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 <dlfcn.h>
  38 #include <urcu/uatomic.h>
  39 #include <urcu/futex.h>
  40 #include <urcu/compiler.h>
  41
  42 #include <lttng/ust-events.h>
  43 #include <lttng/ust-abi.h>
  44 #include <lttng/ust.h>
  45 #include <lttng/ust-error.h>
  46 #include <lttng/ust-ctl.h>
  47 #include <urcu/tls-compat.h>
  48 #include <ust-comm.h>
  49 #include <usterr-signal-safe.h>
  50 #include <helper.h>
  51 #include "tracepoint-internal.h"
  52 #include "lttng-tracer-core.h"
  53 #include "compat.h"
  54 #include "../libringbuffer/tlsfixup.h"
  55 #include "lttng-ust-baddr.h"
  56
  57 /*
  58  * Has lttng ust comm constructor been called ?
  59  */
  60 static int initialized;
  61
  62 /*
  63  * The ust_lock/ust_unlock lock is used as a communication thread mutex.
  64  * Held when handling a command, also held by fork() to deal with
  65  * removal of threads, and by exit path.
  66  */
  67
  68 /* Should the ust comm thread quit ? */
  69 static int lttng_ust_comm_should_quit;
  70
  71 /*
  72  * Wait for either of these before continuing to the main
  73  * program:
  74  * - the register_done message from sessiond daemon
  75  *   (will let the sessiond daemon enable sessions before main
  76  *   starts.)
  77  * - sessiond daemon is not reachable.
  78  * - timeout (ensuring applications are resilient to session
  79  *   daemon problems).
  80  */
  81 static sem_t constructor_wait;
  82 /*
  83  * Doing this for both the global and local sessiond.
  84  */
  85 static int sem_count = { 2 };
  86
  87 /*
  88  * Counting nesting within lttng-ust. Used to ensure that calling fork()
  89  * from liblttng-ust does not execute the pre/post fork handlers.
  90  */
  91 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
  92
  93 /*
  94  * Info about socket and associated listener thread.
  95  */
  96 struct sock_info {
  97         const char *name;
  98         pthread_t ust_listener; /* listener thread */
  99         int root_handle;
 100         int constructor_sem_posted;
 101         int allowed;
 102         int global;
 103         int thread_active;
 104
 105         char sock_path[PATH_MAX];
 106         int socket;
 107         int notify_socket;
 108
 109         char wait_shm_path[PATH_MAX];
 110         char *wait_shm_mmap;
 111         int session_enabled;
 112 };
 113
 114 /* Socket from app (connect) to session daemon (listen) for communication */
 115 struct sock_info global_apps = {
 116         .name = "global",
 117         .global = 1,
 118
 119         .root_handle = -1,
 120         .allowed = 1,
 121         .thread_active = 0,
 122
 123         .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
 124         .socket = -1,
 125         .notify_socket = -1,
 126
 127         .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
 128
 129         .session_enabled = 0,
 130 };
 131
 132 /* TODO: allow global_apps_sock_path override */
 133
 134 struct sock_info local_apps = {
 135         .name = "local",
 136         .global = 0,
 137         .root_handle = -1,
 138         .allowed = 0,   /* Check setuid bit first */
 139         .thread_active = 0,
 140
 141         .socket = -1,
 142         .notify_socket = -1,
 143
 144         .session_enabled = 0,
 145 };
 146
 147 static int wait_poll_fallback;
 148
 149 static const char *cmd_name_mapping[] = {
 150         [ LTTNG_UST_RELEASE ] = "Release",
 151         [ LTTNG_UST_SESSION ] = "Create Session",
 152         [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
 153
 154         [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
 155         [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
 156         [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
 157         [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
 158
 159         /* Session FD commands */
 160         [ LTTNG_UST_CHANNEL ] = "Create Channel",
 161         [ LTTNG_UST_SESSION_START ] = "Start Session",
 162         [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
 163
 164         /* Channel FD commands */
 165         [ LTTNG_UST_STREAM ] = "Create Stream",
 166         [ LTTNG_UST_EVENT ] = "Create Event",
 167
 168         /* Event and Channel FD commands */
 169         [ LTTNG_UST_CONTEXT ] = "Create Context",
 170         [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
 171
 172         /* Event, Channel and Session commands */
 173         [ LTTNG_UST_ENABLE ] = "Enable",
 174         [ LTTNG_UST_DISABLE ] = "Disable",
 175
 176         /* Tracepoint list commands */
 177         [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
 178         [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
 179
 180         /* Event FD commands */
 181         [ LTTNG_UST_FILTER ] = "Create Filter",
 182         [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
 183 };
 184
 185 static const char *str_timeout;
 186 static int got_timeout_env;
 187
 188 extern void lttng_ring_buffer_client_overwrite_init(void);
 189 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
 190 extern void lttng_ring_buffer_client_discard_init(void);
 191 extern void lttng_ring_buffer_client_discard_rt_init(void);
 192 extern void lttng_ring_buffer_metadata_client_init(void);
 193 extern void lttng_ring_buffer_client_overwrite_exit(void);
 194 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
 195 extern void lttng_ring_buffer_client_discard_exit(void);
 196 extern void lttng_ring_buffer_client_discard_rt_exit(void);
 197 extern void lttng_ring_buffer_metadata_client_exit(void);
 198
 199 /*
 200  * Returns the HOME directory path. Caller MUST NOT free(3) the returned
 201  * pointer.
 202  */
 203 static
 204 const char *get_lttng_home_dir(void)
 205 {
 206        const char *val;
 207
 208        val = (const char *) getenv("LTTNG_HOME");
 209        if (val != NULL) {
 210                return val;
 211        }
 212        return (const char *) getenv("HOME");
 213 }
 214
 215 /*
 216  * Force a read (imply TLS fixup for dlopen) of TLS variables.
