src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * SPDX-License-Identifier: GPL-2.0-only
   5  *
   6  */
   7
   8 #define _LGPL_SOURCE
   9 #include <fcntl.h>
  10 #include <stdlib.h>
  11 #include <stdio.h>
  12 #include <string.h>
  13 #include <unistd.h>
  14 #include <inttypes.h>
  15 #include <sys/types.h>
  16
  17 #include <common/common.h>
  18 #include <common/hashtable/utils.h>
  19 #include <common/trace-chunk.h>
  20 #include <common/kernel-ctl/kernel-ctl.h>
  21 #include <common/kernel-ctl/kernel-ioctl.h>
  22 #include <common/sessiond-comm/sessiond-comm.h>
  23 #include <common/tracker.h>
  24 #include <common/utils.h>
  25 #include <lttng/event.h>
  26 #include <lttng/lttng-error.h>
  27 #include <lttng/tracker.h>
  28
  29 #include <lttng/userspace-probe.h>
  30 #include <lttng/userspace-probe-internal.h>
  31 #include <lttng/condition/event-rule.h>
  32 #include <lttng/condition/event-rule-internal.h>
  33 #include <lttng/event-rule/event-rule.h>
  34 #include <lttng/event-rule/event-rule-internal.h>
  35 #include <lttng/event-rule/uprobe-internal.h>
  36
  37 #include "lttng-sessiond.h"
  38 #include "lttng-syscall.h"
  39 #include "condition-internal.h"
  40 #include "consumer.h"
  41 #include "kernel.h"
  42 #include "kernel-consumer.h"
  43 #include "kern-modules.h"
  44 #include "utils.h"
  45 #include "rotate.h"
  46 #include "modprobe.h"
  47 #include "tracker.h"
  48 #include "notification-thread-commands.h"
  49
  50 /*
  51  * Key used to reference a channel between the sessiond and the consumer. This
  52  * is only read and updated with the session_list lock held.
  53  */
  54 static uint64_t next_kernel_channel_key;
  55
  56 static const char *module_proc_lttng = "/proc/lttng";
  57
  58 static int kernel_tracer_fd = -1;
  59 static int kernel_tracer_event_notifier_group_fd = -1;
  60 static int kernel_tracer_event_notifier_group_notification_fd = -1;
  61 static struct cds_lfht *kernel_token_to_event_notifier_rule_ht;
  62
  63 /*
  64  * Add context on a kernel channel.
  65  *
  66  * Assumes the ownership of ctx.
  67  */
  68 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  69                 struct ltt_kernel_context *ctx)
  70 {
  71         int ret;
  72
  73         assert(chan);
  74         assert(ctx);
  75
  76         DBG("Adding context to channel %s", chan->channel->name);
  77         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  78         if (ret < 0) {
  79                 switch (-ret) {
  80                 case ENOSYS:
  81                         /* Exists but not available for this kernel */
  82                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  83                         goto error;
  84                 case EEXIST:
  85                         /* If EEXIST, we just ignore the error */
  86                         ret = 0;
  87                         goto end;
  88                 default:
  89                         PERROR("add context ioctl");
  90                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  91                         goto error;
  92                 }
  93         }
  94         ret = 0;
  95
  96 end:
  97         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  98         ctx->in_list = true;
  99         ctx = NULL;
 100 error:
 101         if (ctx) {
 102                 trace_kernel_destroy_context(ctx);
 103         }
 104         return ret;
 105 }
 106
 107 /*
 108  * Create a new kernel session, register it to the kernel tracer and add it to
 109  * the session daemon session.
 110  */
 111 int kernel_create_session(struct ltt_session *session)
 112 {
 113         int ret;
 114         struct ltt_kernel_session *lks;
 115
 116         assert(session);
 117
 118         /* Allocate data structure */
 119         lks = trace_kernel_create_session();
 120         if (lks == NULL) {
 121                 ret = -1;
 122                 goto error;
 123         }
 124
 125         /* Kernel tracer session creation */
 126         ret = kernctl_create_session(kernel_tracer_fd);
 127         if (ret < 0) {
 128                 PERROR("ioctl kernel create session");
 129                 goto error;
 130         }
 131
 132         lks->fd = ret;
 133         /* Prevent fd duplication after execlp() */
 134         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 135         if (ret < 0) {
 136                 PERROR("fcntl session fd");
 137         }
 138
 139         lks->id = session->id;
 140         lks->consumer_fds_sent = 0;
 141         session->kernel_session = lks;
 142
 143         DBG("Kernel session created (fd: %d)", lks->fd);
 144
 145         /*
 146          * This is necessary since the creation time is present in the session
 147          * name when it is generated.
 148          */
 149         if (session->has_auto_generated_name) {
 150                 ret = kernctl_session_set_name(lks->fd, DEFAULT_SESSION_NAME);
 151         } else {
 152                 ret = kernctl_session_set_name(lks->fd, session->name);
 153         }
 154         if (ret) {
 155                 WARN("Could not set kernel session name for session %" PRIu64 " name: %s",
 156                         session->id, session->name);
 157         }
 158
 159         ret = kernctl_session_set_creation_time(lks->fd, session->creation_time);
 160         if (ret) {
 161                 WARN("Could not set kernel session creation time for session %" PRIu64 " name: %s",
 162                         session->id, session->name);
 163         }
 164
 165         return 0;
 166
 167 error:
 168         if (lks) {
 169                 trace_kernel_destroy_session(lks);
 170                 trace_kernel_free_session(lks);
 171         }
 172         return ret;
 173 }
 174
 175 /*
 176  * Create a kernel channel, register it to the kernel tracer and add it to the
 177  * kernel session.
 178  */
 179 int kernel_create_channel(struct ltt_kernel_session *session,
 180                 struct lttng_channel *chan)
 181 {
 182         int ret;
 183         struct ltt_kernel_channel *lkc;
 184
 185         assert(session);
 186         assert(chan);
 187
 188         /* Allocate kernel channel */
 189         lkc = trace_kernel_create_channel(chan);
 190         if (lkc == NULL) {
 191                 goto error;
 192         }
 193
 194         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 195                         chan->name, lkc->channel->attr.overwrite,
 196                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 197                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 198                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 199
 200         /* Kernel tracer channel creation */
 201         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 202         if (ret < 0) {
 203                 PERROR("ioctl kernel create channel");
 204                 goto error;
 205         }
 206
 207         /* Setup the channel fd */
 208         lkc->fd = ret;
 209         /* Prevent fd duplication after execlp() */
 210         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 211         if (ret < 0) {
 212                 PERROR("fcntl session fd");
 213         }
 214
 215         /* Add channel to session */
 216         cds_list_add(&lkc->list, &session->channel_list.head);
 217         session->channel_count++;
 218         lkc->session = session;
 219         lkc->key = ++next_kernel_channel_key;
 220
 221         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 222                         lkc->channel->name, lkc->fd, lkc->key);
 223
 224         return 0;
 225
 226 error:
 227         if (lkc) {
 228                 free(lkc->channel);
 229                 free(lkc);
 230         }
 231         return -1;
 232 }
 233
 234 /*
 235  * Create a kernel event notifier group, register it to the kernel tracer and
 236  * add it to the kernel session.
 237  */
 238 static int kernel_create_event_notifier_group(int *event_notifier_group_fd)
 239 {
 240         int ret;
 241         int local_fd = -1;
 242
 243         assert(event_notifier_group_fd);
 244
 245         /* Kernel event notifier group creation. */
 246         ret = kernctl_create_event_notifier_group(kernel_tracer_fd);
 247         if (ret < 0) {
 248                 PERROR("Failed to create kernel event notifier group");
 249                 ret = -1;
 250                 goto error;
 251         }
 252
 253         local_fd = ret;
 254
 255         /* Prevent fd duplication after execlp(). */
 256         ret = fcntl(local_fd, F_SETFD, FD_CLOEXEC);
 257         if (ret < 0) {
 258                 PERROR("Failed to set FD_CLOEXEC on kernel event notifier group file descriptor: fd = %d",
 259                                 local_fd);
 260                 goto error;
 261         }
 262
 263         DBG("Created kernel event notifier group: fd = %d", local_fd);
 264         *event_notifier_group_fd = local_fd;
 265         local_fd = -1;
 266         ret = 0;
 267 error:
 268         if (local_fd >= 0) {
 269                 ret = close(local_fd);
 270                 if (ret) {
 271                         PERROR("Failed to close kernel event notifier group file descriptor: fd = %d",
 272                                         local_fd);
 273                 }
 274         }
 275
 276         return ret;
 277 }
 278
 279 /*
 280  * Compute the offset of the instrumentation byte in the binary based on the
 281  * function probe location using the ELF lookup method.
 282  *
 283  * Returns 0 on success and set the offset out parameter to the offset of the
 284  * elf symbol
 285  * Returns -1 on error
 286  */
 287 static
 288 int extract_userspace_probe_offset_function_elf(
 289                 const struct lttng_userspace_probe_location *probe_location,
 290                 uid_t uid, gid_t gid, uint64_t *offset)
 291 {
 292         int fd;
 293         int ret = 0;
 294         const char *symbol = NULL;
 295         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 296         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 297
 298         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 299                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
 300
 301         lookup = lttng_userspace_probe_location_get_lookup_method(
 302                         probe_location);
 303         if (!lookup) {
 304                 ret = -1;
 305                 goto end;
 306         }
 307
 308         lookup_method_type =
 309                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 310
 311         assert(lookup_method_type ==
 312                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
 313
 314         symbol = lttng_userspace_probe_location_function_get_function_name(
 315                         probe_location);
 316         if (!symbol) {
 317                 ret = -1;
 318                 goto end;
 319         }
 320
 321         fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
 322         if (fd < 0) {
 323                 ret = -1;
 324                 goto end;
 325         }
 326
 327         ret = run_as_extract_elf_symbol_offset(fd, symbol, uid, gid, offset);
 328         if (ret < 0) {
 329                 DBG("userspace probe offset calculation failed for "
 330                                 "function %s", symbol);
 331                 goto end;
 332         }
 333
 334         DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
 335 end:
 336         return ret;
 337 }
 338
 339 /*
 340  * Compute the offsets of the instrumentation bytes in the binary based on the
 341  * tracepoint probe location using the SDT lookup method. This function
 342  * allocates the offsets buffer, the caller must free it.
