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