 217  */
 218 static
 219 void lttng_fixup_nest_count_tls(void)
 220 {
 221         asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
 222 }
 223
 224 int lttng_get_notify_socket(void *owner)
 225 {
 226         struct sock_info *info = owner;
 227
 228         return info->notify_socket;
 229 }
 230
 231 static
 232 void print_cmd(int cmd, int handle)
 233 {
 234         const char *cmd_name = "Unknown";
 235
 236         if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
 237                         && cmd_name_mapping[cmd]) {
 238                 cmd_name = cmd_name_mapping[cmd];
 239         }
 240         DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
 241                 cmd_name, cmd,
 242                 lttng_ust_obj_get_name(handle), handle);
 243 }
 244
 245 static
 246 int setup_local_apps(void)
 247 {
 248         const char *home_dir;
 249         uid_t uid;
 250
 251         uid = getuid();
 252         /*
 253          * Disallow per-user tracing for setuid binaries.
 254          */
 255         if (uid != geteuid()) {
 256                 assert(local_apps.allowed == 0);
 257                 return 0;
 258         }
 259         home_dir = get_lttng_home_dir();
 260         if (!home_dir) {
 261                 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
 262                 assert(local_apps.allowed == 0);
 263                 return -ENOENT;
 264         }
 265         local_apps.allowed = 1;
 266         snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
 267                 home_dir,
 268                 LTTNG_DEFAULT_HOME_RUNDIR,
 269                 LTTNG_UST_SOCK_FILENAME);
 270         snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
 271                 LTTNG_UST_WAIT_FILENAME,
 272                 uid);
 273         return 0;
 274 }
 275
 276 /*
 277  * Get notify_sock timeout, in ms.
 278  * -1: don't wait. 0: wait forever. >0: timeout, in ms.
 279  */
 280 static
 281 long get_timeout(void)
 282 {
 283         long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 284
 285         if (!got_timeout_env) {
 286                 str_timeout = getenv("LTTNG_UST_REGISTER_TIMEOUT");
 287                 got_timeout_env = 1;
 288         }
 289         if (str_timeout)
 290                 constructor_delay_ms = strtol(str_timeout, NULL, 10);
 291         return constructor_delay_ms;
 292 }
 293
 294 static
 295 long get_notify_sock_timeout(void)
 296 {
 297         return get_timeout();
 298 }
 299
 300 /*
 301  * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
 302  */
 303 static
 304 int get_constructor_timeout(struct timespec *constructor_timeout)
 305 {
 306         long constructor_delay_ms;
 307         int ret;
 308
 309         constructor_delay_ms = get_timeout();
 310
 311         switch (constructor_delay_ms) {
 312         case -1:/* fall-through */
 313         case 0:
 314                 return constructor_delay_ms;
 315         default:
 316                 break;
 317         }
 318
 319         /*
 320          * If we are unable to find the current time, don't wait.
 321          */
 322         ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
 323         if (ret) {
 324                 return -1;
 325         }
 326         constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
 327         constructor_timeout->tv_nsec +=
 328                 (constructor_delay_ms % 1000UL) * 1000000UL;
 329         if (constructor_timeout->tv_nsec >= 1000000000UL) {
 330                 constructor_timeout->tv_sec++;
 331                 constructor_timeout->tv_nsec -= 1000000000UL;
 332         }
 333         return 1;
 334 }
 335
 336 static
 337 int register_to_sessiond(int socket, enum ustctl_socket_type type)
 338 {
 339         return ustcomm_send_reg_msg(socket,
 340                 type,
 341                 CAA_BITS_PER_LONG,
 342                 lttng_alignof(uint8_t) * CHAR_BIT,
 343                 lttng_alignof(uint16_t) * CHAR_BIT,
 344                 lttng_alignof(uint32_t) * CHAR_BIT,
 345                 lttng_alignof(uint64_t) * CHAR_BIT,
 346                 lttng_alignof(unsigned long) * CHAR_BIT);
 347 }
 348
 349 static
 350 int send_reply(int sock, struct ustcomm_ust_reply *lur)
 351 {
 352         ssize_t len;
 353
 354         len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
 355         switch (len) {
 356         case sizeof(*lur):
 357                 DBG("message successfully sent");
 358                 return 0;
 359         default:
 360                 if (len == -ECONNRESET) {
 361                         DBG("remote end closed connection");
 362                         return 0;
 363                 }
 364                 if (len < 0)
 365                         return len;
 366                 DBG("incorrect message size: %zd", len);
 367                 return -EINVAL;
 368         }
 369 }
 370
 371 static
 372 int handle_register_done(struct sock_info *sock_info)
 373 {
 374         int ret;
 375
 376         if (sock_info->constructor_sem_posted)
 377                 return 0;
 378         sock_info->constructor_sem_posted = 1;
 379         if (uatomic_read(&sem_count) <= 0) {
 380                 return 0;
 381         }
 382         ret = uatomic_add_return(&sem_count, -1);
 383         if (ret == 0) {
 384                 ret = sem_post(&constructor_wait);
 385                 assert(!ret);
 386         }
 387         return 0;
 388 }
 389
 390 static
 391 int handle_message(struct sock_info *sock_info,
 392                 int sock, struct ustcomm_ust_msg *lum)
 393 {
 394         int ret = 0;
 395         const struct lttng_ust_objd_ops *ops;
 396         struct ustcomm_ust_reply lur;
 397         union ust_args args;
 398         ssize_t len;
 399
 400         ust_lock();
 401
 402         memset(&lur, 0, sizeof(lur));
 403
 404         if (lttng_ust_comm_should_quit) {
 405                 ret = -LTTNG_UST_ERR_EXITING;
 406                 goto end;
 407         }
 408
 409         ops = objd_ops(lum->handle);
 410         if (!ops) {
 411                 ret = -ENOENT;
 412                 goto end;
 413         }
 414
 415         switch (lum->cmd) {
 416         case LTTNG_UST_REGISTER_DONE:
 417                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 418                         ret = handle_register_done(sock_info);
 419                 else
 420                         ret = -EINVAL;
 421                 break;
 422         case LTTNG_UST_RELEASE:
 423                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 424                         ret = -EPERM;
 425                 else
 426                         ret = lttng_ust_objd_unref(lum->handle, 1);
 427                 break;
 428         case LTTNG_UST_FILTER:
 429         {
 430                 /* Receive filter data */
 431                 struct lttng_ust_filter_bytecode_node *bytecode;
 432
 433                 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
 434                         ERR("Filter data size is too large: %u bytes",
 435                                 lum->u.filter.data_size);
 436                         ret = -EINVAL;
 437                         goto error;
 438                 }
 439
 440                 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
 441                         ERR("Filter reloc offset %u is not within data",
 442                                 lum->u.filter.reloc_offset);
 443                         ret = -EINVAL;
 444                         goto error;
 445                 }
 446
 447                 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
 448                 if (!bytecode) {
 449                         ret = -ENOMEM;
 450                         goto error;
 451                 }
 452                 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
 453                                 lum->u.filter.data_size);
 454                 switch (len) {
 455                 case 0: /* orderly shutdown */
 456                         ret = 0;
 457                         free(bytecode);
 458                         goto error;
 459                 default:
 460                         if (len == lum->u.filter.data_size) {
 461                                 DBG("filter data received");
 462                                 break;
