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
  37 /*
  38  * Add context on a kernel channel.
  39  *
  40  * Assumes the ownership of ctx.
  41  */
  42 int kernel_add_channel_context(struct ltt_kernel_channel *chan,
  43                 struct ltt_kernel_context *ctx)
  44 {
  45         int ret;
  46
  47         assert(chan);
  48         assert(ctx);
  49
  50         DBG("Adding context to channel %s", chan->channel->name);
  51         ret = kernctl_add_context(chan->fd, &ctx->ctx);
  52         if (ret < 0) {
  53                 switch (-ret) {
  54                 case ENOSYS:
  55                         /* Exists but not available for this kernel */
  56                         ret = LTTNG_ERR_KERN_CONTEXT_UNAVAILABLE;
  57                         goto error;
  58                 case EEXIST:
  59                         /* If EEXIST, we just ignore the error */
  60                         ret = 0;
  61                         goto end;
  62                 default:
  63                         PERROR("add context ioctl");
  64                         ret = LTTNG_ERR_KERN_CONTEXT_FAIL;
  65                         goto error;
  66                 }
  67         }
  68
  69 end:
  70         cds_list_add_tail(&ctx->list, &chan->ctx_list);
  71         ctx->in_list = true;
  72         ctx = NULL;
  73 error:
  74         if (ctx) {
  75                 trace_kernel_destroy_context(ctx);
  76         }
  77         return ret;
  78 }
  79
  80 /*
  81  * Create a new kernel session, register it to the kernel tracer and add it to
  82  * the session daemon session.
  83  */
  84 int kernel_create_session(struct ltt_session *session, int tracer_fd)
  85 {
  86         int ret;
  87         struct ltt_kernel_session *lks;
  88
  89         assert(session);
  90
  91         /* Allocate data structure */
  92         lks = trace_kernel_create_session();
  93         if (lks == NULL) {
  94                 ret = -1;
  95                 goto error;
  96         }
  97
  98         /* Kernel tracer session creation */
  99         ret = kernctl_create_session(tracer_fd);
 100         if (ret < 0) {
 101                 PERROR("ioctl kernel create session");
 102                 goto error;
 103         }
 104
 105         lks->fd = ret;
 106         /* Prevent fd duplication after execlp() */
 107         ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 108         if (ret < 0) {
 109                 PERROR("fcntl session fd");
 110         }
 111
 112         lks->id = session->id;
 113         lks->consumer_fds_sent = 0;
 114         session->kernel_session = lks;
 115
 116         DBG("Kernel session created (fd: %d)", lks->fd);
 117
 118         return 0;
 119
 120 error:
 121         if (lks) {
 122                 trace_kernel_destroy_session(lks);
 123         }
 124         return ret;
 125 }
 126
 127 /*
 128  * Create a kernel channel, register it to the kernel tracer and add it to the
 129  * kernel session.
 130  */
 131 int kernel_create_channel(struct ltt_kernel_session *session,
 132                 struct lttng_channel *chan)
 133 {
 134         int ret;
 135         struct ltt_kernel_channel *lkc;
 136
 137         assert(session);
 138         assert(chan);
 139
 140         /* Allocate kernel channel */
 141         lkc = trace_kernel_create_channel(chan);
 142         if (lkc == NULL) {
 143                 goto error;
 144         }
 145
 146         DBG3("Kernel create channel %s with attr: %d, %" PRIu64 ", %" PRIu64 ", %u, %u, %d, %d",
 147                         chan->name, lkc->channel->attr.overwrite,
 148                         lkc->channel->attr.subbuf_size, lkc->channel->attr.num_subbuf,
 149                         lkc->channel->attr.switch_timer_interval, lkc->channel->attr.read_timer_interval,
 150                         lkc->channel->attr.live_timer_interval, lkc->channel->attr.output);
 151
 152         /* Kernel tracer channel creation */
 153         ret = kernctl_create_channel(session->fd, &lkc->channel->attr);
 154         if (ret < 0) {
 155                 PERROR("ioctl kernel create channel");
 156                 goto error;
 157         }
 158
 159         /* Setup the channel fd */
 160         lkc->fd = ret;
 161         /* Prevent fd duplication after execlp() */
 162         ret = fcntl(lkc->fd, F_SETFD, FD_CLOEXEC);
 163         if (ret < 0) {
 164                 PERROR("fcntl session fd");
 165         }
 166
 167         /* Add channel to session */
 168         cds_list_add(&lkc->list, &session->channel_list.head);
 169         session->channel_count++;
 170         lkc->session = session;
 171
 172         DBG("Kernel channel %s created (fd: %d)", lkc->channel->name, lkc->fd);
 173
 174         return 0;
 175
 176 error:
 177         if (lkc) {
 178                 free(lkc->channel);
 179                 free(lkc);
 180         }
 181         return -1;
 182 }
 183
 184 /*
 185  * Create a kernel event, enable it to the kernel tracer and add it to the
 186  * channel event list of the kernel session.
 187  * We own filter_expression and filter.
