Florian Westphal wrote: > Willem de Bruijn <willemdebruijn.kernel@xxxxxxxxx> wrote: > > > I named the copypasta as nf_skb_get_hash. If placed in sk_buff.h: > > > net_get_hash_net()? > > > skb_get_hash()? > > > > Still passing an skb too, so skb_get_hash_net()? > > Sounds good to me. > > > > And if either of that exists, maybe then use > > > skb_get_hash_symmetric_net(net, skb) > > > > If symmetric is equally good for nft, that would be preferable, as it > > avoids the extra function. But I suppose it aliases the two flow > > directions, which may be exactly what you don't want? > > It would actually be fine, but the more important part is that > skb->hash is set. > > For the trace id, I want a stable identifier that won't change > (e.g. when nat rewrites addresses). > > This currently works because skb_get_hash computes it at most once. Probably not relevant to these skbs, that don't have an skb->sk. But in case skbs coming from the TCP stack are also in scope: can sk_rethink_txhash cause problems? > skb_get_hash_symmetric_net() will be used from nft_hash.c as > __skb_get_hash_symmetric "replacement". > > Pablo, you can drop this patch, I will try the 'pass net to dissector' > route.
