X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fsocket.c;h=3a7aa94036d6123ab16c9cc7306f5af71b6a08f4;hp=075b51966f81b948c404a83b250791f97853b0a9;hb=3611581ebdabbe3a1d6a9b5310a0b59792279d7d;hpb=25366e094c749a613eb065829c424ef72c285e82 diff --git a/src/socket.c b/src/socket.c index 075b51966..3a7aa9403 100644 --- a/src/socket.c +++ b/src/socket.c @@ -1,4 +1,4 @@ -/*-*- Mode: C; c-basic-offset: 8 -*-*/ +/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/ /*** This file is part of systemd. @@ -27,9 +27,11 @@ #include #include #include +#include #include "unit.h" #include "socket.h" +#include "netinet/tcp.h" #include "log.h" #include "load-dropin.h" #include "load-fragment.h" @@ -39,6 +41,9 @@ #include "missing.h" #include "special.h" #include "bus-errors.h" +#include "label.h" +#include "exit-status.h" +#include "def.h" static const UnitActiveState state_translation_table[_SOCKET_STATE_MAX] = { [SOCKET_DEAD] = UNIT_INACTIVE, @@ -52,7 +57,7 @@ static const UnitActiveState state_translation_table[_SOCKET_STATE_MAX] = { [SOCKET_STOP_POST] = UNIT_DEACTIVATING, [SOCKET_FINAL_SIGTERM] = UNIT_DEACTIVATING, [SOCKET_FINAL_SIGKILL] = UNIT_DEACTIVATING, - [SOCKET_MAINTENANCE] = UNIT_MAINTENANCE + [SOCKET_FAILED] = UNIT_FAILED }; static void socket_init(Unit *u) { @@ -74,6 +79,8 @@ static void socket_init(Unit *u) { s->mark = -1; exec_context_init(&s->exec_context); + s->exec_context.std_output = u->meta.manager->default_std_output; + s->exec_context.std_error = u->meta.manager->default_std_error; s->control_command_id = _SOCKET_EXEC_COMMAND_INVALID; } @@ -115,18 +122,67 @@ static void socket_done(Unit *u) { s->service = NULL; + free(s->tcp_congestion); + s->tcp_congestion = NULL; + free(s->bind_to_device); s->bind_to_device = NULL; unit_unwatch_timer(u, &s->timer_watch); /* Make sure no service instance refers to us anymore. */ - LIST_FOREACH(units_per_type, i, u->meta.manager->units_per_type[UNIT_SERVICE]) { + LIST_FOREACH(units_by_type, i, u->meta.manager->units_by_type[UNIT_SERVICE]) { Service *service = (Service *) i; - if (service->socket == s) - service->socket = NULL; + if (service->accept_socket == s) + service->accept_socket = NULL; + + set_remove(service->configured_sockets, s); + } +} + +static int socket_instantiate_service(Socket *s) { + char *prefix, *name; + int r; + Unit *u; + + assert(s); + + /* This fills in s->service if it isn't filled in yet. For + * Accept=yes sockets we create the next connection service + * here. For Accept=no this is mostly a NOP since the service + * is figured out at load time anyway. */ + + if (s->service) + return 0; + + assert(s->accept); + + if (!(prefix = unit_name_to_prefix(s->meta.id))) + return -ENOMEM; + + r = asprintf(&name, "%s@%u.service", prefix, s->n_accepted); + free(prefix); + + if (r < 0) + return -ENOMEM; + + r = manager_load_unit(s->meta.manager, name, NULL, NULL, &u); + free(name); + + if (r < 0) + return r; + +#ifdef HAVE_SYSV_COMPAT + if (SERVICE(u)->sysv_path) { + log_error("Using SysV services for socket activation is not supported. Refusing."); + return -ENOENT; } +#endif + + u->meta.no_gc = true; + s->service = SERVICE(u); + return 0; } static bool have_non_accept_socket(Socket *s) { @@ -160,12 +216,22 @@ static int socket_verify(Socket *s) { return -EINVAL; } + if (s->accept && have_non_accept_socket(s)) { + log_error("%s configured for accepting sockets, but sockets are non-accepting. Refusing.", s->meta.id); + return -EINVAL; + } + if (s->accept && s->max_connections <= 0) { log_error("%s's MaxConnection setting too small. Refusing.", s->meta.id); return -EINVAL; } - if (s->exec_context.pam_name && s->kill_mode != KILL_CONTROL_GROUP) { + if (s->accept && s->service) { + log_error("Explicit service configuration for accepting sockets not supported on %s. Refusing.", s->meta.id); + return -EINVAL; + } + + if (s->exec_context.pam_name && s->exec_context.kill_mode != KILL_CONTROL_GROUP) { log_error("%s has PAM enabled. Kill mode must be set to 'control-group'. Refusing.", s->meta.id); return -EINVAL; } @@ -183,8 +249,7 @@ static bool socket_needs_mount(Socket *s, const char *prefix) { if (p->type == SOCKET_SOCKET) { if (socket_address_needs_mount(&p->address, prefix)) return true; - } else { - assert(p->type == SOCKET_FIFO); + } else if (p->type == SOCKET_FIFO || p->type == SOCKET_SPECIAL) { if (path_startswith(p->path, prefix)) return true; } @@ -218,7 +283,7 @@ static int socket_add_mount_links(Socket *s) { assert(s); - LIST_FOREACH(units_per_type, other, s->meta.manager->units_per_type[UNIT_MOUNT]) + LIST_FOREACH(units_by_type, other, s->meta.manager->units_by_type[UNIT_MOUNT]) if ((r = socket_add_one_mount_link(s, (Mount*) other)) < 0) return r; @@ -247,9 +312,13 @@ static int socket_add_default_dependencies(Socket *s) { int r; assert(s); - if (s->meta.manager->running_as == MANAGER_SYSTEM) + if (s->meta.manager->running_as == MANAGER_SYSTEM) { + if ((r = unit_add_dependency_by_name(UNIT(s), UNIT_BEFORE, SPECIAL_SOCKETS_TARGET, NULL, true)) < 0) + return r; + if ((r = unit_add_two_dependencies_by_name(UNIT(s), UNIT_AFTER, UNIT_REQUIRES, SPECIAL_SYSINIT_TARGET, NULL, true)) < 0) return r; + } return unit_add_two_dependencies_by_name(UNIT(s), UNIT_BEFORE, UNIT_CONFLICTS, SPECIAL_SHUTDOWN_TARGET, NULL, true); } @@ -268,8 +337,10 @@ static int socket_load(Unit *u) { if (u->meta.load_state == UNIT_LOADED) { if (have_non_accept_socket(s)) { - if ((r = unit_load_related_unit(u, ".service", (Unit**) &s->service)) < 0) - return r; + + if (!s->service) + if ((r = unit_load_related_unit(u, ".service", (Unit**) &s->service)) < 0) + return r; if ((r = unit_add_dependency(u, UNIT_BEFORE, UNIT(s->service), true)) < 0) return r; @@ -284,7 +355,7 @@ static int socket_load(Unit *u) { if ((r = unit_add_exec_dependencies(u, &s->exec_context)) < 0) return r; - if ((r = unit_add_default_cgroup(u)) < 0) + if ((r = unit_add_default_cgroups(u)) < 0) return r; if (s->meta.default_dependencies) @@ -295,7 +366,10 @@ static int socket_load(Unit *u) { return socket_verify(s); } -static const char* listen_lookup(int type) { +static const char* listen_lookup(int family, int type) { + + if (family == AF_NETLINK) + return "ListenNetlink"; if (type == SOCK_STREAM) return "ListenStream"; @@ -326,19 +400,23 @@ static void socket_dump(Unit *u, FILE *f, const char *prefix) { "%sSocket State: %s\n" "%sBindIPv6Only: %s\n" "%sBacklog: %u\n" - "%sKillMode: %s\n" "%sSocketMode: %04o\n" "%sDirectoryMode: %04o\n" "%sKeepAlive: %s\n" - "%sFreeBind: %s\n", + "%sFreeBind: %s\n" + "%sTransparent: %s\n" + "%sBroadcast: %s\n" + "%sTCPCongestion: %s\n", prefix, socket_state_to_string(s->state), prefix, socket_address_bind_ipv6_only_to_string(s->bind_ipv6_only), prefix, s->backlog, - prefix, kill_mode_to_string(s->kill_mode), prefix, s->socket_mode, prefix, s->directory_mode, prefix, yes_no(s->keep_alive), - prefix, yes_no(s->free_bind)); + prefix, yes_no(s->free_bind), + prefix, yes_no(s->transparent), + prefix, yes_no(s->broadcast), + prefix, strna(s->tcp_congestion)); if (s->control_pid > 0) fprintf(f, @@ -394,21 +472,35 @@ static void socket_dump(Unit *u, FILE *f, const char *prefix) { "%sMark: %i\n", prefix, s->mark); + if (s->mq_maxmsg > 0) + fprintf(f, + "%sMessageQueueMaxMessages: %li\n", + prefix, s->mq_maxmsg); + + if (s->mq_msgsize > 0) + fprintf(f, + "%sMessageQueueMessageSize: %li\n", + prefix, s->mq_msgsize); + LIST_FOREACH(port, p, s->ports) { if (p->type == SOCKET_SOCKET) { const char *t; int r; - char *k; + char *k = NULL; if ((r = socket_address_print(&p->address, &k)) < 0) t = strerror(-r); else t = k; - fprintf(f, "%s%s: %s\n", prefix, listen_lookup(p->address.type), k); + fprintf(f, "%s%s: %s\n", prefix, listen_lookup(socket_address_family(&p->address), p->address.type), t); free(k); - } else + } else if (p->type == SOCKET_SPECIAL) + fprintf(f, "%sListenSpecial: %s\n", prefix, p->path); + else if (p->type == SOCKET_MQUEUE) + fprintf(f, "%sListenMessageQueue: %s\n", prefix, p->path); + else fprintf(f, "%sListenFIFO: %s\n", prefix, p->path); } @@ -457,8 +549,7 @@ static int instance_from_socket(int fd, unsigned nr, char **instance) { b = ntohl(remote.in.sin_addr.s_addr); if (asprintf(&r, - "%u-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u", - nr, + "%u.