[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>
diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index 5798830..aaf053b 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -282,6 +282,13 @@
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 @@
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 @@
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 @@
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 @@
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);