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 <common/common.h>
  33 #include <common/kernel-ctl/kernel-ctl.h>
  34 #include <common/sessiond-comm/sessiond-comm.h>
  35 #include <common/sessiond-comm/relayd.h>
  36 #include <common/compat/fcntl.h>
  37 #include <common/pipe.h>
  38 #include <common/relayd/relayd.h>
  39 #include <common/utils.h>
  40 #include <common/consumer-stream.h>
  41
  42 #include "kernel-consumer.h"
  43
  44 extern struct lttng_consumer_global_data consumer_data;
  45 extern int consumer_poll_timeout;
  46 extern volatile int consumer_quit;
  47
  48 /*
  49  * Take a snapshot for a specific fd
  50  *
  51  * Returns 0 on success, < 0 on error
  52  */
  53 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  54 {
  55         int ret = 0;
  56         int infd = stream->wait_fd;
  57
  58         ret = kernctl_snapshot(infd);
  59         if (ret != 0) {
  60                 errno = -ret;
  61                 perror("Getting sub-buffer snapshot.");
  62         }
  63
  64         return ret;
  65 }
  66
  67 /*
  68  * Get the produced position
  69  *
  70  * Returns 0 on success, < 0 on error
  71  */
  72 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  73                 unsigned long *pos)
  74 {
  75         int ret;
  76         int infd = stream->wait_fd;
  77
  78         ret = kernctl_snapshot_get_produced(infd, pos);
  79         if (ret != 0) {
  80                 errno = -ret;
  81                 perror("kernctl_snapshot_get_produced");
  82         }
  83
  84         return ret;
  85 }
  86
  87 /*
  88  * Get the consumerd position
  89  *
  90  * Returns 0 on success, < 0 on error
  91  */
  92 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  93                 unsigned long *pos)
  94 {
  95         int ret;
  96         int infd = stream->wait_fd;
  97
  98         ret = kernctl_snapshot_get_consumed(infd, pos);
  99         if (ret != 0) {
 100                 errno = -ret;
 101                 perror("kernctl_snapshot_get_consumed");
 102         }
 103
 104         return ret;
 105 }
 106
 107 /*
 108  * Take a snapshot of all the stream of a channel
 109  *
 110  * Returns 0 on success, < 0 on error
 111  */
 112 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 113                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 114 {
 115         int ret;
 116         unsigned long consumed_pos, produced_pos;
 117         struct lttng_consumer_channel *channel;
 118         struct lttng_consumer_stream *stream;
 119
 120         DBG("Kernel consumer snapshot channel %lu", key);
 121
 122         rcu_read_lock();
 123
 124         channel = consumer_find_channel(key);
 125         if (!channel) {
 126                 ERR("No channel found for key %lu", key);
 127                 ret = -1;
 128                 goto end;
 129         }
 130
 131         /* Splice is not supported yet for channel snapshot. */
 132         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 133                 ERR("Unsupported output %d", channel->output);
 134                 ret = -1;
 135                 goto end;
 136         }
 137
 138         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 139                 /*
 140                  * Lock stream because we are about to change its state.
 141                  */
 142                 pthread_mutex_lock(&stream->lock);
 143
 144                 /*
 145                  * Assign the received relayd ID so we can use it for streaming. The streams
 146                  * are not visible to anyone so this is OK to change it.
