• Pablo Neira Ayuso's avatar
    netfilter: nft_reject_bridge: don't use IP stack to reject traffic · 523b929d
    Pablo Neira Ayuso authored
    
    
    If the packet is received via the bridge stack, this cannot reject
    packets from the IP stack.
    
    This adds functions to build the reject packet and send it from the
    bridge stack. Comments and assumptions on this patch:
    
    1) Validate the IPv4 and IPv6 headers before further processing,
       given that the packet comes from the bridge stack, we cannot assume
       they are clean. Truncated packets are dropped, we follow similar
       approach in the existing iptables match/target extensions that need
       to inspect layer 4 headers that is not available. This also includes
       packets that are directed to multicast and broadcast ethernet
       addresses.
    
    2) br_deliver() is exported to inject the reject packet via
       bridge localout -> postrouting. So the approach is similar to what
       we already do in the iptables reject target. The reject packet is
       sent to the bridge port from which we have received the original
       packet.
    
    3) The reject packet is forged based on the original packet. The TTL
       is set based on sysctl_ip_default_ttl for IPv4 and per-net
       ipv6.devconf_all hoplimit for IPv6.
    
    Signed-off-by: default avatarPablo Neira Ayuso <pablo@netfilter.org>
    523b929d