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 _LGPL_SOURCE
  19 #include <fcntl.h>
  20 #include <stdlib.h>
  21 #include <stdio.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <inttypes.h>
  25
  26 #include <common/common.h>
  27 #include <common/kernel-ctl/kernel-ctl.h>
  28 #include <common/kernel-ctl/kernel-ioctl.h>
  29 #include <common/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kernel-consumer.h"
  34 #include "kern-modules.h"
  35 #include "utils.h"
  36 #include "rotate.h"
  37
  38 /*
  39  * Key used to reference a channel between the sessiond and the consumer. This
  40  * is only read and updated with the session_list lock held.
  41  */
  42 static uint64_t next_kernel_channel_key;
  43
  44 /*
  45  * Add context on a kernel channel.
  46  *
  47  * Assumes the ownership of ctx.
  48  */
  49 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  50                 struct ltt_kernel_context *ctx)
  51 {
  52         int ret;
  53
  54         assert(chan);
  55         assert(ctx);
  56
  57         DBG("Adding context to channel %s", chan->channel->name);
  58         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  59         if (ret < 0) {
  60                 switch (-ret) {
  61                 case ENOSYS:
  62                         /* Exists but not available for this kernel */
  63                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  64                         goto error;
  65                 case EEXIST:
  66                         /* If EEXIST, we just ignore the error */
  67                         ret = 0;
  68                         goto end;
  69                 default:
  70                         PERROR("add context ioctl");
  71                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  72                         goto error;
  73                 }
  74         }
  75         ret = 0;
  76
  77 end:
  78         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  79         ctx->in_list = true;
  80         ctx = NULL;
  81 error:
  82         if (ctx) {
  83                 trace_kernel_destroy_context(ctx);
  84         }
  85         return ret;
  86 }
  87
  88 /*
  89  * Create a new kernel session, register it to the kernel tracer and add it to
  90  * the session daemon session.
  91  */
  92 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  93 {
  94         int ret;
  95         struct ltt_kernel_session *lks;
  96
  97         assert(session);
  98
  99         /* Allocate data structure */
 100         lks = trace_kernel_create_session();
 101         if (lks == NULL) {
 102                 ret = -1;
 103                 goto error;
 104         }
 105
 106         /* Kernel tracer session creation */
 107         ret = kernctl_create_session(tracer_fd);
 108         if (ret < 0) {
 109                 PERROR("ioctl kernel create session");
 110                 goto error;
 111         }
 112
 113         lks->fd = ret;
 114         /* Prevent fd duplication after execlp() */
 115         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 116         if (ret < 0) {
 117                 PERROR("fcntl session fd");
 118         }
 119
 120         lks->id = session->id;
 121         lks->consumer_fds_sent = 0;
 122         session->kernel_session = lks;
 123
 124         DBG("Kernel session created (fd: %d)", lks->fd);
 125
 126         return 0;
 127
 128 error:
 129         if (lks) {
 130                 trace_kernel_destroy_session(lks);
 131         }
 132         return ret;
 133 }
 134
 135 /*
 136  * Create a kernel channel, register it to the kernel tracer and add it to the
 137  * kernel session.
 138  */
 139 int kernel_create_channel(struct ltt_kernel_session *session,
 140                 struct lttng_channel *chan)
 141 {
 142         int ret;
 143         struct ltt_kernel_channel *lkc;
 144
 145         assert(session);
 146         assert(chan);
 147
 148         /* Allocate kernel channel */
 149         lkc = trace_kernel_create_channel(chan);
 150         if (lkc == NULL) {
 151                 goto error;
 152         }
 153
 154         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 155                         chan->name, lkc->channel->attr.overwrite,
 156                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 157                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 158                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 159
 160         /* Kernel tracer channel creation */
 161         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 162         if (ret < 0) {
 163                 PERROR("ioctl kernel create channel");
 164                 goto error;
 165         }
 166
 167         /* Setup the channel fd */
 168         lkc->fd = ret;
 169         /* Prevent fd duplication after execlp() */
 170         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 171         if (ret < 0) {
 172                 PERROR("fcntl session fd");
 173         }
 174
 175         /* Add channel to session */
 176         cds_list_add(&lkc->list, &session->channel_list.head);
 177         session->channel_count++;
 178         lkc->session = session;
 179         lkc->key = ++next_kernel_channel_key;
 180
 181         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 182                         lkc->channel->name, lkc->fd, lkc->key);
 183
 184         return 0;
 185
 186 error:
 187         if (lkc) {
 188                 free(lkc->channel);
 189                 free(lkc);
 190         }
 191         return -1;
 192 }
 193
 194 /*
 195  * Create a kernel event, enable it to the kernel tracer and add it to the
 196  * channel event list of the kernel session.
