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 <stddef.h>
  25 #include <stdint.h>
  26 #include <sys/types.h>
  27 #include <sys/socket.h>
  28 #include <sys/mman.h>
  29 #include <sys/stat.h>
  30 #include <sys/types.h>
  31 #include <sys/wait.h>
  32 #include <dlfcn.h>
  33 #include <fcntl.h>
  34 #include <unistd.h>
  35 #include <errno.h>
  36 #include <pthread.h>
  37 #include <semaphore.h>
  38 #include <time.h>
  39 #include <assert.h>
  40 #include <signal.h>
  41 #include <limits.h>
  42 #include <urcu/uatomic.h>
  43 #include <urcu/futex.h>
  44 #include <urcu/compiler.h>
  45
  46 #include <lttng/ust-events.h>
  47 #include <lttng/ust-abi.h>
  48 #include <lttng/ust.h>
  49 #include <lttng/ust-error.h>
  50 #include <lttng/ust-ctl.h>
  51 #include <urcu/tls-compat.h>
  52 #include <ust-comm.h>
  53 #include <ust-fd.h>
  54 #include <usterr-signal-safe.h>
  55 #include <helper.h>
  56 #include "tracepoint-internal.h"
  57 #include "lttng-tracer-core.h"
  58 #include "compat.h"
  59 #include "../libringbuffer/rb-init.h"
  60 #include "lttng-ust-statedump.h"
  61 #include "clock.h"
  62 #include "../libringbuffer/getcpu.h"
  63 #include "getenv.h"
  64
  65 /* Concatenate lttng ust shared library name with its major version number. */
  66 #define LTTNG_UST_LIB_SO_NAME "liblttng-ust.so." __ust_stringify(CONFIG_LTTNG_UST_LIBRARY_VERSION_MAJOR)
  67
  68 /*
  69  * Has lttng ust comm constructor been called ?
  70  */
  71 static int initialized;
  72
  73 /*
  74  * The ust_lock/ust_unlock lock is used as a communication thread mutex.
  75  * Held when handling a command, also held by fork() to deal with
  76  * removal of threads, and by exit path.
  77  *
  78  * The UST lock is the centralized mutex across UST tracing control and
  79  * probe registration.
  80  *
  81  * ust_exit_mutex must never nest in ust_mutex.
  82  *
  83  * ust_fork_mutex must never nest in ust_mutex.
  84  *
  85  * ust_mutex_nest is a per-thread nesting counter, allowing the perf
  86  * counter lazy initialization called by events within the statedump,
  87  * which traces while the ust_mutex is held.
  88  *
  89  * ust_lock nests within the dynamic loader lock (within glibc) because
  90  * it is taken within the library constructor.
  91  *
  92  * The ust fd tracker lock nests within the ust_mutex.
  93  */
  94 static pthread_mutex_t ust_mutex = PTHREAD_MUTEX_INITIALIZER;
  95
  96 /* Allow nesting the ust_mutex within the same thread. */
  97 static DEFINE_URCU_TLS(int, ust_mutex_nest);
  98
  99 /*
 100  * ust_exit_mutex protects thread_active variable wrt thread exit. It
 101  * cannot be done by ust_mutex because pthread_cancel(), which takes an
 102  * internal libc lock, cannot nest within ust_mutex.
 103  *
 104  * It never nests within a ust_mutex.
 105  */
 106 static pthread_mutex_t ust_exit_mutex = PTHREAD_MUTEX_INITIALIZER;
 107
 108 /*
 109  * ust_fork_mutex protects base address statedump tracing against forks. It
 110  * prevents the dynamic loader lock to be taken (by base address statedump
 111  * tracing) while a fork is happening, thus preventing deadlock issues with
 112  * the dynamic loader lock.
 113  */
 114 static pthread_mutex_t ust_fork_mutex = PTHREAD_MUTEX_INITIALIZER;
 115
 116 /* Should the ust comm thread quit ? */
 117 static int lttng_ust_comm_should_quit;
 118
 119 /*
 120  * This variable can be tested by applications to check whether
 121  * lttng-ust is loaded. They simply have to define their own
 122  * "lttng_ust_loaded" weak symbol, and test it. It is set to 1 by the
 123  * library constructor.
 124  */
 125 int lttng_ust_loaded __attribute__((weak));
 126
 127 /*
 128  * Return 0 on success, -1 if should quit.
 129  * The lock is taken in both cases.
 130  * Signal-safe.
 131  */
 132 int ust_lock(void)
 133 {
 134         sigset_t sig_all_blocked, orig_mask;
 135         int ret, oldstate;
 136
 137         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
 138         if (ret) {
 139                 ERR("pthread_setcancelstate: %s", strerror(ret));
 140         }
 141         if (oldstate != PTHREAD_CANCEL_ENABLE) {
 142                 ERR("pthread_setcancelstate: unexpected oldstate");
 143         }
 144         sigfillset(&sig_all_blocked);
 145         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 146         if (ret) {
 147                 ERR("pthread_sigmask: %s", strerror(ret));
 148         }
 149         if (!URCU_TLS(ust_mutex_nest)++)
 150                 pthread_mutex_lock(&ust_mutex);
 151         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 152         if (ret) {
 153                 ERR("pthread_sigmask: %s", strerror(ret));
 154         }
 155         if (lttng_ust_comm_should_quit) {
 156                 return -1;
 157         } else {
 158                 return 0;
 159         }
 160 }
 161
 162 /*
 163  * ust_lock_nocheck() can be used in constructors/destructors, because
 164  * they are already nested within the dynamic loader lock, and therefore
 165  * have exclusive access against execution of liblttng-ust destructor.
 166  * Signal-safe.
 167  */
 168 void ust_lock_nocheck(void)
 169 {
 170         sigset_t sig_all_blocked, orig_mask;
 171         int ret, oldstate;
 172
 173         ret = pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, &oldstate);
 174         if (ret) {
 175                 ERR("pthread_setcancelstate: %s", strerror(ret));
 176         }
 177         if (oldstate != PTHREAD_CANCEL_ENABLE) {
 178                 ERR("pthread_setcancelstate: unexpected oldstate");
 179         }
 180         sigfillset(&sig_all_blocked);
 181         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 182         if (ret) {
 183                 ERR("pthread_sigmask: %s", strerror(ret));
 184         }
 185         if (!URCU_TLS(ust_mutex_nest)++)
 186                 pthread_mutex_lock(&ust_mutex);
 187         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 188         if (ret) {
 189                 ERR("pthread_sigmask: %s", strerror(ret));
 190         }
 191 }
 192
 193 /*
 194  * Signal-safe.
 195  */
 196 void ust_unlock(void)
 197 {
 198         sigset_t sig_all_blocked, orig_mask;
 199         int ret, oldstate;
 200
 201         sigfillset(&sig_all_blocked);
 202         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_mask);
 203         if (ret) {
 204                 ERR("pthread_sigmask: %s", strerror(ret));
 205         }
 206         if (!--URCU_TLS(ust_mutex_nest))
 207                 pthread_mutex_unlock(&ust_mutex);
 208         ret = pthread_sigmask(SIG_SETMASK, &orig_mask, NULL);
 209         if (ret) {
 210                 ERR("pthread_sigmask: %s", strerror(ret));
 211         }
 212         ret = pthread_setcancelstate(PTHREAD_CANCEL_ENABLE, &oldstate);
 213         if (ret) {
 214                 ERR("pthread_setcancelstate: %s", strerror(ret));
 215         }
 216         if (oldstate != PTHREAD_CANCEL_DISABLE) {
 217                 ERR("pthread_setcancelstate: unexpected oldstate");
 218         }
 219 }
 220
 221 /*
 222  * Wait for either of these before continuing to the main
 223  * program:
 224  * - the register_done message from sessiond daemon
 225  *   (will let the sessiond daemon enable sessions before main
 226  *   starts.)
 227  * - sessiond daemon is not reachable.
 228  * - timeout (ensuring applications are resilient to session
 229  *   daemon problems).
 230  */
 231 static sem_t constructor_wait;
 232 /*
 233  * Doing this for both the global and local sessiond.
 234  */
 235 enum {
 236         sem_count_initial_value = 4,
 237 };
 238
 239 static int sem_count = sem_count_initial_value;
 240
 241 /*
 242  * Counting nesting within lttng-ust. Used to ensure that calling fork()
 243  * from liblttng-ust does not execute the pre/post fork handlers.
 244  */
 245 static DEFINE_URCU_TLS(int, lttng_ust_nest_count);
 246
 247 /*
 248  * Info about socket and associated listener thread.
