src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _GNU_SOURCE
  19 #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/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kern-modules.h"
  34
  35 /*
  36  * Add context on a kernel channel.
  37  */
  38 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  39                 struct lttng_kernel_context *ctx)
  40 {
  41         int ret;
  42
  43         assert(chan);
  44         assert(ctx);
  45
  46         DBG("Adding context to channel %s", chan->channel->name);
  47         ret = kernctl_add_context(chan->fd, ctx);
  48         if (ret < 0) {
  49                 if (errno != EEXIST) {
  50                         PERROR("add context ioctl");
  51                 } else {
  52                         /* If EEXIST, we just ignore the error */
  53                         ret = 0;
  54                 }
  55                 goto error;
  56         }
  57
  58         chan->ctx = zmalloc(sizeof(struct lttng_kernel_context));
  59         if (chan->ctx == NULL) {
  60                 PERROR("zmalloc event context");
  61                 goto error;
  62         }
  63
  64         memcpy(chan->ctx, ctx, sizeof(struct lttng_kernel_context));
  65
  66         return 0;
  67
  68 error:
  69         return ret;
  70 }
  71
  72 /*
  73  * Create a new kernel session, register it to the kernel tracer and add it to
  74  * the session daemon session.
  75  */
  76 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  77 {
  78         int ret;
  79         struct ltt_kernel_session *lks;
  80
  81         assert(session);
  82
  83         /* Allocate data structure */
  84         lks = trace_kernel_create_session();
  85         if (lks == NULL) {
  86                 ret = -1;
  87                 goto error;
  88         }
  89
  90         /* Kernel tracer session creation */
  91         ret = kernctl_create_session(tracer_fd);
  92         if (ret < 0) {
  93                 PERROR("ioctl kernel create session");
  94                 goto error;
  95         }
  96
  97         lks->fd = ret;
  98         /* Prevent fd duplication after execlp() */
  99         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 100         if (ret < 0) {
 101                 PERROR("fcntl session fd");
 102         }
 103
 104         lks->id = session->id;
 105         lks->consumer_fds_sent = 0;
 106         session->kernel_session = lks;
 107
 108         DBG("Kernel session created (fd: %d)", lks->fd);
 109
 110         return 0;
 111
 112 error:
 113         return ret;
 114 }
 115
 116 /*
 117  * Create a kernel channel, register it to the kernel tracer and add it to the
 118  * kernel session.
 119  */
 120 int kernel_create_channel(struct ltt_kernel_session *session,
 121                 struct lttng_channel *chan)
 122 {
 123         int ret;
 124         struct ltt_kernel_channel *lkc;
 125
 126         assert(session);
 127         assert(chan);
 128
 129         /* Allocate kernel channel */
 130         lkc = trace_kernel_create_channel(chan);
 131         if (lkc == NULL) {
 132                 goto error;
 133         }
 134
 135         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d",
 136                         chan->name, lkc->channel->attr.overwrite,
 137                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 138                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 139                         lkc->channel->attr.output);
 140
 141         /* Kernel tracer channel creation */
 142         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 143         if (ret < 0) {
 144                 PERROR("ioctl kernel create channel");
 145                 goto error;
 146         }
 147
 148         /* Setup the channel fd */
 149         lkc->fd = ret;
 150         /* Prevent fd duplication after execlp() */
 151         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 152         if (ret < 0) {
 153                 PERROR("fcntl session fd");
 154         }
 155
 156         /* Add channel to session */
 157         cds_list_add(&lkc->list, &session->channel_list.head);
 158         session->channel_count++;
 159         lkc->session = session;
 160
 161         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 162
 163         return 0;
 164
 165 error:
 166         return -1;
 167 }
 168
 169 /*
 170  * Create a kernel event, enable it to the kernel tracer and add it to the
 171  * channel event list of the kernel session.
 172  */
 173 int kernel_create_event(struct lttng_event *ev,
 174                 struct ltt_kernel_channel *channel)
 175 {
 176         int ret;
 177         struct ltt_kernel_event *event;
 178
 179         assert(ev);
 180         assert(channel);
 181
 182         event = trace_kernel_create_event(ev);
 183         if (event == NULL) {
 184                 ret = -1;
 185                 goto error;
 186         }
 187
 188         ret = kernctl_create_event(channel->fd, event->event);
 189         if (ret < 0) {
 190                 switch (errno) {
 191                 case EEXIST:
 192                         break;
 193                 case ENOSYS:
 194                         WARN("Event type not implemented");
 195                         break;
 196                 default:
 197                         PERROR("create event ioctl");
 198                 }
 199                 ret = -errno;
 200                 goto free_event;
 201         }
 202
 203         /*
 204          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 205          */
 206         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 207                 DBG2("Kernel event syscall creation success");
 208                 /*
 209                  * We use fd == -1 to ensure that we never trigger a close of fd
 210                  * 0.