 463                         } else if (len < 0) {
 464                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 465                                 if (len == -ECONNRESET) {
 466                                         ERR("%s remote end closed connection", sock_info->name);
 467                                         ret = len;
 468                                         free(bytecode);
 469                                         goto error;
 470                                 }
 471                                 ret = len;
 472                                 free(bytecode);
 473                                 goto end;
 474                         } else {
 475                                 DBG("incorrect filter data message size: %zd", len);
 476                                 ret = -EINVAL;
 477                                 free(bytecode);
 478                                 goto end;
 479                         }
 480                 }
 481                 bytecode->bc.len = lum->u.filter.data_size;
 482                 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
 483                 bytecode->bc.seqnum = lum->u.filter.seqnum;
 484                 if (ops->cmd) {
 485                         ret = ops->cmd(lum->handle, lum->cmd,
 486                                         (unsigned long) bytecode,
 487                                         &args, sock_info);
 488                         if (ret) {
 489                                 free(bytecode);
 490                         }
 491                         /* don't free bytecode if everything went fine. */
 492                 } else {
 493                         ret = -ENOSYS;
 494                         free(bytecode);
 495                 }
 496                 break;
 497         }
 498         case LTTNG_UST_EXCLUSION:
 499         {
 500                 /* Receive exclusion names */
 501                 struct lttng_ust_excluder_node *node;
 502                 unsigned int count;
 503
 504                 count = lum->u.exclusion.count;
 505                 if (count == 0) {
 506                         /* There are no names to read */
 507                         ret = 0;
 508                         goto error;
 509                 }
 510                 node = zmalloc(sizeof(*node) +
 511                                 count * LTTNG_UST_SYM_NAME_LEN);
 512                 if (!node) {
 513                         ret = -ENOMEM;
 514                         goto error;
 515                 }
 516                 node->excluder.count = count;
 517                 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
 518                                 count * LTTNG_UST_SYM_NAME_LEN);
 519                 switch (len) {
 520                 case 0: /* orderly shutdown */
 521                         ret = 0;
 522                         free(node);
 523                         goto error;
 524                 default:
 525                         if (len == count * LTTNG_UST_SYM_NAME_LEN) {
 526                                 DBG("Exclusion data received");
 527                                 break;
 528                         } else if (len < 0) {
 529                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 530                                 if (len == -ECONNRESET) {
 531                                         ERR("%s remote end closed connection", sock_info->name);
 532                                         ret = len;
 533                                         free(node);
 534                                         goto error;
 535                                 }
 536                                 ret = len;
 537                                 free(node);
 538                                 goto end;
 539                         } else {
 540                                 DBG("Incorrect exclusion data message size: %zd", len);
 541                                 ret = -EINVAL;
 542                                 free(node);
 543                                 goto end;
 544                         }
 545                 }
 546                 if (ops->cmd) {
 547                         ret = ops->cmd(lum->handle, lum->cmd,
 548                                         (unsigned long) node,
 549                                         &args, sock_info);
 550                         if (ret) {
 551                                 free(node);
 552                         }
 553                         /* Don't free exclusion data if everything went fine. */
 554                 } else {
 555                         ret = -ENOSYS;
 556                         free(node);
 557                 }
 558                 break;
 559         }
 560         case LTTNG_UST_CHANNEL:
 561         {
 562                 void *chan_data;
 563                 int wakeup_fd;
 564
 565                 len = ustcomm_recv_channel_from_sessiond(sock,
 566                                 &chan_data, lum->u.channel.len,
 567                                 &wakeup_fd);
 568                 switch (len) {
 569                 case 0: /* orderly shutdown */
 570                         ret = 0;
 571                         goto error;
 572                 default:
 573                         if (len == lum->u.channel.len) {
 574                                 DBG("channel data received");
 575                                 break;
 576                         } else if (len < 0) {
 577                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 578                                 if (len == -ECONNRESET) {
 579                                         ERR("%s remote end closed connection", sock_info->name);
 580                                         ret = len;
 581                                         goto error;
 582                                 }
 583                                 ret = len;
 584                                 goto end;
 585                         } else {
 586                                 DBG("incorrect channel data message size: %zd", len);
 587                                 ret = -EINVAL;
 588                                 goto end;
 589                         }
 590                 }
 591                 args.channel.chan_data = chan_data;
 592                 args.channel.wakeup_fd = wakeup_fd;
 593                 if (ops->cmd)
 594                         ret = ops->cmd(lum->handle, lum->cmd,
 595                                         (unsigned long) &lum->u,
 596                                         &args, sock_info);
 597                 else
 598                         ret = -ENOSYS;
 599                 break;
 600         }
 601         case LTTNG_UST_STREAM:
 602         {
 603                 /* Receive shm_fd, wakeup_fd */
 604                 ret = ustcomm_recv_stream_from_sessiond(sock,
 605                         &lum->u.stream.len,
 606                         &args.stream.shm_fd,
 607                         &args.stream.wakeup_fd);
 608                 if (ret) {
 609                         goto end;
 610                 }
 611                 if (ops->cmd)
 612                         ret = ops->cmd(lum->handle, lum->cmd,
 613                                         (unsigned long) &lum->u,
 614                                         &args, sock_info);
 615                 else
 616                         ret = -ENOSYS;
 617                 break;
 618         }
 619         default:
 620                 if (ops->cmd)
 621                         ret = ops->cmd(lum->handle, lum->cmd,
 622                                         (unsigned long) &lum->u,
 623                                         &args, sock_info);
 624                 else
 625                         ret = -ENOSYS;
 626                 break;
 627         }
 628
 629 end:
 630         lur.handle = lum->handle;
 631         lur.cmd = lum->cmd;
 632         lur.ret_val = ret;
 633         if (ret >= 0) {
 634                 lur.ret_code = LTTNG_UST_OK;
 635         } else {
 636                 /*
 637                  * Use -LTTNG_UST_ERR as wildcard for UST internal
 638                  * error that are not caused by the transport, except if
 639                  * we already have a more precise error message to
 640                  * report.
