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 _LGPL_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/compat/endian.h>
  39 #include <common/pipe.h>
  40 #include <common/relayd/relayd.h>
  41 #include <common/utils.h>
  42 #include <common/consumer-stream.h>
  43 #include <common/index/index.h>
  44 #include <common/consumer-timer.h>
  45
  46 #include "kernel-consumer.h"
  47
  48 extern struct lttng_consumer_global_data consumer_data;
  49 extern int consumer_poll_timeout;
  50 extern volatile int consumer_quit;
  51
  52 /*
  53  * Take a snapshot for a specific fd
  54  *
  55  * Returns 0 on success, < 0 on error
  56  */
  57 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  58 {
  59         int ret = 0;
  60         int infd = stream->wait_fd;
  61
  62         ret = kernctl_snapshot(infd);
  63         if (ret != 0) {
  64                 PERROR("Getting sub-buffer snapshot.");
  65                 ret = -errno;
  66         }
  67
  68         return ret;
  69 }
  70
  71 /*
  72  * Get the produced position
  73  *
  74  * Returns 0 on success, < 0 on error
  75  */
  76 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  77                 unsigned long *pos)
  78 {
  79         int ret;
  80         int infd = stream->wait_fd;
  81
  82         ret = kernctl_snapshot_get_produced(infd, pos);
  83         if (ret != 0) {
  84                 PERROR("kernctl_snapshot_get_produced");
  85                 ret = -errno;
  86         }
  87
  88         return ret;
  89 }
  90
  91 /*
  92  * Get the consumerd position
  93  *
  94  * Returns 0 on success, < 0 on error
  95  */
  96 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
  97                 unsigned long *pos)
  98 {
  99         int ret;
 100         int infd = stream->wait_fd;
 101
 102         ret = kernctl_snapshot_get_consumed(infd, pos);
 103         if (ret != 0) {
 104                 PERROR("kernctl_snapshot_get_consumed");
 105                 ret = -errno;
 106         }
 107
 108         return ret;
 109 }
 110
 111 /*
 112  * Take a snapshot of all the stream of a channel
 113  *
 114  * Returns 0 on success, < 0 on error
 115  */
 116 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
 117                 uint64_t relayd_id, uint64_t nb_packets_per_stream,
 118                 struct lttng_consumer_local_data *ctx)
 119 {
 120         int ret;
 121         unsigned long consumed_pos, produced_pos;
 122         struct lttng_consumer_channel *channel;
 123         struct lttng_consumer_stream *stream;
 124
 125         DBG("Kernel consumer snapshot channel %" PRIu64, key);
 126
 127         rcu_read_lock();
 128
 129         channel = consumer_find_channel(key);
 130         if (!channel) {
 131                 ERR("No channel found for key %" PRIu64, key);
 132                 ret = -1;
 133                 goto end;
 134         }
 135
 136         /* Splice is not supported yet for channel snapshot. */
 137         if (channel->output != CONSUMER_CHANNEL_MMAP) {
 138                 ERR("Unsupported output %d", channel->output);
 139                 ret = -1;
 140                 goto end;
 141         }
 142
 143         cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
 144
 145                 health_code_update();
 146
 147                 /*
 148                  * Lock stream because we are about to change its state.
 149                  */
 150                 pthread_mutex_lock(&stream->lock);
 151
 152                 /*
 153                  * Assign the received relayd ID so we can use it for streaming. The streams
 154                  * are not visible to anyone so this is OK to change it.
 155                  */
 156                 stream->net_seq_idx = relayd_id;
 157                 channel->relayd_id = relayd_id;
 158                 if (relayd_id != (uint64_t) -1ULL) {
 159                         ret = consumer_send_relayd_stream(stream, path);
 160                         if (ret < 0) {
 161                                 ERR("sending stream to relayd");
 162                                 goto end_unlock;
 163                         }
 164                 } else {
 165                         ret = utils_create_stream_file(path, stream->name,
 166                                         stream->chan->tracefile_size,
 167                                         stream->tracefile_count_current,
 168                                         stream->uid, stream->gid, NULL);
 169                         if (ret < 0) {
 170                                 ERR("utils_create_stream_file");
 171                                 goto end_unlock;
 172                         }
 173
 174                         stream->out_fd = ret;
 175                         stream->tracefile_size_current = 0;
 176
 177                         DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
 178                                         path, stream->name, stream->key);
 179                 }
 180                 if (relayd_id != -1ULL) {
 181                         ret = consumer_send_relayd_streams_sent(relayd_id);
 182                         if (ret < 0) {
 183                                 ERR("sending streams sent to relayd");
 184                                 goto end_unlock;
 185                         }
 186                 }
 187
 188                 ret = kernctl_buffer_flush(stream->wait_fd);
 189                 if (ret < 0) {
 190                         ERR("Failed to flush kernel stream");
 191                         ret = -errno;
 192                         goto end_unlock;
 193                 }
 194
 195                 ret = lttng_kconsumer_take_snapshot(stream);
 196                 if (ret < 0) {
 197                         ERR("Taking kernel snapshot");
 198                         goto end_unlock;
 199                 }
 200
 201                 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
 202                 if (ret < 0) {
 203                         ERR("Produced kernel snapshot position");
 204                         goto end_unlock;
 205                 }
 206
 207                 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
 208                 if (ret < 0) {
 209                         ERR("Consumerd kernel snapshot position");
 210                         goto end_unlock;
 211                 }
 212
 213                 if (stream->max_sb_size == 0) {
 214                         ret = kernctl_get_max_subbuf_size(stream->wait_fd,
 215                                         &stream->max_sb_size);
 216                         if (ret < 0) {
 217                                 ERR("Getting kernel max_sb_size");
 218                                 ret = -errno;
 219                                 goto end_unlock;
 220                         }
 221                 }
 222
 223                 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
 224                                 produced_pos, nb_packets_per_stream,
 225                                 stream->max_sb_size);
 226
 227                 while (consumed_pos < produced_pos) {
 228                         ssize_t read_len;
 229                         unsigned long len, padded_len;
 230
 231                         health_code_update();
 232
 233                         DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
 234
