X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=src%2Fudev%2Fudevd.c;h=e54bfec3b347e60f1f1f38956b2af1cafa5d0599;hp=42cf994b0d8d7887cf9aafb8316e5d4e00d3db75;hb=671174136525ddf208cdbe75d6d6bd159afa961f;hpb=f45928521249bbaf5dbea84933ae2fcaf5354080 diff --git a/src/udev/udevd.c b/src/udev/udevd.c index 42cf994b0..e54bfec3b 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 +47,8 @@ #include #include "udev.h" +#include "udev-util.h" +#include "rtnl-util.h" #include "sd-daemon.h" #include "cgroup-util.h" #include "dev-setup.h" @@ -54,13 +58,11 @@ static bool debug; void udev_main_log(struct udev *udev, int priority, const char *file, int line, const char *fn, - const char *format, va_list args) -{ + const char *format, va_list args) { log_metav(priority, file, line, fn, format, args); } 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 }; @@ -72,10 +74,12 @@ static bool reload; static int children; static int children_max; static int exec_delay; +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); -char *udev_cgroup; +static char *udev_cgroup; static bool udev_exit; enum event_state { @@ -98,13 +102,9 @@ 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) -{ +static inline struct event *node_to_event(struct udev_list_node *node) { return container_of(node, struct event, node); } @@ -126,6 +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 */ @@ -134,48 +135,37 @@ struct worker_message { int exitcode; }; -static inline struct worker *node_to_worker(struct udev_list_node *node) -{ +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\n", udev_device_get_seqnum(event->dev), event->exitcode); - } udev_device_unref(event->dev); free(event); } -static struct worker *worker_ref(struct worker *worker) -{ +static struct worker *worker_ref(struct worker *worker) { worker->refcount++; return worker; } -static void worker_cleanup(struct worker *worker) -{ +static void worker_cleanup(struct worker *worker) { udev_list_node_remove(&worker->node); udev_monitor_unref(worker->monitor); children--; free(worker); } -static void worker_unref(struct worker *worker) -{ +static void worker_unref(struct worker *worker) { worker->refcount--; if (worker->refcount > 0) return; - log_debug("worker [%u] cleaned up\n", worker->pid); + log_debug("worker [%u] cleaned up", worker->pid); worker_cleanup(worker); } -static void worker_list_cleanup(struct udev *udev) -{ +static void worker_list_cleanup(struct udev *udev) { struct udev_list_node *loop, *tmp; udev_list_node_foreach_safe(loop, tmp, &worker_list) { @@ -185,8 +175,7 @@ static void worker_list_cleanup(struct udev *udev) } } -static void worker_new(struct event *event) -{ +static void worker_new(struct event *event) { struct udev *udev = event->udev; struct worker *worker; struct udev_monitor *worker_monitor; @@ -200,7 +189,7 @@ 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; @@ -214,6 +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; @@ -225,7 +215,6 @@ 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); @@ -235,30 +224,30 @@ static void worker_new(struct event *event) sigfillset(&mask); fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); if (fd_signal < 0) { - log_error("error creating signalfd %m\n"); + log_error("error creating signalfd %m"); rc = 2; goto out; } fd_ep = epoll_create1(EPOLL_CLOEXEC); if (fd_ep < 0) { - log_error("error creating epoll fd: %m\n"); + log_error("error creating epoll fd: %m"); rc = 3; 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; if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 || epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) { - log_error("fail to add fds to epoll: %m\n"); + log_error("fail to add fds to epoll: %m"); rc = 4; goto out; } @@ -267,14 +256,15 @@ static void worker_new(struct event *event) prctl(PR_SET_PDEATHSIG, SIGTERM); /* reset OOM score, we only protect the main daemon */ - write_one_line_file("/proc/self/oom_score_adj", "0"); + write_string_file("/proc/self/oom_score_adj", "0"); for (;;) { struct udev_event *udev_event; struct worker_message msg; - int err; + int fd_lock = -1; + int