Skip to contentSkip to conversation

SUNRPC: Fix socket waits for write buffer space

PATCH In review [email protected] 2 participants 4 messages Mar 15, 2022 – Mar 16, 2022

Subsystems (3)
lore
Share & tools
lore

[email protected]

author maintainer
#0 lore raw

PATCH 1/3 SUNRPC: Fix socket waits for write buffer space

From: Trond Myklebust <[email protected]>

The socket layer requires that we use the socket lock to protect changes to the sock->sk_write_pending field and others.

---
 net/sunrpc/xprtsock.c | 54 +++++++++++++++++++++++++++++++------------
 1 file changed, 39 insertions(+), 15 deletions(-)
net/sunrpc/xprtsock.c +39 -15
index 7e39f87cde2d..786df8c0cda3 100644
@@ -763,12 +763,12 @@ xs_stream_start_connect(struct sock_xprt *transport)
 /**
  * xs_nospace - handle transmit was incomplete
  * @req: pointer to RPC request
+ * @transport: pointer to struct sock_xprt
  *
  */
-static int xs_nospace(struct rpc_rqst *req)
+static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 {
-	struct rpc_xprt *xprt = req->rq_xprt;
-	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
+	struct rpc_xprt *xprt = &transport->xprt;
 	struct sock *sk = transport->inet;
 	int ret = -EAGAIN;
 
@@ -779,25 +779,49 @@ static int xs_nospace(struct rpc_rqst *req)
 
 	/* Don't race with disconnect */
 	if (xprt_connected(xprt)) {
+		struct socket_wq *wq;
+
+		rcu_read_lock();
+		wq = rcu_dereference(sk->sk_wq);
+		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
+		rcu_read_unlock();
+
 		/* wait for more buffer space */
+		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 		sk->sk_write_pending++;
 		xprt_wait_for_buffer_space(xprt);
 	} else
 		ret = -ENOTCONN;
 
 	spin_unlock(&xprt->transport_lock);
+	return ret;
+}
 
-	/* Race breaker in case memory is freed before above code is called */
-	if (ret == -EAGAIN) {
-		struct socket_wq *wq;
+static int xs_sock_nospace(struct rpc_rqst *req)
+{
+	struct sock_xprt *transport =
+		container_of(req->rq_xprt, struct sock_xprt, xprt);
+	struct sock *sk = transport->inet;
+	int ret = -EAGAIN;
 
-		rcu_read_lock();
-		wq = rcu_dereference(sk->sk_wq);
-		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
-		rcu_read_unlock();
+	lock_sock(sk);
+	if (!sock_writeable(sk))
+		ret = xs_nospace(req, transport);
+	release_sock(sk);
+	return ret;
+}
 
-		sk->sk_write_space(sk);
-	}
+static int xs_stream_nospace(struct rpc_rqst *req)
+{
+	struct sock_xprt *transport =
+		container_of(req->rq_xprt, struct sock_xprt, xprt);
+	struct sock *sk = transport->inet;
+	int ret = -EAGAIN;
+
+	lock_sock(sk);
+	if (!sk_stream_memory_free(sk))
+		ret = xs_nospace(req, transport);
+	release_sock(sk);
 	return ret;
 }
 
@@ -887,7 +911,7 @@ static int xs_local_send_request(struct rpc_rqst *req)
 	case -ENOBUFS:
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_stream_nospace(req);
 		break;
 	default:
 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
@@ -963,7 +987,7 @@ static int xs_udp_send_request(struct rpc_rqst *req)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_sock_nospace(req);
 		break;
 	case -ENETUNREACH:
 	case -ENOBUFS:
@@ -1083,7 +1107,7 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_stream_nospace(req);
 		break;
 	case -ECONNRESET:
 	case -ECONNREFUSED:
Signature · 1 line
2.35.1
author maintainer
#1 lore raw

PATCH 2/3 SUNRPC: Replace internal use of SOCKWQ_ASYNC_NOSPACE

From: Trond Myklebust <[email protected]>

The socket's SOCKWQ_ASYNC_NOSPACE can be cleared by various actors in the socket layer, so replace it with our own flag in the transport sock_state field.

