[PATCH] udev/mount.sh: try to kill active processes before umount

[PATCH] udev/mount.sh: try to kill active processes before umount

[Steffen Sledz]

* umount will fail if there are processes accessing files at the
  device. Therefor try to kill these processes using fuser if
  available.

Signed-off-by: Steffen Sledz <sledz@dresearch.de>
---
 recipes/udev/files/mount.sh        |   21 ++++++++++++++++++++-
 recipes/udev/files/slugos/mount.sh |   21 ++++++++++++++++++++-
 2 files changed, 40 insertions(+), 2 deletions(-)

diff --git a/recipes/udev/files/mount.sh b/recipes/udev/files/mount.sh
index be8b3df..79c6891 100644
--- a/recipes/udev/files/mount.sh
+++ b/recipes/udev/files/mount.sh
@@ -8,6 +8,7 @@
 MOUNT="/bin/mount"
 PMOUNT="/usr/bin/pmount"
 UMOUNT="/bin/umount"
+FUSER="/usr/bin/fuser"
 name="`basename "$DEVNAME"`"
 
 for line in `cat /etc/udev/mount.blacklist | grep -v ^#`
@@ -64,7 +65,25 @@ fi
 if [ "$ACTION" = "remove" ] && [ -x "$UMOUNT" ] && [ -n "$DEVNAME" ]; then
 	for mnt in `cat /proc/mounts | grep "$DEVNAME" | cut -f 2 -d " " `
 	do
-		$UMOUNT $mnt
+		if [ -x "$FUSER" ]; then
+			$FUSER -k -KILL -m $mnt
+			for try in `seq 10`
+			do
+				$FUSER -m $mnt || break
+				sleep 1
+				logger "mount.sh/automount" "$try secs waited for processes active at $mnt to finish on SIGKILL"
+			done
+
+			if $FUSER -m $mnt 
+			then
+				logger "mount.sh/automount" "Could not kill all processes using files at $mnt, try forced umount"
+				$UMOUNT -f $mnt
+			else
+				$UMOUNT $mnt
+			fi
+		else
+			$UMOUNT $mnt
+		fi
 	done
 	
 	# Remove empty directories from auto-mounter
diff --git a/recipes/udev/files/slugos/mount.sh b/recipes/udev/files/slugos/mount.sh
index 0990a7e..f67c3ff 100644
--- a/recipes/udev/files/slugos/mount.sh
+++ b/recipes/udev/files/slugos/mount.sh
@@ -8,6 +8,7 @@
 MOUNT="/bin/mount"
 PMOUNT="/usr/bin/pmount"
 UMOUNT="/bin/umount"
+FUSER="/usr/bin/fuser"
 name="`basename "$DEVNAME"`"
 
 if ( blkid "$DEVNAME" | grep -q 'TYPE="mdraid"' )
@@ -75,7 +76,25 @@ fi
 if [ "$ACTION" = "remove" ] && [ -x "$UMOUNT" ] && [ -n "$DEVNAME" ]; then
 	for mnt in `cat /proc/mounts | grep "$DEVNAME" | cut -f 2 -d " " `
 	do
-		$UMOUNT $mnt
+		if [ -x "$FUSER" ]; then
+			$FUSER -k -KILL -m $mnt
+			for try in `seq 10`
+			do
+				$FUSER -m $mnt || break
+				sleep 1
+				logger "mount.sh/automount" "$try secs waited for processes active at $mnt to finish on SIGKILL"
+			done
+
+			if $FUSER -m $mnt 
+			then
+				logger "mount.sh/automount" "Could not kill all processes using files at $mnt, try forced umount"
+				$UMOUNT -f $mnt
+			else
+				$UMOUNT $mnt
+			fi
+		else
+			$UMOUNT $mnt
+		fi
 	done
 	
 	# Remove empty directories from auto-mounter
-- 
1.6.4.2

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Wouldn't a umount -l be a lot better in this case?

