src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _LGPL_SOURCE
  19 #include <fcntl.h>
  20 #include <stdlib.h>
  21 #include <stdio.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <inttypes.h>
  25
  26 #include <common/common.h>
  27 #include <common/kernel-ctl/kernel-ctl.h>
  28 #include <common/kernel-ctl/kernel-ioctl.h>
  29 #include <common/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kernel-consumer.h"
  34 #include "kern-modules.h"
  35 #include "utils.h"
  36 #include "rotate.h"
  37
  38 /*
  39  * Key used to reference a channel between the sessiond and the consumer. This
  40  * is only read and updated with the session_list lock held.
  41  */
  42 static uint64_t next_kernel_channel_key;
  43
  44 #include <lttng/userspace-probe.h>
  45 #include <lttng/userspace-probe-internal.h>
  46 /*
  47  * Add context on a kernel channel.
  48  *
  49  * Assumes the ownership of ctx.
  50  */
  51 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  52                 struct ltt_kernel_context *ctx)
  53 {
  54         int ret;
  55
  56         assert(chan);
  57         assert(ctx);
  58
  59         DBG("Adding context to channel %s", chan->channel->name);
  60         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  61         if (ret < 0) {
  62                 switch (-ret) {
  63                 case ENOSYS:
  64                         /* Exists but not available for this kernel */
  65                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  66                         goto error;
  67                 case EEXIST:
  68                         /* If EEXIST, we just ignore the error */
  69                         ret = 0;
  70                         goto end;
  71                 default:
  72                         PERROR("add context ioctl");
  73                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  74                         goto error;
  75                 }
  76         }
  77         ret = 0;
  78
  79 end:
  80         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  81         ctx->in_list = true;
  82         ctx = NULL;
  83 error:
  84         if (ctx) {
  85                 trace_kernel_destroy_context(ctx);
  86         }
  87         return ret;
  88 }
  89
  90 /*
  91  * Create a new kernel session, register it to the kernel tracer and add it to
  92  * the session daemon session.
  93  */
  94 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  95 {
  96         int ret;
  97         struct ltt_kernel_session *lks;
  98
  99         assert(session);
 100
 101         /* Allocate data structure */
 102         lks = trace_kernel_create_session();
 103         if (lks == NULL) {
 104                 ret = -1;
 105                 goto error;
 106         }
 107
 108         /* Kernel tracer session creation */
 109         ret = kernctl_create_session(tracer_fd);
 110         if (ret < 0) {
 111                 PERROR("ioctl kernel create session");
 112                 goto error;
 113         }
 114
 115         lks->fd = ret;
 116         /* Prevent fd duplication after execlp() */
 117         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 118         if (ret < 0) {
 119                 PERROR("fcntl session fd");
 120         }
 121
 122         lks->id = session->id;
 123         lks->consumer_fds_sent = 0;
 124         session->kernel_session = lks;
 125
 126         DBG("Kernel session created (fd: %d)", lks->fd);
 127
 128         return 0;
 129
 130 error:
 131         if (lks) {
 132                 trace_kernel_destroy_session(lks);
 133         }
 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)
 143 {
 144         int ret;
 145         struct ltt_kernel_channel *lkc;
 146
 147         assert(session);
 148         assert(chan);
 149
 150         /* Allocate kernel channel */
 151         lkc = trace_kernel_create_channel(chan);
 152         if (lkc == NULL) {
 153                 goto error;
 154         }
 155
 156         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 157                         chan->name, 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.live_timer_interval, 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         lkc->session = session;
 181         lkc->key = ++next_kernel_channel_key;
 182
 183         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 184                         lkc->channel->name, lkc->fd, lkc->key);
 185
 186         return 0;
 187
 188 error:
 189         if (lkc) {
 190                 free(lkc->channel);
 191                 free(lkc);
 192         }
 193         return -1;
 194 }
 195
 196 /*
 197  * Compute the offset of the instrumentation byte in the binary based on the
 198  * function probe location using the ELF lookup method.
