Re: [PATCH bpf v2 03/15] bpf, sockmap: Don't let an SK_PASS verdict cover the next message

[Jiayuan Chen <jiayuan.chen@linux.dev>]

On 11/4/26 8:28 AM, Alexei Starovoitov wrote:
> On Tue, Nov 03, 2026 at 01:37 PM netdev-bot+sashiko@kernel.org <netdev-bot+sashiko@kernel.org> wrote:
>>> +		if (!psock->apply_bytes)
>>> +			psock->eval = __SK_NONE;
>> When psock->apply_bytes is not 0, eval stays __SK_PASS during the push.
>> The budget is still debited only after tcp_bpf_push() returns. Can a
>> second sender on the same socket still get past the prog here?
> Yes, as today. It's already explained in the commit log.
>
>> But B's message is sized against the budget before A's in-flight bytes
>> are removed from it. Would it be better to reserve the budget before the
>> push? That means debiting tosend, and ending the verdict when the budget
>> reaches 0, before calling tcp_bpf_push(), instead of subtracting after a
>> failure.
> No. That breaks the single sender. After -EAGAIN the verdict has to
> cover the bytes that the push didn't send. A refund after the failed
> push can land on B's verdict.
>
>> The shared-cork problem with two PASS senders already existed. The kfree()
>> variant used to need a third sender to end a byte-limited verdict. The
>> early reset makes it reachable with only two senders.
> yeah. It's an old bug, not from this patch.
> Two senders are enough today too. B's first message goes out with A's
> verdict, and the code after the switch resets eval. The prog runs on B's
> second message. SK_REDIRECT frees the cork that A still pushes.
> The log of patch 13 says that a cork is not safe for two senders.
> Jiayuan proposed to serialize the senders in the reply to that patch.


I will pick up thoes pre-existing issues and organize others reported by sashiko.


>> With the early reset, can sender A's bytes be charged against a verdict
>> that another sender installed?
> Not a bug. eval, apply_bytes and sk_redir belong to the socket, not to
> a thread. B's verdict ends early and the prog runs again for the rest.