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

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