err = 0; - log_debug("seq %llu running\n", udev_device_get_seqnum(dev)); + log_debug("seq %llu running", udev_device_get_seqnum(dev)); udev_event = udev_event_new(dev); if (udev_event == NULL) { rc = 5; @@ -287,29 +277,64 @@ static void worker_new(struct event *event) if (exec_delay > 0) udev_event->exec_delay = exec_delay; + /* + * Take a shared lock on the device node; this establishes + * a concept of device "ownership" to serialize device + * access. External processes holding an exclusive lock will + * cause udev to skip the event handling; in the case udev + * acquired the lock, the external process can block until + * udev has finished its event handling. + */ + if (!streq_ptr(udev_device_get_action(dev), "remove") && + streq_ptr("block", udev_device_get_subsystem(dev)) && + !startswith(udev_device_get_sysname(dev), "dm-") && + !startswith(udev_device_get_sysname(dev), "md")) { + 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; + } + } + } + + /* needed for renaming netifs */ + udev_event->rtnl = rtnl; + /* apply rules, create node, symlinks */ - err = udev_event_execute_rules(udev_event, rules, &sigmask_orig); + udev_event_execute_rules(udev_event, event_timeout_usec, event_timeout_warn_usec, rules, &sigmask_orig); - if (err == 0) - udev_event_execute_run(udev_event, &sigmask_orig); + udev_event_execute_run(udev_event, event_timeout_usec, event_timeout_warn_usec, &sigmask_orig); + + /* in case rtnl was initialized */ + rtnl = sd_rtnl_ref(udev_event->rtnl); /* 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); - log_debug("seq %llu processed with %i\n", udev_device_get_seqnum(dev), err); + log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err); udev_device_unref(dev); dev = NULL; @@ -331,7 +356,7 @@ static void worker_new(struct event *event) if (fdcount < 0) { if (errno == EINTR) continue; - log_error("failed to poll: %m\n"); + log_error("failed to poll: %m"); goto out; } @@ -356,10 +381,8 @@ 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); @@ -373,7 +396,7 @@ out: udev_monitor_unref(worker_monitor); event->state = EVENT_QUEUED; free(worker); - log_error("fork of child failed: %m\n"); + log_error("fork of child failed: %m"); break; default: /* close monitor, but keep address around */ @@ -382,17 +405,17 @@ 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); children++; - log_debug("seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid); + log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid); break; } } -static void event_run(struct event *event) -{ +static void event_run(struct event *event) { struct udev_list_node *loop; udev_list_node_foreach(loop, &worker_list) { @@ -404,7 +427,7 @@ static void event_run(struct event *event) count = udev_monitor_send_device(monitor, worker->monitor, event->dev); if (count < 0) { - log_error("worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count); + log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count); kill(worker->pid, SIGKILL); worker->state = WORKER_KILLED; continue; @@ -413,13 +436,14 @@ 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; } if (children >= children_max) { if (children_max > 1) - log_debug("maximum number (%i) of children reached\n", children); + log_debug("maximum number (%i) of children reached", children); return; } @@ -427,11 +451,10 @@ static void event_run(struct event *event) worker_new(event); } -static int event_queue_insert(struct udev_device *dev) -{ +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; @@ -444,13 +467,8 @@ 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 - udev_queue_export_device_queued(udev_queue_export, dev); - log_debug("seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev), + log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev), udev_device_get_action(dev), udev_device_get_subsystem(dev)); event->state = EVENT_QUEUED; @@ -458,8 +476,7 @@ static int event_queue_insert(struct udev_device *dev) return 0; } -static void worker_kill(struct udev *udev) -{ +static void worker_kill(struct udev *udev) { struct udev_list_node *loop; udev_list_node_foreach(loop, &worker_list) { @@ -474,8 +491,7 @@ static void worker_kill(struct udev *udev) } /* lookup event for identical, parent, child device */ -static bool is_devpath_busy(struct event *event) -{ +static bool is_devpath_busy(struct event *event) { struct udev_list_node *loop; size_t common; @@ -527,12 +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; @@ -552,8 +562,7 @@ static bool is_devpath_busy(struct event *event) return false; } -static void event_queue_start(struct udev *udev) -{ +static void event_queue_start(struct udev *udev) { struct udev_list_node *loop; udev_list_node_foreach(loop, &event_list) { @@ -570,8 +579,7 @@ static void event_queue_start(struct udev *udev) } } -static void event_queue_cleanup(struct udev *udev, enum event_state match_type) -{ +static void event_queue_cleanup(struct udev *udev, enum event_state match_type) { struct udev_list_node *loop, *tmp; udev_list_node_foreach_safe(loop, tmp, &event_list) { @@ -580,12 +588,11 @@ 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); } } -static void worker_returned(int fd_worker) -{ +static void worker_returned(int fd_worker) { for (;;) { struct worker_message msg; ssize_t size; @@ -605,7 +612,7 @@ 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) @@ -617,8 +624,7 @@ static void worker_returned(int fd_worker) } /* receive the udevd message from userspace */ -static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) -{ +static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) { struct udev *udev = udev_ctrl_get_udev(uctrl); struct udev_ctrl_connection *ctrl_conn; struct udev_ctrl_msg *ctrl_msg = NULL; @@ -635,24 +641,24 @@ 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\n", 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); } if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) { - log_debug("udevd message (STOP_EXEC_QUEUE) received\n"); + log_debug("udevd message (STOP_EXEC_QUEUE) received"); stop_exec_queue = true; } if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) { - log_debug("udevd message (START_EXEC_QUEUE) received\n"); + log_debug("udevd message (START_EXEC_QUEUE) received"); stop_exec_queue = false; } if (udev_ctrl_get_reload(ctrl_msg) > 0) { - log_debug("udevd message (RELOAD) received\n"); + log_debug("udevd message (RELOAD) received"); reload = true; } @@ -669,14 +675,14 @@ static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) val[0] = '\0'; val = &val[1]; if (val[0] == '\0') { - log_debug("udevd message (ENV) received, unset '%s'\n", key); + log_debug("udevd message (ENV) received, unset '%s'", key); udev_add_property(udev, key, NULL); } else { - log_debug("udevd message (ENV) received, set '%s=%s'\n", key, val); + log_debug("udevd message (ENV) received, set '%s=%s'", key, val); udev_add_property(udev, key, val); } } else { - log_error("wrong key format '%s'\n", key); + log_error("wrong key format '%s'", key); } free(key); } @@ -685,15 +691,15 @@ static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) i = udev_ctrl_get_set_children_max(ctrl_msg); if (i >= 0) { - log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i); + log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i); children_max = i; } if (udev_ctrl_get_ping(ctrl_msg) > 0) - log_debug("udevd message (SYNC) received\n"); + log_debug("udevd message (SYNC) received"); if (udev_ctrl_get_exit(ctrl_msg) > 0) { - log_debug("udevd message (EXIT) received\n"); + log_debug("udevd message (EXIT) received"); udev_exit = true; /* keep reference to block the client until we exit */ udev_ctrl_connection_ref(ctrl_conn); @@ -703,20 +709,124 @@ out: return udev_ctrl_connection_unref(ctrl_conn); } -/* read inotify messages */ -static int handle_inotify(struct udev *udev) -{ +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(udev_device_get_sysname(dev), "dm-")) { + 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_CLOEXEC|O_NOFOLLOW|O_NONBLOCK); + if (fd >= 0) { + r = flock(fd, LOCK_EX|LOCK_NB); + if (r >= 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); + if (r < 0) + return r; + + 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) { - log_error("error