%u.%u.%u:%u-%u.%u.%u.%u:%u", a >> 24, (a >> 16) & 0xFF, (a >> 8) & 0xFF, a & 0xFF, ntohs(local.in.sin_port), b >> 24, (b >> 16) & 0xFF, (b >> 8) & 0xFF, b & 0xFF, @@ -480,8 +571,7 @@ static int instance_from_socket(int fd, unsigned nr, char **instance) { *b = remote.in6.sin6_addr.s6_addr+12; if (asprintf(&r, - "%u-%u.%u.%u.%u:%u-%u.%u.%u.%u:%u", - nr, + "%u.%u.%u.%u:%u-%u.%u.%u.%u:%u", a[0], a[1], a[2], a[3], ntohs(local.in6.sin6_port), b[0], b[1], b[2], b[3], @@ -491,8 +581,7 @@ static int instance_from_socket(int fd, unsigned nr, char **instance) { char a[INET6_ADDRSTRLEN], b[INET6_ADDRSTRLEN]; if (asprintf(&r, - "%u-%s:%u-%s:%u", - nr, + "%s:%u-%s:%u", inet_ntop(AF_INET6, &local.in6.sin6_addr, a, sizeof(a)), ntohs(local.in6.sin6_port), inet_ntop(AF_INET6, &remote.in6.sin6_addr, b, sizeof(b)), @@ -562,20 +651,26 @@ static void socket_apply_socket_options(Socket *s, int fd) { log_warning("SO_KEEPALIVE failed: %m"); } + if (s->broadcast) { + int one = 1; + if (setsockopt(fd, SOL_SOCKET, SO_BROADCAST, &one, sizeof(one)) < 0) + log_warning("SO_BROADCAST failed: %m"); + } + if (s->priority >= 0) if (setsockopt(fd, SOL_SOCKET, SO_PRIORITY, &s->priority, sizeof(s->priority)) < 0) log_warning("SO_PRIORITY failed: %m"); if (s->receive_buffer > 0) { int value = (int) s->receive_buffer; - if (setsockopt(fd, SOL_SOCKET, SO_RCVBUF, &value, sizeof(value)) < 0) - log_warning("SO_RCVBUF failed: %m"); + if (setsockopt(fd, SOL_SOCKET, SO_RCVBUFFORCE, &value, sizeof(value)) < 0) + log_warning("SO_RCVBUFFORCE failed: %m"); } if (s->send_buffer > 0) { int value = (int) s->send_buffer; - if (setsockopt(fd, SOL_SOCKET, SO_SNDBUF, &value, sizeof(value)) < 0) - log_warning("SO_SNDBUF failed: %m"); + if (setsockopt(fd, SOL_SOCKET, SO_SNDBUFFORCE, &value, sizeof(value)) < 0) + log_warning("SO_SNDBUFFORCE failed: %m"); } if (s->mark >= 0) @@ -590,14 +685,24 @@ static void socket_apply_socket_options(Socket *s, int fd) { int r, x; r = setsockopt(fd, IPPROTO_IP, IP_TTL, &s->ip_ttl, sizeof(s->ip_ttl)); - x = setsockopt(fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &s->ip_ttl, sizeof(s->ip_ttl)); + + if (socket_ipv6_is_supported()) + x = setsockopt(fd, IPPROTO_IPV6, IPV6_UNICAST_HOPS, &s->ip_ttl, sizeof(s->ip_ttl)); + else { + x = -1; + errno = EAFNOSUPPORT; + } if (r < 0 && x < 0) log_warning("IP_TTL/IPV6_UNICAST_HOPS failed: %m"); } + + if (s->tcp_congestion) + if (setsockopt(fd, SOL_TCP, TCP_CONGESTION, s->tcp_congestion, strlen(s->tcp_congestion)+1) < 0) + log_warning("TCP_CONGESTION failed: %m"); } -static void socket_apply_pipe_options(Socket *s, int fd) { +static void socket_apply_fifo_options(Socket *s, int fd) { assert(s); assert(fd >= 0); @@ -606,9 +711,172 @@ static void socket_apply_pipe_options(Socket *s, int fd) { log_warning("F_SETPIPE_SZ: %m"); } +static int fifo_address_create( + const char *path, + mode_t directory_mode, + mode_t socket_mode, + int *_fd) { + + int fd = -1, r = 0; + struct stat st; + mode_t old_mask; + + assert(path); + assert(_fd); + + mkdir_parents(path, directory_mode); + + if ((r = label_fifofile_set(path)) < 0) + goto fail; + + /* Enforce the right access mode for the fifo */ + old_mask = umask(~ socket_mode); + + /* Include the original umask in our mask */ + umask(~socket_mode | old_mask); + + r = mkfifo(path, socket_mode); + umask(old_mask); + + if (r < 0 && errno != EEXIST) { + r = -errno; + goto fail; + } + + if ((fd = open(path, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW)) < 0) { + r = -errno; + goto fail; + } + + label_file_clear(); + + if (fstat(fd, &st) < 0) { + r = -errno; + goto fail; + } + + if (!S_ISFIFO(st.st_mode) || + (st.st_mode & 0777) != (socket_mode & ~old_mask) || + st.st_uid != getuid() || + st.st_gid != getgid()) { + + r = -EEXIST; + goto fail; + } + + *_fd = fd; + return 0; + +fail: + label_file_clear(); + + if (fd >= 0) + close_nointr_nofail(fd); + + return