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 #include <sys/types.h>
  26
  27 #include <common/common.h>
  28 #include <common/trace-chunk.h>
  29 #include <common/kernel-ctl/kernel-ctl.h>
  30 #include <common/kernel-ctl/kernel-ioctl.h>
  31 #include <common/sessiond-comm/sessiond-comm.h>
  32
  33 #include "lttng-sessiond.h"
  34 #include "lttng-syscall.h"
  35 #include "consumer.h"
  36 #include "kernel.h"
  37 #include "kernel-consumer.h"
  38 #include "kern-modules.h"
  39 #include "utils.h"
  40 #include "rotate.h"
  41 #include "modprobe.h"
  42
  43 /*
  44  * Key used to reference a channel between the sessiond and the consumer. This
  45  * is only read and updated with the session_list lock held.
  46  */
  47 static uint64_t next_kernel_channel_key;
  48
  49 static const char *module_proc_lttng = "/proc/lttng";
  50
  51 static int kernel_tracer_fd = -1;
  52
  53 #include <lttng/userspace-probe.h>
  54 #include <lttng/userspace-probe-internal.h>
  55 /*
  56  * Add context on a kernel channel.
  57  *
  58  * Assumes the ownership of ctx.
  59  */
  60 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  61                 struct ltt_kernel_context *ctx)
  62 {
  63         int ret;
  64
  65         assert(chan);
  66         assert(ctx);
  67
  68         DBG("Adding context to channel %s", chan->channel->name);
  69         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  70         if (ret < 0) {
  71                 switch (-ret) {
  72                 case ENOSYS:
  73                         /* Exists but not available for this kernel */
  74                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  75                         goto error;
  76                 case EEXIST:
  77                         /* If EEXIST, we just ignore the error */
  78                         ret = 0;
  79                         goto end;
  80                 default:
  81                         PERROR("add context ioctl");
  82                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  83                         goto error;
  84                 }
  85         }
  86         ret = 0;
  87
  88 end:
  89         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  90         ctx->in_list = true;
  91         ctx = NULL;
  92 error:
  93         if (ctx) {
  94                 trace_kernel_destroy_context(ctx);
  95         }
  96         return ret;
  97 }
  98
  99 /*
 100  * Create a new kernel session, register it to the kernel tracer and add it to
 101  * the session daemon session.
 102  */
 103 int kernel_create_session(struct ltt_session *session)
 104 {
 105         int ret;
 106         struct ltt_kernel_session *lks;
 107
 108         assert(session);
 109
 110         /* Allocate data structure */
 111         lks = trace_kernel_create_session();
 112         if (lks == NULL) {
 113                 ret = -1;
 114                 goto error;
 115         }
 116
 117         /* Kernel tracer session creation */
 118         ret = kernctl_create_session(kernel_tracer_fd);
 119         if (ret < 0) {
 120                 PERROR("ioctl kernel create session");
 121                 goto error;
 122         }
 123
 124         lks->fd = ret;
 125         /* Prevent fd duplication after execlp() */
 126         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 127         if (ret < 0) {
 128                 PERROR("fcntl session fd");
 129         }
 130
 131         lks->id = session->id;
 132         lks->consumer_fds_sent = 0;
 133         session->kernel_session = lks;
 134
 135         DBG("Kernel session created (fd: %d)", lks->fd);
 136
 137         /*
 138          * This is necessary since the creation time is present in the session
 139          * name when it is generated.
 140          */
 141         if (session->has_auto_generated_name) {
 142                 ret = kernctl_session_set_name(lks->fd, DEFAULT_SESSION_NAME);
 143         } else {
 144                 ret = kernctl_session_set_name(lks->fd, session->name);
 145         }
 146         if (ret) {
 147                 WARN("Could not set kernel session name for session %" PRIu64 " name: %s",
 148                         session->id, session->name);
 149         }
 150
 151         ret = kernctl_session_set_creation_time(lks->fd, session->creation_time);
 152         if (ret) {
 153                 WARN("Could not set kernel session creation time for session %" PRIu64 " name: %s",
 154                         session->id, session->name);
 155         }
 156
 157         return 0;
 158
 159 error:
 160         if (lks) {
 161                 trace_kernel_destroy_session(lks);
 162                 trace_kernel_free_session(lks);
 163         }
 164         return ret;
 165 }
 166
 167 /*
 168  * Create a kernel channel, register it to the kernel tracer and add it to the
 169  * kernel session.
 170  */
 171 int kernel_create_channel(struct ltt_kernel_session *session,
 172                 struct lttng_channel *chan)
 173 {
 174         int ret;
 175         struct ltt_kernel_channel *lkc;
 176
 177         assert(session);
 178         assert(chan);
 179
 180         /* Allocate kernel channel */
 181         lkc = trace_kernel_create_channel(chan);
 182         if (lkc == NULL) {
 183                 goto error;
 184         }
 185
 186         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 187                         chan->name, lkc->channel->attr.overwrite,
 188                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 189                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 190                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 191
 192         /* Kernel tracer channel creation */
 193         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 194         if (ret < 0) {
 195                 PERROR("ioctl kernel create channel");
 196                 goto error;
 197         }
 198
 199         /* Setup the channel fd */
 200         lkc->fd = ret;
 201         /* Prevent fd duplication after execlp() */
 202         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 203         if (ret < 0) {
 204                 PERROR("fcntl session fd");
 205         }
 206
 207         /* Add channel to session */
 208         cds_list_add(&lkc->list, &session->channel_list.head);
 209         session->channel_count++;
 210         lkc->session = session;
 211         lkc->key = ++next_kernel_channel_key;
 212
 213         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 214                         lkc->channel->name, lkc->fd, lkc->key);
 215
 216         return 0;
 217
 218 error:
 219         if (lkc) {
 220                 free(lkc->channel);
 221                 free(lkc);
 222         }
 223         return -1;
 224 }
 225
 226 /*
 227  * Compute the offset of the instrumentation byte in the binary based on the
 228  * function probe location using the ELF lookup method.
