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/relayd/relayd.h>
  38 #include <common/utils.h>
  39
  40 #include "kernel-consumer.h"
  41
  42 extern struct lttng_consumer_global_data consumer_data;
  43 extern int consumer_poll_timeout;
  44 extern volatile int consumer_quit;
  45
  46 /*
  47  * Take a snapshot for a specific fd
  48  *
  49  * Returns 0 on success, < 0 on error
  50  */
  51 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
  52 {
  53         int ret = 0;
  54         int infd = stream->wait_fd;
  55
  56         ret = kernctl_snapshot(infd);
  57         if (ret != 0) {
  58                 errno = -ret;
  59                 perror("Getting sub-buffer snapshot.");
  60         }
  61
  62         return ret;
  63 }
  64
  65 /*
  66  * Get the produced position
  67  *
  68  * Returns 0 on success, < 0 on error
  69  */
  70 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
  71                 unsigned long *pos)
  72 {
  73         int ret;
  74         int infd = stream->wait_fd;
  75
  76         ret = kernctl_snapshot_get_produced(infd, pos);
  77         if (ret != 0) {
  78                 errno = -ret;
  79                 perror("kernctl_snapshot_get_produced");
  80         }
  81
  82         return ret;
  83 }
  84
  85 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
  86                 int sock, struct pollfd *consumer_sockpoll)
  87 {
  88         ssize_t ret;
  89         enum lttng_error_code ret_code = LTTNG_OK;
  90         struct lttcomm_consumer_msg msg;
  91
  92         ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
  93         if (ret != sizeof(msg)) {
  94                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
  95                 return ret;
  96         }
  97         if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
  98                 /*
  99                  * Notify the session daemon that the command is completed.
 100                  *
 101                  * On transport layer error, the function call will print an error
 102                  * message so handling the returned code is a bit useless since we
 103                  * return an error code anyway.
 104                  */
 105                 (void) consumer_send_status_msg(sock, ret_code);
 106                 return -ENOENT;
 107         }
 108
 109         /* relayd needs RCU read-side protection */
 110         rcu_read_lock();
 111
 112         switch (msg.cmd_type) {
 113         case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
 114         {
 115                 /* Session daemon status message are handled in the following call. */
 116                 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
 117                                 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
 118                                 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
 119                 goto end_nosignal;
 120         }
 121         case LTTNG_CONSUMER_ADD_CHANNEL:
 122         {
 123                 struct lttng_consumer_channel *new_channel;
 124
 125                 /* First send a status message before receiving the fds. */
 126                 ret = consumer_send_status_msg(sock, ret_code);
 127                 if (ret < 0) {
 128                         /* Somehow, the session daemon is not responding anymore. */
 129                         goto end_nosignal;
 130                 }
 131
 132                 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
 133                 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
 134                                 msg.u.channel.session_id, msg.u.channel.pathname,
 135                                 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
 136                                 msg.u.channel.relayd_id, msg.u.channel.output,
 137                                 msg.u.channel.tracefile_size,
 138                                 msg.u.channel.tracefile_count);
 139                 if (new_channel == NULL) {
 140                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 141                         goto end_nosignal;
 142                 }
 143                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 144
 145                 /* Translate and save channel type. */
 146                 switch (msg.u.channel.type) {
 147                 case CONSUMER_CHANNEL_TYPE_DATA:
 148                 case CONSUMER_CHANNEL_TYPE_METADATA:
 149                         new_channel->type = msg.u.channel.type;
 150                         break;
 151                 default:
 152                         assert(0);
 153                         goto end_nosignal;
 154                 };
 155
 156                 if (ctx->on_recv_channel != NULL) {
 157                         ret = ctx->on_recv_channel(new_channel);
 158                         if (ret == 0) {
 159                                 consumer_add_channel(new_channel, ctx);
 160                         } else if (ret < 0) {
 161                                 goto end_nosignal;
 162                         }
 163                 } else {
 164                         consumer_add_channel(new_channel, ctx);
 165                 }
 166                 goto end_nosignal;
 167         }
 168         case LTTNG_CONSUMER_ADD_STREAM:
 169         {
 170                 int fd, stream_pipe;
 171                 struct consumer_relayd_sock_pair *relayd = NULL;
 172                 struct lttng_consumer_stream *new_stream;
 173                 struct lttng_consumer_channel *channel;
 174                 int alloc_ret = 0;
 175
 176                 /*
 177                  * Get stream's channel reference. Needed when adding the stream to the
 178                  * global hash table.