r; +} + +static int special_address_create( + const char *path, + int *_fd) { + + int fd = -1, r = 0; + struct stat st; + + assert(path); + assert(_fd); + + if ((fd = open(path, O_RDONLY|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW)) < 0) { + r = -errno; + goto fail; + } + + if (fstat(fd, &st) < 0) { + r = -errno; + goto fail; + } + + /* Check whether this is a /proc, /sys or /dev file or char device */ + if (!S_ISREG(st.st_mode) && !S_ISCHR(st.st_mode)) { + r = -EEXIST; + goto fail; + } + + *_fd = fd; + return 0; + +fail: + if (fd >= 0) + close_nointr_nofail(fd); + + return r; +} + +static int mq_address_create( + const char *path, + mode_t mq_mode, + long maxmsg, + long msgsize, + int *_fd) { + + int fd = -1, r = 0; + struct stat st; + mode_t old_mask; + struct mq_attr _attr, *attr = NULL; + + assert(path); + assert(_fd); + + if (maxmsg > 0 && msgsize > 0) { + zero(_attr); + _attr.mq_flags = O_NONBLOCK; + _attr.mq_maxmsg = maxmsg; + _attr.mq_msgsize = msgsize; + attr = &_attr; + } + + /* Enforce the right access mode for the mq */ + old_mask = umask(~ mq_mode); + + /* Include the original umask in our mask */ + umask(~mq_mode | old_mask); + + fd = mq_open(path, O_RDONLY|O_CLOEXEC|O_NONBLOCK|O_CREAT, mq_mode, attr); + umask(old_mask); + + if (fd < 0 && errno != EEXIST) { + r = -errno; + goto fail; + } + + if (fstat(fd, &st) < 0) { + r = -errno; + goto fail; + } + + if ((st.st_mode & 0777) != (mq_mode & ~old_mask) || + st.st_uid != getuid() || + st.st_gid != getgid()) { + + r = -EEXIST; + goto fail; + } + + *_fd = fd; + return 0; + +fail: + if (fd >= 0) + close_nointr_nofail(fd); + + return r; +} + static int socket_open_fds(Socket *s) { SocketPort *p; int r; + char *label = NULL; + bool know_label = false; assert(s); @@ -619,55 +887,71 @@ static int socket_open_fds(Socket *s) { if (p->type == SOCKET_SOCKET) { + if (!know_label) { + + if ((r = socket_instantiate_service(s)) < 0) + return r; + + if (s->service && s->service->exec_command[SERVICE_EXEC_START]) + if ((r = label_get_socket_label_from_exe(s->service->exec_command[SERVICE_EXEC_START]->path, &label)) < 0) { + if (r != -EPERM) + return r; + } + + know_label = true; + } + if ((r = socket_address_listen( &p->address, s->backlog, s->bind_ipv6_only, s->bind_to_device, s->free_bind, + s->transparent, s->directory_mode, s->socket_mode, + label, &p->fd)) < 0) goto rollback; socket_apply_socket_options(s, p->fd); - } else { - struct stat st; - assert(p->type == SOCKET_FIFO); + } else if (p->type == SOCKET_SPECIAL) { - mkdir_parents(p->path, s->directory_mode); - - if (mkfifo(p->path, s->socket_mode) < 0 && errno != EEXIST) { - r = -errno; + if ((r = special_address_create( + p->path, + &p->fd)) < 0) goto rollback; - } - if ((p->fd = open(p->path, O_RDWR|O_CLOEXEC|O_NOCTTY|O_NONBLOCK|O_NOFOLLOW)) < 0) { - r = -errno; - goto rollback; - } + } else if (p->type == SOCKET_FIFO) { - if (fstat(p->fd, &st) < 0) { - r = -errno; + if ((r = fifo_address_create( + p->path, + s->directory_mode, + s->socket_mode, + &p->fd)) < 0) goto rollback; - } - /* FIXME verify user, access mode */ + socket_apply_fifo_options(s, p->fd); + } else if (p->type == SOCKET_MQUEUE) { - if (!S_ISFIFO(st.st_mode)) { - r = -EEXIST; + if ((r = mq_address_create( + p->path, + s->socket_mode, + s->mq_maxmsg, + s->mq_msgsize, + &p->fd)) < 0) goto rollback; - } - - socket_apply_pipe_options(s, p->fd); - } + } else + assert_not_reached("Unknown port type"); } + label_free(label); return 0; rollback: socket_close_fds(s); + label_free(label); return r; } @@ -748,7 +1032,7 @@ static void socket_set_state(Socket *s, SocketState state) { socket_state_to_string(old_state), socket_state_to_string(state)); - unit_notify(UNIT(s), state_translation_table[old_state], state_translation_table[state]); + unit_notify(UNIT(s), state_translation_table[old_state], state_translation_table[state], true); } static int socket_coldplug(Unit *u) { @@ -851,7 +1135,7 @@ static void socket_enter_dead(Socket *s, bool success) { if (!success) s->failure = true; - socket_set_state(s, s->failure ? SOCKET_MAINTENANCE : SOCKET_DEAD); + socket_set_state(s, s->failure ? SOCKET_FAILED : SOCKET_DEAD); } static void socket_enter_signal(Socket *s, SocketState state, bool success); @@ -884,33 +1168,49 @@ fail: static void socket_enter_signal(Socket *s, SocketState state, bool success) { int r; - bool sent = false; + Set *pid_set = NULL; + bool wait_for_exit = false; assert(s); if (!success) s->failure = true; - if (s->kill_mode != KILL_NONE) { - int sig = (state == SOCKET_STOP_PRE_SIGTERM || state == SOCKET_FINAL_SIGTERM) ? SIGTERM : SIGKILL; + if (s->exec_context.kill_mode != KILL_NONE) { + int sig = (state == SOCKET_STOP_PRE_SIGTERM || state == SOCKET_FINAL_SIGTERM) ? s->exec_context.kill_signal : SIGKILL; - if (s->kill_mode == KILL_CONTROL_GROUP) { + if (s->control_pid > 0) { + if (kill_and_sigcont(s->control_pid, sig) < 0 && errno != ESRCH) - if ((r = cgroup_bonding_kill_list(s->meta.cgroup_bondings, sig)) < 0) { - if (r != -EAGAIN && r != -ESRCH) - goto fail; - } else - sent = true; + log_warning("Failed to kill control process %li: %m", (long) s->control_pid); + else + wait_for_exit = true; } - if (!sent && s->control_pid > 0) - if (kill(s->kill_mode == KILL_PROCESS ? s->control_pid : -s->control_pid, sig) < 0 && errno != ESRCH) { - r = -errno; + if (s->exec_context.kill_mode == KILL_CONTROL_GROUP) { + + if (!(pid_set = set_new(trivial_hash_func, trivial_compare_func))) { + r = -ENOMEM; goto fail; } + + /* Exclude the control pid from being killed via the cgroup */ + if (s->control_pid > 0) + if ((r = set_put(pid_set, LONG_TO_PTR(s->control_pid))) < 0) + goto fail; + + if ((r = cgroup_bonding_kill_list(s->meta.cgroup_bondings, sig, true, pid_set)) < 0) { + if (r != -EAGAIN && r != -ESRCH && r != -ENOENT) + log_warning("Failed to kill control group: %s", strerror(-r)); + } else if (r > 0) + wait_for_exit = true; + + set_free(pid_set); + pid_set = NULL; + } } - if (sent && s->control_pid > 0) { + if (wait_for_exit) { if ((r = unit_watch_timer(UNIT(s), s->timeout_usec, &s->timer_watch)) < 0) goto fail; @@ -929,6 +1229,9 @@ fail: socket_enter_stop_post(s, false); else socket_enter_dead(s, false); + + if (pid_set) + set_free(pid_set); } static void socket_enter_stop_pre(Socket *s, bool success) { @@ -1032,14 +1335,53 @@ static void socket_enter_running(Socket *s, int cfd) { assert(s); dbus_error_init(&error); + /* We don't take connections anymore if we are supposed to + * shut down anyway */ + if (unit_pending_inactive(UNIT(s))) { + log_debug("Suppressing connection request on %s since unit stop is scheduled.", s->meta.id); + + if (cfd >= 0) + close_nointr_nofail(cfd); + else { + /* Flush all sockets by closing and reopening them */ + socket_close_fds(s); + + if ((r = socket_watch_fds(s)) < 0) { + log_warning("%s failed to watch sockets: %s", s->meta.id, strerror(-r)); + socket_enter_stop_pre(s, false); + } + } + + return; + } + if (cfd < 0) { - if ((r = manager_add_job(s->meta.manager, JOB_START, UNIT(s->service), JOB_REPLACE, true, &error, NULL)) < 0) - goto fail; + bool pending = false; + Meta *i; + + /* If there's already a start pending don't bother to + * do anything */ + LIST_FOREACH(units_by_type, i, s->meta.manager->units_by_type[UNIT_SERVICE]) { + Service *service = (Service *) i; + + if (!set_get(service->configured_sockets, s)) + continue; + + if (!unit_pending_active(UNIT(service))) + continue; + + pending = true; + break; + } + + if (!pending) + if ((r = manager_add_job(s->meta.manager, JOB_START, UNIT(s->service), JOB_REPLACE, true, &error, NULL)) < 0) + goto fail; socket_set_state(s, SOCKET_RUNNING); } else { - Unit *u; char *prefix, *instance = NULL, *name; + Service *service; if (s->n_connections >= s->max_connections) { log_warning("Too many incoming connections (%u)", s->n_connections); @@ -1047,7 +1389,10 @@ static void socket_enter_running(Socket *s, int cfd) { return; } - if ((r = instance_from_socket(cfd, s->n_accepted++, &instance)) < 0) + if ((r = socket_instantiate_service(s)) < 0) + goto fail; + + if ((r = instance_from_socket(cfd, s->n_accepted, &instance)) < 0) goto fail; if (!