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                 if (new_channel == NULL) {
 137                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 138                         goto end_nosignal;
 139                 }
 140                 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
 141
 142                 /* Translate and save channel type. */
 143                 switch (msg.u.channel.type) {
 144                 case CONSUMER_CHANNEL_TYPE_DATA:
 145                 case CONSUMER_CHANNEL_TYPE_METADATA:
 146                         new_channel->type = msg.u.channel.type;
 147                         break;
 148                 default:
 149                         assert(0);
 150                         goto end_nosignal;
 151                 };
 152
 153                 if (ctx->on_recv_channel != NULL) {
 154                         ret = ctx->on_recv_channel(new_channel);
 155                         if (ret == 0) {
 156                                 consumer_add_channel(new_channel, ctx);
 157                         } else if (ret < 0) {
 158                                 goto end_nosignal;
 159                         }
 160                 } else {
 161                         consumer_add_channel(new_channel, ctx);
 162                 }
 163                 goto end_nosignal;
 164         }
 165         case LTTNG_CONSUMER_ADD_STREAM:
 166         {
 167                 int fd, stream_pipe;
 168                 struct consumer_relayd_sock_pair *relayd = NULL;
 169                 struct lttng_consumer_stream *new_stream;
 170                 struct lttng_consumer_channel *channel;
 171                 int alloc_ret = 0;
 172
 173                 /*
 174                  * Get stream's channel reference. Needed when adding the stream to the
 175                  * global hash table.
 176                  */
 177                 channel = consumer_find_channel(msg.u.stream.channel_key);
 178                 if (!channel) {
 179                         /*
 180                          * We could not find the channel. Can happen if cpu hotplug
 181                          * happens while tearing down.
 182                          */
 183                         ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
 184                         ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
 185                 }
 186
 187                 /* First send a status message before receiving the fds. */
 188                 ret = consumer_send_status_msg(sock, ret_code);
 189                 if (ret < 0 || ret_code != LTTNG_OK) {
 190                         /*
 191                          * Somehow, the session daemon is not responding anymore or the
 192                          * channel was not found.
 193                          */
 194                         goto end_nosignal;
 195                 }
 196
 197                 /* block */
 198                 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
 199                         rcu_read_unlock();
 200                         return -EINTR;
 201                 }
 202
 203                 /* Get stream file descriptor from socket */
 204                 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
 205                 if (ret != sizeof(fd)) {
 206                         lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
 207                         rcu_read_unlock();
 208                         return ret;
 209                 }
 210
 211                 /*
 212                  * Send status code to session daemon only if the recv works. If the
 213                  * above recv() failed, the session daemon is notified through the
 214                  * error socket and the teardown is eventually done.
 215                  */
 216                 ret = consumer_send_status_msg(sock, ret_code);
 217                 if (ret < 0) {
 218                         /* Somehow, the session daemon is not responding anymore. */
 219                         goto end_nosignal;
 220                 }
 221
 222                 new_stream = consumer_allocate_stream(channel->key,
 223                                 fd,
 224                                 LTTNG_CONSUMER_ACTIVE_STREAM,
 225                                 channel->name,
 226                                 channel->uid,
 227                                 channel->gid,
 228                                 channel->relayd_id,
 229                                 channel->session_id,
 230                                 msg.u.stream.cpu,
 231                                 &alloc_ret,
 232                                 channel->type);
 233                 if (new_stream == NULL) {
 234                         switch (alloc_ret) {
 235                         case -ENOMEM:
 236                         case -EINVAL:
 237                         default:
 238                                 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
 239                                 break;
 240                         }
 241                         goto end_nosignal;
 242                 }
 243                 new_stream->chan = channel;
 244                 new_stream->wait_fd = fd;
 245
 246                 /*
 247                  * The buffer flush is done on the session daemon side for the kernel
 248                  * so no need for the stream "hangup_flush_done" variable to be
 249                  * tracked. This is important for a kernel stream since we don't rely
 250                  * on the flush state of the stream to read data. It's not the case for
 251                  * user space tracing.