 197  * We own filter_expression and filter.
 198  */
 199 int kernel_create_event(struct lttng_event *ev,
 200                 struct ltt_kernel_channel *channel,
 201                 char *filter_expression,
 202                 struct lttng_filter_bytecode *filter)
 203 {
 204         int ret;
 205         struct ltt_kernel_event *event;
 206
 207         assert(ev);
 208         assert(channel);
 209
 210         /* We pass ownership of filter_expression and filter */
 211         event = trace_kernel_create_event(ev, filter_expression,
 212                         filter);
 213         if (event == NULL) {
 214                 ret = -1;
 215                 goto error;
 216         }
 217
 218         ret = kernctl_create_event(channel->fd, event->event);
 219         if (ret < 0) {
 220                 switch (-ret) {
 221                 case EEXIST:
 222                         break;
 223                 case ENOSYS:
 224                         WARN("Event type not implemented");
 225                         break;
 226                 case ENOENT:
 227                         WARN("Event %s not found!", ev->name);
 228                         break;
 229                 default:
 230                         PERROR("create event ioctl");
 231                 }
 232                 goto free_event;
 233         }
 234
 235         event->type = ev->type;
 236         event->fd = ret;
 237         /* Prevent fd duplication after execlp() */
 238         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 239         if (ret < 0) {
 240                 PERROR("fcntl session fd");
 241         }
 242
 243         if (filter) {
 244                 ret = kernctl_filter(event->fd, filter);
 245                 if (ret) {
 246                         goto filter_error;
 247                 }
 248         }
 249
 250         ret = kernctl_enable(event->fd);
 251         if (ret < 0) {
 252                 switch (-ret) {
 253                 case EEXIST:
 254                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 255                         break;
 256                 default:
 257                         PERROR("enable kernel event");
 258                         break;
 259                 }
 260                 goto enable_error;
 261         }
 262
 263         /* Add event to event list */
 264         cds_list_add(&event->list, &channel->events_list.head);
 265         channel->event_count++;
 266
 267         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 268
 269         return 0;
 270
 271 enable_error:
 272 filter_error:
 273         {
 274                 int closeret;
 275
 276                 closeret = close(event->fd);
 277                 if (closeret) {
 278                         PERROR("close event fd");
 279                 }
 280         }
 281 free_event:
 282         free(event);
 283 error:
 284         return ret;
 285 }
 286
 287 /*
 288  * Disable a kernel channel.
 289  */
 290 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 291 {
 292         int ret;
 293
 294         assert(chan);
 295
 296         ret = kernctl_disable(chan->fd);
 297         if (ret < 0) {
 298                 PERROR("disable chan ioctl");
 299                 goto error;
 300         }
 301
 302         chan->enabled = 0;
 303         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 304                         chan->channel->name, chan->fd, chan->key);
 305
 306         return 0;
 307
 308 error:
 309         return ret;
 310 }
 311
 312 /*
 313  * Enable a kernel channel.
 314  */
 315 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 316 {
 317         int ret;
 318
 319         assert(chan);
 320
 321         ret = kernctl_enable(chan->fd);
 322         if (ret < 0 && ret != -EEXIST) {
 323                 PERROR("Enable kernel chan");
 324                 goto error;
 325         }
 326
 327         chan->enabled = 1;
 328         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 329                         chan->channel->name, chan->fd, chan->key);
 330
 331         return 0;
 332
 333 error:
 334         return ret;
 335 }
 336
 337 /*
 338  * Enable a kernel event.
