beldzhang

> That's great, I'd love to hear about your use case. TCP? What is your usual payload size per send (e.g. io_uring request)? > heads up, there is some discussion...

`kernel....: 5.19.5` `liburing..: 2.3-git` tested with READ and READ_FIXED, it's the same. after change `IOSQE_IO_LINK` to `IOSQE_IO_HARDLINK`, this will **NOT** happen. each `ECANCELED` will cause following 2 SQEs failed, SEND...

found notification slot was removed from both kernel and liburing after some checking until 42ee2bd, there is no any flag to check send_zc was support by current running kernel. is...

changed to `io_uring_get_probe()` to check sendzc support, thanks to Jens will testing new api with 6.0-rc4 this week, and will check is there any performance lost with linked-timeout. -- for...

> The first one is the usual request completion with IORING_CQE_F_MORE set in this one cqe.res carry out the right number of sent > and the second one is the...

oh... sorry for not describe clearly what want to say is the "protocol handling", in one request-response ``` recv-protocol-head if (cqe.res == head_length) send-head-and-data if (cqe.res == required_length) next_loop ```...

> Tried it out but doesn't reproduce for me, and I don't immediately see anything wrong. my sda is a USB-SSD disk, and re-tested with a nvme disk on 6.0-rc4,...

very happy to help the community, My Name: Beld Zhang E-Mail: [email protected] thanks to all for the hard works !

> Thanks, I've updated the commit and also included a link to this issue. patched [4d9cb92](https://git.kernel.dk/cgit/linux-block/commit/?h=io_uring-6.0&id=4d9cb92ca41dd8e905a4569ceba4716c2f39c75a) to 6.0-rc4, my testing are ok now. next will check/test error handling of this...

> It might also be a latency problem... yes, it is the latency... post that message too rush before found out in my hardware environment. using a usb-ethernet that has...