 188  */
 189 int kernel_create_event(struct lttng_event *ev,
 190                 struct ltt_kernel_channel *channel,
 191                 char *filter_expression,
 192                 struct lttng_filter_bytecode *filter)
 193 {
 194         int ret;
 195         struct ltt_kernel_event *event;
 196
 197         assert(ev);
 198         assert(channel);
 199
 200         /* We pass ownership of filter_expression and filter */
 201         event = trace_kernel_create_event(ev, filter_expression,
 202                         filter);
 203         if (event == NULL) {
 204                 ret = -1;
 205                 goto error;
 206         }
 207
 208         ret = kernctl_create_event(channel->fd, event->event);
 209         if (ret < 0) {
 210                 switch (-ret) {
 211                 case EEXIST:
 212                         break;
 213                 case ENOSYS:
 214                         WARN("Event type not implemented");
 215                         break;
 216                 case ENOENT:
 217                         WARN("Event %s not found!", ev->name);
 218                         break;
 219                 default:
 220                         PERROR("create event ioctl");
 221                 }
 222                 goto free_event;
 223         }
 224
 225         event->type = ev->type;
 226         event->fd = ret;
 227         /* Prevent fd duplication after execlp() */
 228         ret = fcntl(event->fd, F_SETFD, FD_CLOEXEC);
 229         if (ret < 0) {
 230                 PERROR("fcntl session fd");
 231         }
 232
 233         if (filter) {
 234                 ret = kernctl_filter(event->fd, filter);
 235                 if (ret) {
 236                         goto filter_error;
 237                 }
 238         }
 239
 240         ret = kernctl_enable(event->fd);
 241         if (ret < 0) {
 242                 switch (-ret) {
 243                 case EEXIST:
 244                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 245                         break;
 246                 default:
 247                         PERROR("enable kernel event");
 248                         break;
 249                 }
 250                 goto enable_error;
 251         }
 252
 253         /* Add event to event list */
 254         cds_list_add(&event->list, &channel->events_list.head);
 255         channel->event_count++;
 256
 257         DBG("Event %s created (fd: %d)", ev->name, event->fd);
 258
 259         return 0;
 260
 261 enable_error:
 262 filter_error:
 263         {
 264                 int closeret;
 265
 266                 closeret = close(event->fd);
 267                 if (closeret) {
 268                         PERROR("close event fd");
 269                 }
 270         }
 271 free_event:
 272         free(event);
 273 error:
 274         return ret;
 275 }
 276
 277 /*
 278  * Disable a kernel channel.
 279  */
 280 int kernel_disable_channel(struct ltt_kernel_channel *chan)
 281 {
 282         int ret;
 283
 284         assert(chan);
 285
 286         ret = kernctl_disable(chan->fd);
 287         if (ret < 0) {
 288                 PERROR("disable chan ioctl");
 289                 goto error;
 290         }
 291
 292         chan->enabled = 0;
 293         DBG("Kernel channel %s disabled (fd: %d)", chan->channel->name, chan->fd);
 294
 295         return 0;
 296
 297 error:
 298         return ret;
 299 }
 300
 301 /*
 302  * Enable a kernel channel.
 303  */
 304 int kernel_enable_channel(struct ltt_kernel_channel *chan)
 305 {
 306         int ret;
 307
 308         assert(chan);
 309
 310         ret = kernctl_enable(chan->fd);
 311         if (ret < 0 && ret != -EEXIST) {
 312                 PERROR("Enable kernel chan");
 313                 goto error;
 314         }
 315
 316         chan->enabled = 1;
 317         DBG("Kernel channel %s enabled (fd: %d)", chan->channel->name, chan->fd);
 318
 319         return 0;
 320
 321 error:
 322         return ret;
 323 }
 324
 325 /*
 326  * Enable a kernel event.
 327  */
 328 int kernel_enable_event(struct ltt_kernel_event *event)
 329 {
 330         int ret;
 331
 332         assert(event);
 333
 334         ret = kernctl_enable(event->fd);
 335         if (ret < 0) {
 336                 switch (-ret) {
 337                 case EEXIST:
 338                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 339                         break;
 340                 default:
 341                         PERROR("enable kernel event");
 342                         break;
 343                 }
 344                 goto error;
 345         }
 346
 347         event->enabled = 1;
 348         DBG("Kernel event %s enabled (fd: %d)", event->event->name, event->fd);
 349
 350         return 0;
 351
 352 error:
 353         return ret;
 354 }
 355
 356 /*
 357  * Disable a kernel event.
