src/bin/lttng-sessiond/kernel.c

   1 /*
   2  * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca>
   3  *
   4  * This program is free software; you can redistribute it and/or modify
   5  * it under the terms of the GNU General Public License, version 2 only,
   6  * as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  11  * GNU General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public License along
  14  * with this program; if not, write to the Free Software Foundation, Inc.,
  15  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  16  */
  17
  18 #define _LGPL_SOURCE
  19 #include <fcntl.h>
  20 #include <stdlib.h>
  21 #include <stdio.h>
  22 #include <string.h>
  23 #include <unistd.h>
  24 #include <inttypes.h>
  25
  26 #include <common/common.h>
  27 #include <common/kernel-ctl/kernel-ctl.h>
  28 #include <common/kernel-ctl/kernel-ioctl.h>
  29 #include <common/sessiond-comm/sessiond-comm.h>
  30
  31 #include "consumer.h"
  32 #include "kernel.h"
  33 #include "kernel-consumer.h"
  34 #include "kern-modules.h"
  35 #include "utils.h"
  36 #include "rotate.h"
  37
  38 /*
  39  * Key used to reference a channel between the sessiond and the consumer. This
  40  * is only read and updated with the session_list lock held.
  41  */
  42 static uint64_t next_kernel_channel_key;
  43
  44 #include <lttng/userspace-probe.h>
  45 #include <lttng/userspace-probe-internal.h>
  46 /*
  47  * Add context on a kernel channel.
  48  *
  49  * Assumes the ownership of ctx.
  50  */
  51 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  52                 struct ltt_kernel_context *ctx)
  53 {
  54         int ret;
  55
  56         assert(chan);
  57         assert(ctx);
  58
  59         DBG("Adding context to channel %s", chan->channel->name);
  60         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  61         if (ret < 0) {
  62                 switch (-ret) {
  63                 case ENOSYS:
  64                         /* Exists but not available for this kernel */
  65                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  66                         goto error;
  67                 case EEXIST:
  68                         /* If EEXIST, we just ignore the error */
  69                         ret = 0;
  70                         goto end;
  71                 default:
  72                         PERROR("add context ioctl");
  73                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  74                         goto error;
  75                 }
  76         }
  77         ret = 0;
  78
  79 end:
  80         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  81         ctx->in_list = true;
  82         ctx = NULL;
  83 error:
  84         if (ctx) {
  85                 trace_kernel_destroy_context(ctx);
  86         }
  87         return ret;
  88 }
  89
  90 /*
  91  * Create a new kernel session, register it to the kernel tracer and add it to
  92  * the session daemon session.
  93  */
  94 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  95 {
  96         int ret;
  97         struct ltt_kernel_session *lks;
  98
  99         assert(session);
 100
 101         /* Allocate data structure */
 102         lks = trace_kernel_create_session();
 103         if (lks == NULL) {
 104                 ret = -1;
 105                 goto error;
 106         }
 107
 108         /* Kernel tracer session creation */
 109         ret = kernctl_create_session(tracer_fd);
 110         if (ret < 0) {
 111                 PERROR("ioctl kernel create session");
 112                 goto error;
 113         }
 114
 115         lks->fd = ret;
 116         /* Prevent fd duplication after execlp() */
 117         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 118         if (ret < 0) {
 119                 PERROR("fcntl session fd");
 120         }
 121
 122         lks->id = session->id;
 123         lks->consumer_fds_sent = 0;
 124         session->kernel_session = lks;
 125
 126         DBG("Kernel session created (fd: %d)", lks->fd);
 127
 128         ret = kernctl_session_set_name(lks->fd, session->name);
 129         if (ret) {
 130                 WARN("Could not set kernel session name");
 131         }
 132
 133         ret = kernctl_session_set_creation_time(lks->fd, session->creation_time);
 134         if (ret) {
 135                 WARN("Could not set kernel session creation time");
 136         }
 137
 138         return 0;
 139
 140 error:
 141         if (lks) {
 142                 trace_kernel_destroy_session(lks);
 143         }
 144         return ret;
 145 }
 146
 147 /*
 148  * Create a kernel channel, register it to the kernel tracer and add it to the
 149  * kernel session.
 150  */
 151 int kernel_create_channel(struct ltt_kernel_session *session,
 152                 struct lttng_channel *chan)
 153 {
 154         int ret;
 155         struct ltt_kernel_channel *lkc;
 156
 157         assert(session);
 158         assert(chan);
 159
 160         /* Allocate kernel channel */
 161         lkc = trace_kernel_create_channel(chan);
 162         if (lkc == NULL) {
 163                 goto error;
 164         }
 165
 166         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 167                         chan->name, lkc->channel->attr.overwrite,
 168                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 169                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 170                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 171
 172         /* Kernel tracer channel creation */
 173         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 174         if (ret < 0) {
 175                 PERROR("ioctl kernel create channel");
 176                 goto error;
 177         }
 178
 179         /* Setup the channel fd */
 180         lkc->fd = ret;
 181         /* Prevent fd duplication after execlp() */
 182         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 183         if (ret < 0) {
 184                 PERROR("fcntl session fd");
 185         }
 186
 187         /* Add channel to session */
 188         cds_list_add(&lkc->list, &session->channel_list.head);
 189         session->channel_count++;
 190         lkc->session = session;
 191         lkc->key = ++next_kernel_channel_key;
 192
 193         DBG("Kernel channel %s created (fd: %d, key: %" PRIu64 ")",
 194                         lkc->channel->name, lkc->fd, lkc->key);
 195
 196         return 0;
 197
 198 error:
 199         if (lkc) {
 200                 free(lkc->channel);
 201                 free(lkc);
 202         }
 203         return -1;
 204 }
 205
 206 /*
 207  * Compute the offset of the instrumentation byte in the binary based on the
 208  * function probe location using the ELF lookup method.
