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
  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 #include <sys/types.h>
  27
  28 #include <common/common.h>
  29 #include <common/trace-chunk.h>
  30 #include <common/kernel-ctl/kernel-ctl.h>
  31 #include <common/kernel-ctl/kernel-ioctl.h>
  32 #include <common/sessiond-comm/sessiond-comm.h>
  33
  34 #include "lttng-sessiond.h"
  35 #include "lttng-syscall.h"
  36 #include "consumer.h"
  37 #include "kernel.h"
  38 #include "kernel-consumer.h"
  39 #include "kern-modules.h"
  40 #include "utils.h"
  41 #include "rotate.h"
  42 #include "modprobe.h"
  43
  44 /*
  45  * Key used to reference a channel between the sessiond and the consumer. This
  46  * is only read and updated with the session_list lock held.
  47  */
  48 static uint64_t next_kernel_channel_key;
  49
  50 static const char *module_proc_lttng = "/proc/lttng";
  51
  52 static int kernel_tracer_fd = -1;
  53
  54 #include <lttng/userspace-probe.h>
  55 #include <lttng/userspace-probe-internal.h>
  56 /*
  57  * Add context on a kernel channel.
  58  *
  59  * Assumes the ownership of ctx.
  60  */
  61 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  62                 struct ltt_kernel_context *ctx)
  63 {
  64         int ret;
  65
  66         assert(chan);
  67         assert(ctx);
  68
  69         DBG("Adding context to channel %s", chan->channel->name);
  70         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  71         if (ret < 0) {
  72                 switch (-ret) {
  73                 case ENOSYS:
  74                         /* Exists but not available for this kernel */
  75                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  76                         goto error;
  77                 case EEXIST:
  78                         /* If EEXIST, we just ignore the error */
  79                         ret = 0;
  80                         goto end;
  81                 default:
  82                         PERROR("add context ioctl");
  83                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  84                         goto error;
  85                 }
  86         }
  87         ret = 0;
  88
  89 end:
  90         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  91         ctx->in_list = true;
  92         ctx = NULL;
  93 error:
  94         if (ctx) {
  95                 trace_kernel_destroy_context(ctx);
  96         }
  97         return ret;
  98 }
  99
 100 /*
 101  * Create a new kernel session, register it to the kernel tracer and add it to
 102  * the session daemon session.
 103  */
 104 int kernel_create_session(struct ltt_session *session)
 105 {
 106         int ret;
 107         struct ltt_kernel_session *lks;
 108
 109         assert(session);
 110
 111         /* Allocate data structure */
 112         lks = trace_kernel_create_session();
 113         if (lks == NULL) {
 114                 ret = -1;
 115                 goto error;
 116         }
 117
 118         /* Kernel tracer session creation */
 119         ret = kernctl_create_session(kernel_tracer_fd);
 120         if (ret < 0) {
 121                 PERROR("ioctl kernel create session");
 122                 goto error;
 123         }
 124
 125         lks->fd = ret;
 126         /* Prevent fd duplication after execlp() */
 127         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 128         if (ret < 0) {
 129                 PERROR("fcntl session fd");
 130         }
 131
 132         lks->id = session->id;
 133         lks->consumer_fds_sent = 0;
 134         session->kernel_session = lks;
 135
 136         DBG("Kernel session created (fd: %d)", lks->fd);
 137
 138         /*
 139          * This is necessary since the creation time is present in the session
 140          * name when it is generated.
 141          */
 142         if (session->has_auto_generated_name) {
 143                 ret = kernctl_session_set_name(lks->fd, DEFAULT_SESSION_NAME);
 144         } else {
 145                 ret = kernctl_session_set_name(lks->fd, session->name);
 146         }
 147         if (ret) {
 148                 WARN("Could not set kernel session name for session %" PRIu64 " name: %s",
 149                         session->id, session->name);
 150         }
 151
 152         ret = kernctl_session_set_creation_time(lks->fd, session->creation_time);
 153         if (ret) {
 154                 WARN("Could not set kernel session creation time for session %" PRIu64 " name: %s",
 155                         session->id, session->name);
 156         }
 157
 158         return 0;
 159
 160 error:
 161         if (lks) {
 162                 trace_kernel_destroy_session(lks);
 163                 trace_kernel_free_session(lks);
 164         }
 165         return ret;
 166 }
 167
 168 /*
 169  * Create a kernel channel, register it to the kernel tracer and add it to the
 170  * kernel session.
 171  */
 172 int kernel_create_channel(struct ltt_kernel_session *session,
 173                 struct lttng_channel *chan)
 174 {
 175         int ret;
 176         struct ltt_kernel_channel *lkc;
 177
 178         assert(session);
 179         assert(chan);
 180
 181         /* Allocate kernel channel */
 182         lkc = trace_kernel_create_channel(chan);
 183         if (lkc == NULL) {
 184                 goto error;
 185         }
 186
 187         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 188                         chan->name, lkc->channel->attr.overwrite,
 189                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 190                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 191                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 192
 193         /* Kernel tracer channel creation */
 194         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 195         if (ret < 0) {
 196                 PERROR("ioctl kernel create channel");
 197                 goto error;
 198         }
 199
 200         /* Setup the channel fd */
 201         lkc->fd = ret;
 202         /* Prevent fd duplication after execlp() */
 203         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 204         if (ret < 0) {
 205                 PERROR("fcntl session fd");
 206         }
 207
 208         /* Add channel to session */
 209         cds_list_add(&lkc->list, &session->channel_list.head);
 210         session->channel_count++;
 211         lkc->session = session;
 212         lkc->key = ++next_kernel_channel_key;
 213
 214         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 215                         lkc->channel->name, lkc->fd, lkc->key);
 216
 217         return 0;
 218
 219 error:
 220         if (lkc) {
 221                 free(lkc->channel);
 222                 free(lkc);
 223         }
 224         return -1;
 225 }
 226
 227 /*
 228  * Compute the offset of the instrumentation byte in the binary based on the
 229  * function probe location using the ELF lookup method.
