liblttng-ust/lttng-ust-comm.c

   1 /*
   2  * lttng-ust-comm.c
   3  *
   4  * Copyright (C) 2011 David Goulet <david.goulet@polymtl.ca>
   5  * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   6  *
   7  * This library is free software; you can redistribute it and/or
   8  * modify it under the terms of the GNU Lesser General Public
   9  * License as published by the Free Software Foundation; only
  10  * version 2.1 of the License.
  11  *
  12  * This library is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  15  * Lesser General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU Lesser General Public
  18  * License along with this library; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21
  22 #define _LGPL_SOURCE
  23 #include <sys/types.h>
  24 #include <sys/socket.h>
  25 #include <sys/mman.h>
  26 #include <sys/stat.h>
  27 #include <sys/types.h>
  28 #include <sys/wait.h>
  29 #include <fcntl.h>
  30 #include <unistd.h>
  31 #include <errno.h>
  32 #include <pthread.h>
  33 #include <semaphore.h>
  34 #include <time.h>
  35 #include <assert.h>
  36 #include <signal.h>
  37 #include <urcu/uatomic.h>
  38 #include <urcu/futex.h>
  39 #include <urcu/compiler.h>
  40
  41 #include <lttng/ust-events.h>
  42 #include <lttng/ust-abi.h>
  43 #include <lttng/ust.h>
  44 #include <lttng/ust-error.h>
  45 #include <lttng/ust-ctl.h>
  46 #include <urcu/tls-compat.h>
  47 #include <ust-comm.h>
  48 #include <usterr-signal-safe.h>
  49 #include <helper.h>
  50 #include "tracepoint-internal.h"
  51 #include "lttng-tracer-core.h"
  52 #include "compat.h"
  53 #include "../libringbuffer/tlsfixup.h"
  54
  55 /*
  56  * Has lttng ust comm constructor been called ?
  57  */
  58 static int initialized;
  59
  60 /*
  61  * The ust_lock/ust_unlock lock is used as a communication thread mutex.
  62  * Held when handling a command, also held by fork() to deal with
  63  * removal of threads, and by exit path.
  64  */
  65
  66 /* Should the ust comm thread quit ? */
  67 static int lttng_ust_comm_should_quit;
  68
  69 /*
  70  * Wait for either of these before continuing to the main
  71  * program:
  72  * - the register_done message from sessiond daemon
  73  *   (will let the sessiond daemon enable sessions before main
  74  *   starts.)
  75  * - sessiond daemon is not reachable.
  76  * - timeout (ensuring applications are resilient to session
  77  *   daemon problems).
  78  */
  79 static sem_t constructor_wait;
  80 /*
  81  * Doing this for both the global and local sessiond.
  82  */
  83 static int sem_count = { 2 };
  84
  85 /*
  86  * Counting nesting within lttng-ust. Used to ensure that calling fork()
  87  * from liblttng-ust does not execute the pre/post fork handlers.
  88  */
  89 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
  90
  91 /*
  92  * Info about socket and associated listener thread.
  93  */
  94 struct sock_info {
  95         const char *name;
  96         pthread_t ust_listener; /* listener thread */
  97         int root_handle;
  98         int constructor_sem_posted;
  99         int allowed;
 100         int global;
 101         int thread_active;
 102
 103         char sock_path[PATH_MAX];
 104         int socket;
 105         int notify_socket;
 106
 107         char wait_shm_path[PATH_MAX];
 108         char *wait_shm_mmap;
 109 };
 110
 111 /* Socket from app (connect) to session daemon (listen) for communication */
 112 struct sock_info global_apps = {
 113         .name = "global",
 114         .global = 1,
 115
 116         .root_handle = -1,
 117         .allowed = 1,
 118         .thread_active = 0,
 119
 120         .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
 121         .socket = -1,
 122         .notify_socket = -1,
 123
 124         .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
 125 };
 126
 127 /* TODO: allow global_apps_sock_path override */
 128
 129 struct sock_info local_apps = {
 130         .name = "local",
 131         .global = 0,
 132         .root_handle = -1,
 133         .allowed = 0,   /* Check setuid bit first */
 134         .thread_active = 0,
 135
 136         .socket = -1,
 137         .notify_socket = -1,
 138 };
 139
 140 static int wait_poll_fallback;
 141
 142 static const char *cmd_name_mapping[] = {
 143         [ LTTNG_UST_RELEASE ] = "Release",
 144         [ LTTNG_UST_SESSION ] = "Create Session",
 145         [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
 146
 147         [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
 148         [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
 149         [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
 150         [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
 151
 152         /* Session FD commands */
 153         [ LTTNG_UST_CHANNEL ] = "Create Channel",
 154         [ LTTNG_UST_SESSION_START ] = "Start Session",
 155         [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
 156
 157         /* Channel FD commands */
 158         [ LTTNG_UST_STREAM ] = "Create Stream",
 159         [ LTTNG_UST_EVENT ] = "Create Event",
 160
 161         /* Event and Channel FD commands */
 162         [ LTTNG_UST_CONTEXT ] = "Create Context",
 163         [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
 164
 165         /* Event, Channel and Session commands */
 166         [ LTTNG_UST_ENABLE ] = "Enable",
 167         [ LTTNG_UST_DISABLE ] = "Disable",
 168
 169         /* Tracepoint list commands */
 170         [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
 171         [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
 172
 173         /* Event FD commands */
 174         [ LTTNG_UST_FILTER ] = "Create Filter",
 175         [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
 176 };
 177
 178 static const char *str_timeout;
 179 static int got_timeout_env;
 180
 181 extern void lttng_ring_buffer_client_overwrite_init(void);
 182 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
 183 extern void lttng_ring_buffer_client_discard_init(void);
 184 extern void lttng_ring_buffer_client_discard_rt_init(void);
 185 extern void lttng_ring_buffer_metadata_client_init(void);
 186 extern void lttng_ring_buffer_client_overwrite_exit(void);
 187 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
 188 extern void lttng_ring_buffer_client_discard_exit(void);
 189 extern void lttng_ring_buffer_client_discard_rt_exit(void);
 190 extern void lttng_ring_buffer_metadata_client_exit(void);
 191
 192 /*
 193  * Returns the HOME directory path. Caller MUST NOT free(3) the returned
 194  * pointer.
