X-Git-Url: https://www.chiark.greenend.org.uk/ucgi/~ianmdlvl/git?p=elogind.git;a=blobdiff_plain;f=udev%2Fudevd.c;h=72f9b180d398d37294a8511ad6c9bd72ee045cd3;hp=a711df53d4785c566da08568a5d04ceffc17ee97;hb=081be002eae9bc7fc03dbfd7b99b4d28a7c98f9d;hpb=d59f11e1407ec6fa26e3a6f20b2d404af6978199 diff --git a/udev/udevd.c b/udev/udevd.c index a711df53d..72f9b180d 100644 --- a/udev/udevd.c +++ b/udev/udevd.c @@ -1,674 +1,592 @@ /* - * Copyright (C) 2004-2006 Kay Sievers + * Copyright (C) 2004-2009 Kay Sievers * Copyright (C) 2004 Chris Friesen + * Copyright (C) 2009 Canonical Ltd. + * Copyright (C) 2009 Scott James Remnant * - * This program is free software; you can redistribute it and/or modify it - * under the terms of the GNU General Public License as published by the - * Free Software Foundation version 2 of the License. + * This program is free software: you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation, either version 2 of the License, or + * (at your option) any later version. * - * This program is distributed in the hope that it will be useful, but - * WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * - * You should have received a copy of the GNU General Public License along - * with this program; if not, write to the Free Software Foundation, Inc., - * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA. + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. * + * You should have received a copy of the GNU General Public License + * along with this program. If not, see . */ -#include "config.h" - #include #include #include #include #include #include +#include #include #include -#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 "udev.h" -#include "udev_rules.h" -#include "udev_selinux.h" #define UDEVD_PRIORITY -4 #define UDEV_PRIORITY -2 -/* maximum limit of forked childs */ -#define UDEVD_MAX_CHILDS 256 -/* start to throttle forking if maximum number of running childs in our session is reached */ -#define UDEVD_MAX_CHILDS_RUNNING 16 - -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); } } -struct udevd_uevent_msg { - struct udev *udev; - struct list_head node; - pid_t pid; - int exitstatus; - time_t queue_time; - char *action; - char *devpath; - char *subsystem; - char *driver; - dev_t devt; - unsigned long long seqnum; - char *devpath_old; - char *physdevpath; - unsigned int timeout; - char *envp[UEVENT_NUM_ENVP+1]; - char envbuf[]; -}; - -static int debug_trace; -static struct udev_rules rules; +static bool debug_trace; +static struct udev_rules *rules; +static struct udev_queue_export *udev_queue_export; static struct udev_ctrl *udev_ctrl; -static int uevent_netlink_sock = -1; -static int inotify_fd = -1; -static pid_t sid; - -static int signal_pipe[2] = {-1, -1}; -static volatile int sigchilds_waiting; -static volatile int udev_exit; -static volatile int reload_config; -static int run_exec_q; -static int stop_exec_q; +static struct udev_monitor *monitor; +static int worker_watch[2]; +static pid_t settle_pid; +static bool stop_exec_queue; +static bool reload_config; static int max_childs; -static int max_childs_running; -static char udev_log_env[32]; +static int childs; +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 LIST_HEAD(exec_list); -static LIST_HEAD(running_list); +static struct pollfd pfd[] = { + [FD_NETLINK] = { .events = POLLIN }, + [FD_WORKER] = { .events = POLLIN }, + [FD_SIGNAL] = { .events = POLLIN }, + [FD_INOTIFY] = { .events = POLLIN }, + [FD_CONTROL] = { .events = POLLIN }, +}; -static void asmlinkage udev_event_sig_handler(int signum) +enum event_state { + EVENT_UNDEF, + EVENT_QUEUED, + 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; +}; + +static struct event *node_to_event(struct udev_list_node *node) { - if (signum == SIGALRM) - exit(1); + char *event; + + event = (char *)node; + event -= offsetof(struct event, node); + return (struct event *)event; } -static int udev_event_process(struct udevd_uevent_msg *msg) -{ - struct sigaction act; - struct udevice *udevice; - int i; - int retval; - - /* set signal handlers */ - memset(&act, 0x00, sizeof(act)); - act.sa_handler = (void (*)(int)) udev_event_sig_handler; - sigemptyset (&act.sa_mask); - act.sa_flags = 0; - 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); - - /* trigger timeout to prevent hanging processes */ - alarm(UDEV_EVENT_TIMEOUT); - - /* reconstruct event environment from message */ - for (i = 0; msg->envp[i]; i++) - putenv(msg->envp[i]); - - udevice = udev_device_init(msg->udev); - if (udevice == NULL) - return -1; - strlcpy(udevice->action, msg->action, sizeof(udevice->action)); - sysfs_device_set_values(udevice->udev, udevice->dev, msg->devpath, msg->subsystem, msg->driver); - udevice->devpath_old = msg->devpath_old; - udevice->devt = msg->devt; +enum worker_state { + WORKER_UNDEF, + WORKER_RUNNING, + WORKER_IDLE, + WORKER_KILLED, +}; - retval = udev_device_event(&rules, udevice); +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; +}; - /* rules may change/disable the timeout */ - if (udevice->event_timeout >= 0) - alarm(udevice->event_timeout); +/* passed from worker to main process */ +struct worker_message { + pid_t pid; + int exitcode; +}; - /* run programs collected by RUN-key*/ - if (retval == 0 && !udevice->ignore_device && udev_get_run(msg->udev)) - retval = udev_rules_run(udevice); +static struct worker *node_to_worker(struct udev_list_node *node) +{ + char *worker; - udev_device_cleanup(udevice); - return retval; + worker = (char *)node; + worker -= offsetof(struct worker, node); + return (struct worker *)worker; } -enum event_state { - EVENT_QUEUED, - EVENT_FINISHED, - EVENT_FAILED, -}; - -static void export_event_state(struct udevd_uevent_msg *msg, enum event_state state) +static void event_queue_delete(struct event *event) { - char filename[PATH_SIZE]; - char filename_failed[PATH_SIZE]; - size_t start; - - /* location of queue file */ - snprintf(filename, sizeof(filename), "%s/.udev/queue/%llu", udev_get_dev_path(msg->udev), msg->seqnum); - - /* location of failed file */ - strlcpy(filename_failed, udev_get_dev_path(msg->udev), sizeof(filename_failed)); - strlcat(filename_failed, "/", sizeof(filename_failed)); - start = strlcat(filename_failed, ".udev/failed/", sizeof(filename_failed)); - strlcat(filename_failed, msg->devpath, sizeof(filename_failed)); - path_encode(&filename_failed[start], sizeof(filename_failed) - start); - - switch (state) { - case EVENT_QUEUED: - unlink(filename_failed); - delete_path(msg->udev, filename_failed); - create_path(msg->udev, filename); - selinux_setfscreatecon(msg->udev, filename, NULL, S_IFLNK); - symlink(msg->devpath, filename); - selinux_resetfscreatecon(msg->udev); - break; - case EVENT_FINISHED: - if (msg->devpath_old != NULL) { - /* "move" event - rename failed file to current name, do not delete failed */ - char filename_failed_old[PATH_SIZE]; - - strlcpy(filename_failed_old, udev_get_dev_path(msg->udev), sizeof(filename_failed_old)); - strlcat(filename_failed_old, "/", sizeof(filename_failed_old)); - start = strlcat(filename_failed_old, ".udev/failed/", sizeof(filename_failed_old)); - strlcat(filename_failed_old, msg->devpath_old, sizeof(filename_failed_old)); - path_encode(&filename_failed_old[start], sizeof(filename) - start); - - if (rename(filename_failed_old, filename_failed) == 0) - info(msg->udev, "renamed devpath, moved failed state of '%s' to %s'\n", - msg->devpath_old, msg->devpath); - } else { - unlink(filename_failed); - delete_path(msg->udev, filename_failed); - } + udev_list_node_remove(&event->node); - unlink(filename); - delete_path(msg->udev, filename); - break; - case EVENT_FAILED: - /* move failed event to the failed directory */ - create_path(msg->udev, filename_failed); - rename(filename, filename_failed); + /* mark as failed, if "add" event returns non-zero */ + if (event->exitcode != 0 && strcmp(udev_device_get_action(event->dev), "add") == 0) + udev_queue_export_device_failed(udev_queue_export, event->dev); + else + 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); + free(event); +} - /* clean up possibly empty queue directory */ - delete_path(msg->udev, filename); +static void event_sig_handler(int signum) +{ + switch (signum) { + case SIGALRM: + _exit(1); + break; + case SIGTERM: + worker_exit = true; break; } - - return; } -static void msg_queue_delete(struct udevd_uevent_msg *msg) +static struct worker *worker_ref(struct worker *worker) { - list_del(&msg->node); + worker->refcount++; + return worker; +} - /* mark as failed, if "add" event returns non-zero */ - if (msg->exitstatus && strcmp(msg->action, "add") == 0) - export_event_state(msg, EVENT_FAILED); - else - export_event_state(msg, EVENT_FINISHED); +static void worker_unref(struct worker *worker) +{ + worker->refcount--; + if (worker->refcount > 0) + return; - free(msg); + udev_list_node_remove(&worker->node); + udev_monitor_unref(worker->monitor); + childs--; + info(worker->udev, "worker [%u] cleaned up\n", worker->pid); + free(worker); } -static void udev_event_run(struct udevd_uevent_msg *msg) +static void worker_new(struct event *event) { + struct worker *worker; + struct udev_monitor *worker_monitor; pid_t pid; - int retval; + struct sigaction act; + + /* 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) + return; + /* worker + event reference */ + worker->refcount = 2; + worker->udev = event->udev; pid = fork(); switch (pid) { - case 0: - /* child */ - close(uevent_netlink_sock); + 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); - if (inotify_fd >= 0) - close(inotify_fd); - close(signal_pipe[READ_END]); - close(signal_pipe[WRITE_END]); - 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); - retval = udev_event_process(msg); - info(msg->udev, "seq %llu finished with %i\n", msg->seqnum, retval); + /* set signal handlers */ + memset(&act, 0x00, sizeof(act)); + act.sa_handler = event_sig_handler; + sigemptyset (&act.sa_mask); + act.sa_flags = 0; + sigaction(SIGTERM, &act, NULL); + sigaction(SIGALRM, &act, NULL); + + /* unblock SIGALRM */ + sigfillset(&sigmask); + sigdelset(&sigmask, SIGALRM); + sigprocmask(SIG_SETMASK, &sigmask, NULL); + /* SIGTERM is unblocked in ppoll() */ + sigdelset(&sigmask, SIGTERM); + + /* request TERM signal if parent exits */ + prctl(PR_SET_PDEATHSIG, SIGTERM); + + /* initial device */ + dev = event->dev; + + do { + struct udev_event *udev_event; + struct worker_message msg = {}; + int err; + int failed = 0; + + info(event->udev, "seq %llu running\n", udev_device_get_seqnum(dev)); + udev_event = udev_event_new(dev); + if (udev_event == NULL) + _exit(3); + + /* set timeout to prevent hanging processes */ + alarm(UDEV_EVENT_TIMEOUT); + + /* apply rules, create node, symlinks */ + err = udev_event_execute_rules(udev_event, rules); + + /* rules may change/disable the timeout */ + if (udev_device_get_event_timeout(dev) >= 