 209  *
 210  * Returns 0 on success and set the offset out parameter to the offset of the
 211  * elf symbol
 212  * Returns -1 on error
 213  */
 214 static
 215 int extract_userspace_probe_offset_function_elf(
 216                 const struct lttng_userspace_probe_location *probe_location,
 217                 struct ltt_kernel_session *session, uint64_t *offset)
 218 {
 219         int fd;
 220         int ret = 0;
 221         const char *symbol = NULL;
 222         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 223         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 224
 225         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 226                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_FUNCTION);
 227
 228         lookup = lttng_userspace_probe_location_get_lookup_method(
 229                         probe_location);
 230         if (!lookup) {
 231                 ret = -1;
 232                 goto end;
 233         }
 234
 235         lookup_method_type =
 236                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 237
 238         assert(lookup_method_type ==
 239                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF);
 240
 241         symbol = lttng_userspace_probe_location_function_get_function_name(
 242                         probe_location);
 243         if (!symbol) {
 244                 ret = -1;
 245                 goto end;
 246         }
 247
 248         fd = lttng_userspace_probe_location_function_get_binary_fd(probe_location);
 249         if (fd < 0) {
 250                 ret = -1;
 251                 goto end;
 252         }
 253
 254         ret = run_as_extract_elf_symbol_offset(fd, symbol, session->uid,
 255                         session->gid, offset);
 256         if (ret < 0) {
 257                 DBG("userspace probe offset calculation failed for "
 258                                 "function %s", symbol);
 259                 goto end;
 260         }
 261
 262         DBG("userspace probe elf offset for %s is 0x%jd", symbol, (intmax_t)(*offset));
 263 end:
 264         return ret;
 265 }
 266
 267 /*
 268  * Compute the offsets of the instrumentation bytes in the binary based on the
 269  * tracepoint probe location using the SDT lookup method. This function
 270  * allocates the offsets buffer, the caller must free it.
 271  *
 272  * Returns 0 on success and set the offset out parameter to the offsets of the
 273  * SDT tracepoint.
 274  * Returns -1 on error.
 275  */
 276 static
 277 int extract_userspace_probe_offset_tracepoint_sdt(
 278                 const struct lttng_userspace_probe_location *probe_location,
 279                 struct ltt_kernel_session *session, uint64_t **offsets,
 280                 uint32_t *offsets_count)
 281 {
 282         enum lttng_userspace_probe_location_lookup_method_type lookup_method_type;
 283         const struct lttng_userspace_probe_location_lookup_method *lookup = NULL;
 284         const char *probe_name = NULL, *provider_name = NULL;
 285         int ret = 0;
 286         int fd, i;
 287
 288         assert(lttng_userspace_probe_location_get_type(probe_location) ==
 289                         LTTNG_USERSPACE_PROBE_LOCATION_TYPE_TRACEPOINT);
 290
 291         lookup = lttng_userspace_probe_location_get_lookup_method(probe_location);
 292         if (!lookup) {
 293                 ret = -1;
 294                 goto end;
 295         }
 296
 297         lookup_method_type =
 298                         lttng_userspace_probe_location_lookup_method_get_type(lookup);
 299
 300         assert(lookup_method_type ==
 301                         LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT);
 302
 303
 304         probe_name = lttng_userspace_probe_location_tracepoint_get_probe_name(
 305                         probe_location);
 306         if (!probe_name) {
 307                 ret = -1;
 308                 goto end;
 309         }
 310
 311         provider_name = lttng_userspace_probe_location_tracepoint_get_provider_name(
 312                         probe_location);
 313         if (!provider_name) {
 314                 ret = -1;
 315                 goto end;
 316         }
 317
 318         fd = lttng_userspace_probe_location_tracepoint_get_binary_fd(probe_location);
 319         if (fd < 0) {
 320                 ret = -1;
 321                 goto end;
 322         }
 323
 324         ret = run_as_extract_sdt_probe_offsets(fd, provider_name, probe_name,
 325                         session->uid, session->gid, offsets, offsets_count);
 326         if (ret < 0) {
 327                 DBG("userspace probe offset calculation failed for sdt "
 328                                 "probe %s:%s", provider_name, probe_name);
 329                 goto end;
 330         }
 331
 332         if (*offsets_count == 0) {
 333                 DBG("no userspace probe offset found");
 334                 goto end;
 335         }
 336
 337         DBG("%u userspace probe SDT offsets found for %s:%s at:",
 338                         *offsets_count, provider_name, probe_name);
 339         for (i = 0; i < *offsets_count; i++) {
 340                 DBG("\t0x%jd", (intmax_t)((*offsets)[i]));
 341         }
 342 end:
 343         return ret;
 344 }
 345
 346 /*
 347  * Extract the offsets of the instrumentation point for the different lookup
 348  * methods.