 343  *
 344  * Returns 0 on success and set the offset out parameter to the offsets of the
 345  * SDT tracepoint.
 346  * Returns -1 on error.
 347  */
 348 static
 349 int extract_userspace_probe_offset_tracepoint_sdt(
 350                 const struct lttng_userspace_probe_location *probe_location,
 351                 uid_t uid, gid_t gid, uint64_t **offsets,
 352                 uint32_t *offsets_count)
 353 {
 354         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 355         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 356         const char *probe_name = NULL, *provider_name = NULL;
 357         int ret = 0;
 358         int fd, i;
 359
 360         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 361                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
 362
 363         lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
 364         if (!lookup) {
 365                 ret = -1;
 366                 goto end;
 367         }
 368
 369         lookup_method_type =
 370                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 371
 372         assert(lookup_method_type ==
 373                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
 374
 375
 376         probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
 377                         probe_location);
 378         if (!probe_name) {
 379                 ret = -1;
 380                 goto end;
 381         }
 382
 383         provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
 384                         probe_location);
 385         if (!provider_name) {
 386                 ret = -1;
 387                 goto end;
 388         }
 389
 390         fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
 391         if (fd < 0) {
 392                 ret = -1;
 393                 goto end;
 394         }
 395
 396         ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
 397                         uid, gid, offsets, offsets_count);
 398         if (ret < 0) {
 399                 DBG("userspace probe offset calculation failed for sdt "
 400                                 "probe %s:%s", provider_name, probe_name);
 401                 goto end;
 402         }
 403
 404         if (*offsets_count == 0) {
 405                 DBG("no userspace probe offset found");
 406                 goto end;
 407         }
 408
 409         DBG("%u userspace probe SDT offsets found for %s:%s at:",
 410                         *offsets_count, provider_name, probe_name);
 411         for (i = 0; i < *offsets_count; i++) {
 412                 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
 413         }
 414 end:
 415         return ret;
 416 }
 417
 418 static
 419 int userspace_probe_add_callsite(
 420                 const struct lttng_userspace_probe_location *location,
 421                 uid_t uid, gid_t gid, int fd)
 422 {
 423         const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
 424         enum lttng_userspace_probe_location_lookup_method_type type;
 425         int ret;
 426
 427         lookup_method = lttng_userspace_probe_location_get_lookup_method(location);
 428         if (!lookup_method) {
 429                 ret = -1;
 430                 goto end;
 431         }
 432
 433         type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
 434         switch (type) {
 435         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
 436         {
 437                 struct lttng_kernel_event_callsite callsite;
 438                 uint64_t offset;
 439
 440                 ret = extract_userspace_probe_offset_function_elf(location,
 441                                 uid, gid, &offset);
 442                 if (ret) {
 443                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 444                         goto end;
 445                 }
 446
 447                 callsite.u.uprobe.offset = offset;
 448                 ret = kernctl_add_callsite(fd, &callsite);
 449                 if (ret) {
 450                         WARN("Failed to add callsite to ELF userspace probe.");
 451                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 452                         goto end;
 453                 }
 454                 break;
 455         }
 456         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
 457         {
 458                 int i;
 459                 uint64_t *offsets = NULL;
 460                 uint32_t offsets_count;
 461                 struct lttng_kernel_event_callsite callsite;
 462
 463                 /*
 464                  * This call allocates the offsets buffer. This buffer must be freed
 465                  * by the caller
 466                  */
 467                 ret = extract_userspace_probe_offset_tracepoint_sdt(location,
 468                                 uid, gid, &offsets, &offsets_count);
 469                 if (ret) {
 470                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 471                         goto end;
 472                 }
 473                 for (i = 0; i < offsets_count; i++) {
 474                         callsite.u.uprobe.offset = offsets[i];
 475                         ret = kernctl_add_callsite(fd, &callsite);
 476                         if (ret) {
 477                                 WARN("Failed to add callsite to SDT userspace probe");
 478                                 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 479                                 free(offsets);
 480                                 goto end;
 481                         }
 482                 }
 483                 free(offsets);
 484                 break;
 485         }
 486         default:
 487                 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 488                 goto end;
 489         }
 490 end:
 491         return ret;
 492 }
 493
 494 /*
 495  * Extract the offsets of the instrumentation point for the different lookup
 496  * methods.
 497  */
 498 static
 499 int userspace_probe_event_add_callsites(struct lttng_event *ev,
 500                         struct ltt_kernel_session *session, int fd)
 501 {
 502         int ret;
 503         const struct lttng_userspace_probe_location *location = NULL;
 504
 505         assert(ev);
 506         assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
 507
 508         location = lttng_event_get_userspace_probe_location(ev);
 509         if (!location) {
 510                 ret = -1;
 511                 goto end;
 512         }
 513
 514         ret = userspace_probe_add_callsite(location, session->uid, session->gid,
 515                 fd);
 516         if (ret) {
 517                 WARN("Failed to add callsite to userspace probe event '%s'",
 518                                 ev->name);
 519         }
 520
 521 end:
 522         return ret;
 523 }
 524
 525 /*
 526  * Extract the offsets of the instrumentation point for the different look-up
 527  * methods.
 528  */
 529 static int userspace_probe_event_rule_add_callsites(
 530                 const struct lttng_event_rule *rule,
 531                 const struct lttng_credentials *creds,
 532                 int fd)
 533 {
 534         int ret;
 535         enum lttng_event_rule_status status;
 536         enum lttng_event_rule_type event_rule_type;
 537         const struct lttng_userspace_probe_location *location = NULL;
 538
 539         assert(rule);
 540         assert(creds);
 541
 542         event_rule_type = lttng_event_rule_get_type(rule);
 543         assert(event_rule_type == LTTNG_EVENT_RULE_TYPE_UPROBE);
 544
 545         status = lttng_event_rule_uprobe_get_location(rule, &location);
 546         if (status != LTTNG_EVENT_RULE_STATUS_OK || !location) {
 547                 ret = -1;
 548                 goto end;
 549         }
 550
 551         ret = userspace_probe_add_callsite(location,
 552                         lttng_credentials_get_uid(creds),
 553                         lttng_credentials_get_gid(creds), fd);
 554         if (ret) {
 555                 WARN("Failed to add callsite to user space probe object: fd = %d",
 556                                 fd);
 557         }
 558
 559 end:
 560         return ret;
 561 }
 562
 563 /*
 564  * Create a kernel event, enable it to the kernel tracer and add it to the
 565  * channel event list of the kernel session.
 566  * We own filter_expression and filter.
 567  */
 568 int kernel_create_event(struct lttng_event *ev,
 569                 struct ltt_kernel_channel *channel,
 570                 char *filter_expression,
 571                 struct lttng_filter_bytecode *filter)
 572 {
 573         int err, fd;
 574         enum lttng_error_code ret;
 575         struct ltt_kernel_event *event;
 576
 577         assert(ev);
 578         assert(channel);
 579
 580         /* We pass ownership of filter_expression and filter */
 581         ret = trace_kernel_create_event(ev, filter_expression,
 582                         filter, &event);
 583         if (ret != LTTNG_OK) {
 584                 goto error;
 585         }
 586
 587         fd = kernctl_create_event(channel->fd, event->event);
 588         if (fd < 0) {
 589                 switch (-fd) {
 590                 case EEXIST:
 591                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 592                         break;
 593                 case ENOSYS:
 594                         WARN("Event type not implemented");
 595                         ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
 596                         break;
 597                 case ENOENT:
 598                         WARN("Event %s not found!", ev->name);
 599                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 600                         break;
 601                 default:
 602                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 603                         PERROR("create event ioctl");
 604                 }
 605                 goto free_event;
 606         }
 607
 608         event->type = ev->type;
 609         event->fd = fd;
 610         /* Prevent fd duplication after execlp() */
 611         err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 612         if (err < 0) {
 613                 PERROR("fcntl session fd");
 614         }
 615
 616         if (filter) {
 617                 err = kernctl_filter(event->fd, filter);
 618                 if (err < 0) {
 619                         switch (-err) {
 620                         case ENOMEM:
 621                                 ret = LTTNG_ERR_FILTER_NOMEM;
 622                                 break;
 623                         default:
 624                                 ret = LTTNG_ERR_FILTER_INVAL;
 625                                 break;
 626                         }
 627                         goto filter_error;
 628                 }
 629         }
 630
 631         if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
 632                 ret = userspace_probe_event_add_callsites(ev, channel->session,
 633                                 event->fd);
 634                 if (ret) {
 635                         goto add_callsite_error;
 636                 }
 637         }
 638
 639         err = kernctl_enable(event->fd);
 640         if (err < 0) {
 641                 switch (-err) {
 642                 case EEXIST:
 643                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 644                         break;
 645                 default:
 646                         PERROR("enable kernel event");
 647                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 648                         break;
 649                 }
 650                 goto enable_error;
 651         }
 652
 653         /* Add event to event list */
 654         cds_list_add(&event->list, &channel->events_list.head);
 655         channel->event_count++;
 656
 657         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 658
 659         return 0;
 660
 661 add_callsite_error:
 662 enable_error:
 663 filter_error:
 664         {
 665                 int closeret;
 666
 667                 closeret = close(event->fd);
 668                 if (closeret) {
 669                         PERROR("close event fd");
 670                 }
 671         }
 672 free_event:
 673         free(event);
 674 error:
 675         return ret;
 676 }
 677
 678 /*
 679  * Disable a kernel channel.