0) + alarm(udev_device_get_event_timeout(dev)); + + /* execute RUN= */ + if (err == 0 && !udev_event->ignore_device && udev_get_run(udev_event->udev)) + failed = udev_event_execute_run(udev_event, + &orig_sigmask); + + /* reset alarm */ + 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); + } - logging_close(); - if (retval) - exit(1); + /* send processed event back to libudev listeners */ + udev_monitor_send_device(worker_monitor, NULL, dev); + + /* send back 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: - err(msg->udev, "fork of child failed: %s\n", strerror(errno)); - msg_queue_delete(msg); + udev_monitor_unref(worker_monitor); + event->state = EVENT_QUEUED; + free(worker); + err(event->udev, "fork of child failed: %m\n"); break; default: - /* get SIGCHLD in main loop */ - info(msg->udev, "seq %llu forked, pid [%d], '%s' '%s', %ld seconds old\n", - msg->seqnum, pid, msg->action, msg->subsystem, time(NULL) - msg->queue_time); - msg->pid = pid; + /* 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); + childs++; + info(event->udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid); + break; } } -static void msg_queue_insert(struct udevd_uevent_msg *msg) +static void event_run(struct event *event) { - char filename[PATH_SIZE]; - int fd; - - msg->queue_time = time(NULL); - - export_event_state(msg, EVENT_QUEUED); - info(msg->udev, "seq %llu queued, '%s' '%s'\n", msg->seqnum, msg->action, msg->subsystem); + struct udev_list_node *loop; - strlcpy(filename, udev_get_dev_path(msg->udev), sizeof(filename)); - 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; + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); + ssize_t count; - len = sprintf(str, "%llu\n", msg->seqnum); - write(fd, str, len); - close(fd); - } + if (worker->state != WORKER_IDLE) + continue; - /* run one event after the other in debug mode */ - if (debug_trace) { - list_add_tail(&msg->node, &running_list); - udev_event_run(msg); - waitpid(msg->pid, NULL, 0); - msg_queue_delete(msg); + 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; } - /* run all events with a timeout set immediately */ - if (msg->timeout != 0) { - list_add_tail(&msg->node, &running_list); - udev_event_run(msg); + if (childs >= max_childs) { + info(event->udev, "maximum number (%i) of childs reached\n", childs); return; } - list_add_tail(&msg->node, &exec_list); - run_exec_q = 1; + /* start new worker and pass initial device */ + worker_new(event); } -static int mem_size_mb(void) +static void 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) - return -1; + event = calloc(1, sizeof(struct event)); + if (event == NULL) + return; - 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); - if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) { - memsize = value / 1024; - break; - } - } + 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)); - fclose(f); - return memsize; + event->state = EVENT_QUEUED; + udev_list_node_append(&event->node, &event_list); + + /* run all events with a timeout set immediately */ + if (udev_device_get_timeout(dev) > 0) { + worker_new(event); + return; + } } -static int cpu_count(void) +static void worker_kill(struct udev *udev, int retain) { - FILE* f; - char buf[4096]; - int count = 0; + struct udev_list_node *loop; + int max; - f = fopen("/proc/stat", "r"); - if (f == NULL) - return -1; + if (childs <= retain) + return; - while (fgets(buf, sizeof(buf), f) != NULL) { - if (strncmp(buf, "cpu", 3) == 0 && isdigit(buf[3])) - count++; - } + max = childs - retain; - fclose(f); - if (count == 0) - return -1; - return count; + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); + + if (max-- <= 0) + break; + + if (worker->state == WORKER_KILLED) + continue; + + worker->state = WORKER_KILLED; + kill(worker->pid, SIGTERM); + } } -static int running_processes(void) +static int mem_size_mb(void) { - FILE* f; + FILE *f; char buf[4096]; - int running = -1; + long int memsize = -1; - f = fopen("/proc/stat", "r"); + f = fopen("/proc/meminfo", "r"); if (f == NULL) return -1; while (fgets(buf, sizeof(buf), f) != NULL) { - int value; + long int value; - if (sscanf(buf, "procs_running %u", &value) == 1) { - running = value; + if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) { + memsize = value / 1024; break; } } fclose(f); - return running; + return memsize; } -/* return the number of process es in our session, count only until limit */ -static int running_processes_in_session(pid_t session, int limit) +/* lookup event for identical, parent, child device */ +static bool is_devpath_busy(struct event *event) { - DIR *dir; - struct dirent *dent; - int running = 0; + struct udev_list_node *loop; + size_t common; - dir = opendir("/proc"); - if (!dir) - return -1; + /* check if queue contains events we depend on */ + udev_list_node_foreach(loop, &event_list) { + struct event *loop_event = node_to_event(loop); - /* read process info from /proc */ - for (dent = readdir(dir); dent != NULL; dent = readdir(dir)) { - int f; - char procdir[64]; - char line[256]; - const char *pos; - char state; - pid_t ppid, pgrp, sess; - int len; - - if (!isdigit(dent->d_name[0])) + /* we already found a later event, earlier can not block us, no need to check again */ + if (loop_event->seqnum < event->delaying_seqnum) continue; - snprintf(procdir, sizeof(procdir), "/proc/%s/stat", dent->d_name); - procdir[sizeof(procdir)-1] = '\0'; - - f = open(procdir, O_RDONLY); - if (f == -1) - continue; - - len = read(f, line, sizeof(line)-1); - close(f); - - if (len <= 0) - continue; - else - line[len] = '\0'; - - /* skip ugly program name */ - pos = strrchr(line, ')') + 2; - if (pos == NULL) - continue; + /* event we checked earlier still exists, no need to check again */ + if (loop_event->seqnum == event->delaying_seqnum) + return true; - if (sscanf(pos, "%c %d %d %d ", &state, &ppid, &pgrp, &sess) != 4) - continue; - - /* count only processes in our session */ - if (sess != session) - continue; - - /* count only running, no sleeping processes */ - if (state != 'R') - continue; - - running++; - if (limit > 0 && running >= limit) + /* found ourself, no later event can block us */ + if (loop_event->seqnum >= event->seqnum) break; - } - closedir(dir); - - return running; -} -static int compare_devpath(const char *running, const char *waiting) -{ - int i; - - for (i = 0; i < PATH_SIZE; i++) { - /* identical device event found */ - if (running[i] == '\0' && waiting[i] == '\0') - return 1; - - /* parent device event found */ - if (running[i] == '\0' && waiting[i] == '/') - return 2; - - /* child device event found */ - if (running[i] == '/' && waiting[i] == '\0') - return 3; - - /* no matching event */ - if (running[i] != waiting[i]) - break; - } - - return 0; -} - -/* lookup event for identical, parent, child, or physical device */ -static int devpath_busy(struct udevd_uevent_msg *msg, int limit) -{ - struct udevd_uevent_msg *loop_msg; - int childs_count = 0; - - /* check exec-queue which may still contain delayed events we depend on */ - list_for_each_entry(loop_msg, &exec_list, node) { - /* skip ourself and all later events */ - if (loop_msg->seqnum >= msg->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 our old name */ - if (msg->devpath_old != NULL) - if (strcmp(loop_msg->devpath , msg->devpath_old) == 0) - return 2; - - /* check identical, parent, or child device event */ - if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) { - dbg(msg->udev, "%llu, device event still pending %llu (%s)\n", - msg->seqnum, loop_msg->seqnum, loop_msg->devpath); - return 3; + if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) { + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check for our major:minor number */ - if (msg->devt && loop_msg->devt == msg->devt && - strcmp(msg->subsystem, loop_msg->subsystem) == 0) { - dbg(msg->udev, "%llu, device event still pending %llu (%d:%d)\n", msg->seqnum, - loop_msg->seqnum, major(loop_msg->devt), minor(loop_msg->devt)); - return 4; - } + /* compare devpath */ + common = MIN(loop_event->devpath_len, event->devpath_len); - /* check physical device event (special case of parent) */ - if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0) - if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) { - dbg(msg->udev, "%llu, physical device event still pending %llu (%s)\n", - msg->seqnum, loop_msg->seqnum, loop_msg->devpath); - return 5; - } - } + /* one devpath is contained in the other? */ + if (memcmp(loop_event->devpath, event->devpath, common) != 0) + continue; - /* check run queue for still running events */ - list_for_each_entry(loop_msg, &running_list, node) { - if (limit && childs_count++ > limit) { - dbg(msg->udev, "%llu, maximum number (%i) of childs reached\n", msg->seqnum, childs_count); - return 1; + /* identical device event found */ + if (loop_event->devpath_len == event->devpath_len) { + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check our old name */ - if (msg->devpath_old != NULL) - if (strcmp(loop_msg->devpath , msg->devpath_old) == 0) - return 2; - - /* check identical, parent, or child device event */ - if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) { - dbg(msg->udev, "%llu, device event still running %llu (%s)\n", - msg->seqnum, loop_msg->seqnum, loop_msg->devpath); - return 3; + /* parent device event found */ + if (event->devpath[common] == '/') { + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check for our major:minor number */ - if (msg->devt && loop_msg->devt == msg->devt && - strcmp(msg->subsystem, loop_msg->subsystem) == 0) { - dbg(msg->udev, "%llu, device event still running %llu (%d:%d)\n", msg->seqnum, - loop_msg->seqnum, major(loop_msg->devt), minor(loop_msg->devt)); - return 4; + /* child device event found */ + if (loop_event->devpath[common] == '/') { + event->delaying_seqnum = loop_event->seqnum; + return true; } - /* check physical device event (special case of parent) */ - if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0) - if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) { - dbg(msg->udev, "%llu, physical device event still running %llu (%s)\n", - msg->seqnum, loop_msg->seqnum, loop_msg->devpath); - return 5; - } + /* no matching device */ + continue; } - return 0; + + return false; } -/* serializes events for the identical and parent and child devices */ -static void msg_queue_manager(struct udev *udev) +static void events_start(struct udev *udev) { - struct udevd_uevent_msg *loop_msg; - struct udevd_uevent_msg *tmp_msg; - int running; + struct udev_list_node *loop; - if (list_empty(&exec_list)) - return; + udev_list_node_foreach(loop, &event_list) { + struct event *event = node_to_event(loop); - running = running_processes(); - dbg(udev, "%d processes runnning on system\n", running); - if (running < 0) - running = max_childs_running; - - list_for_each_entry_safe(loop_msg, tmp_msg, &exec_list, node) { - /* check