 179                  */
 180                 channel = consumer_find_channel(msg.u.stream.channel_key);
 181                 if (!channel) {
 182                         /*
 183                          * We could not find the channel. Can happen if cpu hotplug
 184                          * happens while tearing down.
 185                          */
 186                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 187                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 188                 }
 189
 190                 /* First send a status message before receiving the fds. */
 191                 ret = consumer_send_status_msg(sock, ret_code);
 192                 if (ret < 0 || ret_code != LTTNG_OK) {
 193                         /*
 194                          * Somehow, the session daemon is not responding anymore or the
 195                          * channel was not found.
 196                          */
 197                         goto end_nosignal;
 198                 }
 199
 200                 /* block */
 201                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 202                         rcu_read_unlock();
 203                         return -EINTR;
 204                 }
 205
 206                 /* Get stream file descriptor from socket */
 207                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 208                 if (ret != sizeof(fd)) {
 209                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 210                         rcu_read_unlock();
 211                         return ret;
 212                 }
 213
 214                 /*
 215                  * Send status code to session daemon only if the recv works. If the
 216                  * above recv() failed, the session daemon is notified through the
 217                  * error socket and the teardown is eventually done.
 218                  */
 219                 ret = consumer_send_status_msg(sock, ret_code);
 220                 if (ret < 0) {
 221                         /* Somehow, the session daemon is not responding anymore. */
 222                         goto end_nosignal;
 223                 }
 224
 225                 new_stream = consumer_allocate_stream(channel->key,
 226                                 fd,
 227                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 228                                 channel->name,
 229                                 channel->uid,
 230                                 channel->gid,
 231                                 channel->relayd_id,
 232                                 channel->session_id,
 233                                 msg.u.stream.cpu,
 234                                 &alloc_ret,
 235                                 channel->type);
 236                 if (new_stream == NULL) {
 237                         switch (alloc_ret) {
 238                         case -ENOMEM:
 239                         case -EINVAL:
 240                         default:
 241                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 242                                 break;
 243                         }
 244                         goto end_nosignal;
 245                 }
 246                 new_stream->chan = channel;
 247                 new_stream->wait_fd = fd;
 248
 249                 /*
 250                  * The buffer flush is done on the session daemon side for the kernel
 251                  * so no need for the stream "hangup_flush_done" variable to be
 252                  * tracked. This is important for a kernel stream since we don't rely
 253                  * on the flush state of the stream to read data. It's not the case for
 254                  * user space tracing.
 255                  */
 256                 new_stream->hangup_flush_done = 0;
 257
 258                 /* The stream is not metadata. Get relayd reference if exists. */
 259                 relayd = consumer_find_relayd(new_stream->net_seq_idx);
 260                 if (relayd != NULL) {
 261                         /* Add stream on the relayd */
 262                         pthread_mutex_lock(&relayd->ctrl_sock_mutex);
 263                         ret = relayd_add_stream(&relayd->control_sock,
 264                                         new_stream->name, new_stream->chan->pathname,
 265                                         &new_stream->relayd_stream_id);
 266                         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 267                         if (ret < 0) {
 268                                 consumer_del_stream(new_stream, NULL);
 269                                 goto end_nosignal;
 270                         }
 271                 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 272                         ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
 273                                         new_stream->net_seq_idx);
 274                         consumer_del_stream(new_stream, NULL);
 275                         goto end_nosignal;
 276                 }
 277
 278                 if (ctx->on_recv_stream) {
 279                         ret = ctx->on_recv_stream(new_stream);
 280                         if (ret < 0) {
 281                                 consumer_del_stream(new_stream, NULL);
 282                                 goto end_nosignal;
 283                         }
 284                 }
 285
 286                 /* Get the right pipe where the stream will be sent. */
 287                 if (new_stream->metadata_flag) {
 288                         stream_pipe = ctx->consumer_metadata_pipe[1];
 289                 } else {
 290                         stream_pipe = ctx->consumer_data_pipe[1];
 291                 }
 292
 293                 do {
 294                         ret = write(stream_pipe, &new_stream, sizeof(new_stream));
 295                 } while (ret < 0 && errno == EINTR);
 296                 if (ret < 0) {
 297                         PERROR("Consumer write %s stream to pipe %d",
 298                                         new_stream->metadata_flag ? "metadata" : "data",
 299                                         stream_pipe);
 300                         consumer_del_stream(new_stream, NULL);
 301                         goto end_nosignal;
 302                 }
 303
 304                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 305                                 new_stream->name, fd, new_stream->relayd_stream_id);
 306                 break;
 307         }
 308         case LTTNG_CONSUMER_UPDATE_STREAM:
 309         {
 310                 rcu_read_unlock();
 311                 return -ENOSYS;
 312         }
 313         case LTTNG_CONSUMER_DESTROY_RELAYD:
 314         {
 315                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 316                 struct consumer_relayd_sock_pair *relayd;
 317
 318                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 319
 320                 /* Get relayd reference if exists. */
 321                 relayd = consumer_find_relayd(index);
 322                 if (relayd == NULL) {
 323                         DBG("Unable to find relayd %" PRIu64, index);
 324                         ret_code = LTTNG_ERR_NO_CONSUMER;
 325                 }
 326
 327                 /*
 328                  * Each relayd socket pair has a refcount of stream attached to it
 329                  * which tells if the relayd is still active or not depending on the
 330                  * refcount value.