 147                  */
 148                 stream->net_seq_idx = relayd_id;
 149                 channel->relayd_id = relayd_id;
 150                 if (relayd_id != (uint64_t) -1ULL) {
 151                         ret = consumer_send_relayd_stream(stream, path);
 152                         if (ret < 0) {
 153                                 ERR("sending stream to relayd");
 154                                 goto end_unlock;
 155                         }
 156                 } else {
 157                         ret = utils_create_stream_file(path, stream->name,
 158                                         stream->chan->tracefile_size,
 159                                         stream->tracefile_count_current,
 160                                         stream->uid, stream->gid);
 161                         if (ret < 0) {
 162                                 ERR("utils_create_stream_file");
 163                                 goto end_unlock;
 164                         }
 165
 166                         stream->out_fd = ret;
 167                         stream->tracefile_size_current = 0;
 168
 169                         DBG("Kernel consumer snapshot stream %s/%s (%lu)", path,
 170                                         stream->name, stream->key);
 171                 }
 172
 173                 ret = kernctl_buffer_flush(stream->wait_fd);
 174                 if (ret < 0) {
 175                         ERR("Failed to flush kernel stream");
 176                         goto end_unlock;
 177                 }
 178
 179                 ret = lttng_kconsumer_take_snapshot(stream);
 180                 if (ret < 0) {
 181                         ERR("Taking kernel snapshot");
 182                         goto end_unlock;
 183                 }
 184
 185                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 186                 if (ret < 0) {
 187                         ERR("Produced kernel snapshot position");
 188                         goto end_unlock;
 189                 }
 190
 191                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 192                 if (ret < 0) {
 193                         ERR("Consumerd kernel snapshot position");
 194                         goto end_unlock;
 195                 }
 196
 197                 if (stream->max_sb_size == 0) {
 198                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 199                                         &stream->max_sb_size);
 200                         if (ret < 0) {
 201                                 ERR("Getting kernel max_sb_size");
 202                                 goto end_unlock;
 203                         }
 204                 }
 205
 206                 while (consumed_pos < produced_pos) {
 207                         ssize_t read_len;
 208                         unsigned long len, padded_len;
 209
 210                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 211
 212                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 213                         if (ret < 0) {
 214                                 if (errno != EAGAIN) {
 215                                         PERROR("kernctl_get_subbuf snapshot");
 216                                         goto end_unlock;
 217                                 }
 218                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 219                                 consumed_pos += stream->max_sb_size;
 220                                 continue;
 221                         }
 222
 223                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 224                         if (ret < 0) {
 225                                 ERR("Snapshot kernctl_get_subbuf_size");
 226                                 goto error_put_subbuf;
 227                         }
 228
 229                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 230                         if (ret < 0) {
 231                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 232                                 goto error_put_subbuf;
 233                         }
 234
 235                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 236                                         padded_len - len);
 237                         /*
 238                          * We write the padded len in local tracefiles but the data len
 239                          * when using a relay. Display the error but continue processing
 240                          * to try to release the subbuffer.
 241                          */
 242                         if (relayd_id != (uint64_t) -1ULL) {
 243                                 if (read_len != len) {
 244                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 245                                                         read_len, len);
 246                                 }
 247                         } else {
 248                                 if (read_len != padded_len) {
 249                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 250                                                         read_len, padded_len);
 251                                 }
 252                         }
 253
 254                         ret = kernctl_put_subbuf(stream->wait_fd);
 255                         if (ret < 0) {
 256                                 ERR("Snapshot kernctl_put_subbuf");
 257                                 goto end_unlock;
 258                         }
 259                         consumed_pos += stream->max_sb_size;
 260                 }
 261
 262                 if (relayd_id == (uint64_t) -1ULL) {
 263                         ret = close(stream->out_fd);
 264                         if (ret < 0) {
 265                                 PERROR("Kernel consumer snapshot close out_fd");
 266                                 goto end_unlock;
 267                         }
 268                         stream->out_fd = -1;
 269                 } else {
 270                         close_relayd_stream(stream);
 271                         stream->net_seq_idx = (uint64_t) -1ULL;
 272                 }
 273                 pthread_mutex_unlock(&stream->lock);
 274         }
 275
 276         /* All good! */
 277         ret = 0;
 278         goto end;
 279
 280 error_put_subbuf:
 281         ret = kernctl_put_subbuf(stream->wait_fd);
 282         if (ret < 0) {
 283                 ERR("Snapshot kernctl_put_subbuf error path");
 284         }
 285 end_unlock:
 286         pthread_mutex_unlock(&stream->lock);
 287 end:
 288         rcu_read_unlock();
 289         return ret;
 290 }
 291
 292 /*
 293  * Read the whole metadata available for a snapshot.