running processes in our session and possibly throttle */ - if (running >= max_childs_running) { - running = running_processes_in_session(sid, max_childs_running+10); - dbg(udev, "at least %d processes running in session\n", running); - if (running >= max_childs_running) { - dbg(udev, "delay seq %llu, too many processes already running\n", loop_msg->seqnum); - return; - } - } + if (event->state != EVENT_QUEUED) + continue; - /* serialize and wait for parent or child events */ - if (devpath_busy(loop_msg, max_childs) != 0) { - dbg(udev, "delay seq %llu (%s)\n", loop_msg->seqnum, loop_msg->devpath); + /* do not start event if parent or child event is still running */ + if (is_devpath_busy(event)) { + dbg(udev, "delay seq %llu (%s)\n", event->seqnum, event->devpath); continue; } - /* move event to run list */ - list_move_tail(&loop_msg->node, &running_list); - udev_event_run(loop_msg); - running++; - dbg(udev, "moved seq %llu to running list\n", loop_msg->seqnum); + event_run(event); } } -static struct udevd_uevent_msg *get_msg_from_envbuf(struct udev *udev, const char *buf, int buf_size) +static void worker_returned(void) { - int bufpos; - int i; - struct udevd_uevent_msg *msg; - char *physdevdriver_key = NULL; - int maj = 0; - int min = 0; - - msg = malloc(sizeof(struct udevd_uevent_msg) + buf_size); - if (msg == NULL) - return NULL; - memset(msg, 0x00, sizeof(struct udevd_uevent_msg) + buf_size); - msg->udev = udev; - - /* copy environment buffer and reconstruct envp */ - memcpy(msg->envbuf, buf, buf_size); - bufpos = 0; - for (i = 0; (bufpos < buf_size) && (i < UEVENT_NUM_ENVP-2); i++) { - int keylen; - char *key; + while (1) { + struct worker_message msg; + ssize_t size; + struct udev_list_node *loop; - key = &msg->envbuf[bufpos]; - keylen = strlen(key); - msg->envp[i] = key; - bufpos += keylen + 1; - dbg(udev, "add '%s' to msg.envp[%i]\n", msg->envp[i], i); - - /* remember some keys for further processing */ - if (strncmp(key, "ACTION=", 7) == 0) - msg->action = &key[7]; - else if (strncmp(key, "DEVPATH=", 8) == 0) - msg->devpath = &key[8]; - else if (strncmp(key, "SUBSYSTEM=", 10) == 0) - msg->subsystem = &key[10]; - else if (strncmp(key, "DRIVER=", 7) == 0) - msg->driver = &key[7]; - else if (strncmp(key, "SEQNUM=", 7) == 0) - msg->seqnum = strtoull(&key[7], NULL, 10); - else if (strncmp(key, "DEVPATH_OLD=", 12) == 0) - msg->devpath_old = &key[12]; - else if (strncmp(key, "PHYSDEVPATH=", 12) == 0) - msg->physdevpath = &key[12]; - else if (strncmp(key, "PHYSDEVDRIVER=", 14) == 0) - physdevdriver_key = key; - else if (strncmp(key, "MAJOR=", 6) == 0) - maj = strtoull(&key[6], NULL, 10); - else if (strncmp(key, "MINOR=", 6) == 0) - min = strtoull(&key[6], NULL, 10); - else if (strncmp(key, "TIMEOUT=", 8) == 0) - msg->timeout = strtoull(&key[8], NULL, 10); - } - msg->devt = makedev(maj, min); - msg->envp[i++] = "UDEVD_EVENT=1"; + size = recv(pfd[FD_WORKER].fd, &msg, sizeof(struct worker_message), MSG_DONTWAIT); + if (size != sizeof(struct worker_message)) + break; - if (msg->driver == NULL && msg->physdevpath == NULL && physdevdriver_key != NULL) { - /* for older kernels DRIVER is empty for a bus device, export PHYSDEVDRIVER as DRIVER */ - msg->envp[i++] = &physdevdriver_key[7]; - msg->driver = &physdevdriver_key[14]; - } + /* lookup worker who sent the signal */ + udev_list_node_foreach(loop, &worker_list) { + struct worker *worker = node_to_worker(loop); - msg->envp[i] = NULL; + if (worker->pid != msg.pid) + continue; - if (msg->devpath == NULL || msg->action == NULL) { - info(udev, "DEVPATH or ACTION missing, ignore message\n"); - free(msg); - return NULL; + /* 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; + } } - return msg; } /* receive the udevd message from userspace */ @@ -687,46 +605,49 @@ 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); - sprintf(udev_log_env, "UDEV_LOG=%i", i); - putenv(udev_log_env); + 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; - msg_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); if (str != NULL) { - char *key = strdup(str); - char *val; - - val = strchr(str, '='); - if (val != NULL) { - val[0] = '\0'; - val = &val[1]; - if (val[0] == '\0') { - info(udev, "udevd message (ENV) received, unset '%s'\n", key); - unsetenv(str); + char *key; + + key = strdup(str); + if (key != NULL) { + char *val; + + val = strchr(key, '='); + if (val != NULL) { + val[0] = '\0'; + val = &val[1]; + if (val[0] == '\0') { + info(udev, "udevd message (ENV) received, unset '%s'\n", key); + udev_add_property(udev, key, NULL); + } else { + info(udev, "udevd message (ENV) received, set '%s=%s'\n", key, val); + udev_add_property(udev, key, val); + } } else { - info(udev, "udevd message (ENV) received, set '%s=%s'\n", key, val); - setenv(key, val, 1); + err(udev, "wrong key format '%s'\n", key); } - } else { - err(udev, "wrong key format '%s'\n", key); + free(key); } - free(key); + worker_kill(udev, 0); } i = udev_ctrl_get_set_max_childs(ctrl_msg); @@ -735,211 +656,168 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl) max_childs = i; } - i = udev_ctrl_get_set_max_childs_running(ctrl_msg); - if (i > 0) { - info(udev, "udevd message (SET_MAX_CHILDS_RUNNING) received, max_childs_running=%i\n", i); - max_childs_running = 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); } -/* receive the kernel user event message and do some sanity checks */ -static struct udevd_uevent_msg *get_netlink_msg(struct