Re: [PATCH 09/10] nfsd: cap decoded POSIX ACL count to bound sort cost

[Chuck Lever <cel@kernel.org>]

On 5/29/26 2:41 PM, Jeff Layton wrote:
> On Fri, 2026-05-29 at 14:34 -0400, Chuck Lever wrote:
>> [ replaced broken email address for Trond ]
>>
>> On Thu, May 28, 2026, at 5:55 PM, Jeff Layton wrote:
>>
>>> diff --git a/fs/nfsd/nfs4xdr.c b/fs/nfsd/nfs4xdr.c
>>> index c6c50c376b23..5469c6c207ba 100644
>>> --- a/fs/nfsd/nfs4xdr.c
>>> +++ b/fs/nfsd/nfs4xdr.c
>>> @@ -448,6 +448,8 @@ nfsd4_decode_posixacl(struct nfsd4_compoundargs 
>>> *argp, struct posix_acl **acl)
>>>
>>>  	if (xdr_stream_decode_u32(argp->xdr, &count) < 0)
>>>  		return nfserr_bad_xdr;
>>> +	if (count > NFS_ACL_MAX_ENTRIES)
>>> +		return nfserr_resource;
>>
>> nfserr_resource is consistent with other fattr4 decoders, but
>> does not make sense here, IMO. A better choice is nfserr_inval.
>>
> 
> Why not? An ACL that long doesn't violate the spec (as you pointed
> out), the implementation just can't handle it. I do agree that
> nfserr_resource is not the ideal error code, but it's the closest error
> I can see that says "you hit an internal limitation of the server".

Among other reasons, NFS4ERR_RESOURCE does not exist outside of NFSv4.0.
The server is not out of resources in this case.

For NFSv4.1+ NFSD's compound encoder converts it to either
nfserr_rep_too_big or nfserr_rep_too_big_to_cache, neither of which are
close to appropriate.

Rick's NFSv4 POSIX ACL specification uses NFS4ERR_INVAL for just about
every error condition. The specific meaning of INVAL is "invalid input"
which is exactly what we have here.


-- 
Chuck Lever