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