 339  */
 340 int kernel_enable_event(struct ltt_kernel_event *event)
 341 {
 342         int ret;
 343
 344         assert(event);
 345
 346         ret = kernctl_enable(event->fd);
 347         if (ret < 0) {
 348                 switch (-ret) {
 349                 case EEXIST:
 350                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 351                         break;
 352                 default:
 353                         PERROR("enable kernel event");
 354                         break;
 355                 }
 356                 goto error;
 357         }
 358
 359         event->enabled = 1;
 360         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 361
 362         return 0;
 363
 364 error:
 365         return ret;
 366 }
 367
 368 /*
 369  * Disable a kernel event.
 370  */
 371 int kernel_disable_event(struct ltt_kernel_event *event)
 372 {
 373         int ret;
 374
 375         assert(event);
 376
 377         ret = kernctl_disable(event->fd);
 378         if (ret < 0) {
 379                 switch (-ret) {
 380                 case EEXIST:
 381                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 382                         break;
 383                 default:
 384                         PERROR("disable kernel event");
 385                         break;
 386                 }
 387                 goto error;
 388         }
 389
 390         event->enabled = 0;
 391         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 392
 393         return 0;
 394
 395 error:
 396         return ret;
 397 }
 398
 399
 400 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 401 {
 402         int ret;
 403
 404         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 405                         pid, session->id);
 406         ret = kernctl_track_pid(session->fd, pid);
 407         if (!ret) {
 408                 return LTTNG_OK;
 409         }
 410         switch (-ret) {
 411         case EINVAL:
 412                 return LTTNG_ERR_INVALID;
 413         case ENOMEM:
 414                 return LTTNG_ERR_NOMEM;
 415         case EEXIST:
 416                 return LTTNG_ERR_PID_TRACKED;
 417         default:
 418                 return LTTNG_ERR_UNK;
 419         }
 420 }
 421
 422 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 423 {
 424         int ret;
 425
 426         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 427                         pid, session->id);
 428         ret = kernctl_untrack_pid(session->fd, pid);
 429         if (!ret) {
 430                 return LTTNG_OK;
 431         }
 432         switch (-ret) {
 433         case EINVAL:
 434                 return LTTNG_ERR_INVALID;
 435         case ENOMEM:
 436                 return LTTNG_ERR_NOMEM;
 437         case ENOENT:
 438                 return LTTNG_ERR_PID_NOT_TRACKED;
 439         default:
 440                 return LTTNG_ERR_UNK;
 441         }
 442 }
 443
 444 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 445                 int **_pids)
 446 {
 447         int fd, ret;
 448         int pid;
 449         ssize_t nbmem, count = 0;
 450         FILE *fp;
 451         int *pids;
 452
 453         fd = kernctl_list_tracker_pids(session->fd);
 454         if (fd < 0) {
 455                 PERROR("kernel tracker pids list");
 456                 goto error;
 457         }
 458
 459         fp = fdopen(fd, "r");
 460         if (fp == NULL) {
 461                 PERROR("kernel tracker pids list fdopen");
 462                 goto error_fp;
 463         }
 464
 465         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 466         pids = zmalloc(sizeof(*pids) * nbmem);
 467         if (pids == NULL) {
 468                 PERROR("alloc list pids");
 469                 count = -ENOMEM;
 470                 goto end;
 471         }
 472
 473         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 474                 if (count >= nbmem) {
 475                         int *new_pids;
 476                         size_t new_nbmem;
 477
 478                         new_nbmem = nbmem << 1;
 479                         DBG("Reallocating pids list from %zu to %zu entries",
 480                                         nbmem, new_nbmem);
 481                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 482                         if (new_pids == NULL) {
 483                                 PERROR("realloc list events");
 484                                 free(pids);
 485                                 count = -ENOMEM;
 486                                 goto end;
 487                         }
 488                         /* Zero the new memory */
 489                         memset(new_pids + nbmem, 0,
 490                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 491                         nbmem = new_nbmem;
 492                         pids = new_pids;
 493                 }
 494                 pids[count++] = pid;
 495         }
 496
 497         *_pids = pids;
 498         DBG("Kernel list tracker pids done (%zd pids)", count);
 499 end:
 500         ret = fclose(fp);       /* closes both fp and fd */
 501         if (ret) {
 502                 PERROR("fclose");
 503         }
 504         return count;
 505
 506 error_fp:
 507         ret = close(fd);
 508         if (ret) {
 509                 PERROR("close");
 510         }
 511 error:
 512         return -1;
 513 }
 514
 515 /*
 516  * Create kernel metadata, open from the kernel tracer and add it to the
 517  * kernel session.
