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