X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=udev%2Fudevd.c;h=df5c1995bc9f06033c140b02b767b02b31986066;hp=6de1196421eacd096da57fea3c78e95bb3cb1358;hb=0e4fa2abfefd6fd955e8cada7caf2733dabc529e;hpb=f2b937440c682bca3837263a1bafd78e30b17e4d diff --git a/udev/udevd.c b/udev/udevd.c index 6de119642..df5c1995b 100644 --- a/udev/udevd.c +++ b/udev/udevd.c @@ -1,5 +1,5 @@ /* - * Copyright (C) 2004-2008 Kay Sievers + * Copyright (C) 2004-2009 Kay Sievers * Copyright (C) 2004 Chris Friesen * Copyright (C) 2009 Canonical Ltd. * Copyright (C) 2009 Scott James Remnant @@ -24,431 +24,563 @@ #include #include #include +#include #include #include #include #include #include #include +#include +#include +#include +#include +#include #include #include #include #include #include -#ifdef HAVE_INOTIFY #include -#endif +#include #include "udev.h" +#include "sd-daemon.h" #define UDEVD_PRIORITY -4 #define UDEV_PRIORITY -2 -/* maximum limit of forked childs */ -#define UDEVD_MAX_CHILDS 256 - -static int debug; +static bool debug; static void log_fn(struct udev *udev, int priority, const char *file, int line, const char *fn, const char *format, va_list args) { if (debug) { - fprintf(stderr, "[%d] %s: ", (int) getpid(), fn); - vfprintf(stderr, format, args); + char buf[1024]; + struct timeval tv; + struct timezone tz; + + vsnprintf(buf, sizeof(buf), format, args); + gettimeofday(&tv, &tz); + fprintf(stderr, "%llu.%06u [%u] %s: %s", + (unsigned long long) tv.tv_sec, (unsigned int) tv.tv_usec, + (int) getpid(), fn, buf); } else { vsyslog(priority, format, args); } } -static void reap_sigchilds(void); - -static int debug_trace; static struct udev_rules *rules; +static struct udev_queue_export *udev_queue_export; static struct udev_ctrl *udev_ctrl; -static struct udev_monitor *kernel_monitor; -static volatile int sigchilds_waiting; -static volatile int udev_exit; -static volatile int reload_config; -static volatile int signal_received; -static volatile pid_t settle_pid; -static int run_exec_q; -static int stop_exec_q; -static int max_childs; -static int childs; +static struct udev_monitor *monitor; +static int worker_watch[2] = { -1, -1 }; +static pid_t settle_pid; +static bool stop_exec_queue; +static bool reload_config; +static int children; +static int children_max; +static int exec_delay; +static sigset_t orig_sigmask; static struct udev_list_node event_list; +static struct udev_list_node worker_list; +static bool udev_exit; +static volatile sig_atomic_t worker_exit; + +enum poll_fd { + FD_CONTROL, + FD_NETLINK, + FD_INOTIFY, + FD_SIGNAL, + FD_WORKER, +}; + +static struct pollfd pfd[] = { + [FD_NETLINK] = { .events = POLLIN, .fd = -1 }, + [FD_WORKER] = { .events = POLLIN, .fd = -1 }, + [FD_SIGNAL] = { .events = POLLIN, .fd = -1 }, + [FD_INOTIFY] = { .events = POLLIN, .fd = -1 }, + [FD_CONTROL] = { .events = POLLIN, .fd = -1 }, +}; enum event_state { + EVENT_UNDEF, EVENT_QUEUED, - EVENT_FINISHED, - EVENT_FAILED, + EVENT_RUNNING, +}; + +struct event { + struct udev_list_node node; + struct udev *udev; + struct udev_device *dev; + enum event_state state; + int exitcode; + unsigned long long int delaying_seqnum; + unsigned long long int seqnum; + const char *devpath; + size_t devpath_len; + const char *devpath_old; + dev_t devnum; + bool is_block; + int ifindex; }; -static struct udev_event *node_to_event(struct udev_list_node *node) +static struct event *node_to_event(struct udev_list_node *node) { char *event; event = (char *)node; - event -= offsetof(struct udev_event, node); - return (struct udev_event *)event; + event -= offsetof(struct event, node); + return (struct event *)event; } -static void export_event_state(struct udev_event *event, enum event_state state) -{ - char filename[UTIL_PATH_SIZE]; - char filename_failed[UTIL_PATH_SIZE]; - size_t start; - - /* location of queue file */ - snprintf(filename, sizeof(filename), "%s/.udev/queue/%llu", - udev_get_dev_path(event->udev), udev_device_get_seqnum(event->dev)); - - /* location of failed file */ - util_strlcpy(filename_failed, udev_get_dev_path(event->udev), sizeof(filename_failed)); - util_strlcat(filename_failed, "/", sizeof(filename_failed)); - start = util_strlcat(filename_failed, ".udev/failed/", sizeof(filename_failed)); - util_strlcat(filename_failed, udev_device_get_devpath(event->dev), sizeof(filename_failed)); - util_path_encode(&filename_failed[start], sizeof(filename_failed) - start); - - switch (state) { - case EVENT_QUEUED: - if(unlink(filename_failed) == 0) - util_delete_path(event->udev, filename_failed); - util_create_path(event->udev, filename); - udev_selinux_setfscreatecon(event->udev, filename, S_IFLNK); - symlink(udev_device_get_devpath(event->dev), filename); - udev_selinux_resetfscreatecon(event->udev); - break; - case EVENT_FINISHED: - if (udev_device_get_devpath_old(event->dev) != NULL) { - /* "move" event - rename failed file to current name, do not delete failed */ - char filename_failed_old[UTIL_PATH_SIZE]; - - util_strlcpy(filename_failed_old, udev_get_dev_path(event->udev), sizeof(filename_failed_old)); - util_strlcat(filename_failed_old, "/", sizeof(filename_failed_old)); - start = util_strlcat(filename_failed_old, ".udev/failed/", sizeof(filename_failed_old)); - util_strlcat(filename_failed_old, udev_device_get_devpath_old(event->dev), sizeof(filename_failed_old)); - util_path_encode(&filename_failed_old[start], sizeof(filename) - start); - - if (rename(filename_failed_old, filename_failed) == 0) - info(event->udev, "renamed devpath, moved failed state of '%s' to %s'\n", - udev_device_get_devpath_old(event->dev), udev_device_get_devpath(event->dev)); - } else { - if (unlink(filename_failed) == 0) - util_delete_path(event->udev, filename_failed); - } +enum worker_state { + WORKER_UNDEF, + WORKER_RUNNING, + WORKER_IDLE, + WORKER_KILLED, +}; - unlink(filename); +struct worker { + struct udev_list_node node; + struct udev *udev; + int refcount; + pid_t pid; + struct udev_monitor *monitor; + enum worker_state state; + struct event *event; +}; - /* clean up possibly empty queue directory */ - if (udev_list_is_empty(&event_list)) - util_delete_path(event->udev, filename); - break; - case EVENT_FAILED: - /* move failed event to the failed directory */ - util_create_path(event->udev, filename_failed); - rename(filename, filename_failed); - - /* clean up possibly empty queue directory */ - if (udev_list_is_empty(&event_list)) - util_delete_path(event->udev, filename); - break; - } +/* passed from worker to main process */ +struct worker_message { + pid_t pid; + int exitcode; +}; + +static struct worker *node_to_worker(struct udev_list_node *node) +{ + char *worker; - return; + worker = (char *)node; + worker -= offsetof(struct worker, node); + return (struct worker *)worker; } -static