src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _GNU_SOURCE
  19 #include <errno.h>
  20 #include <fcntl.h>
  21 #include <stdlib.h>
  22 #include <stdio.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25 #include <inttypes.h>
  26
  27 #include <common/common.h>
  28 #include <common/kernel-ctl/kernel-ctl.h>
  29 #include <common/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kernel-consumer.h"
  34 #include "kern-modules.h"
  35
  36 /*
  37  * Add context on a kernel channel.
  38  */
  39 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  40                 struct ltt_kernel_context *ctx)
  41 {
  42         int ret;
  43
  44         assert(chan);
  45         assert(ctx);
  46
  47         DBG("Adding context to channel %s", chan->channel->name);
  48         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  49         if (ret < 0) {
  50                 if (errno != EEXIST) {
  51                         PERROR("add context ioctl");
  52                 } else {
  53                         /* If EEXIST, we just ignore the error */
  54                         ret = 0;
  55                 }
  56                 goto error;
  57         }
  58
  59         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  60
  61         return 0;
  62
  63 error:
  64         return ret;
  65 }
  66
  67 /*
  68  * Create a new kernel session, register it to the kernel tracer and add it to
  69  * the session daemon session.
  70  */
  71 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  72 {
  73         int ret;
  74         struct ltt_kernel_session *lks;
  75
  76         assert(session);
  77
  78         /* Allocate data structure */
  79         lks = trace_kernel_create_session();
  80         if (lks == NULL) {
  81                 ret = -1;
  82                 goto error;
  83         }
  84
  85         /* Kernel tracer session creation */
  86         ret = kernctl_create_session(tracer_fd);
  87         if (ret < 0) {
  88                 PERROR("ioctl kernel create session");
  89                 goto error;
  90         }
  91
  92         lks->fd = ret;
  93         /* Prevent fd duplication after execlp() */
  94         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
  95         if (ret < 0) {
  96                 PERROR("fcntl session fd");
  97         }
  98
  99         lks->id = session->id;
 100         lks->consumer_fds_sent = 0;
 101         session->kernel_session = lks;
 102
 103         DBG("Kernel session created (fd: %d)", lks->fd);
 104
 105         return 0;
 106
 107 error:
 108         if (lks) {
 109                 trace_kernel_destroy_session(lks);
 110         }
 111         return ret;
 112 }
 113
 114 /*
 115  * Create a kernel channel, register it to the kernel tracer and add it to the
 116  * kernel session.
 117  */
 118 int kernel_create_channel(struct ltt_kernel_session *session,
 119                 struct lttng_channel *chan)
 120 {
 121         int ret;
 122         struct ltt_kernel_channel *lkc;
 123
 124         assert(session);
 125         assert(chan);
 126
 127         /* Allocate kernel channel */
 128         lkc = trace_kernel_create_channel(chan);
 129         if (lkc == NULL) {
 130                 goto error;
 131         }
 132
 133         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 134                         chan->name, lkc->channel->attr.overwrite,
 135                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 136                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 137                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 138
 139         /* Kernel tracer channel creation */
 140         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 141         if (ret < 0) {
 142                 PERROR("ioctl kernel create channel");
 143                 goto error;
 144         }
 145
 146         /* Setup the channel fd */
 147         lkc->fd = ret;
 148         /* Prevent fd duplication after execlp() */
 149         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 150         if (ret < 0) {
 151                 PERROR("fcntl session fd");
 152         }
 153
 154         /* Add channel to session */
 155         cds_list_add(&lkc->list, &session->channel_list.head);
 156         session->channel_count++;
 157         lkc->session = session;
 158
 159         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 160
 161         return 0;
 162
 163 error:
 164         if (lkc) {
 165                 free(lkc->channel);
 166                 free(lkc);
 167         }
 168         return -1;
 169 }
 170
 171 /*
 172  * Create a kernel event, enable it to the kernel tracer and add it to the
 173  * channel event list of the kernel session.
