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                                         new_stream->chan->tracefile_size,
 267                                         new_stream->chan->tracefile_count);
 268                         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 269                         if (ret < 0) {
 270                                 consumer_del_stream(new_stream, NULL);
 271                                 goto end_nosignal;
 272                         }
 273                 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 274                         ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
 275                                         new_stream->net_seq_idx);
 276                         consumer_del_stream(new_stream, NULL);
 277                         goto end_nosignal;
 278                 }
 279
 280                 if (ctx->on_recv_stream) {
 281                         ret = ctx->on_recv_stream(new_stream);
 282                         if (ret < 0) {
 283                                 consumer_del_stream(new_stream, NULL);
 284                                 goto end_nosignal;
 285                         }
 286                 }
 287
 288                 /* Get the right pipe where the stream will be sent. */
 289                 if (new_stream->metadata_flag) {
 290                         stream_pipe = ctx->consumer_metadata_pipe[1];
 291                 } else {
 292                         stream_pipe = ctx->consumer_data_pipe[1];
 293                 }
 294
 295                 do {
 296                         ret = write(stream_pipe, &new_stream, sizeof(new_stream));
 297                 } while (ret < 0 && errno == EINTR);
 298                 if (ret < 0) {
 299                         PERROR("Consumer write %s stream to pipe %d",
 300                                         new_stream->metadata_flag ? "metadata" : "data",
 301                                         stream_pipe);
 302                         consumer_del_stream(new_stream, NULL);
 303                         goto end_nosignal;
 304                 }
 305
 306                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 307                                 new_stream->name, fd, new_stream->relayd_stream_id);
 308                 break;
 309         }
 310         case LTTNG_CONSUMER_UPDATE_STREAM:
 311         {
 312                 rcu_read_unlock();
 313                 return -ENOSYS;
 314         }
 315         case LTTNG_CONSUMER_DESTROY_RELAYD:
 316         {
 317                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 318                 struct consumer_relayd_sock_pair *relayd;
 319
 320                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 321
 322                 /* Get relayd reference if exists. */
 323                 relayd = consumer_find_relayd(index);
 324                 if (relayd == NULL) {
 325                         DBG("Unable to find relayd %" PRIu64, index);
 326                         ret_code = LTTNG_ERR_NO_CONSUMER;
 327                 }
 328
 329                 /*
 330                  * Each relayd socket pair has a refcount of stream attached to it
 331                  * which tells if the relayd is still active or not depending on the
 332                  * refcount value.
 333                  *
 334                  * This will set the destroy flag of the relayd object and destroy it
 335                  * if the refcount reaches zero when called.
 336                  *
 337                  * The destroy can happen either here or when a stream fd hangs up.
 338                  */
 339                 if (relayd) {
 340                         consumer_flag_relayd_for_destroy(relayd);
 341                 }
 342
 343                 ret = consumer_send_status_msg(sock, ret_code);
 344                 if (ret < 0) {
 345                         /* Somehow, the session daemon is not responding anymore. */
 346                         goto end_nosignal;
 347                 }
 348
 349                 goto end_nosignal;
 350         }
 351         case LTTNG_CONSUMER_DATA_PENDING:
 352         {
 353                 int32_t ret;
 354                 uint64_t id = msg.u.data_pending.session_id;
 355
 356                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 357
 358                 ret = consumer_data_pending(id);
 359
 360                 /* Send back returned value to session daemon */
 361                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 362                 if (ret < 0) {
 363                         PERROR("send data pending ret code");
 364                 }
 365
 366                 /*
 367                  * No need to send back a status message since the data pending
 368                  * returned value is the response.
 369                  */
 370                 break;
 371         }
 372         default:
 373                 goto end_nosignal;
 374         }
 375
 376 end_nosignal:
 377         rcu_read_unlock();
 378
 379         /*
 380          * Return 1 to indicate success since the 0 value can be a socket
 381          * shutdown during the recv() or send() call.
 382          */
 383         return 1;
 384 }
 385
 386 /*
 387  * Consume data on a file descriptor and write it on a trace file.
 388  */
 389 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 390                 struct lttng_consumer_local_data *ctx)
 391 {
 392         unsigned long len, subbuf_size, padding;
 393         int err;
 394         ssize_t ret = 0;
 395         int infd = stream->wait_fd;
 396
 397         DBG("In read_subbuffer (infd : %d)", infd);
 398         /* Get the next subbuffer */
 399         err = kernctl_get_next_subbuf(infd);
 400         if (err != 0) {
 401                 ret = err;
 402                 /*
 403                  * This is a debug message even for single-threaded consumer,
 404                  * because poll() have more relaxed criterions than get subbuf,
 405                  * so get_subbuf may fail for short race windows where poll()
 406                  * would issue wakeups.
