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 error:
 374         rcu_read_unlock();
 375         return ret;
 376 }
 377
 378 /*
 379  * Receive command from session daemon and process it.
 380  *
 381  * Return 1 on success else a negative value or 0.
 382  */
 383 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 384                 int sock, struct pollfd *consumer_sockpoll)
 385 {
 386         ssize_t ret;
 387         enum lttng_error_code ret_code = LTTNG_OK;
 388         struct lttcomm_consumer_msg msg;
 389
 390         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 391         if (ret != sizeof(msg)) {
 392                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 393                 if (ret > 0) {
 394                         ret = -1;
 395                 }
 396                 return ret;
 397         }
 398         if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
 399                 /*
 400                  * Notify the session daemon that the command is completed.
 401                  *
 402                  * On transport layer error, the function call will print an error
 403                  * message so handling the returned code is a bit useless since we
 404                  * return an error code anyway.
 405                  */
 406                 (void) consumer_send_status_msg(sock, ret_code);
 407                 return -ENOENT;
 408         }
 409
 410         /* relayd needs RCU read-side protection */
 411         rcu_read_lock();
 412
 413         switch (msg.cmd_type) {
 414         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 415         {
 416                 /* Session daemon status message are handled in the following call. */
 417                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 418                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 419                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
 420                 goto end_nosignal;
 421         }
 422         case LTTNG_CONSUMER_ADD_CHANNEL:
 423         {
 424                 struct lttng_consumer_channel *new_channel;
 425                 int ret_recv;
 426
 427                 /* First send a status message before receiving the fds. */
 428                 ret = consumer_send_status_msg(sock, ret_code);
 429                 if (ret < 0) {
 430                         /* Somehow, the session daemon is not responding anymore. */
 431                         goto error_fatal;
 432                 }
 433                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 434                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 435                                 msg.u.channel.session_id, msg.u.channel.pathname,
 436                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 437                                 msg.u.channel.relayd_id, msg.u.channel.output,
 438                                 msg.u.channel.tracefile_size,
 439                                 msg.u.channel.tracefile_count, 0,
 440                                 msg.u.channel.monitor);
 441                 if (new_channel == NULL) {
 442                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 443                         goto end_nosignal;
 444                 }
 445                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 446                 switch (msg.u.channel.output) {
 447                 case LTTNG_EVENT_SPLICE:
 448                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 449                         break;
 450                 case LTTNG_EVENT_MMAP:
 451                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 452                         break;
 453                 default:
 454                         ERR("Channel output unknown %d", msg.u.channel.output);
 455                         goto end_nosignal;
 456                 }
 457
 458                 /* Translate and save channel type. */
 459                 switch (msg.u.channel.type) {
 460                 case CONSUMER_CHANNEL_TYPE_DATA:
 461                 case CONSUMER_CHANNEL_TYPE_METADATA:
 462                         new_channel->type = msg.u.channel.type;
 463                         break;
 464                 default:
 465                         assert(0);
 466                         goto end_nosignal;
 467                 };
 468
 469                 if (ctx->on_recv_channel != NULL) {
 470                         ret_recv = ctx->on_recv_channel(new_channel);
 471                         if (ret_recv == 0) {
 472                                 ret = consumer_add_channel(new_channel, ctx);
 473                         } else if (ret_recv < 0) {
 474                                 goto end_nosignal;
 475                         }
 476                 } else {
 477                         ret = consumer_add_channel(new_channel, ctx);
 478                 }
 479
 480                 /* If we received an error in add_channel, we need to report it. */
 481                 if (ret < 0) {
 482                         ret = consumer_send_status_msg(sock, ret);
 483                         if (ret < 0) {
 484                                 goto error_fatal;
 485                         }
 486                         goto end_nosignal;
 487                 }
 488
 489                 goto end_nosignal;
 490         }
 491         case LTTNG_CONSUMER_ADD_STREAM:
 492         {
 493                 int fd;
 494                 struct lttng_pipe *stream_pipe;
 495                 struct lttng_consumer_stream *new_stream;
 496                 struct lttng_consumer_channel *channel;
 497                 int alloc_ret = 0;
 498
 499                 /*
 500                  * Get stream's channel reference. Needed when adding the stream to the
 501                  * global hash table.
