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 lttng_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);
  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         chan->ctx = zmalloc(sizeof(struct lttng_kernel_context));
  60         if (chan->ctx == NULL) {
  61                 PERROR("zmalloc event context");
  62                 goto error;
  63         }
  64
  65         memcpy(chan->ctx, ctx, sizeof(struct lttng_kernel_context));
  66
  67         return 0;
  68
  69 error:
  70         return ret;
  71 }
  72
  73 /*
  74  * Create a new kernel session, register it to the kernel tracer and add it to
  75  * the session daemon session.
  76  */
  77 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  78 {
  79         int ret;
  80         struct ltt_kernel_session *lks;
  81
  82         assert(session);
  83
  84         /* Allocate data structure */
  85         lks = trace_kernel_create_session();
  86         if (lks == NULL) {
  87                 ret = -1;
  88                 goto error;
  89         }
  90
  91         /* Kernel tracer session creation */
  92         ret = kernctl_create_session(tracer_fd);
  93         if (ret < 0) {
  94                 PERROR("ioctl kernel create session");
  95                 goto error;
  96         }
  97
  98         lks->fd = ret;
  99         /* Prevent fd duplication after execlp() */
 100         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 101         if (ret < 0) {
 102                 PERROR("fcntl session fd");
 103         }
 104
 105         lks->id = session->id;
 106         lks->consumer_fds_sent = 0;
 107         session->kernel_session = lks;
 108
 109         DBG("Kernel session created (fd: %d)", lks->fd);
 110
 111         return 0;
 112
 113 error:
 114         if (lks) {
 115                 trace_kernel_destroy_session(lks);
 116         }
 117         return ret;
 118 }
 119
 120 /*
 121  * Create a kernel channel, register it to the kernel tracer and add it to the
 122  * kernel session.
 123  */
 124 int kernel_create_channel(struct ltt_kernel_session *session,
 125                 struct lttng_channel *chan)
 126 {
 127         int ret;
 128         struct ltt_kernel_channel *lkc;
 129
 130         assert(session);
 131         assert(chan);
 132
 133         /* Allocate kernel channel */
 134         lkc = trace_kernel_create_channel(chan);
 135         if (lkc == NULL) {
 136                 goto error;
 137         }
 138
 139         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 140                         chan->name, lkc->channel->attr.overwrite,
 141                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 142                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 143                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 144
 145         /* Kernel tracer channel creation */
 146         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 147         if (ret < 0) {
 148                 PERROR("ioctl kernel create channel");
 149                 goto error;
 150         }
 151
 152         /* Setup the channel fd */
 153         lkc->fd = ret;
 154         /* Prevent fd duplication after execlp() */
 155         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 156         if (ret < 0) {
 157                 PERROR("fcntl session fd");
 158         }
 159
 160         /* Add channel to session */
 161         cds_list_add(&lkc->list, &session->channel_list.head);
 162         session->channel_count++;
 163         lkc->session = session;
 164
 165         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 166
 167         return 0;
 168
 169 error:
 170         if (lkc) {
 171                 free(lkc->channel);
 172                 free(lkc);
 173         }
 174         return -1;
 175 }
 176
 177 /*
 178  * Create a kernel event, enable it to the kernel tracer and add it to the
 179  * channel event list of the kernel session.
 180  */
 181 int kernel_create_event(struct lttng_event *ev,
 182                 struct ltt_kernel_channel *channel)
 183 {
 184         int ret;
 185         struct ltt_kernel_event *event;
 186
 187         assert(ev);
 188         assert(channel);
 189
 190         event = trace_kernel_create_event(ev);
 191         if (event == NULL) {
 192                 ret = -1;
 193                 goto error;
 194         }
 195
 196         ret = kernctl_create_event(channel->fd, event->event);
 197         if (ret < 0) {
 198                 switch (errno) {
 199                 case EEXIST:
 200                         break;
 201                 case ENOSYS:
 202                         WARN("Event type not implemented");
 203                         break;
 204                 case ENOENT:
 205                         WARN("Event %s not found!", ev->name);
 206                         break;
 207                 default:
 208                         PERROR("create event ioctl");
 209                 }
 210                 ret = -errno;
 211                 goto free_event;
 212         }
 213
 214         /*
 215          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 216          */
 217         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 218                 DBG2("Kernel event syscall creation success");
 219                 /*
 220                  * We use fd == -1 to ensure that we never trigger a close of fd
 221                  * 0.
