How io_uring_prep_send_bundle works?

[AomaYple]

Forgive my poor code reading skills, I don't understand the liburing example very well.

I think it works like this. After creating an io_uring_buf_ring, the sent data is added via io_uring_buf_ring_add and then advanced with io_uring_buf_ring_advance.

Then call io_uring_prep_send_bundle, set the buf_group and IOSQE_BUFFER_SELECT.

The io_uring_prep_send_bundle will keep sending the buffers in the io_uring_buf_ring until all have been sent.

Is that correct?

[axboe]

That is correct, or until the socket is full and it can't send more.

[AomaYple]

That is correct, or until the socket is full and it can't send more.

With io_uring_prep_send_bundle, it seems that each socket has to have an io_uring_buf_ring.

Because that's the only way to do it without misplacing the data. If one io_uring_buf_ring is shared, it's possible that data from one socket is being sent to another socket

[axboe]

Sure, if you are using it to serialize writes on a single socket, you'd want one per socket. For the recv side, using per-size grouping makes more sense.