chiark / gitweb /
remove precompiled rules option
[elogind.git] / udevd.c
diff --git a/udevd.c b/udevd.c
index f1e28d9ca3de7ba9e72d87ab5054159a7ebfb840..154c37058958f39c94711b73b365fce8f34edca9 100644 (file)
--- a/udevd.c
+++ b/udevd.c
@@ -1,7 +1,7 @@
 /*
- * udevd.c - hotplug event serializer
+ * udevd.c - event listener and serializer
  *
- * Copyright (C) 2004 Kay Sievers <kay.sievers@vrfy.org>
+ * Copyright (C) 2004-2005 Kay Sievers <kay.sievers@vrfy.org>
  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
  *
  *
  */
 
 #include <stddef.h>
-#include <sys/wait.h>
 #include <signal.h>
 #include <unistd.h>
 #include <errno.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
+#include <ctype.h>
+#include <dirent.h>
+#include <fcntl.h>
+#include <syslog.h>
+#include <sys/select.h>
+#include <sys/wait.h>
 #include <sys/time.h>
 #include <sys/types.h>
 #include <sys/socket.h>
 #include <sys/un.h>
-#include <fcntl.h>
 #include <sys/sysinfo.h>
 #include <sys/stat.h>
+#include <sys/ioctl.h>
+#include <linux/types.h>
+#include <linux/netlink.h>
 
 #include "list.h"
+#include "udev_libc_wrapper.h"
 #include "udev.h"
 #include "udev_version.h"
+#include "udev_rules.h"
 #include "udev_utils.h"
 #include "udevd.h"
 #include "logging.h"
 
 /* global variables*/
-static int udevsendsock;
+struct udev_rules rules;
+static int udevd_sock;
+static int uevent_netlink_sock;
+static int inotify_fd;
+static pid_t sid;
 
-static int pipefds[2];
-static long startup_time;
-static unsigned long long expected_seqnum = 0;
+static int signal_pipe[2] = {-1, -1};
 static volatile int sigchilds_waiting;
 static volatile int run_msg_q;
-static volatile int sig_flag;
+static volatile int udev_exit;
+static volatile int reload_config;
+static int init_phase = 1;
 static int run_exec_q;
+static int stop_exec_q;
+static int event_timeout;
+static int max_childs;
+static int max_childs_running;
+static unsigned long long expected_seqnum;
+static char udev_log[32];
 
 static LIST_HEAD(msg_list);
 static LIST_HEAD(exec_list);
 static LIST_HEAD(running_list);
 
-static void exec_queue_manager(void);
-static void msg_queue_manager(void);
-static void user_sighandler(void);
-static void reap_sigchilds(void);
-char *udev_bin;
 
-#ifdef LOG
-void log_message (int level, const char *format, ...)
+#ifdef USE_LOG
+void log_message(int priority, const char *format, ...)
 {
-       va_list args;
+       va_list args;
+
+       if (priority > udev_log_priority)
+               return;
 
        va_start(args, format);
-       vsyslog(level, format, args);
+       vsyslog(priority, format, args);
        va_end(args);
 }
 #endif
 
-#define msg_dump(msg) \
-       dbg("msg_dump: sequence %llu, '%s', '%s', '%s'", \
-       msg->seqnum, msg->action, msg->devpath, msg->subsystem);
-
 static void msg_dump_queue(void)
 {
 #ifdef DEBUG
-       struct hotplug_msg *msg;
+       struct uevent_msg *msg;
 
-       list_for_each_entry(msg, &msg_list, list)
+       list_for_each_entry(msg, &msg_list, node)
                dbg("sequence %llu in queue", msg->seqnum);
 #endif
 }
 
-static void run_queue_delete(struct hotplug_msg *msg)
+static void msg_queue_delete(struct uevent_msg *msg)
 {
-       list_del(&msg->list);
+       list_del(&msg->node);
        free(msg);
 }
 
 /* orders the message in the queue by sequence number */
-static void msg_queue_insert(struct hotplug_msg *msg)
+static void msg_queue_insert(struct uevent_msg *msg)
 {
-       struct hotplug_msg *loop_msg;
+       struct uevent_msg *loop_msg;
        struct sysinfo info;
 
        if (msg->seqnum == 0) {
                dbg("no SEQNUM, move straight to the exec queue");
-               list_add(&msg->list, &exec_list);
+               list_add_tail(&msg->node, &exec_list);
                run_exec_q = 1;
                return;
        }
 
-       /* sort message by sequence number into list */
-       list_for_each_entry_reverse(loop_msg, &msg_list, list)
-               if (loop_msg->seqnum < msg->seqnum)
-                       break;
-
        /* store timestamp of queuing */
        sysinfo(&info);
        msg->queue_time = info.uptime;
 
-       list_add(&msg->list, &loop_msg->list);
-       dbg("queued message seq %llu", msg->seqnum);
+       /* with the first event we provide a phase of shorter timeout */
+       if (init_phase) {
+               static long init_time;
+
+               if (init_time == 0)
+                       init_time = info.uptime;
+               if (info.uptime - init_time >= UDEVD_INIT_TIME)
+                       init_phase = 0;
+       }
+
+       /* sort message by sequence number into list */
+       list_for_each_entry_reverse(loop_msg, &msg_list, node) {
+               if (loop_msg->seqnum < msg->seqnum)
+                       break;
+
+               if (loop_msg->seqnum == msg->seqnum) {
+                       dbg("ignoring duplicate message seq %llu", msg->seqnum);
+                       free(msg);
+                       return;
+               }
+       }
+       list_add(&msg->node, &loop_msg->node);
+       info("seq %llu queued, '%s' '%s'", msg->seqnum, msg->action, msg->devpath);
 