 358  */
 359 int kernel_disable_event(struct ltt_kernel_event *event)
 360 {
 361         int ret;
 362
 363         assert(event);
 364
 365         ret = kernctl_disable(event->fd);
 366         if (ret < 0) {
 367                 switch (-ret) {
 368                 case EEXIST:
 369                         ret = LTTNG_ERR_KERN_EVENT_EXIST;
 370                         break;
 371                 default:
 372                         PERROR("disable kernel event");
 373                         break;
 374                 }
 375                 goto error;
 376         }
 377
 378         event->enabled = 0;
 379         DBG("Kernel event %s disabled (fd: %d)", event->event->name, event->fd);
 380
 381         return 0;
 382
 383 error:
 384         return ret;
 385 }
 386
 387 static
 388 struct lttng_tracker_list *get_id_tracker_list(struct ltt_kernel_session *session,
 389                 enum lttng_tracker_type tracker_type)
 390 {
 391         switch (tracker_type) {
 392         case LTTNG_TRACKER_PID:
 393                 return session->tracker_list_pid;
 394         case LTTNG_TRACKER_VPID:
 395                 return session->tracker_list_vpid;
 396         case LTTNG_TRACKER_UID:
 397                 return session->tracker_list_uid;
 398         case LTTNG_TRACKER_VUID:
 399                 return session->tracker_list_vuid;
 400         case LTTNG_TRACKER_GID:
 401                 return session->tracker_list_gid;
 402         case LTTNG_TRACKER_VGID:
 403                 return session->tracker_list_vgid;
 404         default:
 405                 return NULL;
 406         }
 407 }
 408
 409 int kernel_track_id(enum lttng_tracker_type tracker_type,
 410                 struct ltt_kernel_session *session,
 411                 struct lttng_tracker_id *id)
 412 {
 413         int ret, value;
 414         struct lttng_tracker_list *tracker_list;
 415         struct lttng_tracker_id *saved_ids;
 416         ssize_t saved_ids_count, i;
 417
 418         ret = lttng_tracker_id_lookup_string(tracker_type,
 419                         id, &value);
 420         if (ret != LTTNG_OK) {
 421                 return ret;
 422         }
 423
 424         tracker_list = get_id_tracker_list(session, tracker_type);
 425         if (!tracker_list) {
 426                 return LTTNG_ERR_INVALID;
 427         }
 428         /* Save list for restore on error. */
 429         saved_ids_count = lttng_tracker_id_get_list(tracker_list, &saved_ids);
 430         if (saved_ids_count < 0) {
 431                 return LTTNG_ERR_INVALID;
 432         }
 433         /* Add to list. */
 434         ret = lttng_tracker_list_add(tracker_list, id);
 435         if (ret != LTTNG_OK) {
 436                 goto end;
 437         }
 438
 439         switch (tracker_type) {
 440         case LTTNG_TRACKER_PID:
 441                 DBG("Kernel track PID %d for session id %" PRIu64 ".",
 442                                 value, session->id);
 443                 ret = kernctl_track_pid(session->fd, value);
 444                 if (!ret) {
 445                         ret = LTTNG_OK;
 446                         goto end;
 447                 }
 448                 break;
 449         case LTTNG_TRACKER_VPID:
 450                 DBG("Kernel track VPID %d for session id %" PRIu64 ".",
 451                                 value, session->id);
 452                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VPID, value);
 453                 if (!ret) {
 454                         ret = LTTNG_OK;
 455                         goto end;
 456                 }
 457                 break;
 458         case LTTNG_TRACKER_UID:
 459                 DBG("Kernel track UID %d for session id %" PRIu64 ".",
 460                                 value, session->id);
 461                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_UID, value);
 462                 if (!ret) {
 463                         ret = LTTNG_OK;
 464                         goto end;
 465                 }
 466                 break;
 467         case LTTNG_TRACKER_GID:
 468                 DBG("Kernel track GID %d for session id %" PRIu64 ".",
 469                                 value, session->id);
 470                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_GID, value);
 471                 if (!ret) {
 472                         ret = LTTNG_OK;
 473                         goto end;
 474                 }
 475                 break;
 476         case LTTNG_TRACKER_VUID:
 477                 DBG("Kernel track VUID %d for session id %" PRIu64 ".",
 478                                 value, session->id);
 479                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VUID, value);
 480                 if (!ret) {
 481                         ret = LTTNG_OK;
 482                         goto end;
 483                 }
 484                 break;
 485         case LTTNG_TRACKER_VGID:
 486                 DBG("Kernel track VGID %d for session id %" PRIu64 ".",
 487                                 value, session->id);
 488                 ret = kernctl_track_id(session->fd, LTTNG_TRACKER_VGID, value);
 489                 if (!ret) {
 490                         ret = LTTNG_OK;
 491                         goto end;
 492                 }
 493                 break;
 494         default:
 495                 ret = -EINVAL;
 496                 break;
 497         }
 498         /* Error handling. */
 499         switch (-ret) {
 500         case EINVAL:
 501                 ret = LTTNG_ERR_INVALID;
 502                 break;
 503         case ENOMEM:
 504                 ret = LTTNG_ERR_NOMEM;
 505                 break;
 506         case EEXIST:
 507                 ret = LTTNG_ERR_ID_TRACKED;
 508                 break;
 509         default:
