/* master.c -- IMAP master process to handle recovery, checkpointing, spawning * * Copyright (c) 1998-2003 Carnegie Mellon University. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in * the documentation and/or other materials provided with the * distribution. * * 3. The name "Carnegie Mellon University" must not be used to * endorse or promote products derived from this software without * prior written permission. For permission or any other legal * details, please contact * Office of Technology Transfer * Carnegie Mellon University * 5000 Forbes Avenue * Pittsburgh, PA 15213-3890 * (412) 268-4387, fax: (412) 268-7395 * tech-transfer@andrew.cmu.edu * * 4. Redistributions of any form whatsoever must retain the following * acknowledgment: * "This product includes software developed by Computing Services * at Carnegie Mellon University (http://www.cmu.edu/computing/)." * * CARNEGIE MELLON UNIVERSITY DISCLAIMS ALL WARRANTIES WITH REGARD TO * THIS SOFTWARE, INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY * AND FITNESS, IN NO EVENT SHALL CARNEGIE MELLON UNIVERSITY BE LIABLE * FOR ANY SPECIAL, INDIRECT OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN * AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING * OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE. */ /* $Id: master.c,v 1.83.2.1 2003/12/18 16:29:54 rjs3 Exp $ */ #include #include #include #include #include #include #include #include #include #ifdef HAVE_UNISTD_H #include #endif #ifdef HAVE_SYS_RESOURCE_H #include #endif #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #ifndef INADDR_NONE #define INADDR_NONE 0xffffffff #endif #ifndef INADDR_ANY #define INADDR_ANY 0x00000000 #endif #if !defined(IPV6_V6ONLY) && defined(IPV6_BINDV6ONLY) #define IPV6_V6ONLY IPV6_BINDV6ONLY #endif #ifdef HAVE_UCDSNMP #include #include #include #include "cyrusMasterMIB.h" #endif #include "masterconf.h" #include "master.h" #include "service.h" #include "lock.h" enum { become_cyrus_early = 1, child_table_size = 10000, child_table_inc = 100 }; static int verbose = 0; static int listen_queue_backlog = 32; static int pidfd = -1; struct service *Services = NULL; int allocservices = 0; int nservices = 0; struct recover { char *name; char *const *exec; }; struct event { char *name; time_t mark; time_t period; char *const *exec; struct event *next; }; static struct event *schedule = NULL; struct centry { pid_t pid; struct service *s; struct centry *next; }; static struct centry *ctable[child_table_size]; static struct centry *cfreelist; void limit_fds(rlim_t); void schedule_event(struct event *a); static char *mystrdup(const char *s) { if (!s) return NULL; return strdup(s); } void fatal(char *msg, int code) { syslog(LOG_CRIT, "%s", msg); syslog(LOG_NOTICE, "exiting"); exit(code); } int become_cyrus(void) { struct passwd *p; int newuid, newgid; int result; static int uid = 0; if (uid) return setuid(uid); p = getpwnam(CYRUS_USER); if (p == NULL) { syslog(LOG_ERR, "no entry in /etc/passwd for user %s", CYRUS_USER); return -1; } /* Save these in case initgroups does a getpw*() */ newuid = p->pw_uid; newgid = p->pw_gid; if (initgroups(CYRUS_USER, newgid)) { syslog(LOG_ERR, "unable to initialize groups for user %s: %s", CYRUS_USER, strerror(errno)); return -1; } if (setgid(newgid)) { syslog(LOG_ERR, "unable to set group id to %d for user %s: %s", newgid, CYRUS_USER, strerror(errno)); return -1; } result = setuid(newuid); /* Only set static uid if successful, else future calls won't reset gid */ if (result == 0) uid = newuid; return result; } void get_prog(char *path, unsigned size, char *const *cmd) { if (cmd[0][0] == '/') { /* master lacks strlcpy, due to no libcyrus */ snprintf(path, size, "%s", cmd[0]); } else snprintf(path, size, "%s/%s", SERVICE_PATH, cmd[0]); } void get_statsock(int filedes[2]) { int r, fdflags; r = pipe(filedes); if (r != 0) { fatal("couldn't create status socket: %m", 1); } /* we don't want the master blocking on reads */ fdflags = fcntl(filedes[0], F_GETFL, 0); if (fdflags != -1) fdflags = fcntl(filedes[0], F_SETFL, fdflags | O_NONBLOCK); if (fdflags == -1) { fatal("unable to set non-blocking: %m", 1); } /* we don't want the services to be able to read from it */ fdflags = fcntl(filedes[0], F_GETFD, 0); if (fdflags != -1) fdflags = fcntl(filedes[0], F_SETFD, fdflags | FD_CLOEXEC); if (fdflags == -1) { fatal("unable to set close-on-exec: %m", 1); } } /* return a new 'centry', either from the freelist or by malloc'ing it */ static struct centry *get_centry(void) { struct centry *t; if (!cfreelist) { /* create child_table_inc more and add them to the freelist */ struct centry *n; int i; n = malloc(child_table_inc * sizeof(struct centry)); cfreelist = n; for (i = 0; i < child_table_inc - 1; i++) { n[i].next = n + (i + 1); } /* i == child_table_inc - 1, last item in block */ n[i].next = NULL; } t = cfreelist; cfreelist = cfreelist->next; return t; } /* see if 'listen' parameter has both hostname and port, or just port */ char *parse_listen(char *listen) { char *cp; char *port = NULL; if ((cp = strrchr(listen,']')) != NULL) { /* ":port" after closing bracket for IP address? */ if (*cp++ != '\0' && *cp == ':') { *cp++ = '\0'; if (*cp != '\0') { port = cp; } } } else if ((cp = strrchr(listen,':')) != NULL) { /* ":port" after hostname? */ *cp++ = '\0'; if (*cp != '\0') { port = cp; } } return port; } char *parse_host(char *listen) { char *cp; /* do we have a hostname, or IP number? */ /* XXX are brackets necessary, like for IPv6 later? */ if (*listen == '[') { listen++; /* skip first bracket */ if ((cp = strrchr(listen,']')) != NULL) { *cp = '\0'; } } return listen; } int verify_service_file(char *const *filename) { char path[PATH_MAX]; struct stat statbuf; get_prog(path, sizeof(path), filename); if (stat(path, &statbuf)) return 0; if (! S_ISREG(statbuf.st_mode)) return 0; return statbuf.st_mode & S_IXUSR; } void service_create(struct service *s) { struct service service0, service; struct addrinfo hints, *res0, *res; int error, nsocket = 0; struct sockaddr_un sunsock; mode_t oldumask; int on = 1; int r; if (s->associate > 0) return; /* service is already activated */ if (s->listen[0] == '/') { /* unix socket */ res0 = (struct addrinfo *)malloc(sizeof(struct addrinfo)); if (!res0) fatal("out of memory", EX_UNAVAILABLE); memset(res0, 0, sizeof(struct addrinfo)); res0->ai_flags = AI_PASSIVE; res0->ai_family = PF_UNIX; if(!strcmp(s->proto, "tcp")) { res0->ai_socktype = SOCK_STREAM; } else { /* udp */ res0->ai_socktype = SOCK_DGRAM; } res0->ai_addr = (struct sockaddr *)&sunsock; res0->ai_addrlen = sizeof(sunsock.sun_family) + strlen(s->listen) + 1; #ifdef SIN6_LEN res0->ai_addrlen += sizeof(sunsock.sun_len); sunsock.sun_len = res0->ai_addrlen; #endif sunsock.sun_family = AF_UNIX; strcpy(sunsock.sun_path, s->listen); unlink(s->listen); } else { /* inet socket */ char *listen, *port; char *listen_addr; memset(&hints, 0, sizeof(hints)); hints.ai_flags = AI_PASSIVE; if (!strcmp(s->proto, "tcp")) { hints.ai_family = PF_UNSPEC; hints.ai_socktype = SOCK_STREAM; } else if (!strcmp(s->proto, "tcp4")) { hints.ai_family = PF_INET; hints.ai_socktype = SOCK_STREAM; } else if (!strcmp(s->proto, "tcp6")) { hints.ai_family = PF_INET6; hints.ai_socktype = SOCK_STREAM; } else if (!strcmp(s->proto, "udp")) { hints.ai_family = PF_UNSPEC; hints.ai_socktype = SOCK_DGRAM; } else if (!strcmp(s->proto, "udp4")) { hints.ai_family = PF_INET; hints.ai_socktype = SOCK_DGRAM; } else if (!strcmp(s->proto, "udp6")) { hints.ai_family = PF_INET6; hints.ai_socktype = SOCK_DGRAM; } else { syslog(LOG_INFO, "invalid proto '%s', disabling %s", s->proto, s->name); s->exec = NULL; return; } /* parse_listen() and resolve_host() are destructive, * so make a work copy of s->listen */ listen = strdup(s->listen); if ((port = parse_listen(listen)) == NULL) { /* listen IS the port */ port = listen; listen_addr = NULL; } else { /* s->listen is now just the address */ listen_addr = parse_host(listen); if (*listen_addr == '\0') listen_addr = NULL; } error = getaddrinfo(listen_addr, port, &hints, &res0); free(listen); if (error) { syslog(LOG_INFO, "%s, disabling %s", gai_strerror(error), s->name); s->exec = NULL; return; } } memcpy(&service0, s, sizeof(struct service)); for (res = res0; res; res = res->ai_next) { if (s->socket > 0) { memcpy(&service, &service0, sizeof(struct service)); s = &service; } s->socket = socket(res->ai_family, res->ai_socktype, res->ai_protocol); if (s->socket < 0) { s->socket = 0; if (verbose > 2) syslog(LOG_ERR, "unable to open %s socket: %m", s->name); continue; } /* allow reuse of address */ r = setsockopt(s->socket, SOL_SOCKET, SO_REUSEADDR, (void *) &on, sizeof(on)); if (r < 0) { syslog(LOG_ERR, "unable to setsocketopt(SO_REUSEADDR): %m"); } #if defined(IPV6_V6ONLY) && !