* [PATCH] get_netlink_msg: unable to receive kernel netlink message:
@ 2007-01-30 9:16 Matthias Schwarzott
2007-01-31 9:24 ` [PATCH] get_netlink_msg: unable to receive kernel netlink Matthias Schwarzott
` (4 more replies)
0 siblings, 5 replies; 6+ messages in thread
From: Matthias Schwarzott @ 2007-01-30 9:16 UTC (permalink / raw)
To: linux-hotplug
[-- Attachment #1: Type: text/plain, Size: 1528 bytes --]
Hi there!
Had this Bug-report floating around since some time:
http://bugs.gentoo.org/show_bug.cgi?id=151414
It occurs, when root-partition has no /dev/console, meaning that kernel could
not open it, and such udevd is started without open filedescriptors 0 1 2.
In that case udevd openes its sockets (netlink and control).
They get fds between 0 and 2.
Later duping /dev/null to 0 1 2 closes the sockets and replaces them
with /dev/null.
Communicating over socket in main-look then gets that error-messages.
One possible solution is:
Checking at startup of udevd that fd 0 1 2 are open, and if not, connect them
to /dev/null.
A. For example by explictely checking each fd and dup /dev/null (attached
patch).
B. By opening /dev/null until open returns 2 meaning fds 0 1 2 are open.
C. Remembering which fd is open, and later only dup/close the ones that were
open.
But in case C one perhaps has to conditionalize each printf(...) and
fprintf(2,...) as the sockets are sure not happy about text-messages.
For gentoo-startup, that does work without /dev/console, only udevd has
problems there, we solved it for now this way in udev-startup script (until
udevd copes with this itself):
# check if /dev/console exists outside tmpfs
[[ -c /dev/console ]] ; local need_redirect=$?
....
mount /dev
....
mknod /dev/null
mknod /dev/console
...
if [[ ${need_redirect} == 1 ]]; then
/sbin/udevd --daemon </dev/console >/dev/console 2>/dev/console
else
/sbin/udevd --daemon
fi
Matthias
--
Matthias Schwarzott (zzam)
[-- Attachment #2: udev-104-std_fd_dup_fix.patch --]
[-- Type: text/x-diff, Size: 596 bytes --]
--- udev-104-orig/udevd.c 2007-01-15 14:56:11.000000000 +0100
+++ udev-104/udevd.c 2007-01-29 13:47:56.000000000 +0100
@@ -948,6 +948,21 @@
int rc = 1;
int maxfd;
+ fd = open("/dev/null", O_RDWR);
+
+ if (fd < 0) {
+ fprintf(stderr, "cannot open /dev/null\n");
+ err("cannot open /dev/null");
+ }
+ if (read(STDIN_FILENO, 0, 0) < 0)
+ dup2(fd, STDIN_FILENO);
+ if (write(STDOUT_FILENO, 0, 0) < 0)
+ dup2(fd, STDOUT_FILENO);
+ if (write(STDERR_FILENO, 0, 0) < 0)
+ dup2(fd, STDERR_FILENO);
+ if (fd > 2)
+ close(fd);
+
logging_init("udevd");
udev_config_init();
selinux_init();
[-- Attachment #3: Type: text/plain, Size: 347 bytes --]
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
[-- Attachment #4: Type: text/plain, Size: 226 bytes --]
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] get_netlink_msg: unable to receive kernel netlink
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
@ 2007-01-31 9:24 ` Matthias Schwarzott
2007-03-09 14:09 ` Matthias Schwarzott
` (3 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Matthias Schwarzott @ 2007-01-31 9:24 UTC (permalink / raw)
To: linux-hotplug
On Dienstag, 30. Januar 2007, Matthias Schwarzott wrote:
> Hi there!
>
> Had this Bug-report floating around since some time:
> http://bugs.gentoo.org/show_bug.cgi?id\x151414
>
> It occurs, when root-partition has no /dev/console, meaning that kernel
> could not open it, and such udevd is started without open filedescriptors 0
> 1 2.
>
The error condition can also be reproduced by starting udevd with this
command-line:
killall udevd
udevd --daemon <&- >&- 2>&-
This will fill syslog with the message from subject.
