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
  26 #include <common/common.h>
  27 #include <common/kernel-ctl/kernel-ctl.h>
  28 #include <common/sessiond-comm/sessiond-comm.h>
  29
  30 #include "kernel.h"
  31 #include "kern-modules.h"
  32
  33 /*
  34  * Add context on a kernel channel.
  35  */
  36 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  37                 struct lttng_kernel_context *ctx)
  38 {
  39         int ret;
  40
  41         DBG("Adding context to channel %s", chan->channel->name);
  42         ret = kernctl_add_context(chan->fd, ctx);
  43         if (ret < 0) {
  44                 if (errno != EEXIST) {
  45                         PERROR("add context ioctl");
  46                 } else {
  47                         /* If EEXIST, we just ignore the error */
  48                         ret = 0;
  49                 }
  50                 goto error;
  51         }
  52
  53         chan->ctx = zmalloc(sizeof(struct lttng_kernel_context));
  54         if (chan->ctx == NULL) {
  55                 PERROR("zmalloc event context");
  56                 goto error;
  57         }
  58
  59         memcpy(chan->ctx, ctx, sizeof(struct lttng_kernel_context));
  60
  61         return 0;
  62
  63 error:
  64         return ret;
  65 }
  66
  67 /*
  68  * Add context on a kernel event.
  69  */
  70 int kernel_add_event_context(struct ltt_kernel_event *event,
  71                 struct lttng_kernel_context *ctx)
  72 {
  73         int ret;
  74
  75         DBG("Adding context to event %s", event->event->name);
  76         ret = kernctl_add_context(event->fd, ctx);
  77         if (ret < 0) {
  78                 PERROR("add context ioctl");
  79                 goto error;
  80         }
  81
  82         event->ctx = zmalloc(sizeof(struct lttng_kernel_context));
  83         if (event->ctx == NULL) {
  84                 PERROR("zmalloc event context");
  85                 goto error;
  86         }
  87
  88         memcpy(event->ctx, ctx, sizeof(struct lttng_kernel_context));
  89
  90         return 0;
  91
  92 error:
  93         return ret;
  94 }
  95
  96 /*
  97  * Create a new kernel session, register it to the kernel tracer and add it to
  98  * the session daemon session.
  99  */
 100 int kernel_create_session(struct ltt_session *session, int tracer_fd)
 101 {
 102         int ret;
 103         struct ltt_kernel_session *lks;
 104
 105         /* Allocate data structure */
 106         lks = trace_kernel_create_session(session->path);
 107         if (lks == NULL) {
 108                 ret = -1;
 109                 goto error;
 110         }
 111
 112         /* Kernel tracer session creation */
 113         ret = kernctl_create_session(tracer_fd);
 114         if (ret < 0) {
 115                 PERROR("ioctl kernel create session");
 116                 goto error;
 117         }
 118
 119         lks->fd = ret;
 120         /* Prevent fd duplication after execlp() */
 121         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 122         if (ret < 0) {
 123                 PERROR("fcntl session fd");
 124         }
 125
 126         lks->consumer_fds_sent = 0;
 127         session->kernel_session = lks;
 128
 129         DBG("Kernel session created (fd: %d)", lks->fd);
 130
 131         return 0;
 132
 133 error:
 134         return ret;
 135 }
 136
 137 /*
 138  * Create a kernel channel, register it to the kernel tracer and add it to the
 139  * kernel session.
 140  */
 141 int kernel_create_channel(struct ltt_kernel_session *session,
 142                 struct lttng_channel *chan, char *path)
 143 {
 144         int ret;
 145         struct ltt_kernel_channel *lkc;
 146
 147         /* Allocate kernel channel */
 148         lkc = trace_kernel_create_channel(chan, path);
 149         if (lkc == NULL) {
 150                 goto error;
 151         }
 152
 153         DBG3("Kernel create channel %s in %s with attr: %d, %zu, %zu, %u, %u, %d",
 154                         chan->name, path, lkc->channel->attr.overwrite,
 155                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 156                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 157                         lkc->channel->attr.output);
 158
 159         /* Kernel tracer channel creation */
 160         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 161         if (ret < 0) {
 162                 PERROR("ioctl kernel create channel");
 163                 goto error;
 164         }
 165
 166         /* Setup the channel fd */
 167         lkc->fd = ret;
 168         /* Prevent fd duplication after execlp() */
 169         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 170         if (ret < 0) {
 171                 PERROR("fcntl session fd");
 172         }
 173
 174         /* Add channel to session */
 175         cds_list_add(&lkc->list, &session->channel_list.head);
 176         session->channel_count++;
 177
 178         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 179
 180         return 0;
 181
 182 error:
 183         return -1;
 184 }
 185
 186 /*
 187  * Create a kernel event, enable it to the kernel tracer and add it to the
 188  * channel event list of the kernel session.
