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 static struct lttng_tracker_list *get_id_tracker_list(
 683                 struct ltt_kernel_session *session,
 684                 enum lttng_tracker_type tracker_type)
 685 {
 686         switch (tracker_type) {
 687         case LTTNG_TRACKER_PID:
 688                 return session->tracker_list_pid;
 689         case LTTNG_TRACKER_VPID:
 690                 return session->tracker_list_vpid;
 691         case LTTNG_TRACKER_UID:
 692                 return session->tracker_list_uid;
 693         case LTTNG_TRACKER_VUID:
 694                 return session->tracker_list_vuid;
 695         case LTTNG_TRACKER_GID:
 696                 return session->tracker_list_gid;
 697         case LTTNG_TRACKER_VGID:
 698                 return session->tracker_list_vgid;
 699         default:
 700                 return NULL;
 701         }
 702 }
 703
 704 int kernel_track_id(enum lttng_tracker_type tracker_type,
 705                 struct ltt_kernel_session *session,
 706                 const struct lttng_tracker_id *id)
 707 {
 708         int ret, value;
 709         struct lttng_tracker_list *tracker_list;
 710         struct lttng_tracker_id **saved_ids;
 711         ssize_t saved_ids_count;
 712
 713         ret = lttng_tracker_id_lookup_string(tracker_type, id, &value);
 714         if (ret != LTTNG_OK) {
 715                 return ret;
 716         }
 717
 718         tracker_list = get_id_tracker_list(session, tracker_type);
 719         if (!tracker_list) {
 720                 return LTTNG_ERR_INVALID;
 721         }
 722
 723         /* Save list for restore on error. */
 724         saved_ids_count = lttng_tracker_id_get_list(tracker_list, &saved_ids);
 725         if (saved_ids_count < 0) {
 726                 return LTTNG_ERR_INVALID;
 727         }
 728
 729         /* Add to list. */
 730         ret = lttng_tracker_list_add(tracker_list, id);
 731         if (ret != LTTNG_OK) {
 732                 goto end;
 733         }
 734
 735         switch (tracker_type) {
 736         case LTTNG_TRACKER_PID:
 737                 DBG("Kernel track PID %d for session id %" PRIu64 ".", value,
 738                                 session->id);
 739                 ret = kernctl_track_pid(session->fd, value);
 740                 if (!ret) {
 741                         ret = LTTNG_OK;
 742                         goto end;
 743                 }
 744                 break;
 745         case LTTNG_TRACKER_VPID:
 746                 DBG("Kernel track VPID %d for session id %" PRIu64 ".", value,
 747                                 session->id);
 748                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VPID, value);
 749                 if (!ret) {
 750                         ret = LTTNG_OK;
 751                         goto end;
 752                 }
 753                 break;
 754         case LTTNG_TRACKER_UID:
 755                 DBG("Kernel track UID %d for session id %" PRIu64 ".", value,
 756                                 session->id);
 757                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_UID, value);
 758                 if (!ret) {
 759                         ret = LTTNG_OK;
 760                         goto end;
 761                 }
 762                 break;
 763         case LTTNG_TRACKER_GID:
 764                 DBG("Kernel track GID %d for session id %" PRIu64 ".", value,
 765                                 session->id);
 766                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_GID, value);
 767                 if (!ret) {
 768                         ret = LTTNG_OK;
 769                         goto end;
 770                 }
 771                 break;
 772         case LTTNG_TRACKER_VUID:
 773                 DBG("Kernel track VUID %d for session id %" PRIu64 ".", value,
 774                                 session->id);
 775                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VUID, value);
 776                 if (!ret) {
 777                         ret = LTTNG_OK;
 778                         goto end;
 779                 }
 780                 break;
 781         case LTTNG_TRACKER_VGID:
 782                 DBG("Kernel track VGID %d for session id %" PRIu64 ".", value,
 783                                 session->id);
 784                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VGID, value);
 785                 if (!ret) {
 786                         ret = LTTNG_OK;
 787                         goto end;
 788                 }
 789                 break;
 790         default:
 791                 ret = -EINVAL;
 792                 break;
 793         }
 794
 795         /* Error handling. */
 796         switch (-ret) {
 797         case EINVAL:
 798                 ret = LTTNG_ERR_INVALID;
 799                 break;
 800         case ENOMEM:
 801                 ret = LTTNG_ERR_NOMEM;
 802                 break;
 803         case EEXIST:
 804                 ret = LTTNG_ERR_ID_TRACKED;
 805                 break;
 806         default:
 807                 ret = LTTNG_ERR_UNK;
 808                 break;
 809         }
