On Thu, Oct 04 2018, Jan Harkes wrote:

> Same for Coda.
>
> uid/gid/mode don't mean anything, access is based on the directory ACL and the authentication token that is held by the userspace cache manager and ultimately decided by the servers.
>
> Unless someone broke this recently and made permission checks uid based I would expect no change. If this is broken by a recent commit I expect something similar to what NFS is trying to do by allowing the actual check to be passed down.

As with afs, the only permission check I can find that is uid based and
which actually affects coda is the check for use fcntl(F_SETFL) to set
O_NOATIME. I suspect that is irrelevant for coda.
I'll resubmit with the same code for both NFS and code - and probably
AFS.

Thanks,
NeilBrown


>
> Jan
>
> On October 4, 2018 10:10:13 AM EDT, David Howells <dhowells@redhat.com> wrote:
>>NeilBrown <neilb@suse.com> wrote:
>>
>>> diff --git a/fs/afs/security.c b/fs/afs/security.c
>>> index 81dfedb7879f..ac2e39de8bff 100644
>>> --- a/fs/afs/security.c
>>> +++ b/fs/afs/security.c
>>> @@ -349,6 +349,16 @@ int afs_permission(struct inode *inode, int
>>mask)
>>>  	if (mask & MAY_NOT_BLOCK)
>>>  		return -ECHILD;
>>>  
>>> +	/* Short-circuit for owner */
>>> +	if (mask & MAY_ACT_AS_OWNER) {
>>> +		if (inode_owner_or_capable(inode))
>>
>>You don't know that inode->i_uid in meaningful.  You may have noticed
>>that
>>afs_permission() ignores i_uid and i_gid entirely.  It queries the
>>server (if
>>this information is not otherwise cached) to ask what permits the user
>>is
>>granted - where the user identity is defined by the key returned from
>>afs_request_key()[*].
>>
>>So, NAK for the afs piece.
>>
>>David
>>
>>[*] If there's no appropriate key, anonymous permits will be used.