On Mon, Jun 1, 2009 at 21:39, Kay Sievers <kay.sievers@xxxxxxxx> wrote:
> On Mon, Jun 1, 2009 at 18:24, Alan Jenkins <alan-jenkins@xxxxxxxxxxxxxx> wrote:
>
>> I don't think the signal handler should be doing list traversal, in case it
>> interrupts list manipulation in the main loop. Or perhaps the main loop
>> should only allow signals during ppoll().
>
> Make sense. we can block them during list mangling, or just go for
> signalfd, and convert all the signal handling to it.
Here is a version that uses signalfd(), and a socketpair() from the
workers to pass data back. Same numbers so far, but the code looks a
lot simpler.
Kay
diff --git a/configure.ac b/configure.ac
index f1d008e..f126146 100644
--- a/configure.ac
+++ b/configure.ac
@@ -23,10 +23,6 @@ AC_SUBST(LIBUDEV_LT_AGE)
AC_PATH_PROG([XSLTPROC], [xsltproc])
-AC_CHECK_LIB(c, inotify_init,
- [AC_DEFINE([HAVE_INOTIFY], 1, [inotify available])],
- [AC_MSG_WARN([inotify support disabled])])
-
AC_ARG_WITH(udev-prefix,
AS_HELP_STRING([--with-udev-prefix=DIR], [add prefix to internal udev path names]),
[], [with_udev_prefix='${exec_prefix}'])
diff --git a/udev/Makefile.am b/udev/Makefile.am
index 6cd2f23..94989e6 100644
--- a/udev/Makefile.am
+++ b/udev/Makefile.am
@@ -14,7 +14,6 @@ common_ldadd =
common_files = \
udev.h \
- udev-sysdeps.h \
udev-event.c \
udev-watch.c \
udev-node.c \
diff --git a/udev/lib/libudev-monitor.c b/udev/lib/libudev-monitor.c
index 395a4d2..54c9576 100644
--- a/udev/lib/libudev-monitor.c
+++ b/udev/lib/libudev-monitor.c
@@ -32,7 +32,6 @@ struct udev_monitor {
int refcount;
int sock;
struct sockaddr_nl snl;
- struct sockaddr_nl snl_peer;
struct sockaddr_un sun;
socklen_t addrlen;
struct udev_list_node filter_subsystem_list;
@@ -171,8 +170,8 @@ struct udev_monitor *udev_monitor_new_from_netlink(struct udev *udev, const char
return NULL;
if (name == NULL)
- return NULL;
- if (strcmp(name, "kernel") == 0)
+ group = 0;
+ else if (strcmp(name, "kernel") == 0)
group = UDEV_MONITOR_KERNEL;
else if (strcmp(name, "udev") == 0)
group = UDEV_MONITOR_UDEV;
@@ -193,8 +192,6 @@ struct udev_monitor *udev_monitor_new_from_netlink(struct udev *udev, const char
udev_monitor->snl.nl_family = AF_NETLINK;
udev_monitor->snl.nl_groups = group;
- udev_monitor->snl_peer.nl_family = AF_NETLINK;
- udev_monitor->snl_peer.nl_groups = UDEV_MONITOR_UDEV;
dbg(udev, "monitor %p created with NETLINK_KOBJECT_UEVENT (%u)\n", udev_monitor, group);
return udev_monitor;
@@ -434,7 +431,6 @@ struct udev_device *udev_monitor_receive_device(struct udev_monitor *udev_monito
struct iovec iov;
char cred_msg[CMSG_SPACE(sizeof(struct ucred))];
struct cmsghdr *cmsg;
- struct sockaddr_nl snl;
struct ucred *cred;
char buf[8192];
ssize_t buflen;
@@ -459,11 +455,6 @@ retry:
smsg.msg_control = cred_msg;
smsg.msg_controllen = sizeof(cred_msg);
- if (udev_monitor->snl.nl_family != 0) {
- smsg.msg_name = &snl;
- smsg.msg_namelen = sizeof(snl);
- }
-
buflen = recvmsg(udev_monitor->sock, &smsg, 0);
if (buflen < 0) {
if (errno != EINTR)
@@ -476,20 +467,6 @@ retry:
return NULL;
}
- if (udev_monitor->snl.nl_family != 0) {
- if (snl.nl_groups == 0) {
- info(udev_monitor->udev, "unicast netlink message ignored\n");
- return NULL;
- }
- if (snl.nl_groups == UDEV_MONITOR_KERNEL) {
- if (snl.nl_pid > 0) {
- info(udev_monitor->udev, "multicast kernel netlink message from pid %d ignored\n", snl.nl_pid);
- return NULL;
- }
- is_kernel = 1;
- }
- }
-
cmsg = CMSG_FIRSTHDR(&smsg);
if (cmsg == NULL || cmsg->cmsg_type != SCM_CREDENTIALS) {
info(udev_monitor->udev, "no sender credentials received, message ignored\n");
@@ -621,7 +598,7 @@ retry:
return udev_device;
}
-int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_device *udev_device)
+int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_device *udev_device, pid_t pid)
{
struct msghdr smsg;
struct iovec iov[2];
@@ -660,6 +637,7 @@ int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_devi
} else if (udev_monitor->snl.nl_family != 0) {
const char *val;
struct udev_monitor_netlink_header nlh;
+ struct sockaddr_nl snl_peer;
/* add versioned header */
@@ -680,11 +658,18 @@ int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_devi
iov[1].iov_base = (char *)buf;
iov[1].iov_len = blen;
