X-Git-Url: https://git.exim.org/exim.git/blobdiff_plain/f254e59264330fee415229fe8af2fa974d1d3a10..ecc7ae95e36be550fa4b47de2d8dfc7115eac9cb:/src/src/daemon.c diff --git a/src/src/daemon.c b/src/src/daemon.c index 265ee2bb6..caed44bb3 100644 --- a/src/src/daemon.c +++ b/src/src/daemon.c @@ -2,8 +2,10 @@ * Exim - an Internet mail transport agent * *************************************************/ -/* Copyright (c) University of Cambridge 1995 - 2018 */ +/* Copyright (c) The Exim Maintainers 2020 - 2022 */ +/* Copyright (c) University of Cambridge 1995 - 2023 */ /* See the file NOTICE for conditions of use and distribution. */ +/* SPDX-License-Identifier: GPL-2.0-or-later */ /* Functions concerned with running Exim as a daemon */ @@ -14,17 +16,20 @@ /* Structure for holding data for each SMTP connection */ typedef struct smtp_slot { - pid_t pid; /* pid of the spawned reception process */ - uschar *host_address; /* address of the client host */ + pid_t pid; /* pid of the spawned reception process */ + uschar * host_address; /* address of the client host */ } smtp_slot; +typedef struct runner_slot { + pid_t pid; /* pid of spawned queue-runner process */ + const uschar *queue_name; /* pointer to the name in the qrunner struct */ +} runner_slot; + /* An empty slot for initializing (Standard C does not allow constructor expressions in assignments except as initializers in declarations). */ static smtp_slot empty_smtp_slot = { .pid = 0, .host_address = NULL }; - - /************************************************* * Local static variables * *************************************************/ @@ -37,12 +42,17 @@ static int accept_retry_count = 0; static int accept_retry_errno; static BOOL accept_retry_select_failed; -static int queue_run_count = 0; -static pid_t *queue_pid_slots = NULL; -static smtp_slot *smtp_slots = NULL; +static int queue_run_count = 0; /* current runners */ + +static unsigned queue_runner_slot_count = 0; +static runner_slot * queue_runner_slots = NULL; +static smtp_slot * smtp_slots = NULL; static BOOL write_pid = TRUE; +#ifndef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS +static uschar * notifier_socket_name; +#endif /************************************************* @@ -58,7 +68,6 @@ Returns: nothing static void sighup_handler(int sig) { -sig = sig; /* Keep picky compilers happy */ sighup_seen = TRUE; signal(SIGHUP, sighup_handler); } @@ -82,13 +91,12 @@ Returns: nothing static void main_sigchld_handler(int sig) { -sig = sig; /* Keep picky compilers happy */ os_non_restarting_signal(SIGCHLD, SIG_DFL); sigchld_seen = TRUE; } -/* SIGTERM handler. Try to get the damon pif file removed +/* SIGTERM handler. Try to get the daemon pid file removed before exiting. */ static void @@ -126,6 +134,33 @@ if (smtp_out) smtp_printf("421 %s\r\n", FALSE, smtp_msg); +/************************************************* +*************************************************/ + +static void +unlink_notifier_socket(void) +{ +#ifndef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS +DEBUG(D_any) debug_printf("unlinking notifier socket %s\n", notifier_socket_name); +Uunlink(notifier_socket_name); +#endif +} + + +static void +close_daemon_sockets(int daemon_notifier_fd, + struct pollfd * fd_polls, int listen_socket_count) +{ +if (daemon_notifier_fd >= 0) + { + (void) close(daemon_notifier_fd); + daemon_notifier_fd = -1; + } + +for (int i = 0; i < listen_socket_count; i++) (void) close(fd_polls[i].fd); +} + + /************************************************* * Handle a connected SMTP call * *************************************************/ @@ -137,7 +172,7 @@ is required so that they can be closed in the sub-process. Take care not to leak store in this process - reset the stacking pool at the end. Arguments: - listen_sockets sockets which are listening for incoming calls + fd_polls sockets which are listening for incoming calls listen_socket_count count of listening sockets accept_socket socket of the current accepted call accepted socket information about the current call @@ -146,7 +181,7 @@ Returns: nothing */ static void -handle_smtp_call(int *listen_sockets, int listen_socket_count, +handle_smtp_call(struct pollfd *fd_polls, int listen_socket_count, int accept_socket, struct sockaddr *accepted) { pid_t pid; @@ -245,7 +280,7 @@ subprocess because it might take time. */ if (smtp_load_reserve >= 0) { load_average = OS_GETLOADAVG(); - if (smtp_reserve_hosts == NULL && load_average > smtp_load_reserve) + if (!smtp_reserve_hosts && load_average > smtp_load_reserve) { DEBUG(D_any) debug_printf("rejecting SMTP connection: load average = %.2f\n", (double)load_average/1000.0); @@ -265,10 +300,10 @@ to provide host-specific limits according to $sender_host address, but because this is in the daemon mainline, only fast expansions (such as inline address checks) should be used. The documentation is full of warnings. */ -if (smtp_accept_max_per_host != NULL) +if (smtp_accept_max_per_host) { uschar *expanded = expand_string(smtp_accept_max_per_host); - if (expanded == NULL) + if (!expanded) { if (!f.expand_string_forcedfail) log_write(0, LOG_MAIN|LOG_PANIC, "expansion of smtp_accept_max_per_host " @@ -280,7 +315,7 @@ if (smtp_accept_max_per_host != NULL) uschar *s = expanded; while (isdigit(*s)) max_for_this_host = max_for_this_host * 10 + *s++ - '0'; - if (*s != 0) + if (*s) log_write(0, LOG_MAIN|LOG_PANIC, "expansion of smtp_accept_max_per_host " "for %s contains non-digit: %s", whofrom->s, expanded); } @@ -290,8 +325,7 @@ if (smtp_accept_max_per_host != NULL) per host_address checks. Note that at this stage smtp_accept_count contains the count of *other* connections, not including this one. */ -if ((max_for_this_host > 0) && - (smtp_accept_count >= max_for_this_host)) +if (max_for_this_host > 0 && smtp_accept_count >= max_for_this_host) { int host_accept_count = 0; int other_host_count = 0; /* keep a count of non matches to optimise */ @@ -308,8 +342,8 @@ if ((max_for_this_host > 0) && early, either by hitting the target, or finding there are not enough connections left to make the target. */ - if ((host_accept_count >= max_for_this_host) || - ((smtp_accept_count - other_host_count) < max_for_this_host)) + if ( host_accept_count >= max_for_this_host + || smtp_accept_count - other_host_count < max_for_this_host) break; } @@ -323,6 +357,7 @@ if ((max_for_this_host > 0) && log_write(L_connection_reject, LOG_MAIN, "Connection from %s refused: too many connections " "from that IP address", whofrom->s); + search_tidyup(); goto ERROR_RETURN; } } @@ -344,7 +379,7 @@ if (LOGGING(smtp_connection)) { uschar *list = hosts_connection_nolog; memset(sender_host_cache, 0, sizeof(sender_host_cache)); - if (list != NULL && verify_check_host(&list) == OK) + if (list && verify_check_host(&list) == OK) save_log_selector &= ~L_smtp_connection; else log_write(L_smtp_connection, LOG_MAIN, "SMTP connection from %s " @@ -355,24 +390,29 @@ if (LOGGING(smtp_connection)) expansion above did a lookup. */ search_tidyup(); -pid = fork(); +pid = exim_fork(US"daemon-accept"); /* Handle the child process */ if (pid == 0) { - int i; int queue_only_reason = 0; int old_pool = store_pool; int save_debug_selector = debug_selector; BOOL local_queue_only; BOOL session_local_queue_only; - #ifdef SA_NOCLDWAIT +#ifdef SA_NOCLDWAIT struct sigaction act; - #endif +#endif smtp_accept_count++; /* So that it includes this