On Wed, Mar 16 2016, Sergio Gelato wrote:
> * Jiri Slaby [2016-03-16 11:58:46 +0100]:
>> From: NeilBrown <neilb@xxxxxxxx>
>>
>> 3.12-stable review patch. If anyone has any objections, please let me know.
>
>> If a non-inherited ACL is set on a directory, nfsd will try to set the Posix
>> default ACL to NULL. This gets converted to "" by generic_setxattr().
>> As "" is not a valid posix acl attribute value, this results in an error.
>>
>> So instead of setting the xattr to NULL, remove it.
>
> There is similar code in nfsd_set_posix_acl() further down in the same source
> file which skips the vfs_removexattr() call for default ACLs on non-directories
> (there shouldn't be too many of these) and ignores ENODATA returns from
> vfs_removexattr(). Are these precautions guaranteed to be unnecessary here in
> set_nfsv4_acl_one() ?
Those are not precautions, they are optimisations. The vfs_removexattr
call would not be harmful, but would be unnecessary.
So we don't need to worry about that for set_nfsv4_acl_one - we only
apply optimisations to -stable if they are very significant.
Thanks,
NeilBrown
>
>> Fixes: ba1816b40a ("nfsd: fix NFS regression")
>> Signed-off-by: NeilBrown <neilb@xxxxxxxx>
>> Cc: Sergio Gelato <Sergio.Gelato@xxxxxxxxxxx>
>> Signed-off-by: Jiri Slaby <jslaby@xxxxxxx>
>> ---
>> fs/nfsd/vfs.c | 2 +-
>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/fs/nfsd/vfs.c b/fs/nfsd/vfs.c
>> index fafac65804d6..e5f146c7c871 100644
>> --- a/fs/nfsd/vfs.c
>> +++ b/fs/nfsd/vfs.c
>> @@ -510,7 +510,7 @@ set_nfsv4_acl_one(struct dentry *dentry, struct posix_acl *pacl, char *key)
>> int error = 0;
>>
>> if (!pacl)
>> - return vfs_setxattr(dentry, key, NULL, 0, 0);
>> + return vfs_removexattr(dentry, key);
>>
>> buflen = posix_acl_xattr_size(pacl->a_count);
>> buf = kmalloc(buflen, GFP_KERNEL);
>> --
>> 2.7.3
>>
Attachment:
signature.asc
Description: PGP signature