On 06/07/2019 17:29, wenxu wrote: > > 在 2019/7/6 20:02, Nikolay Aleksandrov 写道: >> On 05/07/2019 16:16, wenxu@xxxxxxxxx wrote: >>> From: wenxu <wenxu@xxxxxxxxx> >>> >>> This patch provide a meta vlan to set the vlan tag of the packet. >>> >>> for q-in-q outer vlan id 20: >>> meta vlan set 0x88a8:20 >>> >>> set the default 0x8100 vlan type with vlan id 20 >>> meta vlan set 20 >>> >>> Signed-off-by: wenxu <wenxu@xxxxxxxxx> >>> --- >>> include/net/netfilter/nft_meta.h | 5 ++++- >>> include/uapi/linux/netfilter/nf_tables.h | 4 ++++ >>> net/netfilter/nft_meta.c | 25 +++++++++++++++++++++++++ >>> 3 files changed, 33 insertions(+), 1 deletion(-) >>> >> The patch looks fine, just a note: you'll only be able to work with the >> outer tag, so guessing to achieve a double-tagged frame you'll have to >> add another NFT_META_VLAN_INNER(?) and will have to organize them one >> after another. > > yes, It's just set/mangle the meta vlan data. > I think it's a good idear for stacked tagged with NFT_META_VLAN_INNER, in > this case it should check the the proto of vlan tag already on packet > There are three case: > 1. packet already contain a vlan header with proto 0x88a8, it should push the inner vlan tag > 2. there is no outer vlan tag, maybe we should add a inner 0x8100 tag > and outer 0x88a8 tag? > 3. packet already contain a vlan tag with proto 0x8100, push the vlan tag and add outer 0x88a8 tag ? > I'm more inclined to make it simpler, i.e. always push inner vlan. That way you can arrange them anyway you like, also skb_vlan_push() already takes care of all that. It'll push the vlan if hwaccel is present or make it hwaccel if it isn't. In fact I'd suggest using skb_vlan_push(), so you don't need another NFT_META_VLAN_INNER, this one can be re-used. If you chain 2 of these you'll get properly double-tagged frame.