process */ + /* If the listen backlog was over the monitoring level, log it. */ + + if (smtp_listen_backlog > smtp_backlog_monitor) + log_write(0, LOG_MAIN, "listen backlog %d I=[%s]:%d", + smtp_listen_backlog, interface_address, interface_port); + /* May have been modified for the subprocess */ *log_selector = save_log_selector; @@ -424,7 +464,7 @@ if (pid == 0) extensive comment before the reception loop in exim.c for a fuller explanation of this logic. */ - for (i = 0; i < listen_socket_count; i++) (void)close(listen_sockets[i]); + close_daemon_sockets(daemon_notifier_fd, fd_polls, listen_socket_count); /* Set FD_CLOEXEC on the SMTP socket. We don't want any rogue child processes to be able to communicate with them, under any circumstances. */ @@ -433,15 +473,16 @@ if (pid == 0) (void)fcntl(dup_accept_socket, F_SETFD, fcntl(dup_accept_socket, F_GETFD) | FD_CLOEXEC); - #ifdef SA_NOCLDWAIT +#ifdef SA_NOCLDWAIT act.sa_handler = SIG_IGN; sigemptyset(&(act.sa_mask)); act.sa_flags = SA_NOCLDWAIT; sigaction(SIGCHLD, &act, NULL); - #else +#else signal(SIGCHLD, SIG_IGN); - #endif +#endif signal(SIGTERM, SIG_DFL); + signal(SIGINT, SIG_DFL); /* Attempt to get an id from the sending machine via the RFC 1413 protocol. We do this in the sub-process in order not to hold up the @@ -525,7 +566,7 @@ if (pid == 0) } if (message_id[0] == 0) continue; /* No message was accepted */ } - else + else /* bad smtp_setup_msg() */ { if (smtp_out) { @@ -645,16 +686,17 @@ if (pid == 0) { pid_t dpid; - /* Before forking, ensure that the C output buffer is flushed. Otherwise - anything that it in it will get duplicated, leading to duplicate copies - of the pending output. */ - - mac_smtp_fflush(); + /* We used to flush smtp_out before forking so that buffered data was not + duplicated, but now we want to pipeline the responses for data and quit. + Instead, hard-close the fd underlying smtp_out right after fork to discard + the data buffer. */ - if ((dpid = fork()) == 0) + if ((dpid = exim_fork(US"daemon-accept-delivery")) == 0) { (void)fclose(smtp_in); + (void)close(fileno(smtp_out)); (void)fclose(smtp_out); + smtp_in = smtp_out = NULL; /* Don't ever molest the parent's SSL connection, but do clean up the data structures if necessary. */ @@ -668,6 +710,7 @@ if (pid == 0) signal(SIGHUP, SIG_DFL); signal(SIGCHLD, SIG_DFL); signal(SIGTERM, SIG_DFL); + signal(SIGINT, SIG_DFL); if (geteuid() != root_uid && !deliver_drop_privilege) { @@ -755,10 +798,8 @@ else (void)close(dup_accept_socket); the incoming host address and an expanded active_hostname. */ log_close_all(); -interface_address = -sender_host_address = NULL; +interface_address = sender_host_name = sender_host_address = NULL; store_reset(reset_point); -sender_host_address = NULL; } @@ -884,24 +925,34 @@ while ((pid = waitpid(-1, &status, WNOHANG)) > 0) /* If it wasn't an accepting process, see if it was a queue-runner process that we are tracking. */ - if (queue_pid_slots) - { - int max = atoi(CS expand_string(queue_run_max)); - for (int i = 0; i < max; i++) - if (queue_pid_slots[i] == pid) + if (queue_runner_slots) + for (unsigned i = 0; i < queue_runner_slot_count; i++) + { + runner_slot * r = queue_runner_slots + i; + if (r->pid == pid) { - queue_pid_slots[i] = 0; + r->pid = 0; /* free up the slot */ + if (--queue_run_count < 0) queue_run_count = 0; DEBUG(D_any) debug_printf("%d queue-runner process%s now running\n", - queue_run_count, (queue_run_count == 1)? "" : "es"); + queue_run_count, queue_run_count == 1 ? "" : "es"); + + for (qrunner ** p = &qrunners, * q = qrunners; q; p = &q->next, q = *p) + if (q->name == r->queue_name) + { + if (q->interval) /* a periodic queue run */ + q->run_count--; + else /* a one-time run */ + *p = q->next; /* drop this qrunner */ + break; + } break; } - } + } } } - static void set_pid_file_path(void) { @@ -910,39 +961,145 @@ if (override_pid_file_path) if (!*pid_file_path) pid_file_path = string_sprintf("%s/exim-daemon.pid", spool_directory); + +if (pid_file_path[0] != '/') + log_write(0, LOG_PANIC_DIE, "pid file path %s must be absolute\n", pid_file_path); } -/* Remove the daemon's pidfile. Note: runs with root privilege, -as a direct child of the daemon. Does not return. */ +enum pid_op { PID_WRITE, PID_CHECK, PID_DELETE }; -void -delete_pid_file(void) +/* Do various pid file operations as safe as possible. Ideally we'd just +drop the privileges for creation of the pid file and not care at all about removal of +the file. FIXME. +Returns: true on success, false + errno==EACCES otherwise +*/ + +static BOOL +operate_on_pid_file(const enum pid_op operation, const pid_t pid) { -uschar * daemon_pid = string_sprintf("%d\n", (int)getppid()); -FILE * f; +char pid_line[sizeof(int) * 3 + 2]; +const int pid_len = snprintf(pid_line, sizeof(pid_line), "%d\n", (int)pid); +BOOL lines_match = FALSE; +uschar * path, * base, * dir; + +const int dir_flags = O_RDONLY | O_NONBLOCK; +const int base_flags = O_NOFOLLOW | O_NONBLOCK; +const mode_t base_mode = 0644; +struct stat sb; +int cwd_fd = -1, dir_fd = -1, base_fd = -1; +BOOL success = FALSE; +errno = EACCES; set_pid_file_path(); -if ((f = Ufopen(pid_file_path, "rb"))) +if (!f.running_in_test_harness && real_uid != root_uid && real_uid != exim_uid) goto cleanup; +if (pid_len < 2 || pid_len >= (int)sizeof(pid_line)) goto cleanup; + +path = string_copy(pid_file_path); +if ((base = Ustrrchr(path, '/')) == NULL) /* should not happen, but who knows */ + log_write(0, LOG_MAIN|LOG_PANIC_DIE, "pid file path \"%s\" does not contain a '/'", pid_file_path); + +dir = base != path ? path : US"/"; +*base++ = '\0'; + +if (!dir || !*dir || *dir != '/') goto cleanup; +if (!base || !*base || Ustrchr(base, '/') != NULL) goto cleanup; + +cwd_fd = open(".", dir_flags); +if (cwd_fd < 0 || fstat(cwd_fd, &sb) != 0 || !S_ISDIR(sb.st_mode)) goto cleanup; +dir_fd = open(CS dir, dir_flags); +if (dir_fd < 0 || fstat(dir_fd, &sb) != 0 || !S_ISDIR(sb.st_mode)) goto cleanup; + +/* emulate openat */ +if (fchdir(dir_fd) != 0) goto cleanup; +base_fd = open(CS base, O_RDONLY | base_flags); +if (fchdir(cwd_fd) != 0) + log_write(0, LOG_MAIN|LOG_PANIC_DIE, "can't return to previous working dir: %s", strerror(errno)); + +if (base_fd >= 0) { - if ( fgets(CS big_buffer, big_buffer_size, f) - && Ustrcmp(daemon_pid, big_buffer) == 0 - ) - if (Uunlink(pid_file_path) == 0) + char line[sizeof(pid_line)]; + ssize_t len = -1; + + if (fstat(base_fd, &sb) != 0 || !S_ISREG(sb.st_mode)) goto cleanup; + if ((sb.st_mode & 07777) != base_mode || sb.st_nlink != 1) goto cleanup; + if (sb.st_size < 2 || sb.st_size >= (off_t)sizeof(line)) goto cleanup; + + len = read(base_fd, line, sizeof(line)); + if (len != (ssize_t)sb.st_size) goto cleanup; + line[len] = '\0'; + + if (strspn(line, "0123456789") != (size_t)len-1) goto cleanup; + if (line[len-1] != '\n') goto cleanup; + lines_match = len == pid_len && strcmp(line, pid_line) == 0; + } + +if (operation == PID_WRITE) + { + if (!lines_match) + { + if (base_fd >= 0) { - DEBUG(D_any) - debug_printf("%s unlink: %s\n", pid_file_path, strerror(errno)); - } - else - DEBUG(D_any) - debug_printf("unlinked %s\n", pid_file_path); - fclose(f); + int error = -1; + /* emulate unlinkat */ + if (fchdir(dir_fd) != 0) goto cleanup; + error = unlink(CS base); + if (fchdir(cwd_fd) != 0) + log_write(0, LOG_MAIN|LOG_PANIC_DIE, "can't return to previous working dir: %s", strerror(errno)); + if (error) goto