 230  *
 231  * Returns 0 on success and set the offset out parameter to the offset of the
 232  * elf symbol
 233  * Returns -1 on error
 234  */
 235 static
 236 int extract_userspace_probe_offset_function_elf(
 237                 const struct lttng_userspace_probe_location *probe_location,
 238                 struct ltt_kernel_session *session, uint64_t *offset)
 239 {
 240         int fd;
 241         int ret = 0;
 242         const char *symbol = NULL;
 243         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 244         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 245
 246         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 247                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
 248
 249         lookup = lttng_userspace_probe_location_get_lookup_method(
 250                         probe_location);
 251         if (!lookup) {
 252                 ret = -1;
 253                 goto end;
 254         }
 255
 256         lookup_method_type =
 257                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 258
 259         assert(lookup_method_type ==
 260                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
 261
 262         symbol = lttng_userspace_probe_location_function_get_function_name(
 263                         probe_location);
 264         if (!symbol) {
 265                 ret = -1;
 266                 goto end;
 267         }
 268
 269         fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
 270         if (fd < 0) {
 271                 ret = -1;
 272                 goto end;
 273         }
 274
 275         ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
 276                         session->gid, offset);
 277         if (ret < 0) {
 278                 DBG("userspace probe offset calculation failed for "
 279                                 "function %s", symbol);
 280                 goto end;
 281         }
 282
 283         DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
 284 end:
 285         return ret;
 286 }
 287
 288 /*
 289  * Compute the offsets of the instrumentation bytes in the binary based on the
 290  * tracepoint probe location using the SDT lookup method. This function
 291  * allocates the offsets buffer, the caller must free it.
 292  *
 293  * Returns 0 on success and set the offset out parameter to the offsets of the
 294  * SDT tracepoint.
 295  * Returns -1 on error.
 296  */
 297 static
 298 int extract_userspace_probe_offset_tracepoint_sdt(
 299                 const struct lttng_userspace_probe_location *probe_location,
 300                 struct ltt_kernel_session *session, uint64_t **offsets,
 301                 uint32_t *offsets_count)
 302 {
 303         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 304         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 305         const char *probe_name = NULL, *provider_name = NULL;
 306         int ret = 0;
 307         int fd, i;
 308
 309         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 310                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
 311
 312         lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
 313         if (!lookup) {
 314                 ret = -1;
 315                 goto end;
 316         }
 317
 318         lookup_method_type =
 319                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 320
 321         assert(lookup_method_type ==
 322                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
 323
 324
 325         probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
 326                         probe_location);
 327         if (!probe_name) {
 328                 ret = -1;
 329                 goto end;
 330         }
 331
 332         provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
 333                         probe_location);
 334         if (!provider_name) {
 335                 ret = -1;
 336                 goto end;
 337         }
 338
 339         fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
 340         if (fd < 0) {
 341                 ret = -1;
 342                 goto end;
 343         }
 344
 345         ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
 346                         session->uid, session->gid, offsets, offsets_count);
 347         if (ret < 0) {
 348                 DBG("userspace probe offset calculation failed for sdt "
 349                                 "probe %s:%s", provider_name, probe_name);
 350                 goto end;
 351         }
 352
 353         if (*offsets_count == 0) {
 354                 DBG("no userspace probe offset found");
 355                 goto end;
 356         }
 357
 358         DBG("%u userspace probe SDT offsets found for %s:%s at:",
 359                         *offsets_count, provider_name, probe_name);
 360         for (i = 0; i < *offsets_count; i++) {
 361                 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
 362         }
 363 end:
 364         return ret;
 365 }
 366
 367 /*
 368  * Extract the offsets of the instrumentation point for the different lookup
 369  * methods.
 370  */
 371 static
 372 int userspace_probe_add_callsites(struct lttng_event *ev,
 373                         struct ltt_kernel_session *session, int fd)
 374 {
 375         const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
 376         enum lttng_userspace_probe_location_lookup_method_type type;
 377         const struct lttng_userspace_probe_location *location = NULL;
 378         int ret;
 379
 380         assert(ev);
 381         assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
 382
 383         location = lttng_event_get_userspace_probe_location(ev);
 384         if (!location) {
 385                 ret = -1;
 386                 goto end;
 387         }
 388         lookup_method =
 389                         lttng_userspace_probe_location_get_lookup_method(location);
 390         if (!lookup_method) {
 391                 ret = -1;
 392                 goto end;
 393         }
 394
 395         type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
 396         switch (type) {
 397         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
 398         {
 399                 struct lttng_kernel_event_callsite callsite;
 400                 uint64_t offset;
 401
 402                 ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
 403                 if (ret) {
 404                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 405                         goto end;
 406                 }
 407
 408                 callsite.u.uprobe.offset = offset;
 409                 ret = kernctl_add_callsite(fd, &callsite);
 410                 if (ret) {
 411                         WARN("Adding callsite to userspace probe "
 412                                         "event %s failed.", ev->name);
 413                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 414                         goto end;
 415                 }
 416                 break;
 417         }
 418         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
 419         {
 420                 int i;
 421                 uint64_t *offsets = NULL;
 422                 uint32_t offsets_count;
 423                 struct lttng_kernel_event_callsite callsite;
 424
 425                 /*
 426                  * This call allocates the offsets buffer. This buffer must be freed
 427                  * by the caller
 428                  */
 429                 ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
 430                                 &offsets, &offsets_count);
 431                 if (ret) {
 432                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 433                         goto end;
 434                 }
 435                 for (i = 0; i < offsets_count; i++) {
 436                         callsite.u.uprobe.offset = offsets[i];
 437                         ret = kernctl_add_callsite(fd, &callsite);
 438                         if (ret) {
 439                                 WARN("Adding callsite to userspace probe "
 440                                                 "event %s failed.", ev->name);
 441                                 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 442                                 free(offsets);
 443                                 goto end;
 444                         }
 445                 }
 446                 free(offsets);
 447                 break;
 448         }
 449         default:
 450                 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 451                 goto end;
 452         }
 453 end:
 454         return ret;
 455 }
 456
 457 /*
 458  * Create a kernel event, enable it to the kernel tracer and add it to the
 459  * channel event list of the kernel session.