 199  *
 200  * Returns 0 on success and set the offset out parameter to the offset of the
 201  * elf symbol
 202  * Returns -1 on error
 203  */
 204 static
 205 int extract_userspace_probe_offset_function_elf(
 206                 const struct lttng_userspace_probe_location *probe_location,
 207                 struct ltt_kernel_session *session, uint64_t *offset)
 208 {
 209         int fd;
 210         int ret = 0;
 211         const char *symbol = NULL;
 212         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 213         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 214
 215         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 216                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
 217
 218         lookup = lttng_userspace_probe_location_get_lookup_method(
 219                         probe_location);
 220         if (!lookup) {
 221                 ret = -1;
 222                 goto end;
 223         }
 224
 225         lookup_method_type =
 226                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 227
 228         assert(lookup_method_type ==
 229                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
 230
 231         symbol = lttng_userspace_probe_location_function_get_function_name(
 232                         probe_location);
 233         if (!symbol) {
 234                 ret = -1;
 235                 goto end;
 236         }
 237
 238         fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
 239         if (fd < 0) {
 240                 ret = -1;
 241                 goto end;
 242         }
 243
 244         ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
 245                         session->gid, offset);
 246         if (ret < 0) {
 247                 DBG("userspace probe offset calculation failed for "
 248                                 "function %s", symbol);
 249                 goto end;
 250         }
 251
 252         DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
 253 end:
 254         return ret;
 255 }
 256
 257 /*
 258  * Compute the offsets of the instrumentation bytes in the binary based on the
 259  * tracepoint probe location using the SDT lookup method. This function
 260  * allocates the offsets buffer, the caller must free it.
 261  *
 262  * Returns 0 on success and set the offset out parameter to the offsets of the
 263  * SDT tracepoint.
 264  * Returns -1 on error.
 265  */
 266 static
 267 int extract_userspace_probe_offset_tracepoint_sdt(
 268                 const struct lttng_userspace_probe_location *probe_location,
 269                 struct ltt_kernel_session *session, uint64_t **offsets,
 270                 uint32_t *offsets_count)
 271 {
 272         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 273         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 274         const char *probe_name = NULL, *provider_name = NULL;
 275         int ret = 0;
 276         int fd, i;
 277
 278         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 279                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
 280
 281         lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
 282         if (!lookup) {
 283                 ret = -1;
 284                 goto end;
 285         }
 286
 287         lookup_method_type =
 288                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 289
 290         assert(lookup_method_type ==
 291                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
 292
 293
 294         probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
 295                         probe_location);
 296         if (!probe_name) {
 297                 ret = -1;
 298                 goto end;
 299         }
 300
 301         provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
 302                         probe_location);
 303         if (!provider_name) {
 304                 ret = -1;
 305                 goto end;
 306         }
 307
 308         fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
 309         if (fd < 0) {
 310                 ret = -1;
 311                 goto end;
 312         }
 313
 314         ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
 315                         session->uid, session->gid, offsets, offsets_count);
 316         if (ret < 0) {
 317                 DBG("userspace probe offset calculation failed for sdt "
 318                                 "probe %s:%s", provider_name, probe_name);
 319                 goto end;
 320         }
 321
 322         if (*offsets_count == 0) {
 323                 DBG("no userspace probe offset found");
 324                 goto end;
 325         }
 326
 327         DBG("%u userspace probe SDT offsets found for %s:%s at:",
 328                         *offsets_count, provider_name, probe_name);
 329         for (i = 0; i < *offsets_count; i++) {
 330                 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
 331         }
 332 end:
 333         return ret;
 334 }
 335
 336 /*
 337  * Extract the offsets of the instrumentation point for the different lookup
 338  * methods.
 339  */
 340 static
 341 int userspace_probe_add_callsites(struct lttng_event *ev,
 342                         struct ltt_kernel_session *session, int fd)
 343 {
 344         const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
 345         enum lttng_userspace_probe_location_lookup_method_type type;
 346         const struct lttng_userspace_probe_location *location = NULL;
 347         int ret;
 348
 349         assert(ev);
 350         assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
 351
 352         location = lttng_event_get_userspace_probe_location(ev);
 353         if (!location) {
 354                 ret = -1;
 355                 goto end;
 356         }
 357         lookup_method =
 358                         lttng_userspace_probe_location_get_lookup_method(location);
 359         if (!lookup_method) {
 360                 ret = -1;
 361                 goto end;
 362         }
 363
 364         type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
 365         switch (type) {
 366         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
 367         {
 368                 struct lttng_kernel_event_callsite callsite;
 369                 uint64_t offset;
 370
 371                 ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
 372                 if (ret) {
 373                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 374                         goto end;
 375                 }
 376
 377                 callsite.u.uprobe.offset = offset;
 378                 ret = kernctl_add_callsite(fd, &callsite);
 379                 if (ret) {
 380                         WARN("Adding callsite to userspace probe "
 381                                         "event %s failed.", ev->name);
 382                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 383                         goto end;
 384                 }
 385                 break;
 386         }
 387         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
 388         {
 389                 int i;
 390                 uint64_t *offsets = NULL;
 391                 uint32_t offsets_count;
 392                 struct lttng_kernel_event_callsite callsite;
 393
 394                 /*
 395                  * This call allocates the offsets buffer. This buffer must be freed
 396                  * by the caller
 397                  */
 398                 ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
 399                                 &offsets, &offsets_count);
 400                 if (ret) {
 401                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 402                         goto end;
 403                 }
 404                 for (i = 0; i < offsets_count; i++) {
 405                         callsite.u.uprobe.offset = offsets[i];
 406                         ret = kernctl_add_callsite(fd, &callsite);
 407                         if (ret) {
 408                                 WARN("Adding callsite to userspace probe "
 409                                                 "event %s failed.", ev->name);
 410                                 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 411                                 free(offsets);
 412                                 goto end;
 413                         }
 414                 }
 415                 free(offsets);
 416                 break;
 417         }
 418         default:
 419                 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 420                 goto end;
 421         }
 422 end:
 423         return ret;
 424 }
 425
 426 /*
 427  * Create a kernel event, enable it to the kernel tracer and add it to the
 428  * channel event list of the kernel session.