 195  */
 196 static
 197 const char *get_lttng_home_dir(void)
 198 {
 199        const char *val;
 200
 201        val = (const char *) getenv("LTTNG_HOME");
 202        if (val != NULL) {
 203                return val;
 204        }
 205        return (const char *) getenv("HOME");
 206 }
 207
 208 /*
 209  * Force a read (imply TLS fixup for dlopen) of TLS variables.
 210  */
 211 static
 212 void lttng_fixup_nest_count_tls(void)
 213 {
 214         asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
 215 }
 216
 217 int lttng_get_notify_socket(void *owner)
 218 {
 219         struct sock_info *info = owner;
 220
 221         return info->notify_socket;
 222 }
 223
 224 static
 225 void print_cmd(int cmd, int handle)
 226 {
 227         const char *cmd_name = "Unknown";
 228
 229         if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
 230                         && cmd_name_mapping[cmd]) {
 231                 cmd_name = cmd_name_mapping[cmd];
 232         }
 233         DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
 234                 cmd_name, cmd,
 235                 lttng_ust_obj_get_name(handle), handle);
 236 }
 237
 238 static
 239 int setup_local_apps(void)
 240 {
 241         const char *home_dir;
 242         uid_t uid;
 243
 244         uid = getuid();
 245         /*
 246          * Disallow per-user tracing for setuid binaries.
 247          */
 248         if (uid != geteuid()) {
 249                 assert(local_apps.allowed == 0);
 250                 return 0;
 251         }
 252         home_dir = get_lttng_home_dir();
 253         if (!home_dir) {
 254                 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
 255                 assert(local_apps.allowed == 0);
 256                 return -ENOENT;
 257         }
 258         local_apps.allowed = 1;
 259         snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
 260                 home_dir,
 261                 LTTNG_DEFAULT_HOME_RUNDIR,
 262                 LTTNG_UST_SOCK_FILENAME);
 263         snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
 264                 LTTNG_UST_WAIT_FILENAME,
 265                 uid);
 266         return 0;
 267 }
 268
 269 /*
 270  * Get notify_sock timeout, in ms.
 271  * -1: don't wait. 0: wait forever. >0: timeout, in ms.
 272  */
 273 static
 274 long get_timeout(void)
 275 {
 276         long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 277
 278         if (!got_timeout_env) {
 279                 str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
 280                 got_timeout_env = 1;
 281         }
 282         if (str_timeout)
 283                 constructor_delay_ms = strtol(str_timeout, NULL, 10);
 284         return constructor_delay_ms;
 285 }
 286
 287 static
 288 long get_notify_sock_timeout(void)
 289 {
 290         return get_timeout();
 291 }
 292
 293 /*
 294  * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
 295  */
 296 static
 297 int get_constructor_timeout(struct timespec *constructor_timeout)
 298 {
 299         long constructor_delay_ms;
 300         int ret;
 301
 302         constructor_delay_ms = get_timeout();
 303
 304         switch (constructor_delay_ms) {
 305         case -1:/* fall-through */
 306         case 0:
 307                 return constructor_delay_ms;
 308         default:
 309                 break;
 310         }
 311
 312         /*
 313          * If we are unable to find the current time, don't wait.
 314          */
 315         ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
 316         if (ret) {
 317                 return -1;
 318         }
 319         constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
 320         constructor_timeout->tv_nsec +=
 321                 (constructor_delay_ms % 1000UL) * 1000000UL;
 322         if (constructor_timeout->tv_nsec >= 1000000000UL) {
 323                 constructor_timeout->tv_sec++;
 324                 constructor_timeout->tv_nsec -= 1000000000UL;
 325         }
 326         return 1;
 327 }
 328
 329 static
 330 int register_to_sessiond(int socket, enum ustctl_socket_type type)
 331 {
 332         return ustcomm_send_reg_msg(socket,
 333                 type,
 334                 CAA_BITS_PER_LONG,
 335                 lttng_alignof(uint8_t) * CHAR_BIT,
 336                 lttng_alignof(uint16_t) * CHAR_BIT,
 337                 lttng_alignof(uint32_t) * CHAR_BIT,
 338                 lttng_alignof(uint64_t) * CHAR_BIT,
 339                 lttng_alignof(unsigned long) * CHAR_BIT);
 340 }
 341
 342 static
 343 int send_reply(int sock, struct ustcomm_ust_reply *lur)
 344 {
 345         ssize_t len;
 346
 347         len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
 348         switch (len) {
 349         case sizeof(*lur):
 350                 DBG("message successfully sent");
 351                 return 0;
 352         default:
 353                 if (len == -ECONNRESET) {
 354                         DBG("remote end closed connection");
 355                         return 0;
 356                 }
 357                 if (len < 0)
 358                         return len;
 359                 DBG("incorrect message size: %zd", len);
 360                 return -EINVAL;
 361         }
 362 }
 363
 364 static
 365 int handle_register_done(struct sock_info *sock_info)
 366 {
 367         int ret;
 368
 369         if (sock_info->constructor_sem_posted)
 370                 return 0;
 371         sock_info->constructor_sem_posted = 1;
 372         if (uatomic_read(&sem_count) <= 0) {
 373                 return 0;
 374         }
 375         ret = uatomic_add_return(&sem_count, -1);
 376         if (ret == 0) {
 377                 ret = sem_post(&constructor_wait);
 378                 assert(!ret);
 379         }
 380         return 0;
 381 }
 382
 383 static
 384 int handle_message(struct sock_info *sock_info,
 385                 int sock, struct ustcomm_ust_msg *lum)
 386 {
 387         int ret = 0;
 388         const struct lttng_ust_objd_ops *ops;
 389         struct ustcomm_ust_reply lur;
 390         union ust_args args;
 391         ssize_t len;
 392
 393         ust_lock();
 394
 395         memset(&lur, 0, sizeof(lur));
 396
 397         if (lttng_ust_comm_should_quit) {
 398                 ret = -LTTNG_UST_ERR_EXITING;
 399                 goto end;
 400         }
 401
 402         ops = objd_ops(lum->handle);
 403         if (!ops) {
 404                 ret = -ENOENT;
 405                 goto end;
 406         }
 407
 408         switch (lum->cmd) {
 409         case LTTNG_UST_REGISTER_DONE:
 410                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 411                         ret = handle_register_done(sock_info);
 412                 else
 413                         ret = -EINVAL;
 414                 break;
 415         case LTTNG_UST_RELEASE:
 416                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 417                         ret = -EPERM;
 418                 else
 419                         ret = lttng_ust_objd_unref(lum->handle, 1);