 229  *
 230  * Returns 0 on success and set the offset out parameter to the offset of the
 231  * elf symbol
 232  * Returns -1 on error
 233  */
 234 static
 235 int extract_userspace_probe_offset_function_elf(
 236                 const struct lttng_userspace_probe_location *probe_location,
 237                 struct ltt_kernel_session *session, uint64_t *offset)
 238 {
 239         int fd;
 240         int ret = 0;
 241         const char *symbol = NULL;
 242         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 243         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 244
 245         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 246                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
 247
 248         lookup = lttng_userspace_probe_location_get_lookup_method(
 249                         probe_location);
 250         if (!lookup) {
 251                 ret = -1;
 252                 goto end;
 253         }
 254
 255         lookup_method_type =
 256                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 257
 258         assert(lookup_method_type ==
 259                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
 260
 261         symbol = lttng_userspace_probe_location_function_get_function_name(
 262                         probe_location);
 263         if (!symbol) {
 264                 ret = -1;
 265                 goto end;
 266         }
 267
 268         fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
 269         if (fd < 0) {
 270                 ret = -1;
 271                 goto end;
 272         }
 273
 274         ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
 275                         session->gid, offset);
 276         if (ret < 0) {
 277                 DBG("userspace probe offset calculation failed for "
 278                                 "function %s", symbol);
 279                 goto end;
 280         }
 281
 282         DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
 283 end:
 284         return ret;
 285 }
 286
 287 /*
 288  * Compute the offsets of the instrumentation bytes in the binary based on the
 289  * tracepoint probe location using the SDT lookup method. This function
 290  * allocates the offsets buffer, the caller must free it.
 291  *
 292  * Returns 0 on success and set the offset out parameter to the offsets of the
 293  * SDT tracepoint.
 294  * Returns -1 on error.
 295  */
 296 static
 297 int extract_userspace_probe_offset_tracepoint_sdt(
 298                 const struct lttng_userspace_probe_location *probe_location,
 299                 struct ltt_kernel_session *session, uint64_t **offsets,
 300                 uint32_t *offsets_count)
 301 {
 302         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 303         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 304         const char *probe_name = NULL, *provider_name = NULL;
 305         int ret = 0;
 306         int fd, i;
 307
 308         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 309                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
 310
 311         lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
 312         if (!lookup) {
 313                 ret = -1;
 314                 goto end;
 315         }
 316
 317         lookup_method_type =
 318                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 319
 320         assert(lookup_method_type ==
 321                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
 322
 323
 324         probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
 325                         probe_location);
 326         if (!probe_name) {
 327                 ret = -1;
 328                 goto end;
 329         }
 330
 331         provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
 332                         probe_location);
 333         if (!provider_name) {
 334                 ret = -1;
 335                 goto end;
 336         }
 337
 338         fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
 339         if (fd < 0) {
 340                 ret = -1;
 341                 goto end;
 342         }
 343
 344         ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
 345                         session->uid, session->gid, offsets, offsets_count);
 346         if (ret < 0) {
 347                 DBG("userspace probe offset calculation failed for sdt "
 348                                 "probe %s:%s", provider_name, probe_name);
 349                 goto end;
 350         }
 351
 352         if (*offsets_count == 0) {
 353                 DBG("no userspace probe offset found");
 354                 goto end;
 355         }
 356
 357         DBG("%u userspace probe SDT offsets found for %s:%s at:",
 358                         *offsets_count, provider_name, probe_name);
 359         for (i = 0; i < *offsets_count; i++) {
 360                 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
 361         }
 362 end:
 363         return ret;
 364 }
 365
 366 /*
 367  * Extract the offsets of the instrumentation point for the different lookup
 368  * methods.
 369  */
 370 static
 371 int userspace_probe_add_callsites(struct lttng_event *ev,
 372                         struct ltt_kernel_session *session, int fd)
 373 {
 374         const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
 375         enum lttng_userspace_probe_location_lookup_method_type type;
 376         const struct lttng_userspace_probe_location *location = NULL;
 377         int ret;
 378
 379         assert(ev);
 380         assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
 381
 382         location = lttng_event_get_userspace_probe_location(ev);
 383         if (!location) {
 384                 ret = -1;
 385                 goto end;
 386         }
 387         lookup_method =
 388                         lttng_userspace_probe_location_get_lookup_method(location);
 389         if (!lookup_method) {
 390                 ret = -1;
 391                 goto end;
 392         }
 393
 394         type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
 395         switch (type) {
 396         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
 397         {
 398                 struct lttng_kernel_event_callsite callsite;
 399                 uint64_t offset;
 400
 401                 ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
 402                 if (ret) {
 403                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 404                         goto end;
 405                 }
 406
 407                 callsite.u.uprobe.offset = offset;
 408                 ret = kernctl_add_callsite(fd, &callsite);
 409                 if (ret) {
 410                         WARN("Adding callsite to userspace probe "
 411                                         "event %s failed.", ev->name);
 412                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 413                         goto end;
 414                 }
 415                 break;
 416         }
 417         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
 418         {
 419                 int i;
 420                 uint64_t *offsets = NULL;
 421                 uint32_t offsets_count;
 422                 struct lttng_kernel_event_callsite callsite;
 423
 424                 /*
 425                  * This call allocates the offsets buffer. This buffer must be freed
 426                  * by the caller
 427                  */
 428                 ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
 429                                 &offsets, &offsets_count);
 430                 if (ret) {
 431                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 432                         goto end;
 433                 }
 434                 for (i = 0; i < offsets_count; i++) {
 435                         callsite.u.uprobe.offset = offsets[i];
 436                         ret = kernctl_add_callsite(fd, &callsite);
 437                         if (ret) {
 438                                 WARN("Adding callsite to userspace probe "
 439                                                 "event %s failed.", ev->name);
 440                                 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 441                                 free(offsets);
 442                                 goto end;
 443                         }
 444                 }
 445                 free(offsets);
 446                 break;
 447         }
 448         default:
 449                 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 450                 goto end;
 451         }
 452 end:
 453         return ret;
 454 }
 455
 456 /*
 457  * Create a kernel event, enable it to the kernel tracer and add it to the
 458  * channel event list of the kernel session.
