src/bin/lttng-consumerd/lttng-consumerd.c

   1 /*
   2  * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
   3  *                      Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
   4  *
   5  * This program is free software; you can redistribute it and/or modify
   6  * it under the terms of the GNU General Public License, version 2 only,
   7  * as published by the Free Software Foundation.
   8  *
   9  * This program is distributed in the hope that it will be useful,
  10  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  12  * GNU General Public License for more details.
  13  *
  14  * You should have received a copy of the GNU General Public License along
  15  * with this program; if not, write to the Free Software Foundation, Inc.,
  16  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
  17  */
  18
  19 #define _LGPL_SOURCE
  20 #include <fcntl.h>
  21 #include <getopt.h>
  22 #include <grp.h>
  23 #include <limits.h>
  24 #include <pthread.h>
  25 #include <signal.h>
  26 #include <stdio.h>
  27 #include <stdlib.h>
  28 #include <string.h>
  29 #include <sys/ipc.h>
  30 #include <sys/resource.h>
  31 #include <sys/shm.h>
  32 #include <sys/socket.h>
  33 #include <sys/stat.h>
  34 #include <sys/types.h>
  35 #include <urcu/list.h>
  36 #include <poll.h>
  37 #include <unistd.h>
  38 #include <sys/mman.h>
  39 #include <assert.h>
  40 #include <urcu/compiler.h>
  41 #include <ulimit.h>
  42
  43 #include <common/defaults.h>
  44 #include <common/common.h>
  45 #include <common/consumer/consumer.h>
  46 #include <common/consumer/consumer-timer.h>
  47 #include <common/compat/poll.h>
  48 #include <common/compat/getenv.h>
  49 #include <common/sessiond-comm/sessiond-comm.h>
  50 #include <common/utils.h>
  51
  52 #include "lttng-consumerd.h"
  53 #include "health-consumerd.h"
  54
  55 /* threads (channel handling, poll, metadata, sessiond) */
  56
  57 static pthread_t channel_thread, data_thread, metadata_thread,
  58                 sessiond_thread, metadata_timer_thread, health_thread;
  59
  60 /* to count the number of times the user pressed ctrl+c */
  61 static int sigintcount = 0;
  62
  63 /* Argument variables */
  64 int lttng_opt_quiet;    /* not static in error.h */
  65 int lttng_opt_verbose;  /* not static in error.h */
  66 int lttng_opt_mi;       /* not static in error.h */
  67
  68 static int opt_daemon;
  69 static const char *progname;
  70 static char command_sock_path[PATH_MAX]; /* Global command socket path */
  71 static char error_sock_path[PATH_MAX]; /* Global error path */
  72 static enum lttng_consumer_type opt_type = LTTNG_CONSUMER_KERNEL;
  73
  74 /* the liblttngconsumerd context */
  75 static struct lttng_consumer_local_data *ctx;
  76
  77 /* Consumerd health monitoring */
  78 struct health_app *health_consumerd;
  79
  80 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
  81
  82 int lttng_consumer_ready = NR_LTTNG_CONSUMER_READY;
  83
  84 enum lttng_consumer_type lttng_consumer_get_type(void)
  85 {
  86         if (!ctx) {
  87                 return LTTNG_CONSUMER_UNKNOWN;
  88         }
  89         return ctx->type;
  90 }
  91
  92 /*
  93  * Signal handler for the daemon
  94  */
  95 static void sighandler(int sig)
  96 {
  97         if (sig == SIGINT && sigintcount++ == 0) {
  98                 DBG("ignoring first SIGINT");
  99                 return;
 100         }
 101
 102         /*
 103          * Ignore SIGPIPE because it should not stop the consumer whenever a
 104          * SIGPIPE is caught through a FD operation.
 105          */
 106         if (sig == SIGPIPE) {
 107                 return;
 108         }
 109
 110         if (ctx) {
 111                 lttng_consumer_should_exit(ctx);
 112         }
 113 }
 114
 115 /*
 116  * Setup signal handler for :
 117  *      SIGINT, SIGTERM, SIGPIPE
 118  */
 119 static int set_signal_handler(void)
 120 {
 121         int ret = 0;
 122         struct sigaction sa;
 123         sigset_t sigset;
 124
 125         if ((ret = sigemptyset(&sigset)) < 0) {
 126                 PERROR("sigemptyset");
 127                 return ret;
 128         }
 129
 130         sa.sa_handler = sighandler;
 131         sa.sa_mask = sigset;
 132         sa.sa_flags = 0;
 133         if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
 134                 PERROR("sigaction");
 135                 return ret;
 136         }
 137
 138         if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
 139                 PERROR("sigaction");
 140                 return ret;
 141         }
 142
 143         if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
 144                 PERROR("sigaction");
 145                 return ret;
 146         }
 147
 148         return ret;
 149 }
 150
 151 /*
 152  * Usage function on stream file.