 211                  */
 212                 event->fd = -1;
 213                 goto add_list;
 214         }
 215
 216         event->fd = ret;
 217         /* Prevent fd duplication after execlp() */
 218         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 219         if (ret < 0) {
 220                 PERROR("fcntl session fd");
 221         }
 222
 223 add_list:
 224         /* Add event to event list */
 225         cds_list_add(&event->list, &channel->events_list.head);
 226         channel->event_count++;
 227
 228         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 229
 230         return 0;
 231
 232 free_event:
 233         free(event);
 234 error:
 235         return ret;
 236 }
 237
 238 /*
 239  * Disable a kernel channel.
 240  */
 241 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 242 {
 243         int ret;
 244
 245         assert(chan);
 246
 247         ret = kernctl_disable(chan->fd);
 248         if (ret < 0) {
 249                 PERROR("disable chan ioctl");
 250                 ret = errno;
 251                 goto error;
 252         }
 253
 254         chan->enabled = 0;
 255         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 256
 257         return 0;
 258
 259 error:
 260         return ret;
 261 }
 262
 263 /*
 264  * Enable a kernel channel.
 265  */
 266 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 267 {
 268         int ret;
 269
 270         assert(chan);
 271
 272         ret = kernctl_enable(chan->fd);
 273         if (ret < 0 && errno != EEXIST) {
 274                 PERROR("Enable kernel chan");
 275                 goto error;
 276         }
 277
 278         chan->enabled = 1;
 279         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 280
 281         return 0;
 282
 283 error:
 284         return ret;
 285 }
 286
 287 /*
 288  * Enable a kernel event.
 289  */
 290 int kernel_enable_event(struct ltt_kernel_event *event)
 291 {
 292         int ret;
 293
 294         assert(event);
 295
 296         ret = kernctl_enable(event->fd);
 297         if (ret < 0) {
 298                 switch (errno) {
 299                 case EEXIST:
 300                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 301                         break;
 302                 default:
 303                         PERROR("enable kernel event");
 304                         break;
 305                 }
 306                 goto error;
 307         }
 308
 309         event->enabled = 1;
 310         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 311
 312         return 0;
 313
 314 error:
 315         return ret;
 316 }
 317
 318 /*
 319  * Disable a kernel event.
 320  */
 321 int kernel_disable_event(struct ltt_kernel_event *event)
 322 {
 323         int ret;
 324
 325         assert(event);
 326
 327         ret = kernctl_disable(event->fd);
 328         if (ret < 0) {
 329                 switch (errno) {
 330                 case EEXIST:
 331                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 332                         break;
 333                 default:
 334                         PERROR("disable kernel event");
 335                         break;
 336                 }
 337                 goto error;
 338         }
 339
 340         event->enabled = 0;
 341         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 342
 343         return 0;
 344
 345 error:
 346         return ret;
 347 }
 348
 349 /*
 350  * Create kernel metadata, open from the kernel tracer and add it to the
 351  * kernel session.
 352  */
 353 int kernel_open_metadata(struct ltt_kernel_session *session)
 354 {
 355         int ret;
 356         struct ltt_kernel_metadata *lkm;
 357
 358         assert(session);
 359
 360         /* Allocate kernel metadata */
 361         lkm = trace_kernel_create_metadata();
 362         if (lkm == NULL) {
 363                 goto error;
 364         }
 365
 366         /* Kernel tracer metadata creation */
 367         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 368         if (ret < 0) {
 369                 goto error;
 370         }
 371
 372         lkm->fd = ret;
 373         /* Prevent fd duplication after execlp() */
 374         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 375         if (ret < 0) {
 376                 PERROR("fcntl session fd");
 377         }
 378
 379         session->metadata = lkm;
 380
 381         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 382
 383         return 0;
 384
 385 error:
 386         return -1;
 387 }
 388
 389 /*
 390  * Start tracing session.
 391  */
 392 int kernel_start_session(struct ltt_kernel_session *session)
 393 {
 394         int ret;
 395
 396         assert(session);
 397
 398         ret = kernctl_start_session(session->fd);
 399         if (ret < 0) {
 400                 PERROR("ioctl start session");
 401                 goto error;
 402         }
 403
 404         DBG("Kernel session started");
 405
 406         return 0;
 407
 408 error:
 409         return ret;
 410 }
 411
 412 /*
 413  * Make a kernel wait to make sure in-flight probe have completed.
