Re: [Evms-devel] EVMS 2.5.4 snapshot on Linux 2.6.20

[Srini <v.v...@gm...>]

On further testing, we found that currently, snapshots are not allowed
to be deleted if the source volume is mounted on the machine and
resulted in an error "Device or Resource busy". Asking the users to
unmount the source v volume just to delete the snapshot would be a
rather cumbersome behavior. This again was observed only on
post-2.6.16 kernel versions and not till kernel 2.6.16. The following
patch that has been applied on the 2.6.17 kernel version seems to be
the reasoning behind this:

http://lkml.org/lkml/2006/6/21/335

Once we were able to reverse this patch completely, we were able to
delete the snapshot without necessarily unmounting the source volume.

Not sure of whether this particular situation is being used by any
specific EVMS user but just wanted to sound off an alarm bell in case
someone is hitting this error....

Cheers

srini

On 2/12/07, Srini <v.v...@gm...> wrote:
> Hi all,
>
> We have resolved the issue and just thought of documenting the reason
> here for folks who hit the same issue in the future....
>
> It looks like that post-Linux 2.6.16 kernel, there was a patch that
> was applied in the kernel device mapper code to remove the chunk size
> argument when activating the snapshot - the reasoning behind that
> after creating the snapshot the chunk size cannot be changed. This
> patch is available at :
> http://kernel.org/pub/linux/kernel/people/agk/patches/2.6/2.6.19/dm-snapshot-allow-zero-chunk_size.patch
>
> While EVMS still seems to pass the chunk size while activating the
> snapshot, device mapper refuses to accept it owing to the above patch
> being applied on all kernels starting from 2.6.17.
>
> On revoking the patch on our kernel tree, the snapshots have started
> working properly. Am not sure of whether there has to be a patch on
> EVMS side to not to send the chunk size for activating the snapshot.
>
> Hope it helps,
>
> Srini
>
>
> On 2/6/07, V VS <v.v...@gm...> wrote:
> > Hi all,
> >
> > We are  using EVMS 2.5.4 version on Linux kernel version 2.6.16 along. It
> > was working fine. For some of the other performance reasons, we had to move
> > to Linux 2.6.20 version - specifically we are currently on 2.6.20-rc6
> > version.
> >
> > We are observing that the snapshots are not getting created successfully in
> > this kernel version. Have anyone successfully created a snapshot in this
> > kernel version yet? We also tried to backtrack the kernel version to
> > 2.6.19.1 but in vain - snapshots are not getting created but the rest of the
> > EVMS is working fine.
> >
> > When we create the snapshots, we observe that the snapshot object is getting
> > successfully created (when we do a query:output) but the corresponding node
> > is not getting created in (/dev/evms/.nodes) that we want to mount for
> > accessing the snapshot contents.
> >
> > We tried to do the following couple of items after going through *somewhat
> > related issues* reported in the EVMS mail archives but we did not observe
> > any change in the behavior:
> >
> > a) Going by (
> > http://marc.theaimsgroup.com/?l=evms-devel&m=112699668724855&w=2)
> > as suggested by Kevin, we modified the drivers in the kernel source tree
> > drivers/dm/dm-snap.c and changed the value of SNAPSHOT_PAGES to be that of
> > 64 instead of the currently 256.
> > b) Going by
> > (http://marc.theaimsgroup.com/?l=evms-devel&m=114536935000036&w=2
> > ), we removed the O_DIRECT flag in plugins/disk/localdskmgr.c while opening
> > the node.
> >
> > Both these changes in the 2.5.4 code base refused to make any changes in the
> > observed behavior. This is what we attempted to do for creating the
> > snapshots. We have also attached a detailed ('debug' log) for the EVMS
> > commands that we used.
> >
> > We would really appreciate if someone can give us a few thoughts on what we
> > are missing to do as we have been raking our brains on this for quite some
> > time now....
> >
>
> ... snipped ...
>