src/common/kernel-consumer/kernel-consumer.c

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _GNU_SOURCE
  20 #include <assert.h>
  21 #include <poll.h>
  22 #include <pthread.h>
  23 #include <stdlib.h>
  24 #include <string.h>
  25 #include <sys/mman.h>
  26 #include <sys/socket.h>
  27 #include <sys/types.h>
  28 #include <inttypes.h>
  29 #include <unistd.h>
  30 #include <sys/stat.h>
  31
  32 #include <bin/lttng-consumerd/health-consumerd.h>
  33 #include <common/common.h>
  34 #include <common/kernel-ctl/kernel-ctl.h>
  35 #include <common/sessiond-comm/sessiond-comm.h>
  36 #include <common/sessiond-comm/relayd.h>
  37 #include <common/compat/fcntl.h>
  38 #include <common/pipe.h>
  39 #include <common/relayd/relayd.h>
  40 #include <common/utils.h>
  41 #include <common/consumer-stream.h>
  42 #include <common/index/index.h>
  43 #include <common/consumer-timer.h>
  44
  45 #include "kernel-consumer.h"
  46
  47 extern struct lttng_consumer_global_data consumer_data;
  48 extern int consumer_poll_timeout;
  49 extern volatile int consumer_quit;
  50
  51 /*
  52  * Take a snapshot for a specific fd
  53  *
  54  * Returns 0 on success, < 0 on error
  55  */
  56 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  57 {
  58         int ret = 0;
  59         int infd = stream->wait_fd;
  60
  61         ret = kernctl_snapshot(infd);
  62         if (ret != 0) {
  63                 perror("Getting sub-buffer snapshot.");
  64                 ret = -errno;
  65         }
  66
  67         return ret;
  68 }
  69
  70 /*
  71  * Get the produced position
  72  *
  73  * Returns 0 on success, < 0 on error
  74  */
  75 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  76                 unsigned long *pos)
  77 {
  78         int ret;
  79         int infd = stream->wait_fd;
  80
  81         ret = kernctl_snapshot_get_produced(infd, pos);
  82         if (ret != 0) {
  83                 perror("kernctl_snapshot_get_produced");
  84                 ret = -errno;
  85         }
  86
  87         return ret;
  88 }
  89
  90 /*
  91  * Get the consumerd position
  92  *
  93  * Returns 0 on success, < 0 on error
  94  */
  95 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  96                 unsigned long *pos)
  97 {
  98         int ret;
  99         int infd = stream->wait_fd;
 100
 101         ret = kernctl_snapshot_get_consumed(infd, pos);
 102         if (ret != 0) {
 103                 perror("kernctl_snapshot_get_consumed");
 104                 ret = -errno;
 105         }
 106
 107         return ret;
 108 }
 109
 110 /*
 111  * Take a snapshot of all the stream of a channel
 112  *
 113  * Returns 0 on success, < 0 on error
 114  */
 115 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 116                 uint64_t relayd_id, uint64_t max_stream_size,
 117                 struct lttng_consumer_local_data *ctx)
 118 {
 119         int ret;
 120         unsigned long consumed_pos, produced_pos;
 121         struct lttng_consumer_channel *channel;
 122         struct lttng_consumer_stream *stream;
 123
 124         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 125
 126         rcu_read_lock();
 127
 128         channel = consumer_find_channel(key);
 129         if (!channel) {
 130                 ERR("No channel found for key %" PRIu64, key);
 131                 ret = -1;
 132                 goto end;
 133         }
 134
 135         /* Splice is not supported yet for channel snapshot. */
 136         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 137                 ERR("Unsupported output %d", channel->output);
 138                 ret = -1;
 139                 goto end;
 140         }
 141
 142         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 143
 144                 health_code_update();
 145
 146                 /*
 147                  * Lock stream because we are about to change its state.
 148                  */
 149                 pthread_mutex_lock(&stream->lock);
 150
 151                 /*
 152                  * Assign the received relayd ID so we can use it for streaming. The streams
 153                  * are not visible to anyone so this is OK to change it.
