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