 510                 ret = LTTNG_ERR_UNK;
 511                 break;
 512         }
 513         if (lttng_tracker_id_set_list(tracker_list, saved_ids, saved_ids_count) != LTTNG_OK) {
 514                 ERR("Error on tracker add error handling.\n");
 515         }
 516 end:
 517         for (i = 0; i < saved_ids_count; i++) {
 518                 free(saved_ids[i].string);
 519         }
 520         free(saved_ids);
 521         return ret;
 522 }
 523
 524 int kernel_untrack_id(enum lttng_tracker_type tracker_type,
 525                 struct ltt_kernel_session *session,
 526                 struct lttng_tracker_id *id)
 527 {
 528         int ret, value;
 529         struct lttng_tracker_list *tracker_list;
 530         struct lttng_tracker_id *saved_ids;
 531         ssize_t saved_ids_count, i;
 532
 533         ret = lttng_tracker_id_lookup_string(tracker_type,
 534                         id, &value);
 535         if (ret != LTTNG_OK) {
 536                 return ret;
 537         }
 538
 539         tracker_list = get_id_tracker_list(session, tracker_type);
 540         if (!tracker_list) {
 541                 return LTTNG_ERR_INVALID;
 542         }
 543         /* Save list for restore on error. */
 544         saved_ids_count = lttng_tracker_id_get_list(tracker_list, &saved_ids);
 545         if (saved_ids_count < 0) {
 546                 return LTTNG_ERR_INVALID;
 547         }
 548         /* Remove from list. */
 549         ret = lttng_tracker_list_remove(tracker_list, id);
 550         if (ret != LTTNG_OK) {
 551                 goto end;
 552         }
 553
 554         switch (tracker_type) {
 555         case LTTNG_TRACKER_PID:
 556                 DBG("Kernel untrack PID %d for session id %" PRIu64 ".",
 557                                 value, session->id);
 558                 ret = kernctl_untrack_pid(session->fd, value);
 559                 if (!ret) {
 560                         ret = LTTNG_OK;
 561                         goto end;
 562                 }
 563                 break;
 564         case LTTNG_TRACKER_VPID:
 565                 DBG("Kernel untrack VPID %d for session id %" PRIu64 ".",
 566                                 value, session->id);
 567                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_VPID, value);
 568                 if (!ret) {
 569                         ret = LTTNG_OK;
 570                         goto end;
 571                 }
 572                 break;
 573         case LTTNG_TRACKER_UID:
 574                 DBG("Kernel untrack UID %d for session id %" PRIu64 ".",
 575                                 value, session->id);
 576                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_UID, value);
 577                 if (!ret) {
 578                         ret = LTTNG_OK;
 579                         goto end;
 580                 }
 581                 break;
 582         case LTTNG_TRACKER_GID:
 583                 DBG("Kernel untrack GID %d for session id %" PRIu64 ".",
 584                                 value, session->id);
 585                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_GID, value);
 586                 if (!ret) {
 587                         ret = LTTNG_OK;
 588                         goto end;
 589                 }
 590                 break;
 591         case LTTNG_TRACKER_VUID:
 592                 DBG("Kernel untrack VUID %d for session id %" PRIu64 ".",
 593                                 value, session->id);
 594                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_VUID, value);
 595                 if (!ret) {
 596                         ret = LTTNG_OK;
 597                         goto end;
 598                 }
 599                 break;
 600         case LTTNG_TRACKER_VGID:
 601                 DBG("Kernel untrack VGID %d for session id %" PRIu64 ".",
 602                                 value, session->id);
 603                 ret = kernctl_untrack_id(session->fd, LTTNG_TRACKER_VGID, value);
 604                 if (!ret) {
 605                         ret = LTTNG_OK;
 606                         goto end;
 607                 }
 608                 break;
 609         default:
 610                 ret = -EINVAL;
 611                 break;
 612         }
 613
 614         /* Error handling. */
 615         switch (-ret) {
 616         case EINVAL:
 617                 ret = LTTNG_ERR_INVALID;
 618                 break;
 619         case ENOMEM:
 620                 ret = LTTNG_ERR_NOMEM;
 621                 break;
 622         case EEXIST:
 623                 ret = LTTNG_ERR_ID_TRACKED;
 624                 break;
 625         default:
 626                 ret = LTTNG_ERR_UNK;
 627                 break;
 628         }
 629
 630         if (lttng_tracker_id_set_list(tracker_list, saved_ids, saved_ids_count) != LTTNG_OK) {
 631                 ERR("Error on tracker remove error handling.\n");
 632         }
 633 end:
 634         for (i = 0; i < saved_ids_count; i++) {
 635                 free(saved_ids[i].string);
 636         }
 637         free(saved_ids);
 638         return ret;
 639 }
 640
 641 /*
 642  * Called with session lock held.
 643  */
 644 ssize_t kernel_list_tracker_ids(enum lttng_tracker_type tracker_type,
 645                 struct ltt_kernel_session *session,
 646                 struct lttng_tracker_id **_ids)
 647 {
 648         struct lttng_tracker_list *tracker_list;
 649
 650         tracker_list = get_id_tracker_list(session, tracker_type);
 651         if (!tracker_list) {
 652                 return -LTTNG_ERR_INVALID;
 653         }
 654         return lttng_tracker_id_get_list(tracker_list, _ids);
 655 }
 656
 657 /*
 658  * Create kernel metadata, open from the kernel tracer and add it to the
 659  * kernel session.