 414  */
 415 void kernel_wait_quiescent(int fd)
 416 {
 417         int ret;
 418
 419         DBG("Kernel quiescent wait on %d", fd);
 420
 421         ret = kernctl_wait_quiescent(fd);
 422         if (ret < 0) {
 423                 PERROR("wait quiescent ioctl");
 424                 ERR("Kernel quiescent wait failed");
 425         }
 426 }
 427
 428 /*
 429  * Kernel calibrate
 430  */
 431 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 432 {
 433         int ret;
 434
 435         assert(calibrate);
 436
 437         ret = kernctl_calibrate(fd, calibrate);
 438         if (ret < 0) {
 439                 PERROR("calibrate ioctl");
 440                 return -1;
 441         }
 442
 443         return 0;
 444 }
 445
 446
 447 /*
 448  *  Force flush buffer of metadata.
 449  */
 450 int kernel_metadata_flush_buffer(int fd)
 451 {
 452         int ret;
 453
 454         ret = kernctl_buffer_flush(fd);
 455         if (ret < 0) {
 456                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 457         }
 458
 459         return 0;
 460 }
 461
 462 /*
 463  * Force flush buffer for channel.
 464  */
 465 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 466 {
 467         int ret;
 468         struct ltt_kernel_stream *stream;
 469
 470         assert(channel);
 471
 472         DBG("Flush buffer for channel %s", channel->channel->name);
 473
 474         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 475                 DBG("Flushing channel stream %d", stream->fd);
 476                 ret = kernctl_buffer_flush(stream->fd);
 477                 if (ret < 0) {
 478                         PERROR("ioctl");
 479                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 480                                         stream->fd, ret);
 481                 }
 482         }
 483
 484         return 0;
 485 }
 486
 487 /*
 488  * Stop tracing session.
 489  */
 490 int kernel_stop_session(struct ltt_kernel_session *session)
 491 {
 492         int ret;
 493
 494         assert(session);
 495
 496         ret = kernctl_stop_session(session->fd);
 497         if (ret < 0) {
 498                 goto error;
 499         }
 500
 501         DBG("Kernel session stopped");
 502
 503         return 0;
 504
 505 error:
 506         return ret;
 507 }
 508
 509 /*
 510  * Open stream of channel, register it to the kernel tracer and add it
 511  * to the stream list of the channel.
 512  *
 513  * Return the number of created stream. Else, a negative value.
 514  */
 515 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 516 {
 517         int ret, count = 0;
 518         struct ltt_kernel_stream *lks;
 519
 520         assert(channel);
 521
 522         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 523                 lks = trace_kernel_create_stream(channel->channel->name, count);
 524                 if (lks == NULL) {
 525                         ret = close(ret);
 526                         if (ret) {
 527                                 PERROR("close");
 528                         }
 529                         goto error;
 530                 }
 531
 532                 lks->fd = ret;
 533                 /* Prevent fd duplication after execlp() */
 534                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 535                 if (ret < 0) {
 536                         PERROR("fcntl session fd");
 537                 }
 538
 539                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 540                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 541
 542                 /* Add stream to channe stream list */
 543                 cds_list_add(&lks->list, &channel->stream_list.head);
 544                 channel->stream_count++;
 545
 546                 /* Increment counter which represent CPU number. */
 547                 count++;
 548
 549                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 550                                 lks->state);
 551         }
 552
 553         return channel->stream_count;
 554
 555 error:
 556         return -1;
 557 }
 558
 559 /*
 560  * Open the metadata stream and set it to the kernel session.
 561  */
 562 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 563 {
 564         int ret;
 565
 566         assert(session);
 567
 568         ret = kernctl_create_stream(session->metadata->fd);
 569         if (ret < 0) {
 570                 PERROR("kernel create metadata stream");
 571                 goto error;
 572         }
 573
 574         DBG("Kernel metadata stream created (fd: %d)", ret);
 575         session->metadata_stream_fd = ret;
 576         /* Prevent fd duplication after execlp() */
 577         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 578         if (ret < 0) {
 579                 PERROR("fcntl session fd");
 580         }
 581
 582         return 0;
 583
 584 error:
 585         return -1;
 586 }
 587
 588 /*
 589  * Get the event list from the kernel tracer and return the number of elements.