 420                 break;
 421         case LTTNG_UST_FILTER:
 422         {
 423                 /* Receive filter data */
 424                 struct lttng_ust_filter_bytecode_node *bytecode;
 425
 426                 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
 427                         ERR("Filter data size is too large: %u bytes",
 428                                 lum->u.filter.data_size);
 429                         ret = -EINVAL;
 430                         goto error;
 431                 }
 432
 433                 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
 434                         ERR("Filter reloc offset %u is not within data",
 435                                 lum->u.filter.reloc_offset);
 436                         ret = -EINVAL;
 437                         goto error;
 438                 }
 439
 440                 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
 441                 if (!bytecode) {
 442                         ret = -ENOMEM;
 443                         goto error;
 444                 }
 445                 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
 446                                 lum->u.filter.data_size);
 447                 switch (len) {
 448                 case 0: /* orderly shutdown */
 449                         ret = 0;
 450                         free(bytecode);
 451                         goto error;
 452                 default:
 453                         if (len == lum->u.filter.data_size) {
 454                                 DBG("filter data received");
 455                                 break;
 456                         } else if (len < 0) {
 457                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 458                                 if (len == -ECONNRESET) {
 459                                         ERR("%s remote end closed connection", sock_info->name);
 460                                         ret = len;
 461                                         free(bytecode);
 462                                         goto error;
 463                                 }
 464                                 ret = len;
 465                                 free(bytecode);
 466                                 goto end;
 467                         } else {
 468                                 DBG("incorrect filter data message size: %zd", len);
 469                                 ret = -EINVAL;
 470                                 free(bytecode);
 471                                 goto end;
 472                         }
 473                 }
 474                 bytecode->bc.len = lum->u.filter.data_size;
 475                 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
 476                 bytecode->bc.seqnum = lum->u.filter.seqnum;
 477                 if (ops->cmd) {
 478                         ret = ops->cmd(lum->handle, lum->cmd,
 479                                         (unsigned long) bytecode,
 480                                         &args, sock_info);
 481                         if (ret) {
 482                                 free(bytecode);
 483                         }
 484                         /* don't free bytecode if everything went fine. */
 485                 } else {
 486                         ret = -ENOSYS;
 487                         free(bytecode);
 488                 }
 489                 break;
 490         }
 491         case LTTNG_UST_CHANNEL:
 492         {
 493                 void *chan_data;
 494                 int wakeup_fd;
 495
 496                 len = ustcomm_recv_channel_from_sessiond(sock,
 497                                 &chan_data, lum->u.channel.len,
 498                                 &wakeup_fd);
 499                 switch (len) {
 500                 case 0: /* orderly shutdown */
 501                         ret = 0;
 502                         goto error;
 503                 default:
 504                         if (len == lum->u.channel.len) {
 505                                 DBG("channel data received");
 506                                 break;
 507                         } else if (len < 0) {
 508                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 509                                 if (len == -ECONNRESET) {
 510                                         ERR("%s remote end closed connection", sock_info->name);
 511                                         ret = len;
 512                                         goto error;
 513                                 }
 514                                 ret = len;
 515                                 goto end;
 516                         } else {
 517                                 DBG("incorrect channel data message size: %zd", len);
 518                                 ret = -EINVAL;
 519                                 goto end;
 520                         }
 521                 }
 522                 args.channel.chan_data = chan_data;
 523                 args.channel.wakeup_fd = wakeup_fd;
 524                 if (ops->cmd)
 525                         ret = ops->cmd(lum->handle, lum->cmd,
 526                                         (unsigned long) &lum->u,
 527                                         &args, sock_info);
 528                 else
 529                         ret = -ENOSYS;
 530                 break;
 531         }
 532         case LTTNG_UST_STREAM:
 533         {
 534                 /* Receive shm_fd, wakeup_fd */
 535                 ret = ustcomm_recv_stream_from_sessiond(sock,
 536                         &lum->u.stream.len,
 537                         &args.stream.shm_fd,
 538                         &args.stream.wakeup_fd);
 539                 if (ret) {
 540                         goto end;
 541                 }
 542                 if (ops->cmd)
 543                         ret = ops->cmd(lum->handle, lum->cmd,
 544                                         (unsigned long) &lum->u,
 545                                         &args, sock_info);
 546                 else
 547                         ret = -ENOSYS;
 548                 break;
 549         }
 550         default:
 551                 if (ops->cmd)
 552                         ret = ops->cmd(lum->handle, lum->cmd,
 553                                         (unsigned long) &lum->u,
 554                                         &args, sock_info);
 555                 else
 556                         ret = -ENOSYS;
 557                 break;
 558         }
 559
 560 end:
 561         lur.handle = lum->handle;
 562         lur.cmd = lum->cmd;
 563         lur.ret_val = ret;
 564         if (ret >= 0) {
 565                 lur.ret_code = LTTNG_UST_OK;
 566         } else {
 567                 /*
 568                  * Use -LTTNG_UST_ERR as wildcard for UST internal
 569                  * error that are not caused by the transport, except if
 570                  * we already have a more precise error message to
 571                  * report.