 154                  */
 155                 stream->net_seq_idx = relayd_id;
 156                 channel->relayd_id = relayd_id;
 157                 if (relayd_id != (uint64_t) -1ULL) {
 158                         ret = consumer_send_relayd_stream(stream, path);
 159                         if (ret < 0) {
 160                                 ERR("sending stream to relayd");
 161                                 goto end_unlock;
 162                         }
 163                 } else {
 164                         ret = utils_create_stream_file(path, stream->name,
 165                                         stream->chan->tracefile_size,
 166                                         stream->tracefile_count_current,
 167                                         stream->uid, stream->gid, NULL);
 168                         if (ret < 0) {
 169                                 ERR("utils_create_stream_file");
 170                                 goto end_unlock;
 171                         }
 172
 173                         stream->out_fd = ret;
 174                         stream->tracefile_size_current = 0;
 175
 176                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 177                                         path, stream->name, stream->key);
 178                 }
 179                 if (relayd_id != -1ULL) {
 180                         ret = consumer_send_relayd_streams_sent(relayd_id);
 181                         if (ret < 0) {
 182                                 ERR("sending streams sent to relayd");
 183                                 goto end_unlock;
 184                         }
 185                 }
 186
 187                 ret = kernctl_buffer_flush(stream->wait_fd);
 188                 if (ret < 0) {
 189                         ERR("Failed to flush kernel stream");
 190                         ret = -errno;
 191                         goto end_unlock;
 192                 }
 193
 194                 ret = lttng_kconsumer_take_snapshot(stream);
 195                 if (ret < 0) {
 196                         ERR("Taking kernel snapshot");
 197                         goto end_unlock;
 198                 }
 199
 200                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 201                 if (ret < 0) {
 202                         ERR("Produced kernel snapshot position");
 203                         goto end_unlock;
 204                 }
 205
 206                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 207                 if (ret < 0) {
 208                         ERR("Consumerd kernel snapshot position");
 209                         goto end_unlock;
 210                 }
 211
 212                 if (stream->max_sb_size == 0) {
 213                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 214                                         &stream->max_sb_size);
 215                         if (ret < 0) {
 216                                 ERR("Getting kernel max_sb_size");
 217                                 ret = -errno;
 218                                 goto end_unlock;
 219                         }
 220                 }
 221
 222                 /*
 223                  * The original value is sent back if max stream size is larger than
 224                  * the possible size of the snapshot. Also, we asume that the session
 225                  * daemon should never send a maximum stream size that is lower than
 226                  * subbuffer size.
 227                  */
 228                 consumed_pos = consumer_get_consumed_maxsize(consumed_pos,
 229                                 produced_pos, max_stream_size);
 230
 231                 while (consumed_pos < produced_pos) {
 232                         ssize_t read_len;
 233                         unsigned long len, padded_len;
 234
 235                         health_code_update();
 236
 237                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 238
 239                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 240                         if (ret < 0) {
 241                                 if (errno != EAGAIN) {
 242                                         PERROR("kernctl_get_subbuf snapshot");
 243                                         ret = -errno;
 244                                         goto end_unlock;
 245                                 }
 246                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 247                                 consumed_pos += stream->max_sb_size;
 248                                 continue;
 249                         }
 250
 251                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 252                         if (ret < 0) {
 253                                 ERR("Snapshot kernctl_get_subbuf_size");
 254                                 ret = -errno;
 255                                 goto error_put_subbuf;
 256                         }
 257
 258                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 259                         if (ret < 0) {
 260                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 261                                 ret = -errno;
 262                                 goto error_put_subbuf;
 263                         }
 264
 265                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 266                                         padded_len - len, NULL);
 267                         /*
 268                          * We write the padded len in local tracefiles but the data len
 269                          * when using a relay. Display the error but continue processing
 270                          * to try to release the subbuffer.
 271                          */
 272                         if (relayd_id != (uint64_t) -1ULL) {
 273                                 if (read_len != len) {
 274                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 275                                                         read_len, len);
 276                                 }
 277                         } else {
 278                                 if (read_len != padded_len) {
 279                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 280                                                         read_len, padded_len);
 281                                 }
 282                         }
 283
 284                         ret = kernctl_put_subbuf(stream->wait_fd);
 285                         if (ret < 0) {
 286                                 ERR("Snapshot kernctl_put_subbuf");
 287                                 ret = -errno;
 288                                 goto end_unlock;
 289                         }
 290                         consumed_pos += stream->max_sb_size;
 291                 }
 292
 293                 if (relayd_id == (uint64_t) -1ULL) {
 294                         if (stream->out_fd >= 0) {
 295                                 ret = close(stream->out_fd);
 296                                 if (ret < 0) {
 297                                         PERROR("Kernel consumer snapshot close out_fd");
 298                                         goto end_unlock;
 299                                 }
 300                                 stream->out_fd = -1;
 301                         }
 302                 } else {
 303                         close_relayd_stream(stream);
 304                         stream->net_seq_idx = (uint64_t) -1ULL;
 305                 }
 306                 pthread_mutex_unlock(&stream->lock);
 307         }
 308
 309         /* All good! */
 310         ret = 0;
 311         goto end;
 312
 313 error_put_subbuf:
 314         ret = kernctl_put_subbuf(stream->wait_fd);
 315         if (ret < 0) {
 316                 ret = -errno;
 317                 ERR("Snapshot kernctl_put_subbuf error path");
 318         }
 319 end_unlock:
 320         pthread_mutex_unlock(&stream->lock);
 321 end:
 322         rcu_read_unlock();
 323         return ret;
 324 }
 325
 326 /*
 327  * Read the whole metadata available for a snapshot.