 460  * We own filter_expression and filter.
 461  */
 462 int kernel_create_event(struct lttng_event *ev,
 463                 struct ltt_kernel_channel *channel,
 464                 char *filter_expression,
 465                 struct lttng_filter_bytecode *filter)
 466 {
 467         int err, fd;
 468         enum lttng_error_code ret;
 469         struct ltt_kernel_event *event;
 470
 471         assert(ev);
 472         assert(channel);
 473
 474         /* We pass ownership of filter_expression and filter */
 475         ret = trace_kernel_create_event(ev, filter_expression,
 476                         filter, &event);
 477         if (ret != LTTNG_OK) {
 478                 goto error;
 479         }
 480
 481         fd = kernctl_create_event(channel->fd, event->event);
 482         if (fd < 0) {
 483                 switch (-fd) {
 484                 case EEXIST:
 485                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 486                         break;
 487                 case ENOSYS:
 488                         WARN("Event type not implemented");
 489                         ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
 490                         break;
 491                 case ENOENT:
 492                         WARN("Event %s not found!", ev->name);
 493                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 494                         break;
 495                 default:
 496                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 497                         PERROR("create event ioctl");
 498                 }
 499                 goto free_event;
 500         }
 501
 502         event->type = ev->type;
 503         event->fd = fd;
 504         /* Prevent fd duplication after execlp() */
 505         err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 506         if (err < 0) {
 507                 PERROR("fcntl session fd");
 508         }
 509
 510         if (filter) {
 511                 err = kernctl_filter(event->fd, filter);
 512                 if (err < 0) {
 513                         switch (-err) {
 514                         case ENOMEM:
 515                                 ret = LTTNG_ERR_FILTER_NOMEM;
 516                                 break;
 517                         default:
 518                                 ret = LTTNG_ERR_FILTER_INVAL;
 519                                 break;
 520                         }
 521                         goto filter_error;
 522                 }
 523         }
 524
 525         if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
 526                 ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
 527                 if (ret) {
 528                         goto add_callsite_error;
 529                 }
 530         }
 531
 532         err = kernctl_enable(event->fd);
 533         if (err < 0) {
 534                 switch (-err) {
 535                 case EEXIST:
 536                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 537                         break;
 538                 default:
 539                         PERROR("enable kernel event");
 540                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 541                         break;
 542                 }
 543                 goto enable_error;
 544         }
 545
 546         /* Add event to event list */
 547         cds_list_add(&event->list, &channel->events_list.head);
 548         channel->event_count++;
 549
 550         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 551
 552         return 0;
 553
 554 add_callsite_error:
 555 enable_error:
 556 filter_error:
 557         {
 558                 int closeret;
 559
 560                 closeret = close(event->fd);
 561                 if (closeret) {
 562                         PERROR("close event fd");
 563                 }
 564         }
 565 free_event:
 566         free(event);
 567 error:
 568         return ret;
 569 }
 570
 571 /*
 572  * Disable a kernel channel.
 573  */
 574 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 575 {
 576         int ret;
 577
 578         assert(chan);
 579
 580         ret = kernctl_disable(chan->fd);
 581         if (ret < 0) {
 582                 PERROR("disable chan ioctl");
 583                 goto error;
 584         }
 585
 586         chan->enabled = 0;
 587         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 588                         chan->channel->name, chan->fd, chan->key);
 589
 590         return 0;
 591
 592 error:
 593         return ret;
 594 }
 595
 596 /*
 597  * Enable a kernel channel.
 598  */
 599 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 600 {
 601         int ret;
 602
 603         assert(chan);
 604
 605         ret = kernctl_enable(chan->fd);
 606         if (ret < 0 && ret != -EEXIST) {
 607                 PERROR("Enable kernel chan");
 608                 goto error;
 609         }
 610
 611         chan->enabled = 1;
 612         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 613                         chan->channel->name, chan->fd, chan->key);
 614
 615         return 0;
 616
 617 error:
 618         return ret;
 619 }
 620
 621 /*
 622  * Enable a kernel event.
 623  */
 624 int kernel_enable_event(struct ltt_kernel_event *event)
 625 {
 626         int ret;
 627
 628         assert(event);
 629
 630         ret = kernctl_enable(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("enable kernel event");
 638                         break;
 639                 }
 640                 goto error;
 641         }
 642
 643         event->enabled = 1;
 644         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 645
 646         return 0;
 647
 648 error:
 649         return ret;
 650 }
 651
 652 /*
 653  * Disable a kernel event.