 189  */
 190 int kernel_create_event(struct lttng_event *ev,
 191                 struct ltt_kernel_channel *channel)
 192 {
 193         int ret;
 194         struct ltt_kernel_event *event;
 195
 196         event = trace_kernel_create_event(ev);
 197         if (event == NULL) {
 198                 ret = -1;
 199                 goto error;
 200         }
 201
 202         ret = kernctl_create_event(channel->fd, event->event);
 203         if (ret < 0) {
 204                 switch (errno) {
 205                 case EEXIST:
 206                         break;
 207                 case ENOSYS:
 208                         WARN("Event type not implemented");
 209                         break;
 210                 default:
 211                         PERROR("create event ioctl");
 212                 }
 213                 ret = -errno;
 214                 goto free_event;
 215         }
 216
 217         /*
 218          * LTTNG_KERNEL_SYSCALL event creation will return 0 on success.
 219          */
 220         if (ret == 0 && event->event->instrumentation == LTTNG_KERNEL_SYSCALL) {
 221                 DBG2("Kernel event syscall creation success");
 222                 /*
 223                  * We use fd == -1 to ensure that we never trigger a close of fd
 224                  * 0.
 225                  */
 226                 event->fd = -1;
 227                 goto add_list;
 228         }
 229
 230         event->fd = ret;
 231         /* Prevent fd duplication after execlp() */
 232         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 233         if (ret < 0) {
 234                 PERROR("fcntl session fd");
 235         }
 236
 237 add_list:
 238         /* Add event to event list */
 239         cds_list_add(&event->list, &channel->events_list.head);
 240         channel->event_count++;
 241
 242         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 243
 244         return 0;
 245
 246 free_event:
 247         free(event);
 248 error:
 249         return ret;
 250 }
 251
 252 /*
 253  * Disable a kernel channel.
 254  */
 255 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 256 {
 257         int ret;
 258
 259         ret = kernctl_disable(chan->fd);
 260         if (ret < 0) {
 261                 PERROR("disable chan ioctl");
 262                 ret = errno;
 263                 goto error;
 264         }
 265
 266         chan->enabled = 0;
 267         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 268
 269         return 0;
 270
 271 error:
 272         return ret;
 273 }
 274
 275 /*
 276  * Enable a kernel channel.
 277  */
 278 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 279 {
 280         int ret;
 281
 282         ret = kernctl_enable(chan->fd);
 283         if (ret < 0 && errno != EEXIST) {
 284                 PERROR("Enable kernel chan");
 285                 goto error;
 286         }
 287
 288         chan->enabled = 1;
 289         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 290
 291         return 0;
 292
 293 error:
 294         return ret;
 295 }
 296
 297 /*
 298  * Enable a kernel event.
 299  */
 300 int kernel_enable_event(struct ltt_kernel_event *event)
 301 {
 302         int ret;
 303
 304         ret = kernctl_enable(event->fd);
 305         if (ret < 0) {
 306                 switch (errno) {
 307                 case EEXIST:
 308                         ret = LTTCOMM_KERN_EVENT_EXIST;
 309                         break;
 310                 default:
 311                         PERROR("enable kernel event");
 312                         break;
 313                 }
 314                 goto error;
 315         }
 316
 317         event->enabled = 1;
 318         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 319
 320         return 0;
 321
 322 error:
 323         return ret;
 324 }
 325
 326 /*
 327  * Disable a kernel event.
 328  */
 329 int kernel_disable_event(struct ltt_kernel_event *event)
 330 {
 331         int ret;
 332
 333         ret = kernctl_disable(event->fd);
 334         if (ret < 0) {
 335                 switch (errno) {
 336                 case EEXIST:
 337                         ret = LTTCOMM_KERN_EVENT_EXIST;
 338                         break;
 339                 default:
 340                         PERROR("disable kernel event");
 341                         break;
 342                 }
 343                 goto error;
 344         }
 345
 346         event->enabled = 0;
 347         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 348
 349         return 0;
 350
 351 error:
 352         return ret;
 353 }
 354
 355 /*
 356  * Create kernel metadata, open from the kernel tracer and add it to the
 357  * kernel session.
