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/kernel-ctl/kernel-ioctl.h>
  30 #include <common/sessiond-comm/sessiond-comm.h>
  31
  32 #include "consumer.h"
  33 #include "kernel.h"
  34 #include "kernel-consumer.h"
  35 #include "kern-modules.h"
  36 #include "utils.h"
  37
  38 /*
  39  * Add context on a kernel channel.
  40  */
  41 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  42                 struct ltt_kernel_context *ctx)
  43 {
  44         int ret;
  45
  46         assert(chan);
  47         assert(ctx);
  48
  49         DBG("Adding context to channel %s", chan->channel->name);
  50         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  51         if (ret < 0) {
  52                 if (errno != EEXIST) {
  53                         PERROR("add context ioctl");
  54                 } else {
  55                         /* If EEXIST, we just ignore the error */
  56                         ret = 0;
  57                 }
  58                 goto error;
  59         }
  60
  61         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  62
  63         return 0;
  64
  65 error:
  66         return ret;
  67 }
  68
  69 /*
  70  * Create a new kernel session, register it to the kernel tracer and add it to
  71  * the session daemon session.
  72  */
  73 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  74 {
  75         int ret;
  76         struct ltt_kernel_session *lks;
  77
  78         assert(session);
  79
  80         /* Allocate data structure */
  81         lks = trace_kernel_create_session();
  82         if (lks == NULL) {
  83                 ret = -1;
  84                 goto error;
  85         }
  86
  87         /* Kernel tracer session creation */
  88         ret = kernctl_create_session(tracer_fd);
  89         if (ret < 0) {
  90                 PERROR("ioctl kernel create session");
  91                 goto error;
  92         }
  93
  94         lks->fd = ret;
  95         /* Prevent fd duplication after execlp() */
  96         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
  97         if (ret < 0) {
  98                 PERROR("fcntl session fd");
  99         }
 100
 101         lks->id = session->id;
 102         lks->consumer_fds_sent = 0;
 103         session->kernel_session = lks;
 104
 105         DBG("Kernel session created (fd: %d)", lks->fd);
 106
 107         return 0;
 108
 109 error:
 110         if (lks) {
 111                 trace_kernel_destroy_session(lks);
 112         }
 113         return ret;
 114 }
 115
 116 /*
 117  * Create a kernel channel, register it to the kernel tracer and add it to the
 118  * kernel session.
 119  */
 120 int kernel_create_channel(struct ltt_kernel_session *session,
 121                 struct lttng_channel *chan)
 122 {
 123         int ret;
 124         struct ltt_kernel_channel *lkc;
 125
 126         assert(session);
 127         assert(chan);
 128
 129         /* Allocate kernel channel */
 130         lkc = trace_kernel_create_channel(chan);
 131         if (lkc == NULL) {
 132                 goto error;
 133         }
 134
 135         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 136                         chan->name, lkc->channel->attr.overwrite,
 137                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 138                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 139                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 140
 141         /* Kernel tracer channel creation */
 142         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 143         if (ret < 0) {
 144                 PERROR("ioctl kernel create channel");
 145                 goto error;
 146         }
 147
 148         /* Setup the channel fd */
 149         lkc->fd = ret;
 150         /* Prevent fd duplication after execlp() */
 151         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 152         if (ret < 0) {
 153                 PERROR("fcntl session fd");
 154         }
 155
 156         /* Add channel to session */
 157         cds_list_add(&lkc->list, &session->channel_list.head);
 158         session->channel_count++;
 159         lkc->session = session;
 160
 161         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 162
 163         return 0;
 164
 165 error:
 166         if (lkc) {
 167                 free(lkc->channel);
 168                 free(lkc);
 169         }
 170         return -1;
 171 }
 172
 173 /*
 174  * Create a kernel event, enable it to the kernel tracer and add it to the
 175  * channel event list of the kernel session.
 176  * We own filter_expression and filter.