 459  * We own filter_expression and filter.
 460  */
 461 int kernel_create_event(struct lttng_event *ev,
 462                 struct ltt_kernel_channel *channel,
 463                 char *filter_expression,
 464                 struct lttng_filter_bytecode *filter)
 465 {
 466         int err, fd;
 467         enum lttng_error_code ret;
 468         struct ltt_kernel_event *event;
 469
 470         assert(ev);
 471         assert(channel);
 472
 473         /* We pass ownership of filter_expression and filter */
 474         ret = trace_kernel_create_event(ev, filter_expression,
 475                         filter, &event);
 476         if (ret != LTTNG_OK) {
 477                 goto error;
 478         }
 479
 480         fd = kernctl_create_event(channel->fd, event->event);
 481         if (fd < 0) {
 482                 switch (-fd) {
 483                 case EEXIST:
 484                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 485                         break;
 486                 case ENOSYS:
 487                         WARN("Event type not implemented");
 488                         ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
 489                         break;
 490                 case ENOENT:
 491                         WARN("Event %s not found!", ev->name);
 492                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 493                         break;
 494                 default:
 495                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 496                         PERROR("create event ioctl");
 497                 }
 498                 goto free_event;
 499         }
 500
 501         event->type = ev->type;
 502         event->fd = fd;
 503         /* Prevent fd duplication after execlp() */
 504         err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 505         if (err < 0) {
 506                 PERROR("fcntl session fd");
 507         }
 508
 509         if (filter) {
 510                 err = kernctl_filter(event->fd, filter);
 511                 if (err < 0) {
 512                         switch (-err) {
 513                         case ENOMEM:
 514                                 ret = LTTNG_ERR_FILTER_NOMEM;
 515                                 break;
 516                         default:
 517                                 ret = LTTNG_ERR_FILTER_INVAL;
 518                                 break;
 519                         }
 520                         goto filter_error;
 521                 }
 522         }
 523
 524         if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
 525                 ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
 526                 if (ret) {
 527                         goto add_callsite_error;
 528                 }
 529         }
 530
 531         err = kernctl_enable(event->fd);
 532         if (err < 0) {
 533                 switch (-err) {
 534                 case EEXIST:
 535                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 536                         break;
 537                 default:
 538                         PERROR("enable kernel event");
 539                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 540                         break;
 541                 }
 542                 goto enable_error;
 543         }
 544
 545         /* Add event to event list */
 546         cds_list_add(&event->list, &channel->events_list.head);
 547         channel->event_count++;
 548
 549         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 550
 551         return 0;
 552
 553 add_callsite_error:
 554 enable_error:
 555 filter_error:
 556         {
 557                 int closeret;
 558
 559                 closeret = close(event->fd);
 560                 if (closeret) {
 561                         PERROR("close event fd");
 562                 }
 563         }
 564 free_event:
 565         free(event);
 566 error:
 567         return ret;
 568 }
 569
 570 /*
 571  * Disable a kernel channel.
 572  */
 573 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 574 {
 575         int ret;
 576
 577         assert(chan);
 578
 579         ret = kernctl_disable(chan->fd);
 580         if (ret < 0) {
 581                 PERROR("disable chan ioctl");
 582                 goto error;
 583         }
 584
 585         chan->enabled = 0;
 586         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 587                         chan->channel->name, chan->fd, chan->key);
 588
 589         return 0;
 590
 591 error:
 592         return ret;
 593 }
 594
 595 /*
 596  * Enable a kernel channel.
 597  */
 598 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 599 {
 600         int ret;
 601
 602         assert(chan);
 603
 604         ret = kernctl_enable(chan->fd);
 605         if (ret < 0 && ret != -EEXIST) {
 606                 PERROR("Enable kernel chan");
 607                 goto error;
 608         }
 609
 610         chan->enabled = 1;
 611         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 612                         chan->channel->name, chan->fd, chan->key);
 613
 614         return 0;
 615
 616 error:
 617         return ret;
 618 }
 619
 620 /*
 621  * Enable a kernel event.
 622  */
 623 int kernel_enable_event(struct ltt_kernel_event *event)
 624 {
 625         int ret;
 626
 627         assert(event);
 628
 629         ret = kernctl_enable(event->fd);
 630         if (ret < 0) {
 631                 switch (-ret) {
 632                 case EEXIST:
 633                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 634                         break;
 635                 default:
 636                         PERROR("enable kernel event");
 637                         break;
 638                 }
 639                 goto error;
 640         }
 641
 642         event->enabled = 1;
 643         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 644
 645         return 0;
 646
 647 error:
 648         return ret;
 649 }
 650
 651 /*
 652  * Disable a kernel event.
