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