(prefix = unit_name_to_prefix(s->meta.id))) { @@ -1065,21 +1410,31 @@ static void socket_enter_running(Socket *s, int cfd) { goto fail; } - r = manager_load_unit(s->meta.manager, name, NULL, NULL, &u); - free(name); - - if (r < 0) + if ((r = unit_add_name(UNIT(s->service), name)) < 0) { + free(name); goto fail; + } + + service = s->service; + s->service = NULL; + s->n_accepted ++; + + service->meta.no_gc = false; - if ((r = service_set_socket_fd(SERVICE(u), cfd, s)) < 0) + unit_choose_id(UNIT(service), name); + free(name); + + if ((r = service_set_socket_fd(service, cfd, s)) < 0) goto fail; cfd = -1; - s->n_connections ++; - if ((r = manager_add_job(u->meta.manager, JOB_START, u, JOB_REPLACE, true, &error, NULL)) < 0) + if ((r = manager_add_job(s->meta.manager, JOB_START, UNIT(service), JOB_REPLACE, true, &error, NULL)) < 0) goto fail; + + /* Notify clients about changed counters */ + unit_add_to_dbus_queue(UNIT(s)); } return; @@ -1145,18 +1500,29 @@ static int socket_start(Unit *u) { /* Cannot run this without the service being around */ if (s->service) { - if (s->service->meta.load_state != UNIT_LOADED) + if (s->service->meta.load_state != UNIT_LOADED) { + log_error("Socket service %s not loaded, refusing.", s->service->meta.id); return -ENOENT; + } - /* If the service is alredy actvie we cannot start the + /* If the service is already active we cannot start the * socket */ if (s->service->state != SERVICE_DEAD && - s->service->state != SERVICE_MAINTENANCE && - s->service->state != SERVICE_AUTO_RESTART) + s->service->state != SERVICE_FAILED && + s->service->state != SERVICE_AUTO_RESTART) { + log_error("Socket service %s already active, refusing.", s->service->meta.id); return -EBUSY; + } + +#ifdef HAVE_SYSV_COMPAT + if (s->service->sysv_path) { + log_error("Using SysV services for socket activation is not supported. Refusing."); + return -ENOENT; + } +#endif } - assert(s->state == SOCKET_DEAD || s->state == SOCKET_MAINTENANCE); + assert(s->state == SOCKET_DEAD || s->state == SOCKET_FAILED); s->failure = false; socket_enter_start_pre(s); @@ -1168,21 +1534,23 @@ static int socket_stop(Unit *u) { assert(s); - /* We cannot fulfill this request right now, try again later - * please! */ - if (s->state == SOCKET_START_PRE || - s->state == SOCKET_START_POST) - return -EAGAIN; - /* Already on it */ if (s->state == SOCKET_STOP_PRE || s->state == SOCKET_STOP_PRE_SIGTERM || s->state == SOCKET_STOP_PRE_SIGKILL || s->state == SOCKET_STOP_POST || s->state == SOCKET_FINAL_SIGTERM || - s->state == SOCKET_FINAL_SIGTERM) + s->state == SOCKET_FINAL_SIGKILL) return 0; + /* If there's already something running we go directly into + * kill mode. */ + if (s->state == SOCKET_START_PRE || + s->state == SOCKET_START_POST) { + socket_enter_signal(s, SOCKET_STOP_PRE_SIGTERM, true); + return -EAGAIN; + } + assert(s->state == SOCKET_LISTENING || s->state == SOCKET_RUNNING); socket_enter_stop_pre(s, true); @@ -1223,9 +1591,14 @@ static int socket_serialize(Unit *u, FILE *f, FDSet *fds) { if ((r = socket_address_print(&p->address, &t)) < 0) return r; - unit_serialize_item_format(u, f, "socket", "%i %i %s", copy, p->address.type, t); + if (socket_address_family(&p->address) == AF_NETLINK) + unit_serialize_item_format(u, f, "netlink", "%i %s", copy, t); + else + unit_serialize_item_format(u, f, "socket", "%i %i %s", copy, p->address.type, t); free(t); - } else { + } else if (p->type == SOCKET_SPECIAL) + unit_serialize_item_format(u, f, "special", "%i %s", copy, p->path); + else { assert(p->type == SOCKET_FIFO); unit_serialize_item_format(u, f, "fifo", "%i %s", copy, p->path); } @@ -1236,7 +1609,6 @@ static int socket_serialize(Unit *u, FILE *f, FDSet *fds) { static int socket_deserialize_item(Unit *u, const char *key, const char *value, FDSet *fds) { Socket *s = SOCKET(u); - int r; assert(u); assert(key); @@ -1261,14 +1633,14 @@ static int socket_deserialize_item(Unit *u, const char *key, const char *value, } else if (streq(key, "n-accepted")) { unsigned