[PATCH] Fix #734193: divisibility issues with striped mirrors

[PATCH] Fix #734193: divisibility issues with striped mirrors

[Petr Rockai]

Hi,

the divisibility requirement in the allocation code is needlessly (and
incorrectly) strict, accounting not only for stripe layout but also for
mirrors (which do not divide extents among themselves, though). The
attached patch should remedy that. Regression test included.

Petr

-------------- next part --------------
A non-text attachment was scrubbed...
Name: lvcreate-striped-mirror.diff
Type: text/x-diff
Size: 3241 bytes
Desc: not available
URL: <http://listman.redhat.com/archives/lvm-devel/attachments/20110915/a62743b3/attachment.bin>
-------------- next part --------------

-- 
id' Ash = Ash; id' Dust = Dust; id' _ = undefined

[PATCH] Fix #734193: divisibility issues with striped mirrors

[Milan Broz]

On 09/15/2011 03:37 PM, Petr Rockai wrote:
> the divisibility requirement in the allocation code is needlessly (and
> incorrectly) strict, accounting not only for stripe layout but also for
> mirrors (which do not divide extents among themselves, though). The
> attached patch should remedy that. Regression test included.

The former reason for this line was, I think it still applies
https://bugzilla.redhat.com/show_bug.cgi?id=707779

If so, ack.

Milan

[PATCH] Fix #734193: divisibility issues with striped mirrors

[Petr Rockai]

Milan Broz <mbroz@redhat.com> writes:

> On 09/15/2011 03:37 PM, Petr Rockai wrote:
>> the divisibility requirement in the allocation code is needlessly (and
>> incorrectly) strict, accounting not only for stripe layout but also for
>> mirrors (which do not divide extents among themselves, though). The
>> attached patch should remedy that. Regression test included.
>
> The former reason for this line was, I think it still applies
> https://bugzilla.redhat.com/show_bug.cgi?id=707779
>
> If so, ack.

Yes, it's required that extent count is divisible by stripe count: any
other setup would make little sense (since part of the LV would have a
different stripe count than the rest, which sounds like a really bad
idea, even if the allocator could cope with the request).

The 707779 case is accounted for by the last test in t-lvconvert-mirror,
which I think you added. Since it still passes after this patch, I am
going ahead and checking it in.

Yours,
   Petr

-- 
id' Ash = Ash; id' Dust = Dust; id' _ = undefined