udev *udev) +/* read inotify messages */ +static int handle_inotify(struct udev *udev) { - struct udevd_uevent_msg *msg; - int bufpos; - ssize_t size; - static char buffer[UEVENT_BUFFER_SIZE+512]; - char *pos; - - size = recv(uevent_netlink_sock, &buffer, sizeof(buffer), 0); - if (size < 0) { - if (errno != EINTR) - err(udev, "unable to receive kernel netlink message: %s\n", strerror(errno)); - return NULL; - } + int nbytes, pos; + char *buf; + struct inotify_event *ev; - if ((size_t)size > sizeof(buffer)-1) - size = sizeof(buffer)-1; - buffer[size] = '\0'; - dbg(udev, "uevent_size=%zi\n", size); - - /* start of event payload */ - bufpos = strlen(buffer)+1; - msg = get_msg_from_envbuf(udev, &buffer[bufpos], size-bufpos); - if (msg == NULL) - return NULL; - - /* validate message */ - pos = strchr(buffer, '@'); - if (pos == NULL) { - err(udev, "invalid uevent '%s'\n", buffer); - free(msg); - return NULL; - } - pos[0] = '\0'; + if ((ioctl(pfd[FD_INOTIFY].fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0)) + return 0; - if (msg->action == NULL) { - info(udev, "no ACTION in payload found, skip event '%s'\n", buffer); - free(msg); - return NULL; + buf = malloc(nbytes); + if (buf == NULL) { + err(udev, "error getting buffer for inotify\n"); + return -1; } - if (strcmp(msg->action, buffer) != 0) { - err(udev, "ACTION in payload does not match uevent, skip event '%s'\n", buffer); - free(msg); - return NULL; - } + nbytes = read(pfd[FD_INOTIFY].fd, buf, nbytes); - return msg; -} + for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) { + struct udev_device *dev; -static void asmlinkage sig_handler(int signum) -{ - 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; - } + ev = (struct inotify_event *)(buf + pos); + if (ev->len) { + dbg(udev, "inotify event: %x for %s\n", ev->mask, ev->name); + reload_config = true; + continue; + } - /* write to pipe, which will wakeup select() in our mainloop */ - write(signal_pipe[WRITE_END], "", 1); -} + 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)); + 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_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"); + close(fd); + } + if (ev->mask & IN_IGNORED) + udev_watch_end(udev, dev); -static void udev_done(int pid, int exitstatus) -{ - /* find msg associated with pid and delete it */ - struct udevd_uevent_msg *msg; - - list_for_each_entry(msg, &running_list, node) { - if (msg->pid == pid) { - info(msg->udev, "seq %llu, pid [%d] exit with %i, %ld seconds old\n", msg->seqnum, msg->pid, - exitstatus, time(NULL) - msg->queue_time); - msg->exitstatus = exitstatus; - msg_queue_delete(msg); - - /* there may be events waiting with the same devpath */ - run_exec_q = 1; - return; + udev_device_unref(dev); } - } -} - -static void reap_sigchilds(void) -{ - pid_t pid; - int status; - 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); } + + free(buf); + return 0; } -static int init_uevent_netlink_sock(struct udev *udev) +static void handle_signal(struct udev *udev, int signo) { - struct sockaddr_nl snl; - const int buffersize = 16 * 1024 * 1024; - int retval; - - memset(&snl, 0x00, sizeof(struct sockaddr_nl)); - snl.nl_family = AF_NETLINK; - snl.nl_pid = getpid(); - snl.nl_groups = 1; - - uevent_netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT); - if (uevent_netlink_sock == -1) { - err(udev, "error getting socket: %s\n", strerror(errno)); - return -1; - } - - /* set receive buffersize */ - setsockopt(uevent_netlink_sock, SOL_SOCKET, SO_RCVBUFFORCE, &buffersize, sizeof(buffersize)); - - retval = bind(uevent_netlink_sock, (struct sockaddr *) &snl, sizeof(struct sockaddr_nl)); - if (retval < 0) { - err(udev, "bind failed: %s\n", strerror(errno)); - close(uevent_netlink_sock); - uevent_netlink_sock = -1; - return -1; + switch (signo) { + case SIGINT: + case SIGTERM: + udev_exit = true; + break; + case SIGCHLD: + while (1) { + 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; } - return 0; } -static void export_initial_seqnum(struct udev *udev) +static void startup_log(struct udev *udev) { - char filename[PATH_SIZE]; - int fd; - char seqnum[32]; - ssize_t len = 0; - - strlcpy(filename, udev_get_sys_path(udev), sizeof(filename)); - 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; - } - strlcpy(filename, udev_get_dev_path(udev), sizeof(filename)); - strlcat(filename, "/.udev/uevent_seqnum", sizeof(filename)); - create_path(udev, filename); - fd = open(filename, O_WRONLY|O_TRUNC|O_CREAT, 0644); - if (fd >= 0) { - write(fd, seqnum, len); - close(fd); + FILE *f; + char path[UTIL_PATH_SIZE]; + struct stat statbuf; + + f = fopen("/dev/kmsg", "w"); + if (f != NULL) + fprintf(f, "<6>udev: starting version " VERSION "\n"); + + util_strscpyl(path, sizeof(path), udev_get_sys_path(udev), "/class/mem/null", NULL); + if (lstat(path, &statbuf) == 0 && S_ISDIR(statbuf.st_mode)) { + const char *depr_str = + "udev: missing sysfs features; please update the kernel " + "or disable the kernel's CONFIG_SYSFS_DEPRECATED option; " + "udev may fail to work correctly"; + + if (f != NULL) + fprintf(f, "<3>%s\n", depr_str); + err(udev, "%s\n", depr_str); + sleep(15); } + + if (f != NULL) + fclose(f); } int main(int argc, char *argv[]) { struct udev *udev; - int retval; int fd; - struct sigaction act; - fd_set readfds; + sigset_t mask; const char *value; - int daemonize = 0; + int daemonize = false; + int resolve_names = 1; static const struct option