 349  */
 350 static
 351 int userspace_probe_add_callsites(struct lttng_event *ev,
 352                         struct ltt_kernel_session *session, int fd)
 353 {
 354         const struct lttng_userspace_probe_location_lookup_method *lookup_method = NULL;
 355         enum lttng_userspace_probe_location_lookup_method_type type;
 356         const struct lttng_userspace_probe_location *location = NULL;
 357         int ret;
 358
 359         assert(ev);
 360         assert(ev->type == LTTNG_EVENT_USERSPACE_PROBE);
 361
 362         location = lttng_event_get_userspace_probe_location(ev);
 363         if (!location) {
 364                 ret = -1;
 365                 goto end;
 366         }
 367         lookup_method =
 368                         lttng_userspace_probe_location_get_lookup_method(location);
 369         if (!lookup_method) {
 370                 ret = -1;
 371                 goto end;
 372         }
 373
 374         type = lttng_userspace_probe_location_lookup_method_get_type(lookup_method);
 375         switch (type) {
 376         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_FUNCTION_ELF:
 377         {
 378                 struct lttng_kernel_event_callsite callsite;
 379                 uint64_t offset;
 380
 381                 ret = extract_userspace_probe_offset_function_elf(location, session, &offset);
 382                 if (ret) {
 383                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 384                         goto end;
 385                 }
 386
 387                 callsite.u.uprobe.offset = offset;
 388                 ret = kernctl_add_callsite(fd, &callsite);
 389                 if (ret) {
 390                         WARN("Adding callsite to userspace probe "
 391                                         "event %s failed.", ev->name);
 392                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 393                         goto end;
 394                 }
 395                 break;
 396         }
 397         case LTTNG_USERSPACE_PROBE_LOCATION_LOOKUP_METHOD_TYPE_TRACEPOINT_SDT:
 398         {
 399                 int i;
 400                 uint64_t *offsets = NULL;
 401                 uint32_t offsets_count;
 402                 struct lttng_kernel_event_callsite callsite;
 403
 404                 /*
 405                  * This call allocates the offsets buffer. This buffer must be freed
 406                  * by the caller
 407                  */
 408                 ret = extract_userspace_probe_offset_tracepoint_sdt(location, session,
 409                                 &offsets, &offsets_count);
 410                 if (ret) {
 411                         ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 412                         goto end;
 413                 }
 414                 for (i = 0; i < offsets_count; i++) {
 415                         callsite.u.uprobe.offset = offsets[i];
 416                         ret = kernctl_add_callsite(fd, &callsite);
 417                         if (ret) {
 418                                 WARN("Adding callsite to userspace probe "
 419                                                 "event %s failed.", ev->name);
 420                                 ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 421                                 free(offsets);
 422                                 goto end;
 423                         }
 424                 }
 425                 free(offsets);
 426                 break;
 427         }
 428         default:
 429                 ret = LTTNG_ERR_PROBE_LOCATION_INVAL;
 430                 goto end;
 431         }
 432 end:
 433         return ret;
 434 }
 435
 436 /*
 437  * Create a kernel event, enable it to the kernel tracer and add it to the
 438  * channel event list of the kernel session.
 439  * We own filter_expression and filter.
 440  */
 441 int kernel_create_event(struct lttng_event *ev,
 442                 struct ltt_kernel_channel *channel,
 443                 char *filter_expression,
 444                 struct lttng_filter_bytecode *filter)
 445 {
 446         int err, fd;
 447         enum lttng_error_code ret;
 448         struct ltt_kernel_event *event;
 449
 450         assert(ev);
 451         assert(channel);
 452
 453         /* We pass ownership of filter_expression and filter */
 454         ret = trace_kernel_create_event(ev, filter_expression,
 455                         filter, &event);
 456         if (ret != LTTNG_OK) {
 457                 goto error;
 458         }
 459
 460         fd = kernctl_create_event(channel->fd, event->event);
 461         if (fd < 0) {
 462                 switch (-fd) {
 463                 case EEXIST:
 464                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 465                         break;
 466                 case ENOSYS:
 467                         WARN("Event type not implemented");
 468                         ret = LTTNG_ERR_KERN_EVENT_ENOSYS;
 469                         break;
 470                 case ENOENT:
 471                         WARN("Event %s not found!", ev->name);
 472                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 473                         break;
 474                 default:
 475                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 476                         PERROR("create event ioctl");
 477                 }
 478                 goto free_event;
 479         }
 480
 481         event->type = ev->type;
 482         event->fd = fd;
 483         /* Prevent fd duplication after execlp() */
 484         err = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 485         if (err < 0) {
 486                 PERROR("fcntl session fd");
 487         }
 488
 489         if (filter) {
 490                 err = kernctl_filter(event->fd, filter);
 491                 if (err < 0) {
 492                         switch (-err) {
 493                         case ENOMEM:
 494                                 ret = LTTNG_ERR_FILTER_NOMEM;
 495                                 break;
 496                         default:
 497                                 ret = LTTNG_ERR_FILTER_INVAL;
 498                                 break;
 499                         }
 500                         goto filter_error;
 501                 }
 502         }
 503
 504         if (ev->type == LTTNG_EVENT_USERSPACE_PROBE) {
 505                 ret = userspace_probe_add_callsites(ev, channel->session, event->fd);
 506                 if (ret) {
 507                         goto add_callsite_error;
 508                 }
 509         }
 510
 511         err = kernctl_enable(event->fd);
 512         if (err < 0) {
 513                 switch (-err) {
 514                 case EEXIST:
 515                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 516                         break;
 517                 default:
 518                         PERROR("enable kernel event");
 519                         ret = LTTNG_ERR_KERN_ENABLE_FAIL;
 520                         break;
 521                 }
 522                 goto enable_error;
 523         }
 524
 525         /* Add event to event list */
 526         cds_list_add(&event->list, &channel->events_list.head);
 527         channel->event_count++;
 528
 529         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 530
 531         return 0;
 532
 533 add_callsite_error:
 534 enable_error:
 535 filter_error:
 536         {
 537                 int closeret;
 538
 539                 closeret = close(event->fd);
 540                 if (closeret) {
 541                         PERROR("close event fd");
 542                 }
 543         }
 544 free_event:
 545         free(event);
 546 error:
 547         return ret;
 548 }
 549
 550 /*
 551  * Disable a kernel channel.