 810
 811         if (lttng_tracker_id_set_list(tracker_list, saved_ids,
 812                             saved_ids_count) != LTTNG_OK) {
 813                 ERR("Error on tracker add error handling.\n");
 814         }
 815 end:
 816         lttng_tracker_ids_destroy(saved_ids, saved_ids_count);
 817         free(saved_ids);
 818         return ret;
 819 }
 820
 821 int kernel_untrack_id(enum lttng_tracker_type tracker_type,
 822                 struct ltt_kernel_session *session,
 823                 const struct lttng_tracker_id *id)
 824 {
 825         int ret, value;
 826         struct lttng_tracker_list *tracker_list;
 827         struct lttng_tracker_id **saved_ids;
 828         ssize_t saved_ids_count;
 829
 830         ret = lttng_tracker_id_lookup_string(tracker_type, id, &value);
 831         if (ret != LTTNG_OK) {
 832                 return ret;
 833         }
 834
 835         tracker_list = get_id_tracker_list(session, tracker_type);
 836         if (!tracker_list) {
 837                 return LTTNG_ERR_INVALID;
 838         }
 839         /* Save list for restore on error. */
 840         saved_ids_count = lttng_tracker_id_get_list(tracker_list, &saved_ids);
 841         if (saved_ids_count < 0) {
 842                 return LTTNG_ERR_INVALID;
 843         }
 844         /* Remove from list. */
 845         ret = lttng_tracker_list_remove(tracker_list, id);
 846         if (ret != LTTNG_OK) {
 847                 goto end;
 848         }
 849
 850         switch (tracker_type) {
 851         case LTTNG_TRACKER_PID:
 852                 DBG("Kernel untrack PID %d for session id %" PRIu64 ".", value,
 853                                 session->id);
 854                 ret = kernctl_untrack_pid(session->fd, value);
 855                 if (!ret) {
 856                         ret = LTTNG_OK;
 857                         goto end;
 858                 }
 859                 break;
 860         case LTTNG_TRACKER_VPID:
 861                 DBG("Kernel untrack VPID %d for session id %" PRIu64 ".", value,
 862                                 session->id);
 863                 ret = kernctl_untrack_id(
 864                                 session->fd, LTTNG_TRACKER_VPID, value);
 865                 if (!ret) {
 866                         ret = LTTNG_OK;
 867                         goto end;
 868                 }
 869                 break;
 870         case LTTNG_TRACKER_UID:
 871                 DBG("Kernel untrack UID %d for session id %" PRIu64 ".", value,
 872                                 session->id);
 873                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_UID, value);
 874                 if (!ret) {
 875                         ret = LTTNG_OK;
 876                         goto end;
 877                 }
 878                 break;
 879         case LTTNG_TRACKER_GID:
 880                 DBG("Kernel untrack GID %d for session id %" PRIu64 ".", value,
 881                                 session->id);
 882                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_GID, value);
 883                 if (!ret) {
 884                         ret = LTTNG_OK;
 885                         goto end;
 886                 }
 887                 break;
 888         case LTTNG_TRACKER_VUID:
 889                 DBG("Kernel untrack VUID %d for session id %" PRIu64 ".", value,
 890                                 session->id);
 891                 ret = kernctl_untrack_id(
 892                                 session->fd, LTTNG_TRACKER_VUID, value);
 893                 if (!ret) {
 894                         ret = LTTNG_OK;
 895                         goto end;
 896                 }
 897                 break;
 898         case LTTNG_TRACKER_VGID:
 899                 DBG("Kernel untrack VGID %d for session id %" PRIu64 ".", value,
 900                                 session->id);
 901                 ret = kernctl_untrack_id(
 902                                 session->fd, LTTNG_TRACKER_VGID, value);
 903                 if (!ret) {
 904                         ret = LTTNG_OK;
 905                         goto end;
 906                 }
 907                 break;
 908         default:
 909                 ret = -EINVAL;
 910                 break;
 911         }
 912
 913         /* Error handling. */
 914         switch (-ret) {
 915         case EINVAL:
 916                 ret = LTTNG_ERR_INVALID;
 917                 break;
 918         case ENOMEM:
 919                 ret = LTTNG_ERR_NOMEM;
 920                 break;
 921         case EEXIST:
 922                 ret = LTTNG_ERR_ID_TRACKED;
 923                 break;
 924         default:
 925                 ret = LTTNG_ERR_UNK;
 926                 break;
 927         }
 928
 929         if (lttng_tracker_id_set_list(tracker_list, saved_ids,
 930                             saved_ids_count) != LTTNG_OK) {
 931                 ERR("Error on tracker remove error handling.\n");
 932         }
 933 end:
 934         lttng_tracker_ids_destroy(saved_ids, saved_ids_count);
 935         free(saved_ids);
 936         return ret;
 937 }
 938
 939 /*
 940  * Called with session lock held.