cleanup; + (void)close(base_fd); + base_fd = -1; + } + /* emulate openat */ + if (fchdir(dir_fd) != 0) goto cleanup; + base_fd = open(CS base, O_WRONLY | O_CREAT | O_EXCL | base_flags, base_mode); + if (fchdir(cwd_fd) != 0) + log_write(0, LOG_MAIN|LOG_PANIC_DIE, "can't return to previous working dir: %s", strerror(errno)); + if (base_fd < 0) goto cleanup; + if (fchmod(base_fd, base_mode) != 0) goto cleanup; + if (write(base_fd, pid_line, pid_len) != pid_len) goto cleanup; + DEBUG(D_any) debug_printf("pid written to %s\n", pid_file_path); + } } else - DEBUG(D_any) - debug_printf("%s\n", string_open_failed(errno, "pid file %s", - pid_file_path)); -exim_exit(EXIT_SUCCESS, US"pid file remover"); + { + if (!lines_match) goto cleanup; + if (operation == PID_DELETE) + { + int error = -1; + /* emulate unlinkat */ + if (fchdir(dir_fd) != 0) goto cleanup; + error = unlink(CS base); + if (fchdir(cwd_fd) != 0) + log_write(0, LOG_MAIN|LOG_PANIC_DIE, "can't return to previous working dir: %s", strerror(errno)); + if (error) goto cleanup; + } + } + +success = TRUE; +errno = 0; + +cleanup: +if (cwd_fd >= 0) (void)close(cwd_fd); +if (dir_fd >= 0) (void)close(dir_fd); +if (base_fd >= 0) (void)close(base_fd); +return success; +} + + +/* Remove the daemon's pidfile. Note: runs with root privilege, +as a direct child of the daemon. Does not return. */ + +void +delete_pid_file(void) +{ +const BOOL success = operate_on_pid_file(PID_DELETE, getppid()); + +DEBUG(D_any) + debug_printf("delete pid file %s %s: %s\n", pid_file_path, + success ? "success" : "failure", strerror(errno)); + +exim_exit(EXIT_SUCCESS); } @@ -954,9 +1111,21 @@ daemon_die(void) { int pid; +DEBUG(D_any) debug_printf("SIGTERM/SIGINT seen\n"); +#if !defined(DISABLE_TLS) && (defined(EXIM_HAVE_INOTIFY) || defined(EXIM_HAVE_KEVENT)) +tls_watch_invalidate(); +#endif + +if (daemon_notifier_fd >= 0) + { + close(daemon_notifier_fd); + daemon_notifier_fd = -1; + unlink_notifier_socket(); + } + if (f.running_in_test_harness || write_pid) { - if ((pid = fork()) == 0) + if ((pid = exim_fork(US"daemon-del-pidfile")) == 0) { if (override_pid_file_path) (void)child_exec_exim(CEE_EXEC_PANIC, FALSE, NULL, FALSE, 3, @@ -969,7 +1138,7 @@ if (f.running_in_test_harness || write_pid) if (pid > 0) child_close(pid, 1); } -exim_exit(EXIT_SUCCESS, US"daemon"); +exim_exit(EXIT_SUCCESS); } @@ -977,66 +1146,133 @@ exim_exit(EXIT_SUCCESS, US"daemon"); * Listener socket for local work prompts * *************************************************/ +ssize_t +daemon_client_sockname(struct sockaddr_un * sup, uschar ** sname) +{ +#ifdef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS +sup->sun_path[0] = 0; /* Abstract local socket addr - Linux-specific? */ +return offsetof(struct sockaddr_un, sun_path) + 1 + + snprintf(sup->sun_path+1, sizeof(sup->sun_path)-1, "exim_%d", getpid()); +#else +*sname = string_sprintf("%s/p_%d", spool_directory, getpid()); +return offsetof(struct sockaddr_un, sun_path) + + snprintf(sup->sun_path, sizeof(sup->sun_path), "%s", CS *sname); +#endif +} + +ssize_t +daemon_notifier_sockname(struct sockaddr_un * sup) +{ +#ifdef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS +sup->sun_path[0] = 0; /* Abstract local socket addr - Linux-specific? */ +return offsetof(struct sockaddr_un, sun_path) + 1 + + snprintf(sup->sun_path+1, sizeof(sup->sun_path)-1, "%s", + CS expand_string(notifier_socket)); +#else +notifier_socket_name = expand_string(notifier_socket); +return offsetof(struct sockaddr_un, sun_path) + + snprintf(sup->sun_path, sizeof(sup->sun_path), "%s", + CS notifier_socket_name); +#endif +} + + static void daemon_notifier_socket(void) { int fd; const uschar * where; -struct sockaddr_un sun = {.sun_family = AF_UNIX}; -int len; +struct sockaddr_un sa_un = {.sun_family = AF_UNIX}; +ssize_t len; + +if (!f.notifier_socket_en) + { + DEBUG(D_any) debug_printf("-oY used so not creating notifier socket\n"); + return; + } +if (override_local_interfaces && !override_pid_file_path) + { + DEBUG(D_any) + debug_printf("-oX used without -oP so not creating notifier socket\n"); + return; + } +if (!notifier_socket || !*notifier_socket) + { + DEBUG(D_any) debug_printf("no name for notifier socket\n"); + return; + } DEBUG(D_any) debug_printf("creating notifier socket\n"); -where = US"socket"; #ifdef SOCK_CLOEXEC -if ((fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0)) < 0) - goto bad; +if ((fd = socket(PF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0)) < 0) + { where = US"socket"; goto bad; } #else -if ((fd = socket(AF_UNIX, SOCK_DGRAM, 0)) < 0) - goto bad; +if ((fd = socket(PF_UNIX, SOCK_DGRAM, 0)) < 0) + { where = US"socket"; goto bad; } (void)fcntl(fd, F_SETFD, fcntl(fd, F_GETFD) | FD_CLOEXEC); #endif -sun.sun_path[0] = 0; /* Abstract local socket addr - Linux-specific? */ -len = offsetof(struct sockaddr_un, sun_path) + 1 - + snprintf(sun.sun_path+1, sizeof(sun.sun_path)-1, "%s", NOTIFIER_SOCKET_NAME); +len = daemon_notifier_sockname(&sa_un); + +#ifdef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS +DEBUG(D_any) debug_printf(" @%s\n", sa_un.sun_path+1); +#else /* filesystem-visible and persistent; will neeed removal */ +DEBUG(D_any) debug_printf(" %s\n", sa_un.sun_path); +#endif -where = US"bind"; -if (bind(fd, (const struct sockaddr *)&sun, len) < 0) - goto bad; +if (bind(fd, (const struct sockaddr *)&sa_un, (socklen_t)len) < 0) + { where = US"bind"; goto bad; } -where = US"SO_PASSCRED"; +#ifdef SO_PASSCRED /* Linux */ if (setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &on, sizeof(on)) < 0) - goto bad; + { where = US"SO_PASSCRED"; goto bad2; } +#elif defined(LOCAL_CREDS) /* FreeBSD-ish */ +if (setsockopt(fd, 0, LOCAL_CREDS, &on, sizeof(on)) < 0) + { where = US"LOCAL_CREDS"; goto bad2; } +#endif /* debug_printf("%s: fd %d\n", __FUNCTION__, fd); */ daemon_notifier_fd = fd; return; +bad2: +#ifndef EXIM_HAVE_ABSTRACT_UNIX_SOCKETS + Uunlink(sa_un.sun_path); +#endif bad: - log_write(0, LOG_MAIN|LOG_PANIC, "%s: %s: %s", + log_write(0, LOG_MAIN|LOG_PANIC, "%s %s: %s", __FUNCTION__, where, strerror(errno)); + close(fd); + return; } +/* Data for notifier-triggered queue runs */ + static uschar queuerun_msgid[MESSAGE_ID_LENGTH+1]; +static const uschar * queuerun_msg_qname; + + +/* The notifier socket has something to read. Pull the message from it, decode +and do the action. + +Return TRUE if a sigalrm should be emulated */ -/* Return TRUE if a sigalrm should be emulated */ static BOOL daemon_notification(void) { uschar buf[256], cbuf[256]; -struct sockaddr_un sun; +struct sockaddr_un sa_un; struct iovec iov = {.iov_base = buf, .iov_len = sizeof(buf)-1}; -struct msghdr msg = { .msg_name = &sun, - .msg_namelen = sizeof(sun), +struct msghdr msg = { .msg_name = &sa_un, + .msg_namelen = sizeof(sa_un), .msg_iov = &iov, .msg_iovlen = 1, .msg_control = cbuf, .msg_controllen = sizeof(cbuf) }; ssize_t sz; -struct cmsghdr * cp; buf[sizeof(buf)-1] = 0; if ((sz = recvmsg(daemon_notifier_fd, &msg, 0)) <= 0) return FALSE; @@ -1045,53 +1281,76 @@ if (sz >= sizeof(buf)) return FALSE; #ifdef notdef debug_printf("addrlen %d\n", msg.msg_namelen); #endif -DEBUG(D_queue_run) debug_printf("%s from addr%s '%s'\n", __FUNCTION__, - *sun.sun_path ? "" : " abstract", sun.sun_path+ (*sun.sun_path ? 0 : 1)); +DEBUG(D_queue_run) + if (msg.msg_namelen > 0) + { + BOOL abstract = !*sa_un.sun_path; + char * name = sa_un.sun_path + (abstract ? 