X-Git-Url: https://pintos-os.org/cgi-bin/gitweb.cgi?a=blobdiff_plain;f=lib%2Fdaemon.c;h=081912b128c230c59550608059ea4cdc9a8b25a0;hb=26ad129e69fc7c800630dbd541dc2dcc8150c3a4;hp=61754de6ec1b9ef44639ebcc096d9563cfba16d4;hpb=55368fb83674be5327639322ab3ffb0192782da2;p=openvswitch diff --git a/lib/daemon.c b/lib/daemon.c index 61754de6..081912b1 100644 --- a/lib/daemon.c +++ b/lib/daemon.c @@ -22,9 +22,11 @@ #include #include #include +#include "command-line.h" #include "fatal-signal.h" #include "dirs.h" #include "lockfile.h" +#include "process.h" #include "socket-util.h" #include "timeval.h" #include "util.h" @@ -44,17 +46,21 @@ static bool overwrite_pidfile; /* Should we chdir to "/"? */ static bool chdir_ = true; -/* File descriptors used by daemonize_start() and daemonize_complete(). */ -static int daemonize_fds[2]; +/* File descriptor used by daemonize_start() and daemonize_complete(). */ +static int daemonize_fd = -1; + +/* --monitor: Should a supervisory process monitor the daemon and restart it if + * it dies due to an error signal? */ +static bool monitor; /* Returns the file name that would be used for a pidfile if 'name' were * provided to set_pidfile(). The caller must free the returned string. */ char * make_pidfile_name(const char *name) { - return (!name ? xasprintf("%s/%s.pid", ovs_rundir, program_name) - : *name == '/' ? xstrdup(name) - : xasprintf("%s/%s", ovs_rundir, name)); + return (!name + ? xasprintf("%s/%s.pid", ovs_rundir, program_name) + : abs_file_name(ovs_rundir, name)); } /* Sets up a following call to daemonize() to create a pidfile named 'name'. @@ -117,6 +123,14 @@ get_detach(void) return detach; } +/* Sets up a following call to daemonize() to fork a supervisory process to + * monitor the daemon and restart it if it dies due to an error signal. */ +void +daemon_set_monitor(void) +{ + monitor = true; +} + /* If a pidfile has been configured and that pidfile already exists and is * locked by a running process, returns the pid of the running process. * Otherwise, returns 0. */ @@ -220,6 +234,157 @@ daemonize(void) daemonize_complete(); } +static pid_t +fork_and_wait_for_startup(int *fdp) +{ + int fds[2]; + pid_t pid; + + if (pipe(fds) < 0) { + ovs_fatal(errno, "pipe failed"); + } + + pid = fork(); + if (pid > 0) { + /* Running in parent process. */ + char c; + + close(fds[1]); + fatal_signal_fork(); + if (read(fds[0], &c, 1) != 1) { + int retval; + int status; + + do { + retval = waitpid(pid, &status, 0); + } while (retval == -1 && errno == EINTR); + + if (retval == pid + && WIFEXITED(status) + && WEXITSTATUS(status)) { + /* Child exited with an error. Convey the same error to + * our parent process as a courtesy. */ + exit(WEXITSTATUS(status)); + } + + ovs_fatal(errno, "fork child failed to signal startup"); + } + close(fds[0]); + *fdp = -1; + } else if (!pid) { + /* Running in child process. */ + close(fds[0]); + time_postfork(); + lockfile_postfork(); + *fdp = fds[1]; + } else { + ovs_fatal(errno, "could not fork"); + } + + return pid; +} + +static void +fork_notify_startup(int fd) +{ + if (fd != -1) { + size_t bytes_written; + int error; + + error = write_fully(fd, "", 1, &bytes_written); + if (error) { + ovs_fatal(error, "could not write to pipe"); + } + + close(fd); + } +} + +static bool +should_restart(int status) +{ + if (WIFSIGNALED(status)) { + static const int error_signals[] = { + SIGABRT, SIGALRM, SIGBUS, SIGFPE, SIGILL, SIGPIPE, SIGSEGV, + SIGXCPU, SIGXFSZ + }; + + size_t i; + + for (i = 0; i < ARRAY_SIZE(error_signals); i++) { + if (error_signals[i] == WTERMSIG(status)) { + return true; + } + } + } + return false; +} + +static void +monitor_daemon(pid_t daemon_pid) +{ + /* XXX Should limit the rate at which we restart the daemon. */ + /* XXX Should log daemon's stderr output at startup time. */ + const char *saved_program_name; + char *status_msg; + + saved_program_name = program_name; + program_name = xasprintf("monitor(%s)", program_name); + status_msg = xstrdup("healthy"); + for (;;) { + int retval; + int status; + + proctitle_set("%s: monitoring pid %lu (%s)", + saved_program_name, (unsigned long int) daemon_pid, + status_msg); + + do { + retval = waitpid(daemon_pid, &status, 0); + } while (retval == -1 && errno == EINTR); + + if (retval == -1) { + ovs_fatal(errno, "waitpid failed"); + } else if (retval == daemon_pid) { + char *s = process_status_msg(status); + free(status_msg); + status_msg = xasprintf("pid %lu died, %s", + (unsigned long int) daemon_pid, s); + free(s); + + if (should_restart(status)) { + VLOG_ERR("%s, restarting", status_msg); + daemon_pid = fork_and_wait_for_startup(&daemonize_fd); + if (!daemon_pid) { + break; + } + } else { + VLOG_INFO("%s, exiting", status_msg); + exit(0); + } + } + } + free(status_msg); + + /* Running in new daemon process. */ + proctitle_restore(); + free((char *) program_name); + program_name = saved_program_name; +} + +/* Close stdin, stdout, stderr. If we're started from e.g. an SSH session, + * then this keeps us from holding that session open artificially. */ +static void +close_standard_fds(void) +{ + int null_fd = get_null_fd(); + if (null_fd >= 0) { + dup2(null_fd, STDIN_FILENO); + dup2(null_fd, STDOUT_FILENO); + dup2(null_fd, STDERR_FILENO); + } +} + /* If daemonization is configured, then starts daemonization, by forking and * returning in the child process. The parent process hangs around until the * child lets it know either that it completed startup successfully (by calling @@ -228,51 +393,31 @@ daemonize(void) void daemonize_start(void) { - if (detach) { - pid_t pid; + daemonize_fd = -1; - if (pipe(daemonize_fds) < 0) { - ovs_fatal(errno, "pipe failed"); + if (detach) { + if (fork_and_wait_for_startup(&daemonize_fd) > 0) { + /* Running in parent process. */ + exit(0); } + /* Running in daemon or monitor process. */ + } - pid = fork(); - if (pid > 0) { - /* Running in parent process. */ - char c; - - close(daemonize_fds[1]); - fatal_signal_fork(); - if (read(daemonize_fds[0], &c, 1) != 1) { - int retval; - int status; - - do { - retval = waitpid(pid, &status, 0); - } while (retval == -1 && errno == EINTR); - - if (retval == pid - && WIFEXITED(status) - && WEXITSTATUS(status)) { - /* Child exited with an error. Convey the same error to - * our parent process as a courtesy. */ - exit(WEXITSTATUS(status)); - } + if (monitor) { + int saved_daemonize_fd = daemonize_fd; + pid_t daemon_pid; - ovs_fatal(errno, "daemon child failed to signal startup"); - } - exit(0); - } else if (!pid) { - /* Running in child process. */ - close(daemonize_fds[0]); - make_pidfile(); - time_postfork(); - lockfile_postfork(); - } else { - ovs_fatal(errno, "could not fork"); + daemon_pid = fork_and_wait_for_startup(&daemonize_fd); + if (daemon_pid > 0) { + /* Running in monitor process. */ + fork_notify_startup(saved_daemonize_fd); + close_standard_fds(); + monitor_daemon(daemon_pid); } - } else { - make_pidfile(); + /* Running in daemon process. */ } + + make_pidfile(); } /* If daemonization is configured, then this function notifies the parent @@ -280,31 +425,14 @@ daemonize_start(void) void daemonize_complete(void) { - if (detach) { - size_t bytes_written; - int null_fd; - int error; + fork_notify_startup(daemonize_fd); - error = write_fully(daemonize_fds[1], "", 1, &bytes_written); - if (error) { - ovs_fatal(error, "could not write to pipe"); - } - - close(daemonize_fds[1]); + if (detach) { setsid(); if (chdir_) { ignore(chdir("/")); } - - /* Close stdin, stdout, stderr. Otherwise if we're started from - * e.g. an SSH session then we tend to hold that session open - * artificially. */ - null_fd = get_null_fd(); - if (null_fd >= 0) { - dup2(null_fd, STDIN_FILENO); - dup2(null_fd, STDOUT_FILENO); - dup2(null_fd, STDERR_FILENO); - } + close_standard_fds(); } }