Dirk Morris <dmor...@metaloft.com> wrote: > On 8/7/19 4:45 PM, Florian Westphal wrote: > > So Pablos suggestion above should work just fine. > > Dirk, can you spin a v2 with that change? > > > > Yes, will do tomorrow.
Thanks. > Also, just an idea, I also played around with just adding > u32 id to struct nf_conn and just calculating the hash inside > __nf_conntack_alloc when initialized or even lazily in nf_ct_get_id. > This seems to work fine and you don't have to worry about anything changing > and only calculate the hash once. > > I'm presuming this method was avoided for some reason, like keeping the struct > size to a minimum. Yes, exactly. If we go for storing id in the struct we could also just use a random value rather than computing a hash.