(defined(__FreeBSD__) && __FreeBSD__ < 3) if (res->ai_family == AF_INET6) { r = setsockopt(s->socket, IPPROTO_IPV6, IPV6_V6ONLY, (void *) &on, sizeof(on)); if (r < 0) { syslog(LOG_ERR, "unable to setsocketopt(IPV6_V6ONLY): %m"); } } #endif oldumask = umask((mode_t) 0); /* for linux */ r = bind(s->socket, res->ai_addr, res->ai_addrlen); umask(oldumask); if (r < 0) { syslog(LOG_ERR, "unable to bind socket for service %s: %m", s->name); close(s->socket); s->socket = 0; continue; } if (s->listen[0] == '/') { /* unix socket */ /* for DUX, where this isn't the default. (harmlessly fails on some systems) */ chmod(s->listen, (mode_t) 0777); } if ((!strcmp(s->proto, "tcp") || !strcmp(s->proto, "tcp4") || !strcmp(s->proto, "tcp6")) && listen(s->socket, listen_queue_backlog) < 0) { syslog(LOG_ERR, "unable to listen to %s socket: %m", s->name); close(s->socket); s->socket = 0; continue; } s->ready_workers = 0; s->associate = nsocket; get_statsock(s->stat); if (s == &service) { if (nservices == allocservices) { Services = realloc(Services, (allocservices+=5) * sizeof(struct service)); if (!Services) fatal("out of memory", EX_UNAVAILABLE); } memcpy(&Services[nservices++], s, sizeof(struct service)); } nsocket++; } if (res0) freeaddrinfo(res0); if (nsocket <= 0) { syslog(LOG_ERR, "unable to create %s listener socket: %m", s->name); s->exec = NULL; return; } } void run_startup(char **cmd) { pid_t pid; int status; char path[PATH_MAX]; switch (pid = fork()) { case -1: syslog(LOG_CRIT, "can't fork process to run startup"); fatal("can't run startup", 1); break; case 0: /* Child - Release our pidfile lock. */ if(pidfd != -1) close(pidfd); if (become_cyrus() != 0) { syslog(LOG_ERR, "can't change to the cyrus user"); exit(1); } limit_fds(256); get_prog(path, sizeof(path), cmd); syslog(LOG_DEBUG, "about to exec %s", path); execv(path, cmd); syslog(LOG_ERR, "can't exec %s for startup: %m", path); exit(EX_OSERR); default: if (waitpid(pid, &status, 0) < 0) { syslog(LOG_ERR, "waitpid(): %m"); } else if (status != 0) { if (WIFEXITED(status)) { syslog(LOG_ERR, "process %d exited, status %d\n", pid, WEXITSTATUS(status)); } if (WIFSIGNALED(status)) { syslog(LOG_ERR, "process %d exited, signaled to death by %d\n", pid, WTERMSIG(status)); } } break; } } void fcntl_unset(int fd, int flag) { int fdflags = fcntl(fd, F_GETFD, 0); if (fdflags != -1) fdflags = fcntl(STATUS_FD, F_SETFD, fdflags & ~flag); if (fdflags == -1) { syslog(LOG_ERR, "fcntl(): unable to unset %d: %m", flag); } } void spawn_service(struct service *s) { /* Note that there is logic that depends on this being 2 */ const int FORKRATE_INTERVAL = 2; pid_t p; int i; char path[PATH_MAX]; static char name_env[100], name_env2[100]; struct centry *c; time_t now = time(NULL); int msg; /* update our fork rate */ if(now - s->last_interval_start >= FORKRATE_INTERVAL) { int interval; s->forkrate = (s->interval_forks/2) + (s->forkrate/2); s->interval_forks = 0; s->last_interval_start += FORKRATE_INTERVAL; /* if there is an even wider window, however, we need * to account for a good deal of zeros, we can do this at once */ interval = now - s->last_interval_start; if(interval > 2) { int skipped_intervals = interval / FORKRATE_INTERVAL; /* avoid a > 30 bit right shift) */ if(skipped_intervals > 30) s->forkrate = 0; else { /* divide by 2^(skipped_intervals). * this is the logic mentioned in the comment above */ s->forkrate >>= skipped_intervals; s->last_interval_start = now; } } } /* If we've been busy lately, we will refuse to fork! */ /* (We schedule a wakeup call for sometime soon though to be * sure that we don't wait to do the fork that is required forever! */ if(s->maxforkrate && s->forkrate >= s->maxforkrate) { struct event *evt = (struct event *) malloc(sizeof(struct event)); if (!evt) fatal("out of memory", EX_UNAVAILABLE); memset(evt, 0, sizeof(struct event)); evt->name = "forkrate wakeup call"; evt->mark = time(NULL) + FORKRATE_INTERVAL; schedule_event(evt); return; } switch (p = fork()) { case -1: syslog(LOG_ERR, "can't fork process to run service %s: %m", s->name); break; case 0: /* Child - Release our pidfile lock. */ if(pidfd != -1) close(pidfd); if (become_cyrus() != 0) { syslog(LOG_ERR, "can't change to the cyrus user"); exit(1); } get_prog(path, sizeof(path), s->exec); if (dup2(s->stat[1], STATUS_FD) < 0) { syslog(LOG_ERR, "can't duplicate status fd: %m"); exit(1); } if (dup2(s->socket, LISTEN_FD) < 0) { syslog(LOG_ERR, "can't duplicate listener fd: %m"); exit(1); } fcntl_unset(STATUS_FD, FD_CLOEXEC); fcntl_unset(LISTEN_FD, FD_CLOEXEC); /* close all listeners */ for (i = 0; i < nservices; i++) { if (Services[i].socket > 0) close(Services[i].socket); if (Services[i].stat[0] > 0) close(Services[i].stat[0]); if (Services[i].stat[1] > 0) close(Services[i].stat[1]); } limit_fds(s->maxfds); syslog(LOG_DEBUG, "about to exec %s", path); /* add service name to environment */ snprintf(name_env, sizeof(name_env), "CYRUS_SERVICE=%s", s->name); putenv(name_env); snprintf(name_env2, sizeof(name_env2), "CYRUS_ID=%d", s->associate); putenv(name_env2); execv(path, s->exec); syslog(LOG_ERR, "couldn't exec %s: %m", path); if (write(STATUS_FD, &msg, sizeof(msg)) != sizeof(msg)) { syslog(LOG_ERR, "unable to tell master %x: %m", msg); } exit(EX_OSERR); default: /* parent */ s->ready_workers++; s->interval_forks++; s->nforks++; s->nactive++; /* add to child table */ c = get_centry(); c->pid = p; c->s = s; c->next = ctable[p % child_table_size]; ctable[p % child_table_size] = c; break; } } void schedule_event(struct event *a) { struct event *ptr; if (!schedule || a->mark < schedule->mark) { a->next = schedule; schedule = a; return; } for (ptr = schedule; ptr->next && ptr->next->mark <= a->mark; ptr = ptr->next) ; /* insert a */ a->next = ptr->next; ptr->next = a; } void spawn_schedule(time_t now) { struct event *a, *b; int i; char path[PATH_MAX]; pid_t p; struct centry *c; a = NULL; /* update schedule accordingly */ while (schedule && schedule->mark <= now) { /* delete from schedule, insert into a */ struct event *ptr = schedule; /* delete */ schedule = schedule->next; /* insert */ ptr->next = a; a = ptr; } /* run all events */ while (a && a != schedule) { /* if a->exec is NULL, we just used the event to wake up, * so we actually don't need to exec anything at the moment */ if(a->exec) { switch (p = fork()) { case -1: syslog(LOG_CRIT, "can't fork process to run event %s", a->name); break; case 0: /* Child - Release our pidfile lock. */ if(pidfd != -1) close(pidfd); if (become_cyrus() != 0) { syslog(LOG_ERR, "can't change to the cyrus user"); exit(1); } /* close all listeners */ for (i = 0; i < nservices; i++) { if (Services[i].socket > 0) close(Services[i].socket); if (Services[i].stat[0] > 0) close(Services[i].stat[0]); if (Services[i].stat[1] > 0) close(Services[i].stat[1]); } limit_fds(256); get_prog(path, sizeof(path), a->exec); syslog(LOG_DEBUG, "about to exec %s", path); execv(path, a->exec); syslog(LOG_ERR, "can't exec %s on schedule: %m", path); exit(EX_OSERR); break; default: /* we don't wait for it to complete */ /* add to child table */ c = get_centry(); c->pid = p; c->s = NULL; c->next = ctable[p % child_table_size]; ctable[p % child_table_size] = c; break; } } /* a->exec */ /* reschedule as needed */ b = a->next; if (a->period) { a->mark += a->period; /* reschedule a */ schedule_event(a); } else { free(a); } /* examine next event */ a = b; } } void reap_child(void) { int status; pid_t pid; struct centry *c; while ((pid = waitpid((pid_t) -1, &status, WNOHANG)) > 0) { if (WIFEXITED(status)) { syslog(LOG_DEBUG, "process %d exited, status %d", pid, WEXITSTATUS(status)); } if (WIFSIGNALED(status)) { syslog(LOG_ERR, "process %d exited, signaled to death by %d", pid, WTERMSIG(status)); } /* account for the child */ c = ctable[pid % child_table_size]; if (c && c->pid == pid) { /* first thing in the linked list */ /* decrement active count for service */ if (c->s) c->s->nactive--; ctable[pid % child_table_size] = c->next; c->next = cfreelist; cfreelist = c; } else { /* not the first thing in the linked list */ while (c->next) { if (c->next->pid == pid) break; c = c->next; } if (c->next) { struct centry *t; t = c->next; /* decrement active count for service */ if (t->s) t->s->nactive--; c->next = t->next; /* remove node */ t->next = cfreelist; /* add to freelist */ cfreelist = t; } else { /* yikes! don't know about this child! */ syslog(LOG_ERR, "process %d not recognized", pid); } } } } static int gotsigchld = 0; void sigchld_handler(int sig __attribute__((unused))) { gotsigchld = 1; } static int gotsighup = 0; void sighup_handler(int sig __attribute__((unused))) { gotsighup = 1; } void sigterm_handler(int sig __attribute__((unused))) { struct sigaction action; /* send all the other processes SIGTERM, then exit */ sigemptyset(&action.sa_mask); action.sa_flags = 0; action.sa_handler = SIG_IGN; if (sigaction(SIGTERM, &action, (struct sigaction *) 0) < 0) { syslog(LOG_ERR, "sigaction: %m"); exit(1); } /* kill my process group */ if (kill(0, SIGTERM) < 0) { syslog(LOG_ERR, "sigterm_handler: kill(0, SIGTERM): %m"); } #if HAVE_UCDSNMP /* tell master agent we're exiting */ snmp_shutdown("cyrusMaster"); #endif syslog(LOG_INFO, "exiting on SIGTERM/SIGINT"); exit(0); } void sigalrm_handler(int sig __attribute__((unused))) { return; } void sighandler_setup(void) { struct sigaction action; sigemptyset(&action.sa_mask); action.sa_flags = 0; action.sa_handler = sighup_handler; #ifdef SA_RESTART action.sa_flags |= SA_RESTART; #endif if (sigaction(SIGHUP, &action, NULL) < 0) { fatal("unable to install signal handler for SIGHUP: %m", 1); } action.sa_handler = sigalrm_handler; if (sigaction(SIGALRM, &action, NULL) < 0) { fatal("unable to install signal handler for SIGALRM: %m", 1); } /* Handle SIGTERM and SIGINT the same way -- kill * off our children! */ action.sa_handler = sigterm_handler; if (sigaction(SIGTERM, &action, NULL) < 0) { fatal("unable to install signal handler for SIGTERM: %m", 1); } if (sigaction(SIGINT, &action, NULL) < 0) { fatal("unable to install signal handler for SIGINT: %m", 1); } action.sa_flags |= SA_NOCLDSTOP; action.sa_handler = sigchld_handler; if (sigaction(SIGCHLD, &action, NULL) < 0) { fatal("unable to install signal handler for SIGCHLD: %m", 1); } } void process_msg(struct service *s, int msg) { switch (msg) { case MASTER_SERVICE_AVAILABLE: s->ready_workers++; break; case MASTER_SERVICE_UNAVAILABLE: s->ready_workers--; break; case MASTER_SERVICE_CONNECTION: s->nconnections++; break; default: syslog(LOG_ERR, "unrecognized message for service '%s': %x", s->name, msg); break; } if (verbose) syslog(LOG_DEBUG, "service %s now has %d workers\n", s->name, s->ready_workers); } static char **tokenize(char *p) { char **tokens = NULL; /* allocated in increments of 10 */ int i = 0; if (!p || !