[Cluster-devel] [GFS2 PATCH] gfs2-utils: Fix hang on withdraw

[Cluster-devel] [GFS2 PATCH] gfs2-utils: Fix hang on withdraw

[Abhijith Das]

Issuing a withdraw on a gfs2 filesystem causes a hang. When
gfs_controld was removed, the userspace functionality that
completes a withdraw operation went away. This causes gfs2
kernel to hang waiting for a withdraw completion ack from
userspace.

This patchset introduces a uevent-based shell script to do
the job that gfs_controld used to do on withdraw. An 'offline'
uevent triggers the execution of this script. This script
suspends the device associated with the filesystem and signals
a completed withdraw to the kernel.

Resolves: rhbz#1225634
Signed-off-by: Abhi Das <adas@redhat.com>
---
 gfs2/scripts/82-gfs2-withdraw.rules |  2 ++
 gfs2/scripts/gfs2_wd_udev.sh        | 30 ++++++++++++++++++++++++++++++
 2 files changed, 32 insertions(+)
 create mode 100644 gfs2/scripts/82-gfs2-withdraw.rules
 create mode 100755 gfs2/scripts/gfs2_wd_udev.sh

diff --git a/gfs2/scripts/82-gfs2-withdraw.rules b/gfs2/scripts/82-gfs2-withdraw.rules
new file mode 100644
index 0000000..c1ab02b
--- /dev/null
+++ b/gfs2/scripts/82-gfs2-withdraw.rules
@@ -0,0 +1,2 @@
+SUBSYSTEM=="gfs2", ACTION=="offline", RUN+="/bin/sh /sbin/gfs2_wd_udev.sh"
+
diff --git a/gfs2/scripts/gfs2_wd_udev.sh b/gfs2/scripts/gfs2_wd_udev.sh
new file mode 100755
index 0000000..ac3ce35
--- /dev/null
+++ b/gfs2/scripts/gfs2_wd_udev.sh
@@ -0,0 +1,30 @@
+#!/bin/sh
+#
+# Do not run this script manually. This script is called by udev on a gfs2
+# withdraw uevent and is used to complete the withdraw action and notify the
+# kernel.
+#
+
+# Sanity checks
+if [ "$SUBSYSTEM" != "gfs2" ] || [ "$LOCKPROTO" != "lock_dlm" ] ||
+       [ -z "$DEVPATH" ] || [ "$ACTION" != "offline" ]
+then
+    exit 1 # Nothing to do here
+fi
+
+# Try and suspend the device
+SYSFS_TOPDIR="/sys"$DEVPATH
+DM_NAME=$(cat "$SYSFS_TOPDIR/device/dm/name")
+DM_DEV="/dev/mapper/"$DM_NAME
+
+if [ -z "$DM_DEV" ]
+then
+    /usr/bin/dmsetup suspend $DM_DEV
+fi
+
+# Signal completion of withdraw
+WD_ACK="$SYSFS_TOPDIR/lock_module/withdraw"
+if [ -f "$WD_ACK" ]
+then
+    echo "1" > $WD_ACK
+fi
-- 
2.4.3

[Cluster-devel] [GFS2 PATCH] gfs2-utils: Fix hang on withdraw

[Andrew Price]

Hi Abhi,

On 03/08/15 20:30, Abhijith Das wrote:
> Issuing a withdraw on a gfs2 filesystem causes a hang. When
> gfs_controld was removed, the userspace functionality that
> completes a withdraw operation went away. This causes gfs2
> kernel to hang waiting for a withdraw completion ack from
> userspace.
>
> This patchset introduces a uevent-based shell script to do
> the job that gfs_controld used to do on withdraw. An 'offline'
> uevent triggers the execution of this script. This script
> suspends the device associated with the filesystem and signals
> a completed withdraw to the kernel.
>
> Resolves: rhbz#1225634
> Signed-off-by: Abhi Das <adas@redhat.com>
> ---
>   gfs2/scripts/82-gfs2-withdraw.rules |  2 ++
>   gfs2/scripts/gfs2_wd_udev.sh        | 30 ++++++++++++++++++++++++++++++
>   2 files changed, 32 insertions(+)
>   create mode 100644 gfs2/scripts/82-gfs2-withdraw.rules
>   create mode 100755 gfs2/scripts/gfs2_wd_udev.sh

Even if they're not getting installed in 'make install' the build system 
will need to know about them, in gfs2/scripts/Makefile.am:

noinst_SCRIPTS = \
	82-gfs2-withdraw.rules \
	gfs2_wd_udev.sh

I think some documentation in README.build would be useful too, so that 
package maintainers know why and where to install them.