        /* run msg queue manager */
        run_msg_q = 1;
 
-       return ;
+       return;
+}
+
+static void asmlinkage udev_event_sig_handler(int signum)
+{
+       if (signum == SIGALRM)
+               exit(1);
+}
+
+static int udev_event_process(struct uevent_msg *msg)
+{
+       struct sigaction act;
+       struct udevice udev;
+       struct name_entry *name_loop;
+       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);
+
+       /* trigger timeout to prevent hanging processes */
+       alarm(UDEV_ALARM_TIMEOUT);
+
+       /* reconstruct env from message */
+       for (i = 0; msg->envp[i]; i++)
+               putenv(msg->envp[i]);
+
+       udev_init_device(&udev, msg->devpath, msg->subsystem, msg->action);
+       retval = udev_process_event(&rules, &udev);
+
+       /* run programs collected by RUN-key*/
+       if (!retval) {
+               list_for_each_entry(name_loop, &udev.run_list, node) {
+                       if (strncmp(name_loop->name, "socket:", strlen("socket:")) == 0)
+                               pass_env_to_socket(&name_loop->name[strlen("socket:")], msg->devpath, msg->action);
+                       else
+                               run_program(name_loop->name, udev.subsystem, NULL, 0, NULL, (udev_log_priority >= LOG_INFO));
+               }
+       }
+
+       udev_cleanup_device(&udev);
+
+       return 0;
 }
 
-/* forks event and removes event from run queue when finished */
-static void udev_run(struct hotplug_msg *msg)
+/* runs event and removes event from run queue when finished */
+static void udev_event_run(struct uevent_msg *msg)
 {
-       char *const argv[] = { "udev", msg->subsystem, NULL };
        pid_t pid;
+       struct sysinfo info;
 
        pid = fork();
        switch (pid) {
        case 0:
                /* child */
-               close(udevsendsock);
+               close(uevent_netlink_sock);
+               close(udevd_sock);
+               if (inotify_fd > 0)
+                       close(inotify_fd);
+               close(signal_pipe[READ_END]);
+               close(signal_pipe[WRITE_END]);
                logging_close();
-               execve(udev_bin, argv, msg->envp);
-               dbg("exec of child failed");
-               _exit(1);
-               break;
+
+               logging_init("udevd-event");
+               setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
+               udev_event_process(msg);
+               info("seq %llu finished", msg->seqnum);
+
+               logging_close();
+               exit(0);
        case -1:
-               dbg("fork of child failed");
-               run_queue_delete(msg);
-               /* note: we never managed to run, so we had no impact on 
-                * running_with_devpath(), so don't bother setting run_exec_q
-                */
+               err("fork of child failed");
+               msg_queue_delete(msg);
                break;
        default:
                /* get SIGCHLD in main loop */
-               dbg("==> exec seq %llu [%d] working at '%s'", msg->seqnum, pid, msg->devpath);
+               sysinfo(&info);
+               info("seq %llu forked, pid [%d], '%s' '%s', %ld seconds old",
+                    msg->seqnum, pid,  msg->action, msg->subsystem, info.uptime - msg->queue_time);
                msg->pid = pid;
        }
 }
 
+static int running_processes(void)
+{
+       int f;
+       static char buf[4096];
+       int len;
+       int running;
+       const char *pos;
+
+       f = open("/proc/stat", O_RDONLY);
+       if (f == -1)
+               return -1;
+
+       len = read(f, buf, sizeof(buf)-1);
+       close(f);
+
+       if (len <= 0)
+               return -1;
+       else
+               buf[len] = '\0';
+
+       pos = strstr(buf, "procs_running ");
+       if (pos == NULL)
+               return -1;
+
+       if (sscanf(pos, "procs_running %u", &running) != 1)
+               return -1;
+
+       return running;
+}
+
+/* return the number of process es in our session, count only until limit */
+static int running_processes_in_session(pid_t session, int limit)
+{
+       DIR *dir;
+       struct dirent *dent;
+       int running = 0;
+
+       dir = opendir("/proc");
+       if (!dir)
+               return -1;
+
+       /* 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]))
+                       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;
+
+               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)
+                       break;
+       }
+       closedir(dir);
+
+       return running;
+}
+
 static int compare_devpath(const char *running, const char *waiting)
 {
        int i;
 
-       for (i = 0; i < DEVPATH_SIZE; i++) {
+       for (i = 0; i < PATH_SIZE; i++) {
                /* identical device event found */
                if (running[i] == '\0' && waiting[i] == '\0')
                        return 1;
@@ -182,54 +359,86 @@ static int compare_devpath(const char *running, const char *waiting)
 }
 
 /* returns still running task for the same device, its parent or its physical device */
-static struct hotplug_msg *running_with_devpath(struct hotplug_msg *msg)
+static int running_with_devpath(struct uevent_msg *msg, int limit)
 {
-       struct hotplug_msg *loop_msg;
+       struct uevent_msg *loop_msg;
+       int childs_count = 0;
 
        if (msg->devpath == NULL)
-               return NULL;
+               return 0;
+
+       /* skip any events with a timeout set */
+       if (msg->timeout != 0)
+               return 0;
 