 680  */
 681 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 682 {
 683         int ret;
 684
 685         assert(chan);
 686
 687         ret = kernctl_disable(chan->fd);
 688         if (ret < 0) {
 689                 PERROR("disable chan ioctl");
 690                 goto error;
 691         }
 692
 693         chan->enabled = 0;
 694         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 695                         chan->channel->name, chan->fd, chan->key);
 696
 697         return 0;
 698
 699 error:
 700         return ret;
 701 }
 702
 703 /*
 704  * Enable a kernel channel.
 705  */
 706 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 707 {
 708         int ret;
 709
 710         assert(chan);
 711
 712         ret = kernctl_enable(chan->fd);
 713         if (ret < 0 && ret != -EEXIST) {
 714                 PERROR("Enable kernel chan");
 715                 goto error;
 716         }
 717
 718         chan->enabled = 1;
 719         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 720                         chan->channel->name, chan->fd, chan->key);
 721
 722         return 0;
 723
 724 error:
 725         return ret;
 726 }
 727
 728 /*
 729  * Enable a kernel event.
 730  */
 731 int kernel_enable_event(struct ltt_kernel_event *event)
 732 {
 733         int ret;
 734
 735         assert(event);
 736
 737         ret = kernctl_enable(event->fd);
 738         if (ret < 0) {
 739                 switch (-ret) {
 740                 case EEXIST:
 741                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 742                         break;
 743                 default:
 744                         PERROR("enable kernel event");
 745                         break;
 746                 }
 747                 goto error;
 748         }
 749
 750         event->enabled = 1;
 751         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 752
 753         return 0;
 754
 755 error:
 756         return ret;
 757 }
 758
 759 /*
 760  * Disable a kernel event.
 761  */
 762 int kernel_disable_event(struct ltt_kernel_event *event)
 763 {
 764         int ret;
 765
 766         assert(event);
 767
 768         ret = kernctl_disable(event->fd);
 769         if (ret < 0) {
 770                 switch (-ret) {
 771                 case EEXIST:
 772                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 773                         break;
 774                 default:
 775                         PERROR("disable kernel event");
 776                         break;
 777                 }
 778                 goto error;
 779         }
 780
 781         event->enabled = 0;
 782         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 783
 784         return 0;
 785
 786 error:
 787         return ret;
 788 }
 789
 790 /*
 791  * Disable a kernel event notifier.
 792  */
 793 static
 794 int kernel_disable_event_notifier_rule(struct ltt_kernel_event_notifier_rule *event)
 795 {
 796         int ret;
 797
 798         assert(event);
 799
 800         rcu_read_lock();
 801         cds_lfht_del(kernel_token_to_event_notifier_rule_ht, &event->ht_node);
 802         rcu_read_unlock();
 803
 804         ret = kernctl_disable(event->fd);
 805         if (ret < 0) {
 806                 switch (-ret) {
 807                 case EEXIST:
 808                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 809                         break;
 810                 default:
 811                         PERROR("Failed to disable kernel event notifier: fd = %d, token = %" PRIu64,
 812                                         event->fd, event->token);
 813                         break;
 814                 }
 815                 goto error;
 816         }
 817
 818         event->enabled = 0;
 819         DBG("Disabled kernel event notifier: fd = %d, token = %" PRIu64,
 820                         event->fd, event->token);
 821
 822 error:
 823         return ret;
 824 }
 825
 826 static
 827 struct process_attr_tracker *_kernel_get_process_attr_tracker(
 828                 struct ltt_kernel_session *session,
 829                 enum lttng_process_attr process_attr)
 830 {
 831         switch (process_attr) {
 832         case LTTNG_PROCESS_ATTR_PROCESS_ID:
 833                 return session->tracker_pid;
 834         case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
 835                 return session->tracker_vpid;
 836         case LTTNG_PROCESS_ATTR_USER_ID:
 837                 return session->tracker_uid;
 838         case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
 839                 return session->tracker_vuid;
 840         case LTTNG_PROCESS_ATTR_GROUP_ID:
 841                 return session->tracker_gid;
 842         case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
 843                 return session->tracker_vgid;
 844         default:
 845                 return NULL;
 846         }
 847 }
 848
 849 const struct process_attr_tracker *kernel_get_process_attr_tracker(
 850                 struct ltt_kernel_session *session,
 851                 enum lttng_process_attr process_attr)
 852 {
 853         return (const struct process_attr_tracker *)
 854                         _kernel_get_process_attr_tracker(session, process_attr);
 855 }
 856
 857 enum lttng_error_code kernel_process_attr_tracker_set_tracking_policy(
 858                 struct ltt_kernel_session *session,
 859                 enum lttng_process_attr process_attr,
 860                 enum lttng_tracking_policy policy)
 861 {
 862         int ret;
 863         enum lttng_error_code ret_code = LTTNG_OK;
 864         struct process_attr_tracker *tracker =
 865                         _kernel_get_process_attr_tracker(session, process_attr);
 866         enum lttng_tracking_policy previous_policy;
 867
 868         if (!tracker) {
 869                 ret_code = LTTNG_ERR_INVALID;
 870                 goto end;
 871         }
 872
 873         previous_policy = process_attr_tracker_get_tracking_policy(tracker);
 874         ret = process_attr_tracker_set_tracking_policy(tracker, policy);
 875         if (ret) {
 876                 ret_code = LTTNG_ERR_UNK;
 877                 goto end;
 878         }
 879
 880         if (previous_policy == policy) {
 881                 goto end;
 882         }
 883
 884         switch (policy) {
 885         case LTTNG_TRACKING_POLICY_INCLUDE_ALL:
 886                 if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
 887                         /*
 888                          * Maintain a special case for the process ID process
 889                          * attribute tracker as it was the only supported
 890                          * attribute prior to 2.12.
 891                          */
 892                         ret = kernctl_track_pid(session->fd, -1);
 893                 } else {
 894                         ret = kernctl_track_id(session->fd, process_attr, -1);
 895                 }
 896                 break;
 897         case LTTNG_TRACKING_POLICY_EXCLUDE_ALL:
 898         case LTTNG_TRACKING_POLICY_INCLUDE_SET:
 899                 /* fall-through. */
 900                 if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
 901                         /*
 902                          * Maintain a special case for the process ID process
 903                          * attribute tracker as it was the only supported
 904                          * attribute prior to 2.12.
 905                          */
 906                         ret = kernctl_untrack_pid(session->fd, -1);
 907                 } else {
 908                         ret = kernctl_untrack_id(session->fd, process_attr, -1);
 909                 }
 910                 break;
 911         default:
 912                 abort();
 913         }
 914         /* kern-ctl error handling */
 915         switch (-ret) {
 916         case 0:
 917                 ret_code = LTTNG_OK;
 918                 break;
 919         case EINVAL:
 920                 ret_code = LTTNG_ERR_INVALID;
 921                 break;
 922         case ENOMEM:
 923                 ret_code = LTTNG_ERR_NOMEM;
 924                 break;
 925         case EEXIST:
 926                 ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
 927                 break;
 928         default:
 929                 ret_code = LTTNG_ERR_UNK;
 930                 break;
 931         }
 932 end:
 933         return ret_code;
 934 }
 935
 936 enum lttng_error_code kernel_process_attr_tracker_inclusion_set_add_value(
 937                 struct ltt_kernel_session *session,
 938                 enum lttng_process_attr process_attr,
 939                 const struct process_attr_value *value)
 940 {
 941         int ret, integral_value;
 942         enum lttng_error_code ret_code;
 943         struct process_attr_tracker *tracker;
 944         enum process_attr_tracker_status status;
 945
 946         /*
 947          * Convert process attribute tracker value to the integral
 948          * representation required by the kern-ctl API.
 949          */
 950         switch (process_attr) {
 951         case LTTNG_PROCESS_ATTR_PROCESS_ID:
 952         case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
 953                 integral_value = (int) value->value.pid;
 954                 break;
 955         case LTTNG_PROCESS_ATTR_USER_ID:
 956         case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
 957                 if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_USER_NAME) {
 958                         uid_t uid;
 959
 960                         ret_code = utils_user_id_from_name(
 961                                         value->value.user_name, &uid);
 962                         if (ret_code != LTTNG_OK) {
 963                                 goto end;
 964                         }
 965                         integral_value = (int) uid;
 966                 } else {
 967                         integral_value = (int) value->value.uid;
 968                 }
 969                 break;
 970         case LTTNG_PROCESS_ATTR_GROUP_ID:
 971         case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
 972                 if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_GROUP_NAME) {
 973                         gid_t gid;
 974
 975                         ret_code = utils_group_id_from_name(
 976                                         value->value.group_name, &gid);
 977                         if (ret_code != LTTNG_OK) {
 978                                 goto end;
 979                         }
 980                         integral_value = (int) gid;
 981                 } else {
 982                         integral_value = (int) value->value.gid;
 983                 }
 984                 break;
 985         default:
 986                 ret_code = LTTNG_ERR_INVALID;
 987                 goto end;
 988         }
 989
 990         tracker = _kernel_get_process_attr_tracker(session, process_attr);
 991         if (!tracker) {
 992                 ret_code = LTTNG_ERR_INVALID;
 993                 goto end;
 994         }
 995
 996         status = process_attr_tracker_inclusion_set_add_value(tracker, value);
 997         if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
 998                 switch (status) {
 999                 case PROCESS_ATTR_TRACKER_STATUS_EXISTS:
1000                         ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
1001                         break;
1002                 case PROCESS_ATTR_TRACKER_STATUS_INVALID_TRACKING_POLICY:
1003                         ret_code = LTTNG_ERR_PROCESS_ATTR_TRACKER_INVALID_TRACKING_POLICY;
1004                         break;
1005                 case PROCESS_ATTR_TRACKER_STATUS_ERROR:
1006                 default:
1007                         ret_code = LTTNG_ERR_UNK;
1008                         break;
1009                 }
1010                 goto end;
1011         }
1012
1013         DBG("Kernel track %s %d for session id %" PRIu64,
1014                         lttng_process_attr_to_string(process_attr),
1015                         integral_value, session->id);
1016         if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
1017                 /*
1018                  * Maintain a special case for the process ID process attribute
1019                  * tracker as it was the only supported attribute prior to 2.12.
