Occasional rndc crashes in tcp_connect_cb() on OpenBSD

Even with !4089 (merged) merged, we still seem to be getting occasional rndc crashes on OpenBSD:

While backtrace collection in GitLab CI is broken for OpenBSD¹, the core dumps themselves seem to be usable once you load them into GDB with the right executables.

Here are some bits I personally found important:

(gdb) bt
#0  0x000005ee0feb049f in uv_handle_get_data () from /usr/local/lib/libuv.so.2.1
#1  0x000005ee6570b50e in tcp_connect_cb (uvreq=0x5edcf49e378, status=-48) at netmgr/tcp.c:156
#2  0x000005ee6570b408 in isc__nm_async_tcpconnect (ev0=Variable "ev0" is not available.
) at netmgr/tcp.c:136
#3  0x000005ee6570abbe in process_queue (worker=0x5edec985000, queue=0x5edec986080) at netmgr/netmgr.c:618
#4  0x000005ee0feb4548 in uv__async_io () from /usr/local/lib/libuv.so.2.1
#5  0x000005ee0fec59dc in uv__io_poll () from /usr/local/lib/libuv.so.2.1
#6  0x000005ee0feb4be0 in uv_run () from /usr/local/lib/libuv.so.2.1
#7  0x000005ee657077aa in nm_thread (worker0=0x5edec985000) at netmgr/netmgr.c:488
...
(gdb) frame 1
#1  0x000005ee6570b50e in tcp_connect_cb (uvreq=0x5edcf49e378, status=-48) at netmgr/tcp.c:156
156		sock = uv_handle_get_data((uv_handle_t *)uvreq->handle);
(gdb) print uvreq
$3 = (uv_connect_t *) 0x5edcf49e378
(gdb) print uvreq->handle
$4 = (uv_stream_t *) 0x0

Output of thread apply all bt full attached anyway.

That is because source binary identification in GDB on OpenBSD works differently than on all other systems and I see no obvious way to make it work. ↩

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information