-       list_for_each_entry(loop_msg, &running_list, list) {
+       list_for_each_entry(loop_msg, &running_list, node) {
+               if (limit && childs_count++ > limit) {
+                       dbg("%llu, maximum number (%i) of child reached", msg->seqnum, childs_count);
+                       return 1;
+               }
                if (loop_msg->devpath == NULL)
                        continue;
 
                /* return running parent/child device event */
-               if (compare_devpath(loop_msg->devpath, msg->devpath) != 0)
-                       return loop_msg;
+               if (compare_devpath(loop_msg->devpath, msg->devpath) != 0) {
+                       dbg("%llu, child device event still running %llu (%s)",
+                           msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
+                       return 2;
+               }
 
                /* return running physical device event */
                if (msg->physdevpath && msg->action && strcmp(msg->action, "add") == 0)
-                       if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0)
-                               return loop_msg;
+                       if (compare_devpath(loop_msg->devpath, msg->physdevpath) != 0) {
+                               dbg("%llu, physical device event still running %llu (%s)",
+                                   msg->seqnum, loop_msg->seqnum, loop_msg->devpath);
+                               return 3;
+                       }
        }
 
-       return NULL;
+       return 0;
 }
 
 /* exec queue management routine executes the events and serializes events in the same sequence */
 static void exec_queue_manager(void)
 {
-       struct hotplug_msg *loop_msg;
-       struct hotplug_msg *tmp_msg;
-       struct hotplug_msg *msg;
+       struct uevent_msg *loop_msg;
+       struct uevent_msg *tmp_msg;
+       int running;
 
-       list_for_each_entry_safe(loop_msg, tmp_msg, &exec_list, list) {
-               msg = running_with_devpath(loop_msg);
-               if (!msg) {
+       if (list_empty(&exec_list))
+               return;
+
+       running = running_processes();
+       dbg("%d processes runnning on system", 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("at least %d processes running in session", running);
+                       if (running >= max_childs_running) {
+                               dbg("delay seq %llu, cause too many processes already running",
+                                   loop_msg->seqnum);
+                               return;
+                       }
+               }
+
+               if (running_with_devpath(loop_msg, max_childs) == 0) {
                        /* move event to run list */
-                       list_move_tail(&loop_msg->list, &running_list);
-                       udev_run(loop_msg);
+                       list_move_tail(&loop_msg->node, &running_list);
+                       udev_event_run(loop_msg);
+                       running++;
                        dbg("moved seq %llu to running list", loop_msg->seqnum);
-               } else {
-                       dbg("delay seq %llu (%s), cause seq %llu (%s) is still running",
-                           loop_msg->seqnum, loop_msg->devpath, msg->seqnum, msg->devpath);
-               }
+               } else
+                       dbg("delay seq %llu (%s)", loop_msg->seqnum, loop_msg->devpath);
        }
 }
 
-static void msg_move_exec(struct hotplug_msg *msg)
+static void msg_move_exec(struct uevent_msg *msg)
 {
-       list_move_tail(&msg->list, &exec_list);
+       list_move_tail(&msg->node, &exec_list);
        run_exec_q = 1;
        expected_seqnum = msg->seqnum+1;
        dbg("moved seq %llu to exec, next expected is %llu",
@@ -239,58 +448,105 @@ static void msg_move_exec(struct hotplug_msg *msg)
 /* msg queue management routine handles the timeouts and dispatches the events */
 static void msg_queue_manager(void)
 {
-       struct hotplug_msg *loop_msg;
-       struct hotplug_msg *tmp_msg;
+       struct uevent_msg *loop_msg;
+       struct uevent_msg *tmp_msg;
        struct sysinfo info;
        long msg_age = 0;
-       static int timeout = EVENT_INIT_TIMEOUT_SEC;
-       static int init = 1;
+       int timeout = event_timeout;
 
        dbg("msg queue manager, next expected is %llu", expected_seqnum);
 recheck:
-       list_for_each_entry_safe(loop_msg, tmp_msg, &msg_list, list) {
+       sysinfo(&info);
+       list_for_each_entry_safe(loop_msg, tmp_msg, &msg_list, node) {
                /* move event with expected sequence to the exec list */
                if (loop_msg->seqnum == expected_seqnum) {
                        msg_move_exec(loop_msg);
                        continue;
                }
 
-               /* see if we are in the initialization phase and wait for the very first events */
-               if (init && (info.uptime - startup_time >= INIT_TIME_SEC)) {
-                       init = 0;
-                       timeout = EVENT_TIMEOUT_SEC;
-                       dbg("initialization phase passed, set timeout to %i seconds", EVENT_TIMEOUT_SEC);
+               /* limit timeout during initialization phase */
+               if (init_phase) {
+                       if (timeout > UDEVD_INIT_EVENT_TIMEOUT)
+                               timeout = UDEVD_INIT_EVENT_TIMEOUT;
+                       dbg("initialization phase, timeout %i seconds", timeout);
                }
 
                /* move event with expired timeout to the exec list */
-               sysinfo(&info);
                msg_age = info.uptime - loop_msg->queue_time;
                dbg("seq %llu is %li seconds old", loop_msg->seqnum, msg_age);
                if (msg_age >= timeout) {
                        msg_move_exec(loop_msg);
                        goto recheck;
-               } else {
+               } else
                        break;
-               }
        }
 
        msg_dump_queue();
 