 518  */
 519 int kernel_open_metadata(struct ltt_kernel_session *session)
 520 {
 521         int ret;
 522         struct ltt_kernel_metadata *lkm = NULL;
 523
 524         assert(session);
 525
 526         /* Allocate kernel metadata */
 527         lkm = trace_kernel_create_metadata();
 528         if (lkm == NULL) {
 529                 goto error;
 530         }
 531
 532         /* Kernel tracer metadata creation */
 533         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 534         if (ret < 0) {
 535                 goto error_open;
 536         }
 537
 538         lkm->fd = ret;
 539         lkm->key = ++next_kernel_channel_key;
 540         /* Prevent fd duplication after execlp() */
 541         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 542         if (ret < 0) {
 543                 PERROR("fcntl session fd");
 544         }
 545
 546         session->metadata = lkm;
 547
 548         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 549
 550         return 0;
 551
 552 error_open:
 553         trace_kernel_destroy_metadata(lkm);
 554 error:
 555         return -1;
 556 }
 557
 558 /*
 559  * Start tracing session.
 560  */
 561 int kernel_start_session(struct ltt_kernel_session *session)
 562 {
 563         int ret;
 564
 565         assert(session);
 566
 567         ret = kernctl_start_session(session->fd);
 568         if (ret < 0) {
 569                 PERROR("ioctl start session");
 570                 goto error;
 571         }
 572
 573         DBG("Kernel session started");
 574
 575         return 0;
 576
 577 error:
 578         return ret;
 579 }
 580
 581 /*
 582  * Make a kernel wait to make sure in-flight probe have completed.
 583  */
 584 void kernel_wait_quiescent(int fd)
 585 {
 586         int ret;
 587
 588         DBG("Kernel quiescent wait on %d", fd);
 589
 590         ret = kernctl_wait_quiescent(fd);
 591         if (ret < 0) {
 592                 PERROR("wait quiescent ioctl");
 593                 ERR("Kernel quiescent wait failed");
 594         }
 595 }
 596
 597 /*
 598  *  Force flush buffer of metadata.
 599  */
 600 int kernel_metadata_flush_buffer(int fd)
 601 {
 602         int ret;
 603
 604         DBG("Kernel flushing metadata buffer on fd %d", fd);
 605
 606         ret = kernctl_buffer_flush(fd);
 607         if (ret < 0) {
 608                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 609         }
 610
 611         return 0;
 612 }
 613
 614 /*
 615  * Force flush buffer for channel.
 616  */
 617 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 618 {
 619         int ret;
 620         struct ltt_kernel_stream *stream;
 621
 622         assert(channel);
 623
 624         DBG("Flush buffer for channel %s", channel->channel->name);
 625
 626         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 627                 DBG("Flushing channel stream %d", stream->fd);
 628                 ret = kernctl_buffer_flush(stream->fd);
 629                 if (ret < 0) {
 630                         PERROR("ioctl");
 631                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 632                                         stream->fd, ret);
 633                 }
 634         }
 635
 636         return 0;
 637 }
 638
 639 /*
 640  * Stop tracing session.
 641  */
 642 int kernel_stop_session(struct ltt_kernel_session *session)
 643 {
 644         int ret;
 645
 646         assert(session);
 647
 648         ret = kernctl_stop_session(session->fd);
 649         if (ret < 0) {
 650                 goto error;
 651         }
 652
 653         DBG("Kernel session stopped");
 654
 655         return 0;
 656
 657 error:
 658         return ret;
 659 }
 660
 661 /*
 662  * Open stream of channel, register it to the kernel tracer and add it
 663  * to the stream list of the channel.
 664  *
 665  * Note: given that the streams may appear in random order wrt CPU
 666  * number (e.g. cpu hotplug), the index value of the stream number in
 667  * the stream name is not necessarily linked to the CPU number.