 572                  */
 573                 if (ret > -LTTNG_UST_ERR) {
 574                         /* Translate code to UST error. */
 575                         switch (ret) {
 576                         case -EEXIST:
 577                                 lur.ret_code = -LTTNG_UST_ERR_EXIST;
 578                                 break;
 579                         case -EINVAL:
 580                                 lur.ret_code = -LTTNG_UST_ERR_INVAL;
 581                                 break;
 582                         case -ENOENT:
 583                                 lur.ret_code = -LTTNG_UST_ERR_NOENT;
 584                                 break;
 585                         case -EPERM:
 586                                 lur.ret_code = -LTTNG_UST_ERR_PERM;
 587                                 break;
 588                         case -ENOSYS:
 589                                 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
 590                                 break;
 591                         default:
 592                                 lur.ret_code = -LTTNG_UST_ERR;
 593                                 break;
 594                         }
 595                 } else {
 596                         lur.ret_code = ret;
 597                 }
 598         }
 599         if (ret >= 0) {
 600                 switch (lum->cmd) {
 601                 case LTTNG_UST_TRACER_VERSION:
 602                         lur.u.version = lum->u.version;
 603                         break;
 604                 case LTTNG_UST_TRACEPOINT_LIST_GET:
 605                         memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
 606                         break;
 607                 }
 608         }
 609         DBG("Return value: %d", lur.ret_val);
 610         ret = send_reply(sock, &lur);
 611         if (ret < 0) {
 612                 DBG("error sending reply");
 613                 goto error;
 614         }
 615
 616         /*
 617          * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
 618          * after the reply.
 619          */
 620         if (lur.ret_code == LTTNG_UST_OK) {
 621                 switch (lum->cmd) {
 622                 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
 623                         len = ustcomm_send_unix_sock(sock,
 624                                 &args.field_list.entry,
 625                                 sizeof(args.field_list.entry));
 626                         if (len < 0) {
 627                                 ret = len;
 628                                 goto error;
 629                         }
 630                         if (len != sizeof(args.field_list.entry)) {
 631                                 ret = -EINVAL;
 632                                 goto error;
 633                         }
 634                 }
 635         }
 636
 637 error:
 638         ust_unlock();
 639         return ret;
 640 }
 641
 642 static
 643 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
 644 {
 645         int ret;
 646
 647         if (sock_info->root_handle != -1) {
 648                 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
 649                 if (ret) {
 650                         ERR("Error unref root handle");
 651                 }
 652                 sock_info->root_handle = -1;
 653         }
 654         sock_info->constructor_sem_posted = 0;
 655
 656         /*
 657          * wait_shm_mmap, socket and notify socket are used by listener
 658          * threads outside of the ust lock, so we cannot tear them down
 659          * ourselves, because we cannot join on these threads. Leave
 660          * responsibility of cleaning up these resources to the OS
 661          * process exit.
 662          */
 663         if (exiting)
 664                 return;
 665
 666         if (sock_info->socket != -1) {
 667                 ret = ustcomm_close_unix_sock(sock_info->socket);
 668                 if (ret) {
 669                         ERR("Error closing ust cmd socket");
 670                 }
 671                 sock_info->socket = -1;
 672         }
 673         if (sock_info->notify_socket != -1) {
 674                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
 675                 if (ret) {
 676                         ERR("Error closing ust notify socket");
 677                 }
 678                 sock_info->notify_socket = -1;
 679         }
 680         if (sock_info->wait_shm_mmap) {
 681                 ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
 682                 if (ret) {
 683                         ERR("Error unmapping wait shm");
 684                 }
 685                 sock_info->wait_shm_mmap = NULL;
 686         }
 687 }
 688
 689 /*
 690  * Using fork to set umask in the child process (not multi-thread safe).
 691  * We deal with the shm_open vs ftruncate race (happening when the
 692  * sessiond owns the shm and does not let everybody modify it, to ensure
 693  * safety against shm_unlink) by simply letting the mmap fail and
 694  * retrying after a few seconds.
 695  * For global shm, everybody has rw access to it until the sessiond
 696  * starts.
 697  */
 698 static
 699 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
 700 {
 701         int wait_shm_fd, ret;
 702         pid_t pid;
 703
 704         /*
 705          * Try to open read-only.
 706          */
 707         wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
 708         if (wait_shm_fd >= 0) {
 709                 int32_t tmp_read;
 710                 ssize_t len;
 711                 size_t bytes_read = 0;
 712
 713                 /*
 714                  * Try to read the fd. If unable to do so, try opening
 715                  * it in write mode.
 716                  */
 717                 do {
 718                         len = read(wait_shm_fd,
 719                                 &((char *) &tmp_read)[bytes_read],
 720                                 sizeof(tmp_read) - bytes_read);
 721                         if (len > 0) {
 722                                 bytes_read += len;
 723                         }
 724                 } while ((len < 0 && errno == EINTR)
 725                         || (len > 0 && bytes_read < sizeof(tmp_read)));
 726                 if (bytes_read != sizeof(tmp_read)) {
 727                         ret = close(wait_shm_fd);
 728                         if (ret) {
 729                                 ERR("close wait_shm_fd");
 730                         }
 731                         goto open_write;
 732                 }
 733                 goto end;
 734         } else if (wait_shm_fd < 0 && errno != ENOENT) {
 735                 /*
 736                  * Real-only open did not work, and it's not because the
 737                  * entry was not present. It's a failure that prohibits
 738                  * using shm.
