src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _GNU_SOURCE
  19 #include <errno.h>
  20 #include <fcntl.h>
  21 #include <stdlib.h>
  22 #include <stdio.h>
  23 #include <string.h>
  24 #include <unistd.h>
  25 #include <inttypes.h>
  26
  27 #include <common/common.h>
  28 #include <common/kernel-ctl/kernel-ctl.h>
  29 #include <common/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kernel-consumer.h"
  34 #include "kern-modules.h"
  35
  36 /*
  37  * Add context on a kernel channel.
  38  */
  39 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  40                 struct lttng_kernel_context *ctx)
  41 {
  42         int ret;
  43
  44         assert(chan);
  45         assert(ctx);
  46
  47         DBG("Adding context to channel %s", chan->channel->name);
  48         ret = kernctl_add_context(chan->fd, ctx);
  49         if (ret < 0) {
  50                 if (errno != EEXIST) {
  51                         PERROR("add context ioctl");
  52                 } else {
  53                         /* If EEXIST, we just ignore the error */
  54                         ret = 0;
  55                 }
  56                 goto error;
  57         }
  58
  59         chan->ctx = zmalloc(sizeof(struct lttng_kernel_context));
  60         if (chan->ctx == NULL) {
  61                 PERROR("zmalloc event context");
  62                 goto error;
  63         }
  64
  65         memcpy(chan->ctx, ctx, sizeof(struct lttng_kernel_context));
  66
  67         return 0;
  68
  69 error:
  70         return ret;
  71 }
  72
  73 /*
  74  * Create a new kernel session, register it to the kernel tracer and add it to
  75  * the session daemon session.
  76  */
  77 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  78 {
  79         int ret;
  80         struct ltt_kernel_session *lks;
  81
  82         assert(session);
  83
  84         /* Allocate data structure */
  85         lks = trace_kernel_create_session();
  86         if (lks == NULL) {
  87                 ret = -1;
  88                 goto error;
  89         }
  90
  91         /* Kernel tracer session creation */
  92         ret = kernctl_create_session(tracer_fd);
  93         if (ret < 0) {
  94                 PERROR("ioctl kernel create session");
  95                 goto error;
  96         }
  97
  98         lks->fd = ret;
  99         /* Prevent fd duplication after execlp() */
 100         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 101         if (ret < 0) {
 102                 PERROR("fcntl session fd");
 103         }
 104
 105         lks->id = session->id;
 106         lks->consumer_fds_sent = 0;
 107         session->kernel_session = lks;
 108
 109         DBG("Kernel session created (fd: %d)", lks->fd);
 110
 111         return 0;
 112
 113 error:
 114         if (lks) {
 115                 trace_kernel_destroy_session(lks);
 116         }
 117         return ret;
 118 }
 119
 120 /*
 121  * Create a kernel channel, register it to the kernel tracer and add it to the
 122  * kernel session.
 123  */
 124 int kernel_create_channel(struct ltt_kernel_session *session,
 125                 struct lttng_channel *chan)
 126 {
 127         int ret;
 128         struct ltt_kernel_channel *lkc;
 129
 130         assert(session);
 131         assert(chan);
 132
 133         /* Allocate kernel channel */
 134         lkc = trace_kernel_create_channel(chan);
 135         if (lkc == NULL) {
 136                 goto error;
 137         }
 138
 139         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 140                         chan->name, lkc->channel->attr.overwrite,
 141                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 142                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 143                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 144
 145         /* Kernel tracer channel creation */
 146         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 147         if (ret < 0) {
 148                 PERROR("ioctl kernel create channel");
 149                 goto error;
 150         }
 151
 152         /* Setup the channel fd */
 153         lkc->fd = ret;
 154         /* Prevent fd duplication after execlp() */
 155         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 156         if (ret < 0) {
 157                 PERROR("fcntl session fd");
 158         }
 159
 160         /* Add channel to session */
 161         cds_list_add(&lkc->list, &session->channel_list.head);
 162         session->channel_count++;
 163         lkc->session = session;
 164
 165         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 166
 167         return 0;
 168
 169 error:
 170         if (lkc) {
 171                 free(lkc->channel);
 172                 free(lkc);
 173         }
 174         return -1;
 175 }
 176
 177 /*
 178  * Create a kernel event, enable it to the kernel tracer and add it to the
 179  * channel event list of the kernel session.
