core: allocate a kdbus bus for each systemd instance, if we can

10 files changed, 194 insertions, 138 deletions

diff --git a/src/core/main.c b/src/core/main.c
index 69d3a4384..6c3d9bfb1 100644
--- a/src/core/main.c
+++ b/src/core/main.c
@@ -1531,7 +1531,7 @@ int main(int argc, char *argv[]) {
         if (arg_running_as == SYSTEMD_SYSTEM)
                 bump_rlimit_nofile(&saved_rlimit_nofile);
 
-        r = manager_new(arg_running_as, !!serialization, &m);
+        r = manager_new(arg_running_as, &m);
         if (r < 0) {
                 log_error("Failed to allocate manager object: %s", strerror(-r));
                 goto finish;
diff --git a/src/core/manager.c b/src/core/manager.c
index 65cb73cc9..a168589e3 100644
--- a/src/core/manager.c
+++ b/src/core/manager.c
@@ -76,6 +76,7 @@
 #include "dbus-unit.h"
 #include "dbus-job.h"
 #include "dbus-manager.h"
+#include "bus-kernel.h"
 
 /* As soon as 5s passed since a unit was added to our GC queue, make sure to run a gc sweep */
 #define GC_QUEUE_USEC_MAX (10*USEC_PER_SEC)
@@ -408,10 +409,45 @@ static int manager_default_environment(Manager *m) {
         return 0;
 }
 
-int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) {
+static int manager_setup_kdbus(Manager *m) {
+        _cleanup_free_ char *p = NULL;
+
+        assert(m);
+
+        if (m->kdbus_fd >= 0)
+                return 0;
+
+        /* If there's already a bus address set, don't set up kdbus */
+        if (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"))
+                return 0;
+
+        m->kdbus_fd = bus_kernel_create(m->running_as == SYSTEMD_SYSTEM ? "system" : "user", &p);
+        if (m->kdbus_fd < 0) {
+                log_debug("Failed to set up kdbus: %s", strerror(-m->kdbus_fd));
+                return m->kdbus_fd;
+        }
+
+        log_info("Successfully set up kdbus on %s", p);
+        return 0;
+}
+
+static int manager_connect_bus(Manager *m, bool reexecuting) {
+        bool try_bus_connect;
+
+        assert(m);
+
+        try_bus_connect =
+                m->kdbus_fd >= 0 ||
+                reexecuting ||
+                (m->running_as == SYSTEMD_USER && getenv("DBUS_SESSION_BUS_ADDRESS"));
+
+        /* Try to connect to the busses, if possible. */
+        return bus_init(m, try_bus_connect);
+}
+
+int manager_new(SystemdRunningAs running_as, Manager **_m) {
         Manager *m;
-        int r = -ENOMEM;
-        bool try_bus_connect = false;
+        int r;
 
         assert(_m);
         assert(running_as >= 0);
@@ -431,7 +467,7 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) {
 
         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 
-        m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd = m->dev_autofs_fd = m->private_listen_fd = -1;
+        m->pin_cgroupfs_fd = m->notify_fd = m->signal_fd = m->time_change_fd = m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = -1;
         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 
         r = manager_default_environment(m);
@@ -496,18 +532,6 @@ int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **_m) {
                 goto fail;
         }
 
-        if (running_as == SYSTEMD_SYSTEM)
-                try_bus_connect = reexecuting;
-        else if (getenv("DBUS_SESSION_BUS_ADDRESS"))
-                try_bus_connect = true;
-        else
-                log_debug("Skipping DBus session bus connection attempt - no DBUS_SESSION_BUS_ADDRESS set...");
-
-        /* Try to connect to the busses, if possible. */
-        r = bus_init(m, try_bus_connect);
-        if (r < 0)
-                goto fail;
-
         m->taint_usr = dir_is_empty("/usr") > 0;
 