 941  */
 942 ssize_t kernel_list_tracker_ids(enum lttng_tracker_type tracker_type,
 943                 struct ltt_kernel_session *session,
 944                 struct lttng_tracker_id ***_ids)
 945 {
 946         struct lttng_tracker_list *tracker_list;
 947
 948         tracker_list = get_id_tracker_list(session, tracker_type);
 949         if (!tracker_list) {
 950                 return -LTTNG_ERR_INVALID;
 951         }
 952         return lttng_tracker_id_get_list(tracker_list, _ids);
 953 }
 954
 955 /*
 956  * Create kernel metadata, open from the kernel tracer and add it to the
 957  * kernel session.
 958  */
 959 int kernel_open_metadata(struct ltt_kernel_session *session)
 960 {
 961         int ret;
 962         struct ltt_kernel_metadata *lkm = NULL;
 963
 964         assert(session);
 965
 966         /* Allocate kernel metadata */
 967         lkm = trace_kernel_create_metadata();
 968         if (lkm == NULL) {
 969                 goto error;
 970         }
 971
 972         /* Kernel tracer metadata creation */
 973         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 974         if (ret < 0) {
 975                 goto error_open;
 976         }
 977
 978         lkm->fd = ret;
 979         lkm->key = ++next_kernel_channel_key;
 980         /* Prevent fd duplication after execlp() */
 981         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 982         if (ret < 0) {
 983                 PERROR("fcntl session fd");
 984         }
 985
 986         session->metadata = lkm;
 987
 988         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 989
 990         return 0;
 991
 992 error_open:
 993         trace_kernel_destroy_metadata(lkm);
 994 error:
 995         return -1;
 996 }
 997
 998 /*
 999  * Start tracing session.
1000  */
1001 int kernel_start_session(struct ltt_kernel_session *session)
1002 {
1003         int ret;
1004
1005         assert(session);
1006
1007         ret = kernctl_start_session(session->fd);
1008         if (ret < 0) {
1009                 PERROR("ioctl start session");
1010                 goto error;
1011         }
1012
1013         DBG("Kernel session started");
1014
1015         return 0;
1016
1017 error:
1018         return ret;
1019 }
1020
1021 /*
1022  * Make a kernel wait to make sure in-flight probe have completed.
1023  */
1024 void kernel_wait_quiescent(void)
1025 {
1026         int ret;
1027         int fd = kernel_tracer_fd;
1028
1029         DBG("Kernel quiescent wait on %d", fd);
1030
1031         ret = kernctl_wait_quiescent(fd);
1032         if (ret < 0) {
1033                 PERROR("wait quiescent ioctl");
1034                 ERR("Kernel quiescent wait failed");
1035         }
1036 }
1037
1038 /*
1039  *  Force flush buffer of metadata.
1040  */
1041 int kernel_metadata_flush_buffer(int fd)
1042 {
1043         int ret;
1044
1045         DBG("Kernel flushing metadata buffer on fd %d", fd);
1046
1047         ret = kernctl_buffer_flush(fd);
1048         if (ret < 0) {
1049                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
1050         }
1051
1052         return 0;
1053 }
1054
1055 /*
1056  * Force flush buffer for channel.
1057  */
1058 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
1059 {
1060         int ret;
1061         struct ltt_kernel_stream *stream;
1062
1063         assert(channel);
1064
1065         DBG("Flush buffer for channel %s", channel->channel->name);
1066
1067         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
1068                 DBG("Flushing channel stream %d", stream->fd);
1069                 ret = kernctl_buffer_flush(stream->fd);
1070                 if (ret < 0) {
1071                         PERROR("ioctl");
1072                         ERR("Fail to flush buffer for stream %d (ret: %d)",
1073                                         stream->fd, ret);
1074                 }
1075         }
1076
1077         return 0;
1078 }
1079
1080 /*
1081  * Stop tracing session.
1082  */
1083 int kernel_stop_session(struct ltt_kernel_session *session)
1084 {
1085         int ret;
1086
1087         assert(session);
1088
1089         ret = kernctl_stop_session(session->fd);
1090         if (ret < 0) {
1091                 goto error;
1092         }
1093
1094         DBG("Kernel session stopped");
1095
1096         return 0;
1097
1098 error:
1099         return ret;
1100 }
1101
1102 /*
1103  * Open stream of channel, register it to the kernel tracer and add it
1104  * to the stream list of the channel.
1105  *
1106  * Note: given that the streams may appear in random order wrt CPU
1107  * number (e.g. cpu hotplug), the index value of the stream number in
1108  * the stream name is not necessarily linked to the CPU number.
1109  *
1110  * Return the number of created stream. Else, a negative value.