 252                  */
 253                 new_stream->hangup_flush_done = 0;
 254
 255                 /* The stream is not metadata. Get relayd reference if exists. */
 256                 relayd = consumer_find_relayd(new_stream->net_seq_idx);
 257                 if (relayd != NULL) {
 258                         /* Add stream on the relayd */
 259                         pthread_mutex_lock(&relayd->ctrl_sock_mutex);
 260                         ret = relayd_add_stream(&relayd->control_sock,
 261                                         new_stream->name, new_stream->chan->pathname,
 262                                         &new_stream->relayd_stream_id);
 263                         pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
 264                         if (ret < 0) {
 265                                 consumer_del_stream(new_stream, NULL);
 266                                 goto end_nosignal;
 267                         }
 268                 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
 269                         ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
 270                                         new_stream->net_seq_idx);
 271                         consumer_del_stream(new_stream, NULL);
 272                         goto end_nosignal;
 273                 }
 274
 275                 if (ctx->on_recv_stream) {
 276                         ret = ctx->on_recv_stream(new_stream);
 277                         if (ret < 0) {
 278                                 consumer_del_stream(new_stream, NULL);
 279                                 goto end_nosignal;
 280                         }
 281                 }
 282
 283                 /* Get the right pipe where the stream will be sent. */
 284                 if (new_stream->metadata_flag) {
 285                         stream_pipe = ctx->consumer_metadata_pipe[1];
 286                 } else {
 287                         stream_pipe = ctx->consumer_data_pipe[1];
 288                 }
 289
 290                 do {
 291                         ret = write(stream_pipe, &new_stream, sizeof(new_stream));
 292                 } while (ret < 0 && errno == EINTR);
 293                 if (ret < 0) {
 294                         PERROR("Consumer write %s stream to pipe %d",
 295                                         new_stream->metadata_flag ? "metadata" : "data",
 296                                         stream_pipe);
 297                         consumer_del_stream(new_stream, NULL);
 298                         goto end_nosignal;
 299                 }
 300
 301                 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
 302                                 new_stream->name, fd, new_stream->relayd_stream_id);
 303                 break;
 304         }
 305         case LTTNG_CONSUMER_UPDATE_STREAM:
 306         {
 307                 rcu_read_unlock();
 308                 return -ENOSYS;
 309         }
 310         case LTTNG_CONSUMER_DESTROY_RELAYD:
 311         {
 312                 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
 313                 struct consumer_relayd_sock_pair *relayd;
 314
 315                 DBG("Kernel consumer destroying relayd %" PRIu64, index);
 316
 317                 /* Get relayd reference if exists. */
 318                 relayd = consumer_find_relayd(index);
 319                 if (relayd == NULL) {
 320                         DBG("Unable to find relayd %" PRIu64, index);
 321                         ret_code = LTTNG_ERR_NO_CONSUMER;
 322                 }
 323
 324                 /*
 325                  * Each relayd socket pair has a refcount of stream attached to it
 326                  * which tells if the relayd is still active or not depending on the
 327                  * refcount value.
 328                  *
 329                  * This will set the destroy flag of the relayd object and destroy it
 330                  * if the refcount reaches zero when called.
 331                  *
 332                  * The destroy can happen either here or when a stream fd hangs up.
 333                  */
 334                 if (relayd) {
 335                         consumer_flag_relayd_for_destroy(relayd);
 336                 }
 337
 338                 ret = consumer_send_status_msg(sock, ret_code);
 339                 if (ret < 0) {
 340                         /* Somehow, the session daemon is not responding anymore. */
 341                         goto end_nosignal;
 342                 }
 343
 344                 goto end_nosignal;
 345         }
 346         case LTTNG_CONSUMER_DATA_PENDING:
 347         {
 348                 int32_t ret;
 349                 uint64_t id = msg.u.data_pending.session_id;
 350
 351                 DBG("Kernel consumer data pending command for id %" PRIu64, id);
 352
 353                 ret = consumer_data_pending(id);
 354
 355                 /* Send back returned value to session daemon */
 356                 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
 357                 if (ret < 0) {
 358                         PERROR("send data pending ret code");
 359                 }
 360
 361                 /*
 362                  * No need to send back a status message since the data pending
 363                  * returned value is the response.
 364                  */
 365                 break;
 366         }
 367         default:
 368                 goto end_nosignal;
 369         }
 370
 371 end_nosignal:
 372         rcu_read_unlock();
 373
 374         /*
 375          * Return 1 to indicate success since the 0 value can be a socket
 376          * shutdown during the recv() or send() call.
 377          */
 378         return 1;
 379 }
 380
 381 /*
 382  * Consume data on a file descriptor and write it on a trace file.
 383  */
 384 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
 385                 struct lttng_consumer_local_data *ctx)
 386 {
 387         unsigned long len, subbuf_size, padding;
 388         int err;
 389         ssize_t ret = 0;
 390         int infd = stream->wait_fd;
 391
 392         DBG("In read_subbuffer (infd : %d)", infd);
 393         /* Get the next subbuffer */
 394         err = kernctl_get_next_subbuf(infd);
 395         if (err != 0) {
 396                 ret = err;
 397                 /*
 398                  * This is a debug message even for single-threaded consumer,
 399                  * because poll() have more relaxed criterions than get subbuf,
 400                  * so get_subbuf may fail for short race windows where poll()
 401                  * would issue wakeups.
 402                  */
 403                 DBG("Reserving sub buffer failed (everything is normal, "
 404                                 "it is due to concurrency)");
 405                 goto end;
 406         }
 407
 408         /* Get the full subbuffer size including padding */
 409         err = kernctl_get_padded_subbuf_size(infd, &len);
 410         if (err != 0) {
 411                 errno = -err;
 412                 perror("Getting sub-buffer len failed.");
 413                 ret = err;
 414                 goto end;
 415         }
 416
 417         switch (stream->chan->output) {
 418         case LTTNG_EVENT_SPLICE:
 419
 420                 /*
 421                  * XXX: The lttng-modules splice "actor" does not handle copying
 422                  * partial pages hence only using the subbuffer size without the
 423                  * padding makes the splice fail.