 177  */
 178 int kernel_create_event(struct lttng_event *ev,
 179                 struct ltt_kernel_channel *channel)
 180 {
 181         int ret;
 182         struct ltt_kernel_event *event;
 183
 184         assert(ev);
 185         assert(channel);
 186
 187         event = trace_kernel_create_event(ev);
 188         if (event == NULL) {
 189                 ret = -1;
 190                 goto error;
 191         }
 192
 193         ret = kernctl_create_event(channel->fd, event->event);
 194         if (ret < 0) {
 195                 switch (errno) {
 196                 case EEXIST:
 197                         break;
 198                 case ENOSYS:
 199                         WARN("Event type not implemented");
 200                         break;
 201                 case ENOENT:
 202                         WARN("Event %s not found!", ev->name);
 203                         break;
 204                 default:
 205                         PERROR("create event ioctl");
 206                 }
 207                 ret = -errno;
 208                 goto free_event;
 209         }
 210
 211         /*
 212          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 213          */
 214         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 215                 DBG2("Kernel event syscall creation success");
 216                 /*
 217                  * We use fd == -1 to ensure that we never trigger a close of fd
 218                  * 0.
 219                  */
 220                 event->fd = -1;
 221                 goto add_list;
 222         }
 223
 224         event->fd = ret;
 225         /* Prevent fd duplication after execlp() */
 226         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 227         if (ret < 0) {
 228                 PERROR("fcntl session fd");
 229         }
 230
 231 add_list:
 232         /* Add event to event list */
 233         cds_list_add(&event->list, &channel->events_list.head);
 234         channel->event_count++;
 235
 236         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 237
 238         return 0;
 239
 240 free_event:
 241         free(event);
 242 error:
 243         return ret;
 244 }
 245
 246 /*
 247  * Disable a kernel channel.
 248  */
 249 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 250 {
 251         int ret;
 252
 253         assert(chan);
 254
 255         ret = kernctl_disable(chan->fd);
 256         if (ret < 0) {
 257                 PERROR("disable chan ioctl");
 258                 ret = errno;
 259                 goto error;
 260         }
 261
 262         chan->enabled = 0;
 263         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 264
 265         return 0;
 266
 267 error:
 268         return ret;
 269 }
 270
 271 /*
 272  * Enable a kernel channel.
 273  */
 274 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 275 {
 276         int ret;
 277
 278         assert(chan);
 279
 280         ret = kernctl_enable(chan->fd);
 281         if (ret < 0 && errno != EEXIST) {
 282                 PERROR("Enable kernel chan");
 283                 goto error;
 284         }
 285
 286         chan->enabled = 1;
 287         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 288
 289         return 0;
 290
 291 error:
 292         return ret;
 293 }
 294
 295 /*
 296  * Enable a kernel event.
 297  */
 298 int kernel_enable_event(struct ltt_kernel_event *event)
 299 {
 300         int ret;
 301
 302         assert(event);
 303
 304         ret = kernctl_enable(event->fd);
 305         if (ret < 0) {
 306                 switch (errno) {
 307                 case EEXIST:
 308                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 309                         break;
 310                 default:
 311                         PERROR("enable kernel event");
 312                         break;
 313                 }
 314                 goto error;
 315         }
 316
 317         event->enabled = 1;
 318         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 319
 320         return 0;
 321
 322 error:
 323         return ret;
 324 }
 325
 326 /*
 327  * Disable a kernel event.
 328  */
 329 int kernel_disable_event(struct ltt_kernel_event *event)
 330 {
 331         int ret;
 332
 333         assert(event);
 334
 335         ret = kernctl_disable(event->fd);
 336         if (ret < 0) {
 337                 switch (errno) {
 338                 case EEXIST:
 339                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 340                         break;
 341                 default:
 342                         PERROR("disable kernel event");
 343                         break;
 344                 }
 345                 goto error;
 346         }
 347
 348         event->enabled = 0;
 349         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 350
 351         return 0;
 352
 353 error:
 354         return ret;
 355 }
 356
 357 int kernel_enable_syscall(const char *syscall_name,
 358                 struct ltt_kernel_channel *channel)
 359 {
 360         return kernctl_enable_syscall(channel->fd, syscall_name);
 361 }
 362
 363 int kernel_disable_syscall(const char *syscall_name,
 364                 struct ltt_kernel_channel *channel)
 365 {
 366         return kernctl_disable_syscall(channel->fd, syscall_name);
 367 }
 368
 369 /*
 370  * Create kernel metadata, open from the kernel tracer and add it to the
 371  * kernel session.