1111  */
1112 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
1113 {
1114         int ret;
1115         struct ltt_kernel_stream *lks;
1116
1117         assert(channel);
1118
1119         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
1120                 lks = trace_kernel_create_stream(channel->channel->name,
1121                                 channel->stream_count);
1122                 if (lks == NULL) {
1123                         ret = close(ret);
1124                         if (ret) {
1125                                 PERROR("close");
1126                         }
1127                         goto error;
1128                 }
1129
1130                 lks->fd = ret;
1131                 /* Prevent fd duplication after execlp() */
1132                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
1133                 if (ret < 0) {
1134                         PERROR("fcntl session fd");
1135                 }
1136
1137                 lks->tracefile_size = channel->channel->attr.tracefile_size;
1138                 lks->tracefile_count = channel->channel->attr.tracefile_count;
1139
1140                 /* Add stream to channel stream list */
1141                 cds_list_add(&lks->list, &channel->stream_list.head);
1142                 channel->stream_count++;
1143
1144                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
1145                                 lks->state);
1146         }
1147
1148         return channel->stream_count;
1149
1150 error:
1151         return -1;
1152 }
1153
1154 /*
1155  * Open the metadata stream and set it to the kernel session.
1156  */
1157 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
1158 {
1159         int ret;
1160
1161         assert(session);
1162
1163         ret = kernctl_create_stream(session->metadata->fd);
1164         if (ret < 0) {
1165                 PERROR("kernel create metadata stream");
1166                 goto error;
1167         }
1168
1169         DBG("Kernel metadata stream created (fd: %d)", ret);
1170         session->metadata_stream_fd = ret;
1171         /* Prevent fd duplication after execlp() */
1172         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
1173         if (ret < 0) {
1174                 PERROR("fcntl session fd");
1175         }
1176
1177         return 0;
1178
1179 error:
1180         return -1;
1181 }
1182
1183 /*
1184  * Get the event list from the kernel tracer and return the number of elements.
1185  */
1186 ssize_t kernel_list_events(struct lttng_event **events)
1187 {
1188         int fd, ret;
1189         char *event;
1190         size_t nbmem, count = 0;
1191         FILE *fp;
1192         struct lttng_event *elist;
1193
1194         assert(events);
1195
1196         fd = kernctl_tracepoint_list(kernel_tracer_fd);
1197         if (fd < 0) {
1198                 PERROR("kernel tracepoint list");
1199                 goto error;
1200         }
1201
1202         fp = fdopen(fd, "r");
1203         if (fp == NULL) {
1204                 PERROR("kernel tracepoint list fdopen");
1205                 goto error_fp;
1206         }
1207
1208         /*
1209          * Init memory size counter
1210          * See kernel-ctl.h for explanation of this value
1211          */
1212         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1213         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1214         if (elist == NULL) {
1215                 PERROR("alloc list events");
1216                 count = -ENOMEM;
1217                 goto end;
1218         }
1219
1220         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1221                 if (count >= nbmem) {
1222                         struct lttng_event *new_elist;
1223                         size_t new_nbmem;
1224
1225                         new_nbmem = nbmem << 1;
1226                         DBG("Reallocating event list from %zu to %zu bytes",
1227                                         nbmem, new_nbmem);
1228                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1229                         if (new_elist == NULL) {
1230                                 PERROR("realloc list events");
1231                                 free(event);
1232                                 free(elist);
1233                                 count = -ENOMEM;
1234                                 goto end;
1235                         }
1236                         /* Zero the new memory */
1237                         memset(new_elist + nbmem, 0,
1238                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1239                         nbmem = new_nbmem;
1240                         elist = new_elist;
1241                 }
1242                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1243                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1244                 elist[count].enabled = -1;
1245                 count++;
1246                 free(event);
1247         }
1248
1249         *events = elist;
1250         DBG("Kernel list events done (%zu events)", count);
1251 end:
1252         ret = fclose(fp);       /* closes both fp and fd */
1253         if (ret) {
1254                 PERROR("fclose");
1255         }
1256         return count;
1257
1258 error_fp:
1259         ret = close(fd);
1260         if (ret) {
1261                 PERROR("close");
1262         }
1263 error:
1264         return -1;
1265 }
1266
1267 /*
1268  * Get kernel version and validate it.