 429  * We own filter_expression and filter.
 430  */
 431 int kernel_create_event(struct lttng_event *ev,
 432                 struct ltt_kernel_channel *channel,
 433                 char *filter_expression,
 434                 struct lttng_filter_bytecode *filter)
 435 {
 436         int err, fd;
 437         enum lttng_error_code ret;
 438         struct ltt_kernel_event *event;
 439
 440         assert(ev);
 441         assert(channel);
 442
 443         /* We pass ownership of filter_expression and filter */
 444         ret = trace_kernel_create_event(ev, filter_expression,
 445                         filter, &event);
 446         if (ret != LTTNG_OK) {
 447                 goto error;
 448         }
 449
 450         fd = kernctl_create_event(channel->fd, event->event);
 451         if (fd < 0) {
 452                 switch (-fd) {
 453                 case EEXIST:
 454                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 455                         break;
 456                 case ENOSYS:
 457                         WARN("Event type not implemented");
 458                         ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
 459                         break;
 460                 case ENOENT:
 461                         WARN("Event %s not found!", ev->name);
 462                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 463                         break;
 464                 default:
 465                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 466                         PERROR("create event ioctl");
 467                 }
 468                 goto free_event;
 469         }
 470
 471         event->type = ev->type;
 472         event->fd = fd;
 473         /* Prevent fd duplication after execlp() */
 474         err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 475         if (err < 0) {
 476                 PERROR("fcntl session fd");
 477         }
 478
 479         if (filter) {
 480                 err = kernctl_filter(event->fd, filter);
 481                 if (err < 0) {
 482                         switch (-err) {
 483                         case ENOMEM:
 484                                 ret = LTTNG_ERR_FILTER_NOMEM;
 485                                 break;
 486                         default:
 487                                 ret = LTTNG_ERR_FILTER_INVAL;
 488                                 break;
 489                         }
 490                         goto filter_error;
 491                 }
 492         }
 493
 494         if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
 495                 ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
 496                 if (ret) {
 497                         goto add_callsite_error;
 498                 }
 499         }
 500
 501         err = kernctl_enable(event->fd);
 502         if (err < 0) {
 503                 switch (-err) {
 504                 case EEXIST:
 505                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 506                         break;
 507                 default:
 508                         PERROR("enable kernel event");
 509                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 510                         break;
 511                 }
 512                 goto enable_error;
 513         }
 514
 515         /* Add event to event list */
 516         cds_list_add(&event->list, &channel->events_list.head);
 517         channel->event_count++;
 518
 519         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 520
 521         return 0;
 522
 523 add_callsite_error:
 524 enable_error:
 525 filter_error:
 526         {
 527                 int closeret;
 528
 529                 closeret = close(event->fd);
 530                 if (closeret) {
 531                         PERROR("close event fd");
 532                 }
 533         }
 534 free_event:
 535         free(event);
 536 error:
 537         return ret;
 538 }
 539
 540 /*
 541  * Disable a kernel channel.
 542  */
 543 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 544 {
 545         int ret;
 546
 547         assert(chan);
 548
 549         ret = kernctl_disable(chan->fd);
 550         if (ret < 0) {
 551                 PERROR("disable chan ioctl");
 552                 goto error;
 553         }
 554
 555         chan->enabled = 0;
 556         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 557                         chan->channel->name, chan->fd, chan->key);
 558
 559         return 0;
 560
 561 error:
 562         return ret;
 563 }
 564
 565 /*
 566  * Enable a kernel channel.
 567  */
 568 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 569 {
 570         int ret;
 571
 572         assert(chan);
 573
 574         ret = kernctl_enable(chan->fd);
 575         if (ret < 0 && ret != -EEXIST) {
 576                 PERROR("Enable kernel chan");
 577                 goto error;
 578         }
 579
 580         chan->enabled = 1;
 581         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 582                         chan->channel->name, chan->fd, chan->key);
 583
 584         return 0;
 585
 586 error:
 587         return ret;
 588 }
 589
 590 /*
 591  * Enable a kernel event.
