udp msg_zerocopy

PATCH Superseded Willem de Bruijn 1 participant 6 messages Nov 30, 2018

Subsystems (6)

Networking Network drivers Kselftest IPv4/IPv6 +2 more subsystems

lore

Share & tools

lore

Willem de Bruijn

author

2018-11-30 07:33 UTC #0 lore raw

v3 cover udp msg_zerocopy

From: Willem de Bruijn <[email protected]>

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

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

Willem de Bruijn (3):
udp: msg_zerocopy
udp: elide zerocopy operation in hot path
selftests: extend zerocopy tests to udp

 include/linux/skbuff.h                      | 13 +++++++----
 net/core/skbuff.c                           | 15 +++++++++----
 net/core/sock.c                             |  5 ++++-
 net/ipv4/ip_output.c                        | 25 +++++++++++++++++++--
 net/ipv4/tcp.c                              |  2 +-
 net/ipv6/ip6_output.c                       | 25 +++++++++++++++++++--
 tools/testing/selftests/net/msg_zerocopy.c  |  3 ++-
 tools/testing/selftests/net/msg_zerocopy.sh |  2 ++
 tools/testing/selftests/net/udpgso_bench.sh |  3 +++
 9 files changed, 78 insertions(+), 15 deletions(-)

Signature · 1 line

2.20.0.rc0.387.gc7a69e6b6c-goog

Willem de Bruijn

author

2018-11-30 07:33 UTC #1 lore raw

PATCH v3 1/3 udp: msg_zerocopy

From: Willem de Bruijn <[email protected]>

Extend zerocopy to udp sockets. Allow setting sockopt SO_ZEROCOPY and interpret flag MSG_ZEROCOPY.

This patch was previously part of the zerocopy RFC patchsets. Zerocopy is not effective at small MTU. With segmentation offload building larger datagrams, the benefit of page flipping outweights the cost of generating a completion notification.

tools/testing/selftests/net/msg_zerocopy.sh after applying follow-on test patch and making skb_orphan_frags_rx same as skb_orphan_frags:

    ipv4 udp -t 1
    tx=191312 (11938 MB) txc=0 zc=n
    rx=191312 (11938 MB)
    ipv4 udp -z -t 1
    tx=304507 (19002 MB) txc=304507 zc=y
    rx=304507 (19002 MB)
    ok
    ipv6 udp -t 1
    tx=174485 (10888 MB) txc=0 zc=n
    rx=174485 (10888 MB)
    ipv6 udp -z -t 1
    tx=294801 (18396 MB) txc=294801 zc=y
    rx=294801 (18396 MB)
    ok

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

Signed-off-by Willem de Bruijn <[email protected]>

---
 include/linux/skbuff.h |  1 +
 net/core/skbuff.c      |  6 ++++++
 net/core/sock.c        |  5 ++++-
 net/ipv4/ip_output.c   | 23 ++++++++++++++++++++++-
 net/ipv6/ip6_output.c  | 23 ++++++++++++++++++++++-
 5 files changed, 55 insertions(+), 3 deletions(-)

include/linux/skbuff.h +1 -0

index 73902acf2b71..04f52e719571 100644
@@ -485,6 +485,7 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg);
 
 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);

net/core/skbuff.c +6 -0

index 02cd7ae3d0fb..435bac91d293 100644
@@ -1105,6 +1105,12 @@ 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)

net/core/sock.c +4 -1

index 6d7e189e3cd9..f5bb89785e47 100644
@@ -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 +22 -1

index c09219e7f230..9602746d7175 100644
@@ -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;

@@ -916,6 +917,19 @@ 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;
+		if (rt->dst.dev->features & NETIF_F_SG &&
+		    csummode == CHECKSUM_PARTIAL) {
+			paged = true;
+		} else {
+			uarg->zerocopy = 0;
+			skb_zcopy_set(skb, uarg);
+		}
+	}
+
 	cork->length += length;
 
 	/* So, what's going on in the loop below?