 294  *
 295  * Returns 0 on success, < 0 on error
 296  */
 297 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 298                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 299 {
 300         int ret, use_relayd = 0;
 301         ssize_t ret_read;
 302         struct lttng_consumer_channel *metadata_channel;
 303         struct lttng_consumer_stream *metadata_stream;
 304
 305         assert(ctx);
 306
 307         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 308                         key, path);
 309
 310         rcu_read_lock();
 311
 312         metadata_channel = consumer_find_channel(key);
 313         if (!metadata_channel) {
 314                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 315                 ret = -1;
 316                 goto error;
 317         }
 318
 319         metadata_stream = metadata_channel->metadata_stream;
 320         assert(metadata_stream);
 321
 322         /* Flag once that we have a valid relayd for the stream. */
 323         if (relayd_id != (uint64_t) -1ULL) {
 324                 use_relayd = 1;
 325         }
 326
 327         if (use_relayd) {
 328                 ret = consumer_send_relayd_stream(metadata_stream, path);
 329                 if (ret < 0) {
 330                         goto error;
 331                 }
 332         } else {
 333                 ret = utils_create_stream_file(path, metadata_stream->name,
 334                                 metadata_stream->chan->tracefile_size,
 335                                 metadata_stream->tracefile_count_current,
 336                                 metadata_stream->uid, metadata_stream->gid);
 337                 if (ret < 0) {
 338                         goto error;
 339                 }
 340                 metadata_stream->out_fd = ret;
 341         }
 342
 343         do {
 344                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 345                 if (ret_read < 0) {
 346                         if (ret_read != -EPERM) {
 347                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %ld)",
 348                                                 ret_read);
 349                                 goto error;
 350                         }
 351                         /* ret_read is negative at this point so we will exit the loop. */
 352                         continue;
 353                 }
 354         } while (ret_read >= 0);
 355
 356         if (use_relayd) {
 357                 close_relayd_stream(metadata_stream);
 358                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 359         } else {
 360                 ret = close(metadata_stream->out_fd);
 361                 if (ret < 0) {
 362                         PERROR("Kernel consumer snapshot metadata close out_fd");
 363                         /*
 364                          * Don't go on error here since the snapshot was successful at this
 365                          * point but somehow the close failed.
 366                          */
 367                 }
 368                 metadata_stream->out_fd = -1;
 369         }
 370
 371         ret = 0;
 372
 373         cds_list_del(&metadata_stream->send_node);
 374         consumer_stream_destroy(metadata_stream, NULL);
 375         metadata_channel->metadata_stream = NULL;
 376 error:
 377         rcu_read_unlock();
 378         return ret;
 379 }
 380
 381 /*
 382  * Receive command from session daemon and process it.
 383  *
 384  * Return 1 on success else a negative value or 0.
 385  */
 386 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 387                 int sock, struct pollfd *consumer_sockpoll)
 388 {
 389         ssize_t ret;
 390         enum lttng_error_code ret_code = LTTNG_OK;
 391         struct lttcomm_consumer_msg msg;
 392
 393         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 394         if (ret != sizeof(msg)) {
 395                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 396                 if (ret > 0) {
 397                         ret = -1;
 398                 }
 399                 return ret;
 400         }
 401         if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
 402                 /*
 403                  * Notify the session daemon that the command is completed.
 404                  *
 405                  * On transport layer error, the function call will print an error
 406                  * message so handling the returned code is a bit useless since we
 407                  * return an error code anyway.
 408                  */
 409                 (void) consumer_send_status_msg(sock, ret_code);
 410                 return -ENOENT;
 411         }
 412
 413         /* relayd needs RCU read-side protection */
 414         rcu_read_lock();
 415
 416         switch (msg.cmd_type) {
 417         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 418         {
 419                 /* Session daemon status message are handled in the following call. */
 420                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 421                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 422                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
 423                 goto end_nosignal;
 424         }
 425         case LTTNG_CONSUMER_ADD_CHANNEL:
 426         {
 427                 struct lttng_consumer_channel *new_channel;
 428                 int ret_recv;
 429
 430                 /* First send a status message before receiving the fds. */
 431                 ret = consumer_send_status_msg(sock, ret_code);
 432                 if (ret < 0) {
 433                         /* Somehow, the session daemon is not responding anymore. */
 434                         goto error_fatal;
 435                 }
 436                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 437                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 438                                 msg.u.channel.session_id, msg.u.channel.pathname,
 439                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 440                                 msg.u.channel.relayd_id, msg.u.channel.output,
 441                                 msg.u.channel.tracefile_size,
 442                                 msg.u.channel.tracefile_count, 0,
 443                                 msg.u.channel.monitor);
 444                 if (new_channel == NULL) {
 445                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 446                         goto end_nosignal;
 447                 }
 448                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 449                 switch (msg.u.channel.output) {
 450                 case LTTNG_EVENT_SPLICE:
 451                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 452                         break;
 453                 case LTTNG_EVENT_MMAP:
 454                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 455                         break;
 456                 default:
 457                         ERR("Channel output unknown %d", msg.u.channel.output);
 458                         goto end_nosignal;
 459                 }
 460
 461                 /* Translate and save channel type. */
 462                 switch (msg.u.channel.type) {
 463                 case CONSUMER_CHANNEL_TYPE_DATA:
 464                 case CONSUMER_CHANNEL_TYPE_METADATA:
 465                         new_channel->type = msg.u.channel.type;
 466                         break;
 467                 default:
 468                         assert(0);
 469                         goto end_nosignal;
 470                 };
 471
 472                 if (ctx->on_recv_channel != NULL) {
 473                         ret_recv = ctx->on_recv_channel(new_channel);
 474                         if (ret_recv == 0) {
 475                                 ret = consumer_add_channel(new_channel, ctx);
 476                         } else if (ret_recv < 0) {
 477                                 goto end_nosignal;
 478                         }
 479                 } else {
 480                         ret = consumer_add_channel(new_channel, ctx);
 481                 }
 482
 483                 /* If we received an error in add_channel, we need to report it. */
 484                 if (ret < 0) {
 485                         ret = consumer_send_status_msg(sock, ret);
 486                         if (ret < 0) {
 487                                 goto error_fatal;
 488                         }
 489                         goto end_nosignal;
 490                 }
 491
 492                 goto end_nosignal;
 493         }
 494         case LTTNG_CONSUMER_ADD_STREAM:
 495         {
 496                 int fd;
 497                 struct lttng_pipe *stream_pipe;
 498                 struct lttng_consumer_stream *new_stream;
 499                 struct lttng_consumer_channel *channel;
 500                 int alloc_ret = 0;
 501
 502                 /*
 503                  * Get stream's channel reference. Needed when adding the stream to the
 504                  * global hash table.
 505                  */
 506                 channel = consumer_find_channel(msg.u.stream.channel_key);
 507                 if (!channel) {
 508                         /*
 509                          * We could not find the channel. Can happen if cpu hotplug
 510                          * happens while tearing down.
 511                          */
 512                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 513                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 514                 }
 515
 516                 /* First send a status message before receiving the fds. */
 517                 ret = consumer_send_status_msg(sock, ret_code);
 518                 if (ret < 0) {
 519                         /* Somehow, the session daemon is not responding anymore. */
 520                         goto error_fatal;
 521                 }
 522                 if (ret_code != LTTNG_OK) {
 523                         /* Channel was not found. */
 524                         goto end_nosignal;
 525                 }
 526
 527                 /* Blocking call */
 528                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 529                         rcu_read_unlock();
 530                         return -EINTR;
 531                 }
 532
 533                 /* Get stream file descriptor from socket */
 534                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 535                 if (ret != sizeof(fd)) {
 536                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 537                         rcu_read_unlock();
 538                         return ret;
 539                 }
 540
 541                 /*
 542                  * Send status code to session daemon only if the recv works. If the
 543                  * above recv() failed, the session daemon is notified through the
 544                  * error socket and the teardown is eventually done.
 545                  */
 546                 ret = consumer_send_status_msg(sock, ret_code);
 547                 if (ret < 0) {
 548                         /* Somehow, the session daemon is not responding anymore. */
 549                         goto end_nosignal;
 550                 }
 551
 552                 new_stream = consumer_allocate_stream(channel->key,
 553                                 fd,
 554                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 555                                 channel->name,
 556                                 channel->uid,
 557                                 channel->gid,
 558                                 channel->relayd_id,
 559                                 channel->session_id,
 560                                 msg.u.stream.cpu,
 561                                 &alloc_ret,
 562                                 channel->type,
 563                                 channel->monitor);
 564                 if (new_stream == NULL) {
 565                         switch (alloc_ret) {
 566                         case -ENOMEM:
 567                         case -EINVAL:
 568                         default:
 569                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 570                                 break;
 571                         }
 572                         goto end_nosignal;
 573                 }
 574
 575                 new_stream->chan = channel;
 576                 new_stream->wait_fd = fd;
 577                 switch (channel->output) {
 578                 case CONSUMER_CHANNEL_SPLICE:
 579                         new_stream->output = LTTNG_EVENT_SPLICE;
 580                         break;
 581                 case CONSUMER_CHANNEL_MMAP:
 582                         new_stream->output = LTTNG_EVENT_MMAP;
 583                         break;
 584                 default:
 585                         ERR("Stream output unknown %d", channel->output);
 586                         goto end_nosignal;
 587                 }
 588
 589                 /*
 590                  * We've just assigned the channel to the stream so increment the
 591                  * refcount right now. We don't need to increment the refcount for
 592                  * streams in no monitor because we handle manually the cleanup of
 593                  * those. It is very important to make sure there is NO prior
 594                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 595                  */
 596                 if (channel->monitor) {
 597                         uatomic_inc(&new_stream->chan->refcount);
 598                 }
 599
 600                 /*
 601                  * The buffer flush is done on the session daemon side for the kernel
 602                  * so no need for the stream "hangup_flush_done" variable to be
 603                  * tracked. This is important for a kernel stream since we don't rely
 604                  * on the flush state of the stream to read data. It's not the case for
 605                  * user space tracing.