 180  */
 181 int kernel_create_event(struct lttng_event *ev,
 182                 struct ltt_kernel_channel *channel)
 183 {
 184         int ret;
 185         struct ltt_kernel_event *event;
 186
 187         assert(ev);
 188         assert(channel);
 189
 190         event = trace_kernel_create_event(ev);
 191         if (event == NULL) {
 192                 ret = -1;
 193                 goto error;
 194         }
 195
 196         ret = kernctl_create_event(channel->fd, event->event);
 197         if (ret < 0) {
 198                 switch (errno) {
 199                 case EEXIST:
 200                         break;
 201                 case ENOSYS:
 202                         WARN("Event type not implemented");
 203                         break;
 204                 default:
 205                         PERROR("create event ioctl");
 206                 }
 207                 ret = -errno;
 208                 goto free_event;
 209         }
 210
 211         /*
 212          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 213          */
 214         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 215                 DBG2("Kernel event syscall creation success");
 216                 /*
 217                  * We use fd == -1 to ensure that we never trigger a close of fd
 218                  * 0.
 219                  */
 220                 event->fd = -1;
 221                 goto add_list;
 222         }
 223
 224         event->fd = ret;
 225         /* Prevent fd duplication after execlp() */
 226         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 227         if (ret < 0) {
 228                 PERROR("fcntl session fd");
 229         }
 230
 231 add_list:
 232         /* Add event to event list */
 233         cds_list_add(&event->list, &channel->events_list.head);
 234         channel->event_count++;
 235
 236         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 237
 238         return 0;
 239
 240 free_event:
 241         free(event);
 242 error:
 243         return ret;
 244 }
 245
 246 /*
 247  * Disable a kernel channel.
 248  */
 249 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 250 {
 251         int ret;
 252
 253         assert(chan);
 254
 255         ret = kernctl_disable(chan->fd);
 256         if (ret < 0) {
 257                 PERROR("disable chan ioctl");
 258                 ret = errno;
 259                 goto error;
 260         }
 261
 262         chan->enabled = 0;
 263         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 264
 265         return 0;
 266
 267 error:
 268         return ret;
 269 }
 270
 271 /*
 272  * Enable a kernel channel.
 273  */
 274 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 275 {
 276         int ret;
 277
 278         assert(chan);
 279
 280         ret = kernctl_enable(chan->fd);
 281         if (ret < 0 && errno != EEXIST) {
 282                 PERROR("Enable kernel chan");
 283                 goto error;
 284         }
 285
 286         chan->enabled = 1;
 287         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 288
 289         return 0;
 290
 291 error:
 292         return ret;
 293 }
 294
 295 /*
 296  * Enable a kernel event.
 297  */
 298 int kernel_enable_event(struct ltt_kernel_event *event)
 299 {
 300         int ret;
 301
 302         assert(event);
 303
 304         ret = kernctl_enable(event->fd);
 305         if (ret < 0) {
 306                 switch (errno) {
 307                 case EEXIST:
 308                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 309                         break;
 310                 default:
 311                         PERROR("enable kernel event");
 312                         break;
 313                 }
 314                 goto error;
 315         }
 316
 317         event->enabled = 1;
 318         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 319
 320         return 0;
 321
 322 error:
 323         return ret;
 324 }
 325
 326 /*
 327  * Disable a kernel event.
 328  */
 329 int kernel_disable_event(struct ltt_kernel_event *event)
 330 {
 331         int ret;
 332
 333         assert(event);
 334
 335         ret = kernctl_disable(event->fd);
 336         if (ret < 0) {
 337                 switch (errno) {
 338                 case EEXIST:
 339                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 340                         break;
 341                 default:
 342                         PERROR("disable kernel event");
 343                         break;
 344                 }
 345                 goto error;
 346         }
 347
 348         event->enabled = 0;
 349         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 350
 351         return 0;
 352
 353 error:
 354         return ret;
 355 }
 356
 357 /*
 358  * Create kernel metadata, open from the kernel tracer and add it to the
 359  * kernel session.
 360  */
 361 int kernel_open_metadata(struct ltt_kernel_session *session)
 362 {
 363         int ret;
 364         struct ltt_kernel_metadata *lkm = NULL;
 365
 366         assert(session);
 367
 368         /* Allocate kernel metadata */
 369         lkm = trace_kernel_create_metadata();
 370         if (lkm == NULL) {
 371                 goto error;
 372         }
 373
 374         /* Kernel tracer metadata creation */
 375         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 376         if (ret < 0) {
 377                 goto error_open;
 378         }
 379
 380         lkm->fd = ret;
 381         /* Prevent fd duplication after execlp() */
 382         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 383         if (ret < 0) {
 384                 PERROR("fcntl session fd");
 385         }
 386
 387         session->metadata = lkm;
 388
 389         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 390
 391         return 0;
 392
 393 error_open:
 394         trace_kernel_destroy_metadata(lkm);
 395 error:
 396         return -1;
 397 }
 398
 399 /*
 400  * Start tracing session.
