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  * We own filter_expression and filter.
 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 /*
 356  * Create kernel metadata, open from the kernel tracer and add it to the
 357  * kernel session.
 358  */
 359 int kernel_open_metadata(struct ltt_kernel_session *session)
 360 {
 361         int ret;
 362         struct ltt_kernel_metadata *lkm = NULL;
 363
 364         assert(session);
 365
 366         /* Allocate kernel metadata */
 367         lkm = trace_kernel_create_metadata();
 368         if (lkm == NULL) {
 369                 goto error;
 370         }
 371
 372         /* Kernel tracer metadata creation */
 373         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 374         if (ret < 0) {
 375                 goto error_open;
 376         }
 377
 378         lkm->fd = ret;
 379         /* Prevent fd duplication after execlp() */
 380         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 381         if (ret < 0) {
 382                 PERROR("fcntl session fd");
 383         }
 384
 385         session->metadata = lkm;
 386
 387         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 388
 389         return 0;
 390
 391 error_open:
 392         trace_kernel_destroy_metadata(lkm);
 393 error:
 394         return -1;
 395 }
 396
 397 /*
 398  * Start tracing session.
 399  */
 400 int kernel_start_session(struct ltt_kernel_session *session)
 401 {
 402         int ret;
 403
 404         assert(session);
 405
 406         ret = kernctl_start_session(session->fd);
 407         if (ret < 0) {
 408                 PERROR("ioctl start session");
 409                 goto error;
 410         }
 411
 412         DBG("Kernel session started");
 413
 414         return 0;
 415
 416 error:
 417         return ret;
 418 }
 419
 420 /*
 421  * Make a kernel wait to make sure in-flight probe have completed.
 422  */
 423 void kernel_wait_quiescent(int fd)
 424 {
 425         int ret;
 426
 427         DBG("Kernel quiescent wait on %d", fd);
 428
 429         ret = kernctl_wait_quiescent(fd);
 430         if (ret < 0) {
 431                 PERROR("wait quiescent ioctl");
 432                 ERR("Kernel quiescent wait failed");
 433         }
 434 }
 435
 436 /*
 437  * Kernel calibrate
 438  */
 439 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 440 {
 441         int ret;
 442
 443         assert(calibrate);
 444
 445         ret = kernctl_calibrate(fd, calibrate);
 446         if (ret < 0) {
 447                 PERROR("calibrate ioctl");
 448                 return -1;
 449         }
 450
 451         return 0;
 452 }
 453
 454
 455 /*
 456  *  Force flush buffer of metadata.
 457  */
 458 int kernel_metadata_flush_buffer(int fd)
 459 {
 460         int ret;
 461
 462         DBG("Kernel flushing metadata buffer on fd %d", fd);
 463
 464         ret = kernctl_buffer_flush(fd);
 465         if (ret < 0) {
 466                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 467         }
 468
 469         return 0;
 470 }
 471
 472 /*
 473  * Force flush buffer for channel.
 474  */
 475 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 476 {
 477         int ret;
 478         struct ltt_kernel_stream *stream;
 479
 480         assert(channel);
 481
 482         DBG("Flush buffer for channel %s", channel->channel->name);
 483
 484         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 485                 DBG("Flushing channel stream %d", stream->fd);
 486                 ret = kernctl_buffer_flush(stream->fd);
 487                 if (ret < 0) {
 488                         PERROR("ioctl");
 489                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 490                                         stream->fd, ret);
 491                 }
 492         }
 493
 494         return 0;
 495 }
 496
 497 /*
 498  * Stop tracing session.
 499  */
 500 int kernel_stop_session(struct ltt_kernel_session *session)
 501 {
 502         int ret;
 503
 504         assert(session);
 505
 506         ret = kernctl_stop_session(session->fd);
 507         if (ret < 0) {
 508                 goto error;
 509         }
 510
 511         DBG("Kernel session stopped");
 512
 513         return 0;
 514
 515 error:
 516         return ret;
 517 }
 518
 519 /*
 520  * Open stream of channel, register it to the kernel tracer and add it
 521  * to the stream list of the channel.