 222                  */
 223                 event->fd = -1;
 224                 goto add_list;
 225         }
 226
 227         event->fd = ret;
 228         /* Prevent fd duplication after execlp() */
 229         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 230         if (ret < 0) {
 231                 PERROR("fcntl session fd");
 232         }
 233
 234 add_list:
 235         /* Add event to event list */
 236         cds_list_add(&event->list, &channel->events_list.head);
 237         channel->event_count++;
 238
 239         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 240
 241         return 0;
 242
 243 free_event:
 244         free(event);
 245 error:
 246         return ret;
 247 }
 248
 249 /*
 250  * Disable a kernel channel.
 251  */
 252 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 253 {
 254         int ret;
 255
 256         assert(chan);
 257
 258         ret = kernctl_disable(chan->fd);
 259         if (ret < 0) {
 260                 PERROR("disable chan ioctl");
 261                 ret = errno;
 262                 goto error;
 263         }
 264
 265         chan->enabled = 0;
 266         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 267
 268         return 0;
 269
 270 error:
 271         return ret;
 272 }
 273
 274 /*
 275  * Enable a kernel channel.
 276  */
 277 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 278 {
 279         int ret;
 280
 281         assert(chan);
 282
 283         ret = kernctl_enable(chan->fd);
 284         if (ret < 0 && errno != EEXIST) {
 285                 PERROR("Enable kernel chan");
 286                 goto error;
 287         }
 288
 289         chan->enabled = 1;
 290         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 291
 292         return 0;
 293
 294 error:
 295         return ret;
 296 }
 297
 298 /*
 299  * Enable a kernel event.
 300  */
 301 int kernel_enable_event(struct ltt_kernel_event *event)
 302 {
 303         int ret;
 304
 305         assert(event);
 306
 307         ret = kernctl_enable(event->fd);
 308         if (ret < 0) {
 309                 switch (errno) {
 310                 case EEXIST:
 311                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 312                         break;
 313                 default:
 314                         PERROR("enable kernel event");
 315                         break;
 316                 }
 317                 goto error;
 318         }
 319
 320         event->enabled = 1;
 321         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 322
 323         return 0;
 324
 325 error:
 326         return ret;
 327 }
 328
 329 /*
 330  * Disable a kernel event.
 331  */
 332 int kernel_disable_event(struct ltt_kernel_event *event)
 333 {
 334         int ret;
 335
 336         assert(event);
 337
 338         ret = kernctl_disable(event->fd);
 339         if (ret < 0) {
 340                 switch (errno) {
 341                 case EEXIST:
 342                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 343                         break;
 344                 default:
 345                         PERROR("disable kernel event");
 346                         break;
 347                 }
 348                 goto error;
 349         }
 350
 351         event->enabled = 0;
 352         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 353
 354         return 0;
 355
 356 error:
 357         return ret;
 358 }
 359
 360 /*
 361  * Create kernel metadata, open from the kernel tracer and add it to the
 362  * kernel session.
 363  */
 364 int kernel_open_metadata(struct ltt_kernel_session *session)
 365 {
 366         int ret;
 367         struct ltt_kernel_metadata *lkm = NULL;
 368
 369         assert(session);
 370
 371         /* Allocate kernel metadata */
 372         lkm = trace_kernel_create_metadata();
 373         if (lkm == NULL) {
 374                 goto error;
 375         }
 376
 377         /* Kernel tracer metadata creation */
 378         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 379         if (ret < 0) {
 380                 goto error_open;
 381         }
 382
 383         lkm->fd = ret;
 384         /* Prevent fd duplication after execlp() */
 385         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 386         if (ret < 0) {
 387                 PERROR("fcntl session fd");
 388         }
 389
 390         session->metadata = lkm;
 391
 392         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 393
 394         return 0;
 395
 396 error_open:
 397         trace_kernel_destroy_metadata(lkm);
 398 error:
 399         return -1;
 400 }
 401
 402 /*
 403  * Start tracing session.
