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