 592  */
 593 int kernel_enable_event(struct ltt_kernel_event *event)
 594 {
 595         int ret;
 596
 597         assert(event);
 598
 599         ret = kernctl_enable(event->fd);
 600         if (ret < 0) {
 601                 switch (-ret) {
 602                 case EEXIST:
 603                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 604                         break;
 605                 default:
 606                         PERROR("enable kernel event");
 607                         break;
 608                 }
 609                 goto error;
 610         }
 611
 612         event->enabled = 1;
 613         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 614
 615         return 0;
 616
 617 error:
 618         return ret;
 619 }
 620
 621 /*
 622  * Disable a kernel event.
 623  */
 624 int kernel_disable_event(struct ltt_kernel_event *event)
 625 {
 626         int ret;
 627
 628         assert(event);
 629
 630         ret = kernctl_disable(event->fd);
 631         if (ret < 0) {
 632                 switch (-ret) {
 633                 case EEXIST:
 634                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 635                         break;
 636                 default:
 637                         PERROR("disable kernel event");
 638                         break;
 639                 }
 640                 goto error;
 641         }
 642
 643         event->enabled = 0;
 644         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 645
 646         return 0;
 647
 648 error:
 649         return ret;
 650 }
 651
 652
 653 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 654 {
 655         int ret;
 656
 657         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 658                         pid, session->id);
 659         ret = kernctl_track_pid(session->fd, pid);
 660         if (!ret) {
 661                 return LTTNG_OK;
 662         }
 663         switch (-ret) {
 664         case EINVAL:
 665                 return LTTNG_ERR_INVALID;
 666         case ENOMEM:
 667                 return LTTNG_ERR_NOMEM;
 668         case EEXIST:
 669                 return LTTNG_ERR_PID_TRACKED;
 670         default:
 671                 return LTTNG_ERR_UNK;
 672         }
 673 }
 674
 675 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 676 {
 677         int ret;
 678
 679         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 680                         pid, session->id);
 681         ret = kernctl_untrack_pid(session->fd, pid);
 682         if (!ret) {
 683                 return LTTNG_OK;
 684         }
 685         switch (-ret) {
 686         case EINVAL:
 687                 return LTTNG_ERR_INVALID;
 688         case ENOMEM:
 689                 return LTTNG_ERR_NOMEM;
 690         case ENOENT:
 691                 return LTTNG_ERR_PID_NOT_TRACKED;
 692         default:
 693                 return LTTNG_ERR_UNK;
 694         }
 695 }
 696
 697 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 698                 int **_pids)
 699 {
 700         int fd, ret;
 701         int pid;
 702         ssize_t nbmem, count = 0;
 703         FILE *fp;
 704         int *pids;
 705
 706         fd = kernctl_list_tracker_pids(session->fd);
 707         if (fd < 0) {
 708                 PERROR("kernel tracker pids list");
 709                 goto error;
 710         }
 711
 712         fp = fdopen(fd, "r");
 713         if (fp == NULL) {
 714                 PERROR("kernel tracker pids list fdopen");
 715                 goto error_fp;
 716         }
 717
 718         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 719         pids = zmalloc(sizeof(*pids) * nbmem);
 720         if (pids == NULL) {
 721                 PERROR("alloc list pids");
 722                 count = -ENOMEM;
 723                 goto end;
 724         }
 725
 726         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 727                 if (count >= nbmem) {
 728                         int *new_pids;
 729                         size_t new_nbmem;
 730
 731                         new_nbmem = nbmem << 1;
 732                         DBG("Reallocating pids list from %zu to %zu entries",
 733                                         nbmem, new_nbmem);
 734                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 735                         if (new_pids == NULL) {
 736                                 PERROR("realloc list events");
 737                                 free(pids);
 738                                 count = -ENOMEM;
 739                                 goto end;
 740                         }
 741                         /* Zero the new memory */
 742                         memset(new_pids + nbmem, 0,
 743                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 744                         nbmem = new_nbmem;
 745                         pids = new_pids;
 746                 }
 747                 pids[count++] = pid;
 748         }
 749
 750         *_pids = pids;
 751         DBG("Kernel list tracker pids done (%zd pids)", count);
 752 end:
 753         ret = fclose(fp);       /* closes both fp and fd */
 754         if (ret) {
 755                 PERROR("fclose");
 756         }
 757         return count;
 758
 759 error_fp:
 760         ret = close(fd);
 761         if (ret) {
 762                 PERROR("close");
 763         }
 764 error:
 765         return -1;
 766 }
 767
 768 /*
 769  * Create kernel metadata, open from the kernel tracer and add it to the
 770  * kernel session.