 235                         ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
 236                         if (ret < 0) {
 237                                 if (errno != EAGAIN) {
 238                                         PERROR("kernctl_get_subbuf snapshot");
 239                                         ret = -errno;
 240                                         goto end_unlock;
 241                                 }
 242                                 DBG("Kernel consumer get subbuf failed. Skipping it.");
 243                                 consumed_pos += stream->max_sb_size;
 244                                 continue;
 245                         }
 246
 247                         ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
 248                         if (ret < 0) {
 249                                 ERR("Snapshot kernctl_get_subbuf_size");
 250                                 ret = -errno;
 251                                 goto error_put_subbuf;
 252                         }
 253
 254                         ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
 255                         if (ret < 0) {
 256                                 ERR("Snapshot kernctl_get_padded_subbuf_size");
 257                                 ret = -errno;
 258                                 goto error_put_subbuf;
 259                         }
 260
 261                         read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
 262                                         padded_len - len, NULL);
 263                         /*
 264                          * We write the padded len in local tracefiles but the data len
 265                          * when using a relay. Display the error but continue processing
 266                          * to try to release the subbuffer.
 267                          */
 268                         if (relayd_id != (uint64_t) -1ULL) {
 269                                 if (read_len != len) {
 270                                         ERR("Error sending to the relay (ret: %zd != len: %lu)",
 271                                                         read_len, len);
 272                                 }
 273                         } else {
 274                                 if (read_len != padded_len) {
 275                                         ERR("Error writing to tracefile (ret: %zd != len: %lu)",
 276                                                         read_len, padded_len);
 277                                 }
 278                         }
 279
 280                         ret = kernctl_put_subbuf(stream->wait_fd);
 281                         if (ret < 0) {
 282                                 ERR("Snapshot kernctl_put_subbuf");
 283                                 ret = -errno;
 284                                 goto end_unlock;
 285                         }
 286                         consumed_pos += stream->max_sb_size;
 287                 }
 288
 289                 if (relayd_id == (uint64_t) -1ULL) {
 290                         if (stream->out_fd >= 0) {
 291                                 ret = close(stream->out_fd);
 292                                 if (ret < 0) {
 293                                         PERROR("Kernel consumer snapshot close out_fd");
 294                                         goto end_unlock;
 295                                 }
 296                                 stream->out_fd = -1;
 297                         }
 298                 } else {
 299                         close_relayd_stream(stream);
 300                         stream->net_seq_idx = (uint64_t) -1ULL;
 301                 }
 302                 pthread_mutex_unlock(&stream->lock);
 303         }
 304
 305         /* All good! */
 306         ret = 0;
 307         goto end;
 308
 309 error_put_subbuf:
 310         ret = kernctl_put_subbuf(stream->wait_fd);
 311         if (ret < 0) {
 312                 ret = -errno;
 313                 ERR("Snapshot kernctl_put_subbuf error path");
 314         }
 315 end_unlock:
 316         pthread_mutex_unlock(&stream->lock);
 317 end:
 318         rcu_read_unlock();
 319         return ret;
 320 }
 321
 322 /*
 323  * Read the whole metadata available for a snapshot.
 324  *
 325  * Returns 0 on success, < 0 on error
 326  */
 327 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
 328                 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
 329 {
 330         int ret, use_relayd = 0;
 331         ssize_t ret_read;
 332         struct lttng_consumer_channel *metadata_channel;
 333         struct lttng_consumer_stream *metadata_stream;
 334
 335         assert(ctx);
 336
 337         DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
 338                         key, path);
 339
 340         rcu_read_lock();
 341
 342         metadata_channel = consumer_find_channel(key);
 343         if (!metadata_channel) {
 344                 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
 345                 ret = -1;
 346                 goto error;
 347         }
 348
 349         metadata_stream = metadata_channel->metadata_stream;
 350         assert(metadata_stream);
 351
 352         /* Flag once that we have a valid relayd for the stream. */
 353         if (relayd_id != (uint64_t) -1ULL) {
 354                 use_relayd = 1;
 355         }
 356
 357         if (use_relayd) {
 358                 ret = consumer_send_relayd_stream(metadata_stream, path);
 359                 if (ret < 0) {
 360                         goto error;
 361                 }
 362         } else {
 363                 ret = utils_create_stream_file(path, metadata_stream->name,
 364                                 metadata_stream->chan->tracefile_size,
 365                                 metadata_stream->tracefile_count_current,
 366                                 metadata_stream->uid, metadata_stream->gid, NULL);
 367                 if (ret < 0) {
 368                         goto error;
 369                 }
 370                 metadata_stream->out_fd = ret;
 371         }
 372
 373         do {
 374                 health_code_update();
 375
 376                 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
 377                 if (ret_read < 0) {
 378                         if (ret_read != -EAGAIN) {
 379                                 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
 380                                                 ret_read);
 381                                 goto error;
 382                         }
 383                         /* ret_read is negative at this point so we will exit the loop. */
 384                         continue;
 385                 }
 386         } while (ret_read >= 0);
 387
 388         if (use_relayd) {
 389                 close_relayd_stream(metadata_stream);
 390                 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
 391         } else {
 392                 if (metadata_stream->out_fd >= 0) {
 393                         ret = close(metadata_stream->out_fd);
 394                         if (ret < 0) {
 395                                 PERROR("Kernel consumer snapshot metadata close out_fd");
 396                                 /*
 397                                  * Don't go on error here since the snapshot was successful at this
 398                                  * point but somehow the close failed.