k; - if ((r = safe_atou(value, &k)) < 0) + if (safe_atou(value, &k) < 0) log_debug("Failed to parse n-accepted value %s", value); else s->n_accepted += k; } else if (streq(key, "control-pid")) { pid_t pid; - if ((r = parse_pid(value, &pid)) < 0) + if (parse_pid(value, &pid) < 0) log_debug("Failed to parse control-pid value %s", value); else s->control_pid = pid; @@ -1290,7 +1662,28 @@ static int socket_deserialize_item(Unit *u, const char *key, const char *value, else { LIST_FOREACH(port, p, s->ports) - if (streq(p->path, value+skip)) + if (p->type == SOCKET_FIFO && + streq_ptr(p->path, value+skip)) + break; + + if (p) { + if (p->fd >= 0) + close_nointr_nofail(p->fd); + p->fd = fdset_remove(fds, fd); + } + } + + } else if (streq(key, "special")) { + int fd, skip = 0; + SocketPort *p; + + if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd)) + log_debug("Failed to parse special value %s", value); + else { + + LIST_FOREACH(port, p, s->ports) + if (p->type == SOCKET_SPECIAL && + streq_ptr(p->path, value+skip)) break; if (p) { @@ -1319,6 +1712,25 @@ static int socket_deserialize_item(Unit *u, const char *key, const char *value, } } + } else if (streq(key, "netlink")) { + int fd, skip = 0; + SocketPort *p; + + if (sscanf(value, "%i %n", &fd, &skip) < 1 || fd < 0 || !fdset_contains(fds, fd)) + log_debug("Failed to parse socket value %s", value); + else { + + LIST_FOREACH(port, p, s->ports) + if (socket_address_is_netlink(&p->address, value+skip)) + break; + + if (p) { + if (p->fd >= 0) + close_nointr_nofail(p->fd); + p->fd = fdset_remove(fds, fd); + } + } + } else log_debug("Unknown serialization key '%s'", key); @@ -1358,7 +1770,12 @@ static void socket_fd_event(Unit *u, int fd, uint32_t events, Watch *w) { log_debug("Incoming traffic on %s", u->meta.id); if (events != EPOLLIN) { - log_error("Got invalid poll event on socket."); + + if (events & EPOLLHUP) + log_error("%s: Got POLLHUP on a listening socket. The service probably invoked shutdown() on it, and should better not do that.", u->meta.id); + else + log_error("%s: Got unexpected poll event (0x%x) on socket.", u->meta.id, events); + goto fail; } @@ -1400,12 +1817,17 @@ static void socket_sigchld_event(Unit *u, pid_t pid, int code, int status) { s->control_pid = 0; success = is_clean_exit(code, status); - s->failure = s->failure || !success; - if (s->control_command) - exec_status_exit(&s->control_command->exec_status, pid, code, status); + if (s->control_command) { + exec_status_exit(&s->control_command->exec_status, &s->exec_context, pid, code, status); - log_debug("%s control process exited, code=%s status=%i", u->meta.id, sigchld_code_to_string(code), status); + if (s->control_command->ignore) + success = true; + } + + log_full(success ? LOG_DEBUG : LOG_NOTICE, + "%s control process exited, code=%s status=%i", u->meta.id, sigchld_code_to_string(code), status); + s->failure = s->failure || !success; if (s->control_command && s->control_command->command_next && success) { log_debug("%s running next command for state %s", u->meta.id, socket_state_to_string(s->state)); @@ -1451,6 +1873,9 @@ static void socket_sigchld_event(Unit *u, pid_t pid, int code, int status) { assert_not_reached("Uh, control process died at wrong time."); } } + + /* Notify clients about changed exit status */ + unit_add_to_dbus_queue(u); } static void socket_timer_event(Unit *u, uint64_t elapsed, Watch *w) { @@ -1465,6 +1890,7 @@ static void socket_timer_event(Unit *u, uint64_t elapsed, Watch *w) { case SOCKET_START_PRE: log_warning("%s starting timed out. Terminating.", u->meta.id); socket_enter_signal(s, SOCKET_FINAL_SIGTERM, false); + break; case SOCKET_START_POST: log_warning("%s starting timed out. Stopping.", u->meta.id); @@ -1477,8 +1903,13 @@ static void socket_timer_event(Unit *u, uint64_t elapsed, Watch *w) { break; case SOCKET_STOP_PRE_SIGTERM: - log_warning("%s stopping timed out. Killing.", u->meta.id); - socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, false); + if (s->exec_context.send_sigkill) { + log_warning("%s