        /* set timeout for remaining queued events */
-       if (list_empty(&msg_list) == 0) {
+       if (!list_empty(&msg_list)) {
                struct itimerval itv = {{0, 0}, {timeout - msg_age, 0}};
-               dbg("next event expires in %li seconds", timeout - msg_age);
+               info("next event expires in %li seconds", timeout - msg_age);
                setitimer(ITIMER_REAL, &itv, NULL);
        }
 }
 
-/* receive the udevsend message and do some sanity checks */
-static struct hotplug_msg *get_udevsend_msg(void)
+static struct uevent_msg *get_msg_from_envbuf(const char *buf, int buf_size)
 {
-       static struct udevsend_msg usend_msg;
-       struct hotplug_msg *msg;
        int bufpos;
        int i;
+       struct uevent_msg *msg;
+       int major = 0;
+       int minor = 0;
+
+       msg = malloc(sizeof(struct uevent_msg) + buf_size);
+       if (msg == NULL)
+               return NULL;
+       memset(msg, 0x00, sizeof(struct uevent_msg) + buf_size);
+
+       /* 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;
+
+               key = &msg->envbuf[bufpos];
+               keylen = strlen(key);
+               msg->envp[i] = key;
+               bufpos += keylen + 1;
+               dbg("add '%s' to msg.envp[%i]", 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, "SEQNUM=", 7) == 0)
+                       msg->seqnum = strtoull(&key[7], NULL, 10);
+               else if (strncmp(key, "PHYSDEVPATH=", 12) == 0)
+                       msg->physdevpath = &key[12];
+               else if (strncmp(key, "MAJOR=", 6) == 0)
+                       major = strtoull(&key[6], NULL, 10);
+               else if (strncmp(key, "MINOR=", 6) == 0)
+                       minor = strtoull(&key[6], NULL, 10);
+               else if (strncmp(key, "TIMEOUT=", 8) == 0)
+                       msg->timeout = strtoull(&key[8], NULL, 10);
+       }
+       msg->devt = makedev(major, minor);
+       msg->envp[i++] = "UDEVD_EVENT=1";
+       msg->envp[i] = NULL;
+
+       return msg;
+}
+
+/* receive the udevd message from userspace */
+static struct uevent_msg *get_udevd_msg(void)
+{
+       static struct udevd_msg usend_msg;
+       struct uevent_msg *msg;
        ssize_t size;
        struct msghdr smsg;
        struct cmsghdr *cmsg;
@@ -298,10 +554,11 @@ static struct hotplug_msg *get_udevsend_msg(void)
        struct ucred *cred;
        char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
        int envbuf_size;
+       int *intval;
 
-       memset(&usend_msg, 0x00, sizeof(struct udevsend_msg));
+       memset(&usend_msg, 0x00, sizeof(struct udevd_msg));
        iov.iov_base = &usend_msg;
-       iov.iov_len = sizeof(struct udevsend_msg);
+       iov.iov_len = sizeof(struct udevd_msg);
 
        memset(&smsg, 0x00, sizeof(struct msghdr));
        smsg.msg_iov = &iov;
@@ -309,114 +566,162 @@ static struct hotplug_msg *get_udevsend_msg(void)
        smsg.msg_control = cred_msg;
        smsg.msg_controllen = sizeof(cred_msg);
 
-       size = recvmsg(udevsendsock, &smsg, 0);
+       size = recvmsg(udevd_sock, &smsg, 0);
        if (size <  0) {
                if (errno != EINTR)
-                       dbg("unable to receive udevsend message");
+                       err("unable to receive udevd message");
                return NULL;
        }
        cmsg = CMSG_FIRSTHDR(&smsg);
        cred = (struct ucred *) CMSG_DATA(cmsg);
 
        if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
-               dbg("no sender credentials received, message ignored");
+               err("no sender credentials received, message ignored");
                return NULL;
        }
 
        if (cred->uid != 0) {
-               dbg("sender uid=%i, message ignored", cred->uid);
+               err("sender uid=%i, message ignored", cred->uid);
                return NULL;
        }
 
        if (strncmp(usend_msg.magic, UDEV_MAGIC, sizeof(UDEV_MAGIC)) != 0 ) {
-               dbg("message magic '%s' doesn't match, ignore it", usend_msg.magic);
+               err("message magic '%s' doesn't match, ignore it", usend_msg.magic);
                return NULL;
        }
 
