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