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 #define _GNU_SOURCE
  24 #include <sys/types.h>
  25 #include <sys/socket.h>
  26 #include <sys/mman.h>
  27 #include <sys/stat.h>
  28 #include <sys/types.h>
  29 #include <sys/wait.h>
  30 #include <dlfcn.h>
  31 #include <fcntl.h>
  32 #include <unistd.h>
  33 #include <errno.h>
  34 #include <pthread.h>
  35 #include <semaphore.h>
  36 #include <time.h>
  37 #include <assert.h>
  38 #include <signal.h>
  39 #include <limits.h>
  40 #include <urcu/uatomic.h>
  41 #include <urcu/futex.h>
  42 #include <urcu/compiler.h>
  43
  44 #include <lttng/ust-events.h>
  45 #include <lttng/ust-abi.h>
  46 #include <lttng/ust.h>
  47 #include <lttng/ust-error.h>
  48 #include <lttng/ust-ctl.h>
  49 #include <urcu/tls-compat.h>
  50 #include <ust-comm.h>
  51 #include <ust-fd.h>
  52 #include <usterr-signal-safe.h>
  53 #include <helper.h>
  54 #include "tracepoint-internal.h"
  55 #include "lttng-tracer-core.h"
  56 #include "compat.h"
  57 #include "../libringbuffer/rb-init.h"
  58 #include "lttng-ust-statedump.h"
  59 #include "clock.h"
  60 #include "../libringbuffer/getcpu.h"
  61 #include "getenv.h"
  62
  63 /* Concatenate lttng ust shared library name with its major version number. */
  64 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." LTTNG_UST_LIBRARY_VERSION_MAJOR
  65
  66 /*
  67  * Has lttng ust comm constructor been called ?
  68  */
  69 static int initialized;
  70
  71 /*
  72  * The ust_lock/ust_unlock lock is used as a communication thread mutex.
  73  * Held when handling a command, also held by fork() to deal with
  74  * removal of threads, and by exit path.
  75  *
  76  * The UST lock is the centralized mutex across UST tracing control and
  77  * probe registration.
  78  *
  79  * ust_exit_mutex must never nest in ust_mutex.
  80  *
  81  * ust_fork_mutex must never nest in ust_mutex.
  82  *
  83  * ust_mutex_nest is a per-thread nesting counter, allowing the perf
  84  * counter lazy initialization called by events within the statedump,
  85  * which traces while the ust_mutex is held.
  86  *
  87  * ust_lock nests within the dynamic loader lock (within glibc) because
  88  * it is taken within the library constructor.
  89  *
  90  * The ust fd tracker lock nests within the ust_mutex.
  91  */
  92 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
  93
  94 /* Allow nesting the ust_mutex within the same thread. */
  95 static DEFINE_URCU_TLS(int, ust_mutex_nest);
  96
  97 /*
  98  * ust_exit_mutex protects thread_active variable wrt thread exit. It
  99  * cannot be done by ust_mutex because pthread_cancel(), which takes an
 100  * internal libc lock, cannot nest within ust_mutex.
 101  *
 102  * It never nests within a ust_mutex.
 103  */
 104 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
 105
 106 /*
 107  * ust_fork_mutex protects base address statedump tracing against forks. It
 108  * prevents the dynamic loader lock to be taken (by base address statedump
 109  * tracing) while a fork is happening, thus preventing deadlock issues with
 110  * the dynamic loader lock.
 111  */
 112 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
 113
 114 /* Should the ust comm thread quit ? */
 115 static int lttng_ust_comm_should_quit;
 116
 117 /*
 118  * This variable can be tested by applications to check whether
 119  * lttng-ust is loaded. They simply have to define their own
 120  * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
 121  * library constructor.
 122  */
 123 int lttng_ust_loaded __attribute__((weak));
 124
 125 /*
 126  * Return 0 on success, -1 if should quit.
 127  * The lock is taken in both cases.
 128  * Signal-safe.
 129  */
 130 int ust_lock(void)
 131 {
 132         sigset_t sig_all_blocked, orig_mask;
 133         int ret, oldstate;
 134
 135         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
 136         if (ret) {
 137                 ERR("pthread_setcancelstate: %s", strerror(ret));
 138         }
 139         if (oldstate != PTHREAD_CANCEL_ENABLE) {
 140                 ERR("pthread_setcancelstate: unexpected oldstate");
 141         }
 142         sigfillset(&sig_all_blocked);
 143         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 144         if (ret) {
 145                 ERR("pthread_sigmask: %s", strerror(ret));
 146         }
 147         if (!URCU_TLS(ust_mutex_nest)++)
 148                 pthread_mutex_lock(&ust_mutex);
 149         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 150         if (ret) {
 151                 ERR("pthread_sigmask: %s", strerror(ret));
 152         }
 153         if (lttng_ust_comm_should_quit) {
 154                 return -1;
 155         } else {
 156                 return 0;
 157         }
 158 }
 159
 160 /*
 161  * ust_lock_nocheck() can be used in constructors/destructors, because
 162  * they are already nested within the dynamic loader lock, and therefore
 163  * have exclusive access against execution of liblttng-ust destructor.
 164  * Signal-safe.
 165  */
 166 void ust_lock_nocheck(void)
 167 {
 168         sigset_t sig_all_blocked, orig_mask;
 169         int ret, oldstate;
 170
 171         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
 172         if (ret) {
 173                 ERR("pthread_setcancelstate: %s", strerror(ret));
 174         }
 175         if (oldstate != PTHREAD_CANCEL_ENABLE) {
 176                 ERR("pthread_setcancelstate: unexpected oldstate");
 177         }
 178         sigfillset(&sig_all_blocked);
 179         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 180         if (ret) {
 181                 ERR("pthread_sigmask: %s", strerror(ret));
 182         }
 183         if (!URCU_TLS(ust_mutex_nest)++)
 184                 pthread_mutex_lock(&ust_mutex);
 185         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 186         if (ret) {
 187                 ERR("pthread_sigmask: %s", strerror(ret));
 188         }
 189 }
 190
 191 /*
 192  * Signal-safe.
 193  */
 194 void ust_unlock(void)
 195 {
 196         sigset_t sig_all_blocked, orig_mask;
 197         int ret, oldstate;
 198
 199         sigfillset(&sig_all_blocked);
 200         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 201         if (ret) {
 202                 ERR("pthread_sigmask: %s", strerror(ret));
 203         }
 204         if (!--URCU_TLS(ust_mutex_nest))
 205                 pthread_mutex_unlock(&ust_mutex);
 206         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 207         if (ret) {
 208                 ERR("pthread_sigmask: %s", strerror(ret));
 209         }
 210         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
 211         if (ret) {
 212                 ERR("pthread_setcancelstate: %s", strerror(ret));
 213         }
 214         if (oldstate != PTHREAD_CANCEL_DISABLE) {
 215                 ERR("pthread_setcancelstate: unexpected oldstate");
 216         }
 217 }
 218
 219 /*
 220  * Wait for either of these before continuing to the main
 221  * program:
 222  * - the register_done message from sessiond daemon
 223  *   (will let the sessiond daemon enable sessions before main
 224  *   starts.)
 225  * - sessiond daemon is not reachable.
 226  * - timeout (ensuring applications are resilient to session
 227  *   daemon problems).
 228  */
 229 static sem_t constructor_wait;
 230 /*
 231  * Doing this for both the global and local sessiond.
 232  */
 233 enum {
 234         sem_count_initial_value = 4,
 235 };
 236
 237 static int sem_count = sem_count_initial_value;
 238
 239 /*
 240  * Counting nesting within lttng-ust. Used to ensure that calling fork()
 241  * from liblttng-ust does not execute the pre/post fork handlers.
 242  */
 243 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
 244
 245 /*
 246  * Info about socket and associated listener thread.
