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 #define _LGPL_SOURCE
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <stdlib.h>
  23 #include <stdio.h>
  24 #include <string.h>
  25 #include <unistd.h>
  26 #include <inttypes.h>
  27
  28 #include <common/common.h>
  29 #include <common/kernel-ctl/kernel-ctl.h>
  30 #include <common/kernel-ctl/kernel-ioctl.h>
  31 #include <common/sessiond-comm/sessiond-comm.h>
  32
  33 #include "consumer.h"
  34 #include "kernel.h"
  35 #include "kernel-consumer.h"
  36 #include "kern-modules.h"
  37 #include "utils.h"
  38
  39 /*
  40  * Add context on a kernel channel.
  41  */
  42 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  43                 struct ltt_kernel_context *ctx)
  44 {
  45         int ret;
  46
  47         assert(chan);
  48         assert(ctx);
  49
  50         DBG("Adding context to channel %s", chan->channel->name);
  51         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  52         if (ret < 0) {
  53                 if (errno != EEXIST) {
  54                         PERROR("add context ioctl");
  55                 } else {
  56                         /* If EEXIST, we just ignore the error */
  57                         ret = 0;
  58                 }
  59                 goto error;
  60         }
  61
  62         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  63
  64         return 0;
  65
  66 error:
  67         return ret;
  68 }
  69
  70 /*
  71  * Create a new kernel session, register it to the kernel tracer and add it to
  72  * the session daemon session.
  73  */
  74 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  75 {
  76         int ret;
  77         struct ltt_kernel_session *lks;
  78
  79         assert(session);
  80
  81         /* Allocate data structure */
  82         lks = trace_kernel_create_session();
  83         if (lks == NULL) {
  84                 ret = -1;
  85                 goto error;
  86         }
  87
  88         /* Kernel tracer session creation */
  89         ret = kernctl_create_session(tracer_fd);
  90         if (ret < 0) {
  91                 PERROR("ioctl kernel create session");
  92                 goto error;
  93         }
  94
  95         lks->fd = ret;
  96         /* Prevent fd duplication after execlp() */
  97         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
  98         if (ret < 0) {
  99                 PERROR("fcntl session fd");
 100         }
 101
 102         lks->id = session->id;
 103         lks->consumer_fds_sent = 0;
 104         session->kernel_session = lks;
 105
 106         DBG("Kernel session created (fd: %d)", lks->fd);
 107
 108         return 0;
 109
 110 error:
 111         if (lks) {
 112                 trace_kernel_destroy_session(lks);
 113         }
 114         return ret;
 115 }
 116
 117 /*
 118  * Create a kernel channel, register it to the kernel tracer and add it to the
 119  * kernel session.
 120  */
 121 int kernel_create_channel(struct ltt_kernel_session *session,
 122                 struct lttng_channel *chan)
 123 {
 124         int ret;
 125         struct ltt_kernel_channel *lkc;
 126
 127         assert(session);
 128         assert(chan);
 129
 130         /* Allocate kernel channel */
 131         lkc = trace_kernel_create_channel(chan);
 132         if (lkc == NULL) {
 133                 goto error;
 134         }
 135
 136         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 137                         chan->name, lkc->channel->attr.overwrite,
 138                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 139                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 140                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 141
 142         /* Kernel tracer channel creation */
 143         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 144         if (ret < 0) {
 145                 PERROR("ioctl kernel create channel");
 146                 goto error;
 147         }
 148
 149         /* Setup the channel fd */
 150         lkc->fd = ret;
 151         /* Prevent fd duplication after execlp() */
 152         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 153         if (ret < 0) {
 154                 PERROR("fcntl session fd");
 155         }
 156
 157         /* Add channel to session */
 158         cds_list_add(&lkc->list, &session->channel_list.head);
 159         session->channel_count++;
 160         lkc->session = session;
 161
 162         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 163
 164         return 0;
 165
 166 error:
 167         if (lkc) {
 168                 free(lkc->channel);
 169                 free(lkc);
 170         }
 171         return -1;
 172 }
 173
 174 /*
 175  * Create a kernel event, enable it to the kernel tracer and add it to the
 176  * channel event list of the kernel session.
