lib/daemon.c

   1 /*
   2  * Copyright (c) 2008, 2009, 2010 Nicira Networks.
   3  *
   4  * Licensed under the Apache License, Version 2.0 (the "License");
   5  * you may not use this file except in compliance with the License.
   6  * You may obtain a copy of the License at:
   7  *
   8  *     http://www.apache.org/licenses/LICENSE-2.0
   9  *
  10  * Unless required by applicable law or agreed to in writing, software
  11  * distributed under the License is distributed on an "AS IS" BASIS,
  12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
  13  * See the License for the specific language governing permissions and
  14  * limitations under the License.
  15  */
  16
  17 #include <config.h>
  18 #include "daemon.h"
  19 #include <errno.h>
  20 #include <fcntl.h>
  21 #include <stdlib.h>
  22 #include <string.h>
  23 #include <sys/wait.h>
  24 #include <unistd.h>
  25 #include "command-line.h"
  26 #include "fatal-signal.h"
  27 #include "dirs.h"
  28 #include "lockfile.h"
  29 #include "process.h"
  30 #include "socket-util.h"
  31 #include "timeval.h"
  32 #include "util.h"
  33
  34 #define THIS_MODULE VLM_daemon
  35 #include "vlog.h"
  36
  37 /* Should we run in the background? */
  38 static bool detach;
  39
  40 /* Name of pidfile (null if none). */
  41 static char *pidfile;
  42
  43 /* Create pidfile even if one already exists and is locked? */
  44 static bool overwrite_pidfile;
  45
  46 /* Should we chdir to "/"? */
  47 static bool chdir_ = true;
  48
  49 /* File descriptor used by daemonize_start() and daemonize_complete(). */
  50 static int daemonize_fd = -1;
  51
  52 /* --monitor: Should a supervisory process monitor the daemon and restart it if
  53  * it dies due to an error signal? */
  54 static bool monitor;
  55
  56 /* Returns the file name that would be used for a pidfile if 'name' were
  57  * provided to set_pidfile().  The caller must free the returned string. */
  58 char *
  59 make_pidfile_name(const char *name)
  60 {
  61     return (!name
  62             ? xasprintf("%s/%s.pid", ovs_rundir, program_name)
  63             : abs_file_name(ovs_rundir, name));
  64 }
  65
  66 /* Sets up a following call to daemonize() to create a pidfile named 'name'.
  67  * If 'name' begins with '/', then it is treated as an absolute path.
  68  * Otherwise, it is taken relative to RUNDIR, which is $(prefix)/var/run by
  69  * default.
  70  *
  71  * If 'name' is null, then program_name followed by ".pid" is used. */
  72 void
  73 set_pidfile(const char *name)
  74 {
  75     free(pidfile);
  76     pidfile = make_pidfile_name(name);
  77 }
  78
  79 /* Returns an absolute path to the configured pidfile, or a null pointer if no
  80  * pidfile is configured.  The caller must not modify or free the returned
  81  * string. */
  82 const char *
  83 get_pidfile(void)
  84 {
  85     return pidfile;
  86 }
  87
  88 /* Sets that we do not chdir to "/". */
  89 void
  90 set_no_chdir(void)
  91 {
  92     chdir_ = false;
  93 }
  94
  95 /* Will we chdir to "/" as part of daemonizing? */
  96 bool
  97 is_chdir_enabled(void)
  98 {
  99     return chdir_;
 100 }
 101
 102 /* Normally, die_if_already_running() will terminate the program with a message
 103  * if a locked pidfile already exists.  If this function is called,
 104  * die_if_already_running() will merely log a warning. */
 105 void
 106 ignore_existing_pidfile(void)
 107 {
 108     overwrite_pidfile = true;
 109 }
 110
 111 /* Sets up a following call to daemonize() to detach from the foreground
 112  * session, running this process in the background.  */
 113 void
 114 set_detach(void)
 115 {
 116     detach = true;
 117 }
 118
 119 /* Will daemonize() really detach? */
 120 bool
 121 get_detach(void)
 122 {
 123     return detach;
 124 }
 125
 126 /* Sets up a following call to daemonize() to fork a supervisory process to
 127  * monitor the daemon and restart it if it dies due to an error signal.  */
 128 void
 129 daemon_set_monitor(void)
 130 {
 131     monitor = true;
 132 }
 133
 134 /* If a pidfile has been configured and that pidfile already exists and is
 135  * locked by a running process, returns the pid of the running process.