 249  */
 250 struct sock_info {
 251         const char *name;
 252         pthread_t ust_listener; /* listener thread */
 253         int root_handle;
 254         int registration_done;
 255         int allowed;
 256         int global;
 257         int thread_active;
 258
 259         char sock_path[PATH_MAX];
 260         int socket;
 261         int notify_socket;
 262
 263         char wait_shm_path[PATH_MAX];
 264         char *wait_shm_mmap;
 265         /* Keep track of lazy state dump not performed yet. */
 266         int statedump_pending;
 267         int initial_statedump_done;
 268         /* Keep procname for statedump */
 269         char procname[LTTNG_UST_PROCNAME_LEN];
 270 };
 271
 272 /* Socket from app (connect) to session daemon (listen) for communication */
 273 struct sock_info global_apps = {
 274         .name = "global",
 275         .global = 1,
 276
 277         .root_handle = -1,
 278         .registration_done = 0,
 279         .allowed = 0,
 280         .thread_active = 0,
 281
 282         .sock_path = LTTNG_DEFAULT_RUNDIR "/" LTTNG_UST_SOCK_FILENAME,
 283         .socket = -1,
 284         .notify_socket = -1,
 285
 286         .wait_shm_path = "/" LTTNG_UST_WAIT_FILENAME,
 287
 288         .statedump_pending = 0,
 289         .initial_statedump_done = 0,
 290         .procname[0] = '\0'
 291 };
 292
 293 /* TODO: allow global_apps_sock_path override */
 294
 295 struct sock_info local_apps = {
 296         .name = "local",
 297         .global = 0,
 298         .root_handle = -1,
 299         .registration_done = 0,
 300         .allowed = 0,   /* Check setuid bit first */
 301         .thread_active = 0,
 302
 303         .socket = -1,
 304         .notify_socket = -1,
 305
 306         .statedump_pending = 0,
 307         .initial_statedump_done = 0,
 308         .procname[0] = '\0'
 309 };
 310
 311 static int wait_poll_fallback;
 312
 313 static const char *cmd_name_mapping[] = {
 314         [ LTTNG_UST_RELEASE ] = "Release",
 315         [ LTTNG_UST_SESSION ] = "Create Session",
 316         [ LTTNG_UST_TRACER_VERSION ] = "Get Tracer Version",
 317
 318         [ LTTNG_UST_TRACEPOINT_LIST ] = "Create Tracepoint List",
 319         [ LTTNG_UST_WAIT_QUIESCENT ] = "Wait for Quiescent State",
 320         [ LTTNG_UST_REGISTER_DONE ] = "Registration Done",
 321         [ LTTNG_UST_TRACEPOINT_FIELD_LIST ] = "Create Tracepoint Field List",
 322
 323         /* Session FD commands */
 324         [ LTTNG_UST_CHANNEL ] = "Create Channel",
 325         [ LTTNG_UST_SESSION_START ] = "Start Session",
 326         [ LTTNG_UST_SESSION_STOP ] = "Stop Session",
 327
 328         /* Channel FD commands */
 329         [ LTTNG_UST_STREAM ] = "Create Stream",
 330         [ LTTNG_UST_EVENT ] = "Create Event",
 331
 332         /* Event and Channel FD commands */
 333         [ LTTNG_UST_CONTEXT ] = "Create Context",
 334         [ LTTNG_UST_FLUSH_BUFFER ] = "Flush Buffer",
 335
 336         /* Event, Channel and Session commands */
 337         [ LTTNG_UST_ENABLE ] = "Enable",
 338         [ LTTNG_UST_DISABLE ] = "Disable",
 339
 340         /* Tracepoint list commands */
 341         [ LTTNG_UST_TRACEPOINT_LIST_GET ] = "List Next Tracepoint",
 342         [ LTTNG_UST_TRACEPOINT_FIELD_LIST_GET ] = "List Next Tracepoint Field",
 343
 344         /* Event FD commands */
 345         [ LTTNG_UST_FILTER ] = "Create Filter",
 346         [ LTTNG_UST_EXCLUSION ] = "Add exclusions to event",
 347 };
 348
 349 static const char *str_timeout;
 350 static int got_timeout_env;
 351
 352 extern void lttng_ring_buffer_client_overwrite_init(void);
 353 extern void lttng_ring_buffer_client_overwrite_rt_init(void);
 354 extern void lttng_ring_buffer_client_discard_init(void);
 355 extern void lttng_ring_buffer_client_discard_rt_init(void);
 356 extern void lttng_ring_buffer_metadata_client_init(void);
 357 extern void lttng_ring_buffer_client_overwrite_exit(void);
 358 extern void lttng_ring_buffer_client_overwrite_rt_exit(void);
 359 extern void lttng_ring_buffer_client_discard_exit(void);
 360 extern void lttng_ring_buffer_client_discard_rt_exit(void);
 361 extern void lttng_ring_buffer_metadata_client_exit(void);
 362
 363 static char *get_map_shm(struct sock_info *sock_info);
 364
 365 ssize_t lttng_ust_read(int fd, void *buf, size_t len)
 366 {
 367         ssize_t ret;
 368         size_t copied = 0, to_copy = len;
 369
 370         do {
 371                 ret = read(fd, buf + copied, to_copy);
 372                 if (ret > 0) {
 373                         copied += ret;
 374                         to_copy -= ret;
 375                 }
 376         } while ((ret > 0 && to_copy > 0)
 377                 || (ret < 0 && errno == EINTR));
 378         if (ret > 0) {
 379                 ret = copied;
 380         }
 381         return ret;
 382 }
 383 /*
 384  * Returns the HOME directory path. Caller MUST NOT free(3) the returned
 385  * pointer.
 386  */
 387 static
 388 const char *get_lttng_home_dir(void)
 389 {
 390        const char *val;
 391
 392        val = (const char *) lttng_getenv("LTTNG_HOME");
 393        if (val != NULL) {
 394                return val;
 395        }
 396        return (const char *) lttng_getenv("HOME");
 397 }
 398
 399 /*
 400  * Force a read (imply TLS fixup for dlopen) of TLS variables.
 401  */
 402 static
 403 void lttng_fixup_nest_count_tls(void)
 404 {
 405         asm volatile ("" : : "m" (URCU_TLS(lttng_ust_nest_count)));
 406 }
 407
 408 static
 409 void lttng_fixup_ust_mutex_nest_tls(void)
 410 {
 411         asm volatile ("" : : "m" (URCU_TLS(ust_mutex_nest)));
 412 }
 413
 414 /*
 415  * Fixup urcu bp TLS.
 416  */
 417 static
 418 void lttng_fixup_urcu_bp_tls(void)
 419 {
 420         rcu_read_lock();
 421         rcu_read_unlock();
 422 }
 423
 424 void lttng_ust_fixup_tls(void)
 425 {
 426         lttng_fixup_urcu_bp_tls();
 427         lttng_fixup_ringbuffer_tls();
 428         lttng_fixup_vtid_tls();
 429         lttng_fixup_nest_count_tls();
 430         lttng_fixup_procname_tls();
 431         lttng_fixup_ust_mutex_nest_tls();
 432         lttng_ust_fixup_perf_counter_tls();
 433         lttng_ust_fixup_fd_tracker_tls();
 434         lttng_fixup_cgroup_ns_tls();
 435         lttng_fixup_ipc_ns_tls();
 436         lttng_fixup_net_ns_tls();
 437         lttng_fixup_time_ns_tls();
 438         lttng_fixup_uts_ns_tls();
 439 }
 440
 441 int lttng_get_notify_socket(void *owner)
 442 {
 443         struct sock_info *info = owner;
 444
 445         return info->notify_socket;
 446 }
 447
 448
 449 LTTNG_HIDDEN
 450 char* lttng_ust_sockinfo_get_procname(void *owner)
 451 {
 452         struct sock_info *info = owner;
 453
 454         return info->procname;
 455 }
 456
 457 static
 458 void print_cmd(int cmd, int handle)
 459 {
 460         const char *cmd_name = "Unknown";
 461
 462         if (cmd >= 0 && cmd < LTTNG_ARRAY_SIZE(cmd_name_mapping)
 463                         && cmd_name_mapping[cmd]) {
 464                 cmd_name = cmd_name_mapping[cmd];
 465         }
 466         DBG("Message Received \"%s\" (%d), Handle \"%s\" (%d)",
 467                 cmd_name, cmd,
 468                 lttng_ust_obj_get_name(handle), handle);
 469 }
 470
 471 static
 472 int setup_global_apps(void)
 473 {
 474         int ret = 0;
 475         assert(!global_apps.wait_shm_mmap);
 476
 477         global_apps.wait_shm_mmap = get_map_shm(&global_apps);
 478         if (!global_apps.wait_shm_mmap) {
 479                 WARN("Unable to get map shm for global apps. Disabling LTTng-UST global tracing.");
 480                 global_apps.allowed = 0;
 481                 ret = -EIO;
 482                 goto error;
 483         }
 484
 485         global_apps.allowed = 1;
 486         lttng_ust_getprocname(global_apps.procname);
 487 error:
 488         return ret;
 489 }
 490 static
 491 int setup_local_apps(void)
 492 {
 493         int ret = 0;
 494         const char *home_dir;
 495         uid_t uid;
 496
 497         assert(!local_apps.wait_shm_mmap);
 498
 499         uid = getuid();
 500         /*
 501          * Disallow per-user tracing for setuid binaries.
 502          */
 503         if (uid != geteuid()) {
 504                 assert(local_apps.allowed == 0);
 505                 ret = 0;
 506                 goto end;
 507         }
 508         home_dir = get_lttng_home_dir();
 509         if (!home_dir) {
 510                 WARN("HOME environment variable not set. Disabling LTTng-UST per-user tracing.");
 511                 assert(local_apps.allowed == 0);
 512                 ret = -ENOENT;
 513                 goto end;
 514         }
 515         local_apps.allowed = 1;
 516         snprintf(local_apps.sock_path, PATH_MAX, "%s/%s/%s",
 517                 home_dir,
 518                 LTTNG_DEFAULT_HOME_RUNDIR,
 519                 LTTNG_UST_SOCK_FILENAME);
 520         snprintf(local_apps.wait_shm_path, PATH_MAX, "/%s-%u",
 521                 LTTNG_UST_WAIT_FILENAME,
 522                 uid);
 523
 524         local_apps.wait_shm_mmap = get_map_shm(&local_apps);
 525         if (!local_apps.wait_shm_mmap) {
 526                 WARN("Unable to get map shm for local apps. Disabling LTTng-UST per-user tracing.");
 527                 local_apps.allowed = 0;
 528                 ret = -EIO;
 529                 goto end;
 530         }
 531
 532         lttng_ust_getprocname(local_apps.procname);
 533 end:
 534         return ret;
 535 }
 536
 537 /*
 538  * Get socket timeout, in ms.
 539  * -1: wait forever. 0: don't wait. >0: timeout, in ms.