 641                  */
 642                 if (ret > -LTTNG_UST_ERR) {
 643                         /* Translate code to UST error. */
 644                         switch (ret) {
 645                         case -EEXIST:
 646                                 lur.ret_code = -LTTNG_UST_ERR_EXIST;
 647                                 break;
 648                         case -EINVAL:
 649                                 lur.ret_code = -LTTNG_UST_ERR_INVAL;
 650                                 break;
 651                         case -ENOENT:
 652                                 lur.ret_code = -LTTNG_UST_ERR_NOENT;
 653                                 break;
 654                         case -EPERM:
 655                                 lur.ret_code = -LTTNG_UST_ERR_PERM;
 656                                 break;
 657                         case -ENOSYS:
 658                                 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
 659                                 break;
 660                         default:
 661                                 lur.ret_code = -LTTNG_UST_ERR;
 662                                 break;
 663                         }
 664                 } else {
 665                         lur.ret_code = ret;
 666                 }
 667         }
 668         if (ret >= 0) {
 669                 switch (lum->cmd) {
 670                 case LTTNG_UST_TRACER_VERSION:
 671                         lur.u.version = lum->u.version;
 672                         break;
 673                 case LTTNG_UST_TRACEPOINT_LIST_GET:
 674                         memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
 675                         break;
 676                 }
 677         }
 678         DBG("Return value: %d", lur.ret_val);
 679         ret = send_reply(sock, &lur);
 680         if (ret < 0) {
 681                 DBG("error sending reply");
 682                 goto error;
 683         }
 684
 685         /*
 686          * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
 687          * after the reply.
 688          */
 689         if (lur.ret_code == LTTNG_UST_OK) {
 690                 switch (lum->cmd) {
 691                 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
 692                         len = ustcomm_send_unix_sock(sock,
 693                                 &args.field_list.entry,
 694                                 sizeof(args.field_list.entry));
 695                         if (len < 0) {
 696                                 ret = len;
 697                                 goto error;
 698                         }
 699                         if (len != sizeof(args.field_list.entry)) {
 700                                 ret = -EINVAL;
 701                                 goto error;
 702                         }
 703                 }
 704         }
 705
 706 error:
 707         ust_unlock();
 708         return ret;
 709 }
 710
 711 static
 712 void handle_pending_statedumps(struct sock_info *sock_info)
 713 {
 714         int ctor_passed = sock_info->constructor_sem_posted;
 715
 716         if (ctor_passed && sock_info->session_enabled) {
 717                 sock_info->session_enabled = 0;
 718                 lttng_handle_pending_statedumps(&lttng_ust_baddr_statedump);
 719         }
 720 }
 721
 722 static
 723 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
 724 {
 725         int ret;
 726
 727         if (sock_info->root_handle != -1) {
 728                 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
 729                 if (ret) {
 730                         ERR("Error unref root handle");
 731                 }
 732                 sock_info->root_handle = -1;
 733         }
 734         sock_info->constructor_sem_posted = 0;
 735
 736         /*
 737          * wait_shm_mmap, socket and notify socket are used by listener
 738          * threads outside of the ust lock, so we cannot tear them down
 739          * ourselves, because we cannot join on these threads. Leave
 740          * responsibility of cleaning up these resources to the OS
 741          * process exit.
 742          */
 743         if (exiting)
 744                 return;
 745
 746         if (sock_info->socket != -1) {
 747                 ret = ustcomm_close_unix_sock(sock_info->socket);
 748                 if (ret) {
 749                         ERR("Error closing ust cmd socket");
 750                 }
 751                 sock_info->socket = -1;
 752         }
 753         if (sock_info->notify_socket != -1) {
 754                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
 755                 if (ret) {
 756                         ERR("Error closing ust notify socket");
 757                 }
 758                 sock_info->notify_socket = -1;
 759         }
 760         if (sock_info->wait_shm_mmap) {
 761                 ret = munmap(sock_info->wait_shm_mmap, sysconf(_SC_PAGE_SIZE));
 762                 if (ret) {
 763                         ERR("Error unmapping wait shm");
 764                 }
 765                 sock_info->wait_shm_mmap = NULL;
 766         }
 767 }
 768
 769 /*
 770  * Using fork to set umask in the child process (not multi-thread safe).
 771  * We deal with the shm_open vs ftruncate race (happening when the
 772  * sessiond owns the shm and does not let everybody modify it, to ensure
 773  * safety against shm_unlink) by simply letting the mmap fail and
 774  * retrying after a few seconds.
 775  * For global shm, everybody has rw access to it until the sessiond
 776  * starts.
 777  */
 778 static
 779 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
 780 {
 781         int wait_shm_fd, ret;
 782         pid_t pid;
 783
 784         /*
 785          * Try to open read-only.
 786          */
 787         wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
 788         if (wait_shm_fd >= 0) {
 789                 int32_t tmp_read;
 790                 ssize_t len;
 791                 size_t bytes_read = 0;
 792
 793                 /*
 794                  * Try to read the fd. If unable to do so, try opening
 795                  * it in write mode.