 552  */
 553 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 554 {
 555         int ret;
 556
 557         assert(chan);
 558
 559         ret = kernctl_disable(chan->fd);
 560         if (ret < 0) {
 561                 PERROR("disable chan ioctl");
 562                 goto error;
 563         }
 564
 565         chan->enabled = 0;
 566         DBG("Kernel channel %s disabled (fd: %d, key: %" PRIu64 ")",
 567                         chan->channel->name, chan->fd, chan->key);
 568
 569         return 0;
 570
 571 error:
 572         return ret;
 573 }
 574
 575 /*
 576  * Enable a kernel channel.
 577  */
 578 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 579 {
 580         int ret;
 581
 582         assert(chan);
 583
 584         ret = kernctl_enable(chan->fd);
 585         if (ret < 0 && ret != -EEXIST) {
 586                 PERROR("Enable kernel chan");
 587                 goto error;
 588         }
 589
 590         chan->enabled = 1;
 591         DBG("Kernel channel %s enabled (fd: %d, key: %" PRIu64 ")",
 592                         chan->channel->name, chan->fd, chan->key);
 593
 594         return 0;
 595
 596 error:
 597         return ret;
 598 }
 599
 600 /*
 601  * Enable a kernel event.
 602  */
 603 int kernel_enable_event(struct ltt_kernel_event *event)
 604 {
 605         int ret;
 606
 607         assert(event);
 608
 609         ret = kernctl_enable(event->fd);
 610         if (ret < 0) {
 611                 switch (-ret) {
 612                 case EEXIST:
 613                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 614                         break;
 615                 default:
 616                         PERROR("enable kernel event");
 617                         break;
 618                 }
 619                 goto error;
 620         }
 621
 622         event->enabled = 1;
 623         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 624
 625         return 0;
 626
 627 error:
 628         return ret;
 629 }
 630
 631 /*
 632  * Disable a kernel event.
 633  */
 634 int kernel_disable_event(struct ltt_kernel_event *event)
 635 {
 636         int ret;
 637
 638         assert(event);
 639
 640         ret = kernctl_disable(event->fd);
 641         if (ret < 0) {
 642                 switch (-ret) {
 643                 case EEXIST:
 644                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 645                         break;
 646                 default:
 647                         PERROR("disable kernel event");
 648                         break;
 649                 }
 650                 goto error;
 651         }
 652
 653         event->enabled = 0;
 654         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 655
 656         return 0;
 657
 658 error:
 659         return ret;
 660 }
 661
 662
 663 int kernel_track_pid(struct ltt_kernel_session *session, int pid)
 664 {
 665         int ret;
 666
 667         DBG("Kernel track PID %d for session id %" PRIu64 ".",
 668                         pid, session->id);
 669         ret = kernctl_track_pid(session->fd, pid);
 670         if (!ret) {
 671                 return LTTNG_OK;
 672         }
 673         switch (-ret) {
 674         case EINVAL:
 675                 return LTTNG_ERR_INVALID;
 676         case ENOMEM:
 677                 return LTTNG_ERR_NOMEM;
 678         case EEXIST:
 679                 return LTTNG_ERR_PID_TRACKED;
 680         default:
 681                 return LTTNG_ERR_UNK;
 682         }
 683 }
 684
 685 int kernel_untrack_pid(struct ltt_kernel_session *session, int pid)
 686 {
 687         int ret;
 688
 689         DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 690                         pid, session->id);
 691         ret = kernctl_untrack_pid(session->fd, pid);
 692         if (!ret) {
 693                 return LTTNG_OK;
 694         }
 695         switch (-ret) {
 696         case EINVAL:
 697                 return LTTNG_ERR_INVALID;
 698         case ENOMEM:
 699                 return LTTNG_ERR_NOMEM;
 700         case ENOENT:
 701                 return LTTNG_ERR_PID_NOT_TRACKED;
 702         default:
 703                 return LTTNG_ERR_UNK;
 704         }
 705 }
 706
 707 ssize_t kernel_list_tracker_pids(struct ltt_kernel_session *session,
 708                 int **_pids)
 709 {
 710         int fd, ret;
 711         int pid;
 712         ssize_t nbmem, count = 0;
 713         FILE *fp;
 714         int *pids;
 715
 716         fd = kernctl_list_tracker_pids(session->fd);
 717         if (fd < 0) {
 718                 PERROR("kernel tracker pids list");
 719                 goto error;
 720         }
 721
 722         fp = fdopen(fd, "r");
 723         if (fp == NULL) {
 724                 PERROR("kernel tracker pids list fdopen");
 725                 goto error_fp;
 726         }
 727
 728         nbmem = KERNEL_TRACKER_PIDS_INIT_LIST_SIZE;
 729         pids = zmalloc(sizeof(*pids) * nbmem);
 730         if (pids == NULL) {
 731                 PERROR("alloc list pids");
 732                 count = -ENOMEM;
 733                 goto end;
 734         }
 735
 736         while (fscanf(fp, "process { pid = %u; };\n", &pid) == 1) {
 737                 if (count >= nbmem) {
 738                         int *new_pids;
 739                         size_t new_nbmem;
 740
 741                         new_nbmem = nbmem << 1;
 742                         DBG("Reallocating pids list from %zu to %zu entries",
 743                                         nbmem, new_nbmem);
 744                         new_pids = realloc(pids, new_nbmem * sizeof(*new_pids));
 745                         if (new_pids == NULL) {
 746                                 PERROR("realloc list events");
 747                                 free(pids);
 748                                 count = -ENOMEM;
 749                                 goto end;
 750                         }
 751                         /* Zero the new memory */
 752                         memset(new_pids + nbmem, 0,
 753                                 (new_nbmem - nbmem) * sizeof(*new_pids));
 754                         nbmem = new_nbmem;
 755                         pids = new_pids;
 756                 }
 757                 pids[count++] = pid;
 758         }
 759
 760         *_pids = pids;
 761         DBG("Kernel list tracker pids done (%zd pids)", count);
 762 end:
 763         ret = fclose(fp);       /* closes both fp and fd */
 764         if (ret) {
 765                 PERROR("fclose");
 766         }
 767         return count;
 768
 769 error_fp:
 770         ret = close(fd);
 771         if (ret) {
 772                 PERROR("close");
 773         }
 774 error:
 775         return -1;
 776 }
 777
 778 /*
 779  * Create kernel metadata, open from the kernel tracer and add it to the
 780  * kernel session.
