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