */
1112                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1113                                 }
1114                                 ret = -errno;
1115                                 goto end;
1116                         }
1117                         goto end;
1118                 }
1119         } else {
1120                 write_index = 0;
1121         }
1122
1123         switch (stream->chan->output) {
1124         case CONSUMER_CHANNEL_SPLICE:
1125                 /*
1126                  * XXX: The lttng-modules splice "actor" does not handle copying
1127                  * partial pages hence only using the subbuffer size without the
1128                  * padding makes the splice fail.
1129                  */
1130                 subbuf_size = len;
1131                 padding = 0;
1132
1133                 /* splice the subbuffer to the tracefile */
1134                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1135                                 padding, &index);
1136                 /*
1137                  * XXX: Splice does not support network streaming so the return value
1138                  * is simply checked against subbuf_size and not like the mmap() op.
1139                  */
1140                 if (ret != subbuf_size) {
1141                         /*
1142                          * display the error but continue processing to try
1143                          * to release the subbuffer
1144                          */
1145                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1146                                         ret, subbuf_size);
1147                         write_index = 0;
1148                 }
1149                 break;
1150         case CONSUMER_CHANNEL_MMAP:
1151                 /* Get subbuffer size without padding */
1152                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1153                 if (err != 0) {
1154                         PERROR("Getting sub-buffer len failed.");
1155                         err = kernctl_put_subbuf(infd);
1156                         if (err != 0) {
1157                                 if (errno == EFAULT) {
1158                                         PERROR("Error in unreserving sub buffer\n");
1159                                 } else if (errno == EIO) {
1160                                         /* Should never happen with newer LTTng versions */
1161                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1162                                 }
1163                                 ret = -errno;
1164                                 goto end;
1165                         }
1166                         ret = -errno;
1167                         goto end;
1168                 }
1169
1170                 /* Make sure the tracer is not gone mad on us! */
1171                 assert(len >= subbuf_size);
1172
1173                 padding = len - subbuf_size;
1174
1175                 /* write the subbuffer to the tracefile */
1176                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1177                                 padding, &index);
1178                 /*
1179                  * The mmap operation should write subbuf_size amount of data when
1180                  * network streaming or the full padding (len) size when we are _not_
1181                  * streaming.
1182                  */
1183                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1184                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1185                         /*
1186                          * Display the error but continue processing to try to release the
1187                          * subbuffer. This is a DBG statement since this is possible to
1188                          * happen without being a critical error.
1189                          */
1190                         DBG("Error writing to tracefile "
1191                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1192                                         ret, len, subbuf_size);
1193                         write_index = 0;
1194                 }
1195                 break;
1196         default:
1197                 ERR("Unknown output method");
1198                 ret = -EPERM;
1199         }
1200
1201         err = kernctl_put_next_subbuf(infd);
1202         if (err != 0) {
1203                 if (errno == EFAULT) {
1204                         PERROR("Error in unreserving sub buffer\n");
1205                 } else if (errno == EIO) {
1206                         /* Should never happen with newer LTTng versions */
1207                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1208                 }
1209                 ret = -errno;
1210                 goto end;
1211         }
1212
1213         /* Write index if needed. */
1214         if (!write_index) {
1215                 goto end;
1216         }
1217
1218         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1219                 /*
1220                  * In live, block until all the metadata is sent.
1221                  */
1222                 pthread_mutex_lock(&stream->metadata_timer_lock);
1223                 assert(!stream->missed_metadata_flush);
1224                 stream->waiting_on_metadata = true;
1225                 pthread_mutex_unlock(&stream->metadata_timer_lock);
1226
1227                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1228
1229                 pthread_mutex_lock(&stream->metadata_timer_lock);
1230                 stream->waiting_on_metadata = false;
1231                 if (stream->missed_metadata_flush) {
1232                         stream->missed_metadata_flush = false;
1233                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1234                         (void) consumer_flush_kernel_index(stream);
1235                 } else {
1236                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1237                 }
1238                 if (err < 0) {
1239                         goto end;
1240                 }
1241         }
1242
1243         err = consumer_stream_write_index(stream, &index);
1244         if (err < 0) {
1245                 goto end;
1246         }
1247
1248 end:
1249         return ret;
1250 }
1251
1252 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1253 {
1254         int ret;
1255
1256         assert(stream);
1257
1258         /*
1259          * Don't create anything if this is set for streaming or should not be
1260          * monitored.
1261          */
1262         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1263                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1264                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1265                                 stream->uid, stream->gid, NULL);
1266                 if (ret < 0) {
1267                         goto error;
1268                 }
1269                 stream->out_fd = ret;
1270                 stream->tracefile_size_current = 0;
1271
1272                 if (!stream->metadata_flag) {
1273                         ret = index_create_file(stream->chan->pathname,
1274                                         stream->name, stream->uid, stream->gid,
1275                                         stream->chan->tracefile_size,
1276                                         stream->tracefile_count_current);
1277                         if (ret < 0) {
1278                                 goto error;
1279                         }
1280                         stream->index_fd = ret;
1281                 }
1282         }
1283
1284         if (stream->output == LTTNG_EVENT_MMAP) {
1285                 /* get the len of the mmap region */
1286                 unsigned long mmap_len;
1287
1288                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1289                 if (ret != 0) {
1290                         PERROR("kernctl_get_mmap_len");
1291                         ret = -errno;
1292                         goto error_close_fd;
1293                 }
1294                 stream->mmap_len = (size_t) mmap_len;
1295
1296                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1297                                 MAP_PRIVATE, stream->wait_fd, 0);
1298                 if (stream->mmap_base == MAP_FAILED) {
1299                         PERROR("Error mmaping");
1300                         ret = -1;
1301                         goto error_close_fd;
1302                 }
1303         }
1304
1305         /* we return 0 to let the library handle the FD internally */
1306         return 0;
1307
1308 error_close_fd:
1309         if (stream->out_fd >= 0) {
1310                 int err;
1311
1312                 err = close(stream->out_fd);
1313                 assert(!err);
1314                 stream->out_fd = -1;
1315         }
1316 error:
1317         return ret;
1318 }
1319
1320 /*
1321  * Check if data is still being extracted from the buffers for a specific
1322  * stream. Consumer data lock MUST be acquired before calling this function
1323  * and the stream lock.
1324  *
1325  * Return 1 if the traced data are still getting read else 0 meaning that the
1326  * data is available for trace viewer reading.
1327  */
1328 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1329 {
1330         int ret;
1331
1332         assert(stream);
1333
1334         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1335                 ret = 0;
1336                 goto end;
1337         }
1338
1339         ret = kernctl_get_next_subbuf(stream->wait_fd);
1340         if (ret == 0) {
1341                 /* There is still data so let's put back this subbuffer. */
1342                 ret = kernctl_put_subbuf(stream->wait_fd);
1343                 assert(ret == 0);
1344                 ret = 1;   /* Data is pending */
1345                 goto end;
1346         }
1347
1348         /* Data is NOT pending and ready to be read. */
1349         ret = 0;
1350
1351 end:
1352         return ret;
1353 }