Testing different scenarios in autopkgtest scripts, and especially "eofcat" Python that reads from the 9P mount in the guest, I experimented with setting a larger time.sleep(0.2) instead of 0.05, and separately, removing the O_NONBLOCK flag from its stdin FD. Neither showed any difference - the hang occurs.
I'm about to test with enabling internal.h's CONFIG_NETFS_DEBUG=y to capture more clues from the actual IO dispatch functions.
Testing different scenarios in autopkgtest scripts, and especially "eofcat" Python that reads from the 9P mount in the guest, I experimented with setting a larger time.sleep(0.2) instead of 0.05, and separately, removing the O_NONBLOCK flag from its stdin FD. Neither showed any difference - the hang occurs.
I'm about to test with enabling internal.h's CONFIG_ NETFS_DEBUG= y to capture more clues from the actual IO dispatch functions.