 247  */
 248 struct sock_info {
 249         const char *name;
 250         pthread_t ust_listener; /* listener thread */
 251         int root_handle;
 252         int registration_done;
 253         int allowed;
 254         int global;
 255         int thread_active;
 256
 257         char sock_path[PATH_MAX];
 258         int socket;
 259         int notify_socket;
 260
 261         char wait_shm_path[PATH_MAX];
 262         char *wait_shm_mmap;
 263         /* Keep track of lazy state dump not performed yet. */
 264         int statedump_pending;
 265         int initial_statedump_done;
 266 };
 267
 268 /* Socket from app (connect) to session daemon (listen) for communication */
 269 struct sock_info global_apps = {
 270         .name = "global",
 271         .global = 1,
 272
 273         .root_handle = -1,
 274         .registration_done = 0,
 275         .allowed = 0,
 276         .thread_active = 0,
 277
 278         .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
 279         .socket = -1,
 280         .notify_socket = -1,
 281
 282         .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
 283
 284         .statedump_pending = 0,
 285         .initial_statedump_done = 0,
 286 };
 287
 288 /* TODO: allow global_apps_sock_path override */
 289
 290 struct sock_info local_apps = {
 291         .name = "local",
 292         .global = 0,
 293         .root_handle = -1,
 294         .registration_done = 0,
 295         .allowed = 0,   /* Check setuid bit first */
 296         .thread_active = 0,
 297
 298         .socket = -1,
 299         .notify_socket = -1,
 300
 301         .statedump_pending = 0,
 302         .initial_statedump_done = 0,
 303 };
 304
 305 static int wait_poll_fallback;
 306
 307 static const char *cmd_name_mapping[] = {
 308         [ LTTNG_UST_RELEASE ] = "Release",
 309         [ LTTNG_UST_SESSION ] = "Create Session",
 310         [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
 311
 312         [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
 313         [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
 314         [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
 315         [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
 316
 317         /* Session FD commands */
 318         [ LTTNG_UST_CHANNEL ] = "Create Channel",
 319         [ LTTNG_UST_SESSION_START ] = "Start Session",
 320         [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
 321
 322         /* Channel FD commands */
 323         [ LTTNG_UST_STREAM ] = "Create Stream",
 324         [ LTTNG_UST_EVENT ] = "Create Event",
 325
 326         /* Event and Channel FD commands */
 327         [ LTTNG_UST_CONTEXT ] = "Create Context",
 328         [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
 329
 330         /* Event, Channel and Session commands */
 331         [ LTTNG_UST_ENABLE ] = "Enable",
 332         [ LTTNG_UST_DISABLE ] = "Disable",
 333
 334         /* Tracepoint list commands */
 335         [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
 336         [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
 337
 338         /* Event FD commands */
 339         [ LTTNG_UST_FILTER ] = "Create Filter",
 340         [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
 341 };
 342
 343 static const char *str_timeout;
 344 static int got_timeout_env;
 345
 346 extern void lttng_ring_buffer_client_overwrite_init(void);
 347 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
 348 extern void lttng_ring_buffer_client_discard_init(void);
 349 extern void lttng_ring_buffer_client_discard_rt_init(void);
 350 extern void lttng_ring_buffer_metadata_client_init(void);
 351 extern void lttng_ring_buffer_client_overwrite_exit(void);
 352 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
 353 extern void lttng_ring_buffer_client_discard_exit(void);
 354 extern void lttng_ring_buffer_client_discard_rt_exit(void);
 355 extern void lttng_ring_buffer_metadata_client_exit(void);
 356
 357 static char *get_map_shm(struct sock_info *sock_info);
 358
 359 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
 360 {
 361         ssize_t ret;
 362         size_t copied = 0, to_copy = len;
 363
 364         do {
 365                 ret = read(fd, buf + copied, to_copy);
 366                 if (ret > 0) {
 367                         copied += ret;
 368                         to_copy -= ret;
 369                 }
 370         } while ((ret > 0 && to_copy > 0)
 371                 || (ret < 0 && errno == EINTR));
 372         if (ret > 0) {
 373                 ret = copied;
 374         }
 375         return ret;
 376 }
 377 /*
 378  * Returns the HOME directory path. Caller MUST NOT free(3) the returned
 379  * pointer.
 380  */
 381 static
 382 const char *get_lttng_home_dir(void)
 383 {
 384        const char *val;
 385
 386        val = (const char *) lttng_getenv("LTTNG_HOME");
 387        if (val != NULL) {
 388                return val;
 389        }
 390        return (const char *) lttng_getenv("HOME");
 391 }
 392
 393 /*
 394  * Force a read (imply TLS fixup for dlopen) of TLS variables.
 395  */
 396 static
 397 void lttng_fixup_nest_count_tls(void)
 398 {
 399         asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
 400 }
 401
 402 static
 403 void lttng_fixup_ust_mutex_nest_tls(void)
 404 {
 405         asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
 406 }
 407
 408 /*
 409  * Fixup urcu bp TLS.
 410  */
 411 static
 412 void lttng_fixup_urcu_bp_tls(void)
 413 {
 414         rcu_read_lock();
 415         rcu_read_unlock();
 416 }
 417
 418 void lttng_ust_fixup_tls(void)
 419 {
 420         lttng_fixup_urcu_bp_tls();
 421         lttng_fixup_ringbuffer_tls();
 422         lttng_fixup_vtid_tls();
 423         lttng_fixup_nest_count_tls();
 424         lttng_fixup_procname_tls();
 425         lttng_fixup_ust_mutex_nest_tls();
 426         lttng_ust_fixup_perf_counter_tls();
 427         lttng_ust_fixup_fd_tracker_tls();
 428 }
 429
 430 int lttng_get_notify_socket(void *owner)
 431 {
 432         struct sock_info *info = owner;
 433
 434         return info->notify_socket;
 435 }
 436
 437 static
 438 void print_cmd(int cmd, int handle)
 439 {
 440         const char *cmd_name = "Unknown";
 441
 442         if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
 443                         && cmd_name_mapping[cmd]) {
 444                 cmd_name = cmd_name_mapping[cmd];
 445         }
 446         DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
 447                 cmd_name, cmd,
 448                 lttng_ust_obj_get_name(handle), handle);
 449 }
 450
 451 static
 452 int setup_global_apps(void)
 453 {
 454         int ret = 0;
 455         assert(!global_apps.wait_shm_mmap);
 456
 457         global_apps.wait_shm_mmap = get_map_shm(&global_apps);
 458         if (!global_apps.wait_shm_mmap) {
 459                 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
 460                 global_apps.allowed = 0;
 461                 ret = -EIO;
 462                 goto error;
 463         }
 464
 465         global_apps.allowed = 1;
 466 error:
 467         return ret;
 468 }
 469 static
 470 int setup_local_apps(void)
 471 {
 472         int ret = 0;
 473         const char *home_dir;
 474         uid_t uid;
 475
 476         assert(!local_apps.wait_shm_mmap);
 477
 478         uid = getuid();
 479         /*
 480          * Disallow per-user tracing for setuid binaries.
 481          */
 482         if (uid != geteuid()) {
 483                 assert(local_apps.allowed == 0);
 484                 ret = 0;
 485                 goto end;
 486         }
 487         home_dir = get_lttng_home_dir();
 488         if (!home_dir) {
 489                 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
 490                 assert(local_apps.allowed == 0);
 491                 ret = -ENOENT;
 492                 goto end;
 493         }
 494         local_apps.allowed = 1;
 495         snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
 496                 home_dir,
 497                 LTTNG_DEFAULT_HOME_RUNDIR,
 498                 LTTNG_UST_SOCK_FILENAME);
 499         snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
 500                 LTTNG_UST_WAIT_FILENAME,
 501                 uid);
 502
 503         local_apps.wait_shm_mmap = get_map_shm(&local_apps);
 504         if (!local_apps.wait_shm_mmap) {
 505                 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
 506                 local_apps.allowed = 0;
 507                 ret = -EIO;
 508                 goto end;
 509         }
 510 end:
 511         return ret;
 512 }
 513
 514 /*
 515  * Get socket timeout, in ms.
 516  * -1: wait forever. 0: don't wait. >0: timeout, in ms.
 517  */
 518 static
 519 long get_timeout(void)
 520 {
 521         long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 522
 523         if (!got_timeout_env) {
 524                 str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
 525                 got_timeout_env = 1;
 526         }
 527         if (str_timeout)
 528                 constructor_delay_ms = strtol(str_timeout, NULL, 10);
 529         /* All negative values are considered as "-1". */
 530         if (constructor_delay_ms < -1)
 531                 constructor_delay_ms = -1;
 532         return constructor_delay_ms;
 533 }
 534
 535 /* Timeout for notify socket send and recv. */
 536 static
 537 long get_notify_sock_timeout(void)
 538 {
 539         return get_timeout();
 540 }
 541
 542 /* Timeout for connecting to cmd and notify sockets. */
 543 static
 544 long get_connect_sock_timeout(void)
 545 {
 546         return get_timeout();
 547 }
 548
 549 /*
 550  * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
 551  */
 552 static
 553 int get_constructor_timeout(struct timespec *constructor_timeout)
 554 {
 555         long constructor_delay_ms;
 556         int ret;
 557
 558         constructor_delay_ms = get_timeout();
 559
 560         switch (constructor_delay_ms) {
 561         case -1:/* fall-through */
 562         case 0:
 563                 return constructor_delay_ms;
 564         default:
 565                 break;
 566         }
 567
 568         /*
 569          * If we are unable to find the current time, don't wait.