 358  */
 359 int kernel_open_metadata(struct ltt_kernel_session *session, char *path)
 360 {
 361         int ret;
 362         struct ltt_kernel_metadata *lkm;
 363
 364         /* Allocate kernel metadata */
 365         lkm = trace_kernel_create_metadata(path);
 366         if (lkm == NULL) {
 367                 goto error;
 368         }
 369
 370         /* Kernel tracer metadata creation */
 371         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 372         if (ret < 0) {
 373                 goto error;
 374         }
 375
 376         lkm->fd = ret;
 377         /* Prevent fd duplication after execlp() */
 378         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 379         if (ret < 0) {
 380                 PERROR("fcntl session fd");
 381         }
 382
 383         session->metadata = lkm;
 384
 385         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 386
 387         return 0;
 388
 389 error:
 390         return -1;
 391 }
 392
 393 /*
 394  * Start tracing session.
 395  */
 396 int kernel_start_session(struct ltt_kernel_session *session)
 397 {
 398         int ret;
 399
 400         ret = kernctl_start_session(session->fd);
 401         if (ret < 0) {
 402                 PERROR("ioctl start session");
 403                 goto error;
 404         }
 405
 406         DBG("Kernel session started");
 407
 408         return 0;
 409
 410 error:
 411         return ret;
 412 }
 413
 414 /*
 415  * Make a kernel wait to make sure in-flight probe have completed.
 416  */
 417 void kernel_wait_quiescent(int fd)
 418 {
 419         int ret;
 420
 421         DBG("Kernel quiescent wait on %d", fd);
 422
 423         ret = kernctl_wait_quiescent(fd);
 424         if (ret < 0) {
 425                 PERROR("wait quiescent ioctl");
 426                 ERR("Kernel quiescent wait failed");
 427         }
 428 }
 429
 430 /*
 431  * Kernel calibrate
 432  */
 433 int kernel_calibrate(int fd, struct lttng_kernel_calibrate *calibrate)
 434 {
 435         int ret;
 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         DBG("Flush buffer for channel %s", channel->channel->name);
 471
 472         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 473                 DBG("Flushing channel stream %d", stream->fd);
 474                 ret = kernctl_buffer_flush(stream->fd);
 475                 if (ret < 0) {
 476                         PERROR("ioctl");
 477                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 478                                         stream->fd, ret);
 479                 }
 480         }
 481
 482         return 0;
 483 }
 484
 485 /*
 486  * Stop tracing session.
 487  */
 488 int kernel_stop_session(struct ltt_kernel_session *session)
 489 {
 490         int ret;
 491
 492         ret = kernctl_stop_session(session->fd);
 493         if (ret < 0) {
 494                 goto error;
 495         }
 496
 497         DBG("Kernel session stopped");
 498
 499         return 0;
 500
 501 error:
 502         return ret;
 503 }
 504
 505 /*
 506  * Open stream of channel, register it to the kernel tracer and add it
 507  * to the stream list of the channel.
 508  *
 509  * Return the number of created stream. Else, a negative value.
 510  */
 511 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 512 {
 513         int ret, count = 0;
 514         struct ltt_kernel_stream *lks;
 515
 516         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 517                 lks = trace_kernel_create_stream(channel->channel->name, count);
 518                 if (lks == NULL) {
 519                         ret = close(ret);
 520                         if (ret) {
 521                                 PERROR("close");
 522                         }
 523                         goto error;
 524                 }
 525
 526                 lks->fd = ret;
 527                 /* Prevent fd duplication after execlp() */
 528                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 529                 if (ret < 0) {
 530                         PERROR("fcntl session fd");
 531                 }
 532
 533                 /* Add stream to channe stream list */
 534                 cds_list_add(&lks->list, &channel->stream_list.head);
 535                 channel->stream_count++;
 536
 537                 /* Increment counter which represent CPU number. */
 538                 count++;
 539
 540                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 541                                 lks->state);
 542         }
 543
 544         return channel->stream_count;
 545
 546 error:
 547         return -1;
 548 }
 549
 550 /*
 551  * Open the metadata stream and set it to the kernel session.
 552  */
 553 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 554 {
 555         int ret;
 556
 557         ret = kernctl_create_stream(session->metadata->fd);
 558         if (ret < 0) {
 559                 PERROR("kernel create metadata stream");
 560                 goto error;
 561         }
 562
 563         DBG("Kernel metadata stream created (fd: %d)", ret);
 564         session->metadata_stream_fd = ret;
 565         /* Prevent fd duplication after execlp() */
 566         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 567         if (ret < 0) {
 568                 PERROR("fcntl session fd");
 569         }
 570
 571         return 0;
 572
 573 error:
 574         return -1;
 575 }
 576
 577 /*
 578  * Get the event list from the kernel tracer and return the number of elements.