+ /* we will always get ECONNREFUSED when sending to the muticast group */
+ memset(&snl_peer, 0x00, sizeof(struct sockaddr_nl));
+ snl_peer.nl_family = AF_NETLINK;
+ if (pid > 0)
+ snl_peer.nl_pid = pid;
+ else
+ snl_peer.nl_groups = UDEV_MONITOR_UDEV;
+
memset(&smsg, 0x00, sizeof(struct msghdr));
smsg.msg_iov = iov;
smsg.msg_iovlen = 2;
- /* no destination besides the muticast group, we will always get ECONNREFUSED */
- smsg.msg_name = &udev_monitor->snl_peer;
+ smsg.msg_name = &snl_peer;
smsg.msg_namelen = sizeof(struct sockaddr_nl);
} else {
return -1;
diff --git a/udev/lib/libudev-private.h b/udev/lib/libudev-private.h
index dc02a84..e600802 100644
--- a/udev/lib/libudev-private.h
+++ b/udev/lib/libudev-private.h
@@ -86,7 +86,7 @@ int udev_device_delete_db(struct udev_device *udev_device);
int udev_device_rename_db(struct udev_device *udev_device, const char *devpath);
/* libudev-monitor - netlink/unix socket communication */
-int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_device *udev_device);
+int udev_monitor_send_device(struct udev_monitor *udev_monitor, struct udev_device *udev_device, pid_t pid);
int udev_monitor_set_receive_buffer_size(struct udev_monitor *udev_monitor, int size);
/* libudev-ctrl - daemon runtime setup */
diff --git a/udev/udev-event.c b/udev/udev-event.c
index d521251..8ab262a 100644
--- a/udev/udev-event.c
+++ b/udev/udev-event.c
@@ -734,18 +734,13 @@ int udev_event_execute_run(struct udev_event *event)
monitor = udev_monitor_new_from_socket(event->udev, &cmd[strlen("socket:")]);
if (monitor == NULL)
continue;
- udev_monitor_send_device(monitor, event->dev);
+ udev_monitor_send_device(monitor, event->dev, 0);
udev_monitor_unref(monitor);
} else {
char program[UTIL_PATH_SIZE];
char **envp;
udev_event_apply_format(event, cmd, program, sizeof(program));
- if (event->trace)
- fprintf(stderr, "run %s (%llu) '%s'\n",
- udev_device_get_syspath(event->dev),
- udev_device_get_seqnum(event->dev),
- program);
envp = udev_device_get_properties_envp(event->dev);
if (util_run_program(event->udev, program, envp, NULL, 0, NULL) != 0) {
if (!udev_list_entry_get_flag(list_entry))
diff --git a/udev/udev-sysdeps.h b/udev/udev-sysdeps.h
deleted file mode 100644
index 35671ba..0000000
--- a/udev/udev-sysdeps.h
+++ /dev/null
@@ -1,44 +0,0 @@
-/*
- * wrapping of libc features and kernel interfaces
- *
- * Copyright (C) 2005-2008 Kay Sievers <kay.sievers@xxxxxxxx>
- *
- * 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, see <http://www.gnu.org/licenses/>.
- */
-
-#ifndef _UDEV_SYSDEPS_H_
-#define _UDEV_SYSDEPS_H_
-
-#include <stdint.h>
-#include <errno.h>
-
-#ifndef HAVE_INOTIFY
-static inline int inotify_init(void)
-{
- errno = ENOSYS;
- return -1;
-}
-
-static inline int inotify_add_watch(int fd, const char *name, uint32_t mask)
-{
- return -1;
-}
-
-#define IN_CREATE 0
-#define IN_DELETE 0
-#define IN_MOVE 0
-#define IN_CLOSE_WRITE 0
-
-#endif /* HAVE_INOTIFY */
-#endif
diff --git a/udev/udev-watch.c b/udev/udev-watch.c
index 53492e5..5a49c96 100644
--- a/udev/udev-watch.c
+++ b/udev/udev-watch.c
@@ -26,27 +26,24 @@
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
-#ifdef HAVE_INOTIFY
#include <sys/inotify.h>
-#endif
#include "udev.h"
-int inotify_fd = -1;
+static int inotify_fd = -1;
/* inotify descriptor, will be shared with rules directory;
* set to cloexec since we need our children to be able to add
* watches for us
*/
-void udev_watch_init(struct udev *udev)
+int udev_watch_init(struct udev *udev)
{
inotify_fd = inotify_init();
if (inotify_fd >= 0)
util_set_fd_cloexec(inotify_fd);
- else if (errno == ENOSYS)
- info(udev, "unable to use inotify, udevd will not monitor rule files changes\n");
else
err(udev, "inotify_init failed: %m\n");
+ return inotify_fd;
}
/* move any old watches directory out of the way, and then restore
diff --git a/udev/udev.h b/udev/udev.h
index 8f2c1c6..7187975 100644
--- a/udev/udev.h
+++ b/udev/udev.h
@@ -22,7 +22,6 @@
#include <sys/types.h>
#include <sys/param.h>
-#include "udev-sysdeps.h"
#include "lib/libudev.h"
#include "lib/libudev-private.h"
@@ -53,7 +52,6 @@ static inline void logging_close(void)
}
struct udev_event {
- struct udev_list_node node;
struct udev *udev;
struct udev_device *dev;