 404  */
 405 int kernel_start_session(struct ltt_kernel_session *session)
 406 {
 407         int ret;
 408
 409         assert(session);
 410
 411         ret = kernctl_start_session(session->fd);
 412         if (ret < 0) {
 413                 PERROR("ioctl start session");
 414                 goto error;
 415         }
 416
 417         DBG("Kernel session started");
 418
 419         return 0;
 420
 421 error:
 422         return ret;
 423 }
 424
 425 /*
 426  * Make a kernel wait to make sure in-flight probe have completed.
 427  */
 428 void kernel_wait_quiescent(int fd)
 429 {
 430         int ret;
 431
 432         DBG("Kernel quiescent wait on %d", fd);
 433
 434         ret = kernctl_wait_quiescent(fd);
 435         if (ret < 0) {
 436                 PERROR("wait quiescent ioctl");
 437                 ERR("Kernel quiescent wait failed");
 438         }
 439 }
 440
 441 /*
 442  * Kernel calibrate
 443  */
 444 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 445 {
 446         int ret;
 447
 448         assert(calibrate);
 449
 450         ret = kernctl_calibrate(fd, calibrate);
 451         if (ret < 0) {
 452                 PERROR("calibrate ioctl");
 453                 return -1;
 454         }
 455
 456         return 0;
 457 }
 458
 459
 460 /*
 461  *  Force flush buffer of metadata.
 462  */
 463 int kernel_metadata_flush_buffer(int fd)
 464 {
 465         int ret;
 466
 467         DBG("Kernel flushing metadata buffer on fd %d", fd);
 468
 469         ret = kernctl_buffer_flush(fd);
 470         if (ret < 0) {
 471                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 472         }
 473
 474         return 0;
 475 }
 476
 477 /*
 478  * Force flush buffer for channel.
 479  */
 480 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 481 {
 482         int ret;
 483         struct ltt_kernel_stream *stream;
 484
 485         assert(channel);
 486
 487         DBG("Flush buffer for channel %s", channel->channel->name);
 488
 489         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 490                 DBG("Flushing channel stream %d", stream->fd);
 491                 ret = kernctl_buffer_flush(stream->fd);
 492                 if (ret < 0) {
 493                         PERROR("ioctl");
 494                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 495                                         stream->fd, ret);
 496                 }
 497         }
 498
 499         return 0;
 500 }
 501
 502 /*
 503  * Stop tracing session.
 504  */
 505 int kernel_stop_session(struct ltt_kernel_session *session)
 506 {
 507         int ret;
 508
 509         assert(session);
 510
 511         ret = kernctl_stop_session(session->fd);
 512         if (ret < 0) {
 513                 goto error;
 514         }
 515
 516         DBG("Kernel session stopped");
 517
 518         return 0;
 519
 520 error:
 521         return ret;
 522 }
 523
 524 /*
 525  * Open stream of channel, register it to the kernel tracer and add it
 526  * to the stream list of the channel.
 527  *
 528  * Return the number of created stream. Else, a negative value.
 529  */
 530 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 531 {
 532         int ret, count = 0;
 533         struct ltt_kernel_stream *lks;
 534
 535         assert(channel);
 536
 537         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 538                 lks = trace_kernel_create_stream(channel->channel->name, count);
 539                 if (lks == NULL) {
 540                         ret = close(ret);
 541                         if (ret) {
 542                                 PERROR("close");
 543                         }
 544                         goto error;
 545                 }
 546
 547                 lks->fd = ret;
 548                 /* Prevent fd duplication after execlp() */
 549                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 550                 if (ret < 0) {
 551                         PERROR("fcntl session fd");
 552                 }
 553
 554                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 555                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 556
 557                 /* Add stream to channe stream list */
 558                 cds_list_add(&lks->list, &channel->stream_list.head);
 559                 channel->stream_count++;
 560
 561                 /* Increment counter which represent CPU number. */
 562                 count++;
 563
 564                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 565                                 lks->state);
 566         }
 567
 568         return channel->stream_count;
 569
 570 error:
 571         return -1;
 572 }
 573
 574 /*
 575  * Open the metadata stream and set it to the kernel session.