1269  */
1270 int kernel_validate_version(struct lttng_kernel_tracer_version *version,
1271                 struct lttng_kernel_tracer_abi_version *abi_version)
1272 {
1273         int ret;
1274
1275         ret = kernctl_tracer_version(kernel_tracer_fd, version);
1276         if (ret < 0) {
1277                 ERR("Failed to retrieve the lttng-modules version");
1278                 goto error;
1279         }
1280
1281         /* Validate version */
1282         if (version->major != VERSION_MAJOR) {
1283                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1284                         version->major, VERSION_MAJOR);
1285                 goto error_version;
1286         }
1287         ret = kernctl_tracer_abi_version(kernel_tracer_fd, abi_version);
1288         if (ret < 0) {
1289                 ERR("Failed to retrieve lttng-modules ABI version");
1290                 goto error;
1291         }
1292         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1293                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1294                         abi_version->major, abi_version->minor,
1295                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1296                 goto error;
1297         }
1298         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1299                         version->major, version->minor,
1300                         abi_version->major, abi_version->minor);
1301         return 0;
1302
1303 error_version:
1304         ret = -1;
1305
1306 error:
1307         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1308         return ret;
1309 }
1310
1311 /*
1312  * Kernel work-arounds called at the start of sessiond main().
1313  */
1314 int init_kernel_workarounds(void)
1315 {
1316         int ret;
1317         FILE *fp;
1318
1319         /*
1320          * boot_id needs to be read once before being used concurrently
1321          * to deal with a Linux kernel race. A fix is proposed for
1322          * upstream, but the work-around is needed for older kernels.
1323          */
1324         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1325         if (!fp) {
1326                 goto end_boot_id;
1327         }
1328         while (!feof(fp)) {
1329                 char buf[37] = "";
1330
1331                 ret = fread(buf, 1, sizeof(buf), fp);
1332                 if (ret < 0) {
1333                         /* Ignore error, we don't really care */
1334                 }
1335         }
1336         ret = fclose(fp);
1337         if (ret) {
1338                 PERROR("fclose");
1339         }
1340 end_boot_id:
1341         return 0;
1342 }
1343
1344 /*
1345  * Teardown of a kernel session, keeping data required by destroy notifiers.
1346  */
1347 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1348 {
1349         struct lttng_trace_chunk *trace_chunk;
1350
1351         if (ksess == NULL) {
1352                 DBG3("No kernel session when tearing down session");
1353                 return;
1354         }
1355
1356         DBG("Tearing down kernel session");
1357         trace_chunk = ksess->current_trace_chunk;
1358
1359         /*
1360          * Destroy channels on the consumer if at least one FD has been sent and we
1361          * are in no output mode because the streams are in *no* monitor mode so we
1362          * have to send a command to clean them up or else they leaked.
1363          */
1364         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1365                 int ret;
1366                 struct consumer_socket *socket;
1367                 struct lttng_ht_iter iter;
1368
1369                 /* For each consumer socket. */
1370                 rcu_read_lock();
1371                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1372                                 socket, node.node) {
1373                         struct ltt_kernel_channel *chan;
1374
1375                         /* For each channel, ask the consumer to destroy it. */
1376                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1377                                 ret = kernel_consumer_destroy_channel(socket, chan);
1378                                 if (ret < 0) {
1379                                         /* Consumer is probably dead. Use next socket. */
1380                                         continue;
1381                                 }
1382                         }
1383                 }
1384                 rcu_read_unlock();
1385         }
1386
1387         /* Close any relayd session */
1388         consumer_output_send_destroy_relayd(ksess->consumer);
1389
1390         trace_kernel_destroy_session(ksess);
1391         lttng_trace_chunk_put(trace_chunk);
1392 }
1393
1394 /* Teardown of data required by destroy notifiers. */
1395 void kernel_free_session(struct ltt_kernel_session *ksess)
1396 {
1397         if (ksess == NULL) {
1398                 return;
1399         }
1400         trace_kernel_free_session(ksess);
1401 }
1402
1403 /*
1404  * Destroy a kernel channel object. It does not do anything on the tracer side.
1405  */
1406 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1407 {
1408         struct ltt_kernel_session *ksess = NULL;
1409
1410         assert(kchan);
1411         assert(kchan->channel);
1412
1413         DBG3("Kernel destroy channel %s", kchan->channel->name);
1414
1415         /* Update channel count of associated session. */
1416         if (kchan->session) {
1417                 /* Keep pointer reference so we can update it after the destroy. */
1418                 ksess = kchan->session;
1419         }
1420
1421         trace_kernel_destroy_channel(kchan);
1422
1423         /*
1424          * At this point the kernel channel is not visible anymore. This is safe
1425          * since in order to work on a visible kernel session, the tracing session
1426          * lock (ltt_session.lock) MUST be acquired.
1427          */
1428         if (ksess) {
1429                 ksess->channel_count--;
1430         }
1431 }
1432
1433 /*
1434  * Take a snapshot for a given kernel session.
