X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fudev%2Fudevd.c;h=81e3f69a3a7cc3e9e59c47cd3efe87662af04f99;hp=d7c172cc75aea0c6039fb242341e6cc199ded9ac;hb=e9fc29f4ecc9509ccc02eb8a014341e26c0d7831;hpb=9f6445e34a57c270f013c9416c123e56261553dd diff --git a/src/udev/udevd.c b/src/udev/udevd.c index d7c172cc7..81e3f69a3 100644 --- a/src/udev/udevd.c +++ b/src/udev/udevd.c @@ -31,12 +31,14 @@ #include #include #include +#include #include #include #include #include #include #include +#include #include #include #include @@ -45,6 +47,7 @@ #include #include "udev.h" +#include "udev-util.h" #include "sd-daemon.h" #include "cgroup-util.h" #include "dev-setup.h" @@ -60,7 +63,6 @@ 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 +141,9 @@ 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); } @@ -200,7 +197,7 @@ static void worker_new(struct event *event) udev_monitor_allow_unicast_sender(worker_monitor, monitor); udev_monitor_enable_receiving(worker_monitor); - worker = calloc(1, sizeof(struct worker)); + worker = new0(struct worker, 1); if (worker == NULL) { udev_monitor_unref(worker_monitor); return; @@ -225,7 +222,6 @@ 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); @@ -247,12 +243,12 @@ static void worker_new(struct event *event) goto out; } - memset(&ep_signal, 0, sizeof(struct epoll_event)); + memzero(&ep_signal, sizeof(struct epoll_event)); ep_signal.events = EPOLLIN; ep_signal.data.fd = fd_signal; fd_monitor = udev_monitor_get_fd(worker_monitor); - memset(&ep_monitor, 0, sizeof(struct epoll_event)); + memzero(&ep_monitor, sizeof(struct epoll_event)); ep_monitor.events = EPOLLIN; ep_monitor.data.fd = fd_monitor; @@ -272,7 +268,8 @@ static void worker_new(struct event *event) for (;;) { struct udev_event *udev_event; struct worker_message msg; - int err; + int fd_lock = -1; + int err = 0; log_debug("seq %llu running", udev_device_get_seqnum(dev)); udev_event = udev_event_new(dev); @@ -287,25 +284,63 @@ static void worker_new(struct event *event) if (exec_delay > 0) udev_event->exec_delay = exec_delay; + /* + * Take a "read lock" on the device node; this establishes + * a concept of device "ownership" to serialize device + * access. External processes holding a "write lock" will + * cause udev to skip the event handling; in the case udev + * acquired the lock, the external process will block until + * udev has finished its event handling. + */ + + /* + * since we make check - device seems unused - we try + * ioctl to deactivate - and device is found to be opened + * sure, you try to take a write lock + * if you get it udev is out + * if you can't get it, udev is busy + * we cannot deactivate openned device (as it is in-use) + * maybe we should just exclude dm from that thing entirely + * IMHO this sounds like a good plan for this moment + */ + if (streq_ptr("block", udev_device_get_subsystem(dev)) && + !startswith("dm-", udev_device_get_sysname(dev))) { + struct udev_device *d = dev; + + if (streq_ptr("partition", udev_device_get_devtype(d))) + d = udev_device_get_parent(d); + + if (d) { + fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); + if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) { + log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d)); + err = -EWOULDBLOCK; + fd_lock = safe_close(fd_lock); + goto skip; + } + } + } + /* apply rules, create node, symlinks */ - err = udev_event_execute_rules(udev_event, rules, &sigmask_orig); + udev_event_execute_rules(udev_event, rules, &sigmask_orig); - if (err == 0) - udev_event_execute_run(udev_event, &sigmask_orig); + udev_event_execute_run(udev_event, &sigmask_orig); /* apply/restore inotify watch */ - if (err == 0 && udev_event->inotify_watch) { + if (udev_event->inotify_watch) { udev_watch_begin(udev, dev); udev_device_update_db(dev); } + safe_close(fd_lock); + /* send processed event back to libudev listeners */ udev_monitor_send_device(worker_monitor, NULL, dev); +skip: /* send udevd the result of the event execution */ - memset(&msg, 0, sizeof(struct worker_message)); - if (err != 0) - msg.exitcode = err; + memzero(&msg, sizeof(struct worker_message)); + msg.exitcode = err; msg.pid = getpid(); send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0); @@ -356,10 +391,8 @@ static void worker_new(struct event *event) } out: udev_device_unref(dev); - if (fd_signal >= 0) - close(fd_signal); - if (fd_ep >= 0) - close(fd_ep); + safe_close(fd_signal); + safe_close(fd_ep); close(fd_inotify); close(worker_watch[WRITE_END]); udev_rules_unref(rules); @@ -431,7 +464,7 @@ static int event_queue_insert(struct udev_device *dev) { struct event *event; - event = calloc(1, sizeof(struct event)); + event = new0(struct event, 1); if (event == NULL) return -1; @@ -449,7 +482,6 @@ 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 +612,7 @@ 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 +637,7 @@ 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) @@ -635,7 +667,7 @@ static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) i = udev_ctrl_get_set_log_level(ctrl_msg); if (i >= 0) { - log_debug("udevd message (SET_LOG_PRIORITY) received, log_priority=%i", i); + log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i); log_set_max_level(i); udev_set_log_priority(udev, i); worker_kill(udev); @@ -703,20 +735,120 @@ out: return udev_ctrl_connection_unref(ctrl_conn); } -/* read inotify messages */ +static int synthesize_change(struct udev_device *dev) { + char filename[UTIL_PATH_SIZE]; + int r; + + if (streq_ptr("block", udev_device_get_subsystem(dev)) && + streq_ptr("disk", udev_device_get_devtype(dev)) && + !startswith("dm-", udev_device_get_sysname(dev))) { + bool part_table_read = false; + bool has_partitions = false; + int fd; + struct udev *udev = udev_device_get_udev(dev); + _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL; + struct udev_list_entry *item; + + /* + * Try to re-read the partition table. This only succeeds if + * none of the devices is busy. The kernel returns 0 if no + * partition table is found, and we will not get an event for + * the disk. + */ + fd = open(udev_device_get_devnode(dev), O_RDONLY|O_EXCL|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); + if (fd >= 0) { + r = ioctl(fd, BLKRRPART, 0); + close(fd); + if (r >= 0) + part_table_read = true; + } + + /* search for partitions */ + e = udev_enumerate_new(udev); + if (!e) + return -ENOMEM; + + r = udev_enumerate_add_match_parent(e, dev); + if (r < 0) + return r; + + r = udev_enumerate_add_match_subsystem(e, "block"); + if (r < 0) + return r; + + r = udev_enumerate_scan_devices(e); + + udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) { + _cleanup_udev_device_unref_ struct udev_device *d = NULL; + + d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item)); + if (!d) + continue; + + if (!streq_ptr("partition", udev_device_get_devtype(d))) + continue; + + has_partitions = true; + break; + } + + /* + * We have partitions and re-read the table, the kernel already sent + * out a "change" event for the disk, and "remove/add" for all + * partitions. + */ + if (part_table_read && has_partitions) + return 0; + + /* + * We have partitions but re-reading the partition table did not + * work, synthesize "change" for the disk and all partitions. + */ + log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev)); + strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL); + write_string_file(filename, "change"); + + udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) { + _cleanup_udev_device_unref_ struct udev_device *d = NULL; + + d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item)); + if (!d) + continue; + + if (!streq_ptr("partition", udev_device_get_devtype(d))) + continue; + + log_debug("device %s closed, synthesising partition '%s' 'change'", + udev_device_get_devnode(dev), udev_device_get_devnode(d)); + strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL); + write_string_file(filename, "change"); + } + + return 0; + } + + log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev)); + strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL); + write_string_file(filename, "change"); + + return 0; +} + static int handle_inotify(struct udev *udev) { int nbytes, pos; char *buf; struct inotify_event *ev; + int r; - if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0)) - return 0; + r = ioctl(fd_inotify, FIONREAD, &nbytes); + if (r < 0 || nbytes <= 0) + return -errno; buf = malloc(nbytes); - if (buf == NULL) { + if (!buf) { log_error("error getting buffer for inotify"); - return -1; + return -ENOMEM; } nbytes = read(fd_inotify, buf, nbytes); @@ -726,27 +858,16 @@ static int handle_inotify(struct udev *udev) ev = (struct inotify_event *)(buf + pos); dev = udev_watch_lookup(udev, ev->wd); - if (dev != NULL) { - log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev)); - if (ev->mask & IN_CLOSE_WRITE) { - char filename[UTIL_PATH_SIZE]; - int fd; - - log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev)); - strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL); - fd = open(filename, O_WRONLY); - if (fd >= 0) { - if (write(fd, "change", 6) < 0) - log_debug("error writing uevent: %m"); - close(fd); - } - } - if (ev->mask & IN_IGNORED) - udev_watch_end(udev, dev); + if (!dev) + continue; - udev_device_unref(dev); - } + log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev)); + if (ev->mask & IN_CLOSE_WRITE) + synthesize_change(dev); + else if (ev->mask & IN_IGNORED) + udev_watch_end(udev, dev); + udev_device_unref(dev); } free(buf); @@ -797,7 +918,8 @@ 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 +1198,7 @@ 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; @@ -1159,23 +1274,23 @@ int main(int argc, char *argv[]) goto exit; } - memset(&ep_ctrl, 0, sizeof(struct epoll_event)); + memzero(&ep_ctrl, sizeof(struct epoll_event)); ep_ctrl.events = EPOLLIN; ep_ctrl.data.fd = fd_ctrl; - memset(&ep_inotify, 0, sizeof(struct epoll_event)); + memzero(&ep_inotify, sizeof(struct epoll_event)); ep_inotify.events = EPOLLIN; ep_inotify.data.fd = fd_inotify; - memset(&ep_signal, 0, sizeof(struct epoll_event)); + memzero(&ep_signal, sizeof(struct epoll_event)); ep_signal.events = EPOLLIN; ep_signal.data.fd = fd_signal; - memset(&ep_netlink, 0, sizeof(struct epoll_event)); + memzero(&ep_netlink, sizeof(struct epoll_event)); ep_netlink.events = EPOLLIN; ep_netlink.data.fd = fd_netlink; - memset(&ep_worker, 0, sizeof(struct epoll_event)); + memzero(&ep_worker, sizeof(struct epoll_event)); ep_worker.events = EPOLLIN; ep_worker.data.fd = fd_worker; @@ -1241,12 +1356,12 @@ 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 +1370,20 @@ 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 +1410,18 @@ 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 : ""); 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 +1444,7 @@ 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 +1517,8 @@ 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 +1533,6 @@ 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();