per-file NOCOW in 3.7

per-file NOCOW in 3.7

[Marios Titas]

According to the tentative changelog in the wiki, 3.7 will properly
support per-file NOCOW. I think this is misleading. In particular, it
still does not work correctly in the case when the NOCOW flag of a new
flag is being inherited from the directory. I think it would be better
to remove that item from the changelog until the the fix [1] is
merged.

[1] https://patchwork.kernel.org/patch/1438431/

Re: per-file NOCOW in 3.7

[David Sterba]

On Wed, Nov 21, 2012 at 10:48:08AM -0500, Marios Titas wrote:
> According to the tentative changelog in the wiki, 3.7 will properly
> support per-file NOCOW. I think this is misleading. In particular, it
> still does not work correctly in the case when the NOCOW flag of a new
> flag is being inherited from the directory. I think it would be better
> to remove that item from the changelog until the the fix [1] is
> merged.
> [1] https://patchwork.kernel.org/patch/1438431/

You're right, the patch is not in 3.7-rc nor in btrfs-next, and fixes a
real bug.

Technically it is a regresssion within the 3.7 cycle (so it could even
go to a late -rc), in any case it's a small change that could also go
through the stable tree.

david

Re: per-file NOCOW in 3.7

[Liu Bo]

On Wed, Nov 21, 2012 at 06:25:11PM +0100, David Sterba wrote:
> On Wed, Nov 21, 2012 at 10:48:08AM -0500, Marios Titas wrote:
> > According to the tentative changelog in the wiki, 3.7 will properly
> > support per-file NOCOW. I think this is misleading. In particular, it
> > still does not work correctly in the case when the NOCOW flag of a new
> > flag is being inherited from the directory. I think it would be better
> > to remove that item from the changelog until the the fix [1] is
> > merged.
> > [1] https://patchwork.kernel.org/patch/1438431/
> 
> You're right, the patch is not in 3.7-rc nor in btrfs-next, and fixes a
> real bug.
> 
> Technically it is a regresssion within the 3.7 cycle (so it could even
> go to a late -rc), in any case it's a small change that could also go
> through the stable tree.
> 
> david

Well, I'll resend it for merge with Marios's reported-by and your Reviewed-by.

Don't worry, it's a good fix though.

thanks,
liubo

Re: per-file NOCOW in 3.7

[Sergei Trofimovich]

[-- Attachment #1: Type: text/plain, Size: 1089 bytes --]

> > > According to the tentative changelog in the wiki, 3.7 will properly
> > > support per-file NOCOW. I think this is misleading. In particular, it
> > > still does not work correctly in the case when the NOCOW flag of a new
> > > flag is being inherited from the directory. I think it would be better
> > > to remove that item from the changelog until the the fix [1] is
> > > merged.
> > > [1] https://patchwork.kernel.org/patch/1438431/
> > 
> > You're right, the patch is not in 3.7-rc nor in btrfs-next, and fixes a
> > real bug.
> > 
> > Technically it is a regresssion within the 3.7 cycle (so it could even
> > go to a late -rc), in any case it's a small change that could also go
> > through the stable tree.
> > 
> > david
> 
> Well, I'll resend it for merge with Marios's reported-by and your Reviewed-by.
> 
> Don't worry, it's a good fix though.

I'm using your patch on 2 boxes for 2 months.
It is the only way to get decent performance
on my workload.

Thanks for the patch!

Tested-by: Sergei Trofimovich <slyich@gmail.com>

-- 

  Sergei

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]