 540  */
 541 static
 542 long get_timeout(void)
 543 {
 544         long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 545
 546         if (!got_timeout_env) {
 547                 str_timeout = lttng_getenv("LTTNG_UST_REGISTER_TIMEOUT");
 548                 got_timeout_env = 1;
 549         }
 550         if (str_timeout)
 551                 constructor_delay_ms = strtol(str_timeout, NULL, 10);
 552         /* All negative values are considered as "-1". */
 553         if (constructor_delay_ms < -1)
 554                 constructor_delay_ms = -1;
 555         return constructor_delay_ms;
 556 }
 557
 558 /* Timeout for notify socket send and recv. */
 559 static
 560 long get_notify_sock_timeout(void)
 561 {
 562         return get_timeout();
 563 }
 564
 565 /* Timeout for connecting to cmd and notify sockets. */
 566 static
 567 long get_connect_sock_timeout(void)
 568 {
 569         return get_timeout();
 570 }
 571
 572 /*
 573  * Return values: -1: wait forever. 0: don't wait. 1: timeout wait.
 574  */
 575 static
 576 int get_constructor_timeout(struct timespec *constructor_timeout)
 577 {
 578         long constructor_delay_ms;
 579         int ret;
 580
 581         constructor_delay_ms = get_timeout();
 582
 583         switch (constructor_delay_ms) {
 584         case -1:/* fall-through */
 585         case 0:
 586                 return constructor_delay_ms;
 587         default:
 588                 break;
 589         }
 590
 591         /*
 592          * If we are unable to find the current time, don't wait.
 593          */
 594         ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
 595         if (ret) {
 596                 /* Don't wait. */
 597                 return 0;
 598         }
 599         constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
 600         constructor_timeout->tv_nsec +=
 601                 (constructor_delay_ms % 1000UL) * 1000000UL;
 602         if (constructor_timeout->tv_nsec >= 1000000000UL) {
 603                 constructor_timeout->tv_sec++;
 604                 constructor_timeout->tv_nsec -= 1000000000UL;
 605         }
 606         /* Timeout wait (constructor_delay_ms). */
 607         return 1;
 608 }
 609
 610 static
 611 void get_allow_blocking(void)
 612 {
 613         const char *str_allow_blocking =
 614                 lttng_getenv("LTTNG_UST_ALLOW_BLOCKING");
 615
 616         if (str_allow_blocking) {
 617                 DBG("%s environment variable is set",
 618                         "LTTNG_UST_ALLOW_BLOCKING");
 619                 lttng_ust_ringbuffer_set_allow_blocking();
 620         }
 621 }
 622
 623 static
 624 int register_to_sessiond(int socket, enum ustctl_socket_type type)
 625 {
 626         return ustcomm_send_reg_msg(socket,
 627                 type,
 628                 CAA_BITS_PER_LONG,
 629                 lttng_alignof(uint8_t) * CHAR_BIT,
 630                 lttng_alignof(uint16_t) * CHAR_BIT,
 631                 lttng_alignof(uint32_t) * CHAR_BIT,
 632                 lttng_alignof(uint64_t) * CHAR_BIT,
 633                 lttng_alignof(unsigned long) * CHAR_BIT);
 634 }
 635
 636 static
 637 int send_reply(int sock, struct ustcomm_ust_reply *lur)
 638 {
 639         ssize_t len;
 640
 641         len = ustcomm_send_unix_sock(sock, lur, sizeof(*lur));
 642         switch (len) {
 643         case sizeof(*lur):
 644                 DBG("message successfully sent");
 645                 return 0;
 646         default:
 647                 if (len == -ECONNRESET) {
 648                         DBG("remote end closed connection");
 649                         return 0;
 650                 }
 651                 if (len < 0)
 652                         return len;
 653                 DBG("incorrect message size: %zd", len);
 654                 return -EINVAL;
 655         }
 656 }
 657
 658 static
 659 void decrement_sem_count(unsigned int count)
 660 {
 661         int ret;
 662
 663         assert(uatomic_read(&sem_count) >= count);
 664
 665         if (uatomic_read(&sem_count) <= 0) {
 666                 return;
 667         }
 668
 669         ret = uatomic_add_return(&sem_count, -count);
 670         if (ret == 0) {
 671                 ret = sem_post(&constructor_wait);
 672                 assert(!ret);
 673         }
 674 }
 675
 676 static
 677 int handle_register_done(struct sock_info *sock_info)
 678 {
 679         if (sock_info->registration_done)
 680                 return 0;
 681         sock_info->registration_done = 1;
 682
 683         decrement_sem_count(1);
 684         if (!sock_info->statedump_pending) {
 685                 sock_info->initial_statedump_done = 1;
 686                 decrement_sem_count(1);
 687         }
 688
 689         return 0;
 690 }
 691
 692 static
 693 int handle_register_failed(struct sock_info *sock_info)
 694 {
 695         if (sock_info->registration_done)
 696                 return 0;
 697         sock_info->registration_done = 1;
 698         sock_info->initial_statedump_done = 1;
 699
 700         decrement_sem_count(2);
 701
 702         return 0;
 703 }
 704
 705 /*
 706  * Only execute pending statedump after the constructor semaphore has
 707  * been posted by the current listener thread. This means statedump will
 708  * only be performed after the "registration done" command is received
 709  * from this thread's session daemon.
 710  *
 711  * This ensures we don't run into deadlock issues with the dynamic
 712  * loader mutex, which is held while the constructor is called and
 713  * waiting on the constructor semaphore. All operations requiring this
 714  * dynamic loader lock need to be postponed using this mechanism.
 715  *
 716  * In a scenario with two session daemons connected to the application,
 717  * it is possible that the first listener thread which receives the
 718  * registration done command issues its statedump while the dynamic
 719  * loader lock is still held by the application constructor waiting on
 720  * the semaphore. It will however be allowed to proceed when the
 721  * second session daemon sends the registration done command to the
 722  * second listener thread. This situation therefore does not produce
 723  * a deadlock.
 724  */
 725 static
 726 void handle_pending_statedump(struct sock_info *sock_info)
 727 {
 728         if (sock_info->registration_done && sock_info->statedump_pending) {
 729                 sock_info->statedump_pending = 0;
 730                 pthread_mutex_lock(&ust_fork_mutex);
 731                 lttng_handle_pending_statedump(sock_info);
 732                 pthread_mutex_unlock(&ust_fork_mutex);
 733
 734                 if (!sock_info->initial_statedump_done) {
 735                         sock_info->initial_statedump_done = 1;
 736                         decrement_sem_count(1);
 737                 }
 738         }
 739 }
 740
 741 static
 742 int handle_message(struct sock_info *sock_info,
 743                 int sock, struct ustcomm_ust_msg *lum)
 744 {
 745         int ret = 0;
 746         const struct lttng_ust_objd_ops *ops;
 747         struct ustcomm_ust_reply lur;
 748         union ust_args args;
 749         char ctxstr[LTTNG_UST_SYM_NAME_LEN];    /* App context string. */
 750         ssize_t len;
 751
 752         memset(&lur, 0, sizeof(lur));
 753
 754         if (ust_lock()) {
 755                 ret = -LTTNG_UST_ERR_EXITING;
 756                 goto error;
 757         }
 758
 759         ops = objd_ops(lum->handle);
 760         if (!ops) {
 761                 ret = -ENOENT;
 762                 goto error;
 763         }
 764
 765         switch (lum->cmd) {
 766         case LTTNG_UST_REGISTER_DONE:
 767                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 768                         ret = handle_register_done(sock_info);
 769                 else
 770                         ret = -EINVAL;
 771                 break;
 772         case LTTNG_UST_RELEASE:
 773                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 774                         ret = -EPERM;
 775                 else
 776                         ret = lttng_ust_objd_unref(lum->handle, 1);
 777                 break;
 778         case LTTNG_UST_FILTER:
 779         {
 780                 /* Receive filter data */
 781                 struct lttng_ust_filter_bytecode_node *bytecode;
 782
 783                 if (lum->u.filter.data_size > FILTER_BYTECODE_MAX_LEN) {
 784                         ERR("Filter data size is too large: %u bytes",
 785                                 lum->u.filter.data_size);
 786                         ret = -EINVAL;
 787                         goto error;
 788                 }
 789
 790                 if (lum->u.filter.reloc_offset > lum->u.filter.data_size) {
 791                         ERR("Filter reloc offset %u is not within data",
 792                                 lum->u.filter.reloc_offset);
 793                         ret = -EINVAL;
 794                         goto error;
 795                 }
 796
 797                 bytecode = zmalloc(sizeof(*bytecode) + lum->u.filter.data_size);
 798                 if (!bytecode) {
 799                         ret = -ENOMEM;
 800                         goto error;
 801                 }
 802                 len = ustcomm_recv_unix_sock(sock, bytecode->bc.data,
 803                                 lum->u.filter.data_size);
 804                 switch (len) {
 805                 case 0: /* orderly shutdown */
 806                         ret = 0;
 807                         free(bytecode);
 808                         goto error;
 809                 default:
 810                         if (len == lum->u.filter.data_size) {
 811                                 DBG("filter data received");
 812                                 break;
 813                         } else if (len < 0) {
 814                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 815                                 if (len == -ECONNRESET) {
 816                                         ERR("%s remote end closed connection", sock_info->name);
 817                                         ret = len;
 818                                         free(bytecode);
 819                                         goto error;
 820                                 }
 821                                 ret = len;
 822                                 free(bytecode);
 823                                 goto error;
 824                         } else {
 825                                 DBG("incorrect filter data message size: %zd", len);
 826                                 ret = -EINVAL;
 827                                 free(bytecode);
 828                                 goto error;
 829                         }
 830                 }
 831                 bytecode->bc.len = lum->u.filter.data_size;
 832                 bytecode->bc.reloc_offset = lum->u.filter.reloc_offset;
 833                 bytecode->bc.seqnum = lum->u.filter.seqnum;
 834                 if (ops->cmd) {
 835                         ret = ops->cmd(lum->handle, lum->cmd,
 836                                         (unsigned long) bytecode,
 837                                         &args, sock_info);
 838                         if (ret) {
 839                                 free(bytecode);
 840                         }
 841                         /* don't free bytecode if everything went fine. */
 842                 } else {
 843                         ret = -ENOSYS;
 844                         free(bytecode);
 845                 }
 846                 break;
 847         }
 848         case LTTNG_UST_EXCLUSION:
 849         {
 850                 /* Receive exclusion names */
 851                 struct lttng_ust_excluder_node *node;
 852                 unsigned int count;
 853
 854                 count = lum->u.exclusion.count;
 855                 if (count == 0) {