 502                  */
 503                 channel = consumer_find_channel(msg.u.stream.channel_key);
 504                 if (!channel) {
 505                         /*
 506                          * We could not find the channel. Can happen if cpu hotplug
 507                          * happens while tearing down.
 508                          */
 509                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 510                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 511                 }
 512
 513                 /* First send a status message before receiving the fds. */
 514                 ret = consumer_send_status_msg(sock, ret_code);
 515                 if (ret < 0) {
 516                         /* Somehow, the session daemon is not responding anymore. */
 517                         goto error_fatal;
 518                 }
 519                 if (ret_code != LTTNG_OK) {
 520                         /* Channel was not found. */
 521                         goto end_nosignal;
 522                 }
 523
 524                 /* Blocking call */
 525                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 526                         rcu_read_unlock();
 527                         return -EINTR;
 528                 }
 529
 530                 /* Get stream file descriptor from socket */
 531                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 532                 if (ret != sizeof(fd)) {
 533                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 534                         rcu_read_unlock();
 535                         return ret;
 536                 }
 537
 538                 /*
 539                  * Send status code to session daemon only if the recv works. If the
 540                  * above recv() failed, the session daemon is notified through the
 541                  * error socket and the teardown is eventually done.
 542                  */
 543                 ret = consumer_send_status_msg(sock, ret_code);
 544                 if (ret < 0) {
 545                         /* Somehow, the session daemon is not responding anymore. */
 546                         goto end_nosignal;
 547                 }
 548
 549                 new_stream = consumer_allocate_stream(channel->key,
 550                                 fd,
 551                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 552                                 channel->name,
 553                                 channel->uid,
 554                                 channel->gid,
 555                                 channel->relayd_id,
 556                                 channel->session_id,
 557                                 msg.u.stream.cpu,
 558                                 &alloc_ret,
 559                                 channel->type);
 560                 if (new_stream == NULL) {
 561                         switch (alloc_ret) {
 562                         case -ENOMEM:
 563                         case -EINVAL:
 564                         default:
 565                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 566                                 break;
 567                         }
 568                         goto end_nosignal;
 569                 }
 570
 571                 new_stream->chan = channel;
 572                 new_stream->wait_fd = fd;
 573                 switch (channel->output) {
 574                 case CONSUMER_CHANNEL_SPLICE:
 575                         new_stream->output = LTTNG_EVENT_SPLICE;
 576                         break;
 577                 case CONSUMER_CHANNEL_MMAP:
 578                         new_stream->output = LTTNG_EVENT_MMAP;
 579                         break;
 580                 default:
 581                         ERR("Stream output unknown %d", channel->output);
 582                         goto end_nosignal;
 583                 }
 584
 585                 /*
 586                  * We've just assigned the channel to the stream so increment the
 587                  * refcount right now. We don't need to increment the refcount for
 588                  * streams in no monitor because we handle manually the cleanup of
 589                  * those. It is very important to make sure there is NO prior
 590                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 591                  */
 592                 if (channel->monitor) {
 593                         uatomic_inc(&new_stream->chan->refcount);
 594                 }
 595
 596                 /*
 597                  * The buffer flush is done on the session daemon side for the kernel
 598                  * so no need for the stream "hangup_flush_done" variable to be
 599                  * tracked. This is important for a kernel stream since we don't rely
 600                  * on the flush state of the stream to read data. It's not the case for
 601                  * user space tracing.