 653  */
 654 int kernel_disable_event(struct ltt_kernel_event *event)
 655 {
 656         int ret;
 657
 658         assert(event);
 659
 660         ret = kernctl_disable(event->fd);
 661         if (ret < 0) {
 662                 switch (-ret) {
 663                 case EEXIST:
 664                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 665                         break;
 666                 default:
 667                         PERROR("disable kernel event");
 668                         break;
 669                 }
 670                 goto error;
 671         }
 672
 673         event->enabled = 0;
 674         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 675
 676         return 0;
 677
 678 error:
 679         return ret;
 680 }
 681
 682
 683 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 684 {
 685         int ret;
 686
 687         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 688                         pid, session->id);
 689         ret = kernctl_track_pid(session->fd, pid);
 690         if (!ret) {
 691                 return LTTNG_OK;
 692         }
 693         switch (-ret) {
 694         case EINVAL:
 695                 return LTTNG_ERR_INVALID;
 696         case ENOMEM:
 697                 return LTTNG_ERR_NOMEM;
 698         case EEXIST:
 699                 return LTTNG_ERR_PID_TRACKED;
 700         default:
 701                 return LTTNG_ERR_UNK;
 702         }
 703 }
 704
 705 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 706 {
 707         int ret;
 708
 709         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 710                         pid, session->id);
 711         ret = kernctl_untrack_pid(session->fd, pid);
 712         if (!ret) {
 713                 return LTTNG_OK;
 714         }
 715         switch (-ret) {
 716         case EINVAL:
 717                 return LTTNG_ERR_INVALID;
 718         case ENOMEM:
 719                 return LTTNG_ERR_NOMEM;
 720         case ENOENT:
 721                 return LTTNG_ERR_PID_NOT_TRACKED;
 722         default:
 723                 return LTTNG_ERR_UNK;
 724         }
 725 }
 726
 727 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 728                 int **_pids)
 729 {
 730         int fd, ret;
 731         int pid;
 732         ssize_t nbmem, count = 0;
 733         FILE *fp;
 734         int *pids;
 735
 736         fd = kernctl_list_tracker_pids(session->fd);
 737         if (fd < 0) {
 738                 PERROR("kernel tracker pids list");
 739                 goto error;
 740         }
 741
 742         fp = fdopen(fd, "r");
 743         if (fp == NULL) {
 744                 PERROR("kernel tracker pids list fdopen");
 745                 goto error_fp;
 746         }
 747
 748         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 749         pids = zmalloc(sizeof(*pids) * nbmem);
 750         if (pids == NULL) {
 751                 PERROR("alloc list pids");
 752                 count = -ENOMEM;
 753                 goto end;
 754         }
 755
 756         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 757                 if (count >= nbmem) {
 758                         int *new_pids;
 759                         size_t new_nbmem;
 760
 761                         new_nbmem = nbmem << 1;
 762                         DBG("Reallocating pids list from %zu to %zu entries",
 763                                         nbmem, new_nbmem);
 764                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 765                         if (new_pids == NULL) {
 766                                 PERROR("realloc list events");
 767                                 free(pids);
 768                                 count = -ENOMEM;
 769                                 goto end;
 770                         }
 771                         /* Zero the new memory */
 772                         memset(new_pids + nbmem, 0,
 773                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 774                         nbmem = new_nbmem;
 775                         pids = new_pids;
 776                 }
 777                 pids[count++] = pid;
 778         }
 779
 780         *_pids = pids;
 781         DBG("Kernel list tracker pids done (%zd pids)", count);
 782 end:
 783         ret = fclose(fp);       /* closes both fp and fd */
 784         if (ret) {
 785                 PERROR("fclose");
 786         }
 787         return count;
 788
 789 error_fp:
 790         ret = close(fd);
 791         if (ret) {
 792                 PERROR("close");
 793         }
 794 error:
 795         return -1;
 796 }
 797
 798 /*
 799  * Create kernel metadata, open from the kernel tracer and add it to the
 800  * kernel session.
 801  */
 802 int kernel_open_metadata(struct ltt_kernel_session *session)
 803 {
 804         int ret;
 805         struct ltt_kernel_metadata *lkm = NULL;
 806
 807         assert(session);
 808
 809         /* Allocate kernel metadata */
 810         lkm = trace_kernel_create_metadata();
 811         if (lkm == NULL) {
 812                 goto error;
 813         }
 814
 815         /* Kernel tracer metadata creation */
 816         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 817         if (ret < 0) {
 818                 goto error_open;
 819         }
 820
 821         lkm->fd = ret;
 822         lkm->key = ++next_kernel_channel_key;
 823         /* Prevent fd duplication after execlp() */
 824         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 825         if (ret < 0) {
 826                 PERROR("fcntl session fd");
 827         }
 828
 829         session->metadata = lkm;
 830
 831         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 832
 833         return 0;
 834
 835 error_open:
 836         trace_kernel_destroy_metadata(lkm);
 837 error:
 838         return -1;
 839 }
 840
 841 /*
 842  * Start tracing session.
 843  */
 844 int kernel_start_session(struct ltt_kernel_session *session)
 845 {
 846         int ret;
 847
 848         assert(session);
 849
 850         ret = kernctl_start_session(session->fd);
 851         if (ret < 0) {
 852                 PERROR("ioctl start session");
 853                 goto error;
 854         }
 855
 856         DBG("Kernel session started");
 857
 858         return 0;
 859
 860 error:
 861         return ret;
 862 }
 863
 864 /*
 865  * Make a kernel wait to make sure in-flight probe have completed.
 866  */
 867 void kernel_wait_quiescent(void)
 868 {
 869         int ret;
 870         int fd = kernel_tracer_fd;
 871
 872         DBG("Kernel quiescent wait on %d", fd);
 873
 874         ret = kernctl_wait_quiescent(fd);
 875         if (ret < 0) {
 876                 PERROR("wait quiescent ioctl");
 877                 ERR("Kernel quiescent wait failed");
 878         }
 879 }
 880
 881 /*
 882  *  Force flush buffer of metadata.