 136  * Otherwise, returns 0. */
 137 static pid_t
 138 already_running(void)
 139 {
 140     pid_t pid = 0;
 141     if (pidfile) {
 142         int fd = open(pidfile, O_RDWR);
 143         if (fd >= 0) {
 144             struct flock lck;
 145             lck.l_type = F_WRLCK;
 146             lck.l_whence = SEEK_SET;
 147             lck.l_start = 0;
 148             lck.l_len = 0;
 149             if (fcntl(fd, F_GETLK, &lck) != -1 && lck.l_type != F_UNLCK) {
 150                 pid = lck.l_pid;
 151             }
 152             close(fd);
 153         }
 154     }
 155     return pid;
 156 }
 157
 158 /* If a locked pidfile exists, issue a warning message and, unless
 159  * ignore_existing_pidfile() has been called, terminate the program. */
 160 void
 161 die_if_already_running(void)
 162 {
 163     pid_t pid = already_running();
 164     if (pid) {
 165         if (!overwrite_pidfile) {
 166             ovs_fatal(0, "%s: already running as pid %ld",
 167                       get_pidfile(), (long int) pid);
 168         } else {
 169             VLOG_WARN("%s: %s already running as pid %ld",
 170                       get_pidfile(), program_name, (long int) pid);
 171         }
 172     }
 173 }
 174
 175 /* If a pidfile has been configured, creates it and stores the running process'
 176  * pid init.  Ensures that the pidfile will be deleted when the process
 177  * exits. */
 178 static void
 179 make_pidfile(void)
 180 {
 181     if (pidfile) {
 182         /* Create pidfile via temporary file, so that observers never see an
 183          * empty pidfile or an unlocked pidfile. */
 184         long int pid = getpid();
 185         char *tmpfile;
 186         int fd;
 187
 188         tmpfile = xasprintf("%s.tmp%ld", pidfile, pid);
 189         fatal_signal_add_file_to_unlink(tmpfile);
 190         fd = open(tmpfile, O_CREAT | O_WRONLY | O_TRUNC, 0666);
 191         if (fd >= 0) {
 192             struct flock lck;
 193             lck.l_type = F_WRLCK;
 194             lck.l_whence = SEEK_SET;
 195             lck.l_start = 0;
 196             lck.l_len = 0;
 197             if (fcntl(fd, F_SETLK, &lck) != -1) {
 198                 char *text = xasprintf("%ld\n", pid);
 199                 if (write(fd, text, strlen(text)) == strlen(text)) {
 200                     fatal_signal_add_file_to_unlink(pidfile);
 201                     if (rename(tmpfile, pidfile) < 0) {
 202                         VLOG_ERR("failed to rename \"%s\" to \"%s\": %s",
 203                                  tmpfile, pidfile, strerror(errno));
 204                         fatal_signal_remove_file_to_unlink(pidfile);
 205                         close(fd);
 206                     } else {
 207                         /* Keep 'fd' open to retain the lock. */
 208                     }
 209                     free(text);
 210                 } else {
 211                     VLOG_ERR("%s: write failed: %s", tmpfile, strerror(errno));
 212                     close(fd);
 213                 }
 214             } else {
 215                 VLOG_ERR("%s: fcntl failed: %s", tmpfile, strerror(errno));
 216                 close(fd);
 217             }
 218         } else {
 219             VLOG_ERR("%s: create failed: %s", tmpfile, strerror(errno));
 220         }
 221         fatal_signal_remove_file_to_unlink(tmpfile);
 222         free(tmpfile);
 223     }
 224     free(pidfile);
 225     pidfile = NULL;
 226 }
 227
 228 /* If configured with set_pidfile() or set_detach(), creates the pid file and
 229  * detaches from the foreground session.  */
 230 void
 231 daemonize(void)
 232 {
 233     daemonize_start();
 234     daemonize_complete();
 235 }
 236
 237 static pid_t
 238 fork_and_wait_for_startup(int *fdp)
 239 {
 240     int fds[2];
 241     pid_t pid;
 242
 243     if (pipe(fds) < 0) {
 244         ovs_fatal(errno, "pipe failed");
 245     }
 246
 247     pid = fork();
 248     if (pid > 0) {
 249         /* Running in parent process. */
 250         char c;
 251
 252         close(fds[1]);
