Jakub Sitnicki wrote:

On Mon, Jul 19, 2021 at 11:48 PM CEST, John Fastabend wrote:

quoted

We don't want strparser to run and pass skbs into skmsg handlers when
the psock is null. We just sk_drop them in this case. When removing
a live socket from map it means extra drops that we do not need to
incur. Move the zap below strparser close to avoid this condition.

This way we stop the stream parser first stopping it from processing
packets and then delete the psock.

Fixes: a136678c0bdbb ("bpf: sk_msg, zap ingress queue on psock down")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
---

To confirm my understanding - the extra drops can happen because
currently we are racing to clear SK_PSOCK_TX_ENABLED flag in
sk_psock_drop with sk_psock_verdict_apply, which checks the flag before
pushing skb onto psock->ingress_skb queue (or possibly straight into
psock->ingress_msg queue on no redirect).

Correct. If strparser hands a skb to the sk_psock_* handlers then before
they enqueue the packet the flag is checked and the packet will be
dropped in this case. I noticed this while testing. So rather than
letting packets get into sk_psock_* handlers this patch just stops the
strparser.