 399                                  */
 400                         }
 401                         metadata_stream->out_fd = -1;
 402                 }
 403         }
 404
 405         ret = 0;
 406
 407         cds_list_del(&metadata_stream->send_node);
 408         consumer_stream_destroy(metadata_stream, NULL);
 409         metadata_channel->metadata_stream = NULL;
 410 error:
 411         rcu_read_unlock();
 412         return ret;
 413 }
 414
 415 /*
 416  * Receive command from session daemon and process it.
 417  *
 418  * Return 1 on success else a negative value or 0.
 419  */
 420 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
 421                 int sock, struct pollfd *consumer_sockpoll)
 422 {
 423         ssize_t ret;
 424         enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
 425         struct lttcomm_consumer_msg msg;
 426
 427         health_code_update();
 428
 429         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
 430         if (ret != sizeof(msg)) {
 431                 if (ret > 0) {
 432                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 433                         ret = -1;
 434                 }
 435                 return ret;
 436         }
 437
 438         health_code_update();
 439
 440         /* Deprecated command */
 441         assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
 442
 443         health_code_update();
 444
 445         /* relayd needs RCU read-side protection */
 446         rcu_read_lock();
 447
 448         switch (msg.cmd_type) {
 449         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 450         {
 451                 /* Session daemon status message are handled in the following call. */
 452                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 453                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 454                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
 455                                  msg.u.relayd_sock.relayd_session_id);
 456                 goto end_nosignal;
 457         }
 458         case LTTNG_CONSUMER_ADD_CHANNEL:
 459         {
 460                 struct lttng_consumer_channel *new_channel;
 461                 int ret_recv;
 462
 463                 health_code_update();
 464
 465                 /* First send a status message before receiving the fds. */
 466                 ret = consumer_send_status_msg(sock, ret_code);
 467                 if (ret < 0) {
 468                         /* Somehow, the session daemon is not responding anymore. */
 469                         goto error_fatal;
 470                 }
 471
 472                 health_code_update();
 473
 474                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 475                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 476                                 msg.u.channel.session_id, msg.u.channel.pathname,
 477                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 478                                 msg.u.channel.relayd_id, msg.u.channel.output,
 479                                 msg.u.channel.tracefile_size,
 480                                 msg.u.channel.tracefile_count, 0,
 481                                 msg.u.channel.monitor,
 482                                 msg.u.channel.live_timer_interval,
 483                                 NULL, NULL);
 484                 if (new_channel == NULL) {
 485                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 486                         goto end_nosignal;
 487                 }
 488                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 489                 switch (msg.u.channel.output) {
 490                 case LTTNG_EVENT_SPLICE:
 491                         new_channel->output = CONSUMER_CHANNEL_SPLICE;
 492                         break;
 493                 case LTTNG_EVENT_MMAP:
 494                         new_channel->output = CONSUMER_CHANNEL_MMAP;
 495                         break;
 496                 default:
 497                         ERR("Channel output unknown %d", msg.u.channel.output);
 498                         goto end_nosignal;
 499                 }
 500
 501                 /* Translate and save channel type. */
 502                 switch (msg.u.channel.type) {
 503                 case CONSUMER_CHANNEL_TYPE_DATA:
 504                 case CONSUMER_CHANNEL_TYPE_METADATA:
 505                         new_channel->type = msg.u.channel.type;
 506                         break;
 507                 default:
 508                         assert(0);
 509                         goto end_nosignal;
 510                 };
 511
 512                 health_code_update();
 513
 514                 if (ctx->on_recv_channel != NULL) {
 515                         ret_recv = ctx->on_recv_channel(new_channel);
 516                         if (ret_recv == 0) {
 517                                 ret = consumer_add_channel(new_channel, ctx);
 518                         } else if (ret_recv < 0) {
 519                                 goto end_nosignal;
 520                         }
 521                 } else {
 522                         ret = consumer_add_channel(new_channel, ctx);
 523                 }
 524                 if (CONSUMER_CHANNEL_TYPE_DATA) {
 525                         consumer_timer_live_start(new_channel,
 526                                         msg.u.channel.live_timer_interval);
 527                 }
 528
 529                 health_code_update();
 530
 531                 /* If we received an error in add_channel, we need to report it. */
 532                 if (ret < 0) {
 533                         ret = consumer_send_status_msg(sock, ret);
 534                         if (ret < 0) {
 535                                 goto error_fatal;
 536                         }
 537                         goto end_nosignal;
 538                 }
 539
 540                 goto end_nosignal;
 541         }
 542         case LTTNG_CONSUMER_ADD_STREAM:
 543         {
 544                 int fd;
 545                 struct lttng_pipe *stream_pipe;
 546                 struct lttng_consumer_stream *new_stream;
 547                 struct lttng_consumer_channel *channel;
 548                 int alloc_ret = 0;
 549
 550                 /*
 551                  * Get stream's channel reference. Needed when adding the stream to the
 552                  * global hash table.
