core: add RemoveIPC= setting

This adds the boolean RemoveIPC= setting to service, socket, mount and swap
units (i.e.  all unit types that may invoke processes). if turned on, and the
unit's user/group is not root, all IPC objects of the user/group are removed
when the service is shut down. The life-cycle of the IPC objects is hence bound
to the unit life-cycle.

This is particularly relevant for units with dynamic users, as it is essential
that no objects owned by the dynamic users survive the service exiting. In
fact, this patch adds code to imply RemoveIPC= if DynamicUser= is set.

In order to communicate the UID/GID of an executed process back to PID 1 this
adds a new "user lookup" socket pair, that is inherited into the forked
processes, and closed before the exec(). This is needed since we cannot do NSS
from PID 1 due to deadlock risks, However need to know the used UID/GID in
order to clean up IPC owned by it if the unit shuts down.

1 files changed, 409 insertions, 2 deletions

diff --git a/src/core/manager.c b/src/core/manager.c
index c20e185d78..6f2477eef4 100644
--- a/src/core/manager.c
+++ b/src/core/manager.c
@@ -45,6 +45,7 @@
 #include "bus-error.h"
 #include "bus-kernel.h"
 #include "bus-util.h"
+#include "clean-ipc.h"
 #include "dbus-job.h"
 #include "dbus-manager.h"
 #include "dbus-unit.h"
@@ -81,6 +82,7 @@
 #include "transaction.h"
 #include "umask-util.h"
 #include "unit-name.h"
+#include "user-util.h"
 #include "util.h"
 #include "virt.h"
 #include "watchdog.h"
@@ -98,6 +100,7 @@ static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, ui
 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
+static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
 static int manager_run_generators(Manager *m);
@@ -590,6 +593,8 @@ int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
                 m->dev_autofs_fd = m->private_listen_fd = m->kdbus_fd = m->cgroup_inotify_fd =
                 m->ask_password_inotify_fd = -1;
 
+        m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
+
         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 
         m->have_ask_password = -EINVAL; /* we don't know */
@@ -812,6 +817,59 @@ static int manager_setup_cgroups_agent(Manager *m) {
         return 0;
 }
 
+static int manager_setup_user_lookup_fd(Manager *m) {
+        int r;
+
+        assert(m);
+
+        /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
+         * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
+         * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
+         * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
+         * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
+         * hence we establish this communication channel so that forked off processes can pass their UID/GID
+         * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
+         * datagram, along with their unit name, so that we can share one communication socket pair among all units for
+         * this purpose.
+         *
+         * You might wonder why we need a communication channel for this that is independent of the usual notification
+         * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
+         * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
+         * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
+         *
+         * Note that this function is called under two circumstances: when we first initialize (in which case we
+         * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
+         * (in which case the socket pair already exists but we still need to allocate the event source for it). */
+
+        if (m->user_lookup_fds[0] < 0) {
+
+                /* Free all secondary fields */
+                safe_close_pair(m->user_lookup_fds);
+                m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
+
+                if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
+                        return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
+
+                (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
+        }
+
+        if (!m->user_lookup_event_source) {
+                r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
+                if (r < 0)
+                        return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
+
+                /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
+                 * resolutions */
+                r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
+                if (r < 0)
+                        return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
+
+                (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
+        }
+
+        return 0;
+}
+
 static int manager_connect_bus(Manager *m, bool reexecuting) {
         bool try_bus_connect;
 
@@ -853,8 +911,7 @@ enum {
         _GC_OFFSET_MAX
 };
 
-static void unit_gc_mark_good(Unit *u, unsigned gc_marker)
-{
+static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
         Iterator i;
         Unit *other;
 
@@ -1021,12 +1078,14 @@ Manager* manager_free(Manager *m) {
         sd_event_source_unref(m->time_change_event_source);
         sd_event_source_unref(m->jobs_in_progress_event_source);
         sd_event_source_unref(m->run_queue_event_source);
+        sd_event_source_unref(m->user_lookup_event_source);
 
         safe_close(m->signal_fd);
         safe_close(m->notify_fd);
         safe_close(m->cgroups_agent_fd);
         safe_close(m->time_change_fd);
         safe_close(m->kdbus_fd);
+        safe_close_pair(m->user_lookup_fds);
 
         manager_close_ask_password(m);
 
@@ -1052,6 +1111,9 @@ Manager* manager_free(Manager *m) {
         assert(hashmap_isempty(m->units_requiring_mounts_for));
         hashmap_free(m->units_requiring_mounts_for);
 
+        hashmap_free(m->uid_refs);
+        hashmap_free(m->gid_refs);
+
         free(m);
         return NULL;
 }
@@ -1221,6 +1283,10 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
         if (q < 0 && r == 0)
                 r = q;
 
+        q = manager_setup_user_lookup_fd(m);
+        if (q < 0 && r == 0)
+                r = q;
+
         /* We might have deserialized the kdbus control fd, but if we
          * didn't, then let's create the bus now. */
         manager_connect_bus(m, !!serialization);
@@ -1232,6 +1298,10 @@ int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
         /* Release any dynamic users no longer referenced */
         dynamic_user_vacuum(m, true);
 
