libust/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 #include <sys/types.h>
  23 #include <sys/socket.h>
  24 #include <sys/prctl.h>
  25 #include <unistd.h>
  26 #include <errno.h>
  27 #include <pthread.h>
  28 #include <semaphore.h>
  29 #include <time.h>
  30 #include <assert.h>
  31 #include <urcu/uatomic.h>
  32
  33 #include <lttng-ust-comm.h>
  34 #include <ust/usterr-signal-safe.h>
  35 #include <ust/lttng-ust-abi.h>
  36 #include <ust/tracepoint.h>
  37
  38 /*
  39  * Has lttng ust comm constructor been called ?
  40  */
  41 static int initialized;
  42
  43 /*
  44  * communication thread mutex. Held when handling a command, also held
  45  * by fork() to deal with removal of threads, and by exit path.
  46  */
  47 static pthread_mutex_t lttng_ust_comm_mutex = PTHREAD_MUTEX_INITIALIZER;
  48
  49 /* Should the ust comm thread quit ? */
  50 static int lttng_ust_comm_should_quit;
  51
  52 /*
  53  * Wait for either of these before continuing to the main
  54  * program:
  55  * - the register_done message from sessiond daemon
  56  *   (will let the sessiond daemon enable sessions before main
  57  *   starts.)
  58  * - sessiond daemon is not reachable.
  59  * - timeout (ensuring applications are resilient to session
  60  *   daemon problems).
  61  */
  62 static sem_t constructor_wait;
  63 /*
  64  * Doing this for both the global and local sessiond.
  65  */
  66 static int sem_count = { 2 };
  67
  68 /*
  69  * Info about socket and associated listener thread.
  70  */
  71 struct sock_info {
  72         const char *name;
  73         char sock_path[PATH_MAX];
  74         int socket;
  75         pthread_t ust_listener; /* listener thread */
  76         int root_handle;
  77         int constructor_sem_posted;;
  78 };
  79
  80 /* Socket from app (connect) to session daemon (listen) for communication */
  81 struct sock_info global_apps = {
  82         .name = "global",
  83         .sock_path = DEFAULT_GLOBAL_APPS_UNIX_SOCK,
  84         .socket = -1,
  85         .root_handle = -1,
  86 };
  87
  88 /* TODO: allow global_apps_sock_path override */
  89
  90 struct sock_info local_apps = {
  91         .name = "local",
  92         .socket = -1,
  93         .root_handle = -1,
  94 };
  95
  96 extern void ltt_ring_buffer_client_overwrite_init(void);
  97 extern void ltt_ring_buffer_client_discard_init(void);
  98 extern void ltt_ring_buffer_metadata_client_init(void);
  99 extern void ltt_ring_buffer_client_overwrite_exit(void);
 100 extern void ltt_ring_buffer_client_discard_exit(void);
 101 extern void ltt_ring_buffer_metadata_client_exit(void);
 102
 103 static
 104 int setup_local_apps_socket(void)
 105 {
 106         const char *home_dir;
 107
 108         home_dir = (const char *) getenv("HOME");
 109         if (!home_dir)
 110                 return -ENOENT;
 111         snprintf(local_apps.sock_path, PATH_MAX,
 112                  DEFAULT_HOME_APPS_UNIX_SOCK, home_dir);
 113         return 0;
 114 }
 115
 116 static
 117 int register_app_to_sessiond(int socket)
 118 {
 119         ssize_t ret;
 120         int prctl_ret;
 121         struct {
 122                 uint32_t major;
 123                 uint32_t minor;
 124                 pid_t pid;
 125                 uid_t uid;
 126                 char name[16];  /* process name */
 127         } reg_msg;
 128
 129         reg_msg.major = LTTNG_UST_COMM_VERSION_MAJOR;
 130         reg_msg.minor = LTTNG_UST_COMM_VERSION_MINOR;
 131         reg_msg.pid = getpid();
 132         reg_msg.uid = getuid();
 133         prctl_ret = prctl(PR_GET_NAME, (unsigned long) reg_msg.name, 0, 0, 0);
 134         if (prctl_ret) {
 135                 ERR("Error executing prctl");
 136                 return -errno;
 137         }
 138
 139         ret = lttcomm_send_unix_sock(socket, &reg_msg, sizeof(reg_msg));
 140         if (ret >= 0 && ret != sizeof(reg_msg))
 141                 return -EIO;
 142         return ret;
 143 }
 144
 145 static
 146 int send_reply(int sock, struct lttcomm_ust_reply *lur)