 372  */
 373 int kernel_open_metadata(struct ltt_kernel_session *session)
 374 {
 375         int ret;
 376         struct ltt_kernel_metadata *lkm = NULL;
 377
 378         assert(session);
 379
 380         /* Allocate kernel metadata */
 381         lkm = trace_kernel_create_metadata();
 382         if (lkm == NULL) {
 383                 goto error;
 384         }
 385
 386         /* Kernel tracer metadata creation */
 387         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 388         if (ret < 0) {
 389                 goto error_open;
 390         }
 391
 392         lkm->fd = ret;
 393         /* Prevent fd duplication after execlp() */
 394         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 395         if (ret < 0) {
 396                 PERROR("fcntl session fd");
 397         }
 398
 399         session->metadata = lkm;
 400
 401         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 402
 403         return 0;
 404
 405 error_open:
 406         trace_kernel_destroy_metadata(lkm);
 407 error:
 408         return -1;
 409 }
 410
 411 /*
 412  * Start tracing session.
 413  */
 414 int kernel_start_session(struct ltt_kernel_session *session)
 415 {
 416         int ret;
 417
 418         assert(session);
 419
 420         ret = kernctl_start_session(session->fd);
 421         if (ret < 0) {
 422                 PERROR("ioctl start session");
 423                 goto error;
 424         }
 425
 426         DBG("Kernel session started");
 427
 428         return 0;
 429
 430 error:
 431         return ret;
 432 }
 433
 434 /*
 435  * Make a kernel wait to make sure in-flight probe have completed.
 436  */
 437 void kernel_wait_quiescent(int fd)
 438 {
 439         int ret;
 440
 441         DBG("Kernel quiescent wait on %d", fd);
 442
 443         ret = kernctl_wait_quiescent(fd);
 444         if (ret < 0) {
 445                 PERROR("wait quiescent ioctl");
 446                 ERR("Kernel quiescent wait failed");
 447         }
 448 }
 449
 450 /*
 451  * Kernel calibrate
 452  */
 453 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 454 {
 455         int ret;
 456
 457         assert(calibrate);
 458
 459         ret = kernctl_calibrate(fd, calibrate);
 460         if (ret < 0) {
 461                 PERROR("calibrate ioctl");
 462                 return -1;
 463         }
 464
 465         return 0;
 466 }
 467
 468
 469 /*
 470  *  Force flush buffer of metadata.
 471  */
 472 int kernel_metadata_flush_buffer(int fd)
 473 {
 474         int ret;
 475
 476         DBG("Kernel flushing metadata buffer on fd %d", fd);
 477
 478         ret = kernctl_buffer_flush(fd);
 479         if (ret < 0) {
 480                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 481         }
 482
 483         return 0;
 484 }
 485
 486 /*
 487  * Force flush buffer for channel.
 488  */
 489 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 490 {
 491         int ret;
 492         struct ltt_kernel_stream *stream;
 493
 494         assert(channel);
 495
 496         DBG("Flush buffer for channel %s", channel->channel->name);
 497
 498         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 499                 DBG("Flushing channel stream %d", stream->fd);
 500                 ret = kernctl_buffer_flush(stream->fd);
 501                 if (ret < 0) {
 502                         PERROR("ioctl");
 503                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 504                                         stream->fd, ret);
 505                 }
 506         }
 507
 508         return 0;
 509 }
 510
 511 /*
 512  * Stop tracing session.
 513  */
 514 int kernel_stop_session(struct ltt_kernel_session *session)
 515 {
 516         int ret;
 517
 518         assert(session);
 519
 520         ret = kernctl_stop_session(session->fd);
 521         if (ret < 0) {
 522                 goto error;
 523         }
 524
 525         DBG("Kernel session stopped");
 526
 527         return 0;
 528
 529 error:
 530         return ret;
 531 }
 532
 533 /*
 534  * Open stream of channel, register it to the kernel tracer and add it
 535  * to the stream list of the channel.
 536  *
 537  * Return the number of created stream. Else, a negative value.