On 22-06-10 12:08, Steffen Sledz wrote:
> * umount will fail if there are processes accessing files at the
>   device. Therefor try to kill these processes using fuser if
>   available.
> 
> Signed-off-by: Steffen Sledz <sledz@dresearch.de>
> ---
>  recipes/udev/files/mount.sh        |   21 ++++++++++++++++++++-
>  recipes/udev/files/slugos/mount.sh |   21 ++++++++++++++++++++-
>  2 files changed, 40 insertions(+), 2 deletions(-)
> 
> diff --git a/recipes/udev/files/mount.sh b/recipes/udev/files/mount.sh
> index be8b3df..79c6891 100644
> --- a/recipes/udev/files/mount.sh
> +++ b/recipes/udev/files/mount.sh
> @@ -8,6 +8,7 @@
>  MOUNT="/bin/mount"
>  PMOUNT="/usr/bin/pmount"
>  UMOUNT="/bin/umount"
> +FUSER="/usr/bin/fuser"
>  name="`basename "$DEVNAME"`"
>  
>  for line in `cat /etc/udev/mount.blacklist | grep -v ^#`
> @@ -64,7 +65,25 @@ fi
>  if [ "$ACTION" = "remove" ] && [ -x "$UMOUNT" ] && [ -n "$DEVNAME" ]; then
>  	for mnt in `cat /proc/mounts | grep "$DEVNAME" | cut -f 2 -d " " `
>  	do
> -		$UMOUNT $mnt
> +		if [ -x "$FUSER" ]; then
> +			$FUSER -k -KILL -m $mnt
> +			for try in `seq 10`
> +			do
> +				$FUSER -m $mnt || break
> +				sleep 1
> +				logger "mount.sh/automount" "$try secs waited for processes active at $mnt to finish on SIGKILL"
> +			done
> +
> +			if $FUSER -m $mnt 
> +			then
> +				logger "mount.sh/automount" "Could not kill all processes using files at $mnt, try forced umount"
> +				$UMOUNT -f $mnt
> +			else
> +				$UMOUNT $mnt
> +			fi
> +		else
> +			$UMOUNT $mnt
> +		fi
>  	done
>  	
>  	# Remove empty directories from auto-mounter
> diff --git a/recipes/udev/files/slugos/mount.sh b/recipes/udev/files/slugos/mount.sh
> index 0990a7e..f67c3ff 100644
> --- a/recipes/udev/files/slugos/mount.sh
> +++ b/recipes/udev/files/slugos/mount.sh
> @@ -8,6 +8,7 @@
>  MOUNT="/bin/mount"
>  PMOUNT="/usr/bin/pmount"
>  UMOUNT="/bin/umount"
> +FUSER="/usr/bin/fuser"
>  name="`basename "$DEVNAME"`"
>  
>  if ( blkid "$DEVNAME" | grep -q 'TYPE="mdraid"' )
> @@ -75,7 +76,25 @@ fi
>  if [ "$ACTION" = "remove" ] && [ -x "$UMOUNT" ] && [ -n "$DEVNAME" ]; then
>  	for mnt in `cat /proc/mounts | grep "$DEVNAME" | cut -f 2 -d " " `
>  	do
> -		$UMOUNT $mnt
> +		if [ -x "$FUSER" ]; then
> +			$FUSER -k -KILL -m $mnt
> +			for try in `seq 10`
> +			do
> +				$FUSER -m $mnt || break
> +				sleep 1
> +				logger "mount.sh/automount" "$try secs waited for processes active at $mnt to finish on SIGKILL"
> +			done
> +
> +			if $FUSER -m $mnt 
> +			then
> +				logger "mount.sh/automount" "Could not kill all processes using files at $mnt, try forced umount"
> +				$UMOUNT -f $mnt
> +			else
> +				$UMOUNT $mnt
> +			fi
> +		else
> +			$UMOUNT $mnt
> +		fi
>  	done
>  	
>  	# Remove empty directories from auto-mounter

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMIJhGMkyGM64RGpERAo+WAJsEfFfYOKZzeSChA8FBf6SmLSlEbgCdHP2z
d2kNPcFp4RiRHac8Zqh8Hb4=
=HT9O
-----END PGP SIGNATURE-----

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[Steffen Sledz]

Am 22.06.2010 13:02, schrieb Koen Kooi:
> Wouldn't a umount -l be a lot better in this case?