 153  */
 154 static void usage(FILE *fp)
 155 {
 156         fprintf(fp, "Usage: %s OPTIONS\n\nOptions:\n", progname);
 157         fprintf(fp, "  -h, --help                         "
 158                         "Display this usage.\n");
 159         fprintf(fp, "  -c, --consumerd-cmd-sock PATH      "
 160                         "Specify path for the command socket\n");
 161         fprintf(fp, "  -e, --consumerd-err-sock PATH      "
 162                         "Specify path for the error socket\n");
 163         fprintf(fp, "  -d, --daemonize                    "
 164                         "Start as a daemon.\n");
 165         fprintf(fp, "  -q, --quiet                        "
 166                         "No output at all.\n");
 167         fprintf(fp, "  -v, --verbose                      "
 168                         "Verbose mode. Activate DBG() macro.\n");
 169         fprintf(fp, "  -V, --version                      "
 170                         "Show version number.\n");
 171         fprintf(fp, "  -g, --group NAME                   "
 172                         "Specify the tracing group name. (default: tracing)\n");
 173         fprintf(fp, "  -k, --kernel                       "
 174                         "Consumer kernel buffers (default).\n");
 175         fprintf(fp, "  -u, --ust                          "
 176                         "Consumer UST buffers.%s\n",
 177 #ifdef HAVE_LIBLTTNG_UST_CTL
 178                         ""
 179 #else
 180                         " (support not compiled in)"
 181 #endif
 182                         );
 183 }
 184
 185 /*
 186  * daemon argument parsing
 187  */
 188 static int parse_args(int argc, char **argv)
 189 {
 190         int c, ret = 0;
 191
 192         static struct option long_options[] = {
 193                 { "consumerd-cmd-sock", 1, 0, 'c' },
 194                 { "consumerd-err-sock", 1, 0, 'e' },
 195                 { "daemonize", 0, 0, 'd' },
 196                 { "group", 1, 0, 'g' },
 197                 { "help", 0, 0, 'h' },
 198                 { "quiet", 0, 0, 'q' },
 199                 { "verbose", 0, 0, 'v' },
 200                 { "version", 0, 0, 'V' },
 201                 { "kernel", 0, 0, 'k' },
 202 #ifdef HAVE_LIBLTTNG_UST_CTL
 203                 { "ust", 0, 0, 'u' },
 204 #endif
 205                 { NULL, 0, 0, 0 }
 206         };
 207
 208         while (1) {
 209                 int option_index = 0;
 210                 c = getopt_long(argc, argv, "dhqvVku" "c:e:g:",
 211                                 long_options, &option_index);
 212                 if (c == -1) {
 213                         break;
 214                 }
 215
 216                 switch (c) {
 217                 case 0:
 218                         fprintf(stderr, "option %s",
 219                                 long_options[option_index].name);
 220                         if (optarg) {
 221                                 fprintf(stderr, " with arg %s\n", optarg);
 222                                 ret = -1;
 223                                 goto end;
 224                         }
 225                         break;
 226                 case 'c':
 227                         if (lttng_is_setuid_setgid()) {
 228                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 229                                         "-c, --consumerd-cmd-sock");
 230                         } else {
 231                                 snprintf(command_sock_path, PATH_MAX, "%s", optarg);
 232                         }
 233                         break;
 234                 case 'e':
 235                         if (lttng_is_setuid_setgid()) {
 236                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 237                                         "-e, --consumerd-err-sock");
 238                         } else {
 239                                 snprintf(error_sock_path, PATH_MAX, "%s", optarg);
 240                         }
 241                         break;
 242                 case 'd':
 243                         opt_daemon = 1;
 244                         break;
 245                 case 'g':
 246                         if (lttng_is_setuid_setgid()) {
 247                                 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
 248                                         "-g, --group");
 249                         } else {
 250                                 tracing_group_name = optarg;
 251                         }
 252                         break;
 253                 case 'h':
 254                         usage(stdout);
 255                         exit(EXIT_SUCCESS);
 256                 case 'q':
 257                         lttng_opt_quiet = 1;
 258                         break;
 259                 case 'v':
 260                         lttng_opt_verbose = 1;
 261                         break;
 262                 case 'V':
 263                         fprintf(stdout, "%s\n", VERSION);
 264                         exit(EXIT_SUCCESS);
 265                 case 'k':
 266                         opt_type = LTTNG_CONSUMER_KERNEL;
 267                         break;
 268 #ifdef HAVE_LIBLTTNG_UST_CTL
 269                 case 'u':
 270 # if (CAA_BITS_PER_LONG == 64)
 271                         opt_type = LTTNG_CONSUMER64_UST;
 272 # elif (CAA_BITS_PER_LONG == 32)
 273                         opt_type = LTTNG_CONSUMER32_UST;
 274 # else
 275 #  error "Unknown bitness"
 276 # endif
 277                         break;
 278 #endif
 279                 default:
 280                         usage(stderr);
 281                         ret = -1;
 282                         goto end;
 283                 }
 284         }
 285 end:
 286         return ret;
 287 }
 288
 289 /*
 290  * Set open files limit to unlimited. This daemon can open a large number of
 291  * file descriptors in order to consumer multiple kernel traces.