 883  */
 884 int kernel_metadata_flush_buffer(int fd)
 885 {
 886         int ret;
 887
 888         DBG("Kernel flushing metadata buffer on fd %d", fd);
 889
 890         ret = kernctl_buffer_flush(fd);
 891         if (ret < 0) {
 892                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 893         }
 894
 895         return 0;
 896 }
 897
 898 /*
 899  * Force flush buffer for channel.
 900  */
 901 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 902 {
 903         int ret;
 904         struct ltt_kernel_stream *stream;
 905
 906         assert(channel);
 907
 908         DBG("Flush buffer for channel %s", channel->channel->name);
 909
 910         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 911                 DBG("Flushing channel stream %d", stream->fd);
 912                 ret = kernctl_buffer_flush(stream->fd);
 913                 if (ret < 0) {
 914                         PERROR("ioctl");
 915                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 916                                         stream->fd, ret);
 917                 }
 918         }
 919
 920         return 0;
 921 }
 922
 923 /*
 924  * Stop tracing session.
 925  */
 926 int kernel_stop_session(struct ltt_kernel_session *session)
 927 {
 928         int ret;
 929
 930         assert(session);
 931
 932         ret = kernctl_stop_session(session->fd);
 933         if (ret < 0) {
 934                 goto error;
 935         }
 936
 937         DBG("Kernel session stopped");
 938
 939         return 0;
 940
 941 error:
 942         return ret;
 943 }
 944
 945 /*
 946  * Open stream of channel, register it to the kernel tracer and add it
 947  * to the stream list of the channel.
 948  *
 949  * Note: given that the streams may appear in random order wrt CPU
 950  * number (e.g. cpu hotplug), the index value of the stream number in
 951  * the stream name is not necessarily linked to the CPU number.
 952  *
 953  * Return the number of created stream. Else, a negative value.
 954  */
 955 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 956 {
 957         int ret;
 958         struct ltt_kernel_stream *lks;
 959
 960         assert(channel);
 961
 962         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 963                 lks = trace_kernel_create_stream(channel->channel->name,
 964                                 channel->stream_count);
 965                 if (lks == NULL) {
 966                         ret = close(ret);
 967                         if (ret) {
 968                                 PERROR("close");
 969                         }
 970                         goto error;
 971                 }
 972
 973                 lks->fd = ret;
 974                 /* Prevent fd duplication after execlp() */
 975                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 976                 if (ret < 0) {
 977                         PERROR("fcntl session fd");
 978                 }
 979
 980                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 981                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 982
 983                 /* Add stream to channel stream list */
 984                 cds_list_add(&lks->list, &channel->stream_list.head);
 985                 channel->stream_count++;
 986
 987                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 988                                 lks->state);
 989         }
 990
 991         return channel->stream_count;
 992
 993 error:
 994         return -1;
 995 }
 996
 997 /*
 998  * Open the metadata stream and set it to the kernel session.
 999  */
1000 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
1001 {
1002         int ret;
1003
1004         assert(session);
1005
1006         ret = kernctl_create_stream(session->metadata->fd);
1007         if (ret < 0) {
1008                 PERROR("kernel create metadata stream");
1009                 goto error;
1010         }
1011
1012         DBG("Kernel metadata stream created (fd: %d)", ret);
1013         session->metadata_stream_fd = ret;
1014         /* Prevent fd duplication after execlp() */
1015         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
1016         if (ret < 0) {
1017                 PERROR("fcntl session fd");
1018         }
1019
1020         return 0;
1021
1022 error:
1023         return -1;
1024 }
1025
1026 /*
1027  * Get the event list from the kernel tracer and return the number of elements.
1028  */
1029 ssize_t kernel_list_events(struct lttng_event **events)
1030 {
1031         int fd, ret;
1032         char *event;
1033         size_t nbmem, count = 0;
1034         FILE *fp;
1035         struct lttng_event *elist;
1036
1037         assert(events);
1038
1039         fd = kernctl_tracepoint_list(kernel_tracer_fd);
1040         if (fd < 0) {
1041                 PERROR("kernel tracepoint list");
1042                 goto error;
1043         }
1044
1045         fp = fdopen(fd, "r");
1046         if (fp == NULL) {
1047                 PERROR("kernel tracepoint list fdopen");
1048                 goto error_fp;
1049         }
1050
1051         /*
1052          * Init memory size counter
1053          * See kernel-ctl.h for explanation of this value
1054          */
1055         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1056         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1057         if (elist == NULL) {
1058                 PERROR("alloc list events");
1059                 count = -ENOMEM;
1060                 goto end;
1061         }
1062
1063         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1064                 if (count >= nbmem) {
1065                         struct lttng_event *new_elist;
1066                         size_t new_nbmem;
1067
1068                         new_nbmem = nbmem << 1;
1069                         DBG("Reallocating event list from %zu to %zu bytes",
1070                                         nbmem, new_nbmem);
1071                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1072                         if (new_elist == NULL) {
1073                                 PERROR("realloc list events");
1074                                 free(event);
1075                                 free(elist);
1076                                 count = -ENOMEM;
1077                                 goto end;
1078                         }
1079                         /* Zero the new memory */
1080                         memset(new_elist + nbmem, 0,
1081                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1082                         nbmem = new_nbmem;
1083                         elist = new_elist;
1084                 }
1085                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1086                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1087                 elist[count].enabled = -1;
1088                 count++;
1089                 free(event);
1090         }
1091
1092         *events = elist;
1093         DBG("Kernel list events done (%zu events)", count);
1094 end:
1095         ret = fclose(fp);       /* closes both fp and fd */
1096         if (ret) {
1097                 PERROR("fclose");
1098         }
1099         return count;
1100
1101 error_fp:
1102         ret = close(fd);
1103         if (ret) {
1104                 PERROR("close");
1105         }
1106 error:
1107         return -1;
1108 }
1109
1110 /*
1111  * Get kernel version and validate it.
