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
  41 #include "kernel-consumer.h"
  42
  43 extern struct lttng_consumer_global_data consumer_data;
  44 extern int consumer_poll_timeout;
  45 extern volatile int consumer_quit;
  46
  47 /*
  48  * Take a snapshot for a specific fd
  49  *
  50  * Returns 0 on success, < 0 on error
  51  */
  52 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  53 {
  54         int ret = 0;
  55         int infd = stream->wait_fd;
  56
  57         ret = kernctl_snapshot(infd);
  58         if (ret != 0) {
  59                 errno = -ret;
  60                 perror("Getting sub-buffer snapshot.");
  61         }
  62
  63         return ret;
  64 }
  65
  66 /*
  67  * Get the produced position
  68  *
  69  * Returns 0 on success, < 0 on error
  70  */
  71 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  72                 unsigned long *pos)
  73 {
  74         int ret;
  75         int infd = stream->wait_fd;
  76
  77         ret = kernctl_snapshot_get_produced(infd, pos);
  78         if (ret != 0) {
  79                 errno = -ret;
  80                 perror("kernctl_snapshot_get_produced");
  81         }
  82
  83         return ret;
  84 }
  85
  86 /*
  87  * Receive command from session daemon and process it.
  88  *
  89  * Return 1 on success else a negative value or 0.
  90  */
  91 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
  92                 int sock, struct pollfd *consumer_sockpoll)
  93 {
  94         ssize_t ret;
  95         enum lttng_error_code ret_code = LTTNG_OK;
  96         struct lttcomm_consumer_msg msg;
  97
  98         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
  99         if (ret != sizeof(msg)) {
 100                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
 101                 if (ret > 0) {
 102                         ret = -1;
 103                 }
 104                 return ret;
 105         }
 106         if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
 107                 /*
 108                  * Notify the session daemon that the command is completed.
 109                  *
 110                  * On transport layer error, the function call will print an error
 111                  * message so handling the returned code is a bit useless since we
 112                  * return an error code anyway.
 113                  */
 114                 (void) consumer_send_status_msg(sock, ret_code);
 115                 return -ENOENT;
 116         }
 117
 118         /* relayd needs RCU read-side protection */
 119         rcu_read_lock();
 120
 121         switch (msg.cmd_type) {
 122         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 123         {
 124                 /* Session daemon status message are handled in the following call. */
 125                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 126                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 127                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
 128                 goto end_nosignal;
 129         }
 130         case LTTNG_CONSUMER_ADD_CHANNEL:
 131         {
 132                 struct lttng_consumer_channel *new_channel;
 133                 int ret_recv;
 134
 135                 /* First send a status message before receiving the fds. */
 136                 ret = consumer_send_status_msg(sock, ret_code);
 137                 if (ret < 0) {
 138                         /* Somehow, the session daemon is not responding anymore. */
 139                         goto error_fatal;
 140                 }
 141                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 142                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 143                                 msg.u.channel.session_id, msg.u.channel.pathname,
 144                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 145                                 msg.u.channel.relayd_id, msg.u.channel.output,
 146                                 msg.u.channel.tracefile_size,
 147                                 msg.u.channel.tracefile_count,
 148                                 msg.u.channel.monitor);
 149                 if (new_channel == NULL) {
 150                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 151                         goto end_nosignal;
 152                 }
 153                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 154
 155                 /* Translate and save channel type. */
 156                 switch (msg.u.channel.type) {
 157                 case CONSUMER_CHANNEL_TYPE_DATA:
 158                 case CONSUMER_CHANNEL_TYPE_METADATA:
 159                         new_channel->type = msg.u.channel.type;
 160                         break;
 161                 default:
 162                         assert(0);
 163                         goto end_nosignal;
 164                 };
 165
 166                 if (ctx->on_recv_channel != NULL) {
 167                         ret_recv = ctx->on_recv_channel(new_channel);
 168                         if (ret_recv == 0) {
 169                                 ret = consumer_add_channel(new_channel, ctx);
 170                         } else if (ret_recv < 0) {
 171                                 goto end_nosignal;
 172                         }
 173                 } else {
 174                         ret = consumer_add_channel(new_channel, ctx);
 175                 }
 176
 177                 /* If we received an error in add_channel, we need to report it. */
 178                 if (ret < 0) {
 179                         ret = consumer_send_status_msg(sock, ret);
 180                         if (ret < 0) {
 181                                 goto error_fatal;
 182                         }
 183                         goto end_nosignal;
 184                 }
 185
 186                 goto end_nosignal;
 187         }
 188         case LTTNG_CONSUMER_ADD_STREAM:
 189         {
 190                 int fd;
 191                 struct lttng_pipe *stream_pipe;
 192                 struct consumer_relayd_sock_pair *relayd = NULL;
 193                 struct lttng_consumer_stream *new_stream;
 194                 struct lttng_consumer_channel *channel;
 195                 int alloc_ret = 0;
 196
 197                 /*
 198                  * Get stream's channel reference. Needed when adding the stream to the
 199                  * global hash table.