 292  */
 293 static void set_ulimit(void)
 294 {
 295         int ret;
 296         struct rlimit lim;
 297
 298         /* The kernel does not allowed an infinite limit for open files */
 299         lim.rlim_cur = 65535;
 300         lim.rlim_max = 65535;
 301
 302         ret = setrlimit(RLIMIT_NOFILE, &lim);
 303         if (ret < 0) {
 304                 PERROR("failed to set open files limit");
 305         }
 306 }
 307
 308 /*
 309  * main
 310  */
 311 int main(int argc, char **argv)
 312 {
 313         int ret = 0, retval = 0;
 314         void *status;
 315         struct lttng_consumer_local_data *tmp_ctx;
 316
 317         if (set_signal_handler()) {
 318                 retval = -1;
 319                 goto exit_set_signal_handler;
 320         }
 321
 322         /* Parse arguments */
 323         progname = argv[0];
 324         if (parse_args(argc, argv)) {
 325                 retval = -1;
 326                 goto exit_options;
 327         }
 328
 329         /* Daemonize */
 330         if (opt_daemon) {
 331                 int i;
 332
 333                 /*
 334                  * fork
 335                  * child: setsid, close FD 0, 1, 2, chdir /
 336                  * parent: exit (if fork is successful)
 337                  */
 338                 ret = daemon(0, 0);
 339                 if (ret < 0) {
 340                         PERROR("daemon");
 341                         retval = -1;
 342                         goto exit_options;
 343                 }
 344                 /*
 345                  * We are in the child. Make sure all other file
 346                  * descriptors are closed, in case we are called with
 347                  * more opened file descriptors than the standard ones.
 348                  */
 349                 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
 350                         (void) close(i);
 351                 }
 352         }
 353
 354         /*
 355          * Starting from here, we can create threads. This needs to be after
 356          * lttng_daemonize due to RCU.
 357          */
 358
 359         health_consumerd = health_app_create(NR_HEALTH_CONSUMERD_TYPES);
 360         if (!health_consumerd) {
 361                 retval = -1;
 362                 goto exit_health_consumerd_cleanup;
 363         }
 364
 365         /* Set up max poll set size */
 366         if (lttng_poll_set_max_size()) {
 367                 retval = -1;
 368                 goto exit_init_data;
 369         }
 370
 371         if (*command_sock_path == '\0') {
 372                 switch (opt_type) {
 373                 case LTTNG_CONSUMER_KERNEL:
 374                         ret = snprintf(command_sock_path, PATH_MAX,
 375                                         DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
 376                                         DEFAULT_LTTNG_RUNDIR);
 377                         if (ret < 0) {
 378                                 retval = -1;
 379                                 goto exit_init_data;
 380                         }
 381                         break;
 382                 case LTTNG_CONSUMER64_UST:
 383                         ret = snprintf(command_sock_path, PATH_MAX,
 384                                         DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH,
 385                                         DEFAULT_LTTNG_RUNDIR);
 386                         if (ret < 0) {
 387                                 retval = -1;
 388                                 goto exit_init_data;
 389                         }
 390                         break;
 391                 case LTTNG_CONSUMER32_UST:
 392                         ret = snprintf(command_sock_path, PATH_MAX,
 393                                         DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH,
 394                                         DEFAULT_LTTNG_RUNDIR);
 395                         if (ret < 0) {
 396                                 retval = -1;
 397                                 goto exit_init_data;
 398                         }
 399                         break;
 400                 default:
 401                         ERR("Unknown consumerd type");
 402                         retval = -1;
 403                         goto exit_init_data;
 404                 }
 405         }
 406
 407         /* Init */
 408         if (lttng_consumer_init()) {