 739                  */
 740                 ERR("Error opening shm %s", sock_info->wait_shm_path);
 741                 goto end;
 742         }
 743
 744 open_write:
 745         /*
 746          * If the open failed because the file did not exist, or because
 747          * the file was not truncated yet, try creating it ourself.
 748          */
 749         URCU_TLS(lttng_ust_nest_count)++;
 750         pid = fork();
 751         URCU_TLS(lttng_ust_nest_count)--;
 752         if (pid > 0) {
 753                 int status;
 754
 755                 /*
 756                  * Parent: wait for child to return, in which case the
 757                  * shared memory map will have been created.
 758                  */
 759                 pid = wait(&status);
 760                 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 761                         wait_shm_fd = -1;
 762                         goto end;
 763                 }
 764                 /*
 765                  * Try to open read-only again after creation.
 766                  */
 767                 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
 768                 if (wait_shm_fd < 0) {
 769                         /*
 770                          * Real-only open did not work. It's a failure
 771                          * that prohibits using shm.
 772                          */
 773                         ERR("Error opening shm %s", sock_info->wait_shm_path);
 774                         goto end;
 775                 }
 776                 goto end;
 777         } else if (pid == 0) {
 778                 int create_mode;
 779
 780                 /* Child */
 781                 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
 782                 if (sock_info->global)
 783                         create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
 784                 /*
 785                  * We're alone in a child process, so we can modify the
 786                  * process-wide umask.
 787                  */
 788                 umask(~create_mode);
 789                 /*
 790                  * Try creating shm (or get rw access).
 791                  * We don't do an exclusive open, because we allow other
 792                  * processes to create+ftruncate it concurrently.
 793                  */
 794                 wait_shm_fd = shm_open(sock_info->wait_shm_path,
 795                                 O_RDWR | O_CREAT, create_mode);
 796                 if (wait_shm_fd >= 0) {
 797                         ret = ftruncate(wait_shm_fd, mmap_size);
 798                         if (ret) {
 799                                 PERROR("ftruncate");
 800                                 _exit(EXIT_FAILURE);
 801                         }
 802                         _exit(EXIT_SUCCESS);
 803                 }
 804                 /*
 805                  * For local shm, we need to have rw access to accept
 806                  * opening it: this means the local sessiond will be
 807                  * able to wake us up. For global shm, we open it even
 808                  * if rw access is not granted, because the root.root
 809                  * sessiond will be able to override all rights and wake
 810                  * us up.
 811                  */
 812                 if (!sock_info->global && errno != EACCES) {
 813                         ERR("Error opening shm %s", sock_info->wait_shm_path);
 814                         _exit(EXIT_FAILURE);
 815                 }
 816                 /*
 817                  * The shm exists, but we cannot open it RW. Report
 818                  * success.
 819                  */
 820                 _exit(EXIT_SUCCESS);
 821         } else {
 822                 return -1;
 823         }
 824 end:
 825         if (wait_shm_fd >= 0 && !sock_info->global) {
 826                 struct stat statbuf;
 827
 828                 /*
 829                  * Ensure that our user is the owner of the shm file for
 830                  * local shm. If we do not own the file, it means our
 831                  * sessiond will not have access to wake us up (there is
 832                  * probably a rogue process trying to fake our
 833                  * sessiond). Fallback to polling method in this case.
 834                  */
 835                 ret = fstat(wait_shm_fd, &statbuf);
 836                 if (ret) {
 837                         PERROR("fstat");
 838                         goto error_close;
 839                 }
 840                 if (statbuf.st_uid != getuid())
 841                         goto error_close;
 842         }
 843         return wait_shm_fd;
 844
 845 error_close:
 846         ret = close(wait_shm_fd);
 847         if (ret) {
 848                 PERROR("Error closing fd");
 849         }
 850         return -1;
 851 }
 852
 853 static
 854 char *get_map_shm(struct sock_info *sock_info)
 855 {
 856         size_t mmap_size = sysconf(_SC_PAGE_SIZE);
 857         int wait_shm_fd, ret;
 858         char *wait_shm_mmap;
 859
 860         wait_shm_fd = get_wait_shm(sock_info, mmap_size);
 861         if (wait_shm_fd < 0) {
 862                 goto error;
 863         }
 864         wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
 865                   MAP_SHARED, wait_shm_fd, 0);
 866         /* close shm fd immediately after taking the mmap reference */
 867         ret = close(wait_shm_fd);
 868         if (ret) {
 869                 PERROR("Error closing fd");
 870         }
 871         if (wait_shm_mmap == MAP_FAILED) {
 872                 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
 873                 goto error;
 874         }
 875         return wait_shm_mmap;
 876
 877 error:
 878         return NULL;
 879 }
 880
 881 static
 882 void wait_for_sessiond(struct sock_info *sock_info)
 883 {
 884         int ret;
 885
 886         ust_lock();
 887         if (lttng_ust_comm_should_quit) {
 888                 goto quit;
 889         }
 890         if (wait_poll_fallback) {
 891                 goto error;
 892         }
 893         if (!sock_info->wait_shm_mmap) {
 894                 sock_info->wait_shm_mmap = get_map_shm(sock_info);
 895                 if (!sock_info->wait_shm_mmap)
 896                         goto error;
 897         }
 898         ust_unlock();
 899
 900         DBG("Waiting for %s apps sessiond", sock_info->name);
 901         /* Wait for futex wakeup */
 902         if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
 903                 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
 904                         FUTEX_WAIT, 0, NULL, NULL, 0);
 905                 if (ret < 0) {
 906                         if (errno == EFAULT) {
 907                                 wait_poll_fallback = 1;
 908                                 DBG(
 909 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
 910 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
 911 "Please upgrade your kernel "
 912 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
 913 "mainline). LTTng-UST will use polling mode fallback.");
 914                                 if (ust_debug())
 915                                         PERROR("futex");
 916                         }
 917                 }
 918         }
 919         return;
 920
 921 quit:
 922         ust_unlock();
 923         return;
 924
 925 error:
 926         ust_unlock();
 927         return;
 928 }
 929
 930 /*
 931  * This thread does not allocate any resource, except within
 932  * handle_message, within mutex protection. This mutex protects against
 933  * fork and exit.