 660  */
 661 int kernel_open_metadata(struct ltt_kernel_session *session)
 662 {
 663         int ret;
 664         struct ltt_kernel_metadata *lkm = NULL;
 665
 666         assert(session);
 667
 668         /* Allocate kernel metadata */
 669         lkm = trace_kernel_create_metadata();
 670         if (lkm == NULL) {
 671                 goto error;
 672         }
 673
 674         /* Kernel tracer metadata creation */
 675         ret = kernctl_open_metadata(session->fd, &lkm->conf->attr);
 676         if (ret < 0) {
 677                 goto error_open;
 678         }
 679
 680         lkm->fd = ret;
 681         /* Prevent fd duplication after execlp() */
 682         ret = fcntl(lkm->fd, F_SETFD, FD_CLOEXEC);
 683         if (ret < 0) {
 684                 PERROR("fcntl session fd");
 685         }
 686
 687         session->metadata = lkm;
 688
 689         DBG("Kernel metadata opened (fd: %d)", lkm->fd);
 690
 691         return 0;
 692
 693 error_open:
 694         trace_kernel_destroy_metadata(lkm);
 695 error:
 696         return -1;
 697 }
 698
 699 /*
 700  * Start tracing session.
 701  */
 702 int kernel_start_session(struct ltt_kernel_session *session)
 703 {
 704         int ret;
 705
 706         assert(session);
 707
 708         ret = kernctl_start_session(session->fd);
 709         if (ret < 0) {
 710                 PERROR("ioctl start session");
 711                 goto error;
 712         }
 713
 714         DBG("Kernel session started");
 715
 716         return 0;
 717
 718 error:
 719         return ret;
 720 }
 721
 722 /*
 723  * Make a kernel wait to make sure in-flight probe have completed.
 724  */
 725 void kernel_wait_quiescent(int fd)
 726 {
 727         int ret;
 728
 729         DBG("Kernel quiescent wait on %d", fd);
 730
 731         ret = kernctl_wait_quiescent(fd);
 732         if (ret < 0) {
 733                 PERROR("wait quiescent ioctl");
 734                 ERR("Kernel quiescent wait failed");
 735         }
 736 }
 737
 738 /*
 739  *  Force flush buffer of metadata.
 740  */
 741 int kernel_metadata_flush_buffer(int fd)
 742 {
 743         int ret;
 744
 745         DBG("Kernel flushing metadata buffer on fd %d", fd);
 746
 747         ret = kernctl_buffer_flush(fd);
 748         if (ret < 0) {
 749                 ERR("Fail to flush metadata buffers %d (ret: %d)", fd, ret);
 750         }
 751
 752         return 0;
 753 }
 754
 755 /*
 756  * Force flush buffer for channel.
 757  */
 758 int kernel_flush_buffer(struct ltt_kernel_channel *channel)
 759 {
 760         int ret;
 761         struct ltt_kernel_stream *stream;
 762
 763         assert(channel);
 764
 765         DBG("Flush buffer for channel %s", channel->channel->name);
 766
 767         cds_list_for_each_entry(stream, &channel->stream_list.head, list) {
 768                 DBG("Flushing channel stream %d", stream->fd);
 769                 ret = kernctl_buffer_flush(stream->fd);
 770                 if (ret < 0) {
 771                         PERROR("ioctl");
 772                         ERR("Fail to flush buffer for stream %d (ret: %d)",
 773                                         stream->fd, ret);
 774                 }
 775         }
 776
 777         return 0;
 778 }
 779
 780 /*
 781  * Stop tracing session.
 782  */
 783 int kernel_stop_session(struct ltt_kernel_session *session)
 784 {
 785         int ret;
 786
 787         assert(session);
 788
 789         ret = kernctl_stop_session(session->fd);
 790         if (ret < 0) {
 791                 goto error;
 792         }
 793
 794         DBG("Kernel session stopped");
 795
 796         return 0;
 797
 798 error:
 799         return ret;
 800 }
 801
 802 /*
 803  * Open stream of channel, register it to the kernel tracer and add it
 804  * to the stream list of the channel.
 805  *
 806  * Note: given that the streams may appear in random order wrt CPU
 807  * number (e.g. cpu hotplug), the index value of the stream number in
 808  * the stream name is not necessarily linked to the CPU number.
 809  *
 810  * Return the number of created stream. Else, a negative value.