 328  *
 329  * Returns 0 on success, < 0 on error
 330  */
 331 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 332                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 333 {
 334         int ret, use_relayd = 0;
 335         ssize_t ret_read;
 336         struct lttng_consumer_channel *metadata_channel;
 337         struct lttng_consumer_stream *metadata_stream;
 338
 339         assert(ctx);
 340
 341         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 342                         key, path);
 343
 344         rcu_read_lock();
 345
 346         metadata_channel = consumer_find_channel(key);
 347         if (!metadata_channel) {
 348                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 349                 ret = -1;
 350                 goto error;
 351         }
 352
 353         metadata_stream = metadata_channel->metadata_stream;
 354         assert(metadata_stream);
 355
 356         /* Flag once that we have a valid relayd for the stream. */
 357         if (relayd_id != (uint64_t) -1ULL) {
 358                 use_relayd = 1;
 359         }
 360
 361         if (use_relayd) {
 362                 ret = consumer_send_relayd_stream(metadata_stream, path);
 363                 if (ret < 0) {
 364                         goto error;
 365                 }
 366         } else {
 367                 ret = utils_create_stream_file(path, metadata_stream->name,
 368                                 metadata_stream->chan->tracefile_size,
 369                                 metadata_stream->tracefile_count_current,
 370                                 metadata_stream->uid, metadata_stream->gid, NULL);
 371                 if (ret < 0) {
 372                         goto error;
 373                 }
 374                 metadata_stream->out_fd = ret;
 375         }
 376
 377         do {
 378                 health_code_update();
 379
 380                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 381                 if (ret_read < 0) {
 382                         if (ret_read != -EAGAIN) {
 383                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 384                                                 ret_read);
 385                                 goto error;
 386                         }
 387                         /* ret_read is negative at this point so we will exit the loop. */
 388                         continue;
 389                 }
 390         } while (ret_read >= 0);
 391
 392         if (use_relayd) {
 393                 close_relayd_stream(metadata_stream);
 394                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 395         } else {
 396                 if (metadata_stream->out_fd >= 0) {
 397                         ret = close(metadata_stream->out_fd);
 398                         if (ret < 0) {
 399                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 400                                 /*
 401                                  * Don't go on error here since the snapshot was successful at this
 402                                  * point but somehow the close failed.
 403                                  */
 404                         }
 405                         metadata_stream->out_fd = -1;
 406                 }
 407         }
 408
 409         ret = 0;
 410
 411         cds_list_del(&metadata_stream->send_node);
 412         consumer_stream_destroy(metadata_stream, NULL);
 413         metadata_channel->metadata_stream = NULL;
 414 error:
 415         rcu_read_unlock();
 416         return ret;
 417 }
 418
 419 /*
 420  * Receive command from session daemon and process it.
 421  *
 422  * Return 1 on success else a negative value or 0.
 423  */
 424 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 425                 int sock, struct pollfd *consumer_sockpoll)
 426 {
 427         ssize_t ret;
 428         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 429         struct lttcomm_consumer_msg msg;
 430
 431         health_code_update();
 432
 433         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 434         if (ret != sizeof(msg)) {
 435                 if (ret > 0) {
 436                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 437                         ret = -1;
 438                 }
 439                 return ret;
 440         }
 441
 442         health_code_update();
 443
 444         if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
 445                 /*
 446                  * Notify the session daemon that the command is completed.
 447                  *
 448                  * On transport layer error, the function call will print an error
 449                  * message so handling the returned code is a bit useless since we
 450                  * return an error code anyway.
 451                  */
 452                 (void) consumer_send_status_msg(sock, ret_code);
 453                 return -ENOENT;
 454         }
 455
 456         health_code_update();
 457
 458         /* relayd needs RCU read-side protection */
 459         rcu_read_lock();
 460
 461         switch (msg.cmd_type) {
 462         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 463         {
 464                 /* Session daemon status message are handled in the following call. */
 465                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 466                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 467                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 468                                  msg.u.relayd_sock.relayd_session_id);
 469                 goto end_nosignal;
 470         }
 471         case LTTNG_CONSUMER_ADD_CHANNEL:
 472         {
 473                 struct lttng_consumer_channel *new_channel;
 474                 int ret_recv;
 475
 476                 health_code_update();
 477
 478                 /* First send a status message before receiving the fds. */
 479                 ret = consumer_send_status_msg(sock, ret_code);
 480                 if (ret < 0) {
 481                         /* Somehow, the session daemon is not responding anymore. */
 482                         goto error_fatal;
 483                 }
 484
 485                 health_code_update();
 486
 487                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 488                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 489                                 msg.u.channel.session_id, msg.u.channel.pathname,
 490                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 491                                 msg.u.channel.relayd_id, msg.u.channel.output,
 492                                 msg.u.channel.tracefile_size,
 493                                 msg.u.channel.tracefile_count, 0,
 494                                 msg.u.channel.monitor,
 495                                 msg.u.channel.live_timer_interval);
 496                 if (new_channel == NULL) {
 497                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 498                         goto end_nosignal;
 499                 }
 500                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 501                 switch (msg.u.channel.output) {
 502                 case LTTNG_EVENT_SPLICE:
 503                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 504                         break;
 505                 case LTTNG_EVENT_MMAP:
 506                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 507                         break;
 508                 default:
 509                         ERR("Channel output unknown %d", msg.u.channel.output);
 510                         goto end_nosignal;