*p) return NULL; /* sanity check */ while (*p) { while (*p && isspace((int) *p)) p++; /* skip whitespace */ if (!(i % 10)) tokens = realloc(tokens, (i+10) * sizeof(char *)); if (!tokens) return NULL; /* got a token */ tokens[i++] = p; while (*p && !isspace((int) *p)) p++; /* p is whitespace or end of cmd */ if (*p) *p++ = '\0'; } /* add a NULL on the end */ if (!(i % 10)) tokens = realloc(tokens, (i+1) * sizeof(char *)); if (!tokens) return NULL; tokens[i] = NULL; return tokens; } void add_start(const char *name, struct entry *e, void *rock __attribute__((unused))) { char *cmd = mystrdup(masterconf_getstring(e, "cmd", NULL)); char buf[256]; char **tok; if (!cmd) { snprintf(buf, sizeof(buf), "unable to find command for %s", name); fatal(buf, EX_CONFIG); } tok = tokenize(cmd); if (!tok) fatal("out of memory", EX_UNAVAILABLE); run_startup(tok); free(tok); free(cmd); } void add_service(const char *name, struct entry *e, void *rock) { int ignore_err = (int) rock; char *cmd = mystrdup(masterconf_getstring(e, "cmd", NULL)); int prefork = masterconf_getint(e, "prefork", 0); int babysit = masterconf_getswitch(e, "babysit", 0); int maxforkrate = masterconf_getint(e, "maxforkrate", 0); char *listen = mystrdup(masterconf_getstring(e, "listen", NULL)); char *proto = mystrdup(masterconf_getstring(e, "proto", "tcp")); char *max = mystrdup(masterconf_getstring(e, "maxchild", "-1")); rlim_t maxfds = (rlim_t) masterconf_getint(e, "maxfds", 256); int i, j; if(babysit && prefork == 0) prefork = 1; if(babysit && maxforkrate == 0) maxforkrate = 10; /* reasonable safety */ if (!cmd || !listen) { char buf[256]; snprintf(buf, sizeof(buf), "unable to find command or port for service '%s'", name); if (ignore_err) { syslog(LOG_WARNING, "WARNING: %s -- ignored", buf); return; } fatal(buf, EX_CONFIG); } /* see if we have an existing entry for this service */ for (i = 0; i < nservices; i++) { if (Services[i].associate > 0) continue; if (Services[i].name && !strcmp(Services[i].name, name)) break; } /* we have duplicate service names in the config file */ if ((i < nservices) && Services[i].exec) { char buf[256]; snprintf(buf, sizeof(buf), "multiple entries for service '%s'", name); if (ignore_err) { syslog(LOG_WARNING, "WARNING: %s -- ignored", buf); return; } fatal(buf, EX_CONFIG); } if ((i < nservices) && !strcmp(Services[i].listen, listen) && !strcmp(Services[i].proto, proto)) { /* we found an existing entry and the port paramters are the same */ Services[i].exec = tokenize(cmd); if (!Services[i].exec) fatal("out of memory", EX_UNAVAILABLE); /* is this service actually there? */ if (!verify_service_file(Services[i].exec)) { char buf[1024]; snprintf(buf, sizeof(buf), "cannot find executable for service '%s'", name); /* if it is not, we're misconfigured, die. */ fatal(buf, EX_CONFIG); } Services[i].maxforkrate = maxforkrate; Services[i].maxfds = maxfds; if (!strcmp(Services[i].proto, "tcp") || !strcmp(Services[i].proto, "tcp4") || !strcmp(Services[i].proto, "tcp6")) { Services[i].desired_workers = prefork; Services[i].babysit = babysit; Services[i].max_workers = atoi(max); if (Services[i].max_workers == -1) { Services[i].max_workers = INT_MAX; } } else { /* udp */ if (prefork > 1) prefork = 1; Services[i].desired_workers = prefork; Services[i].max_workers = 1; } for (j = 0; j < nservices; j++) { if (Services[j].associate > 0 && Services[j].name && !strcmp(Services[j].name, name)) { Services[j].maxforkrate = Services[i].maxforkrate; Services[j].exec = Services[i].exec; Services[j].desired_workers = Services[i].desired_workers; Services[j].babysit = Services[i].babysit; Services[j].max_workers = Services[i].max_workers; } } if (verbose > 2) syslog(LOG_DEBUG, "reconfig: service '%s' (%s, %s:%s, %d, %d)", Services[i].name, cmd, Services[i].proto, Services[i].listen, Services[i].desired_workers, Services[i].max_workers); } else { /* either we don't have an existing entry or we are changing * the port parameters, so create a new service */ if (nservices == allocservices) { Services = realloc(Services, (allocservices+=5) * sizeof(struct service)); if (!Services) fatal("out of memory", EX_UNAVAILABLE); } Services[nservices].name = strdup(name); Services[nservices].listen = listen; Services[nservices].proto = proto; Services[nservices].exec = tokenize(cmd); if (!Services[nservices].exec) fatal("out of memory", EX_UNAVAILABLE); /* is this service actually there? */ if (!verify_service_file(Services[i].exec)) { char buf[1024]; snprintf(buf, sizeof(buf), "cannot find executable for service '%s'", name); /* if it is not, we're misconfigured, die. */ fatal(buf, EX_CONFIG); } Services[nservices].socket = 0; Services[nservices].saddr = NULL; Services[nservices].ready_workers = 0; Services[nservices].maxforkrate = maxforkrate; Services[nservices].maxfds = maxfds; if(!strcmp(Services[nservices].proto, "tcp") || !strcmp(Services[nservices].proto, "tcp4") || !strcmp(Services[nservices].proto, "tcp6")) { Services[nservices].desired_workers = prefork; Services[nservices].babysit = babysit; Services[nservices].max_workers = atoi(max); if (Services[nservices].max_workers == -1) { Services[nservices].max_workers = INT_MAX; } } else { if (prefork > 1) prefork = 1; Services[nservices].desired_workers = prefork; Services[nservices].max_workers = 1; } memset(Services[nservices].stat, 0, sizeof(Services[nservices].stat)); Services[nservices].last_interval_start = time(NULL); Services[nservices].interval_forks = 0; Services[nservices].forkrate = 0; Services[nservices].nforks = 0; Services[nservices].nactive = 0; Services[nservices].nconnections = 0; Services[nservices].associate = 0; if (verbose > 2) syslog(LOG_DEBUG, "add: service '%s' (%s, %s:%s, %d, %d, %d)", Services[nservices].name, cmd, Services[nservices].proto, Services[nservices].listen, Services[nservices].desired_workers, Services[nservices].max_workers, (int) Services[nservices].maxfds); nservices++; } free(max); } void add_event(const char *name, struct entry *e, void *rock) { int ignore_err = (int) rock; char *cmd = mystrdup(masterconf_getstring(e, "cmd", NULL)); int period = 60 * masterconf_getint(e, "period", 0); int at = masterconf_getint(e, "at", -1), hour, min; time_t now = time(NULL); struct event *evt; if (!cmd) { char buf[256]; snprintf(buf, sizeof(buf), "unable to find command or port for event '%s'", name); if (ignore_err) { syslog(LOG_WARNING, "WARNING: %s -- ignored", buf); return; } fatal(buf, EX_CONFIG); } evt = (struct event *) malloc(sizeof(struct event)); if (!evt) fatal("out of memory", EX_UNAVAILABLE); evt->name = strdup(name); if (at >= 0 && ((hour = at / 100) <= 23) && ((min = at % 100) <= 59)) { struct tm *tm = localtime(&now); period = 86400; /* 24 hours */ tm->tm_hour = hour; tm->tm_min = min; tm->tm_sec = 0; if ((evt->mark = mktime(tm)) < now) { /* already missed it, so schedule for next day */ evt->mark += period; } } else { evt->mark = now; } evt->period = period; evt->exec = tokenize(cmd); if (!evt->exec) fatal("out of memory", EX_UNAVAILABLE); schedule_event(evt); } #ifdef HAVE_SETRLIMIT #ifdef RLIMIT_NOFILE # define RLIMIT_NUMFDS RLIMIT_NOFILE #else # ifdef RLIMIT_OFILE # define RLIMIT_NUMFDS RLIMIT_OFILE # endif #endif void limit_fds(rlim_t x) { struct rlimit rl; int r; rl.rlim_cur = x; rl.rlim_max = x; if (setrlimit(RLIMIT_NUMFDS, &rl) < 0) { syslog(LOG_ERR, "setrlimit: Unable to set file descriptors limit to %ld: %m", x); #ifdef HAVE_GETRLIMIT if (!getrlimit(RLIMIT_NUMFDS, &rl)) { syslog(LOG_ERR, "retrying with %ld (current max)", rl.rlim_max); rl.rlim_cur = rl.rlim_max; if (setrlimit(RLIMIT_NUMFDS, &rl) < 0) { syslog(LOG_ERR, "setrlimit: Unable to set file descriptors limit to %ld: %m", x); } } } if (verbose > 1) { r = getrlimit(RLIMIT_NUMFDS, &rl); syslog(LOG_DEBUG, "set maximum file descriptors to %ld/%ld", rl.rlim_cur, rl.rlim_max); } #else } #endif /* HAVE_GETRLIMIT */ } #else void limit_fds(rlim_t x) { } #endif /* HAVE_SETRLIMIT */ void reread_conf(void) { int i; struct event *ptr; /* disable all services - they will be re-enabled if they appear in config file */ for (i = 0; i < nservices; i++) Services[i].exec = NULL; /* read services */ masterconf_getsection("SERVICES", &add_service, (void*) 1); for (i = 0; i < nservices; i++) { if (!Services[i].exec && Services[i].socket) { /* cleanup newly disabled services */ if (verbose > 2) syslog(LOG_DEBUG, "disable: service %s socket %d pipe %d %d", Services[i].name, Services[i].socket, Services[i].stat[0], Services[i].stat[1]); if (Services[i].associate == 0) { free(Services[i].name); free(Services[i].listen); free(Services[i].proto); } Services[i].name = NULL; Services[i].desired_workers = 0; Services[i].nforks = 0; Services[i].nactive = 0; Services[i].nconnections = 0; /* close all listeners */ if (Services[i].socket > 0) { shutdown(Services[i].socket, SHUT_RDWR); close(Services[i].socket); } Services[i].socket = 0; Services[i].saddr = NULL; if (Services[i].stat[0] > 0) close(Services[i].stat[0]); if (Services[i].stat[1] > 0) close(Services[i].stat[1]); memset(Services[i].stat, 0, sizeof(Services[i].stat)); } else if (Services[i].exec && !Services[i].socket) { /* initialize new services */ service_create(&Services[i]); if (verbose > 2) syslog(LOG_DEBUG, "init: service %s socket %d pipe %d %d", Services[i].name, Services[i].socket, Services[i].stat[0], Services[i].stat[1]); } } /* remove existing events */ while (schedule) { ptr = schedule; schedule = schedule->next; free(ptr->name); free((char**) ptr->exec); free(ptr); } schedule = NULL; /* read events */ masterconf_getsection("EVENTS", &add_event, (void*) 1); } int main(int argc, char **argv) { const char *default_pidfile = MASTER_PIDFILE; const char *lock_suffix = ".lock"; const char *pidfile = default_pidfile; char *pidfile_lock = NULL; int startup_pipe[2] = { -1, -1 }; int pidlock_fd = -1; int i, opt, close_std = 1, daemon_mode = 0; extern int optind; extern char *optarg; int fd; fd_set rfds; char *p = NULL; p = getenv("CYRUS_VERBOSE"); if (p) verbose = atoi(p) + 1; while ((opt = getopt(argc, argv, "p:l:Dd")) != EOF) { switch (opt) { case 'l': /* user defined listen queue backlog */ listen_queue_backlog = atoi(optarg); break; case 'p': /* Set the pidfile name */ pidfile = optarg; break; case 'd': /* Daemon Mode */ if(!close_std) fatal("Unable to both be debug and daemon mode", EX_CONFIG); daemon_mode = 1; break; case 'D': /* Debug Mode */ if(daemon_mode) fatal("Unable to be both debug and daemon mode", EX_CONFIG); close_std = 0; break; default: break; } } masterconf_init("master"); /* zero out the children table */ memset(&ctable, 0, sizeof(struct centry *) * child_table_size); if (close_std) { /* close stdin/out/err */ for (fd = 0; fd < 3; fd++) { close(fd); if (open("/dev/null", O_RDWR, 0) != fd) fatal("couldn't open /dev/null: %m", 2); } } /* we reserve fds 3 and 4 for children to communicate with us, so they better be available. */ for (fd = 3; fd < 5; fd++) { close(fd); if (dup(0) != fd) fatal("couldn't dup fd 0: %m", 2); } /* Pidfile Algorithm in Daemon Mode. This is a little subtle because * we want to ensure that we can report an error to our parent if the * child fails to lock the pidfile. * * [A] Create/lock pidfile.lock. If locked, exit(failure). * [A] Create a pipe * [A] Fork [B] * [A] Block on reading exit code from pipe * [B] Create/lock pidfile. If locked, write failure code to pipe and * exit(failure) * [B] write pid to pidfile * [B] write success code to pipe & finish starting up * [A] unlink pidfile.lock and exit(code read from pipe) * */ if(daemon_mode) { /* Daemonize */ pid_t pid = -1; pidfile_lock = malloc(strlen(pidfile) + strlen(lock_suffix) + 1); if(!pidfile_lock) fatal("out of memory", EX_TEMPFAIL); strcpy(pidfile_lock, pidfile); strcat(pidfile_lock, lock_suffix); pidlock_fd = open(pidfile_lock, O_CREAT|O_TRUNC|O_RDWR, 0644); if(pidlock_fd == -1) { syslog(LOG_ERR, "can't open pidfile lock: %s (%m)", pidfile_lock); exit(EX_OSERR); } else { if(lock_nonblocking(pidlock_fd)) { syslog(LOG_ERR, "can't get exclusive lock on %s", pidfile_lock); exit(EX_TEMPFAIL); } } if(pipe(startup_pipe) == -1) { syslog(LOG_ERR, "can't create startup pipe (%m)"); exit(EX_OSERR); } do { pid = fork(); if ((pid == -1) && (errno == EAGAIN)) { syslog(LOG_WARNING, "master fork failed (sleeping): %m"); sleep(5); } } while ((pid == -1) && (errno == EAGAIN)); if (pid == -1) { fatal("fork error", EX_OSERR); } else if (pid != 0) { int exit_code; /* Parent, wait for child */ if(read(startup_pipe[0], &exit_code, sizeof(exit_code)) == -1) { syslog(LOG_ERR, "could not read from startup_pipe (%m)"); unlink(pidfile_lock); exit(EX_OSERR); } else { unlink(pidfile_lock); exit(exit_code); } } /* Child! */ close(startup_pipe[0]); free(pidfile_lock); /* * We're now running in the child. Lose our controlling terminal * and obtain a new process group. */ if (setsid() == -1) { int exit_result = EX_OSERR; /* Tell our parent that we failed. */ write(startup_pipe[1], &exit_result, sizeof(exit_result)); fatal("setsid failure", EX_OSERR); } } limit_fds(RLIM_INFINITY); /* Write out the pidfile */ pidfd = open(pidfile, O_CREAT|O_RDWR, 0644); if(pidfd == -1) { int exit_result = EX_OSERR; /* Tell our parent that we failed. */ write(startup_pipe[1], &exit_result, sizeof(exit_result)); syslog(LOG_ERR, "can't open pidfile: %m"); exit(EX_OSERR); } else { char buf[100]; if(lock_nonblocking(pidfd)) { int exit_result = EX_OSERR; /* Tell our parent that we failed. */ write(startup_pipe[1], &exit_result, sizeof(exit_result)); fatal("cannot get exclusive lock on pidfile (is another master still