 811  */
 812 int kernel_open_channel_stream(struct ltt_kernel_channel *channel)
 813 {
 814         int ret;
 815         struct ltt_kernel_stream *lks;
 816
 817         assert(channel);
 818
 819         while ((ret = kernctl_create_stream(channel->fd)) >= 0) {
 820                 lks = trace_kernel_create_stream(channel->channel->name,
 821                                 channel->stream_count);
 822                 if (lks == NULL) {
 823                         ret = close(ret);
 824                         if (ret) {
 825                                 PERROR("close");
 826                         }
 827                         goto error;
 828                 }
 829
 830                 lks->fd = ret;
 831                 /* Prevent fd duplication after execlp() */
 832                 ret = fcntl(lks->fd, F_SETFD, FD_CLOEXEC);
 833                 if (ret < 0) {
 834                         PERROR("fcntl session fd");
 835                 }
 836
 837                 lks->tracefile_size = channel->channel->attr.tracefile_size;
 838                 lks->tracefile_count = channel->channel->attr.tracefile_count;
 839
 840                 /* Add stream to channel stream list */
 841                 cds_list_add(&lks->list, &channel->stream_list.head);
 842                 channel->stream_count++;
 843
 844                 DBG("Kernel stream %s created (fd: %d, state: %d)", lks->name, lks->fd,
 845                                 lks->state);
 846         }
 847
 848         return channel->stream_count;
 849
 850 error:
 851         return -1;
 852 }
 853
 854 /*
 855  * Open the metadata stream and set it to the kernel session.
 856  */
 857 int kernel_open_metadata_stream(struct ltt_kernel_session *session)
 858 {
 859         int ret;
 860
 861         assert(session);
 862
 863         ret = kernctl_create_stream(session->metadata->fd);
 864         if (ret < 0) {
 865                 PERROR("kernel create metadata stream");
 866                 goto error;
 867         }
 868
 869         DBG("Kernel metadata stream created (fd: %d)", ret);
 870         session->metadata_stream_fd = ret;
 871         /* Prevent fd duplication after execlp() */
 872         ret = fcntl(session->metadata_stream_fd, F_SETFD, FD_CLOEXEC);
 873         if (ret < 0) {
 874                 PERROR("fcntl session fd");
 875         }
 876
 877         return 0;
 878
 879 error:
 880         return -1;
 881 }
 882
 883 /*
 884  * Get the event list from the kernel tracer and return the number of elements.
 885  */
 886 ssize_t kernel_list_events(int tracer_fd, struct lttng_event **events)
 887 {
 888         int fd, ret;
 889         char *event;
 890         size_t nbmem, count = 0;
 891         FILE *fp;
 892         struct lttng_event *elist;
 893
 894         assert(events);
 895
 896         fd = kernctl_tracepoint_list(tracer_fd);
 897         if (fd < 0) {
 898                 PERROR("kernel tracepoint list");
 899                 goto error;
 900         }
 901
 902         fp = fdopen(fd, "r");
 903         if (fp == NULL) {
 904                 PERROR("kernel tracepoint list fdopen");
 905                 goto error_fp;
 906         }
 907
 908         /*
 909          * Init memory size counter
 910          * See kernel-ctl.h for explanation of this value
 911          */
 912         nbmem = KERNEL_EVENT_INIT_LIST_SIZE;
 913         elist = zmalloc(sizeof(struct lttng_event) * nbmem);
 914         if (elist == NULL) {
 915                 PERROR("alloc list events");
 916                 count = -ENOMEM;
 917                 goto end;
 918         }
 919
 920         while (fscanf(fp, "event { name = %m[^;]; };\n", &event) == 1) {
 921                 if (count >= nbmem) {
 922                         struct lttng_event *new_elist;
 923                         size_t new_nbmem;
 924
 925                         new_nbmem = nbmem << 1;
 926                         DBG("Reallocating event list from %zu to %zu bytes",
 927                                         nbmem, new_nbmem);
 928                         new_elist = realloc(elist, new_nbmem * sizeof(struct lttng_event));
 929                         if (new_elist == NULL) {
 930                                 PERROR("realloc list events");
 931                                 free(event);
 932                                 free(elist);
 933                                 count = -ENOMEM;
 934                                 goto end;
 935                         }
 936                         /* Zero the new memory */
 937                         memset(new_elist + nbmem, 0,
 938                                 (new_nbmem - nbmem) * sizeof(struct lttng_event));
 939                         nbmem = new_nbmem;
 940                         elist = new_elist;
 941                 }
 942                 strncpy(elist[count].name, event, LTTNG_SYMBOL_NAME_LEN);
 943                 elist[count].name[LTTNG_SYMBOL_NAME_LEN - 1] = '\0';
 944                 elist[count].enabled = -1;
 945                 count++;
 946                 free(event);
 947         }
 948
 949         *events = elist;
 950         DBG("Kernel list events done (%zu events)", count);
 951 end:
 952         ret = fclose(fp);       /* closes both fp and fd */
 953         if (ret) {
 954                 PERROR("fclose");
 955         }
 956         return count;
 957
 958 error_fp:
 959         ret = close(fd);
 960         if (ret) {
 961                 PERROR("close");
 962         }
 963 error:
 964         return -1;
 965 }
 966
 967 /*
 968  * Get kernel version and validate it.
