Stable backport of cifs nlink workaround?

Stable backport of cifs nlink workaround?

[Sitsofe Wheeler]

Hi,

Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
([CIFS] Provide sane values for nlink) could be backported to the stable
3.2 kernel?

The reason I ask is because if you Using a Debain Wheezy box with a
3.2.41-2+deb7u2 kernel the following is problematic:

Mount a share using cifs from a Windows 2008 machine on to 2008share
Use Samba 3.6 to export the share as //linux/2008share
Using a Windows 2012 machine browse to \\linux\2008share

because the directories will appear to be files to the Windows 2012
machine.

On the Debian box doing
stat 2008share/dir
says
Device: 18h/24d	Inode: 844424932383132  Links: 1

Using a Fedora 20 box with a 3.13.3-201.fc20.x86_64 kernel to do the
same says:
Device: 27h/39d	Inode: 844424932383132  Links: 2

In https://bugzilla.kernel.org/show_bug.cgi?id=52791#c4 Jeff mentions
that if there were demand a RHEL 6 customer could ask for a backport so
it looks like the change is isolated enough to be put into stable...

https://bugzilla.samba.org/show_bug.cgi?id=9346 also talks about the
issue.

-- 
Sitsofe | http://sucs.org/~sits/

Re: Stable backport of cifs nlink workaround?

[Steve French]

The patch is a little larger than usual for stable but for those where
the backport is trivial, seems reasonable candidate for stable.

On Tue, Feb 25, 2014 at 9:24 AM, Sitsofe Wheeler <sitsofe@yahoo.com> wrote:
> Hi,
>
> Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
> ([CIFS] Provide sane values for nlink) could be backported to the stable
> 3.2 kernel?
>
> The reason I ask is because if you Using a Debain Wheezy box with a
> 3.2.41-2+deb7u2 kernel the following is problematic:
>
> Mount a share using cifs from a Windows 2008 machine on to 2008share
> Use Samba 3.6 to export the share as //linux/2008share
> Using a Windows 2012 machine browse to \\linux\2008share
>
> because the directories will appear to be files to the Windows 2012
> machine.
>
> On the Debian box doing
> stat 2008share/dir
> says
> Device: 18h/24d Inode: 844424932383132  Links: 1
>
> Using a Fedora 20 box with a 3.13.3-201.fc20.x86_64 kernel to do the
> same says:
> Device: 27h/39d Inode: 844424932383132  Links: 2
>
> In https://bugzilla.kernel.org/show_bug.cgi?id=52791#c4 Jeff mentions
> that if there were demand a RHEL 6 customer could ask for a backport so
> it looks like the change is isolated enough to be put into stable...
>
> https://bugzilla.samba.org/show_bug.cgi?id=9346 also talks about the
> issue.
>
> --
> Sitsofe | http://sucs.org/~sits/



-- 
Thanks,

Steve

Re: Stable backport of cifs nlink workaround?

[Greg Kroah-Hartman]

On Tue, Feb 25, 2014 at 03:24:43PM +0000, Sitsofe Wheeler wrote:
> Hi,
> 
> Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
> ([CIFS] Provide sane values for nlink) could be backported to the stable
> 3.2 kernel?

Why just 3.2?  What's wrong with all of the other kernels before 3.12
that do not have this patch in it?

thanks,

greg k-h

Re: Stable backport of cifs nlink workaround?

[Jeff Layton]

On Tue, 25 Feb 2014 08:11:56 -0800
Greg Kroah-Hartman <gregkh@linuxfoundation.org> wrote:

> On Tue, Feb 25, 2014 at 03:24:43PM +0000, Sitsofe Wheeler wrote:
> > Hi,
> > 
> > Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
> > ([CIFS] Provide sane values for nlink) could be backported to the stable
> > 3.2 kernel?
> 
> Why just 3.2?  What's wrong with all of the other kernels before 3.12
> that do not have this patch in it?
> 
> thanks,
> 
> greg k-h

Yeah, you'd probably need to apply it to every stable kernel > 3.2 or
you'd technically have a regression. I've no real objection to putting
that fix in, but backporting to older kernels may not be trivial. There
have been other changes in this area over the years...

-- 
Jeff Layton <jlayton@redhat.com>

Re: Stable backport of cifs nlink workaround?

[Sitsofe Wheeler]

On Tue, Feb 25, 2014 at 09:43:06AM -0800, Jeff Layton wrote:
> On Tue, 25 Feb 2014 08:11:56 -0800
> Greg Kroah-Hartman <gregkh@linuxfoundation.org> wrote:
> 
> > On Tue, Feb 25, 2014 at 03:24:43PM +0000, Sitsofe Wheeler wrote:
> > > 
> > > Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
> > > ([CIFS] Provide sane values for nlink) could be backported to the stable
> > > 3.2 kernel?
> > 
> > Why just 3.2?  What's wrong with all of the other kernels before 3.12
> > that do not have this patch in it?
> 
> Yeah, you'd probably need to apply it to every stable kernel > 3.2 or
> you'd technically have a regression. I've no real objection to putting
> that fix in, but backporting to older kernels may not be trivial. There
> have been other changes in this area over the years...

There's no reason why it should only be 3.2 only - the only reason I
picked that one is because it is the current Debian stable kernel. 

-- 
Sitsofe | http://sucs.org/~sits/

Re: Stable backport of cifs nlink workaround?

[Greg Kroah-Hartman]

On Tue, Feb 25, 2014 at 09:43:06AM -0800, Jeff Layton wrote:
> On Tue, 25 Feb 2014 08:11:56 -0800
> Greg Kroah-Hartman <gregkh@linuxfoundation.org> wrote:
> 
> > On Tue, Feb 25, 2014 at 03:24:43PM +0000, Sitsofe Wheeler wrote:
> > > Hi,
> > > 
> > > Is there any chance that 74d290da476f672ad756634d12aa707375d3564d
> > > ([CIFS] Provide sane values for nlink) could be backported to the stable
> > > 3.2 kernel?
> > 
> > Why just 3.2?  What's wrong with all of the other kernels before 3.12
> > that do not have this patch in it?
> > 
> > thanks,
> > 
> > greg k-h
> 
> Yeah, you'd probably need to apply it to every stable kernel > 3.2 or
> you'd technically have a regression. I've no real objection to putting
> that fix in, but backporting to older kernels may not be trivial. There
> have been other changes in this area over the years...

It doesn't apply to 3.10 or 3.4-stable trees, so I'll wait for someone
to provide a working backport before I can commit it.

thanks,

greg k-h