sendto(), recvfrom()

sendto(), recvfrom()

[Nathan Ringo]

Would adding IORING_OP_{SENDTO,RECVFROM} be a reasonable first kernel
contribution? I'd like to write a program with io_uring that's listening
on a UDP socket, so recvfrom() at least is important to my use-case.

--
Nathan Ringo

Re: sendto(), recvfrom()

[Jens Axboe]

On 6/26/20 2:47 PM, Nathan Ringo wrote:
> Would adding IORING_OP_{SENDTO,RECVFROM} be a reasonable first kernel
> contribution? I'd like to write a program with io_uring that's listening
> on a UDP socket, so recvfrom() at least is important to my use-case.

Certainly. When you hack it up, be sure to also provide a liburing
regression test case for test/ that tests the feature.

-- 
Jens Axboe

Re: sendto(), recvfrom()

[Jann Horn]

On Fri, Jun 26, 2020 at 10:47 PM Nathan Ringo <nathan@remexre.xyz> wrote:
> Would adding IORING_OP_{SENDTO,RECVFROM} be a reasonable first kernel
> contribution? I'd like to write a program with io_uring that's listening
> on a UDP socket, so recvfrom() at least is important to my use-case.

What is the benefit compared to IORING_OP_RECVMSG and
IORING_OP_SENDMSG, which already exist (and provide a superset of
sendto/recvfrom)?

Re: sendto(), recvfrom()

[Nathan Ringo]

On Friday, June 26, 2020 4:08 PM, Jann Horn <jannh@google.com> wrote:
> What is the benefit compared to IORING_OP_RECVMSG and
> IORING_OP_SENDMSG, which already exist (and provide a superset of
> sendto/recvfrom)?

Ah, I didn't realize this was the case, whoops. I guess I'm showing off
my lack of familiarity with Unix; I remembered sendmsg as being specific
to Unix domain sockets. I'll just use that, then.

--
Nathan Ringo