How to reuse io_uring_buf_ring after a failed/incomplete send

[stonebrakert6]

TLDR:
Question: What is the status of io_uring_buf_ring's head in case io_uring_prep_send_bundle fails? Can I assume that head will be set to 0 in case of error?

Full read:
I am using ring with the following flags IORING_SETUP_SINGLE_ISSUER | IORING_SETUP_DEFER_TASKRUN , a ring will only be used by a single thread.

Use-case:

1. Maintain a pre-initialized pool of io_uring_buf_ring buffer-rings. I use buf_ring to "serialize" send data.
2. When a socket wants to send something, I Get a buffer-ring from the above pool
3. Add data(i.e buffers) in the buffer-ring. Buffers are always added to tail, while kernel code updates the head.
4. Then do the work i.e io_uring_prep_send_bundle, io_uring_submit etc.
5. On successful send I can Put the buffer-ring back to the pool for reuse with a different send. Because I know that head == tail?
6. But what happens if the send was only partial/fails? e.g I wanted to send 4MB data, but after sending 2 MB, the peer sent RST.

In this case if I try to reuse the buffer-ring for some other send i.e add buffers to it(which would always be added at tail), it might send some additional garbage(because when I added these buffers, head != tail).
Questions:

1. Is this understanding correct? Feel free to correct me if my understanding about usage is incorrect.
2. Technically I can do the following on send failure
   1. tail = io_uring_buf_ring_head
   2. Put buf-ring back to the pool

But that is a system call, is this the best we can do?

Can't the kernel code set head = 0 on failure(I don't know if it already does that or not)? And so we can call io_uring_buf_ring_init(which just sets tail to 0) and we're done?

Reference:
1431 and I don't think the following reply by Jens holds good for my use-case?

If whatever send(s) you have using the ring are not active, then you can manipulate the ring entries as you see fit.

[stonebrakert6]

Please let me know if my usage is incorrect or not natural/idiomatic.

[noteflakes]

FWIW it seems to me that buffer rings are more useful for multishot read/recv than for send. I did some benchmarks and vectorized send seems to have equivalent performance to send_bundle. It is of course much simpler to set up.