udev: remove seqnum API and all assumptions about seqnums

The way the kernel namespaces have been implemented breaks assumptions
udev made regarding uevent sequence numbers. Creating devices in a
namespace "steals" uevents and its sequence numbers from the host. It
confuses the "udevadmin settle" logic, which might block until util a
timeout is reached, even when no uevent is pending.

Remove any assumptions about sequence numbers and deprecate libudev's
API exposing these numbers; none of that can reliably be used anymore
when namespaces are involved.

3 files changed, 51 insertions, 141 deletions

diff --git a/src/udev/udev-ctrl.c b/src/udev/udev-ctrl.c
index 1e91ec20a1..74bbd3a99e 100644
--- a/src/udev/udev-ctrl.c
+++ b/src/udev/udev-ctrl.c
@@ -275,7 +275,7 @@ static int ctrl_send(struct udev_ctrl *uctrl, enum udev_ctrl_msg_type type, int
 
                 pfd[0].fd = uctrl->sock;
                 pfd[0].events = POLLIN;
-                r = poll(pfd, 1, timeout * 1000);
+                r = poll(pfd, 1, timeout * MSEC_PER_SEC);
                 if (r  < 0) {
                         if (errno == EINTR)
                                 continue;
diff --git a/src/udev/udevadm-settle.c b/src/udev/udevadm-settle.c
index 927ea2a7f2..65fc35f9a0 100644
--- a/src/udev/udevadm-settle.c
+++ b/src/udev/udevadm-settle.c
@@ -41,42 +41,28 @@
 static void help(void) {
         printf("Usage: udevadm settle OPTIONS\n"
                "  -t,--timeout=<seconds>     maximum time to wait for events\n"
-               "  -s,--seq-start=<seqnum>    first seqnum to wait for\n"
-               "  -e,--seq-end=<seqnum>      last seqnum to wait for\n"
                "  -E,--exit-if-exists=<file> stop waiting if file exists\n"
-               "  -q,--quiet                 do not print list after timeout\n"
                "  -h,--help\n\n");
 }
 
 static int adm_settle(struct udev *udev, int argc, char *argv[])
 {
         static const struct option options[] = {
-                { "seq-start",      required_argument, NULL, 's' },
-                { "seq-end",        required_argument, NULL, 'e' },
+                { "seq-start",      required_argument, NULL, '\0' }, /* removed */
+                { "seq-end",        required_argument, NULL, '\0' }, /* removed */
                 { "timeout",        required_argument, NULL, 't' },
                 { "exit-if-exists", required_argument, NULL, 'E' },
-                { "quiet",          no_argument,       NULL, 'q' },
+                { "quiet",          no_argument,       NULL, 'q' },  /* removed */
                 { "help",           no_argument,       NULL, 'h' },
                 {}
         };
-        usec_t start_usec = now(CLOCK_MONOTONIC);
-        usec_t start = 0;
-        usec_t end = 0;
-        int quiet = 0;
         const char *exists = NULL;
         unsigned int timeout = 120;
         struct pollfd pfd[1] = { {.fd = -1}, };
-        _cleanup_udev_queue_unref_ struct udev_queue *udev_queue = NULL;
         int rc = EXIT_FAILURE, c;
 
-        while ((c = getopt_long(argc, argv, "s:e:t:E:qh", options, NULL)) >= 0)
+        while ((c = getopt_long(argc, argv, "s:e:t:E:qh", options, NULL)) >= 0) {
                 switch (c) {
-                case 's':
-                        start = strtoull(optarg, NULL, 0);
-                        break;
-                case 'e':
-                        end = strtoull(optarg, NULL, 0);
-                        break;
                 case 't': {
                         int r;
 