Matthias
--
Matthias Schwarzott (zzam)
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys - and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] get_netlink_msg: unable to receive kernel netlink
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
2007-01-31 9:24 ` [PATCH] get_netlink_msg: unable to receive kernel netlink Matthias Schwarzott
@ 2007-03-09 14:09 ` Matthias Schwarzott
2007-03-10 21:15 ` Kay Sievers
` (2 subsequent siblings)
4 siblings, 0 replies; 6+ messages in thread
From: Matthias Schwarzott @ 2007-03-09 14:09 UTC (permalink / raw)
To: linux-hotplug
On Mittwoch, 31. Januar 2007, Matthias Schwarzott wrote:
> >
> > Had this Bug-report floating around since some time:
> > http://bugs.gentoo.org/show_bug.cgi?id\x151414
> >
> > It occurs, when root-partition has no /dev/console, meaning that kernel
> > could not open it, and such udevd is started without open filedescriptors
> > 0 1 2.
>
> The error condition can also be reproduced by starting udevd with this
> command-line:
>
> killall udevd
> udevd --daemon <&- >&- 2>&-
>
> This will fill syslog with the message from subject.
This is just to bring this issue back to mind.
Does nobody cosider this bug worth fixing it, as I consider it ugly to have to
explicitly redirect stdin, stdout and stderr just for executing a daemon that
already has the goal to close these fds.
Greetings
Matthias
--
Matthias Schwarzott (zzam)
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] get_netlink_msg: unable to receive kernel netlink
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
2007-01-31 9:24 ` [PATCH] get_netlink_msg: unable to receive kernel netlink Matthias Schwarzott
2007-03-09 14:09 ` Matthias Schwarzott
@ 2007-03-10 21:15 ` Kay Sievers
2007-03-13 19:29 ` Matthias Schwarzott
2007-03-13 21:12 ` Kay Sievers
4 siblings, 0 replies; 6+ messages in thread
From: Kay Sievers @ 2007-03-10 21:15 UTC (permalink / raw)
To: linux-hotplug
On 3/9/07, Matthias Schwarzott <zzam@gentoo.org> wrote:
> On Mittwoch, 31. Januar 2007, Matthias Schwarzott wrote:
> > >
> > > Had this Bug-report floating around since some time:
> > > http://bugs.gentoo.org/show_bug.cgi?id\x151414
> > >
> > > It occurs, when root-partition has no /dev/console, meaning that kernel
> > > could not open it, and such udevd is started without open filedescriptors
> > > 0 1 2.
> >
> > The error condition can also be reproduced by starting udevd with this
> > command-line:
> >
>
> > killall udevd
> > udevd --daemon <&- >&- 2>&-
> >
> > This will fill syslog with the message from subject.
>
> This is just to bring this issue back to mind.
>
> Does nobody cosider this bug worth fixing it, as I consider it ugly to have to
> explicitly redirect stdin, stdout and stderr just for executing a daemon that
> already has the goal to close these fds.
Aren't you expected to connect the init script to the (dead) console anyway?
exec 0<> $CONSOLE 1>&0 2>&0
Kay
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] get_netlink_msg: unable to receive kernel netlink
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
` (2 preceding siblings ...)
2007-03-10 21:15 ` Kay Sievers
@ 2007-03-13 19:29 ` Matthias Schwarzott
2007-03-13 21:12 ` Kay Sievers
4 siblings, 0 replies; 6+ messages in thread
From: Matthias Schwarzott @ 2007-03-13 19:29 UTC (permalink / raw)
To: linux-hotplug
On Samstag, 10. März 2007, Kay Sievers wrote:
> On 3/9/07, Matthias Schwarzott <zzam@gentoo.org> wrote:
> >
> > Does nobody cosider this bug worth fixing it, as I consider it ugly to
> > have to explicitly redirect stdin, stdout and stderr just for executing a
> > daemon that already has the goal to close these fds.
>
> Aren't you expected to connect the init script to the (dead) console
> anyway? exec 0<> $CONSOLE 1>&0 2>&0
Normally the kernel connects /dev/console and the init-process.
But only if /dev/console (or $CONSOLE) exists on the root-partition.
If it does not fds 0, 1 and 2 are closed.
In that case I have to add
# exec 0<> $CONSOLE 1>&0 2>&
to every init-script?
And for that even create $CONSOLE before executing udev, as that does not
exist in empty tmpfs.
For all other init-scripts/daemons it works without, but not for udev.
If this bug is solved we still can boot from a partition with completely
destroyed/cleaned /dev, without crashing bootup at udevd startup.
Matthias
--
Matthias Schwarzott (zzam)
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CIDÞVDEV
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] get_netlink_msg: unable to receive kernel netlink
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
` (3 preceding siblings ...)