 602                  */
 603                 new_stream->hangup_flush_done = 0;
 604
 605                 if (ctx->on_recv_stream) {
 606                         ret = ctx->on_recv_stream(new_stream);
 607                         if (ret < 0) {
 608                                 consumer_stream_free(new_stream);
 609                                 goto end_nosignal;
 610                         }
 611                 }
 612
 613                 if (new_stream->metadata_flag) {
 614                         channel->metadata_stream = new_stream;
 615                 }
 616
 617                 /* Do not monitor this stream. */
 618                 if (!channel->monitor) {
 619                         DBG("Kernel consumer add stream %s in no monitor mode with "
 620                                         "relayd id %" PRIu64, new_stream->name,
 621                                         new_stream->net_seq_idx);
 622                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 623                         break;
 624                 }
 625
 626                 ret = consumer_send_relayd_stream(new_stream, NULL);
 627                 if (ret < 0) {
 628                         consumer_stream_free(new_stream);
 629                         goto end_nosignal;
 630                 }
 631
 632                 /* Get the right pipe where the stream will be sent. */
 633                 if (new_stream->metadata_flag) {
 634                         stream_pipe = ctx->consumer_metadata_pipe;
 635                 } else {
 636                         stream_pipe = ctx->consumer_data_pipe;
 637                 }
 638
 639                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 640                 if (ret < 0) {
 641                         ERR("Consumer write %s stream to pipe %d",
 642                                         new_stream->metadata_flag ? "metadata" : "data",
 643                                         lttng_pipe_get_writefd(stream_pipe));
 644                         consumer_stream_free(new_stream);
 645                         goto end_nosignal;
 646                 }
 647
 648                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 649                                 new_stream->name, fd, new_stream->relayd_stream_id);
 650                 break;
 651         }
 652         case LTTNG_CONSUMER_UPDATE_STREAM:
 653         {
 654                 rcu_read_unlock();
 655                 return -ENOSYS;
 656         }
 657         case LTTNG_CONSUMER_DESTROY_RELAYD:
 658         {
 659                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 660                 struct consumer_relayd_sock_pair *relayd;
 661
 662                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 663
 664                 /* Get relayd reference if exists. */
 665                 relayd = consumer_find_relayd(index);
 666                 if (relayd == NULL) {
 667                         DBG("Unable to find relayd %" PRIu64, index);
 668                         ret_code = LTTNG_ERR_NO_CONSUMER;
 669                 }
 670
 671                 /*
 672                  * Each relayd socket pair has a refcount of stream attached to it
 673                  * which tells if the relayd is still active or not depending on the
 674                  * refcount value.
 675                  *
 676                  * This will set the destroy flag of the relayd object and destroy it
 677                  * if the refcount reaches zero when called.
 678                  *
 679                  * The destroy can happen either here or when a stream fd hangs up.
 680                  */
 681                 if (relayd) {
 682                         consumer_flag_relayd_for_destroy(relayd);
 683                 }
 684
 685                 ret = consumer_send_status_msg(sock, ret_code);
 686                 if (ret < 0) {
 687                         /* Somehow, the session daemon is not responding anymore. */
 688                         goto error_fatal;
 689                 }
 690
 691                 goto end_nosignal;
 692         }
 693         case LTTNG_CONSUMER_DATA_PENDING:
 694         {
 695                 int32_t ret;
 696                 uint64_t id = msg.u.data_pending.session_id;
 697
 698                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 699
 700                 ret = consumer_data_pending(id);
 701
 702                 /* Send back returned value to session daemon */
 703                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 704                 if (ret < 0) {
 705                         PERROR("send data pending ret code");
 706                         goto error_fatal;
 707                 }
 708
 709                 /*
 710                  * No need to send back a status message since the data pending
 711                  * returned value is the response.