 177  * We own filter_expression and filter.
 178  */
 179 int kernel_create_event(struct lttng_event *ev,
 180                 struct ltt_kernel_channel *channel,
 181                 char *filter_expression,
 182                 struct lttng_filter_bytecode *filter)
 183 {
 184         int ret;
 185         struct ltt_kernel_event *event;
 186
 187         assert(ev);
 188         assert(channel);
 189
 190         /* We pass ownership of filter_expression and filter */
 191         event = trace_kernel_create_event(ev, filter_expression,
 192                         filter);
 193         if (event == NULL) {
 194                 ret = -1;
 195                 goto error;
 196         }
 197
 198         ret = kernctl_create_event(channel->fd, event->event);
 199         if (ret < 0) {
 200                 switch (errno) {
 201                 case EEXIST:
 202                         break;
 203                 case ENOSYS:
 204                         WARN("Event type not implemented");
 205                         break;
 206                 case ENOENT:
 207                         WARN("Event %s not found!", ev->name);
 208                         break;
 209                 default:
 210                         PERROR("create event ioctl");
 211                 }
 212                 ret = -errno;
 213                 goto free_event;
 214         }
 215
 216         event->type = ev->type;
 217         event->fd = ret;
 218         /* Prevent fd duplication after execlp() */
 219         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 220         if (ret < 0) {
 221                 PERROR("fcntl session fd");
 222         }
 223
 224         if (filter) {
 225                 ret = kernctl_filter(event->fd, filter);
 226                 if (ret) {
 227                         goto filter_error;
 228                 }
 229         }
 230
 231         ret = kernctl_enable(event->fd);
 232         if (ret < 0) {
 233                 switch (errno) {
 234                 case EEXIST:
 235                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 236                         break;
 237                 default:
 238                         PERROR("enable kernel event");
 239                         break;
 240                 }
 241                 goto enable_error;
 242         }
 243
 244         /* Add event to event list */
 245         cds_list_add(&event->list, &channel->events_list.head);
 246         channel->event_count++;
 247
 248         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 249
 250         return 0;
 251
 252 enable_error:
 253 filter_error:
 254         {
 255                 int closeret;
 256
 257                 closeret = close(event->fd);
 258                 if (closeret) {
 259                         PERROR("close event fd");
 260                 }
 261         }
 262 free_event:
 263         free(event);
 264 error:
 265         return ret;
 266 }
 267
 268 /*
 269  * Disable a kernel channel.
 270  */
 271 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 272 {
 273         int ret;
 274
 275         assert(chan);
 276
 277         ret = kernctl_disable(chan->fd);
 278         if (ret < 0) {
 279                 PERROR("disable chan ioctl");
 280                 ret = errno;
 281                 goto error;
 282         }
 283
 284         chan->enabled = 0;
 285         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 286
 287         return 0;
 288
 289 error:
 290         return ret;
 291 }
 292
 293 /*
 294  * Enable a kernel channel.
 295  */
 296 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 297 {
 298         int ret;
 299
 300         assert(chan);
 301
 302         ret = kernctl_enable(chan->fd);
 303         if (ret < 0 && errno != EEXIST) {
 304                 PERROR("Enable kernel chan");
 305                 goto error;
 306         }
 307
 308         chan->enabled = 1;
 309         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 310
 311         return 0;
 312
 313 error:
 314         return ret;
 315 }
 316
 317 /*
 318  * Enable a kernel event.
 319  */
 320 int kernel_enable_event(struct ltt_kernel_event *event)
 321 {
 322         int ret;
 323
 324         assert(event);
 325
 326         ret = kernctl_enable(event->fd);
 327         if (ret < 0) {
 328                 switch (errno) {
 329                 case EEXIST:
 330                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 331                         break;
 332                 default:
 333                         PERROR("enable kernel event");
 334                         break;
 335                 }
 336                 goto error;
 337         }
 338
 339         event->enabled = 1;
 340         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 341
 342         return 0;
 343
 344 error:
 345         return ret;
 346 }
 347
 348 /*
 349  * Disable a kernel event.