 579  */
 580 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 581 {
 582         int fd, pos, ret;
 583         char *event;
 584         size_t nbmem, count = 0;
 585         ssize_t size;
 586         FILE *fp;
 587         struct lttng_event *elist;
 588
 589         fd = kernctl_tracepoint_list(tracer_fd);
 590         if (fd < 0) {
 591                 PERROR("kernel tracepoint list");
 592                 goto error;
 593         }
 594
 595         fp = fdopen(fd, "r");
 596         if (fp == NULL) {
 597                 PERROR("kernel tracepoint list fdopen");
 598                 goto error_fp;
 599         }
 600
 601         /*
 602          * Init memory size counter
 603          * See kernel-ctl.h for explanation of this value
 604          */
 605         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 606         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 607         if (elist == NULL) {
 608                 PERROR("alloc list events");
 609                 count = -ENOMEM;
 610                 goto end;
 611         }
 612
 613         while ((size = fscanf(fp, "event { name = %m[^;]; };%n\n", &event, &pos)) == 1) {
 614                 if (count >= nbmem) {
 615                         struct lttng_event *new_elist;
 616
 617                         DBG("Reallocating event list from %zu to %zu bytes", nbmem,
 618                                         nbmem * 2);
 619                         /* Double the size */
 620                         nbmem <<= 1;
 621                         new_elist = realloc(elist, nbmem * sizeof(struct lttng_event));
 622                         if (new_elist == NULL) {
 623                                 PERROR("realloc list events");
 624                                 free(event);
 625                                 free(elist);
 626                                 count = -ENOMEM;
 627                                 goto end;
 628                         }
 629                         elist = new_elist;
 630                 }
 631                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 632                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 633                 elist[count].enabled = -1;
 634                 count++;
 635                 free(event);
 636         }
 637
 638         *events = elist;
 639         DBG("Kernel list events done (%zu events)", count);
 640 end:
 641         ret = fclose(fp);       /* closes both fp and fd */
 642         if (ret) {
 643                 PERROR("fclose");
 644         }
 645         return count;
 646
 647 error_fp:
 648         ret = close(fd);
 649         if (ret) {
 650                 PERROR("close");
 651         }
 652 error:
 653         return -1;
 654 }
 655
 656 /*
 657  * Get kernel version and validate it.
 658  */
 659 int kernel_validate_version(int tracer_fd)
 660 {
 661         int ret;
 662         struct lttng_kernel_tracer_version version;
 663
 664         ret = kernctl_tracer_version(tracer_fd, &version);
 665         if (ret < 0) {
 666                 ERR("Failed at getting the lttng-modules version");
 667                 goto error;
 668         }
 669
 670         /* Validate version */
 671         if (version.major != KERN_MODULES_PRE_MAJOR
 672                 && version.major != KERN_MODULES_MAJOR) {
 673                 goto error_version;
 674         }
 675
 676         DBG2("Kernel tracer version validated (major version %d)", version.major);
 677         return 0;
 678
 679 error_version:
 680         ERR("Kernel major version %d is not compatible (supporting <= %d)",
 681                         version.major, KERN_MODULES_MAJOR)
 682         ret = -1;
 683
 684 error:
 685         return ret;
 686 }
 687
 688 /*
 689  * Kernel work-arounds called at the start of sessiond main().
 690  */
 691 int init_kernel_workarounds(void)
 692 {
 693         int ret;
 694         FILE *fp;
 695
 696         /*
 697          * boot_id needs to be read once before being used concurrently
 698          * to deal with a Linux kernel race. A fix is proposed for
 699          * upstream, but the work-around is needed for older kernels.
 700          */
 701         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
 702         if (!fp) {
 703                 goto end_boot_id;
 704         }
 705         while (!feof(fp)) {
 706                 char buf[37] = "";
 707
 708                 ret = fread(buf, 1, sizeof(buf), fp);
 709                 if (ret < 0) {
 710                         /* Ignore error, we don't really care */
 711                 }
 712         }
 713         ret = fclose(fp);
 714         if (ret) {
 715                 PERROR("fclose");
 716         }
 717 end_boot_id:
 718         return 0;
 719 }