 401  */
 402 int kernel_start_session(struct ltt_kernel_session *session)
 403 {
 404         int ret;
 405
 406         assert(session);
 407
 408         ret = kernctl_start_session(session->fd);
 409         if (ret < 0) {
 410                 PERROR("ioctl start session");
 411                 goto error;
 412         }
 413
 414         DBG("Kernel session started");
 415
 416         return 0;
 417
 418 error:
 419         return ret;
 420 }
 421
 422 /*
 423  * Make a kernel wait to make sure in-flight probe have completed.
 424  */
 425 void kernel_wait_quiescent(int fd)
 426 {
 427         int ret;
 428
 429         DBG("Kernel quiescent wait on %d", fd);
 430
 431         ret = kernctl_wait_quiescent(fd);
 432         if (ret < 0) {
 433                 PERROR("wait quiescent ioctl");
 434                 ERR("Kernel quiescent wait failed");
 435         }
 436 }
 437
 438 /*
 439  * Kernel calibrate
 440  */
 441 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 442 {
 443         int ret;
 444
 445         assert(calibrate);
 446
 447         ret = kernctl_calibrate(fd, calibrate);
 448         if (ret < 0) {
 449                 PERROR("calibrate ioctl");
 450                 return -1;
 451         }
 452
 453         return 0;
 454 }
 455
 456
 457 /*
 458  *  Force flush buffer of metadata.
 459  */
 460 int kernel_metadata_flush_buffer(int fd)
 461 {
 462         int ret;
 463
 464         DBG("Kernel flushing metadata buffer on fd %d", fd);
 465
 466         ret = kernctl_buffer_flush(fd);
 467         if (ret < 0) {
 468                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 469         }
 470
 471         return 0;
 472 }
 473
 474 /*
 475  * Force flush buffer for channel.
 476  */
 477 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 478 {
 479         int ret;
 480         struct ltt_kernel_stream *stream;
 481
 482         assert(channel);
 483
 484         DBG("Flush buffer for channel %s", channel->channel->name);
 485
 486         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 487                 DBG("Flushing channel stream %d", stream->fd);
 488                 ret = kernctl_buffer_flush(stream->fd);
 489                 if (ret < 0) {
 490                         PERROR("ioctl");
 491                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 492                                         stream->fd, ret);
 493                 }
 494         }
 495
 496         return 0;
 497 }
 498
 499 /*
 500  * Stop tracing session.
 501  */
 502 int kernel_stop_session(struct ltt_kernel_session *session)
 503 {
 504         int ret;
 505
 506         assert(session);
 507
 508         ret = kernctl_stop_session(session->fd);
 509         if (ret < 0) {
 510                 goto error;
 511         }
 512
 513         DBG("Kernel session stopped");
 514
 515         return 0;
 516
 517 error:
 518         return ret;
 519 }
 520
 521 /*
 522  * Open stream of channel, register it to the kernel tracer and add it
 523  * to the stream list of the channel.
 524  *
 525  * Return the number of created stream. Else, a negative value.
 526  */
 527 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 528 {
 529         int ret, count = 0;
 530         struct ltt_kernel_stream *lks;
 531
 532         assert(channel);
 533
 534         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 535                 lks = trace_kernel_create_stream(channel->channel->name, count);
 536                 if (lks == NULL) {
 537                         ret = close(ret);
 538                         if (ret) {
 539                                 PERROR("close");
 540                         }
 541                         goto error;
 542                 }
 543
 544                 lks->fd = ret;
 545                 /* Prevent fd duplication after execlp() */
 546                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 547                 if (ret < 0) {
 548                         PERROR("fcntl session fd");
 549                 }
 550
 551                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 552                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 553
 554                 /* Add stream to channe stream list */
 555                 cds_list_add(&lks->list, &channel->stream_list.head);
 556                 channel->stream_count++;
 557
 558                 /* Increment counter which represent CPU number. */
 559                 count++;
 560
 561                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 562                                 lks->state);
 563         }
 564
 565         return channel->stream_count;
 566
 567 error:
 568         return -1;
 569 }
 570
 571 /*
 572  * Open the metadata stream and set it to the kernel session.
