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