 712                  */
 713                 break;
 714         }
 715         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 716         {
 717                 if (msg.u.snapshot_channel.metadata == 1) {
 718                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 719                                         msg.u.snapshot_channel.pathname,
 720                                         msg.u.snapshot_channel.relayd_id, ctx);
 721                         if (ret < 0) {
 722                                 ERR("Snapshot metadata failed");
 723                                 ret_code = LTTNG_ERR_KERN_META_FAIL;
 724                         }
 725                 } else {
 726                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 727                                         msg.u.snapshot_channel.pathname,
 728                                         msg.u.snapshot_channel.relayd_id, ctx);
 729                         if (ret < 0) {
 730                                 ERR("Snapshot channel failed");
 731                                 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
 732                         }
 733                 }
 734
 735                 ret = consumer_send_status_msg(sock, ret_code);
 736                 if (ret < 0) {
 737                         /* Somehow, the session daemon is not responding anymore. */
 738                         goto end_nosignal;
 739                 }
 740                 break;
 741         }
 742         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 743         {
 744                 uint64_t key = msg.u.destroy_channel.key;
 745                 struct lttng_consumer_channel *channel;
 746
 747                 channel = consumer_find_channel(key);
 748                 if (!channel) {
 749                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 750                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 751                 }
 752
 753                 ret = consumer_send_status_msg(sock, ret_code);
 754                 if (ret < 0) {
 755                         /* Somehow, the session daemon is not responding anymore. */
 756                         goto end_nosignal;
 757                 }
 758
 759                 /*
 760                  * This command should ONLY be issued for channel with streams set in
 761                  * no monitor mode.
 762                  */
 763                 assert(!channel->monitor);
 764
 765                 /*
 766                  * The refcount should ALWAYS be 0 in the case of a channel in no
 767                  * monitor mode.
 768                  */
 769                 assert(!uatomic_sub_return(&channel->refcount, 1));
 770
 771                 consumer_del_channel(channel);
 772
 773                 goto end_nosignal;
 774         }
 775         default:
 776                 goto end_nosignal;
 777         }
 778
 779 end_nosignal:
 780         rcu_read_unlock();
 781
 782         /*
 783          * Return 1 to indicate success since the 0 value can be a socket
 784          * shutdown during the recv() or send() call.
 785          */
 786         return 1;
 787
 788 error_fatal:
 789         rcu_read_unlock();
 790         /* This will issue a consumer stop. */
 791         return -1;
 792 }
 793
 794 /*
 795  * Consume data on a file descriptor and write it on a trace file.
 796  */
 797 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 798                 struct lttng_consumer_local_data *ctx)
 799 {
 800         unsigned long len, subbuf_size, padding;
 801         int err;
 802         ssize_t ret = 0;
 803         int infd = stream->wait_fd;
 804
 805         DBG("In read_subbuffer (infd : %d)", infd);
 806         /* Get the next subbuffer */
 807         err = kernctl_get_next_subbuf(infd);
 808         if (err != 0) {
 809                 ret = err;
 810                 /*
 811                  * This is a debug message even for single-threaded consumer,
 812                  * because poll() have more relaxed criterions than get subbuf,
 813                  * so get_subbuf may fail for short race windows where poll()
 814                  * would issue wakeups.
 815                  */
 816                 DBG("Reserving sub buffer failed (everything is normal, "
 817                                 "it is due to concurrency)");
 818                 goto end;
 819         }
 820
 821         /* Get the full subbuffer size including padding */
 822         err = kernctl_get_padded_subbuf_size(infd, &len);
 823         if (err != 0) {
 824                 errno = -err;
 825                 perror("Getting sub-buffer len failed.");
 826                 ret = err;
 827                 goto end;
 828         }
 829
 830         switch (stream->chan->output) {
 831         case CONSUMER_CHANNEL_SPLICE:
 832                 /*
 833                  * XXX: The lttng-modules splice "actor" does not handle copying
 834                  * partial pages hence only using the subbuffer size without the
 835                  * padding makes the splice fail.
 836                  */
 837                 subbuf_size = len;
 838                 padding = 0;
 839
 840                 /* splice the subbuffer to the tracefile */
 841                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 842                                 padding);
 843                 /*
 844                  * XXX: Splice does not support network streaming so the return value
 845                  * is simply checked against subbuf_size and not like the mmap() op.