 570          */
 571         ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
 572         if (ret) {
 573                 /* Don't wait. */
 574                 return 0;
 575         }
 576         constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
 577         constructor_timeout->tv_nsec +=
 578                 (constructor_delay_ms % 1000UL) * 1000000UL;
 579         if (constructor_timeout->tv_nsec >= 1000000000UL) {
 580                 constructor_timeout->tv_sec++;
 581                 constructor_timeout->tv_nsec -= 1000000000UL;
 582         }
 583         /* Timeout wait (constructor_delay_ms). */
 584         return 1;
 585 }
 586
 587 static
 588 void get_allow_blocking(void)
 589 {
 590         const char *str_allow_blocking =
 591                 lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
 592
 593         if (str_allow_blocking) {
 594                 DBG("%s environment variable is set",
 595                         "LTTNG_UST_ALLOW_BLOCKING");
 596                 lttng_ust_ringbuffer_set_allow_blocking();
 597         }
 598 }
 599
 600 static
 601 int register_to_sessiond(int socket, enum ustctl_socket_type type)
 602 {
 603         return ustcomm_send_reg_msg(socket,
 604                 type,
 605                 CAA_BITS_PER_LONG,
 606                 lttng_alignof(uint8_t) * CHAR_BIT,
 607                 lttng_alignof(uint16_t) * CHAR_BIT,
 608                 lttng_alignof(uint32_t) * CHAR_BIT,
 609                 lttng_alignof(uint64_t) * CHAR_BIT,
 610                 lttng_alignof(unsigned long) * CHAR_BIT);
 611 }
 612
 613 static
 614 int send_reply(int sock, struct ustcomm_ust_reply *lur)
 615 {
 616         ssize_t len;
 617
 618         len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
 619         switch (len) {
 620         case sizeof(*lur):
 621                 DBG("message successfully sent");
 622                 return 0;
 623         default:
 624                 if (len == -ECONNRESET) {
 625                         DBG("remote end closed connection");
 626                         return 0;
 627                 }
 628                 if (len < 0)
 629                         return len;
 630                 DBG("incorrect message size: %zd", len);
 631                 return -EINVAL;
 632         }
 633 }
 634
 635 static
 636 void decrement_sem_count(unsigned int count)
 637 {
 638         int ret;
 639
 640         assert(uatomic_read(&sem_count) >= count);
 641
 642         if (uatomic_read(&sem_count) <= 0) {
 643                 return;
 644         }
 645
 646         ret = uatomic_add_return(&sem_count, -count);
 647         if (ret == 0) {
 648                 ret = sem_post(&constructor_wait);
 649                 assert(!ret);
 650         }
 651 }
 652
 653 static
 654 int handle_register_done(struct sock_info *sock_info)
 655 {
 656         if (sock_info->registration_done)
 657                 return 0;
 658         sock_info->registration_done = 1;
 659
 660         decrement_sem_count(1);
 661         if (!sock_info->statedump_pending) {
 662                 sock_info->initial_statedump_done = 1;
 663                 decrement_sem_count(1);
 664         }
 665
 666         return 0;
 667 }
 668
 669 static
 670 int handle_register_failed(struct sock_info *sock_info)
 671 {
 672         if (sock_info->registration_done)
 673                 return 0;
 674         sock_info->registration_done = 1;
 675         sock_info->initial_statedump_done = 1;
 676
 677         decrement_sem_count(2);
 678
 679         return 0;
 680 }
 681
 682 /*
 683  * Only execute pending statedump after the constructor semaphore has
 684  * been posted by the current listener thread. This means statedump will
 685  * only be performed after the "registration done" command is received
 686  * from this thread's session daemon.
 687  *
 688  * This ensures we don't run into deadlock issues with the dynamic
 689  * loader mutex, which is held while the constructor is called and
 690  * waiting on the constructor semaphore. All operations requiring this
 691  * dynamic loader lock need to be postponed using this mechanism.
 692  *
 693  * In a scenario with two session daemons connected to the application,
 694  * it is possible that the first listener thread which receives the
 695  * registration done command issues its statedump while the dynamic
 696  * loader lock is still held by the application constructor waiting on
 697  * the semaphore. It will however be allowed to proceed when the
 698  * second session daemon sends the registration done command to the
 699  * second listener thread. This situation therefore does not produce
 700  * a deadlock.
 701  */
 702 static
 703 void handle_pending_statedump(struct sock_info *sock_info)
 704 {
 705         if (sock_info->registration_done && sock_info->statedump_pending) {
 706                 sock_info->statedump_pending = 0;
 707                 pthread_mutex_lock(&ust_fork_mutex);
 708                 lttng_handle_pending_statedump(sock_info);
 709                 pthread_mutex_unlock(&ust_fork_mutex);
 710
 711                 if (!sock_info->initial_statedump_done) {
 712                         sock_info->initial_statedump_done = 1;
 713                         decrement_sem_count(1);
 714                 }
 715         }
 716 }
 717
 718 static
 719 int handle_message(struct sock_info *sock_info,
 720                 int sock, struct ustcomm_ust_msg *lum)
 721 {
 722         int ret = 0;
 723         const struct lttng_ust_objd_ops *ops;
 724         struct ustcomm_ust_reply lur;
 725         union ust_args args;
 726         char ctxstr[LTTNG_UST_SYM_NAME_LEN];    /* App context string. */
 727         ssize_t len;
 728
 729         memset(&lur, 0, sizeof(lur));
 730
 731         if (ust_lock()) {
 732                 ret = -LTTNG_UST_ERR_EXITING;
 733                 goto error;
 734         }
 735
 736         ops = objd_ops(lum->handle);
 737         if (!ops) {
 738                 ret = -ENOENT;
 739                 goto error;
 740         }
 741
 742         switch (lum->cmd) {
 743         case LTTNG_UST_REGISTER_DONE:
 744                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 745                         ret = handle_register_done(sock_info);
 746                 else
 747                         ret = -EINVAL;
 748                 break;
 749         case LTTNG_UST_RELEASE:
 750                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 751                         ret = -EPERM;
 752                 else
 753                         ret = lttng_ust_objd_unref(lum->handle, 1);
 754                 break;
 755         case LTTNG_UST_FILTER:
 756         {
 757                 /* Receive filter data */
 758                 struct lttng_ust_filter_bytecode_node *bytecode;
 759
 760                 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
 761                         ERR("Filter data size is too large: %u bytes",
 762                                 lum->u.filter.data_size);
 763                         ret = -EINVAL;
 764                         goto error;
 765                 }
 766
 767                 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
 768                         ERR("Filter reloc offset %u is not within data",
 769                                 lum->u.filter.reloc_offset);
 770                         ret = -EINVAL;
 771                         goto error;
 772                 }
 773
 774                 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
 775                 if (!bytecode) {
 776                         ret = -ENOMEM;
 777                         goto error;
 778                 }
 779                 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
 780                                 lum->u.filter.data_size);
 781                 switch (len) {
 782                 case 0: /* orderly shutdown */
 783                         ret = 0;
 784                         free(bytecode);
 785                         goto error;
 786                 default:
 787                         if (len == lum->u.filter.data_size) {
 788                                 DBG("filter data received");
 789                                 break;
 790                         } else if (len < 0) {
 791                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 792                                 if (len == -ECONNRESET) {
 793                                         ERR("%s remote end closed connection", sock_info->name);
 794                                         ret = len;
 795                                         free(bytecode);
 796                                         goto error;
 797                                 }
 798                                 ret = len;
 799                                 free(bytecode);
 800                                 goto error;
 801                         } else {
 802                                 DBG("incorrect filter data message size: %zd", len);
 803                                 ret = -EINVAL;
 804                                 free(bytecode);
 805                                 goto error;
 806                         }
 807                 }
 808                 bytecode->bc.len = lum->u.filter.data_size;
 809                 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
 810                 bytecode->bc.seqnum = lum->u.filter.seqnum;
 811                 if (ops->cmd) {
 812                         ret = ops->cmd(lum->handle, lum->cmd,
 813                                         (unsigned long) bytecode,
 814                                         &args, sock_info);
 815                         if (ret) {
 816                                 free(bytecode);
 817                         }
 818                         /* don't free bytecode if everything went fine. */
 819                 } else {
 820                         ret = -ENOSYS;
 821                         free(bytecode);
 822                 }
 823                 break;
 824         }
 825         case LTTNG_UST_EXCLUSION:
 826         {
 827                 /* Receive exclusion names */
 828                 struct lttng_ust_excluder_node *node;
 829                 unsigned int count;
 830
 831                 count = lum->u.exclusion.count;
 832                 if (count == 0) {
 833                         /* There are no names to read */
 834                         ret = 0;
 835                         goto error;
 836                 }
 837                 node = zmalloc(sizeof(*node) +
 838                                 count * LTTNG_UST_SYM_NAME_LEN);
 839                 if (!node) {
 840                         ret = -ENOMEM;
 841                         goto error;
 842                 }
 843                 node->excluder.count = count;
 844                 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
 845                                 count * LTTNG_UST_SYM_NAME_LEN);
 846                 switch (len) {
 847                 case 0: /* orderly shutdown */
 848                         ret = 0;
 849                         free(node);
 850                         goto error;
 851                 default:
 852                         if (len == count * LTTNG_UST_SYM_NAME_LEN) {
 853                                 DBG("Exclusion data received");
 854                                 break;
 855                         } else if (len < 0) {
 856                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 857                                 if (len == -ECONNRESET) {