---
 include/linux/sunrpc/xprtsock.h |  1 +
 net/sunrpc/xprtsock.c           | 22 ++++------------------
 2 files changed, 5 insertions(+), 18 deletions(-)
include/linux/sunrpc/xprtsock.h +1 -0
index 8c2a712cb242..121162a95863 100644
@@ -89,5 +89,6 @@ struct sock_xprt {
 #define XPRT_SOCK_WAKE_WRITE	(5)
 #define XPRT_SOCK_WAKE_PENDING	(6)
 #define XPRT_SOCK_WAKE_DISCONNECT	(7)
+#define XPRT_SOCK_NOSPACE	(8)
 
 #endif /* _LINUX_SUNRPC_XPRTSOCK_H */
net/sunrpc/xprtsock.c +4 -18
index 786df8c0cda3..23cdb841f056 100644
@@ -779,14 +779,8 @@ static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 
 	/* Don't race with disconnect */
 	if (xprt_connected(xprt)) {
-		struct socket_wq *wq;
-
-		rcu_read_lock();
-		wq = rcu_dereference(sk->sk_wq);
-		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
-		rcu_read_unlock();
-
 		/* wait for more buffer space */
+		set_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 		sk->sk_write_pending++;
 		xprt_wait_for_buffer_space(xprt);
@@ -1148,6 +1142,7 @@ static void xs_sock_reset_state_flags(struct rpc_xprt *xprt)
 	clear_bit(XPRT_SOCK_WAKE_ERROR, &transport->sock_state);
 	clear_bit(XPRT_SOCK_WAKE_WRITE, &transport->sock_state);
 	clear_bit(XPRT_SOCK_WAKE_DISCONNECT, &transport->sock_state);
+	clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state);
 }
 
 static void xs_run_error_worker(struct sock_xprt *transport, unsigned int nr)
@@ -1494,7 +1489,6 @@ static void xs_tcp_state_change(struct sock *sk)
 
 static void xs_write_space(struct sock *sk)
 {
-	struct socket_wq *wq;
 	struct sock_xprt *transport;
 	struct rpc_xprt *xprt;
 
@@ -1505,15 +1499,10 @@ static void xs_write_space(struct sock *sk)
 	if (unlikely(!(xprt = xprt_from_sock(sk))))
 		return;
 	transport = container_of(xprt, struct sock_xprt, xprt);
-	rcu_read_lock();
-	wq = rcu_dereference(sk->sk_wq);
-	if (!wq || test_and_clear_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags) == 0)
-		goto out;
-
+	if (!test_and_clear_bit(XPRT_SOCK_NOSPACE, &transport->sock_state))
+		return;
 	xs_run_error_worker(transport, XPRT_SOCK_WAKE_WRITE);
 	sk->sk_write_pending--;
-out:
-	rcu_read_unlock();
 }
 
 /**
@@ -1854,7 +1843,6 @@ static int xs_local_finish_connecting(struct rpc_xprt *xprt,
 		sk->sk_user_data = xprt;
 		sk->sk_data_ready = xs_data_ready;
 		sk->sk_write_space = xs_udp_write_space;
-		sock_set_flag(sk, SOCK_FASYNC);
 		sk->sk_error_report = xs_error_report;
 
 		xprt_clear_connected(xprt);
@@ -2048,7 +2036,6 @@ static void xs_udp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
 		sk->sk_user_data = xprt;
 		sk->sk_data_ready = xs_data_ready;
 		sk->sk_write_space = xs_udp_write_space;
-		sock_set_flag(sk, SOCK_FASYNC);
 
 		xprt_set_connected(xprt);
 
@@ -2215,7 +2202,6 @@ static int xs_tcp_finish_connecting(struct rpc_xprt *xprt, struct socket *sock)
 		sk->sk_data_ready = xs_data_ready;
 		sk->sk_state_change = xs_tcp_state_change;
 		sk->sk_write_space = xs_tcp_write_space;
-		sock_set_flag(sk, SOCK_FASYNC);
 		sk->sk_error_report = xs_error_report;
 
 		/* socket options */
Signature · 1 line
2.35.1
author maintainer
#2 lore raw

PATCH 3/3 SUNRPC: Improve accuracy of socket ENOBUFS determination

From: Trond Myklebust <[email protected]>

The current code checks for whether or not the socket is in a writeable state after we get an EAGAIN. That is racy, since we've dropped the socket lock, so the amount of free buffer may have changed.