1020                  */
1021                 ret = kernctl_track_pid(session->fd, integral_value);
1022         } else {
1023                 ret = kernctl_track_id(
1024                                 session->fd, process_attr, integral_value);
1025         }
1026         if (ret == 0) {
1027                 ret_code = LTTNG_OK;
1028                 goto end;
1029         }
1030
1031         kernel_wait_quiescent();
1032
1033         /* kern-ctl error handling */
1034         switch (-ret) {
1035         case 0:
1036                 ret_code = LTTNG_OK;
1037                 break;
1038         case EINVAL:
1039                 ret_code = LTTNG_ERR_INVALID;
1040                 break;
1041         case ENOMEM:
1042                 ret_code = LTTNG_ERR_NOMEM;
1043                 break;
1044         case EEXIST:
1045                 ret_code = LTTNG_ERR_PROCESS_ATTR_EXISTS;
1046                 break;
1047         default:
1048                 ret_code = LTTNG_ERR_UNK;
1049                 break;
1050         }
1051
1052         /* Attempt to remove the value from the tracker. */
1053         status = process_attr_tracker_inclusion_set_remove_value(
1054                         tracker, value);
1055         if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
1056                 ERR("Failed to roll-back the tracking of kernel %s process attribute %d while handling a kern-ctl error",
1057                                 lttng_process_attr_to_string(process_attr),
1058                                 integral_value);
1059         }
1060 end:
1061         return ret_code;
1062 }
1063
1064 enum lttng_error_code kernel_process_attr_tracker_inclusion_set_remove_value(
1065                 struct ltt_kernel_session *session,
1066                 enum lttng_process_attr process_attr,
1067                 const struct process_attr_value *value)
1068 {
1069         int ret, integral_value;
1070         enum lttng_error_code ret_code;
1071         struct process_attr_tracker *tracker;
1072         enum process_attr_tracker_status status;
1073
1074         /*
1075          * Convert process attribute tracker value to the integral
1076          * representation required by the kern-ctl API.
1077          */
1078         switch (process_attr) {
1079         case LTTNG_PROCESS_ATTR_PROCESS_ID:
1080         case LTTNG_PROCESS_ATTR_VIRTUAL_PROCESS_ID:
1081                 integral_value = (int) value->value.pid;
1082                 break;
1083         case LTTNG_PROCESS_ATTR_USER_ID:
1084         case LTTNG_PROCESS_ATTR_VIRTUAL_USER_ID:
1085                 if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_USER_NAME) {
1086                         uid_t uid;
1087
1088                         ret_code = utils_user_id_from_name(
1089                                         value->value.user_name, &uid);
1090                         if (ret_code != LTTNG_OK) {
1091                                 goto end;
1092                         }
1093                         integral_value = (int) uid;
1094                 } else {
1095                         integral_value = (int) value->value.uid;
1096                 }
1097                 break;
1098         case LTTNG_PROCESS_ATTR_GROUP_ID:
1099         case LTTNG_PROCESS_ATTR_VIRTUAL_GROUP_ID:
1100                 if (value->type == LTTNG_PROCESS_ATTR_VALUE_TYPE_GROUP_NAME) {
1101                         gid_t gid;
1102
1103                         ret_code = utils_group_id_from_name(
1104                                         value->value.group_name, &gid);
1105                         if (ret_code != LTTNG_OK) {
1106                                 goto end;
1107                         }
1108                         integral_value = (int) gid;
1109                 } else {
1110                         integral_value = (int) value->value.gid;
1111                 }
1112                 break;
1113         default:
1114                 ret_code = LTTNG_ERR_INVALID;
1115                 goto end;
1116         }
1117
1118         tracker = _kernel_get_process_attr_tracker(session, process_attr);
1119         if (!tracker) {
1120                 ret_code = LTTNG_ERR_INVALID;
1121                 goto end;
1122         }
1123
1124         status = process_attr_tracker_inclusion_set_remove_value(
1125                         tracker, value);
1126         if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
1127                 switch (status) {
1128                 case PROCESS_ATTR_TRACKER_STATUS_MISSING:
1129                         ret_code = LTTNG_ERR_PROCESS_ATTR_MISSING;
1130                         break;
1131                 case PROCESS_ATTR_TRACKER_STATUS_INVALID_TRACKING_POLICY:
1132                         ret_code = LTTNG_ERR_PROCESS_ATTR_TRACKER_INVALID_TRACKING_POLICY;
1133                         break;
1134                 case PROCESS_ATTR_TRACKER_STATUS_ERROR:
1135                 default:
1136                         ret_code = LTTNG_ERR_UNK;
1137                         break;
1138                 }
1139                 goto end;
1140         }
1141
1142         DBG("Kernel track %s %d for session id %" PRIu64,
1143                         lttng_process_attr_to_string(process_attr),
1144                         integral_value, session->id);
1145         if (process_attr == LTTNG_PROCESS_ATTR_PROCESS_ID) {
1146                 /*
1147                  * Maintain a special case for the process ID process attribute
1148                  * tracker as it was the only supported attribute prior to 2.12.
1149                  */
1150                 ret = kernctl_untrack_pid(session->fd, integral_value);
1151         } else {
1152                 ret = kernctl_untrack_id(
1153                                 session->fd, process_attr, integral_value);
1154         }
1155         if (ret == 0) {
1156                 ret_code = LTTNG_OK;
1157                 goto end;
1158         }
1159         kernel_wait_quiescent();
1160
1161         /* kern-ctl error handling */
1162         switch (-ret) {
1163         case 0:
1164                 ret_code = LTTNG_OK;
1165                 break;
1166         case EINVAL:
1167                 ret_code = LTTNG_ERR_INVALID;
1168                 break;
1169         case ENOMEM:
1170                 ret_code = LTTNG_ERR_NOMEM;
1171                 break;
1172         case ENOENT:
1173                 ret_code = LTTNG_ERR_PROCESS_ATTR_MISSING;
1174                 break;
1175         default:
1176                 ret_code = LTTNG_ERR_UNK;
1177                 break;
1178         }
1179
1180         /* Attempt to add the value to the tracker. */
1181         status = process_attr_tracker_inclusion_set_add_value(
1182                         tracker, value);
1183         if (status != PROCESS_ATTR_TRACKER_STATUS_OK) {
1184                 ERR("Failed to roll-back the tracking of kernel %s process attribute %d while handling a kern-ctl error",
1185                                 lttng_process_attr_to_string(process_attr),
1186                                 integral_value);
1187         }
1188 end:
1189         return ret_code;
1190 }
1191
1192 /*
1193  * Create kernel metadata, open from the kernel tracer and add it to the
1194  * kernel session.
1195  */
1196 int kernel_open_metadata(struct ltt_kernel_session *session)
1197 {
1198         int ret;
1199         struct ltt_kernel_metadata *lkm = NULL;
1200
1201         assert(session);
1202
1203         /* Allocate kernel metadata */
1204         lkm = trace_kernel_create_metadata();
1205         if (lkm == NULL) {
1206                 goto error;
1207         }
1208
1209         /* Kernel tracer metadata creation */
1210         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
1211         if (ret < 0) {
1212                 goto error_open;
1213         }
1214
1215         lkm->fd = ret;
1216         lkm->key = ++next_kernel_channel_key;
1217         /* Prevent fd duplication after execlp() */
1218         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
1219         if (ret < 0) {
1220                 PERROR("fcntl session fd");
1221         }
1222
1223         session->metadata = lkm;
1224
1225         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
1226
1227         return 0;
1228
1229 error_open:
1230         trace_kernel_destroy_metadata(lkm);
1231 error:
1232         return -1;
1233 }
1234
1235 /*
1236  * Start tracing session.
1237  */
1238 int kernel_start_session(struct ltt_kernel_session *session)
1239 {
1240         int ret;
1241
1242         assert(session);
1243
1244         ret = kernctl_start_session(session->fd);
1245         if (ret < 0) {
1246                 PERROR("ioctl start session");
1247                 goto error;
1248         }
1249
1250         DBG("Kernel session started");
1251
1252         return 0;
1253
1254 error:
1255         return ret;
1256 }
1257
1258 /*
1259  * Make a kernel wait to make sure in-flight probe have completed.
1260  */
1261 void kernel_wait_quiescent(void)
1262 {
1263         int ret;
1264         int fd = kernel_tracer_fd;
1265
1266         DBG("Kernel quiescent wait on %d", fd);
1267
1268         ret = kernctl_wait_quiescent(fd);
1269         if (ret < 0) {
1270                 PERROR("wait quiescent ioctl");
1271                 ERR("Kernel quiescent wait failed");
1272         }
1273 }
1274
1275 /*
1276  *  Force flush buffer of metadata.
1277  */
1278 int kernel_metadata_flush_buffer(int fd)
1279 {
1280         int ret;
1281
1282         DBG("Kernel flushing metadata buffer on fd %d", fd);
1283
1284         ret = kernctl_buffer_flush(fd);
1285         if (ret < 0) {
1286                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
1287         }
1288
1289         return 0;
1290 }
1291
1292 /*
1293  * Force flush buffer for channel.
1294  */
1295 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
1296 {
1297         int ret;
1298         struct ltt_kernel_stream *stream;
1299
1300         assert(channel);
1301
1302         DBG("Flush buffer for channel %s", channel->channel->name);
1303
1304         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
1305                 DBG("Flushing channel stream %d", stream->fd);
1306                 ret = kernctl_buffer_flush(stream->fd);
1307                 if (ret < 0) {
1308                         PERROR("ioctl");
1309                         ERR("Fail to flush buffer for stream %d (ret: %d)",
1310                                         stream->fd, ret);
1311                 }
1312         }
1313
1314         return 0;
1315 }
1316
1317 /*
1318  * Stop tracing session.