 668  *
 669  * Return the number of created stream. Else, a negative value.
 670  */
 671 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 672 {
 673         int ret;
 674         struct ltt_kernel_stream *lks;
 675
 676         assert(channel);
 677
 678         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 679                 lks = trace_kernel_create_stream(channel->channel->name,
 680                                 channel->stream_count);
 681                 if (lks == NULL) {
 682                         ret = close(ret);
 683                         if (ret) {
 684                                 PERROR("close");
 685                         }
 686                         goto error;
 687                 }
 688
 689                 lks->fd = ret;
 690                 /* Prevent fd duplication after execlp() */
 691                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 692                 if (ret < 0) {
 693                         PERROR("fcntl session fd");
 694                 }
 695
 696                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 697                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 698
 699                 /* Add stream to channel stream list */
 700                 cds_list_add(&lks->list, &channel->stream_list.head);
 701                 channel->stream_count++;
 702
 703                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 704                                 lks->state);
 705         }
 706
 707         return channel->stream_count;
 708
 709 error:
 710         return -1;
 711 }
 712
 713 /*
 714  * Open the metadata stream and set it to the kernel session.
 715  */
 716 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 717 {
 718         int ret;
 719
 720         assert(session);
 721
 722         ret = kernctl_create_stream(session->metadata->fd);
 723         if (ret < 0) {
 724                 PERROR("kernel create metadata stream");
 725                 goto error;
 726         }
 727
 728         DBG("Kernel metadata stream created (fd: %d)", ret);
 729         session->metadata_stream_fd = ret;
 730         /* Prevent fd duplication after execlp() */
 731         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 732         if (ret < 0) {
 733                 PERROR("fcntl session fd");
 734         }
 735
 736         return 0;
 737
 738 error:
 739         return -1;
 740 }
 741
 742 /*
 743  * Get the event list from the kernel tracer and return the number of elements.
 744  */
 745 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 746 {
 747         int fd, ret;
 748         char *event;
 749         size_t nbmem, count = 0;
 750         FILE *fp;
 751         struct lttng_event *elist;
 752
 753         assert(events);
 754
 755         fd = kernctl_tracepoint_list(tracer_fd);
 756         if (fd < 0) {
 757                 PERROR("kernel tracepoint list");
 758                 goto error;
 759         }
 760
 761         fp = fdopen(fd, "r");
 762         if (fp == NULL) {
 763                 PERROR("kernel tracepoint list fdopen");
 764                 goto error_fp;
 765         }
 766
 767         /*
 768          * Init memory size counter
 769          * See kernel-ctl.h for explanation of this value
 770          */
 771         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 772         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 773         if (elist == NULL) {
 774                 PERROR("alloc list events");
 775                 count = -ENOMEM;
 776                 goto end;
 777         }
 778
 779         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 780                 if (count >= nbmem) {
 781                         struct lttng_event *new_elist;
 782                         size_t new_nbmem;
 783
 784                         new_nbmem = nbmem << 1;
 785                         DBG("Reallocating event list from %zu to %zu bytes",
 786                                         nbmem, new_nbmem);
 787                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 788                         if (new_elist == NULL) {
 789                                 PERROR("realloc list events");
 790                                 free(event);
 791                                 free(elist);
 792                                 count = -ENOMEM;
 793                                 goto end;
 794                         }
 795                         /* Zero the new memory */
 796                         memset(new_elist + nbmem, 0,
 797                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 798                         nbmem = new_nbmem;
 799                         elist = new_elist;
 800                 }
 801                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 802                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 803                 elist[count].enabled = -1;
 804                 count++;
 805                 free(event);
 806         }
 807
 808         *events = elist;
 809         DBG("Kernel list events done (%zu events)", count);
 810 end:
 811         ret = fclose(fp);       /* closes both fp and fd */
 812         if (ret) {
 813                 PERROR("fclose");
 814         }
 815         return count;
 816
 817 error_fp:
 818         ret = close(fd);
 819         if (ret) {
 820                 PERROR("close");
 821         }
 822 error:
 823         return -1;
 824 }
 825
 826 /*
 827  * Get kernel version and validate it.