         *_m = m;
@@ -694,6 +718,8 @@ void manager_free(Manager *m) {
                 close_nointr_nofail(m->notify_fd);
         if (m->time_change_fd >= 0)
                 close_nointr_nofail(m->time_change_fd);
+        if (m->kdbus_fd >= 0)
+                close_nointr_nofail(m->kdbus_fd);
 
         manager_close_idle_pipe(m);
 
@@ -889,6 +915,11 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
                         r = q;
         }
 
+        /* We might have deserialized the kdbus control fd, but if we
+         * didn't, then let's create the bus now. */
+        manager_setup_kdbus(m);
+        manager_connect_bus(m, !!serialization);
+
         /* Third, fire things up! */
         q = manager_coldplug(m);
         if (q < 0)
@@ -1979,11 +2010,23 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
                         _cleanup_free_ char *ce;
 
                         ce = cescape(*e);
-                        if (ce)
-                                fprintf(f, "env=%s\n", *e);
+                        if (!ce)
+                                return -ENOMEM;
+
+                        fprintf(f, "env=%s\n", *e);
                 }
         }
 
+        if (m->kdbus_fd >= 0) {
+                int copy;
+
+                copy = fdset_put_dup(fds, m->kdbus_fd);
+                if (copy < 0)
+                        return copy;
+
+                fprintf(f, "kdbus-fd=%i\n", copy);
+        }
+
         bus_serialize(m, f);
 
         fputc('\n', f);
@@ -2074,7 +2117,8 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
                 } else if (startswith(l, "taint-usr=")) {
                         int b;
 
-                        if ((b = parse_boolean(l+10)) < 0)
+                        b = parse_boolean(l+10);
+                        if (b < 0)
                                 log_debug("Failed to parse taint /usr flag %s", l+10);
                         else
                                 m->taint_usr = m->taint_usr || b;
@@ -2121,6 +2165,19 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
 
                         strv_free(m->environment);
                         m->environment = e;
+
+                } else if (startswith(l, "kdbus-fd=")) {
+                        int fd;
+
+                        if (safe_atoi(l + 9, &fd) < 0 || !fdset_contains(fds, fd))
+                                log_debug("Failed to parse kdbus fd: %s", l + 9);
+                        else {
+                                if (m->kdbus_fd >= 0)
+                                        close_nointr_nofail(m->kdbus_fd);
+
+                                m->kdbus_fd = fdset_remove(fds, fd);
+                        }
+
                 } else if (bus_deserialize_item(m, l) == 0)
                         log_debug("Unknown serialization item '%s'", l);
         }
diff --git a/src/core/manager.h b/src/core/manager.h
index bf05812fc..fdf60ae69 100644
--- a/src/core/manager.h
+++ b/src/core/manager.h
@@ -234,9 +234,12 @@ struct Manager {
          * them. It's a hashmap with a path string as key and a Set as
          * value where Unit objects are contained. */
         Hashmap *units_requiring_mounts_for;
+
+        /* Reference to the kdbus bus control fd */
+        int kdbus_fd;
 };
 
-int manager_new(SystemdRunningAs running_as, bool reexecuting, Manager **m);
+int manager_new(SystemdRunningAs running_as, Manager **m);
 void manager_free(Manager *m);
 
 int manager_enumerate(Manager *m);
diff --git a/src/libsystemd-bus/bus-kernel.c b/src/libsystemd-bus/bus-kernel.c
index 69143434b..1cab2544e 100644
--- a/src/libsystemd-bus/bus-kernel.c
+++ b/src/libsystemd-bus/bus-kernel.c
@@ -605,39 +605,6 @@ static int bus_kernel_translate_message(sd_bus *bus, struct kdbus_msg *k) {
         return translate[found->type - _KDBUS_ITEM_KERNEL_BASE](bus, k, found);
 }
 
