Re: [PATCH net-next 17/27] OVS: remove assumptions about VLAN_TAG_PRESENT bit

On Tue, Dec 13, 2016 at 11:40:10AM +0100, Jiri Benc wrote:

On Tue, 13 Dec 2016 01:12:38 +0100 (CET), Michał Mirosław wrote:

quoted

@@ -850,20 +848,11 @@ static int validate_vlan_from_nlattrs(const struct sw_flow_match *match,
 		return -EINVAL;
 	}
 
-	if (a[OVS_KEY_ATTR_VLAN])
-		tci = nla_get_be16(a[OVS_KEY_ATTR_VLAN]);
-
-	if (!(tci & htons(VLAN_TAG_PRESENT))) {
-		if (tci) {
-			OVS_NLERR(log, "%s TCI does not have VLAN_TAG_PRESENT bit set.",
-				  (inner) ? "C-VLAN" : "VLAN");
-			return -EINVAL;
-		} else if (nla_len(a[OVS_KEY_ATTR_ENCAP])) {
-			/* Corner case for truncated VLAN header. */
-			OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
-				  (inner) ? "C-VLAN" : "VLAN");
-			return -EINVAL;
-		}
+	if (!a[OVS_KEY_ATTR_VLAN] && nla_len(a[OVS_KEY_ATTR_ENCAP])) {
+		/* Corner case for truncated VLAN header. */
+		OVS_NLERR(log, "Truncated %s header has non-zero encap attribute.",
+			(inner) ? "C-VLAN" : "VLAN");
+		return -EINVAL;

This looks wrong. The zero value of nla_get_be16(a[OVS_KEY_ATTR_VLAN])
together with empty a[OVS_KEY_ATTR_ENCAP] means truncated VLAN header
and this needs to be preserved.

In other words, you need to check also !nla_get_be16(a[OVS_KEY_ATTR_VLAN])
here.


Overall, this whole patch looks very suspicious from the very
beginning. The xors used are strong indication that something is not
right.

And indeed, you're breaking uAPI compatibility. Previously, the
VLAG_TAG_PRESENT bit set in OVS_KEY_ATTR_VLAN caused the flow to match
on packets with VLAN tag present. After this patch, it causes the flow
to match only on those packets that have a certain value of
VLAN_CFI_MASK in their VLAN tag (I haven't bother deciphering what
value is checked after all these xors, it's as well possible that it
will also match on packets _without_ any VLAN header).

You have to preserve the old meaning of VLAN_TAG_PRESENT in
OVS_KEY_ATTR_VLAN. When doing flow lookups, VLAN_TAG_PRESENT must match
on and only on packets that have VLAN tag present, irrespective of their
VLAN_CFI_MASK.

If you want to introduce support for lookups on VLAN_CFI_MASK, you'll
have to do that by introducing a new netlink attribute.

Hmm. In that case I'll just mask the CFI bit on entry to OVS.
Otherwise it's for me like to stab in a dark at a large beast
I know nothing about.

Best Regards,
Michał Mirosław