 858                                         ERR("%s remote end closed connection", sock_info->name);
 859                                         ret = len;
 860                                         free(node);
 861                                         goto error;
 862                                 }
 863                                 ret = len;
 864                                 free(node);
 865                                 goto error;
 866                         } else {
 867                                 DBG("Incorrect exclusion data message size: %zd", len);
 868                                 ret = -EINVAL;
 869                                 free(node);
 870                                 goto error;
 871                         }
 872                 }
 873                 if (ops->cmd) {
 874                         ret = ops->cmd(lum->handle, lum->cmd,
 875                                         (unsigned long) node,
 876                                         &args, sock_info);
 877                         if (ret) {
 878                                 free(node);
 879                         }
 880                         /* Don't free exclusion data if everything went fine. */
 881                 } else {
 882                         ret = -ENOSYS;
 883                         free(node);
 884                 }
 885                 break;
 886         }
 887         case LTTNG_UST_CHANNEL:
 888         {
 889                 void *chan_data;
 890                 int wakeup_fd;
 891
 892                 len = ustcomm_recv_channel_from_sessiond(sock,
 893                                 &chan_data, lum->u.channel.len,
 894                                 &wakeup_fd);
 895                 switch (len) {
 896                 case 0: /* orderly shutdown */
 897                         ret = 0;
 898                         goto error;
 899                 default:
 900                         if (len == lum->u.channel.len) {
 901                                 DBG("channel data received");
 902                                 break;
 903                         } else if (len < 0) {
 904                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 905                                 if (len == -ECONNRESET) {
 906                                         ERR("%s remote end closed connection", sock_info->name);
 907                                         ret = len;
 908                                         goto error;
 909                                 }
 910                                 ret = len;
 911                                 goto error;
 912                         } else {
 913                                 DBG("incorrect channel data message size: %zd", len);
 914                                 ret = -EINVAL;
 915                                 goto error;
 916                         }
 917                 }
 918                 args.channel.chan_data = chan_data;
 919                 args.channel.wakeup_fd = wakeup_fd;
 920                 if (ops->cmd)
 921                         ret = ops->cmd(lum->handle, lum->cmd,
 922                                         (unsigned long) &lum->u,
 923                                         &args, sock_info);
 924                 else
 925                         ret = -ENOSYS;
 926                 if (args.channel.wakeup_fd >= 0) {
 927                         int close_ret;
 928
 929                         lttng_ust_lock_fd_tracker();
 930                         close_ret = close(args.channel.wakeup_fd);
 931                         lttng_ust_unlock_fd_tracker();
 932                         args.channel.wakeup_fd = -1;
 933                         if (close_ret)
 934                                 PERROR("close");
 935                 }
 936                 free(args.channel.chan_data);
 937                 break;
 938         }
 939         case LTTNG_UST_STREAM:
 940         {
 941                 int close_ret;
 942
 943                 /* Receive shm_fd, wakeup_fd */
 944                 ret = ustcomm_recv_stream_from_sessiond(sock,
 945                         NULL,
 946                         &args.stream.shm_fd,
 947                         &args.stream.wakeup_fd);
 948                 if (ret) {
 949                         goto error;
 950                 }
 951
 952                 if (ops->cmd)
 953                         ret = ops->cmd(lum->handle, lum->cmd,
 954                                         (unsigned long) &lum->u,
 955                                         &args, sock_info);
 956                 else
 957                         ret = -ENOSYS;
 958                 if (args.stream.shm_fd >= 0) {
 959                         lttng_ust_lock_fd_tracker();
 960                         close_ret = close(args.stream.shm_fd);
 961                         lttng_ust_unlock_fd_tracker();
 962                         args.stream.shm_fd = -1;
 963                         if (close_ret)
 964                                 PERROR("close");
 965                 }
 966                 if (args.stream.wakeup_fd >= 0) {
 967                         lttng_ust_lock_fd_tracker();
 968                         close_ret = close(args.stream.wakeup_fd);
 969                         lttng_ust_unlock_fd_tracker();
 970                         args.stream.wakeup_fd = -1;
 971                         if (close_ret)
 972                                 PERROR("close");
 973                 }
 974                 break;
 975         }
 976         case LTTNG_UST_CONTEXT:
 977                 switch (lum->u.context.ctx) {
 978                 case LTTNG_UST_CONTEXT_APP_CONTEXT:
 979                 {
 980                         char *p;
 981                         size_t ctxlen, recvlen;
 982
 983                         ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
 984                                         + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
 985                         if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
 986                                 ERR("Application context string length size is too large: %zu bytes",
 987                                         ctxlen);
 988                                 ret = -EINVAL;
 989                                 goto error;
 990                         }
 991                         strcpy(ctxstr, "$app.");
 992                         p = &ctxstr[strlen("$app.")];
 993                         recvlen = ctxlen - strlen("$app.");
 994                         len = ustcomm_recv_unix_sock(sock, p, recvlen);
 995                         switch (len) {
 996                         case 0: /* orderly shutdown */
 997                                 ret = 0;
 998                                 goto error;
 999                         default:
1000                                 if (len == recvlen) {
1001                                         DBG("app context data received");
1002                                         break;
1003                                 } else if (len < 0) {
1004                                         DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1005                                         if (len == -ECONNRESET) {
1006                                                 ERR("%s remote end closed connection", sock_info->name);
1007                                                 ret = len;
1008                                                 goto error;
1009                                         }
1010                                         ret = len;
1011                                         goto error;
1012                                 } else {
1013                                         DBG("incorrect app context data message size: %zd", len);
1014                                         ret = -EINVAL;
1015                                         goto error;
1016                                 }
1017                         }
1018                         /* Put : between provider and ctxname. */
1019                         p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1020                         args.app_context.ctxname = ctxstr;
1021                         break;
1022                 }
1023                 default:
1024                         break;
1025                 }
1026                 if (ops->cmd) {
1027                         ret = ops->cmd(lum->handle, lum->cmd,
1028                                         (unsigned long) &lum->u,
1029                                         &args, sock_info);
1030                 } else {
1031                         ret = -ENOSYS;
1032                 }
1033                 break;
1034         default:
1035                 if (ops->cmd)
1036                         ret = ops->cmd(lum->handle, lum->cmd,
1037                                         (unsigned long) &lum->u,
1038                                         &args, sock_info);
1039                 else
1040                         ret = -ENOSYS;
1041                 break;
1042         }
1043
1044         lur.handle = lum->handle;
1045         lur.cmd = lum->cmd;
1046         lur.ret_val = ret;
1047         if (ret >= 0) {
1048                 lur.ret_code = LTTNG_UST_OK;
1049         } else {
1050                 /*
1051                  * Use -LTTNG_UST_ERR as wildcard for UST internal
1052                  * error that are not caused by the transport, except if
1053                  * we already have a more precise error message to
1054                  * report.
1055                  */
1056                 if (ret > -LTTNG_UST_ERR) {
1057                         /* Translate code to UST error. */
1058                         switch (ret) {
1059                         case -EEXIST:
1060                                 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1061                                 break;
1062                         case -EINVAL:
1063                                 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1064                                 break;
1065                         case -ENOENT:
1066                                 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1067                                 break;
1068                         case -EPERM:
1069                                 lur.ret_code = -LTTNG_UST_ERR_PERM;
1070                                 break;
1071                         case -ENOSYS:
1072                                 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1073                                 break;
1074                         default:
1075                                 lur.ret_code = -LTTNG_UST_ERR;
1076                                 break;
1077                         }
1078                 } else {
1079                         lur.ret_code = ret;
1080                 }
1081         }
1082         if (ret >= 0) {
1083                 switch (lum->cmd) {
1084                 case LTTNG_UST_TRACER_VERSION:
1085                         lur.u.version = lum->u.version;
1086                         break;
1087                 case LTTNG_UST_TRACEPOINT_LIST_GET:
1088                         memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1089                         break;
1090                 }
1091         }
1092         DBG("Return value: %d", lur.ret_val);
1093
1094         ust_unlock();
1095
1096         /*
1097          * Performed delayed statedump operations outside of the UST
1098          * lock. We need to take the dynamic loader lock before we take
1099          * the UST lock internally within handle_pending_statedump().