 781  */
 782 int kernel_open_metadata(struct ltt_kernel_session *session)
 783 {
 784         int ret;
 785         struct ltt_kernel_metadata *lkm = NULL;
 786
 787         assert(session);
 788
 789         /* Allocate kernel metadata */
 790         lkm = trace_kernel_create_metadata();
 791         if (lkm == NULL) {
 792                 goto error;
 793         }
 794
 795         /* Kernel tracer metadata creation */
 796         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 797         if (ret < 0) {
 798                 goto error_open;
 799         }
 800
 801         lkm->fd = ret;
 802         lkm->key = ++next_kernel_channel_key;
 803         /* Prevent fd duplication after execlp() */
 804         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 805         if (ret < 0) {
 806                 PERROR("fcntl session fd");
 807         }
 808
 809         session->metadata = lkm;
 810
 811         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 812
 813         return 0;
 814
 815 error_open:
 816         trace_kernel_destroy_metadata(lkm);
 817 error:
 818         return -1;
 819 }
 820
 821 /*
 822  * Start tracing session.
 823  */
 824 int kernel_start_session(struct ltt_kernel_session *session)
 825 {
 826         int ret;
 827
 828         assert(session);
 829
 830         ret = kernctl_start_session(session->fd);
 831         if (ret < 0) {
 832                 PERROR("ioctl start session");
 833                 goto error;
 834         }
 835
 836         DBG("Kernel session started");
 837
 838         return 0;
 839
 840 error:
 841         return ret;
 842 }
 843
 844 /*
 845  * Make a kernel wait to make sure in-flight probe have completed.
 846  */
 847 void kernel_wait_quiescent(int fd)
 848 {
 849         int ret;
 850
 851         DBG("Kernel quiescent wait on %d", fd);
 852
 853         ret = kernctl_wait_quiescent(fd);
 854         if (ret < 0) {
 855                 PERROR("wait quiescent ioctl");
 856                 ERR("Kernel quiescent wait failed");
 857         }
 858 }
 859
 860 /*
 861  *  Force flush buffer of metadata.
 862  */
 863 int kernel_metadata_flush_buffer(int fd)
 864 {
 865         int ret;
 866
 867         DBG("Kernel flushing metadata buffer on fd %d", fd);
 868
 869         ret = kernctl_buffer_flush(fd);
 870         if (ret < 0) {
 871                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 872         }
 873
 874         return 0;
 875 }
 876
 877 /*
 878  * Force flush buffer for channel.
 879  */
 880 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 881 {
 882         int ret;
 883         struct ltt_kernel_stream *stream;
 884
 885         assert(channel);
 886
 887         DBG("Flush buffer for channel %s", channel->channel->name);
 888
 889         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 890                 DBG("Flushing channel stream %d", stream->fd);
 891                 ret = kernctl_buffer_flush(stream->fd);
 892                 if (ret < 0) {
 893                         PERROR("ioctl");
 894                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 895                                         stream->fd, ret);
 896                 }
 897         }
 898
 899         return 0;
 900 }
 901
 902 /*
 903  * Stop tracing session.
 904  */
 905 int kernel_stop_session(struct ltt_kernel_session *session)
 906 {
 907         int ret;
 908
 909         assert(session);
 910
 911         ret = kernctl_stop_session(session->fd);
 912         if (ret < 0) {
 913                 goto error;
 914         }
 915
 916         DBG("Kernel session stopped");
 917
 918         return 0;
 919
 920 error:
 921         return ret;
 922 }
 923
 924 /*
 925  * Open stream of channel, register it to the kernel tracer and add it
 926  * to the stream list of the channel.
 927  *
 928  * Note: given that the streams may appear in random order wrt CPU
 929  * number (e.g. cpu hotplug), the index value of the stream number in
 930  * the stream name is not necessarily linked to the CPU number.
 931  *
 932  * Return the number of created stream. Else, a negative value.