 856                         /* There are no names to read */
 857                         ret = 0;
 858                         goto error;
 859                 }
 860                 node = zmalloc(sizeof(*node) +
 861                                 count * LTTNG_UST_SYM_NAME_LEN);
 862                 if (!node) {
 863                         ret = -ENOMEM;
 864                         goto error;
 865                 }
 866                 node->excluder.count = count;
 867                 len = ustcomm_recv_unix_sock(sock, node->excluder.names,
 868                                 count * LTTNG_UST_SYM_NAME_LEN);
 869                 switch (len) {
 870                 case 0: /* orderly shutdown */
 871                         ret = 0;
 872                         free(node);
 873                         goto error;
 874                 default:
 875                         if (len == count * LTTNG_UST_SYM_NAME_LEN) {
 876                                 DBG("Exclusion data received");
 877                                 break;
 878                         } else if (len < 0) {
 879                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 880                                 if (len == -ECONNRESET) {
 881                                         ERR("%s remote end closed connection", sock_info->name);
 882                                         ret = len;
 883                                         free(node);
 884                                         goto error;
 885                                 }
 886                                 ret = len;
 887                                 free(node);
 888                                 goto error;
 889                         } else {
 890                                 DBG("Incorrect exclusion data message size: %zd", len);
 891                                 ret = -EINVAL;
 892                                 free(node);
 893                                 goto error;
 894                         }
 895                 }
 896                 if (ops->cmd) {
 897                         ret = ops->cmd(lum->handle, lum->cmd,
 898                                         (unsigned long) node,
 899                                         &args, sock_info);
 900                         if (ret) {
 901                                 free(node);
 902                         }
 903                         /* Don't free exclusion data if everything went fine. */
 904                 } else {
 905                         ret = -ENOSYS;
 906                         free(node);
 907                 }
 908                 break;
 909         }
 910         case LTTNG_UST_CHANNEL:
 911         {
 912                 void *chan_data;
 913                 int wakeup_fd;
 914
 915                 len = ustcomm_recv_channel_from_sessiond(sock,
 916                                 &chan_data, lum->u.channel.len,
 917                                 &wakeup_fd);
 918                 switch (len) {
 919                 case 0: /* orderly shutdown */
 920                         ret = 0;
 921                         goto error;
 922                 default:
 923                         if (len == lum->u.channel.len) {
 924                                 DBG("channel data received");
 925                                 break;
 926                         } else if (len < 0) {
 927                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 928                                 if (len == -ECONNRESET) {
 929                                         ERR("%s remote end closed connection", sock_info->name);
 930                                         ret = len;
 931                                         goto error;
 932                                 }
 933                                 ret = len;
 934                                 goto error;
 935                         } else {
 936                                 DBG("incorrect channel data message size: %zd", len);
 937                                 ret = -EINVAL;
 938                                 goto error;
 939                         }
 940                 }
 941                 args.channel.chan_data = chan_data;
 942                 args.channel.wakeup_fd = wakeup_fd;
 943                 if (ops->cmd)
 944                         ret = ops->cmd(lum->handle, lum->cmd,
 945                                         (unsigned long) &lum->u,
 946                                         &args, sock_info);
 947                 else
 948                         ret = -ENOSYS;
 949                 break;
 950         }
 951         case LTTNG_UST_STREAM:
 952         {
 953                 /* Receive shm_fd, wakeup_fd */
 954                 ret = ustcomm_recv_stream_from_sessiond(sock,
 955                         NULL,
 956                         &args.stream.shm_fd,
 957                         &args.stream.wakeup_fd);
 958                 if (ret) {
 959                         goto error;
 960                 }
 961
 962                 if (ops->cmd)
 963                         ret = ops->cmd(lum->handle, lum->cmd,
 964                                         (unsigned long) &lum->u,
 965                                         &args, sock_info);
 966                 else
 967                         ret = -ENOSYS;
 968                 break;
 969         }
 970         case LTTNG_UST_CONTEXT:
 971                 switch (lum->u.context.ctx) {
 972                 case LTTNG_UST_CONTEXT_APP_CONTEXT:
 973                 {
 974                         char *p;
 975                         size_t ctxlen, recvlen;
 976
 977                         ctxlen = strlen("$app.") + lum->u.context.u.app_ctx.provider_name_len - 1
 978                                         + strlen(":") + lum->u.context.u.app_ctx.ctx_name_len;
 979                         if (ctxlen >= LTTNG_UST_SYM_NAME_LEN) {
 980                                 ERR("Application context string length size is too large: %zu bytes",
 981                                         ctxlen);
 982                                 ret = -EINVAL;
 983                                 goto error;
 984                         }
 985                         strcpy(ctxstr, "$app.");
 986                         p = &ctxstr[strlen("$app.")];
 987                         recvlen = ctxlen - strlen("$app.");
 988                         len = ustcomm_recv_unix_sock(sock, p, recvlen);
 989                         switch (len) {
 990                         case 0: /* orderly shutdown */
 991                                 ret = 0;
 992                                 goto error;
 993                         default:
 994                                 if (len == recvlen) {
 995                                         DBG("app context data received");
 996                                         break;
 997                                 } else if (len < 0) {
 998                                         DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
 999                                         if (len == -ECONNRESET) {
1000                                                 ERR("%s remote end closed connection", sock_info->name);
1001                                                 ret = len;
1002                                                 goto error;
1003                                         }
1004                                         ret = len;
1005                                         goto error;
1006                                 } else {
1007                                         DBG("incorrect app context data message size: %zd", len);
1008                                         ret = -EINVAL;
1009                                         goto error;
1010                                 }
1011                         }
1012                         /* Put : between provider and ctxname. */
1013                         p[lum->u.context.u.app_ctx.provider_name_len - 1] = ':';
1014                         args.app_context.ctxname = ctxstr;
1015                         break;
1016                 }
1017                 default:
1018                         break;
1019                 }
1020                 if (ops->cmd) {
1021                         ret = ops->cmd(lum->handle, lum->cmd,
1022                                         (unsigned long) &lum->u,
1023                                         &args, sock_info);
1024                 } else {
1025                         ret = -ENOSYS;
1026                 }
1027                 break;
1028         default:
1029                 if (ops->cmd)
1030                         ret = ops->cmd(lum->handle, lum->cmd,
1031                                         (unsigned long) &lum->u,
1032                                         &args, sock_info);
1033                 else
1034                         ret = -ENOSYS;
1035                 break;
1036         }
1037
1038         lur.handle = lum->handle;
1039         lur.cmd = lum->cmd;
1040         lur.ret_val = ret;
1041         if (ret >= 0) {
1042                 lur.ret_code = LTTNG_UST_OK;
1043         } else {
1044                 /*
1045                  * Use -LTTNG_UST_ERR as wildcard for UST internal
1046                  * error that are not caused by the transport, except if
1047                  * we already have a more precise error message to
1048                  * report.
1049                  */
1050                 if (ret > -LTTNG_UST_ERR) {
1051                         /* Translate code to UST error. */
1052                         switch (ret) {
1053                         case -EEXIST:
1054                                 lur.ret_code = -LTTNG_UST_ERR_EXIST;
1055                                 break;
1056                         case -EINVAL:
1057                                 lur.ret_code = -LTTNG_UST_ERR_INVAL;
1058                                 break;
1059                         case -ENOENT:
1060                                 lur.ret_code = -LTTNG_UST_ERR_NOENT;
1061                                 break;
1062                         case -EPERM:
1063                                 lur.ret_code = -LTTNG_UST_ERR_PERM;
1064                                 break;
1065                         case -ENOSYS:
1066                                 lur.ret_code = -LTTNG_UST_ERR_NOSYS;
1067                                 break;
1068                         default:
1069                                 lur.ret_code = -LTTNG_UST_ERR;
1070                                 break;
1071                         }
1072                 } else {
1073                         lur.ret_code = ret;
1074                 }
1075         }
1076         if (ret >= 0) {
1077                 switch (lum->cmd) {
1078                 case LTTNG_UST_TRACER_VERSION:
1079                         lur.u.version = lum->u.version;
1080                         break;
1081                 case LTTNG_UST_TRACEPOINT_LIST_GET:
1082                         memcpy(&lur.u.tracepoint, &lum->u.tracepoint, sizeof(lur.u.tracepoint));
1083                         break;
1084                 }
1085         }
1086         DBG("Return value: %d", lur.ret_val);
1087
1088         ust_unlock();
1089
1090         /*
1091          * Performed delayed statedump operations outside of the UST
1092          * lock. We need to take the dynamic loader lock before we take
1093          * the UST lock internally within handle_pending_statedump().
1094           */
1095         handle_pending_statedump(sock_info);
1096
1097         if (ust_lock()) {
1098                 ret = -LTTNG_UST_ERR_EXITING;
1099                 goto error;
1100         }
1101
1102         ret = send_reply(sock, &lur);
1103         if (ret < 0) {
1104                 DBG("error sending reply");
1105                 goto error;
1106         }
1107
1108         /*
1109          * LTTNG_UST_TRACEPOINT_FIELD_LIST_GET needs to send the field
1110          * after the reply.
