Re: [PATCH][RFC]: ingress socket filter by mark

[Eric Dumazet <eric.dumazet@gmail.com>]

jamal a écrit :
> Maciej forced me to dig into this ;->
> 
> at the socket level if a packet arrives with a different mark than
> what we bind to, drop it. I have tested this patch and it drops a packet
> with mismatching mark.
> 
> There are several approaches - and i think the patch suggestion i have
> made here maybe too strict. I assume that if someone binds to a mark,
> they want to not only send packets with that mark but receive
> only if that mark is set. 
> A looser check would be something along the line accept as well if mark
> is not set i.e
> if (sk->sk_mark && skb->mark && sk->sk_mark != skb->mark)
> 
> Alternatively i could add one bit in the socket flags and have it so
> that check is made only if app has been explicit:
> if (sock_flag(sk, SOCK_CHK_SOMARK) && sk->sk_mark != skb->mark) drop
> 
> Another approach  is to set sock filter from app. I dont like this
> approach because it will be the least usable from app level and would be
> the least simple from kernel level.
> 
> cheers,
> jamal
> 

I vote for extending BPF, and not adding the price of a compare
for each packet. Only users wanting mark filtering should pay the price.