 606                  */
 607                 new_stream->hangup_flush_done = 0;
 608
 609                 if (ctx->on_recv_stream) {
 610                         ret = ctx->on_recv_stream(new_stream);
 611                         if (ret < 0) {
 612                                 consumer_stream_free(new_stream);
 613                                 goto end_nosignal;
 614                         }
 615                 }
 616
 617                 if (new_stream->metadata_flag) {
 618                         channel->metadata_stream = new_stream;
 619                 }
 620
 621                 /* Do not monitor this stream. */
 622                 if (!channel->monitor) {
 623                         DBG("Kernel consumer add stream %s in no monitor mode with "
 624                                         "relayd id %" PRIu64, new_stream->name,
 625                                         new_stream->net_seq_idx);
 626                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 627                         break;
 628                 }
 629
 630                 /* Send stream to relayd if the stream has an ID. */
 631                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 632                         ret = consumer_send_relayd_stream(new_stream, NULL);
 633                         if (ret < 0) {
 634                                 consumer_stream_free(new_stream);
 635                                 goto end_nosignal;
 636                         }
 637                 }
 638
 639                 /* Get the right pipe where the stream will be sent. */
 640                 if (new_stream->metadata_flag) {
 641                         stream_pipe = ctx->consumer_metadata_pipe;
 642                 } else {
 643                         stream_pipe = ctx->consumer_data_pipe;
 644                 }
 645
 646                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 647                 if (ret < 0) {
 648                         ERR("Consumer write %s stream to pipe %d",
 649                                         new_stream->metadata_flag ? "metadata" : "data",
 650                                         lttng_pipe_get_writefd(stream_pipe));
 651                         consumer_stream_free(new_stream);
 652                         goto end_nosignal;
 653                 }
 654
 655                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 656                                 new_stream->name, fd, new_stream->relayd_stream_id);
 657                 break;
 658         }
 659         case LTTNG_CONSUMER_UPDATE_STREAM:
 660         {
 661                 rcu_read_unlock();
 662                 return -ENOSYS;
 663         }
 664         case LTTNG_CONSUMER_DESTROY_RELAYD:
 665         {
 666                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 667                 struct consumer_relayd_sock_pair *relayd;
 668
 669                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 670
 671                 /* Get relayd reference if exists. */
 672                 relayd = consumer_find_relayd(index);
 673                 if (relayd == NULL) {
 674                         DBG("Unable to find relayd %" PRIu64, index);
 675                         ret_code = LTTNG_ERR_NO_CONSUMER;
 676                 }
 677
 678                 /*
 679                  * Each relayd socket pair has a refcount of stream attached to it
 680                  * which tells if the relayd is still active or not depending on the
 681                  * refcount value.
 682                  *
 683                  * This will set the destroy flag of the relayd object and destroy it
 684                  * if the refcount reaches zero when called.
 685                  *
 686                  * The destroy can happen either here or when a stream fd hangs up.
 687                  */
 688                 if (relayd) {
 689                         consumer_flag_relayd_for_destroy(relayd);
 690                 }
 691
 692                 ret = consumer_send_status_msg(sock, ret_code);
 693                 if (ret < 0) {
 694                         /* Somehow, the session daemon is not responding anymore. */
 695                         goto error_fatal;
 696                 }
 697
 698                 goto end_nosignal;
 699         }
 700         case LTTNG_CONSUMER_DATA_PENDING:
 701         {
 702                 int32_t ret;
 703                 uint64_t id = msg.u.data_pending.session_id;
 704
 705                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 706
 707                 ret = consumer_data_pending(id);
 708
 709                 /* Send back returned value to session daemon */
 710                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 711                 if (ret < 0) {
 712                         PERROR("send data pending ret code");
 713                         goto error_fatal;
 714                 }
 715
 716                 /*
 717                  * No need to send back a status message since the data pending
 718                  * returned value is the response.