getting buffer for inotify\n"); - return -1; + if (!buf) { + log_error("error getting buffer for inotify"); + return -ENOMEM; } nbytes = read(fd_inotify, buf, nbytes); @@ -726,35 +836,23 @@ 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\n", 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'\n", 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\n"); - 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); return 0; } -static void handle_signal(struct udev *udev, int signo) -{ +static void handle_signal(struct udev *udev, int signo) { switch (signo) { case SIGINT: case SIGTERM: @@ -775,28 +873,30 @@ static void handle_signal(struct udev *udev, int signo) if (worker->pid != pid) continue; - log_debug("worker [%u] exit\n", pid); + log_debug("worker [%u] exit", pid); if (WIFEXITED(status)) { if (WEXITSTATUS(status) != 0) - log_error("worker [%u] exit with return code %i\n", pid, WEXITSTATUS(status)); + log_error("worker [%u] exit with return code %i", + pid, WEXITSTATUS(status)); } else if (WIFSIGNALED(status)) { - log_error("worker [%u] terminated by signal %i (%s)\n", - pid, WTERMSIG(status), strsignal(WTERMSIG(status))); + log_error("worker [%u] terminated by signal %i (%s)", + pid, WTERMSIG(status), strsignal(WTERMSIG(status))); } else if (WIFSTOPPED(status)) { - log_error("worker [%u] stopped\n", pid); + log_error("worker [%u] stopped", pid); } else if (WIFCONTINUED(status)) { - log_error("worker [%u] continued\n", pid); + log_error("worker [%u] continued", pid); } else { - log_error("worker [%u] exit with status 0x%04x\n", pid, status); + log_error("worker [%u] exit with status 0x%04x", pid, status); } if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { if (worker->event) { - log_error("worker [%u] failed while handling '%s'\n", + 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); } @@ -812,79 +912,7 @@ static void handle_signal(struct udev *udev, int signo) } } -static void static_dev_create_from_modules(struct udev *udev) -{ - struct utsname kernel; - char modules[UTIL_PATH_SIZE]; - char buf[4096]; - FILE *f; - - if (uname(&kernel) < 0) { - log_error("uname failed: %m"); - return; - } - - strscpyl(modules, sizeof(modules), ROOTPREFIX "/lib/modules/", kernel.release, "/modules.devname", NULL); - f = fopen(modules, "re"); - if (f == NULL) - return; - - while (fgets(buf, sizeof(buf), f) != NULL) { - char *s; - const char *modname; - const char *devname; - const char *devno; - int maj, min; - char type; - mode_t mode; - char filename[UTIL_PATH_SIZE]; - - if (buf[0] == '#') - continue; - - modname = buf; - s = strchr(modname, ' '); - if (s == NULL) - continue; - s[0] = '\0'; - - devname = &s[1]; - s = strchr(devname, ' '); - if (s == NULL) - continue; - s[0] = '\0'; - - devno = &s[1]; - s = strchr(devno, ' '); - if (s == NULL) - s = strchr(devno, '\n'); - if (s != NULL) - s[0] = '\0'; - if (sscanf(devno, "%c%u:%u", &type, &maj, &min) != 3) - continue; - - mode = 0600; - if (type == 'c') - mode |= S_IFCHR; - else if (type == 'b') - mode |= S_IFBLK; - else - continue; - - strscpyl(filename, sizeof(filename), "/dev/", devname, NULL); - mkdir_parents_label(filename, 0755); - label_context_set(filename, mode); - log_debug("mknod '%s' %c%u:%u\n", filename, type, maj, min); - if (mknod(filename, mode, makedev(maj, min)) < 0 && errno == EEXIST) - utimensat(AT_FDCWD, filename, NULL, 0); - label_context_clear(); - } - - fclose(f); -} - -static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) -{ +static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) { int ctrl = -1, netlink = -1; int fd, n; @@ -913,7 +941,7 @@ static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) if (ctrl < 0 || netlink < 0) return -1; - log_debug("ctrl=%i netlink=%i\n", ctrl, netlink); + log_debug("ctrl=%i netlink=%i", ctrl, netlink); *rctrl = ctrl; *rnetlink = netlink; return 0; @@ -925,18 +953,22 @@ static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) * udev.children-max= events are fully serialized if set to 1 * udev.exec-delay= delay execution of every executed program */ -static void kernel_cmdline_options(struct udev *udev) -{ - char *line, *w, *state; +static void kernel_cmdline_options(struct udev *udev) { + _cleanup_free_ char *line = NULL; + const char *word, *state; size_t l; + int r; - if (read_one_line_file("/proc/cmdline", &line) < 0) + r = proc_cmdline(&line); + if (r < 0) + log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r)); + 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; @@ -956,16 +988,16 @@ static void kernel_cmdline_options(struct udev *udev) children_max = strtoul(opt + 18, NULL, 0); } else if (startswith(opt, "udev.exec-delay=")) { 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); } - - free(line); } -int main(int argc, char *argv[]) -{ +int main(int argc, char *argv[]) { struct udev *udev; sigset_t mask; int daemonize = false; @@ -975,6 +1007,7 @@ int main(int argc, char *argv[]) { "debug", no_argument, NULL, 'D' }, { "children-max", required_argument, NULL, 'c' }, { "exec-delay", required_argument, NULL, 'e' }, + { "event-timeout", required_argument, NULL, 't' }, { "resolve-names", required_argument, NULL, 'N' }, { "help", no_argument, NULL, 'h' }, { "version", no_argument, NULL, 'V' }, @@ -994,8 +1027,11 @@ int main(int argc, char *argv[]) log_set_target(LOG_TARGET_AUTO); log_parse_environment(); log_open(); + udev_set_log_fn(udev, udev_main_log); - log_debug("version %s\n", VERSION); + log_set_max_level(udev_get_log_priority(udev)); + + log_debug("version %s", VERSION); label_init("/dev"); for (;;) { @@ -1015,6 +1051,10 @@ int main(int argc, char *argv[]) case 'e': exec_delay = strtoul(optarg, NULL, 0); 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; log_set_max_level(LOG_DEBUG); @@ -1029,7 +1069,7 @@ int main(int argc, char *argv[]) resolve_names = -1; } else { fprintf(stderr, "resolve-names must be early, late or never\n"); - log_error("resolve-names must be early, late or never\n"); + log_error("resolve-names must be early, late or never"); goto exit; } break; @@ -1039,6 +1079,7 @@ int main(int argc, char *argv[]) " --debug\n" " --children-max=\n" " --exec-delay=\n" + " --event-timeout=\n" " --resolve-names=early|late|never\n" " --version\n" " --help\n" @@ -1056,7 +1097,7 @@ int main(int argc, char *argv[]) if (getuid() != 0) { fprintf(stderr, "root privileges required\n"); - log_error("root privileges required\n"); + log_error("root privileges required"); goto exit; } @@ -1067,7 +1108,6 @@ int main(int argc, char *argv[]) mkdir("/run/udev", 0755); dev_setup(NULL); - static_dev_create_from_modules(udev); /* before opening new files, make sure std{in,out,err} fds are in a sane state */ if (daemonize) { @@ -1083,7 +1123,7 @@ int main(int argc, char *argv[]) close(fd); } else { fprintf(stderr, "cannot open /dev/null\n"); - log_error("cannot open /dev/null\n"); + log_error("cannot open /dev/null"); } } @@ -1098,13 +1138,13 @@ int main(int argc, char *argv[]) monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink); if (monitor == NULL) { - log_error("error taking over netlink socket\n"); + log_error("error taking over netlink socket"); rc = 3; goto exit; } /* get our own cgroup, we regularly kill everything udev has left behind */ - if (cg_get_by_pid(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0) + if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0) udev_cgroup = NULL; } else { /* open control and netlink socket */ @@ -1120,7 +1160,7 @@ int main(int argc, char *argv[]) monitor = udev_monitor_new_from_netlink(udev, "kernel"); if (monitor == NULL) { fprintf(stderr, "error initializing netlink socket\n"); - log_error("error initializing netlink socket\n"); + log_error("error initializing netlink socket"); rc = 3; goto exit; } @@ -1129,26 +1169,19 @@ int main(int argc, char *argv[]) if (udev_monitor_enable_receiving(monitor) < 0) { fprintf(stderr, "error binding netlink socket\n"); - log_error("error binding netlink socket\n"); + log_error("error binding netlink socket"); rc = 3; goto exit; } if (udev_ctrl_enable_receiving(udev_ctrl) < 0) { fprintf(stderr, "error binding udev control socket\n"); - log_error("error binding udev