stopping timed out. Killing.", u->meta.id); + socket_enter_signal(s, SOCKET_STOP_PRE_SIGKILL, false); + } else { + log_warning("%s stopping timed out. Skipping SIGKILL. Ignoring.", u->meta.id); + socket_enter_stop_post(s, false); + } break; case SOCKET_STOP_PRE_SIGKILL: @@ -1492,12 +1923,17 @@ static void socket_timer_event(Unit *u, uint64_t elapsed, Watch *w) { break; case SOCKET_FINAL_SIGTERM: - log_warning("%s stopping timed out (2). Killing.", u->meta.id); - socket_enter_signal(s, SOCKET_FINAL_SIGKILL, false); + if (s->exec_context.send_sigkill) { + log_warning("%s stopping timed out (2). Killing.", u->meta.id); + socket_enter_signal(s, SOCKET_FINAL_SIGKILL, false); + } else { + log_warning("%s stopping timed out (2). Skipping SIGKILL. Ignoring.", u->meta.id); + socket_enter_dead(s, false); + } break; case SOCKET_FINAL_SIGKILL: - log_warning("%s still around after SIGKILL (2). Entering maintenance mode.", u->meta.id); + log_warning("%s still around after SIGKILL (2). Entering failed mode.", u->meta.id); socket_enter_dead(s, false); break; @@ -1522,7 +1958,7 @@ int socket_collect_fds(Socket *s, int **fds, unsigned *n_fds) { if (p->fd >= 0) rn_fds++; - if (!(rfds = new(int, rn_fds)) < 0) + if (!(rfds = new(int, rn_fds))) return -ENOMEM; k = 0; @@ -1557,7 +1993,7 @@ void socket_connection_unref(Socket *s) { /* The service is dead. Yay! * - * This is strictly for one-onstance-per-connection + * This is strictly for one-instance-per-connection * services. */ assert(s->n_connections > 0); @@ -1566,6 +2002,64 @@ void socket_connection_unref(Socket *s) { log_debug("%s: One connection closed, %u left.", s->meta.id, s->n_connections); } +static void socket_reset_failed(Unit *u) { + Socket *s = SOCKET(u); + + assert(s); + + if (s->state == SOCKET_FAILED) + socket_set_state(s, SOCKET_DEAD); + + s->failure = false; +} + +static int socket_kill(Unit *u, KillWho who, KillMode mode, int signo, DBusError *error) { + Socket *s = SOCKET(u); + int r = 0; + Set *pid_set = NULL; + + assert(s); + + if (who == KILL_MAIN) { + dbus_set_error(error, BUS_ERROR_NO_SUCH_PROCESS, "Socket units have no main processes"); + return -EINVAL; + } + + if (s->control_pid <= 0 && who == KILL_CONTROL) { + dbus_set_error(error, BUS_ERROR_NO_SUCH_PROCESS, "No control process to kill"); + return -ENOENT; + } + + if (who == KILL_CONTROL || who == KILL_ALL) + if (s->control_pid > 0) + if (kill(s->control_pid, signo) < 0) + r = -errno; + + if (who == KILL_ALL && mode == KILL_CONTROL_GROUP) { + int q; + + if (!(pid_set = set_new(trivial_hash_func, trivial_compare_func))) + return -ENOMEM; + + /* Exclude the control pid from being killed via the cgroup */ + if (s->control_pid > 0) + if ((q = set_put(pid_set, LONG_TO_PTR(s->control_pid))) < 0) { + r = q; + goto finish; + } + + if ((q = cgroup_bonding_kill_list(s->meta.cgroup_bondings, signo, false, pid_set)) < 0) + if (q != -EAGAIN && q != -ESRCH && q != -ENOENT) + r = q; + } + +finish: + if (pid_set) + set_free(pid_set); + + return r; +} + static const char* const socket_state_table[_SOCKET_STATE_MAX] = { [SOCKET_DEAD] = "dead", [SOCKET_START_PRE] = "start-pre", @@ -1578,7 +2072,7 @@ static const char* const socket_state_table[_SOCKET_STATE_MAX] = { [SOCKET_STOP_POST] = "stop-post", [SOCKET_FINAL_SIGTERM] = "final-sigterm", [SOCKET_FINAL_SIGKILL] = "final-sigkill", - [SOCKET_MAINTENANCE] = "maintenance" + [SOCKET_FAILED] = "failed" }; DEFINE_STRING_TABLE_LOOKUP(socket_state, SocketState); @@ -1599,6 +2093,8 @@ const UnitVTable socket_vtable = { .done = socket_done, .load = socket_load, + .kill = socket_kill, + .coldplug = socket_coldplug, .dump = socket_dump, @@ -1618,5 +2114,9 @@ const UnitVTable socket_vtable = { .sigchld_event = socket_sigchld_event, .timer_event = socket_timer_event, - .bus_message_handler = bus_socket_message_handler + .reset_failed = socket_reset_failed, + + .bus_interface = "org.freedesktop.systemd1.Socket", + .bus_message_handler = bus_socket_message_handler, + .bus_invalidating_properties = bus_socket_invalidating_properties };