@@ -1005,6 +1019,7 @@ static int __ip_append_data(struct sock *sk,
 			cork->tx_flags = 0;
 			skb_shinfo(skb)->tskey = tskey;
 			tskey = 0;
+			skb_zcopy_set(skb, uarg);
 
 			/*
 			 *	Find where to start putting bytes.

@@ -1067,7 +1082,7 @@ 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;

@@ -1097,6 +1112,10 @@ 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;

@@ -1104,11 +1123,13 @@ static int __ip_append_data(struct sock *sk,
 
 	if (wmem_alloc_delta)
 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
+	sock_zerocopy_put(uarg);
 	return 0;
 
 error_efault:
 	err = -EFAULT;
 error:
+	sock_zerocopy_put_abort(uarg);
 	cork->length -= length;
 	IP_INC_STATS(sock_net(sk), IPSTATS_MIB_OUTDISCARDS);
 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);

net/ipv6/ip6_output.c +22 -1

index 89e0d5118afe..56f09c4c40dc 100644
@@ -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;

@@ -1322,6 +1323,19 @@ 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;
+		if (rt->dst.dev->features & NETIF_F_SG &&
+		    csummode == CHECKSUM_PARTIAL) {
+			paged = true;
+		} else {
+			uarg->zerocopy = 0;
+			skb_zcopy_set(skb, uarg);
+		}
+	}
+
 	/*
 	 * Let's try using as much space as possible.
 	 * Use MTU if total length of the message fits into the MTU.

@@ -1444,6 +1458,7 @@ static int __ip6_append_data(struct sock *sk,
 			cork->tx_flags = 0;
 			skb_shinfo(skb)->tskey = tskey;
 			tskey = 0;
+			skb_zcopy_set(skb, uarg);
 
 			/*
 			 *	Find where to start putting bytes

@@ -1505,7 +1520,7 @@ 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;

@@ -1535,6 +1550,10 @@ 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;

@@ -1542,11 +1561,13 @@ static int __ip6_append_data(struct sock *sk,
 
 	if (wmem_alloc_delta)
 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
+	sock_zerocopy_put(uarg);
 	return 0;
 
 error_efault:
 	err = -EFAULT;
 error:
+	sock_zerocopy_put_abort(uarg);
 	cork->length -= length;
 	IP6_INC_STATS(sock_net(sk), rt->rt6i_idev, IPSTATS_MIB_OUTDISCARDS);
 	refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);

Signature · 1 line

2.20.0.rc0.387.gc7a69e6b6c-goog

Willem de Bruijn

author

2018-11-30 07:36 UTC #2 lore raw

Re: [PATCH net-next v3 1/3] udp: msg_zerocopy

Quoted 29 lines · Willem de Bruijn

From: Willem de Bruijn <[email protected]>
Extend zerocopy to udp sockets. Allow setting sockopt SO_ZEROCOPY and interpret flag MSG_ZEROCOPY.
This patch was previously part of the zerocopy RFC patchsets. Zerocopy is not effective at small MTU. With segmentation offload building larger datagrams, the benefit of page flipping outweights the cost of generating a completion notification.
tools/testing/selftests/net/msg_zerocopy.sh after applying follow-on test patch and making skb_orphan_frags_rx same as skb_orphan_frags:
    ipv4 udp -t 1
    tx=191312 (11938 MB) txc=0 zc=n
    rx=191312 (11938 MB)
    ipv4 udp -z -t 1
    tx=304507 (19002 MB) txc=304507 zc=y
    rx=304507 (19002 MB)
    ok
    ipv6 udp -t 1
    tx=174485 (10888 MB) txc=0 zc=n
    rx=174485 (10888 MB)
    ipv6 udp -z -t 1
    tx=294801 (18396 MB) txc=294801 zc=y
    rx=294801 (18396 MB)
    ok

Quoted 8 lines

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

one additional change that I forgot to add to the changelog

- test sock_flag SOCK_ZEROCOPY directly in __ip_append_data. this is needed since commit 5cf4a8532c99 ("tcp: really ignore MSG_ZEROCOPY if no SO_ZEROCOPY") did the same for tcp.