 719                  */
 720                 break;
 721         }
 722         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 723         {
 724                 if (msg.u.snapshot_channel.metadata == 1) {
 725                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 726                                         msg.u.snapshot_channel.pathname,
 727                                         msg.u.snapshot_channel.relayd_id, ctx);
 728                         if (ret < 0) {
 729                                 ERR("Snapshot metadata failed");
 730                                 ret_code = LTTNG_ERR_KERN_META_FAIL;
 731                         }
 732                 } else {
 733                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 734                                         msg.u.snapshot_channel.pathname,
 735                                         msg.u.snapshot_channel.relayd_id, ctx);
 736                         if (ret < 0) {
 737                                 ERR("Snapshot channel failed");
 738                                 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
 739                         }
 740                 }
 741
 742                 ret = consumer_send_status_msg(sock, ret_code);
 743                 if (ret < 0) {
 744                         /* Somehow, the session daemon is not responding anymore. */
 745                         goto end_nosignal;
 746                 }
 747                 break;
 748         }
 749         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 750         {
 751                 uint64_t key = msg.u.destroy_channel.key;
 752                 struct lttng_consumer_channel *channel;
 753
 754                 channel = consumer_find_channel(key);
 755                 if (!channel) {
 756                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 757                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 758                 }
 759
 760                 ret = consumer_send_status_msg(sock, ret_code);
 761                 if (ret < 0) {
 762                         /* Somehow, the session daemon is not responding anymore. */
 763                         goto end_nosignal;
 764                 }
 765
 766                 /*
 767                  * This command should ONLY be issued for channel with streams set in
 768                  * no monitor mode.
 769                  */
 770                 assert(!channel->monitor);
 771
 772                 /*
 773                  * The refcount should ALWAYS be 0 in the case of a channel in no
 774                  * monitor mode.
 775                  */
 776                 assert(!uatomic_sub_return(&channel->refcount, 1));
 777
 778                 consumer_del_channel(channel);
 779
 780                 goto end_nosignal;
 781         }
 782         default:
 783                 goto end_nosignal;
 784         }
 785
 786 end_nosignal:
 787         rcu_read_unlock();
 788
 789         /*
 790          * Return 1 to indicate success since the 0 value can be a socket
 791          * shutdown during the recv() or send() call.
 792          */
 793         return 1;
 794
 795 error_fatal:
 796         rcu_read_unlock();
 797         /* This will issue a consumer stop. */
 798         return -1;
 799 }
 800
 801 /*
 802  * Consume data on a file descriptor and write it on a trace file.
 803  */
 804 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 805                 struct lttng_consumer_local_data *ctx)
 806 {
 807         unsigned long len, subbuf_size, padding;
 808         int err;
 809         ssize_t ret = 0;
 810         int infd = stream->wait_fd;
 811
 812         DBG("In read_subbuffer (infd : %d)", infd);
 813         /* Get the next subbuffer */
 814         err = kernctl_get_next_subbuf(infd);
 815         if (err != 0) {
 816                 ret = err;
 817                 /*
 818                  * This is a debug message even for single-threaded consumer,
 819                  * because poll() have more relaxed criterions than get subbuf,
 820                  * so get_subbuf may fail for short race windows where poll()
 821                  * would issue wakeups.
 822                  */
 823                 DBG("Reserving sub buffer failed (everything is normal, "
 824                                 "it is due to concurrency)");
 825                 goto end;
 826         }
 827
 828         /* Get the full subbuffer size including padding */
 829         err = kernctl_get_padded_subbuf_size(infd, &len);
 830         if (err != 0) {
 831                 errno = -err;
 832                 perror("Getting sub-buffer len failed.");
 833                 ret = err;
 834                 goto end;
 835         }
 836
 837         switch (stream->chan->output) {
 838         case CONSUMER_CHANNEL_SPLICE:
 839                 /*
 840                  * XXX: The lttng-modules splice "actor" does not handle copying
 841                  * partial pages hence only using the subbuffer size without the
 842                  * padding makes the splice fail.
 843                  */
 844                 subbuf_size = len;
 845                 padding = 0;
 846
 847                 /* splice the subbuffer to the tracefile */
 848                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 849                                 padding);
 850                 /*
 851                  * XXX: Splice does not support network streaming so the return value
 852                  * is simply checked against subbuf_size and not like the mmap() op.