 253         fatal_signal_fork();
 254         if (read(fds[0], &c, 1) != 1) {
 255             int retval;
 256             int status;
 257
 258             do {
 259                 retval = waitpid(pid, &status, 0);
 260             } while (retval == -1 && errno == EINTR);
 261
 262             if (retval == pid
 263                 && WIFEXITED(status)
 264                 && WEXITSTATUS(status)) {
 265                 /* Child exited with an error.  Convey the same error to
 266                  * our parent process as a courtesy. */
 267                 exit(WEXITSTATUS(status));
 268             }
 269
 270             ovs_fatal(errno, "fork child failed to signal startup");
 271         }
 272         close(fds[0]);
 273         *fdp = -1;
 274     } else if (!pid) {
 275         /* Running in child process. */
 276         close(fds[0]);
 277         time_postfork();
 278         lockfile_postfork();
 279         *fdp = fds[1];
 280     } else {
 281         ovs_fatal(errno, "could not fork");
 282     }
 283
 284     return pid;
 285 }
 286
 287 static void
 288 fork_notify_startup(int fd)
 289 {
 290     if (fd != -1) {
 291         size_t bytes_written;
 292         int error;
 293
 294         error = write_fully(fd, "", 1, &bytes_written);
 295         if (error) {
 296             ovs_fatal(error, "could not write to pipe");
 297         }
 298
 299         close(fd);
 300     }
 301 }
 302
 303 static bool
 304 should_restart(int status)
 305 {
 306     if (WIFSIGNALED(status)) {
 307         static const int error_signals[] = {
 308             SIGABRT, SIGALRM, SIGBUS, SIGFPE, SIGILL, SIGPIPE, SIGSEGV,
 309             SIGXCPU, SIGXFSZ
 310         };
 311
 312         size_t i;
 313
 314         for (i = 0; i < ARRAY_SIZE(error_signals); i++) {
 315             if (error_signals[i] == WTERMSIG(status)) {
 316                 return true;
 317             }
 318         }
 319     }
 320     return false;
 321 }
 322
 323 static void
 324 monitor_daemon(pid_t daemon_pid)
 325 {
 326     /* XXX Should limit the rate at which we restart the daemon. */
 327     /* XXX Should log daemon's stderr output at startup time. */
 328     const char *saved_program_name;
 329     char *status_msg;
 330
 331     saved_program_name = program_name;
 332     program_name = xasprintf("monitor(%s)", program_name);
 333     status_msg = xstrdup("healthy");
 334     for (;;) {
 335         int retval;
 336         int status;
 337
 338         proctitle_set("%s: monitoring pid %lu (%s)",
 339                       saved_program_name, (unsigned long int) daemon_pid,
 340                       status_msg);
 341
 342         do {
 343             retval = waitpid(daemon_pid, &status, 0);
 344         } while (retval == -1 && errno == EINTR);
 345
 346         if (retval == -1) {
 347             ovs_fatal(errno, "waitpid failed");
 348         } else if (retval == daemon_pid) {
 349             char *s = process_status_msg(status);
 350             free(status_msg);
 351             status_msg = xasprintf("pid %lu died, %s",
 352                                    (unsigned long int) daemon_pid, s);
 353             free(s);
 354
 355             if (should_restart(status)) {
 356                 VLOG_ERR("%s, restarting", status_msg);
 357                 daemon_pid = fork_and_wait_for_startup(&daemonize_fd);
 358                 if (!daemon_pid) {
 359                     break;
 360                 }
 361             } else {
 362                 VLOG_INFO("%s, exiting", status_msg);
 363                 exit(0);
 364             }
 365         }
 366     }
 367     free(status_msg);
 368
 369     /* Running in new daemon process. */
 370     proctitle_restore();
 371     free((char *) program_name);
 372     program_name = saved_program_name;
 373 }
 374
 375 /* Close stdin, stdout, stderr.  If we're started from e.g. an SSH session,
 376  * then this keeps us from holding that session open artificially. */
 377 static void
 378 close_standard_fds(void)
 379 {