void event_queue_delete(struct udev_event *event) +static void event_queue_delete(struct event *event) { udev_list_node_remove(&event->node); /* mark as failed, if "add" event returns non-zero */ - if (event->exitstatus && strcmp(udev_device_get_action(event->dev), "add") == 0) - export_event_state(event, EVENT_FAILED); + if (event->exitcode != 0 && strcmp(udev_device_get_action(event->dev), "remove") != 0) + udev_queue_export_device_failed(udev_queue_export, event->dev); else - export_event_state(event, EVENT_FINISHED); + udev_queue_export_device_finished(udev_queue_export, event->dev); + info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode); udev_device_unref(event->dev); - udev_event_unref(event); + free(event); +} + +static void event_sig_handler(int signum) +{ + switch (signum) { + case SIGALRM: + _exit(1); + break; + case SIGTERM: + worker_exit = true; + break; + } } -static void asmlinkage event_sig_handler(int signum) +static struct worker *worker_ref(struct worker *worker) { - if (signum == SIGALRM) - exit(1); + worker->refcount++; + return worker; } -static void event_fork(struct udev_event *event) +static void worker_unref(struct worker *worker) { + worker->refcount--; + if (worker->refcount > 0) + return; + + udev_list_node_remove(&worker->node); + udev_monitor_unref(worker->monitor); + children--; + info(worker->udev, "worker [%u] cleaned up\n", worker->pid); + free(worker); +} + +static void worker_new(struct event *event) +{ + struct worker *worker; + struct udev_monitor *worker_monitor; pid_t pid; struct sigaction act; - int err; - if (debug_trace) { - event->trace = 1; - fprintf(stderr, "fork %s (%llu)\n", - udev_device_get_syspath(event->dev), - udev_device_get_seqnum(event->dev)); + /* listen for new events */ + worker_monitor = udev_monitor_new_from_netlink(event->udev, NULL); + if (worker_monitor == NULL) + return; + /* allow the main daemon netlink address to send devices to the worker */ + udev_monitor_allow_unicast_sender(worker_monitor, monitor); + udev_monitor_enable_receiving(worker_monitor); + + worker = calloc(1, sizeof(struct worker)); + if (worker == NULL) { + udev_monitor_unref(worker_monitor); + return; } + /* worker + event reference */ + worker->refcount = 2; + worker->udev = event->udev; pid = fork(); switch (pid) { - case 0: - /* child */ + case 0: { + sigset_t sigmask; + struct udev_device *dev; + struct pollfd pmon = { + .fd = udev_monitor_get_fd(worker_monitor), + .events = POLLIN, + }; + + udev_queue_export_unref(udev_queue_export); + udev_monitor_unref(monitor); udev_ctrl_unref(udev_ctrl); - logging_close(); - logging_init("udevd-event"); + close(pfd[FD_SIGNAL].fd); + close(worker_watch[READ_END]); + udev_log_close(); + udev_log_init("udevd-work"); setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY); /* set signal handlers */ memset(&act, 0x00, sizeof(act)); - act.sa_handler = (void (*)(int)) event_sig_handler; + act.sa_handler = event_sig_handler; sigemptyset (&act.sa_mask); act.sa_flags = 0; + sigaction(SIGTERM, &act, NULL); sigaction(SIGALRM, &act, NULL); - /* reset to default */ - act.sa_handler = SIG_DFL; - sigaction(SIGINT, &act, NULL); - sigaction(SIGTERM, &act, NULL); - sigaction(SIGCHLD, &act, NULL); - sigaction(SIGHUP, &act, NULL); + /* unblock SIGALRM */ + sigfillset(&sigmask); + sigdelset(&sigmask, SIGALRM); + sigprocmask(SIG_SETMASK, &sigmask, NULL); + /* SIGTERM is unblocked in ppoll() */ + sigdelset(&sigmask, SIGTERM); - /* set timeout to prevent hanging processes */ - alarm(UDEV_EVENT_TIMEOUT); + /* request TERM signal if parent exits */ + prctl(PR_SET_PDEATHSIG, SIGTERM); - /* apply rules, create node, symlinks */ - err = udev_event_execute_rules(event, rules); + /* initial device */ + dev = event->dev; - /* rules may change/disable the timeout */ - if (udev_device_get_event_timeout(event->dev) >= 0) - alarm(udev_device_get_event_timeout(event->dev)); + do { + struct udev_event *udev_event; + struct worker_message msg = {}; + int err; + int failed = 0; - /* execute RUN= */ - if (err == 0 && !event->ignore_device && udev_get_run(event->udev)) - udev_event_execute_run(event); + info(event->udev, "seq %llu running\n", udev_device_get_seqnum(dev)); + udev_event = udev_event_new(dev); + if (udev_event == NULL) + _exit(3); - /* apply/restore inotify watch */ - if (err == 0 && event->inotify_watch) { - udev_watch_begin(event->udev, event->dev); - udev_device_update_db(event->dev); - } + /* set timeout to prevent hanging processes */ + alarm(UDEV_EVENT_TIMEOUT); + + if (exec_delay > 0) + udev_event->exec_delay = exec_delay; + + /* apply rules, create node, symlinks */ + err = udev_event_execute_rules(udev_event, rules); - /* send processed event back to the kernel netlink socket */ - udev_monitor_send_device(kernel_monitor, event->dev); + /* rules may change/disable the timeout */ + if (udev_device_get_event_timeout(dev) >= 0) + alarm(udev_device_get_event_timeout(dev)); - info(event->udev, "seq %llu exit with %i\n", udev_device_get_seqnum(event->dev), err); - logging_close(); - if (err != 0) - exit(1); + if (err == 0) + failed = udev_event_execute_run(udev_event, &orig_sigmask); + + alarm(0); + + /* apply/restore inotify watch */ + if (err == 0 && udev_event->inotify_watch) { + udev_watch_begin(udev_event->udev, dev); + udev_device_update_db(dev); + } + + /* send processed event back to libudev listeners */ + udev_monitor_send_device(worker_monitor, NULL, dev); + + /* send udevd the result of the event execution */ + if (err != 0) + msg.exitcode = err; + else if (failed != 0) + msg.exitcode = failed; + msg.pid = getpid(); + send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0); + + info(event->udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err); + udev_event_unref(udev_event); + udev_device_unref(dev); + dev = NULL; + + /* wait for more device messages or signal from udevd */ + while (!worker_exit) { + int fdcount; + + fdcount = ppoll(&pmon, 1, NULL, &sigmask); + if (fdcount < 0) + continue; + + if (pmon.revents & POLLIN) { + dev = udev_monitor_receive_device(worker_monitor); + if (dev != NULL) + break; + } + } + } while (dev != NULL); + + udev_monitor_unref(worker_monitor); + udev_log_close(); exit(0); + } case -1: + udev_monitor_unref(worker_monitor); + event->state = EVENT_QUEUED; + free(worker); err(event->udev, "fork of child failed: %m\n"); - event_queue_delete(event); break; default: - /* get SIGCHLD in main loop */ - info(event->udev, "seq %llu forked, pid [%d], '%s' '%s', %ld seconds old\n", - udev_device_get_seqnum(event->dev), - pid, - udev_device_get_action(event->dev), - udev_device_get_subsystem(event->dev), - time(NULL) - event->queue_time); - event->pid = pid; - childs++; + /* close monitor, but keep address around */ + udev_monitor_disconnect(worker_monitor); + worker->monitor = worker_monitor; + worker->pid = pid; + worker->state = WORKER_RUNNING; + worker->event = event; + event->state = EVENT_RUNNING; + udev_list_node_append(&worker->node, &worker_list); + children++; + info(event->udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid); + break; } } -static void event_queue_insert(struct udev_event *event) +static void event_run(struct event *event, bool force) { - char filename[UTIL_PATH_SIZE]; - int fd; - - event->queue_time = time(NULL); + struct udev_list_node *loop; - export_event_state(event, EVENT_QUEUED); - info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(event->dev), - udev_device_get_action(event->dev), udev_device_get_subsystem(event->dev)); + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); + ssize_t count; - util_strlcpy(filename, udev_get_dev_path(event->udev), sizeof(filename)); - util_strlcat(filename, "/.udev/uevent_seqnum", sizeof(filename)); - fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644); - if (fd >= 0) { - char str[32]; - int len; + if (worker->state != WORKER_IDLE) + continue; - len = sprintf(str, "%llu\n", udev_device_get_seqnum(event->dev)); - write(fd, str, len); - close(fd); + count = udev_monitor_send_device(monitor, worker->monitor, event->dev); + if (count < 0) { + err(event->udev, "worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count); + kill(worker->pid, SIGKILL); + worker->state = WORKER_KILLED; + continue; + } + worker_ref(worker); + worker->event = event; + worker->state = WORKER_RUNNING; + event->state = EVENT_RUNNING; + return; } - - udev_list_node_append(&event->node, &event_list); - run_exec_q = 1; - - /* run all events with a timeout set immediately */ - if (udev_device_get_timeout(event->dev) > 0) { - event_fork(event); + if (!force && children >= children_max) { + if (children_max > 1) + info(event->udev, "maximum number (%i) of children reached\n", children); return; } + + /* start new worker and pass initial device */ + worker_new(event); } -static int mem_size_mb(void) +static int event_queue_insert(struct udev_device *dev) { - FILE* f; - char buf[4096]; - long int memsize = -1; + struct event *event; - f = fopen("/proc/meminfo", "r"); - if (f == NULL) + event = calloc(1, sizeof(struct event)); + if (event == NULL) return -1; - while (fgets(buf, sizeof(buf), f) != NULL) { - long int value; + event->udev = udev_device_get_udev(dev); + event->dev = dev; + event->seqnum = udev_device_get_seqnum(dev); + event->devpath = udev_device_get_devpath(dev); + event->devpath_len = strlen(event->devpath); + event->devpath_old = udev_device_get_devpath_old(dev); + event->devnum = udev_device_get_devnum(dev); + event->is_block = (strcmp("block", udev_device_get_subsystem(dev)) == 0); + event->ifindex = udev_device_get_ifindex(dev); + + udev_queue_export_device_queued(udev_queue_export, dev); + info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev), + udev_device_get_action(dev), udev_device_get_subsystem(dev)); + + event->state = EVENT_QUEUED; + udev_list_node_append(&event->node, &event_list); - if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) { - memsize = value / 1024; - break; - } + /* run all events with a timeout set immediately */ + if (udev_device_get_timeout(dev) > 0) { + event_run(event, true); + return 0; } - fclose(f); - return memsize; + return 0; } -static int compare_devpath(const char *running, const char *waiting) +static void worker_kill(struct udev *udev, int retain) { - int i = 0; + struct udev_list_node *loop; + int max; - while (running[i] != '\0' && running[i] == waiting[i]) - i++; + if (children <= retain) + return; - /* identical device event found */ - if (running[i] == '\0' && waiting[i] == '\0') - return 1; + max = children - retain; - /* parent device event found */ - if (running[i] == '\0' && waiting[i] == '/') - return 2; + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); - /* child device event found */ - if (running[i] == '/' && waiting[i] == '\0') - return 3; + if (max-- <= 0) + break; - /* no matching event */ - return 0; + if (worker->state == WORKER_KILLED) + continue; + + worker->state = WORKER_KILLED; + kill(worker->pid, SIGTERM); + } } -/* lookup event for identical, parent, child, or physical device */ -static int devpath_busy(struct udev_event *event) +/* lookup event for identical, parent, child device */ +static bool is_devpath_busy(struct event *event) { struct udev_list_node *loop; + size_t common; - if (event->delaying_seqnum > 0) { - } /* check if queue contains events we depend on */ udev_list_node_foreach(loop, &event_list) { - struct udev_event *loop_event = node_to_event(loop); + struct event *loop_event = node_to_event(loop); /* we already found a later event, earlier can not block us, no need to check again */ - if (udev_device_get_seqnum(loop_event->dev) < event->delaying_seqnum) + if (loop_event->seqnum < event->delaying_seqnum) continue; /* event we checked earlier still exists, no need to check again */ - if (udev_device_get_seqnum(loop_event->dev) == event->delaying_seqnum) - return 2; + if (loop_event->seqnum == event->delaying_seqnum) + return true; /* found ourself, no later event can block us */ - if (udev_device_get_seqnum(loop_event->dev) >= udev_device_get_seqnum(event->dev)) + if (loop_event->seqnum >= event->seqnum) break; + /* check major/minor */ + if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block) + return true; + + /* check network device ifindex */ + if (event->ifindex != 0 && event->ifindex == loop_event->ifindex) + return true; + /* check our old name */ - if (udev_device_get_devpath_old(event->dev) != NULL) - if (strcmp(udev_device_get_devpath(loop_event->dev), udev_device_get_devpath_old(event->dev)) == 0) { - event->delaying_seqnum = udev_device_get_seqnum(loop_event->dev); - return 3; - } + if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) { + event->delaying_seqnum = loop_event->seqnum; + return true; + } + + /* compare devpath */ + common = MIN(loop_event->devpath_len, event->devpath_len); + + /* one devpath is contained in the other? */ + if (memcmp(loop_event->devpath, event->devpath, common) != 0) + continue; - /* check identical, parent, or child device event */ - if (compare_devpath(udev_device_get_devpath(loop_event->dev), udev_device_get_devpath(event->dev)) != 0) { - dbg(event->udev, "%llu, device event still pending %llu (%s)\n", - udev_device_get_seqnum(event->dev), - udev_device_get_seqnum(loop_event->dev), - udev_device_get_devpath(loop_event->dev)); - event->delaying_seqnum = udev_device_get_seqnum(loop_event->dev); - return 4; + /* identical device event found */ + if (loop_event->devpath_len == event->devpath_len) { + /* devices names might have changed/swapped in the meantime */ + if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block)) + continue; + if (event->ifindex != 0 && event->ifindex != loop_event->ifindex) + continue; + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check for our major:minor number */ - if (major(udev_device_get_devnum(event->dev)) > 0 && - udev_device_get_devnum(loop_event->dev) == udev_device_get_devnum(event->dev) && - strcmp(udev_device_get_subsystem(event->dev), udev_device_get_subsystem(loop_event->dev)) == 0) { - dbg(event->udev, "%llu, device event still pending %llu (%d:%d)\n", - udev_device_get_seqnum(event->dev), - udev_device_get_seqnum(loop_event->dev), - major(udev_device_get_devnum(loop_event->dev)), minor(udev_device_get_devnum(loop_event->dev))); - event->delaying_seqnum = udev_device_get_seqnum(loop_event->dev); - return 5; + /* parent device event found */ + if (event->devpath[common] == '/') { + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check physical device event (special case of parent) */ - if (udev_device_get_physdevpath(event->dev) != NULL && - strcmp(udev_device_get_action(event->dev), "add") == 0) - if (compare_devpath(udev_device_get_devpath(loop_event->dev), - udev_device_get_physdevpath(event->dev)) != 0) { - dbg(event->udev, "%llu, physical device event still pending %llu (%s)\n", - udev_device_get_seqnum(event->dev), - udev_device_get_seqnum(loop_event->dev), - udev_device_get_devpath(loop_event->dev)); - event->delaying_seqnum = udev_device_get_seqnum(loop_event->dev); - return 6; - } + /* child device event found */ + if (loop_event->devpath[common] == '/') { + event->delaying_seqnum = loop_event->seqnum; + return true; + } + + /* no matching device */ + continue; } - return 0; + + return false; } -/* serializes events for the identical and parent and child devices */ -static void event_queue_manager(struct udev *udev) +static void events_start(struct udev *udev) { struct udev_list_node *loop; - struct udev_list_node *tmp; -start_over: - if (udev_list_is_empty(&event_list)) { - if (childs > 0) { - err(udev, "event list empty, but childs count is %i", childs); - childs = 0; - } - return; - } - - udev_list_node_foreach_safe(loop, tmp, &event_list) { - struct udev_event *loop_event = node_to_event(loop); - - if (childs >= max_childs) { - info(udev, "maximum number (%i) of childs reached\n", childs); - break; - } + udev_list_node_foreach(loop, &event_list) { + struct event *event = node_to_event(loop); - if (loop_event->pid != 0) + if (event->state != EVENT_QUEUED) continue; /* do not start event if parent or child event is still running */ - if (devpath_busy(loop_event) != 0) { - dbg(udev, "delay seq %llu (%s)\n", - udev_device_get_seqnum(loop_event->dev), - udev_device_get_devpath(loop_event->dev)); + if (is_devpath_busy(event)) { + dbg(udev, "delay seq %llu (%s)\n", event->seqnum, event->devpath); continue; } - event_fork(loop_event); - dbg(udev, "moved seq %llu to running list\n", udev_device_get_seqnum(loop_event->dev)); + event_run(event, false); + } +} + +static void worker_returned(void) +{ + for (;;) { + struct worker_message msg; + ssize_t size; + struct udev_list_node *loop; + + size = recv(pfd[FD_WORKER].fd, &msg, sizeof(struct worker_message), MSG_DONTWAIT); + if (size != sizeof(struct worker_message)) + break; + + /* lookup worker who sent the signal */ + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); + + if (worker->pid != msg.pid) + continue; - /* retry if events finished in the meantime */ - if (sigchilds_waiting) { - sigchilds_waiting = 0; - reap_sigchilds(); - goto start_over; + /* worker returned */ + worker->event->exitcode = msg.exitcode; + event_queue_delete(worker->event); + worker->event = NULL; + if (worker->state != WORKER_KILLED) + worker->state = WORKER_IDLE; + worker_unref(worker); + break; } } } @@ -469,22 +601,22 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl) if (i >= 0) { info(udev, "udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i); udev_set_log_priority(udev, i); + worker_kill(udev, 0); } if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) { info(udev, "udevd message (STOP_EXEC_QUEUE) received\n"); - stop_exec_q = 1; + stop_exec_queue = true; } if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) { info(udev, "udevd message (START_EXEC_QUEUE) received\n"); - stop_exec_q = 0; - event_queue_manager(udev); + stop_exec_queue = false; } if (udev_ctrl_get_reload_rules(ctrl_msg) > 0) { info(udev, "udevd message (RELOAD_RULES) received\n"); - reload_config = 1; + reload_config = true; } str = udev_ctrl_get_set_env(ctrl_msg); @@ -511,17 +643,20 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl) } free(key); } + worker_kill(udev, 0); } - i = udev_ctrl_get_set_max_childs(ctrl_msg); + i = udev_ctrl_get_set_children_max(ctrl_msg); if (i >= 0) { - info(udev, "udevd message (SET_MAX_CHILDS) received, max_childs=%i\n", i); - max_childs = i; + info(udev, "udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i); + children_max = i; } settle_pid = udev_ctrl_get_settle(ctrl_msg); if (settle_pid > 0) { info(udev, "udevd message (SETTLE) received\n"); + kill(settle_pid, SIGUSR1); + settle_pid = 0; } udev_ctrl_msg_unref(ctrl_msg); } @@ -533,39 +668,43 @@ static int handle_inotify(struct udev *udev) char *buf; struct inotify_event *ev; - if ((ioctl(inotify_fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0)) + if ((ioctl(pfd[FD_INOTIFY].fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0)) return 0; buf = malloc(nbytes); if (buf == NULL) { - err(udev, "error getting buffer for inotify, disable watching\n"); - close(inotify_fd); - inotify_fd = -1; - return 0; + err(udev, "error getting buffer for inotify\n"); + return -1; } - read(inotify_fd, buf, nbytes); + nbytes = read(pfd[FD_INOTIFY].fd, buf, nbytes); for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) { struct udev_device *dev; ev = (struct inotify_event *)(buf + pos); if (ev->len) { - dbg(udev, "inotify event: %x for %s\n", ev->mask, ev->name); - reload_config = 1; + const char *s; + + info(udev, "inotify event: %x for %s\n", ev->mask, ev->name); + s = strstr(ev->name, ".rules"); + if (s == NULL) + continue; + if (strlen(s) != strlen(".rules")) + continue; + reload_config = true; continue; } dev = udev_watch_lookup(udev, ev->wd); if (dev != NULL) { - dbg(udev, "inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev)); + info(udev, "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; info(udev, "device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev)); - util_strlcpy(filename, udev_device_get_syspath(dev), sizeof(filename)); - util_strlcat(filename, "/uevent", sizeof(filename)); + util_strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL); fd = open(filename, O_WRONLY); if (fd < 0 || write(fd, "change", 6) < 0) info(udev, "error