1435  *
1436  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1437  */
1438 enum lttng_error_code kernel_snapshot_record(
1439                 struct ltt_kernel_session *ksess,
1440                 const struct consumer_output *output, int wait,
1441                 uint64_t nb_packets_per_stream)
1442 {
1443         int err, ret, saved_metadata_fd;
1444         enum lttng_error_code status = LTTNG_OK;
1445         struct consumer_socket *socket;
1446         struct lttng_ht_iter iter;
1447         struct ltt_kernel_metadata *saved_metadata;
1448         char *trace_path = NULL;
1449         size_t consumer_path_offset = 0;
1450
1451         assert(ksess);
1452         assert(ksess->consumer);
1453         assert(output);
1454
1455         DBG("Kernel snapshot record started");
1456
1457         /* Save current metadata since the following calls will change it. */
1458         saved_metadata = ksess->metadata;
1459         saved_metadata_fd = ksess->metadata_stream_fd;
1460
1461         rcu_read_lock();
1462
1463         ret = kernel_open_metadata(ksess);
1464         if (ret < 0) {
1465                 status = LTTNG_ERR_KERN_META_FAIL;
1466                 goto error;
1467         }
1468
1469         ret = kernel_open_metadata_stream(ksess);
1470         if (ret < 0) {
1471                 status = LTTNG_ERR_KERN_META_FAIL;
1472                 goto error_open_stream;
1473         }
1474
1475         trace_path = setup_channel_trace_path(ksess->consumer,
1476                         DEFAULT_KERNEL_TRACE_DIR, &consumer_path_offset);
1477         if (!trace_path) {
1478                 status = LTTNG_ERR_INVALID;
1479                 goto error;
1480         }
1481         /* Send metadata to consumer and snapshot everything. */
1482         cds_lfht_for_each_entry(output->socks->ht, &iter.iter,
1483                         socket, node.node) {
1484                 struct ltt_kernel_channel *chan;
1485
1486                 pthread_mutex_lock(socket->lock);
1487                 /* This stream must not be monitored by the consumer. */
1488                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1489                 pthread_mutex_unlock(socket->lock);
1490                 if (ret < 0) {
1491                         status = LTTNG_ERR_KERN_META_FAIL;
1492                         goto error_consumer;
1493                 }
1494
1495                 /* For each channel, ask the consumer to snapshot it. */
1496                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1497                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1498                                         ksess->uid, ksess->gid,
1499                                         &trace_path[consumer_path_offset], wait,
1500                                         nb_packets_per_stream);
1501                         if (status != LTTNG_OK) {
1502                                 (void) kernel_consumer_destroy_metadata(socket,
1503                                                 ksess->metadata);
1504                                 goto error_consumer;
1505                         }
1506                 }
1507
1508                 /* Snapshot metadata, */
1509                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1510                                 1, ksess->uid, ksess->gid, &trace_path[consumer_path_offset],
1511                                 wait, 0);
1512                 if (status != LTTNG_OK) {
1513                         goto error_consumer;
1514                 }
1515
1516                 /*
1517                  * The metadata snapshot is done, ask the consumer to destroy it since
1518                  * it's not monitored on the consumer side.
1519                  */
1520                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1521         }
1522
1523 error_consumer:
1524         /* Close newly opened metadata stream. It's now on the consumer side. */
1525         err = close(ksess->metadata_stream_fd);
1526         if (err < 0) {
1527                 PERROR("close snapshot kernel");
1528         }
1529
1530 error_open_stream:
1531         trace_kernel_destroy_metadata(ksess->metadata);
1532 error:
1533         /* Restore metadata state.*/
1534         ksess->metadata = saved_metadata;
1535         ksess->metadata_stream_fd = saved_metadata_fd;
1536         rcu_read_unlock();
1537         free(trace_path);
1538         return status;
1539 }
1540
1541 /*
1542  * Get the syscall mask array from the kernel tracer.
1543  *
1544  * Return 0 on success else a negative value. In both case, syscall_mask should
1545  * be freed.
1546  */
1547 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1548 {
1549         assert(syscall_mask);
1550         assert(nr_bits);
1551
1552         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1553 }
1554
1555 /*
1556  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1557  * version number.
1558  *
1559  * Return 1 on success, 0 when feature is not supported, negative value in case
1560  * of errors.
1561  */
1562 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1563 {
1564         int ret = 0; // Not supported by default
1565         struct lttng_kernel_tracer_abi_version abi;
1566
1567         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1568         if (ret < 0) {
1569                 ERR("Failed to retrieve lttng-modules ABI version");
1570                 goto error;
1571         }
1572
1573         /*
1574          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1575          */
1576         if (abi.major >= 2 && abi.minor >= 3) {
1577                 /* Supported */
1578                 ret = 1;
1579         } else {
1580                 /* Not supported */
1581                 ret = 0;
1582         }
1583 error:
1584         return ret;
1585 }
1586
1587 /*
1588  * Check for the support of the packet sequence number via abi version number.