 573  */
 574 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 575 {
 576         int ret;
 577
 578         assert(session);
 579
 580         ret = kernctl_create_stream(session->metadata->fd);
 581         if (ret < 0) {
 582                 PERROR("kernel create metadata stream");
 583                 goto error;
 584         }
 585
 586         DBG("Kernel metadata stream created (fd: %d)", ret);
 587         session->metadata_stream_fd = ret;
 588         /* Prevent fd duplication after execlp() */
 589         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 590         if (ret < 0) {
 591                 PERROR("fcntl session fd");
 592         }
 593
 594         return 0;
 595
 596 error:
 597         return -1;
 598 }
 599
 600 /*
 601  * Get the event list from the kernel tracer and return the number of elements.
 602  */
 603 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 604 {
 605         int fd, ret;
 606         char *event;
 607         size_t nbmem, count = 0;
 608         FILE *fp;
 609         struct lttng_event *elist;
 610
 611         assert(events);
 612
 613         fd = kernctl_tracepoint_list(tracer_fd);
 614         if (fd < 0) {
 615                 PERROR("kernel tracepoint list");
 616                 goto error;
 617         }
 618
 619         fp = fdopen(fd, "r");
 620         if (fp == NULL) {
 621                 PERROR("kernel tracepoint list fdopen");
 622                 goto error_fp;
 623         }
 624
 625         /*
 626          * Init memory size counter
 627          * See kernel-ctl.h for explanation of this value
 628          */
 629         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 630         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 631         if (elist == NULL) {
 632                 PERROR("alloc list events");
 633                 count = -ENOMEM;
 634                 goto end;
 635         }
 636
 637         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 638                 if (count >= nbmem) {
 639                         struct lttng_event *new_elist;
 640                         size_t new_nbmem;
 641
 642                         new_nbmem = nbmem << 1;
 643                         DBG("Reallocating event list from %zu to %zu bytes",
 644                                         nbmem, new_nbmem);
 645                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 646                         if (new_elist == NULL) {
 647                                 PERROR("realloc list events");
 648                                 free(event);
 649                                 free(elist);
 650                                 count = -ENOMEM;
 651                                 goto end;
 652                         }
 653                         /* Zero the new memory */
 654                         memset(new_elist + nbmem, 0,
 655                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 656                         nbmem = new_nbmem;
 657                         elist = new_elist;
 658                 }
 659                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 660                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 661                 elist[count].enabled = -1;
 662                 count++;
 663                 free(event);
 664         }
 665
 666         *events = elist;
 667         DBG("Kernel list events done (%zu events)", count);
 668 end:
 669         ret = fclose(fp);       /* closes both fp and fd */
 670         if (ret) {
 671                 PERROR("fclose");
 672         }
 673         return count;
 674
 675 error_fp:
 676         ret = close(fd);
 677         if (ret) {
 678                 PERROR("close");
 679         }
 680 error:
 681         return -1;
 682 }
 683
 684 /*
 685  * Get kernel version and validate it.
 686  */
 687 int kernel_validate_version(int tracer_fd)
 688 {
 689         int ret;
 690         struct lttng_kernel_tracer_version version;
 691
 692         ret = kernctl_tracer_version(tracer_fd, &version);
 693         if (ret < 0) {
 694                 ERR("Failed at getting the lttng-modules version");
 695                 goto error;
 696         }
 697
 698         /* Validate version */
 699         if (version.major != KERN_MODULES_PRE_MAJOR
 700                 && version.major != KERN_MODULES_MAJOR) {
 701                 goto error_version;
 702         }
 703
 704         DBG2("Kernel tracer version validated (major version %d)", version.major);
 705         return 0;
 706
 707 error_version:
 708         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 709                         version.major, KERN_MODULES_MAJOR)
 710         ret = -1;
 711
 712 error:
 713         return ret;
 714 }
 715
 716 /*
 717  * Kernel work-arounds called at the start of sessiond main().
 718  */
 719 int init_kernel_workarounds(void)
 720 {
 721         int ret;
 722         FILE *fp;
 723
 724         /*
 725          * boot_id needs to be read once before being used concurrently
 726          * to deal with a Linux kernel race. A fix is proposed for
 727          * upstream, but the work-around is needed for older kernels.
 728          */
 729         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 730         if (!fp) {
 731                 goto end_boot_id;
 732         }
 733         while (!feof(fp)) {
 734                 char buf[37] = "";
 735
 736                 ret = fread(buf, 1, sizeof(buf), fp);
 737                 if (ret < 0) {
 738                         /* Ignore error, we don't really care */
 739                 }
 740         }
 741         ret = fclose(fp);
 742         if (ret) {
 743                 PERROR("fclose");
 744         }
 745 end_boot_id:
 746         return 0;
 747 }
 748
 749 /*
 750  * Complete teardown of a kernel session.