 350  */
 351 int kernel_disable_event(struct ltt_kernel_event *event)
 352 {
 353         int ret;
 354
 355         assert(event);
 356
 357         ret = kernctl_disable(event->fd);
 358         if (ret < 0) {
 359                 switch (errno) {
 360                 case EEXIST:
 361                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 362                         break;
 363                 default:
 364                         PERROR("disable kernel event");
 365                         break;
 366                 }
 367                 goto error;
 368         }
 369
 370         event->enabled = 0;
 371         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 372
 373         return 0;
 374
 375 error:
 376         return ret;
 377 }
 378
 379
 380 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 381 {
 382         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 383                         pid, session->id);
 384         return kernctl_track_pid(session->fd, pid);
 385 }
 386
 387 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 388 {
 389         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 390                         pid, session->id);
 391         return kernctl_untrack_pid(session->fd, pid);
 392 }
 393
 394 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 395                 int **_pids)
 396 {
 397         int fd, ret;
 398         int pid;
 399         ssize_t nbmem, count = 0;
 400         FILE *fp;
 401         int *pids;
 402
 403         fd = kernctl_list_tracker_pids(session->fd);
 404         if (fd < 0) {
 405                 PERROR("kernel tracker pids list");
 406                 goto error;
 407         }
 408
 409         fp = fdopen(fd, "r");
 410         if (fp == NULL) {
 411                 PERROR("kernel tracker pids list fdopen");
 412                 goto error_fp;
 413         }
 414
 415         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 416         pids = zmalloc(sizeof(*pids) * nbmem);
 417         if (pids == NULL) {
 418                 PERROR("alloc list pids");
 419                 count = -ENOMEM;
 420                 goto end;
 421         }
 422
 423         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 424                 if (count >= nbmem) {
 425                         int *new_pids;
 426                         size_t new_nbmem;
 427
 428                         new_nbmem = nbmem << 1;
 429                         DBG("Reallocating pids list from %zu to %zu entries",
 430                                         nbmem, new_nbmem);
 431                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 432                         if (new_pids == NULL) {
 433                                 PERROR("realloc list events");
 434                                 free(pids);
 435                                 count = -ENOMEM;
 436                                 goto end;
 437                         }
 438                         /* Zero the new memory */
 439                         memset(new_pids + nbmem, 0,
 440                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 441                         nbmem = new_nbmem;
 442                         pids = new_pids;
 443                 }
 444                 pids[count++] = pid;
 445         }
 446
 447         *_pids = pids;
 448         DBG("Kernel list tracker pids done (%zd pids)", count);
 449 end:
 450         ret = fclose(fp);       /* closes both fp and fd */
 451         if (ret) {
 452                 PERROR("fclose");
 453         }
 454         return count;
 455
 456 error_fp:
 457         ret = close(fd);
 458         if (ret) {
 459                 PERROR("close");
 460         }
 461 error:
 462         return -1;
 463 }
 464
 465 /*
 466  * Create kernel metadata, open from the kernel tracer and add it to the
 467  * kernel session.
 468  */
 469 int kernel_open_metadata(struct ltt_kernel_session *session)
 470 {
 471         int ret;
 472         struct ltt_kernel_metadata *lkm = NULL;
 473
 474         assert(session);
 475
 476         /* Allocate kernel metadata */
 477         lkm = trace_kernel_create_metadata();
 478         if (lkm == NULL) {
 479                 goto error;
 480         }
 481
 482         /* Kernel tracer metadata creation */
 483         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 484         if (ret < 0) {
 485                 goto error_open;
 486         }
 487
 488         lkm->fd = ret;
 489         /* Prevent fd duplication after execlp() */
 490         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 491         if (ret < 0) {
 492                 PERROR("fcntl session fd");
 493         }
 494
 495         session->metadata = lkm;
 496
 497         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 498
 499         return 0;
 500
 501 error_open:
 502         trace_kernel_destroy_metadata(lkm);
 503 error:
 504         return -1;
 505 }
 506
 507 /*
 508  * Start tracing session.