1100           */
1101         handle_pending_statedump(sock_info);
1102
1103         if (ust_lock()) {
1104                 ret = -LTTNG_UST_ERR_EXITING;
1105                 goto error;
1106         }
1107
1108         ret = send_reply(sock, &lur);
1109         if (ret < 0) {
1110                 DBG("error sending reply");
1111                 goto error;
1112         }
1113
1114         /*
1115          * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1116          * after the reply.
1117          */
1118         if (lur.ret_code == LTTNG_UST_OK) {
1119                 switch (lum->cmd) {
1120                 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1121                         len = ustcomm_send_unix_sock(sock,
1122                                 &args.field_list.entry,
1123                                 sizeof(args.field_list.entry));
1124                         if (len < 0) {
1125                                 ret = len;
1126                                 goto error;
1127                         }
1128                         if (len != sizeof(args.field_list.entry)) {
1129                                 ret = -EINVAL;
1130                                 goto error;
1131                         }
1132                 }
1133         }
1134
1135 error:
1136         ust_unlock();
1137
1138         return ret;
1139 }
1140
1141 static
1142 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1143 {
1144         int ret;
1145
1146         if (sock_info->root_handle != -1) {
1147                 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1148                 if (ret) {
1149                         ERR("Error unref root handle");
1150                 }
1151                 sock_info->root_handle = -1;
1152         }
1153         sock_info->registration_done = 0;
1154         sock_info->initial_statedump_done = 0;
1155
1156         /*
1157          * wait_shm_mmap, socket and notify socket are used by listener
1158          * threads outside of the ust lock, so we cannot tear them down
1159          * ourselves, because we cannot join on these threads. Leave
1160          * responsibility of cleaning up these resources to the OS
1161          * process exit.
1162          */
1163         if (exiting)
1164                 return;
1165
1166         if (sock_info->socket != -1) {
1167                 ret = ustcomm_close_unix_sock(sock_info->socket);
1168                 if (ret) {
1169                         ERR("Error closing ust cmd socket");
1170                 }
1171                 sock_info->socket = -1;
1172         }
1173         if (sock_info->notify_socket != -1) {
1174                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1175                 if (ret) {
1176                         ERR("Error closing ust notify socket");
1177                 }
1178                 sock_info->notify_socket = -1;
1179         }
1180         if (sock_info->wait_shm_mmap) {
1181                 long page_size;
1182
1183                 page_size = sysconf(_SC_PAGE_SIZE);
1184                 if (page_size <= 0) {
1185                         if (!page_size) {
1186                                 errno = EINVAL;
1187                         }
1188                         PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1189                 } else {
1190                         ret = munmap(sock_info->wait_shm_mmap, page_size);
1191                         if (ret) {
1192                                 ERR("Error unmapping wait shm");
1193                         }
1194                 }
1195                 sock_info->wait_shm_mmap = NULL;
1196         }
1197 }
1198
1199 /*
1200  * Using fork to set umask in the child process (not multi-thread safe).
1201  * We deal with the shm_open vs ftruncate race (happening when the
1202  * sessiond owns the shm and does not let everybody modify it, to ensure
1203  * safety against shm_unlink) by simply letting the mmap fail and
1204  * retrying after a few seconds.
1205  * For global shm, everybody has rw access to it until the sessiond
1206  * starts.
1207  */
1208 static
1209 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1210 {
1211         int wait_shm_fd, ret;
1212         pid_t pid;
1213
1214         /*
1215          * Try to open read-only.
1216          */
1217         wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1218         if (wait_shm_fd >= 0) {
1219                 int32_t tmp_read;
1220                 ssize_t len;
1221                 size_t bytes_read = 0;
1222
1223                 /*
1224                  * Try to read the fd. If unable to do so, try opening
1225                  * it in write mode.
1226                  */
1227                 do {
1228                         len = read(wait_shm_fd,
1229                                 &((char *) &tmp_read)[bytes_read],
1230                                 sizeof(tmp_read) - bytes_read);
1231                         if (len > 0) {
1232                                 bytes_read += len;
1233                         }
1234                 } while ((len < 0 && errno == EINTR)
1235                         || (len > 0 && bytes_read < sizeof(tmp_read)));
1236                 if (bytes_read != sizeof(tmp_read)) {
1237                         ret = close(wait_shm_fd);
1238                         if (ret) {
1239                                 ERR("close wait_shm_fd");
1240                         }
1241                         goto open_write;
1242                 }
1243                 goto end;
1244         } else if (wait_shm_fd < 0 && errno != ENOENT) {
1245                 /*
1246                  * Real-only open did not work, and it's not because the
1247                  * entry was not present. It's a failure that prohibits
1248                  * using shm.
1249                  */
1250                 ERR("Error opening shm %s", sock_info->wait_shm_path);
1251                 goto end;
1252         }
1253
1254 open_write:
1255         /*
1256          * If the open failed because the file did not exist, or because
1257          * the file was not truncated yet, try creating it ourself.
1258          */
1259         URCU_TLS(lttng_ust_nest_count)++;
1260         pid = fork();
1261         URCU_TLS(lttng_ust_nest_count)--;
1262         if (pid > 0) {
1263                 int status;
1264
1265                 /*
1266                  * Parent: wait for child to return, in which case the
1267                  * shared memory map will have been created.
1268                  */
1269                 pid = wait(&status);
1270                 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1271                         wait_shm_fd = -1;
1272                         goto end;
1273                 }
1274                 /*
1275                  * Try to open read-only again after creation.
1276                  */
1277                 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1278                 if (wait_shm_fd < 0) {
1279                         /*
1280                          * Real-only open did not work. It's a failure
1281                          * that prohibits using shm.
1282                          */
1283                         ERR("Error opening shm %s", sock_info->wait_shm_path);
1284                         goto end;
1285                 }
1286                 goto end;
1287         } else if (pid == 0) {
1288                 int create_mode;
1289
1290                 /* Child */
1291                 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1292                 if (sock_info->global)
1293                         create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1294                 /*
1295                  * We're alone in a child process, so we can modify the
1296                  * process-wide umask.
1297                  */
1298                 umask(~create_mode);
1299                 /*
1300                  * Try creating shm (or get rw access).
1301                  * We don't do an exclusive open, because we allow other
1302                  * processes to create+ftruncate it concurrently.
1303                  */
1304                 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1305                                 O_RDWR | O_CREAT, create_mode);
1306                 if (wait_shm_fd >= 0) {
1307                         ret = ftruncate(wait_shm_fd, mmap_size);
1308                         if (ret) {
1309                                 PERROR("ftruncate");
1310                                 _exit(EXIT_FAILURE);
1311                         }
1312                         _exit(EXIT_SUCCESS);
1313                 }
1314                 /*
1315                  * For local shm, we need to have rw access to accept
1316                  * opening it: this means the local sessiond will be
1317                  * able to wake us up. For global shm, we open it even
1318                  * if rw access is not granted, because the root.root
1319                  * sessiond will be able to override all rights and wake
1320                  * us up.
1321                  */
1322                 if (!sock_info->global && errno != EACCES) {
1323                         ERR("Error opening shm %s", sock_info->wait_shm_path);
1324                         _exit(EXIT_FAILURE);
1325                 }
1326                 /*
1327                  * The shm exists, but we cannot open it RW. Report
1328                  * success.
1329                  */
1330                 _exit(EXIT_SUCCESS);
1331         } else {
1332                 return -1;
1333         }
1334 end:
1335         if (wait_shm_fd >= 0 && !sock_info->global) {
1336                 struct stat statbuf;
1337
1338                 /*
1339                  * Ensure that our user is the owner of the shm file for
1340                  * local shm. If we do not own the file, it means our
1341                  * sessiond will not have access to wake us up (there is
1342                  * probably a rogue process trying to fake our
1343                  * sessiond). Fallback to polling method in this case.