 934  * The other moment it allocates resources is at socket connection, which
 935  * is also protected by the mutex.
 936  */
 937 static
 938 void *ust_listener_thread(void *arg)
 939 {
 940         struct sock_info *sock_info = arg;
 941         int sock, ret, prev_connect_failed = 0, has_waited = 0;
 942         long timeout;
 943
 944         /* Restart trying to connect to the session daemon */
 945 restart:
 946         if (prev_connect_failed) {
 947                 /* Wait for sessiond availability with pipe */
 948                 wait_for_sessiond(sock_info);
 949                 if (has_waited) {
 950                         has_waited = 0;
 951                         /*
 952                          * Sleep for 5 seconds before retrying after a
 953                          * sequence of failure / wait / failure. This
 954                          * deals with a killed or broken session daemon.
 955                          */
 956                         sleep(5);
 957                 }
 958                 has_waited = 1;
 959                 prev_connect_failed = 0;
 960         }
 961
 962         if (sock_info->socket != -1) {
 963                 ret = ustcomm_close_unix_sock(sock_info->socket);
 964                 if (ret) {
 965                         ERR("Error closing %s ust cmd socket",
 966                                 sock_info->name);
 967                 }
 968                 sock_info->socket = -1;
 969         }
 970         if (sock_info->notify_socket != -1) {
 971                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
 972                 if (ret) {
 973                         ERR("Error closing %s ust notify socket",
 974                                 sock_info->name);
 975                 }
 976                 sock_info->notify_socket = -1;
 977         }
 978
 979         /*
 980          * Register. We need to perform both connect and sending
 981          * registration message before doing the next connect otherwise
 982          * we may reach unix socket connect queue max limits and block
 983          * on the 2nd connect while the session daemon is awaiting the
 984          * first connect registration message.
 985          */
 986         /* Connect cmd socket */
 987         ret = ustcomm_connect_unix_sock(sock_info->sock_path);
 988         if (ret < 0) {
 989                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
 990                 prev_connect_failed = 1;
 991
 992                 ust_lock();
 993
 994                 if (lttng_ust_comm_should_quit) {
 995                         goto quit;
 996                 }
 997
 998                 /*
 999                  * If we cannot find the sessiond daemon, don't delay
1000                  * constructor execution.
1001                  */
1002                 ret = handle_register_done(sock_info);
1003                 assert(!ret);
1004                 ust_unlock();
1005                 goto restart;
1006         }
1007         sock_info->socket = ret;
1008
1009         ust_lock();
1010
1011         if (lttng_ust_comm_should_quit) {
1012                 goto quit;
1013         }
1014
1015         /*
1016          * Create only one root handle per listener thread for the whole
1017          * process lifetime, so we ensure we get ID which is statically
1018          * assigned to the root handle.
1019          */
1020         if (sock_info->root_handle == -1) {
1021                 ret = lttng_abi_create_root_handle();
1022                 if (ret < 0) {
1023                         ERR("Error creating root handle");
1024                         goto quit;
1025                 }
1026                 sock_info->root_handle = ret;
1027         }
1028
1029         ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1030         if (ret < 0) {
1031                 ERR("Error registering to %s ust cmd socket",
1032                         sock_info->name);
1033                 prev_connect_failed = 1;
1034                 /*
1035                  * If we cannot register to the sessiond daemon, don't
1036                  * delay constructor execution.
1037                  */
1038                 ret = handle_register_done(sock_info);
1039                 assert(!ret);
1040                 ust_unlock();
1041                 goto restart;
1042         }
1043
1044         ust_unlock();
1045
1046         /* Connect notify socket */
1047         ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1048         if (ret < 0) {
1049                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1050                 prev_connect_failed = 1;
1051
1052                 ust_lock();
1053
1054                 if (lttng_ust_comm_should_quit) {
1055                         goto quit;
1056                 }
1057
1058                 /*
1059                  * If we cannot find the sessiond daemon, don't delay
1060                  * constructor execution.
1061                  */
1062                 ret = handle_register_done(sock_info);
1063                 assert(!ret);
1064                 ust_unlock();
1065                 goto restart;
1066         }
1067         sock_info->notify_socket = ret;
1068
1069         timeout = get_notify_sock_timeout();
1070         if (timeout >= 0) {
1071                 /*
1072                  * Give at least 10ms to sessiond to reply to
1073                  * notifications.
1074                  */
1075                 if (timeout < 10)
1076                         timeout = 10;
1077                 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1078                                 timeout);
1079                 if (ret < 0) {
1080                         WARN("Error setting socket receive timeout");
1081                 }
1082                 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1083                                 timeout);
1084                 if (ret < 0) {
1085                         WARN("Error setting socket send timeout");
1086                 }
1087         } else if (timeout < -1) {
1088                 WARN("Unsupported timeout value %ld", timeout);
1089         }
1090
1091         ust_lock();
1092
1093         if (lttng_ust_comm_should_quit) {
1094                 goto quit;
1095         }
1096
1097         ret = register_to_sessiond(sock_info->notify_socket,
1098                         USTCTL_SOCKET_NOTIFY);
1099         if (ret < 0) {
1100                 ERR("Error registering to %s ust notify socket",
1101                         sock_info->name);
1102                 prev_connect_failed = 1;
1103                 /*
1104                  * If we cannot register to the sessiond daemon, don't
1105                  * delay constructor execution.