1 : 0); + int namelen = (int)msg.msg_namelen - abstract ? 1 : 0; + if (*name) + debug_printf("%s from addr '%s%.*s'\n", __FUNCTION__, + abstract ? "@" : "", + namelen, name); + else + debug_printf("%s (from unknown addr)\n", __FUNCTION__); + } + else + debug_printf("%s (from unknown addr)\n", __FUNCTION__); /* Refuse to handle the item unless the peer has good credentials */ #ifdef SCM_CREDENTIALS # define EXIM_SCM_CR_TYPE SCM_CREDENTIALS -#elif defined(SCM_CREDS) +#elif defined(LOCAL_CREDS) && defined(SCM_CREDS) # define EXIM_SCM_CR_TYPE SCM_CREDS #else -# error no SCM creds knowlege + /* The OS has no way to get the creds of the caller (for a unix/datagram socket. + Punt; don't try to check. */ #endif +#ifdef EXIM_SCM_CR_TYPE for (struct cmsghdr * cp = CMSG_FIRSTHDR(&msg); cp; cp = CMSG_NXTHDR(&msg, cp)) if (cp->cmsg_level == SOL_SOCKET && cp->cmsg_type == EXIM_SCM_CR_TYPE) { -#ifdef SCM_CREDENTIALS +# ifdef SCM_CREDENTIALS /* Linux */ struct ucred * cr = (struct ucred *) CMSG_DATA(cp); if (cr->uid && cr->uid != exim_uid) { DEBUG(D_queue_run) debug_printf("%s: sender creds pid %d uid %d gid %d\n", __FUNCTION__, (int)cr->pid, (int)cr->uid, (int)cr->gid); - return FALSE; -#elif defined(SCM_CREDS) - struct cmsgcred * cr = (struct cmsgcred *) CMSG_DATA(cp); - if (cr->cmcred_uid && cr->cmcred_uid != exim_uid) + } +# elif defined(LOCAL_CREDS) /* BSD-ish */ + struct sockcred * cr = (struct sockcred *) CMSG_DATA(cp); + if (cr->sc_uid && cr->sc_uid != exim_uid) { - DEBUG(D_queue_run) debug_printf("%s: sender creds pid %d uid %d gid %d\n", - __FUNCTION__, (int)cr->cmcred_pid, (int)cr->cmcred_uid, (int)cr->cmcred_gid); - return FALSE; -#endif + DEBUG(D_queue_run) debug_printf("%s: sender creds pid ??? uid %d gid %d\n", + __FUNCTION__, (int)cr->sc_uid, (int)cr->sc_gid); } +# endif break; } +#endif buf[sz] = 0; switch (buf[0]) { -#ifdef EXPERIMENTAL_QUEUE_RAMP +#ifndef DISABLE_QUEUE_RAMP case NOTIFY_MSG_QRUN: /* this should be a message_id */ DEBUG(D_queue_run) debug_printf("%s: qrunner trigger: %s\n", __FUNCTION__, buf+1); + memcpy(queuerun_msgid, buf+1, MESSAGE_ID_LENGTH+1); + + for (qrunner * q = qrunners; q; q = q->next) + if (q->name + ? Ustrcmp(q->name, buf+1+MESSAGE_ID_LENGTH+1) == 0 + : !buf[1+MESSAGE_ID_LENGTH+1] + ) + { queuerun_msg_qname = q->name; break; } return TRUE; -#endif /*EXPERIMENTAL_QUEUE_RAMP*/ +#endif case NOTIFY_QUEUE_SIZE_REQ: { @@ -1102,16 +1361,304 @@ switch (buf[0]) debug_printf("%s: queue size request: %s\n", __FUNCTION__, buf); if (sendto(daemon_notifier_fd, buf, len, 0, - (const struct sockaddr *)&sun, msg.msg_namelen) < 0) + (const struct sockaddr *)&sa_un, msg.msg_namelen) < 0) log_write(0, LOG_MAIN|LOG_PANIC, "%s: sendto: %s\n", __FUNCTION__, strerror(errno)); - return FALSE; + break; } + + case NOTIFY_REGEX: + regex_at_daemon(buf); + break; } return FALSE; } + +static void +daemon_inetd_wtimeout(time_t last_connection_time) +{ +time_t resignal_interval = inetd_wait_timeout; + +if (last_connection_time == (time_t)0) + { + DEBUG(D_any) + debug_printf("inetd wait timeout expired, but still not seen first message, ignoring\n"); + } +else + { + time_t now = time(NULL); + if (now == (time_t)-1) + { + DEBUG(D_any) debug_printf("failed to get time: %s\n", strerror(errno)); + } + else if ((now - last_connection_time) >= inetd_wait_timeout) + { + DEBUG(D_any) + debug_printf("inetd wait timeout %d expired, ending daemon\n", + inetd_wait_timeout); + log_write(0, LOG_MAIN, "exim %s daemon terminating, inetd wait timeout reached.\n", + version_string); + daemon_die(); /* Does not return */ + } + else + resignal_interval -= (now - last_connection_time); + } + +sigalrm_seen = FALSE; +ALARM(resignal_interval); +} + + +/* Re-sort the qrunners list, and return the shortest interval. +That could be negatime. +The next-tick times should have been updated by any runs initiated, +though will not be when the global limit on runners was reached. + +Unlikely to have many queues, so insertion-sort. +*/ + +static int +next_qrunner_interval(void) +{ +qrunner * sorted = NULL; +for (qrunner * q = qrunners, * next; q; q = next) + { + next = q->next; + q->next = NULL; + if (sorted) + { + qrunner ** p = &sorted; + for (qrunner * qq; qq = *p; p = &(qq->next)) + if ( q->next_tick < qq->next_tick + || q->next_tick == qq->next_tick && q->interval < qq->interval + ) + { + *p = q; + q->next = qq; + goto INSERTED; + } + *p = q; + INSERTED: ; + } + else + sorted = q; + } +qrunners = sorted; +return qrunners ? qrunners->next_tick - time(NULL) : 0; +} + +static int +daemon_qrun(int local_queue_run_max, struct pollfd * fd_polls, int listen_socket_count) +{ +DEBUG(D_any) debug_printf("%s received\n", +#ifndef DISABLE_QUEUE_RAMP + *queuerun_msgid ? "qrun notification" : +#endif + "SIGALRM"); + +/* Do a full queue run in a child process, if required, unless we already have +enough queue runners on the go. If we are not running as root, a re-exec is +required. In the calling process, restart the alamr timer for the next run. */ + +if (is_multiple_qrun()) + if (local_queue_run_max <= 0 || queue_run_count < local_queue_run_max) + { + qrunner * q = NULL; + +#ifndef DISABLE_QUEUE_RAMP + if (*queuerun_msgid) /* See if we can start another runner for this queue */ + { + for (qrunner * qq = qrunners; qq; qq = qq->next) + if (qq->name == queuerun_msg_qname) + { + q = qq->run_count < qq->run_max ? qq : NULL; + break; + } + } + else +#endif + /* In order of run priority, find the first queue for which we can start + a runner */ + + for (q = qrunners; q; q = q->next) + if (q->run_count < q->run_max) break; + + if (q) + { + pid_t pid; + + /* Bump this queue's next-tick by it's interval */ + + if (q->interval) + { + time_t now = time(NULL); + do ; while ((q->next_tick += q->interval) <= now); + } + + if ((pid = exim_fork(US"queue-runner")) == 0) + { + /* Disable debugging if it's required only for the daemon process. We + leave the above message, because it ties up with the "child ended" + debugging messages. */ + + if (f.debug_daemon) debug_selector = 0; + + /* Close any open listening sockets in the child */ + + close_daemon_sockets(daemon_notifier_fd, + fd_polls, listen_socket_count); + + /* Reset SIGHUP and SIGCHLD in the child in both cases. */ + + signal(SIGHUP, SIG_DFL); + signal(SIGCHLD, SIG_DFL); + signal(SIGTERM, SIG_DFL); + signal(SIGINT, SIG_DFL); + + /* Re-exec if privilege has been given up, unless deliver_drop_ + privilege is set. Reset SIGALRM before exec(). */ + + if (geteuid() != root_uid && !deliver_drop_privilege) + { + uschar opt[8]; + uschar *p = opt; + uschar *extra[7]; + int extracount = 1; + + signal(SIGALRM, SIG_DFL); + queue_name = US""; + + *p++ = '-'; + *p++ = 'q'; + if ( q->queue_2stage +#ifndef DISABLE_QUEUE_RAMP + && !