writing uevent: %m\n"); @@ -579,150 +718,367 @@ static int handle_inotify(struct udev *udev) } - free (buf); + free(buf); return 0; } -static void asmlinkage sig_handler(int signum) +static void handle_signal(struct udev *udev, int signo) { - switch (signum) { - case SIGINT: - case SIGTERM: - udev_exit = 1; - break; - case SIGCHLD: - /* set flag, then write to pipe if needed */ - sigchilds_waiting = 1; - break; - case SIGHUP: - reload_config = 1; - break; + switch (signo) { + case SIGINT: + case SIGTERM: + udev_exit = true; + break; + case SIGCHLD: + for (;;) { + pid_t pid; + int status; + struct udev_list_node *loop, *tmp; + + pid = waitpid(-1, &status, WNOHANG); + if (pid <= 0) + break; + + udev_list_node_foreach_safe(loop, tmp, &worker_list) { + struct worker *worker = node_to_worker(loop); + + if (worker->pid != pid) + continue; + + info(udev, "worker [%u] exit\n", pid); + if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) { + err(udev, "worker [%u] unexpectedly returned with status 0x%04x\n", pid, status); + if (worker->event != NULL) { + err(udev, "worker [%u] failed while handling '%s'\n", pid, worker->event->devpath); + worker->event->exitcode = -32; + event_queue_delete(worker->event); + /* drop reference from running event */ + worker_unref(worker); + } + } + worker_unref(worker); + break; + } + } + break; + case SIGHUP: + reload_config = true; + break; } +} + +static void static_dev_create_from_modules(struct udev *udev) +{ + struct utsname kernel; + char modules[UTIL_PATH_SIZE]; + char buf[4096]; + FILE *f; - signal_received = 1; + uname(&kernel); + util_strscpyl(modules, sizeof(modules), "/lib/modules/", kernel.release, "/modules.devname", NULL); + f = fopen(modules, "r"); + 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; + + if (type == 'c') + mode = 0600 | S_IFCHR; + else if (type == 'b') + mode = 0600 | S_IFBLK; + else + continue; + + util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/", devname, NULL); + util_create_path_selinux(udev, filename); + udev_selinux_setfscreatecon(udev, filename, mode); + info(udev, "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); + udev_selinux_resetfscreatecon(udev); + } + + fclose(f); } -static void udev_done(int pid, int exitstatus) +static int copy_dev_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth) { - struct udev_list_node *loop; + struct dirent *dent; - /* find event associated with pid and delete it */ - udev_list_node_foreach(loop, &event_list) { - struct udev_event *loop_event = node_to_event(loop); - - if (loop_event->pid == pid) { - info(loop_event->udev, "seq %llu cleanup, pid [%d], status %i, %ld seconds old\n", - udev_device_get_seqnum(loop_event->dev), loop_event->pid, - exitstatus, time(NULL) - loop_event->queue_time); - loop_event->exitstatus = exitstatus; - if (debug_trace) - fprintf(stderr, "exit %s (%llu)\n", - udev_device_get_syspath(loop_event->dev), - udev_device_get_seqnum(loop_event->dev)); - event_queue_delete(loop_event); - childs--; - - /* there may be dependent events waiting */ - run_exec_q = 1; - return; + for (dent = readdir(dir_from); dent != NULL; dent = readdir(dir_from)) { + struct stat stats; + + if (dent->d_name[0] == '.') + continue; + if (fstatat(dirfd(dir_from), dent->d_name, &stats, AT_SYMLINK_NOFOLLOW) != 0) + continue; + + if (S_ISBLK(stats.st_mode) || S_ISCHR(stats.st_mode)) { + udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, stats.st_mode & 0777); + if (mknodat(dirfd(dir_to), dent->d_name, stats.st_mode, stats.st_rdev) == 0) { + fchmodat(dirfd(dir_to), dent->d_name, stats.st_mode & 0777, 0); + fchownat(dirfd(dir_to), dent->d_name, stats.st_uid, stats.st_gid, 0); + } else { + utimensat(dirfd(dir_to), dent->d_name, NULL, 0); + } + udev_selinux_resetfscreatecon(udev); + } else if (S_ISLNK(stats.st_mode)) { + char target[UTIL_PATH_SIZE]; + ssize_t len; + + len = readlinkat(dirfd(dir_from), dent->d_name, target, sizeof(target)); + if (len <= 0 || len == (ssize_t)sizeof(target)) + continue; + target[len] = '\0'; + udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFLNK); + if (symlinkat(target, dirfd(dir_to), dent->d_name) < 0 && errno == EEXIST) + utimensat(dirfd(dir_to), dent->d_name, NULL, AT_SYMLINK_NOFOLLOW); + udev_selinux_resetfscreatecon(udev); + } else if (S_ISDIR(stats.st_mode)) { + DIR *dir2_from, *dir2_to; + + if (maxdepth == 0) + continue; + + udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFDIR|0755); + mkdirat(dirfd(dir_to), dent->d_name, 0755); + udev_selinux_resetfscreatecon(udev); + + dir2_to = fdopendir(openat(dirfd(dir_to), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC)); + if (dir2_to == NULL) + continue; + + dir2_from = fdopendir(openat(dirfd(dir_from), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC)); + if (dir2_from == NULL) { + closedir(dir2_to); + continue; + } + + copy_dev_dir(udev, dir2_from, dir2_to, maxdepth-1); + + closedir(dir2_to); + closedir(dir2_from); } } + + return 0; } -static void reap_sigchilds(void) +static void static_dev_create_links(struct udev *udev, DIR *dir) { - pid_t pid; - int status; + struct stdlinks { + const char *link; + const char *target; + }; + static const struct stdlinks stdlinks[] = { + { "core", "/proc/kcore" }, + { "fd", "/proc/self/fd" }, + { "stdin", "/proc/self/fd/0" }, + { "stdout", "/proc/self/fd/1" }, + { "stderr", "/proc/self/fd/2" }, + }; + unsigned int i; - while (1) { - pid = waitpid(-1, &status, WNOHANG); - if (pid <= 0) - break; - if (WIFEXITED(status)) - status = WEXITSTATUS(status); - else if (WIFSIGNALED(status)) - status = WTERMSIG(status) + 128; - else - status = 0; - udev_done(pid, status); + for (i = 0; i < ARRAY_SIZE(stdlinks); i++) { + struct stat sb; + + if (stat(stdlinks[i].target, &sb) == 0) { + udev_selinux_setfscreateconat(udev, dirfd(dir), stdlinks[i].link, S_IFLNK); + if (symlinkat(stdlinks[i].target, dirfd(dir), stdlinks[i].link) < 0 && errno == EEXIST) + utimensat(dirfd(dir), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW); + udev_selinux_resetfscreatecon(udev); + } } } -static void cleanup_queue_dir(struct udev *udev) +static void static_dev_create_from_devices(struct udev *udev, DIR *dir) +{ + DIR *dir_from; + + dir_from = opendir(LIBEXECDIR "/devices"); + if (dir_from == NULL) + return; + copy_dev_dir(udev, dir_from, dir, 8); + closedir(dir_from); +} + +static void static_dev_create(struct udev *udev) { - char dirname[UTIL_PATH_SIZE]; - char filename[UTIL_PATH_SIZE]; DIR *dir; - util_strlcpy(filename, udev_get_dev_path(udev), sizeof(filename)); - util_strlcat(filename, "/.udev/uevent_seqnum", sizeof(filename)); - unlink(filename); + dir = opendir(udev_get_dev_path(udev)); + if (dir == NULL) + return; - util_strlcpy(dirname, udev_get_dev_path(udev), sizeof(dirname)); - util_strlcat(dirname, "/.udev/queue", sizeof(dirname)); - dir = opendir(dirname); - if (dir != NULL) { - while (1) { - struct dirent *dent; + static_dev_create_links(udev, dir); + static_dev_create_from_devices(udev, dir); - dent = readdir(dir); - if (dent == NULL || dent->d_name[0] == '\0') - break; - if (dent->d_name[0] == '.') - continue; - util_strlcpy(filename, dirname, sizeof(filename)); - util_strlcat(filename, "/", sizeof(filename)); - util_strlcat(filename, dent->d_name, sizeof(filename)); - unlink(filename); + closedir(dir); +} + +static int mem_size_mb(void) +{ + FILE *f; + char buf[4096]; + long int memsize = -1; + + f = fopen("/proc/meminfo", "r"); + if (f == NULL) + return -1; + + while (fgets(buf, sizeof(buf), f) != NULL) { + long int value; + + if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) { + memsize = value / 1024; + break; } - closedir(dir); - rmdir(dirname); } + + fclose(f); + return memsize; } -static void export_initial_seqnum(struct udev *udev) +static int convert_db(struct udev *udev) { char filename[UTIL_PATH_SIZE]; - int fd; - char seqnum[32]; - ssize_t len = 0; - - util_strlcpy(filename, udev_get_sys_path(udev), sizeof(filename)); - util_strlcat(filename, "/kernel/uevent_seqnum", sizeof(filename)); - fd = open(filename, O_RDONLY); - if (fd >= 0) { - len = read(fd, seqnum, sizeof(seqnum)-1); - close(fd); - } - if (len <= 0) { - strcpy(seqnum, "0\n"); - len = 3; - } - util_strlcpy(filename, udev_get_dev_path(udev), sizeof(filename)); - util_strlcat(filename, "/.udev/uevent_seqnum", sizeof(filename)); + FILE *f; + struct udev_enumerate *udev_enumerate; + struct udev_list_entry *list_entry; + + /* current database */ + util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/data", NULL); + if (access(filename, F_OK) >= 0) + return 0; + + /* make sure we do not get here again */ util_create_path(udev, filename); - fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644); - if (fd >= 0) { - write(fd, seqnum, len); - close(fd); + mkdir(filename, 0755); + + /* old database */ + util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/db", NULL); + if (access(filename, F_OK) < 0) + return 0; + + f = fopen("/dev/kmsg", "w"); + if (f != NULL) { + fprintf(f, "<30>udev[%u]: converting old udev database\n", getpid()); + fclose(f); } + + udev_enumerate = udev_enumerate_new(udev); + if (udev_enumerate == NULL) + return -1; + udev_enumerate_scan_devices(udev_enumerate); + udev_list_entry_foreach(list_entry, udev_enumerate_get_list_entry(udev_enumerate)) { + struct udev_device *device; + + device = udev_device_new_from_syspath(udev, udev_list_entry_get_name(list_entry)); + if (device == NULL) + continue; + + /* try to find the old database for devices without a current one */ + if (udev_device_read_db(device, NULL) < 0) { + bool have_db; + const char *id; + struct stat stats; + char devpath[UTIL_PATH_SIZE]; + char from[UTIL_PATH_SIZE]; + + have_db = false; + + /* find database in old location */ + id = udev_device_get_id_filename(device); + util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", id, NULL); + if (lstat(from, &stats) == 0) { + if (!have_db) { + udev_device_read_db(device, from); + have_db = true; + } + unlink(from); + } + + /* find old database with $subsys:$sysname name */ + util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), + "/.udev/db/", udev_device_get_subsystem(device), ":", + udev_device_get_sysname(device), NULL); + if (lstat(from, &stats) == 0) { + if (!have_db) { + udev_device_read_db(device, from); + have_db = true; + } + unlink(from); + } + + /* find old database with the encoded devpath name */ + util_path_encode(udev_device_get_devpath(device), devpath, sizeof(devpath)); + util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", devpath, NULL); + if (lstat(from, &stats) == 0) { + if (!have_db) { + udev_device_read_db(device, from); + have_db = true; + } + unlink(from); + } + + /* write out new database */ + if (have_db) + udev_device_update_db(device); + } + udev_device_unref(device); + } + udev_enumerate_unref(udev_enumerate); + return 0; } int main(int argc, char *argv[]) { struct udev *udev; int fd; - struct sigaction act; - const char *value; - int daemonize = 0; + FILE *f; + sigset_t mask; + int daemonize = false; int resolve_names = 1; static const struct option options[] = { { "daemon", no_argument, NULL, 'd' }, - { "debug-trace", no_argument, NULL, 't' }, { "debug", no_argument, NULL, 'D' }, + { "children-max", required_argument, NULL, 'c' }, + { "exec-delay", required_argument, NULL, 'e' }, + { "resolve-names", required_argument, NULL, 'N' }, { "help", no_argument, NULL, 'h' }, { "version", no_argument, NULL, 'V' }, - { "resolve-names", required_argument, NULL, 'N' }, {} }; int rc = 1; @@ -731,27 +1087,53 @@ int main(int argc, char *argv[]) if (udev == NULL) goto exit; - logging_init("udevd"); + udev_log_init("udevd"); udev_set_log_fn(udev, log_fn); info(udev, "version %s\n", VERSION); udev_selinux_init(udev); - while (1) { + /* make sure, that our runtime dir exists and is writable */ + if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) < 0) { + /* try to create our own subdirectory, do not create parent directories */ + mkdir(udev_get_run_config_path(udev), 0755); + + if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) >= 0) { + /* directory seems writable now */ + udev_set_run_path(udev, udev_get_run_config_path(udev)); + } else { + /* fall back to /dev/.udev */ + char filename[UTIL_PATH_SIZE]; + + util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev", NULL); + if (udev_set_run_path(udev, filename) == NULL) + goto exit; + mkdir(udev_get_run_path(udev), 0755); + } + } + /* relabel runtime dir only if it resides below /dev */ + if (strncmp(udev_get_run_path(udev), udev_get_dev_path(udev), strlen(udev_get_dev_path(udev))) == 0) + udev_selinux_lsetfilecon(udev, udev_get_run_path(udev), 0755); + info(udev, "runtime dir '%s'\n", udev_get_run_path(udev)); + + for (;;) { int option; - option = getopt_long(argc, argv, "dDthV", options, NULL); + option = getopt_long(argc, argv, "c:deDtN:hV", options, NULL); if (option == -1) break; switch (option) { case 'd': - daemonize = 1; + daemonize = true; break; - case 't': - debug_trace = 1; + case 'c': + children_max = strtoul(optarg, NULL, 0); + break; + case 'e': + exec_delay = strtoul(optarg, NULL, 0); break; case 'D': - debug = 1; + debug = true; if (udev_get_log_priority(udev) < LOG_INFO) udev_set_log_priority(udev, LOG_INFO); break; @@ -769,8 +1151,15 @@ int main(int argc, char *argv[]) } break; case 'h': - printf("Usage: udevd [--help] [--daemon] [--debug-trace] [--debug] " - "[--resolve-names=early|late|never] [--version]\n"); + printf("Usage: udevd