 424                  */
 425                 subbuf_size = len;
 426                 padding = 0;
 427
 428                 /* splice the subbuffer to the tracefile */
 429                 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
 430                                 padding);
 431                 /*
 432                  * XXX: Splice does not support network streaming so the return value
 433                  * is simply checked against subbuf_size and not like the mmap() op.
 434                  */
 435                 if (ret != subbuf_size) {
 436                         /*
 437                          * display the error but continue processing to try
 438                          * to release the subbuffer
 439                          */
 440                         ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
 441                                         ret, subbuf_size);
 442                 }
 443                 break;
 444         case LTTNG_EVENT_MMAP:
 445                 /* Get subbuffer size without padding */
 446                 err = kernctl_get_subbuf_size(infd, &subbuf_size);
 447                 if (err != 0) {
 448                         errno = -err;
 449                         perror("Getting sub-buffer len failed.");
 450                         ret = err;
 451                         goto end;
 452                 }
 453
 454                 /* Make sure the tracer is not gone mad on us! */
 455                 assert(len >= subbuf_size);
 456
 457                 padding = len - subbuf_size;
 458
 459                 /* write the subbuffer to the tracefile */
 460                 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
 461                                 padding);
 462                 /*
 463                  * The mmap operation should write subbuf_size amount of data when
 464                  * network streaming or the full padding (len) size when we are _not_
 465                  * streaming.
 466                  */
 467                 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
 468                                 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
 469                         /*
 470                          * Display the error but continue processing to try to release the
 471                          * subbuffer
 472                          */
 473                         ERR("Error writing to tracefile "
 474                                         "(ret: %zd != len: %lu != subbuf_size: %lu)",
 475                                         ret, len, subbuf_size);
 476                 }
 477                 break;
 478         default:
 479                 ERR("Unknown output method");
 480                 ret = -1;
 481         }
 482
 483         err = kernctl_put_next_subbuf(infd);
 484         if (err != 0) {
 485                 errno = -err;
 486                 if (errno == EFAULT) {
 487                         perror("Error in unreserving sub buffer\n");
 488                 } else if (errno == EIO) {
 489                         /* Should never happen with newer LTTng versions */
 490                         perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
 491                 }
 492
 493                 ret = -err;
 494                 goto end;
 495         }
 496
 497 end:
 498         return ret;
 499 }
 500
 501 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
 502 {
 503         int ret;
 504         char full_path[PATH_MAX];
 505
 506         assert(stream);
 507
 508         ret = snprintf(full_path, sizeof(full_path), "%s/%s",
 509                         stream->chan->pathname, stream->name);
 510         if (ret < 0) {
 511                 PERROR("snprintf on_recv_stream");
 512                 goto error;
 513         }
 514
 515         /* Opening the tracefile in write mode */
 516         if (stream->net_seq_idx == (uint64_t) -1ULL) {
 517                 ret = run_as_open(full_path, O_WRONLY | O_CREAT | O_TRUNC,
 518                                 S_IRWXU|S_IRWXG|S_IRWXO, stream->uid, stream->gid);
 519                 if (ret < 0) {
 520                         PERROR("open kernel stream path %s", full_path);
 521                         goto error;
 522                 }
 523                 stream->out_fd = ret;
 524         }
 525
 526         if (stream->output == LTTNG_EVENT_MMAP) {
 527                 /* get the len of the mmap region */
 528                 unsigned long mmap_len;
 529
 530                 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
 531                 if (ret != 0) {
 532                         errno = -ret;
 533                         PERROR("kernctl_get_mmap_len");
 534                         goto error_close_fd;
 535                 }
 536                 stream->mmap_len = (size_t) mmap_len;
 537
 538                 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
 539                                 MAP_PRIVATE, stream->wait_fd, 0);
 540                 if (stream->mmap_base == MAP_FAILED) {
 541                         PERROR("Error mmaping");
 542                         ret = -1;
 543                         goto error_close_fd;
 544                 }
 545         }
 546
 547         /* we return 0 to let the library handle the FD internally */
 548         return 0;
 549
 550 error_close_fd:
 551         {
 552                 int err;
 553
 554                 err = close(stream->out_fd);
 555                 assert(!err);
 556         }
 557 error:
 558         return ret;
 559 }
 560
 561 /*
 562  * Check if data is still being extracted from the buffers for a specific
 563  * stream. Consumer data lock MUST be acquired before calling this function
 564  * and the stream lock.
 565  *
 566  * Return 1 if the traced data are still getting read else 0 meaning that the
 567  * data is available for trace viewer reading.
 568  */
 569 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
 570 {
 571         int ret;
 572
 573         assert(stream);
 574
 575         ret = kernctl_get_next_subbuf(stream->wait_fd);
 576         if (ret == 0) {
 577                 /* There is still data so let's put back this subbuffer. */
 578                 ret = kernctl_put_subbuf(stream->wait_fd);
 579                 assert(ret == 0);
 580                 ret = 1;   /* Data is pending */
 581                 goto end;
 582         }
 583
 584         /* Data is NOT pending and ready to be read. */
 585         ret = 0;
 586
 587 end:
 588         return ret;
 589 }