nfsd4: fix response size estimation for OP_SEQUENCE
authorJ. Bruce Fields <bfields@redhat.com>
Thu, 21 Aug 2014 19:04:31 +0000 (15:04 -0400)
committerJ. Bruce Fields <bfields@redhat.com>
Tue, 21 Oct 2014 13:10:50 +0000 (09:10 -0400)
We added this new estimator function but forgot to hook it up.  The
effect is that NFSv4.1 (and greater) won't do zero-copy reads.

The estimate was also wrong by 8 bytes.

Fixes: ccae70a9ee41 "nfsd4: estimate sequence response size"
Cc: stable@vger.kernel.org
Reported-by: Chuck Lever <chucklever@gmail.com>
Signed-off-by: J. Bruce Fields <bfields@redhat.com>
fs/nfsd/nfs4proc.c

index cdeb3cfd6f32b2d412c8478283c8ffc732d6dc06..f4bd578bed551451d15852e36e254d2e585cba11 100644 (file)
@@ -1589,7 +1589,8 @@ static inline u32 nfsd4_rename_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op
 static inline u32 nfsd4_sequence_rsize(struct svc_rqst *rqstp,
                                       struct nfsd4_op *op)
 {
-       return NFS4_MAX_SESSIONID_LEN + 20;
+       return (op_encode_hdr_size
+               + XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + 5) * sizeof(__be32);
 }
 
 static inline u32 nfsd4_setattr_rsize(struct svc_rqst *rqstp, struct nfsd4_op *op)
@@ -1893,6 +1894,7 @@ static struct nfsd4_operation nfsd4_ops[] = {
                .op_func = (nfsd4op_func)nfsd4_sequence,
                .op_flags = ALLOWED_WITHOUT_FH | ALLOWED_AS_FIRST_OP,
                .op_name = "OP_SEQUENCE",
+               .op_rsize_bop = (nfsd4op_rsize)nfsd4_sequence_rsize,
        },
        [OP_DESTROY_CLIENTID] = {
                .op_func = (nfsd4op_func)nfsd4_destroy_clientid,