 771  */
 772 int kernel_open_metadata(struct ltt_kernel_session *session)
 773 {
 774         int ret;
 775         struct ltt_kernel_metadata *lkm = NULL;
 776
 777         assert(session);
 778
 779         /* Allocate kernel metadata */
 780         lkm = trace_kernel_create_metadata();
 781         if (lkm == NULL) {
 782                 goto error;
 783         }
 784
 785         /* Kernel tracer metadata creation */
 786         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 787         if (ret < 0) {
 788                 goto error_open;
 789         }
 790
 791         lkm->fd = ret;
 792         lkm->key = ++next_kernel_channel_key;
 793         /* Prevent fd duplication after execlp() */
 794         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 795         if (ret < 0) {
 796                 PERROR("fcntl session fd");
 797         }
 798
 799         session->metadata = lkm;
 800
 801         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 802
 803         return 0;
 804
 805 error_open:
 806         trace_kernel_destroy_metadata(lkm);
 807 error:
 808         return -1;
 809 }
 810
 811 /*
 812  * Start tracing session.
 813  */
 814 int kernel_start_session(struct ltt_kernel_session *session)
 815 {
 816         int ret;
 817
 818         assert(session);
 819
 820         ret = kernctl_start_session(session->fd);
 821         if (ret < 0) {
 822                 PERROR("ioctl start session");
 823                 goto error;
 824         }
 825
 826         DBG("Kernel session started");
 827
 828         return 0;
 829
 830 error:
 831         return ret;
 832 }
 833
 834 /*
 835  * Make a kernel wait to make sure in-flight probe have completed.
 836  */
 837 void kernel_wait_quiescent(int fd)
 838 {
 839         int ret;
 840
 841         DBG("Kernel quiescent wait on %d", fd);
 842
 843         ret = kernctl_wait_quiescent(fd);
 844         if (ret < 0) {
 845                 PERROR("wait quiescent ioctl");
 846                 ERR("Kernel quiescent wait failed");
 847         }
 848 }
 849
 850 /*
 851  *  Force flush buffer of metadata.
 852  */
 853 int kernel_metadata_flush_buffer(int fd)
 854 {
 855         int ret;
 856
 857         DBG("Kernel flushing metadata buffer on fd %d", fd);
 858
 859         ret = kernctl_buffer_flush(fd);
 860         if (ret < 0) {
 861                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 862         }
 863
 864         return 0;
 865 }
 866
 867 /*
 868  * Force flush buffer for channel.
 869  */
 870 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 871 {
 872         int ret;
 873         struct ltt_kernel_stream *stream;
 874
 875         assert(channel);
 876
 877         DBG("Flush buffer for channel %s", channel->channel->name);
 878
 879         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 880                 DBG("Flushing channel stream %d", stream->fd);
 881                 ret = kernctl_buffer_flush(stream->fd);
 882                 if (ret < 0) {
 883                         PERROR("ioctl");
 884                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 885                                         stream->fd, ret);
 886                 }
 887         }
 888
 889         return 0;
 890 }
 891
 892 /*
 893  * Stop tracing session.
 894  */
 895 int kernel_stop_session(struct ltt_kernel_session *session)
 896 {
 897         int ret;
 898
 899         assert(session);
 900
 901         ret = kernctl_stop_session(session->fd);
 902         if (ret < 0) {
 903                 goto error;
 904         }
 905
 906         DBG("Kernel session stopped");
 907
 908         return 0;
 909
 910 error:
 911         return ret;
 912 }
 913
 914 /*
 915  * Open stream of channel, register it to the kernel tracer and add it
 916  * to the stream list of the channel.
 917  *
 918  * Note: given that the streams may appear in random order wrt CPU
 919  * number (e.g. cpu hotplug), the index value of the stream number in
 920  * the stream name is not necessarily linked to the CPU number.
 921  *
 922  * Return the number of created stream. Else, a negative value.
 923  */
 924 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 925 {
 926         int ret;
 927         struct ltt_kernel_stream *lks;
 928
 929         assert(channel);
 930
 931         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 932                 lks = trace_kernel_create_stream(channel->channel->name,
 933                                 channel->stream_count);
 934                 if (lks == NULL) {
 935                         ret = close(ret);
 936                         if (ret) {
 937                                 PERROR("close");
 938                         }
 939                         goto error;
 940                 }
 941
 942                 lks->fd = ret;
 943                 /* Prevent fd duplication after execlp() */
 944                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 945                 if (ret < 0) {
 946                         PERROR("fcntl session fd");
 947                 }
 948
 949                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 950                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 951
 952                 /* Add stream to channel stream list */
 953                 cds_list_add(&lks->list, &channel->stream_list.head);
 954                 channel->stream_count++;
 955
 956                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 957                                 lks->state);
 958         }
 959
 960         return channel->stream_count;
 961
 962 error:
 963         return -1;
 964 }
 965
 966 /*
 967  * Open the metadata stream and set it to the kernel session.