 654  */
 655 int kernel_disable_event(struct ltt_kernel_event *event)
 656 {
 657         int ret;
 658
 659         assert(event);
 660
 661         ret = kernctl_disable(event->fd);
 662         if (ret < 0) {
 663                 PERROR("Failed to disable kernel event: name = '%s', fd = %d",
 664                                 event->event->name, event->fd);
 665                 goto error;
 666         }
 667
 668         event->enabled = 0;
 669         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 670
 671         return 0;
 672
 673 error:
 674         return ret;
 675 }
 676
 677
 678 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 679 {
 680         int ret;
 681
 682         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 683                         pid, session->id);
 684         ret = kernctl_track_pid(session->fd, pid);
 685         if (!ret) {
 686                 return LTTNG_OK;
 687         }
 688         switch (-ret) {
 689         case EINVAL:
 690                 return LTTNG_ERR_INVALID;
 691         case ENOMEM:
 692                 return LTTNG_ERR_NOMEM;
 693         case EEXIST:
 694                 return LTTNG_ERR_PID_TRACKED;
 695         default:
 696                 return LTTNG_ERR_UNK;
 697         }
 698 }
 699
 700 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 701 {
 702         int ret;
 703
 704         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 705                         pid, session->id);
 706         ret = kernctl_untrack_pid(session->fd, pid);
 707         if (!ret) {
 708                 return LTTNG_OK;
 709         }
 710         switch (-ret) {
 711         case EINVAL:
 712                 return LTTNG_ERR_INVALID;
 713         case ENOMEM:
 714                 return LTTNG_ERR_NOMEM;
 715         case ENOENT:
 716                 return LTTNG_ERR_PID_NOT_TRACKED;
 717         default:
 718                 return LTTNG_ERR_UNK;
 719         }
 720 }
 721
 722 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 723                 int **_pids)
 724 {
 725         int fd, ret;
 726         int pid;
 727         ssize_t nbmem, count = 0;
 728         FILE *fp;
 729         int *pids;
 730
 731         fd = kernctl_list_tracker_pids(session->fd);
 732         if (fd < 0) {
 733                 PERROR("kernel tracker pids list");
 734                 goto error;
 735         }
 736
 737         fp = fdopen(fd, "r");
 738         if (fp == NULL) {
 739                 PERROR("kernel tracker pids list fdopen");
 740                 goto error_fp;
 741         }
 742
 743         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 744         pids = zmalloc(sizeof(*pids) * nbmem);
 745         if (pids == NULL) {
 746                 PERROR("alloc list pids");
 747                 count = -ENOMEM;
 748                 goto end;
 749         }
 750
 751         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 752                 if (count >= nbmem) {
 753                         int *new_pids;
 754                         size_t new_nbmem;
 755
 756                         new_nbmem = nbmem << 1;
 757                         DBG("Reallocating pids list from %zu to %zu entries",
 758                                         nbmem, new_nbmem);
 759                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 760                         if (new_pids == NULL) {
 761                                 PERROR("realloc list events");
 762                                 free(pids);
 763                                 count = -ENOMEM;
 764                                 goto end;
 765                         }
 766                         /* Zero the new memory */
 767                         memset(new_pids + nbmem, 0,
 768                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 769                         nbmem = new_nbmem;
 770                         pids = new_pids;
 771                 }
 772                 pids[count++] = pid;
 773         }
 774
 775         *_pids = pids;
 776         DBG("Kernel list tracker pids done (%zd pids)", count);
 777 end:
 778         ret = fclose(fp);       /* closes both fp and fd */
 779         if (ret) {
 780                 PERROR("fclose");
 781         }
 782         return count;
 783
 784 error_fp:
 785         ret = close(fd);
 786         if (ret) {
 787                 PERROR("close");
 788         }
 789 error:
 790         return -1;
 791 }
 792
 793 /*
 794  * Create kernel metadata, open from the kernel tracer and add it to the
 795  * kernel session.
 796  */
 797 int kernel_open_metadata(struct ltt_kernel_session *session)
 798 {
 799         int ret;
 800         struct ltt_kernel_metadata *lkm = NULL;
 801
 802         assert(session);
 803
 804         /* Allocate kernel metadata */
 805         lkm = trace_kernel_create_metadata();
 806         if (lkm == NULL) {
 807                 goto error;
 808         }
 809
 810         /* Kernel tracer metadata creation */
 811         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 812         if (ret < 0) {
 813                 goto error_open;
 814         }
 815
 816         lkm->fd = ret;
 817         lkm->key = ++next_kernel_channel_key;
 818         /* Prevent fd duplication after execlp() */
 819         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 820         if (ret < 0) {
 821                 PERROR("fcntl session fd");
 822         }
 823
 824         session->metadata = lkm;
 825
 826         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 827
 828         return 0;
 829
 830 error_open:
 831         trace_kernel_destroy_metadata(lkm);
 832 error:
 833         return -1;
 834 }
 835
 836 /*
 837  * Start tracing session.
 838  */
 839 int kernel_start_session(struct ltt_kernel_session *session)
 840 {
 841         int ret;
 842
 843         assert(session);
 844
 845         ret = kernctl_start_session(session->fd);
 846         if (ret < 0) {
 847                 PERROR("ioctl start session");
 848                 goto error;
 849         }
 850
 851         DBG("Kernel session started");
 852
 853         return 0;
 854
 855 error:
 856         return ret;
 857 }
 858
 859 /*
 860  * Make a kernel wait to make sure in-flight probe have completed.
 861  */
 862 void kernel_wait_quiescent(void)
 863 {
 864         int ret;
 865         int fd = kernel_tracer_fd;
 866
 867         DBG("Kernel quiescent wait on %d", fd);
 868
 869         ret = kernctl_wait_quiescent(fd);
 870         if (ret < 0) {
 871                 PERROR("wait quiescent ioctl");
 872                 ERR("Kernel quiescent wait failed");
 873         }
 874 }
 875
 876 /*
 877  *  Force flush buffer of metadata.