1111          */
1112         if (lur.ret_code == LTTNG_UST_OK) {
1113                 switch (lum->cmd) {
1114                 case LTTNG_UST_TRACEPOINT_FIELD_LIST_GET:
1115                         len = ustcomm_send_unix_sock(sock,
1116                                 &args.field_list.entry,
1117                                 sizeof(args.field_list.entry));
1118                         if (len < 0) {
1119                                 ret = len;
1120                                 goto error;
1121                         }
1122                         if (len != sizeof(args.field_list.entry)) {
1123                                 ret = -EINVAL;
1124                                 goto error;
1125                         }
1126                 }
1127         }
1128
1129 error:
1130         ust_unlock();
1131
1132         return ret;
1133 }
1134
1135 static
1136 void cleanup_sock_info(struct sock_info *sock_info, int exiting)
1137 {
1138         int ret;
1139
1140         if (sock_info->root_handle != -1) {
1141                 ret = lttng_ust_objd_unref(sock_info->root_handle, 1);
1142                 if (ret) {
1143                         ERR("Error unref root handle");
1144                 }
1145                 sock_info->root_handle = -1;
1146         }
1147         sock_info->registration_done = 0;
1148         sock_info->initial_statedump_done = 0;
1149
1150         /*
1151          * wait_shm_mmap, socket and notify socket are used by listener
1152          * threads outside of the ust lock, so we cannot tear them down
1153          * ourselves, because we cannot join on these threads. Leave
1154          * responsibility of cleaning up these resources to the OS
1155          * process exit.
1156          */
1157         if (exiting)
1158                 return;
1159
1160         if (sock_info->socket != -1) {
1161                 ret = ustcomm_close_unix_sock(sock_info->socket);
1162                 if (ret) {
1163                         ERR("Error closing ust cmd socket");
1164                 }
1165                 sock_info->socket = -1;
1166         }
1167         if (sock_info->notify_socket != -1) {
1168                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1169                 if (ret) {
1170                         ERR("Error closing ust notify socket");
1171                 }
1172                 sock_info->notify_socket = -1;
1173         }
1174         if (sock_info->wait_shm_mmap) {
1175                 long page_size;
1176
1177                 page_size = sysconf(_SC_PAGE_SIZE);
1178                 if (page_size <= 0) {
1179                         if (!page_size) {
1180                                 errno = EINVAL;
1181                         }
1182                         PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1183                 } else {
1184                         ret = munmap(sock_info->wait_shm_mmap, page_size);
1185                         if (ret) {
1186                                 ERR("Error unmapping wait shm");
1187                         }
1188                 }
1189                 sock_info->wait_shm_mmap = NULL;
1190         }
1191 }
1192
1193 /*
1194  * Using fork to set umask in the child process (not multi-thread safe).
1195  * We deal with the shm_open vs ftruncate race (happening when the
1196  * sessiond owns the shm and does not let everybody modify it, to ensure
1197  * safety against shm_unlink) by simply letting the mmap fail and
1198  * retrying after a few seconds.
1199  * For global shm, everybody has rw access to it until the sessiond
1200  * starts.
1201  */
1202 static
1203 int get_wait_shm(struct sock_info *sock_info, size_t mmap_size)
1204 {
1205         int wait_shm_fd, ret;
1206         pid_t pid;
1207
1208         /*
1209          * Try to open read-only.
1210          */
1211         wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1212         if (wait_shm_fd >= 0) {
1213                 int32_t tmp_read;
1214                 ssize_t len;
1215                 size_t bytes_read = 0;
1216
1217                 /*
1218                  * Try to read the fd. If unable to do so, try opening
1219                  * it in write mode.
1220                  */
1221                 do {
1222                         len = read(wait_shm_fd,
1223                                 &((char *) &tmp_read)[bytes_read],
1224                                 sizeof(tmp_read) - bytes_read);
1225                         if (len > 0) {
1226                                 bytes_read += len;
1227                         }
1228                 } while ((len < 0 && errno == EINTR)
1229                         || (len > 0 && bytes_read < sizeof(tmp_read)));
1230                 if (bytes_read != sizeof(tmp_read)) {
1231                         ret = close(wait_shm_fd);
1232                         if (ret) {
1233                                 ERR("close wait_shm_fd");
1234                         }
1235                         goto open_write;
1236                 }
1237                 goto end;
1238         } else if (wait_shm_fd < 0 && errno != ENOENT) {
1239                 /*
1240                  * Real-only open did not work, and it's not because the
1241                  * entry was not present. It's a failure that prohibits
1242                  * using shm.
1243                  */
1244                 ERR("Error opening shm %s", sock_info->wait_shm_path);
1245                 goto end;
1246         }
1247
1248 open_write:
1249         /*
1250          * If the open failed because the file did not exist, or because
1251          * the file was not truncated yet, try creating it ourself.
1252          */
1253         URCU_TLS(lttng_ust_nest_count)++;
1254         pid = fork();
1255         URCU_TLS(lttng_ust_nest_count)--;
1256         if (pid > 0) {
1257                 int status;
1258
1259                 /*
1260                  * Parent: wait for child to return, in which case the
1261                  * shared memory map will have been created.
1262                  */
1263                 pid = wait(&status);
1264                 if (pid < 0 || !WIFEXITED(status) || WEXITSTATUS(status) != 0) {
1265                         wait_shm_fd = -1;
1266                         goto end;
1267                 }
1268                 /*
1269                  * Try to open read-only again after creation.
1270                  */
1271                 wait_shm_fd = shm_open(sock_info->wait_shm_path, O_RDONLY, 0);
1272                 if (wait_shm_fd < 0) {
1273                         /*
1274                          * Real-only open did not work. It's a failure
1275                          * that prohibits using shm.
1276                          */
1277                         ERR("Error opening shm %s", sock_info->wait_shm_path);
1278                         goto end;
1279                 }
1280                 goto end;
1281         } else if (pid == 0) {
1282                 int create_mode;
1283
1284                 /* Child */
1285                 create_mode = S_IRUSR | S_IWUSR | S_IRGRP;
1286                 if (sock_info->global)
1287                         create_mode |= S_IROTH | S_IWGRP | S_IWOTH;
1288                 /*
1289                  * We're alone in a child process, so we can modify the
1290                  * process-wide umask.
1291                  */
1292                 umask(~create_mode);
1293                 /*
1294                  * Try creating shm (or get rw access).
1295                  * We don't do an exclusive open, because we allow other
1296                  * processes to create+ftruncate it concurrently.
1297                  */
1298                 wait_shm_fd = shm_open(sock_info->wait_shm_path,
1299                                 O_RDWR | O_CREAT, create_mode);
1300                 if (wait_shm_fd >= 0) {
1301                         ret = ftruncate(wait_shm_fd, mmap_size);
1302                         if (ret) {
1303                                 PERROR("ftruncate");
1304                                 _exit(EXIT_FAILURE);
1305                         }
1306                         _exit(EXIT_SUCCESS);
1307                 }
1308                 /*
1309                  * For local shm, we need to have rw access to accept
1310                  * opening it: this means the local sessiond will be
1311                  * able to wake us up. For global shm, we open it even
1312                  * if rw access is not granted, because the root.root
1313                  * sessiond will be able to override all rights and wake
1314                  * us up.
1315                  */
1316                 if (!sock_info->global && errno != EACCES) {
1317                         ERR("Error opening shm %s", sock_info->wait_shm_path);
1318                         _exit(EXIT_FAILURE);
1319                 }
1320                 /*
1321                  * The shm exists, but we cannot open it RW. Report
1322                  * success.
1323                  */
1324                 _exit(EXIT_SUCCESS);
1325         } else {
1326                 return -1;
1327         }
1328 end:
1329         if (wait_shm_fd >= 0 && !sock_info->global) {
1330                 struct stat statbuf;
1331
1332                 /*
1333                  * Ensure that our user is the owner of the shm file for
1334                  * local shm. If we do not own the file, it means our
1335                  * sessiond will not have access to wake us up (there is
1336                  * probably a rogue process trying to fake our
1337                  * sessiond). Fallback to polling method in this case.