-int kdbus_translate_attach_flags(uint64_t mask, uint64_t *kdbus_mask) {
-
-        uint64_t m = 0;
-
-        SET_FLAG(m, KDBUS_ATTACH_CREDS,
-                 !!(mask & (SD_BUS_CREDS_UID|SD_BUS_CREDS_GID|SD_BUS_CREDS_PID|SD_BUS_CREDS_PID_STARTTIME|SD_BUS_CREDS_TID)));
-
-        SET_FLAG(m, KDBUS_ATTACH_COMM,
-                 !!(mask & (SD_BUS_CREDS_COMM|SD_BUS_CREDS_TID_COMM)));
-
-        SET_FLAG(m, KDBUS_ATTACH_EXE,
-                 !!(mask & SD_BUS_CREDS_EXE));
-
-        SET_FLAG(m, KDBUS_ATTACH_CMDLINE,
-                 !!(mask & SD_BUS_CREDS_CMDLINE));
-
-        SET_FLAG(m, KDBUS_ATTACH_CGROUP,
-                 !!(mask & (SD_BUS_CREDS_CGROUP|SD_BUS_CREDS_UNIT|SD_BUS_CREDS_USER_UNIT|SD_BUS_CREDS_SLICE|SD_BUS_CREDS_SESSION|SD_BUS_CREDS_OWNER_UID)));
-
-        SET_FLAG(m, KDBUS_ATTACH_CAPS,
-                 !!(mask & (SD_BUS_CREDS_EFFECTIVE_CAPS|SD_BUS_CREDS_PERMITTED_CAPS|SD_BUS_CREDS_INHERITABLE_CAPS|SD_BUS_CREDS_BOUNDING_CAPS)));
-
-        SET_FLAG(m, KDBUS_ATTACH_SECLABEL,
-                 !!(mask & SD_BUS_CREDS_SELINUX_CONTEXT));
-
-        SET_FLAG(m, KDBUS_ATTACH_AUDIT,
-                 !!(mask & (SD_BUS_CREDS_AUDIT_SESSION_ID|SD_BUS_CREDS_AUDIT_LOGIN_UID)));
-
-        *kdbus_mask = m;
-
-        return 0;
-}
-
 static int bus_kernel_make_message(sd_bus *bus, struct kdbus_msg *k) {
         sd_bus_message *m = NULL;
         struct kdbus_item *d;
@@ -923,52 +890,6 @@ int bus_kernel_read_message(sd_bus *bus) {
         return r < 0 ? r : 1;
 }
 
-int bus_kernel_create(const char *name, char **s) {
-        struct kdbus_cmd_bus_make *make;
-        struct kdbus_item *n;
-        size_t l;
-        int fd;
-        char *p;
-
-        assert(name);
-        assert(s);
-
-        fd = open("/dev/kdbus/control", O_RDWR|O_NOCTTY|O_CLOEXEC);
-        if (fd < 0)
-                return -errno;
-
-        l = strlen(name);
-        make = alloca0(offsetof(struct kdbus_cmd_bus_make, items) +
-                       KDBUS_PART_HEADER_SIZE + sizeof(uint64_t) +
-                       KDBUS_PART_HEADER_SIZE + DECIMAL_STR_MAX(uid_t) + 1 + l + 1);
-
-        n = make->items;
-        n->type = KDBUS_MAKE_NAME;
-        sprintf(n->str, "%lu-%s", (unsigned long) getuid(), name);
-        n->size = KDBUS_PART_HEADER_SIZE + strlen(n->str) + 1;
-
-        make->size = offsetof(struct kdbus_cmd_bus_make, items) + n->size;
-        make->flags = KDBUS_MAKE_POLICY_OPEN;
-        make->bus_flags = 0;
-        make->bloom_size = BLOOM_SIZE;
-        assert_cc(BLOOM_SIZE % 8 == 0);
-
-        p = strjoin("/dev/kdbus/", n->str, "/bus", NULL);
-        if (!p)
-                return -ENOMEM;
-
-        if (ioctl(fd, KDBUS_CMD_BUS_MAKE, make) < 0) {
-                close_nointr_nofail(fd);
-                free(p);
-                return -errno;
-        }
-
-        if (s)
-                *s = p;
-
-        return fd;
-}
-
 int bus_kernel_pop_memfd(sd_bus *bus, void **address, size_t *size) {
         struct memfd_cache *c;
         int fd;
@@ -1061,20 +982,100 @@ void bus_kernel_flush_memfd(sd_bus *b) {
                 close_and_munmap(b->memfd_cache[i].fd, b->memfd_cache[i].address, b->memfd_cache[i].size);
 }
 