 174  */
 175 int kernel_create_event(struct lttng_event *ev,
 176                 struct ltt_kernel_channel *channel)
 177 {
 178         int ret;
 179         struct ltt_kernel_event *event;
 180
 181         assert(ev);
 182         assert(channel);
 183
 184         event = trace_kernel_create_event(ev);
 185         if (event == NULL) {
 186                 ret = -1;
 187                 goto error;
 188         }
 189
 190         ret = kernctl_create_event(channel->fd, event->event);
 191         if (ret < 0) {
 192                 switch (errno) {
 193                 case EEXIST:
 194                         break;
 195                 case ENOSYS:
 196                         WARN("Event type not implemented");
 197                         break;
 198                 case ENOENT:
 199                         WARN("Event %s not found!", ev->name);
 200                         break;
 201                 default:
 202                         PERROR("create event ioctl");
 203                 }
 204                 ret = -errno;
 205                 goto free_event;
 206         }
 207
 208         /*
 209          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 210          */
 211         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 212                 DBG2("Kernel event syscall creation success");
 213                 /*
 214                  * We use fd == -1 to ensure that we never trigger a close of fd
 215                  * 0.
 216                  */
 217                 event->fd = -1;
 218                 goto add_list;
 219         }
 220
 221         event->fd = ret;
 222         /* Prevent fd duplication after execlp() */
 223         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 224         if (ret < 0) {
 225                 PERROR("fcntl session fd");
 226         }
 227
 228 add_list:
 229         /* Add event to event list */
 230         cds_list_add(&event->list, &channel->events_list.head);
 231         channel->event_count++;
 232
 233         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 234
 235         return 0;
 236
 237 free_event:
 238         free(event);
 239 error:
 240         return ret;
 241 }
 242
 243 /*
 244  * Disable a kernel channel.
 245  */
 246 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 247 {
 248         int ret;
 249
 250         assert(chan);
 251
 252         ret = kernctl_disable(chan->fd);
 253         if (ret < 0) {
 254                 PERROR("disable chan ioctl");
 255                 ret = errno;
 256                 goto error;
 257         }
 258
 259         chan->enabled = 0;
 260         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 261
 262         return 0;
 263
 264 error:
 265         return ret;
 266 }
 267
 268 /*
 269  * Enable a kernel channel.
 270  */
 271 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 272 {
 273         int ret;
 274
 275         assert(chan);
 276
 277         ret = kernctl_enable(chan->fd);
 278         if (ret < 0 && errno != EEXIST) {
 279                 PERROR("Enable kernel chan");
 280                 goto error;
 281         }
 282
 283         chan->enabled = 1;
 284         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 285
 286         return 0;
 287
 288 error:
 289         return ret;
 290 }
 291
 292 /*
 293  * Enable a kernel event.
 294  */
 295 int kernel_enable_event(struct ltt_kernel_event *event)
 296 {
 297         int ret;
 298
 299         assert(event);
 300
 301         ret = kernctl_enable(event->fd);
 302         if (ret < 0) {
 303                 switch (errno) {
 304                 case EEXIST:
 305                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 306                         break;
 307                 default:
 308                         PERROR("enable kernel event");
 309                         break;
 310                 }
 311                 goto error;
 312         }
 313
 314         event->enabled = 1;
 315         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 316
 317         return 0;
 318
 319 error:
 320         return ret;
 321 }
 322
 323 /*
 324  * Disable a kernel event.
 325  */
 326 int kernel_disable_event(struct ltt_kernel_event *event)
 327 {
 328         int ret;
 329
 330         assert(event);
 331
 332         ret = kernctl_disable(event->fd);
 333         if (ret < 0) {
 334                 switch (errno) {
 335                 case EEXIST:
 336                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 337                         break;
 338                 default:
 339                         PERROR("disable kernel event");
 340                         break;
 341                 }
 342                 goto error;
 343         }
 344
 345         event->enabled = 0;
 346         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 347
 348         return 0;
 349
 350 error:
 351         return ret;
 352 }
 353
 354 /*
 355  * Create kernel metadata, open from the kernel tracer and add it to the
 356  * kernel session.
