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