[Acl-Devel] What would be the 'proper' function to determine

[John Trostel]

On 26-Apr-2001 Andreas Gruenbacher wrote:
> On Wed, 25 Apr 2001, Robert Watson wrote:
> 
>> On Mon, 23 Apr 2001, Andreas Gruenbacher wrote:
>> 

...snip ...

>>   If the argument type specifies a type of ACL that cannot be associated
>>   with path_p, then the function shall fail.
>> 
>> I wasn't sure I entirely agreed, but it seemed to me that Casey's argument
>> held water.  :-)  We now return EOPNOTSUPP from acl_get_*() if ACLs are
>> not permitted on the object in question.  We also don't have the
>> pathconf() feature working yet, as right now ACLs are implicitly enabled
>> on file systems where the appropriate EAs are available.  I have some
>> patches to make that policy driven using a file system superblock flag,
>> but haven't tested/committed yet.
> 
> I did not realize I was overruled on this, but for compatibility reasons I
> will change the behaviour of acl_get_{fd,file}() to match the one of Irix
> + FreeBSD.
> 
> How about acl_set_{fd,file}()? I presume your interpretation of draft 17
> also means those never do the equivalent of chmod on filesystems without
> ACL support?
> 
> Thanks,
> --Andreas.

It is pretty easy to modify sys_acl_get and sys_acl_set (and consequently
acl_set_file and acl_set_fd) in XFS to return ENOSYS from this call if they are
made on a filesystem which does not support acls and an acl on an acl-enabled
system.  I have a post in to SGI to see if they want to take the change into
XFS.  I seem to remember from the 'standard' that EOPNOTSUPP was the 'proper'
error to report back instead of ENOSYS.  (I choose ENOSYS because SGI uses
ENOSYS in the no_posix_acl.c file implementation of these functions.) Should I
press SGI to convert the return codes to EOPNOTSUPP?

Oh yeah... the XFS acl_set_{fd,file} do _not_ do an equivalent chmod on non
acl-ized filesystems either.

-- 
John M. Trostel
Linux OS Engineer
Connex
jtrostel at connex.com
To Unsubscribe: send mail to majordomo at cyrus.watson.org
with "unsubscribe posix1e" in the body of the message