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