I'm not sure if this would do the right things.

mount.sh is called with REMOVE action when the device *is
disconnected* i.e. the filesystem is no longer available.
So i believe it is not possible to "cleanup all references
to the filesystem as soon as it is not busy anymore".

I think killing all the processes is the only way here.

Steffen

-- 
Steffen Sledz
DResearch Digital Media Systems GmbH
Otto-Schmirgal-Str.3, D-10319 Berlin, Germany
Tel: +49 (30) 515932237 mailto:sledz@DResearch.DE
Fax: +49 (30) 515932299 http://www.DResearch.DE
Geschäftsführer: Dr. Michael Weber, Werner Mögle;
Amtsgericht Berlin Charlottenburg; HRB 54412;
Ust.-IDNr. DE169013825; WEEE Reg.-Nr. DE 85995642

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[Koen Kooi]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 22-06-10 14:01, Steffen Sledz wrote:
> Am 22.06.2010 13:02, schrieb Koen Kooi:
>> Wouldn't a umount -l be a lot better in this case?
> 
> I'm not sure if this would do the right things.
> 
> mount.sh is called with REMOVE action when the device *is
> disconnected* i.e. the filesystem is no longer available.
> So i believe it is not possible to "cleanup all references
> to the filesystem as soon as it is not busy anymore".
> 
> I think killing all the processes is the only way here.

I don't like the idea of killing random processes, especially is that
process is smart enough to "move on" after a timeout.

regards,

Koen
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (Darwin)

iD8DBQFMIMiwMkyGM64RGpERAiFVAKCzGeqyIx8Mh+EHcXz8/w3uK20D0ACfXR++
x3OaWYcazQ9Wc1/cObARf2E=
=xOUQ
-----END PGP SIGNATURE-----

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[John Willis]

> >> Wouldn't a umount -l be a lot better in this case?
> >
> > I'm not sure if this would do the right things.
> >
> > mount.sh is called with REMOVE action when the device *is
> > disconnected* i.e. the filesystem is no longer available.
> > So i believe it is not possible to "cleanup all references to the
> > filesystem as soon as it is not busy anymore".
> >
> > I think killing all the processes is the only way here.
> 
> I don't like the idea of killing random processes, especially is that
process is
> smart enough to "move on" after a timeout.

I concur, the issue at hand is a little 'rock and a hard place' (what is the
'correct' way to deal with processes left over when the file system has most
definitely been pulled with prejudice ;)). Killing processes feels a little
wrong as a lot of apps can recover from filesystem pulls.

That said, we have been experimenting with umount -l on the OpenPandora with
some promising initial results. We are also playing with some other changes
that may also be of wider interest (mountpoints based on SD card labels if
they exist etc.). I'll keep this thread updated with the results of testing,
if it works out well I was going to suggest some of the changes for the
stock mount.sh.

There is little harm in a umount -l in the short term.

Regards,

John

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[Steffen Sledz]

Am 22.06.2010 21:11, schrieb John Willis:
> There is little harm in a umount -l in the short term.

OK, i've prepared a new patch for this.

Because i forgot the --in-reply-to it started a new thread. Sorry.

Steffen

Re: [PATCH] udev/mount.sh: try to kill active processes before umount

[Phil Blundell]

On Tue, 2010-06-22 at 14:01 +0200, Steffen Sledz wrote:
> mount.sh is called with REMOVE action when the device *is
> disconnected* i.e. the filesystem is no longer available.
> So i believe it is not possible to "cleanup all references
> to the filesystem as soon as it is not busy anymore".
> 
> I think killing all the processes is the only way here.

I don't think that's necessarily true.  Any process that is holding a
file open on the removed device will get an I/O error the next time it
tries to access that filesystem.  Obviously, whether or not it handles
that error correctly is open to question, but increasingly many apps do
get it right and it seems a bit sad to kill them all off just in case
one doesn't do the right thing.

In the worst case, an app that chronically ignores I/O errors will just
keep the file descriptor open forever (or until it exits normally).
That doesn't sound like such an awful outcome.

p.