 511                 }
 512
 513                 /* Translate and save channel type. */
 514                 switch (msg.u.channel.type) {
 515                 case CONSUMER_CHANNEL_TYPE_DATA:
 516                 case CONSUMER_CHANNEL_TYPE_METADATA:
 517                         new_channel->type = msg.u.channel.type;
 518                         break;
 519                 default:
 520                         assert(0);
 521                         goto end_nosignal;
 522                 };
 523
 524                 health_code_update();
 525
 526                 if (ctx->on_recv_channel != NULL) {
 527                         ret_recv = ctx->on_recv_channel(new_channel);
 528                         if (ret_recv == 0) {
 529                                 ret = consumer_add_channel(new_channel, ctx);
 530                         } else if (ret_recv < 0) {
 531                                 goto end_nosignal;
 532                         }
 533                 } else {
 534                         ret = consumer_add_channel(new_channel, ctx);
 535                 }
 536                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 537                         consumer_timer_live_start(new_channel,
 538                                         msg.u.channel.live_timer_interval);
 539                 }
 540
 541                 health_code_update();
 542
 543                 /* If we received an error in add_channel, we need to report it. */
 544                 if (ret < 0) {
 545                         ret = consumer_send_status_msg(sock, ret);
 546                         if (ret < 0) {
 547                                 goto error_fatal;
 548                         }
 549                         goto end_nosignal;
 550                 }
 551
 552                 goto end_nosignal;
 553         }
 554         case LTTNG_CONSUMER_ADD_STREAM:
 555         {
 556                 int fd;
 557                 struct lttng_pipe *stream_pipe;
 558                 struct lttng_consumer_stream *new_stream;
 559                 struct lttng_consumer_channel *channel;
 560                 int alloc_ret = 0;
 561
 562                 /*
 563                  * Get stream's channel reference. Needed when adding the stream to the
 564                  * global hash table.
 565                  */
 566                 channel = consumer_find_channel(msg.u.stream.channel_key);
 567                 if (!channel) {
 568                         /*
 569                          * We could not find the channel. Can happen if cpu hotplug
 570                          * happens while tearing down.
 571                          */
 572                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 573                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 574                 }
 575
 576                 health_code_update();
 577
 578                 /* First send a status message before receiving the fds. */
 579                 ret = consumer_send_status_msg(sock, ret_code);
 580                 if (ret < 0) {
 581                         /* Somehow, the session daemon is not responding anymore. */
 582                         goto error_fatal;
 583                 }
 584
 585                 health_code_update();
 586
 587                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 588                         /* Channel was not found. */
 589                         goto end_nosignal;
 590                 }
 591
 592                 /* Blocking call */
 593                 health_poll_entry();
 594                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 595                 health_poll_exit();
 596                 if (ret < 0) {
 597                         rcu_read_unlock();
 598                         return -EINTR;
 599                 }
 600
 601                 health_code_update();
 602
 603                 /* Get stream file descriptor from socket */
 604                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 605                 if (ret != sizeof(fd)) {
 606                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 607                         rcu_read_unlock();
 608                         return ret;
 609                 }
 610
 611                 health_code_update();
 612
 613                 /*
 614                  * Send status code to session daemon only if the recv works. If the
 615                  * above recv() failed, the session daemon is notified through the
 616                  * error socket and the teardown is eventually done.
 617                  */
 618                 ret = consumer_send_status_msg(sock, ret_code);
 619                 if (ret < 0) {
 620                         /* Somehow, the session daemon is not responding anymore. */
 621                         goto end_nosignal;
 622                 }
 623
 624                 health_code_update();
 625
 626                 new_stream = consumer_allocate_stream(channel->key,
 627                                 fd,
 628                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 629                                 channel->name,
 630                                 channel->uid,
 631                                 channel->gid,
 632                                 channel->relayd_id,
 633                                 channel->session_id,
 634                                 msg.u.stream.cpu,
 635                                 &alloc_ret,
 636                                 channel->type,
 637                                 channel->monitor);
 638                 if (new_stream == NULL) {
 639                         switch (alloc_ret) {
 640                         case -ENOMEM:
 641                         case -EINVAL:
 642                         default:
 643                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 644                                 break;
 645                         }
 646                         goto end_nosignal;
 647                 }
 648
 649                 new_stream->chan = channel;
 650                 new_stream->wait_fd = fd;
 651                 switch (channel->output) {
 652                 case CONSUMER_CHANNEL_SPLICE:
 653                         new_stream->output = LTTNG_EVENT_SPLICE;
 654                         break;
 655                 case CONSUMER_CHANNEL_MMAP:
 656                         new_stream->output = LTTNG_EVENT_MMAP;
 657                         break;
 658                 default:
 659                         ERR("Stream output unknown %d", channel->output);
 660                         goto end_nosignal;
 661                 }
 662
 663                 /*
 664                  * We've just assigned the channel to the stream so increment the
 665                  * refcount right now. We don't need to increment the refcount for
 666                  * streams in no monitor because we handle manually the cleanup of
 667                  * those. It is very important to make sure there is NO prior
 668                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 669                  */
 670                 if (channel->monitor) {
 671                         uatomic_inc(&new_stream->chan->refcount);
 672                 }
 673
 674                 /*
 675                  * The buffer flush is done on the session daemon side for the kernel
 676                  * so no need for the stream "hangup_flush_done" variable to be
 677                  * tracked. This is important for a kernel stream since we don't rely
 678                  * on the flush state of the stream to read data. It's not the case for
 679                  * user space tracing.