 969  */
 970 int kernel_validate_version(int tracer_fd)
 971 {
 972         int ret;
 973         struct lttng_kernel_tracer_version version;
 974         struct lttng_kernel_tracer_abi_version abi_version;
 975
 976         ret = kernctl_tracer_version(tracer_fd, &version);
 977         if (ret < 0) {
 978                 ERR("Failed to retrieve the lttng-modules version");
 979                 goto error;
 980         }
 981
 982         /* Validate version */
 983         if (version.major != VERSION_MAJOR) {
 984                 ERR("Kernel tracer major version (%d) is not compatible with lttng-tools major version (%d)",
 985                         version.major, VERSION_MAJOR);
 986                 goto error_version;
 987         }
 988         ret = kernctl_tracer_abi_version(tracer_fd, &abi_version);
 989         if (ret < 0) {
 990                 ERR("Failed to retrieve lttng-modules ABI version");
 991                 goto error;
 992         }
 993         if (abi_version.major != LTTNG_MODULES_ABI_MAJOR_VERSION) {
 994                 ERR("Kernel tracer ABI version (%d.%d) does not match the expected ABI major version (%d.*)",
 995                         abi_version.major, abi_version.minor,
 996                         LTTNG_MODULES_ABI_MAJOR_VERSION);
 997                 goto error;
 998         }
 999         DBG2("Kernel tracer version validated (%d.%d, ABI %d.%d)",
1000                         version.major, version.minor,
1001                         abi_version.major, abi_version.minor);
1002         return 0;
1003
1004 error_version:
1005         ret = -1;
1006
1007 error:
1008         ERR("Kernel tracer version check failed; kernel tracing will not be available");
1009         return ret;
1010 }
1011
1012 /*
1013  * Kernel work-arounds called at the start of sessiond main().
1014  */
1015 int init_kernel_workarounds(void)
1016 {
1017         int ret;
1018         FILE *fp;
1019
1020         /*
1021          * boot_id needs to be read once before being used concurrently
1022          * to deal with a Linux kernel race. A fix is proposed for
1023          * upstream, but the work-around is needed for older kernels.
1024          */
1025         fp = fopen("/proc/sys/kernel/random/boot_id", "r");
1026         if (!fp) {
1027                 goto end_boot_id;
1028         }
1029         while (!feof(fp)) {
1030                 char buf[37] = "";
1031
1032                 ret = fread(buf, 1, sizeof(buf), fp);
1033                 if (ret < 0) {
1034                         /* Ignore error, we don't really care */
1035                 }
1036         }
1037         ret = fclose(fp);
1038         if (ret) {
1039                 PERROR("fclose");
1040         }
1041 end_boot_id:
1042         return 0;
1043 }
1044
1045 /*
1046  * Complete teardown of a kernel session.
1047  */
1048 void kernel_destroy_session(struct ltt_kernel_session *ksess)
1049 {
1050         if (ksess == NULL) {
1051                 DBG3("No kernel session when tearing down session");
1052                 return;
1053         }
1054
1055         DBG("Tearing down kernel session");
1056
1057         /*
1058          * Destroy channels on the consumer if at least one FD has been sent and we
1059          * are in no output mode because the streams are in *no* monitor mode so we
1060          * have to send a command to clean them up or else they leaked.
1061          */
1062         if (!ksess->output_traces && ksess->consumer_fds_sent) {
1063                 int ret;
1064                 struct consumer_socket *socket;
1065                 struct lttng_ht_iter iter;
1066
1067                 /* For each consumer socket. */
1068                 rcu_read_lock();
1069                 cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1070                                 socket, node.node) {
1071                         struct ltt_kernel_channel *chan;
1072
1073                         /* For each channel, ask the consumer to destroy it. */
1074                         cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1075                                 ret = kernel_consumer_destroy_channel(socket, chan);
1076                                 if (ret < 0) {
1077                                         /* Consumer is probably dead. Use next socket. */
1078                                         continue;
1079                                 }
1080                         }
1081                 }
1082                 rcu_read_unlock();
1083         }
1084
1085         /* Close any relayd session */
1086         consumer_output_send_destroy_relayd(ksess->consumer);
1087
1088         trace_kernel_destroy_session(ksess);
1089 }
1090
1091 /*
1092  * Destroy a kernel channel object. It does not do anything on the tracer side.
1093  */
1094 void kernel_destroy_channel(struct ltt_kernel_channel *kchan)
1095 {
1096         struct ltt_kernel_session *ksess = NULL;
1097
1098         assert(kchan);
1099         assert(kchan->channel);
1100
1101         DBG3("Kernel destroy channel %s", kchan->channel->name);
1102
1103         /* Update channel count of associated session. */
1104         if (kchan->session) {
1105                 /* Keep pointer reference so we can update it after the destroy. */
1106                 ksess = kchan->session;
1107         }
1108
1109         trace_kernel_destroy_channel(kchan);
1110
1111         /*
1112          * At this point the kernel channel is not visible anymore. This is safe
1113          * since in order to work on a visible kernel session, the tracing session
1114          * lock (ltt_session.lock) MUST be acquired.