+        /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
+        manager_vacuum_uid_refs(m);
+        manager_vacuum_gid_refs(m);
+
         if (serialization) {
                 assert(m->n_reloading > 0);
                 m->n_reloading--;
@@ -2396,6 +2466,20 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
                 fprintf(f, "cgroups-agent-fd=%i\n", copy);
         }
 
+        if (m->user_lookup_fds[0] >= 0) {
+                int copy0, copy1;
+
+                copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
+                if (copy0 < 0)
+                        return copy0;
+
+                copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
+                if (copy1 < 0)
+                        return copy1;
+
+                fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
+        }
+
         if (m->kdbus_fd >= 0) {
                 int copy;
 
@@ -2412,6 +2496,9 @@ int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
         if (r < 0)
                 return r;
 
+        manager_serialize_uid_refs(m, f);
+        manager_serialize_gid_refs(m, f);
+
         fputc('\n', f);
 
         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
@@ -2578,6 +2665,18 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
                                 m->cgroups_agent_fd = fdset_remove(fds, fd);
                         }
 
+                } else if (startswith(l, "user-lookup=")) {
+                        int fd0, fd1;
+
+                        if (sscanf(l + 12, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
+                                log_debug("Failed to parse user lookup fd: %s", l + 12);
+                        else {
+                                m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
+                                safe_close_pair(m->user_lookup_fds);
+                                m->user_lookup_fds[0] = fdset_remove(fds, fd0);
+                                m->user_lookup_fds[1] = fdset_remove(fds, fd1);
+                        }
+
                 } else if (startswith(l, "kdbus-fd=")) {
                         int fd;
 
@@ -2590,6 +2689,10 @@ int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
 
                 } else if (startswith(l, "dynamic-user="))
                         dynamic_user_deserialize_one(m, l + 13, fds);
+                else if (startswith(l, "destroy-ipc-uid="))
+                        manager_deserialize_uid_refs_one(m, l + 16);
+                else if (startswith(l, "destroy-ipc-gid="))
+                        manager_deserialize_gid_refs_one(m, l + 16);
                 else {
                         int k;
 
@@ -2672,6 +2775,8 @@ int manager_reload(Manager *m) {
         lookup_paths_flush_generator(&m->lookup_paths);
         lookup_paths_free(&m->lookup_paths);
         dynamic_user_vacuum(m, false);
+        m->uid_refs = hashmap_free(m->uid_refs);
+        m->gid_refs = hashmap_free(m->gid_refs);
 
         q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
         if (q < 0 && r >= 0)
@@ -2705,12 +2810,20 @@ int manager_reload(Manager *m) {
         if (q < 0 && r >= 0)
                 r = q;
 
+        q = manager_setup_user_lookup_fd(m);
+        if (q < 0 && r >= 0)
+                r = q;
+
         /* Third, fire things up! */
         manager_coldplug(m);
 
         /* Release any dynamic users no longer referenced */
         dynamic_user_vacuum(m, true);
 
+        /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
+        manager_vacuum_uid_refs(m);
+        manager_vacuum_gid_refs(m);
+
         /* Sync current state of bus names with our set of listening units */
         if (m->api_bus)
                 manager_sync_bus_names(m, m->api_bus);
@@ -3144,6 +3257,300 @@ ManagerState manager_state(Manager *m) {
         return MANAGER_RUNNING;
 }
 