 796                  */
 797                 do {
 798                         len = read(wait_shm_fd,
 799                                 &((char *) &tmp_read)[bytes_read],
 800                                 sizeof(tmp_read) - bytes_read);
 801                         if (len > 0) {
 802                                 bytes_read += len;
 803                         }
 804                 } while ((len < 0 && errno == EINTR)
 805                         || (len > 0 && bytes_read < sizeof(tmp_read)));
 806                 if (bytes_read != sizeof(tmp_read)) {
 807                         ret = close(wait_shm_fd);
 808                         if (ret) {
 809                                 ERR("close wait_shm_fd");
 810                         }
 811                         goto open_write;
 812                 }
 813                 goto end;
 814         } else if (wait_shm_fd < 0 && errno != ENOENT) {
 815                 /*
 816                  * Real-only open did not work, and it's not because the
 817                  * entry was not present. It's a failure that prohibits
 818                  * using shm.
 819                  */
 820                 ERR("Error opening shm %s", sock_info->wait_shm_path);
 821                 goto end;
 822         }
 823
 824 open_write:
 825         /*
 826          * If the open failed because the file did not exist, or because
 827          * the file was not truncated yet, try creating it ourself.
 828          */
 829         URCU_TLS(lttng_ust_nest_count)++;
 830         pid = fork();
 831         URCU_TLS(lttng_ust_nest_count)--;
 832         if (pid > 0) {
 833                 int status;
 834
 835                 /*
 836                  * Parent: wait for child to return, in which case the
 837                  * shared memory map will have been created.
 838                  */
 839                 pid = wait(&status);
 840                 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 841                         wait_shm_fd = -1;
 842                         goto end;
 843                 }
 844                 /*
 845                  * Try to open read-only again after creation.
 846                  */
 847                 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
 848                 if (wait_shm_fd < 0) {
 849                         /*
 850                          * Real-only open did not work. It's a failure
 851                          * that prohibits using shm.
 852                          */
 853                         ERR("Error opening shm %s", sock_info->wait_shm_path);
 854                         goto end;
 855                 }
 856                 goto end;
 857         } else if (pid == 0) {
 858                 int create_mode;
 859
 860                 /* Child */
 861                 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
 862                 if (sock_info->global)
 863                         create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
 864                 /*
 865                  * We're alone in a child process, so we can modify the
 866                  * process-wide umask.
 867                  */
 868                 umask(~create_mode);
 869                 /*
 870                  * Try creating shm (or get rw access).
 871                  * We don't do an exclusive open, because we allow other
 872                  * processes to create+ftruncate it concurrently.
 873                  */
 874                 wait_shm_fd = shm_open(sock_info->wait_shm_path,
 875                                 O_RDWR | O_CREAT, create_mode);
 876                 if (wait_shm_fd >= 0) {
 877                         ret = ftruncate(wait_shm_fd, mmap_size);
 878                         if (ret) {
 879                                 PERROR("ftruncate");
 880                                 _exit(EXIT_FAILURE);
 881                         }
 882                         _exit(EXIT_SUCCESS);
 883                 }
 884                 /*
 885                  * For local shm, we need to have rw access to accept
 886                  * opening it: this means the local sessiond will be
 887                  * able to wake us up. For global shm, we open it even
 888                  * if rw access is not granted, because the root.root
 889                  * sessiond will be able to override all rights and wake
 890                  * us up.
 891                  */
 892                 if (!sock_info->global && errno != EACCES) {
 893                         ERR("Error opening shm %s", sock_info->wait_shm_path);
 894                         _exit(EXIT_FAILURE);
 895                 }
 896                 /*
 897                  * The shm exists, but we cannot open it RW. Report
 898                  * success.
 899                  */
 900                 _exit(EXIT_SUCCESS);
 901         } else {
 902                 return -1;
 903         }
 904 end:
 905         if (wait_shm_fd >= 0 && !sock_info->global) {
 906                 struct stat statbuf;
 907
 908                 /*
 909                  * Ensure that our user is the owner of the shm file for
 910                  * local shm. If we do not own the file, it means our
 911                  * sessiond will not have access to wake us up (there is
 912                  * probably a rogue process trying to fake our
 913                  * sessiond). Fallback to polling method in this case.
 914                  */
 915                 ret = fstat(wait_shm_fd, &statbuf);
 916                 if (ret) {
 917                         PERROR("fstat");
 918                         goto error_close;
 919                 }
 920                 if (statbuf.st_uid != getuid())
 921                         goto error_close;
 922         }
 923         return wait_shm_fd;
 924
 925 error_close:
 926         ret = close(wait_shm_fd);
 927         if (ret) {
 928                 PERROR("Error closing fd");
 929         }
 930         return -1;
 931 }
 932
 933 static
 934 char *get_map_shm(struct sock_info *sock_info)
 935 {
 936         size_t mmap_size = sysconf(_SC_PAGE_SIZE);
 937         int wait_shm_fd, ret;
 938         char *wait_shm_mmap;
 939
 940         wait_shm_fd = get_wait_shm(sock_info, mmap_size);
 941         if (wait_shm_fd < 0) {
 942                 goto error;
 943         }
 944         wait_shm_mmap = mmap(NULL, mmap_size, PROT_READ,
 945                   MAP_SHARED, wait_shm_fd, 0);
 946         /* close shm fd immediately after taking the mmap reference */
 947         ret = close(wait_shm_fd);
 948         if (ret) {
 949                 PERROR("Error closing fd");
 950         }
 951         if (wait_shm_mmap == MAP_FAILED) {
 952                 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
 953                 goto error;
 954         }
 955         return wait_shm_mmap;
 956
 957 error:
 958         return NULL;
 959 }
 960
 961 static
 962 void wait_for_sessiond(struct sock_info *sock_info)
 963 {
 964         int ret;
 965
 966         ust_lock();
 967         if (lttng_ust_comm_should_quit) {
 968                 goto quit;
 969         }
 970         if (wait_poll_fallback) {
 971                 goto error;
 972         }
 973         if (!sock_info->wait_shm_mmap) {
 974                 sock_info->wait_shm_mmap = get_map_shm(sock_info);
 975                 if (!sock_info->wait_shm_mmap)
 976                         goto error;
 977         }
 978         ust_unlock();
 979
 980         DBG("Waiting for %s apps sessiond", sock_info->name);
 981         /* Wait for futex wakeup */
 982         if (uatomic_read((int32_t *) sock_info->wait_shm_mmap) == 0) {
 983                 ret = futex_async((int32_t *) sock_info->wait_shm_mmap,
 984                         FUTEX_WAIT, 0, NULL, NULL, 0);
 985                 if (ret < 0) {
 986                         if (errno == EFAULT) {
 987                                 wait_poll_fallback = 1;
 988                                 DBG(
 989 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
 990 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
 991 "Please upgrade your kernel "
 992 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
 993 "mainline). LTTng-UST will use polling mode fallback.");
 994                                 if (ust_debug())
 995                                         PERROR("futex");
 996                         }
 997                 }
 998         }
 999         return;
1000
1001 quit:
1002         ust_unlock();
1003         return;
1004
1005 error:
1006         ust_unlock();
1007         return;
1008 }
1009
1010 /*
1011  * This thread does not allocate any resource, except within
1012  * handle_message, within mutex protection. This mutex protects against
1013  * fork and exit.