 509  */
 510 int kernel_start_session(struct ltt_kernel_session *session)
 511 {
 512         int ret;
 513
 514         assert(session);
 515
 516         ret = kernctl_start_session(session->fd);
 517         if (ret < 0) {
 518                 PERROR("ioctl start session");
 519                 goto error;
 520         }
 521
 522         DBG("Kernel session started");
 523
 524         return 0;
 525
 526 error:
 527         return ret;
 528 }
 529
 530 /*
 531  * Make a kernel wait to make sure in-flight probe have completed.
 532  */
 533 void kernel_wait_quiescent(int fd)
 534 {
 535         int ret;
 536
 537         DBG("Kernel quiescent wait on %d", fd);
 538
 539         ret = kernctl_wait_quiescent(fd);
 540         if (ret < 0) {
 541                 PERROR("wait quiescent ioctl");
 542                 ERR("Kernel quiescent wait failed");
 543         }
 544 }
 545
 546 /*
 547  * Kernel calibrate
 548  */
 549 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 550 {
 551         int ret;
 552
 553         assert(calibrate);
 554
 555         ret = kernctl_calibrate(fd, calibrate);
 556         if (ret < 0) {
 557                 PERROR("calibrate ioctl");
 558                 return -1;
 559         }
 560
 561         return 0;
 562 }
 563
 564
 565 /*
 566  *  Force flush buffer of metadata.
 567  */
 568 int kernel_metadata_flush_buffer(int fd)
 569 {
 570         int ret;
 571
 572         DBG("Kernel flushing metadata buffer on fd %d", fd);
 573
 574         ret = kernctl_buffer_flush(fd);
 575         if (ret < 0) {
 576                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 577         }
 578
 579         return 0;
 580 }
 581
 582 /*
 583  * Force flush buffer for channel.
 584  */
 585 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 586 {
 587         int ret;
 588         struct ltt_kernel_stream *stream;
 589
 590         assert(channel);
 591
 592         DBG("Flush buffer for channel %s", channel->channel->name);
 593
 594         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 595                 DBG("Flushing channel stream %d", stream->fd);
 596                 ret = kernctl_buffer_flush(stream->fd);
 597                 if (ret < 0) {
 598                         PERROR("ioctl");
 599                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 600                                         stream->fd, ret);
 601                 }
 602         }
 603
 604         return 0;
 605 }
 606
 607 /*
 608  * Stop tracing session.
 609  */
 610 int kernel_stop_session(struct ltt_kernel_session *session)
 611 {
 612         int ret;
 613
 614         assert(session);
 615
 616         ret = kernctl_stop_session(session->fd);
 617         if (ret < 0) {
 618                 goto error;
 619         }
 620
 621         DBG("Kernel session stopped");
 622
 623         return 0;
 624
 625 error:
 626         return ret;
 627 }
 628
 629 /*
 630  * Open stream of channel, register it to the kernel tracer and add it
 631  * to the stream list of the channel.
 632  *
 633  * Return the number of created stream. Else, a negative value.
 634  */
 635 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 636 {
 637         int ret, count = 0;
 638         struct ltt_kernel_stream *lks;
 639
 640         assert(channel);
 641
 642         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 643                 lks = trace_kernel_create_stream(channel->channel->name, count);
 644                 if (lks == NULL) {
 645                         ret = close(ret);
 646                         if (ret) {
 647                                 PERROR("close");
 648                         }
 649                         goto error;
 650                 }
 651
 652                 lks->fd = ret;
 653                 /* Prevent fd duplication after execlp() */
 654                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 655                 if (ret < 0) {
 656                         PERROR("fcntl session fd");
 657                 }
 658
 659                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 660                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 661
 662                 /* Add stream to channe stream list */
 663                 cds_list_add(&lks->list, &channel->stream_list.head);
 664                 channel->stream_count++;
 665
 666                 /* Increment counter which represent CPU number. */
 667                 count++;
 668
 669                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 670                                 lks->state);
 671         }
 672
 673         return channel->stream_count;
 674
 675 error:
 676         return -1;
 677 }
 678
 679 /*
 680  * Open the metadata stream and set it to the kernel session.