 200                  */
 201                 channel = consumer_find_channel(msg.u.stream.channel_key);
 202                 if (!channel) {
 203                         /*
 204                          * We could not find the channel. Can happen if cpu hotplug
 205                          * happens while tearing down.
 206                          */
 207                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 208                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 209                 }
 210
 211                 /* First send a status message before receiving the fds. */
 212                 ret = consumer_send_status_msg(sock, ret_code);
 213                 if (ret < 0) {
 214                         /*
 215                          * Somehow, the session daemon is not responding
 216                          * anymore.
 217                          */
 218                         goto error_fatal;
 219                 }
 220                 if (ret_code != LTTNG_OK) {
 221                         /*
 222                          * Channel was not found.
 223                          */
 224                         goto end_nosignal;
 225                 }
 226
 227                 /* block */
 228                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 229                         rcu_read_unlock();
 230                         return -EINTR;
 231                 }
 232
 233                 /* Get stream file descriptor from socket */
 234                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 235                 if (ret != sizeof(fd)) {
 236                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 237                         rcu_read_unlock();
 238                         return ret;
 239                 }
 240
 241                 /*
 242                  * Send status code to session daemon only if the recv works. If the
 243                  * above recv() failed, the session daemon is notified through the
 244                  * error socket and the teardown is eventually done.
 245                  */
 246                 ret = consumer_send_status_msg(sock, ret_code);
 247                 if (ret < 0) {
 248                         /* Somehow, the session daemon is not responding anymore. */
 249                         goto end_nosignal;
 250                 }
 251
 252                 new_stream = consumer_allocate_stream(channel->key,
 253                                 fd,
 254                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 255                                 channel->name,
 256                                 channel->uid,
 257                                 channel->gid,
 258                                 channel->relayd_id,
 259                                 channel->session_id,
 260                                 msg.u.stream.cpu,
 261                                 &alloc_ret,
 262                                 channel->type);
 263                 if (new_stream == NULL) {
 264                         switch (alloc_ret) {
 265                         case -ENOMEM:
 266                         case -EINVAL:
 267                         default:
 268                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 269                                 break;
 270                         }
 271                         goto end_nosignal;
 272                 }
 273                 new_stream->chan = channel;
 274                 new_stream->wait_fd = fd;
 275
 276                 /*
 277                  * We've just assigned the channel to the stream so increment the
 278                  * refcount right now.
 279                  */
 280                 uatomic_inc(&new_stream->chan->refcount);
 281
 282                 /*
 283                  * The buffer flush is done on the session daemon side for the kernel
 284                  * so no need for the stream "hangup_flush_done" variable to be
 285                  * tracked. This is important for a kernel stream since we don't rely
 286                  * on the flush state of the stream to read data. It's not the case for
 287                  * user space tracing.
 288                  */
 289                 new_stream->hangup_flush_done = 0;
 290
 291                 /* The stream is not metadata. Get relayd reference if exists. */
 292                 relayd = consumer_find_relayd(new_stream->net_seq_idx);
 293                 if (relayd != NULL) {
 294                         /* Add stream on the relayd */
 295                         pthread_mutex_lock(&relayd->ctrl_sock_mutex);
 296                         ret = relayd_add_stream(&relayd->control_sock,
 297                                         new_stream->name, new_stream->chan->pathname,
 298                                         &new_stream->relayd_stream_id,
 299                                         new_stream->chan->tracefile_size,
 300                                         new_stream->chan->tracefile_count);
 301                         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 302                         if (ret < 0) {
 303                                 consumer_del_stream(new_stream, NULL);
 304                                 goto end_nosignal;
 305                         }
 306                 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 307                         ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
 308                                         new_stream->net_seq_idx);
 309                         consumer_del_stream(new_stream, NULL);
 310                         goto end_nosignal;
 311                 }
 312
 313                 if (ctx->on_recv_stream) {
 314                         ret = ctx->on_recv_stream(new_stream);
 315                         if (ret < 0) {
 316                                 consumer_del_stream(new_stream, NULL);
 317                                 goto end_nosignal;
 318                         }
 319                 }
 320
 321                 /* Do not monitor this stream. */
 322                 if (!channel->monitor) {
 323                         DBG("Kernel consumer add stream %s in no monitor mode with"
 324                                         "relayd id %" PRIu64, new_stream->name,
 325                                         new_stream->relayd_stream_id);