 147 {
 148         ssize_t len;
 149
 150         len = lttcomm_send_unix_sock(sock, lur, sizeof(*lur));
 151         switch (len) {
 152         case sizeof(*lur):
 153                 DBG("message successfully sent");
 154                 return 0;
 155         case -1:
 156                 if (errno == ECONNRESET) {
 157                         printf("remote end closed connection\n");
 158                         return 0;
 159                 }
 160                 return -1;
 161         default:
 162                 printf("incorrect message size: %zd\n", len);
 163                 return -1;
 164         }
 165 }
 166
 167 static
 168 int handle_register_done(struct sock_info *sock_info)
 169 {
 170         int ret;
 171
 172         if (sock_info->constructor_sem_posted)
 173                 return 0;
 174         sock_info->constructor_sem_posted = 1;
 175         ret = uatomic_add_return(&sem_count, -1);
 176         if (ret == 0) {
 177                 ret = sem_post(&constructor_wait);
 178                 assert(!ret);
 179         }
 180         return 0;
 181 }
 182
 183 static
 184 int handle_message(struct sock_info *sock_info,
 185                 int sock, struct lttcomm_ust_msg *lum)
 186 {
 187         int ret = 0;
 188         const struct objd_ops *ops;
 189         struct lttcomm_ust_reply lur;
 190
 191         pthread_mutex_lock(&lttng_ust_comm_mutex);
 192
 193         memset(&lur, 0, sizeof(lur));
 194
 195         if (lttng_ust_comm_should_quit) {
 196                 ret = -EPERM;
 197                 goto end;
 198         }
 199
 200         ops = objd_ops(lum->handle);
 201         if (!ops) {
 202                 ret = -ENOENT;
 203                 goto end;
 204         }
 205
 206         switch (lum->cmd) {
 207         case LTTNG_UST_REGISTER_DONE:
 208                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 209                         ret = handle_register_done(sock_info);
 210                 else
 211                         ret = -EINVAL;
 212                 break;
 213         case LTTNG_UST_RELEASE:
 214                 if (lum->handle == LTTNG_UST_ROOT_HANDLE)
 215                         ret = -EPERM;
 216                 else
 217                         ret = objd_unref(lum->handle);
 218                 break;
 219         default:
 220                 if (ops->cmd)
 221                         ret = ops->cmd(lum->handle, lum->cmd,
 222                                         (unsigned long) &lum->u);
 223                 else
 224                         ret = -ENOSYS;
 225                 break;
 226         }
 227
 228 end:
 229         lur.handle = lum->handle;
 230         lur.cmd = lum->cmd;
 231         lur.ret_val = ret;
 232         if (ret >= 0) {
 233                 lur.ret_code = LTTCOMM_OK;
 234         } else {
 235                 lur.ret_code = LTTCOMM_SESSION_FAIL;
 236         }
 237         ret = send_reply(sock, &lur);
 238
 239         pthread_mutex_unlock(&lttng_ust_comm_mutex);
 240         return ret;
 241 }
 242
 243 static
 244 void cleanup_sock_info(struct sock_info *sock_info)
 245 {
 246         int ret;
 247
 248         if (sock_info->socket != -1) {
 249                 ret = close(sock_info->socket);
 250                 if (ret) {
 251                         ERR("Error closing local apps socket");
 252                 }
 253                 sock_info->socket = -1;
 254         }
 255         if (sock_info->root_handle != -1) {
 256                 ret = objd_unref(sock_info->root_handle);
 257                 if (ret) {
 258                         ERR("Error unref root handle");
 259                 }
 260                 sock_info->root_handle = -1;
 261         }
 262 }
 263
 264 /*
 265  * This thread does not allocate any resource, except within
 266  * handle_message, within mutex protection. This mutex protects against
 267  * fork and exit.
 268  * The other moment it allocates resources is at socket connexion, which
 269  * is also protected by the mutex.
