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