 326                         break;
 327                 }
 328
 329                 /* Get the right pipe where the stream will be sent. */
 330                 if (new_stream->metadata_flag) {
 331                         stream_pipe = ctx->consumer_metadata_pipe;
 332                 } else {
 333                         stream_pipe = ctx->consumer_data_pipe;
 334                 }
 335
 336                 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
 337                 if (ret < 0) {
 338                         ERR("Consumer write %s stream to pipe %d",
 339                                         new_stream->metadata_flag ? "metadata" : "data",
 340                                         lttng_pipe_get_writefd(stream_pipe));
 341                         consumer_del_stream(new_stream, NULL);
 342                         goto end_nosignal;
 343                 }
 344
 345                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 346                                 new_stream->name, fd, new_stream->relayd_stream_id);
 347                 break;
 348         }
 349         case LTTNG_CONSUMER_UPDATE_STREAM:
 350         {
 351                 rcu_read_unlock();
 352                 return -ENOSYS;
 353         }
 354         case LTTNG_CONSUMER_DESTROY_RELAYD:
 355         {
 356                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 357                 struct consumer_relayd_sock_pair *relayd;
 358
 359                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 360
 361                 /* Get relayd reference if exists. */
 362                 relayd = consumer_find_relayd(index);
 363                 if (relayd == NULL) {
 364                         DBG("Unable to find relayd %" PRIu64, index);
 365                         ret_code = LTTNG_ERR_NO_CONSUMER;
 366                 }
 367
 368                 /*
 369                  * Each relayd socket pair has a refcount of stream attached to it
 370                  * which tells if the relayd is still active or not depending on the
 371                  * refcount value.
 372                  *
 373                  * This will set the destroy flag of the relayd object and destroy it
 374                  * if the refcount reaches zero when called.
 375                  *
 376                  * The destroy can happen either here or when a stream fd hangs up.
 377                  */
 378                 if (relayd) {
 379                         consumer_flag_relayd_for_destroy(relayd);
 380                 }
 381
 382                 ret = consumer_send_status_msg(sock, ret_code);
 383                 if (ret < 0) {
 384                         /* Somehow, the session daemon is not responding anymore. */
 385                         goto error_fatal;
 386                 }
 387
 388                 goto end_nosignal;
 389         }
 390         case LTTNG_CONSUMER_DATA_PENDING:
 391         {
 392                 int32_t ret;
 393                 uint64_t id = msg.u.data_pending.session_id;
 394
 395                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 396
 397                 ret = consumer_data_pending(id);
 398
 399                 /* Send back returned value to session daemon */
 400                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 401                 if (ret < 0) {
 402                         PERROR("send data pending ret code");
 403                         goto error_fatal;
 404                 }
 405
 406                 /*
 407                  * No need to send back a status message since the data pending
 408                  * returned value is the response.
 409                  */
 410                 break;
 411         }
 412         case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
 413         {
 414                 ret = consumer_send_status_msg(sock, ret_code);
 415                 if (ret < 0) {
 416                         /* Somehow, the session daemon is not responding anymore. */
 417                         goto end_nosignal;
 418                 }
 419                 break;
 420         }
 421         default:
 422                 goto end_nosignal;
 423         }
 424
 425 end_nosignal:
 426         rcu_read_unlock();
 427
 428         /*
 429          * Return 1 to indicate success since the 0 value can be a socket
 430          * shutdown during the recv() or send() call.
 431          */
 432         return 1;
 433
 434 error_fatal:
 435         rcu_read_unlock();
 436         /* This will issue a consumer stop. */
 437         return -1;
 438 }
 439
 440 /*
 441  * Consume data on a file descriptor and write it on a trace file.
 442  */
 443 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 444                 struct lttng_consumer_local_data *ctx)
 445 {
 446         unsigned long len, subbuf_size, padding;
 447         int err;
 448         ssize_t ret = 0;
 449         int infd = stream->wait_fd;
 450
 451         DBG("In read_subbuffer (infd : %d)", infd);
 452         /* Get the next subbuffer */
 453         err = kernctl_get_next_subbuf(infd);
 454         if (err != 0) {
 455                 ret = err;
 456                 /*
 457                  * This is a debug message even for single-threaded consumer,
 458                  * because poll() have more relaxed criterions than get subbuf,
 459                  * so get_subbuf may fail for short race windows where poll()
 460                  * would issue wakeups.
 461                  */
 462                 DBG("Reserving sub buffer failed (everything is normal, "
 463                                 "it is due to concurrency)");
 464                 goto end;
 465         }
 466
 467         /* Get the full subbuffer size including padding */
 468         err = kernctl_get_padded_subbuf_size(infd, &len);
 469         if (err != 0) {
 470                 errno = -err;
 471                 perror("Getting sub-buffer len failed.");
 472                 ret = err;
 473                 goto end;
 474         }
 475
 476         switch (stream->chan->output) {
 477         case LTTNG_EVENT_SPLICE:
 478
 479                 /*
 480                  * XXX: The lttng-modules splice "actor" does not handle copying
 481                  * partial pages hence only using the subbuffer size without the
 482                  * padding makes the splice fail.