control socket\n"); + log_error("error binding udev control socket"); rc = 1; 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\n"); - goto exit; - } + udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024); if (daemonize) { pid_t pid; @@ -1158,7 +1191,7 @@ int main(int argc, char *argv[]) case 0: break; case -1: - log_error("fork of daemon failed: %m\n"); + log_error("fork of daemon failed: %m"); rc = 4; goto exit; default: @@ -1168,12 +1201,12 @@ int main(int argc, char *argv[]) setsid(); - write_one_line_file("/proc/self/oom_score_adj", "-1000"); + write_string_file("/proc/self/oom_score_adj", "-1000"); } else { sd_notify(1, "READY=1"); } - print_kmsg("starting version " VERSION "\n"); + log_info("starting version " VERSION "\n"); if (!debug) { int fd; @@ -1190,7 +1223,7 @@ int main(int argc, char *argv[]) fd_inotify = udev_watch_init(udev); if (fd_inotify < 0) { fprintf(stderr, "error initializing inotify\n"); - log_error("error initializing inotify\n"); + log_error("error initializing inotify"); rc = 4; goto exit; } @@ -1202,7 +1235,7 @@ int main(int argc, char *argv[]) fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC); if (fd_signal < 0) { fprintf(stderr, "error creating signalfd\n"); - log_error("error creating signalfd\n"); + log_error("error creating signalfd"); rc = 5; goto exit; } @@ -1210,7 +1243,7 @@ int main(int argc, char *argv[]) /* unnamed socket from workers to the main daemon */ if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) { fprintf(stderr, "error creating socketpair\n"); - log_error("error creating socketpair\n"); + log_error("error creating socketpair"); rc = 6; goto exit; } @@ -1220,33 +1253,33 @@ int main(int argc, char *argv[]) rules = udev_rules_new(udev, resolve_names); if (rules == NULL) { - log_error("error reading rules\n"); + log_error("error reading rules"); 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; fd_ep = epoll_create1(EPOLL_CLOEXEC); if (fd_ep < 0) { - log_error("error creating epoll fd: %m\n"); + log_error("error creating epoll fd: %m"); goto exit; } if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 || @@ -1254,7 +1287,7 @@ int main(int argc, char *argv[]) epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 || epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 || epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) { - log_error("fail to add fds to epoll: %m\n"); + log_error("fail to add fds to epoll: %m"); goto exit; } @@ -1267,9 +1300,11 @@ int main(int argc, char *argv[]) children_max += CPU_COUNT(&cpu_set) * 2; } } - log_debug("set children_max to %u\n", children_max); + log_debug("set children_max to %u", children_max); - udev_rules_apply_static_dev_perms(rules); + rc = udev_rules_apply_static_dev_perms(rules); + if (rc < 0) + log_error("failed to apply permissions on static device nodes - %s", strerror(-rc)); udev_list_node_init(&event_list); udev_list_node_init(&worker_list); @@ -1304,12 +1339,12 @@ 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; @@ -1318,8 +1353,20 @@ 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; @@ -1329,36 +1376,41 @@ int main(int argc, char *argv[]) /* timeout */ if (udev_exit) { - log_error("timeout, giving up waiting for workers to finish\n"); + log_error("timeout, giving up waiting for workers to finish"); break; } /* kill idle workers */ if (udev_list_node_is_empty(&event_list)) { - log_debug("cleanup idle workers\n"); + log_debug("cleanup idle workers"); worker_kill(udev); } /* 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) > 30 * 1000 * 1000) { - log_error("worker [%u] %s timeout; kill it\n", 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\n", - udev_device_get_seqnum(worker->event->dev), worker->event->devpath); + 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, true); + 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; } } } @@ -1380,7 +1432,7 @@ 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)) @@ -1453,8 +1505,8 @@ 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); @@ -1469,7 +1521,6 @@ 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();