1319  */
1320 int kernel_stop_session(struct ltt_kernel_session *session)
1321 {
1322         int ret;
1323
1324         assert(session);
1325
1326         ret = kernctl_stop_session(session->fd);
1327         if (ret < 0) {
1328                 goto error;
1329         }
1330
1331         DBG("Kernel session stopped");
1332
1333         return 0;
1334
1335 error:
1336         return ret;
1337 }
1338
1339 /*
1340  * Open stream of channel, register it to the kernel tracer and add it
1341  * to the stream list of the channel.
1342  *
1343  * Note: given that the streams may appear in random order wrt CPU
1344  * number (e.g. cpu hotplug), the index value of the stream number in
1345  * the stream name is not necessarily linked to the CPU number.
1346  *
1347  * Return the number of created stream. Else, a negative value.
1348  */
1349 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
1350 {
1351         int ret;
1352         struct ltt_kernel_stream *lks;
1353
1354         assert(channel);
1355
1356         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
1357                 lks = trace_kernel_create_stream(channel->channel->name,
1358                                 channel->stream_count);
1359                 if (lks == NULL) {
1360                         ret = close(ret);
1361                         if (ret) {
1362                                 PERROR("close");
1363                         }
1364                         goto error;
1365                 }
1366
1367                 lks->fd = ret;
1368                 /* Prevent fd duplication after execlp() */
1369                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
1370                 if (ret < 0) {
1371                         PERROR("fcntl session fd");
1372                 }
1373
1374                 lks->tracefile_size = channel->channel->attr.tracefile_size;
1375                 lks->tracefile_count = channel->channel->attr.tracefile_count;
1376
1377                 /* Add stream to channel stream list */
1378                 cds_list_add(&lks->list, &channel->stream_list.head);
1379                 channel->stream_count++;
1380
1381                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
1382                                 lks->state);
1383         }
1384
1385         return channel->stream_count;
1386
1387 error:
1388         return -1;
1389 }
1390
1391 /*
1392  * Open the metadata stream and set it to the kernel session.
1393  */
1394 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
1395 {
1396         int ret;
1397
1398         assert(session);
1399
1400         ret = kernctl_create_stream(session->metadata->fd);
1401         if (ret < 0) {
1402                 PERROR("kernel create metadata stream");
1403                 goto error;
1404         }
1405
1406         DBG("Kernel metadata stream created (fd: %d)", ret);
1407         session->metadata_stream_fd = ret;
1408         /* Prevent fd duplication after execlp() */
1409         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
1410         if (ret < 0) {
1411                 PERROR("fcntl session fd");
1412         }
1413
1414         return 0;
1415
1416 error:
1417         return -1;
1418 }
1419
1420 /*
1421  * Get the event list from the kernel tracer and return the number of elements.
1422  */
1423 ssize_t kernel_list_events(struct lttng_event **events)
1424 {
1425         int fd, ret;
1426         char *event;
1427         size_t nbmem, count = 0;
1428         FILE *fp;
1429         struct lttng_event *elist;
1430
1431         assert(events);
1432
1433         fd = kernctl_tracepoint_list(kernel_tracer_fd);
1434         if (fd < 0) {
1435                 PERROR("kernel tracepoint list");
1436                 goto error;
1437         }
1438
1439         fp = fdopen(fd, "r");
1440         if (fp == NULL) {
1441                 PERROR("kernel tracepoint list fdopen");
1442                 goto error_fp;
1443         }
1444
1445         /*
1446          * Init memory size counter
1447          * See kernel-ctl.h for explanation of this value
1448          */
1449         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1450         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1451         if (elist == NULL) {
1452                 PERROR("alloc list events");
1453                 count = -ENOMEM;
1454                 goto end;
1455         }
1456
1457         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1458                 if (count >= nbmem) {
1459                         struct lttng_event *new_elist;
1460                         size_t new_nbmem;
1461
1462                         new_nbmem = nbmem << 1;
1463                         DBG("Reallocating event list from %zu to %zu bytes",
1464                                         nbmem, new_nbmem);
1465                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1466                         if (new_elist == NULL) {
1467                                 PERROR("realloc list events");
1468                                 free(event);
1469                                 free(elist);
1470                                 count = -ENOMEM;
1471                                 goto end;
1472                         }
1473                         /* Zero the new memory */
1474                         memset(new_elist + nbmem, 0,
1475                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1476                         nbmem = new_nbmem;
1477                         elist = new_elist;
1478                 }
1479                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1480                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1481                 elist[count].enabled = -1;
1482                 count++;
1483                 free(event);
1484         }
1485
1486         *events = elist;
1487         DBG("Kernel list events done (%zu events)", count);
1488 end:
1489         ret = fclose(fp);       /* closes both fp and fd */
1490         if (ret) {
1491                 PERROR("fclose");
1492         }
1493         return count;
1494
1495 error_fp:
1496         ret = close(fd);
1497         if (ret) {
1498                 PERROR("close");
1499         }
1500 error:
1501         return -1;
1502 }
1503
1504 /*
1505  * Get kernel version and validate it.
1506  */
1507 int kernel_validate_version(struct lttng_kernel_tracer_version *version,
1508                 struct lttng_kernel_tracer_abi_version *abi_version)
1509 {
1510         int ret;
1511
1512         ret = kernctl_tracer_version(kernel_tracer_fd, version);
1513         if (ret < 0) {
1514                 ERR("Failed to retrieve the lttng-modules version");
1515                 goto error;
1516         }
1517
1518         /* Validate version */
1519         if (version->major != VERSION_MAJOR) {
1520                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1521                         version->major, VERSION_MAJOR);
1522                 goto error_version;
1523         }
1524         ret = kernctl_tracer_abi_version(kernel_tracer_fd, abi_version);
1525         if (ret < 0) {
1526                 ERR("Failed to retrieve lttng-modules ABI version");
1527                 goto error;
1528         }
1529         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1530                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1531                         abi_version->major, abi_version->minor,
1532                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1533                 goto error;
1534         }
1535         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1536                         version->major, version->minor,
1537                         abi_version->major, abi_version->minor);
1538         return 0;
1539
1540 error_version:
1541         ret = -1;
1542
1543 error:
1544         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1545         return ret;
1546 }
1547
1548 /*
1549  * Kernel work-arounds called at the start of sessiond main().
1550  */
1551 int init_kernel_workarounds(void)
1552 {
1553         int ret;
1554         FILE *fp;
1555
1556         /*
1557          * boot_id needs to be read once before being used concurrently
1558          * to deal with a Linux kernel race. A fix is proposed for
1559          * upstream, but the work-around is needed for older kernels.
1560          */
1561         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1562         if (!fp) {
1563                 goto end_boot_id;
1564         }
1565         while (!feof(fp)) {
1566                 char buf[37] = "";
1567
1568                 ret = fread(buf, 1, sizeof(buf), fp);
1569                 if (ret < 0) {
1570                         /* Ignore error, we don't really care */
1571                 }
1572         }
1573         ret = fclose(fp);
1574         if (ret) {
1575                 PERROR("fclose");
1576         }
1577 end_boot_id:
1578         return 0;
1579 }
1580
1581 /*
1582  * Teardown of a kernel session, keeping data required by destroy notifiers.
1583  */
1584 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1585 {
1586         struct lttng_trace_chunk *trace_chunk;
1587
1588         if (ksess == NULL) {
1589                 DBG3("No kernel session when tearing down session");
1590                 return;
1591         }
1592
1593         DBG("Tearing down kernel session");
1594         trace_chunk = ksess->current_trace_chunk;
1595
1596         /*
1597          * Destroy channels on the consumer if at least one FD has been sent and we
1598          * are in no output mode because the streams are in *no* monitor mode so we
1599          * have to send a command to clean them up or else they leaked.
1600          */
1601         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1602                 int ret;
1603                 struct consumer_socket *socket;
1604                 struct lttng_ht_iter iter;
1605
1606                 /* For each consumer socket. */
1607                 rcu_read_lock();
1608                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1609                                 socket, node.node) {
1610                         struct ltt_kernel_channel *chan;
1611
1612                         /* For each channel, ask the consumer to destroy it. */
1613                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1614                                 ret = kernel_consumer_destroy_channel(socket, chan);
1615                                 if (ret < 0) {
1616                                         /* Consumer is probably dead. Use next socket. */
1617                                         continue;
1618                                 }
1619                         }
1620                 }
1621                 rcu_read_unlock();
1622         }
1623
1624         /* Close any relayd session */
1625         consumer_output_send_destroy_relayd(ksess->consumer);
1626
1627         trace_kernel_destroy_session(ksess);
1628         lttng_trace_chunk_put(trace_chunk);
1629 }
1630
1631 /* Teardown of data required by destroy notifiers. */
1632 void kernel_free_session(struct ltt_kernel_session *ksess)
1633 {
1634         if (ksess == NULL) {
1635                 return;
1636         }
1637         trace_kernel_free_session(ksess);
1638 }
1639
1640 /*
1641  * Destroy a kernel channel object. It does not do anything on the tracer side.
1642  */
1643 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1644 {
1645         struct ltt_kernel_session *ksess = NULL;
1646
1647         assert(kchan);
1648         assert(kchan->channel);
1649
1650         DBG3("Kernel destroy channel %s", kchan->channel->name);
1651
1652         /* Update channel count of associated session. */
1653         if (kchan->session) {
1654                 /* Keep pointer reference so we can update it after the destroy. */
1655                 ksess = kchan->session;
1656         }
1657
1658         trace_kernel_destroy_channel(kchan);
1659
1660         /*
1661          * At this point the kernel channel is not visible anymore. This is safe
1662          * since in order to work on a visible kernel session, the tracing session
1663          * lock (ltt_session.lock) MUST be acquired.
1664          */
1665         if (ksess) {
1666                 ksess->channel_count--;
1667         }
1668 }
1669
1670 /*
1671  * Take a snapshot for a given kernel session.