 680                  */
 681                 new_stream->hangup_flush_done = 0;
 682
 683                 health_code_update();
 684
 685                 if (ctx->on_recv_stream) {
 686                         ret = ctx->on_recv_stream(new_stream);
 687                         if (ret < 0) {
 688                                 consumer_stream_free(new_stream);
 689                                 goto end_nosignal;
 690                         }
 691                 }
 692
 693                 health_code_update();
 694
 695                 if (new_stream->metadata_flag) {
 696                         channel->metadata_stream = new_stream;
 697                 }
 698
 699                 /* Do not monitor this stream. */
 700                 if (!channel->monitor) {
 701                         DBG("Kernel consumer add stream %s in no monitor mode with "
 702                                         "relayd id %" PRIu64, new_stream->name,
 703                                         new_stream->net_seq_idx);
 704                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 705                         break;
 706                 }
 707
 708                 /* Send stream to relayd if the stream has an ID. */
 709                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 710                         ret = consumer_send_relayd_stream(new_stream,
 711                                         new_stream->chan->pathname);
 712                         if (ret < 0) {
 713                                 consumer_stream_free(new_stream);
 714                                 goto end_nosignal;
 715                         }
 716                 }
 717
 718                 /* Get the right pipe where the stream will be sent. */
 719                 if (new_stream->metadata_flag) {
 720                         ret = consumer_add_metadata_stream(new_stream);
 721                         if (ret) {
 722                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 723                                                 new_stream->key);
 724                                 consumer_stream_free(new_stream);
 725                                 goto end_nosignal;
 726                         }
 727                         stream_pipe = ctx->consumer_metadata_pipe;
 728                 } else {
 729                         ret = consumer_add_data_stream(new_stream);
 730                         if (ret) {
 731                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 732                                                 new_stream->key);
 733                                 consumer_stream_free(new_stream);
 734                                 goto end_nosignal;
 735                         }
 736                         stream_pipe = ctx->consumer_data_pipe;
 737                 }
 738
 739                 /* Vitible to other threads */
 740                 new_stream->globally_visible = 1;
 741
 742                 health_code_update();
 743
 744                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 745                 if (ret < 0) {
 746                         ERR("Consumer write %s stream to pipe %d",
 747                                         new_stream->metadata_flag ? "metadata" : "data",
 748                                         lttng_pipe_get_writefd(stream_pipe));
 749                         if (new_stream->metadata_flag) {
 750                                 consumer_del_stream_for_metadata(new_stream);
 751                         } else {
 752                                 consumer_del_stream_for_data(new_stream);
 753                         }
 754                         goto end_nosignal;
 755                 }
 756
 757                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 758                                 new_stream->name, fd, new_stream->relayd_stream_id);
 759                 break;
 760         }
 761         case LTTNG_CONSUMER_STREAMS_SENT:
 762         {
 763                 struct lttng_consumer_channel *channel;
 764
 765                 /*
 766                  * Get stream's channel reference. Needed when adding the stream to the
 767                  * global hash table.
 768                  */
 769                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 770                 if (!channel) {
 771                         /*
 772                          * We could not find the channel. Can happen if cpu hotplug
 773                          * happens while tearing down.
 774                          */
 775                         ERR("Unable to find channel key %" PRIu64,
 776                                         msg.u.sent_streams.channel_key);
 777                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 778                 }
 779
 780                 health_code_update();
 781
 782                 /*
 783                  * Send status code to session daemon.
 784                  */
 785                 ret = consumer_send_status_msg(sock, ret_code);
 786                 if (ret < 0) {
 787                         /* Somehow, the session daemon is not responding anymore. */
 788                         goto end_nosignal;
 789                 }
 790
 791                 health_code_update();
 792
 793                 /*
 794                  * We should not send this message if we don't monitor the
 795                  * streams in this channel.