 407                  */
 408                 DBG("Reserving sub buffer failed (everything is normal, "
 409                                 "it is due to concurrency)");
 410                 goto end;
 411         }
 412
 413         /* Get the full subbuffer size including padding */
 414         err = kernctl_get_padded_subbuf_size(infd, &len);
 415         if (err != 0) {
 416                 errno = -err;
 417                 perror("Getting sub-buffer len failed.");
 418                 ret = err;
 419                 goto end;
 420         }
 421
 422         switch (stream->chan->output) {
 423         case LTTNG_EVENT_SPLICE:
 424
 425                 /*
 426                  * XXX: The lttng-modules splice "actor" does not handle copying
 427                  * partial pages hence only using the subbuffer size without the
 428                  * padding makes the splice fail.
 429                  */
 430                 subbuf_size = len;
 431                 padding = 0;
 432
 433                 /* splice the subbuffer to the tracefile */
 434                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 435                                 padding);
 436                 /*
 437                  * XXX: Splice does not support network streaming so the return value
 438                  * is simply checked against subbuf_size and not like the mmap() op.
 439                  */
 440                 if (ret != subbuf_size) {
 441                         /*
 442                          * display the error but continue processing to try
 443                          * to release the subbuffer
 444                          */
 445                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 446                                         ret, subbuf_size);
 447                 }
 448                 break;
 449         case LTTNG_EVENT_MMAP:
 450                 /* Get subbuffer size without padding */
 451                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 452                 if (err != 0) {
 453                         errno = -err;
 454                         perror("Getting sub-buffer len failed.");
 455                         ret = err;
 456                         goto end;
 457                 }
 458
 459                 /* Make sure the tracer is not gone mad on us! */
 460                 assert(len >= subbuf_size);
 461
 462                 padding = len - subbuf_size;
 463
 464                 /* write the subbuffer to the tracefile */
 465                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 466                                 padding);
 467                 /*
 468                  * The mmap operation should write subbuf_size amount of data when
 469                  * network streaming or the full padding (len) size when we are _not_
 470                  * streaming.
 471                  */
 472                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 473                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 474                         /*
 475                          * Display the error but continue processing to try to release the
 476                          * subbuffer
 477                          */
 478                         ERR("Error writing to tracefile "
 479                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 480                                         ret, len, subbuf_size);
 481                 }
 482                 break;
 483         default:
 484                 ERR("Unknown output method");
 485                 ret = -1;
 486         }
 487
 488         err = kernctl_put_next_subbuf(infd);
 489         if (err != 0) {
 490                 errno = -err;
 491                 if (errno == EFAULT) {
 492                         perror("Error in unreserving sub buffer\n");
 493                 } else if (errno == EIO) {
 494                         /* Should never happen with newer LTTng versions */
 495                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 496                 }
 497
 498                 ret = -err;
 499                 goto end;
 500         }
 501
 502 end:
 503         return ret;
 504 }
 505
 506 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 507 {
 508         int ret;
 509
 510         assert(stream);
 511
 512         /* Don't create anything if this is set for streaming. */
 513         if (stream->net_seq_idx == (uint64_t) -1ULL) {
 514                 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
 515                                 stream->chan->tracefile_size, stream->tracefile_count_current,
 516                                 stream->uid, stream->gid);
 517                 if (ret < 0) {
 518                         goto error;
 519                 }
 520                 stream->out_fd = ret;
 521                 stream->tracefile_size_current = 0;
 522         }
 523
 524         if (stream->output == LTTNG_EVENT_MMAP) {
 525                 /* get the len of the mmap region */
 526                 unsigned long mmap_len;
 527
 528                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 529                 if (ret != 0) {
 530                         errno = -ret;
 531                         PERROR("kernctl_get_mmap_len");
 532                         goto error_close_fd;
 533                 }
 534                 stream->mmap_len = (size_t) mmap_len;
 535
 536                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 537                                 MAP_PRIVATE, stream->wait_fd, 0);
 538                 if (stream->mmap_base == MAP_FAILED) {
 539                         PERROR("Error mmaping");
 540                         ret = -1;
 541                         goto error_close_fd;
 542                 }
 543         }
 544
 545         /* we return 0 to let the library handle the FD internally */
 546         return 0;
 547
 548 error_close_fd:
 549         {
 550                 int err;
 551
 552                 err = close(stream->out_fd);
 553                 assert(!err);
 554         }
 555 error:
 556         return ret;
 557 }
 558
 559 /*
 560  * Check if data is still being extracted from the buffers for a specific
 561  * stream. Consumer data lock MUST be acquired before calling this function
 562  * and the stream lock.
 563  *
 564  * Return 1 if the traced data are still getting read else 0 meaning that the
 565  * data is available for trace viewer reading.
 566  */
 567 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 568 {
 569         int ret;
 570
 571         assert(stream);
 572
 573         ret = kernctl_get_next_subbuf(stream->wait_fd);
 574         if (ret == 0) {
 575                 /* There is still data so let's put back this subbuffer. */
 576                 ret = kernctl_put_subbuf(stream->wait_fd);
 577                 assert(ret == 0);
 578                 ret = 1;   /* Data is pending */
 579                 goto end;
 580         }
 581
 582         /* Data is NOT pending and ready to be read. */
 583         ret = 0;
 584
 585 end:
 586         return ret;
 587 }