1014  * The other moment it allocates resources is at socket connection, which
1015  * is also protected by the mutex.
1016  */
1017 static
1018 void *ust_listener_thread(void *arg)
1019 {
1020         struct sock_info *sock_info = arg;
1021         int sock, ret, prev_connect_failed = 0, has_waited = 0;
1022         long timeout;
1023
1024         /* Restart trying to connect to the session daemon */
1025 restart:
1026         if (prev_connect_failed) {
1027                 /* Wait for sessiond availability with pipe */
1028                 wait_for_sessiond(sock_info);
1029                 if (has_waited) {
1030                         has_waited = 0;
1031                         /*
1032                          * Sleep for 5 seconds before retrying after a
1033                          * sequence of failure / wait / failure. This
1034                          * deals with a killed or broken session daemon.
1035                          */
1036                         sleep(5);
1037                 }
1038                 has_waited = 1;
1039                 prev_connect_failed = 0;
1040         }
1041
1042         if (sock_info->socket != -1) {
1043                 ret = ustcomm_close_unix_sock(sock_info->socket);
1044                 if (ret) {
1045                         ERR("Error closing %s ust cmd socket",
1046                                 sock_info->name);
1047                 }
1048                 sock_info->socket = -1;
1049         }
1050         if (sock_info->notify_socket != -1) {
1051                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1052                 if (ret) {
1053                         ERR("Error closing %s ust notify socket",
1054                                 sock_info->name);
1055                 }
1056                 sock_info->notify_socket = -1;
1057         }
1058
1059         /*
1060          * Register. We need to perform both connect and sending
1061          * registration message before doing the next connect otherwise
1062          * we may reach unix socket connect queue max limits and block
1063          * on the 2nd connect while the session daemon is awaiting the
1064          * first connect registration message.
1065          */
1066         /* Connect cmd socket */
1067         ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1068         if (ret < 0) {
1069                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1070                 prev_connect_failed = 1;
1071
1072                 ust_lock();
1073
1074                 if (lttng_ust_comm_should_quit) {
1075                         goto quit;
1076                 }
1077
1078                 /*
1079                  * If we cannot find the sessiond daemon, don't delay
1080                  * constructor execution.
1081                  */
1082                 ret = handle_register_done(sock_info);
1083                 assert(!ret);
1084                 ust_unlock();
1085                 goto restart;
1086         }
1087         sock_info->socket = ret;
1088
1089         ust_lock();
1090
1091         if (lttng_ust_comm_should_quit) {
1092                 goto quit;
1093         }
1094
1095         /*
1096          * Create only one root handle per listener thread for the whole
1097          * process lifetime, so we ensure we get ID which is statically
1098          * assigned to the root handle.
1099          */
1100         if (sock_info->root_handle == -1) {
1101                 ret = lttng_abi_create_root_handle();
1102                 if (ret < 0) {
1103                         ERR("Error creating root handle");
1104                         goto quit;
1105                 }
1106                 sock_info->root_handle = ret;
1107         }
1108
1109         ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1110         if (ret < 0) {
1111                 ERR("Error registering to %s ust cmd socket",
1112                         sock_info->name);
1113                 prev_connect_failed = 1;
1114                 /*
1115                  * If we cannot register to the sessiond daemon, don't
1116                  * delay constructor execution.
1117                  */
1118                 ret = handle_register_done(sock_info);
1119                 assert(!ret);
1120                 ust_unlock();
1121                 goto restart;
1122         }
1123
1124         ust_unlock();
1125
1126         /* Connect notify socket */
1127         ret = ustcomm_connect_unix_sock(sock_info->sock_path);
1128         if (ret < 0) {
1129                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1130                 prev_connect_failed = 1;
1131
1132                 ust_lock();
1133
1134                 if (lttng_ust_comm_should_quit) {
1135                         goto quit;
1136                 }
1137
1138                 /*
1139                  * If we cannot find the sessiond daemon, don't delay
1140                  * constructor execution.
1141                  */
1142                 ret = handle_register_done(sock_info);
1143                 assert(!ret);
1144                 ust_unlock();
1145                 goto restart;
1146         }
1147         sock_info->notify_socket = ret;
1148
1149         timeout = get_notify_sock_timeout();
1150         if (timeout >= 0) {
1151                 /*
1152                  * Give at least 10ms to sessiond to reply to
1153                  * notifications.
1154                  */
1155                 if (timeout < 10)
1156                         timeout = 10;
1157                 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1158                                 timeout);
1159                 if (ret < 0) {
1160                         WARN("Error setting socket receive timeout");
1161                 }
1162                 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1163                                 timeout);
1164                 if (ret < 0) {
1165                         WARN("Error setting socket send timeout");
1166                 }
1167         } else if (timeout < -1) {
1168                 WARN("Unsupported timeout value %ld", timeout);
1169         }
1170
1171         ust_lock();
1172
1173         if (lttng_ust_comm_should_quit) {
1174                 goto quit;
1175         }
1176
1177         ret = register_to_sessiond(sock_info->notify_socket,
1178                         USTCTL_SOCKET_NOTIFY);
1179         if (ret < 0) {
1180                 ERR("Error registering to %s ust notify socket",
1181                         sock_info->name);
1182                 prev_connect_failed = 1;
1183                 /*
1184                  * If we cannot register to the sessiond daemon, don't
1185                  * delay constructor execution.