 878  */
 879 int kernel_metadata_flush_buffer(int fd)
 880 {
 881         int ret;
 882
 883         DBG("Kernel flushing metadata buffer on fd %d", fd);
 884
 885         ret = kernctl_buffer_flush(fd);
 886         if (ret < 0) {
 887                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 888         }
 889
 890         return 0;
 891 }
 892
 893 /*
 894  * Force flush buffer for channel.
 895  */
 896 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 897 {
 898         int ret;
 899         struct ltt_kernel_stream *stream;
 900
 901         assert(channel);
 902
 903         DBG("Flush buffer for channel %s", channel->channel->name);
 904
 905         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 906                 DBG("Flushing channel stream %d", stream->fd);
 907                 ret = kernctl_buffer_flush(stream->fd);
 908                 if (ret < 0) {
 909                         PERROR("ioctl");
 910                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 911                                         stream->fd, ret);
 912                 }
 913         }
 914
 915         return 0;
 916 }
 917
 918 /*
 919  * Stop tracing session.
 920  */
 921 int kernel_stop_session(struct ltt_kernel_session *session)
 922 {
 923         int ret;
 924
 925         assert(session);
 926
 927         ret = kernctl_stop_session(session->fd);
 928         if (ret < 0) {
 929                 goto error;
 930         }
 931
 932         DBG("Kernel session stopped");
 933
 934         return 0;
 935
 936 error:
 937         return ret;
 938 }
 939
 940 /*
 941  * Open stream of channel, register it to the kernel tracer and add it
 942  * to the stream list of the channel.
 943  *
 944  * Note: given that the streams may appear in random order wrt CPU
 945  * number (e.g. cpu hotplug), the index value of the stream number in
 946  * the stream name is not necessarily linked to the CPU number.
 947  *
 948  * Return the number of created stream. Else, a negative value.
 949  */
 950 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 951 {
 952         int ret;
 953         struct ltt_kernel_stream *lks;
 954
 955         assert(channel);
 956
 957         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 958                 lks = trace_kernel_create_stream(channel->channel->name,
 959                                 channel->stream_count);
 960                 if (lks == NULL) {
 961                         ret = close(ret);
 962                         if (ret) {
 963                                 PERROR("close");
 964                         }
 965                         goto error;
 966                 }
 967
 968                 lks->fd = ret;
 969                 /* Prevent fd duplication after execlp() */
 970                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 971                 if (ret < 0) {
 972                         PERROR("fcntl session fd");
 973                 }
 974
 975                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 976                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 977
 978                 /* Add stream to channel stream list */
 979                 cds_list_add(&lks->list, &channel->stream_list.head);
 980                 channel->stream_count++;
 981
 982                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 983                                 lks->state);
 984         }
 985
 986         return channel->stream_count;
 987
 988 error:
 989         return -1;
 990 }
 991
 992 /*
 993  * Open the metadata stream and set it to the kernel session.
 994  */
 995 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 996 {
 997         int ret;
 998
 999         assert(session);
1000
1001         ret = kernctl_create_stream(session->metadata->fd);
1002         if (ret < 0) {
1003                 PERROR("kernel create metadata stream");
1004                 goto error;
1005         }
1006
1007         DBG("Kernel metadata stream created (fd: %d)", ret);
1008         session->metadata_stream_fd = ret;
1009         /* Prevent fd duplication after execlp() */
1010         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
1011         if (ret < 0) {
1012                 PERROR("fcntl session fd");
1013         }
1014
1015         return 0;
1016
1017 error:
1018         return -1;
1019 }
1020
1021 /*
1022  * Get the event list from the kernel tracer and return the number of elements.
1023  */
1024 ssize_t kernel_list_events(struct lttng_event **events)
1025 {
1026         int fd, ret;
1027         char *event;
1028         size_t nbmem, count = 0;
1029         FILE *fp;
1030         struct lttng_event *elist;
1031
1032         assert(events);
1033
1034         fd = kernctl_tracepoint_list(kernel_tracer_fd);
1035         if (fd < 0) {
1036                 PERROR("kernel tracepoint list");
1037                 goto error;
1038         }
1039
1040         fp = fdopen(fd, "r");
1041         if (fp == NULL) {
1042                 PERROR("kernel tracepoint list fdopen");
1043                 goto error_fp;
1044         }
1045
1046         /*
1047          * Init memory size counter
1048          * See kernel-ctl.h for explanation of this value
1049          */
1050         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1051         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1052         if (elist == NULL) {
1053                 PERROR("alloc list events");
1054                 count = -ENOMEM;
1055                 goto end;
1056         }
1057
1058         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1059                 if (count >= nbmem) {
1060                         struct lttng_event *new_elist;
1061                         size_t new_nbmem;
1062
1063                         new_nbmem = nbmem << 1;
1064                         DBG("Reallocating event list from %zu to %zu bytes",
1065                                         nbmem, new_nbmem);
1066                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1067                         if (new_elist == NULL) {
1068                                 PERROR("realloc list events");
1069                                 free(event);
1070                                 free(elist);
1071                                 count = -ENOMEM;
1072                                 goto end;
1073                         }
1074                         /* Zero the new memory */
1075                         memset(new_elist + nbmem, 0,
1076                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1077                         nbmem = new_nbmem;
1078                         elist = new_elist;
1079                 }
1080                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1081                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1082                 elist[count].enabled = -1;
1083                 count++;
1084                 free(event);
1085         }
1086
1087         *events = elist;
1088         DBG("Kernel list events done (%zu events)", count);
1089 end:
1090         ret = fclose(fp);       /* closes both fp and fd */
1091         if (ret) {
1092                 PERROR("fclose");
1093         }
1094         return count;
1095
1096 error_fp:
1097         ret = close(fd);
1098         if (ret) {
1099                 PERROR("close");
1100         }
1101 error:
1102         return -1;
1103 }
1104
1105 /*
1106  * Get kernel version and validate it.