 357  */
 358 int kernel_open_metadata(struct ltt_kernel_session *session)
 359 {
 360         int ret;
 361         struct ltt_kernel_metadata *lkm = NULL;
 362
 363         assert(session);
 364
 365         /* Allocate kernel metadata */
 366         lkm = trace_kernel_create_metadata();
 367         if (lkm == NULL) {
 368                 goto error;
 369         }
 370
 371         /* Kernel tracer metadata creation */
 372         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 373         if (ret < 0) {
 374                 goto error_open;
 375         }
 376
 377         lkm->fd = ret;
 378         /* Prevent fd duplication after execlp() */
 379         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 380         if (ret < 0) {
 381                 PERROR("fcntl session fd");
 382         }
 383
 384         session->metadata = lkm;
 385
 386         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 387
 388         return 0;
 389
 390 error_open:
 391         trace_kernel_destroy_metadata(lkm);
 392 error:
 393         return -1;
 394 }
 395
 396 /*
 397  * Start tracing session.
 398  */
 399 int kernel_start_session(struct ltt_kernel_session *session)
 400 {
 401         int ret;
 402
 403         assert(session);
 404
 405         ret = kernctl_start_session(session->fd);
 406         if (ret < 0) {
 407                 PERROR("ioctl start session");
 408                 goto error;
 409         }
 410
 411         DBG("Kernel session started");
 412
 413         return 0;
 414
 415 error:
 416         return ret;
 417 }
 418
 419 /*
 420  * Make a kernel wait to make sure in-flight probe have completed.
 421  */
 422 void kernel_wait_quiescent(int fd)
 423 {
 424         int ret;
 425
 426         DBG("Kernel quiescent wait on %d", fd);
 427
 428         ret = kernctl_wait_quiescent(fd);
 429         if (ret < 0) {
 430                 PERROR("wait quiescent ioctl");
 431                 ERR("Kernel quiescent wait failed");
 432         }
 433 }
 434
 435 /*
 436  * Kernel calibrate
 437  */
 438 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 439 {
 440         int ret;
 441
 442         assert(calibrate);
 443
 444         ret = kernctl_calibrate(fd, calibrate);
 445         if (ret < 0) {
 446                 PERROR("calibrate ioctl");
 447                 return -1;
 448         }
 449
 450         return 0;
 451 }
 452
 453
 454 /*
 455  *  Force flush buffer of metadata.
 456  */
 457 int kernel_metadata_flush_buffer(int fd)
 458 {
 459         int ret;
 460
 461         DBG("Kernel flushing metadata buffer on fd %d", fd);
 462
 463         ret = kernctl_buffer_flush(fd);
 464         if (ret < 0) {
 465                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 466         }
 467
 468         return 0;
 469 }
 470
 471 /*
 472  * Force flush buffer for channel.
 473  */
 474 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 475 {
 476         int ret;
 477         struct ltt_kernel_stream *stream;
 478
 479         assert(channel);
 480
 481         DBG("Flush buffer for channel %s", channel->channel->name);
 482
 483         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 484                 DBG("Flushing channel stream %d", stream->fd);
 485                 ret = kernctl_buffer_flush(stream->fd);
 486                 if (ret < 0) {
 487                         PERROR("ioctl");
 488                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 489                                         stream->fd, ret);
 490                 }
 491         }
 492
 493         return 0;
 494 }
 495
 496 /*
 497  * Stop tracing session.
 498  */
 499 int kernel_stop_session(struct ltt_kernel_session *session)
 500 {
 501         int ret;
 502
 503         assert(session);
 504
 505         ret = kernctl_stop_session(session->fd);
 506         if (ret < 0) {
 507                 goto error;
 508         }
 509
 510         DBG("Kernel session stopped");
 511
 512         return 0;
 513
 514 error:
 515         return ret;
 516 }
 517
 518 /*
 519  * Open stream of channel, register it to the kernel tracer and add it
 520  * to the stream list of the channel.
 521  *
 522  * Return the number of created stream. Else, a negative value.