 828  */
 829 int kernel_validate_version(int tracer_fd,
 830                 struct lttng_kernel_tracer_version *version,
 831                 struct lttng_kernel_tracer_abi_version *abi_version)
 832 {
 833         int ret;
 834
 835         ret = kernctl_tracer_version(tracer_fd, version);
 836         if (ret < 0) {
 837                 ERR("Failed to retrieve the lttng-modules version");
 838                 goto error;
 839         }
 840
 841         /* Validate version */
 842         if (version->major != VERSION_MAJOR) {
 843                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
 844                         version->major, VERSION_MAJOR);
 845                 goto error_version;
 846         }
 847         ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
 848         if (ret < 0) {
 849                 ERR("Failed to retrieve lttng-modules ABI version");
 850                 goto error;
 851         }
 852         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
 853                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
 854                         abi_version->major, abi_version->minor,
 855                         LTTNG_MODULES_ABI_MAJOR_VERSION);
 856                 goto error;
 857         }
 858         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
 859                         version->major, version->minor,
 860                         abi_version->major, abi_version->minor);
 861         return 0;
 862
 863 error_version:
 864         ret = -1;
 865
 866 error:
 867         ERR("Kernel tracer version check failed; kernel tracing will not be available");
 868         return ret;
 869 }
 870
 871 /*
 872  * Kernel work-arounds called at the start of sessiond main().
 873  */
 874 int init_kernel_workarounds(void)
 875 {
 876         int ret;
 877         FILE *fp;
 878
 879         /*
 880          * boot_id needs to be read once before being used concurrently
 881          * to deal with a Linux kernel race. A fix is proposed for
 882          * upstream, but the work-around is needed for older kernels.
 883          */
 884         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 885         if (!fp) {
 886                 goto end_boot_id;
 887         }
 888         while (!feof(fp)) {
 889                 char buf[37] = "";
 890
 891                 ret = fread(buf, 1, sizeof(buf), fp);
 892                 if (ret < 0) {
 893                         /* Ignore error, we don't really care */
 894                 }
 895         }
 896         ret = fclose(fp);
 897         if (ret) {
 898                 PERROR("fclose");
 899         }
 900 end_boot_id:
 901         return 0;
 902 }
 903
 904 /*
 905  * Complete teardown of a kernel session.
 906  */
 907 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 908 {
 909         if (ksess == NULL) {
 910                 DBG3("No kernel session when tearing down session");
 911                 return;
 912         }
 913
 914         DBG("Tearing down kernel session");
 915
 916         /*
 917          * Destroy channels on the consumer if at least one FD has been sent and we
 918          * are in no output mode because the streams are in *no* monitor mode so we
 919          * have to send a command to clean them up or else they leaked.
 920          */
 921         if (!ksess->output_traces && ksess->consumer_fds_sent) {
 922                 int ret;
 923                 struct consumer_socket *socket;
 924                 struct lttng_ht_iter iter;
 925
 926                 /* For each consumer socket. */
 927                 rcu_read_lock();
 928                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
 929                                 socket, node.node) {
 930                         struct ltt_kernel_channel *chan;
 931
 932                         /* For each channel, ask the consumer to destroy it. */
 933                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
 934                                 ret = kernel_consumer_destroy_channel(socket, chan);
 935                                 if (ret < 0) {
 936                                         /* Consumer is probably dead. Use next socket. */
 937                                         continue;
 938                                 }
 939                         }
 940                 }
 941                 rcu_read_unlock();
 942         }
 943
 944         /* Close any relayd session */
 945         consumer_output_send_destroy_relayd(ksess->consumer);
 946
 947         trace_kernel_destroy_session(ksess);
 948 }
 949
 950 /*
 951  * Destroy a kernel channel object. It does not do anything on the tracer side.
 952  */
 953 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 954 {
 955         struct ltt_kernel_session *ksess = NULL;
 956
 957         assert(kchan);
 958         assert(kchan->channel);
 959
 960         DBG3("Kernel destroy channel %s", kchan->channel->name);
 961
 962         /* Update channel count of associated session. */
 963         if (kchan->session) {
 964                 /* Keep pointer reference so we can update it after the destroy. */
 965                 ksess = kchan->session;
 966         }
 967
 968         trace_kernel_destroy_channel(kchan);
 969
 970         /*
 971          * At this point the kernel channel is not visible anymore. This is safe
 972          * since in order to work on a visible kernel session, the tracing session
 973          * lock (ltt_session.lock) MUST be acquired.