 553                  */
 554                 channel = consumer_find_channel(msg.u.stream.channel_key);
 555                 if (!channel) {
 556                         /*
 557                          * We could not find the channel. Can happen if cpu hotplug
 558                          * happens while tearing down.
 559                          */
 560                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 561                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 562                 }
 563
 564                 health_code_update();
 565
 566                 /* First send a status message before receiving the fds. */
 567                 ret = consumer_send_status_msg(sock, ret_code);
 568                 if (ret < 0) {
 569                         /* Somehow, the session daemon is not responding anymore. */
 570                         goto error_fatal;
 571                 }
 572
 573                 health_code_update();
 574
 575                 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 576                         /* Channel was not found. */
 577                         goto end_nosignal;
 578                 }
 579
 580                 /* Blocking call */
 581                 health_poll_entry();
 582                 ret = lttng_consumer_poll_socket(consumer_sockpoll);
 583                 health_poll_exit();
 584                 if (ret) {
 585                         goto error_fatal;
 586                 }
 587
 588                 health_code_update();
 589
 590                 /* Get stream file descriptor from socket */
 591                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 592                 if (ret != sizeof(fd)) {
 593                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 594                         rcu_read_unlock();
 595                         return ret;
 596                 }
 597
 598                 health_code_update();
 599
 600                 /*
 601                  * Send status code to session daemon only if the recv works. If the
 602                  * above recv() failed, the session daemon is notified through the
 603                  * error socket and the teardown is eventually done.
 604                  */
 605                 ret = consumer_send_status_msg(sock, ret_code);
 606                 if (ret < 0) {
 607                         /* Somehow, the session daemon is not responding anymore. */
 608                         goto end_nosignal;
 609                 }
 610
 611                 health_code_update();
 612
 613                 new_stream = consumer_allocate_stream(channel->key,
 614                                 fd,
 615                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 616                                 channel->name,
 617                                 channel->uid,
 618                                 channel->gid,
 619                                 channel->relayd_id,
 620                                 channel->session_id,
 621                                 msg.u.stream.cpu,
 622                                 &alloc_ret,
 623                                 channel->type,
 624                                 channel->monitor);
 625                 if (new_stream == NULL) {
 626                         switch (alloc_ret) {
 627                         case -ENOMEM:
 628                         case -EINVAL:
 629                         default:
 630                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 631                                 break;
 632                         }
 633                         goto end_nosignal;
 634                 }
 635
 636                 new_stream->chan = channel;
 637                 new_stream->wait_fd = fd;
 638                 switch (channel->output) {
 639                 case CONSUMER_CHANNEL_SPLICE:
 640                         new_stream->output = LTTNG_EVENT_SPLICE;
 641                         ret = utils_create_pipe(new_stream->splice_pipe);
 642                         if (ret < 0) {
 643                                 goto end_nosignal;
 644                         }
 645                         break;
 646                 case CONSUMER_CHANNEL_MMAP:
 647                         new_stream->output = LTTNG_EVENT_MMAP;
 648                         break;
 649                 default:
 650                         ERR("Stream output unknown %d", channel->output);
 651                         goto end_nosignal;
 652                 }
 653
 654                 /*
 655                  * We've just assigned the channel to the stream so increment the
 656                  * refcount right now. We don't need to increment the refcount for
 657                  * streams in no monitor because we handle manually the cleanup of
 658                  * those. It is very important to make sure there is NO prior
 659                  * consumer_del_stream() calls or else the refcount will be unbalanced.
 660                  */
 661                 if (channel->monitor) {
 662                         uatomic_inc(&new_stream->chan->refcount);
 663                 }
 664
 665                 /*
 666                  * The buffer flush is done on the session daemon side for the kernel
 667                  * so no need for the stream "hangup_flush_done" variable to be
 668                  * tracked. This is important for a kernel stream since we don't rely
 669                  * on the flush state of the stream to read data. It's not the case for
 670                  * user space tracing.