 523  */
 524 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 525 {
 526         int ret, count = 0;
 527         struct ltt_kernel_stream *lks;
 528
 529         assert(channel);
 530
 531         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 532                 lks = trace_kernel_create_stream(channel->channel->name, count);
 533                 if (lks == NULL) {
 534                         ret = close(ret);
 535                         if (ret) {
 536                                 PERROR("close");
 537                         }
 538                         goto error;
 539                 }
 540
 541                 lks->fd = ret;
 542                 /* Prevent fd duplication after execlp() */
 543                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 544                 if (ret < 0) {
 545                         PERROR("fcntl session fd");
 546                 }
 547
 548                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 549                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 550
 551                 /* Add stream to channe stream list */
 552                 cds_list_add(&lks->list, &channel->stream_list.head);
 553                 channel->stream_count++;
 554
 555                 /* Increment counter which represent CPU number. */
 556                 count++;
 557
 558                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 559                                 lks->state);
 560         }
 561
 562         return channel->stream_count;
 563
 564 error:
 565         return -1;
 566 }
 567
 568 /*
 569  * Open the metadata stream and set it to the kernel session.
 570  */
 571 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 572 {
 573         int ret;
 574
 575         assert(session);
 576
 577         ret = kernctl_create_stream(session->metadata->fd);
 578         if (ret < 0) {
 579                 PERROR("kernel create metadata stream");
 580                 goto error;
 581         }
 582
 583         DBG("Kernel metadata stream created (fd: %d)", ret);
 584         session->metadata_stream_fd = ret;
 585         /* Prevent fd duplication after execlp() */
 586         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 587         if (ret < 0) {
 588                 PERROR("fcntl session fd");
 589         }
 590
 591         return 0;
 592
 593 error:
 594         return -1;
 595 }
 596
 597 /*
 598  * Get the event list from the kernel tracer and return the number of elements.
 599  */
 600 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 601 {
 602         int fd, ret;
 603         char *event;
 604         size_t nbmem, count = 0;
 605         FILE *fp;
 606         struct lttng_event *elist;
 607
 608         assert(events);
 609
 610         fd = kernctl_tracepoint_list(tracer_fd);
 611         if (fd < 0) {
 612                 PERROR("kernel tracepoint list");
 613                 goto error;
 614         }
 615
 616         fp = fdopen(fd, "r");
 617         if (fp == NULL) {
 618                 PERROR("kernel tracepoint list fdopen");
 619                 goto error_fp;
 620         }
 621
 622         /*
 623          * Init memory size counter
 624          * See kernel-ctl.h for explanation of this value
 625          */
 626         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 627         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 628         if (elist == NULL) {
 629                 PERROR("alloc list events");
 630                 count = -ENOMEM;
 631                 goto end;
 632         }
 633
 634         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 635                 if (count >= nbmem) {
 636                         struct lttng_event *new_elist;
 637                         size_t new_nbmem;
 638
 639                         new_nbmem = nbmem << 1;
 640                         DBG("Reallocating event list from %zu to %zu bytes",
 641                                         nbmem, new_nbmem);
 642                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 643                         if (new_elist == NULL) {
 644                                 PERROR("realloc list events");
 645                                 free(event);
 646                                 free(elist);
 647                                 count = -ENOMEM;
 648                                 goto end;
 649                         }
 650                         /* Zero the new memory */
 651                         memset(new_elist + nbmem, 0,
 652                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 653                         nbmem = new_nbmem;
 654                         elist = new_elist;
 655                 }
 656                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 657                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 658                 elist[count].enabled = -1;
 659                 count++;
 660                 free(event);
 661         }
 662
 663         *events = elist;
 664         DBG("Kernel list events done (%zu events)", count);
 665 end:
 666         ret = fclose(fp);       /* closes both fp and fd */
 667         if (ret) {
 668                 PERROR("fclose");
 669         }
 670         return count;
 671
 672 error_fp:
 673         ret = close(fd);
 674         if (ret) {
 675                 PERROR("close");
 676         }
 677 error:
 678         return -1;
 679 }
 680
 681 /*
 682  * Get kernel version and validate it.
