Merge branch 'udp-msg_zerocopy'

Willem de Bruijn says:

====================
udp msg_zerocopy

Enable MSG_ZEROCOPY for udp sockets

Patch 1/3 is the main patch, a rework of RFC patch
  http://patchwork.ozlabs.org/patch/899630/
  more details in the patch commit message

Patch 2/3 is an optimization to remove a branch from the UDP hot path
  and refcount_inc/refcount_dec_and_test pair when zerocopy is used.
  This used to be included in the first patch in v2.

Patch 3/3 runs the already existing udp zerocopy tests
  as part of kselftest

See also recent Linux Plumbers presentation
  https://linuxplumbersconf.org/event/2/contributions/106/attachments/104/128/willemdebruijn-lpc2018-udpgso-presentation-20181113.pdf

Changes:
  v1 -> v2
    - Fixup reverse christmas tree violation
  v2 -> v3
    - Split refcount avoidance optimization into separate patch
      - Fix refcount leak on error in fragmented case
        (thanks to Paolo Abeni for pointing this one out!)
      - Fix refcount inc on zero
  v3 -> v4
    - Move skb_zcopy_set below the only kfree_skb that might cause
      a premature uarg destroy before skb_zerocopy_put_abort
      - Move the entire skb_shinfo assignment block, to keep that
	cacheline access in one place
====================

Signed-off-by: David S. Miller <davem@davemloft.net>

include/linux/skbuff.h

@@ -481,10 +481,11 @@ static inline void sock_zerocopy_get(struct ubuf_info *uarg)
 }
 
 void sock_zerocopy_put(struct ubuf_info *uarg);
-void sock_zerocopy_put_abort(struct ubuf_info *uarg);
+void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
 
 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
 
+int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 			     struct msghdr *msg, int len,
 			     struct ubuf_info *uarg);
@@ -1325,9 +1326,13 @@ static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
 	return is_zcopy ? skb_uarg(skb) : NULL;
 }
 
-static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
+static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
+				 bool *have_ref)
 {
 	if (skb && uarg && !skb_zcopy(skb)) {
+		if (unlikely(have_ref && *have_ref))
+			*have_ref = false;
+		else
 			sock_zerocopy_get(uarg);
 		skb_shinfo(skb)->destructor_arg = uarg;
 		skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
@@ -1373,7 +1378,7 @@ static inline void skb_zcopy_abort(struct sk_buff *skb)
 	struct ubuf_info *uarg = skb_zcopy(skb);
 
 	if (uarg) {
-		sock_zerocopy_put_abort(uarg);
+		sock_zerocopy_put_abort(uarg, false);
 		skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
 	}
 }

net/core/skbuff.c

@@ -1089,7 +1089,7 @@ void sock_zerocopy_put(struct ubuf_info *uarg)
 }
 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
 
-void sock_zerocopy_put_abort(struct ubuf_info *uarg)
+void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
 {
 	if (uarg) {
 		struct sock *sk = skb_from_uarg(uarg)->sk;
@@ -1097,6 +1097,7 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg)
 		atomic_dec(&sk->sk_zckey);
 		uarg->len--;
 
+		if (have_uref)
 			sock_zerocopy_put(uarg);
 	}
 }
@@ -1105,6 +1106,12 @@ EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
 				   struct iov_iter *from, size_t length);
 
+int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
+{
+	return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
+}
+EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
+
 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 			     struct msghdr *msg, int len,
 			     struct ubuf_info *uarg)
@@ -1131,7 +1138,7 @@ int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 		return err;
 	}
 
-	skb_zcopy_set(skb, uarg);
+	skb_zcopy_set(skb, uarg, NULL);
 	return skb->len - orig_len;
 }
 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
@@ -1151,7 +1158,7 @@ static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
 			if (skb_copy_ubufs(nskb, GFP_ATOMIC))
 				return -EIO;
 		}
-		skb_zcopy_set(nskb, skb_uarg(orig));
+		skb_zcopy_set(nskb, skb_uarg(orig), NULL);
 	}
 	return 0;
 }

net/core/sock.c

@@ -1018,7 +1018,10 @@ int sock_setsockopt(struct socket *sock, int level, int optname,
 
 	case SO_ZEROCOPY:
 		if (sk->sk_family == PF_INET || sk->sk_family == PF_INET6) {
-			if (sk->sk_protocol != IPPROTO_TCP)
+			if (!((sk->sk_type == SOCK_STREAM &&
+			       sk->sk_protocol == IPPROTO_TCP) ||
+			      (sk->sk_type == SOCK_DGRAM &&
+			       sk->sk_protocol == IPPROTO_UDP)))
 				ret = -ENOTSUPP;
 		} else if (sk->sk_family != PF_RDS) {
 			ret = -ENOTSUPP;

