src/bin/lttng-sessiond/kernel.c

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