 846                  */
 847                 if (ret != subbuf_size) {
 848                         /*
 849                          * display the error but continue processing to try
 850                          * to release the subbuffer
 851                          */
 852                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 853                                         ret, subbuf_size);
 854                 }
 855                 break;
 856         case CONSUMER_CHANNEL_MMAP:
 857                 /* Get subbuffer size without padding */
 858                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 859                 if (err != 0) {
 860                         errno = -err;
 861                         perror("Getting sub-buffer len failed.");
 862                         ret = err;
 863                         goto end;
 864                 }
 865
 866                 /* Make sure the tracer is not gone mad on us! */
 867                 assert(len >= subbuf_size);
 868
 869                 padding = len - subbuf_size;
 870
 871                 /* write the subbuffer to the tracefile */
 872                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 873                                 padding);
 874                 /*
 875                  * The mmap operation should write subbuf_size amount of data when
 876                  * network streaming or the full padding (len) size when we are _not_
 877                  * streaming.
 878                  */
 879                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 880                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 881                         /*
 882                          * Display the error but continue processing to try to release the
 883                          * subbuffer
 884                          */
 885                         ERR("Error writing to tracefile "
 886                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 887                                         ret, len, subbuf_size);
 888                 }
 889                 break;
 890         default:
 891                 ERR("Unknown output method");
 892                 ret = -1;
 893         }
 894
 895         err = kernctl_put_next_subbuf(infd);
 896         if (err != 0) {
 897                 errno = -err;
 898                 if (errno == EFAULT) {
 899                         perror("Error in unreserving sub buffer\n");
 900                 } else if (errno == EIO) {
 901                         /* Should never happen with newer LTTng versions */
 902                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 903                 }
 904
 905                 ret = -err;
 906                 goto end;
 907         }
 908
 909 end:
 910         return ret;
 911 }
 912
 913 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 914 {
 915         int ret;
 916
 917         assert(stream);
 918
 919         /*
 920          * Don't create anything if this is set for streaming or should not be
 921          * monitored.
 922          */
 923         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
 924                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
 925                                 stream->chan->tracefile_size, stream->tracefile_count_current,
 926                                 stream->uid, stream->gid);
 927                 if (ret < 0) {
 928                         goto error;
 929                 }
 930                 stream->out_fd = ret;
 931                 stream->tracefile_size_current = 0;
 932         }
 933
 934         if (stream->output == LTTNG_EVENT_MMAP) {
 935                 /* get the len of the mmap region */
 936                 unsigned long mmap_len;
 937
 938                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 939                 if (ret != 0) {
 940                         errno = -ret;
 941                         PERROR("kernctl_get_mmap_len");
 942                         goto error_close_fd;
 943                 }
 944                 stream->mmap_len = (size_t) mmap_len;
 945
 946                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 947                                 MAP_PRIVATE, stream->wait_fd, 0);
 948                 if (stream->mmap_base == MAP_FAILED) {
 949                         PERROR("Error mmaping");
 950                         ret = -1;
 951                         goto error_close_fd;
 952                 }
 953         }
 954
 955         /* we return 0 to let the library handle the FD internally */
 956         return 0;
 957
 958 error_close_fd:
 959         if (stream->out_fd >= 0) {
 960                 int err;
 961
 962                 err = close(stream->out_fd);
 963                 assert(!err);
 964                 stream->out_fd = -1;
 965         }
 966 error:
 967         return ret;
 968 }
 969
 970 /*
 971  * Check if data is still being extracted from the buffers for a specific
 972  * stream. Consumer data lock MUST be acquired before calling this function
 973  * and the stream lock.
 974  *
 975  * Return 1 if the traced data are still getting read else 0 meaning that the
 976  * data is available for trace viewer reading.
 977  */
 978 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 979 {
 980         int ret;
 981
 982         assert(stream);
 983
 984         ret = kernctl_get_next_subbuf(stream->wait_fd);
 985         if (ret == 0) {
 986                 /* There is still data so let's put back this subbuffer. */
 987                 ret = kernctl_put_subbuf(stream->wait_fd);
 988                 assert(ret == 0);
 989                 ret = 1;   /* Data is pending */
 990                 goto end;
 991         }
 992
 993         /* Data is NOT pending and ready to be read. */
 994         ret = 0;
 995
 996 end:
 997         return ret;
 998 }