1107  */
1108 int kernel_validate_version(struct lttng_kernel_tracer_version *version,
1109                 struct lttng_kernel_tracer_abi_version *abi_version)
1110 {
1111         int ret;
1112
1113         ret = kernctl_tracer_version(kernel_tracer_fd, version);
1114         if (ret < 0) {
1115                 ERR("Failed to retrieve the lttng-modules version");
1116                 goto error;
1117         }
1118
1119         /* Validate version */
1120         if (version->major != VERSION_MAJOR) {
1121                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1122                         version->major, VERSION_MAJOR);
1123                 goto error_version;
1124         }
1125         ret = kernctl_tracer_abi_version(kernel_tracer_fd, abi_version);
1126         if (ret < 0) {
1127                 ERR("Failed to retrieve lttng-modules ABI version");
1128                 goto error;
1129         }
1130         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1131                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1132                         abi_version->major, abi_version->minor,
1133                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1134                 goto error;
1135         }
1136         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1137                         version->major, version->minor,
1138                         abi_version->major, abi_version->minor);
1139         return 0;
1140
1141 error_version:
1142         ret = -1;
1143
1144 error:
1145         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1146         return ret;
1147 }
1148
1149 /*
1150  * Kernel work-arounds called at the start of sessiond main().
1151  */
1152 int init_kernel_workarounds(void)
1153 {
1154         int ret;
1155         FILE *fp;
1156
1157         /*
1158          * boot_id needs to be read once before being used concurrently
1159          * to deal with a Linux kernel race. A fix is proposed for
1160          * upstream, but the work-around is needed for older kernels.
1161          */
1162         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1163         if (!fp) {
1164                 goto end_boot_id;
1165         }
1166         while (!feof(fp)) {
1167                 char buf[37] = "";
1168
1169                 ret = fread(buf, 1, sizeof(buf), fp);
1170                 if (ret < 0) {
1171                         /* Ignore error, we don't really care */
1172                 }
1173         }
1174         ret = fclose(fp);
1175         if (ret) {
1176                 PERROR("fclose");
1177         }
1178 end_boot_id:
1179         return 0;
1180 }
1181
1182 /*
1183  * Teardown of a kernel session, keeping data required by destroy notifiers.
1184  */
1185 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1186 {
1187         struct lttng_trace_chunk *trace_chunk;
1188
1189         if (ksess == NULL) {
1190                 DBG3("No kernel session when tearing down session");
1191                 return;
1192         }
1193
1194         DBG("Tearing down kernel session");
1195         trace_chunk = ksess->current_trace_chunk;
1196
1197         /*
1198          * Destroy channels on the consumer if at least one FD has been sent and we
1199          * are in no output mode because the streams are in *no* monitor mode so we
1200          * have to send a command to clean them up or else they leaked.
1201          */
1202         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1203                 int ret;
1204                 struct consumer_socket *socket;
1205                 struct lttng_ht_iter iter;
1206
1207                 /* For each consumer socket. */
1208                 rcu_read_lock();
1209                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1210                                 socket, node.node) {
1211                         struct ltt_kernel_channel *chan;
1212
1213                         /* For each channel, ask the consumer to destroy it. */
1214                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1215                                 ret = kernel_consumer_destroy_channel(socket, chan);
1216                                 if (ret < 0) {
1217                                         /* Consumer is probably dead. Use next socket. */
1218                                         continue;
1219                                 }
1220                         }
1221                 }
1222                 rcu_read_unlock();
1223         }
1224
1225         /* Close any relayd session */
1226         consumer_output_send_destroy_relayd(ksess->consumer);
1227
1228         trace_kernel_destroy_session(ksess);
1229         lttng_trace_chunk_put(trace_chunk);
1230 }
1231
1232 /* Teardown of data required by destroy notifiers. */
1233 void kernel_free_session(struct ltt_kernel_session *ksess)
1234 {
1235         if (ksess == NULL) {
1236                 return;
1237         }
1238         trace_kernel_free_session(ksess);
1239 }
1240
1241 /*
1242  * Destroy a kernel channel object. It does not do anything on the tracer side.
1243  */
1244 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1245 {
1246         struct ltt_kernel_session *ksess = NULL;
1247
1248         assert(kchan);
1249         assert(kchan->channel);
1250
1251         DBG3("Kernel destroy channel %s", kchan->channel->name);
1252
1253         /* Update channel count of associated session. */
1254         if (kchan->session) {
1255                 /* Keep pointer reference so we can update it after the destroy. */
1256                 ksess = kchan->session;
1257         }
1258
1259         trace_kernel_destroy_channel(kchan);
1260
1261         /*
1262          * At this point the kernel channel is not visible anymore. This is safe
1263          * since in order to work on a visible kernel session, the tracing session
1264          * lock (ltt_session.lock) MUST be acquired.
1265          */
1266         if (ksess) {
1267                 ksess->channel_count--;
1268         }
1269 }
1270
1271 /*
1272  * Take a snapshot for a given kernel session.
