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