options[] = { - { "daemon", 0, NULL, 'd' }, - { "debug-trace", 0, NULL, 't' }, - { "debug", 0, NULL, 'D' }, - { "help", 0, NULL, 'h' }, - { "version", 0, NULL, 'V' }, + { "daemon", no_argument, NULL, 'd' }, + { "debug-trace", no_argument, NULL, 't' }, + { "debug", no_argument, NULL, 'D' }, + { "help", no_argument, NULL, 'h' }, + { "version", no_argument, NULL, 'V' }, + { "resolve-names", required_argument, NULL, 'N' }, {} }; int rc = 1; - int maxfd; udev = udev_new(); if (udev == NULL) goto exit; - logging_init("udevd"); + udev_log_init("udevd"); udev_set_log_fn(udev, log_fn); - - selinux_init(udev); - dbg(udev, "version %s\n", VERSION); + info(udev, "version %s\n", VERSION); + udev_selinux_init(udev); while (1) { int option; @@ -950,18 +828,32 @@ int main(int argc, char *argv[]) switch (option) { case 'd': - daemonize = 1; + daemonize = true; break; case 't': - debug_trace = 1; + debug_trace = true; break; case 'D': - debug = 1; + debug = true; if (udev_get_log_priority(udev) < LOG_INFO) udev_set_log_priority(udev, LOG_INFO); break; + case 'N': + if (strcmp (optarg, "early") == 0) { + resolve_names = 1; + } else if (strcmp (optarg, "late") == 0) { + resolve_names = 0; + } else if (strcmp (optarg, "never") == 0) { + resolve_names = -1; + } else { + fprintf(stderr, "resolve-names must be early, late or never\n"); + err(udev, "resolve-names must be early, late or never\n"); + goto exit; + } + break; case 'h': - printf("Usage: udevd [--help] [--daemon] [--debug-trace] [--debug] [--version]\n"); + printf("Usage: udevd [--help] [--daemon] [--debug-trace] [--debug] " + "[--resolve-names=early|late|never] [--version]\n"); goto exit; case 'V': printf("%s\n", VERSION); @@ -983,8 +875,6 @@ int main(int argc, char *argv[]) fprintf(stderr, "cannot open /dev/null\n"); err(udev, "cannot open /dev/null\n"); } - if (fd > STDIN_FILENO) - dup2(fd, STDIN_FILENO); if (write(STDOUT_FILENO, 0, 0) < 0) dup2(fd, STDOUT_FILENO); if (write(STDERR_FILENO, 0, 0) < 0) @@ -998,54 +888,88 @@ int main(int argc, char *argv[]) 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"); rc = 1; goto exit; } + pfd[FD_CONTROL].fd = udev_ctrl_get_fd(udev_ctrl); - if (init_uevent_netlink_sock(udev) < 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; } - - retval = pipe(signal_pipe); - if (retval < 0) { - err(udev, "error getting pipes: %s\n", strerror(errno)); + 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; } - retval = fcntl(signal_pipe[READ_END], F_GETFL, 0); - if (retval < 0) { - err(udev, "error fcntl on read pipe: %s\n", strerror(errno)); - goto exit; + if (udev_get_rules_path(udev) != NULL) { + inotify_add_watch(pfd[FD_INOTIFY].fd, udev_get_rules_path(udev), + IN_CREATE | 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_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + inotify_add_watch(pfd[FD_INOTIFY].fd, SYSCONFDIR "/udev/rules.d", + IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); + + /* watch dynamic rules directory */ + util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/rules.d", NULL); + if (stat(filename, &statbuf) != 0) { + util_create_path(udev, filename); + udev_selinux_setfscreatecon(udev, filename, S_IFDIR|0755); + mkdir(filename, 0755); + udev_selinux_resetfscreatecon(udev); + } + inotify_add_watch(pfd[FD_INOTIFY].fd, filename, + IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); } - retval = fcntl(signal_pipe[READ_END], F_SETFL, retval | O_NONBLOCK); - if (retval < 0) { - err(udev, "error fcntl on read pipe: %s\n", strerror(errno)); + 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; } - retval = fcntl(signal_pipe[WRITE_END], F_GETFL, 0); - if (retval < 0) { - err(udev, "error fcntl on write pipe: %s\n", strerror(errno)); + /* 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; } - retval = fcntl(signal_pipe[WRITE_END], F_SETFL, retval | O_NONBLOCK); - if (retval < 0) { - err(udev, "error fcntl on write pipe: %s\n", strerror(errno)); + 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; } - /* parse the rules and keep them in memory */ - sysfs_init(); - udev_rules_init(udev, &rules, 1); - - 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 (daemonize) { pid_t pid; @@ -1053,21 +977,22 @@ 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: %s\n", strerror(errno)); + 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; } } + startup_log(udev); + /* redirect std{out,err} */ - if (!debug) { + if (!debug && !debug_trace) { + dup2(fd, STDIN_FILENO); dup2(fd, STDOUT_FILENO); dup2(fd, STDERR_FILENO); } @@ -1079,196 +1004,128 @@ int main(int argc, char *argv[]) chdir("/"); umask(022); - - /* become session leader */ - sid = setsid(); - dbg(udev, "our session is %d\n", sid); + setsid(); /* OOM_DISABLE == -17 */ fd = open("/proc/self/oom_adj", O_RDWR); - if (fd < 0) - err(udev, "error disabling OOM: %s\n", strerror(errno)); - else { + 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 *str = "<6>udevd version " VERSION " started\n"; + /* in trace mode run one event after the other */ + if (debug_trace) { + max_childs = 1; + } else { + int memsize = mem_size_mb(); - write(fd, str, strlen(str)); - close(fd); + if (memsize > 0) + max_childs = 128 + (memsize / 8); + else + max_childs = 128; } - /* 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 */ - inotify_fd = inotify_init(); - 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); - } else { - char filename[PATH_MAX]; - - 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 */ - strlcpy(filename, udev_get_dev_path(udev), sizeof(filename)); - strlcat(filename, "/.udev/rules.d", sizeof(filename)); - inotify_add_watch(inotify_fd, filename, - IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE); - } - } else if (errno == ENOSYS) - err(udev, "the kernel does not support inotify, udevd can't monitor rules file changes\n"); - else - err(udev, "inotify_init failed: %s\n", strerror(errno)); - - /* maximum limit of forked childs */ + /* possibly overwrite maximum limit of executed events */ value = getenv("UDEVD_MAX_CHILDS"); if (value) max_childs = strtoul(value, NULL, 10); - else { - int memsize = mem_size_mb(); - if (memsize > 0) - max_childs = 128 + (memsize / 4); - else - max_childs = UDEVD_MAX_CHILDS; - } info(udev, "initialize max_childs to %u\n", max_childs); - /* start to throttle forking if maximum number of _running_ childs is reached */ - value = getenv("UDEVD_MAX_CHILDS_RUNNING"); - if (value) - max_childs_running = strtoull(value, NULL, 10); - else { - int cpus = cpu_count(); - if (cpus > 0) - max_childs_running = 8 + (8 * cpus); - else - max_childs_running = UDEVD_MAX_CHILDS_RUNNING; - } - info(udev, "initialize max_childs_running to %u\n", max_childs_running); - - /* clear environment for forked event processes */ - clearenv(); - - /* export log_priority , as called programs may want to follow that setting */ - sprintf(udev_log_env, "UDEV_LOG=%i", udev_get_log_priority(udev)); - putenv(udev_log_env); - if (debug_trace) - putenv("DEBUG=1"); - - maxfd = udev_ctrl_get_fd(udev_ctrl); - maxfd = UDEV_MAX(maxfd, uevent_netlink_sock); - maxfd = UDEV_MAX(maxfd, signal_pipe[READ_END]); - maxfd = UDEV_MAX(maxfd, inotify_fd); + udev_list_init(&event_list); + udev_list_init(&worker_list); while (!udev_exit) { - struct udevd_uevent_msg *msg; int fdcount; + int timeout; - FD_ZERO(&readfds); - FD_SET(signal_pipe[READ_END], &readfds); - FD_SET(udev_ctrl_get_fd(udev_ctrl), &readfds); - FD_SET(uevent_netlink_sock, &readfds); - if (inotify_fd >= 0) - FD_SET(inotify_fd, &readfds); - - fdcount = select(maxfd+1, &readfds, NULL, NULL, NULL); - if (fdcount < 0) { - if (errno != EINTR) - err(udev, "error in select: %s\n", strerror(errno)); + /* set timeout to kill idle workers */ + if (udev_list_is_empty(&event_list) && childs > 2) + timeout = 3 * 1000; + else + timeout = -1; + /* wait for events */ + fdcount = poll(pfd, ARRAY_SIZE(pfd), timeout); + if (fdcount < 0) continue; - } - /* get control message */ - if (FD_ISSET(udev_ctrl_get_fd(udev_ctrl), &readfds)) - handle_ctrl_msg(udev_ctrl); + /* timeout - kill idle workers */ + if (fdcount == 0) + worker_kill(udev, 2); - /* get netlink message */ - if (FD_ISSET(uevent_netlink_sock, &readfds)) { - msg = get_netlink_msg(udev); - if (msg) - msg_queue_insert(msg); - } + /* event has finished */ + if (pfd[FD_WORKER].revents & POLLIN) + worker_returned(); - /* received a signal, clear our notification pipe */ - if (FD_ISSET(signal_pipe[READ_END], &readfds)) { - char buf[256]; + /* get kernel uevent */ + if (pfd[FD_NETLINK].revents & POLLIN) { + struct udev_device *dev; - read(signal_pipe[READ_END], &buf, sizeof(buf)); + dev = udev_monitor_receive_device(monitor); + if (dev != NULL) + event_queue_insert(dev); + else + udev_device_unref(dev); } - /* rules directory inotify watch */ - if ((inotify_fd >= 0) && FD_ISSET(inotify_fd, &readfds)) { - int nbytes; + /* start new events */ + if (!udev_list_is_empty(&event_list) && !stop_exec_queue) + events_start(udev); - /* discard all possible events, we can just reload the config */ - if ((ioctl(inotify_fd, FIONREAD, &nbytes) == 0) && nbytes > 0) { - char *buf; + /* get signal */ + if (pfd[FD_SIGNAL].revents & POLLIN) { + struct signalfd_siginfo fdsi; + ssize_t size; - reload_config = 1; - buf = malloc(nbytes); - if (buf == NULL) { - err(udev, "error getting buffer for inotify, disable watching\n"); - close(inotify_fd); - inotify_fd = -1; - } - read(inotify_fd, buf, nbytes); - free(buf); - } + size = read(pfd[FD_SIGNAL].fd, &fdsi, sizeof(struct signalfd_siginfo)); + if (size == sizeof(struct signalfd_siginfo)) + handle_signal(udev, fdsi.ssi_signo); } + /* device node and rules directory inotify watch */ + if (pfd[FD_INOTIFY].revents & POLLIN) + handle_inotify(udev); + + /* + * 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) { - reload_config = 0; - udev_rules_cleanup(&rules); - udev_rules_init(udev, &rules, 1); - } + struct udev_rules *rules_new; - /* forked child has returned */ - if (sigchilds_waiting) { - sigchilds_waiting = 0; - reap_sigchilds(); - } - - if (run_exec_q) { - run_exec_q = 0; - if (!stop_exec_q) - msg_queue_manager(udev); + worker_kill(udev, 0); + rules_new = udev_rules_new(udev, resolve_names); + if (rules_new != NULL) { + udev_rules_unref(rules); + rules = rules_new; + } + reload_config = 0; } } - rc = 0; + udev_queue_export_cleanup(udev_queue_export); + rc = 0; exit: - udev_rules_cleanup(&rules); - sysfs_cleanup(); - selinux_exit(udev); - - if (signal_pipe[READ_END] >= 0) - close(signal_pipe[READ_END]); - if (signal_pipe[WRITE_END] >= 0) - close(signal_pipe[WRITE_END]); - + udev_queue_export_unref(udev_queue_export); + udev_rules_unref(rules); udev_ctrl_unref(udev_ctrl); - if (inotify_fd >= 0) - close(inotify_fd); - if (uevent_netlink_sock >= 0) - close(uevent_netlink_sock); - - logging_close(); + 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); + udev_log_close(); return rc; }