 751  */
 752 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 753 {
 754         if (ksess == NULL) {
 755                 DBG3("No kernel session when tearing down session");
 756                 return;
 757         }
 758
 759         DBG("Tearing down kernel session");
 760
 761         /*
 762          * Destroy channels on the consumer if at least one FD has been sent and we
 763          * are in no output mode because the streams are in *no* monitor mode so we
 764          * have to send a command to clean them up or else they leaked.
 765          */
 766         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 767                 int ret;
 768                 struct consumer_socket *socket;
 769                 struct lttng_ht_iter iter;
 770
 771                 /* For each consumer socket. */
 772                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 773                                 socket, node.node) {
 774                         struct ltt_kernel_channel *chan;
 775
 776                         /* For each channel, ask the consumer to destroy it. */
 777                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 778                                 ret = kernel_consumer_destroy_channel(socket, chan);
 779                                 if (ret < 0) {
 780                                         /* Consumer is probably dead. Use next socket. */
 781                                         continue;
 782                                 }
 783                         }
 784                 }
 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, unsigned int nb_streams)
 830 {
 831         int err, ret, saved_metadata_fd;
 832         struct consumer_socket *socket;
 833         struct lttng_ht_iter iter;
 834         struct ltt_kernel_metadata *saved_metadata;
 835         uint64_t max_size_per_stream = 0;
 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         if (output->max_size > 0 && nb_streams > 0) {
 862                 max_size_per_stream = output->max_size / nb_streams;
 863         }
 864
 865         /* Send metadata to consumer and snapshot everything. */
 866         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 867                         socket, node.node) {
 868                 struct consumer_output *saved_output;
 869                 struct ltt_kernel_channel *chan;
 870
 871                 /*
 872                  * Temporarly switch consumer output for our snapshot output. As long
 873                  * as the session lock is taken, this is safe.
 874                  */
 875                 saved_output = ksess->consumer;
 876                 ksess->consumer = output->consumer;
 877
 878                 pthread_mutex_lock(socket->lock);
 879                 /* This stream must not be monitored by the consumer. */
 880                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 881                 pthread_mutex_unlock(socket->lock);
 882                 /* Put back the saved consumer output into the session. */
 883                 ksess->consumer = saved_output;
 884                 if (ret < 0) {
 885                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 886                         goto error_consumer;
 887                 }
 888
 889                 /* For each channel, ask the consumer to snapshot it. */
 890                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 891                         if (max_size_per_stream &&
 892                                         chan->channel->attr.subbuf_size > max_size_per_stream) {
 893                                 ret = LTTNG_ERR_INVALID;
 894                                 DBG3("Kernel snapshot record maximum stream size %" PRIu64
 895                                                 " is smaller than subbuffer size of %" PRIu64,
 896                                                 max_size_per_stream, chan->channel->attr.subbuf_size);
 897                                 (void) kernel_consumer_destroy_metadata(socket,
 898                                                 ksess->metadata);
 899                                 goto error_consumer;
 900                         }
 901
 902                         pthread_mutex_lock(socket->lock);
 903                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
 904                                         ksess->uid, ksess->gid,
 905                                         DEFAULT_KERNEL_TRACE_DIR, wait,
 906                                         max_size_per_stream);
 907                         pthread_mutex_unlock(socket->lock);
 908                         if (ret < 0) {
 909                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 910                                 (void) kernel_consumer_destroy_metadata(socket,
 911                                                 ksess->metadata);
 912                                 goto error_consumer;
 913                         }
 914                 }
 915
 916                 /* Snapshot metadata, */
 917                 pthread_mutex_lock(socket->lock);
 918                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
 919                                 1, ksess->uid, ksess->gid,
 920                                 DEFAULT_KERNEL_TRACE_DIR, wait, max_size_per_stream);
 921                 pthread_mutex_unlock(socket->lock);
 922                 if (ret < 0) {
 923                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
 924                         goto error_consumer;
 925                 }
 926
 927                 /*
 928                  * The metadata snapshot is done, ask the consumer to destroy it since
 929                  * it's not monitored on the consumer side.
 930                  */
 931                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
 932         }
 933
 934         ret = LTTNG_OK;
 935
 936 error_consumer:
 937         /* Close newly opened metadata stream. It's now on the consumer side. */
 938         err = close(ksess->metadata_stream_fd);
 939         if (err < 0) {
 940                 PERROR("close snapshot kernel");
 941         }
 942
 943 error_open_stream:
 944         trace_kernel_destroy_metadata(ksess->metadata);
 945 error:
 946         /* Restore metadata state.*/
 947         ksess->metadata = saved_metadata;
 948         ksess->metadata_stream_fd = saved_metadata_fd;
 949
 950         rcu_read_unlock();
 951         return ret;
 952 }