 671                  */
 672                 new_stream->hangup_flush_done = 0;
 673
 674                 health_code_update();
 675
 676                 if (ctx->on_recv_stream) {
 677                         ret = ctx->on_recv_stream(new_stream);
 678                         if (ret < 0) {
 679                                 consumer_stream_free(new_stream);
 680                                 goto end_nosignal;
 681                         }
 682                 }
 683
 684                 health_code_update();
 685
 686                 if (new_stream->metadata_flag) {
 687                         channel->metadata_stream = new_stream;
 688                 }
 689
 690                 /* Do not monitor this stream. */
 691                 if (!channel->monitor) {
 692                         DBG("Kernel consumer add stream %s in no monitor mode with "
 693                                         "relayd id %" PRIu64, new_stream->name,
 694                                         new_stream->net_seq_idx);
 695                         cds_list_add(&new_stream->send_node, &channel->streams.head);
 696                         break;
 697                 }
 698
 699                 /* Send stream to relayd if the stream has an ID. */
 700                 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 701                         ret = consumer_send_relayd_stream(new_stream,
 702                                         new_stream->chan->pathname);
 703                         if (ret < 0) {
 704                                 consumer_stream_free(new_stream);
 705                                 goto end_nosignal;
 706                         }
 707                 }
 708
 709                 /* Get the right pipe where the stream will be sent. */
 710                 if (new_stream->metadata_flag) {
 711                         ret = consumer_add_metadata_stream(new_stream);
 712                         if (ret) {
 713                                 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
 714                                                 new_stream->key);
 715                                 consumer_stream_free(new_stream);
 716                                 goto end_nosignal;
 717                         }
 718                         stream_pipe = ctx->consumer_metadata_pipe;
 719                 } else {
 720                         ret = consumer_add_data_stream(new_stream);
 721                         if (ret) {
 722                                 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
 723                                                 new_stream->key);
 724                                 consumer_stream_free(new_stream);
 725                                 goto end_nosignal;
 726                         }
 727                         stream_pipe = ctx->consumer_data_pipe;
 728                 }
 729
 730                 /* Vitible to other threads */
 731                 new_stream->globally_visible = 1;
 732
 733                 health_code_update();
 734
 735                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 736                 if (ret < 0) {
 737                         ERR("Consumer write %s stream to pipe %d",
 738                                         new_stream->metadata_flag ? "metadata" : "data",
 739                                         lttng_pipe_get_writefd(stream_pipe));
 740                         if (new_stream->metadata_flag) {
 741                                 consumer_del_stream_for_metadata(new_stream);
 742                         } else {
 743                                 consumer_del_stream_for_data(new_stream);
 744                         }
 745                         goto end_nosignal;
 746                 }
 747
 748                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 749                                 new_stream->name, fd, new_stream->relayd_stream_id);
 750                 break;
 751         }
 752         case LTTNG_CONSUMER_STREAMS_SENT:
 753         {
 754                 struct lttng_consumer_channel *channel;
 755
 756                 /*
 757                  * Get stream's channel reference. Needed when adding the stream to the
 758                  * global hash table.
 759                  */
 760                 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
 761                 if (!channel) {
 762                         /*
 763                          * We could not find the channel. Can happen if cpu hotplug
 764                          * happens while tearing down.
 765                          */
 766                         ERR("Unable to find channel key %" PRIu64,
 767                                         msg.u.sent_streams.channel_key);
 768                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 769                 }
 770
 771                 health_code_update();
 772
 773                 /*
 774                  * Send status code to session daemon.
 775                  */
 776                 ret = consumer_send_status_msg(sock, ret_code);
 777                 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
 778                         /* Somehow, the session daemon is not responding anymore. */
 779                         goto end_nosignal;
 780                 }
 781
 782                 health_code_update();
 783
 784                 /*
 785                  * We should not send this message if we don't monitor the
 786                  * streams in this channel.
 787                  */
 788                 if (!channel->monitor) {
 789                         break;
 790                 }
 791
 792                 health_code_update();
 793                 /* Send stream to relayd if the stream has an ID. */
 794                 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
 795                         ret = consumer_send_relayd_streams_sent(
 796                                         msg.u.sent_streams.net_seq_idx);
 797                         if (ret < 0) {
 798                                 goto end_nosignal;
 799                         }
 800                 }
 801                 break;
 802         }
 803         case LTTNG_CONSUMER_UPDATE_STREAM:
 804         {
 805                 rcu_read_unlock();
 806                 return -ENOSYS;
 807         }
 808         case LTTNG_CONSUMER_DESTROY_RELAYD:
 809         {
 810                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 811                 struct consumer_relayd_sock_pair *relayd;
 812
 813                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 814
 815                 /* Get relayd reference if exists. */
 816                 relayd = consumer_find_relayd(index);
 817                 if (relayd == NULL) {
 818                         DBG("Unable to find relayd %" PRIu64, index);
 819                         ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
 820                 }
 821
 822                 /*
 823                  * Each relayd socket pair has a refcount of stream attached to it
 824                  * which tells if the relayd is still active or not depending on the
 825                  * refcount value.
 826                  *
 827                  * This will set the destroy flag of the relayd object and destroy it
 828                  * if the refcount reaches zero when called.
 829                  *
 830                  * The destroy can happen either here or when a stream fd hangs up.