 538  */
 539 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 540 {
 541         int ret, count = 0;
 542         struct ltt_kernel_stream *lks;
 543
 544         assert(channel);
 545
 546         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 547                 lks = trace_kernel_create_stream(channel->channel->name, count);
 548                 if (lks == NULL) {
 549                         ret = close(ret);
 550                         if (ret) {
 551                                 PERROR("close");
 552                         }
 553                         goto error;
 554                 }
 555
 556                 lks->fd = ret;
 557                 /* Prevent fd duplication after execlp() */
 558                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 559                 if (ret < 0) {
 560                         PERROR("fcntl session fd");
 561                 }
 562
 563                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 564                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 565
 566                 /* Add stream to channe stream list */
 567                 cds_list_add(&lks->list, &channel->stream_list.head);
 568                 channel->stream_count++;
 569
 570                 /* Increment counter which represent CPU number. */
 571                 count++;
 572
 573                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 574                                 lks->state);
 575         }
 576
 577         return channel->stream_count;
 578
 579 error:
 580         return -1;
 581 }
 582
 583 /*
 584  * Open the metadata stream and set it to the kernel session.
 585  */
 586 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 587 {
 588         int ret;
 589
 590         assert(session);
 591
 592         ret = kernctl_create_stream(session->metadata->fd);
 593         if (ret < 0) {
 594                 PERROR("kernel create metadata stream");
 595                 goto error;
 596         }
 597
 598         DBG("Kernel metadata stream created (fd: %d)", ret);
 599         session->metadata_stream_fd = ret;
 600         /* Prevent fd duplication after execlp() */
 601         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 602         if (ret < 0) {
 603                 PERROR("fcntl session fd");
 604         }
 605
 606         return 0;
 607
 608 error:
 609         return -1;
 610 }
 611
 612 /*
 613  * Get the event list from the kernel tracer and return the number of elements.
 614  */
 615 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 616 {
 617         int fd, ret;
 618         char *event;
 619         size_t nbmem, count = 0;
 620         FILE *fp;
 621         struct lttng_event *elist;
 622
 623         assert(events);
 624
 625         fd = kernctl_tracepoint_list(tracer_fd);
 626         if (fd < 0) {
 627                 PERROR("kernel tracepoint list");
 628                 goto error;
 629         }
 630
 631         fp = fdopen(fd, "r");
 632         if (fp == NULL) {
 633                 PERROR("kernel tracepoint list fdopen");
 634                 goto error_fp;
 635         }
 636
 637         /*
 638          * Init memory size counter
 639          * See kernel-ctl.h for explanation of this value
 640          */
 641         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 642         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 643         if (elist == NULL) {
 644                 PERROR("alloc list events");
 645                 count = -ENOMEM;
 646                 goto end;
 647         }
 648
 649         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 650                 if (count >= nbmem) {
 651                         struct lttng_event *new_elist;
 652                         size_t new_nbmem;
 653
 654                         new_nbmem = nbmem << 1;
 655                         DBG("Reallocating event list from %zu to %zu bytes",
 656                                         nbmem, new_nbmem);
 657                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 658                         if (new_elist == NULL) {
 659                                 PERROR("realloc list events");
 660                                 free(event);
 661                                 free(elist);
 662                                 count = -ENOMEM;
 663                                 goto end;
 664                         }
 665                         /* Zero the new memory */
 666                         memset(new_elist + nbmem, 0,
 667                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 668                         nbmem = new_nbmem;
 669                         elist = new_elist;
 670                 }
 671                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 672                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 673                 elist[count].enabled = -1;
 674                 count++;
 675                 free(event);
 676         }
 677
 678         *events = elist;
 679         DBG("Kernel list events done (%zu events)", count);
 680 end:
 681         ret = fclose(fp);       /* closes both fp and fd */
 682         if (ret) {
 683                 PERROR("fclose");
 684         }
 685         return count;
 686
 687 error_fp:
 688         ret = close(fd);
 689         if (ret) {
 690                 PERROR("close");
 691         }
 692 error:
 693         return -1;
 694 }
 695
 696 /*
 697  * Get kernel version and validate it.