 522  *
 523  * Return the number of created stream. Else, a negative value.
 524  */
 525 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 526 {
 527         int ret, count = 0;
 528         struct ltt_kernel_stream *lks;
 529
 530         assert(channel);
 531
 532         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 533                 lks = trace_kernel_create_stream(channel->channel->name, count);
 534                 if (lks == NULL) {
 535                         ret = close(ret);
 536                         if (ret) {
 537                                 PERROR("close");
 538                         }
 539                         goto error;
 540                 }
 541
 542                 lks->fd = ret;
 543                 /* Prevent fd duplication after execlp() */
 544                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 545                 if (ret < 0) {
 546                         PERROR("fcntl session fd");
 547                 }
 548
 549                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 550                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 551
 552                 /* Add stream to channe stream list */
 553                 cds_list_add(&lks->list, &channel->stream_list.head);
 554                 channel->stream_count++;
 555
 556                 /* Increment counter which represent CPU number. */
 557                 count++;
 558
 559                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 560                                 lks->state);
 561         }
 562
 563         return channel->stream_count;
 564
 565 error:
 566         return -1;
 567 }
 568
 569 /*
 570  * Open the metadata stream and set it to the kernel session.
 571  */
 572 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 573 {
 574         int ret;
 575
 576         assert(session);
 577
 578         ret = kernctl_create_stream(session->metadata->fd);
 579         if (ret < 0) {
 580                 PERROR("kernel create metadata stream");
 581                 goto error;
 582         }
 583
 584         DBG("Kernel metadata stream created (fd: %d)", ret);
 585         session->metadata_stream_fd = ret;
 586         /* Prevent fd duplication after execlp() */
 587         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 588         if (ret < 0) {
 589                 PERROR("fcntl session fd");
 590         }
 591
 592         return 0;
 593
 594 error:
 595         return -1;
 596 }
 597
 598 /*
 599  * Get the event list from the kernel tracer and return the number of elements.
 600  */
 601 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 602 {
 603         int fd, ret;
 604         char *event;
 605         size_t nbmem, count = 0;
 606         FILE *fp;
 607         struct lttng_event *elist;
 608
 609         assert(events);
 610
 611         fd = kernctl_tracepoint_list(tracer_fd);
 612         if (fd < 0) {
 613                 PERROR("kernel tracepoint list");
 614                 goto error;
 615         }
 616
 617         fp = fdopen(fd, "r");
 618         if (fp == NULL) {
 619                 PERROR("kernel tracepoint list fdopen");
 620                 goto error_fp;
 621         }
 622
 623         /*
 624          * Init memory size counter
 625          * See kernel-ctl.h for explanation of this value
 626          */
 627         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 628         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 629         if (elist == NULL) {
 630                 PERROR("alloc list events");
 631                 count = -ENOMEM;
 632                 goto end;
 633         }
 634
 635         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 636                 if (count >= nbmem) {
 637                         struct lttng_event *new_elist;
 638                         size_t new_nbmem;
 639
 640                         new_nbmem = nbmem << 1;
 641                         DBG("Reallocating event list from %zu to %zu bytes",
 642                                         nbmem, new_nbmem);
 643                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 644                         if (new_elist == NULL) {
 645                                 PERROR("realloc list events");
 646                                 free(event);
 647                                 free(elist);
 648                                 count = -ENOMEM;
 649                                 goto end;
 650                         }
 651                         /* Zero the new memory */
 652                         memset(new_elist + nbmem, 0,
 653                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 654                         nbmem = new_nbmem;
 655                         elist = new_elist;
 656                 }
 657                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 658                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 659                 elist[count].enabled = -1;
 660                 count++;
 661                 free(event);
 662         }
 663
 664         *events = elist;
 665         DBG("Kernel list events done (%zu events)", count);
 666 end:
 667         ret = fclose(fp);       /* closes both fp and fd */
 668         if (ret) {
 669                 PERROR("fclose");
 670         }
 671         return count;
 672
 673 error_fp:
 674         ret = close(fd);
 675         if (ret) {
 676                 PERROR("close");
 677         }
 678 error:
 679         return -1;
 680 }
 681
 682 /*
 683  * Get kernel version and validate it.