 933  */
 934 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 935 {
 936         int ret;
 937         struct ltt_kernel_stream *lks;
 938
 939         assert(channel);
 940
 941         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 942                 lks = trace_kernel_create_stream(channel->channel->name,
 943                                 channel->stream_count);
 944                 if (lks == NULL) {
 945                         ret = close(ret);
 946                         if (ret) {
 947                                 PERROR("close");
 948                         }
 949                         goto error;
 950                 }
 951
 952                 lks->fd = ret;
 953                 /* Prevent fd duplication after execlp() */
 954                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 955                 if (ret < 0) {
 956                         PERROR("fcntl session fd");
 957                 }
 958
 959                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 960                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 961
 962                 /* Add stream to channel stream list */
 963                 cds_list_add(&lks->list, &channel->stream_list.head);
 964                 channel->stream_count++;
 965
 966                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 967                                 lks->state);
 968         }
 969
 970         return channel->stream_count;
 971
 972 error:
 973         return -1;
 974 }
 975
 976 /*
 977  * Open the metadata stream and set it to the kernel session.
 978  */
 979 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 980 {
 981         int ret;
 982
 983         assert(session);
 984
 985         ret = kernctl_create_stream(session->metadata->fd);
 986         if (ret < 0) {
 987                 PERROR("kernel create metadata stream");
 988                 goto error;
 989         }
 990
 991         DBG("Kernel metadata stream created (fd: %d)", ret);
 992         session->metadata_stream_fd = ret;
 993         /* Prevent fd duplication after execlp() */
 994         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 995         if (ret < 0) {
 996                 PERROR("fcntl session fd");
 997         }
 998
 999         return 0;
1000
1001 error:
1002         return -1;
1003 }
1004
1005 /*
1006  * Get the event list from the kernel tracer and return the number of elements.
1007  */
1008 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
1009 {
1010         int fd, ret;
1011         char *event;
1012         size_t nbmem, count = 0;
1013         FILE *fp;
1014         struct lttng_event *elist;
1015
1016         assert(events);
1017
1018         fd = kernctl_tracepoint_list(tracer_fd);
1019         if (fd < 0) {
1020                 PERROR("kernel tracepoint list");
1021                 goto error;
1022         }
1023
1024         fp = fdopen(fd, "r");
1025         if (fp == NULL) {
1026                 PERROR("kernel tracepoint list fdopen");
1027                 goto error_fp;
1028         }
1029
1030         /*
1031          * Init memory size counter
1032          * See kernel-ctl.h for explanation of this value
1033          */
1034         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
1035         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
1036         if (elist == NULL) {
1037                 PERROR("alloc list events");
1038                 count = -ENOMEM;
1039                 goto end;
1040         }
1041
1042         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
1043                 if (count >= nbmem) {
1044                         struct lttng_event *new_elist;
1045                         size_t new_nbmem;
1046
1047                         new_nbmem = nbmem << 1;
1048                         DBG("Reallocating event list from %zu to %zu bytes",
1049                                         nbmem, new_nbmem);
1050                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
1051                         if (new_elist == NULL) {
1052                                 PERROR("realloc list events");
1053                                 free(event);
1054                                 free(elist);
1055                                 count = -ENOMEM;
1056                                 goto end;
1057                         }
1058                         /* Zero the new memory */
1059                         memset(new_elist + nbmem, 0,
1060                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
1061                         nbmem = new_nbmem;
1062                         elist = new_elist;
1063                 }
1064                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
1065                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
1066                 elist[count].enabled = -1;
1067                 count++;
1068                 free(event);
1069         }
1070
1071         *events = elist;
1072         DBG("Kernel list events done (%zu events)", count);
1073 end:
1074         ret = fclose(fp);       /* closes both fp and fd */
1075         if (ret) {
1076                 PERROR("fclose");
1077         }
1078         return count;
1079
1080 error_fp:
1081         ret = close(fd);
1082         if (ret) {
1083                 PERROR("close");
1084         }
1085 error:
1086         return -1;
1087 }
1088
1089 /*
1090  * Get kernel version and validate it.
1091  */
1092 int kernel_validate_version(int tracer_fd,
1093                 struct lttng_kernel_tracer_version *version,
1094                 struct lttng_kernel_tracer_abi_version *abi_version)
1095 {
1096         int ret;
1097
1098         ret = kernctl_tracer_version(tracer_fd, version);
1099         if (ret < 0) {
1100                 ERR("Failed to retrieve the lttng-modules version");
1101                 goto error;
1102         }
1103
1104         /* Validate version */
1105         if (version->major != VERSION_MAJOR) {
1106                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
1107                         version->major, VERSION_MAJOR);
1108                 goto error_version;
1109         }
1110         ret = kernctl_tracer_abi_version(tracer_fd, abi_version);
1111         if (ret < 0) {
1112                 ERR("Failed to retrieve lttng-modules ABI version");
1113                 goto error;
1114         }
1115         if (abi_version->major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
1116                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
1117                         abi_version->major, abi_version->minor,
1118                         LTTNG_MODULES_ABI_MAJOR_VERSION);
1119                 goto error;
1120         }
1121         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1122                         version->major, version->minor,
1123                         abi_version->major, abi_version->minor);
1124         return 0;
1125
1126 error_version:
1127         ret = -1;
1128
1129 error:
1130         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1131         return ret;
1132 }
1133
1134 /*
1135  * Kernel work-arounds called at the start of sessiond main().
