Re: [PATCH net v2] tcp: reject devmem tx with fastopen and repair

[Pavel Begunkov <asml.silence@gmail.com>]

On 10/21/26 18:15, Mina Almasry wrote:
> On Wed, Oct 21, 2026 at 8:17 AM Pavel Begunkov <asml.silence@gmail.com> wrote:
...>>> diff --git a/net/ipv4/tcp.c b/net/ipv4/tcp.c
>>> index 87ef6d5cbfeb..2792923f0748 100644
>>> --- a/net/ipv4/tcp.c
>>> +++ b/net/ipv4/tcp.c
>>> @@ -1169,14 +1169,18 @@ int tcp_sendmsg_locked(struct sock *sk, struct msghdr *msg, size_t size)
>>>                        zc = MSG_SPLICE_PAGES;
>>>        }
>>>
>>> -     if (!sockc_err && sockc.dmabuf_id && (zc != MSG_ZEROCOPY || !binding)) {
>>> -             err = -EINVAL;
>>> +     if (sockc.dmabuf_id && (zc != MSG_ZEROCOPY || !binding)) {
>>
>> I wonder why it does sockc_err checks like that for devmem, e.g. instead of
>> failing earlier. Maybe Mina remembers?
> 
> IIRC, Paolo caught an issue where we were subtely changing behavior.
> Like if someone passes in a bad sockc flags, it's usually ignored, but
> if we process the sockc_err before we need it for devmem we change
> behavior for existing users that were passing bad sockc but not using
> devmem.

Got the idea, thanks. Sounds like this chunk can potentially change
it in the same way, though I'd need to look up that thread with
Paolo to be sure.

-- 
Pavel Begunkov