 796                  */
 797                 if (!channel->monitor) {
 798                         break;
 799                 }
 800
 801                 health_code_update();
 802                 /* Send stream to relayd if the stream has an ID. */
 803                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 804                         ret = consumer_send_relayd_streams_sent(
 805                                         msg.u.sent_streams.net_seq_idx);
 806                         if (ret < 0) {
 807                                 goto end_nosignal;
 808                         }
 809                 }
 810                 break;
 811         }
 812         case LTTNG_CONSUMER_UPDATE_STREAM:
 813         {
 814                 rcu_read_unlock();
 815                 return -ENOSYS;
 816         }
 817         case LTTNG_CONSUMER_DESTROY_RELAYD:
 818         {
 819                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 820                 struct consumer_relayd_sock_pair *relayd;
 821
 822                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 823
 824                 /* Get relayd reference if exists. */
 825                 relayd = consumer_find_relayd(index);
 826                 if (relayd == NULL) {
 827                         DBG("Unable to find relayd %" PRIu64, index);
 828                         ret_code = LTTNG_ERR_NO_CONSUMER;
 829                 }
 830
 831                 /*
 832                  * Each relayd socket pair has a refcount of stream attached to it
 833                  * which tells if the relayd is still active or not depending on the
 834                  * refcount value.
 835                  *
 836                  * This will set the destroy flag of the relayd object and destroy it
 837                  * if the refcount reaches zero when called.
 838                  *
 839                  * The destroy can happen either here or when a stream fd hangs up.
 840                  */
 841                 if (relayd) {
 842                         consumer_flag_relayd_for_destroy(relayd);
 843                 }
 844
 845                 health_code_update();
 846
 847                 ret = consumer_send_status_msg(sock, ret_code);
 848                 if (ret < 0) {
 849                         /* Somehow, the session daemon is not responding anymore. */
 850                         goto error_fatal;
 851                 }
 852
 853                 goto end_nosignal;
 854         }
 855         case LTTNG_CONSUMER_DATA_PENDING:
 856         {
 857                 int32_t ret;
 858                 uint64_t id = msg.u.data_pending.session_id;
 859
 860                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 861
 862                 ret = consumer_data_pending(id);
 863
 864                 health_code_update();
 865
 866                 /* Send back returned value to session daemon */
 867                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 868                 if (ret < 0) {
 869                         PERROR("send data pending ret code");
 870                         goto error_fatal;
 871                 }
 872
 873                 /*
 874                  * No need to send back a status message since the data pending
 875                  * returned value is the response.
 876                  */
 877                 break;
 878         }
 879         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 880         {
 881                 if (msg.u.snapshot_channel.metadata == 1) {
 882                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 883                                         msg.u.snapshot_channel.pathname,
 884                                         msg.u.snapshot_channel.relayd_id, ctx);
 885                         if (ret < 0) {
 886                                 ERR("Snapshot metadata failed");
 887                                 ret_code = LTTNG_ERR_KERN_META_FAIL;
 888                         }
 889                 } else {
 890                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 891                                         msg.u.snapshot_channel.pathname,
 892                                         msg.u.snapshot_channel.relayd_id,
 893                                         msg.u.snapshot_channel.max_stream_size,
 894                                         ctx);
 895                         if (ret < 0) {
 896                                 ERR("Snapshot channel failed");
 897                                 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
 898                         }
 899                 }
 900
 901                 health_code_update();
 902
 903                 ret = consumer_send_status_msg(sock, ret_code);
 904                 if (ret < 0) {
 905                         /* Somehow, the session daemon is not responding anymore. */
 906                         goto end_nosignal;
 907                 }
 908                 break;
 909         }
 910         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 911         {
 912                 uint64_t key = msg.u.destroy_channel.key;
 913                 struct lttng_consumer_channel *channel;
 914
 915                 channel = consumer_find_channel(key);
 916                 if (!channel) {
 917                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 918                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 919                 }
 920
 921                 health_code_update();
 922
 923                 ret = consumer_send_status_msg(sock, ret_code);
 924                 if (ret < 0) {
 925                         /* Somehow, the session daemon is not responding anymore. */
 926                         goto end_nosignal;
 927                 }
 928
 929                 health_code_update();
 930
 931                 /*
 932                  * This command should ONLY be issued for channel with streams set in
 933                  * no monitor mode.
 934                  */
 935                 assert(!channel->monitor);
 936
 937                 /*
 938                  * The refcount should ALWAYS be 0 in the case of a channel in no
 939                  * monitor mode.
 940                  */
 941                 assert(!uatomic_sub_return(&channel->refcount, 1));
 942
 943                 consumer_del_channel(channel);
 944
 945                 goto end_nosignal;
 946         }
 947         default:
 948                 goto end_nosignal;
 949         }
 950
 951 end_nosignal:
 952         rcu_read_unlock();
 953
 954         /*
 955          * Return 1 to indicate success since the 0 value can be a socket
 956          * shutdown during the recv() or send() call.
 957          */
 958         health_code_update();
 959         return 1;
 960
 961 error_fatal:
 962         rcu_read_unlock();
 963         /* This will issue a consumer stop. */
 964         return -1;
 965 }
 966
 967 /*
 968  * Populate index values of a kernel stream. Values are set in big endian order.
 969  *
 970  * Return 0 on success or else a negative value.
