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