-       envbuf_size = size - offsetof(struct udevsend_msg, envbuf);
-       dbg("envbuf_size=%i", envbuf_size);
-       msg = malloc(sizeof(struct hotplug_msg) + envbuf_size);
-       if (msg == NULL)
-               return NULL;
-
-       memset(msg, 0x00, sizeof(struct hotplug_msg) + envbuf_size);
+       switch (usend_msg.type) {
+       case UDEVD_UEVENT_UDEVSEND:
+       case UDEVD_UEVENT_INITSEND:
+               info("udevd event message received");
+               envbuf_size = size - offsetof(struct udevd_msg, envbuf);
+               dbg("envbuf_size=%i", envbuf_size);
+               msg = get_msg_from_envbuf(usend_msg.envbuf, envbuf_size);
+               if (msg == NULL)
+                       return NULL;
+               msg->type = usend_msg.type;
+               return msg;
+       case UDEVD_STOP_EXEC_QUEUE:
+               info("udevd message (STOP_EXEC_QUEUE) received");
+               stop_exec_q = 1;
+               break;
+       case UDEVD_START_EXEC_QUEUE:
+               info("udevd message (START_EXEC_QUEUE) received");
+               stop_exec_q = 0;
+               exec_queue_manager();
+               break;
+       case UDEVD_SET_LOG_LEVEL:
+               intval = (int *) usend_msg.envbuf;
+               info("udevd message (SET_LOG_PRIORITY) received, udev_log_priority=%i", *intval);
+               udev_log_priority = *intval;
+               sprintf(udev_log, "UDEV_LOG=%i", udev_log_priority);
+               putenv(udev_log);
+               break;
+       case UDEVD_SET_MAX_CHILDS:
+               intval = (int *) usend_msg.envbuf;
+               info("udevd message (UDEVD_SET_MAX_CHILDS) received, max_childs=%i", *intval);
+               max_childs = *intval;
+               break;
+       case UDEVD_RELOAD_RULES:
+               info("udevd message (RELOAD_RULES) received");
+               reload_config = 1;
+               break;
+       default:
+               dbg("unknown message type");
+       }
+       return NULL;
+}
 
-       /* copy environment buffer and reconstruct envp */
-       memcpy(msg->envbuf, usend_msg.envbuf, envbuf_size);
-       bufpos = 0;
-       for (i = 0; (bufpos < envbuf_size) && (i < HOTPLUG_NUM_ENVP-2); i++) {
-               int keylen;
-               char *key;
+/* receive the kernel user event message and do some sanity checks */
+static struct uevent_msg *get_netlink_msg(void)
+{
+       struct uevent_msg *msg;
+       int bufpos;
+       ssize_t size;
+       static char buffer[UEVENT_BUFFER_SIZE + 512];
+       char *pos;
 
-               key = &msg->envbuf[bufpos];
-               keylen = strlen(key);
-               msg->envp[i] = key;
-               bufpos += keylen + 1;
-               dbg("add '%s' to msg.envp[%i]", msg->envp[i], i);
+       size = recv(uevent_netlink_sock, &buffer, sizeof(buffer), 0);
+       if (size <  0) {
+               if (errno != EINTR)
+                       err("unable to receive udevd message");
+               return NULL;
+       }
 
-               /* remember some keys for further processing */
-               if (strncmp(key, "ACTION=", 7) == 0)
-                       msg->action = &key[7];
+       if ((size_t)size > sizeof(buffer)-1)
+               size = sizeof(buffer)-1;
+       buffer[size] = '\0';
+       dbg("uevent_size=%zi", size);
 
-               if (strncmp(key, "DEVPATH=", 8) == 0)
-                       msg->devpath = &key[8];
+       /* start of event payload */
+       bufpos = strlen(buffer)+1;
+       msg = get_msg_from_envbuf(&buffer[bufpos], size-bufpos);
+       if (msg == NULL)
+               return NULL;
+       msg->type = UDEVD_UEVENT_NETLINK;
 
-               if (strncmp(key, "SUBSYSTEM=", 10) == 0)
-                       msg->subsystem = &key[10];
+       /* validate message */
+       pos = strchr(buffer, '@');
+       if (pos == NULL) {
+               err("invalid uevent '%s'", buffer);
+               free(msg);
+               return NULL;
+       }
+       pos[0] = '\0';
 
-               if (strncmp(key, "SEQNUM=", 7) == 0)
-                       msg->seqnum = strtoull(&key[7], NULL, 10);
+       if (msg->action == NULL) {
+               info("no ACTION in payload found, skip event '%s'", buffer);
+               free(msg);
+               return NULL;
+       }
 
-               if (strncmp(key, "PHYSDEVPATH=", 12) == 0)
-                       msg->physdevpath = &key[12];
+       if (strcmp(msg->action, buffer) != 0) {
+               err("ACTION in payload does not match uevent, skip event '%s'", buffer);
+               free(msg);
+               return NULL;
        }
-       msg->envp[i++] = "UDEVD_EVENT=1";
-       msg->envp[i] = NULL;
 
        return msg;
 }
 
 static void asmlinkage sig_handler(int signum)
 {
-       int rc;
-
        switch (signum) {
                case SIGINT:
                case SIGTERM:
-                       exit(20 + signum);
+                       udev_exit = 1;
                        break;
                case SIGALRM:
                        /* set flag, then write to pipe if needed */
                        run_msg_q = 1;
-                       goto do_write;
                        break;
                case SIGCHLD:
                        /* set flag, then write to pipe if needed */
                        sigchilds_waiting = 1;
-                       goto do_write;
+                       break;
+               case SIGHUP:
+                       reload_config = 1;
                        break;
        }
 
-do_write:
-       /* if pipe is empty, write to pipe to force select to return
-        * immediately when it gets called
-        */
-       if (!sig_flag) {
-               rc = write(pipefds[1],&signum,sizeof(signum));
-               if (rc >= 0)
-                       sig_flag = 1;
-       }
+       /* write to pipe, which will wakeup select() in our mainloop */
+       write(signal_pipe[WRITE_END], "", 1);
 }
 
 static void udev_done(int pid)
 {
        /* find msg associated with pid and delete it */
-       struct hotplug_msg *msg;
+       struct uevent_msg *msg;
+       struct sysinfo info;
 