 853                  */
 854                 if (ret != subbuf_size) {
 855                         /*
 856                          * display the error but continue processing to try
 857                          * to release the subbuffer
 858                          */
 859                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 860                                         ret, subbuf_size);
 861                 }
 862                 break;
 863         case CONSUMER_CHANNEL_MMAP:
 864                 /* Get subbuffer size without padding */
 865                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 866                 if (err != 0) {
 867                         errno = -err;
 868                         perror("Getting sub-buffer len failed.");
 869                         ret = err;
 870                         goto end;
 871                 }
 872
 873                 /* Make sure the tracer is not gone mad on us! */
 874                 assert(len >= subbuf_size);
 875
 876                 padding = len - subbuf_size;
 877
 878                 /* write the subbuffer to the tracefile */
 879                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 880                                 padding);
 881                 /*
 882                  * The mmap operation should write subbuf_size amount of data when
 883                  * network streaming or the full padding (len) size when we are _not_
 884                  * streaming.
 885                  */
 886                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 887                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 888                         /*
 889                          * Display the error but continue processing to try to release the
 890                          * subbuffer
 891                          */
 892                         ERR("Error writing to tracefile "
 893                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 894                                         ret, len, subbuf_size);
 895                 }
 896                 break;
 897         default:
 898                 ERR("Unknown output method");
 899                 ret = -1;
 900         }
 901
 902         err = kernctl_put_next_subbuf(infd);
 903         if (err != 0) {
 904                 errno = -err;
 905                 if (errno == EFAULT) {
 906                         perror("Error in unreserving sub buffer\n");
 907                 } else if (errno == EIO) {
 908                         /* Should never happen with newer LTTng versions */
 909                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 910                 }
 911
 912                 ret = -err;
 913                 goto end;
 914         }
 915
 916 end:
 917         return ret;
 918 }
 919
 920 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 921 {
 922         int ret;
 923
 924         assert(stream);
 925
 926         /*
 927          * Don't create anything if this is set for streaming or should not be
 928          * monitored.
 929          */
 930         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
 931                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
 932                                 stream->chan->tracefile_size, stream->tracefile_count_current,
 933                                 stream->uid, stream->gid);
 934                 if (ret < 0) {
 935                         goto error;
 936                 }
 937                 stream->out_fd = ret;
 938                 stream->tracefile_size_current = 0;
 939         }
 940
 941         if (stream->output == LTTNG_EVENT_MMAP) {
 942                 /* get the len of the mmap region */
 943                 unsigned long mmap_len;
 944
 945                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 946                 if (ret != 0) {
 947                         errno = -ret;
 948                         PERROR("kernctl_get_mmap_len");
 949                         goto error_close_fd;
 950                 }
 951                 stream->mmap_len = (size_t) mmap_len;
 952
 953                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 954                                 MAP_PRIVATE, stream->wait_fd, 0);
 955                 if (stream->mmap_base == MAP_FAILED) {
 956                         PERROR("Error mmaping");
 957                         ret = -1;
 958                         goto error_close_fd;
 959                 }
 960         }
 961
 962         /* we return 0 to let the library handle the FD internally */
 963         return 0;
 964
 965 error_close_fd:
 966         if (stream->out_fd >= 0) {
 967                 int err;
 968
 969                 err = close(stream->out_fd);
 970                 assert(!err);
 971                 stream->out_fd = -1;
 972         }
 973 error:
 974         return ret;
 975 }
 976
 977 /*
 978  * Check if data is still being extracted from the buffers for a specific
 979  * stream. Consumer data lock MUST be acquired before calling this function
 980  * and the stream lock.
 981  *
 982  * Return 1 if the traced data are still getting read else 0 meaning that the
 983  * data is available for trace viewer reading.
 984  */
 985 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 986 {
 987         int ret;
 988
 989         assert(stream);
 990
 991         ret = kernctl_get_next_subbuf(stream->wait_fd);
 992         if (ret == 0) {
 993                 /* There is still data so let's put back this subbuffer. */
 994                 ret = kernctl_put_subbuf(stream->wait_fd);
 995                 assert(ret == 0);
 996                 ret = 1;   /* Data is pending */
 997                 goto end;
 998         }
 999
1000         /* Data is NOT pending and ready to be read. */
1001         ret = 0;
1002
1003 end:
1004         return ret;
1005 }