 590  */
 591 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 592 {
 593         int fd, pos, ret;
 594         char *event;
 595         size_t nbmem, count = 0;
 596         FILE *fp;
 597         struct lttng_event *elist;
 598
 599         assert(events);
 600
 601         fd = kernctl_tracepoint_list(tracer_fd);
 602         if (fd < 0) {
 603                 PERROR("kernel tracepoint list");
 604                 goto error;
 605         }
 606
 607         fp = fdopen(fd, "r");
 608         if (fp == NULL) {
 609                 PERROR("kernel tracepoint list fdopen");
 610                 goto error_fp;
 611         }
 612
 613         /*
 614          * Init memory size counter
 615          * See kernel-ctl.h for explanation of this value
 616          */
 617         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 618         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 619         if (elist == NULL) {
 620                 PERROR("alloc list events");
 621                 count = -ENOMEM;
 622                 goto end;
 623         }
 624
 625         while (fscanf(fp, "event { name = %m[^;]; };%n\n", &event, &pos) == 1) {
 626                 if (count >= nbmem) {
 627                         struct lttng_event *new_elist;
 628
 629                         DBG("Reallocating event list from %zu to %zu bytes", nbmem,
 630                                         nbmem * 2);
 631                         /* Double the size */
 632                         nbmem <<= 1;
 633                         new_elist = realloc(elist, nbmem * sizeof(struct lttng_event));
 634                         if (new_elist == NULL) {
 635                                 PERROR("realloc list events");
 636                                 free(event);
 637                                 free(elist);
 638                                 count = -ENOMEM;
 639                                 goto end;
 640                         }
 641                         elist = new_elist;
 642                 }
 643                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 644                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 645                 elist[count].enabled = -1;
 646                 count++;
 647                 free(event);
 648         }
 649
 650         *events = elist;
 651         DBG("Kernel list events done (%zu events)", count);
 652 end:
 653         ret = fclose(fp);       /* closes both fp and fd */
 654         if (ret) {
 655                 PERROR("fclose");
 656         }
 657         return count;
 658
 659 error_fp:
 660         ret = close(fd);
 661         if (ret) {
 662                 PERROR("close");
 663         }
 664 error:
 665         return -1;
 666 }
 667
 668 /*
 669  * Get kernel version and validate it.
 670  */
 671 int kernel_validate_version(int tracer_fd)
 672 {
 673         int ret;
 674         struct lttng_kernel_tracer_version version;
 675
 676         ret = kernctl_tracer_version(tracer_fd, &version);
 677         if (ret < 0) {
 678                 ERR("Failed at getting the lttng-modules version");
 679                 goto error;
 680         }
 681
 682         /* Validate version */
 683         if (version.major != KERN_MODULES_PRE_MAJOR
 684                 && version.major != KERN_MODULES_MAJOR) {
 685                 goto error_version;
 686         }
 687
 688         DBG2("Kernel tracer version validated (major version %d)", version.major);
 689         return 0;
 690
 691 error_version:
 692         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 693                         version.major, KERN_MODULES_MAJOR)
 694         ret = -1;
 695
 696 error:
 697         return ret;
 698 }
 699
 700 /*
 701  * Kernel work-arounds called at the start of sessiond main().
 702  */
 703 int init_kernel_workarounds(void)
 704 {
 705         int ret;
 706         FILE *fp;
 707
 708         /*
 709          * boot_id needs to be read once before being used concurrently
 710          * to deal with a Linux kernel race. A fix is proposed for
 711          * upstream, but the work-around is needed for older kernels.
 712          */
 713         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 714         if (!fp) {
 715                 goto end_boot_id;
 716         }
 717         while (!feof(fp)) {
 718                 char buf[37] = "";
 719
 720                 ret = fread(buf, 1, sizeof(buf), fp);
 721                 if (ret < 0) {
 722                         /* Ignore error, we don't really care */
 723                 }
 724         }
 725         ret = fclose(fp);
 726         if (ret) {
 727                 PERROR("fclose");
 728         }
 729 end_boot_id:
 730         return 0;
 731 }
 732
 733 /*
 734  * Complete teardown of a kernel session.
 735  */
 736 void kernel_destroy_session(struct ltt_kernel_session *ksess)
 737 {
 738         if (ksess == NULL) {
 739                 DBG3("No kernel session when tearing down session");
 740                 return;
 741         }
 742
 743         DBG("Tearing down kernel session");
 744
 745         /* Close any relayd session */
 746         consumer_output_send_destroy_relayd(ksess->consumer);
 747
 748         trace_kernel_destroy_session(ksess);
 749 }
 750
 751 /*
 752  * Destroy a kernel channel object. It does not do anything on the tracer side.
 753  */
 754 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
 755 {
 756         struct ltt_kernel_session *ksess = NULL;
 757
 758         assert(kchan);
 759         assert(kchan->channel);
 760
 761         DBG3("Kernel destroy channel %s", kchan->channel->name);
 762
 763         /* Update channel count of associated session. */
 764         if (kchan->session) {
 765                 /* Keep pointer reference so we can update it after the destroy. */
 766                 ksess = kchan->session;
 767         }
 768
 769         trace_kernel_destroy_channel(kchan);
 770
 771         /*
 772          * At this point the kernel channel is not visible anymore. This is safe
 773          * since in order to work on a visible kernel session, the tracing session
 774          * lock (ltt_session.lock) MUST be acquired.
 775          */
 776         if (ksess) {
 777                 ksess->channel_count--;
 778         }
 779 }