 698  */
 699 int kernel_validate_version(int tracer_fd)
 700 {
 701         int ret;
 702         struct lttng_kernel_tracer_version version;
 703         struct lttng_kernel_tracer_abi_version abi_version;
 704
 705         ret = kernctl_tracer_version(tracer_fd, &version);
 706         if (ret < 0) {
 707                 ERR("Failed at getting the lttng-modules version");
 708                 goto error;
 709         }
 710
 711         /* Validate version */
 712         if (version.major != VERSION_MAJOR) {
 713                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
 714                         version.major, VERSION_MAJOR);
 715                 goto error_version;
 716         }
 717         ret = kernctl_tracer_abi_version(tracer_fd, &abi_version);
 718         if (ret < 0) {
 719                 ERR("Failed at getting lttng-modules ABI version");
 720                 goto error;
 721         }
 722         if (abi_version.major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
 723                 ERR("Kernel tracer ABI version (%d.%d) is not compatible with expected ABI major version (%d.*)",
 724                         abi_version.major, abi_version.minor,
 725                         LTTNG_MODULES_ABI_MAJOR_VERSION);
 726                 goto error;
 727         }
 728         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
 729                         version.major, version.minor,
 730                         abi_version.major, abi_version.minor);
 731         return 0;
 732
 733 error_version:
 734         ret = -1;
 735
 736 error:
 737         return ret;
 738 }
 739
 740 /*
 741  * Kernel work-arounds called at the start of sessiond main().
 742  */
 743 int init_kernel_workarounds(void)
 744 {
 745         int ret;
 746         FILE *fp;
 747
 748         /*
 749          * boot_id needs to be read once before being used concurrently
 750          * to deal with a Linux kernel race. A fix is proposed for
 751          * upstream, but the work-around is needed for older kernels.
 752          */
 753         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 754         if (!fp) {
 755                 goto end_boot_id;
 756         }
 757         while (!feof(fp)) {
 758                 char buf[37] = "";
 759
 760                 ret = fread(buf, 1, sizeof(buf), fp);
 761                 if (ret < 0) {
 762                         /* Ignore error, we don't really care */
 763                 }
 764         }
 765         ret = fclose(fp);
 766         if (ret) {
 767                 PERROR("fclose");
 768         }
 769 end_boot_id:
 770         return 0;
 771 }
 772
 773 /*
 774  * Complete teardown of a kernel session.
 775  */
 776 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 777 {
 778         if (ksess == NULL) {
 779                 DBG3("No kernel session when tearing down session");
 780                 return;
 781         }
 782
 783         DBG("Tearing down kernel session");
 784
 785         /*
 786          * Destroy channels on the consumer if at least one FD has been sent and we
 787          * are in no output mode because the streams are in *no* monitor mode so we
 788          * have to send a command to clean them up or else they leaked.
 789          */
 790         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 791                 int ret;
 792                 struct consumer_socket *socket;
 793                 struct lttng_ht_iter iter;
 794
 795                 /* For each consumer socket. */
 796                 rcu_read_lock();
 797                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 798                                 socket, node.node) {
 799                         struct ltt_kernel_channel *chan;
 800
 801                         /* For each channel, ask the consumer to destroy it. */
 802                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 803                                 ret = kernel_consumer_destroy_channel(socket, chan);
 804                                 if (ret < 0) {
 805                                         /* Consumer is probably dead. Use next socket. */
 806                                         continue;
 807                                 }
 808                         }
 809                 }
 810                 rcu_read_unlock();
 811         }
 812
 813         /* Close any relayd session */
 814         consumer_output_send_destroy_relayd(ksess->consumer);
 815
 816         trace_kernel_destroy_session(ksess);
 817 }
 818
 819 /*
 820  * Destroy a kernel channel object. It does not do anything on the tracer side.
 821  */
 822 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 823 {
 824         struct ltt_kernel_session *ksess = NULL;
 825
 826         assert(kchan);
 827         assert(kchan->channel);
 828
 829         DBG3("Kernel destroy channel %s", kchan->channel->name);
 830
 831         /* Update channel count of associated session. */
 832         if (kchan->session) {
 833                 /* Keep pointer reference so we can update it after the destroy. */
 834                 ksess = kchan->session;
 835         }
 836
 837         trace_kernel_destroy_channel(kchan);
 838
 839         /*
 840          * At this point the kernel channel is not visible anymore. This is safe
 841          * since in order to work on a visible kernel session, the tracing session
 842          * lock (ltt_session.lock) MUST be acquired.
 843          */
 844         if (ksess) {
 845                 ksess->channel_count--;
 846         }
 847 }
 848
 849 /*
 850  * Take a snapshot for a given kernel session.