1344                  */
1345                 ret = fstat(wait_shm_fd, &statbuf);
1346                 if (ret) {
1347                         PERROR("fstat");
1348                         goto error_close;
1349                 }
1350                 if (statbuf.st_uid != getuid())
1351                         goto error_close;
1352         }
1353         return wait_shm_fd;
1354
1355 error_close:
1356         ret = close(wait_shm_fd);
1357         if (ret) {
1358                 PERROR("Error closing fd");
1359         }
1360         return -1;
1361 }
1362
1363 static
1364 char *get_map_shm(struct sock_info *sock_info)
1365 {
1366         long page_size;
1367         int wait_shm_fd, ret;
1368         char *wait_shm_mmap;
1369
1370         page_size = sysconf(_SC_PAGE_SIZE);
1371         if (page_size <= 0) {
1372                 if (!page_size) {
1373                         errno = EINVAL;
1374                 }
1375                 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1376                 goto error;
1377         }
1378
1379         lttng_ust_lock_fd_tracker();
1380         wait_shm_fd = get_wait_shm(sock_info, page_size);
1381         if (wait_shm_fd < 0) {
1382                 lttng_ust_unlock_fd_tracker();
1383                 goto error;
1384         }
1385
1386         ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1387         if (ret < 0) {
1388                 ret = close(wait_shm_fd);
1389                 if (!ret) {
1390                         PERROR("Error closing fd");
1391                 }
1392                 lttng_ust_unlock_fd_tracker();
1393                 goto error;
1394         }
1395
1396         wait_shm_fd = ret;
1397         lttng_ust_unlock_fd_tracker();
1398
1399         wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1400                   MAP_SHARED, wait_shm_fd, 0);
1401
1402         /* close shm fd immediately after taking the mmap reference */
1403         lttng_ust_lock_fd_tracker();
1404         ret = close(wait_shm_fd);
1405         if (!ret) {
1406                 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1407         } else {
1408                 PERROR("Error closing fd");
1409         }
1410         lttng_ust_unlock_fd_tracker();
1411
1412         if (wait_shm_mmap == MAP_FAILED) {
1413                 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1414                 goto error;
1415         }
1416         return wait_shm_mmap;
1417
1418 error:
1419         return NULL;
1420 }
1421
1422 static
1423 void wait_for_sessiond(struct sock_info *sock_info)
1424 {
1425         /* Use ust_lock to check if we should quit. */
1426         if (ust_lock()) {
1427                 goto quit;
1428         }
1429         if (wait_poll_fallback) {
1430                 goto error;
1431         }
1432         ust_unlock();
1433
1434         assert(sock_info->wait_shm_mmap);
1435
1436         DBG("Waiting for %s apps sessiond", sock_info->name);
1437         /* Wait for futex wakeup */
1438         if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1439                 goto end_wait;
1440
1441         while (futex_async((int32_t *) sock_info->wait_shm_mmap,
1442                         FUTEX_WAIT, 0, NULL, NULL, 0)) {
1443                 switch (errno) {
1444                 case EWOULDBLOCK:
1445                         /* Value already changed. */
1446                         goto end_wait;
1447                 case EINTR:
1448                         /* Retry if interrupted by signal. */
1449                         break;  /* Get out of switch. */
1450                 case EFAULT:
1451                         wait_poll_fallback = 1;
1452                         DBG(
1453 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1454 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1455 "Please upgrade your kernel "
1456 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1457 "mainline). LTTng-UST will use polling mode fallback.");
1458                         if (ust_debug())
1459                                 PERROR("futex");
1460                         goto end_wait;
1461                 }
1462         }
1463 end_wait:
1464         return;
1465
1466 quit:
1467         ust_unlock();
1468         return;
1469
1470 error:
1471         ust_unlock();
1472         return;
1473 }
1474
1475 /*
1476  * This thread does not allocate any resource, except within
1477  * handle_message, within mutex protection. This mutex protects against
1478  * fork and exit.
1479  * The other moment it allocates resources is at socket connection, which
1480  * is also protected by the mutex.
1481  */
1482 static
1483 void *ust_listener_thread(void *arg)
1484 {
1485         struct sock_info *sock_info = arg;
1486         int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1487         long timeout;
1488
1489         lttng_ust_fixup_tls();
1490         /*
1491          * If available, add '-ust' to the end of this thread's
1492          * process name
1493          */
1494         ret = lttng_ust_setustprocname();
1495         if (ret) {
1496                 ERR("Unable to set UST process name");
1497         }
1498
1499         /* Restart trying to connect to the session daemon */
1500 restart:
1501         if (prev_connect_failed) {
1502                 /* Wait for sessiond availability with pipe */
1503                 wait_for_sessiond(sock_info);
1504                 if (has_waited) {
1505                         has_waited = 0;
1506                         /*
1507                          * Sleep for 5 seconds before retrying after a
1508                          * sequence of failure / wait / failure. This
1509                          * deals with a killed or broken session daemon.
1510                          */
1511                         sleep(5);
1512                 } else {
1513                         has_waited = 1;
1514                 }
1515                 prev_connect_failed = 0;
1516         }
1517
1518         if (ust_lock()) {
1519                 goto quit;
1520         }
1521
1522         if (sock_info->socket != -1) {
1523                 /* FD tracker is updated by ustcomm_close_unix_sock() */
1524                 ret = ustcomm_close_unix_sock(sock_info->socket);
1525                 if (ret) {
1526                         ERR("Error closing %s ust cmd socket",
1527                                 sock_info->name);
1528                 }
1529                 sock_info->socket = -1;
1530         }
1531         if (sock_info->notify_socket != -1) {
1532                 /* FD tracker is updated by ustcomm_close_unix_sock() */
1533                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1534                 if (ret) {
1535                         ERR("Error closing %s ust notify socket",
1536                                 sock_info->name);
1537                 }
1538                 sock_info->notify_socket = -1;
1539         }
1540
1541
1542         /*
1543          * Register. We need to perform both connect and sending
1544          * registration message before doing the next connect otherwise
1545          * we may reach unix socket connect queue max limits and block
1546          * on the 2nd connect while the session daemon is awaiting the
1547          * first connect registration message.
1548          */
1549         /* Connect cmd socket */
1550         lttng_ust_lock_fd_tracker();
1551         ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1552                 get_connect_sock_timeout());
1553         if (ret < 0) {
1554                 lttng_ust_unlock_fd_tracker();
1555                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1556                 prev_connect_failed = 1;
1557
1558                 /*
1559                  * If we cannot find the sessiond daemon, don't delay
1560                  * constructor execution.
1561                  */
1562                 ret = handle_register_failed(sock_info);
1563                 assert(!ret);
1564                 ust_unlock();
1565                 goto restart;
1566         }
1567         fd = ret;
1568         ret = lttng_ust_add_fd_to_tracker(fd);
1569         if (ret < 0) {
1570                 ret = close(fd);
1571                 if (ret) {
1572                         PERROR("close on sock_info->socket");
1573                 }
1574                 ret = -1;
1575                 lttng_ust_unlock_fd_tracker();
1576                 ust_unlock();
1577                 goto quit;
1578         }
1579
1580         sock_info->socket = ret;
1581         lttng_ust_unlock_fd_tracker();
1582
1583         ust_unlock();
1584         /*
1585          * Unlock/relock ust lock because connect is blocking (with
1586          * timeout). Don't delay constructors on the ust lock for too
1587          * long.
1588          */
1589         if (ust_lock()) {
1590                 goto quit;
1591         }
1592
1593         /*
1594          * Create only one root handle per listener thread for the whole
1595          * process lifetime, so we ensure we get ID which is statically
1596          * assigned to the root handle.
1597          */
1598         if (sock_info->root_handle == -1) {
1599                 ret = lttng_abi_create_root_handle();
1600                 if (ret < 0) {
1601                         ERR("Error creating root handle");
1602                         goto quit;
1603                 }
1604                 sock_info->root_handle = ret;
1605         }
1606
1607         ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1608         if (ret < 0) {
1609                 ERR("Error registering to %s ust cmd socket",
1610                         sock_info->name);
1611                 prev_connect_failed = 1;
1612                 /*
1613                  * If we cannot register to the sessiond daemon, don't
1614                  * delay constructor execution.
1615                  */
1616                 ret = handle_register_failed(sock_info);
1617                 assert(!ret);
1618                 ust_unlock();
1619                 goto restart;
1620         }
1621
1622         ust_unlock();
1623         /*
1624          * Unlock/relock ust lock because connect is blocking (with
1625          * timeout). Don't delay constructors on the ust lock for too
1626          * long.
1627          */
1628         if (ust_lock()) {
1629                 goto quit;
1630         }
1631
1632         /* Connect notify socket */
1633         lttng_ust_lock_fd_tracker();
1634         ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1635                 get_connect_sock_timeout());
1636         if (ret < 0) {
1637                 lttng_ust_unlock_fd_tracker();
1638                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1639                 prev_connect_failed = 1;
1640
1641                 /*
1642                  * If we cannot find the sessiond daemon, don't delay
1643                  * constructor execution.