1672  *
1673  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1674  */
1675 enum lttng_error_code kernel_snapshot_record(
1676                 struct ltt_kernel_session *ksess,
1677                 const struct consumer_output *output, int wait,
1678                 uint64_t nb_packets_per_stream)
1679 {
1680         int err, ret, saved_metadata_fd;
1681         enum lttng_error_code status = LTTNG_OK;
1682         struct consumer_socket *socket;
1683         struct lttng_ht_iter iter;
1684         struct ltt_kernel_metadata *saved_metadata;
1685         char *trace_path = NULL;
1686         size_t consumer_path_offset = 0;
1687
1688         assert(ksess);
1689         assert(ksess->consumer);
1690         assert(output);
1691
1692         DBG("Kernel snapshot record started");
1693
1694         /* Save current metadata since the following calls will change it. */
1695         saved_metadata = ksess->metadata;
1696         saved_metadata_fd = ksess->metadata_stream_fd;
1697
1698         rcu_read_lock();
1699
1700         ret = kernel_open_metadata(ksess);
1701         if (ret < 0) {
1702                 status = LTTNG_ERR_KERN_META_FAIL;
1703                 goto error;
1704         }
1705
1706         ret = kernel_open_metadata_stream(ksess);
1707         if (ret < 0) {
1708                 status = LTTNG_ERR_KERN_META_FAIL;
1709                 goto error_open_stream;
1710         }
1711
1712         trace_path = setup_channel_trace_path(ksess->consumer,
1713                         DEFAULT_KERNEL_TRACE_DIR, &consumer_path_offset);
1714         if (!trace_path) {
1715                 status = LTTNG_ERR_INVALID;
1716                 goto error;
1717         }
1718         /* Send metadata to consumer and snapshot everything. */
1719         cds_lfht_for_each_entry(output->socks->ht, &iter.iter,
1720                         socket, node.node) {
1721                 struct ltt_kernel_channel *chan;
1722
1723                 pthread_mutex_lock(socket->lock);
1724                 /* This stream must not be monitored by the consumer. */
1725                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1726                 pthread_mutex_unlock(socket->lock);
1727                 if (ret < 0) {
1728                         status = LTTNG_ERR_KERN_META_FAIL;
1729                         goto error_consumer;
1730                 }
1731
1732                 /* For each channel, ask the consumer to snapshot it. */
1733                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1734                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1735                                         ksess->uid, ksess->gid,
1736                                         &trace_path[consumer_path_offset], wait,
1737                                         nb_packets_per_stream);
1738                         if (status != LTTNG_OK) {
1739                                 (void) kernel_consumer_destroy_metadata(socket,
1740                                                 ksess->metadata);
1741                                 goto error_consumer;
1742                         }
1743                 }
1744
1745                 /* Snapshot metadata, */
1746                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1747                                 1, ksess->uid, ksess->gid, &trace_path[consumer_path_offset],
1748                                 wait, 0);
1749                 if (status != LTTNG_OK) {
1750                         goto error_consumer;
1751                 }
1752
1753                 /*
1754                  * The metadata snapshot is done, ask the consumer to destroy it since
1755                  * it's not monitored on the consumer side.
1756                  */
1757                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1758         }
1759
1760 error_consumer:
1761         /* Close newly opened metadata stream. It's now on the consumer side. */
1762         err = close(ksess->metadata_stream_fd);
1763         if (err < 0) {
1764                 PERROR("close snapshot kernel");
1765         }
1766
1767 error_open_stream:
1768         trace_kernel_destroy_metadata(ksess->metadata);
1769 error:
1770         /* Restore metadata state.*/
1771         ksess->metadata = saved_metadata;
1772         ksess->metadata_stream_fd = saved_metadata_fd;
1773         rcu_read_unlock();
1774         free(trace_path);
1775         return status;
1776 }
1777
1778 /*
1779  * Get the syscall mask array from the kernel tracer.
1780  *
1781  * Return 0 on success else a negative value. In both case, syscall_mask should
1782  * be freed.
1783  */
1784 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1785 {
1786         assert(syscall_mask);
1787         assert(nr_bits);
1788
1789         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1790 }
1791
1792 static
1793 int kernel_tracer_abi_greater_or_equal(unsigned int major, unsigned int minor)
1794 {
1795         int ret;
1796         struct lttng_kernel_tracer_abi_version abi;
1797
1798         ret = kernctl_tracer_abi_version(kernel_tracer_fd, &abi);
1799         if (ret < 0) {
1800                 ERR("Failed to retrieve lttng-modules ABI version");
1801                 goto error;
1802         }
1803
1804         ret = abi.major > major || (abi.major == major && abi.minor >= minor);
1805 error:
1806         return ret;
1807 }
1808
1809 /*
1810  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1811  * version number.
1812  *
1813  * Return 1 on success, 0 when feature is not supported, negative value in case
1814  * of errors.
1815  */
1816 int kernel_supports_ring_buffer_snapshot_sample_positions(void)
1817 {
1818         /*
1819          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1820          */
1821         return kernel_tracer_abi_greater_or_equal(2, 3);
1822 }
1823
1824 /*
1825  * Check for the support of the packet sequence number via abi version number.
1826  *
1827  * Return 1 on success, 0 when feature is not supported, negative value in case
1828  * of errors.
1829  */
1830 int kernel_supports_ring_buffer_packet_sequence_number(void)
1831 {
1832         /*
1833          * Packet sequence number was introduced in LTTng 2.8,
1834          * lttng-modules ABI 2.1.
1835          */
1836         return kernel_tracer_abi_greater_or_equal(2, 1);
1837 }
1838
1839 /*
1840  * Check for the support of event notifiers via abi version number.
1841  *
1842  * Return 1 on success, 0 when feature is not supported, negative value in case
1843  * of errors.
1844  */
1845 int kernel_supports_event_notifiers(void)
1846 {
1847         /*
1848          * Event notifiers were introduced in LTTng 2.13, lttng-modules ABI 2.6.
1849          */
1850         return kernel_tracer_abi_greater_or_equal(2, 6);
1851 }
1852
1853 /*
1854  * Rotate a kernel session.
1855  *
1856  * Return LTTNG_OK on success or else an LTTng error code.
1857  */
1858 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1859 {
1860         int ret;
1861         enum lttng_error_code status = LTTNG_OK;
1862         struct consumer_socket *socket;
1863         struct lttng_ht_iter iter;
1864         struct ltt_kernel_session *ksess = session->kernel_session;
1865
1866         assert(ksess);
1867         assert(ksess->consumer);
1868
1869         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1870                         session->name, session->id);
1871
1872         rcu_read_lock();
1873
1874         /*
1875          * Note that this loop will end after one iteration given that there is
1876          * only one kernel consumer.
1877          */
1878         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1879                         socket, node.node) {
1880                 struct ltt_kernel_channel *chan;
1881
1882                 /* For each channel, ask the consumer to rotate it. */
1883                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1884                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1885                                         chan->key, session->name);
1886                         ret = consumer_rotate_channel(socket, chan->key,
1887                                         ksess->uid, ksess->gid, ksess->consumer,
1888                                         /* is_metadata_channel */ false);
1889                         if (ret < 0) {
1890                                 status = LTTNG_ERR_ROTATION_FAIL_CONSUMER;
1891                                 goto error;
1892                         }
1893                 }
1894
1895                 /*
1896                  * Rotate the metadata channel.
1897                  */
1898                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1899                                 ksess->uid, ksess->gid, ksess->consumer,
1900                                 /* is_metadata_channel */ true);
1901                 if (ret < 0) {
1902                         status = LTTNG_ERR_ROTATION_FAIL_CONSUMER;
1903                         goto error;
1904                 }
1905         }
1906
1907 error:
1908         rcu_read_unlock();
1909         return status;
1910 }
1911
1912 enum lttng_error_code kernel_create_channel_subdirectories(
1913                 const struct ltt_kernel_session *ksess)
1914 {
1915         enum lttng_error_code ret = LTTNG_OK;
1916         enum lttng_trace_chunk_status chunk_status;
1917
1918         rcu_read_lock();
1919         assert(ksess->current_trace_chunk);
1920
1921         /*
1922          * Create the index subdirectory which will take care
1923          * of implicitly creating the channel's path.
1924          */
1925         chunk_status = lttng_trace_chunk_create_subdirectory(
1926                         ksess->current_trace_chunk,
1927                         DEFAULT_KERNEL_TRACE_DIR "/" DEFAULT_INDEX_DIR);
1928         if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1929                 ret = LTTNG_ERR_CREATE_DIR_FAIL;
1930                 goto error;
1931         }
1932 error:
1933         rcu_read_unlock();
1934         return ret;
1935 }
1936
1937 /*
1938  * Setup necessary data for kernel tracer action.