1106                  */
1107                 ret = handle_register_done(sock_info);
1108                 assert(!ret);
1109                 ust_unlock();
1110                 goto restart;
1111         }
1112         sock = sock_info->socket;
1113
1114         ust_unlock();
1115
1116         for (;;) {
1117                 ssize_t len;
1118                 struct ustcomm_ust_msg lum;
1119
1120                 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1121                 switch (len) {
1122                 case 0: /* orderly shutdown */
1123                         DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1124                         ust_lock();
1125                         if (lttng_ust_comm_should_quit) {
1126                                 goto quit;
1127                         }
1128                         /*
1129                          * Either sessiond has shutdown or refused us by closing the socket.
1130                          * In either case, we don't want to delay construction execution,
1131                          * and we need to wait before retry.
1132                          */
1133                         prev_connect_failed = 1;
1134                         /*
1135                          * If we cannot register to the sessiond daemon, don't
1136                          * delay constructor execution.
1137                          */
1138                         ret = handle_register_done(sock_info);
1139                         assert(!ret);
1140                         ust_unlock();
1141                         goto end;
1142                 case sizeof(lum):
1143                         print_cmd(lum.cmd, lum.handle);
1144                         ret = handle_message(sock_info, sock, &lum);
1145                         if (ret) {
1146                                 ERR("Error handling message for %s socket", sock_info->name);
1147                         }
1148                         continue;
1149                 default:
1150                         if (len < 0) {
1151                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1152                         } else {
1153                                 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1154                         }
1155                         if (len == -ECONNRESET) {
1156                                 DBG("%s remote end closed connection", sock_info->name);
1157                                 goto end;
1158                         }
1159                         goto end;
1160                 }
1161
1162         }
1163 end:
1164         ust_lock();
1165         if (lttng_ust_comm_should_quit) {
1166                 goto quit;
1167         }
1168         /* Cleanup socket handles before trying to reconnect */
1169         lttng_ust_objd_table_owner_cleanup(sock_info);
1170         ust_unlock();
1171         goto restart;   /* try to reconnect */
1172
1173 quit:
1174         sock_info->thread_active = 0;
1175         ust_unlock();
1176         return NULL;
1177 }
1178
1179 /*
1180  * sessiond monitoring thread: monitor presence of global and per-user
1181  * sessiond by polling the application common named pipe.
1182  */
1183 void __attribute__((constructor)) lttng_ust_init(void)
1184 {
1185         struct timespec constructor_timeout;
1186         sigset_t sig_all_blocked, orig_parent_mask;
1187         pthread_attr_t thread_attr;
1188         int timeout_mode;
1189         int ret;
1190
1191         if (uatomic_xchg(&initialized, 1) == 1)
1192                 return;
1193
1194         /*
1195          * Fixup interdependency between TLS fixup mutex (which happens
1196          * to be the dynamic linker mutex) and ust_lock, taken within
1197          * the ust lock.
1198          */
1199         lttng_fixup_ringbuffer_tls();
1200         lttng_fixup_vtid_tls();
1201         lttng_fixup_nest_count_tls();
1202         lttng_fixup_procname_tls();
1203
1204         /*
1205          * We want precise control over the order in which we construct
1206          * our sub-libraries vs starting to receive commands from
1207          * sessiond (otherwise leading to errors when trying to create
1208          * sessiond before the init functions are completed).
1209          */
1210         init_usterr();
1211         init_tracepoint();
1212         lttng_ring_buffer_metadata_client_init();
1213         lttng_ring_buffer_client_overwrite_init();
1214         lttng_ring_buffer_client_overwrite_rt_init();
1215         lttng_ring_buffer_client_discard_init();
1216         lttng_ring_buffer_client_discard_rt_init();
1217         lttng_context_init();
1218
1219         timeout_mode = get_constructor_timeout(&constructor_timeout);
1220
1221         ret = sem_init(&constructor_wait, 0, 0);
1222         assert(!ret);
1223
1224         ret = setup_local_apps();
1225         if (ret) {
1226                 DBG("local apps setup returned %d", ret);
1227         }
1228
1229         /* A new thread created by pthread_create inherits the signal mask
1230          * from the parent. To avoid any signal being received by the
1231          * listener thread, we block all signals temporarily in the parent,
1232          * while we create the listener thread.
1233          */
1234         sigfillset(&sig_all_blocked);
1235         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1236         if (ret) {
1237                 ERR("pthread_sigmask: %s", strerror(ret));
1238         }
1239
1240         ret = pthread_attr_init(&thread_attr);
1241         if (ret) {
1242                 ERR("pthread_attr_init: %s", strerror(ret));
1243         }
1244         ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1245         if (ret) {
1246                 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1247         }
1248
1249         ust_lock();
1250         ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1251                         ust_listener_thread, &global_apps);
1252         if (ret) {
1253                 ERR("pthread_create global: %s", strerror(ret));
1254         }
1255         global_apps.thread_active = 1;
1256         ust_unlock();
1257
1258         if (local_apps.allowed) {
1259                 ust_lock();
1260                 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1261                                 ust_listener_thread, &local_apps);
1262                 if (ret) {
1263                         ERR("pthread_create local: %s", strerror(ret));
1264                 }
1265                 local_apps.thread_active = 1;
1266                 ust_unlock();
1267         } else {
1268                 handle_register_done(&local_apps);
1269         }
1270         ret = pthread_attr_destroy(&thread_attr);
1271         if (ret) {
1272                 ERR("pthread_attr_destroy: %s", strerror(ret));
1273         }
1274
1275         /* Restore original signal mask in parent */
1276         ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1277         if (ret) {
1278                 ERR("pthread_sigmask: %s", strerror(ret));
1279         }
1280
1281         switch (timeout_mode) {
1282         case 1: /* timeout wait */
1283                 do {
1284                         ret = sem_timedwait(&constructor_wait,
1285                                         &constructor_timeout);
1286                 } while (ret < 0 && errno == EINTR);
1287                 if (ret < 0 && errno == ETIMEDOUT) {
1288                         ERR("Timed out waiting for lttng-sessiond");
1289                 } else {
1290                         assert(!ret);
1291                 }
1292                 break;
1293         case -1:/* wait forever */
1294                 do {
1295                         ret = sem_wait(&constructor_wait);
1296                 } while (ret < 0 && errno == EINTR);
1297                 assert(!ret);
1298                 break;
1299         case 0: /* no timeout */
1300                 break;
1301         }
1302 }
1303
1304 static
1305 void lttng_ust_cleanup(int exiting)
1306 {
1307         cleanup_sock_info(&global_apps, exiting);
1308         if (local_apps.allowed) {
1309                 cleanup_sock_info(&local_apps, exiting);
1310         }
1311         /*
1312          * The teardown in this function all affect data structures
1313          * accessed under the UST lock by the listener thread. This
1314          * lock, along with the lttng_ust_comm_should_quit flag, ensure
1315          * that none of these threads are accessing this data at this
1316          * point.
