EA regression in 3.2?

[Jeremy Allison]

On Fri, Jan 09, 2009 at 04:19:50PM +0100, Corinna Vinschen wrote:
> Hi,
> 
> while testing something in Cygwin on a Samba 3.2.4 share, I suddenly
> got a strange error I never encountered when testing on Samba 3.0.x
> shares.
> 
> Consider a share //server/share.  On the Windows side I try to get
> a handle to the share and immediately ask for read access to the Extended
> Attributes, if available:
> 
>   status = NtCreateFile (&h,
> 			 READ_CONTROL | FILE_READ_ATTRIBUTES | FILE_READ_EA,
> 			 &attr, &io, NULL, 0, FILE_SHARE_VALID_FLAGS,
> 			 FILE_OPEN,
> 			 FILE_OPEN_REPARSE_POINT | FILE_OPEN_FOR_BACKUP_INTENT,
> 			 eabuf, easize);
> 
> What I expect as status code is either STATUS_EAS_NOT_SUPPORTED if the
> share does not support EAs, or STATUS_SUCCESS if the Samba share supports
> EAs (with "ea support = yes" in smb.conf).
> 
> That's what I get from my Samba 3.0.x shares, at least.
> 
> What I get from my Samba 3.2 shares is either STATUS_EAS_NOT_SUPPORTED
> if EAs are not supported, or STATUS_INVALID_PARAMETER, if EAs are
> supported.  That's somewhat surprising and, AFAICS, a bug.
> 
> An interesting fact here is that I only get STATUS_INVALID_PARAMETER
> for the shares's root dir:
> 
>   //server/share             ==> STATUS_INVALID_PARAMETER
>   //server/share/file        ==> STATUS_SUCCESS
>   //server/share/subdir      ==> STATUS_SUCCESS
>   //server/share/subdir/file ==> STATUS_SUCCESS
> 
> So the problem is restricted to the case where I try to open a share's
> root dir.
> 
> Here's the question:  Is that a regression or is that expected
> behaviour in 3.2?  If it's expected, what's the reason?

Can you send a debug level 10 log from smbd showing this please.

Thanks,

Jeremy.