chiark / gitweb /
socket: Support IPv6-less systems with runtime check.
[elogind.git] / src / socket.c
index 78fc049a95773cf7af9615ff07076711144e95b1..aacf9bed9f1f50f106db3cd78b322461fc689ef8 100644 (file)
@@ -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.
@@ -30,6 +30,7 @@
 
 #include "unit.h"
 #include "socket.h"
+#include "netinet/tcp.h"
 #include "log.h"
 #include "load-dropin.h"
 #include "load-fragment.h"
@@ -39,6 +40,7 @@
 #include "missing.h"
 #include "special.h"
 #include "bus-errors.h"
+#include "label.h"
 
 static const UnitActiveState state_translation_table[_SOCKET_STATE_MAX] = {
         [SOCKET_DEAD] = UNIT_INACTIVE,
@@ -52,7 +54,7 @@ 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) {
@@ -115,6 +117,9 @@ 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;
 
@@ -129,6 +134,43 @@ static void socket_done(Unit *u) {
         }
 }
 
+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;
+
+        u->meta.no_gc = true;
+        s->service = SERVICE(u);
+        return 0;
+}
+
 static bool have_non_accept_socket(Socket *s) {
         SocketPort *p;
 
@@ -256,7 +298,7 @@ static int socket_add_default_dependencies(Socket *s) {
                 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);
+        return unit_add_two_dependencies_by_name(UNIT(s), UNIT_BEFORE, UNIT_CONFLICTED_BY, SPECIAL_SHUTDOWN_TARGET, NULL, true);
 }
 
 static int socket_load(Unit *u) {
@@ -334,14 +376,16 @@ static void socket_dump(Unit *u, FILE *f, const char *prefix) {
                 "%sSocketMode: %04o\n"
                 "%sDirectoryMode: %04o\n"
                 "%sKeepAlive: %s\n"
-                "%sFreeBind: %s\n",
+                "%sFreeBind: %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, 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, strna(s->tcp_congestion));
 
         if (s->control_pid > 0)
                 fprintf(f,
@@ -402,14 +446,14 @@ static void socket_dump(Unit *u, FILE *f, const char *prefix) {
                 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(p->address.type), t);
                         free(k);
                 } else
                         fprintf(f, "%sListenFIFO: %s\n", prefix, p->path);
@@ -460,8 +504,7 @@ 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,
@@ -483,8 +526,7 @@ 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],
@@ -494,8 +536,7 @@ 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)),
@@ -593,14 +634,24 @@ 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);
 
@@ -609,9 +660,77 @@ 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) {
+                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 socket_open_fds(Socket *s) {
         SocketPort *p;
         int r;
+        char *label = NULL;
+        bool know_label = false;
 
         assert(s);
 
@@ -622,6 +741,17 @@ 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 ((r = label_get_socket_label_from_exe(s->service->exec_command[SERVICE_EXEC_START]->path, &label)) < 0)
+                                        return r;
+
+                                know_label = true;
+                        }
+
                         if ((r = socket_address_listen(
                                              &p->address,
                                              s->backlog,
@@ -630,47 +760,33 @@ static int socket_open_fds(Socket *s) {
                                              s->free_bind,
                                              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);
-
-                        mkdir_parents(p->path, s->directory_mode);
-
-                        if (mkfifo(p->path, s->socket_mode) < 0 && errno != EEXIST) {
-                                r = -errno;
-                                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 */
-
-                        if (!S_ISFIFO(st.st_mode)) {
-                                r = -EEXIST;
-                                goto rollback;
-                        }
+                        socket_apply_fifo_options(s, p->fd);
 
-                        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;
 }
 
@@ -854,7 +970,7 @@ 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);
@@ -887,7 +1003,8 @@ 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);
 
@@ -897,23 +1014,39 @@ static void socket_enter_signal(Socket *s, SocketState state, bool success) {
         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->exec_context.kill_mode == KILL_CONTROL_GROUP) {
+                if (s->control_pid > 0) {
+                        if (kill(s->exec_context.kill_mode == KILL_PROCESS_GROUP ?
+                                 -s->control_pid :
+                                 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->exec_context.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, 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);
+                }
         }
 
-        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;
 
@@ -932,6 +1065,9 @@ 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) {
@@ -1037,7 +1173,7 @@ static void socket_enter_running(Socket *s, int cfd) {
 
         /* We don't take connections anymore if we are supposed to
          * shut down anyway */
-        if (s->meta.job && s->meta.job->type == JOB_STOP) {
+        if (unit_pending_inactive(UNIT(s))) {
                 if (cfd >= 0)
                         close_nointr_nofail(cfd);
                 else  {
@@ -1059,8 +1195,8 @@ static void socket_enter_running(Socket *s, int cfd) {
 
                 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);
@@ -1068,7 +1204,10 @@ 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))) {
@@ -1086,21 +1225,31 @@ 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;
+
+                unit_choose_id(UNIT(service), name);
+                free(name);
 
-                if ((r = service_set_socket_fd(SERVICE(u), cfd, s)) < 0)
+                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;
@@ -1172,12 +1321,12 @@ static int socket_start(Unit *u) {
                 /* If the service is alredy actvie we cannot start the
                  * socket */
                 if (s->service->state != SERVICE_DEAD &&
-                    s->service->state != SERVICE_MAINTENANCE &&
+                    s->service->state != SERVICE_FAILED &&
                     s->service->state != SERVICE_AUTO_RESTART)
                         return -EBUSY;
         }
 
-        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);
@@ -1259,7 +1408,6 @@ 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);
@@ -1284,14 +1432,14 @@ 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;
@@ -1431,7 +1579,8 @@ static void socket_sigchld_event(Unit *u, pid_t pid, int code, int status) {
                         success = true;
         }
 
-        log_debug("%s control process exited, code=%s status=%i", u->meta.id, sigchld_code_to_string(code), status);
+        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) {
@@ -1478,6 +1627,9 @@ 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) {
@@ -1524,7 +1676,7 @@ static void socket_timer_event(Unit *u, uint64_t elapsed, Watch *w) {
                 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;
 
@@ -1549,7 +1701,7 @@ 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;
@@ -1593,6 +1745,17 @@ 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 const char* const socket_state_table[_SOCKET_STATE_MAX] = {
         [SOCKET_DEAD] = "dead",
         [SOCKET_START_PRE] = "start-pre",
@@ -1605,7 +1768,7 @@ 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);
@@ -1645,5 +1808,9 @@ 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
 };