 409                 retval = -1;
 410                 goto exit_init_data;
 411         }
 412
 413         /* Initialize communication library */
 414         lttcomm_init();
 415         /* Initialize TCP timeout values */
 416         lttcomm_inet_init();
 417
 418         if (!getuid()) {
 419                 /* Set limit for open files */
 420                 set_ulimit();
 421         }
 422
 423         if (run_as_create_worker(argv[0]) < 0) {
 424                 goto exit_init_data;
 425         }
 426
 427         /* create the consumer instance with and assign the callbacks */
 428         ctx = lttng_consumer_create(opt_type, lttng_consumer_read_subbuffer,
 429                 NULL, lttng_consumer_on_recv_stream, NULL);
 430         if (!ctx) {
 431                 retval = -1;
 432                 goto exit_init_data;
 433         }
 434
 435         lttng_consumer_set_command_sock_path(ctx, command_sock_path);
 436         if (*error_sock_path == '\0') {
 437                 switch (opt_type) {
 438                 case LTTNG_CONSUMER_KERNEL:
 439                         ret = snprintf(error_sock_path, PATH_MAX,
 440                                         DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
 441                                         DEFAULT_LTTNG_RUNDIR);
 442                         if (ret < 0) {
 443                                 retval = -1;
 444                                 goto exit_init_data;
 445                         }
 446                         break;
 447                 case LTTNG_CONSUMER64_UST:
 448                         ret = snprintf(error_sock_path, PATH_MAX,
 449                                         DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH,
 450                                         DEFAULT_LTTNG_RUNDIR);
 451                         if (ret < 0) {
 452                                 retval = -1;
 453                                 goto exit_init_data;
 454                         }
 455                         break;
 456                 case LTTNG_CONSUMER32_UST:
 457                         ret = snprintf(error_sock_path, PATH_MAX,
 458                                         DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH,
 459                                         DEFAULT_LTTNG_RUNDIR);
 460                         if (ret < 0) {
 461                                 retval = -1;
 462                                 goto exit_init_data;
 463                         }
 464                         break;
 465                 default:
 466                         ERR("Unknown consumerd type");
 467                         retval = -1;
 468                         goto exit_init_data;
 469                 }
 470         }
 471
 472         /* Connect to the socket created by lttng-sessiond to report errors */
 473         DBG("Connecting to error socket %s", error_sock_path);
 474         ret = lttcomm_connect_unix_sock(error_sock_path);
 475         /*
 476          * Not a fatal error, but all communication with lttng-sessiond will
 477          * fail.
 478          */
 479         if (ret < 0) {
 480                 WARN("Cannot connect to error socket (is lttng-sessiond started?)");
 481         }
 482         lttng_consumer_set_error_sock(ctx, ret);
 483
 484         /*
 485          * Block RT signals used for UST periodical metadata flush and the live
 486          * timer in main, and create a dedicated thread to handle these signals.
 487          */
 488         if (consumer_signal_init()) {
 489                 retval = -1;
 490                 goto exit_init_data;
 491         }
 492
 493         ctx->type = opt_type;
 494
 495         if (utils_create_pipe(health_quit_pipe)) {
 496                 retval = -1;
 497                 goto exit_health_pipe;
 498         }
 499
 500         /* Create thread to manage the client socket */
 501         ret = pthread_create(&health_thread, NULL,
 502                         thread_manage_health, (void *) NULL);
 503         if (ret) {
 504                 errno = ret;
 505                 PERROR("pthread_create health");
 506                 retval = -1;
 507                 goto exit_health_thread;
 508         }
 509
 510         /*
 511          * Wait for health thread to be initialized before letting the
 512          * sessiond thread reply to the sessiond that we are ready.