 968  */
 969 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 970 {
 971         int ret;
 972
 973         assert(session);
 974
 975         ret = kernctl_create_stream(session->metadata->fd);
 976         if (ret < 0) {
 977                 PERROR("kernel create metadata stream");
 978                 goto error;
 979         }
 980
 981         DBG("Kernel metadata stream created (fd: %d)", ret);
 982         session->metadata_stream_fd = ret;
 983         /* Prevent fd duplication after execlp() */
 984         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 985         if (ret < 0) {
 986                 PERROR("fcntl session fd");
 987         }
 988
 989         return 0;
 990
 991 error:
 992         return -1;
 993 }
 994
 995 /*
 996  * Get the event list from the kernel tracer and return the number of elements.
 997  */
 998 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 999 {
1000         int fd, ret;
1001         char *event;
1002         size_t nbmem, count = 0;
1003         FILE *fp;
1004         struct lttng_event *elist;
1005
1006         assert(events);
1007
1008         fd = kernctl_tracepoint_list(tracer_fd);
1009         if (fd < 0) {
1010                 PERROR("kernel tracepoint list");
1011                 goto error;
1012         }
1013
1014         fp = fdopen(fd, "r");
1015         if (fp == NULL) {
1016                 PERROR("kernel tracepoint list fdopen");
1017                 goto error_fp;
1018         }
1019
1020         /*
1021          * Init memory size counter
1022          * See kernel-ctl.h for explanation of this value
1023          */
1024         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1025         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1026         if (elist == NULL) {
1027                 PERROR("alloc list events");
1028                 count = -ENOMEM;
1029                 goto end;
1030         }
1031
1032         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1033                 if (count >= nbmem) {
1034                         struct lttng_event *new_elist;
1035                         size_t new_nbmem;
1036
1037                         new_nbmem = nbmem << 1;
1038                         DBG("Reallocating event list from %zu to %zu bytes",
1039                                         nbmem, new_nbmem);
1040                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1041                         if (new_elist == NULL) {
1042                                 PERROR("realloc list events");
1043                                 free(event);
1044                                 free(elist);
1045                                 count = -ENOMEM;
1046                                 goto end;
1047                         }
1048                         /* Zero the new memory */
1049                         memset(new_elist + nbmem, 0,
1050                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1051                         nbmem = new_nbmem;
1052                         elist = new_elist;
1053                 }
1054                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1055                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1056                 elist[count].enabled = -1;
1057                 count++;
1058                 free(event);
1059         }
1060
1061         *events = elist;
1062         DBG("Kernel list events done (%zu events)", count);
1063 end:
1064         ret = fclose(fp);       /* closes both fp and fd */
1065         if (ret) {
1066                 PERROR("fclose");
1067         }
1068         return count;
1069
1070 error_fp:
1071         ret = close(fd);
1072         if (ret) {
1073                 PERROR("close");
1074         }
1075 error:
1076         return -1;
1077 }
1078
1079 /*
1080  * Get kernel version and validate it.
1081  */
1082 int kernel_validate_version(int tracer_fd,
1083                 struct lttng_kernel_tracer_version *version,
1084                 struct lttng_kernel_tracer_abi_version *abi_version)
1085 {
1086         int ret;
1087
1088         ret = kernctl_tracer_version(tracer_fd, version);
1089         if (ret < 0) {
1090                 ERR("Failed to retrieve the lttng-modules version");
1091                 goto error;
1092         }
1093
1094         /* Validate version */
1095         if (version->major != VERSION_MAJOR) {
1096                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1097                         version->major, VERSION_MAJOR);
1098                 goto error_version;
1099         }
1100         ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
1101         if (ret < 0) {
1102                 ERR("Failed to retrieve lttng-modules ABI version");
1103                 goto error;
1104         }
1105         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1106                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1107                         abi_version->major, abi_version->minor,
1108                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1109                 goto error;
1110         }
1111         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1112                         version->major, version->minor,
1113                         abi_version->major, abi_version->minor);
1114         return 0;
1115
1116 error_version:
1117         ret = -1;
1118
1119 error:
1120         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1121         return ret;
1122 }
1123
1124 /*
1125  * Kernel work-arounds called at the start of sessiond main().
1126  */
1127 int init_kernel_workarounds(void)
1128 {
1129         int ret;
1130         FILE *fp;
1131
1132         /*
1133          * boot_id needs to be read once before being used concurrently
1134          * to deal with a Linux kernel race. A fix is proposed for
1135          * upstream, but the work-around is needed for older kernels.
1136          */
1137         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1138         if (!fp) {
1139                 goto end_boot_id;
1140         }
1141         while (!feof(fp)) {
1142                 char buf[37] = "";
1143
1144                 ret = fread(buf, 1, sizeof(buf), fp);
1145                 if (ret < 0) {
1146                         /* Ignore error, we don't really care */
1147                 }
1148         }
1149         ret = fclose(fp);
1150         if (ret) {
1151                 PERROR("fclose");
1152         }
1153 end_boot_id:
1154         return 0;
1155 }
1156
1157 /*
1158  * Complete teardown of a kernel session.