1112  */
1113 int kernel_validate_version(struct lttng_kernel_tracer_version *version,
1114                 struct lttng_kernel_tracer_abi_version *abi_version)
1115 {
1116         int ret;
1117
1118         ret = kernctl_tracer_version(kernel_tracer_fd, version);
1119         if (ret < 0) {
1120                 ERR("Failed to retrieve the lttng-modules version");
1121                 goto error;
1122         }
1123
1124         /* Validate version */
1125         if (version->major != VERSION_MAJOR) {
1126                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1127                         version->major, VERSION_MAJOR);
1128                 goto error_version;
1129         }
1130         ret = kernctl_tracer_abi_version(kernel_tracer_fd, abi_version);
1131         if (ret < 0) {
1132                 ERR("Failed to retrieve lttng-modules ABI version");
1133                 goto error;
1134         }
1135         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1136                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1137                         abi_version->major, abi_version->minor,
1138                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1139                 goto error;
1140         }
1141         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1142                         version->major, version->minor,
1143                         abi_version->major, abi_version->minor);
1144         return 0;
1145
1146 error_version:
1147         ret = -1;
1148
1149 error:
1150         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1151         return ret;
1152 }
1153
1154 /*
1155  * Kernel work-arounds called at the start of sessiond main().
1156  */
1157 int init_kernel_workarounds(void)
1158 {
1159         int ret;
1160         FILE *fp;
1161
1162         /*
1163          * boot_id needs to be read once before being used concurrently
1164          * to deal with a Linux kernel race. A fix is proposed for
1165          * upstream, but the work-around is needed for older kernels.
1166          */
1167         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1168         if (!fp) {
1169                 goto end_boot_id;
1170         }
1171         while (!feof(fp)) {
1172                 char buf[37] = "";
1173
1174                 ret = fread(buf, 1, sizeof(buf), fp);
1175                 if (ret < 0) {
1176                         /* Ignore error, we don't really care */
1177                 }
1178         }
1179         ret = fclose(fp);
1180         if (ret) {
1181                 PERROR("fclose");
1182         }
1183 end_boot_id:
1184         return 0;
1185 }
1186
1187 /*
1188  * Teardown of a kernel session, keeping data required by destroy notifiers.
1189  */
1190 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1191 {
1192         struct lttng_trace_chunk *trace_chunk;
1193
1194         if (ksess == NULL) {
1195                 DBG3("No kernel session when tearing down session");
1196                 return;
1197         }
1198
1199         DBG("Tearing down kernel session");
1200         trace_chunk = ksess->current_trace_chunk;
1201
1202         /*
1203          * Destroy channels on the consumer if at least one FD has been sent and we
1204          * are in no output mode because the streams are in *no* monitor mode so we
1205          * have to send a command to clean them up or else they leaked.
1206          */
1207         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1208                 int ret;
1209                 struct consumer_socket *socket;
1210                 struct lttng_ht_iter iter;
1211
1212                 /* For each consumer socket. */
1213                 rcu_read_lock();
1214                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1215                                 socket, node.node) {
1216                         struct ltt_kernel_channel *chan;
1217
1218                         /* For each channel, ask the consumer to destroy it. */
1219                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1220                                 ret = kernel_consumer_destroy_channel(socket, chan);
1221                                 if (ret < 0) {
1222                                         /* Consumer is probably dead. Use next socket. */
1223                                         continue;
1224                                 }
1225                         }
1226                 }
1227                 rcu_read_unlock();
1228         }
1229
1230         /* Close any relayd session */
1231         consumer_output_send_destroy_relayd(ksess->consumer);
1232
1233         trace_kernel_destroy_session(ksess);
1234         lttng_trace_chunk_put(trace_chunk);
1235 }
1236
1237 /* Teardown of data required by destroy notifiers. */
1238 void kernel_free_session(struct ltt_kernel_session *ksess)
1239 {
1240         if (ksess == NULL) {
1241                 return;
1242         }
1243         trace_kernel_free_session(ksess);
1244 }
1245
1246 /*
1247  * Destroy a kernel channel object. It does not do anything on the tracer side.
1248  */
1249 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1250 {
1251         struct ltt_kernel_session *ksess = NULL;
1252
1253         assert(kchan);
1254         assert(kchan->channel);
1255
1256         DBG3("Kernel destroy channel %s", kchan->channel->name);
1257
1258         /* Update channel count of associated session. */
1259         if (kchan->session) {
1260                 /* Keep pointer reference so we can update it after the destroy. */
1261                 ksess = kchan->session;
1262         }
1263
1264         trace_kernel_destroy_channel(kchan);
1265
1266         /*
1267          * At this point the kernel channel is not visible anymore. This is safe
1268          * since in order to work on a visible kernel session, the tracing session
1269          * lock (ltt_session.lock) MUST be acquired.
1270          */
1271         if (ksess) {
1272                 ksess->channel_count--;
1273         }
1274 }
1275
1276 /*
1277  * Take a snapshot for a given kernel session.