@@ -91,9 +77,6 @@ static int adm_settle(struct udev *udev, int argc, char *argv[])
                 case 'E':
                         exists = optarg;
                         break;
-                case 'q':
-                        quiet = 1;
-                        break;
                 case 'h':
                         help();
                         exit(EXIT_SUCCESS);
@@ -102,44 +85,13 @@ static int adm_settle(struct udev *udev, int argc, char *argv[])
                 default:
                         assert_not_reached("Unknown argument");
                 }
+        }
 
         if (optind < argc) {
                 fprintf(stderr, "Extraneous argument: '%s'\n", argv[optind]);
                 exit(EXIT_FAILURE);
         }
 
-        udev_queue = udev_queue_new(udev);
-        if (udev_queue == NULL)
-                exit(2);
-
-        if (start > 0) {
-                unsigned long long kernel_seq;
-
-                kernel_seq = udev_queue_get_kernel_seqnum(udev_queue);
-
-                /* unless specified, the last event is the current kernel seqnum */
-                if (end == 0)
-                        end = udev_queue_get_kernel_seqnum(udev_queue);
-
-                if (start > end) {
-                        log_error("seq-start larger than seq-end, ignoring");
-                        start = 0;
-                        end = 0;
-                }
-
-                if (start > kernel_seq || end > kernel_seq) {
-                        log_error("seq-start or seq-end larger than current kernel value, ignoring");
-                        start = 0;
-                        end = 0;
-                }
-                log_debug("start=%llu end=%llu current=%llu", (unsigned long long)start, (unsigned long long)end, kernel_seq);
-        } else {
-                if (end > 0) {
-                        log_error("seq-end needs seq-start parameter, ignoring");
-                        end = 0;
-                }
-        }
-
         /* guarantee that the udev daemon isn't pre-processing */
         if (getuid() == 0) {
                 struct udev_ctrl *uctrl;
@@ -160,73 +112,34 @@ static int adm_settle(struct udev *udev, int argc, char *argv[])
         pfd[0].fd = inotify_init1(IN_CLOEXEC);
         if (pfd[0].fd < 0) {
                 log_error("inotify_init failed: %m");
-        } else {
-                if (inotify_add_watch(pfd[0].fd, "/run/udev" , IN_MOVED_TO) < 0) {
-                        log_error("watching /run/udev failed");
-                        close(pfd[0].fd);
-                        pfd[0].fd = -1;
-                }
+                goto out;
         }
 
-        for (;;) {
-                struct stat statbuf;
+        if (inotify_add_watch(pfd[0].fd, "/run/udev/queue" , IN_DELETE) < 0) {
+                log_debug("watching /run/udev failed");
+                goto out;
+        }
 
-                if (exists != NULL && stat(exists, &statbuf) == 0) {
+        for (;;) {
+                if (exists && access(exists, F_OK) >= 0) {
                         rc = EXIT_SUCCESS;
                         break;
                 }
 
-                if (start > 0) {
-                        /* if asked for, wait for a specific sequence of events */
-                        if (udev_queue_get_seqnum_sequence_is_finished(udev_queue, start, end) == 1) {
-                                rc = EXIT_SUCCESS;
-                                break;
-                        }
-                } else {
-                        /* exit if queue is empty */
-                        if (udev_queue_get_queue_is_empty(udev_queue)) {
-                                rc = EXIT_SUCCESS;
-                                break;
-                        }
-                }
-
-                if (pfd[0].fd >= 0) {
-                        int delay;
-
-                        if (exists != NULL || start > 0)
-                                delay = 100;
-                        else
-                                delay = 1000;
-                        /* wake up after delay, or immediately after the queue is rebuilt */
-                        if (poll(pfd, 1, delay) > 0 && pfd[0].revents & POLLIN) {
-                                char buf[sizeof(struct inotify_event) + PATH_MAX];
-
-                                read(pfd[0].fd, buf, sizeof(buf));
-                        }
-                } else {
-                        sleep(1);
+                /* exit if queue is empty */
+                if (access("/run/udev/queue", F_OK) < 0) {
+                        rc = EXIT_SUCCESS;
+                        break;
                 }
 