 576  */
 577 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 578 {
 579         int ret;
 580
 581         assert(session);
 582
 583         ret = kernctl_create_stream(session->metadata->fd);
 584         if (ret < 0) {
 585                 PERROR("kernel create metadata stream");
 586                 goto error;
 587         }
 588
 589         DBG("Kernel metadata stream created (fd: %d)", ret);
 590         session->metadata_stream_fd = ret;
 591         /* Prevent fd duplication after execlp() */
 592         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 593         if (ret < 0) {
 594                 PERROR("fcntl session fd");
 595         }
 596
 597         return 0;
 598
 599 error:
 600         return -1;
 601 }
 602
 603 /*
 604  * Get the event list from the kernel tracer and return the number of elements.
 605  */
 606 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 607 {
 608         int fd, ret;
 609         char *event;
 610         size_t nbmem, count = 0;
 611         FILE *fp;
 612         struct lttng_event *elist;
 613
 614         assert(events);
 615
 616         fd = kernctl_tracepoint_list(tracer_fd);
 617         if (fd < 0) {
 618                 PERROR("kernel tracepoint list");
 619                 goto error;
 620         }
 621
 622         fp = fdopen(fd, "r");
 623         if (fp == NULL) {
 624                 PERROR("kernel tracepoint list fdopen");
 625                 goto error_fp;
 626         }
 627
 628         /*
 629          * Init memory size counter
 630          * See kernel-ctl.h for explanation of this value
 631          */
 632         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 633         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 634         if (elist == NULL) {
 635                 PERROR("alloc list events");
 636                 count = -ENOMEM;
 637                 goto end;
 638         }
 639
 640         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 641                 if (count >= nbmem) {
 642                         struct lttng_event *new_elist;
 643                         size_t new_nbmem;
 644
 645                         new_nbmem = nbmem << 1;
 646                         DBG("Reallocating event list from %zu to %zu bytes",
 647                                         nbmem, new_nbmem);
 648                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 649                         if (new_elist == NULL) {
 650                                 PERROR("realloc list events");
 651                                 free(event);
 652                                 free(elist);
 653                                 count = -ENOMEM;
 654                                 goto end;
 655                         }
 656                         /* Zero the new memory */
 657                         memset(new_elist + nbmem, 0,
 658                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 659                         nbmem = new_nbmem;
 660                         elist = new_elist;
 661                 }
 662                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 663                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 664                 elist[count].enabled = -1;
 665                 count++;
 666                 free(event);
 667         }
 668
 669         *events = elist;
 670         DBG("Kernel list events done (%zu events)", count);
 671 end:
 672         ret = fclose(fp);       /* closes both fp and fd */
 673         if (ret) {
 674                 PERROR("fclose");
 675         }
 676         return count;
 677
 678 error_fp:
 679         ret = close(fd);
 680         if (ret) {
 681                 PERROR("close");
 682         }
 683 error:
 684         return -1;
 685 }
 686
 687 /*
 688  * Get kernel version and validate it.
 689  */
 690 int kernel_validate_version(int tracer_fd)
 691 {
 692         int ret;
 693         struct lttng_kernel_tracer_version version;
 694
 695         ret = kernctl_tracer_version(tracer_fd, &version);
 696         if (ret < 0) {
 697                 ERR("Failed at getting the lttng-modules version");
 698                 goto error;
 699         }
 700
 701         /* Validate version */
 702         if (version.major != KERN_MODULES_PRE_MAJOR
 703                 && version.major != KERN_MODULES_MAJOR) {
 704                 goto error_version;
 705         }
 706
 707         DBG2("Kernel tracer version validated (major version %d)", version.major);
 708         return 0;
 709
 710 error_version:
 711         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 712                         version.major, KERN_MODULES_MAJOR)
 713         ret = -1;
 714
 715 error:
 716         return ret;
 717 }
 718
 719 /*
 720  * Kernel work-arounds called at the start of sessiond main().