-       list_for_each_entry(msg, &running_list, list) {
+       list_for_each_entry(msg, &running_list, node) {
                if (msg->pid == pid) {
-                       dbg("<== exec seq %llu came back", msg->seqnum);
-                       run_queue_delete(msg);
+                       sysinfo(&info);
+                       if (msg->queue_time)
+                               info("seq %llu, pid [%d] exit, %ld seconds old", msg->seqnum, msg->pid, info.uptime - msg->queue_time);
+                       else
+                               info("seq 0, pid [%d] exit", msg->pid);
+                       msg_queue_delete(msg);
 
                        /* we want to run the exec queue manager since there may
                         * be events waiting with the devpath of the one that
@@ -430,182 +735,340 @@ static void udev_done(int pid)
 
 static void reap_sigchilds(void)
 {
-       while(1) {
-               int pid = waitpid(-1, NULL, WNOHANG);
-               if ((pid == -1) || (pid == 0))
+       pid_t pid;
+
+       while (1) {
+               pid = waitpid(-1, NULL, WNOHANG);
+               if (pid <= 0)
                        break;
                udev_done(pid);
        }
 }
 
-/* just read everything from the pipe and clear the flag,
- * the flags was set in the signal handler
- */
-static void user_sighandler(void)
+static int init_udevd_socket(void)
 {
-       int sig;
-       while(1) {
-               int rc = read(pipefds[0], &sig, sizeof(sig));
-               if (rc < 0)
-                       break;
+       struct sockaddr_un saddr;
+       const int buffersize = 16 * 1024 * 1024;
+       socklen_t addrlen;
+       const int feature_on = 1;
+       int retval;
 
-               sig_flag = 0;
+       memset(&saddr, 0x00, sizeof(saddr));
+       saddr.sun_family = AF_LOCAL;
+       /* use abstract namespace for socket path */
+       strcpy(&saddr.sun_path[1], UDEVD_SOCK_PATH);
+       addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(saddr.sun_path+1) + 1;
+
+       udevd_sock = socket(AF_LOCAL, SOCK_DGRAM, 0);
+       if (udevd_sock == -1) {
+               err("error getting socket, %s", strerror(errno));
+               return -1;
        }
+
+       /* set receive buffersize */
+       setsockopt(udevd_sock, SOL_SOCKET, SO_RCVBUFFORCE, &buffersize, sizeof(buffersize));
+
+       /* the bind takes care of ensuring only one copy running */
+       retval = bind(udevd_sock, (struct sockaddr *) &saddr, addrlen);
+       if (retval < 0) {
+               err("bind failed, %s", strerror(errno));
+               return -1;
+       }
+
+       /* enable receiving of the sender credentials */
+       setsockopt(udevd_sock, SOL_SOCKET, SO_PASSCRED, &feature_on, sizeof(feature_on));
+
+       return 0;
+}
+
+static int init_uevent_netlink_sock(void)
+{
+       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 = 0xffffffff;
+
+       uevent_netlink_sock = socket(PF_NETLINK, SOCK_DGRAM, NETLINK_KOBJECT_UEVENT);
+       if (uevent_netlink_sock == -1) {
+               err("error getting socket, %s", 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("bind failed, %s", strerror(errno));
+               close(uevent_netlink_sock);
+               uevent_netlink_sock = -1;
+               return -1;
+       }
+       return 0;
 }
 
 int main(int argc, char *argv[], char *envp[])
 {
-       struct sysinfo info;
-       int maxsockplus;
-       struct sockaddr_un saddr;
-       socklen_t addrlen;
-       int retval, fd;
-       const int feature_on = 1;
+       int retval;
+       int fd;
        struct sigaction act;
        fd_set readfds;
+       const char *value;
+       int uevent_netlink_active = 0;
+       int daemonize = 0;
+       int i;
+       int rc = 0;
+
+       /* redirect std fd's, if the kernel forks us, we don't have them at all */
+       fd = open("/dev/null", O_RDWR);
+       if (fd >= 0) {
+               if (fd != STDIN_FILENO)
+                       dup2(fd, STDIN_FILENO);
+               if (fd != STDOUT_FILENO)
+                       dup2(fd, STDOUT_FILENO);
+               if (fd != STDERR_FILENO)
+                       dup2(fd, STDERR_FILENO);
+               if (fd > STDERR_FILENO)
+                       close(fd);
+       }
 
        logging_init("udevd");
+       if (fd < 0)
+               err("fatal, could not open /dev/null");
+
+       udev_init_config();
        dbg("version %s", UDEV_VERSION);
 
        if (getuid() != 0) {
-               dbg("need to be root, exit");
+               err("need to be root, exit");
+               goto exit;
+       }
+
+       /* parse commandline options */
+       for (i = 1 ; i < argc; i++) {
+               char *arg = argv[i];
+               if (strcmp(arg, "--daemon") == 0 || strcmp(arg, "-d") == 0) {
+                       info("will daemonize");
+                       daemonize = 1;
+               }
+               if (strcmp(arg, "--stop-exec-queue") == 0) {
+                       info("will not execute events until START_EXEC_QUEUE is received");
+                       stop_exec_q = 1;
+               }
+       }
+
+       /* init sockets to receive events */
+       if (init_udevd_socket() < 0) {
+               if (errno == EADDRINUSE) {
+                       dbg("another udevd running, exit");
+                       rc = 1;
+               } else {
+                       dbg("error initializing udevd socket: %s", strerror(errno));
+                       rc = 2;
+               }
                goto exit;
        }
 