 831                  */
 832                 if (relayd) {
 833                         consumer_flag_relayd_for_destroy(relayd);
 834                 }
 835
 836                 health_code_update();
 837
 838                 ret = consumer_send_status_msg(sock, ret_code);
 839                 if (ret < 0) {
 840                         /* Somehow, the session daemon is not responding anymore. */
 841                         goto error_fatal;
 842                 }
 843
 844                 goto end_nosignal;
 845         }
 846         case LTTNG_CONSUMER_DATA_PENDING:
 847         {
 848                 int32_t ret;
 849                 uint64_t id = msg.u.data_pending.session_id;
 850
 851                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 852
 853                 ret = consumer_data_pending(id);
 854
 855                 health_code_update();
 856
 857                 /* Send back returned value to session daemon */
 858                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 859                 if (ret < 0) {
 860                         PERROR("send data pending ret code");
 861                         goto error_fatal;
 862                 }
 863
 864                 /*
 865                  * No need to send back a status message since the data pending
 866                  * returned value is the response.
 867                  */
 868                 break;
 869         }
 870         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 871         {
 872                 if (msg.u.snapshot_channel.metadata == 1) {
 873                         ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
 874                                         msg.u.snapshot_channel.pathname,
 875                                         msg.u.snapshot_channel.relayd_id, ctx);
 876                         if (ret < 0) {
 877                                 ERR("Snapshot metadata failed");
 878                                 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
 879                         }
 880                 } else {
 881                         ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
 882                                         msg.u.snapshot_channel.pathname,
 883                                         msg.u.snapshot_channel.relayd_id,
 884                                         msg.u.snapshot_channel.nb_packets_per_stream,
 885                                         ctx);
 886                         if (ret < 0) {
 887                                 ERR("Snapshot channel failed");
 888                                 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 889                         }
 890                 }
 891
 892                 health_code_update();
 893
 894                 ret = consumer_send_status_msg(sock, ret_code);
 895                 if (ret < 0) {
 896                         /* Somehow, the session daemon is not responding anymore. */
 897                         goto end_nosignal;
 898                 }
 899                 break;
 900         }
 901         case LTTNG_CONSUMER_DESTROY_CHANNEL:
 902         {
 903                 uint64_t key = msg.u.destroy_channel.key;
 904                 struct lttng_consumer_channel *channel;
 905
 906                 channel = consumer_find_channel(key);
 907                 if (!channel) {
 908                         ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
 909                         ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
 910                 }
 911
 912                 health_code_update();
 913
 914                 ret = consumer_send_status_msg(sock, ret_code);
 915                 if (ret < 0) {
 916                         /* Somehow, the session daemon is not responding anymore. */
 917                         goto end_nosignal;
 918                 }
 919
 920                 health_code_update();
 921
 922                 /* Stop right now if no channel was found. */
 923                 if (!channel) {
 924                         goto end_nosignal;
 925                 }
 926
 927                 /*
 928                  * This command should ONLY be issued for channel with streams set in
 929                  * no monitor mode.
 930                  */
 931                 assert(!channel->monitor);
 932
 933                 /*
 934                  * The refcount should ALWAYS be 0 in the case of a channel in no
 935                  * monitor mode.
 936                  */
 937                 assert(!uatomic_sub_return(&channel->refcount, 1));
 938
 939                 consumer_del_channel(channel);
 940
 941                 goto end_nosignal;
 942         }
 943         default:
 944                 goto end_nosignal;
 945         }
 946
 947 end_nosignal:
 948         rcu_read_unlock();
 949
 950         /*
 951          * Return 1 to indicate success since the 0 value can be a socket
 952          * shutdown during the recv() or send() call.
 953          */
 954         health_code_update();
 955         return 1;
 956
 957 error_fatal:
 958         rcu_read_unlock();
 959         /* This will issue a consumer stop. */
 960         return -1;
 961 }
 962
 963 /*
 964  * Populate index values of a kernel stream. Values are set in big endian order.
 965  *
 966  * Return 0 on success or else a negative value.
 967  */
 968 static int get_index_values(struct ctf_packet_index *index, int infd)
 969 {
 970         int ret;
 971
 972         ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
 973         if (ret < 0) {
 974                 PERROR("kernctl_get_timestamp_begin");
 975                 goto error;
 976         }
 977         index->timestamp_begin = htobe64(index->timestamp_begin);
 978
 979         ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
 980         if (ret < 0) {
 981                 PERROR("kernctl_get_timestamp_end");
 982                 goto error;
 983         }
 984         index->timestamp_end = htobe64(index->timestamp_end);
 985
 986         ret = kernctl_get_events_discarded(infd, &index->events_discarded);
 987         if (ret < 0) {
 988                 PERROR("kernctl_get_events_discarded");
 989                 goto error;
 990         }
 991         index->events_discarded = htobe64(index->events_discarded);
 992
 993         ret = kernctl_get_content_size(infd, &index->content_size);
 994         if (ret < 0) {
 995                 PERROR("kernctl_get_content_size");
 996                 goto error;
 997         }
 998         index->content_size = htobe64(index->content_size);
 999
1000         ret = kernctl_get_packet_size(infd, &index->packet_size);
1001         if (ret < 0) {
1002                 PERROR("kernctl_get_packet_size");
1003                 goto error;
1004         }
1005         index->packet_size = htobe64(index->packet_size);
1006
1007         ret = kernctl_get_stream_id(infd, &index->stream_id);
1008         if (ret < 0) {
1009                 PERROR("kernctl_get_stream_id");
1010                 goto error;
1011         }
1012         index->stream_id = htobe64(index->stream_id);
1013
1014 error:
1015         return ret;
1016 }
1017 /*
1018  * Sync metadata meaning request them to the session daemon and snapshot to the
1019  * metadata thread can consumer them.