 721  */
 722 int init_kernel_workarounds(void)
 723 {
 724         int ret;
 725         FILE *fp;
 726
 727         /*
 728          * boot_id needs to be read once before being used concurrently
 729          * to deal with a Linux kernel race. A fix is proposed for
 730          * upstream, but the work-around is needed for older kernels.
 731          */
 732         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 733         if (!fp) {
 734                 goto end_boot_id;
 735         }
 736         while (!feof(fp)) {
 737                 char buf[37] = "";
 738
 739                 ret = fread(buf, 1, sizeof(buf), fp);
 740                 if (ret < 0) {
 741                         /* Ignore error, we don't really care */
 742                 }
 743         }
 744         ret = fclose(fp);
 745         if (ret) {
 746                 PERROR("fclose");
 747         }
 748 end_boot_id:
 749         return 0;
 750 }
 751
 752 /*
 753  * Complete teardown of a kernel session.
 754  */
 755 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 756 {
 757         if (ksess == NULL) {
 758                 DBG3("No kernel session when tearing down session");
 759                 return;
 760         }
 761
 762         DBG("Tearing down kernel session");
 763
 764         /*
 765          * Destroy channels on the consumer if at least one FD has been sent and we
 766          * are in no output mode because the streams are in *no* monitor mode so we
 767          * have to send a command to clean them up or else they leaked.
 768          */
 769         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 770                 int ret;
 771                 struct consumer_socket *socket;
 772                 struct lttng_ht_iter iter;
 773
 774                 /* For each consumer socket. */
 775                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 776                                 socket, node.node) {
 777                         struct ltt_kernel_channel *chan;
 778
 779                         /* For each channel, ask the consumer to destroy it. */
 780                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 781                                 ret = kernel_consumer_destroy_channel(socket, chan);
 782                                 if (ret < 0) {
 783                                         /* Consumer is probably dead. Use next socket. */
 784                                         continue;
 785                                 }
 786                         }
 787                 }
 788         }
 789
 790         /* Close any relayd session */
 791         consumer_output_send_destroy_relayd(ksess->consumer);
 792
 793         trace_kernel_destroy_session(ksess);
 794 }
 795
 796 /*
 797  * Destroy a kernel channel object. It does not do anything on the tracer side.
 798  */
 799 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 800 {
 801         struct ltt_kernel_session *ksess = NULL;
 802
 803         assert(kchan);
 804         assert(kchan->channel);
 805
 806         DBG3("Kernel destroy channel %s", kchan->channel->name);
 807
 808         /* Update channel count of associated session. */
 809         if (kchan->session) {
 810                 /* Keep pointer reference so we can update it after the destroy. */
 811                 ksess = kchan->session;
 812         }
 813
 814         trace_kernel_destroy_channel(kchan);
 815
 816         /*
 817          * At this point the kernel channel is not visible anymore. This is safe
 818          * since in order to work on a visible kernel session, the tracing session
 819          * lock (ltt_session.lock) MUST be acquired.
 820          */
 821         if (ksess) {
 822                 ksess->channel_count--;
 823         }
 824 }
 825
 826 /*
 827  * Take a snapshot for a given kernel session.
 828  *
 829  * Return 0 on success or else return a LTTNG_ERR code.