Willem de Bruijn

author

2018-11-30 07:33 UTC #3 lore raw

PATCH v3 2/3 udp: elide zerocopy operation in hot path

In reply to Willem de Bruijn

From: Willem de Bruijn <[email protected]>

With MSG_ZEROCOPY, each skb holds a reference to a struct ubuf_info. Release of its last reference triggers a completion notification.

The TCP stack in tcp_sendmsg_locked holds an extra ref independent of the skbs, because it can build, send and free skbs within its loop, possibly reaching refcount zero and freeing the ubuf_info too soon.

The UDP stack currently also takes this extra ref, but does not need it as all skbs are sent after return from __ip(6)_append_data.

Avoid the extra refcount_inc and refcount_dec_and_test, and generally the sock_zerocopy_put in the common path, by passing the initial reference to the first skb.

This approach is taken instead of initializing the refcount to 0, as that would generate error "refcount_t: increment on 0" on the next skb_zcopy_set.

Signed-off-by Willem de Bruijn <[email protected]>

---
 include/linux/skbuff.h | 12 ++++++++----
 net/core/skbuff.c      |  9 +++++----
 net/ipv4/ip_output.c   | 10 +++++-----
 net/ipv4/tcp.c         |  2 +-
 net/ipv6/ip6_output.c  | 10 +++++-----
 5 files changed, 24 insertions(+), 19 deletions(-)

include/linux/skbuff.h +8 -4

index 04f52e719571..75d50ab7997c 100644
@@ -481,7 +481,7 @@ 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);

@@ -1326,10 +1326,14 @@ 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)) {
-		sock_zerocopy_get(uarg);
+		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;
 	}

@@ -1374,7 +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 +5 -4

index 435bac91d293..7cfc2144228a 100644
@@ -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;

@@ -1097,7 +1097,8 @@ void sock_zerocopy_put_abort(struct ubuf_info *uarg)
 		atomic_dec(&sk->sk_zckey);
 		uarg->len--;
 
-		sock_zerocopy_put(uarg);
+		if (have_uref)
+			sock_zerocopy_put(uarg);
 	}
 }
 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);

@@ -1137,7 +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);

@@ -1157,7 +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/ipv4/ip_output.c +5 -5

index 9602746d7175..08ff04d12642 100644
@@ -881,8 +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);

@@ -921,12 +921,13 @@ static int __ip_append_data(struct sock *sk,
 		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);
+			skb_zcopy_set(skb, uarg, &extra_uref);
 		}
 	}

@@ -1019,7 +1020,7 @@ static int __ip_append_data(struct sock *sk,
 			cork->tx_flags = 0;
 			skb_shinfo(skb)->tskey = tskey;
 			tskey = 0;
-			skb_zcopy_set(skb, uarg);
+			skb_zcopy_set(skb, uarg, &extra_uref);
 
 			/*
 			 *	Find where to start putting bytes.

@@ -1123,13 +1124,12 @@ static int __ip_append_data(struct sock *sk,
 
 	if (wmem_alloc_delta)
 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
-	sock_zerocopy_put(uarg);
 	return 0;
 
 error_efault:
 	err = -EFAULT;
 error:
-	sock_zerocopy_put_abort(uarg);
+	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 +1 -1

index 252048776dbb..444cdbff0638 100644
@@ -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 +5 -5

index 56f09c4c40dc..1cc9ae923093 100644
@@ -1258,7 +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) {

@@ -1327,12 +1327,13 @@ static int __ip6_append_data(struct sock *sk,
 		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);
+			skb_zcopy_set(skb, uarg, &extra_uref);
 		}
 	}

@@ -1458,7 +1459,7 @@ static int __ip6_append_data(struct sock *sk,
 			cork->tx_flags = 0;
 			skb_shinfo(skb)->tskey = tskey;
 			tskey = 0;
-			skb_zcopy_set(skb, uarg);
+			skb_zcopy_set(skb, uarg, &extra_uref);
 
 			/*
 			 *	Find where to start putting bytes

@@ -1561,13 +1562,12 @@ static int __ip6_append_data(struct sock *sk,
 
 	if (wmem_alloc_delta)
 		refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
-	sock_zerocopy_put(uarg);
 	return 0;
 
 error_efault:
 	err = -EFAULT;
 error:
-	sock_zerocopy_put_abort(uarg);
+	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);