1278  *
1279  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1280  */
1281 enum lttng_error_code kernel_snapshot_record(
1282                 struct ltt_kernel_session *ksess,
1283                 const struct consumer_output *output, int wait,
1284                 uint64_t nb_packets_per_stream)
1285 {
1286         int err, ret, saved_metadata_fd;
1287         enum lttng_error_code status = LTTNG_OK;
1288         struct consumer_socket *socket;
1289         struct lttng_ht_iter iter;
1290         struct ltt_kernel_metadata *saved_metadata;
1291         char *trace_path = NULL;
1292         size_t consumer_path_offset = 0;
1293
1294         assert(ksess);
1295         assert(ksess->consumer);
1296         assert(output);
1297
1298         DBG("Kernel snapshot record started");
1299
1300         /* Save current metadata since the following calls will change it. */
1301         saved_metadata = ksess->metadata;
1302         saved_metadata_fd = ksess->metadata_stream_fd;
1303
1304         rcu_read_lock();
1305
1306         ret = kernel_open_metadata(ksess);
1307         if (ret < 0) {
1308                 status = LTTNG_ERR_KERN_META_FAIL;
1309                 goto error;
1310         }
1311
1312         ret = kernel_open_metadata_stream(ksess);
1313         if (ret < 0) {
1314                 status = LTTNG_ERR_KERN_META_FAIL;
1315                 goto error_open_stream;
1316         }
1317
1318         trace_path = setup_channel_trace_path(ksess->consumer,
1319                         DEFAULT_KERNEL_TRACE_DIR, &consumer_path_offset);
1320         if (!trace_path) {
1321                 status = LTTNG_ERR_INVALID;
1322                 goto error;
1323         }
1324         /* Send metadata to consumer and snapshot everything. */
1325         cds_lfht_for_each_entry(output->socks->ht, &iter.iter,
1326                         socket, node.node) {
1327                 struct ltt_kernel_channel *chan;
1328
1329                 pthread_mutex_lock(socket->lock);
1330                 /* This stream must not be monitored by the consumer. */
1331                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1332                 pthread_mutex_unlock(socket->lock);
1333                 if (ret < 0) {
1334                         status = LTTNG_ERR_KERN_META_FAIL;
1335                         goto error_consumer;
1336                 }
1337
1338                 /* For each channel, ask the consumer to snapshot it. */
1339                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1340                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1341                                         ksess->uid, ksess->gid,
1342                                         &trace_path[consumer_path_offset], wait,
1343                                         nb_packets_per_stream);
1344                         if (status != LTTNG_OK) {
1345                                 (void) kernel_consumer_destroy_metadata(socket,
1346                                                 ksess->metadata);
1347                                 goto error_consumer;
1348                         }
1349                 }
1350
1351                 /* Snapshot metadata, */
1352                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1353                                 1, ksess->uid, ksess->gid, &trace_path[consumer_path_offset],
1354                                 wait, 0);
1355                 if (status != LTTNG_OK) {
1356                         goto error_consumer;
1357                 }
1358
1359                 /*
1360                  * The metadata snapshot is done, ask the consumer to destroy it since
1361                  * it's not monitored on the consumer side.
1362                  */
1363                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1364         }
1365
1366 error_consumer:
1367         /* Close newly opened metadata stream. It's now on the consumer side. */
1368         err = close(ksess->metadata_stream_fd);
1369         if (err < 0) {
1370                 PERROR("close snapshot kernel");
1371         }
1372
1373 error_open_stream:
1374         trace_kernel_destroy_metadata(ksess->metadata);
1375 error:
1376         /* Restore metadata state.*/
1377         ksess->metadata = saved_metadata;
1378         ksess->metadata_stream_fd = saved_metadata_fd;
1379         rcu_read_unlock();
1380         free(trace_path);
1381         return status;
1382 }
1383
1384 /*
1385  * Get the syscall mask array from the kernel tracer.
1386  *
1387  * Return 0 on success else a negative value. In both case, syscall_mask should
1388  * be freed.
1389  */
1390 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1391 {
1392         assert(syscall_mask);
1393         assert(nr_bits);
1394
1395         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1396 }
1397
1398 /*
1399  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1400  * version number.
1401  *
1402  * Return 1 on success, 0 when feature is not supported, negative value in case
1403  * of errors.
1404  */
1405 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1406 {
1407         int ret = 0; // Not supported by default
1408         struct lttng_kernel_tracer_abi_version abi;
1409
1410         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1411         if (ret < 0) {
1412                 ERR("Failed to retrieve lttng-modules ABI version");
1413                 goto error;
1414         }
1415
1416         /*
1417          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1418          */
1419         if (abi.major >= 2 && abi.minor >= 3) {
1420                 /* Supported */
1421                 ret = 1;
1422         } else {
1423                 /* Not supported */
1424                 ret = 0;
1425         }
1426 error:
1427         return ret;
1428 }
1429
1430 /*
1431  * Check for the support of the packet sequence number via abi version number.
1432  *
1433  * Return 1 on success, 0 when feature is not supported, negative value in case
1434  * of errors.
1435  */
1436 int kernel_supports_ring_buffer_packet_sequence_number(void)
1437 {
1438         int ret = 0; // Not supported by default
1439         struct lttng_kernel_tracer_abi_version abi;
1440
1441         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1442         if (ret < 0) {
1443                 ERR("Failed to retrieve lttng-modules ABI version");
1444                 goto error;
1445         }
1446
1447         /*
1448          * Packet sequence number was introduced in 2.8
1449          */
1450         if (abi.major >= 2 && abi.minor >= 8) {
1451                 /* Supported */
1452                 ret = 1;
1453         } else {
1454                 /* Not supported */
1455                 ret = 0;
1456         }
1457 error:
1458         return ret;
1459 }
1460
1461 /*
1462  * Rotate a kernel session.