running?)", EX_OSERR); } else { int pidfd_flags = fcntl(pidfd, F_GETFD, 0); if (pidfd_flags != -1) pidfd_flags = fcntl(pidfd, F_SETFD, pidfd_flags | FD_CLOEXEC); if (pidfd_flags == -1) { int exit_result = EX_OSERR; /* Tell our parent that we failed. */ write(startup_pipe[1], &exit_result, sizeof(exit_result)); fatal("unable to set close-on-exec for pidfile: %m", EX_OSERR); } /* Write PID */ snprintf(buf, sizeof(buf), "%lu\n", (unsigned long)getpid()); if(lseek(pidfd, 0, SEEK_SET) == -1 || ftruncate(pidfd, 0) == -1 || write(pidfd, buf, strlen(buf)) == -1) { int exit_result = EX_OSERR; /* Tell our parent that we failed. */ write(startup_pipe[1], &exit_result, sizeof(exit_result)); fatal("unable to write to pidfile: %m", EX_OSERR); } fsync(pidfd); } } if(daemon_mode) { int exit_result = 0; /* success! */ if(write(startup_pipe[1], &exit_result, sizeof(exit_result)) == -1) { syslog(LOG_ERR, "could not write success result to startup pipe (%m)"); exit(EX_OSERR); } close(startup_pipe[1]); if(pidlock_fd != -1) close(pidlock_fd); } syslog(LOG_NOTICE, "process started"); #ifdef HAVE_UCDSNMP /* initialize SNMP agent */ /* make us a agentx client. */ ds_set_boolean(DS_APPLICATION_ID, DS_AGENT_ROLE, 1); /* initialize the agent library */ init_agent("cyrusMaster"); init_cyrusMasterMIB(); init_snmp("cyrusMaster"); #endif masterconf_getsection("START", &add_start, NULL); masterconf_getsection("SERVICES", &add_service, NULL); masterconf_getsection("EVENTS", &add_event, NULL); /* set signal handlers */ sighandler_setup(); /* initialize services */ for (i = 0; i < nservices; i++) { service_create(&Services[i]); if (verbose > 2) syslog(LOG_DEBUG, "init: service %s socket %d pipe %d %d", Services[i].name, Services[i].socket, Services[i].stat[0], Services[i].stat[1]); } if (become_cyrus_early) become_cyrus(); /* ok, we're going to start spawning like mad now */ syslog(LOG_NOTICE, "ready for work"); for (;;) { int r, i, msg, maxfd; struct timeval tv, *tvptr; time_t now = time(NULL); #if HAVE_UCDSNMP int blockp = 0; #endif /* run any scheduled processes */ spawn_schedule(now); /* reap first, that way if we need to babysit we will */ if (gotsigchld) { /* order matters here */ gotsigchld = 0; reap_child(); } /* do we have any services undermanned? */ for (i = 0; i < nservices; i++) { if (Services[i].exec /* enabled */ && (Services[i].nactive < Services[i].max_workers) && (Services[i].ready_workers < Services[i].desired_workers)) { spawn_service(&Services[i]); } else if (Services[i].exec && Services[i].babysit && Services[i].nactive == 0) { syslog(LOG_ERR, "lost all children for service: %s. " \ "Applying babysitter.", Services[i].name); spawn_service(&Services[i]); } } if (gotsighup) { syslog(LOG_NOTICE, "got SIGHUP"); gotsighup = 0; reread_conf(); } FD_ZERO(&rfds); maxfd = 0; for (i = 0; i < nservices; i++) { int x = Services[i].stat[0]; int y = Services[i].socket; /* messages */ if (x > 0) { if (verbose > 2) syslog(LOG_DEBUG, "listening for messages from %s", Services[i].name); FD_SET(x, &rfds); } if (x > maxfd) maxfd = x; /* connections */ if (y > 0 && Services[i].ready_workers == 0 && Services[i].nactive < Services[i].max_workers) { if (verbose > 2) syslog(LOG_DEBUG, "listening for connections for %s", Services[i].name); FD_SET(y, &rfds); if (y > maxfd) maxfd = y; } /* paranoia */ if (Services[i].ready_workers < 0) { syslog(LOG_ERR, "%s has %d workers?!?", Services[i].name, Services[i].ready_workers); } } maxfd++; /* need 1 greater than maxfd */ /* how long to wait? - do now so that any scheduled wakeup * calls get accounted for*/ tvptr = NULL; if (schedule) { if (now < schedule->mark) tv.tv_sec = schedule->mark - now; else tv.tv_sec = 0; tv.tv_usec = 0; tvptr = &tv; } #ifdef HAVE_UCDSNMP if (tvptr == NULL) blockp = 1; snmp_select_info(&maxfd, &rfds, tvptr, &blockp); #endif errno = 0; r = select(maxfd, &rfds, NULL, NULL, tvptr); if (r == -1 && errno == EAGAIN) continue; if (r == -1 && errno == EINTR) continue; if (r == -1) { /* uh oh */ fatal("select failed: %m", 1); } #ifdef HAVE_UCDSNMP /* check for SNMP queries */ snmp_read(&rfds); snmp_timeout(); #endif for (i = 0; i < nservices; i++) { int x = Services[i].stat[0]; int y = Services[i].socket; int j; if (FD_ISSET(x, &rfds)) { r = read(x, &msg, sizeof(int)); if (r != sizeof(int)) { syslog(LOG_ERR, "got weird response from child: %x", i); continue; } process_msg(&Services[i], msg); } if (Services[i].exec && Services[i].nactive < Services[i].max_workers) { /* bring us up to desired_workers */ for (j = Services[i].ready_workers; j < Services[i].desired_workers; j++) { spawn_service(&Services[i]); } if (Services[i].ready_workers == 0 && FD_ISSET(y, &rfds)) { /* huh, someone wants to talk to us */ spawn_service(&Services[i]); } } } } }