 971  */
 972 static int get_index_values(struct ctf_packet_index *index, int infd)
 973 {
 974         int ret;
 975
 976         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 977         if (ret < 0) {
 978                 PERROR("kernctl_get_timestamp_begin");
 979                 goto error;
 980         }
 981         index->timestamp_begin = htobe64(index->timestamp_begin);
 982
 983         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 984         if (ret < 0) {
 985                 PERROR("kernctl_get_timestamp_end");
 986                 goto error;
 987         }
 988         index->timestamp_end = htobe64(index->timestamp_end);
 989
 990         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 991         if (ret < 0) {
 992                 PERROR("kernctl_get_events_discarded");
 993                 goto error;
 994         }
 995         index->events_discarded = htobe64(index->events_discarded);
 996
 997         ret = kernctl_get_content_size(infd, &index->content_size);
 998         if (ret < 0) {
 999                 PERROR("kernctl_get_content_size");
1000                 goto error;
1001         }
1002         index->content_size = htobe64(index->content_size);
1003
1004         ret = kernctl_get_packet_size(infd, &index->packet_size);
1005         if (ret < 0) {
1006                 PERROR("kernctl_get_packet_size");
1007                 goto error;
1008         }
1009         index->packet_size = htobe64(index->packet_size);
1010
1011         ret = kernctl_get_stream_id(infd, &index->stream_id);
1012         if (ret < 0) {
1013                 PERROR("kernctl_get_stream_id");
1014                 goto error;
1015         }
1016         index->stream_id = htobe64(index->stream_id);
1017
1018 error:
1019         return ret;
1020 }
1021 /*
1022  * Sync metadata meaning request them to the session daemon and snapshot to the
1023  * metadata thread can consumer them.
1024  *
1025  * Metadata stream lock MUST be acquired.
1026  *
1027  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1028  * is empty or a negative value on error.
1029  */
1030 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1031 {
1032         int ret;
1033
1034         assert(metadata);
1035
1036         ret = kernctl_buffer_flush(metadata->wait_fd);
1037         if (ret < 0) {
1038                 ERR("Failed to flush kernel stream");
1039                 goto end;
1040         }
1041
1042         ret = kernctl_snapshot(metadata->wait_fd);
1043         if (ret < 0) {
1044                 if (errno != EAGAIN) {
1045                         ERR("Sync metadata, taking kernel snapshot failed.");
1046                         goto end;
1047                 }
1048                 DBG("Sync metadata, no new kernel metadata");
1049                 /* No new metadata, exit. */
1050                 ret = ENODATA;
1051                 goto end;
1052         }
1053
1054 end:
1055         return ret;
1056 }
1057
1058 /*
1059  * Consume data on a file descriptor and write it on a trace file.
1060  */
1061 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1062                 struct lttng_consumer_local_data *ctx)
1063 {
1064         unsigned long len, subbuf_size, padding;
1065         int err, write_index = 1;
1066         ssize_t ret = 0;
1067         int infd = stream->wait_fd;
1068         struct ctf_packet_index index;
1069
1070         DBG("In read_subbuffer (infd : %d)", infd);
1071
1072         /* Get the next subbuffer */
1073         err = kernctl_get_next_subbuf(infd);
1074         if (err != 0) {
1075                 /*
1076                  * This is a debug message even for single-threaded consumer,
1077                  * because poll() have more relaxed criterions than get subbuf,
1078                  * so get_subbuf may fail for short race windows where poll()
1079                  * would issue wakeups.
1080                  */
1081                 DBG("Reserving sub buffer failed (everything is normal, "
1082                                 "it is due to concurrency)");
1083                 ret = -errno;
1084                 goto end;
1085         }
1086
1087         /* Get the full subbuffer size including padding */
1088         err = kernctl_get_padded_subbuf_size(infd, &len);
1089         if (err != 0) {
1090                 perror("Getting sub-buffer len failed.");
1091                 ret = -errno;
1092                 goto end;
1093         }
1094
1095         if (!stream->metadata_flag) {
1096                 ret = get_index_values(&index, infd);
1097                 if (ret < 0) {
1098                         goto end;
1099                 }
1100         } else {
1101                 write_index = 0;
1102         }
1103
1104         switch (stream->chan->output) {
1105         case CONSUMER_CHANNEL_SPLICE:
1106                 /*
1107                  * XXX: The lttng-modules splice "actor" does not handle copying
1108                  * partial pages hence only using the subbuffer size without the
1109                  * padding makes the splice fail.
1110                  */
1111                 subbuf_size = len;
1112                 padding = 0;
1113
1114                 /* splice the subbuffer to the tracefile */
1115                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1116                                 padding, &index);
1117                 /*
1118                  * XXX: Splice does not support network streaming so the return value
1119                  * is simply checked against subbuf_size and not like the mmap() op.
