On Jul 10, 2014, at 2:40 AM, Thomas Graf <tg...@noironetworks.com> wrote:
> On 07/09/14 at 03:22pm, Daniele Di Proietto wrote:
>> match_validate() enforce that a mask matching on NDP attributes has also an
>> exact match on ICMPv6 type.
>> The ICMPv6 type, which is 8-bit wide, is stored in the 'tp.src' field of
>> 'struct sw_flow_key', which is 16-bit wide.
>> Therefore, an exact match on ICMPv6 type should only check the first 8 bits.
>>
>> This commit fixes a bug that prevented flows with an exact match on NDP field
>> from being installed
>>
>> Signed-off-by: Daniele Di Proietto <ddiproie...@vmware.com>
>
> Using 0xff is obviously correct but wouldn't this break ABI for users with
> existing 0xffff masks? How about allowing both?
It shouldn’t be possible at all for the mask to be 0xffff.
In the ICMPv6 case, we create the mask from the netlink attribute
with this line:
(I’m assuming this is the only place where we fill a sw_flow_key mask, is that
right?)
datapath/flow_netlink.c:786
SW_FLOW_KEY_PUT(match, tp.src,
htons(icmpv6_key->icmpv6_type), is_mask);
This cannot set tp.src to htons(0xffff), because icmpv6_key->icmpv6_type is
__u8.
The bug has always been there, since a1c564be (datapath: Mega flow
implementation).
datapath/flow.c:201
if (match->mask && (match->mask->key.ipv6.tp.src == htons(0xffff)))
datapath/flow.c:1437
SW_FLOW_KEY_PUT(match, ipv6.tp.src,
htons(icmpv6_key->icmpv6_type), is_mask);
Does this address your concern?
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