1159  */
1160 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1161 {
1162         if (ksess == NULL) {
1163                 DBG3("No kernel session when tearing down session");
1164                 return;
1165         }
1166
1167         DBG("Tearing down kernel session");
1168
1169         /*
1170          * Destroy channels on the consumer if at least one FD has been sent and we
1171          * are in no output mode because the streams are in *no* monitor mode so we
1172          * have to send a command to clean them up or else they leaked.
1173          */
1174         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1175                 int ret;
1176                 struct consumer_socket *socket;
1177                 struct lttng_ht_iter iter;
1178
1179                 /* For each consumer socket. */
1180                 rcu_read_lock();
1181                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1182                                 socket, node.node) {
1183                         struct ltt_kernel_channel *chan;
1184
1185                         /* For each channel, ask the consumer to destroy it. */
1186                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1187                                 ret = kernel_consumer_destroy_channel(socket, chan);
1188                                 if (ret < 0) {
1189                                         /* Consumer is probably dead. Use next socket. */
1190                                         continue;
1191                                 }
1192                         }
1193                 }
1194                 rcu_read_unlock();
1195         }
1196
1197         /* Close any relayd session */
1198         consumer_output_send_destroy_relayd(ksess->consumer);
1199
1200         trace_kernel_destroy_session(ksess);
1201 }
1202
1203 /*
1204  * Destroy a kernel channel object. It does not do anything on the tracer side.
1205  */
1206 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1207 {
1208         struct ltt_kernel_session *ksess = NULL;
1209
1210         assert(kchan);
1211         assert(kchan->channel);
1212
1213         DBG3("Kernel destroy channel %s", kchan->channel->name);
1214
1215         /* Update channel count of associated session. */
1216         if (kchan->session) {
1217                 /* Keep pointer reference so we can update it after the destroy. */
1218                 ksess = kchan->session;
1219         }
1220
1221         trace_kernel_destroy_channel(kchan);
1222
1223         /*
1224          * At this point the kernel channel is not visible anymore. This is safe
1225          * since in order to work on a visible kernel session, the tracing session
1226          * lock (ltt_session.lock) MUST be acquired.
1227          */
1228         if (ksess) {
1229                 ksess->channel_count--;
1230         }
1231 }
1232
1233 /*
1234  * Take a snapshot for a given kernel session.
1235  *
1236  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1237  */
1238 enum lttng_error_code kernel_snapshot_record(struct ltt_kernel_session *ksess,
1239                 struct snapshot_output *output, int wait,
1240                 uint64_t nb_packets_per_stream)
1241 {
1242         int err, ret, saved_metadata_fd;
1243         enum lttng_error_code status = LTTNG_OK;
1244         struct consumer_socket *socket;
1245         struct lttng_ht_iter iter;
1246         struct ltt_kernel_metadata *saved_metadata;
1247         struct ltt_session *session;
1248         uint64_t trace_archive_id;
1249
1250         assert(ksess);
1251         assert(ksess->consumer);
1252         assert(output);
1253
1254         DBG("Kernel snapshot record started");
1255
1256         session = session_find_by_id(ksess->id);
1257         assert(session);
1258         assert(pthread_mutex_trylock(&session->lock));
1259         assert(session_trylock_list());
1260         trace_archive_id = session->current_archive_id;
1261
1262         /* Save current metadata since the following calls will change it. */
1263         saved_metadata = ksess->metadata;
1264         saved_metadata_fd = ksess->metadata_stream_fd;
1265
1266         rcu_read_lock();
1267
1268         ret = kernel_open_metadata(ksess);
1269         if (ret < 0) {
1270                 status = LTTNG_ERR_KERN_META_FAIL;
1271                 goto error;
1272         }
1273
1274         ret = kernel_open_metadata_stream(ksess);
1275         if (ret < 0) {
1276                 status = LTTNG_ERR_KERN_META_FAIL;
1277                 goto error_open_stream;
1278         }
1279
1280         /* Send metadata to consumer and snapshot everything. */
1281         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1282                         socket, node.node) {
1283                 struct consumer_output *saved_output;
1284                 struct ltt_kernel_channel *chan;
1285
1286                 /*
1287                  * Temporarly switch consumer output for our snapshot output. As long
1288                  * as the session lock is taken, this is safe.