 513          */
 514         while (uatomic_read(&lttng_consumer_ready)) {
 515                 usleep(100000);
 516         }
 517         cmm_smp_mb();   /* Read ready before following operations */
 518
 519         /* Create thread to manage channels */
 520         ret = pthread_create(&channel_thread, NULL,
 521                         consumer_thread_channel_poll,
 522                         (void *) ctx);
 523         if (ret) {
 524                 errno = ret;
 525                 PERROR("pthread_create");
 526                 retval = -1;
 527                 goto exit_channel_thread;
 528         }
 529
 530         /* Create thread to manage the polling/writing of trace metadata */
 531         ret = pthread_create(&metadata_thread, NULL,
 532                         consumer_thread_metadata_poll,
 533                         (void *) ctx);
 534         if (ret) {
 535                 errno = ret;
 536                 PERROR("pthread_create");
 537                 retval = -1;
 538                 goto exit_metadata_thread;
 539         }
 540
 541         /* Create thread to manage the polling/writing of trace data */
 542         ret = pthread_create(&data_thread, NULL, consumer_thread_data_poll,
 543                         (void *) ctx);
 544         if (ret) {
 545                 errno = ret;
 546                 PERROR("pthread_create");
 547                 retval = -1;
 548                 goto exit_data_thread;
 549         }
 550
 551         /* Create the thread to manage the receive of fd */
 552         ret = pthread_create(&sessiond_thread, NULL,
 553                         consumer_thread_sessiond_poll,
 554                         (void *) ctx);
 555         if (ret) {
 556                 errno = ret;
 557                 PERROR("pthread_create");
 558                 retval = -1;
 559                 goto exit_sessiond_thread;
 560         }
 561
 562         /*
 563          * Create the thread to manage the UST metadata periodic timer and
 564          * live timer.
 565          */
 566         ret = pthread_create(&metadata_timer_thread, NULL,
 567                         consumer_timer_thread, (void *) ctx);
 568         if (ret) {
 569                 errno = ret;
 570                 PERROR("pthread_create");
 571                 retval = -1;
 572                 goto exit_metadata_timer_thread;
 573         }
 574
 575         ret = pthread_detach(metadata_timer_thread);
 576         if (ret) {
 577                 errno = ret;
 578                 PERROR("pthread_detach");
 579                 retval = -1;
 580                 goto exit_metadata_timer_detach;
 581         }
 582
 583         /*
 584          * This is where we start awaiting program completion (e.g. through
 585          * signal that asks threads to teardown.
 586          */
 587
 588 exit_metadata_timer_detach:
 589 exit_metadata_timer_thread:
 590         ret = pthread_join(sessiond_thread, &status);
 591         if (ret) {
 592                 errno = ret;
 593                 PERROR("pthread_join sessiond_thread");
 594                 retval = -1;
 595         }
 596 exit_sessiond_thread:
 597
 598         ret = pthread_join(data_thread, &status);
 599         if (ret) {
 600                 errno = ret;
 601                 PERROR("pthread_join data_thread");
 602                 retval = -1;
 603         }
 604 exit_data_thread:
 605
 606         ret = pthread_join(metadata_thread, &status);
 607         if (ret) {
 608                 errno = ret;
 609                 PERROR("pthread_join metadata_thread");
 610                 retval = -1;
 611         }
 612 exit_metadata_thread:
 613
 614         ret = pthread_join(channel_thread, &status);
 615         if (ret) {
 616                 errno = ret;
 617                 PERROR("pthread_join channel_thread");
 618                 retval = -1;
 619         }
 620 exit_channel_thread:
 621
 622         ret = pthread_join(health_thread, &status);
 623         if (ret) {
 624                 errno = ret;
 625                 PERROR("pthread_join health_thread");
 626                 retval = -1;
 627         }
 628 exit_health_thread:
 629
 630         utils_close_pipe(health_quit_pipe);
 631 exit_health_pipe:
 632
 633 exit_init_data:
 634         tmp_ctx = ctx;
 635         ctx = NULL;
 636         cmm_barrier();  /* Clear ctx for signal handler. */
 637         lttng_consumer_destroy(tmp_ctx);
 638         lttng_consumer_cleanup();
 639
 640         if (health_consumerd) {
 641                 health_app_destroy(health_consumerd);
 642         }
 643         /* Ensure all prior call_rcu are done. */
 644         rcu_barrier();
 645
 646         run_as_destroy_worker();
 647
 648 exit_health_consumerd_cleanup:
 649 exit_options:
 650 exit_set_signal_handler:
 651
 652         if (!retval) {
 653                 exit(EXIT_SUCCESS);
 654         } else {
 655                 exit(EXIT_FAILURE);
 656         }
 657 }