-       /* daemonize on request */
-       if (argc == 2 && strcmp(argv[1], "-d") == 0) {
+       if (init_uevent_netlink_sock() < 0) {
+               err("uevent socket not available");
+               rc = 3;
+               goto exit;
+       }
+
+       /* parse the rules and keep it in memory */
+       udev_rules_init(&rules, 1);
+
+       if (daemonize) {
                pid_t pid;
 
                pid = fork();
                switch (pid) {
                case 0:
-                       dbg("damonized fork running");
+                       dbg("daemonized fork running");
                        break;
                case -1:
-                       dbg("fork of daemon failed");
+                       err("fork of daemon failed");
+                       rc = 4;
                        goto exit;
                default:
-                       logging_close();
-                       exit(0);
+                       dbg("child [%u] running, parent exits", pid);
+                       goto exit;
                }
        }
 
-       /* make sure we don't lock any path */
-       chdir("/");
-       umask(umask(077) | 022);
+       /* set scheduling priority for the daemon */
+       setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
 
-       /* Set fds to dev/null */
-       fd = open( "/dev/null", O_RDWR );
-       if ( fd < 0 ) {
-               dbg("error opening /dev/null %s", strerror(errno));
-               goto exit;
-       }
-       dup2(fd, 0);
-       dup2(fd, 1);
-       dup2(fd, 2);
-       if (fd > 2) 
-               close(fd);
+       chdir("/");
+       umask(077);
 
        /* become session leader */
-       setsid();
+       sid = setsid();
+       dbg("our session is %d", sid);
+
+       /* OOM_DISABLE == -17 */
+       fd = open("/proc/self/oom_adj", O_RDWR);
+       if (fd < 0)
+               err("error disabling OOM");
+       else {
+               write(fd, "-17", 3);
+               close(fd);
+       }
 
        /* setup signal handler pipe */
-       retval = pipe(pipefds);
+       retval = pipe(signal_pipe);
        if (retval < 0) {
-               dbg("error getting pipes: %s", strerror(errno));
+               err("error getting pipes: %s", strerror(errno));
                goto exit;
        }
-
-       retval = fcntl(pipefds[0], F_SETFL, O_NONBLOCK);
+       retval = fcntl(signal_pipe[READ_END], F_SETFL, O_NONBLOCK);
        if (retval < 0) {
-               dbg("error fcntl on read pipe: %s", strerror(errno));
+               err("error fcntl on read pipe: %s", strerror(errno));
                goto exit;
        }
-       retval = fcntl(pipefds[0], F_SETFD, FD_CLOEXEC);
+       retval = fcntl(signal_pipe[WRITE_END], F_SETFL, O_NONBLOCK);
        if (retval < 0) {
-               dbg("error fcntl on read pipe: %s", strerror(errno));
-               goto exit;
-       }
-
-       retval = fcntl(pipefds[1], F_SETFL, O_NONBLOCK);
-       if (retval < 0) {
-               dbg("error fcntl on write pipe: %s", strerror(errno));
-               goto exit;
-       }
-       retval = fcntl(pipefds[1], F_SETFD, FD_CLOEXEC);
-       if (retval < 0) {
-               dbg("error fcntl on write pipe: %s", strerror(errno));
+               err("error fcntl on write pipe: %s", strerror(errno));
                goto exit;
        }
 
        /* set signal handlers */
-       act.sa_handler = (void (*) (int))sig_handler;
+       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(SIGALRM, &act, NULL);
        sigaction(SIGCHLD, &act, NULL);
-
-       memset(&saddr, 0x00, sizeof(saddr));
-       saddr.sun_family = AF_LOCAL;
-       /* use abstract namespace for socket path */
-       strcpy(&saddr.sun_path[1], UDEVD_SOCK_PATH);
-       addrlen = offsetof(struct sockaddr_un, sun_path) + strlen(saddr.sun_path+1) + 1;
-
-       udevsendsock = socket(AF_LOCAL, SOCK_DGRAM, 0);
-       if (udevsendsock == -1) {
-               dbg("error getting socket, exit");
-               goto exit;
+       sigaction(SIGHUP, &act, NULL);
+
+       /* watch rules directory */
+       inotify_fd = inotify_init();
+       if (inotify_fd > 0)
+               inotify_add_watch(inotify_fd, udev_rules_filename, IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
+
+       /* init of expected_seqnum value */
+       value = getenv("UDEVD_EXPECTED_SEQNUM");
+       if (value) {
+               expected_seqnum = strtoull(value, NULL, 10);
+               info("initialize expected_seqnum to %llu", expected_seqnum);
        }
 
-       /* the bind takes care of ensuring only one copy running */
-       retval = bind(udevsendsock, (struct sockaddr *) &saddr, addrlen);
-       if (retval < 0) {
-               dbg("bind failed, exit");
-               close(udevsendsock);
-               goto exit;
-       }
+       /* timeout to wait for missing events */
+       value = getenv("UDEVD_EVENT_TIMEOUT");
+       if (value)
+               event_timeout = strtoul(value, NULL, 10);
+       else
+               event_timeout = UDEVD_EVENT_TIMEOUT;
+       info("initialize event_timeout to %u", event_timeout);
 
-       /* enable receiving of the sender credentials */
-       setsockopt(udevsendsock, SOL_SOCKET, SO_PASSCRED, &feature_on, sizeof(feature_on));
+       /* maximum limit of forked childs */
+       value = getenv("UDEVD_MAX_CHILDS");
+       if (value)
+               max_childs = strtoul(value, NULL, 10);
+       else
+               max_childs = UDEVD_MAX_CHILDS;
+       info("initialize max_childs to %u", max_childs);
 
