DMARC-Filter: OpenDMARC Filter v1.3.2 mail.kernel.org F39D8206B7
From:   Trond Myklebust <trondmy@gmail.com>
To:     Chuck Lever <chuck.lever@oracle.com>,
        Catalin Marinas <catalin.marinas@arm.com>,
        Cristian Marussi <cristian.marussi@arm.com>
Cc:     linux-nfs@vger.kernel.org
Subject: [PATCH] SUNRPC: Fix RPC receive hangs
Date:   Tue,  4 Dec 2018 14:24:39 -0500
Message-Id: <20181204192439.39259-1-trond.myklebust@hammerspace.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-nfs-owner@vger.kernel.org
Precedence: bulk

The RPC code is occasionally hanging when the receive code fails to
empty the socket buffer due to a partial read of the data. When we
convert that to an EAGAIN, it appears we occasionally leave data in the
socket. The fix is to just keep reading until the socket returns
EAGAIN/EWOULDBLOCK.

Reported-by: Catalin Marinas <catalin.marinas@arm.com>
Reported-by: Cristian Marussi <cristian.marussi@arm.com>
Reported-by: Chuck Lever <chuck.lever@oracle.com>
Signed-off-by: Trond Myklebust <trond.myklebust@hammerspace.com>
---
 net/sunrpc/xprtsock.c | 39 +++++++++++++++++++--------------------
 1 file changed, 19 insertions(+), 20 deletions(-)

diff --git a/net/sunrpc/xprtsock.c b/net/sunrpc/xprtsock.c
index ae77c71c1f64..1c0560dc5557 100644
--- a/net/sunrpc/xprtsock.c
+++ b/net/sunrpc/xprtsock.c
@@ -398,7 +398,7 @@ xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 			goto out;
 		if (ret != want)
-			goto eagain;
+			goto out;
 		seek = 0;
 	} else {
 		seek -= buf->head[0].iov_len;
@@ -418,7 +418,7 @@ xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 			goto out;
 		if (ret != want)
-			goto eagain;
+			goto out;
 		seek = 0;
 	} else {
 		seek -= buf->page_len;
@@ -433,7 +433,7 @@ xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 		if (offset == count || msg->msg_flags & (MSG_EOR|MSG_TRUNC))
 			goto out;
 		if (ret != want)
-			goto eagain;
+			goto out;
 	} else
 		offset += buf->tail[0].iov_len;
 	ret = -EMSGSIZE;
@@ -441,9 +441,6 @@ xs_read_xdr_buf(struct socket *sock, struct msghdr *msg, int flags,
 out:
 	*read = offset - seek_init;
 	return ret;
-eagain:
-	ret = -EAGAIN;
-	goto out;
 sock_err:
 	offset += seek;
 	goto out;
@@ -486,19 +483,18 @@ xs_read_stream_request(struct sock_xprt *transport, struct msghdr *msg,
 	if (transport->recv.offset == transport->recv.len) {
 		if (xs_read_stream_request_done(transport))
 			msg->msg_flags |= MSG_EOR;
-		return transport->recv.copied;
+		return read;
 	}
 
 	switch (ret) {
+	default:
+		break;
 	case -EMSGSIZE:
-		return transport->recv.copied;
+		return read;
 	case 0:
 		return -ESHUTDOWN;
-	default:
-		if (ret < 0)
-			return ret;
 	}
-	return -EAGAIN;
+	return ret;
 }
 
 static size_t
@@ -537,7 +533,7 @@ xs_read_stream_call(struct sock_xprt *transport, struct msghdr *msg, int flags)
 
 	ret = xs_read_stream_request(transport, msg, flags, req);
 	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
-		xprt_complete_bc_request(req, ret);
+		xprt_complete_bc_request(req, transport->recv.copied);
 
 	return ret;
 }
@@ -570,7 +566,7 @@ xs_read_stream_reply(struct sock_xprt *transport, struct msghdr *msg, int flags)
 
 	spin_lock(&xprt->queue_lock);
 	if (msg->msg_flags & (MSG_EOR|MSG_TRUNC))
-		xprt_complete_rqst(req->rq_task, ret);
+		xprt_complete_rqst(req->rq_task, transport->recv.copied);
 	xprt_unpin_rqst(req);
 out:
 	spin_unlock(&xprt->queue_lock);
@@ -591,10 +587,8 @@ xs_read_stream(struct sock_xprt *transport, int flags)
 		if (ret <= 0)
 			goto out_err;
 		transport->recv.offset = ret;
-		if (ret != want) {
-			ret = -EAGAIN;
-			goto out_err;
-		}
+		if (transport->recv.offset != want)
+			return transport->recv.offset;
 		transport->recv.len = be32_to_cpu(transport->recv.fraghdr) &
 			RPC_FRAGMENT_SIZE_MASK;
 		transport->recv.offset -= sizeof(transport->recv.fraghdr);
@@ -602,6 +596,9 @@ xs_read_stream(struct sock_xprt *transport, int flags)
 	}
 
 	switch (be32_to_cpu(transport->recv.calldir)) {
+	default:
+		msg.msg_flags |= MSG_TRUNC;
+		break;
 	case RPC_CALL:
 		ret = xs_read_stream_call(transport, &msg, flags);
 		break;
@@ -616,6 +613,8 @@ xs_read_stream(struct sock_xprt *transport, int flags)
 		goto out_err;
 	read += ret;
 	if (transport->recv.offset < transport->recv.len) {
+		if (!(msg.msg_flags & MSG_TRUNC))
+			return read;
 		ret = xs_read_discard(transport->sock, &msg, flags,
 				transport->recv.len - transport->recv.offset);
 		if (ret <= 0)
@@ -623,7 +622,7 @@ xs_read_stream(struct sock_xprt *transport, int flags)
 		transport->recv.offset += ret;
 		read += ret;
 		if (transport->recv.offset != transport->recv.len)
-			return -EAGAIN;
+			return read;
 	}
 	if (xs_read_stream_request_done(transport)) {
 		trace_xs_stream_read_request(transport);
@@ -653,7 +652,7 @@ static void xs_stream_data_receive(struct sock_xprt *transport)
 	clear_bit(XPRT_SOCK_DATA_READY, &transport->sock_state);
 	for (;;) {
 		ret = xs_read_stream(transport, MSG_DONTWAIT);
-		if (ret <= 0)
+		if (ret < 0)
 			break;
 		read += ret;
 		cond_resched();
-- 
2.19.2