p2p: improve TxOrphanage denial of service bounds #109
Description
mentioned in: bitcoin#31829
"> As soon as the victim package is spotted, the attacker sends their own "honest"-yet-evil cpfp packages, 404kWU total, at mempool minfee. The FIFO behavior kicks in, and the victim's child is dropped from orphanage.
I think @mzumsande made this point a few months ago as well - we don't have many guarantees if new orphans arrive much faster than we can process them. An attacker would need to pay fees for real transactions so that other peers are also sending these transactions. TLDR you could end up dropping transactions simply because the traffic is beyond the size of each peer's buffer.
Higher limits would help here (noted on 5-10x). But also, it seems like a good followup to assign different limits for outbounds vs inbounds? It seems advantageous for outbounds to take much longer to forget orphans." - glozow