1463  *
1464  * Return LTTNG_OK on success or else an LTTng error code.
1465  */
1466 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1467 {
1468         int ret;
1469         enum lttng_error_code status = LTTNG_OK;
1470         struct consumer_socket *socket;
1471         struct lttng_ht_iter iter;
1472         struct ltt_kernel_session *ksess = session->kernel_session;
1473
1474         assert(ksess);
1475         assert(ksess->consumer);
1476
1477         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1478                         session->name, session->id);
1479
1480         rcu_read_lock();
1481
1482         /*
1483          * Note that this loop will end after one iteration given that there is
1484          * only one kernel consumer.
1485          */
1486         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1487                         socket, node.node) {
1488                 struct ltt_kernel_channel *chan;
1489
1490                 /* For each channel, ask the consumer to rotate it. */
1491                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1492                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1493                                         chan->key, session->name);
1494                         ret = consumer_rotate_channel(socket, chan->key,
1495                                         ksess->uid, ksess->gid, ksess->consumer,
1496                                         /* is_metadata_channel */ false);
1497                         if (ret < 0) {
1498                                 status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1499                                 goto error;
1500                         }
1501                 }
1502
1503                 /*
1504                  * Rotate the metadata channel.
1505                  */
1506                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1507                                 ksess->uid, ksess->gid, ksess->consumer,
1508                                 /* is_metadata_channel */ true);
1509                 if (ret < 0) {
1510                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1511                         goto error;
1512                 }
1513         }
1514
1515 error:
1516         rcu_read_unlock();
1517         return status;
1518 }
1519
1520 enum lttng_error_code kernel_create_channel_subdirectories(
1521                 const struct ltt_kernel_session *ksess)
1522 {
1523         enum lttng_error_code ret = LTTNG_OK;
1524         enum lttng_trace_chunk_status chunk_status;
1525
1526         rcu_read_lock();
1527         assert(ksess->current_trace_chunk);
1528
1529         /*
1530          * Create the index subdirectory which will take care
1531          * of implicitly creating the channel's path.
1532          */
1533         chunk_status = lttng_trace_chunk_create_subdirectory(
1534                         ksess->current_trace_chunk,
1535                         DEFAULT_KERNEL_TRACE_DIR "/" DEFAULT_INDEX_DIR);
1536         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1537                 ret = LTTNG_ERR_CREATE_DIR_FAIL;
1538                 goto error;
1539         }
1540 error:
1541         rcu_read_unlock();
1542         return ret;
1543 }
1544
1545 /*
1546  * Setup necessary data for kernel tracer action.
1547  */
1548 LTTNG_HIDDEN
1549 int init_kernel_tracer(void)
1550 {
1551         int ret;
1552         bool is_root = !getuid();
1553
1554         /* Modprobe lttng kernel modules */
1555         ret = modprobe_lttng_control();
1556         if (ret < 0) {
1557                 goto error;
1558         }
1559
1560         /* Open debugfs lttng */
1561         kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1562         if (kernel_tracer_fd < 0) {
1563                 DBG("Failed to open %s", module_proc_lttng);
1564                 goto error_open;
1565         }
1566
1567         /* Validate kernel version */
1568         ret = kernel_validate_version(&kernel_tracer_version,
1569                         &kernel_tracer_abi_version);
1570         if (ret < 0) {
1571                 goto error_version;
1572         }
1573
1574         ret = modprobe_lttng_data();
1575         if (ret < 0) {
1576                 goto error_modules;
1577         }
1578
1579         ret = kernel_supports_ring_buffer_snapshot_sample_positions();
1580         if (ret < 0) {
1581                 goto error_modules;
1582         }
1583
1584         if (ret < 1) {
1585                 WARN("Kernel tracer does not support buffer monitoring. "
1586                         "The monitoring timer of channels in the kernel domain "
1587                         "will be set to 0 (disabled).");
1588         }
1589
1590         DBG("Kernel tracer fd %d", kernel_tracer_fd);
1591
1592         ret = syscall_init_table(kernel_tracer_fd);
1593         if (ret < 0) {
1594                 ERR("Unable to populate syscall table. Syscall tracing won't "
1595                         "work for this session daemon.");
1596         }
1597         return 0;
1598
1599 error_version:
1600         modprobe_remove_lttng_control();
1601         ret = close(kernel_tracer_fd);
1602         if (ret) {
1603                 PERROR("close");
1604         }
1605         kernel_tracer_fd = -1;
1606         return LTTNG_ERR_KERN_VERSION;
1607
1608 error_modules:
1609         ret = close(kernel_tracer_fd);
1610         if (ret) {
1611                 PERROR("close");
1612         }
1613
1614 error_open:
1615         modprobe_remove_lttng_control();
1616
1617 error:
1618         WARN("No kernel tracer available");
1619         kernel_tracer_fd = -1;
1620         if (!is_root) {
1621                 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1622         } else {
1623                 return LTTNG_ERR_KERN_NA;
1624         }
1625 }
1626
1627 LTTNG_HIDDEN
1628 void cleanup_kernel_tracer(void)
1629 {
1630         int ret;
1631
1632         DBG2("Closing kernel fd");
1633         if (kernel_tracer_fd >= 0) {
1634                 ret = close(kernel_tracer_fd);
1635                 if (ret) {
1636                         PERROR("close");
1637                 }
1638                 kernel_tracer_fd = -1;
1639         }
1640         DBG("Unloading kernel modules");
1641         modprobe_remove_lttng_all();
1642         free(syscall_table);
1643 }
1644
1645 LTTNG_HIDDEN
1646 bool kernel_tracer_is_initialized(void)
1647 {
1648         return kernel_tracer_fd >= 0;
1649 }