1020  *
1021  * Metadata stream lock MUST be acquired.
1022  *
1023  * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1024  * is empty or a negative value on error.
1025  */
1026 int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1027 {
1028         int ret;
1029
1030         assert(metadata);
1031
1032         ret = kernctl_buffer_flush(metadata->wait_fd);
1033         if (ret < 0) {
1034                 ERR("Failed to flush kernel stream");
1035                 goto end;
1036         }
1037
1038         ret = kernctl_snapshot(metadata->wait_fd);
1039         if (ret < 0) {
1040                 if (errno != EAGAIN) {
1041                         ERR("Sync metadata, taking kernel snapshot failed.");
1042                         goto end;
1043                 }
1044                 DBG("Sync metadata, no new kernel metadata");
1045                 /* No new metadata, exit. */
1046                 ret = ENODATA;
1047                 goto end;
1048         }
1049
1050 end:
1051         return ret;
1052 }
1053
1054 /*
1055  * Consume data on a file descriptor and write it on a trace file.
1056  */
1057 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
1058                 struct lttng_consumer_local_data *ctx)
1059 {
1060         unsigned long len, subbuf_size, padding;
1061         int err, write_index = 1;
1062         ssize_t ret = 0;
1063         int infd = stream->wait_fd;
1064         struct ctf_packet_index index;
1065
1066         DBG("In read_subbuffer (infd : %d)", infd);
1067
1068         /* Get the next subbuffer */
1069         err = kernctl_get_next_subbuf(infd);
1070         if (err != 0) {
1071                 /*
1072                  * This is a debug message even for single-threaded consumer,
1073                  * because poll() have more relaxed criterions than get subbuf,
1074                  * so get_subbuf may fail for short race windows where poll()
1075                  * would issue wakeups.
1076                  */
1077                 DBG("Reserving sub buffer failed (everything is normal, "
1078                                 "it is due to concurrency)");
1079                 ret = -errno;
1080                 goto end;
1081         }
1082
1083         /* Get the full subbuffer size including padding */
1084         err = kernctl_get_padded_subbuf_size(infd, &len);
1085         if (err != 0) {
1086                 PERROR("Getting sub-buffer len failed.");
1087                 err = kernctl_put_subbuf(infd);
1088                 if (err != 0) {
1089                         if (errno == EFAULT) {
1090                                 PERROR("Error in unreserving sub buffer\n");
1091                         } else if (errno == EIO) {
1092                                 /* Should never happen with newer LTTng versions */
1093                                 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1094                         }
1095                         ret = -errno;
1096                         goto end;
1097                 }
1098                 ret = -errno;
1099                 goto end;
1100         }
1101
1102         if (!stream->metadata_flag) {
1103                 ret = get_index_values(&index, infd);
1104                 if (ret < 0) {
1105                         err = kernctl_put_subbuf(infd);
1106                         if (err != 0) {
1107                                 if (errno == EFAULT) {
1108                                         PERROR("Error in unreserving sub buffer\n");
1109                                 } else if (errno == EIO) {
1110                                         /* Should never happen with newer LTTng versions */
1111                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1112                                 }
1113                                 ret = -errno;
1114                                 goto end;
1115                         }
1116                         goto end;
1117                 }
1118         } else {
1119                 write_index = 0;
1120         }
1121
1122         switch (stream->chan->output) {
1123         case CONSUMER_CHANNEL_SPLICE:
1124                 /*
1125                  * XXX: The lttng-modules splice "actor" does not handle copying
1126                  * partial pages hence only using the subbuffer size without the
1127                  * padding makes the splice fail.
1128                  */
1129                 subbuf_size = len;
1130                 padding = 0;
1131
1132                 /* splice the subbuffer to the tracefile */
1133                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
1134                                 padding, &index);
1135                 /*
1136                  * XXX: Splice does not support network streaming so the return value
1137                  * is simply checked against subbuf_size and not like the mmap() op.
1138                  */
1139                 if (ret != subbuf_size) {
1140                         /*
1141                          * display the error but continue processing to try
1142                          * to release the subbuffer
1143                          */
1144                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1145                                         ret, subbuf_size);
1146                         write_index = 0;
1147                 }
1148                 break;
1149         case CONSUMER_CHANNEL_MMAP:
1150                 /* Get subbuffer size without padding */
1151                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1152                 if (err != 0) {
1153                         PERROR("Getting sub-buffer len failed.");
1154                         err = kernctl_put_subbuf(infd);
1155                         if (err != 0) {
1156                                 if (errno == EFAULT) {
1157                                         PERROR("Error in unreserving sub buffer\n");
1158                                 } else if (errno == EIO) {
1159                                         /* Should never happen with newer LTTng versions */
1160                                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1161                                 }
1162                                 ret = -errno;
1163                                 goto end;
1164                         }
1165                         ret = -errno;
1166                         goto end;
1167                 }
1168
1169                 /* Make sure the tracer is not gone mad on us! */
1170                 assert(len >= subbuf_size);
1171
1172                 padding = len - subbuf_size;
1173
1174                 /* write the subbuffer to the tracefile */
1175                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
1176                                 padding, &index);
1177                 /*
1178                  * The mmap operation should write subbuf_size amount of data when
1179                  * network streaming or the full padding (len) size when we are _not_
1180                  * streaming.
