net: fix possible wrong checksum generation

Pravin Shelar mentioned that GSO could potentially generate
wrong TX checksum if skb has fragments that are overwritten
by the user between the checksum computation and transmit.

He suggested to linearize skbs but this extra copy can be
avoided for normal tcp skbs cooked by tcp_sendmsg().

This patch introduces a new SKB_GSO_SHARED_FRAG flag, set
in skb_shinfo(skb)->gso_type if at least one frag can be
modified by the user.

Typical sources of such possible overwrites are {vm}splice(),
sendfile(), and macvtap/tun/virtio_net drivers.

Tested:

$ netperf -H 7.7.8.84
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
7.7.8.84 () port 0 AF_INET
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 87380  16384  16384    10.00    3959.52

$ netperf -H 7.7.8.84 -t TCP_SENDFILE
TCP SENDFILE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.8.84 ()
port 0 AF_INET
Recv   Send    Send
Socket Socket  Message  Elapsed
Size   Size    Size     Time     Throughput
bytes  bytes   bytes    secs.    10^6bits/sec

 87380  16384  16384    10.00    3216.80

Performance of the SENDFILE is impacted by the extra allocation and
copy, and because we use order-0 pages, while the TCP_STREAM uses
bigger pages.

Reported-by: Pravin Shelar <pshelar@nicira.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

3 files changed, 18 insertions, 9 deletions

diff --git a/drivers/net/macvtap.c b/drivers/net/macvtap.c
index 0f0f9ce3a77..b181dfb3d6d 100644
--- a/drivers/net/macvtap.c
+++ b/drivers/net/macvtap.c
@@ -543,6 +543,7 @@ static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 		skb->data_len += len;
 		skb->len += len;
 		skb->truesize += truesize;
+		skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 		while (len) {
 			int off = base & ~PAGE_MASK;
@@ -598,7 +599,7 @@ static int macvtap_skb_from_vnet_hdr(struct sk_buff *skb,
 
 	if (vnet_hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
 		skb_shinfo(skb)->gso_size = vnet_hdr->gso_size;
-		skb_shinfo(skb)->gso_type = gso_type;
+		skb_shinfo(skb)->gso_type |= gso_type;
 
 		/* Header must be checked, and gso_segs computed. */
 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
diff --git a/drivers/net/tun.c b/drivers/net/tun.c
index c81680dc10e..293ce8dfc9e 100644
--- a/drivers/net/tun.c
+++ b/drivers/net/tun.c
@@ -1005,6 +1005,7 @@ static int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *from,
 		skb->data_len += len;
 		skb->len += len;
 		skb->truesize += truesize;
+		skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
 		atomic_add(truesize, &skb->sk->sk_wmem_alloc);
 		while (len) {
 			int off = base & ~PAGE_MASK;
@@ -1150,16 +1151,18 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 	}
 
 	if (gso.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		unsigned short gso_type = 0;
+
 		pr_debug("GSO!\n");
 		switch (gso.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 		case VIRTIO_NET_HDR_GSO_TCPV4:
-			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
+			gso_type = SKB_GSO_TCPV4;
 			break;
 		case VIRTIO_NET_HDR_GSO_TCPV6:
-			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
+			gso_type = SKB_GSO_TCPV6;
 			break;
 		case VIRTIO_NET_HDR_GSO_UDP:
-			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+			gso_type = SKB_GSO_UDP;
 			break;
 		default:
 			tun->dev->stats.rx_frame_errors++;
@@ -1168,9 +1171,10 @@ static ssize_t tun_get_user(struct tun_struct *tun, struct tun_file *tfile,
 		}
 
 		if (gso.gso_type & VIRTIO_NET_HDR_GSO_ECN)
-			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
+			gso_type |= SKB_GSO_TCP_ECN;
 
 		skb_shinfo(skb)->gso_size = gso.gso_size;
+		skb_shinfo(skb)->gso_type |= gso_type;
 		if (skb_shinfo(skb)->gso_size == 0) {
 			tun->dev->stats.rx_frame_errors++;
 			kfree_skb(skb);
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 701408a1ded..58914c8ea68 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -220,6 +220,7 @@ static void set_skb_frag(struct sk_buff *skb, struct page *page,
 	skb->len += size;
 	skb->truesize += PAGE_SIZE;
 	skb_shinfo(skb)->nr_frags++;
+	skb_shinfo(skb)->gso_type |= SKB_GSO_SHARED_FRAG;
 	*len -= size;
 }
 
@@ -379,16 +380,18 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 		 ntohs(skb->protocol), skb->len, skb->pkt_type);
 
 	if (hdr->hdr.gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		unsigned short gso_type = 0;
+
 		pr_debug("GSO!\n");
 		switch (hdr->hdr.gso_type & ~VIRTIO_NET_HDR_GSO_ECN) {
 		case VIRTIO_NET_HDR_GSO_TCPV4:
-			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
+			gso_type = SKB_GSO_TCPV4;
 			break;
 		case VIRTIO_NET_HDR_GSO_UDP:
-			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+			gso_type = SKB_GSO_UDP;
 			break;
 		case VIRTIO_NET_HDR_GSO_TCPV6:
-			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
+			gso_type = SKB_GSO_TCPV6;
 			break;
 		default:
 			net_warn_ratelimited("%s: bad gso type %u.\n",
@@ -397,7 +400,7 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 		}
 
 		if (hdr->hdr.gso_type & VIRTIO_NET_HDR_GSO_ECN)
-			skb_shinfo(skb)->gso_type |= SKB_GSO_TCP_ECN;
+			gso_type |= SKB_GSO_TCP_ECN;
 
 		skb_shinfo(skb)->gso_size = hdr->hdr.gso_size;
 		if (skb_shinfo(skb)->gso_size == 0) {
@@ -405,6 +408,7 @@ static void receive_buf(struct receive_queue *rq, void *buf, unsigned int len)
 			goto frame_err;
 		}
 
+		skb_shinfo(skb)->gso_type |= gso_type;
 		/* Header must be checked, and gso_segs computed. */
 		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
 		skb_shinfo(skb)->gso_segs = 0;