struct udev_device *dev_parent;
@@ -64,10 +62,6 @@ struct udev_event {
uid_t uid;
gid_t gid;
struct udev_list_node run_list;
- pid_t pid;
- int exitstatus;
- time_t queue_time;
- unsigned long long int delaying_seqnum;
unsigned int group_final:1;
unsigned int owner_final:1;
unsigned int mode_final:1;
@@ -76,7 +70,6 @@ struct udev_event {
unsigned int run_final:1;
unsigned int ignore_device:1;
unsigned int inotify_watch:1;
- unsigned int trace:1;
};
struct udev_watch {
@@ -101,8 +94,7 @@ int udev_event_apply_subsys_kernel(struct udev_event *event, const char *string,
char *result, size_t maxsize, int read_value);
/* udev-watch.c */
-extern int inotify_fd;
-void udev_watch_init(struct udev *udev);
+int udev_watch_init(struct udev *udev);
void udev_watch_restore(struct udev *udev);
void udev_watch_begin(struct udev *udev, struct udev_device *dev);
void udev_watch_end(struct udev *udev, struct udev_device *dev);
diff --git a/udev/udevd.c b/udev/udevd.c
index 37b547a..7e6d244 100644
--- a/udev/udevd.c
+++ b/udev/udevd.c
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2004-2008 Kay Sievers <kay.sievers@xxxxxxxx>
+ * Copyright (C) 2004-2009 Kay Sievers <kay.sievers@xxxxxxxx>
* Copyright (C) 2004 Chris Friesen <chris_friesen@xxxxxxxxxxxx>
* Copyright (C) 2009 Canonical Ltd.
* Copyright (C) 2009 Scott James Remnant <scott@xxxxxxxxxxxx>
@@ -30,23 +30,21 @@
#include <time.h>
#include <getopt.h>
#include <dirent.h>
+#include <sys/prctl.h>
+#include <sys/socket.h>
+#include <sys/signalfd.h>
#include <sys/select.h>
#include <sys/poll.h>
#include <sys/wait.h>
#include <sys/stat.h>
#include <sys/ioctl.h>
-#ifdef HAVE_INOTIFY
#include <sys/inotify.h>
-#endif
#include "udev.h"
#define UDEVD_PRIORITY -4
#define UDEV_PRIORITY -2
-/* maximum limit of forked childs */
-#define UDEVD_MAX_CHILDS 256
-
static int debug;
static void log_fn(struct udev *udev, int priority,
@@ -61,163 +59,311 @@ static void log_fn(struct udev *udev, int priority,
}
}
-static void reap_sigchilds(void);
-
static int debug_trace;
static struct udev_rules *rules;
static struct udev_queue_export *udev_queue_export;
static struct udev_ctrl *udev_ctrl;
-static struct udev_monitor *kernel_monitor;
-static volatile sig_atomic_t sigchilds_waiting;
-static volatile sig_atomic_t udev_exit;
-static volatile sig_atomic_t reload_config;
-static volatile sig_atomic_t signal_received;
-static volatile pid_t settle_pid;
-static int run_exec_q;
-static int stop_exec_q;
+static struct udev_monitor *monitor;
+static int worker_socket[2];
+static pid_t settle_pid;
+static int stop_exec_queue;
+static int reload_config;
static int max_childs;
static int childs;
static struct udev_list_node event_list;
+static struct udev_list_node worker_list;
+static int udev_exit;
+static volatile sig_atomic_t worker_exit;
+
+enum poll_fd {
+ FD_CONTROL,
+ FD_NETLINK,
+ FD_INOTIFY,
+ FD_SIGNAL,
+ FD_WORKER,
+};
+
+static struct pollfd pfd[] = {
+ [FD_NETLINK] = { .events = POLLIN },
+ [FD_WORKER] = { .events = POLLIN },
+ [FD_SIGNAL] = { .events = POLLIN },
+ [FD_INOTIFY] = { .events = POLLIN },
+ [FD_CONTROL] = { .events = POLLIN },
+};
+
+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;
+};
-static struct udev_event *node_to_event(struct udev_list_node *node)
+static struct event *node_to_event(struct udev_list_node *node)
{
char *event;
event = (char *)node;
- event -= offsetof(struct udev_event, node);
- return (struct udev_event *)event;
+ event -= offsetof(struct event, node);
+ return (struct event *)event;
+}
+
+enum worker_state {
+ WORKER_UNDEF,
+ WORKER_RUNNING,
+ WORKER_IDLE,
+ WORKER_KILLED,
+};
+
+struct worker {
+ struct udev_list_node node;
+ pid_t pid;
+ enum worker_state state;
+ struct event *event;
+};
+
+/* passed from worker to main process */
+struct worker_message {
+ pid_t pid;
+ int exitcode;
+};
+
+static struct worker *node_to_worker(struct udev_list_node *node)
+{
+ char *worker;
+
+ worker = (char *)node;
+ worker -= offsetof(struct worker, node);
+ return (struct worker *)worker;
}
-static void event_queue_delete(struct udev_event *event)
+static void event_queue_delete(struct event *event)
{
udev_list_node_remove(&event->node);
/* mark as failed, if "add" event returns non-zero */