Instead, let's check whether the socket is writeable before we try to write to it. If that was the case, we do expect the message to be at least partially sent unless we're in a low memory situation.

---
 net/sunrpc/xprtsock.c | 53 +++++++++++++++----------------------------
 1 file changed, 18 insertions(+), 35 deletions(-)
net/sunrpc/xprtsock.c +18 -35
index 23cdb841f056..87cc97d6c677 100644
@@ -805,13 +805,15 @@ static int xs_sock_nospace(struct rpc_rqst *req)
 	return ret;
 }
 
-static int xs_stream_nospace(struct rpc_rqst *req)
+static int xs_stream_nospace(struct rpc_rqst *req, bool vm_wait)
 {
 	struct sock_xprt *transport =
 		container_of(req->rq_xprt, struct sock_xprt, xprt);
 	struct sock *sk = transport->inet;
 	int ret = -EAGAIN;
 
+	if (vm_wait)
+		return -ENOBUFS;
 	lock_sock(sk);
 	if (!sk_stream_memory_free(sk))
 		ret = xs_nospace(req, transport);
@@ -869,6 +871,7 @@ static int xs_local_send_request(struct rpc_rqst *req)
 	struct msghdr msg = {
 		.msg_flags	= XS_SENDMSG_FLAGS,
 	};
+	bool vm_wait;
 	unsigned int sent;
 	int status;
 
@@ -881,15 +884,14 @@ static int xs_local_send_request(struct rpc_rqst *req)
 	xs_pktdump("packet data:",
 			req->rq_svec->iov_base, req->rq_svec->iov_len);
 
+	vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
+
 	req->rq_xtime = ktime_get();
 	status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 				   transport->xmit.offset, rm, &sent);
 	dprintk("RPC:       %s(%u) = %d\n",
 			__func__, xdr->len - transport->xmit.offset, status);
 
-	if (status == -EAGAIN && sock_writeable(transport->inet))
-		status = -ENOBUFS;
-
 	if (likely(sent > 0) || status == 0) {
 		transport->xmit.offset += sent;
 		req->rq_bytes_sent = transport->xmit.offset;
@@ -899,13 +901,12 @@ static int xs_local_send_request(struct rpc_rqst *req)
 			return 0;
 		}
 		status = -EAGAIN;
+		vm_wait = false;
 	}
 
 	switch (status) {
-	case -ENOBUFS:
-		break;
 	case -EAGAIN:
-		status = xs_stream_nospace(req);
+		status = xs_stream_nospace(req, vm_wait);
 		break;
 	default:
 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
@@ -1023,7 +1024,7 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 	struct msghdr msg = {
 		.msg_flags	= XS_SENDMSG_FLAGS,
 	};
-	bool vm_wait = false;
+	bool vm_wait;
 	unsigned int sent;
 	int status;
 
@@ -1048,7 +1049,10 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 	 * called sendmsg(). */
 	req->rq_xtime = ktime_get();
 	tcp_sock_set_cork(transport->inet, true);
-	while (1) {
+
+	vm_wait = sk_stream_is_writeable(transport->inet) ? true : false;
+
+	do {
 		status = xprt_sock_sendmsg(transport->sock, &msg, xdr,
 					   transport->xmit.offset, rm, &sent);
 
@@ -1069,31 +1073,10 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 
 		WARN_ON_ONCE(sent == 0 && status == 0);
 
-		if (status == -EAGAIN ) {
-			/*
-			 * Return EAGAIN if we're sure we're hitting the
-			 * socket send buffer limits.
-			 */
-			if (test_bit(SOCK_NOSPACE, &transport->sock->flags))
-				break;
-			/*
-			 * Did we hit a memory allocation failure?
-			 */
-			if (sent == 0) {
-				status = -ENOBUFS;
-				if (vm_wait)
-					break;
-				/* Retry, knowing now that we're below the
-				 * socket send buffer limit
-				 */
-				vm_wait = true;
-			}
-			continue;
-		}
-		if (status < 0)
-			break;
-		vm_wait = false;
-	}
+		if (sent > 0)
+			vm_wait = false;
+
+	} while (status == 0);
 
 	switch (status) {
 	case -ENOTSOCK:
@@ -1101,7 +1084,7 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_stream_nospace(req);
+		status = xs_stream_nospace(req, vm_wait);
 		break;
 	case -ECONNRESET:
 	case -ECONNREFUSED:
Signature · 1 line
2.35.1
#3 lore raw
In reply to [email protected]
Quoted 11 lines · On Mar 15, 2022, at 12:28 PM, [email protected] wrote:

From: Trond Myklebust <[email protected]>

The socket layer requires that we use the socket lock to protect changes to the sock->sk_write_pending field and others.

