[PATCH] RPC: skip over transport-specific heads automatically

 Add a generic mechanism for skipping over transport-specific headers
 when constructing an RPC request.  This removes another "xprt->stream"
 dependency.

 Test-plan:
 Write-intensive workload on a single mount point (try both UDP and
 TCP).

 Signed-off-by: Chuck Lever <cel@netapp.com>
 Signed-off-by: Trond Myklebust <Trond.Myklebust@netapp.com>

3 files changed, 21 insertions, 14 deletions

diff --git a/net/sunrpc/auth_gss/auth_gss.c b/net/sunrpc/auth_gss/auth_gss.c
index 53a030acdf75..d2b08f16c257 100644
--- a/net/sunrpc/auth_gss/auth_gss.c
+++ b/net/sunrpc/auth_gss/auth_gss.c
@@ -844,10 +844,8 @@ gss_marshal(struct rpc_task *task, u32 *p)
 
         /* We compute the checksum for the verifier over the xdr-encoded bytes
          * starting with the xid and ending at the end of the credential: */
-        iov.iov_base = req->rq_snd_buf.head[0].iov_base;
-        if (task->tk_client->cl_xprt->stream)
-                /* See clnt.c:call_header() */
-                iov.iov_base += 4;
+        iov.iov_base = xprt_skip_transport_header(task->tk_xprt,
+                                        req->rq_snd_buf.head[0].iov_base);
         iov.iov_len = (u8 *)p - (u8 *)iov.iov_base;
         xdr_buf_from_iov(&iov, &verf_buf);
 
diff --git a/net/sunrpc/clnt.c b/net/sunrpc/clnt.c
index 4677959d2834..cc1b773a79d3 100644
--- a/net/sunrpc/clnt.c
+++ b/net/sunrpc/clnt.c
@@ -1075,13 +1075,12 @@ static u32 *
 call_header(struct rpc_task *task)
 {
         struct rpc_clnt *clnt = task->tk_client;
-        struct rpc_xprt *xprt = clnt->cl_xprt;
         struct rpc_rqst *req = task->tk_rqstp;
         u32             *p = req->rq_svec[0].iov_base;
 
         /* FIXME: check buffer size? */
-        if (xprt->stream)
-                *p++ = 0;               /* fill in later */
+
+        p = xprt_skip_transport_header(task->tk_xprt, p);
         *p++ = req->rq_xid;             /* XID */
         *p++ = htonl(RPC_CALL);         /* CALL */
         *p++ = htonl(RPC_VERSION);      /* RPC version */
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 57988300640a..aaf053b1a0c4 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -282,6 +282,13 @@ static int xs_udp_send_request(struct rpc_task *task)
         return status;
 }
 
+static inline void xs_encode_tcp_record_marker(struct xdr_buf *buf)
+{
+        u32 reclen = buf->len - sizeof(rpc_fraghdr);
+        rpc_fraghdr *base = buf->head[0].iov_base;
+        *base = htonl(RPC_LAST_STREAM_FRAGMENT | reclen);
+}
+
 /**
  * xs_tcp_send_request - write an RPC request to a TCP socket
  * @task: address of RPC task that manages the state of an RPC request
@@ -301,11 +308,9 @@ static int xs_tcp_send_request(struct rpc_task *task)
         struct rpc_rqst *req = task->tk_rqstp;
         struct rpc_xprt *xprt = req->rq_xprt;
         struct xdr_buf *xdr = &req->rq_snd_buf;
-        u32 *marker = req->rq_svec[0].iov_base;
         int status, retry = 0;
 
-        /* Write the record marker */
-        *marker = htonl(0x80000000|(req->rq_slen-sizeof(*marker)));
+        xs_encode_tcp_record_marker(&req->rq_snd_buf);
 
         xs_pktdump("packet data:",
                                 req->rq_svec->iov_base,
@@ -503,16 +508,19 @@ static inline void xs_tcp_read_fraghdr(struct rpc_xprt *xprt, skb_reader_t *desc
         xprt->tcp_offset += used;
         if (used != len)
                 return;
+
         xprt->tcp_reclen = ntohl(xprt->tcp_recm);
-        if (xprt->tcp_reclen & 0x80000000)
+        if (xprt->tcp_reclen & RPC_LAST_STREAM_FRAGMENT)
                 xprt->tcp_flags |= XPRT_LAST_FRAG;
         else
                 xprt->tcp_flags &= ~XPRT_LAST_FRAG;
-        xprt->tcp_reclen &= 0x7fffffff;
+        xprt->tcp_reclen &= RPC_FRAGMENT_SIZE_MASK;
+
         xprt->tcp_flags &= ~XPRT_COPY_RECM;
         xprt->tcp_offset = 0;
+
         /* Sanity check of the record length */
-        if (xprt->tcp_reclen < 4) {
+        if (unlikely(xprt->tcp_reclen < 4)) {
                 dprintk("RPC:      invalid TCP record fragment length\n");
                 xprt_disconnect(xprt);
                 return;
@@ -1065,6 +1073,7 @@ int xs_setup_udp(struct rpc_xprt *xprt, struct rpc_timeout *to)
 
         xprt->prot = IPPROTO_UDP;
         xprt->port = XS_MAX_RESVPORT;
+        xprt->tsh_size = 0;
         xprt->stream = 0;
         xprt->nocong = 0;
         xprt->cwnd = RPC_INITCWND;
@@ -1105,11 +1114,12 @@ int xs_setup_tcp(struct rpc_xprt *xprt, struct rpc_timeout *to)
 
         xprt->prot = IPPROTO_TCP;
         xprt->port = XS_MAX_RESVPORT;
+        xprt->tsh_size = sizeof(rpc_fraghdr) / sizeof(u32);
         xprt->stream = 1;
         xprt->nocong = 1;
         xprt->cwnd = RPC_MAXCWND(xprt);
         xprt->resvport = capable(CAP_NET_BIND_SERVICE) ? 1 : 0;
-        xprt->max_payload = (1U << 31) - 1;
+        xprt->max_payload = RPC_MAX_FRAGMENT_SIZE;
 
         INIT_WORK(&xprt->connect_worker, xs_tcp_connect_worker, xprt);