- if (event->exitstatus && strcmp(udev_device_get_action(event->dev), "add") == 0)
+ if (event->exitcode && 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);
udev_device_unref(event->dev);
- udev_event_unref(event);
+ free(event);
}
static void event_sig_handler(int signum)
{
- if (signum == SIGALRM)
+ switch (signum) {
+ case SIGALRM:
_exit(1);
+ break;
+ case SIGTERM:
+ worker_exit = 1;
+ break;
+ }
}
-static void event_fork(struct udev_event *event)
+static void worker_new(struct event *event)
{
+ struct worker *worker;
pid_t pid;
struct sigaction act;
- int err;
-
-#if 0
- /* single process, no forking, just for testing/profiling */
- err = udev_event_execute_rules(event, rules);
- if (err == 0 && !event->ignore_device && udev_get_run(event->udev))
- udev_event_execute_run(event);
- info(event->udev, "seq %llu exit with %i\n", udev_device_get_seqnum(event->dev), err);
- event_queue_delete(event);
- return;
-#endif
- if (debug_trace) {
- event->trace = 1;
- fprintf(stderr, "fork %s (%llu)\n",
- udev_device_get_syspath(event->dev),
- udev_device_get_seqnum(event->dev));
- }
+ worker = calloc(1, sizeof(struct worker));
+ if (worker == NULL)
+ return;
pid = fork();
switch (pid) {
- case 0:
- /* child */
+ case 0: {
+ sigset_t mask;
+ struct udev_device *dev;
+
udev_queue_export_unref(udev_queue_export);
udev_ctrl_unref(udev_ctrl);
+ udev_monitor_unref(monitor);
+ close(pfd[FD_SIGNAL].fd);
+ close(worker_socket[READ_END]);
logging_close();
- logging_init("udevd-event");
+ logging_init("udevd-work");
setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
+ /* re-open socket to listen to udevd only, and send back libudev events */
+ monitor = udev_monitor_new_from_netlink(event->udev, NULL);
+ if (monitor == NULL)
+ _exit(2);
+ udev_monitor_enable_receiving(monitor);
+
/* 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);
- /* reset to default */
- act.sa_handler = SIG_DFL;
- sigaction(SIGINT, &act, NULL);
- sigaction(SIGTERM, &act, NULL);
- sigaction(SIGCHLD, &act, NULL);
- sigaction(SIGHUP, &act, NULL);
+ /* unblock signals */
+ sigfillset(&mask);
+ sigdelset(&mask, SIGTERM);
+ sigdelset(&mask, SIGALRM);
+ sigprocmask(SIG_SETMASK, &mask, NULL);
- /* set timeout to prevent hanging processes */
- alarm(UDEV_EVENT_TIMEOUT);
+ /* request TERM signal if main daemon exits */
+ prctl(PR_SET_PDEATHSIG, SIGTERM);
- /* apply rules, create node, symlinks */
- err = udev_event_execute_rules(event, rules);
+ /* initial device */
+ dev = event->dev;
- /* rules may change/disable the timeout */
- if (udev_device_get_event_timeout(event->dev) >= 0)
- alarm(udev_device_get_event_timeout(event->dev));
+ while (!worker_exit) {
+ struct udev_event *udev_event;
+ struct worker_message msg;
+ int err;
- /* execute RUN= */
- if (err == 0 && !event->ignore_device && udev_get_run(event->udev))
- udev_event_execute_run(event);
+ udev_event = udev_event_new(dev);
+ if (udev_event == NULL)
+ _exit(3);
- /* apply/restore inotify watch */
- if (err == 0 && event->inotify_watch) {
- udev_watch_begin(event->udev, event->dev);
- udev_device_update_db(event->dev);
- }
+ /* set timeout to prevent hanging processes */
+ alarm(UDEV_EVENT_TIMEOUT);
+
+ /* 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))
+ udev_event_execute_run(udev_event);
+
+ /* 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);
+ }
+
+ /* send processed event back to libudev listeners */
+ udev_monitor_send_device(monitor, dev, 0);
+
+ info(event->udev, "seq %llu finished with %i\n", udev_device_get_seqnum(dev), err);
+ udev_device_unref(dev);
+ udev_event_unref(udev_event);
+
+ /* send back the result of the event execution */
+ msg.exitcode = err;
+ msg.pid = getpid();
+ send(worker_socket[WRITE_END], &msg, sizeof(struct worker_message), 0);
- /* send processed event back to the kernel netlink socket */
- udev_monitor_send_device(kernel_monitor, event->dev);
+ /* wait for more device messages from udevd */
+ do
+ dev = udev_monitor_receive_device(monitor);
+ while (!worker_exit && dev == NULL);
+ }
- info(event->udev, "seq %llu exit with %i\n", udev_device_get_seqnum(event->dev), err);
+ udev_monitor_unref(monitor);
logging_close();
- if (err != 0)
- exit(1);
exit(0);
+ }
case -1:
+ event->state = EVENT_QUEUED;