1136  */
1137 int init_kernel_workarounds(void)
1138 {
1139         int ret;
1140         FILE *fp;
1141
1142         /*
1143          * boot_id needs to be read once before being used concurrently
1144          * to deal with a Linux kernel race. A fix is proposed for
1145          * upstream, but the work-around is needed for older kernels.
1146          */
1147         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1148         if (!fp) {
1149                 goto end_boot_id;
1150         }
1151         while (!feof(fp)) {
1152                 char buf[37] = "";
1153
1154                 ret = fread(buf, 1, sizeof(buf), fp);
1155                 if (ret < 0) {
1156                         /* Ignore error, we don't really care */
1157                 }
1158         }
1159         ret = fclose(fp);
1160         if (ret) {
1161                 PERROR("fclose");
1162         }
1163 end_boot_id:
1164         return 0;
1165 }
1166
1167 /*
1168  * Complete teardown of a kernel session.
1169  */
1170 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1171 {
1172         if (ksess == NULL) {
1173                 DBG3("No kernel session when tearing down session");
1174                 return;
1175         }
1176
1177         DBG("Tearing down kernel session");
1178
1179         /*
1180          * Destroy channels on the consumer if at least one FD has been sent and we
1181          * are in no output mode because the streams are in *no* monitor mode so we
1182          * have to send a command to clean them up or else they leaked.
1183          */
1184         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1185                 int ret;
1186                 struct consumer_socket *socket;
1187                 struct lttng_ht_iter iter;
1188
1189                 /* For each consumer socket. */
1190                 rcu_read_lock();
1191                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1192                                 socket, node.node) {
1193                         struct ltt_kernel_channel *chan;
1194
1195                         /* For each channel, ask the consumer to destroy it. */
1196                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1197                                 ret = kernel_consumer_destroy_channel(socket, chan);
1198                                 if (ret < 0) {
1199                                         /* Consumer is probably dead. Use next socket. */
1200                                         continue;
1201                                 }
1202                         }
1203                 }
1204                 rcu_read_unlock();
1205         }
1206
1207         /* Close any relayd session */
1208         consumer_output_send_destroy_relayd(ksess->consumer);
1209
1210         trace_kernel_destroy_session(ksess);
1211 }
1212
1213 /*
1214  * Destroy a kernel channel object. It does not do anything on the tracer side.
1215  */
1216 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1217 {
1218         struct ltt_kernel_session *ksess = NULL;
1219
1220         assert(kchan);
1221         assert(kchan->channel);
1222
1223         DBG3("Kernel destroy channel %s", kchan->channel->name);
1224
1225         /* Update channel count of associated session. */
1226         if (kchan->session) {
1227                 /* Keep pointer reference so we can update it after the destroy. */
1228                 ksess = kchan->session;
1229         }
1230
1231         trace_kernel_destroy_channel(kchan);
1232
1233         /*
1234          * At this point the kernel channel is not visible anymore. This is safe
1235          * since in order to work on a visible kernel session, the tracing session
1236          * lock (ltt_session.lock) MUST be acquired.
1237          */
1238         if (ksess) {
1239                 ksess->channel_count--;
1240         }
1241 }
1242
1243 /*
1244  * Take a snapshot for a given kernel session.
1245  *
1246  * Return LTTNG_OK on success or else return a LTTNG_ERR code.
1247  */
1248 enum lttng_error_code kernel_snapshot_record(struct ltt_kernel_session *ksess,
1249                 struct snapshot_output *output, int wait,
1250                 uint64_t nb_packets_per_stream)
1251 {
1252         int err, ret, saved_metadata_fd;
1253         enum lttng_error_code status = LTTNG_OK;
1254         struct consumer_socket *socket;
1255         struct lttng_ht_iter iter;
1256         struct ltt_kernel_metadata *saved_metadata;
1257         struct ltt_session *session = NULL;
1258         uint64_t trace_archive_id;
1259
1260         assert(ksess);
1261         assert(ksess->consumer);
1262         assert(output);
1263
1264         DBG("Kernel snapshot record started");
1265
1266         session = session_find_by_id(ksess->id);
1267         assert(session);
1268         assert(pthread_mutex_trylock(&session->lock));
1269         assert(session_trylock_list());
1270         trace_archive_id = session->current_archive_id;
1271
1272         /* Save current metadata since the following calls will change it. */
1273         saved_metadata = ksess->metadata;
1274         saved_metadata_fd = ksess->metadata_stream_fd;
1275
1276         rcu_read_lock();
1277
1278         ret = kernel_open_metadata(ksess);
1279         if (ret < 0) {
1280                 status = LTTNG_ERR_KERN_META_FAIL;
1281                 goto error;
1282         }
1283
1284         ret = kernel_open_metadata_stream(ksess);
1285         if (ret < 0) {
1286                 status = LTTNG_ERR_KERN_META_FAIL;
1287                 goto error_open_stream;
1288         }
1289
1290         /* Send metadata to consumer and snapshot everything. */
1291         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1292                         socket, node.node) {
1293                 struct consumer_output *saved_output;
1294                 struct ltt_kernel_channel *chan;
1295
1296                 /*
1297                  * Temporarly switch consumer output for our snapshot output. As long
1298                  * as the session lock is taken, this is safe.
