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