+ free(worker);
err(event->udev, "fork of child failed: %m\n");
- event_queue_delete(event);
break;
default:
- /* get SIGCHLD in main loop */
- info(event->udev, "seq %llu forked, pid [%d], '%s' '%s', %ld seconds old\n",
- udev_device_get_seqnum(event->dev),
- pid,
- udev_device_get_action(event->dev),
- udev_device_get_subsystem(event->dev),
- time(NULL) - event->queue_time);
- event->pid = pid;
+ event->state = EVENT_RUNNING;
+ worker->event = event;
+ worker->pid = pid;
+ worker->state = WORKER_RUNNING;
+ udev_list_node_append(&worker->node, &worker_list);
childs++;
+ break;
}
}
-static void event_queue_insert(struct udev_event *event)
+static void event_run(struct event *event)
{
- event->queue_time = time(NULL);
+ struct udev_list_node *loop;
- udev_queue_export_device_queued(udev_queue_export, event->dev);
- info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(event->dev),
- udev_device_get_action(event->dev), udev_device_get_subsystem(event->dev));
+ udev_list_node_foreach(loop, &worker_list) {
+ struct worker *worker = node_to_worker(loop);
+ ssize_t count;
+ if (worker->state != WORKER_IDLE)
+ continue;
+
+ worker->event = event;
+ worker->state = WORKER_RUNNING;
+ event->state = EVENT_RUNNING;
+ count = udev_monitor_send_device(monitor, event->dev, worker->pid);
+ if (count < 0) {
+ err(event->udev, "worker [%u] did not accept message, kill it\n", worker->pid);
+ event->state = EVENT_QUEUED;
+ worker->state = WORKER_KILLED;
+ kill(worker->pid, SIGKILL);
+ continue;
+ }
+ return;
+ }
+
+ if (childs >= max_childs) {
+ info(event->udev, "maximum number (%i) of childs reached\n", childs);
+ return;
+ }
+
+ /* start new worker and pass initial device */
+ worker_new(event);
+}
+
+static void event_queue_insert(struct udev_device *dev)
+{
+ struct event *event;
+
+ event = calloc(1, sizeof(struct event));
+ if (event == NULL)
+ return;
+
+ event->udev = udev_device_get_udev(dev);
+ event->dev = dev;
+ udev_queue_export_device_queued(udev_queue_export, dev);
+ info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
+ udev_device_get_action(dev), udev_device_get_subsystem(dev));
+
+ event->state = EVENT_QUEUED;
udev_list_node_append(&event->node, &event_list);
- run_exec_q = 1;
/* run all events with a timeout set immediately */
- if (udev_device_get_timeout(event->dev) > 0) {
- event_fork(event);
+ if (udev_device_get_timeout(dev) > 0) {
+ worker_new(event);
return;
}
}
+static void worker_kill_idle(void)
+{
+ struct udev_list_node *loop;
+
+ udev_list_node_foreach(loop, &worker_list) {
+ struct worker *worker = node_to_worker(loop);
+
+ if (worker->state != WORKER_IDLE)
+ continue;
+
+ worker->state = WORKER_KILLED;
+ kill(worker->pid, SIGTERM);
+ }
+}
+
static int mem_size_mb(void)
{
FILE *f;
@@ -265,13 +411,13 @@ static int compare_devpath(const char *running, const char *waiting)
}
/* lookup event for identical, parent, child device */
-static int devpath_busy(struct udev_event *event)
+static int devpath_busy(struct event *event)
{
struct udev_list_node *loop;
/* check if queue contains events we depend on */
udev_list_node_foreach(loop, &event_list) {
- struct udev_event *loop_event = node_to_event(loop);
+ struct event *loop_event = node_to_event(loop);
/* we already found a later event, earlier can not block us, no need to check again */
if (udev_device_get_seqnum(loop_event->dev) < event->delaying_seqnum)
@@ -305,52 +451,6 @@ static int devpath_busy(struct udev_event *event)
return 0;
}
-/* serializes events for the identical and parent and child devices */
-static void event_queue_manager(struct udev *udev)
-{
- struct udev_list_node *loop;
- struct udev_list_node *tmp;
-
-start_over:
- if (udev_list_is_empty(&event_list)) {
- if (childs > 0) {
- err(udev, "event list empty, but childs count is %i", childs);
- childs = 0;
- }
- return;
- }
-
- udev_list_node_foreach_safe(loop, tmp, &event_list) {
- struct udev_event *loop_event = node_to_event(loop);
-
- if (childs >= max_childs) {
- info(udev, "maximum number (%i) of childs reached\n", childs);
- break;
- }
-
- if (loop_event->pid != 0)
- continue;
-
- /* do not start event if parent or child event is still running */
- if (devpath_busy(loop_event) != 0) {
- dbg(udev, "delay seq %llu (%s)\n",