 684  */
 685 int kernel_validate_version(int tracer_fd)
 686 {
 687         int ret;
 688         struct lttng_kernel_tracer_version version;
 689
 690         ret = kernctl_tracer_version(tracer_fd, &version);
 691         if (ret < 0) {
 692                 ERR("Failed at getting the lttng-modules version");
 693                 goto error;
 694         }
 695
 696         /* Validate version */
 697         if (version.major != KERN_MODULES_PRE_MAJOR
 698                 && version.major != KERN_MODULES_MAJOR) {
 699                 goto error_version;
 700         }
 701
 702         DBG2("Kernel tracer version validated (major version %d)", version.major);
 703         return 0;
 704
 705 error_version:
 706         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 707                         version.major, KERN_MODULES_MAJOR)
 708         ret = -1;
 709
 710 error:
 711         return ret;
 712 }
 713
 714 /*
 715  * Kernel work-arounds called at the start of sessiond main().
 716  */
 717 int init_kernel_workarounds(void)
 718 {
 719         int ret;
 720         FILE *fp;
 721
 722         /*
 723          * boot_id needs to be read once before being used concurrently
 724          * to deal with a Linux kernel race. A fix is proposed for
 725          * upstream, but the work-around is needed for older kernels.
 726          */
 727         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 728         if (!fp) {
 729                 goto end_boot_id;
 730         }
 731         while (!feof(fp)) {
 732                 char buf[37] = "";
 733
 734                 ret = fread(buf, 1, sizeof(buf), fp);
 735                 if (ret < 0) {
 736                         /* Ignore error, we don't really care */
 737                 }
 738         }
 739         ret = fclose(fp);
 740         if (ret) {
 741                 PERROR("fclose");
 742         }
 743 end_boot_id:
 744         return 0;
 745 }
 746
 747 /*
 748  * Complete teardown of a kernel session.
 749  */
 750 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 751 {
 752         if (ksess == NULL) {
 753                 DBG3("No kernel session when tearing down session");
 754                 return;
 755         }
 756
 757         DBG("Tearing down kernel session");
 758
 759         /*
 760          * Destroy channels on the consumer if at least one FD has been sent and we
 761          * are in no output mode because the streams are in *no* monitor mode so we
 762          * have to send a command to clean them up or else they leaked.
 763          */
 764         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 765                 int ret;
 766                 struct consumer_socket *socket;
 767                 struct lttng_ht_iter iter;
 768
 769                 /* For each consumer socket. */
 770                 rcu_read_lock();
 771                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 772                                 socket, node.node) {
 773                         struct ltt_kernel_channel *chan;
 774
 775                         /* For each channel, ask the consumer to destroy it. */
 776                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 777                                 ret = kernel_consumer_destroy_channel(socket, chan);
 778                                 if (ret < 0) {
 779                                         /* Consumer is probably dead. Use next socket. */
 780                                         continue;
 781                                 }
 782                         }
 783                 }
 784                 rcu_read_unlock();
 785         }
 786
 787         /* Close any relayd session */
 788         consumer_output_send_destroy_relayd(ksess->consumer);
 789
 790         trace_kernel_destroy_session(ksess);
 791 }
 792
 793 /*
 794  * Destroy a kernel channel object. It does not do anything on the tracer side.
 795  */
 796 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 797 {
 798         struct ltt_kernel_session *ksess = NULL;
 799
 800         assert(kchan);
 801         assert(kchan->channel);
 802
 803         DBG3("Kernel destroy channel %s", kchan->channel->name);
 804
 805         /* Update channel count of associated session. */
 806         if (kchan->session) {
 807                 /* Keep pointer reference so we can update it after the destroy. */
 808                 ksess = kchan->session;
 809         }
 810
 811         trace_kernel_destroy_channel(kchan);
 812
 813         /*
 814          * At this point the kernel channel is not visible anymore. This is safe
 815          * since in order to work on a visible kernel session, the tracing session
 816          * lock (ltt_session.lock) MUST be acquired.
