Re: [PATCH v1 12/42] NFSD: Update the NFSv3 FSSTAT3res encoder to use struct xdr_stream

[Chuck Lever <chuck.lever@oracle.com>]

> On Mar 2, 2021, at 3:45 PM, J. Bruce Fields <bfields@fieldses.org> wrote:
> 
> On Mon, Mar 01, 2021 at 10:16:31AM -0500, Chuck Lever wrote:
>> Signed-off-by: Chuck Lever <chuck.lever@oracle.com>
>> ---
>> fs/nfsd/nfs3xdr.c |   58 ++++++++++++++++++++++++++++++++++++++++-------------
>> 1 file changed, 44 insertions(+), 14 deletions(-)
>> 
>> diff --git a/fs/nfsd/nfs3xdr.c b/fs/nfsd/nfs3xdr.c
>> index e159e4557428..e4a569e7216d 100644
>> --- a/fs/nfsd/nfs3xdr.c
>> +++ b/fs/nfsd/nfs3xdr.c
>> @@ -17,6 +17,13 @@
>> #define NFSDDBG_FACILITY		NFSDDBG_XDR
>> 
>> 
>> +/*
>> + * Force construction of an empty post-op attr
>> + */
>> +static const struct svc_fh nfs3svc_null_fh = {
>> +	.fh_no_wcc	= true,
>> +};
>> +
>> /*
>>  * Mapping of S_IF* types to NFS file types
>>  */
>> @@ -1392,27 +1399,50 @@ nfs3svc_encode_entry_plus(void *cd, const char *name,
>> 	return encode_entry(cd, name, namlen, offset, ino, d_type, 1);
>> }
>> 
>> +static bool
>> +svcxdr_encode_fsstat3resok(struct xdr_stream *xdr,
>> +			   const struct nfsd3_fsstatres *resp)
>> +{
>> +	const struct kstatfs *s = &resp->stats;
>> +	u64 bs = s->f_bsize;
>> +	__be32 *p;
>> +
>> +	p = xdr_reserve_space(xdr, XDR_UNIT * 13);
>> +	if (!p)
>> +		return false;
>> +	p = xdr_encode_hyper(p, bs * s->f_blocks);	/* total bytes */
>> +	p = xdr_encode_hyper(p, bs * s->f_bfree);	/* free bytes */
>> +	p = xdr_encode_hyper(p, bs * s->f_bavail);	/* user available bytes */
>> +	p = xdr_encode_hyper(p, s->f_files);		/* total inodes */
>> +	p = xdr_encode_hyper(p, s->f_ffree);		/* free inodes */
>> +	p = xdr_encode_hyper(p, s->f_ffree);		/* user available inodes */
>> +	*p = cpu_to_be32(resp->invarsec);		/* mean unchanged time */
>> +
>> +	return true;
>> +}
>> +
>> /* FSSTAT */
>> int
>> nfs3svc_encode_fsstatres(struct svc_rqst *rqstp, __be32 *p)
>> {
>> +	struct xdr_stream *xdr = &rqstp->rq_res_stream;
>> 	struct nfsd3_fsstatres *resp = rqstp->rq_resp;
>> -	struct kstatfs	*s = &resp->stats;
>> -	u64		bs = s->f_bsize;
>> 
>> -	*p++ = resp->status;
>> -	*p++ = xdr_zero;	/* no post_op_attr */
>> -
>> -	if (resp->status == 0) {
>> -		p = xdr_encode_hyper(p, bs * s->f_blocks);	/* total bytes */
>> -		p = xdr_encode_hyper(p, bs * s->f_bfree);	/* free bytes */
>> -		p = xdr_encode_hyper(p, bs * s->f_bavail);	/* user available bytes */
>> -		p = xdr_encode_hyper(p, s->f_files);	/* total inodes */
>> -		p = xdr_encode_hyper(p, s->f_ffree);	/* free inodes */
>> -		p = xdr_encode_hyper(p, s->f_ffree);	/* user available inodes */
>> -		*p++ = htonl(resp->invarsec);	/* mean unchanged time */
>> +	if (!svcxdr_encode_nfsstat3(xdr, resp->status))
>> +		return 0;
>> +	switch (resp->status) {
>> +	case nfs_ok:
>> +		if (!svcxdr_encode_post_op_attr(rqstp, xdr, &nfs3svc_null_fh))
>> +			return 0;
>> +		if (!svcxdr_encode_fsstat3resok(xdr, resp))
>> +			return 0;
>> +		break;
>> +	default:
>> +		if (!svcxdr_encode_post_op_attr(rqstp, xdr, &nfs3svc_null_fh))
>> +			return 0;
> 
> Dumb question: will this result in the same xdr on the wire as the above
> that just hard-coded xdr_zero?

It should result in an xdr_zero. The point here is to document what that
xdr_zero means: basically that it is an absent post_op_attr.

The simpler form would be

     if (xdr_stream_encode_item_absent(xdr) < 0)
           return 0;

Which has the same amount of boilerplatiness but is less explanatory.


> I feel like there's a lot of biolerplate error handling.  In the v4 case
> I centralized some of it after a fuzzer found an oops in some of that
> copy-pasted boilerplate:
> 
> 	b7571e4cd39a nfsd4: skip encoder in trivial error cases
> 	bac966d60652 nfsd4: individual encoders no longer see error cases
> 
> I dunno, maybe that was overkill.

True, errors will be extremely rare and usually the result of a coding
error. The problem is that an encoder error can result in:

- A general protection fault that makes the server unusable, resulting
  in a temporary denial of service or at worst corruption of recently
  written data, or
- The server returning garbage or corrupted data along with an NFS_OK
  result.

So I think we need to take a little extra care here. If there's a more
efficient and easier way to do that checking, I'm all ears.


> --b.
> 
>> 	}
>> -	return xdr_ressize_check(rqstp, p);
>> +
>> +	return 1;
>> }
>> 
>> /* FSINFO */

--
Chuck Lever