 270  */
 271 static
 272 void *ust_listener_thread(void *arg)
 273 {
 274         struct sock_info *sock_info = arg;
 275         int sock, ret;
 276
 277         /* Restart trying to connect to the session daemon */
 278 restart:
 279         pthread_mutex_lock(&lttng_ust_comm_mutex);
 280
 281         if (lttng_ust_comm_should_quit) {
 282                 pthread_mutex_unlock(&lttng_ust_comm_mutex);
 283                 goto quit;
 284         }
 285
 286         if (sock_info->socket != -1) {
 287                 ret = close(sock_info->socket);
 288                 if (ret) {
 289                         ERR("Error closing %s apps socket", sock_info->name);
 290                 }
 291                 sock_info->socket = -1;
 292         }
 293
 294         /* Check for sessiond availability with pipe TODO */
 295
 296         /* Register */
 297         ret = lttcomm_connect_unix_sock(sock_info->sock_path);
 298         if (ret < 0) {
 299                 ERR("Error connecting to %s apps socket", sock_info->name);
 300                 /*
 301                  * If we cannot find the sessiond daemon, don't delay
 302                  * constructor execution.
 303                  */
 304                 ret = handle_register_done(sock_info);
 305                 assert(!ret);
 306                 pthread_mutex_unlock(&lttng_ust_comm_mutex);
 307                 sleep(5);
 308                 goto restart;
 309         }
 310
 311         sock_info->socket = sock = ret;
 312
 313         /*
 314          * Create only one root handle per listener thread for the whole
 315          * process lifetime.
 316          */
 317         if (sock_info->root_handle == -1) {
 318                 ret = lttng_abi_create_root_handle();
 319                 if (ret) {
 320                         ERR("Error creating root handle");
 321                         pthread_mutex_unlock(&lttng_ust_comm_mutex);
 322                         goto quit;
 323                 }
 324                 sock_info->root_handle = ret;
 325         }
 326
 327         ret = register_app_to_sessiond(sock);
 328         if (ret < 0) {
 329                 ERR("Error registering to %s apps socket", sock_info->name);
 330                 /*
 331                  * If we cannot register to the sessiond daemon, don't
 332                  * delay constructor execution.
 333                  */
 334                 ret = handle_register_done(sock_info);
 335                 assert(!ret);
 336                 pthread_mutex_unlock(&lttng_ust_comm_mutex);
 337                 sleep(5);
 338                 goto restart;
 339         }
 340         pthread_mutex_unlock(&lttng_ust_comm_mutex);
 341
 342         for (;;) {
 343                 ssize_t len;
 344                 struct lttcomm_ust_msg lum;
 345
 346                 len = lttcomm_recv_unix_sock(sock, &lum, sizeof(lum));
 347                 switch (len) {
 348                 case 0: /* orderly shutdown */
 349                         DBG("%s ltt-sessiond has performed an orderly shutdown\n", sock_info->name);
 350                         goto end;
 351                 case sizeof(lum):
 352                         DBG("message received\n");
 353                         ret = handle_message(sock_info, sock, &lum);
 354                         if (ret < 0) {
 355                                 ERR("Error handling message for %s socket", sock_info->name);
 356                         }
 357                         continue;
 358                 case -1:
 359                         if (errno == ECONNRESET) {
 360                                 ERR("%s remote end closed connection\n", sock_info->name);
 361                                 goto end;
 362                         }
 363                         goto end;
 364                 default:
 365                         ERR("incorrect message size (%s socket): %zd\n", sock_info->name, len);
 366                         continue;
 367                 }
 368
 369         }
 370 end:
 371         goto restart;   /* try to reconnect */
 372 quit:
 373         return NULL;
 374 }
 375
 376 /*
 377  * Return values: -1: don't wait. 0: wait forever. 1: timeout wait.
 378  */
 379 static
 380 int get_timeout(struct timespec *constructor_timeout)
 381 {
 382         long constructor_delay_ms = LTTNG_UST_DEFAULT_CONSTRUCTOR_TIMEOUT_MS;
 383         char *str_delay;
 384         int ret;
 385
 386         str_delay = getenv("UST_REGISTER_TIMEOUT");
 387         if (str_delay) {
 388                 constructor_delay_ms = strtol(str_delay, NULL, 10);
 389         }
 390
 391         switch (constructor_delay_ms) {
 392         case -1:/* fall-through */
 393         case 0:
 394                 return constructor_delay_ms;
 395         default:
 396                 break;
 397         }
 398
 399         /*
 400          * If we are unable to find the current time, don't wait.