 681  */
 682 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 683 {
 684         int ret;
 685
 686         assert(session);
 687
 688         ret = kernctl_create_stream(session->metadata->fd);
 689         if (ret < 0) {
 690                 PERROR("kernel create metadata stream");
 691                 goto error;
 692         }
 693
 694         DBG("Kernel metadata stream created (fd: %d)", ret);
 695         session->metadata_stream_fd = ret;
 696         /* Prevent fd duplication after execlp() */
 697         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 698         if (ret < 0) {
 699                 PERROR("fcntl session fd");
 700         }
 701
 702         return 0;
 703
 704 error:
 705         return -1;
 706 }
 707
 708 /*
 709  * Get the event list from the kernel tracer and return the number of elements.
 710  */
 711 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 712 {
 713         int fd, ret;
 714         char *event;
 715         size_t nbmem, count = 0;
 716         FILE *fp;
 717         struct lttng_event *elist;
 718
 719         assert(events);
 720
 721         fd = kernctl_tracepoint_list(tracer_fd);
 722         if (fd < 0) {
 723                 PERROR("kernel tracepoint list");
 724                 goto error;
 725         }
 726
 727         fp = fdopen(fd, "r");
 728         if (fp == NULL) {
 729                 PERROR("kernel tracepoint list fdopen");
 730                 goto error_fp;
 731         }
 732
 733         /*
 734          * Init memory size counter
 735          * See kernel-ctl.h for explanation of this value
 736          */
 737         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 738         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 739         if (elist == NULL) {
 740                 PERROR("alloc list events");
 741                 count = -ENOMEM;
 742                 goto end;
 743         }
 744
 745         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 746                 if (count >= nbmem) {
 747                         struct lttng_event *new_elist;
 748                         size_t new_nbmem;
 749
 750                         new_nbmem = nbmem << 1;
 751                         DBG("Reallocating event list from %zu to %zu bytes",
 752                                         nbmem, new_nbmem);
 753                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 754                         if (new_elist == NULL) {
 755                                 PERROR("realloc list events");
 756                                 free(event);
 757                                 free(elist);
 758                                 count = -ENOMEM;
 759                                 goto end;
 760                         }
 761                         /* Zero the new memory */
 762                         memset(new_elist + nbmem, 0,
 763                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 764                         nbmem = new_nbmem;
 765                         elist = new_elist;
 766                 }
 767                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 768                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 769                 elist[count].enabled = -1;
 770                 count++;
 771                 free(event);
 772         }
 773
 774         *events = elist;
 775         DBG("Kernel list events done (%zu events)", count);
 776 end:
 777         ret = fclose(fp);       /* closes both fp and fd */
 778         if (ret) {
 779                 PERROR("fclose");
 780         }
 781         return count;
 782
 783 error_fp:
 784         ret = close(fd);
 785         if (ret) {
 786                 PERROR("close");
 787         }
 788 error:
 789         return -1;
 790 }
 791
 792 /*
 793  * Get kernel version and validate it.
 794  */
 795 int kernel_validate_version(int tracer_fd)
 796 {
 797         int ret;
 798         struct lttng_kernel_tracer_version version;
 799         struct lttng_kernel_tracer_abi_version abi_version;
 800
 801         ret = kernctl_tracer_version(tracer_fd, &version);
 802         if (ret < 0) {
 803                 ERR("Failed at getting the lttng-modules version");
 804                 goto error;
 805         }
 806
 807         /* Validate version */
 808         if (version.major != VERSION_MAJOR) {
 809                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
 810                         version.major, VERSION_MAJOR);
 811                 goto error_version;
 812         }
 813         ret = kernctl_tracer_abi_version(tracer_fd, &abi_version);
 814         if (ret < 0) {
 815                 ERR("Failed at getting lttng-modules ABI version");
 816                 goto error;
 817         }
 818         if (abi_version.major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
 819                 ERR("Kernel tracer ABI version (%d.%d) is not compatible with expected ABI major version (%d.*)",
 820                         abi_version.major, abi_version.minor,
 821                         LTTNG_MODULES_ABI_MAJOR_VERSION);
 822                 goto error;
 823         }
 824         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
 825                         version.major, version.minor,
 826                         abi_version.major, abi_version.minor);
 827         return 0;
 828
 829 error_version:
 830         ret = -1;
 831
 832 error:
 833         return ret;
 834 }
 835
 836 /*
 837  * Kernel work-arounds called at the start of sessiond main().