1115          */
1116         if (ksess) {
1117                 ksess->channel_count--;
1118         }
1119 }
1120
1121 /*
1122  * Take a snapshot for a given kernel session.
1123  *
1124  * Return 0 on success or else return a LTTNG_ERR code.
1125  */
1126 int kernel_snapshot_record(struct ltt_kernel_session *ksess,
1127                 struct snapshot_output *output, int wait,
1128                 uint64_t nb_packets_per_stream)
1129 {
1130         int err, ret, saved_metadata_fd;
1131         struct consumer_socket *socket;
1132         struct lttng_ht_iter iter;
1133         struct ltt_kernel_metadata *saved_metadata;
1134
1135         assert(ksess);
1136         assert(ksess->consumer);
1137         assert(output);
1138
1139         DBG("Kernel snapshot record started");
1140
1141         /* Save current metadata since the following calls will change it. */
1142         saved_metadata = ksess->metadata;
1143         saved_metadata_fd = ksess->metadata_stream_fd;
1144
1145         rcu_read_lock();
1146
1147         ret = kernel_open_metadata(ksess);
1148         if (ret < 0) {
1149                 ret = LTTNG_ERR_KERN_META_FAIL;
1150                 goto error;
1151         }
1152
1153         ret = kernel_open_metadata_stream(ksess);
1154         if (ret < 0) {
1155                 ret = LTTNG_ERR_KERN_META_FAIL;
1156                 goto error_open_stream;
1157         }
1158
1159         /* Send metadata to consumer and snapshot everything. */
1160         cds_lfht_for_each_entry(ksess->consumer->socks->ht, &iter.iter,
1161                         socket, node.node) {
1162                 struct consumer_output *saved_output;
1163                 struct ltt_kernel_channel *chan;
1164
1165                 /*
1166                  * Temporarly switch consumer output for our snapshot output. As long
1167                  * as the session lock is taken, this is safe.
1168                  */
1169                 saved_output = ksess->consumer;
1170                 ksess->consumer = output->consumer;
1171
1172                 pthread_mutex_lock(socket->lock);
1173                 /* This stream must not be monitored by the consumer. */
1174                 ret = kernel_consumer_add_metadata(socket, ksess, 0);
1175                 pthread_mutex_unlock(socket->lock);
1176                 /* Put back the saved consumer output into the session. */
1177                 ksess->consumer = saved_output;
1178                 if (ret < 0) {
1179                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1180                         goto error_consumer;
1181                 }
1182
1183                 /* For each channel, ask the consumer to snapshot it. */
1184                 cds_list_for_each_entry(chan, &ksess->channel_list.head, list) {
1185                         ret = consumer_snapshot_channel(socket, chan->fd, output, 0,
1186                                         ksess->uid, ksess->gid,
1187                                         DEFAULT_KERNEL_TRACE_DIR, wait,
1188                                         nb_packets_per_stream);
1189                         if (ret < 0) {
1190                                 ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1191                                 (void) kernel_consumer_destroy_metadata(socket,
1192                                                 ksess->metadata);
1193                                 goto error_consumer;
1194                         }
1195                 }
1196
1197                 /* Snapshot metadata, */
1198                 ret = consumer_snapshot_channel(socket, ksess->metadata->fd, output,
1199                                 1, ksess->uid, ksess->gid,
1200                                 DEFAULT_KERNEL_TRACE_DIR, wait, 0);
1201                 if (ret < 0) {
1202                         ret = LTTNG_ERR_KERN_CONSUMER_FAIL;
1203                         goto error_consumer;
1204                 }
1205
1206                 /*
1207                  * The metadata snapshot is done, ask the consumer to destroy it since
1208                  * it's not monitored on the consumer side.
1209                  */
1210                 (void) kernel_consumer_destroy_metadata(socket, ksess->metadata);
1211         }
1212
1213         ret = LTTNG_OK;
1214
1215 error_consumer:
1216         /* Close newly opened metadata stream. It's now on the consumer side. */
1217         err = close(ksess->metadata_stream_fd);
1218         if (err < 0) {
1219                 PERROR("close snapshot kernel");
1220         }
1221
1222 error_open_stream:
1223         trace_kernel_destroy_metadata(ksess->metadata);
1224 error:
1225         /* Restore metadata state.*/
1226         ksess->metadata = saved_metadata;
1227         ksess->metadata_stream_fd = saved_metadata_fd;
1228
1229         rcu_read_unlock();
1230         return ret;
1231 }
1232
1233 /*
1234  * Get the syscall mask array from the kernel tracer.
1235  *
1236  * Return 0 on success else a negative value. In both case, syscall_mask should
1237  * be freed.
1238  */
1239 int kernel_syscall_mask(int chan_fd, char **syscall_mask, uint32_t *nr_bits)
1240 {
1241         assert(syscall_mask);
1242         assert(nr_bits);
1243
1244         return kernctl_syscall_mask(chan_fd, syscall_mask, nr_bits);
1245 }