1589  *
1590  * Return 1 on success, 0 when feature is not supported, negative value in case
1591  * of errors.
1592  */
1593 int kernel_supports_ring_buffer_packet_sequence_number(void)
1594 {
1595         int ret = 0; // Not supported by default
1596         struct lttng_kernel_tracer_abi_version abi;
1597
1598         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1599         if (ret < 0) {
1600                 ERR("Failed to retrieve lttng-modules ABI version");
1601                 goto error;
1602         }
1603
1604         /*
1605          * Packet sequence number was introduced in LTTng 2.8,
1606          * lttng-modules ABI 2.1.
1607          */
1608         if (abi.major >= 2 && abi.minor >= 1) {
1609                 /* Supported */
1610                 ret = 1;
1611         } else {
1612                 /* Not supported */
1613                 ret = 0;
1614         }
1615 error:
1616         return ret;
1617 }
1618
1619 /*
1620  * Rotate a kernel session.
1621  *
1622  * Return LTTNG_OK on success or else an LTTng error code.
1623  */
1624 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1625 {
1626         int ret;
1627         enum lttng_error_code status = LTTNG_OK;
1628         struct consumer_socket *socket;
1629         struct lttng_ht_iter iter;
1630         struct ltt_kernel_session *ksess = session->kernel_session;
1631
1632         assert(ksess);
1633         assert(ksess->consumer);
1634
1635         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1636                         session->name, session->id);
1637
1638         rcu_read_lock();
1639
1640         /*
1641          * Note that this loop will end after one iteration given that there is
1642          * only one kernel consumer.
1643          */
1644         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1645                         socket, node.node) {
1646                 struct ltt_kernel_channel *chan;
1647
1648                 /* For each channel, ask the consumer to rotate it. */
1649                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1650                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1651                                         chan->key, session->name);
1652                         ret = consumer_rotate_channel(socket, chan->key,
1653                                         ksess->uid, ksess->gid, ksess->consumer,
1654                                         /* is_metadata_channel */ false);
1655                         if (ret < 0) {
1656                                 status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1657                                 goto error;
1658                         }
1659                 }
1660
1661                 /*
1662                  * Rotate the metadata channel.
1663                  */
1664                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1665                                 ksess->uid, ksess->gid, ksess->consumer,
1666                                 /* is_metadata_channel */ true);
1667                 if (ret < 0) {
1668                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1669                         goto error;
1670                 }
1671         }
1672
1673 error:
1674         rcu_read_unlock();
1675         return status;
1676 }
1677
1678 enum lttng_error_code kernel_create_channel_subdirectories(
1679                 const struct ltt_kernel_session *ksess)
1680 {
1681         enum lttng_error_code ret = LTTNG_OK;
1682         enum lttng_trace_chunk_status chunk_status;
1683
1684         rcu_read_lock();
1685         assert(ksess->current_trace_chunk);
1686
1687         /*
1688          * Create the index subdirectory which will take care
1689          * of implicitly creating the channel's path.
1690          */
1691         chunk_status = lttng_trace_chunk_create_subdirectory(
1692                         ksess->current_trace_chunk,
1693                         DEFAULT_KERNEL_TRACE_DIR "/" DEFAULT_INDEX_DIR);
1694         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1695                 ret = LTTNG_ERR_CREATE_DIR_FAIL;
1696                 goto error;
1697         }
1698 error:
1699         rcu_read_unlock();
1700         return ret;
1701 }
1702
1703 /*
1704  * Setup necessary data for kernel tracer action.