 817          */
 818         if (ksess) {
 819                 ksess->channel_count--;
 820         }
 821 }
 822
 823 /*
 824  * Take a snapshot for a given kernel session.
 825  *
 826  * Return 0 on success or else return a LTTNG_ERR code.
 827  */
 828 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 829                 struct snapshot_output *output, int wait,
 830                 uint64_t nb_packets_per_stream)
 831 {
 832         int err, ret, saved_metadata_fd;
 833         struct consumer_socket *socket;
 834         struct lttng_ht_iter iter;
 835         struct ltt_kernel_metadata *saved_metadata;
 836
 837         assert(ksess);
 838         assert(ksess->consumer);
 839         assert(output);
 840
 841         DBG("Kernel snapshot record started");
 842
 843         /* Save current metadata since the following calls will change it. */
 844         saved_metadata = ksess->metadata;
 845         saved_metadata_fd = ksess->metadata_stream_fd;
 846
 847         rcu_read_lock();
 848
 849         ret = kernel_open_metadata(ksess);
 850         if (ret < 0) {
 851                 ret = LTTNG_ERR_KERN_META_FAIL;
 852                 goto error;
 853         }
 854
 855         ret = kernel_open_metadata_stream(ksess);
 856         if (ret < 0) {
 857                 ret = LTTNG_ERR_KERN_META_FAIL;
 858                 goto error_open_stream;
 859         }
 860
 861         /* Send metadata to consumer and snapshot everything. */
 862         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 863                         socket, node.node) {
 864                 struct consumer_output *saved_output;
 865                 struct ltt_kernel_channel *chan;
 866
 867                 /*
 868                  * Temporarly switch consumer output for our snapshot output. As long
 869                  * as the session lock is taken, this is safe.
 870                  */
 871                 saved_output = ksess->consumer;
 872                 ksess->consumer = output->consumer;
 873
 874                 pthread_mutex_lock(socket->lock);
 875                 /* This stream must not be monitored by the consumer. */
 876                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 877                 pthread_mutex_unlock(socket->lock);
 878                 /* Put back the saved consumer output into the session. */
 879                 ksess->consumer = saved_output;
 880                 if (ret < 0) {
 881                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 882                         goto error_consumer;
 883                 }
 884
 885                 /* For each channel, ask the consumer to snapshot it. */
 886                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 887                         pthread_mutex_lock(socket->lock);
 888                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
 889                                         ksess->uid, ksess->gid,
 890                                         DEFAULT_KERNEL_TRACE_DIR, wait,
 891                                         nb_packets_per_stream);
 892                         pthread_mutex_unlock(socket->lock);
 893                         if (ret < 0) {
 894                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 895                                 (void) kernel_consumer_destroy_metadata(socket,
 896                                                 ksess->metadata);
 897                                 goto error_consumer;
 898                         }
 899                 }
 900
 901                 /* Snapshot metadata, */
 902                 pthread_mutex_lock(socket->lock);
 903                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
 904                                 1, ksess->uid, ksess->gid,
 905                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0);
 906                 pthread_mutex_unlock(socket->lock);
 907                 if (ret < 0) {
 908                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 909                         goto error_consumer;
 910                 }
 911
 912                 /*
 913                  * The metadata snapshot is done, ask the consumer to destroy it since
 914                  * it's not monitored on the consumer side.
 915                  */
 916                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
 917         }
 918
 919         ret = LTTNG_OK;
 920
 921 error_consumer:
 922         /* Close newly opened metadata stream. It's now on the consumer side. */
 923         err = close(ksess->metadata_stream_fd);
 924         if (err < 0) {
 925                 PERROR("close snapshot kernel");
 926         }
 927
 928 error_open_stream:
 929         trace_kernel_destroy_metadata(ksess->metadata);
 930 error:
 931         /* Restore metadata state.*/
 932         ksess->metadata = saved_metadata;
 933         ksess->metadata_stream_fd = saved_metadata_fd;
 934
 935         rcu_read_unlock();
 936         return ret;
 937 }