chiark / gitweb /
udev: timeout - warn after a third of the timeout before killing
[elogind.git] / src / udev / udevd.c
index b75145eebfbbcfaca30e5a41db8e8ad62798557a..e54bfec3b347e60f1f1f38956b2af1cafa5d0599 100644 (file)
@@ -48,6 +48,7 @@
 
 #include "udev.h"
 #include "udev-util.h"
+#include "rtnl-util.h"
 #include "sd-daemon.h"
 #include "cgroup-util.h"
 #include "dev-setup.h"
@@ -73,7 +74,8 @@ static bool reload;
 static int children;
 static int children_max;
 static int exec_delay;
-static usec_t event_timeout_usec = 30 * USEC_PER_SEC;
+static usec_t event_timeout_usec = 180 * USEC_PER_SEC;
+static usec_t event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
 static sigset_t sigmask_orig;
 static UDEV_LIST(event_list);
 static UDEV_LIST(worker_list);
@@ -100,9 +102,6 @@ struct event {
         dev_t devnum;
         int ifindex;
         bool is_block;
-#ifdef HAVE_FIRMWARE
-        bool nodelay;
-#endif
 };
 
 static inline struct event *node_to_event(struct udev_list_node *node) {
@@ -127,6 +126,7 @@ struct worker {
         enum worker_state state;
         struct event *event;
         usec_t event_start_usec;
+        bool event_warned;
 };
 
 /* passed from worker to main process */
@@ -203,6 +203,7 @@ static void worker_new(struct event *event) {
         case 0: {
                 struct udev_device *dev = NULL;
                 int fd_monitor;
+                _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
                 struct epoll_event ep_signal, ep_monitor;
                 sigset_t mask;
                 int rc = EXIT_SUCCESS;
@@ -304,10 +305,16 @@ static void worker_new(struct event *event) {
                                 }
                         }
 
+                        /* needed for renaming netifs */
+                        udev_event->rtnl = rtnl;
+
                         /* apply rules, create node, symlinks */
-                        udev_event_execute_rules(udev_event, event_timeout_usec, rules, &sigmask_orig);
+                        udev_event_execute_rules(udev_event, event_timeout_usec, event_timeout_warn_usec, rules, &sigmask_orig);
+
+                        udev_event_execute_run(udev_event, event_timeout_usec, event_timeout_warn_usec, &sigmask_orig);
 
-                        udev_event_execute_run(udev_event, event_timeout_usec, &sigmask_orig);
+                        /* in case rtnl was initialized */
+                        rtnl = sd_rtnl_ref(udev_event->rtnl);
 
                         /* apply/restore inotify watch */
                         if (udev_event->inotify_watch) {
@@ -398,6 +405,7 @@ out:
                 worker->pid = pid;
                 worker->state = WORKER_RUNNING;
                 worker->event_start_usec = now(CLOCK_MONOTONIC);
+                worker->event_warned = false;
                 worker->event = event;
                 event->state = EVENT_RUNNING;
                 udev_list_node_append(&worker->node, &worker_list);
@@ -428,6 +436,7 @@ static void event_run(struct event *event) {
                 worker->event = event;
                 worker->state = WORKER_RUNNING;
                 worker->event_start_usec = now(CLOCK_MONOTONIC);
+                worker->event_warned = false;
                 event->state = EVENT_RUNNING;
                 return;
         }
@@ -458,10 +467,6 @@ static int event_queue_insert(struct udev_device *dev) {
         event->devnum = udev_device_get_devnum(dev);
         event->is_block = streq("block", udev_device_get_subsystem(dev));
         event->ifindex = udev_device_get_ifindex(dev);
-#ifdef HAVE_FIRMWARE
-        if (streq(udev_device_get_subsystem(dev), "firmware"))
-                event->nodelay = true;
-#endif
 
         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
              udev_device_get_action(dev), udev_device_get_subsystem(dev));
@@ -538,12 +543,6 @@ static bool is_devpath_busy(struct event *event) {
                         return true;
                 }
 
-#ifdef HAVE_FIRMWARE
-                /* allow to bypass the dependency tracking */
-                if (event->nodelay)
-                        continue;
-#endif
-
                 /* parent device event found */
                 if (event->devpath[common] == '/') {
                         event->delaying_seqnum = loop_event->seqnum;
@@ -956,7 +955,7 @@ static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
  */
 static void kernel_cmdline_options(struct udev *udev) {
         _cleanup_free_ char *line = NULL;
-        char *w, *state;
+        const char *word, *state;
         size_t l;
         int r;
 
@@ -966,10 +965,10 @@ static void kernel_cmdline_options(struct udev *udev) {
         if (r <= 0)
                 return;
 
-        FOREACH_WORD_QUOTED(w, l, line, state) {
+        FOREACH_WORD_QUOTED(word, l, line, state) {
                 char *s, *opt;
 
-                s = strndup(w, l);
+                s = strndup(word, l);
                 if (!s)
                         break;
 
@@ -991,6 +990,7 @@ static void kernel_cmdline_options(struct udev *udev) {
                         exec_delay = strtoul(opt + 16, NULL, 0);
                 } else if (startswith(opt, "udev.event-timeout=")) {
                         event_timeout_usec = strtoul(opt + 16, NULL, 0) * USEC_PER_SEC;
+                        event_timeout_warn_usec = (event_timeout_usec / 3) ? : 1;
                 }
 
                 free(s);
@@ -1053,6 +1053,7 @@ int main(int argc, char *argv[]) {
                         break;
                 case 't':
                         event_timeout_usec = strtoul(optarg, NULL, 0) * USEC_PER_SEC;
+                        event_timeout_warn_usec = (event_timeout_usec / 3) ? : 1;
                         break;
                 case 'D':
                         debug = true;
@@ -1205,7 +1206,7 @@ int main(int argc, char *argv[]) {
                 sd_notify(1, "READY=1");
         }
 
-        print_kmsg("starting version " VERSION "\n");
+        log_info("starting version " VERSION "\n");
 
         if (!debug) {
                 int fd;
@@ -1388,21 +1389,29 @@ int main(int argc, char *argv[]) {
                         /* check for hanging events */
                         udev_list_node_foreach(loop, &worker_list) {
                                 struct worker *worker = node_to_worker(loop);
+                                usec_t ts;
 
                                 if (worker->state != WORKER_RUNNING)
                                         continue;
 
-                                if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > event_timeout_usec) {
-                                        log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
-                                        kill(worker->pid, SIGKILL);
-                                        worker->state = WORKER_KILLED;
-
-                                        /* drop reference taken for state 'running' */
-                                        worker_unref(worker);
-                                        log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
-                                        worker->event->exitcode = -64;
-                                        event_queue_delete(worker->event);
-                                        worker->event = NULL;
+                                ts = now(CLOCK_MONOTONIC);
+
+                                if ((ts - worker->event_start_usec) > event_timeout_warn_usec) {
+                                        if ((ts - worker->event_start_usec) > event_timeout_usec) {
+                                                log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
+                                                kill(worker->pid, SIGKILL);
+                                                worker->state = WORKER_KILLED;
+
+                                                /* drop reference taken for state 'running' */
+                                                worker_unref(worker);
+                                                log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
+                                                worker->event->exitcode = -64;
+                                                event_queue_delete(worker->event);
+                                                worker->event = NULL;
+                                        } else if (!worker->event_warned) {
+                                                log_warning("worker [%u] %s is taking a long time", worker->pid, worker->event->devpath);
+                                                worker->event_warned = true;
+                                        }
                                 }
                         }