 830  */
 831 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 832                 struct snapshot_output *output, int wait, unsigned int nb_streams)
 833 {
 834         int err, ret, saved_metadata_fd;
 835         struct consumer_socket *socket;
 836         struct lttng_ht_iter iter;
 837         struct ltt_kernel_metadata *saved_metadata;
 838         uint64_t max_size_per_stream = 0;
 839
 840         assert(ksess);
 841         assert(ksess->consumer);
 842         assert(output);
 843
 844         DBG("Kernel snapshot record started");
 845
 846         /* Save current metadata since the following calls will change it. */
 847         saved_metadata = ksess->metadata;
 848         saved_metadata_fd = ksess->metadata_stream_fd;
 849
 850         rcu_read_lock();
 851
 852         ret = kernel_open_metadata(ksess);
 853         if (ret < 0) {
 854                 ret = LTTNG_ERR_KERN_META_FAIL;
 855                 goto error;
 856         }
 857
 858         ret = kernel_open_metadata_stream(ksess);
 859         if (ret < 0) {
 860                 ret = LTTNG_ERR_KERN_META_FAIL;
 861                 goto error_open_stream;
 862         }
 863
 864         if (output->max_size > 0 && nb_streams > 0) {
 865                 max_size_per_stream = output->max_size / nb_streams;
 866         }
 867
 868         /* Send metadata to consumer and snapshot everything. */
 869         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 870                         socket, node.node) {
 871                 struct consumer_output *saved_output;
 872                 struct ltt_kernel_channel *chan;
 873
 874                 /*
 875                  * Temporarly switch consumer output for our snapshot output. As long
 876                  * as the session lock is taken, this is safe.
 877                  */
 878                 saved_output = ksess->consumer;
 879                 ksess->consumer = output->consumer;
 880
 881                 pthread_mutex_lock(socket->lock);
 882                 /* This stream must not be monitored by the consumer. */
 883                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 884                 pthread_mutex_unlock(socket->lock);
 885                 /* Put back the saved consumer output into the session. */
 886                 ksess->consumer = saved_output;
 887                 if (ret < 0) {
 888                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 889                         goto error_consumer;
 890                 }
 891
 892                 /* For each channel, ask the consumer to snapshot it. */
 893                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 894                         if (max_size_per_stream &&
 895                                         chan->channel->attr.subbuf_size > max_size_per_stream) {
 896                                 ret = LTTNG_ERR_INVALID;
 897                                 DBG3("Kernel snapshot record maximum stream size %" PRIu64
 898                                                 " is smaller than subbuffer size of %" PRIu64,
 899                                                 max_size_per_stream, chan->channel->attr.subbuf_size);
 900                                 (void) kernel_consumer_destroy_metadata(socket,
 901                                                 ksess->metadata);
 902                                 goto error_consumer;
 903                         }
 904
 905                         pthread_mutex_lock(socket->lock);
 906                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
 907                                         ksess->uid, ksess->gid,
 908                                         DEFAULT_KERNEL_TRACE_DIR, wait,
 909                                         max_size_per_stream);
 910                         pthread_mutex_unlock(socket->lock);
 911                         if (ret < 0) {
 912                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 913                                 (void) kernel_consumer_destroy_metadata(socket,
 914                                                 ksess->metadata);
 915                                 goto error_consumer;
 916                         }
 917                 }
 918
 919                 /* Snapshot metadata, */
 920                 pthread_mutex_lock(socket->lock);
 921                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
 922                                 1, ksess->uid, ksess->gid,
 923                                 DEFAULT_KERNEL_TRACE_DIR, wait, max_size_per_stream);
 924                 pthread_mutex_unlock(socket->lock);
 925                 if (ret < 0) {
 926                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 927                         goto error_consumer;
 928                 }
 929
 930                 /*
 931                  * The metadata snapshot is done, ask the consumer to destroy it since
 932                  * it's not monitored on the consumer side.
 933                  */
 934                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
 935         }
 936
 937         ret = LTTNG_OK;
 938
 939 error_consumer:
 940         /* Close newly opened metadata stream. It's now on the consumer side. */
 941         err = close(ksess->metadata_stream_fd);
 942         if (err < 0) {
 943                 PERROR("close snapshot kernel");
 944         }
 945
 946 error_open_stream:
 947         trace_kernel_destroy_metadata(ksess->metadata);
 948 error:
 949         /* Restore metadata state.*/
 950         ksess->metadata = saved_metadata;
 951         ksess->metadata_stream_fd = saved_metadata_fd;
 952
 953         rcu_read_unlock();
 954         return ret;
 955 }