1186                  */
1187                 ret = handle_register_done(sock_info);
1188                 assert(!ret);
1189                 ust_unlock();
1190                 goto restart;
1191         }
1192         sock = sock_info->socket;
1193
1194         ust_unlock();
1195
1196         for (;;) {
1197                 ssize_t len;
1198                 struct ustcomm_ust_msg lum;
1199
1200                 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1201                 switch (len) {
1202                 case 0: /* orderly shutdown */
1203                         DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1204                         ust_lock();
1205                         if (lttng_ust_comm_should_quit) {
1206                                 goto quit;
1207                         }
1208                         /*
1209                          * Either sessiond has shutdown or refused us by closing the socket.
1210                          * In either case, we don't want to delay construction execution,
1211                          * and we need to wait before retry.
1212                          */
1213                         prev_connect_failed = 1;
1214                         /*
1215                          * If we cannot register to the sessiond daemon, don't
1216                          * delay constructor execution.
1217                          */
1218                         ret = handle_register_done(sock_info);
1219                         assert(!ret);
1220                         ust_unlock();
1221                         goto end;
1222                 case sizeof(lum):
1223                         print_cmd(lum.cmd, lum.handle);
1224                         ret = handle_message(sock_info, sock, &lum);
1225                         if (ret) {
1226                                 ERR("Error handling message for %s socket", sock_info->name);
1227                         } else {
1228                                 handle_pending_statedumps(sock_info);
1229                         }
1230                         continue;
1231                 default:
1232                         if (len < 0) {
1233                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1234                         } else {
1235                                 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1236                         }
1237                         if (len == -ECONNRESET) {
1238                                 DBG("%s remote end closed connection", sock_info->name);
1239                                 goto end;
1240                         }
1241                         goto end;
1242                 }
1243
1244         }
1245 end:
1246         ust_lock();
1247         if (lttng_ust_comm_should_quit) {
1248                 goto quit;
1249         }
1250         /* Cleanup socket handles before trying to reconnect */
1251         lttng_ust_objd_table_owner_cleanup(sock_info);
1252         ust_unlock();
1253         goto restart;   /* try to reconnect */
1254
1255 quit:
1256         sock_info->thread_active = 0;
1257         ust_unlock();
1258         return NULL;
1259 }
1260
1261 /*
1262  * sessiond monitoring thread: monitor presence of global and per-user
1263  * sessiond by polling the application common named pipe.
1264  */
1265 void __attribute__((constructor)) lttng_ust_init(void)
1266 {
1267         struct timespec constructor_timeout;
1268         sigset_t sig_all_blocked, orig_parent_mask;
1269         pthread_attr_t thread_attr;
1270         int timeout_mode;
1271         int ret;
1272
1273         if (uatomic_xchg(&initialized, 1) == 1)
1274                 return;
1275
1276         /*
1277          * Fixup interdependency between TLS fixup mutex (which happens
1278          * to be the dynamic linker mutex) and ust_lock, taken within
1279          * the ust lock.
1280          */
1281         lttng_fixup_ringbuffer_tls();
1282         lttng_fixup_vtid_tls();
1283         lttng_fixup_nest_count_tls();
1284         lttng_fixup_procname_tls();
1285
1286         /*
1287          * We want precise control over the order in which we construct
1288          * our sub-libraries vs starting to receive commands from
1289          * sessiond (otherwise leading to errors when trying to create
1290          * sessiond before the init functions are completed).
1291          */
1292         init_usterr();
1293         init_tracepoint();
1294         lttng_ring_buffer_metadata_client_init();
1295         lttng_ring_buffer_client_overwrite_init();
1296         lttng_ring_buffer_client_overwrite_rt_init();
1297         lttng_ring_buffer_client_discard_init();
1298         lttng_ring_buffer_client_discard_rt_init();
1299         lttng_context_init();
1300
1301         timeout_mode = get_constructor_timeout(&constructor_timeout);
1302
1303         ret = sem_init(&constructor_wait, 0, 0);
1304         assert(!ret);
1305
1306         ret = setup_local_apps();
1307         if (ret) {
1308                 DBG("local apps setup returned %d", ret);
1309         }
1310
1311         /* A new thread created by pthread_create inherits the signal mask
1312          * from the parent. To avoid any signal being received by the
1313          * listener thread, we block all signals temporarily in the parent,
1314          * while we create the listener thread.
1315          */
1316         sigfillset(&sig_all_blocked);
1317         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1318         if (ret) {
1319                 ERR("pthread_sigmask: %s", strerror(ret));
1320         }
1321
1322         ret = pthread_attr_init(&thread_attr);
1323         if (ret) {
1324                 ERR("pthread_attr_init: %s", strerror(ret));
1325         }
1326         ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1327         if (ret) {
1328                 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1329         }
1330
1331         ust_lock();
1332         ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1333                         ust_listener_thread, &global_apps);
1334         if (ret) {
1335                 ERR("pthread_create global: %s", strerror(ret));
1336         }
1337         global_apps.thread_active = 1;
1338         ust_unlock();
1339
1340         if (local_apps.allowed) {
1341                 ust_lock();
1342                 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1343                                 ust_listener_thread, &local_apps);
1344                 if (ret) {
1345                         ERR("pthread_create local: %s", strerror(ret));
1346                 }
1347                 local_apps.thread_active = 1;
1348                 ust_unlock();
1349         } else {
1350                 handle_register_done(&local_apps);
1351         }
1352         ret = pthread_attr_destroy(&thread_attr);
1353         if (ret) {
1354                 ERR("pthread_attr_destroy: %s", strerror(ret));
1355         }
1356
1357         /* Restore original signal mask in parent */
1358         ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1359         if (ret) {
1360                 ERR("pthread_sigmask: %s", strerror(ret));
1361         }
1362
1363         switch (timeout_mode) {
1364         case 1: /* timeout wait */
1365                 do {
1366                         ret = sem_timedwait(&constructor_wait,
1367                                         &constructor_timeout);
1368                 } while (ret < 0 && errno == EINTR);
1369                 if (ret < 0 && errno == ETIMEDOUT) {
1370                         ERR("Timed out waiting for lttng-sessiond");
1371                 } else {
1372                         assert(!ret);
1373                 }
1374                 break;
1375         case -1:/* wait forever */
1376                 do {
1377                         ret = sem_wait(&constructor_wait);
1378                 } while (ret < 0 && errno == EINTR);
1379                 assert(!ret);
1380                 break;
1381         case 0: /* no timeout */
1382                 break;
1383         }
1384 }
1385
1386 static
1387 void lttng_ust_cleanup(int exiting)
1388 {
1389         cleanup_sock_info(&global_apps, exiting);
1390         if (local_apps.allowed) {
1391                 cleanup_sock_info(&local_apps, exiting);
1392         }
1393         /*
1394          * The teardown in this function all affect data structures
1395          * accessed under the UST lock by the listener thread. This
1396          * lock, along with the lttng_ust_comm_should_quit flag, ensure
1397          * that none of these threads are accessing this data at this
1398          * point.