1273  *
1274  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1275  */
1276 enum lttng_error_code kernel_snapshot_record(
1277                 struct ltt_kernel_session *ksess,
1278                 const struct consumer_output *output, int wait,
1279                 uint64_t nb_packets_per_stream)
1280 {
1281         int err, ret, saved_metadata_fd;
1282         enum lttng_error_code status = LTTNG_OK;
1283         struct consumer_socket *socket;
1284         struct lttng_ht_iter iter;
1285         struct ltt_kernel_metadata *saved_metadata;
1286         char *trace_path = NULL;
1287
1288         assert(ksess);
1289         assert(ksess->consumer);
1290         assert(output);
1291
1292         DBG("Kernel snapshot record started");
1293
1294         /* Save current metadata since the following calls will change it. */
1295         saved_metadata = ksess->metadata;
1296         saved_metadata_fd = ksess->metadata_stream_fd;
1297
1298         rcu_read_lock();
1299
1300         ret = kernel_open_metadata(ksess);
1301         if (ret < 0) {
1302                 status = LTTNG_ERR_KERN_META_FAIL;
1303                 goto error;
1304         }
1305
1306         ret = kernel_open_metadata_stream(ksess);
1307         if (ret < 0) {
1308                 status = LTTNG_ERR_KERN_META_FAIL;
1309                 goto error_open_stream;
1310         }
1311
1312         trace_path = setup_channel_trace_path(ksess->consumer,
1313                         DEFAULT_KERNEL_TRACE_DIR);
1314         if (!trace_path) {
1315                 status = LTTNG_ERR_INVALID;
1316                 goto error;
1317         }
1318         /* Send metadata to consumer and snapshot everything. */
1319         cds_lfht_for_each_entry(output->socks->ht, &iter.iter,
1320                         socket, node.node) {
1321                 struct ltt_kernel_channel *chan;
1322
1323                 pthread_mutex_lock(socket->lock);
1324                 /* This stream must not be monitored by the consumer. */
1325                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1326                 pthread_mutex_unlock(socket->lock);
1327                 if (ret < 0) {
1328                         status = LTTNG_ERR_KERN_META_FAIL;
1329                         goto error_consumer;
1330                 }
1331
1332                 /* For each channel, ask the consumer to snapshot it. */
1333                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1334                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1335                                         ksess->uid, ksess->gid,
1336                                         trace_path, wait,
1337                                         nb_packets_per_stream);
1338                         if (status != LTTNG_OK) {
1339                                 (void) kernel_consumer_destroy_metadata(socket,
1340                                                 ksess->metadata);
1341                                 goto error_consumer;
1342                         }
1343                 }
1344
1345                 /* Snapshot metadata, */
1346                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1347                                 1, ksess->uid, ksess->gid, trace_path, wait, 0);
1348                 if (status != LTTNG_OK) {
1349                         goto error_consumer;
1350                 }
1351
1352                 /*
1353                  * The metadata snapshot is done, ask the consumer to destroy it since
1354                  * it's not monitored on the consumer side.
1355                  */
1356                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1357         }
1358
1359 error_consumer:
1360         /* Close newly opened metadata stream. It's now on the consumer side. */
1361         err = close(ksess->metadata_stream_fd);
1362         if (err < 0) {
1363                 PERROR("close snapshot kernel");
1364         }
1365
1366 error_open_stream:
1367         trace_kernel_destroy_metadata(ksess->metadata);
1368 error:
1369         /* Restore metadata state.*/
1370         ksess->metadata = saved_metadata;
1371         ksess->metadata_stream_fd = saved_metadata_fd;
1372         rcu_read_unlock();
1373         free(trace_path);
1374         return status;
1375 }
1376
1377 /*
1378  * Get the syscall mask array from the kernel tracer.
1379  *
1380  * Return 0 on success else a negative value. In both case, syscall_mask should
1381  * be freed.
1382  */
1383 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1384 {
1385         assert(syscall_mask);
1386         assert(nr_bits);
1387
1388         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1389 }
1390
1391 /*
1392  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1393  * version number.
1394  *
1395  * Return 1 on success, 0 when feature is not supported, negative value in case
1396  * of errors.
1397  */
1398 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1399 {
1400         int ret = 0; // Not supported by default
1401         struct lttng_kernel_tracer_abi_version abi;
1402
1403         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1404         if (ret < 0) {
1405                 ERR("Failed to retrieve lttng-modules ABI version");
1406                 goto error;
1407         }
1408
1409         /*
1410          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1411          */
1412         if (abi.major >= 2 && abi.minor >= 3) {
1413                 /* Supported */
1414                 ret = 1;
1415         } else {
1416                 /* Not supported */
1417                 ret = 0;
1418         }
1419 error:
1420         return ret;
1421 }
1422
1423 /*
1424  * Check for the support of the packet sequence number via abi version number.
1425  *
1426  * Return 1 on success, 0 when feature is not supported, negative value in case
1427  * of errors.
1428  */
1429 int kernel_supports_ring_buffer_packet_sequence_number(void)
1430 {
1431         int ret = 0; // Not supported by default
1432         struct lttng_kernel_tracer_abi_version abi;
1433
1434         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1435         if (ret < 0) {
1436                 ERR("Failed to retrieve lttng-modules ABI version");
1437                 goto error;
1438         }
1439
1440         /*
1441          * Packet sequence number was introduced in LTTng 2.8,
1442          * lttng-modules ABI 2.1.
1443          */
1444         if (abi.major >= 2 && abi.minor >= 1) {
1445                 /* Supported */
1446                 ret = 1;
1447         } else {
1448                 /* Not supported */
1449                 ret = 0;
1450         }
1451 error:
1452         return ret;
1453 }
1454
1455 /*
1456  * Rotate a kernel session.
1457  *
1458  * Return LTTNG_OK on success or else an LTTng error code.
