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

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