2007-03-13 19:29 ` Matthias Schwarzott
@ 2007-03-13 21:12 ` Kay Sievers
4 siblings, 0 replies; 6+ messages in thread
From: Kay Sievers @ 2007-03-13 21:12 UTC (permalink / raw)
To: linux-hotplug
[-- Attachment #1: Type: text/plain, Size: 1587 bytes --]
On Tue, 2007-03-13 at 20:29 +0100, Matthias Schwarzott wrote:
> On Samstag, 10. März 2007, Kay Sievers wrote:
> > On 3/9/07, Matthias Schwarzott <zzam@gentoo.org> wrote:
> > >
> > > Does nobody cosider this bug worth fixing it, as I consider it ugly to
> > > have to explicitly redirect stdin, stdout and stderr just for executing a
> > > daemon that already has the goal to close these fds.
> >
> > Aren't you expected to connect the init script to the (dead) console
> > anyway? exec 0<> $CONSOLE 1>&0 2>&0
>
> Normally the kernel connects /dev/console and the init-process.
> But only if /dev/console (or $CONSOLE) exists on the root-partition.
> If it does not fds 0, 1 and 2 are closed.
> In that case I have to add
> # exec 0<> $CONSOLE 1>&0 2>&
> to every init-script?
Usually only in the single-user script forked by init. But sure, you
would need the device-node, to do that.
> And for that even create $CONSOLE before executing udev, as that does not
> exist in empty tmpfs.
Yes, but you are expected to have created /dev/null and some other stuff
too, in the empty tmpfs, before you start udevd.
If you don't do the 'exec', in case of a missing /dev/console, your
terminal is not connected to the init script, which maybe nice to do for
various other reasons.
> For all other init-scripts/daemons it works without, but not for udev.
> If this bug is solved we still can boot from a partition with completely
> destroyed/cleaned /dev, without crashing bootup at udevd startup.
Fine. Does the attached work for you?
Thanks,
Kay
[-- Attachment #2: 012fd.patch --]
[-- Type: text/x-patch, Size: 1256 bytes --]
diff --git a/udevd.c b/udevd.c
index 23f5fd6..961ceb5 100644
--- a/udevd.c
+++ b/udevd.c
@@ -989,6 +989,19 @@ int main(int argc, char *argv[], char *e
goto exit;
}
+ /* make sure std{in,out,err} fd's are in a sane state */
+ fd = open("/dev/null", O_RDWR);
+ if (fd < 0) {
+ fprintf(stderr, "cannot open /dev/null\n");
+ err("cannot open /dev/null");
+ }
+ 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)
+ dup2(fd, STDERR_FILENO);
+
/* init sockets to receive events */
if (init_udevd_socket() < 0) {
if (errno == EADDRINUSE) {
@@ -1064,17 +1077,12 @@ int main(int argc, char *argv[], char *e
}
}
- /* redirect std fd's */
- fd = open("/dev/null", O_RDWR);
- if (fd >= 0) {
- dup2(fd, STDIN_FILENO);
- if (!verbose)
- dup2(fd, STDOUT_FILENO);
- dup2(fd, STDERR_FILENO);
- if (fd > STDERR_FILENO)
- close(fd);
- } else
- err("error opening /dev/null: %s", strerror(errno));
+ /* redirect std{out,err} fd's */
+ if (!verbose)
+ dup2(fd, STDOUT_FILENO);
+ dup2(fd, STDERR_FILENO);
+ if (fd > STDERR_FILENO)
+ close(fd);
/* set scheduling priority for the daemon */
setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
[-- Attachment #3: Type: text/plain, Size: 345 bytes --]
-------------------------------------------------------------------------
Take Surveys. Earn Cash. Influence the Future of IT
Join SourceForge.net's Techsay panel and you'll get the chance to share your
opinions on IT & business topics through brief surveys-and earn cash
http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV
[-- Attachment #4: Type: text/plain, Size: 226 bytes --]
_______________________________________________
Linux-hotplug-devel mailing list http://linux-hotplug.sourceforge.net
Linux-hotplug-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/linux-hotplug-devel
^ permalink raw reply related [flat|nested] 6+ messages in thread
end of thread, other threads:[~2007-03-13 21:12 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2007-01-30 9:16 [PATCH] get_netlink_msg: unable to receive kernel netlink message: Matthias Schwarzott
2007-01-31 9:24 ` [PATCH] get_netlink_msg: unable to receive kernel netlink Matthias Schwarzott
2007-03-09 14:09 ` Matthias Schwarzott
2007-03-10 21:15 ` Kay Sievers
2007-03-13 19:29 ` Matthias Schwarzott
2007-03-13 21:12 ` Kay Sievers
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).