From f75f613d259e9332be8e9657fa37e73f7aabcb8a Mon Sep 17 00:00:00 2001 From: Franck Bui Date: Mon, 18 Mar 2019 20:59:36 +0100 Subject: [PATCH] core: reduce the number of stalled PIDs from the watched processes list when possible Some PIDs can remain in the watched list even though their processes have exited since a long time. It can easily happen if the main process of a forking service manages to spawn a child before the control process exits for example. However when a pid is about to be mapped to a unit by calling unit_watch_pid(), the caller usually knows if the pid should belong to this unit exclusively: if we just forked() off a child, then we can be sure that its PID is otherwise unused. In this case we take this opportunity to remove any stalled PIDs from the watched process list. If we learnt about a PID in any other form (for example via PID file, via searching, MAINPID= and so on), then we can't assume anything. --- src/core/cgroup.c | 2 +- src/core/dbus-scope.c | 2 +- src/core/manager.c | 10 ++++++++++ src/core/manager.h | 2 ++ src/core/mount.c | 5 ++--- src/core/service.c | 16 ++++++++-------- src/core/socket.c | 7 +++---- src/core/swap.c | 5 ++--- src/core/unit.c | 8 +++++++- src/core/unit.h | 2 +- src/test/test-watch-pid.c | 12 ++++++------ 11 files changed, 43 insertions(+), 28 deletions(-) diff --git a/src/core/cgroup.c b/src/core/cgroup.c index 8db044befa1..ad67ba04389 100644 --- a/src/core/cgroup.c +++ b/src/core/cgroup.c @@ -2319,7 +2319,7 @@ static int unit_watch_pids_in_path(Unit *u, const char *path) { pid_t pid; while ((r = cg_read_pid(f, &pid)) > 0) { - r = unit_watch_pid(u, pid); + r = unit_watch_pid(u, pid, false); if (r < 0 && ret >= 0) ret = r; } diff --git a/src/core/dbus-scope.c b/src/core/dbus-scope.c index bb807df2e92..8eb915e5084 100644 --- a/src/core/dbus-scope.c +++ b/src/core/dbus-scope.c @@ -106,7 +106,7 @@ static int bus_scope_set_transient_property( return r; if (!UNIT_WRITE_FLAGS_NOOP(flags)) { - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, false); if (r < 0 && r != -EEXIST) return r; } diff --git a/src/core/manager.c b/src/core/manager.c index 22d1fda2dcd..33785d905f7 100644 --- a/src/core/manager.c +++ b/src/core/manager.c @@ -2129,6 +2129,16 @@ void manager_clear_jobs(Manager *m) { job_finish_and_invalidate(j, JOB_CANCELED, false, false); } +void manager_unwatch_pid(Manager *m, pid_t pid) { + assert(m); + + /* First let's drop the unit keyed as "pid". */ + (void) hashmap_remove(m->watch_pids, PID_TO_PTR(pid)); + + /* Then, let's also drop the array keyed by -pid. */ + free(hashmap_remove(m->watch_pids, PID_TO_PTR(-pid))); +} + static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) { Manager *m = userdata; Job *j; diff --git a/src/core/manager.h b/src/core/manager.h index bce8020cfd4..c43c587a79d 100644 --- a/src/core/manager.h +++ b/src/core/manager.h @@ -437,6 +437,8 @@ int manager_get_dump_string(Manager *m, char **ret); void manager_clear_jobs(Manager *m); +void manager_unwatch_pid(Manager *m, pid_t pid); + unsigned manager_dispatch_load_queue(Manager *m); int manager_default_environment(Manager *m); diff --git a/src/core/mount.c b/src/core/mount.c index a991bf8794d..4fb4b0f81e3 100644 --- a/src/core/mount.c +++ b/src/core/mount.c @@ -704,7 +704,7 @@ static int mount_coldplug(Unit *u) { pid_is_unwaited(m->control_pid) && MOUNT_STATE_WITH_PROCESS(new_state)) { - r = unit_watch_pid(UNIT(m), m->control_pid); + r = unit_watch_pid(UNIT(m), m->control_pid, false); if (r < 0) return r; @@ -810,9 +810,8 @@ static int mount_spawn(Mount *m, ExecCommand *c, pid_t *_pid) { if (r < 0) return r; - r = unit_watch_pid(UNIT(m), pid); + r = unit_watch_pid(UNIT(m), pid, true); if (r < 0) - /* FIXME: we need to do something here */ return r; *_pid = pid; diff --git a/src/core/service.c b/src/core/service.c index 4682e86f8ee..883ca7bbfeb 100644 --- a/src/core/service.c +++ b/src/core/service.c @@ -990,7 +990,7 @@ static int service_load_pid_file(Service *s, bool may_warn) { if (r < 0) return r; - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, false); if (r < 0) /* FIXME: we need to do something here */ return log_unit_warning_errno(UNIT(s), r, "Failed to watch PID "PID_FMT" for service: %m", pid); @@ -1020,7 +1020,7 @@ static void service_search_main_pid(Service *s) { if (service_set_main_pid(s, pid) < 0) return; - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, false); if (r < 0) /* FIXME: we need to do something here */ log_unit_warning_errno(UNIT(s), r, "Failed to watch PID "PID_FMT" from: %m", pid); @@ -1154,7 +1154,7 @@ static int service_coldplug(Unit *u) { SERVICE_RUNNING, SERVICE_RELOAD, SERVICE_STOP, SERVICE_STOP_WATCHDOG, SERVICE_STOP_SIGTERM, SERVICE_STOP_SIGKILL, SERVICE_STOP_POST, SERVICE_FINAL_SIGTERM, SERVICE_FINAL_SIGKILL))) { - r = unit_watch_pid(UNIT(s), s->main_pid); + r = unit_watch_pid(UNIT(s), s->main_pid, false); if (r < 0) return r; } @@ -1166,7 +1166,7 @@ static int service_coldplug(Unit *u) { SERVICE_RELOAD, SERVICE_STOP, SERVICE_STOP_WATCHDOG, SERVICE_STOP_SIGTERM, SERVICE_STOP_SIGKILL, SERVICE_STOP_POST, SERVICE_FINAL_SIGTERM, SERVICE_FINAL_SIGKILL)) { - r = unit_watch_pid(UNIT(s), s->control_pid); + r = unit_watch_pid(UNIT(s), s->control_pid, false); if (r < 0) return r; } @@ -1566,8 +1566,8 @@ static int service_spawn( s->exec_fd_event_source = TAKE_PTR(exec_fd_source); s->exec_fd_hot = false; - r = unit_watch_pid(UNIT(s), pid); - if (r < 0) /* FIXME: we need to do something here */ + r = unit_watch_pid(UNIT(s), pid, true); + if (r < 0) return r; *_pid = pid; @@ -3705,7 +3705,7 @@ static void service_notify_message( if (r > 0) { service_set_main_pid(s, new_main_pid); - r = unit_watch_pid(UNIT(s), new_main_pid); + r = unit_watch_pid(UNIT(s), new_main_pid, false); if (r < 0) log_unit_warning_errno(UNIT(s), r, "Failed to watch new main PID "PID_FMT" for service: %m", new_main_pid); @@ -3923,7 +3923,7 @@ static void service_bus_name_owner_change( log_unit_debug(u, "D-Bus name %s is now owned by process " PID_FMT, name, pid); service_set_main_pid(s, pid); - unit_watch_pid(UNIT(s), pid); + unit_watch_pid(UNIT(s), pid, false); } } } diff --git a/src/core/socket.c b/src/core/socket.c index fd550290bea..1396842b269 100644 --- a/src/core/socket.c +++ b/src/core/socket.c @@ -1852,7 +1852,7 @@ static int socket_coldplug(Unit *u) { SOCKET_FINAL_SIGTERM, SOCKET_FINAL_SIGKILL)) { - r = unit_watch_pid(UNIT(s), s->control_pid); + r = unit_watch_pid(UNIT(s), s->control_pid, false); if (r < 0) return r; @@ -1938,9 +1938,8 @@ static int socket_spawn(Socket *s, ExecCommand *c, pid_t *_pid) { if (r < 0) return r; - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, true); if (r < 0) - /* FIXME: we need to do something here */ return