 683  */
 684 int kernel_validate_version(int tracer_fd)
 685 {
 686         int ret;
 687         struct lttng_kernel_tracer_version version;
 688
 689         ret = kernctl_tracer_version(tracer_fd, &version);
 690         if (ret < 0) {
 691                 ERR("Failed at getting the lttng-modules version");
 692                 goto error;
 693         }
 694
 695         /* Validate version */
 696         if (version.major != KERN_MODULES_PRE_MAJOR
 697                 && version.major != KERN_MODULES_MAJOR) {
 698                 goto error_version;
 699         }
 700
 701         DBG2("Kernel tracer version validated (major version %d)", version.major);
 702         return 0;
 703
 704 error_version:
 705         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 706                         version.major, KERN_MODULES_MAJOR)
 707         ret = -1;
 708
 709 error:
 710         return ret;
 711 }
 712
 713 /*
 714  * Kernel work-arounds called at the start of sessiond main().
 715  */
 716 int init_kernel_workarounds(void)
 717 {
 718         int ret;
 719         FILE *fp;
 720
 721         /*
 722          * boot_id needs to be read once before being used concurrently
 723          * to deal with a Linux kernel race. A fix is proposed for
 724          * upstream, but the work-around is needed for older kernels.
 725          */
 726         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 727         if (!fp) {
 728                 goto end_boot_id;
 729         }
 730         while (!feof(fp)) {
 731                 char buf[37] = "";
 732
 733                 ret = fread(buf, 1, sizeof(buf), fp);
 734                 if (ret < 0) {
 735                         /* Ignore error, we don't really care */
 736                 }
 737         }
 738         ret = fclose(fp);
 739         if (ret) {
 740                 PERROR("fclose");
 741         }
 742 end_boot_id:
 743         return 0;
 744 }
 745
 746 /*
 747  * Complete teardown of a kernel session.
 748  */
 749 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 750 {
 751         if (ksess == NULL) {
 752                 DBG3("No kernel session when tearing down session");
 753                 return;
 754         }
 755
 756         DBG("Tearing down kernel session");
 757
 758         /*
 759          * Destroy channels on the consumer if at least one FD has been sent and we
 760          * are in no output mode because the streams are in *no* monitor mode so we
 761          * have to send a command to clean them up or else they leaked.
 762          */
 763         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 764                 int ret;
 765                 struct consumer_socket *socket;
 766                 struct lttng_ht_iter iter;
 767
 768                 /* For each consumer socket. */
 769                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 770                                 socket, node.node) {
 771                         struct ltt_kernel_channel *chan;
 772
 773                         /* For each channel, ask the consumer to destroy it. */
 774                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 775                                 ret = kernel_consumer_destroy_channel(socket, chan);
 776                                 if (ret < 0) {
 777                                         /* Consumer is probably dead. Use next socket. */
 778                                         continue;
 779                                 }
 780                         }
 781                 }
 782         }
 783
 784         /* Close any relayd session */
 785         consumer_output_send_destroy_relayd(ksess->consumer);
 786
 787         trace_kernel_destroy_session(ksess);
 788 }
 789
 790 /*
 791  * Destroy a kernel channel object. It does not do anything on the tracer side.
 792  */
 793 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 794 {
 795         struct ltt_kernel_session *ksess = NULL;
 796
 797         assert(kchan);
 798         assert(kchan->channel);
 799
 800         DBG3("Kernel destroy channel %s", kchan->channel->name);
 801
 802         /* Update channel count of associated session. */
 803         if (kchan->session) {
 804                 /* Keep pointer reference so we can update it after the destroy. */
 805                 ksess = kchan->session;
 806         }
 807
 808         trace_kernel_destroy_channel(kchan);
 809
 810         /*
 811          * At this point the kernel channel is not visible anymore. This is safe
 812          * since in order to work on a visible kernel session, the tracing session
 813          * lock (ltt_session.lock) MUST be acquired.
 814          */
 815         if (ksess) {
 816                 ksess->channel_count--;
 817         }
 818 }
 819
 820 /*
 821  * Take a snapshot for a given kernel session.
 822  *
 823  * Return 0 on success or else return a LTTNG_ERR code.