net/ipv4/ip_output.c

@@ -867,6 +867,7 @@ static int __ip_append_data(struct sock *sk,
 			    unsigned int flags)
 {
 	struct inet_sock *inet = inet_sk(sk);
+	struct ubuf_info *uarg = NULL;
 	struct sk_buff *skb;
 
 	struct ip_options *opt = cork->opt;
@@ -880,8 +881,8 @@ static int __ip_append_data(struct sock *sk,
 	int csummode = CHECKSUM_NONE;
 	struct rtable *rt = (struct rtable *)cork->dst;
 	unsigned int wmem_alloc_delta = 0;
+	bool paged, extra_uref;
 	u32 tskey = 0;
-	bool paged;
 
 	skb = skb_peek_tail(queue);
 
@@ -916,6 +917,20 @@ static int __ip_append_data(struct sock *sk,
 	    (!exthdrlen || (rt->dst.dev->features & NETIF_F_HW_ESP_TX_CSUM)))
 		csummode = CHECKSUM_PARTIAL;
 
+	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
+		uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
+		if (!uarg)
+			return -ENOBUFS;
+		extra_uref = true;
+		if (rt->dst.dev->features & NETIF_F_SG &&
+		    csummode == CHECKSUM_PARTIAL) {
+			paged = true;
+		} else {
+			uarg->zerocopy = 0;
+			skb_zcopy_set(skb, uarg, &extra_uref);
+		}
+	}
+
 	cork->length += length;
 
 	/* So, what's going on in the loop below?
@@ -1001,12 +1016,6 @@ static int __ip_append_data(struct sock *sk,
 			skb->csum = 0;
 			skb_reserve(skb, hh_len);
 
-			/* only the initial fragment is time stamped */
-			skb_shinfo(skb)->tx_flags = cork->tx_flags;
-			cork->tx_flags = 0;
-			skb_shinfo(skb)->tskey = tskey;
-			tskey = 0;
-
 			/*
 			 *	Find where to start putting bytes.
 			 */
@@ -1039,6 +1048,13 @@ static int __ip_append_data(struct sock *sk,
 			exthdrlen = 0;
 			csummode = CHECKSUM_NONE;
 
+			/* only the initial fragment is time stamped */
+			skb_shinfo(skb)->tx_flags = cork->tx_flags;
+			cork->tx_flags = 0;
+			skb_shinfo(skb)->tskey = tskey;
+			tskey = 0;
+			skb_zcopy_set(skb, uarg, &extra_uref);
+
 			if ((flags & MSG_CONFIRM) && !skb_prev)
 				skb_set_dst_pending_confirm(skb, 1);
 
@@ -1068,7 +1084,7 @@ static int __ip_append_data(struct sock *sk,
 				err = -EFAULT;
 				goto error;
 			}
-		} else {
+		} else if (!uarg || !uarg->zerocopy) {
 			int i = skb_shinfo(skb)->nr_frags;
 
 			err = -ENOMEM;
@@ -1098,6 +1114,10 @@ static int __ip_append_data(struct sock *sk,
 			skb->data_len += copy;
 			skb->truesize += copy;
 			wmem_alloc_delta += copy;
+		} else {
+			err = skb_zerocopy_iter_dgram(skb, from, copy);
+			if (err < 0)
+				goto error;
 		}
 		offset += copy;
 		length -= copy;
@@ -1110,6 +1130,7 @@ static int __ip_append_data(struct sock *sk,
 error_efault:
 	err = -EFAULT;
 error:
+	sock_zerocopy_put_abort(uarg, extra_uref);
 	cork->length -= length;
 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);

net/ipv4/tcp.c

@@ -1423,7 +1423,7 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
 	if (copied + copied_syn)
 		goto out;
 out_err:
-	sock_zerocopy_put_abort(uarg);
+	sock_zerocopy_put_abort(uarg, true);
 	err = sk_stream_error(sk, flags, err);
 	/* make sure we wake any epoll edge trigger waiter */
 	if (unlikely(skb_queue_len(&sk->sk_write_queue) == 0 &&

net/ipv6/ip6_output.c

@@ -1245,6 +1245,7 @@ static int __ip6_append_data(struct sock *sk,
 {
 	struct sk_buff *skb, *skb_prev = NULL;
 	unsigned int maxfraglen, fragheaderlen, mtu, orig_mtu, pmtu;
+	struct ubuf_info *uarg = NULL;
 	int exthdrlen = 0;
 	int dst_exthdrlen = 0;
 	int hh_len;
@@ -1257,7 +1258,7 @@ static int __ip6_append_data(struct sock *sk,
 	int csummode = CHECKSUM_NONE;
 	unsigned int maxnonfragsize, headersize;
 	unsigned int wmem_alloc_delta = 0;
-	bool paged;
+	bool paged, extra_uref;
 
 	skb = skb_peek_tail(queue);
 	if (!skb) {
@@ -1322,6 +1323,20 @@ static int __ip6_append_data(struct sock *sk,
 	    rt->dst.dev->features & (NETIF_F_IPV6_CSUM | NETIF_F_HW_CSUM))
 		csummode = CHECKSUM_PARTIAL;
 