1399          */
1400         lttng_ust_abi_exit();
1401         lttng_ust_events_exit();
1402         lttng_context_exit();
1403         lttng_ring_buffer_client_discard_rt_exit();
1404         lttng_ring_buffer_client_discard_exit();
1405         lttng_ring_buffer_client_overwrite_rt_exit();
1406         lttng_ring_buffer_client_overwrite_exit();
1407         lttng_ring_buffer_metadata_client_exit();
1408         exit_tracepoint();
1409         if (!exiting) {
1410                 /* Reinitialize values for fork */
1411                 sem_count = 2;
1412                 lttng_ust_comm_should_quit = 0;
1413                 initialized = 0;
1414         }
1415 }
1416
1417 void __attribute__((destructor)) lttng_ust_exit(void)
1418 {
1419         int ret;
1420
1421         /*
1422          * Using pthread_cancel here because:
1423          * A) we don't want to hang application teardown.
1424          * B) the thread is not allocating any resource.
1425          */
1426
1427         /*
1428          * Require the communication thread to quit. Synchronize with
1429          * mutexes to ensure it is not in a mutex critical section when
1430          * pthread_cancel is later called.
1431          */
1432         ust_lock();
1433         lttng_ust_comm_should_quit = 1;
1434
1435         /* cancel threads */
1436         if (global_apps.thread_active) {
1437                 ret = pthread_cancel(global_apps.ust_listener);
1438                 if (ret) {
1439                         ERR("Error cancelling global ust listener thread: %s",
1440                                 strerror(ret));
1441                 } else {
1442                         global_apps.thread_active = 0;
1443                 }
1444         }
1445         if (local_apps.thread_active) {
1446                 ret = pthread_cancel(local_apps.ust_listener);
1447                 if (ret) {
1448                         ERR("Error cancelling local ust listener thread: %s",
1449                                 strerror(ret));
1450                 } else {
1451                         local_apps.thread_active = 0;
1452                 }
1453         }
1454         ust_unlock();
1455
1456         /*
1457          * Do NOT join threads: use of sys_futex makes it impossible to
1458          * join the threads without using async-cancel, but async-cancel
1459          * is delivered by a signal, which could hit the target thread
1460          * anywhere in its code path, including while the ust_lock() is
1461          * held, causing a deadlock for the other thread. Let the OS
1462          * cleanup the threads if there are stalled in a syscall.
1463          */
1464         lttng_ust_cleanup(1);
1465 }
1466
1467 /*
1468  * We exclude the worker threads across fork and clone (except
1469  * CLONE_VM), because these system calls only keep the forking thread
1470  * running in the child.  Therefore, we don't want to call fork or clone
1471  * in the middle of an tracepoint or ust tracing state modification.
1472  * Holding this mutex protects these structures across fork and clone.
1473  */
1474 void ust_before_fork(sigset_t *save_sigset)
1475 {
1476         /*
1477          * Disable signals. This is to avoid that the child intervenes
1478          * before it is properly setup for tracing. It is safer to
1479          * disable all signals, because then we know we are not breaking
1480          * anything by restoring the original mask.
1481          */
1482         sigset_t all_sigs;
1483         int ret;
1484
1485         if (URCU_TLS(lttng_ust_nest_count))
1486                 return;
1487         /* Disable signals */
1488         sigfillset(&all_sigs);
1489         ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
1490         if (ret == -1) {
1491                 PERROR("sigprocmask");
1492         }
1493         ust_lock();
1494         rcu_bp_before_fork();
1495 }
1496
1497 static void ust_after_fork_common(sigset_t *restore_sigset)
1498 {
1499         int ret;
1500
1501         DBG("process %d", getpid());
1502         ust_unlock();
1503         /* Restore signals */
1504         ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
1505         if (ret == -1) {
1506                 PERROR("sigprocmask");
1507         }
1508 }
1509
1510 void ust_after_fork_parent(sigset_t *restore_sigset)
1511 {
1512         if (URCU_TLS(lttng_ust_nest_count))
1513                 return;
1514         DBG("process %d", getpid());
1515         rcu_bp_after_fork_parent();
1516         /* Release mutexes and reenable signals */
1517         ust_after_fork_common(restore_sigset);
1518 }
1519
1520 /*
1521  * After fork, in the child, we need to cleanup all the leftover state,
1522  * except the worker thread which already magically disappeared thanks
1523  * to the weird Linux fork semantics. After tyding up, we call
1524  * lttng_ust_init() again to start over as a new PID.
1525  *
1526  * This is meant for forks() that have tracing in the child between the
1527  * fork and following exec call (if there is any).
1528  */
1529 void ust_after_fork_child(sigset_t *restore_sigset)
1530 {
1531         if (URCU_TLS(lttng_ust_nest_count))
1532                 return;
1533         DBG("process %d", getpid());
1534         /* Release urcu mutexes */
1535         rcu_bp_after_fork_child();
1536         lttng_ust_cleanup(0);
1537         lttng_context_vtid_reset();
1538         /* Release mutexes and reenable signals */
1539         ust_after_fork_common(restore_sigset);
1540         lttng_ust_init();
1541 }
1542
1543 void lttng_ust_sockinfo_session_enabled(void *owner)
1544 {
1545         struct sock_info *sock_info = owner;
1546         sock_info->session_enabled = 1;
1547 }