1644                  */
1645                 ret = handle_register_failed(sock_info);
1646                 assert(!ret);
1647                 ust_unlock();
1648                 goto restart;
1649         }
1650
1651         fd = ret;
1652         ret = lttng_ust_add_fd_to_tracker(fd);
1653         if (ret < 0) {
1654                 ret = close(fd);
1655                 if (ret) {
1656                         PERROR("close on sock_info->notify_socket");
1657                 }
1658                 ret = -1;
1659                 lttng_ust_unlock_fd_tracker();
1660                 ust_unlock();
1661                 goto quit;
1662         }
1663
1664         sock_info->notify_socket = ret;
1665         lttng_ust_unlock_fd_tracker();
1666
1667         ust_unlock();
1668         /*
1669          * Unlock/relock ust lock because connect is blocking (with
1670          * timeout). Don't delay constructors on the ust lock for too
1671          * long.
1672          */
1673         if (ust_lock()) {
1674                 goto quit;
1675         }
1676
1677         timeout = get_notify_sock_timeout();
1678         if (timeout >= 0) {
1679                 /*
1680                  * Give at least 10ms to sessiond to reply to
1681                  * notifications.
1682                  */
1683                 if (timeout < 10)
1684                         timeout = 10;
1685                 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1686                                 timeout);
1687                 if (ret < 0) {
1688                         WARN("Error setting socket receive timeout");
1689                 }
1690                 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1691                                 timeout);
1692                 if (ret < 0) {
1693                         WARN("Error setting socket send timeout");
1694                 }
1695         } else if (timeout < -1) {
1696                 WARN("Unsupported timeout value %ld", timeout);
1697         }
1698
1699         ret = register_to_sessiond(sock_info->notify_socket,
1700                         USTCTL_SOCKET_NOTIFY);
1701         if (ret < 0) {
1702                 ERR("Error registering to %s ust notify socket",
1703                         sock_info->name);
1704                 prev_connect_failed = 1;
1705                 /*
1706                  * If we cannot register to the sessiond daemon, don't
1707                  * delay constructor execution.
1708                  */
1709                 ret = handle_register_failed(sock_info);
1710                 assert(!ret);
1711                 ust_unlock();
1712                 goto restart;
1713         }
1714         sock = sock_info->socket;
1715
1716         ust_unlock();
1717
1718         for (;;) {
1719                 ssize_t len;
1720                 struct ustcomm_ust_msg lum;
1721
1722                 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1723                 switch (len) {
1724                 case 0: /* orderly shutdown */
1725                         DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1726                         if (ust_lock()) {
1727                                 goto quit;
1728                         }
1729                         /*
1730                          * Either sessiond has shutdown or refused us by closing the socket.
1731                          * In either case, we don't want to delay construction execution,
1732                          * and we need to wait before retry.
1733                          */
1734                         prev_connect_failed = 1;
1735                         /*
1736                          * If we cannot register to the sessiond daemon, don't
1737                          * delay constructor execution.
1738                          */
1739                         ret = handle_register_failed(sock_info);
1740                         assert(!ret);
1741                         ust_unlock();
1742                         goto end;
1743                 case sizeof(lum):
1744                         print_cmd(lum.cmd, lum.handle);
1745                         ret = handle_message(sock_info, sock, &lum);
1746                         if (ret) {
1747                                 ERR("Error handling message for %s socket",
1748                                         sock_info->name);
1749                                 /*
1750                                  * Close socket if protocol error is
1751                                  * detected.
1752                                  */
1753                                 goto end;
1754                         }
1755                         continue;
1756                 default:
1757                         if (len < 0) {
1758                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1759                         } else {
1760                                 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1761                         }
1762                         if (len == -ECONNRESET) {
1763                                 DBG("%s remote end closed connection", sock_info->name);
1764                                 goto end;
1765                         }
1766                         goto end;
1767                 }
1768
1769         }
1770 end:
1771         if (ust_lock()) {
1772                 goto quit;
1773         }
1774         /* Cleanup socket handles before trying to reconnect */
1775         lttng_ust_objd_table_owner_cleanup(sock_info);
1776         ust_unlock();
1777         goto restart;   /* try to reconnect */
1778
1779 quit:
1780         ust_unlock();
1781
1782         pthread_mutex_lock(&ust_exit_mutex);
1783         sock_info->thread_active = 0;
1784         pthread_mutex_unlock(&ust_exit_mutex);
1785         return NULL;
1786 }
1787
1788 /*
1789  * Weak symbol to call when the ust malloc wrapper is not loaded.
1790  */
1791 __attribute__((weak))
1792 void lttng_ust_malloc_wrapper_init(void)
1793 {
1794 }
1795
1796 /*
1797  * sessiond monitoring thread: monitor presence of global and per-user
1798  * sessiond by polling the application common named pipe.
1799  */
1800 void __attribute__((constructor)) lttng_ust_init(void)
1801 {
1802         struct timespec constructor_timeout;
1803         sigset_t sig_all_blocked, orig_parent_mask;
1804         pthread_attr_t thread_attr;
1805         int timeout_mode;
1806         int ret;
1807         void *handle;
1808
1809         if (uatomic_xchg(&initialized, 1) == 1)
1810                 return;
1811
1812         /*
1813          * Fixup interdependency between TLS fixup mutex (which happens
1814          * to be the dynamic linker mutex) and ust_lock, taken within
1815          * the ust lock.
1816          */
1817         lttng_ust_fixup_tls();
1818
1819         lttng_ust_loaded = 1;
1820
1821         /*
1822          * We need to ensure that the liblttng-ust library is not unloaded to avoid
1823          * the unloading of code used by the ust_listener_threads as we can not
1824          * reliably know when they exited. To do that, manually load
1825          * liblttng-ust.so to increment the dynamic loader's internal refcount for
1826          * this library so it never becomes zero, thus never gets unloaded from the
1827          * address space of the process. Since we are already running in the
1828          * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
1829          * simply increment the refcount and no additionnal work is needed by the
1830          * dynamic loader as the shared library is already loaded in the address
1831          * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
1832          * unloading of the UST library if its refcount becomes zero (which should
1833          * never happen). Do the return value check but discard the handle at the
1834          * end of the function as it's not needed.
1835          */
1836         handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
1837         if (!handle) {
1838                 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
1839         }
1840
1841         /*
1842          * We want precise control over the order in which we construct
1843          * our sub-libraries vs starting to receive commands from
1844          * sessiond (otherwise leading to errors when trying to create
1845          * sessiond before the init functions are completed).
1846          */
1847         init_usterr();
1848         lttng_ust_getenv_init();        /* Needs init_usterr() to be completed. */
1849         init_tracepoint();
1850         lttng_ust_init_fd_tracker();
1851         lttng_ust_clock_init();
1852         lttng_ust_getcpu_init();
1853         lttng_ust_statedump_init();
1854         lttng_ring_buffer_metadata_client_init();
1855         lttng_ring_buffer_client_overwrite_init();
1856         lttng_ring_buffer_client_overwrite_rt_init();
1857         lttng_ring_buffer_client_discard_init();
1858         lttng_ring_buffer_client_discard_rt_init();
1859         lttng_perf_counter_init();
1860         /*
1861          * Invoke ust malloc wrapper init before starting other threads.
1862          */
1863         lttng_ust_malloc_wrapper_init();
1864
1865         timeout_mode = get_constructor_timeout(&constructor_timeout);
1866
1867         get_allow_blocking();
1868
1869         ret = sem_init(&constructor_wait, 0, 0);
1870         if (ret) {
1871                 PERROR("sem_init");
1872         }
1873
1874         ret = setup_global_apps();
1875         if (ret) {
1876                 assert(global_apps.allowed == 0);
1877                 DBG("global apps setup returned %d", ret);
1878         }
1879
1880         ret = setup_local_apps();
1881         if (ret) {
1882                 assert(local_apps.allowed == 0);
1883                 DBG("local apps setup returned %d", ret);
1884         }
1885
1886         /* A new thread created by pthread_create inherits the signal mask
1887          * from the parent. To avoid any signal being received by the
1888          * listener thread, we block all signals temporarily in the parent,
1889          * while we create the listener thread.