1120                  */
1121                 if (ret != subbuf_size) {
1122                         /*
1123                          * display the error but continue processing to try
1124                          * to release the subbuffer
1125                          */
1126                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1127                                         ret, subbuf_size);
1128                         write_index = 0;
1129                 }
1130                 break;
1131         case CONSUMER_CHANNEL_MMAP:
1132                 /* Get subbuffer size without padding */
1133                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1134                 if (err != 0) {
1135                         perror("Getting sub-buffer len failed.");
1136                         ret = -errno;
1137                         goto end;
1138                 }
1139
1140                 /* Make sure the tracer is not gone mad on us! */
1141                 assert(len >= subbuf_size);
1142
1143                 padding = len - subbuf_size;
1144
1145                 /* write the subbuffer to the tracefile */
1146                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1147                                 padding, &index);
1148                 /*
1149                  * The mmap operation should write subbuf_size amount of data when
1150                  * network streaming or the full padding (len) size when we are _not_
1151                  * streaming.
1152                  */
1153                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1154                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1155                         /*
1156                          * Display the error but continue processing to try to release the
1157                          * subbuffer. This is a DBG statement since this is possible to
1158                          * happen without being a critical error.
1159                          */
1160                         DBG("Error writing to tracefile "
1161                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1162                                         ret, len, subbuf_size);
1163                         write_index = 0;
1164                 }
1165                 break;
1166         default:
1167                 ERR("Unknown output method");
1168                 ret = -EPERM;
1169         }
1170
1171         err = kernctl_put_next_subbuf(infd);
1172         if (err != 0) {
1173                 if (errno == EFAULT) {
1174                         perror("Error in unreserving sub buffer\n");
1175                 } else if (errno == EIO) {
1176                         /* Should never happen with newer LTTng versions */
1177                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
1178                 }
1179                 ret = -errno;
1180                 goto end;
1181         }
1182
1183         /* Write index if needed. */
1184         if (!write_index) {
1185                 goto end;
1186         }
1187
1188         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1189                 /*
1190                  * In live, block until all the metadata is sent.
1191                  */
1192                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1193                 if (err < 0) {
1194                         goto end;
1195                 }
1196         }
1197
1198         err = consumer_stream_write_index(stream, &index);
1199         if (err < 0) {
1200                 goto end;
1201         }
1202
1203 end:
1204         return ret;
1205 }
1206
1207 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1208 {
1209         int ret;
1210
1211         assert(stream);
1212
1213         /*
1214          * Don't create anything if this is set for streaming or should not be
1215          * monitored.
1216          */
1217         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1218                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1219                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1220                                 stream->uid, stream->gid, NULL);
1221                 if (ret < 0) {
1222                         goto error;
1223                 }
1224                 stream->out_fd = ret;
1225                 stream->tracefile_size_current = 0;
1226
1227                 if (!stream->metadata_flag) {
1228                         ret = index_create_file(stream->chan->pathname,
1229                                         stream->name, stream->uid, stream->gid,
1230                                         stream->chan->tracefile_size,
1231                                         stream->tracefile_count_current);
1232                         if (ret < 0) {
1233                                 goto error;
1234                         }
1235                         stream->index_fd = ret;
1236                 }
1237         }
1238
1239         if (stream->output == LTTNG_EVENT_MMAP) {
1240                 /* get the len of the mmap region */
1241                 unsigned long mmap_len;
1242
1243                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1244                 if (ret != 0) {
1245                         PERROR("kernctl_get_mmap_len");
1246                         ret = -errno;
1247                         goto error_close_fd;
1248                 }
1249                 stream->mmap_len = (size_t) mmap_len;
1250
1251                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1252                                 MAP_PRIVATE, stream->wait_fd, 0);
1253                 if (stream->mmap_base == MAP_FAILED) {
1254                         PERROR("Error mmaping");
1255                         ret = -1;
1256                         goto error_close_fd;
1257                 }
1258         }
1259
1260         /* we return 0 to let the library handle the FD internally */
1261         return 0;
1262
1263 error_close_fd:
1264         if (stream->out_fd >= 0) {
1265                 int err;
1266
1267                 err = close(stream->out_fd);
1268                 assert(!err);
1269                 stream->out_fd = -1;
1270         }
1271 error:
1272         return ret;
1273 }
1274
1275 /*
1276  * Check if data is still being extracted from the buffers for a specific
1277  * stream. Consumer data lock MUST be acquired before calling this function
1278  * and the stream lock.
1279  *
1280  * Return 1 if the traced data are still getting read else 0 meaning that the
1281  * data is available for trace viewer reading.
1282  */
1283 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1284 {
1285         int ret;
1286
1287         assert(stream);
1288
1289         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1290                 ret = 0;
1291                 goto end;
1292         }
1293
1294         ret = kernctl_get_next_subbuf(stream->wait_fd);
1295         if (ret == 0) {
1296                 /* There is still data so let's put back this subbuffer. */
1297                 ret = kernctl_put_subbuf(stream->wait_fd);
1298                 assert(ret == 0);
1299                 ret = 1;   /* Data is pending */
1300                 goto end;
1301         }
1302
1303         /* Data is NOT pending and ready to be read. */
1304         ret = 0;
1305
1306 end:
1307         return ret;
1308 }