> diff --git a/gfs2/scripts/82-gfs2-withdraw.rules b/gfs2/scripts/82-gfs2-withdraw.rules
> new file mode 100644
> index 0000000..c1ab02b
> --- /dev/null
> +++ b/gfs2/scripts/82-gfs2-withdraw.rules
> @@ -0,0 +1,2 @@
> +SUBSYSTEM=="gfs2", ACTION=="offline", RUN+="/bin/sh /sbin/gfs2_wd_udev.sh"

Perhaps that should be /usr/sbin instead of /sbin, after the whole 
"UsrMove" thing.

Thanks,
Andy

> +
> diff --git a/gfs2/scripts/gfs2_wd_udev.sh b/gfs2/scripts/gfs2_wd_udev.sh
> new file mode 100755
> index 0000000..ac3ce35
> --- /dev/null
> +++ b/gfs2/scripts/gfs2_wd_udev.sh
> @@ -0,0 +1,30 @@
> +#!/bin/sh
> +#
> +# Do not run this script manually. This script is called by udev on a gfs2
> +# withdraw uevent and is used to complete the withdraw action and notify the
> +# kernel.
> +#
> +
> +# Sanity checks
> +if [ "$SUBSYSTEM" != "gfs2" ] || [ "$LOCKPROTO" != "lock_dlm" ] ||
> +       [ -z "$DEVPATH" ] || [ "$ACTION" != "offline" ]
> +then
> +    exit 1 # Nothing to do here
> +fi
> +
> +# Try and suspend the device
> +SYSFS_TOPDIR="/sys"$DEVPATH
> +DM_NAME=$(cat "$SYSFS_TOPDIR/device/dm/name")
> +DM_DEV="/dev/mapper/"$DM_NAME
> +
> +if [ -z "$DM_DEV" ]
> +then
> +    /usr/bin/dmsetup suspend $DM_DEV
> +fi
> +
> +# Signal completion of withdraw
> +WD_ACK="$SYSFS_TOPDIR/lock_module/withdraw"
> +if [ -f "$WD_ACK" ]
> +then
> +    echo "1" > $WD_ACK
> +fi
>

[Cluster-devel] [GFS2 PATCH] gfs2-utils: Fix hang on withdraw

[Abhijith Das]

----- Original Message -----
> From: "Andrew Price" <anprice@redhat.com>
> To: "Abhijith Das" <adas@redhat.com>, "cluster-devel" <cluster-devel@redhat.com>
> Sent: Tuesday, August 4, 2015 6:06:16 AM
> Subject: Re: [Cluster-devel] [GFS2 PATCH] gfs2-utils: Fix hang on withdraw
> 
> Hi Abhi,
> 
> On 03/08/15 20:30, Abhijith Das wrote:
> > Issuing a withdraw on a gfs2 filesystem causes a hang. When
> > gfs_controld was removed, the userspace functionality that
> > completes a withdraw operation went away. This causes gfs2
> > kernel to hang waiting for a withdraw completion ack from
> > userspace.
> >
> > This patchset introduces a uevent-based shell script to do
> > the job that gfs_controld used to do on withdraw. An 'offline'
> > uevent triggers the execution of this script. This script
> > suspends the device associated with the filesystem and signals
> > a completed withdraw to the kernel.
> >
> > Resolves: rhbz#1225634
> > Signed-off-by: Abhi Das <adas@redhat.com>
> > ---
> >   gfs2/scripts/82-gfs2-withdraw.rules |  2 ++
> >   gfs2/scripts/gfs2_wd_udev.sh        | 30 ++++++++++++++++++++++++++++++
> >   2 files changed, 32 insertions(+)
> >   create mode 100644 gfs2/scripts/82-gfs2-withdraw.rules
> >   create mode 100755 gfs2/scripts/gfs2_wd_udev.sh
> 
> Even if they're not getting installed in 'make install' the build system
> will need to know about them, in gfs2/scripts/Makefile.am:
> 
> noinst_SCRIPTS = \
> 	82-gfs2-withdraw.rules \
> 	gfs2_wd_udev.sh
> 
> I think some documentation in README.build would be useful too, so that
> package maintainers know why and where to install them.
> 
> > diff --git a/gfs2/scripts/82-gfs2-withdraw.rules
> > b/gfs2/scripts/82-gfs2-withdraw.rules
> > new file mode 100644
> > index 0000000..c1ab02b
> > --- /dev/null
> > +++ b/gfs2/scripts/82-gfs2-withdraw.rules
> > @@ -0,0 +1,2 @@
> > +SUBSYSTEM=="gfs2", ACTION=="offline", RUN+="/bin/sh /sbin/gfs2_wd_udev.sh"
> 
> Perhaps that should be /usr/sbin instead of /sbin, after the whole
> "UsrMove" thing.
> 

Ok, Thanks! I'll make these changes and re-post.

Cheers!
--Abhi