Applied the three patches in this series on top of my RPC-with-TLS prototype. Works nicely!

Quoted 113 lines

---
net/sunrpc/xprtsock.c | 54 +++++++++++++++++++++++++++++++------------

1 file changed, 39 insertions(+), 15 deletions(-)
net/sunrpc/xprtsock.c +39 -15
index 7e39f87cde2d..786df8c0cda3 100644
@@ -763,12 +763,12 @@ xs_stream_start_connect(struct sock_xprt *transport)
 /**
 * xs_nospace - handle transmit was incomplete
 * @req: pointer to RPC request
+ * @transport: pointer to struct sock_xprt
 *
 */
-static int xs_nospace(struct rpc_rqst *req)
+static int xs_nospace(struct rpc_rqst *req, struct sock_xprt *transport)
 {
-	struct rpc_xprt *xprt = req->rq_xprt;
-	struct sock_xprt *transport = container_of(xprt, struct sock_xprt, xprt);
+	struct rpc_xprt *xprt = &transport->xprt;
 	struct sock *sk = transport->inet;
 	int ret = -EAGAIN;
 
@@ -779,25 +779,49 @@ static int xs_nospace(struct rpc_rqst *req)
 
 	/* Don't race with disconnect */
 	if (xprt_connected(xprt)) {
+		struct socket_wq *wq;
+
+		rcu_read_lock();
+		wq = rcu_dereference(sk->sk_wq);
+		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
+		rcu_read_unlock();
+
 		/* wait for more buffer space */
+		set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
 		sk->sk_write_pending++;
 		xprt_wait_for_buffer_space(xprt);
 	} else
 		ret = -ENOTCONN;
 
 	spin_unlock(&xprt->transport_lock);
+	return ret;
+}
 
-	/* Race breaker in case memory is freed before above code is called */
-	if (ret == -EAGAIN) {
-		struct socket_wq *wq;
+static int xs_sock_nospace(struct rpc_rqst *req)
+{
+	struct sock_xprt *transport =
+		container_of(req->rq_xprt, struct sock_xprt, xprt);
+	struct sock *sk = transport->inet;
+	int ret = -EAGAIN;
 
-		rcu_read_lock();
-		wq = rcu_dereference(sk->sk_wq);
-		set_bit(SOCKWQ_ASYNC_NOSPACE, &wq->flags);
-		rcu_read_unlock();
+	lock_sock(sk);
+	if (!sock_writeable(sk))
+		ret = xs_nospace(req, transport);
+	release_sock(sk);
+	return ret;
+}
 
-		sk->sk_write_space(sk);
-	}
+static int xs_stream_nospace(struct rpc_rqst *req)
+{
+	struct sock_xprt *transport =
+		container_of(req->rq_xprt, struct sock_xprt, xprt);
+	struct sock *sk = transport->inet;
+	int ret = -EAGAIN;
+
+	lock_sock(sk);
+	if (!sk_stream_memory_free(sk))
+		ret = xs_nospace(req, transport);
+	release_sock(sk);
 	return ret;
 }
 
@@ -887,7 +911,7 @@ static int xs_local_send_request(struct rpc_rqst *req)
 	case -ENOBUFS:
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_stream_nospace(req);
 		break;
 	default:
 		dprintk("RPC:       sendmsg returned unrecognized error %d\n",
@@ -963,7 +987,7 @@ static int xs_udp_send_request(struct rpc_rqst *req)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_sock_nospace(req);
 		break;
 	case -ENETUNREACH:
 	case -ENOBUFS:
@@ -1083,7 +1107,7 @@ static int xs_tcp_send_request(struct rpc_rqst *req)
 		/* Should we call xs_close() here? */
 		break;
 	case -EAGAIN:
-		status = xs_nospace(req);
+		status = xs_stream_nospace(req);
 		break;
 	case -ECONNRESET:
 	case -ECONNREFUSED:
Signature · 1 line
2.35.1

--
Chuck Lever