1181                  */
1182                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1183                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1184                         /*
1185                          * Display the error but continue processing to try to release the
1186                          * subbuffer. This is a DBG statement since this is possible to
1187                          * happen without being a critical error.
1188                          */
1189                         DBG("Error writing to tracefile "
1190                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
1191                                         ret, len, subbuf_size);
1192                         write_index = 0;
1193                 }
1194                 break;
1195         default:
1196                 ERR("Unknown output method");
1197                 ret = -EPERM;
1198         }
1199
1200         err = kernctl_put_next_subbuf(infd);
1201         if (err != 0) {
1202                 if (errno == EFAULT) {
1203                         PERROR("Error in unreserving sub buffer\n");
1204                 } else if (errno == EIO) {
1205                         /* Should never happen with newer LTTng versions */
1206                         PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1207                 }
1208                 ret = -errno;
1209                 goto end;
1210         }
1211
1212         /* Write index if needed. */
1213         if (!write_index) {
1214                 goto end;
1215         }
1216
1217         if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1218                 /*
1219                  * In live, block until all the metadata is sent.
1220                  */
1221                 pthread_mutex_lock(&stream->metadata_timer_lock);
1222                 assert(!stream->missed_metadata_flush);
1223                 stream->waiting_on_metadata = true;
1224                 pthread_mutex_unlock(&stream->metadata_timer_lock);
1225
1226                 err = consumer_stream_sync_metadata(ctx, stream->session_id);
1227
1228                 pthread_mutex_lock(&stream->metadata_timer_lock);
1229                 stream->waiting_on_metadata = false;
1230                 if (stream->missed_metadata_flush) {
1231                         stream->missed_metadata_flush = false;
1232                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1233                         (void) consumer_flush_kernel_index(stream);
1234                 } else {
1235                         pthread_mutex_unlock(&stream->metadata_timer_lock);
1236                 }
1237                 if (err < 0) {
1238                         goto end;
1239                 }
1240         }
1241
1242         err = consumer_stream_write_index(stream, &index);
1243         if (err < 0) {
1244                 goto end;
1245         }
1246
1247 end:
1248         return ret;
1249 }
1250
1251 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1252 {
1253         int ret;
1254
1255         assert(stream);
1256
1257         /*
1258          * Don't create anything if this is set for streaming or should not be
1259          * monitored.
1260          */
1261         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
1262                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1263                                 stream->chan->tracefile_size, stream->tracefile_count_current,
1264                                 stream->uid, stream->gid, NULL);
1265                 if (ret < 0) {
1266                         goto error;
1267                 }
1268                 stream->out_fd = ret;
1269                 stream->tracefile_size_current = 0;
1270
1271                 if (!stream->metadata_flag) {
1272                         ret = index_create_file(stream->chan->pathname,
1273                                         stream->name, stream->uid, stream->gid,
1274                                         stream->chan->tracefile_size,
1275                                         stream->tracefile_count_current);
1276                         if (ret < 0) {
1277                                 goto error;
1278                         }
1279                         stream->index_fd = ret;
1280                 }
1281         }
1282
1283         if (stream->output == LTTNG_EVENT_MMAP) {
1284                 /* get the len of the mmap region */
1285                 unsigned long mmap_len;
1286
1287                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1288                 if (ret != 0) {
1289                         PERROR("kernctl_get_mmap_len");
1290                         ret = -errno;
1291                         goto error_close_fd;
1292                 }
1293                 stream->mmap_len = (size_t) mmap_len;
1294
1295                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1296                                 MAP_PRIVATE, stream->wait_fd, 0);
1297                 if (stream->mmap_base == MAP_FAILED) {
1298                         PERROR("Error mmaping");
1299                         ret = -1;
1300                         goto error_close_fd;
1301                 }
1302         }
1303
1304         /* we return 0 to let the library handle the FD internally */
1305         return 0;
1306
1307 error_close_fd:
1308         if (stream->out_fd >= 0) {
1309                 int err;
1310
1311                 err = close(stream->out_fd);
1312                 assert(!err);
1313                 stream->out_fd = -1;
1314         }
1315 error:
1316         return ret;
1317 }
1318
1319 /*
1320  * Check if data is still being extracted from the buffers for a specific
1321  * stream. Consumer data lock MUST be acquired before calling this function
1322  * and the stream lock.
1323  *
1324  * Return 1 if the traced data are still getting read else 0 meaning that the
1325  * data is available for trace viewer reading.
1326  */
1327 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1328 {
1329         int ret;
1330
1331         assert(stream);
1332
1333         if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1334                 ret = 0;
1335                 goto end;
1336         }
1337
1338         ret = kernctl_get_next_subbuf(stream->wait_fd);
1339         if (ret == 0) {
1340                 /* There is still data so let's put back this subbuffer. */
1341                 ret = kernctl_put_subbuf(stream->wait_fd);
1342                 assert(ret == 0);
1343                 ret = 1;   /* Data is pending */
1344                 goto end;
1345         }
1346
1347         /* Data is NOT pending and ready to be read. */
1348         ret = 0;
1349
1350 end:
1351         return ret;
1352 }