1289                  */
1290                 saved_output = ksess->consumer;
1291                 ksess->consumer = output->consumer;
1292
1293                 pthread_mutex_lock(socket->lock);
1294                 /* This stream must not be monitored by the consumer. */
1295                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1296                 pthread_mutex_unlock(socket->lock);
1297                 /* Put back the saved consumer output into the session. */
1298                 ksess->consumer = saved_output;
1299                 if (ret < 0) {
1300                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1301                         goto error_consumer;
1302                 }
1303
1304                 /* For each channel, ask the consumer to snapshot it. */
1305                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1306                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1307                                         ksess->uid, ksess->gid,
1308                                         DEFAULT_KERNEL_TRACE_DIR, wait,
1309                                         nb_packets_per_stream,
1310                                         trace_archive_id);
1311                         if (status != LTTNG_OK) {
1312                                 (void) kernel_consumer_destroy_metadata(socket,
1313                                                 ksess->metadata);
1314                                 goto error_consumer;
1315                         }
1316                 }
1317
1318                 /* Snapshot metadata, */
1319                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1320                                 1, ksess->uid, ksess->gid,
1321                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0,
1322                                 trace_archive_id);
1323                 if (status != LTTNG_OK) {
1324                         goto error_consumer;
1325                 }
1326
1327                 /*
1328                  * The metadata snapshot is done, ask the consumer to destroy it since
1329                  * it's not monitored on the consumer side.
1330                  */
1331                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1332         }
1333
1334 error_consumer:
1335         /* Close newly opened metadata stream. It's now on the consumer side. */
1336         err = close(ksess->metadata_stream_fd);
1337         if (err < 0) {
1338                 PERROR("close snapshot kernel");
1339         }
1340
1341 error_open_stream:
1342         trace_kernel_destroy_metadata(ksess->metadata);
1343 error:
1344         /* Restore metadata state.*/
1345         ksess->metadata = saved_metadata;
1346         ksess->metadata_stream_fd = saved_metadata_fd;
1347
1348         rcu_read_unlock();
1349         return status;
1350 }
1351
1352 /*
1353  * Get the syscall mask array from the kernel tracer.
1354  *
1355  * Return 0 on success else a negative value. In both case, syscall_mask should
1356  * be freed.
1357  */
1358 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1359 {
1360         assert(syscall_mask);
1361         assert(nr_bits);
1362
1363         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1364 }
1365
1366 /*
1367  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1368  * version number.
1369  *
1370  * Return 1 on success, 0 when feature is not supported, negative value in case
1371  * of errors.
1372  */
1373 int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd)
1374 {
1375         int ret = 0; // Not supported by default
1376         struct lttng_kernel_tracer_abi_version abi;
1377
1378         ret = kernctl_tracer_abi_version(tracer_fd, &abi);
1379         if (ret < 0) {
1380                 ERR("Failed to retrieve lttng-modules ABI version");
1381                 goto error;
1382         }
1383
1384         /*
1385          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1386          */
1387         if (abi.major >= 2 && abi.minor >= 3) {
1388                 /* Supported */
1389                 ret = 1;
1390         } else {
1391                 /* Not supported */
1392                 ret = 0;
1393         }
1394 error:
1395         return ret;
1396 }
1397
1398 /*
1399  * Rotate a kernel session.
1400  *
1401  * Return LTTNG_OK on success or else an LTTng error code.
1402  */
1403 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1404 {
1405         int ret;
1406         enum lttng_error_code status = LTTNG_OK;
1407         struct consumer_socket *socket;
1408         struct lttng_ht_iter iter;
1409         struct ltt_kernel_session *ksess = session->kernel_session;
1410
1411         assert(ksess);
1412         assert(ksess->consumer);
1413
1414         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1415                         session->name, session->id);
1416
1417         rcu_read_lock();
1418
1419         /*
1420          * Note that this loop will end after one iteration given that there is
1421          * only one kernel consumer.
1422          */
1423         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1424                         socket, node.node) {
1425                 struct ltt_kernel_channel *chan;
1426
1427                 /* For each channel, ask the consumer to rotate it. */
1428                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1429                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1430                                         chan->key, session->name);
1431                         ret = consumer_rotate_channel(socket, chan->key,
1432                                         ksess->uid, ksess->gid, ksess->consumer,
1433                                         ksess->consumer->subdir,
1434                                         /* is_metadata_channel */ false,
1435                                         session->current_archive_id);
1436                         if (ret < 0) {
1437                                 status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1438                                 goto error;
1439                         }
1440                 }
1441
1442                 /*
1443                  * Rotate the metadata channel.
1444                  */
1445                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1446                                 ksess->uid, ksess->gid, ksess->consumer,
1447                                 ksess->consumer->subdir,
1448                                 /* is_metadata_channel */ true,
1449                                 session->current_archive_id);
1450                 if (ret < 0) {
1451                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1452                         goto error;
1453                 }
1454         }
1455
1456 error:
1457         rcu_read_unlock();
1458         return status;
1459 }