On Fri, Apr 21 2017, J. Bruce Fields wrote:

> On Thu, Apr 20, 2017 at 12:19:35PM -0400, J. Bruce Fields wrote:
>> On Tue, Apr 18, 2017 at 01:13:51PM -0400, J. Bruce Fields wrote:
>> > On Tue, Apr 18, 2017 at 10:25:20AM +1000, NeilBrown wrote:
>> > >  I can't say that I like this patch at all.
>> > > 
>> > > The problem is that:
>> > > 
>> > > 	pages = size / PAGE_SIZE + 1; /* extra page as we hold both request and reply.
>> > > 				       * We assume one is at most one page
>> > > 				       */
>> > > 
>> > > this assumption is never verified.
>> > > To my mind, the "obvious" way to verify this assumption is that an
>> > > attempt to generate a multi-page reply should fail if there was a
>> > > multi-page request.
>> > 
>> > A third option, by the way, which Ari Kauppi argued for, is adding a
>> > null check each time we increment rq_next_page, since we seem to arrange
>> > for the page array to always be NULL-terminated.
>> > 
>> > > Failing if there was a little bit of extra noise at the end of the
>> > > request seems harsher than necessary, and could result in a regression.
>> > 
>> > You're worrying there might be a weird old client out there somewhere?
>> > I guess it seems like a small enough risk to me.  I'm more worried the
>> > extra garbage might violate assumptions elsewhere in the code.
>> > 
>> > But, this looks good too:
>> 
>> But, I'm not too happy about putting that NFSv2/v3-specific check in
>> common rpc code.
>
> Well, but it should work just as well in nfsd_dispatch, I think?
> (Untested).  So, maybe that's simplest as a first step:
>
> diff --git a/fs/nfsd/nfssvc.c b/fs/nfsd/nfssvc.c
> index 31e1f9593457..b6298d30a01f 100644
> --- a/fs/nfsd/nfssvc.c
> +++ b/fs/nfsd/nfssvc.c
> @@ -759,6 +759,22 @@ nfsd_dispatch(struct svc_rqst *rqstp, __be32 *statp)
>  				rqstp->rq_vers, rqstp->rq_proc);
>  	proc = rqstp->rq_procinfo;
>  
> +	if (rqstp->rq_vers < 4 &&
> +	    (proc->pc_xdrressize == 0
> +			|| proc->pc_xdrressize > XDR_QUADLEN(PAGE_SIZE))
> +		&& rqstp->rq_arg.len > PAGE_SIZE) {
> +		/*
> +		 * NFSv2 and v3 assume that an operation may have either a
> +		 * large argument, or a large reply, but never both.
> +		 *
> +		 * NFSv4 may handle compounds with both argument and
> +		 * reply larger than a reply; it has more xdr careful
> +		 * xdr decoding which can handle such calls safely.
> +		 */
> +		dprintk("nfsd: NFSv%d argument too large\n", rqstp->rq_vers);
> +		*statp = rpc_garbage_args;
> +		return 1;
> +	}
>  	/*
>  	 * Give the xdr decoder a chance to change this if it wants
>  	 * (necessary in the NFSv4.0 compound case)

I like this.  I think this should be the basis of what goes to -stable,
and other improvements should stay in mainline.

The only change I would suggest would be to be explicit about where the
nfsacl protocol fits with this.

We could change "rqstp->rq_vers < 4" to
 "rqstp->rq_prog == NFS_PROGRAM && rqstp->rq_vers < 4"
or we could change the text:
 NFSv2 and v3 assume ...
to
 NFSv2 and v3, along with NFSASL, assume ...

and possibly change "rqstp->rq_vers < 4" to "!nfsd_v4client(rqstp)".

I believe none of this applies to lockd as none of that code ever looks
beyond a single page.

Thanks,
NeilBrown