1338                  */
1339                 ret = fstat(wait_shm_fd, &statbuf);
1340                 if (ret) {
1341                         PERROR("fstat");
1342                         goto error_close;
1343                 }
1344                 if (statbuf.st_uid != getuid())
1345                         goto error_close;
1346         }
1347         return wait_shm_fd;
1348
1349 error_close:
1350         ret = close(wait_shm_fd);
1351         if (ret) {
1352                 PERROR("Error closing fd");
1353         }
1354         return -1;
1355 }
1356
1357 static
1358 char *get_map_shm(struct sock_info *sock_info)
1359 {
1360         long page_size;
1361         int wait_shm_fd, ret;
1362         char *wait_shm_mmap;
1363
1364         page_size = sysconf(_SC_PAGE_SIZE);
1365         if (page_size <= 0) {
1366                 if (!page_size) {
1367                         errno = EINVAL;
1368                 }
1369                 PERROR("Error in sysconf(_SC_PAGE_SIZE)");
1370                 goto error;
1371         }
1372
1373         lttng_ust_lock_fd_tracker();
1374         wait_shm_fd = get_wait_shm(sock_info, page_size);
1375         if (wait_shm_fd < 0) {
1376                 lttng_ust_unlock_fd_tracker();
1377                 goto error;
1378         }
1379
1380         ret = lttng_ust_add_fd_to_tracker(wait_shm_fd);
1381         if (ret < 0) {
1382                 ret = close(wait_shm_fd);
1383                 if (!ret) {
1384                         PERROR("Error closing fd");
1385                 }
1386                 lttng_ust_unlock_fd_tracker();
1387                 goto error;
1388         }
1389
1390         wait_shm_fd = ret;
1391         lttng_ust_unlock_fd_tracker();
1392
1393         wait_shm_mmap = mmap(NULL, page_size, PROT_READ,
1394                   MAP_SHARED, wait_shm_fd, 0);
1395
1396         /* close shm fd immediately after taking the mmap reference */
1397         lttng_ust_lock_fd_tracker();
1398         ret = close(wait_shm_fd);
1399         if (!ret) {
1400                 lttng_ust_delete_fd_from_tracker(wait_shm_fd);
1401         } else {
1402                 PERROR("Error closing fd");
1403         }
1404         lttng_ust_unlock_fd_tracker();
1405
1406         if (wait_shm_mmap == MAP_FAILED) {
1407                 DBG("mmap error (can be caused by race with sessiond). Fallback to poll mode.");
1408                 goto error;
1409         }
1410         return wait_shm_mmap;
1411
1412 error:
1413         return NULL;
1414 }
1415
1416 static
1417 void wait_for_sessiond(struct sock_info *sock_info)
1418 {
1419         /* Use ust_lock to check if we should quit. */
1420         if (ust_lock()) {
1421                 goto quit;
1422         }
1423         if (wait_poll_fallback) {
1424                 goto error;
1425         }
1426         ust_unlock();
1427
1428         assert(sock_info->wait_shm_mmap);
1429
1430         DBG("Waiting for %s apps sessiond", sock_info->name);
1431         /* Wait for futex wakeup */
1432         if (uatomic_read((int32_t *) sock_info->wait_shm_mmap))
1433                 goto end_wait;
1434
1435         while (futex_async((int32_t *) sock_info->wait_shm_mmap,
1436                         FUTEX_WAIT, 0, NULL, NULL, 0)) {
1437                 switch (errno) {
1438                 case EWOULDBLOCK:
1439                         /* Value already changed. */
1440                         goto end_wait;
1441                 case EINTR:
1442                         /* Retry if interrupted by signal. */
1443                         break;  /* Get out of switch. */
1444                 case EFAULT:
1445                         wait_poll_fallback = 1;
1446                         DBG(
1447 "Linux kernels 2.6.33 to 3.0 (with the exception of stable versions) "
1448 "do not support FUTEX_WAKE on read-only memory mappings correctly. "
1449 "Please upgrade your kernel "
1450 "(fix is commit 9ea71503a8ed9184d2d0b8ccc4d269d05f7940ae in Linux kernel "
1451 "mainline). LTTng-UST will use polling mode fallback.");
1452                         if (ust_debug())
1453                                 PERROR("futex");
1454                         goto end_wait;
1455                 }
1456         }
1457 end_wait:
1458         return;
1459
1460 quit:
1461         ust_unlock();
1462         return;
1463
1464 error:
1465         ust_unlock();
1466         return;
1467 }
1468
1469 /*
1470  * This thread does not allocate any resource, except within
1471  * handle_message, within mutex protection. This mutex protects against
1472  * fork and exit.
1473  * The other moment it allocates resources is at socket connection, which
1474  * is also protected by the mutex.
1475  */
1476 static
1477 void *ust_listener_thread(void *arg)
1478 {
1479         struct sock_info *sock_info = arg;
1480         int sock, ret, prev_connect_failed = 0, has_waited = 0, fd;
1481         long timeout;
1482
1483         lttng_ust_fixup_tls();
1484         /*
1485          * If available, add '-ust' to the end of this thread's
1486          * process name
1487          */
1488         ret = lttng_ust_setustprocname();
1489         if (ret) {
1490                 ERR("Unable to set UST process name");
1491         }
1492
1493         /* Restart trying to connect to the session daemon */
1494 restart:
1495         if (prev_connect_failed) {
1496                 /* Wait for sessiond availability with pipe */
1497                 wait_for_sessiond(sock_info);
1498                 if (has_waited) {
1499                         has_waited = 0;
1500                         /*
1501                          * Sleep for 5 seconds before retrying after a
1502                          * sequence of failure / wait / failure. This
1503                          * deals with a killed or broken session daemon.
1504                          */
1505                         sleep(5);
1506                 } else {
1507                         has_waited = 1;
1508                 }
1509                 prev_connect_failed = 0;
1510         }
1511
1512         if (ust_lock()) {
1513                 goto quit;
1514         }
1515
1516         if (sock_info->socket != -1) {
1517                 /* FD tracker is updated by ustcomm_close_unix_sock() */
1518                 ret = ustcomm_close_unix_sock(sock_info->socket);
1519                 if (ret) {
1520                         ERR("Error closing %s ust cmd socket",
1521                                 sock_info->name);
1522                 }
1523                 sock_info->socket = -1;
1524         }
1525         if (sock_info->notify_socket != -1) {
1526                 /* FD tracker is updated by ustcomm_close_unix_sock() */
1527                 ret = ustcomm_close_unix_sock(sock_info->notify_socket);
1528                 if (ret) {
1529                         ERR("Error closing %s ust notify socket",
1530                                 sock_info->name);
1531                 }
1532                 sock_info->notify_socket = -1;
1533         }
1534
1535
1536         /*
1537          * Register. We need to perform both connect and sending
1538          * registration message before doing the next connect otherwise
1539          * we may reach unix socket connect queue max limits and block
1540          * on the 2nd connect while the session daemon is awaiting the
1541          * first connect registration message.
1542          */
1543         /* Connect cmd socket */
1544         lttng_ust_lock_fd_tracker();
1545         ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1546                 get_connect_sock_timeout());
1547         if (ret < 0) {
1548                 lttng_ust_unlock_fd_tracker();
1549                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1550                 prev_connect_failed = 1;
1551
1552                 /*
1553                  * If we cannot find the sessiond daemon, don't delay
1554                  * constructor execution.
1555                  */
1556                 ret = handle_register_failed(sock_info);
1557                 assert(!ret);
1558                 ust_unlock();
1559                 goto restart;
1560         }
1561         fd = ret;
1562         ret = lttng_ust_add_fd_to_tracker(fd);
1563         if (ret < 0) {
1564                 ret = close(fd);
1565                 if (ret) {
1566                         PERROR("close on sock_info->socket");
1567                 }
1568                 ret = -1;
1569                 lttng_ust_unlock_fd_tracker();
1570                 ust_unlock();
1571                 goto quit;
1572         }
1573
1574         sock_info->socket = ret;
1575         lttng_ust_unlock_fd_tracker();
1576
1577         ust_unlock();
1578         /*
1579          * Unlock/relock ust lock because connect is blocking (with
1580          * timeout). Don't delay constructors on the ust lock for too
1581          * long.
1582          */
1583         if (ust_lock()) {
1584                 goto quit;
1585         }
1586
1587         /*
1588          * Create only one root handle per listener thread for the whole
1589          * process lifetime, so we ensure we get ID which is statically
1590          * assigned to the root handle.
1591          */
1592         if (sock_info->root_handle == -1) {
1593                 ret = lttng_abi_create_root_handle();
1594                 if (ret < 0) {
1595                         ERR("Error creating root handle");
1596                         goto quit;
1597                 }
1598                 sock_info->root_handle = ret;
1599         }
1600
1601         ret = register_to_sessiond(sock_info->socket, USTCTL_SOCKET_CMD);
1602         if (ret < 0) {
1603                 ERR("Error registering to %s ust cmd socket",
1604                         sock_info->name);
1605                 prev_connect_failed = 1;
1606                 /*
1607                  * If we cannot register to the sessiond daemon, don't
1608                  * delay constructor execution.
1609                  */
1610                 ret = handle_register_failed(sock_info);
1611                 assert(!ret);
1612                 ust_unlock();
1613                 goto restart;
1614         }
1615
1616         ust_unlock();
1617         /*
1618          * Unlock/relock ust lock because connect is blocking (with
1619          * timeout). Don't delay constructors on the ust lock for too
1620          * long.
1621          */
1622         if (ust_lock()) {
1623                 goto quit;
1624         }
1625
1626         /* Connect notify socket */
1627         lttng_ust_lock_fd_tracker();
1628         ret = ustcomm_connect_unix_sock(sock_info->sock_path,
1629                 get_connect_sock_timeout());
1630         if (ret < 0) {
1631                 lttng_ust_unlock_fd_tracker();
1632                 DBG("Info: sessiond not accepting connections to %s apps socket", sock_info->name);
1633                 prev_connect_failed = 1;
1634
1635                 /*
1636                  * If we cannot find the sessiond daemon, don't delay
1637                  * constructor execution.
1638                  */
1639                 ret = handle_register_failed(sock_info);
1640                 assert(!ret);
1641                 ust_unlock();
1642                 goto restart;
1643         }
1644
1645         fd = ret;
1646         ret = lttng_ust_add_fd_to_tracker(fd);
1647         if (ret < 0) {
1648                 ret = close(fd);
1649                 if (ret) {
1650                         PERROR("close on sock_info->notify_socket");
1651                 }
1652                 ret = -1;
1653                 lttng_ust_unlock_fd_tracker();
1654                 ust_unlock();
1655                 goto quit;
1656         }
1657
1658         sock_info->notify_socket = ret;
1659         lttng_ust_unlock_fd_tracker();
1660
1661         ust_unlock();
1662         /*
1663          * Unlock/relock ust lock because connect is blocking (with
1664          * timeout). Don't delay constructors on the ust lock for too
1665          * long.
1666          */
1667         if (ust_lock()) {
1668                 goto quit;
1669         }
1670
1671         timeout = get_notify_sock_timeout();
1672         if (timeout >= 0) {
1673                 /*
1674                  * Give at least 10ms to sessiond to reply to
1675                  * notifications.
1676                  */
1677                 if (timeout < 10)
1678                         timeout = 10;
1679                 ret = ustcomm_setsockopt_rcv_timeout(sock_info->notify_socket,
1680                                 timeout);
1681                 if (ret < 0) {
1682                         WARN("Error setting socket receive timeout");
1683                 }
1684                 ret = ustcomm_setsockopt_snd_timeout(sock_info->notify_socket,
1685                                 timeout);
1686                 if (ret < 0) {
1687                         WARN("Error setting socket send timeout");
1688                 }
1689         } else if (timeout < -1) {
1690                 WARN("Unsupported timeout value %ld", timeout);
1691         }
1692
1693         ret = register_to_sessiond(sock_info->notify_socket,
1694                         USTCTL_SOCKET_NOTIFY);
1695         if (ret < 0) {
1696                 ERR("Error registering to %s ust notify socket",
1697                         sock_info->name);
1698                 prev_connect_failed = 1;
1699                 /*
1700                  * If we cannot register to the sessiond daemon, don't
1701                  * delay constructor execution.