1705  */
1706 LTTNG_HIDDEN
1707 int init_kernel_tracer(void)
1708 {
1709         int ret;
1710         bool is_root = !getuid();
1711
1712         /* Modprobe lttng kernel modules */
1713         ret = modprobe_lttng_control();
1714         if (ret < 0) {
1715                 goto error;
1716         }
1717
1718         /* Open debugfs lttng */
1719         kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1720         if (kernel_tracer_fd < 0) {
1721                 DBG("Failed to open %s", module_proc_lttng);
1722                 goto error_open;
1723         }
1724
1725         /* Validate kernel version */
1726         ret = kernel_validate_version(&kernel_tracer_version,
1727                         &kernel_tracer_abi_version);
1728         if (ret < 0) {
1729                 goto error_version;
1730         }
1731
1732         ret = modprobe_lttng_data();
1733         if (ret < 0) {
1734                 goto error_modules;
1735         }
1736
1737         ret = kernel_supports_ring_buffer_snapshot_sample_positions();
1738         if (ret < 0) {
1739                 goto error_modules;
1740         }
1741
1742         if (ret < 1) {
1743                 WARN("Kernel tracer does not support buffer monitoring. "
1744                         "The monitoring timer of channels in the kernel domain "
1745                         "will be set to 0 (disabled).");
1746         }
1747
1748         DBG("Kernel tracer fd %d", kernel_tracer_fd);
1749
1750         ret = syscall_init_table(kernel_tracer_fd);
1751         if (ret < 0) {
1752                 ERR("Unable to populate syscall table. Syscall tracing won't "
1753                         "work for this session daemon.");
1754         }
1755         return 0;
1756
1757 error_version:
1758         modprobe_remove_lttng_control();
1759         ret = close(kernel_tracer_fd);
1760         if (ret) {
1761                 PERROR("close");
1762         }
1763         kernel_tracer_fd = -1;
1764         return LTTNG_ERR_KERN_VERSION;
1765
1766 error_modules:
1767         ret = close(kernel_tracer_fd);
1768         if (ret) {
1769                 PERROR("close");
1770         }
1771
1772 error_open:
1773         modprobe_remove_lttng_control();
1774
1775 error:
1776         WARN("No kernel tracer available");
1777         kernel_tracer_fd = -1;
1778         if (!is_root) {
1779                 return LTTNG_ERR_NEED_ROOT_SESSIOND;
1780         } else {
1781                 return LTTNG_ERR_KERN_NA;
1782         }
1783 }
1784
1785 LTTNG_HIDDEN
1786 void cleanup_kernel_tracer(void)
1787 {
1788         int ret;
1789
1790         DBG2("Closing kernel fd");
1791         if (kernel_tracer_fd >= 0) {
1792                 ret = close(kernel_tracer_fd);
1793                 if (ret) {
1794                         PERROR("close");
1795                 }
1796                 kernel_tracer_fd = -1;
1797         }
1798         DBG("Unloading kernel modules");
1799         modprobe_remove_lttng_all();
1800         free(syscall_table);
1801 }
1802
1803 LTTNG_HIDDEN
1804 bool kernel_tracer_is_initialized(void)
1805 {
1806         return kernel_tracer_fd >= 0;
1807 }
1808
1809 /*
1810  *  Clear a kernel session.
1811  *
1812  * Return LTTNG_OK on success or else an LTTng error code.
1813  */
1814 enum lttng_error_code kernel_clear_session(struct ltt_session *session)
1815 {
1816         int ret;
1817         enum lttng_error_code status = LTTNG_OK;
1818         struct consumer_socket *socket;
1819         struct lttng_ht_iter iter;
1820         struct ltt_kernel_session *ksess = session->kernel_session;
1821
1822         assert(ksess);
1823         assert(ksess->consumer);
1824
1825         DBG("Clear kernel session %s (session %" PRIu64 ")",
1826                         session->name, session->id);
1827
1828         rcu_read_lock();
1829
1830         if (ksess->active) {
1831                 ERR("Expecting inactive session %s (%" PRIu64 ")", session->name, session->id);
1832                 status = LTTNG_ERR_FATAL;
1833                 goto end;
1834         }
1835
1836         /*
1837          * Note that this loop will end after one iteration given that there is
1838          * only one kernel consumer.
1839          */
1840         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1841                         socket, node.node) {
1842                 struct ltt_kernel_channel *chan;
1843
1844                 /* For each channel, ask the consumer to clear it. */
1845                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1846                         DBG("Clear kernel channel %" PRIu64 ", session %s",
1847                                         chan->key, session->name);
1848                         ret = consumer_clear_channel(socket, chan->key);
1849                         if (ret < 0) {
1850                                 goto error;
1851                         }
1852                 }
1853
1854                 if (!ksess->metadata) {
1855                         /*
1856                          * Nothing to do for the metadata.
1857                          * This is a snapshot session.
1858                          * The metadata is genererated on the fly.
1859                          */
1860                         continue;
1861                 }
1862
1863                 /*
1864                  * Clear the metadata channel.
1865                  * Metadata channel is not cleared per se but we still need to
1866                  * perform a rotation operation on it behind the scene.
1867                  */
1868                 ret = consumer_clear_channel(socket, ksess->metadata->key);
1869                 if (ret < 0) {
1870                         goto error;
1871                 }
1872         }
1873
1874         goto end;
1875 error:
1876         switch (-ret) {
1877         case LTTCOMM_CONSUMERD_RELAYD_CLEAR_DISALLOWED:
1878               status = LTTNG_ERR_CLEAR_RELAY_DISALLOWED;
1879               break;
1880         default:
1881               status = LTTNG_ERR_CLEAR_FAIL_CONSUMER;
1882               break;
1883         }
1884 end:
1885         rcu_read_unlock();
1886         return status;
1887 }