 331                  *
 332                  * This will set the destroy flag of the relayd object and destroy it
 333                  * if the refcount reaches zero when called.
 334                  *
 335                  * The destroy can happen either here or when a stream fd hangs up.
 336                  */
 337                 if (relayd) {
 338                         consumer_flag_relayd_for_destroy(relayd);
 339                 }
 340
 341                 ret = consumer_send_status_msg(sock, ret_code);
 342                 if (ret < 0) {
 343                         /* Somehow, the session daemon is not responding anymore. */
 344                         goto end_nosignal;
 345                 }
 346
 347                 goto end_nosignal;
 348         }
 349         case LTTNG_CONSUMER_DATA_PENDING:
 350         {
 351                 int32_t ret;
 352                 uint64_t id = msg.u.data_pending.session_id;
 353
 354                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 355
 356                 ret = consumer_data_pending(id);
 357
 358                 /* Send back returned value to session daemon */
 359                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 360                 if (ret < 0) {
 361                         PERROR("send data pending ret code");
 362                 }
 363
 364                 /*
 365                  * No need to send back a status message since the data pending
 366                  * returned value is the response.
 367                  */
 368                 break;
 369         }
 370         default:
 371                 goto end_nosignal;
 372         }
 373
 374 end_nosignal:
 375         rcu_read_unlock();
 376
 377         /*
 378          * Return 1 to indicate success since the 0 value can be a socket
 379          * shutdown during the recv() or send() call.
 380          */
 381         return 1;
 382 }
 383
 384 /*
 385  * Consume data on a file descriptor and write it on a trace file.
 386  */
 387 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 388                 struct lttng_consumer_local_data *ctx)
 389 {
 390         unsigned long len, subbuf_size, padding;
 391         int err;
 392         ssize_t ret = 0;
 393         int infd = stream->wait_fd;
 394
 395         DBG("In read_subbuffer (infd : %d)", infd);
 396         /* Get the next subbuffer */
 397         err = kernctl_get_next_subbuf(infd);
 398         if (err != 0) {
 399                 ret = err;
 400                 /*
 401                  * This is a debug message even for single-threaded consumer,
 402                  * because poll() have more relaxed criterions than get subbuf,
 403                  * so get_subbuf may fail for short race windows where poll()
 404                  * would issue wakeups.
 405                  */
 406                 DBG("Reserving sub buffer failed (everything is normal, "
 407                                 "it is due to concurrency)");
 408                 goto end;
 409         }
 410
 411         /* Get the full subbuffer size including padding */
 412         err = kernctl_get_padded_subbuf_size(infd, &len);
 413         if (err != 0) {
 414                 errno = -err;
 415                 perror("Getting sub-buffer len failed.");
 416                 ret = err;
 417                 goto end;
 418         }
 419
 420         switch (stream->chan->output) {
 421         case LTTNG_EVENT_SPLICE:
 422
 423                 /*
 424                  * XXX: The lttng-modules splice "actor" does not handle copying
 425                  * partial pages hence only using the subbuffer size without the
 426                  * padding makes the splice fail.
 427                  */
 428                 subbuf_size = len;
 429                 padding = 0;
 430
 431                 /* splice the subbuffer to the tracefile */
 432                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 433                                 padding);
 434                 /*
 435                  * XXX: Splice does not support network streaming so the return value
 436                  * is simply checked against subbuf_size and not like the mmap() op.