1702                  */
1703                 ret = handle_register_failed(sock_info);
1704                 assert(!ret);
1705                 ust_unlock();
1706                 goto restart;
1707         }
1708         sock = sock_info->socket;
1709
1710         ust_unlock();
1711
1712         for (;;) {
1713                 ssize_t len;
1714                 struct ustcomm_ust_msg lum;
1715
1716                 len = ustcomm_recv_unix_sock(sock, &lum, sizeof(lum));
1717                 switch (len) {
1718                 case 0: /* orderly shutdown */
1719                         DBG("%s lttng-sessiond has performed an orderly shutdown", sock_info->name);
1720                         if (ust_lock()) {
1721                                 goto quit;
1722                         }
1723                         /*
1724                          * Either sessiond has shutdown or refused us by closing the socket.
1725                          * In either case, we don't want to delay construction execution,
1726                          * and we need to wait before retry.
1727                          */
1728                         prev_connect_failed = 1;
1729                         /*
1730                          * If we cannot register to the sessiond daemon, don't
1731                          * delay constructor execution.
1732                          */
1733                         ret = handle_register_failed(sock_info);
1734                         assert(!ret);
1735                         ust_unlock();
1736                         goto end;
1737                 case sizeof(lum):
1738                         print_cmd(lum.cmd, lum.handle);
1739                         ret = handle_message(sock_info, sock, &lum);
1740                         if (ret) {
1741                                 ERR("Error handling message for %s socket",
1742                                         sock_info->name);
1743                                 /*
1744                                  * Close socket if protocol error is
1745                                  * detected.
1746                                  */
1747                                 goto end;
1748                         }
1749                         continue;
1750                 default:
1751                         if (len < 0) {
1752                                 DBG("Receive failed from lttng-sessiond with errno %d", (int) -len);
1753                         } else {
1754                                 DBG("incorrect message size (%s socket): %zd", sock_info->name, len);
1755                         }
1756                         if (len == -ECONNRESET) {
1757                                 DBG("%s remote end closed connection", sock_info->name);
1758                                 goto end;
1759                         }
1760                         goto end;
1761                 }
1762
1763         }
1764 end:
1765         if (ust_lock()) {
1766                 goto quit;
1767         }
1768         /* Cleanup socket handles before trying to reconnect */
1769         lttng_ust_objd_table_owner_cleanup(sock_info);
1770         ust_unlock();
1771         goto restart;   /* try to reconnect */
1772
1773 quit:
1774         ust_unlock();
1775
1776         pthread_mutex_lock(&ust_exit_mutex);
1777         sock_info->thread_active = 0;
1778         pthread_mutex_unlock(&ust_exit_mutex);
1779         return NULL;
1780 }
1781
1782 /*
1783  * Weak symbol to call when the ust malloc wrapper is not loaded.
1784  */
1785 __attribute__((weak))
1786 void lttng_ust_malloc_wrapper_init(void)
1787 {
1788 }
1789
1790 /*
1791  * sessiond monitoring thread: monitor presence of global and per-user
1792  * sessiond by polling the application common named pipe.
1793  */
1794 void __attribute__((constructor)) lttng_ust_init(void)
1795 {
1796         struct timespec constructor_timeout;
1797         sigset_t sig_all_blocked, orig_parent_mask;
1798         pthread_attr_t thread_attr;
1799         int timeout_mode;
1800         int ret;
1801         void *handle;
1802
1803         if (uatomic_xchg(&initialized, 1) == 1)
1804                 return;
1805
1806         /*
1807          * Fixup interdependency between TLS fixup mutex (which happens
1808          * to be the dynamic linker mutex) and ust_lock, taken within
1809          * the ust lock.
1810          */
1811         lttng_ust_fixup_tls();
1812
1813         lttng_ust_loaded = 1;
1814
1815         /*
1816          * We need to ensure that the liblttng-ust library is not unloaded to avoid
1817          * the unloading of code used by the ust_listener_threads as we can not
1818          * reliably know when they exited. To do that, manually load
1819          * liblttng-ust.so to increment the dynamic loader's internal refcount for
1820          * this library so it never becomes zero, thus never gets unloaded from the
1821          * address space of the process. Since we are already running in the
1822          * constructor of the LTTNG_UST_LIB_SO_NAME library, calling dlopen will
1823          * simply increment the refcount and no additionnal work is needed by the
1824          * dynamic loader as the shared library is already loaded in the address
1825          * space. As a safe guard, we use the RTLD_NODELETE flag to prevent
1826          * unloading of the UST library if its refcount becomes zero (which should
1827          * never happen). Do the return value check but discard the handle at the
1828          * end of the function as it's not needed.
1829          */
1830         handle = dlopen(LTTNG_UST_LIB_SO_NAME, RTLD_LAZY | RTLD_NODELETE);
1831         if (!handle) {
1832                 ERR("dlopen of liblttng-ust shared library (%s).", LTTNG_UST_LIB_SO_NAME);
1833         }
1834
1835         /*
1836          * We want precise control over the order in which we construct
1837          * our sub-libraries vs starting to receive commands from
1838          * sessiond (otherwise leading to errors when trying to create
1839          * sessiond before the init functions are completed).
1840          */
1841         init_usterr();
1842         lttng_ust_getenv_init();        /* Needs init_usterr() to be completed. */
1843         init_tracepoint();
1844         lttng_ust_init_fd_tracker();
1845         lttng_ust_clock_init();
1846         lttng_ust_getcpu_init();
1847         lttng_ust_statedump_init();
1848         lttng_ring_buffer_metadata_client_init();
1849         lttng_ring_buffer_client_overwrite_init();
1850         lttng_ring_buffer_client_overwrite_rt_init();
1851         lttng_ring_buffer_client_discard_init();
1852         lttng_ring_buffer_client_discard_rt_init();
1853         lttng_perf_counter_init();
1854         /*
1855          * Invoke ust malloc wrapper init before starting other threads.
1856          */
1857         lttng_ust_malloc_wrapper_init();
1858
1859         timeout_mode = get_constructor_timeout(&constructor_timeout);
1860
1861         get_allow_blocking();
1862
1863         ret = sem_init(&constructor_wait, 0, 0);
1864         if (ret) {
1865                 PERROR("sem_init");
1866         }
1867
1868         ret = setup_global_apps();
1869         if (ret) {
1870                 assert(global_apps.allowed == 0);
1871                 DBG("global apps setup returned %d", ret);
1872         }
1873
1874         ret = setup_local_apps();
1875         if (ret) {
1876                 assert(local_apps.allowed == 0);
1877                 DBG("local apps setup returned %d", ret);
1878         }
1879
1880         /* A new thread created by pthread_create inherits the signal mask
1881          * from the parent. To avoid any signal being received by the
1882          * listener thread, we block all signals temporarily in the parent,
1883          * while we create the listener thread.
1884          */
1885         sigfillset(&sig_all_blocked);
1886         ret = pthread_sigmask(SIG_SETMASK, &sig_all_blocked, &orig_parent_mask);
1887         if (ret) {
1888                 ERR("pthread_sigmask: %s", strerror(ret));
1889         }
1890
1891         ret = pthread_attr_init(&thread_attr);
1892         if (ret) {
1893                 ERR("pthread_attr_init: %s", strerror(ret));
1894         }
1895         ret = pthread_attr_setdetachstate(&thread_attr, PTHREAD_CREATE_DETACHED);
1896         if (ret) {
1897                 ERR("pthread_attr_setdetachstate: %s", strerror(ret));
1898         }
1899
1900         if (global_apps.allowed) {
1901                 pthread_mutex_lock(&ust_exit_mutex);
1902                 ret = pthread_create(&global_apps.ust_listener, &thread_attr,
1903                                 ust_listener_thread, &global_apps);
1904                 if (ret) {
1905                         ERR("pthread_create global: %s", strerror(ret));
1906                 }
1907                 global_apps.thread_active = 1;
1908                 pthread_mutex_unlock(&ust_exit_mutex);
1909         } else {
1910                 handle_register_done(&global_apps);
1911         }
1912
1913         if (local_apps.allowed) {
1914                 pthread_mutex_lock(&ust_exit_mutex);
1915                 ret = pthread_create(&local_apps.ust_listener, &thread_attr,
1916                                 ust_listener_thread, &local_apps);
1917                 if (ret) {
1918                         ERR("pthread_create local: %s", strerror(ret));
1919                 }
1920                 local_apps.thread_active = 1;
1921                 pthread_mutex_unlock(&ust_exit_mutex);
1922         } else {
1923                 handle_register_done(&local_apps);
1924         }
1925         ret = pthread_attr_destroy(&thread_attr);
1926         if (ret) {
1927                 ERR("pthread_attr_destroy: %s", strerror(ret));
1928         }
1929
1930         /* Restore original signal mask in parent */
1931         ret = pthread_sigmask(SIG_SETMASK, &orig_parent_mask, NULL);
1932         if (ret) {
1933                 ERR("pthread_sigmask: %s", strerror(ret));
1934         }
1935
1936         switch (timeout_mode) {
1937         case 1: /* timeout wait */
1938                 do {
1939                         ret = sem_timedwait(&constructor_wait,
1940                                         &constructor_timeout);
1941                 } while (ret < 0 && errno == EINTR);
1942                 if (ret < 0) {
1943                         switch (errno) {
1944                         case ETIMEDOUT:
1945                                 ERR("Timed out waiting for lttng-sessiond");
1946                                 break;
1947                         case EINVAL:
1948                                 PERROR("sem_timedwait");
1949                                 break;
1950                         default:
1951                                 ERR("Unexpected error \"%s\" returned by sem_timedwait",
1952                                         strerror(errno));
1953                         }
1954                 }
1955                 break;
1956         case -1:/* wait forever */
1957                 do {
1958                         ret = sem_wait(&constructor_wait);
1959                 } while (ret < 0 && errno == EINTR);
1960                 if (ret < 0) {
1961                         switch (errno) {
1962                         case EINVAL:
1963                                 PERROR("sem_wait");
1964                                 break;
1965                         default:
1966                                 ERR("Unexpected error \"%s\" returned by sem_wait",
1967                                         strerror(errno));
1968                         }
1969                 }
1970                 break;
1971         case 0: /* no timeout */
1972                 break;
1973         }
1974 }
1975
1976 static
1977 void lttng_ust_cleanup(int exiting)
1978 {
1979         cleanup_sock_info(&global_apps, exiting);
1980         cleanup_sock_info(&local_apps, exiting);
1981         local_apps.allowed = 0;
1982         global_apps.allowed = 0;
1983         /*
1984          * The teardown in this function all affect data structures
1985          * accessed under the UST lock by the listener thread. This
1986          * lock, along with the lttng_ust_comm_should_quit flag, ensure
1987          * that none of these threads are accessing this data at this
1988          * point.