OPTIONS\n" + " --daemon\n" + " --debug\n" + " --children-max=\n" + " --exec-delay=\n" + " --resolve-names=early|late|never\n" + " --version\n" + " --help\n" + "\n"); goto exit; case 'V': printf("%s\n", VERSION); @@ -780,13 +1169,55 @@ int main(int argc, char *argv[]) } } + /* + * read the kernel commandline, in case we need to get into debug mode + * udev.log-priority= syslog priority + * udev.children-max= events are fully serialized if set to 1 + * + */ + f = fopen("/proc/cmdline", "r"); + if (f != NULL) { + char cmdline[4096]; + + if (fgets(cmdline, sizeof(cmdline), f) != NULL) { + char *pos; + + pos = strstr(cmdline, "udev.log-priority="); + if (pos != NULL) { + pos += strlen("udev.log-priority="); + udev_set_log_priority(udev, util_log_priority(pos)); + } + + pos = strstr(cmdline, "udev.children-max="); + if (pos != NULL) { + pos += strlen("udev.children-max="); + children_max = strtoul(pos, NULL, 0); + } + + pos = strstr(cmdline, "udev.exec-delay="); + if (pos != NULL) { + pos += strlen("udev.exec-delay="); + exec_delay = strtoul(pos, NULL, 0); + } + } + fclose(f); + } + if (getuid() != 0) { fprintf(stderr, "root privileges required\n"); err(udev, "root privileges required\n"); goto exit; } - /* make sure std{in,out,err} fd's are in a sane state */ + /* set umask before creating any file/directory */ + chdir("/"); + umask(022); + + /* create standard links, copy static nodes, create nodes from modules */ + static_dev_create(udev); + static_dev_create_from_modules(udev); + + /* before opening new files, make sure std{in,out,err} fds are in a sane state */ fd = open("/dev/null", O_RDWR); if (fd < 0) { fprintf(stderr, "cannot open /dev/null\n"); @@ -797,39 +1228,108 @@ int main(int argc, char *argv[]) if (write(STDERR_FILENO, 0, 0) < 0) dup2(fd, STDERR_FILENO); - /* init control socket, bind() ensures, that only one udevd instance is running */ - udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH); + /* udevadm control socket */ + if (sd_listen_fds(true) == 1 && sd_is_socket(SD_LISTEN_FDS_START, AF_LOCAL, SOCK_DGRAM, -1)) + udev_ctrl = udev_ctrl_new_from_fd(udev, SD_LISTEN_FDS_START); + else + udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH); if (udev_ctrl == NULL) { - fprintf(stderr, "error initializing control socket"); - err(udev, "error initializing udevd socket"); + fprintf(stderr, "error initializing udev control socket"); + err(udev, "error initializing udev control socket"); rc = 1; goto exit; } - if (udev_ctrl_enable_receiving(udev_ctrl) < 0) { - fprintf(stderr, "error binding control socket, seems udevd is already running\n"); - err(udev, "error binding control socket, seems udevd is already running\n"); + fprintf(stderr, "error binding udev control socket\n"); + err(udev, "error binding udev control socket\n"); rc = 1; goto exit; } + pfd[FD_CONTROL].fd = udev_ctrl_get_fd(udev_ctrl); - kernel_monitor = udev_monitor_new_from_netlink(udev, "kernel"); - if (kernel_monitor == NULL || udev_monitor_enable_receiving(kernel_monitor) < 0) { + monitor = udev_monitor_new_from_netlink(udev, "kernel"); + if (monitor == NULL || udev_monitor_enable_receiving(monitor) < 0) { fprintf(stderr, "error initializing netlink socket\n"); err(udev, "error initializing netlink socket\n"); rc = 3; goto exit; } - udev_monitor_set_receive_buffer_size(kernel_monitor, 128*1024*1024); + udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024); + pfd[FD_NETLINK].fd = udev_monitor_get_fd(monitor); + + pfd[FD_INOTIFY].fd = udev_watch_init(udev); + if (pfd[FD_INOTIFY].fd < 0) { + fprintf(stderr, "error initializing inotify\n"); + err(udev, "error initializing inotify\n"); + rc = 4; + goto exit; + } + + if (udev_get_rules_path(udev) != NULL) { + inotify_add_watch(pfd[FD_INOTIFY].fd, udev_get_rules_path(udev), + IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + } else { + char filename[UTIL_PATH_SIZE]; + struct stat statbuf; + + inotify_add_watch(pfd[FD_INOTIFY].fd, LIBEXECDIR "/rules.d", + IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + inotify_add_watch(pfd[FD_INOTIFY].fd, SYSCONFDIR "/udev/rules.d", + IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + + /* watch dynamic rules directory */ + util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/rules.d", NULL); + if (stat(filename, &statbuf) != 0) { + util_create_path(udev, filename); + mkdir(filename, 0755); + } + inotify_add_watch(pfd[FD_INOTIFY].fd, filename, + IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + } + udev_watch_restore(udev); + + /* block and listen to all signals on signalfd */ + sigfillset(&mask); + sigprocmask(SIG_SETMASK, &mask, &orig_sigmask); + pfd[FD_SIGNAL].fd = signalfd(-1, &mask, 0); + if (pfd[FD_SIGNAL].fd < 0) { + fprintf(stderr, "error getting signalfd\n"); + err(udev, "error getting signalfd\n"); + rc = 5; + goto exit; + } + + /* unnamed socket from workers to the main daemon */ + if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) { + fprintf(stderr, "error getting socketpair\n"); + err(udev, "error getting socketpair\n"); + rc = 6; + goto exit; + } + pfd[FD_WORKER].fd = worker_watch[READ_END]; rules = udev_rules_new(udev, resolve_names); if (rules == NULL) { err(udev, "error reading rules\n"); goto exit; } - udev_list_init(&event_list); - cleanup_queue_dir(udev); - export_initial_seqnum(udev); + + udev_queue_export = udev_queue_export_new(udev); + if (udev_queue_export == NULL) { + err(udev, "error creating queue file\n"); + goto exit; + } + + /* if needed, convert old database from earlier udev version */ + convert_db(udev); + + if (!debug) { + dup2(fd, STDIN_FILENO); + dup2(fd, STDOUT_FILENO); + dup2(fd, STDERR_FILENO); + } + if (fd > STDERR_FILENO) + close(fd); if (daemonize) { pid_t pid; @@ -837,218 +1337,151 @@ int main(int argc, char *argv[]) pid = fork(); switch (pid) { case 0: - dbg(udev, "daemonized fork running\n"); break; case -1: err(udev, "fork of daemon failed: %m\n"); rc = 4; goto exit; default: - dbg(udev, "child [%u] running, parent exits\n", pid); rc = 0; goto exit; } + } else { + sd_notify(1, "READY=1"); } - /* redirect std{out,err} */ - if (!debug && !debug_trace) { - dup2(fd, STDIN_FILENO); - dup2(fd, STDOUT_FILENO); - dup2(fd, STDERR_FILENO); - } - if (fd > STDERR_FILENO) - close(fd); - - /* set scheduling priority for the daemon */ + /* set scheduling priority for the main daemon process */ setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY); - chdir("/"); - umask(022); setsid(); - /* OOM_DISABLE == -17 */ - fd = open("/proc/self/oom_adj", O_RDWR); - if (fd < 0) - err(udev, "error disabling OOM: %m\n"); - else { - write(fd, "-17", 3); - close(fd); - } - - fd = open("/dev/kmsg", O_WRONLY); - if (fd > 0) { - const char *ver_str = "<6>udev: starting version " VERSION "\n"; - char path[UTIL_PATH_SIZE]; - struct stat