The fill-up logic in iofd_txqueue_dequeue() was incorrectly decrementing next->io_len twice: once when removing a buffer that didn't fit, and again when assigning to msg_iovlen. This may cause next->io_len to reach 0 and trigger an assert at osmo_io.c:274:
Assert failed next->io_len > 0
Add separate osmo_io_backpressure_test to verify queue management under network backpressure conditions. Use separate .ok files for poll vs io_uring backends since they produce different I/O patterns. This unit test reproduces the assert when the fix is not applied.
Change-Id: Ia96e25d11d8d683fa9dd0c56bdac257992edd782 Fixes: 3c2a02db4 ("osmo-io: Put together message buffers when dequeued from tx queue") Related: OS#6997