 483                  */
 484                 subbuf_size = len;
 485                 padding = 0;
 486
 487                 /* splice the subbuffer to the tracefile */
 488                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 489                                 padding);
 490                 /*
 491                  * XXX: Splice does not support network streaming so the return value
 492                  * is simply checked against subbuf_size and not like the mmap() op.
 493                  */
 494                 if (ret != subbuf_size) {
 495                         /*
 496                          * display the error but continue processing to try
 497                          * to release the subbuffer
 498                          */
 499                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 500                                         ret, subbuf_size);
 501                 }
 502                 break;
 503         case LTTNG_EVENT_MMAP:
 504                 /* Get subbuffer size without padding */
 505                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 506                 if (err != 0) {
 507                         errno = -err;
 508                         perror("Getting sub-buffer len failed.");
 509                         ret = err;
 510                         goto end;
 511                 }
 512
 513                 /* Make sure the tracer is not gone mad on us! */
 514                 assert(len >= subbuf_size);
 515
 516                 padding = len - subbuf_size;
 517
 518                 /* write the subbuffer to the tracefile */
 519                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 520                                 padding);
 521                 /*
 522                  * The mmap operation should write subbuf_size amount of data when
 523                  * network streaming or the full padding (len) size when we are _not_
 524                  * streaming.
 525                  */
 526                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 527                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 528                         /*
 529                          * Display the error but continue processing to try to release the
 530                          * subbuffer
 531                          */
 532                         ERR("Error writing to tracefile "
 533                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 534                                         ret, len, subbuf_size);
 535                 }
 536                 break;
 537         default:
 538                 ERR("Unknown output method");
 539                 ret = -1;
 540         }
 541
 542         err = kernctl_put_next_subbuf(infd);
 543         if (err != 0) {
 544                 errno = -err;
 545                 if (errno == EFAULT) {
 546                         perror("Error in unreserving sub buffer\n");
 547                 } else if (errno == EIO) {
 548                         /* Should never happen with newer LTTng versions */
 549                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 550                 }
 551
 552                 ret = -err;
 553                 goto end;
 554         }
 555
 556 end:
 557         return ret;
 558 }
 559
 560 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 561 {
 562         int ret;
 563
 564         assert(stream);
 565
 566         /*
 567          * Don't create anything if this is set for streaming or should not be
 568          * monitored.
 569          */
 570         if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
 571                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
 572                                 stream->chan->tracefile_size, stream->tracefile_count_current,
 573                                 stream->uid, stream->gid);
 574                 if (ret < 0) {
 575                         goto error;
 576                 }
 577                 stream->out_fd = ret;
 578                 stream->tracefile_size_current = 0;
 579         }
 580
 581         if (stream->output == LTTNG_EVENT_MMAP) {
 582                 /* get the len of the mmap region */
 583                 unsigned long mmap_len;
 584
 585                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 586                 if (ret != 0) {
 587                         errno = -ret;
 588                         PERROR("kernctl_get_mmap_len");
 589                         goto error_close_fd;
 590                 }
 591                 stream->mmap_len = (size_t) mmap_len;
 592
 593                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 594                                 MAP_PRIVATE, stream->wait_fd, 0);
 595                 if (stream->mmap_base == MAP_FAILED) {
 596                         PERROR("Error mmaping");
 597                         ret = -1;
 598                         goto error_close_fd;
 599                 }
 600         }
 601
 602         /* we return 0 to let the library handle the FD internally */
 603         return 0;
 604
 605 error_close_fd:
 606         {
 607                 int err;
 608
 609                 err = close(stream->out_fd);
 610                 assert(!err);
 611         }
 612 error:
 613         return ret;
 614 }
 615
 616 /*
 617  * Check if data is still being extracted from the buffers for a specific
 618  * stream. Consumer data lock MUST be acquired before calling this function
 619  * and the stream lock.
 620  *
 621  * Return 1 if the traced data are still getting read else 0 meaning that the
 622  * data is available for trace viewer reading.
 623  */
 624 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 625 {
 626         int ret;
 627
 628         assert(stream);
 629
 630         ret = kernctl_get_next_subbuf(stream->wait_fd);
 631         if (ret == 0) {
 632                 /* There is still data so let's put back this subbuffer. */
 633                 ret = kernctl_put_subbuf(stream->wait_fd);
 634                 assert(ret == 0);
 635                 ret = 1;   /* Data is pending */
 636                 goto end;
 637         }
 638
 639         /* Data is NOT pending and ready to be read. */
 640         ret = 0;
 641
 642 end:
 643         return ret;
 644 }