- udev_device_get_seqnum(loop_event->dev),
- udev_device_get_devpath(loop_event->dev));
- continue;
- }
-
- event_fork(loop_event);
- dbg(udev, "moved seq %llu to running list\n", udev_device_get_seqnum(loop_event->dev));
-
- /* retry if events finished in the meantime */
- if (sigchilds_waiting) {
- sigchilds_waiting = 0;
- reap_sigchilds();
- goto start_over;
- }
- }
-}
-
/* receive the udevd message from userspace */
static void handle_ctrl_msg(struct udev_ctrl *uctrl)
{
@@ -371,13 +471,12 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl)
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 = 1;
}
if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
info(udev, "udevd message (START_EXEC_QUEUE) received\n");
- stop_exec_q = 0;
- event_queue_manager(udev);
+ stop_exec_queue = 0;
}
if (udev_ctrl_get_reload_rules(ctrl_msg) > 0) {
@@ -420,6 +519,8 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl)
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);
}
@@ -427,22 +528,20 @@ static void handle_ctrl_msg(struct udev_ctrl *uctrl)
/* read inotify messages */
static int handle_inotify(struct udev *udev)
{
- int nbytes, pos;
+ ssize_t nbytes, pos;
char *buf;
struct inotify_event *ev;
- if ((ioctl(inotify_fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
+ if ((ioctl(pfd[FD_INOTIFY].fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
return 0;
buf = malloc(nbytes);
if (buf == NULL) {
err(udev, "error getting buffer for inotify, disable watching\n");
- close(inotify_fd);
- inotify_fd = -1;
- return 0;
+ return -1;
}
- read(inotify_fd, buf, nbytes);
+ nbytes = read(pfd[FD_INOTIFY].fd, buf, nbytes);
for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
struct udev_device *dev;
@@ -480,71 +579,6 @@ static int handle_inotify(struct udev *udev)
return 0;
}
-static void 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;
- }
-
- signal_received = 1;
-}
-
-static void udev_done(int pid, int exitstatus)
-{
- struct udev_list_node *loop;
-
- /* find event associated with pid and delete it */
- udev_list_node_foreach(loop, &event_list) {
- struct udev_event *loop_event = node_to_event(loop);
-
- if (loop_event->pid == pid) {
- info(loop_event->udev, "seq %llu cleanup, pid [%d], status %i, %ld seconds old\n",
- udev_device_get_seqnum(loop_event->dev), loop_event->pid,
- exitstatus, time(NULL) - loop_event->queue_time);
- loop_event->exitstatus = exitstatus;
- if (debug_trace)
- fprintf(stderr, "exit %s (%llu)\n",
- udev_device_get_syspath(loop_event->dev),
- udev_device_get_seqnum(loop_event->dev));
- event_queue_delete(loop_event);
- childs--;
-
- /* there may be dependent events waiting */
- run_exec_q = 1;
- return;
- }
- }
-}
-
-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);
- }
-}
-
static void startup_log(struct udev *udev)
{
FILE *f;
@@ -576,7 +610,7 @@ int main(int argc, char *argv[])
{
struct udev *udev;
int fd;
- struct sigaction act;
+ sigset_t mask;
const char *value;
int daemonize = 0;
int resolve_names = 1;
@@ -669,29 +703,76 @@ 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);
- kernel_monitor = udev_monitor_new_from_netlink(udev, "kernel");
- if (kernel_monitor == NULL || udev_monitor_enable_receiving(kernel_monitor) < 0) {
+ monitor = udev_monitor_new_from_netlink(udev, "kernel");
+ if (monitor == NULL || udev_monitor_enable_receiving(monitor) < 0) {
fprintf(stderr, "error initializing netlink socket\n");
err(udev, "error initializing netlink socket\n");
rc = 3;
goto exit;
}
- udev_monitor_set_receive_buffer_size(kernel_monitor, 128*1024*1024);
+ udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
+ pfd[FD_NETLINK].fd = udev_monitor_get_fd(monitor);
+
+ pfd[FD_INOTIFY].fd = udev_watch_init(udev);
+ if (pfd[FD_INOTIFY].fd < 0) {
+ fprintf(stderr, "error initializing inotify\n");
+ err(udev, "error initializing inotify\n");
+ rc = 4;
+ goto exit;
+ }
+
+ if (udev_get_rules_path(udev) != NULL) {
+ inotify_add_watch(pfd[FD_NETLINK].fd, udev_get_rules_path(udev),
+ IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
+ } else {
+ char filename[UTIL_PATH_SIZE];
+
+ inotify_add_watch(pfd[FD_NETLINK].fd, UDEV_PREFIX "/lib/udev/rules.d",
+ IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