1989          */
1990         lttng_ust_abi_exit();
1991         lttng_ust_events_exit();
1992         lttng_perf_counter_exit();
1993         lttng_ring_buffer_client_discard_rt_exit();
1994         lttng_ring_buffer_client_discard_exit();
1995         lttng_ring_buffer_client_overwrite_rt_exit();
1996         lttng_ring_buffer_client_overwrite_exit();
1997         lttng_ring_buffer_metadata_client_exit();
1998         lttng_ust_statedump_destroy();
1999         exit_tracepoint();
2000         if (!exiting) {
2001                 /* Reinitialize values for fork */
2002                 sem_count = sem_count_initial_value;
2003                 lttng_ust_comm_should_quit = 0;
2004                 initialized = 0;
2005         }
2006 }
2007
2008 void __attribute__((destructor)) lttng_ust_exit(void)
2009 {
2010         int ret;
2011
2012         /*
2013          * Using pthread_cancel here because:
2014          * A) we don't want to hang application teardown.
2015          * B) the thread is not allocating any resource.
2016          */
2017
2018         /*
2019          * Require the communication thread to quit. Synchronize with
2020          * mutexes to ensure it is not in a mutex critical section when
2021          * pthread_cancel is later called.
2022          */
2023         ust_lock_nocheck();
2024         lttng_ust_comm_should_quit = 1;
2025         ust_unlock();
2026
2027         pthread_mutex_lock(&ust_exit_mutex);
2028         /* cancel threads */
2029         if (global_apps.thread_active) {
2030                 ret = pthread_cancel(global_apps.ust_listener);
2031                 if (ret) {
2032                         ERR("Error cancelling global ust listener thread: %s",
2033                                 strerror(ret));
2034                 } else {
2035                         global_apps.thread_active = 0;
2036                 }
2037         }
2038         if (local_apps.thread_active) {
2039                 ret = pthread_cancel(local_apps.ust_listener);
2040                 if (ret) {
2041                         ERR("Error cancelling local ust listener thread: %s",
2042                                 strerror(ret));
2043                 } else {
2044                         local_apps.thread_active = 0;
2045                 }
2046         }
2047         pthread_mutex_unlock(&ust_exit_mutex);
2048
2049         /*
2050          * Do NOT join threads: use of sys_futex makes it impossible to
2051          * join the threads without using async-cancel, but async-cancel
2052          * is delivered by a signal, which could hit the target thread
2053          * anywhere in its code path, including while the ust_lock() is
2054          * held, causing a deadlock for the other thread. Let the OS
2055          * cleanup the threads if there are stalled in a syscall.
2056          */
2057         lttng_ust_cleanup(1);
2058 }
2059
2060 static
2061 void ust_context_ns_reset(void)
2062 {
2063         lttng_context_pid_ns_reset();
2064         lttng_context_cgroup_ns_reset();
2065         lttng_context_ipc_ns_reset();
2066         lttng_context_mnt_ns_reset();
2067         lttng_context_net_ns_reset();
2068         lttng_context_user_ns_reset();
2069         lttng_context_time_ns_reset();
2070         lttng_context_uts_ns_reset();
2071 }
2072
2073 static
2074 void ust_context_vuids_reset(void)
2075 {
2076         lttng_context_vuid_reset();
2077         lttng_context_veuid_reset();
2078         lttng_context_vsuid_reset();
2079 }
2080
2081 static
2082 void ust_context_vgids_reset(void)
2083 {
2084         lttng_context_vgid_reset();
2085         lttng_context_vegid_reset();
2086         lttng_context_vsgid_reset();
2087 }
2088
2089 /*
2090  * We exclude the worker threads across fork and clone (except
2091  * CLONE_VM), because these system calls only keep the forking thread
2092  * running in the child.  Therefore, we don't want to call fork or clone
2093  * in the middle of an tracepoint or ust tracing state modification.
2094  * Holding this mutex protects these structures across fork and clone.
2095  */
2096 void ust_before_fork(sigset_t *save_sigset)
2097 {
2098         /*
2099          * Disable signals. This is to avoid that the child intervenes
2100          * before it is properly setup for tracing. It is safer to
2101          * disable all signals, because then we know we are not breaking
2102          * anything by restoring the original mask.
2103          */
2104         sigset_t all_sigs;
2105         int ret;
2106
2107         /* Fixup lttng-ust TLS. */
2108         lttng_ust_fixup_tls();
2109
2110         if (URCU_TLS(lttng_ust_nest_count))
2111                 return;
2112         /* Disable signals */
2113         sigfillset(&all_sigs);
2114         ret = sigprocmask(SIG_BLOCK, &all_sigs, save_sigset);
2115         if (ret == -1) {
2116                 PERROR("sigprocmask");
2117         }
2118
2119         pthread_mutex_lock(&ust_fork_mutex);
2120
2121         ust_lock_nocheck();
2122         urcu_bp_before_fork();
2123         lttng_ust_lock_fd_tracker();
2124         lttng_perf_lock();
2125 }
2126
2127 static void ust_after_fork_common(sigset_t *restore_sigset)
2128 {
2129         int ret;
2130
2131         DBG("process %d", getpid());
2132         lttng_perf_unlock();
2133         lttng_ust_unlock_fd_tracker();
2134         ust_unlock();
2135
2136         pthread_mutex_unlock(&ust_fork_mutex);
2137
2138         /* Restore signals */
2139         ret = sigprocmask(SIG_SETMASK, restore_sigset, NULL);
2140         if (ret == -1) {
2141                 PERROR("sigprocmask");
2142         }
2143 }
2144
2145 void ust_after_fork_parent(sigset_t *restore_sigset)
2146 {
2147         if (URCU_TLS(lttng_ust_nest_count))
2148                 return;
2149         DBG("process %d", getpid());
2150         urcu_bp_after_fork_parent();
2151         /* Release mutexes and reenable signals */
2152         ust_after_fork_common(restore_sigset);
2153 }
2154
2155 /*
2156  * After fork, in the child, we need to cleanup all the leftover state,
2157  * except the worker thread which already magically disappeared thanks
2158  * to the weird Linux fork semantics. After tyding up, we call
2159  * lttng_ust_init() again to start over as a new PID.
2160  *
2161  * This is meant for forks() that have tracing in the child between the
2162  * fork and following exec call (if there is any).
2163  */
2164 void ust_after_fork_child(sigset_t *restore_sigset)
2165 {
2166         if (URCU_TLS(lttng_ust_nest_count))
2167                 return;
2168         lttng_context_vpid_reset();
2169         lttng_context_vtid_reset();
2170         lttng_context_procname_reset();
2171         ust_context_ns_reset();
2172         ust_context_vuids_reset();
2173         ust_context_vgids_reset();
2174         DBG("process %d", getpid());
2175         /* Release urcu mutexes */
2176         urcu_bp_after_fork_child();
2177         lttng_ust_cleanup(0);
2178         /* Release mutexes and reenable signals */
2179         ust_after_fork_common(restore_sigset);
2180         lttng_ust_init();
2181 }
2182
2183 void ust_after_setns(void)
2184 {
2185         ust_context_ns_reset();
2186         ust_context_vuids_reset();
2187         ust_context_vgids_reset();
2188 }
2189
2190 void ust_after_unshare(void)
2191 {
2192         ust_context_ns_reset();
2193         ust_context_vuids_reset();
2194         ust_context_vgids_reset();
2195 }
2196
2197 void ust_after_setuid(void)
2198 {
2199         ust_context_vuids_reset();
2200 }
2201
2202 void ust_after_seteuid(void)
2203 {
2204         ust_context_vuids_reset();
2205 }
2206
2207 void ust_after_setreuid(void)
2208 {
2209         ust_context_vuids_reset();
2210 }
2211
2212 void ust_after_setresuid(void)
2213 {
2214         ust_context_vuids_reset();
2215 }
2216
2217 void ust_after_setgid(void)
2218 {
2219         ust_context_vgids_reset();
2220 }
2221
2222 void ust_after_setegid(void)
2223 {
2224         ust_context_vgids_reset();
2225 }
2226
2227 void ust_after_setregid(void)
2228 {
2229         ust_context_vgids_reset();
2230 }
2231
2232 void ust_after_setresgid(void)
2233 {
2234         ust_context_vgids_reset();
2235 }
2236
2237 void lttng_ust_sockinfo_session_enabled(void *owner)
2238 {
2239         struct sock_info *sock_info = owner;
2240         sock_info->statedump_pending = 1;
2241 }