1299                  */
1300                 saved_output = ksess->consumer;
1301                 ksess->consumer = output->consumer;
1302
1303                 pthread_mutex_lock(socket->lock);
1304                 /* This stream must not be monitored by the consumer. */
1305                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1306                 pthread_mutex_unlock(socket->lock);
1307                 /* Put back the saved consumer output into the session. */
1308                 ksess->consumer = saved_output;
1309                 if (ret < 0) {
1310                         status = LTTNG_ERR_KERN_META_FAIL;
1311                         goto error_consumer;
1312                 }
1313
1314                 /* For each channel, ask the consumer to snapshot it. */
1315                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1316                         status = consumer_snapshot_channel(socket, chan->key, output, 0,
1317                                         ksess->uid, ksess->gid,
1318                                         DEFAULT_KERNEL_TRACE_DIR, wait,
1319                                         nb_packets_per_stream,
1320                                         trace_archive_id);
1321                         if (status != LTTNG_OK) {
1322                                 (void) kernel_consumer_destroy_metadata(socket,
1323                                                 ksess->metadata);
1324                                 goto error_consumer;
1325                         }
1326                 }
1327
1328                 /* Snapshot metadata, */
1329                 status = consumer_snapshot_channel(socket, ksess->metadata->key, output,
1330                                 1, ksess->uid, ksess->gid,
1331                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0,
1332                                 trace_archive_id);
1333                 if (status != LTTNG_OK) {
1334                         goto error_consumer;
1335                 }
1336
1337                 /*
1338                  * The metadata snapshot is done, ask the consumer to destroy it since
1339                  * it's not monitored on the consumer side.
1340                  */
1341                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1342         }
1343
1344 error_consumer:
1345         /* Close newly opened metadata stream. It's now on the consumer side. */
1346         err = close(ksess->metadata_stream_fd);
1347         if (err < 0) {
1348                 PERROR("close snapshot kernel");
1349         }
1350
1351 error_open_stream:
1352         trace_kernel_destroy_metadata(ksess->metadata);
1353 error:
1354         /* Restore metadata state.*/
1355         ksess->metadata = saved_metadata;
1356         ksess->metadata_stream_fd = saved_metadata_fd;
1357         if (session) {
1358                 session_put(session);
1359         }
1360         rcu_read_unlock();
1361         return status;
1362 }
1363
1364 /*
1365  * Get the syscall mask array from the kernel tracer.
1366  *
1367  * Return 0 on success else a negative value. In both case, syscall_mask should
1368  * be freed.
1369  */
1370 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1371 {
1372         assert(syscall_mask);
1373         assert(nr_bits);
1374
1375         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1376 }
1377
1378 /*
1379  * Check for the support of the RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS via abi
1380  * version number.
1381  *
1382  * Return 1 on success, 0 when feature is not supported, negative value in case
1383  * of errors.
1384  */
1385 int kernel_supports_ring_buffer_snapshot_sample_positions(int tracer_fd)
1386 {
1387         int ret = 0; // Not supported by default
1388         struct lttng_kernel_tracer_abi_version abi;
1389
1390         ret = kernctl_tracer_abi_version(tracer_fd, &abi);
1391         if (ret < 0) {
1392                 ERR("Failed to retrieve lttng-modules ABI version");
1393                 goto error;
1394         }
1395
1396         /*
1397          * RING_BUFFER_SNAPSHOT_SAMPLE_POSITIONS was introduced in 2.3
1398          */
1399         if (abi.major >= 2 && abi.minor >= 3) {
1400                 /* Supported */
1401                 ret = 1;
1402         } else {
1403                 /* Not supported */
1404                 ret = 0;
1405         }
1406 error:
1407         return ret;
1408 }
1409
1410 /*
1411  * Rotate a kernel session.
1412  *
1413  * Return LTTNG_OK on success or else an LTTng error code.
1414  */
1415 enum lttng_error_code kernel_rotate_session(struct ltt_session *session)
1416 {
1417         int ret;
1418         enum lttng_error_code status = LTTNG_OK;
1419         struct consumer_socket *socket;
1420         struct lttng_ht_iter iter;
1421         struct ltt_kernel_session *ksess = session->kernel_session;
1422
1423         assert(ksess);
1424         assert(ksess->consumer);
1425
1426         DBG("Rotate kernel session %s started (session %" PRIu64 ")",
1427                         session->name, session->id);
1428
1429         rcu_read_lock();
1430
1431         /*
1432          * Note that this loop will end after one iteration given that there is
1433          * only one kernel consumer.
1434          */
1435         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1436                         socket, node.node) {
1437                 struct ltt_kernel_channel *chan;
1438
1439                 /* For each channel, ask the consumer to rotate it. */
1440                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1441                         DBG("Rotate kernel channel %" PRIu64 ", session %s",
1442                                         chan->key, session->name);
1443                         ret = consumer_rotate_channel(socket, chan->key,
1444                                         ksess->uid, ksess->gid, ksess->consumer,
1445                                         ksess->consumer->domain_subdir,
1446                                         /* is_metadata_channel */ false,
1447                                         session->current_archive_id);
1448                         if (ret < 0) {
1449                                 status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1450                                 goto error;
1451                         }
1452                 }
1453
1454                 /*
1455                  * Rotate the metadata channel.
1456                  */
1457                 ret = consumer_rotate_channel(socket, ksess->metadata->key,
1458                                 ksess->uid, ksess->gid, ksess->consumer,
1459                                 ksess->consumer->domain_subdir,
1460                                 /* is_metadata_channel */ true,
1461                                 session->current_archive_id);
1462                 if (ret < 0) {
1463                         status = LTTNG_ERR_KERN_CONSUMER_FAIL;
1464                         goto error;
1465                 }
1466         }
1467
1468 error:
1469         rcu_read_unlock();
1470         return status;
1471 }