+ inotify_add_watch(pfd[FD_NETLINK].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);
+ inotify_add_watch(pfd[FD_NETLINK].fd, filename,
+ IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
+ }
+ udev_watch_restore(udev);
+
+ /* block and listen to all signals on signalfd */
+ sigfillset(&mask);
+ sigprocmask(SIG_SETMASK, &mask, NULL);
+ pfd[FD_SIGNAL].fd = signalfd(-1, &mask, 0);
+ if (pfd[FD_SIGNAL].fd < 0) {
+ fprintf(stderr, "error getting signalfd\n");
+ err(udev, "error getting signalfd\n");
+ rc = 5;
+ goto exit;
+ }
+
+ /* unnamed socket from workers to the main daemon */
+ if (socketpair(AF_LOCAL, SOCK_DGRAM, 0, worker_socket) < 0) {
+ fprintf(stderr, "error getting socketpair\n");
+ err(udev, "error getting socketpair\n");
+ rc = 6;
+ goto exit;
+ }
+ pfd[FD_WORKER].fd = worker_socket[READ_END];
rules = udev_rules_new(udev, resolve_names);
if (rules == NULL) {
err(udev, "error reading rules\n");
goto exit;
}
- udev_list_init(&event_list);
+
udev_queue_export = udev_queue_export_new(udev);
if (udev_queue_export == NULL) {
err(udev, "error creating queue file\n");
@@ -704,19 +785,19 @@ int main(int argc, char *argv[])
pid = fork();
switch (pid) {
case 0:
- dbg(udev, "daemonized fork running\n");
break;
case -1:
err(udev, "fork of daemon failed: %m\n");
rc = 4;
goto exit;
default:
- dbg(udev, "child [%u] running, parent exits\n", pid);
rc = 0;
goto exit;
}
}
+ startup_log(udev);
+
/* redirect std{out,err} */
if (!debug && !debug_trace) {
dup2(fd, STDIN_FILENO);
@@ -742,159 +823,189 @@ int main(int argc, char *argv[])
close(fd);
}
- startup_log(udev);
-
- /* set signal handlers */
- memset(&act, 0x00, sizeof(struct sigaction));
- act.sa_handler = sig_handler;
- sigemptyset(&act.sa_mask);
- act.sa_flags = SA_RESTART;
- sigaction(SIGINT, &act, NULL);
- sigaction(SIGTERM, &act, NULL);
- sigaction(SIGCHLD, &act, NULL);
- sigaction(SIGHUP, &act, NULL);
-
- /* watch rules directory */
- udev_watch_init(udev);
- if (inotify_fd >= 0) {
- if (udev_get_rules_path(udev) != NULL) {
- inotify_add_watch(inotify_fd, udev_get_rules_path(udev),
- IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
- } else {
- char filename[UTIL_PATH_SIZE];
-
- inotify_add_watch(inotify_fd, UDEV_PREFIX "/lib/udev/rules.d",
- IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
- inotify_add_watch(inotify_fd, SYSCONFDIR "/udev/rules.d",
- IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
-
- /* watch dynamic rules directory */
- util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/rules.d", NULL);
- inotify_add_watch(inotify_fd, filename,
- IN_CREATE | IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
- }
-
- udev_watch_restore(udev);
- }
-
/* in trace mode run one event after the other */
if (debug_trace) {
max_childs = 1;
} else {
int memsize = mem_size_mb();
+
if (memsize > 0)
max_childs = 128 + (memsize / 4);
else
- max_childs = UDEVD_MAX_CHILDS;
+ max_childs = 256;
}
+
/* possibly overwrite maximum limit of executed events */
value = getenv("UDEVD_MAX_CHILDS");
if (value)
max_childs = strtoul(value, NULL, 10);
info(udev, "initialize max_childs to %u\n", max_childs);
+ udev_list_init(&event_list);
+ udev_list_init(&worker_list);
+
while (!udev_exit) {
- sigset_t blocked_mask, orig_mask;
- struct pollfd pfd[4];
- struct pollfd *ctrl_poll, *monitor_poll, *inotify_poll = NULL;
- int nfds = 0;
int fdcount;
+ int timeout;
- sigfillset(&blocked_mask);
- sigprocmask(SIG_SETMASK, &blocked_mask, &orig_mask);
- if (signal_received) {
- sigprocmask(SIG_SETMASK, &orig_mask, NULL);
- goto handle_signals;
- }
-
- ctrl_poll = &pfd[nfds++];
- ctrl_poll->fd = udev_ctrl_get_fd(udev_ctrl);
- ctrl_poll->events = POLLIN;
-
- monitor_poll = &pfd[nfds++];
- monitor_poll->fd = udev_monitor_get_fd(kernel_monitor);
- monitor_poll->events = POLLIN;
+ /* set timeout to kill idle workers */
+ if (udev_list_is_empty(&event_list) && !udev_list_is_empty(&worker_list))
+ timeout = 10 * 1000;
+ else
+ timeout = -1;
+ /* wait for events */
+ fdcount = poll(pfd, ARRAY_SIZE(pfd), timeout);
+ if (fdcount < 0)
+ continue;
- if (inotify_fd >= 0) {
- inotify_poll = &pfd[nfds++];
- inotify_poll->fd = inotify_fd;
- inotify_poll->events = POLLIN;
+ /* timeout - kill idle workers */
+ if (fdcount == 0)
+ worker_kill_idle();