 380     int null_fd = get_null_fd();
 381     if (null_fd >= 0) {
 382         dup2(null_fd, STDIN_FILENO);
 383         dup2(null_fd, STDOUT_FILENO);
 384         dup2(null_fd, STDERR_FILENO);
 385     }
 386 }
 387
 388 /* If daemonization is configured, then starts daemonization, by forking and
 389  * returning in the child process.  The parent process hangs around until the
 390  * child lets it know either that it completed startup successfully (by calling
 391  * daemon_complete()) or that it failed to start up (by exiting with a nonzero
 392  * exit code). */
 393 void
 394 daemonize_start(void)
 395 {
 396     daemonize_fd = -1;
 397
 398     if (detach) {
 399         if (fork_and_wait_for_startup(&daemonize_fd) > 0) {
 400             /* Running in parent process. */
 401             exit(0);
 402         }
 403         /* Running in daemon or monitor process. */
 404     }
 405
 406     if (monitor) {
 407         int saved_daemonize_fd = daemonize_fd;
 408         pid_t daemon_pid;
 409
 410         daemon_pid = fork_and_wait_for_startup(&daemonize_fd);
 411         if (daemon_pid > 0) {
 412             /* Running in monitor process. */
 413             fork_notify_startup(saved_daemonize_fd);
 414             close_standard_fds();
 415             monitor_daemon(daemon_pid);
 416         }
 417         /* Running in daemon process. */
 418     }
 419
 420     make_pidfile();
 421 }
 422
 423 /* If daemonization is configured, then this function notifies the parent
 424  * process that the child process has completed startup successfully. */
 425 void
 426 daemonize_complete(void)
 427 {
 428     fork_notify_startup(daemonize_fd);
 429
 430     if (detach) {
 431         setsid();
 432         if (chdir_) {
 433             ignore(chdir("/"));
 434         }
 435         close_standard_fds();
 436     }
 437 }
 438
 439 void
 440 daemon_usage(void)
 441 {
 442     printf(
 443         "\nDaemon options:\n"
 444         "  --detach                run in background as daemon\n"
 445         "  --no-chdir              do not chdir to '/'\n"
 446         "  --pidfile[=FILE]        create pidfile (default: %s/%s.pid)\n"
 447         "  --overwrite-pidfile     with --pidfile, start even if already "
 448                                    "running\n",
 449         ovs_rundir, program_name);
 450 }
 451
 452 /* Opens and reads a PID from 'pidfile'.  Returns the nonnegative PID if
 453  * successful, otherwise a negative errno value. */
 454 pid_t
 455 read_pidfile(const char *pidfile)
 456 {
 457     char line[128];
 458     struct flock lck;
 459     FILE *file;
 460     int error;
 461
 462     file = fopen(pidfile, "r");
 463     if (!file) {
 464         error = errno;
 465         VLOG_WARN("%s: open: %s", pidfile, strerror(error));
 466         goto error;
 467     }
 468
 469     lck.l_type = F_WRLCK;
 470     lck.l_whence = SEEK_SET;
 471     lck.l_start = 0;
 472     lck.l_len = 0;
 473     if (fcntl(fileno(file), F_GETLK, &lck)) {
 474         error = errno;
 475         VLOG_WARN("%s: fcntl: %s", pidfile, strerror(error));
 476         goto error;
 477     }
 478     if (lck.l_type == F_UNLCK) {
 479         error = ESRCH;
 480         VLOG_WARN("%s: pid file is not locked", pidfile);
 481         goto error;
 482     }
 483
 484     if (!fgets(line, sizeof line, file)) {
 485         if (ferror(file)) {
 486             error = errno;
 487             VLOG_WARN("%s: read: %s", pidfile, strerror(error));
 488         } else {
 489             error = ESRCH;
 490             VLOG_WARN("%s: read: unexpected end of file", pidfile);
 491         }
 492         goto error;
 493     }
 494
 495     if (lck.l_pid != strtoul(line, NULL, 10)) {
 496         error = ESRCH;
 497         VLOG_WARN("l_pid (%ld) != %s pid (%s)",
 498                    (long int) lck.l_pid, pidfile, line);
 499         goto error;
 500     }
 501
 502     fclose(file);
 503     return lck.l_pid;
 504
 505 error:
 506     if (file) {
 507         fclose(file);
 508     }
 509     return -error;
 510 }