1317          */
1318         lttng_ust_abi_exit();
1319         lttng_ust_events_exit();
1320         lttng_context_exit();
1321         lttng_ring_buffer_client_discard_rt_exit();
1322         lttng_ring_buffer_client_discard_exit();
1323         lttng_ring_buffer_client_overwrite_rt_exit();
1324         lttng_ring_buffer_client_overwrite_exit();
1325         lttng_ring_buffer_metadata_client_exit();
1326         exit_tracepoint();
1327         if (!exiting) {
1328                 /* Reinitialize values for fork */
1329                 sem_count = 2;
1330                 lttng_ust_comm_should_quit = 0;
1331                 initialized = 0;
1332         }
1333 }
1334
1335 void __attribute__((destructor)) lttng_ust_exit(void)
1336 {
1337         int ret;
1338
1339         /*
1340          * Using pthread_cancel here because:
1341          * A) we don't want to hang application teardown.
1342          * B) the thread is not allocating any resource.
1343          */
1344
1345         /*
1346          * Require the communication thread to quit. Synchronize with
1347          * mutexes to ensure it is not in a mutex critical section when
1348          * pthread_cancel is later called.
1349          */
1350         ust_lock();
1351         lttng_ust_comm_should_quit = 1;
1352
1353         /* cancel threads */
1354         if (global_apps.thread_active) {
1355                 ret = pthread_cancel(global_apps.ust_listener);
1356                 if (ret) {
1357                         ERR("Error cancelling global ust listener thread: %s",
1358                                 strerror(ret));
1359                 } else {
1360                         global_apps.thread_active = 0;
1361                 }
1362         }
1363         if (local_apps.thread_active) {
1364                 ret = pthread_cancel(local_apps.ust_listener);
1365                 if (ret) {
1366                         ERR("Error cancelling local ust listener thread: %s",
1367                                 strerror(ret));
1368                 } else {
1369                         local_apps.thread_active = 0;
1370                 }
1371         }
1372         ust_unlock();
1373
1374         /*
1375          * Do NOT join threads: use of sys_futex makes it impossible to
1376          * join the threads without using async-cancel, but async-cancel
1377          * is delivered by a signal, which could hit the target thread
1378          * anywhere in its code path, including while the ust_lock() is
1379          * held, causing a deadlock for the other thread. Let the OS
1380          * cleanup the threads if there are stalled in a syscall.
1381          */
1382         lttng_ust_cleanup(1);
1383 }
1384
1385 /*
1386  * We exclude the worker threads across fork and clone (except
1387  * CLONE_VM), because these system calls only keep the forking thread
1388  * running in the child.  Therefore, we don't want to call fork or clone
1389  * in the middle of an tracepoint or ust tracing state modification.
1390  * Holding this mutex protects these structures across fork and clone.
1391  */
1392 void ust_before_fork(sigset_t *save_sigset)
1393 {
1394         /*
1395          * Disable signals. This is to avoid that the child intervenes
1396          * before it is properly setup for tracing. It is safer to
1397          * disable all signals, because then we know we are not breaking
1398          * anything by restoring the original mask.
1399          */
1400         sigset_t all_sigs;
1401         int ret;
1402
1403         if (URCU_TLS(lttng_ust_nest_count))
1404                 return;
1405         /* Disable signals */
1406         sigfillset(&all_sigs);
1407         ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
1408         if (ret == -1) {
1409                 PERROR("sigprocmask");
1410         }
1411         ust_lock();
1412         rcu_bp_before_fork();
1413 }
1414
1415 static void ust_after_fork_common(sigset_t *restore_sigset)
1416 {
1417         int ret;
1418
1419         DBG("process %d", getpid());
1420         ust_unlock();
1421         /* Restore signals */
1422         ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
1423         if (ret == -1) {
1424                 PERROR("sigprocmask");
1425         }
1426 }
1427
1428 void ust_after_fork_parent(sigset_t *restore_sigset)
1429 {
1430         if (URCU_TLS(lttng_ust_nest_count))
1431                 return;
1432         DBG("process %d", getpid());
1433         rcu_bp_after_fork_parent();
1434         /* Release mutexes and reenable signals */
1435         ust_after_fork_common(restore_sigset);
1436 }
1437
1438 /*
1439  * After fork, in the child, we need to cleanup all the leftover state,
1440  * except the worker thread which already magically disappeared thanks
1441  * to the weird Linux fork semantics. After tyding up, we call
1442  * lttng_ust_init() again to start over as a new PID.
1443  *
1444  * This is meant for forks() that have tracing in the child between the
1445  * fork and following exec call (if there is any).
1446  */
1447 void ust_after_fork_child(sigset_t *restore_sigset)
1448 {
1449         if (URCU_TLS(lttng_ust_nest_count))
1450                 return;
1451         DBG("process %d", getpid());
1452         /* Release urcu mutexes */
1453         rcu_bp_after_fork_child();
1454         lttng_ust_cleanup(0);
1455         lttng_context_vtid_reset();
1456         /* Release mutexes and reenable signals */
1457         ust_after_fork_common(restore_sigset);
1458         lttng_ust_init();
1459 }