I won't apply this version, please send v2 with the changes below.

On Fri, 06 Nov 2026 05:02:13 +0200, Eva Kurchatova 
<[email protected]> wrote:
> Over memory limits, drop the incoming skb, relying on MPTCP-level
> retransmissions.

Item 1 of the cover letter reply: put mainline 8ab24fdebc369 ("mptcp:
close TOCTOU race while computing rcv_wnd") and da23be77e1292 ("mptcp:
allow subflow rcv wnd to shrink") before this patch.

Why: our tree lacks both. TCP never lets the subflow window shrink,
so data that TCP has acked but that is out of order at MPTCP level
inflates the MPTCP window we announce. A peer that behaves correctly
can then send more than our rcvbuf holds. Until now that excess
waited in the subflow queue; with this patch it is dropped and only
the sender's MPTCP retransmit timer brings it back. Mainline had the
two window fixes two months before it started to drop here.

-- 
Konstantin Khorenko,
Virtuozzo Linux Kernel Team

_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel

Reply via email to