r; *_pid = pid; @@ -2009,7 +2008,7 @@ static int socket_chown(Socket *s, pid_t *_pid) { _exit(EXIT_SUCCESS); } - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, true); if (r < 0) goto fail; diff --git a/src/core/swap.c b/src/core/swap.c index b836fdd3046..28acef23739 100644 --- a/src/core/swap.c +++ b/src/core/swap.c @@ -550,7 +550,7 @@ static int swap_coldplug(Unit *u) { pid_is_unwaited(s->control_pid) && SWAP_STATE_WITH_PROCESS(new_state)) { - r = unit_watch_pid(UNIT(s), s->control_pid); + r = unit_watch_pid(UNIT(s), s->control_pid, false); if (r < 0) return r; @@ -657,9 +657,8 @@ static int swap_spawn(Swap *s, ExecCommand *c, pid_t *_pid) { if (r < 0) goto fail; - r = unit_watch_pid(UNIT(s), pid); + r = unit_watch_pid(UNIT(s), pid, true); if (r < 0) - /* FIXME: we need to do something here */ goto fail; *_pid = pid; diff --git a/src/core/unit.c b/src/core/unit.c index 445cf6695c8..5620fd90178 100644 --- a/src/core/unit.c +++ b/src/core/unit.c @@ -2525,7 +2525,7 @@ void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlag unit_add_to_gc_queue(u); } -int unit_watch_pid(Unit *u, pid_t pid) { +int unit_watch_pid(Unit *u, pid_t pid, bool exclusive) { int r; assert(u); @@ -2533,6 +2533,12 @@ int unit_watch_pid(Unit *u, pid_t pid) { /* Watch a specific PID */ + /* Caller might be sure that this PID belongs to this unit only. Let's take this + * opportunity to remove any stalled references to this PID as they can be created + * easily (when watching a process which is not our direct child). */ + if (exclusive) + manager_unwatch_pid(u->manager, pid); + r = set_ensure_allocated(&u->pids, NULL); if (r < 0) return r; diff --git a/src/core/unit.h b/src/core/unit.h index 35b936771c9..28878b5f4f4 100644 --- a/src/core/unit.h +++ b/src/core/unit.h @@ -678,7 +678,7 @@ typedef enum UnitNotifyFlags { void unit_notify(Unit *u, UnitActiveState os, UnitActiveState ns, UnitNotifyFlags flags); -int unit_watch_pid(Unit *u, pid_t pid); +int unit_watch_pid(Unit *u, pid_t pid, bool exclusive); void unit_unwatch_pid(Unit *u, pid_t pid); void unit_unwatch_all_pids(Unit *u); diff --git a/src/test/test-watch-pid.c b/src/test/test-watch-pid.c index 2c6ca0a1a29..04d7e1fbd85 100644 --- a/src/test/test-watch-pid.c +++ b/src/test/test-watch-pid.c @@ -42,25 +42,25 @@ int main(int argc, char *argv[]) { assert_se(hashmap_isempty(m->watch_pids)); assert_se(manager_get_unit_by_pid(m, 4711) == NULL); - assert_se(unit_watch_pid(a, 4711) >= 0); + assert_se(unit_watch_pid(a, 4711, false) >= 0); assert_se(manager_get_unit_by_pid(m, 4711) == a); - assert_se(unit_watch_pid(a, 4711) >= 0); + assert_se(unit_watch_pid(a, 4711, false) >= 0); assert_se(manager_get_unit_by_pid(m, 4711) == a); - assert_se(unit_watch_pid(b, 4711) >= 0); + assert_se(unit_watch_pid(b, 4711, false) >= 0); u = manager_get_unit_by_pid(m, 4711); assert_se(u == a || u == b); - assert_se(unit_watch_pid(b, 4711) >= 0); + assert_se(unit_watch_pid(b, 4711, false) >= 0); u = manager_get_unit_by_pid(m, 4711); assert_se(u == a || u == b); - assert_se(unit_watch_pid(c, 4711) >= 0); + assert_se(unit_watch_pid(c, 4711, false) >= 0); u = manager_get_unit_by_pid(m, 4711); assert_se(u == a || u == b || u == c); - assert_se(unit_watch_pid(c, 4711) >= 0); + assert_se(unit_watch_pid(c, 4711, false) >= 0); u = manager_get_unit_by_pid(m, 4711); assert_se(u == a || u == b || u == c); -- 2.39.2