*queuerun_msgid +#endif + ) *p++ = 'q'; + if (q->queue_run_first_delivery) *p++ = 'i'; + if (q->queue_run_force) *p++ = 'f'; + if (q->deliver_force_thaw) *p++ = 'f'; + if (q->queue_run_local) *p++ = 'l'; + *p = 0; + + extra[0] = q->name + ? string_sprintf("%sG%s", opt, q->name) : opt; + +#ifndef DISABLE_QUEUE_RAMP + if (*queuerun_msgid) + { + log_write(0, LOG_MAIN, "notify triggered queue run"); + extra[extracount++] = queuerun_msgid; /* Trigger only the */ + extra[extracount++] = queuerun_msgid; /* one message */ + } +#endif + + /* If -R or -S were on the original command line, ensure they get + passed on. */ + + if (deliver_selectstring) + { + extra[extracount++] = f.deliver_selectstring_regex ? US"-Rr" : US"-R"; + extra[extracount++] = deliver_selectstring; + } + + if (deliver_selectstring_sender) + { + extra[extracount++] = f.deliver_selectstring_sender_regex + ? US"-Sr" : US"-S"; + extra[extracount++] = deliver_selectstring_sender; + } + + /* Overlay this process with a new execution. */ + + (void)child_exec_exim(CEE_EXEC_PANIC, FALSE, NULL, FALSE, extracount, + extra[0], extra[1], extra[2], extra[3], extra[4], extra[5], extra[6]); + + /* Control never returns here. */ + } + + /* No need to re-exec; SIGALRM remains set to the default handler */ + +#ifndef DISABLE_QUEUE_RAMP + if (*queuerun_msgid) + { + log_write(0, LOG_MAIN, "notify triggered queue run"); + f.queue_2stage = FALSE; + queue_run(q, queuerun_msgid, queuerun_msgid, FALSE); + } + else +#endif + queue_run(q, NULL, NULL, FALSE); + exim_underbar_exit(EXIT_SUCCESS); + } + + if (pid < 0) + { + log_write(0, LOG_MAIN|LOG_PANIC, "daemon: fork of queue-runner " + "process failed: %s", strerror(errno)); + log_close_all(); + } + else + { + for (int i = 0; i < local_queue_run_max; ++i) + if (queue_runner_slots[i].pid <= 0) + { + queue_runner_slots[i].pid = pid; + queue_runner_slots[i].queue_name = q->name; + q->run_count++; + queue_run_count++; + break; + } + DEBUG(D_any) debug_printf("%d queue-runner process%s running\n", + queue_run_count, queue_run_count == 1 ? "" : "es"); + } + } + } + +sigalrm_seen = FALSE; +#ifndef DISABLE_QUEUE_RAMP +if (*queuerun_msgid) /* it was a fast-ramp kick */ + *queuerun_msgid = 0; +else /* periodic or one-time queue run */ +#endif + { /* Impose a minimum 1s tick, even when a run was outstanding */ + int interval = next_qrunner_interval(); + if (interval <= 0) interval = 1; + + if (qrunners) /* there are still periodic qrunners */ + { + ALARM(interval); + return interval; + } + } +return 0; +} + + + + +const uschar * +describe_queue_runners(void) +{ +gstring * g = NULL; + +if (!is_multiple_qrun()) return US"no queue runs"; + +for (qrunner * q = qrunners; q; q = q->next) + { + g = string_catn(g, US"-q", 2); + if (q->name) g = string_append(g, 3, US"G", q->name, US"/"); + g = string_cat(g, readconf_printtime(q->interval)); + g = string_catn(g, US" ", 1); + } +gstring_trim(g, 1); +gstring_release_unused(g); +return string_from_gstring(g); +} + + /************************************************* * Exim Daemon Mainline * *************************************************/ @@ -1138,28 +1685,60 @@ There are no arguments to this function, and it never returns. */ void daemon_go(void) { -struct passwd *pw; -int *listen_sockets = NULL; -int listen_socket_count = 0; -ip_address_item *addresses = NULL; +struct passwd * pw; +struct pollfd * fd_polls, * tls_watch_poll = NULL, * dnotify_poll = NULL; +int listen_socket_count = 0, poll_fd_count; +ip_address_item * addresses = NULL; time_t last_connection_time = (time_t)0; -int local_queue_run_max = atoi(CS expand_string(queue_run_max)); +int local_queue_run_max = 0; +BOOL queue_run_max_has_dollar; + +if (is_multiple_qrun()) + + /* Nuber of runner-tracking structs needed: If the option queue_run_max has + no expandable elements then it is the overall maximum; else we assume it + depends on the queue name, and add them up to get the maximum. + Evaluate both that and the individual limits. */ + + if (Ustrchr(queue_run_max, '$') != NULL) + { + for (qrunner * q = qrunners; q; q = q->next) + { + queue_name = q->name; + local_queue_run_max += + (q->run_max = atoi(CS expand_string(queue_run_max))); + } + queue_name = US""; + } + else + { + local_queue_run_max = atoi(CS expand_string(queue_run_max)); + for (qrunner * q = qrunners; q; q = q->next) + q->run_max = local_queue_run_max; + } + +process_purpose = US"daemon"; /* If any debugging options are set, turn on the D_pid bit so that all debugging lines get the pid added. */ DEBUG(D_any|D_v) debug_selector |= D_pid; +/* Allocate enough pollstructs for inetd mode plus the ancillary sockets; +also used when there are no listen sockets. */ + +fd_polls = store_get(sizeof(struct pollfd) * 3, GET_UNTAINTED); + if (f.inetd_wait_mode) { listen_socket_count = 1; - listen_sockets = store_get(sizeof(int), FALSE); (void) close(3); if (dup2(0, 3) == -1) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "failed to dup inetd socket safely away: %s", strerror(errno)); - listen_sockets[0] = 3; + fd_polls[0].fd = 3; + fd_polls[0].events = POLLIN; (void) close(0); (void) close(1); (void) close(2); @@ -1196,11 +1775,11 @@ if (f.inetd_wait_mode || f.daemon_listen) for those OS for which this is necessary the first time it is called (in order to perform an "open" on the kernel memory file). */ - #ifdef LOAD_AVG_NEEDS_ROOT +#ifdef LOAD_AVG_NEEDS_ROOT if (queue_only_load >= 0 || smtp_load_reserve >= 0 || (deliver_queue_load_max >= 0 && deliver_drop_privilege)) (void)os_getloadavg(); - #endif +#endif } @@ -1296,7 +1875,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) list = override_local_interfaces; sep = 0; - while ((s = string_nextinlist(&list, &sep, big_buffer, big_buffer_size))) + while ((s = string_nextinlist(&list, &sep, NULL, 0))) { uschar joinstr[4]; gstring ** gp = Ustrpbrk(s, ".:") ? &new_local_interfaces : &new_smtp_port; @@ -1334,13 +1913,13 @@ if (f.daemon_listen && !f.inetd_wait_mode) list = daemon_smtp_port; sep = 0; - while ((s = string_nextinlist(&list, &sep, big_buffer, big_buffer_size))) + while ((s = string_nextinlist(&list, &sep, NULL, 0))) pct++; - default_smtp_port = store_get((pct+1) * sizeof(int), FALSE); + default_smtp_port = store_get((pct+1) * sizeof(int), GET_UNTAINTED); list = daemon_smtp_port; sep = 0; for (pct = 0; - (s = string_nextinlist(&list, &sep, big_buffer, big_buffer_size)); + (s = string_nextinlist(&list, &sep, NULL, 0)); pct++) { if (isdigit(*s)) @@ -1364,6 +1943,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) list = tls_in.on_connect_ports; sep = 0; + /* the list isn't expanded so cannot be tainted. If it ever is we will trap here */ while ((s = string_nextinlist(&list, &sep, big_buffer, big_buffer_size))) if (!isdigit(*s)) { @@ -1424,7 +2004,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) ipa->port = default_smtp_port[0]; for (int i = 1; default_smtp_port[i] > 0; i++) { - ip_address_item *new = store_get(sizeof(ip_address_item), FALSE); + ip_address_item * new = store_get(sizeof(ip_address_item), GET_UNTAINTED); memcpy(new->address, ipa->address, Ustrlen(ipa->address) + 1); new->port = default_smtp_port[i]; @@ -1478,11 +2058,16 @@ if (f.daemon_listen && !f.inetd_wait_mode) } } - /* Get a vector to remember all the sockets in */ + /* Get a vector to remember all the sockets in. + Two extra elements for the ancillary sockets */ for (ipa = addresses; ipa; ipa = ipa->next) listen_socket_count++; - listen_sockets = store_get(sizeof(int) * listen_socket_count, FALSE); + fd_polls = store_get(sizeof(struct pollfd) * (listen_socket_count + 2), + GET_UNTAINTED); + for (struct pollfd * p = fd_polls; p < fd_polls + listen_socket_count + 2; + p++) + { p->fd = -1; p->events = POLLIN; } } /* daemon_listen but not inetd_wait_mode */ @@ -1505,7 +2090,7 @@ if (f.daemon_listen) if (smtp_accept_max > 0) { - smtp_slots = store_get(smtp_accept_max * sizeof(smtp_slot), FALSE); + smtp_slots = store_get(smtp_accept_max * sizeof(smtp_slot), GET_UNTAINTED); for (int i = 0; i < smtp_accept_max; i++) smtp_slots[i] = empty_smtp_slot; } } @@ -1542,15 +2127,19 @@ if (f.background_daemon) daemon as the result of a SIGHUP. In this case, there is no need to do anything, because the controlling terminal has long gone. Otherwise, fork, in case current process is a process group leader (see 'man setsid' for an - explanation) before calling setsid(). */ + explanation) before calling setsid(). + All other forks want daemon_listen cleared. Rather than blow a register, jsut + restore it here. */ if (getppid() != 1) { - pid_t pid = fork(); + BOOL daemon_listen = f.daemon_listen; + pid_t pid = exim_fork(US"daemon"); if (pid < 0) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "fork() failed when starting daemon: %s", strerror(errno)); if (pid > 0) exit(EXIT_SUCCESS); /* in parent process, just exit */ (void)setsid(); /* release controlling terminal */ + f.daemon_listen = daemon_listen; } } @@ -1573,8 +2162,8 @@ if (f.daemon_listen && !f.inetd_wait_mode) for (ipa = addresses, sk = 0; sk < listen_socket_count; ipa = ipa->next, sk++) { BOOL wildcard; - ip_address_item *ipa2; - int af; + ip_address_item * ipa2; + int fd, af; if (Ustrchr(ipa->address, ':') != NULL) { @@ -1587,7 +2176,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) wildcard = ipa->address[0] == 0; } - if ((listen_sockets[sk] = ip_socket(SOCK_STREAM, af)) < 0) + if ((fd_polls[sk].fd = fd = ip_socket(SOCK_STREAM, af)) < 0) { if (check_special_case(0, addresses, ipa, FALSE)) { @@ -1596,7 +2185,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) goto SKIP_SOCKET; } log_write(0, LOG_PANIC_DIE, "IPv%c socket creation failed: %s", - (af == AF_INET6)? '6' : '4', strerror(errno)); + af == AF_INET6 ? '6' : '4', strerror(errno)); } /* If this is an IPv6 wildcard socket, set IPV6_V6ONLY if that option is @@ -1605,8 +2194,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) #ifdef IPV6_V6ONLY if (af == AF_INET6 && wildcard && - setsockopt(listen_sockets[sk], IPPROTO_IPV6, IPV6_V6ONLY, CS (&on), - sizeof(on)) < 0) + setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY, &on, sizeof(on)) < 0) log_write(0, LOG_MAIN, "Setting IPV6_V6ONLY on daemon's IPv6 wildcard " "socket failed (%s): carrying on without it", strerror(errno)); #endif /* IPV6_V6ONLY */ @@ -1615,16 +2203,14 @@ if (f.daemon_listen && !f.inetd_wait_mode) is being handled. Without this, a connection will prevent reuse of the smtp port for listening. */ - if (setsockopt(listen_sockets[sk], SOL_SOCKET, SO_REUSEADDR, - US (&on), sizeof(on)) < 0) + if (setsockopt(fd, SOL_SOCKET, SO_REUSEADDR, &on, sizeof(on)) < 0) log_write(0, LOG_MAIN|LOG_PANIC_DIE, "setting SO_REUSEADDR on socket " "failed when starting daemon: %s", strerror(errno)); /* Set TCP_NODELAY; Exim does its own buffering. There is a switch to disable this because it breaks some broken clients. */ - if (tcp_nodelay) setsockopt(listen_sockets[sk], IPPROTO_TCP, TCP_NODELAY, - US (&on), sizeof(on)); + if (tcp_nodelay) setsockopt(fd, IPPROTO_TCP, TCP_NODELAY, &on, sizeof(on)); /* Now bind the socket to the required port; if Exim is being restarted it may not always be possible to bind immediately, even with SO_REUSEADDR @@ -1642,12 +2228,12 @@ if (f.daemon_listen && !f.inetd_wait_mode) for(;;) { uschar *msg, *addr; - if (ip_bind(listen_sockets[sk], af, ipa->address, ipa->port) >= 0) break; + if (ip_bind(fd, af, ipa->address, ipa->port) >= 0) break; if (check_special_case(errno, addresses, ipa, TRUE)) { DEBUG(D_any) debug_printf("wildcard IPv4 bind() failed after IPv6 " "listen() success; EADDRINUSE ignored\n"); - (void)close(listen_sockets[sk]); + (void)close(fd); goto SKIP_SOCKET; } msg = US strerror(errno); @@ -1675,30 +2261,30 @@ if (f.daemon_listen && !f.inetd_wait_mode) else debug_printf("listening on %s port %d\n", ipa->address, ipa->port); + /* Start listening on the bound socket, establishing the maximum backlog of + connections that is allowed. On success, add to the set of sockets for select + and continue to the next address. */ + #if defined(TCP_FASTOPEN) && !defined(__APPLE__) if ( f.tcp_fastopen_ok - && setsockopt(listen_sockets[sk], IPPROTO_TCP, TCP_FASTOPEN, + && setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &smtp_connect_backlog, sizeof(smtp_connect_backlog))) { DEBUG(D_any) debug_printf("setsockopt FASTOPEN: %s\n", strerror(errno)); f.tcp_fastopen_ok = FALSE; } #endif - - /* Start listening on the bound socket, establishing the maximum backlog of - connections that is allowed. On success, continue to the next address. */ - - if (listen(listen_sockets[sk], smtp_connect_backlog) >= 0) + if (listen(fd, smtp_connect_backlog) >= 0) { #if defined(TCP_FASTOPEN) && defined(__APPLE__) if ( f.tcp_fastopen_ok - && setsockopt(listen_sockets[sk], IPPROTO_TCP, TCP_FASTOPEN, - &on, sizeof(on))) + && setsockopt(fd, IPPROTO_TCP, TCP_FASTOPEN, &on, sizeof(on))) { DEBUG(D_any) debug_printf("setsockopt FASTOPEN: %s\n", strerror(errno)); f.tcp_fastopen_ok = FALSE; } #endif + fd_polls[sk].fd = fd; continue; } @@ -1716,7 +2302,7 @@ if (f.daemon_listen && !f.inetd_wait_mode) DEBUG(D_any) debug_printf("wildcard IPv4 listen() failed after IPv6 " "listen() success; EADDRINUSE ignored\n"); - (void)close(listen_sockets[sk]); + (void)close(fd); /* Come here if there has been a problem with the socket which we are going to ignore. We remove the address from the chain, and back up the @@ -1757,19 +2343,11 @@ The variable daemon_write_pid is used to control this. */ if (f.running_in_test_harness || write_pid) { - FILE *f; - - set_pid_file_path(); - if ((f = modefopen(pid_file_path, "wb", 0644))) - { - (void)fprintf(f, "%d\n", (int)getpid()); - (void)fclose(f); - DEBUG(D_any) debug_printf("pid written to %s\n", pid_file_path); - } - else - DEBUG(D_any) - debug_printf("%s\n", string_open_failed(errno, "pid file %s", - pid_file_path)); + const enum pid_op operation = (f.running_in_test_harness + || real_uid == root_uid + || (real_uid == exim_uid && !override_pid_file_path)) ? PID_WRITE : PID_CHECK; + if (!operate_on_pid_file(operation, getpid())) + DEBUG(D_any) debug_printf("%s pid file %s: %s\n", (operation == PID_WRITE) ? "write" : "check", pid_file_path, strerror(errno)); } /* Set up the handler for SIGHUP, which causes a restart of the daemon. */ @@ -1796,10 +2374,11 @@ originator_login = (pw = getpwuid(exim_uid)) /* Get somewhere to keep the list of queue-runner pids if we are keeping track of them (and also if we are doing queue runs). */ -if (queue_interval > 0 && local_queue_run_max > 0) +if (is_multiple_qrun() && local_queue_run_max > 0) { - queue_pid_slots = store_get(local_queue_run_max * sizeof(pid_t), FALSE); - for (int i = 0; i < local_queue_run_max; i++) queue_pid_slots[i] = 0; + queue_runner_slot_count = local_queue_run_max; + queue_runner_slots = store_get(local_queue_run_max * sizeof(runner_slot), GET_UNTAINTED); + memset(queue_runner_slots, 0, local_queue_run_max * sizeof(runner_slot)); } /* Set up the handler for termination of child processes, and the one @@ -1810,11 +2389,15 @@ os_non_restarting_signal(SIGCHLD, main_sigchld_handler); sigterm_seen = FALSE; os_non_restarting_signal(SIGTERM, main_sigterm_handler); +os_non_restarting_signal(SIGINT, main_sigterm_handler); /* If we are to run the queue periodically, pretend the alarm has just gone -off. This will cause the first queue-runner to get kicked off straight away. */ +off. This will cause the first queue-runner to get kicked off straight away. +Get an initial sort of the list of queues, to