1890          */
1891         sigfillset(&sig_all_blocked);
1892         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1893         if (ret) {
1894                 ERR("pthread_sigmask: %s", strerror(ret));
1895         }
1896
1897         ret = pthread_attr_init(&thread_attr);
1898         if (ret) {
1899                 ERR("pthread_attr_init: %s", strerror(ret));
1900         }
1901         ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1902         if (ret) {
1903                 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1904         }
1905
1906         if (global_apps.allowed) {
1907                 pthread_mutex_lock(&ust_exit_mutex);
1908                 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1909                                 ust_listener_thread, &global_apps);
1910                 if (ret) {
1911                         ERR("pthread_create global: %s", strerror(ret));
1912                 }
1913                 global_apps.thread_active = 1;
1914                 pthread_mutex_unlock(&ust_exit_mutex);
1915         } else {
1916                 handle_register_done(&global_apps);
1917         }
1918
1919         if (local_apps.allowed) {
1920                 pthread_mutex_lock(&ust_exit_mutex);
1921                 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1922                                 ust_listener_thread, &local_apps);
1923                 if (ret) {
1924                         ERR("pthread_create local: %s", strerror(ret));
1925                 }
1926                 local_apps.thread_active = 1;
1927                 pthread_mutex_unlock(&ust_exit_mutex);
1928         } else {
1929                 handle_register_done(&local_apps);
1930         }
1931         ret = pthread_attr_destroy(&thread_attr);
1932         if (ret) {
1933                 ERR("pthread_attr_destroy: %s", strerror(ret));
1934         }
1935
1936         /* Restore original signal mask in parent */
1937         ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1938         if (ret) {
1939                 ERR("pthread_sigmask: %s", strerror(ret));
1940         }
1941
1942         switch (timeout_mode) {
1943         case 1: /* timeout wait */
1944                 do {
1945                         ret = sem_timedwait(&constructor_wait,
1946                                         &constructor_timeout);
1947                 } while (ret < 0 && errno == EINTR);
1948                 if (ret < 0) {
1949                         switch (errno) {
1950                         case ETIMEDOUT:
1951                                 ERR("Timed out waiting for lttng-sessiond");
1952                                 break;
1953                         case EINVAL:
1954                                 PERROR("sem_timedwait");
1955                                 break;
1956                         default:
1957                                 ERR("Unexpected error \"%s\" returned by sem_timedwait",
1958                                         strerror(errno));
1959                         }
1960                 }
1961                 break;
1962         case -1:/* wait forever */
1963                 do {
1964                         ret = sem_wait(&constructor_wait);
1965                 } while (ret < 0 && errno == EINTR);
1966                 if (ret < 0) {
1967                         switch (errno) {
1968                         case EINVAL:
1969                                 PERROR("sem_wait");
1970                                 break;
1971                         default:
1972                                 ERR("Unexpected error \"%s\" returned by sem_wait",
1973                                         strerror(errno));
1974                         }
1975                 }
1976                 break;
1977         case 0: /* no timeout */
1978                 break;
1979         }
1980 }
1981
1982 static
1983 void lttng_ust_cleanup(int exiting)
1984 {
1985         cleanup_sock_info(&global_apps, exiting);
1986         cleanup_sock_info(&local_apps, exiting);
1987         local_apps.allowed = 0;
1988         global_apps.allowed = 0;
1989         /*
1990          * The teardown in this function all affect data structures
1991          * accessed under the UST lock by the listener thread. This
1992          * lock, along with the lttng_ust_comm_should_quit flag, ensure
1993          * that none of these threads are accessing this data at this
1994          * point.
1995          */
1996         lttng_ust_abi_exit();
1997         lttng_ust_events_exit();
1998         lttng_perf_counter_exit();
1999         lttng_ring_buffer_client_discard_rt_exit();
2000         lttng_ring_buffer_client_discard_exit();
2001         lttng_ring_buffer_client_overwrite_rt_exit();
2002         lttng_ring_buffer_client_overwrite_exit();
2003         lttng_ring_buffer_metadata_client_exit();
2004         lttng_ust_statedump_destroy();
2005         exit_tracepoint();
2006         if (!exiting) {
2007                 /* Reinitialize values for fork */
2008                 sem_count = sem_count_initial_value;
2009                 lttng_ust_comm_should_quit = 0;
2010                 initialized = 0;
2011         }
2012 }
2013
2014 void __attribute__((destructor)) lttng_ust_exit(void)
2015 {
2016         int ret;
2017
2018         /*
2019          * Using pthread_cancel here because:
2020          * A) we don't want to hang application teardown.
2021          * B) the thread is not allocating any resource.
2022          */
2023
2024         /*
2025          * Require the communication thread to quit. Synchronize with
2026          * mutexes to ensure it is not in a mutex critical section when
2027          * pthread_cancel is later called.
2028          */
2029         ust_lock_nocheck();
2030         lttng_ust_comm_should_quit = 1;
2031         ust_unlock();
2032
2033         pthread_mutex_lock(&ust_exit_mutex);
2034         /* cancel threads */
2035         if (global_apps.thread_active) {
2036                 ret = pthread_cancel(global_apps.ust_listener);
2037                 if (ret) {
2038                         ERR("Error cancelling global ust listener thread: %s",
2039                                 strerror(ret));
2040                 } else {
2041                         global_apps.thread_active = 0;
2042                 }
2043         }
2044         if (local_apps.thread_active) {
2045                 ret = pthread_cancel(local_apps.ust_listener);
2046                 if (ret) {
2047                         ERR("Error cancelling local ust listener thread: %s",
2048                                 strerror(ret));
2049                 } else {
2050                         local_apps.thread_active = 0;
2051                 }
2052         }
2053         pthread_mutex_unlock(&ust_exit_mutex);
2054
2055         /*
2056          * Do NOT join threads: use of sys_futex makes it impossible to
2057          * join the threads without using async-cancel, but async-cancel
2058          * is delivered by a signal, which could hit the target thread
2059          * anywhere in its code path, including while the ust_lock() is
2060          * held, causing a deadlock for the other thread. Let the OS
2061          * cleanup the threads if there are stalled in a syscall.
2062          */
2063         lttng_ust_cleanup(1);
2064 }
2065
2066 /*
2067  * We exclude the worker threads across fork and clone (except
2068  * CLONE_VM), because these system calls only keep the forking thread
2069  * running in the child.  Therefore, we don't want to call fork or clone
2070  * in the middle of an tracepoint or ust tracing state modification.
2071  * Holding this mutex protects these structures across fork and clone.
2072  */
2073 void ust_before_fork(sigset_t *save_sigset)
2074 {
2075         /*
2076          * Disable signals. This is to avoid that the child intervenes
2077          * before it is properly setup for tracing. It is safer to
2078          * disable all signals, because then we know we are not breaking
2079          * anything by restoring the original mask.
2080          */
2081         sigset_t all_sigs;
2082         int ret;
2083
2084         /* Fixup lttng-ust TLS. */
2085         lttng_ust_fixup_tls();
2086
2087         if (URCU_TLS(lttng_ust_nest_count))
2088                 return;
2089         /* Disable signals */
2090         sigfillset(&all_sigs);
2091         ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2092         if (ret == -1) {
2093                 PERROR("sigprocmask");
2094         }
2095
2096         pthread_mutex_lock(&ust_fork_mutex);
2097
2098         ust_lock_nocheck();
2099         rcu_bp_before_fork();
2100         lttng_ust_lock_fd_tracker();
2101         lttng_perf_lock();
2102 }
2103
2104 static void ust_after_fork_common(sigset_t *restore_sigset)
2105 {
2106         int ret;
2107
2108         DBG("process %d", getpid());
2109         lttng_perf_unlock();
2110         lttng_ust_unlock_fd_tracker();
2111         ust_unlock();
2112
2113         pthread_mutex_unlock(&ust_fork_mutex);
2114
2115         /* Restore signals */
2116         ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2117         if (ret == -1) {
2118                 PERROR("sigprocmask");
2119         }
2120 }
2121
2122 void ust_after_fork_parent(sigset_t *restore_sigset)
2123 {
2124         if (URCU_TLS(lttng_ust_nest_count))
2125                 return;
2126         DBG("process %d", getpid());
2127         rcu_bp_after_fork_parent();
2128         /* Release mutexes and reenable signals */
2129         ust_after_fork_common(restore_sigset);
2130 }
2131
2132 /*
2133  * After fork, in the child, we need to cleanup all the leftover state,
2134  * except the worker thread which already magically disappeared thanks
2135  * to the weird Linux fork semantics. After tyding up, we call
2136  * lttng_ust_init() again to start over as a new PID.
2137  *
2138  * This is meant for forks() that have tracing in the child between the
2139  * fork and following exec call (if there is any).
2140  */
2141 void ust_after_fork_child(sigset_t *restore_sigset)
2142 {
2143         if (URCU_TLS(lttng_ust_nest_count))
2144                 return;
2145         lttng_context_vpid_reset();
2146         lttng_context_vtid_reset();
2147         lttng_context_procname_reset();
2148         DBG("process %d", getpid());
2149         /* Release urcu mutexes */
2150         rcu_bp_after_fork_child();
2151         lttng_ust_cleanup(0);
2152         /* Release mutexes and reenable signals */
2153         ust_after_fork_common(restore_sigset);
2154         lttng_ust_init();
2155 }
2156
2157 void lttng_ust_sockinfo_session_enabled(void *owner)
2158 {
2159         struct sock_info *sock_info = owner;
2160         sock_info->statedump_pending = 1;
2161 }