+
+ /* event has finished */
+ if (pfd[FD_WORKER].revents & POLLIN) {
+ while (1) {
+ struct worker_message msg;
+ ssize_t size;
+ struct udev_list_node *loop;
+
+ size = recv(pfd[FD_WORKER].fd, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
+ if (size != sizeof(struct worker_message))
+ break;
+
+ /* lookup worker who sent the signal */
+ udev_list_node_foreach(loop, &worker_list) {
+ struct worker *worker = node_to_worker(loop);
+
+ if (worker->pid != msg.pid)
+ continue;
+
+ /* worker returned */
+ worker->event->exitcode = msg.exitcode;
+ event_queue_delete(worker->event);
+ worker->event = NULL;
+ worker->state = WORKER_IDLE;
+ break;
+ }
+ }
}
- fdcount = ppoll(pfd, nfds, NULL, &orig_mask);
- sigprocmask(SIG_SETMASK, &orig_mask, NULL);
- if (fdcount < 0) {
- if (errno == EINTR)
- goto handle_signals;
- err(udev, "error in select: %m\n");
- continue;
+ /* get kernel uevent */
+ if (pfd[FD_NETLINK].revents & POLLIN) {
+ struct udev_device *dev;
+
+ dev = udev_monitor_receive_device(monitor);
+ if (dev != NULL)
+ event_queue_insert(dev);
+ else
+ udev_device_unref(dev);
}
/* get control message */
- if (ctrl_poll->revents & POLLIN)
+ if (pfd[FD_CONTROL].revents & POLLIN)
handle_ctrl_msg(udev_ctrl);
- /* get kernel uevent */
- if (monitor_poll->revents & POLLIN) {
- struct udev_device *dev;
+ /* start new events */
+ if (!udev_list_is_empty(&event_list) && !stop_exec_queue) {
+ struct udev_list_node *loop;
- dev = udev_monitor_receive_device(kernel_monitor);
- if (dev != NULL) {
- struct udev_event *event;
+ udev_list_node_foreach(loop, &event_list) {
+ struct event *event = node_to_event(loop);
+
+ /* skip running events */
+ if (event->state != EVENT_QUEUED)
+ continue;
+
+ /* do not start event if parent or child event is still running */
+ if (devpath_busy(event) != 0) {
+ dbg(udev, "delay seq %llu (%s)\n",
+ udev_device_get_seqnum(event->dev),
+ udev_device_get_devpath(event->dev));
+ continue;
+ }
- event = udev_event_new(dev);
- if (event != NULL)
- event_queue_insert(event);
- else
- udev_device_unref(dev);
+ event_run(event);
}
}
- /* rules directory inotify watch */
- if (inotify_poll && (inotify_poll->revents & POLLIN))
- handle_inotify(udev);
+ /* get signal */
+ if (pfd[FD_SIGNAL].revents & POLLIN) {
+ struct signalfd_siginfo fdsi;
+ ssize_t size;
+
+ size = read(pfd[FD_SIGNAL].fd, &fdsi, sizeof(struct signalfd_siginfo));
+ if (size != sizeof(struct signalfd_siginfo))
+ continue;
+ switch (fdsi.ssi_signo) {
+ case SIGINT:
+ case SIGTERM:
+ udev_exit = 1;
+ break;
+ case SIGCHLD:
+ while (1) {
+ pid_t pid;
+ struct udev_list_node *loop, *tmp;
+
+ pid = waitpid(-1, NULL, 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;
+
+ /* fail event, if worker died unexpectedly */
+ if (worker->event != NULL) {
+ worker->event->exitcode = 127;
+ event_queue_delete(worker->event);
+ }
+
+ udev_list_node_remove(&worker->node);
+ free(worker);
+ childs--;
+ break;
+ }
+ }
+ break;
+ case SIGHUP:
+ reload_config = 1;
+ break;
+ }
+ }
-handle_signals:
- signal_received = 0;
+ /* device node and rules directory inotify watch */
+ if (pfd[FD_INOTIFY].revents & POLLIN)
+ handle_inotify(udev);
/* rules changed, set by inotify or a HUP signal */
if (reload_config) {
struct udev_rules *rules_new;
- reload_config = 0;
+ worker_kill_idle();
rules_new = udev_rules_new(udev, resolve_names);
if (rules_new != NULL) {
udev_rules_unref(rules);
rules = rules_new;
}
- }
-
- if (sigchilds_waiting) {
- sigchilds_waiting = 0;
- reap_sigchilds();
- }
-
- if (run_exec_q) {
- run_exec_q = 0;
- if (!stop_exec_q)
- event_queue_manager(udev);
- }
-
- if (settle_pid > 0) {
- kill(settle_pid, SIGUSR1);
- settle_pid = 0;
+ reload_config = 0;
}
}
+
udev_queue_export_cleanup(udev_queue_export);
rc = 0;
exit:
-
udev_queue_export_unref(udev_queue_export);
udev_rules_unref(rules);
udev_ctrl_unref(udev_ctrl);
- if (inotify_fd >= 0)
- close(inotify_fd);
- udev_monitor_unref(kernel_monitor);
+ if (pfd[FD_SIGNAL].fd >= 0)
+ close(pfd[FD_SIGNAL].fd);
+ if (worker_socket[READ_END] >= 0)
+ close(worker_socket[READ_END]);
+ if (worker_socket[WRITE_END] >= 0)
+ close(worker_socket[WRITE_END]);
+ udev_monitor_unref(monitor);
udev_selinux_exit(udev);
udev_unref(udev);
logging_close();