Signature · 1 line

2.20.0.rc0.387.gc7a69e6b6c-goog

Willem de Bruijn

author

2018-11-30 07:44 UTC #4 lore raw

Re: [PATCH net-next v3 2/3] udp: elide zerocopy operation in hot path

Quoted 64 lines · Willem de Bruijn

From: Willem de Bruijn <[email protected]>

With MSG_ZEROCOPY, each skb holds a reference to a struct ubuf_info. Release of its last reference triggers a completion notification.

The UDP stack currently also takes this extra ref, but does not need it as all skbs are sent after return from __ip(6)_append_data.

Avoid the extra refcount_inc and refcount_dec_and_test, and generally the sock_zerocopy_put in the common path, by passing the initial reference to the first skb.

This approach is taken instead of initializing the refcount to 0, as that would generate error "refcount_t: increment on 0" on the next skb_zcopy_set.

Signed-off-by Willem de Bruijn <[email protected]>

---

net/ipv4/ip_output.c +4 -3

index 9602746d7175..08ff04d12642 100644
@@ -881,8 +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);

@@ -921,12 +921,13 @@ static int __ip_append_data(struct sock *sk,
                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);
+                       skb_zcopy_set(skb, uarg, &extra_uref);
                }
        }

@@ -1019,7 +1020,7 @@ static int __ip_append_data(struct sock *sk,
                        cork->tx_flags = 0;
                        skb_shinfo(skb)->tskey = tskey;
                        tskey = 0;
-                       skb_zcopy_set(skb, uarg);
+                       skb_zcopy_set(skb, uarg, &extra_uref);
 
                        /*
                         *      Find where to start putting bytes.

@@ -1123,13 +1124,12 @@ static int __ip_append_data(struct sock *sk,

net/ipv4/ip_output.c static int __ip_append_data(struct sock *sk, · line 1132

Quoted 10 lines

net/ipv4/ip_output.c +1 -2

        if (wmem_alloc_delta)
                refcount_add(wmem_alloc_delta, &sk->sk_wmem_alloc);
-       sock_zerocopy_put(uarg);
        return 0;
 
 error_efault:
        err = -EFAULT;
 error:
-       sock_zerocopy_put_abort(uarg);
+       sock_zerocopy_put_abort(uarg, extra_uref);

I'll need another revision. Sorry for the spam.

In the draft patch I suggested that the skb_zcopy_set needs to be moved below getfrag, so that the uarg is not freed on the only error path that calls kfree_skb inside the main loop.

This is still needed. Else sock_zerocopy_put_abort here is reached with a non-NULL, but already freed uarg.

Will send a v4. Will let this sit for at least a day in case anyone else has comments.

Willem de Bruijn

author

2018-11-30 07:33 UTC #5 lore raw

PATCH v3 3/3 selftests: extend zerocopy tests to udp

In reply to Willem de Bruijn

From: Willem de Bruijn <[email protected]>

Both msg_zerocopy and udpgso_bench have udp zerocopy variants. Exercise these as part of the standard kselftest run.

With udp, msg_zerocopy has no control channel. Ensure that the receiver exits after the sender by accounting for the initial delay in starting them (in msg_zerocopy.sh).

Signed-off-by Willem de Bruijn <[email protected]>

---
 tools/testing/selftests/net/msg_zerocopy.c  | 3 ++-
 tools/testing/selftests/net/msg_zerocopy.sh | 2 ++
 tools/testing/selftests/net/udpgso_bench.sh | 3 +++
 3 files changed, 7 insertions(+), 1 deletion(-)

tools/testing/selftests/net/msg_zerocopy.c +2 -1

index 406cc70c571d..4b02933cab8a 100644
@@ -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 +2 -0

index c43c6debda06..825ffec85cea 100755
@@ -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 +3 -0

index 0f0628613f81..5670a9ffd8eb 100755
@@ -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() {

Signature · 1 line

2.20.0.rc0.387.gc7a69e6b6c-goog