-       /* possible override of udev binary, used for testing */
-       udev_bin = getenv("UDEV_BIN");
-       if (udev_bin != NULL)
-               dbg("udev binary is set to '%s'", udev_bin);
+       /* 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
-               udev_bin = UDEV_BIN;
+               max_childs_running = UDEVD_MAX_CHILDS_RUNNING;
+       info("initialize max_childs_running to %u", max_childs_running);
 
-       /* handle special startup timeout*/
-       sysinfo(&info);
-       startup_time = info.uptime;
+       /* clear environment for forked event processes */
+       clearenv();
 
-       FD_ZERO(&readfds);
-       FD_SET(udevsendsock, &readfds);
-       FD_SET(pipefds[0], &readfds);
-       maxsockplus = udevsendsock+1;
-       while (1) {
-               struct hotplug_msg *msg;
+       /* export log_priority , as called programs may want to follow that setting */
+       sprintf(udev_log, "UDEV_LOG=%i", udev_log_priority);
+       putenv(udev_log);
+
+       while (!udev_exit) {
+               struct uevent_msg *msg;
+               int fdcount;
 
-               fd_set workreadfds = readfds;
-               retval = select(maxsockplus, &workreadfds, NULL, NULL, NULL);
+               FD_ZERO(&readfds);
+               FD_SET(signal_pipe[READ_END], &readfds);
+               FD_SET(udevd_sock, &readfds);
+               FD_SET(uevent_netlink_sock, &readfds);
+               if (inotify_fd > 0)
+                       FD_SET(inotify_fd, &readfds);
 
-               if (retval < 0) {
+               fdcount = select(UDEV_MAX(uevent_netlink_sock, inotify_fd)+1, &readfds, NULL, NULL, NULL);
+               if (fdcount < 0) {
                        if (errno != EINTR)
                                dbg("error in select: %s", strerror(errno));
                        continue;
                }
 
-               if (FD_ISSET(udevsendsock, &workreadfds)) {
-                       msg = get_udevsend_msg();
-                       if (msg)
+               /* get user socket message */
+               if (FD_ISSET(udevd_sock, &readfds)) {
+                       msg = get_udevd_msg();
+                       if (msg) {
+                               /* discard kernel messages if netlink is active */
+                               if (uevent_netlink_active && msg->type == UDEVD_UEVENT_UDEVSEND && msg->seqnum != 0) {
+                                       info("skip uevent_helper message with SEQNUM, netlink is active");
+                                       free(msg);
+                               } else
+                                       msg_queue_insert(msg);
+                       }
+               }
+
+               /* get kernel netlink message */
+               if (FD_ISSET(uevent_netlink_sock, &readfds)) {
+                       msg = get_netlink_msg();
+                       if (msg) {
                                msg_queue_insert(msg);
+                               /* disable udevsend with first netlink message */
+                               if (!uevent_netlink_active) {
+                                       info("uevent_nl message received, disable udevsend messages");
+                                       uevent_netlink_active = 1;
+                               }
+                       }
                }
 
-               if (FD_ISSET(pipefds[0], &workreadfds))
-                       user_sighandler();
+               /* received a signal, clear our notification pipe */
+               if (FD_ISSET(signal_pipe[READ_END], &readfds)) {
+                       char buf[256];
+
+                       read(signal_pipe[READ_END], &buf, sizeof(buf));
+               }
 
+               /* rules directory inotify watch */
+               if ((inotify_fd > 0) && FD_ISSET(inotify_fd, &readfds)) {
+                       int nbytes;
+
+                       /* discard all possible events, we can just reload the config */
+                       if ((ioctl(inotify_fd, FIONREAD, &nbytes) == 0) && nbytes) {
+                               char *buf;
+
+                               reload_config = 1;
+                               buf = malloc(nbytes);
+                               if (!buf) {
+                                       err("error getting buffer for inotify, disable watching");
+                                       close(inotify_fd);
+                                       inotify_fd = -1;
+                               }
+                               read(inotify_fd, buf, nbytes);
+                               free(buf);
+                       }
+               }
+
+               /* rules changed, set by inotify or a signal*/
+               if (reload_config) {
+                       reload_config = 0;
+                       udev_rules_close(&rules);
+                       udev_rules_init(&rules, 1);
+               }
+
+               /* forked child have returned */
                if (sigchilds_waiting) {
                        sigchilds_waiting = 0;
                        reap_sigchilds();
@@ -624,11 +1087,27 @@ int main(int argc, char *argv[], char *envp[])
                        }
 
                        run_exec_q = 0;
-                       exec_queue_manager();
+                       if (!stop_exec_q)
+                               exec_queue_manager();
                }
        }
 
 exit:
+       udev_rules_close(&rules);
+
+       if (signal_pipe[READ_END] > 0)
+               close(signal_pipe[READ_END]);
+       if (signal_pipe[WRITE_END] > 0)
+               close(signal_pipe[WRITE_END]);
+
+       if (udevd_sock > 0)
+               close(udevd_sock);
+       if (inotify_fd > 0)
+               close(inotify_fd);
+       if (uevent_netlink_sock > 0)
+               close(uevent_netlink_sock);
+
        logging_close();
-       return 1;
+
+       return rc;
 }