+	if (flags & MSG_ZEROCOPY && length && sock_flag(sk, SOCK_ZEROCOPY)) {
+		uarg = sock_zerocopy_realloc(sk, length, skb_zcopy(skb));
+		if (!uarg)
+			return -ENOBUFS;
+		extra_uref = true;
+		if (rt->dst.dev->features & NETIF_F_SG &&
+		    csummode == CHECKSUM_PARTIAL) {
+			paged = true;
+		} else {
+			uarg->zerocopy = 0;
+			skb_zcopy_set(skb, uarg, &extra_uref);
+		}
+	}
+
 	/*
 	 * Let's try using as much space as possible.
 	 * Use MTU if total length of the message fits into the MTU.
@@ -1440,12 +1455,6 @@ static int __ip6_append_data(struct sock *sk,
 			skb_reserve(skb, hh_len + sizeof(struct frag_hdr) +
 				    dst_exthdrlen);
 
-			/* Only the initial fragment is time stamped */
-			skb_shinfo(skb)->tx_flags = cork->tx_flags;
-			cork->tx_flags = 0;
-			skb_shinfo(skb)->tskey = tskey;
-			tskey = 0;
-
 			/*
 			 *	Find where to start putting bytes
 			 */
@@ -1477,6 +1486,13 @@ static int __ip6_append_data(struct sock *sk,
 			exthdrlen = 0;
 			dst_exthdrlen = 0;
 
+			/* Only the initial fragment is time stamped */
+			skb_shinfo(skb)->tx_flags = cork->tx_flags;
+			cork->tx_flags = 0;
+			skb_shinfo(skb)->tskey = tskey;
+			tskey = 0;
+			skb_zcopy_set(skb, uarg, &extra_uref);
+
 			if ((flags & MSG_CONFIRM) && !skb_prev)
 				skb_set_dst_pending_confirm(skb, 1);
 
@@ -1506,7 +1522,7 @@ static int __ip6_append_data(struct sock *sk,
 				err = -EFAULT;
 				goto error;
 			}
-		} else {
+		} else if (!uarg || !uarg->zerocopy) {
 			int i = skb_shinfo(skb)->nr_frags;
 
 			err = -ENOMEM;
@@ -1536,6 +1552,10 @@ static int __ip6_append_data(struct sock *sk,
 			skb->data_len += copy;
 			skb->truesize += copy;
 			wmem_alloc_delta += copy;
+		} else {
+			err = skb_zerocopy_iter_dgram(skb, from, copy);
+			if (err < 0)
+				goto error;
 		}
 		offset += copy;
 		length -= copy;
@@ -1548,6 +1568,7 @@ static int __ip6_append_data(struct sock *sk,
 error_efault:
 	err = -EFAULT;
 error:
+	sock_zerocopy_put_abort(uarg, extra_uref);
 	cork->length -= length;
 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);

tools/testing/selftests/net/msg_zerocopy.c

@@ -651,12 +651,13 @@ static void do_flush_datagram(int fd, int type)
 
 static void do_rx(int domain, int type, int protocol)
 {
+	const int cfg_receiver_wait_ms = 400;
 	uint64_t tstop;
 	int fd;
 
 	fd = do_setup_rx(domain, type, protocol);
 
-	tstop = gettimeofday_ms() + cfg_runtime_ms;
+	tstop = gettimeofday_ms() + cfg_runtime_ms + cfg_receiver_wait_ms;
 	do {
 		if (type == SOCK_STREAM)
 			do_flush_tcp(fd);

tools/testing/selftests/net/msg_zerocopy.sh

@@ -25,6 +25,8 @@ readonly path_sysctl_mem="net.core.optmem_max"
 if [[ "$#" -eq "0" ]]; then
 	$0 4 tcp -t 1
 	$0 6 tcp -t 1
+	$0 4 udp -t 1
+	$0 6 udp -t 1
 	echo "OK. All tests passed"
 	exit 0
 fi

tools/testing/selftests/net/udpgso_bench.sh

@@ -35,6 +35,9 @@ run_udp() {
 
 	echo "udp gso"
 	run_in_netns ${args} -S 0
+
+	echo "udp gso zerocopy"
+	run_in_netns ${args} -S 0 -z
 }
 
 run_tcp() {