1939  */
1940 LTTNG_HIDDEN
1941 int init_kernel_tracer(void)
1942 {
1943         int ret;
1944         bool is_root = !getuid();
1945
1946         /* Modprobe lttng kernel modules */
1947         ret = modprobe_lttng_control();
1948         if (ret < 0) {
1949                 goto error;
1950         }
1951
1952         /* Open debugfs lttng */
1953         kernel_tracer_fd = open(module_proc_lttng, O_RDWR);
1954         if (kernel_tracer_fd < 0) {
1955                 DBG("Failed to open %s", module_proc_lttng);
1956                 goto error_open;
1957         }
1958
1959         /* Validate kernel version */
1960         ret = kernel_validate_version(&kernel_tracer_version,
1961                         &kernel_tracer_abi_version);
1962         if (ret < 0) {
1963                 goto error_version;
1964         }
1965
1966         ret = modprobe_lttng_data();
1967         if (ret < 0) {
1968                 goto error_modules;
1969         }
1970
1971         ret = kernel_supports_ring_buffer_snapshot_sample_positions();
1972         if (ret < 0) {
1973                 goto error_modules;
1974         }
1975         if (ret < 1) {
1976                 WARN("Kernel tracer does not support buffer monitoring. "
1977                         "The monitoring timer of channels in the kernel domain "
1978                         "will be set to 0 (disabled).");
1979         }
1980
1981         ret = kernel_supports_event_notifiers();
1982         if (ret < 0) {
1983                 ERR("Failed to check for kernel tracer event notifier support");
1984                 goto error_modules;
1985         }
1986         ret = kernel_create_event_notifier_group(&kernel_tracer_event_notifier_group_fd);
1987         if (ret < 0) {
1988                 /* This is not fatal. */
1989                 WARN("Failed to create kernel event notifier group");
1990                 kernel_tracer_event_notifier_group_fd = -1;
1991         } else {
1992                 const enum lttng_error_code error_code_ret =
1993                                 kernel_create_event_notifier_group_notification_fd(
1994                                                 &kernel_tracer_event_notifier_group_notification_fd);
1995
1996                 if (error_code_ret != LTTNG_OK) {
1997                         goto error_modules;
1998                 }
1999
2000                 kernel_token_to_event_notifier_rule_ht = cds_lfht_new(
2001                                 DEFAULT_HT_SIZE, 1, 0,
2002                                 CDS_LFHT_AUTO_RESIZE | CDS_LFHT_ACCOUNTING,
2003                                 NULL);
2004                 if (!kernel_token_to_event_notifier_rule_ht) {
2005                         goto error_token_ht;
2006                 }
2007         }
2008
2009         DBG("Kernel tracer initialized: kernel tracer fd = %d, event notifier group fd = %d, event notifier group notification fd = %d",
2010                         kernel_tracer_fd, kernel_tracer_event_notifier_group_fd,
2011                         kernel_tracer_event_notifier_group_notification_fd);
2012
2013         ret = syscall_init_table(kernel_tracer_fd);
2014         if (ret < 0) {
2015                 ERR("Unable to populate syscall table. Syscall tracing won't "
2016                         "work for this session daemon.");
2017         }
2018
2019         return 0;
2020
2021 error_version:
2022         modprobe_remove_lttng_control();
2023         ret = close(kernel_tracer_fd);
2024         if (ret) {
2025                 PERROR("Failed to close kernel tracer file descriptor: fd = %d",
2026                                 kernel_tracer_fd);
2027         }
2028         kernel_tracer_fd = -1;
2029         return LTTNG_ERR_KERN_VERSION;
2030
2031
2032 error_token_ht:
2033         ret = close(kernel_tracer_event_notifier_group_notification_fd);
2034         if (ret) {
2035                 PERROR("Failed to close kernel tracer event notifier group notification file descriptor: fd = %d",
2036                                 kernel_tracer_event_notifier_group_notification_fd);
2037         }
2038
2039 error_modules:
2040         ret = close(kernel_tracer_event_notifier_group_fd);
2041         if (ret) {
2042                 PERROR("Failed to close kernel tracer event notifier group file descriptor: fd = %d",
2043                                 kernel_tracer_event_notifier_group_fd);
2044         }
2045
2046         ret = close(kernel_tracer_fd);
2047         if (ret) {
2048                 PERROR("Failed to close kernel tracer file descriptor: fd = %d",
2049                                 kernel_tracer_fd);
2050         }
2051
2052 error_open:
2053         modprobe_remove_lttng_control();
2054
2055 error:
2056         WARN("No kernel tracer available");
2057         kernel_tracer_fd = -1;
2058         if (!is_root) {
2059                 return LTTNG_ERR_NEED_ROOT_SESSIOND;
2060         } else {
2061                 return LTTNG_ERR_KERN_NA;
2062         }
2063 }
2064
2065 LTTNG_HIDDEN
2066 void cleanup_kernel_tracer(void)
2067 {
2068         DBG2("Closing kernel event notifier group notification file descriptor");
2069         if (kernel_tracer_event_notifier_group_notification_fd >= 0) {
2070                 int ret = notification_thread_command_remove_tracer_event_source(
2071                                 notification_thread_handle,
2072                                 kernel_tracer_event_notifier_group_notification_fd);
2073                 if (ret != LTTNG_OK) {
2074                         ERR("Failed to remove kernel event notifier notification from notification thread");
2075                 }
2076
2077                 ret = close(kernel_tracer_event_notifier_group_notification_fd);
2078                 if (ret) {
2079                         PERROR("Failed to close kernel event notifier group notification file descriptor: fd = %d",
2080                                         kernel_tracer_event_notifier_group_notification_fd);
2081                 }
2082
2083                 kernel_tracer_event_notifier_group_notification_fd = -1;
2084         }
2085
2086         if (kernel_token_to_event_notifier_rule_ht) {
2087                 const int ret = cds_lfht_destroy(
2088                                 kernel_token_to_event_notifier_rule_ht, NULL);
2089                 assert(ret == 0);
2090         }
2091
2092         DBG2("Closing kernel event notifier group file descriptor");
2093         if (kernel_tracer_event_notifier_group_fd >= 0) {
2094                 const int ret = close(kernel_tracer_event_notifier_group_fd);
2095
2096                 if (ret) {
2097                         PERROR("Failed to close kernel event notifier group file descriptor: fd = %d",
2098                                         kernel_tracer_event_notifier_group_fd);
2099                 }
2100
2101                 kernel_tracer_event_notifier_group_fd = -1;
2102         }
2103
2104         DBG2("Closing kernel fd");
2105         if (kernel_tracer_fd >= 0) {
2106                 const int ret = close(kernel_tracer_fd);
2107
2108                 if (ret) {
2109                         PERROR("Failed to close kernel tracer file descriptor: fd = %d",
2110                                         kernel_tracer_fd);
2111                 }
2112
2113                 kernel_tracer_fd = -1;
2114         }
2115
2116         DBG("Unloading kernel modules");
2117         modprobe_remove_lttng_all();
2118         free(syscall_table);
2119 }
2120
2121 LTTNG_HIDDEN
2122 bool kernel_tracer_is_initialized(void)
2123 {
2124         return kernel_tracer_fd >= 0;
2125 }
2126
2127 /*
2128  *  Clear a kernel session.
2129  *
2130  * Return LTTNG_OK on success or else an LTTng error code.
2131  */
2132 enum lttng_error_code kernel_clear_session(struct ltt_session *session)
2133 {
2134         int ret;
2135         enum lttng_error_code status = LTTNG_OK;
2136         struct consumer_socket *socket;
2137         struct lttng_ht_iter iter;
2138         struct ltt_kernel_session *ksess = session->kernel_session;
2139
2140         assert(ksess);
2141         assert(ksess->consumer);
2142
2143         DBG("Clear kernel session %s (session %" PRIu64 ")",
2144                         session->name, session->id);
2145
2146         rcu_read_lock();
2147
2148         if (ksess->active) {
2149                 ERR("Expecting inactive session %s (%" PRIu64 ")", session->name, session->id);
2150                 status = LTTNG_ERR_FATAL;
2151                 goto end;
2152         }
2153
2154         /*
2155          * Note that this loop will end after one iteration given that there is
2156          * only one kernel consumer.
2157          */
2158         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
2159                         socket, node.node) {
2160                 struct ltt_kernel_channel *chan;
2161
2162                 /* For each channel, ask the consumer to clear it. */
2163                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
2164                         DBG("Clear kernel channel %" PRIu64 ", session %s",
2165                                         chan->key, session->name);
2166                         ret = consumer_clear_channel(socket, chan->key);
2167                         if (ret < 0) {
2168                                 goto error;
2169                         }
2170                 }
2171
2172                 if (!ksess->metadata) {
2173                         /*
2174                          * Nothing to do for the metadata.
2175                          * This is a snapshot session.
2176                          * The metadata is genererated on the fly.
2177                          */
2178                         continue;
2179                 }
2180
2181                 /*
2182                  * Clear the metadata channel.
2183                  * Metadata channel is not cleared per se but we still need to
2184                  * perform a rotation operation on it behind the scene.