 824  */
 825 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 826                 struct snapshot_output *output, int wait, unsigned int nb_streams)
 827 {
 828         int err, ret, saved_metadata_fd;
 829         struct consumer_socket *socket;
 830         struct lttng_ht_iter iter;
 831         struct ltt_kernel_metadata *saved_metadata;
 832         uint64_t max_size_per_stream = 0;
 833
 834         assert(ksess);
 835         assert(ksess->consumer);
 836         assert(output);
 837
 838         DBG("Kernel snapshot record started");
 839
 840         /* Save current metadata since the following calls will change it. */
 841         saved_metadata = ksess->metadata;
 842         saved_metadata_fd = ksess->metadata_stream_fd;
 843
 844         rcu_read_lock();
 845
 846         ret = kernel_open_metadata(ksess);
 847         if (ret < 0) {
 848                 ret = LTTNG_ERR_KERN_META_FAIL;
 849                 goto error;
 850         }
 851
 852         ret = kernel_open_metadata_stream(ksess);
 853         if (ret < 0) {
 854                 ret = LTTNG_ERR_KERN_META_FAIL;
 855                 goto error_open_stream;
 856         }
 857
 858         if (output->max_size > 0 && nb_streams > 0) {
 859                 max_size_per_stream = output->max_size / nb_streams;
 860         }
 861
 862         /* Send metadata to consumer and snapshot everything. */
 863         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 864                         socket, node.node) {
 865                 struct consumer_output *saved_output;
 866                 struct ltt_kernel_channel *chan;
 867
 868                 /*
 869                  * Temporarly switch consumer output for our snapshot output. As long
 870                  * as the session lock is taken, this is safe.
 871                  */
 872                 saved_output = ksess->consumer;
 873                 ksess->consumer = output->consumer;
 874
 875                 pthread_mutex_lock(socket->lock);
 876                 /* This stream must not be monitored by the consumer. */
 877                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 878                 pthread_mutex_unlock(socket->lock);
 879                 /* Put back the saved consumer output into the session. */
 880                 ksess->consumer = saved_output;
 881                 if (ret < 0) {
 882                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 883                         goto error_consumer;
 884                 }
 885
 886                 /* For each channel, ask the consumer to snapshot it. */
 887                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 888                         if (max_size_per_stream &&
 889                                         chan->channel->attr.subbuf_size > max_size_per_stream) {
 890                                 ret = LTTNG_ERR_INVALID;
 891                                 DBG3("Kernel snapshot record maximum stream size %" PRIu64
 892                                                 " is smaller than subbuffer size of %" PRIu64,
 893                                                 max_size_per_stream, chan->channel->attr.subbuf_size);
 894                                 (void) kernel_consumer_destroy_metadata(socket,
 895                                                 ksess->metadata);
 896                                 goto error_consumer;
 897                         }
 898
 899                         pthread_mutex_lock(socket->lock);
 900                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
 901                                         ksess->uid, ksess->gid,
 902                                         DEFAULT_KERNEL_TRACE_DIR, wait,
 903                                         max_size_per_stream);
 904                         pthread_mutex_unlock(socket->lock);
 905                         if (ret < 0) {
 906                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 907                                 (void) kernel_consumer_destroy_metadata(socket,
 908                                                 ksess->metadata);
 909                                 goto error_consumer;
 910                         }
 911                 }
 912
 913                 /* Snapshot metadata, */
 914                 pthread_mutex_lock(socket->lock);
 915                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
 916                                 1, ksess->uid, ksess->gid,
 917                                 DEFAULT_KERNEL_TRACE_DIR, wait, max_size_per_stream);
 918                 pthread_mutex_unlock(socket->lock);
 919                 if (ret < 0) {
 920                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 921                         goto error_consumer;
 922                 }
 923
 924                 /*
 925                  * The metadata snapshot is done, ask the consumer to destroy it since
 926                  * it's not monitored on the consumer side.
 927                  */
 928                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
 929         }
 930
 931         ret = LTTNG_OK;
 932
 933 error_consumer:
 934         /* Close newly opened metadata stream. It's now on the consumer side. */
 935         err = close(ksess->metadata_stream_fd);
 936         if (err < 0) {
 937                 PERROR("close snapshot kernel");
 938         }
 939
 940 error_open_stream:
 941         trace_kernel_destroy_metadata(ksess->metadata);
 942 error:
 943         /* Restore metadata state.*/
 944         ksess->metadata = saved_metadata;
 945         ksess->metadata_stream_fd = saved_metadata_fd;
 946
 947         rcu_read_unlock();
 948         return ret;
 949 }