 838  */
 839 int init_kernel_workarounds(void)
 840 {
 841         int ret;
 842         FILE *fp;
 843
 844         /*
 845          * boot_id needs to be read once before being used concurrently
 846          * to deal with a Linux kernel race. A fix is proposed for
 847          * upstream, but the work-around is needed for older kernels.
 848          */
 849         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 850         if (!fp) {
 851                 goto end_boot_id;
 852         }
 853         while (!feof(fp)) {
 854                 char buf[37] = "";
 855
 856                 ret = fread(buf, 1, sizeof(buf), fp);
 857                 if (ret < 0) {
 858                         /* Ignore error, we don't really care */
 859                 }
 860         }
 861         ret = fclose(fp);
 862         if (ret) {
 863                 PERROR("fclose");
 864         }
 865 end_boot_id:
 866         return 0;
 867 }
 868
 869 /*
 870  * Complete teardown of a kernel session.
 871  */
 872 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 873 {
 874         if (ksess == NULL) {
 875                 DBG3("No kernel session when tearing down session");
 876                 return;
 877         }
 878
 879         DBG("Tearing down kernel session");
 880
 881         /*
 882          * Destroy channels on the consumer if at least one FD has been sent and we
 883          * are in no output mode because the streams are in *no* monitor mode so we
 884          * have to send a command to clean them up or else they leaked.
 885          */
 886         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 887                 int ret;
 888                 struct consumer_socket *socket;
 889                 struct lttng_ht_iter iter;
 890
 891                 /* For each consumer socket. */
 892                 rcu_read_lock();
 893                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 894                                 socket, node.node) {
 895                         struct ltt_kernel_channel *chan;
 896
 897                         /* For each channel, ask the consumer to destroy it. */
 898                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 899                                 ret = kernel_consumer_destroy_channel(socket, chan);
 900                                 if (ret < 0) {
 901                                         /* Consumer is probably dead. Use next socket. */
 902                                         continue;
 903                                 }
 904                         }
 905                 }
 906                 rcu_read_unlock();
 907         }
 908
 909         /* Close any relayd session */
 910         consumer_output_send_destroy_relayd(ksess->consumer);
 911
 912         trace_kernel_destroy_session(ksess);
 913 }
 914
 915 /*
 916  * Destroy a kernel channel object. It does not do anything on the tracer side.
 917  */
 918 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 919 {
 920         struct ltt_kernel_session *ksess = NULL;
 921
 922         assert(kchan);
 923         assert(kchan->channel);
 924
 925         DBG3("Kernel destroy channel %s", kchan->channel->name);
 926
 927         /* Update channel count of associated session. */
 928         if (kchan->session) {
 929                 /* Keep pointer reference so we can update it after the destroy. */
 930                 ksess = kchan->session;
 931         }
 932
 933         trace_kernel_destroy_channel(kchan);
 934
 935         /*
 936          * At this point the kernel channel is not visible anymore. This is safe
 937          * since in order to work on a visible kernel session, the tracing session
 938          * lock (ltt_session.lock) MUST be acquired.
 939          */
 940         if (ksess) {
 941                 ksess->channel_count--;
 942         }
 943 }
 944
 945 /*
 946  * Take a snapshot for a given kernel session.
 947  *
 948  * Return 0 on success or else return a LTTNG_ERR code.