prioritize the initial q-runs */ + -sigalrm_seen = (queue_interval > 0); +if ((sigalrm_seen = is_multiple_qrun())) + (void) next_qrunner_interval(); /* Log the start up of a daemon - at least one of listening or queue running must be set up. */ @@ -1843,19 +2426,16 @@ else if (f.daemon_listen) int smtps_ports = 0; ip_address_item * ipa; uschar * p; - uschar * qinfo = queue_interval > 0 - ? string_sprintf("-q%s", readconf_printtime(queue_interval)) - : US"no queue runs"; + const uschar * qinfo = describe_queue_runners(); /* Build a list of listening addresses in big_buffer, but limit it to 10 items. The style is for backwards compatibility. - It is now possible to have some ports listening for SMTPS (the old, - deprecated protocol that starts TLS without using STARTTLS), and others - listening for standard SMTP. Keep their listings separate. */ + It is possible to have some ports listening for SMTPS (as opposed to TLS + startted by STARTTLS), and others listening for standard SMTP. Keep their + listings separate. */ for (int j = 0, i; j < 2; j++) - { for (i = 0, ipa = addresses; i < 10 && ipa; i++, ipa = ipa->next) { /* First time round, look for SMTP ports; second time round, look for @@ -1893,11 +2473,11 @@ else if (f.daemon_listen) && Ustrcmp(ipa->address, i2->address) == 0 ) { /* found; append port to list */ - for (p = i2->log; *p; ) p++; /* end of existing string */ + for (p = i2->log; *p; ) p++; /* end of existing string { */ if (*--p == '}') *p = '\0'; /* drop EOL */ while (isdigit(*--p)) ; /* char before port */ - i2->log = *p == ':' /* no list yet? */ + i2->log = *p == ':' /* no list yet? { */ ? string_sprintf("%.*s{%s,%d}", (int)(p - i2->log + 1), i2->log, p+1, ipa->port) : string_sprintf("%s,%d}", i2->log, ipa->port); @@ -1909,7 +2489,6 @@ else if (f.daemon_listen) } } } - } p = big_buffer; for (int j = 0, i; j < 2; j++) @@ -1945,11 +2524,9 @@ else if (f.daemon_listen) version_string, qinfo, big_buffer); } -else +else /* no listening sockets, only queue-runs */ { - uschar * s = *queue_name - ? string_sprintf("-qG%s/%s", queue_name, readconf_printtime(queue_interval)) - : string_sprintf("-q%s", readconf_printtime(queue_interval)); + const uschar * s = describe_queue_runners(); log_write(0, LOG_MAIN, "exim %s daemon started: pid=%d, %s, not listening for SMTP", version_string, getpid(), s); @@ -1981,6 +2558,27 @@ malware_init(); #ifdef SUPPORT_SPF spf_init(); #endif +#ifndef DISABLE_TLS +tls_daemon_init(); +#endif + +/* Add ancillary sockets to the set for select */ + +poll_fd_count = listen_socket_count; +#ifndef DISABLE_TLS +if (tls_watch_fd >= 0) + { + tls_watch_poll = &fd_polls[poll_fd_count++]; + tls_watch_poll->fd = tls_watch_fd; + tls_watch_poll->events = POLLIN; + } +#endif +if (daemon_notifier_fd >= 0) + { + dnotify_poll = &fd_polls[poll_fd_count++]; + dnotify_poll->fd = daemon_notifier_fd; + dnotify_poll->events = POLLIN; + } /* Close the log so it can be renamed and moved. In the few cases below where this long-running process writes to the log (always exceptional conditions), it @@ -2002,14 +2600,7 @@ report_time_since(×tamp_startup, US"daemon loop start"); /* testcase 0022 * for (;;) { - #if HAVE_IPV6 - struct sockaddr_in6 accepted; - #else - struct sockaddr_in accepted; - #endif - - EXIM_SOCKLEN_T len; - pid_t pid; + int nolisten_sleep = 60; if (sigterm_seen) daemon_die(); /* Does not return */ @@ -2021,188 +2612,11 @@ for (;;) The other option is that we have an inetd wait timeout specified to -bw. */ if (sigalrm_seen) - { if (inetd_wait_timeout > 0) - { - time_t resignal_interval = inetd_wait_timeout; - - if (last_connection_time == (time_t)0) - { - DEBUG(D_any) - debug_printf("inetd wait timeout expired, but still not seen first message, ignoring\n"); - } - else - { - time_t now = time(NULL); - if (now == (time_t)-1) - { - DEBUG(D_any) debug_printf("failed to get time: %s\n", strerror(errno)); - } - else - { - if ((now - last_connection_time) >= inetd_wait_timeout) - { - DEBUG(D_any) - debug_printf("inetd wait timeout %d expired, ending daemon\n", - inetd_wait_timeout); - log_write(0, LOG_MAIN, "exim %s daemon terminating, inetd wait timeout reached.\n", - version_string); - exit(EXIT_SUCCESS); - } - else - { - resignal_interval -= (now - last_connection_time); - } - } - } - - sigalrm_seen = FALSE; - ALARM(resignal_interval); - } - + daemon_inetd_wtimeout(last_connection_time); /* Might not return */ else - { - DEBUG(D_any) debug_printf("%s received\n", -#ifdef EXPERIMENTAL_QUEUE_RAMP - *queuerun_msgid ? "qrun notification" : -#endif - "SIGALRM"); - - /* Do a full queue run in a child process, if required, unless we already - have enough queue runners on the go. If we are not running as root, a - re-exec is required. */ - - if (queue_interval > 0 && - (local_queue_run_max <= 0 || queue_run_count < local_queue_run_max)) - { - if ((pid = fork()) == 0) - { - DEBUG(D_any) debug_printf("Starting queue-runner: pid %d\n", - (int)getpid()); - - /* Disable debugging if it's required only for the daemon process. We - leave the above message, because it ties up with the "child ended" - debugging messages. */ - - if (f.debug_daemon) debug_selector = 0; - - /* Close any open listening sockets in the child */ - - if (daemon_notifier_fd >= 0) - (void) close(daemon_notifier_fd); - for (int sk = 0; sk < listen_socket_count; sk++) - (void) close(listen_sockets[sk]); - - /* Reset SIGHUP and SIGCHLD in the child in both cases. */ - - signal(SIGHUP, SIG_DFL); - signal(SIGCHLD, SIG_DFL); - signal(SIGTERM, SIG_DFL); - - /* Re-exec if privilege has been given up, unless deliver_drop_ - privilege is set. Reset SIGALRM before exec(). */ - - if (geteuid() != root_uid && !deliver_drop_privilege) - { - uschar opt[8]; - uschar *p = opt; - uschar *extra[7]; - int extracount = 1; - - signal(SIGALRM, SIG_DFL); - *p++ = '-'; - *p++ = 'q'; - if ( f.queue_2stage -#ifdef EXPERIMENTAL_QUEUE_RAMP - && !*queuerun_msgid -#endif - ) *p++ = 'q'; - if (f.queue_run_first_delivery) *p++ = 'i'; - if (f.queue_run_force) *p++ = 'f'; - if (f.deliver_force_thaw) *p++ = 'f'; - if (f.queue_run_local) *p++ = 'l'; - *p = 0; - extra[0] = *queue_name - ? string_sprintf("%sG%s", opt, queue_name) : opt; - -#ifdef EXPERIMENTAL_QUEUE_RAMP - if (*queuerun_msgid) - { - extra[extracount++] = queuerun_msgid; /* Trigger only the */ - extra[extracount++] = queuerun_msgid; /* one message */ - } -#endif - - /* If -R or -S were on the original command line, ensure they get - passed on. */ - - if (deliver_selectstring) - { - extra[extracount++] = f.deliver_selectstring_regex ? US"-Rr" : US"-R"; - extra[extracount++] = deliver_selectstring; - } - - if (deliver_selectstring_sender) - { - extra[extracount++] = f.deliver_selectstring_sender_regex - ? US"-Sr" : US"-S"; - extra[extracount++] = deliver_selectstring_sender; - } - - /* Overlay this process with a new execution. */ - - (void)child_exec_exim(CEE_EXEC_PANIC, FALSE, NULL, FALSE, extracount, - extra[0], extra[1], extra[2], extra[3], extra[4], extra[5], extra[6]); - - /* Control never returns here. */ - } - - /* No need to re-exec; SIGALRM remains set to the default handler */ - -#ifdef EXPERIMENTAL_QUEUE_RAMP - if (*queuerun_msgid) - { - f.queue_2stage = FALSE; - queue_run(queuerun_msgid, queuerun_msgid, FALSE); - } - else -#endif - queue_run(NULL, NULL, FALSE); - exim_underbar_exit(EXIT_SUCCESS); - } - - if (pid < 0) - { - log_write(0, LOG_MAIN|LOG_PANIC, "daemon: fork of queue-runner " - "process failed: %s", strerror(errno)); - log_close_all(); - } - else - { - for (int i = 0; i < local_queue_run_max; ++i) - if (queue_pid_slots[i] <= 0) - { - queue_pid_slots[i] = pid; - queue_run_count++; - break; - } - DEBUG(D_any) debug_printf("%d queue-runner process%s running\n", - queue_run_count, queue_run_count == 1 ? "" : "es"); - } - } - - /* Reset the alarm clock */ - - sigalrm_seen = FALSE; -#ifdef EXPERIMENTAL_QUEUE_RAMP - if (*queuerun_msgid) - *queuerun_msgid = 0; - else -#endif - ALARM(queue_interval); - } - - } /* sigalrm_seen */ + nolisten_sleep = + daemon_qrun(local_queue_run_max, fd_polls, listen_socket_count); /* Sleep till a connection happens if listening, and handle the connection if @@ -2216,19 +2630,8 @@ for (;;) if (f.daemon_listen) { - int lcount, select_errno; - int max_socket = 0; + int lcount; BOOL select_failed = FALSE; - fd_set select_listen; - - FD_ZERO(&select_listen); - if (daemon_notifier_fd >= 0) - FD_SET(daemon_notifier_fd, &select_listen); - for (int sk = 0; sk < listen_socket_count; sk++) - { - FD_SET(listen_sockets[sk], &select_listen); - if (listen_sockets[sk] > max_socket) max_socket = listen_sockets[sk]; - } DEBUG(D_any) debug_printf("Listening...\n"); @@ -2245,8 +2648,7 @@ for (;;) errno = EINTR; } else - lcount = select(max_socket + 1, (SELECT_ARG2_TYPE *)&select_listen, - NULL, NULL, NULL); + lcount = poll(fd_polls, poll_fd_count, -1); if (lcount < 0) { @@ -2261,14 +2663,23 @@ for (;;) old one had just finished. Preserve the errno from any select() failure for the use of the common select/accept error processing below. */ - select_errno = errno; - handle_ending_processes(); - errno = select_errno; + { + int select_errno = errno; + handle_ending_processes(); #ifndef DISABLE_TLS - /* Create or rotate any required keys */ - tls_daemon_init(); + { + int old_tfd; + /* Create or rotate any required keys; handle (delayed) filewatch event */ + + if ((old_tfd = tls_daemon_tick()) >= 0) + for (struct pollfd * p = &fd_polls[listen_socket_count]; + p < fd_polls + poll_fd_count; p++) + if (p->fd == old_tfd) { p->fd = tls_watch_fd ; break; } + } #endif + errno = select_errno; + } /* Loop for all the sockets that are currently ready to go. If select actually failed, we have set the count to 1 and select_failed=TRUE, so as @@ -2277,23 +2688,63 @@ for (;;) while (lcount-- > 0) { int accept_socket = -1; +#if HAVE_IPV6 + struct sockaddr_in6 accepted; +#else + struct sockaddr_in accepted; +#endif if (!select_failed) { - if ( daemon_notifier_fd >= 0 - && FD_ISSET(daemon_notifier_fd, &select_listen)) +#if !defined(DISABLE_TLS) && (defined(EXIM_HAVE_INOTIFY) || defined(EXIM_HAVE_KEVENT)) + if (tls_watch_poll && tls_watch_poll->revents & POLLIN) + { + tls_watch_poll->revents = 0; + tls_watch_trigger_time = time(NULL); /* Set up delayed event */ + tls_watch_discard_event(tls_watch_fd); + break; /* to top of daemon loop */ + } +#endif + if (dnotify_poll && dnotify_poll->revents & POLLIN) { - FD_CLR(daemon_notifier_fd, &select_listen); + dnotify_poll->revents = 0; sigalrm_seen = daemon_notification(); break; /* to top of daemon loop */ } - for (int sk = 0; sk < listen_socket_count; sk++) - if (FD_ISSET(listen_sockets[sk], &select_listen)) + for (struct pollfd * p = fd_polls; p < fd_polls + listen_socket_count; + p++) + if (p->revents & POLLIN) { - len = sizeof(accepted); - accept_socket = accept(listen_sockets[sk], - (struct sockaddr *)&accepted, &len); - FD_CLR(listen_sockets[sk], &select_listen); + EXIM_SOCKLEN_T alen = sizeof(accepted); +#if defined(__FreeBSD__) && defined(SO_LISTENQLEN) + int backlog; + socklen_t blen = sizeof(backlog); + + if ( smtp_backlog_monitor > 0 + && getsockopt(p->fd, SOL_SOCKET, SO_LISTENQLEN, &backlog, &blen) == 0) + { + DEBUG(D_interface) + debug_printf("listen fd %d queue curr %d\n", p->fd, backlog); + smtp_listen_backlog = backlog; + } + +#elif defined(TCP_INFO) && defined(EXIM_HAVE_TCPI_UNACKED) + struct tcp_info ti; + socklen_t tlen = sizeof(ti); + + /* If monitoring the backlog is wanted, grab for later logging */ + + smtp_listen_backlog = 0; + if ( smtp_backlog_monitor > 0 + && getsockopt(p->fd, IPPROTO_TCP, TCP_INFO, &ti, &tlen) == 0) + { + DEBUG(D_interface) debug_printf("listen fd %d queue max %u curr %u\n", + p->fd, ti.tcpi_sacked, ti.tcpi_unacked); + smtp_listen_backlog = ti.tcpi_unacked; + } +#endif + p->revents = 0; + accept_socket = accept(p->fd, (struct sockaddr *)&accepted, &alen); break; } } @@ -2315,48 +2766,46 @@ for (;;) accept_retry_errno = errno; accept_retry_select_failed = select_failed; } - else - { - if (errno != accept_retry_errno || - select_failed != accept_retry_select_failed || - accept_retry_count >= 50) - { - log_write(0, LOG_MAIN | ((accept_retry_count >= 50)? LOG_PANIC : 0), - "%d %s() failure%s: %s", - accept_retry_count, - accept_retry_select_failed? "select" : "accept", - (accept_retry_count == 1)? "" : "s", - strerror(accept_retry_errno)); - log_close_all(); - accept_retry_count = 0; - accept_retry_errno = errno; - accept_retry_select_failed = select_failed; - } - } + else if ( errno != accept_retry_errno + || select_failed != accept_retry_select_failed + || accept_retry_count >= 50) + { + log_write(0, LOG_MAIN | (accept_retry_count >= 50 ? LOG_PANIC : 0), + "%d %s() failure%s: %s", + accept_retry_count, + accept_retry_select_failed ? "select" : "accept", + accept_retry_count == 1 ? "" : "s", + strerror(accept_retry_errno)); + log_close_all(); + accept_retry_count = 0; + accept_retry_errno = errno; + accept_retry_select_failed = select_failed; + } accept_retry_count++; } - - else - { - if (accept_retry_count > 0) - { - log_write(0, LOG_MAIN, "%d %s() failure%s: %s", - accept_retry_count, - accept_retry_select_failed? "select" : "accept", - (accept_retry_count == 1)? "" : "s", - strerror(accept_retry_errno)); - log_close_all(); - accept_retry_count = 0; - } - } + else if (accept_retry_count > 0) + { + log_write(0, LOG_MAIN, "%d %s() failure%s: %s", + accept_retry_count, + accept_retry_select_failed ? "select" : "accept", + accept_retry_count == 1 ? "" : "s", + strerror(accept_retry_errno)); + log_close_all(); + accept_retry_count = 0; + } /* If select/accept succeeded, deal with the connection. */ if (accept_socket >= 0) { +#ifdef TCP_QUICKACK /* Avoid pure-ACKs while in tls protocol pingpong phase */ + /* Unfortunately we cannot be certain to do this before a TLS-on-connect + Client Hello arrives and is acked. We do it as early as possible. */ + (void) setsockopt(accept_socket, IPPROTO_TCP, TCP_QUICKACK, US &off, sizeof(off)); +#endif if (inetd_wait_timeout) last_connection_time = time(NULL); - handle_smtp_call(listen_sockets, listen_socket_count, accept_socket, + handle_smtp_call(fd_polls, listen_socket_count, accept_socket, (struct sockaddr *)&accepted); } } @@ -2371,10 +2820,8 @@ for (;;) else { - struct timeval tv; - tv.tv_sec = queue_interval; - tv.tv_usec = 0; - select(0, NULL, NULL, NULL, &tv); + struct pollfd p; + poll(&p, 0, nolisten_sleep * 1000); handle_ending_processes(); } @@ -2399,8 +2846,8 @@ for (;;) { log_write(0, LOG_MAIN, "pid %d: SIGHUP received: re-exec daemon", getpid()); - for (int sk = 0; sk < listen_socket_count; sk++) - (void)close(listen_sockets[sk]); + close_daemon_sockets(daemon_notifier_fd, fd_polls, listen_socket_count); + unlink_notifier_socket(); ALARM_CLR(0); signal(SIGHUP, SIG_IGN); sighup_argv[0] = exim_path;