-                if (timeout > 0) {
-                        usec_t age_usec;
+                /* wake up when "queue" file is deleted */
+                if (poll(pfd, 1, 100) > 0 && pfd[0].revents & POLLIN) {
+                        char buf[sizeof(struct inotify_event) + PATH_MAX];
 
-                        age_usec = now(CLOCK_MONOTONIC) - start_usec;
-                        if (age_usec / (1000 * 1000) >= timeout) {
-                                struct udev_list_entry *list_entry;
-
-                                if (!quiet && udev_queue_get_queued_list_entry(udev_queue) != NULL) {
-                                        log_debug("timeout waiting for udev queue");
-                                        printf("\nudevadm settle - timeout of %i seconds reached, the event queue contains:\n", timeout);
-                                        udev_list_entry_foreach(list_entry, udev_queue_get_queued_list_entry(udev_queue))
-                                                printf("  %s (%s)\n",
-                                                udev_list_entry_get_name(list_entry),
-                                                udev_list_entry_get_value(list_entry));
-                                }
-
-                                break;
-                        }
+                        read(pfd[0].fd, buf, sizeof(buf));
                 }
         }
+
 out:
         if (pfd[0].fd >= 0)
                 close(pfd[0].fd);
@@ -236,5 +149,5 @@ out:
 const struct udevadm_cmd udevadm_settle = {
         .name = "settle",
         .cmd = adm_settle,
-        .help = "wait for the event queue to finish",
+        .help = "wait for pending udev events",
 };
diff --git a/src/udev/udevd.c b/src/udev/udevd.c
index f21c227d17..f9ee36870b 100644
--- a/src/udev/udevd.c
+++ b/src/udev/udevd.c
@@ -60,7 +60,6 @@ void udev_main_log(struct udev *udev, int priority,
 }
 
 static struct udev_rules *rules;
-static struct udev_queue_export *udev_queue_export;
 static struct udev_ctrl *udev_ctrl;
 static struct udev_monitor *monitor;
 static int worker_watch[2] = { -1, -1 };
@@ -139,14 +138,9 @@ static inline struct worker *node_to_worker(struct udev_list_node *node)
         return container_of(node, struct worker, node);
 }
 
-static void event_queue_delete(struct event *event, bool export)
+static void event_queue_delete(struct event *event)
 {
         udev_list_node_remove(&event->node);
-
-        if (export) {
-                udev_queue_export_device_finished(udev_queue_export, event->dev);
-                log_debug("seq %llu done with %i", udev_device_get_seqnum(event->dev), event->exitcode);
-        }
         udev_device_unref(event->dev);
         free(event);
 }
@@ -225,7 +219,6 @@ static void worker_new(struct event *event)
                 free(worker);
                 worker_list_cleanup(udev);
                 event_queue_cleanup(udev, EVENT_UNDEF);
-                udev_queue_export_unref(udev_queue_export);
                 udev_monitor_unref(monitor);
                 udev_ctrl_unref(udev_ctrl);
                 close(fd_signal);
@@ -449,7 +442,6 @@ static int event_queue_insert(struct udev_device *dev)
                 event->nodelay = true;
 #endif
 
-        udev_queue_export_device_queued(udev_queue_export, dev);
         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 
@@ -580,7 +572,7 @@ static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
                 if (match_type != EVENT_UNDEF && match_type != event->state)
                         continue;
 
-                event_queue_delete(event, false);
+                event_queue_delete(event);
         }
 }
 
@@ -605,7 +597,7 @@ static void worker_returned(int fd_worker)
                         /* worker returned */
                         if (worker->event) {
                                 worker->event->exitcode = msg.exitcode;
-                                event_queue_delete(worker->event, true);
+                                event_queue_delete(worker->event);
                                 worker->event = NULL;
                         }
                         if (worker->state != WORKER_KILLED)
@@ -797,7 +789,8 @@ static void handle_signal(struct udev *udev, int signo)
                                                 log_error("worker [%u] failed while handling '%s'",
                                                           pid, worker->event->devpath);
                                                 worker->event->exitcode = -32;
-                                                event_queue_delete(worker->event, true);
+                                                event_queue_delete(worker->event);
+
                                                 /* drop reference taken for state 'running' */
                                                 worker_unref(worker);
                                         }
@@ -1076,14 +1069,7 @@ int main(int argc, char *argv[])
                 goto exit;
         }
 