 949  */
 950 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 951                 struct snapshot_output *output, int wait,
 952                 uint64_t nb_packets_per_stream)
 953 {
 954         int err, ret, saved_metadata_fd;
 955         struct consumer_socket *socket;
 956         struct lttng_ht_iter iter;
 957         struct ltt_kernel_metadata *saved_metadata;
 958
 959         assert(ksess);
 960         assert(ksess->consumer);
 961         assert(output);
 962
 963         DBG("Kernel snapshot record started");
 964
 965         /* Save current metadata since the following calls will change it. */
 966         saved_metadata = ksess->metadata;
 967         saved_metadata_fd = ksess->metadata_stream_fd;
 968
 969         rcu_read_lock();
 970
 971         ret = kernel_open_metadata(ksess);
 972         if (ret < 0) {
 973                 ret = LTTNG_ERR_KERN_META_FAIL;
 974                 goto error;
 975         }
 976
 977         ret = kernel_open_metadata_stream(ksess);
 978         if (ret < 0) {
 979                 ret = LTTNG_ERR_KERN_META_FAIL;
 980                 goto error_open_stream;
 981         }
 982
 983         /* Send metadata to consumer and snapshot everything. */
 984         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 985                         socket, node.node) {
 986                 struct consumer_output *saved_output;
 987                 struct ltt_kernel_channel *chan;
 988
 989                 /*
 990                  * Temporarly switch consumer output for our snapshot output. As long
 991                  * as the session lock is taken, this is safe.
 992                  */
 993                 saved_output = ksess->consumer;
 994                 ksess->consumer = output->consumer;
 995
 996                 pthread_mutex_lock(socket->lock);
 997                 /* This stream must not be monitored by the consumer. */
 998                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
 999                 pthread_mutex_unlock(socket->lock);
1000                 /* Put back the saved consumer output into the session. */
1001                 ksess->consumer = saved_output;
1002                 if (ret < 0) {
1003                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1004                         goto error_consumer;
1005                 }
1006
1007                 /* For each channel, ask the consumer to snapshot it. */
1008                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1009                         pthread_mutex_lock(socket->lock);
1010                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
1011                                         ksess->uid, ksess->gid,
1012                                         DEFAULT_KERNEL_TRACE_DIR, wait,
1013                                         nb_packets_per_stream);
1014                         pthread_mutex_unlock(socket->lock);
1015                         if (ret < 0) {
1016                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1017                                 (void) kernel_consumer_destroy_metadata(socket,
1018                                                 ksess->metadata);
1019                                 goto error_consumer;
1020                         }
1021                 }
1022
1023                 /* Snapshot metadata, */
1024                 pthread_mutex_lock(socket->lock);
1025                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
1026                                 1, ksess->uid, ksess->gid,
1027                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0);
1028                 pthread_mutex_unlock(socket->lock);
1029                 if (ret < 0) {
1030                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1031                         goto error_consumer;
1032                 }
1033
1034                 /*
1035                  * The metadata snapshot is done, ask the consumer to destroy it since
1036                  * it's not monitored on the consumer side.
1037                  */
1038                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1039         }
1040
1041         ret = LTTNG_OK;
1042
1043 error_consumer:
1044         /* Close newly opened metadata stream. It's now on the consumer side. */
1045         err = close(ksess->metadata_stream_fd);
1046         if (err < 0) {
1047                 PERROR("close snapshot kernel");
1048         }
1049
1050 error_open_stream:
1051         trace_kernel_destroy_metadata(ksess->metadata);
1052 error:
1053         /* Restore metadata state.*/
1054         ksess->metadata = saved_metadata;
1055         ksess->metadata_stream_fd = saved_metadata_fd;
1056
1057         rcu_read_unlock();
1058         return ret;
1059 }
1060
1061 /*
1062  * Get the syscall mask array from the kernel tracer.
1063  *
1064  * Return 0 on success else a negative value. In both case, syscall_mask should
1065  * be freed.
1066  */
1067 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1068 {
1069         assert(syscall_mask);
1070         assert(nr_bits);
1071
1072         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1073 }