 401          */
 402         ret = clock_gettime(CLOCK_REALTIME, constructor_timeout);
 403         if (ret) {
 404                 return -1;
 405         }
 406         constructor_timeout->tv_sec += constructor_delay_ms / 1000UL;
 407         constructor_timeout->tv_nsec +=
 408                 (constructor_delay_ms % 1000UL) * 1000000UL;
 409         if (constructor_timeout->tv_nsec >= 1000000000UL) {
 410                 constructor_timeout->tv_sec++;
 411                 constructor_timeout->tv_nsec -= 1000000000UL;
 412         }
 413         return 1;
 414 }
 415
 416 /*
 417  * sessiond monitoring thread: monitor presence of global and per-user
 418  * sessiond by polling the application common named pipe.
 419  */
 420 /* TODO */
 421
 422 void __attribute__((constructor)) lttng_ust_init(void)
 423 {
 424         struct timespec constructor_timeout;
 425         int timeout_mode;
 426         int ret;
 427
 428         if (uatomic_xchg(&initialized, 1) == 1)
 429                 return;
 430
 431         /*
 432          * We want precise control over the order in which we construct
 433          * our sub-libraries vs starting to receive commands from
 434          * sessiond (otherwise leading to errors when trying to create
 435          * sessiond before the init functions are completed).
 436          */
 437         init_usterr();
 438         init_tracepoint();
 439         ltt_ring_buffer_metadata_client_init();
 440         ltt_ring_buffer_client_overwrite_init();
 441         ltt_ring_buffer_client_discard_init();
 442
 443         timeout_mode = get_timeout(&constructor_timeout);
 444
 445         ret = sem_init(&constructor_wait, 0, 0);
 446         assert(!ret);
 447
 448         ret = setup_local_apps_socket();
 449         if (ret) {
 450                 ERR("Error setting up to local apps socket");
 451         }
 452
 453         ret = pthread_create(&global_apps.ust_listener, NULL,
 454                         ust_listener_thread, &global_apps);
 455         ret = pthread_create(&local_apps.ust_listener, NULL,
 456                         ust_listener_thread, &local_apps);
 457
 458         switch (timeout_mode) {
 459         case 1: /* timeout wait */
 460                 do {
 461                         ret = sem_timedwait(&constructor_wait,
 462                                         &constructor_timeout);
 463                 } while (ret < 0 && errno == EINTR);
 464                 if (ret < 0 && errno == ETIMEDOUT) {
 465                         ERR("Timed out waiting for ltt-sessiond");
 466                 } else {
 467                         assert(!ret);
 468                 }
 469                 break;
 470         case -1:/* wait forever */
 471                 do {
 472                         ret = sem_wait(&constructor_wait);
 473                 } while (ret < 0 && errno == EINTR);
 474                 assert(!ret);
 475                 break;
 476         case 0: /* no timeout */
 477                 break;
 478         }
 479 }
 480
 481 void __attribute__((destructor)) lttng_ust_exit(void)
 482 {
 483         int ret;
 484
 485         /*
 486          * Using pthread_cancel here because:
 487          * A) we don't want to hang application teardown.
 488          * B) the thread is not allocating any resource.
 489          */
 490
 491         /*
 492          * Require the communication thread to quit. Synchronize with
 493          * mutexes to ensure it is not in a mutex critical section when
 494          * pthread_cancel is later called.
 495          */
 496         pthread_mutex_lock(&lttng_ust_comm_mutex);
 497         lttng_ust_comm_should_quit = 1;
 498         pthread_mutex_unlock(&lttng_ust_comm_mutex);
 499
 500 #if 0
 501         ret = pthread_cancel(global_apps.ust_listener);
 502         if (ret) {
 503                 ERR("Error cancelling global ust listener thread");
 504         }
 505 #endif //0
 506
 507         cleanup_sock_info(&global_apps);
 508
 509         ret = pthread_cancel(local_apps.ust_listener);
 510         if (ret) {
 511                 ERR("Error cancelling local ust listener thread");
 512         }
 513
 514         cleanup_sock_info(&local_apps);
 515
 516         lttng_ust_abi_exit();
 517         ltt_events_exit();
 518         ltt_ring_buffer_client_discard_exit();
 519         ltt_ring_buffer_client_overwrite_exit();
 520         ltt_ring_buffer_metadata_client_exit();
 521         exit_tracepoint();
 522 }