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

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