2185                  */
2186                 ret = consumer_clear_channel(socket, ksess->metadata->key);
2187                 if (ret < 0) {
2188                         goto error;
2189                 }
2190         }
2191
2192         goto end;
2193 error:
2194         switch (-ret) {
2195         case LTTCOMM_CONSUMERD_RELAYD_CLEAR_DISALLOWED:
2196               status = LTTNG_ERR_CLEAR_RELAY_DISALLOWED;
2197               break;
2198         default:
2199               status = LTTNG_ERR_CLEAR_FAIL_CONSUMER;
2200               break;
2201         }
2202 end:
2203         rcu_read_unlock();
2204         return status;
2205 }
2206
2207 enum lttng_error_code kernel_create_event_notifier_group_notification_fd(
2208                 int *event_notifier_group_notification_fd)
2209 {
2210         int local_fd = -1, ret;
2211         enum lttng_error_code error_code_ret;
2212
2213         assert(event_notifier_group_notification_fd);
2214
2215         ret = kernctl_create_event_notifier_group_notification_fd(
2216                         kernel_tracer_event_notifier_group_fd);
2217         if (ret < 0) {
2218                 PERROR("Failed to create kernel event notifier group notification file descriptor");
2219                 error_code_ret = LTTNG_ERR_EVENT_NOTIFIER_GROUP_NOTIFICATION_FD;
2220                 goto error;
2221         }
2222
2223         local_fd = ret;
2224
2225         /* Prevent fd duplication after execlp(). */
2226         ret = fcntl(local_fd, F_SETFD, FD_CLOEXEC);
2227         if (ret < 0) {
2228                 PERROR("Failed to set FD_CLOEXEC on kernel event notifier group notification file descriptor: fd = %d",
2229                                 local_fd);
2230                 error_code_ret = LTTNG_ERR_EVENT_NOTIFIER_GROUP_NOTIFICATION_FD;
2231                 goto error;
2232         }
2233
2234         DBG("Created kernel notifier group notification file descriptor: fd = %d",
2235                         local_fd);
2236         error_code_ret = LTTNG_OK;
2237         *event_notifier_group_notification_fd = local_fd;
2238         local_fd = -1;
2239
2240 error:
2241         if (local_fd >= 0) {
2242                 ret = close(local_fd);
2243                 if (ret) {
2244                         PERROR("Failed to close kernel event notifier group notification file descriptor: fd = %d",
2245                                         local_fd);
2246                 }
2247         }
2248
2249         return error_code_ret;
2250 }
2251
2252 enum lttng_error_code kernel_destroy_event_notifier_group_notification_fd(
2253                 int event_notifier_group_notification_fd)
2254 {
2255         enum lttng_error_code ret_code = LTTNG_OK;
2256
2257         DBG("Closing event notifier group notification file descriptor: fd = %d",
2258                         event_notifier_group_notification_fd);
2259         if (event_notifier_group_notification_fd >= 0) {
2260                 const int ret = close(event_notifier_group_notification_fd);
2261                 if (ret) {
2262                         PERROR("Failed to close event notifier group notification file descriptor: fd = %d",
2263                                         event_notifier_group_notification_fd);
2264                 }
2265         }
2266
2267         return ret_code;
2268 }
2269
2270 static
2271 unsigned long hash_trigger(const struct lttng_trigger *trigger)
2272 {
2273         const struct lttng_condition *condition =
2274                         lttng_trigger_get_const_condition(trigger);
2275
2276         return lttng_condition_hash(condition);
2277 }
2278
2279 static
2280 int match_trigger(struct cds_lfht_node *node, const void *key)
2281 {
2282         const struct ltt_kernel_event_notifier_rule *event_notifier_rule;
2283         const struct lttng_trigger *trigger = key;
2284
2285         event_notifier_rule = caa_container_of(node,
2286                         const struct ltt_kernel_event_notifier_rule, ht_node);
2287
2288         return lttng_trigger_is_equal(trigger, event_notifier_rule->trigger);
2289 }
2290
2291 static enum lttng_error_code kernel_create_event_notifier_rule(
2292                 struct lttng_trigger *trigger,
2293                 const struct lttng_credentials *creds, uint64_t token)
2294 {
2295         int err, fd, ret = 0;
2296         enum lttng_error_code error_code_ret;
2297         enum lttng_condition_status condition_status;
2298         enum lttng_condition_type condition_type;
2299         enum lttng_event_rule_type event_rule_type;
2300         struct ltt_kernel_event_notifier_rule *event_notifier_rule;
2301         struct lttng_kernel_event_notifier kernel_event_notifier = {};
2302         const struct lttng_condition *condition = NULL;
2303         const struct lttng_event_rule *event_rule = NULL;
2304
2305         assert(trigger);
2306
2307         condition = lttng_trigger_get_const_condition(trigger);
2308         assert(condition);
2309
2310         condition_type = lttng_condition_get_type(condition);
2311         assert(condition_type == LTTNG_CONDITION_TYPE_EVENT_RULE_HIT);
2312
2313         /* Does not acquire a reference. */
2314         condition_status = lttng_condition_event_rule_get_rule(
2315                         condition, &event_rule);
2316         assert(condition_status == LTTNG_CONDITION_STATUS_OK);
2317         assert(event_rule);
2318
2319         event_rule_type = lttng_event_rule_get_type(event_rule);
2320         assert(event_rule_type != LTTNG_EVENT_RULE_TYPE_UNKNOWN);
2321
2322         error_code_ret = trace_kernel_create_event_notifier_rule(trigger, token,
2323                         &event_notifier_rule);
2324         if (error_code_ret != LTTNG_OK) {
2325                 goto error;
2326         }
2327
2328         error_code_ret = trace_kernel_init_event_notifier_from_event_rule(
2329                         event_rule, &kernel_event_notifier);
2330         if (error_code_ret != LTTNG_OK) {
2331                 goto error;
2332         }
2333
2334         kernel_event_notifier.event.token = event_notifier_rule->token;
2335
2336         fd = kernctl_create_event_notifier(
2337                         kernel_tracer_event_notifier_group_fd,
2338                         &kernel_event_notifier);
2339         if (fd < 0) {
2340                 switch (-fd) {
2341                 case EEXIST:
2342                         error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
2343                         break;
2344                 case ENOSYS:
2345                         WARN("Failed to create kernel event notifier: not notifier type not implemented");
2346                         error_code_ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
2347                         break;
2348                 case ENOENT:
2349                         WARN("Failed to create kernel event notifier: not found: name = '%s'",
2350                                         kernel_event_notifier.event.name);
2351                         error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
2352                         break;
2353                 default:
2354                         PERROR("Failed to create kernel event notifier: error code = %d, name = '%s'",
2355                                         fd, kernel_event_notifier.event.name);
2356                         error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
2357                 }
2358                 goto free_event;
2359         }
2360
2361         event_notifier_rule->fd = fd;
2362         /* Prevent fd duplication after execlp(). */
2363         err = fcntl(event_notifier_rule->fd, F_SETFD, FD_CLOEXEC);
2364         if (err < 0) {
2365                 PERROR("Failed to set FD_CLOEXEC on kernel event notifier file descriptor: fd = %d",
2366                                 fd);
2367                 error_code_ret = LTTNG_ERR_FATAL;
2368                 goto set_cloexec_error;
2369         }
2370
2371         if (event_notifier_rule->filter) {
2372                 err = kernctl_filter(event_notifier_rule->fd, event_notifier_rule->filter);
2373                 if (err < 0) {
2374                         switch (-err) {
2375                         case ENOMEM:
2376                                 error_code_ret = LTTNG_ERR_FILTER_NOMEM;
2377                                 break;
2378                         default:
2379                                 error_code_ret = LTTNG_ERR_FILTER_INVAL;
2380                                 break;
2381                         }
2382                         goto filter_error;
2383                 }
2384         }
2385
2386         if (lttng_event_rule_get_type(event_rule) ==
2387                         LTTNG_EVENT_RULE_TYPE_UPROBE) {
2388                 ret = userspace_probe_event_rule_add_callsites(
2389                                 event_rule, creds, event_notifier_rule->fd);
2390                 if (ret) {
2391                         error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
2392                         goto add_callsite_error;
2393                 }
2394         }
2395
2396         err = kernctl_enable(event_notifier_rule->fd);
2397         if (err < 0) {
2398                 switch (-err) {
2399                 case EEXIST:
2400                         error_code_ret = LTTNG_ERR_KERN_EVENT_EXIST;
2401                         break;
2402                 default:
2403                         PERROR("enable kernel event notifier");
2404                         error_code_ret = LTTNG_ERR_KERN_ENABLE_FAIL;
2405                         break;
2406                 }
2407                 goto enable_error;
2408         }
2409
2410         /* Add trigger to kernel token mapping in the hash table. */
2411         rcu_read_lock();
2412         cds_lfht_add(kernel_token_to_event_notifier_rule_ht, hash_trigger(trigger),
2413                         &event_notifier_rule->ht_node);
2414         rcu_read_unlock();
2415
2416         DBG("Created kernel event notifier: name = '%s', fd = %d",
2417                         kernel_event_notifier.event.name,
2418                         event_notifier_rule->fd);
2419
2420         return LTTNG_OK;
2421
2422 add_callsite_error:
2423 enable_error:
2424 set_cloexec_error:
2425 filter_error:
2426         {
2427                 const int close_ret = close(event_notifier_rule->fd);
2428
2429                 if (close_ret) {
2430                         PERROR("Failed to close kernel event notifier file descriptor: fd = %d",
2431                                         event_notifier_rule->fd);
2432                 }
2433         }
2434 free_event:
2435         free(event_notifier_rule);
2436 error:
2437         return error_code_ret;
2438 }
2439
2440 enum lttng_error_code kernel_register_event_notifier(
2441                 struct lttng_trigger *trigger,
2442                 const struct lttng_credentials *cmd_creds)
2443 {
2444         enum lttng_error_code ret;
2445         enum lttng_condition_status status;
2446         enum lttng_domain_type domain_type;
2447         const struct lttng_event_rule *event_rule;
2448         const struct lttng_condition *const condition =
2449                         lttng_trigger_get_const_condition(trigger);
2450         const uint64_t token = lttng_trigger_get_tracer_token(trigger);
2451
2452         assert(condition);
2453
2454         /* Does not acquire a reference to the event rule. */
2455         status = lttng_condition_event_rule_get_rule(
2456                         condition, &event_rule);
2457         assert(status == LTTNG_CONDITION_STATUS_OK);
2458
2459         domain_type = lttng_event_rule_get_domain_type(event_rule);
2460         assert(domain_type == LTTNG_DOMAIN_KERNEL);
2461
2462         ret = kernel_create_event_notifier_rule(trigger, cmd_creds, token);
2463         if (ret != LTTNG_OK) {
2464                 ERR("Failed to create kernel event notifier rule");
2465         }
2466
2467         return ret;
2468 }
2469
2470 enum lttng_error_code kernel_unregister_event_notifier(
2471                 const struct lttng_trigger *trigger)
2472 {
2473         struct ltt_kernel_event_notifier_rule *token_event_rule_element;
2474         struct cds_lfht_node *node;
2475         struct cds_lfht_iter iter;
2476         enum lttng_error_code error_code_ret;
2477         int ret;
2478
2479         rcu_read_lock();
2480
2481         cds_lfht_lookup(kernel_token_to_event_notifier_rule_ht,
2482                         hash_trigger(trigger), match_trigger, trigger, &iter);
2483
2484         node = cds_lfht_iter_get_node(&iter);
2485         if (!node) {
2486                 error_code_ret = LTTNG_ERR_TRIGGER_NOT_FOUND;
2487                 goto error;
2488         }
2489
2490         token_event_rule_element = caa_container_of(node,
2491                         struct ltt_kernel_event_notifier_rule, ht_node);
2492
2493         ret = kernel_disable_event_notifier_rule(token_event_rule_element);
2494         if (ret) {
2495                 error_code_ret = LTTNG_ERR_FATAL;
2496                 goto error;
2497         }
2498
2499         trace_kernel_destroy_event_notifier_rule(token_event_rule_element);
2500         error_code_ret = LTTNG_OK;
2501
2502 error:
2503         rcu_read_unlock();
2504
2505         return error_code_ret;
2506 }
2507
2508 int kernel_get_notification_fd(void)
2509 {
2510         return kernel_tracer_event_notifier_group_notification_fd;
2511 }