[PATCH net v2] net: fix wrong mac_len calculation for vlans

[Nikolay Aleksandrov <nikolay@redhat.com>]

After 1e785f48d29a ("net: Start with correct mac_len in
skb_network_protocol") skb->mac_len is used as a start of the
calculation in skb_network_protocol() but that is not always correct. If
skb->protocol == 8021Q/AD, usually the vlan header is already inserted
in the skb (i.e. vlan reorder hdr == 0). Usually when the packet enters
dev_hard_xmit it has mac_len == 0 so we take 2 bytes from the
destination mac address (skb->data + VLAN_HLEN) as a type in
skb_network_protocol() and return vlan_depth == 4. In the case where TSO is
off, then the mac_len is set but it's == 18 (ETH_HLEN + VLAN_HLEN), so
skb_network_protocol() returns a type from inside the packet and
offset == 22. Also make vlan_depth unsigned as suggested before.
As suggested by Eric Dumazet, move the while() loop in the if() so we
can avoid additional testing in fast path.

Here are few netperf tests + debug printk's to illustrate:
cat netperf.tso-on.reorder-on.bugged
- Vlan -> device (reorder on, default, this case is okay)
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
192.168.3.1 () 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    7111.54
[   81.605435] skb->len 65226 skb->gso_size 1448 skb->proto 0x800
skb->mac_len 0 vlan_depth 0 type 0x800

- Vlan -> device (reorder off, bad)
cat netperf.tso-on.reorder-off.bugged
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
192.168.3.1 () 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     241.35
[  204.578332] skb->len 1518 skb->gso_size 0 skb->proto 0x8100
skb->mac_len 0 vlan_depth 4 type 0x5301
0x5301 are the last two bytes of the destination mac.

And if we stop TSO, we may get even the following:
[   83.343156] skb->len 2966 skb->gso_size 1448 skb->proto 0x8100
skb->mac_len 18 vlan_depth 22 type 0xb84
Because mac_len already accounts for VLAN_HLEN.

After the fix:
cat netperf.tso-on.reorder-off.fixed
MIGRATED TCP STREAM TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to
192.168.3.1 () 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    10.01    5001.46
[   81.888489] skb->len 65230 skb->gso_size 1448 skb->proto 0x8100
skb->mac_len 0 vlan_depth 18 type 0x800

CC: Vlad Yasevich <vyasevic@redhat.com>
CC: Eric Dumazet <eric.dumazet@gmail.com>
CC: Daniel Borkman <dborkman@redhat.com>
CC: David S. Miller <davem@davemloft.net>

Fixes:1e785f48d29a ("net: Start with correct mac_len in
skb_network_protocol")
Signed-off-by: Nikolay Aleksandrov <nikolay@redhat.com>
---
v2: Move the while() loop in the if() to avoid additional testing in fast
path as suggested by Eric Dumazet.

Note: A good side-effect of this patch is that in the mac_len > 0
case we'll directly look in the last vlan header in the while loop later
(mac_len - VLAN_HLEN), so it should take only a single iteration.
Also I'm working on getting mac_len always correct for net-next, so we can
get rid of this later. My plan currently is to make dev_hard_header()
adjust mac_len according to the return value from create(), but I'll have
to look into if that's enough and if it's feasible. Any suggestions on this
are very welcome.

 net/core/dev.c | 35 +++++++++++++++++++++++++----------
 1 file changed, 25 insertions(+), 10 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index 9abc503b19b7..fb8b0546485b 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -2283,8 +2283,8 @@ EXPORT_SYMBOL(skb_checksum_help);
 
 __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
 {
+	unsigned int vlan_depth = skb->mac_len;
 	__be16 type = skb->protocol;
-	int vlan_depth = skb->mac_len;
 
 	/* Tunnel gso handlers can set protocol to ethernet. */
 	if (type == htons(ETH_P_TEB)) {
@@ -2297,15 +2297,30 @@ __be16 skb_network_protocol(struct sk_buff *skb, int *depth)
 		type = eth->h_proto;
 	}
 
-	while (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
-		struct vlan_hdr *vh;
-
-		if (unlikely(!pskb_may_pull(skb, vlan_depth + VLAN_HLEN)))
-			return 0;
-
-		vh = (struct vlan_hdr *)(skb->data + vlan_depth);
-		type = vh->h_vlan_encapsulated_proto;
-		vlan_depth += VLAN_HLEN;
+	/* if skb->protocol is 802.1Q/AD then the header should already be
+	 * present at mac_len - VLAN_HLEN (if mac_len > 0), or at
+	 * ETH_HLEN otherwise
+	 */
+	if (type == htons(ETH_P_8021Q) || type == htons(ETH_P_8021AD)) {
+		if (vlan_depth) {
+			if (unlikely(WARN_ON(vlan_depth < VLAN_HLEN)))
+				return 0;
+			vlan_depth -= VLAN_HLEN;
+		} else {
+			vlan_depth = ETH_HLEN;
+		}
+		do {
+			struct vlan_hdr *vh;
+
+			if (unlikely(!pskb_may_pull(skb,
+						    vlan_depth + VLAN_HLEN)))
+				return 0;
+
+			vh = (struct vlan_hdr *)(skb->data + vlan_depth);
+			type = vh->h_vlan_encapsulated_proto;
+			vlan_depth += VLAN_HLEN;
+		} while (type == htons(ETH_P_8021Q) ||
+			 type == htons(ETH_P_8021AD));
 	}
 
 	*depth = vlan_depth;
-- 
1.8.5.3