statbuf; - - write(fd, ver_str, strlen(ver_str)); - util_strlcpy(path, udev_get_sys_path(udev), sizeof(path)); - util_strlcat(path, "/class/mem/null", sizeof(path)); - if (lstat(path, &statbuf) == 0) { - if (S_ISDIR(statbuf.st_mode)) { - const char *depr_str = - "<6>udev: deprecated sysfs layout; update the kernel or " - "disable CONFIG_SYSFS_DEPRECATED; some udev features will " - "not work correctly\n"; - - write(fd, depr_str, strlen(depr_str)); - } - } - close(fd); + f = fopen("/dev/kmsg", "w"); + if (f != NULL) { + fprintf(f, "<30>udev[%u]: starting version " VERSION "\n", getpid()); + fclose(f); } - /* set signal handlers */ - memset(&act, 0x00, sizeof(struct sigaction)); - act.sa_handler = (void (*)(int)) sig_handler; - sigemptyset(&act.sa_mask); - act.sa_flags = SA_RESTART; - sigaction(SIGINT, &act, NULL); - sigaction(SIGTERM, &act, NULL); - sigaction(SIGCHLD, &act, NULL); - sigaction(SIGHUP, &act, NULL); - - /* watch rules directory */ - udev_watch_init(udev); - if (inotify_fd >= 0) { - if (udev_get_rules_path(udev) != NULL) { - inotify_add_watch(inotify_fd, udev_get_rules_path(udev), - IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + fd = open("/proc/self/oom_score_adj", O_RDWR); + if (fd < 0) { + /* Fallback to old interface */ + fd = open("/proc/self/oom_adj", O_RDWR); + if (fd < 0) { + err(udev, "error disabling OOM: %m\n"); } else { - char filename[UTIL_PATH_SIZE]; - - inotify_add_watch(inotify_fd, UDEV_PREFIX "/lib/udev/rules.d", - IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); - inotify_add_watch(inotify_fd, SYSCONFDIR "/udev/rules.d", - IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); - - /* watch dynamic rules directory */ - util_strlcpy(filename, udev_get_dev_path(udev), sizeof(filename)); - util_strlcat(filename, "/.udev/rules.d", sizeof(filename)); - inotify_add_watch(inotify_fd, filename, - IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + /* OOM_DISABLE == -17 */ + write(fd, "-17", 3); + close(fd); } - - udev_watch_restore(udev); + } else { + write(fd, "-1000", 5); + close(fd); } - /* in trace mode run one event after the other */ - if (debug_trace) { - max_childs = 1; - } else { + if (children_max <= 0) { int memsize = mem_size_mb(); + + /* set value depending on the amount of RAM */ if (memsize > 0) - max_childs = 128 + (memsize / 4); + children_max = 128 + (memsize / 8); else - max_childs = UDEVD_MAX_CHILDS; + children_max = 128; } - /* possibly overwrite maximum limit of executed events */ - value = getenv("UDEVD_MAX_CHILDS"); - if (value) - max_childs = strtoul(value, NULL, 10); - info(udev, "initialize max_childs to %u\n", max_childs); + info(udev, "set children_max to %u\n", children_max); + + udev_rules_apply_static_dev_perms(rules); + + udev_list_init(&event_list); + udev_list_init(&worker_list); while (!udev_exit) { - sigset_t blocked_mask, orig_mask; - struct pollfd pfd[4]; - struct pollfd *ctrl_poll, *monitor_poll, *inotify_poll = NULL; - int nfds = 0; int fdcount; + int timeout; - sigfillset(&blocked_mask); - sigprocmask(SIG_SETMASK, &blocked_mask, &orig_mask); - if (signal_received) { - sigprocmask(SIG_SETMASK, &orig_mask, NULL); - goto handle_signals; - } + /* set timeout to kill idle workers */ + if (udev_list_is_empty(&event_list) && children > 2) + timeout = 3 * 1000; + else + timeout = -1; + /* wait for events */ + fdcount = poll(pfd, ARRAY_SIZE(pfd), timeout); + if (fdcount < 0) + continue; - ctrl_poll = &pfd[nfds++]; - ctrl_poll->fd = udev_ctrl_get_fd(udev_ctrl); - ctrl_poll->events = POLLIN; + /* timeout - kill idle workers */ + if (fdcount == 0) + worker_kill(udev, 2); - monitor_poll = &pfd[nfds++]; - monitor_poll->fd = udev_monitor_get_fd(kernel_monitor); - monitor_poll->events = POLLIN; + /* event has finished */ + if (pfd[FD_WORKER].revents & POLLIN) + worker_returned(); - if (inotify_fd >= 0) { - inotify_poll = &pfd[nfds++]; - inotify_poll->fd = inotify_fd; - inotify_poll->events = POLLIN; - } + /* get kernel uevent */ + if (pfd[FD_NETLINK].revents & POLLIN) { + struct udev_device *dev; - fdcount = ppoll(pfd, nfds, NULL, &orig_mask); - sigprocmask(SIG_SETMASK, &orig_mask, NULL); - if (fdcount < 0) { - if (errno == EINTR) - goto handle_signals; - err(udev, "error in select: %m\n"); - continue; + dev = udev_monitor_receive_device(monitor); + if (dev != NULL) + if (event_queue_insert(dev) < 0) + udev_device_unref(dev); } - /* get control message */ - if (ctrl_poll->revents & POLLIN) - handle_ctrl_msg(udev_ctrl); + /* start new events */ + if (!udev_list_is_empty(&event_list) && !stop_exec_queue) + events_start(udev); - /* get kernel uevent */ - if (monitor_poll->revents & POLLIN) { - struct udev_device *dev; + /* get signal */ + if (pfd[FD_SIGNAL].revents & POLLIN) { + struct signalfd_siginfo fdsi; + ssize_t size; - dev = udev_monitor_receive_device(kernel_monitor); - if (dev != NULL) { - struct udev_event *event; - - event = udev_event_new(dev); - if (event != NULL) - event_queue_insert(event); - else - udev_device_unref(dev); - } + size = read(pfd[FD_SIGNAL].fd, &fdsi, sizeof(struct signalfd_siginfo)); + if (size == sizeof(struct signalfd_siginfo)) + handle_signal(udev, fdsi.ssi_signo); } - /* rules directory inotify watch */ - if (inotify_poll && (inotify_poll->revents & POLLIN)) + /* device node and rules directory inotify watch */ + if (pfd[FD_INOTIFY].revents & POLLIN) handle_inotify(udev); -handle_signals: - signal_received = 0; + /* + * get control message + * + * This needs to be after the inotify handling, to make sure, + * that the settle signal is send back after the possibly generated + * "change" events by the inotify device node watch. + */ + if (pfd[FD_CONTROL].revents & POLLIN) + handle_ctrl_msg(udev_ctrl); /* rules changed, set by inotify or a HUP signal */ if (reload_config) { struct udev_rules *rules_new; - reload_config = 0; + worker_kill(udev, 0); rules_new = udev_rules_new(udev, resolve_names); if (rules_new != NULL) { udev_rules_unref(rules); rules = rules_new; } - } - - if (sigchilds_waiting) { - sigchilds_waiting = 0; - reap_sigchilds(); - } - - if (run_exec_q) { - run_exec_q = 0; - if (!stop_exec_q) - event_queue_manager(udev); - } - - if (settle_pid > 0) { - kill(settle_pid, SIGUSR1); - settle_pid = 0; + reload_config = 0; } } - cleanup_queue_dir(udev); + + udev_queue_export_cleanup(udev_queue_export); rc = 0; exit: + udev_queue_export_unref(udev_queue_export); udev_rules_unref(rules); udev_ctrl_unref(udev_ctrl); - if (inotify_fd >= 0) - close(inotify_fd); - udev_monitor_unref(kernel_monitor); + if (pfd[FD_SIGNAL].fd >= 0) + close(pfd[FD_SIGNAL].fd); + if (worker_watch[READ_END] >= 0) + close(worker_watch[READ_END]); + if (worker_watch[WRITE_END] >= 0) + close(worker_watch[WRITE_END]); + udev_monitor_unref(monitor); udev_selinux_exit(udev); udev_unref(udev); - logging_close(); + udev_log_close(); return rc; }