+#define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
+
+static void manager_unref_uid_internal(
+                Manager *m,
+                Hashmap **uid_refs,
+                uid_t uid,
+                bool destroy_now,
+                int (*_clean_ipc)(uid_t uid)) {
+
+        uint32_t c, n;
+
+        assert(m);
+        assert(uid_refs);
+        assert(uid_is_valid(uid));
+        assert(_clean_ipc);
+
+        /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
+         * that uid_t and gid_t are actually defined the same way, with the same validity rules.
+         *
+         * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
+         * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
+         * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
+         * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
+
+        assert_cc(sizeof(uid_t) == sizeof(gid_t));
+        assert_cc(UID_INVALID == (uid_t) GID_INVALID);
+
+        if (uid == 0) /* We don't keep track of root, and will never destroy it */
+                return;
+
+        c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
+
+        n = c & ~DESTROY_IPC_FLAG;
+        assert(n > 0);
+        n--;
+
+        if (destroy_now && n == 0) {
+                hashmap_remove(*uid_refs, UID_TO_PTR(uid));
+
+                if (c & DESTROY_IPC_FLAG) {
+                        log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
+                                  _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
+                                  uid);
+                        (void) _clean_ipc(uid);
+                }
+        } else {
+                c = n | (c & DESTROY_IPC_FLAG);
+                assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
+        }
+}
+
+void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
+        manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
+}
+
+void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
+        manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
+}
+
+static int manager_ref_uid_internal(
+                Manager *m,
+                Hashmap **uid_refs,
+                uid_t uid,
+                bool clean_ipc) {
+
+        uint32_t c, n;
+        int r;
+
+        assert(m);
+        assert(uid_refs);
+        assert(uid_is_valid(uid));
+
+        /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
+         * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
+
+        assert_cc(sizeof(uid_t) == sizeof(gid_t));
+        assert_cc(UID_INVALID == (uid_t) GID_INVALID);
+
+        if (uid == 0) /* We don't keep track of root, and will never destroy it */
+                return 0;
+
+        r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
+        if (r < 0)
+                return r;
+
+        c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
+
+        n = c & ~DESTROY_IPC_FLAG;
+        n++;
+
+        if (n & DESTROY_IPC_FLAG) /* check for overflow */
+                return -EOVERFLOW;
+
+        c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
+
+        return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
+}
+
+int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
+        return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
+}
+
+int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
+        return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
+}
+
+static void manager_vacuum_uid_refs_internal(
+                Manager *m,
+                Hashmap **uid_refs,
+                int (*_clean_ipc)(uid_t uid)) {
+
+        Iterator i;
+        void *p, *k;
+
+        assert(m);
+        assert(uid_refs);
+        assert(_clean_ipc);
+
+        HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
+                uint32_t c, n;
+                uid_t uid;
+
+                uid = PTR_TO_UID(k);
+                c = PTR_TO_UINT32(p);
+
+                n = c & ~DESTROY_IPC_FLAG;
+                if (n > 0)
+                        continue;
+
+                if (c & DESTROY_IPC_FLAG) {
+                        log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
+                                  _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
+                                  uid);
+                        (void) _clean_ipc(uid);
+                }
+
+                assert_se(hashmap_remove(*uid_refs, k) == p);
+        }
+}
+
+void manager_vacuum_uid_refs(Manager *m) {
+        manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
+}
+
+void manager_vacuum_gid_refs(Manager *m) {
+        manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
+}
+
+static void manager_serialize_uid_refs_internal(
+                Manager *m,
+                FILE *f,
+                Hashmap **uid_refs,
+                const char *field_name) {
+
+        Iterator i;
+        void *p, *k;
+
+        assert(m);
+        assert(f);
+        assert(uid_refs);
+        assert(field_name);
+
+        /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
+         * of it is better rebuild after a reload/reexec. */
+
+        HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
+                uint32_t c;
+                uid_t uid;
+
+                uid = PTR_TO_UID(k);
+                c = PTR_TO_UINT32(p);
+
+                if (!(c & DESTROY_IPC_FLAG))
+                        continue;
+
+                fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
+        }
+}
+
+void manager_serialize_uid_refs(Manager *m, FILE *f) {
+        manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
+}
+
+void manager_serialize_gid_refs(Manager *m, FILE *f) {
+        manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
+}
+
+static void manager_deserialize_uid_refs_one_internal(
+                Manager *m,
+                Hashmap** uid_refs,
+                const char *value) {
+
+        uid_t uid;
+        uint32_t c;
+        int r;
+
+        assert(m);
+        assert(uid_refs);
+        assert(value);
+
+        r = parse_uid(value, &uid);
+        if (r < 0 || uid == 0) {
+                log_debug("Unable to parse UID reference serialization");
+                return;
+        }
+
+        r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
+        if (r < 0) {
+                log_oom();
+                return;
+        }
+
+        c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
+        if (c & DESTROY_IPC_FLAG)
+                return;
+
+        c |= DESTROY_IPC_FLAG;
+
+        r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
+        if (r < 0) {
+                log_debug("Failed to add UID reference entry");
+                return;
+        }
+}
+
+void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
+        manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
+}
+
+void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
+        manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
+}
+
+int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
+        struct buffer {
+                uid_t uid;
+                gid_t gid;
+                char unit_name[UNIT_NAME_MAX+1];
+        } _packed_ buffer;
+
+        Manager *m = userdata;
+        ssize_t l;
+        size_t n;
+        Unit *u;
+
+        assert_se(source);
+        assert_se(m);
+
+        /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
+         * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
+         * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
+
+        l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
+        if (l < 0) {
+                if (errno == EINTR || errno == EAGAIN)
+                        return 0;
+
+                return log_error_errno(errno, "Failed to read from user lookup fd: %m");
+        }
+
+        if ((size_t) l <= offsetof(struct buffer, unit_name)) {
+                log_warning("Received too short user lookup message, ignoring.");
+                return 0;
+        }
+
+        if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
+                log_warning("Received too long user lookup message, ignoring.");
+                return 0;
+        }
+
+        if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
+                log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
+                return 0;
+        }
+
+        n = (size_t) l - offsetof(struct buffer, unit_name);
+        if (memchr(buffer.unit_name, 0, n)) {
+                log_warning("Received lookup message with embedded NUL character, ignoring.");
+                return 0;
+        }
+
+        buffer.unit_name[n] = 0;
+        u = manager_get_unit(m, buffer.unit_name);
+        if (!u) {
+                log_debug("Got user lookup message but unit doesn't exist, ignoring.");
+                return 0;
+        }
+
+        log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
+
+        unit_notify_user_lookup(u, buffer.uid, buffer.gid);
+        return 0;
+}
+
 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
         [MANAGER_INITIALIZING] = "initializing",
         [MANAGER_STARTING] = "starting",