-        udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
-
-        /* create queue file before signalling 'ready', to make sure we block 'settle' */
-        udev_queue_export = udev_queue_export_new(udev);
-        if (udev_queue_export == NULL) {
-                log_error("error creating queue file");
-                goto exit;
-        }
+        udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
 
         if (daemonize) {
                 pid_t pid;
@@ -1241,12 +1227,12 @@ int main(int argc, char *argv[])
                         worker_kill(udev);
 
                         /* exit after all has cleaned up */
-                        if (udev_list_node_is_empty(&event_list) && udev_list_node_is_empty(&worker_list))
+                        if (udev_list_node_is_empty(&event_list) && children == 0)
                                 break;
 
                         /* timeout at exit for workers to finish */
-                        timeout = 30 * 1000;
-                } else if (udev_list_node_is_empty(&event_list) && !children) {
+                        timeout = 30 * MSEC_PER_SEC;
+                } else if (udev_list_node_is_empty(&event_list) && children == 0) {
                         /* we are idle */
                         timeout = -1;
 
@@ -1255,8 +1241,20 @@ int main(int argc, char *argv[])
                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
                 } else {
                         /* kill idle or hanging workers */
-                        timeout = 3 * 1000;
+                        timeout = 3 * MSEC_PER_SEC;
                 }
+
+                /* tell settle that we are busy or idle */
+                if (!udev_list_node_is_empty(&event_list)) {
+                        int fd;
+
+                        fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
+                        if (fd >= 0)
+                                close(fd);
+                } else {
+                        unlink("/run/udev/queue");
+                }
+
                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
                 if (fdcount < 0)
                         continue;
@@ -1283,18 +1281,18 @@ int main(int argc, char *argv[])
                                 if (worker->state != WORKER_RUNNING)
                                         continue;
 
-                                if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * 1000 * 1000) {
+                                if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * USEC_PER_SEC) {
                                         log_error("worker [%u] %s timeout; kill it", worker->pid,
                                             worker->event ? worker->event->devpath : "<idle>");
                                         kill(worker->pid, SIGKILL);
                                         worker->state = WORKER_KILLED;
+
                                         /* drop reference taken for state 'running' */
                                         worker_unref(worker);
                                         if (worker->event) {
-                                                log_error("seq %llu '%s' killed",
-                                                          udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
+                                                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, true);
+                                                event_queue_delete(worker->event);
                                                 worker->event = NULL;
                                         }
                                 }
@@ -1317,7 +1315,7 @@ int main(int argc, char *argv[])
                 }
 
                 /* check for changed config, every 3 seconds at most */
-                if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * 1000 * 1000) {
+                if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
                         if (udev_rules_check_timestamp(rules))
                                 reload = true;
                         if (udev_builtin_validate(udev))
@@ -1390,8 +1388,8 @@ int main(int argc, char *argv[])
 
         rc = EXIT_SUCCESS;
 exit:
-        udev_queue_export_cleanup(udev_queue_export);
         udev_ctrl_cleanup(udev_ctrl);
+        unlink("/run/udev/queue");
 exit_daemonize:
         if (fd_ep >= 0)
                 close(fd_ep);
@@ -1406,7 +1404,6 @@ exit_daemonize:
         if (worker_watch[WRITE_END] >= 0)
                 close(worker_watch[WRITE_END]);
         udev_monitor_unref(monitor);
-        udev_queue_export_unref(udev_queue_export);
         udev_ctrl_connection_unref(ctrl_conn);
         udev_ctrl_unref(udev_ctrl);
         label_finish();