 437                  */
 438                 if (ret != subbuf_size) {
 439                         /*
 440                          * display the error but continue processing to try
 441                          * to release the subbuffer
 442                          */
 443                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 444                                         ret, subbuf_size);
 445                 }
 446                 break;
 447         case LTTNG_EVENT_MMAP:
 448                 /* Get subbuffer size without padding */
 449                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 450                 if (err != 0) {
 451                         errno = -err;
 452                         perror("Getting sub-buffer len failed.");
 453                         ret = err;
 454                         goto end;
 455                 }
 456
 457                 /* Make sure the tracer is not gone mad on us! */
 458                 assert(len >= subbuf_size);
 459
 460                 padding = len - subbuf_size;
 461
 462                 /* write the subbuffer to the tracefile */
 463                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 464                                 padding);
 465                 /*
 466                  * The mmap operation should write subbuf_size amount of data when
 467                  * network streaming or the full padding (len) size when we are _not_
 468                  * streaming.
 469                  */
 470                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 471                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 472                         /*
 473                          * Display the error but continue processing to try to release the
 474                          * subbuffer
 475                          */
 476                         ERR("Error writing to tracefile "
 477                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 478                                         ret, len, subbuf_size);
 479                 }
 480                 break;
 481         default:
 482                 ERR("Unknown output method");
 483                 ret = -1;
 484         }
 485
 486         err = kernctl_put_next_subbuf(infd);
 487         if (err != 0) {
 488                 errno = -err;
 489                 if (errno == EFAULT) {
 490                         perror("Error in unreserving sub buffer\n");
 491                 } else if (errno == EIO) {
 492                         /* Should never happen with newer LTTng versions */
 493                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 494                 }
 495
 496                 ret = -err;
 497                 goto end;
 498         }
 499
 500 end:
 501         return ret;
 502 }
 503
 504 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 505 {
 506         int ret;
 507
 508         assert(stream);
 509
 510         /* Don't create anything if this is set for streaming. */
 511         if (stream->net_seq_idx == (uint64_t) -1ULL) {
 512                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
 513                                 stream->chan->tracefile_size, stream->tracefile_count_current,
 514                                 stream->uid, stream->gid);
 515                 if (ret < 0) {
 516                         goto error;
 517                 }
 518                 stream->out_fd = ret;
 519                 stream->tracefile_size_current = 0;
 520         }
 521
 522         if (stream->output == LTTNG_EVENT_MMAP) {
 523                 /* get the len of the mmap region */
 524                 unsigned long mmap_len;
 525
 526                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 527                 if (ret != 0) {
 528                         errno = -ret;
 529                         PERROR("kernctl_get_mmap_len");
 530                         goto error_close_fd;
 531                 }
 532                 stream->mmap_len = (size_t) mmap_len;
 533
 534                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 535                                 MAP_PRIVATE, stream->wait_fd, 0);
 536                 if (stream->mmap_base == MAP_FAILED) {
 537                         PERROR("Error mmaping");
 538                         ret = -1;
 539                         goto error_close_fd;
 540                 }
 541         }
 542
 543         /* we return 0 to let the library handle the FD internally */
 544         return 0;
 545
 546 error_close_fd:
 547         {
 548                 int err;
 549
 550                 err = close(stream->out_fd);
 551                 assert(!err);
 552         }
 553 error:
 554         return ret;
 555 }
 556
 557 /*
 558  * Check if data is still being extracted from the buffers for a specific
 559  * stream. Consumer data lock MUST be acquired before calling this function
 560  * and the stream lock.
 561  *
 562  * Return 1 if the traced data are still getting read else 0 meaning that the
 563  * data is available for trace viewer reading.
 564  */
 565 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 566 {
 567         int ret;
 568
 569         assert(stream);
 570
 571         ret = kernctl_get_next_subbuf(stream->wait_fd);
 572         if (ret == 0) {
 573                 /* There is still data so let's put back this subbuffer. */
 574                 ret = kernctl_put_subbuf(stream->wait_fd);
 575                 assert(ret == 0);
 576                 ret = 1;   /* Data is pending */
 577                 goto end;
 578         }
 579
 580         /* Data is NOT pending and ready to be read. */
 581         ret = 0;
 582
 583 end:
 584         return ret;
 585 }