1459  */
1460 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1461 {
1462         int ret;
1463         enum lttng_error_code status = LTTNG_OK;
1464         struct consumer_socket *socket;
1465         struct lttng_ht_iter iter;
1466         struct ltt_kernel_session *ksess = session->kernel_session;
1467
1468         assert(ksess);
1469         assert(ksess->consumer);
1470
1471         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1472                         session->name, session->id);
1473
1474         rcu_read_lock();
1475
1476         /*
1477          * Note that this loop will end after one iteration given that there is
1478          * only one kernel consumer.
1479          */
1480         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1481                         socket, node.node) {
1482                 struct ltt_kernel_channel *chan;
1483
1484                 /* For each channel, ask the consumer to rotate it. */
1485                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1486                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1487                                         chan->key, session->name);
1488                         ret = consumer_rotate_channel(socket, chan->key,
1489                                         ksess->uid, ksess->gid, ksess->consumer,
1490                                         /* is_metadata_channel */ false);
1491                         if (ret < 0) {
1492                                 status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1493                                 goto error;
1494                         }
1495                 }
1496
1497                 /*
1498                  * Rotate the metadata channel.
1499                  */
1500                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1501                                 ksess->uid, ksess->gid, ksess->consumer,
1502                                 /* is_metadata_channel */ true);
1503                 if (ret < 0) {
1504                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1505                         goto error;
1506                 }
1507         }
1508
1509 error:
1510         rcu_read_unlock();
1511         return status;
1512 }
1513
1514 enum lttng_error_code kernel_create_channel_subdirectories(
1515                 const struct ltt_kernel_session *ksess)
1516 {
1517         enum lttng_error_code ret = LTTNG_OK;
1518         enum lttng_trace_chunk_status chunk_status;
1519
1520         rcu_read_lock();
1521         assert(ksess->current_trace_chunk);
1522
1523         /*
1524          * Create the index subdirectory which will take care
1525          * of implicitly creating the channel's path.
1526          */
1527         chunk_status = lttng_trace_chunk_create_subdirectory(
1528                         ksess->current_trace_chunk,
1529                         DEFAULT_KERNEL_TRACE_DIR "/" DEFAULT_INDEX_DIR);
1530         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1531                 ret = LTTNG_ERR_CREATE_DIR_FAIL;
1532                 goto error;
1533         }
1534 error:
1535         rcu_read_unlock();
1536         return ret;
1537 }
1538
1539 /*
1540  * Setup necessary data for kernel tracer action.
1541  */
1542 LTTNG_HIDDEN
1543 int init_kernel_tracer(void)
1544 {
1545         int ret;
1546         bool is_root = !getuid();
1547
1548         /* Modprobe lttng kernel modules */
1549         ret = modprobe_lttng_control();
1550         if (ret < 0) {
1551                 goto error;
1552         }
1553
1554         /* Open debugfs lttng */
1555         kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1556         if (kernel_tracer_fd < 0) {
1557                 DBG("Failed to open %s", module_proc_lttng);
1558                 goto error_open;
1559         }
1560
1561         /* Validate kernel version */
1562         ret = kernel_validate_version(&kernel_tracer_version,
1563                         &kernel_tracer_abi_version);
1564         if (ret < 0) {
1565                 goto error_version;
1566         }
1567
1568         ret = modprobe_lttng_data();
1569         if (ret < 0) {
1570                 goto error_modules;
1571         }
1572
1573         ret = kernel_supports_ring_buffer_snapshot_sample_positions();
1574         if (ret < 0) {
1575                 goto error_modules;
1576         }
1577
1578         if (ret < 1) {
1579                 WARN("Kernel tracer does not support buffer monitoring. "
1580                         "The monitoring timer of channels in the kernel domain "
1581                         "will be set to 0 (disabled).");
1582         }
1583
1584         DBG("Kernel tracer fd %d", kernel_tracer_fd);
1585
1586         ret = syscall_init_table(kernel_tracer_fd);
1587         if (ret < 0) {
1588                 ERR("Unable to populate syscall table. Syscall tracing won't "
1589                         "work for this session daemon.");
1590         }
1591         return 0;
1592
1593 error_version:
1594         modprobe_remove_lttng_control();
1595         ret = close(kernel_tracer_fd);
1596         if (ret) {
1597                 PERROR("close");
1598         }
1599         kernel_tracer_fd = -1;
1600         return LTTNG_ERR_KERN_VERSION;
1601
1602 error_modules:
1603         ret = close(kernel_tracer_fd);
1604         if (ret) {
1605                 PERROR("close");
1606         }
1607
1608 error_open:
1609         modprobe_remove_lttng_control();
1610
1611 error:
1612         WARN("No kernel tracer available");
1613         kernel_tracer_fd = -1;
1614         if (!is_root) {
1615                 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1616         } else {
1617                 return LTTNG_ERR_KERN_NA;
1618         }
1619 }
1620
1621 LTTNG_HIDDEN
1622 void cleanup_kernel_tracer(void)
1623 {
1624         int ret;
1625
1626         DBG2("Closing kernel fd");
1627         if (kernel_tracer_fd >= 0) {
1628                 ret = close(kernel_tracer_fd);
1629                 if (ret) {
1630                         PERROR("close");
1631                 }
1632                 kernel_tracer_fd = -1;
1633         }
1634         DBG("Unloading kernel modules");
1635         modprobe_remove_lttng_all();
1636         free(syscall_table);
1637 }
1638
1639 LTTNG_HIDDEN
1640 bool kernel_tracer_is_initialized(void)
1641 {
1642         return kernel_tracer_fd >= 0;
1643 }