-int kdbus_translate_request_name_flags(uint64_t sd_bus_flags, uint64_t *kdbus_flags) {
+int kdbus_translate_request_name_flags(uint64_t flags, uint64_t *kdbus_flags) {
+        uint64_t f = 0;
 
-        assert_return(kdbus_flags != NULL, -EINVAL);
+        assert(kdbus_flags);
 
-        *kdbus_flags = 0;
+        if (flags & SD_BUS_NAME_ALLOW_REPLACEMENT)
+                f |= KDBUS_NAME_ALLOW_REPLACEMENT;
 
-        if (sd_bus_flags & SD_BUS_NAME_ALLOW_REPLACEMENT)
-                *kdbus_flags |= KDBUS_NAME_ALLOW_REPLACEMENT;
+        if (flags & SD_BUS_NAME_REPLACE_EXISTING)
+                f |= KDBUS_NAME_REPLACE_EXISTING;
 
-        if (sd_bus_flags & SD_BUS_NAME_REPLACE_EXISTING)
-                *kdbus_flags |= KDBUS_NAME_REPLACE_EXISTING;
+        if (!(flags & SD_BUS_NAME_DO_NOT_QUEUE))
+                f |= KDBUS_NAME_QUEUE;
 
-        if (!(sd_bus_flags & SD_BUS_NAME_DO_NOT_QUEUE))
-                *kdbus_flags |= KDBUS_NAME_QUEUE;
+        *kdbus_flags = f;
+        return 0;
+}
+
+int kdbus_translate_attach_flags(uint64_t mask, uint64_t *kdbus_mask) {
+        uint64_t m = 0;
+
+        assert(kdbus_mask);
+
+        if (mask & (SD_BUS_CREDS_UID|SD_BUS_CREDS_GID|SD_BUS_CREDS_PID|SD_BUS_CREDS_PID_STARTTIME|SD_BUS_CREDS_TID))
+                m |= KDBUS_ATTACH_CREDS;
+
+        if (mask & (SD_BUS_CREDS_COMM|SD_BUS_CREDS_TID_COMM))
+                m |= KDBUS_ATTACH_COMM;
+
+        if (mask & SD_BUS_CREDS_EXE)
+                m |= KDBUS_ATTACH_EXE;
+
+        if (mask & SD_BUS_CREDS_CMDLINE)
+                m |= KDBUS_ATTACH_CMDLINE;
+
+        if (mask & (SD_BUS_CREDS_CGROUP|SD_BUS_CREDS_UNIT|SD_BUS_CREDS_USER_UNIT|SD_BUS_CREDS_SLICE|SD_BUS_CREDS_SESSION|SD_BUS_CREDS_OWNER_UID))
+                m |= KDBUS_ATTACH_CGROUP;
+
+        if (mask & (SD_BUS_CREDS_EFFECTIVE_CAPS|SD_BUS_CREDS_PERMITTED_CAPS|SD_BUS_CREDS_INHERITABLE_CAPS|SD_BUS_CREDS_BOUNDING_CAPS))
+                m |= KDBUS_ATTACH_CAPS;
+
+        if (mask & SD_BUS_CREDS_SELINUX_CONTEXT)
+                m |= KDBUS_ATTACH_SECLABEL;
 