 974          */
 975         if (ksess) {
 976                 ksess->channel_count--;
 977         }
 978 }
 979
 980 /*
 981  * Take a snapshot for a given kernel session.
 982  *
 983  * Return 0 on success or else return a LTTNG_ERR code.
 984  */
 985 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
 986                 struct snapshot_output *output, int wait,
 987                 uint64_t nb_packets_per_stream)
 988 {
 989         int err, ret, saved_metadata_fd;
 990         struct consumer_socket *socket;
 991         struct lttng_ht_iter iter;
 992         struct ltt_kernel_metadata *saved_metadata;
 993         struct ltt_session *session;
 994         uint64_t trace_archive_id;
 995
 996         assert(ksess);
 997         assert(ksess->consumer);
 998         assert(output);
 999
1000         DBG("Kernel snapshot record started");
1001
1002         session = session_find_by_id(ksess->id);
1003         assert(session);
1004         assert(pthread_mutex_trylock(&session->lock));
1005         assert(session_trylock_list());
1006         trace_archive_id = session->current_archive_id;
1007
1008         /* Save current metadata since the following calls will change it. */
1009         saved_metadata = ksess->metadata;
1010         saved_metadata_fd = ksess->metadata_stream_fd;
1011
1012         rcu_read_lock();
1013
1014         ret = kernel_open_metadata(ksess);
1015         if (ret < 0) {
1016                 ret = LTTNG_ERR_KERN_META_FAIL;
1017                 goto error;
1018         }
1019
1020         ret = kernel_open_metadata_stream(ksess);
1021         if (ret < 0) {
1022                 ret = LTTNG_ERR_KERN_META_FAIL;
1023                 goto error_open_stream;
1024         }
1025
1026         /* Send metadata to consumer and snapshot everything. */
1027         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1028                         socket, node.node) {
1029                 struct consumer_output *saved_output;
1030                 struct ltt_kernel_channel *chan;
1031
1032                 /*
1033                  * Temporarly switch consumer output for our snapshot output. As long
1034                  * as the session lock is taken, this is safe.
1035                  */
1036                 saved_output = ksess->consumer;
1037                 ksess->consumer = output->consumer;
1038
1039                 pthread_mutex_lock(socket->lock);
1040                 /* This stream must not be monitored by the consumer. */
1041                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1042                 pthread_mutex_unlock(socket->lock);
1043                 /* Put back the saved consumer output into the session. */
1044                 ksess->consumer = saved_output;
1045                 if (ret < 0) {
1046                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1047                         goto error_consumer;
1048                 }
1049
1050                 /* For each channel, ask the consumer to snapshot it. */
1051                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1052                         ret = consumer_snapshot_channel(socket, chan->key, output, 0,
1053                                         ksess->uid, ksess->gid,
1054                                         DEFAULT_KERNEL_TRACE_DIR, wait,
1055                                         nb_packets_per_stream,
1056                                         trace_archive_id);
1057                         if (ret < 0) {
1058                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1059                                 (void) kernel_consumer_destroy_metadata(socket,
1060                                                 ksess->metadata);
1061                                 goto error_consumer;
1062                         }
1063                 }
1064
1065                 /* Snapshot metadata, */
1066                 ret = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1067                                 1, ksess->uid, ksess->gid,
1068                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0,
1069                                 trace_archive_id);
1070                 if (ret < 0) {
1071                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1072                         goto error_consumer;
1073                 }
1074
1075                 /*
1076                  * The metadata snapshot is done, ask the consumer to destroy it since
1077                  * it's not monitored on the consumer side.