 851  *
 852  * Return 0 on success or else return a LTTNG_ERR code.
 853  */
 854 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 855                 struct snapshot_output *output, int wait,
 856                 uint64_t nb_packets_per_stream)
 857 {
 858         int err, ret, saved_metadata_fd;
 859         struct consumer_socket *socket;
 860         struct lttng_ht_iter iter;
 861         struct ltt_kernel_metadata *saved_metadata;
 862
 863         assert(ksess);
 864         assert(ksess->consumer);
 865         assert(output);
 866
 867         DBG("Kernel snapshot record started");
 868
 869         /* Save current metadata since the following calls will change it. */
 870         saved_metadata = ksess->metadata;
 871         saved_metadata_fd = ksess->metadata_stream_fd;
 872
 873         rcu_read_lock();
 874
 875         ret = kernel_open_metadata(ksess);
 876         if (ret < 0) {
 877                 ret = LTTNG_ERR_KERN_META_FAIL;
 878                 goto error;
 879         }
 880
 881         ret = kernel_open_metadata_stream(ksess);
 882         if (ret < 0) {
 883                 ret = LTTNG_ERR_KERN_META_FAIL;
 884                 goto error_open_stream;
 885         }
 886
 887         /* Send metadata to consumer and snapshot everything. */
 888         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 889                         socket, node.node) {
 890                 struct consumer_output *saved_output;
 891                 struct ltt_kernel_channel *chan;
 892
 893                 /*
 894                  * Temporarly switch consumer output for our snapshot output. As long
 895                  * as the session lock is taken, this is safe.
 896                  */
 897                 saved_output = ksess->consumer;
 898                 ksess->consumer = output->consumer;
 899
 900                 pthread_mutex_lock(socket->lock);
 901                 /* This stream must not be monitored by the consumer. */
 902                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 903                 pthread_mutex_unlock(socket->lock);
 904                 /* Put back the saved consumer output into the session. */
 905                 ksess->consumer = saved_output;
 906                 if (ret < 0) {
 907                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 908                         goto error_consumer;
 909                 }
 910
 911                 /* For each channel, ask the consumer to snapshot it. */
 912                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 913                         pthread_mutex_lock(socket->lock);
 914                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
 915                                         ksess->uid, ksess->gid,
 916                                         DEFAULT_KERNEL_TRACE_DIR, wait,
 917                                         nb_packets_per_stream);
 918                         pthread_mutex_unlock(socket->lock);
 919                         if (ret < 0) {
 920                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 921                                 (void) kernel_consumer_destroy_metadata(socket,
 922                                                 ksess->metadata);
 923                                 goto error_consumer;
 924                         }
 925                 }
 926
 927                 /* Snapshot metadata, */
 928                 pthread_mutex_lock(socket->lock);
 929                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
 930                                 1, ksess->uid, ksess->gid,
 931                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0);
 932                 pthread_mutex_unlock(socket->lock);
 933                 if (ret < 0) {
 934                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 935                         goto error_consumer;
 936                 }
 937
 938                 /*
 939                  * The metadata snapshot is done, ask the consumer to destroy it since
 940                  * it's not monitored on the consumer side.
 941                  */
 942                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
 943         }
 944
 945         ret = LTTNG_OK;
 946
 947 error_consumer:
 948         /* Close newly opened metadata stream. It's now on the consumer side. */
 949         err = close(ksess->metadata_stream_fd);
 950         if (err < 0) {
 951                 PERROR("close snapshot kernel");
 952         }
 953
 954 error_open_stream:
 955         trace_kernel_destroy_metadata(ksess->metadata);
 956 error:
 957         /* Restore metadata state.*/
 958         ksess->metadata = saved_metadata;
 959         ksess->metadata_stream_fd = saved_metadata_fd;
 960
 961         rcu_read_unlock();
 962         return ret;
 963 }
 964
 965 /*
 966  * Get the syscall mask array from the kernel tracer.
 967  *
 968  * Return 0 on success else a negative value. In both case, syscall_mask should
 969  * be freed.
 970  */
 971 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
 972 {
 973         assert(syscall_mask);
 974         assert(nr_bits);
 975
 976         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
 977 }