+        if (mask & (SD_BUS_CREDS_AUDIT_SESSION_ID|SD_BUS_CREDS_AUDIT_LOGIN_UID))
+                m |= KDBUS_ATTACH_AUDIT;
+
+        *kdbus_mask = m;
         return 0;
 }
+
+int bus_kernel_create(const char *name, char **s) {
+        struct kdbus_cmd_bus_make *make;
+        struct kdbus_item *n;
+        int fd;
+
+        assert(name);
+        assert(s);
+
+        fd = open("/dev/kdbus/control", O_RDWR|O_NOCTTY|O_CLOEXEC);
+        if (fd < 0)
+                return -errno;
+
+        make = alloca0(ALIGN8(offsetof(struct kdbus_cmd_bus_make, items) +
+                              offsetof(struct kdbus_item, str) +
+                              DECIMAL_STR_MAX(uid_t) + 1 + strlen(name) + 1));
+
+        n = make->items;
+        sprintf(n->str, "%lu-%s", (unsigned long) getuid(), name);
+        n->size = offsetof(struct kdbus_item, str) + strlen(n->str) + 1;
+        n->type = KDBUS_MAKE_NAME;
+
+        make->size = ALIGN8(offsetof(struct kdbus_cmd_bus_make, items) + n->size);
+        make->flags = KDBUS_MAKE_POLICY_OPEN;
+        make->bus_flags = 0;
+        make->bloom_size = BLOOM_SIZE;
+        assert_cc(BLOOM_SIZE % 8 == 0);
+
+        if (ioctl(fd, KDBUS_CMD_BUS_MAKE, make) < 0) {
+                close_nointr_nofail(fd);
+                return -errno;
+        }
+
+        if (s) {
+                char *p;
+
+                p = strjoin("/dev/kdbus/", n->str, "/bus", NULL);
+                if (!p) {
+                        close_nointr_nofail(fd);
+                        return -ENOMEM;
+                }
+
+                *s = p;
+        }
+
+        return fd;
+}
diff --git a/src/libsystemd-bus/kdbus.h b/src/libsystemd-bus/kdbus.h
index c9377ea27..bfd299bd2 100644
--- a/src/libsystemd-bus/kdbus.h
+++ b/src/libsystemd-bus/kdbus.h
@@ -297,11 +297,6 @@ enum {
 enum {
 	_KDBUS_MAKE_NULL,
 	KDBUS_MAKE_NAME,
-	KDBUS_MAKE_CRED,	/* allow translator services which connect
-				 * to the bus on behalf of somebody else,
-				 * allow specifying the credentials of the
-				 * client to connect on behalf on. Needs
-				 * privileges */
 };
 
 struct kdbus_cmd_bus_make {
diff --git a/src/libsystemd-bus/sd-bus.c b/src/libsystemd-bus/sd-bus.c
index 8ffc9a9ef..0d5deb6c6 100644
--- a/src/libsystemd-bus/sd-bus.c
+++ b/src/libsystemd-bus/sd-bus.c
@@ -986,15 +986,13 @@ _public_ int sd_bus_open_system(sd_bus **ret) {
                 return r;
 
         e = secure_getenv("DBUS_SYSTEM_BUS_ADDRESS");
-        if (e) {
+        if (e)
                 r = sd_bus_set_address(b, e);
-                if (r < 0)
-                        goto fail;
-        } else {
-                b->sockaddr.un.sun_family = AF_UNIX;
-                strncpy(b->sockaddr.un.sun_path, "/run/dbus/system_bus_socket", sizeof(b->sockaddr.un.sun_path));
-                b->sockaddr_size = offsetof(struct sockaddr_un, sun_path) + sizeof("/run/dbus/system_bus_socket") - 1;
-        }
+        else
+                r = sd_bus_set_address(b, "kernel:path=/dev/kdbus/0-system/bus;unix:path=/run/dbus/system_bus_socket");
+
+        if (r < 0)
+                goto fail;
 
         b->bus_client = true;
 
@@ -1013,7 +1011,6 @@ fail:
 _public_ int sd_bus_open_user(sd_bus **ret) {
         const char *e;
         sd_bus *b;
-        size_t l;
         int r;
 
         assert_return(ret, -EINVAL);
@@ -1029,20 +1026,23 @@ _public_ int sd_bus_open_user(sd_bus **ret) {
                         goto fail;
         } else {
                 e = secure_getenv("XDG_RUNTIME_DIR");
-                if (!e) {
-                        r = -ENOENT;
-                        goto fail;
-                }
+                if (e) {
+                        _cleanup_free_ char *ee = NULL;
+
+                        ee = bus_address_escape(e);
+                        if (!ee) {
+                                r = -ENOENT;
+                                goto fail;
+                        }
 