1078                  */
1079                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1080         }
1081
1082         ret = LTTNG_OK;
1083
1084 error_consumer:
1085         /* Close newly opened metadata stream. It's now on the consumer side. */
1086         err = close(ksess->metadata_stream_fd);
1087         if (err < 0) {
1088                 PERROR("close snapshot kernel");
1089         }
1090
1091 error_open_stream:
1092         trace_kernel_destroy_metadata(ksess->metadata);
1093 error:
1094         /* Restore metadata state.*/
1095         ksess->metadata = saved_metadata;
1096         ksess->metadata_stream_fd = saved_metadata_fd;
1097
1098         rcu_read_unlock();
1099         return ret;
1100 }
1101
1102 /*
1103  * Get the syscall mask array from the kernel tracer.
1104  *
1105  * Return 0 on success else a negative value. In both case, syscall_mask should
1106  * be freed.
1107  */
1108 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1109 {
1110         assert(syscall_mask);
1111         assert(nr_bits);
1112
1113         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1114 }
1115
1116 /*
1117  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1118  * version number.
1119  *
1120  * Return 1 on success, 0 when feature is not supported, negative value in case
1121  * of errors.
1122  */
1123 int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd)
1124 {
1125         int ret = 0; // Not supported by default
1126         struct lttng_kernel_tracer_abi_version abi;
1127
1128         ret = kernctl_tracer_abi_version(tracer_fd, &abi);
1129         if (ret < 0) {
1130                 ERR("Failed to retrieve lttng-modules ABI version");
1131                 goto error;
1132         }
1133
1134         /*
1135          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1136          */
1137         if (abi.major >= 2 && abi.minor >= 3) {
1138                 /* Supported */
1139                 ret = 1;
1140         } else {
1141                 /* Not supported */
1142                 ret = 0;
1143         }
1144 error:
1145         return ret;
1146 }
1147
1148 /*
1149  * Rotate a kernel session.
1150  *
1151  * Return 0 on success or else return a LTTNG_ERR code.
1152  */
1153 int kernel_rotate_session(struct ltt_session *session)
1154 {
1155         int ret;
1156         struct consumer_socket *socket;
1157         struct lttng_ht_iter iter;
1158         struct ltt_kernel_session *ksess = session->kernel_session;
1159
1160         assert(ksess);
1161         assert(ksess->consumer);
1162
1163         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1164                         session->name, session->id);
1165
1166         rcu_read_lock();
1167
1168         /*
1169          * Note that this loop will end after one iteration given that there is
1170          * only one kernel consumer.
1171          */
1172         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1173                         socket, node.node) {
1174                 struct ltt_kernel_channel *chan;
1175
1176                 /*
1177                  * Account the metadata channel first to make sure the
1178                  * number of channels waiting for a rotation cannot
1179                  * reach 0 before we complete the iteration over all
1180                  * the channels.
1181                  */
1182                 ret = rotate_add_channel_pending(ksess->metadata->key,
1183                                 LTTNG_DOMAIN_KERNEL, session);
1184                 if (ret < 0) {
1185                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1186                         goto error;
1187                 }
1188
1189                 /* For each channel, ask the consumer to rotate it. */
1190                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1191                         ret = rotate_add_channel_pending(chan->key,
1192                                         LTTNG_DOMAIN_KERNEL, session);
1193                         if (ret < 0) {
1194                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1195                                 goto error;
1196                         }
1197
1198                         DBG("Rotate channel %" PRIu64 ", session %s", chan->key, session->name);
1199                         ret = consumer_rotate_channel(socket, chan->key,
1200                                         ksess->uid, ksess->gid, ksess->consumer,
1201                                         ksess->consumer->subdir,
1202                                         /* is_metadata_channel */ false,
1203                                         session->current_archive_id,
1204                                         &session->rotate_pending_relay);
1205                         if (ret < 0) {
1206                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1207                                 goto error;
1208                         }
1209                 }
1210
1211                 /*
1212                  * Rotate the metadata channel.
1213                  */
1214                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1215                                 ksess->uid, ksess->gid, ksess->consumer,
1216                                 ksess->consumer->subdir,
1217                                 /* is_metadata_channel */ true,
1218                                 session->current_archive_id,
1219                                 &session->rotate_pending_relay);
1220                 if (ret < 0) {
1221                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1222                         goto error;
1223                 }
1224         }
1225
1226         ret = LTTNG_OK;
1227
1228 error:
1229         rcu_read_unlock();
1230         return ret;
1231 }