-                l = strlen(e);
-                if (l + 4 > sizeof(b->sockaddr.un.sun_path)) {
-                        r = -E2BIG;
+                        asprintf(&b->address, "kernel:path=/dev/kdbus/%lu-user/bus;unix:path=%s/bus", (unsigned long) getuid(), ee);
+                } else
+                        asprintf(&b->address, "kernel:path=/dev/kdbus/%lu-user/bus", (unsigned long) getuid());
+
+                if (!b->address) {
+                        r = -ENOMEM;
                         goto fail;
                 }
-
-                b->sockaddr.un.sun_family = AF_UNIX;
-                memcpy(mempcpy(b->sockaddr.un.sun_path, e, l), "/bus", 4);
-                b->sockaddr_size = offsetof(struct sockaddr_un, sun_path) + l + 4;
         }
 
         b->bus_client = true;
diff --git a/src/test/test-cgroup-mask.c b/src/test/test-cgroup-mask.c
index ecf041f86..0f145d6bc 100644
--- a/src/test/test-cgroup-mask.c
+++ b/src/test/test-cgroup-mask.c
@@ -40,7 +40,7 @@ static int test_cgroup_mask(void) {
 
         /* Prepare the manager. */
         assert_se(set_unit_path(TEST_DIR) >= 0);
-        r = manager_new(SYSTEMD_USER, false, &m);
+        r = manager_new(SYSTEMD_USER, &m);
         if (r == -EPERM || r == -EACCES) {
                 puts("manager_new: Permission denied. Skipping test.");
                 return EXIT_TEST_SKIP;
diff --git a/src/test/test-engine.c b/src/test/test-engine.c
index 20ae103a1..0f3862226 100644
--- a/src/test/test-engine.c
+++ b/src/test/test-engine.c
@@ -33,7 +33,7 @@ int main(int argc, char *argv[]) {
 
         assert_se(set_unit_path("test") >= 0);
 
-        assert_se(manager_new(SYSTEMD_SYSTEM, false, &m) >= 0);
+        assert_se(manager_new(SYSTEMD_SYSTEM, &m) >= 0);
 
         printf("Load1:\n");
         assert_se(manager_load_unit(m, "a.service", NULL, NULL, &a) >= 0);
diff --git a/src/test/test-sched-prio.c b/src/test/test-sched-prio.c
index 208fca854..56e11f8da 100644
--- a/src/test/test-sched-prio.c
+++ b/src/test/test-sched-prio.c
@@ -34,7 +34,7 @@ int main(int argc, char *argv[]) {
 
         /* prepare the test */
         assert_se(set_unit_path(TEST_DIR) >= 0);
-        r = manager_new(SYSTEMD_USER, false, &m);
+        r = manager_new(SYSTEMD_USER, &m);
         if (r == -EPERM || r == -EACCES || r == -EADDRINUSE || r == -EHOSTDOWN) {
                 printf("Skipping test: manager_new: %s", strerror(-r));
                 return EXIT_TEST_SKIP;
diff --git a/src/test/test-unit-name.c b/src/test/test-unit-name.c
index bde643ad5..b6209b1d1 100644
--- a/src/test/test-unit-name.c
+++ b/src/test/test-unit-name.c
@@ -124,7 +124,7 @@ static int test_unit_printf(void) {
         assert_se((root = getpwnam("root")));
         assert_se(asprintf(&root_uid, "%d", (int) root->pw_uid) > 0);
 
-        r = manager_new(SYSTEMD_USER, false, &m);
+        r = manager_new(SYSTEMD_USER, &m);
         if (r == -EPERM || r == -EACCES || r == -EADDRINUSE) {
                 puts("manager_new: Permission denied. Skipping test.");
                 return EXIT_TEST_SKIP;