DF-2666 decisive guest run — stock INVARIANTS kernel #0
(DragonFly dfbsd 6.5-DEVELOPMENT #0: Thu Jul  2 06:02:54 UTC 2026 root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64)

Attempt 1 (sizing, aborted): ./df2666_trigger 2000 — ~1050 reconnect cycles
completed cleanly (~1 s/cycle: hz/2 teardown sleeps per kdmsg reconnect);
run aborted for runtime, mount unmounted cleanly, no guest damage.

Attempt 2 (decisive): ./df2666_trigger 200 — full output:

OPENFILES_BEFORE=-4294967199 iters=200
OPENFILES_AFTER=-4294967199 DELTA=0
RECLUSTER_OK=200 RECLUSTER_EINVAL=0 RECLUSTER_OTHER=0 LAST_ERRNO=0 (Undefined error: 0)
LEAK_INDICATOR: DELTA - 1 (one ref legally held by iocom->msg_fp) = -1
BADFD_RC=22 (Invalid argument) [expected EINVAL, no leak]
UNMOUNT_OK
OPENFILES_POST_UNMOUNT=-4294967201 (FINAL_DELTA=-2)
TRIGGER_DONE

(kern.openfiles is int-backed and has wrapped negative on this long-lived
guest; the value is still live — it moved 0 during the 200 calls and -2
across unmount — so the deltas are meaningful.)

New console lines during the run: 610 (mount banner + 200 cycles of
"HAMMER2: VOLDATA DUMP" / "HAMMER2: INITIATE SPANs" / "HAMMER2: CONN WAS
TERMINATED" — the teardown-simulated CONN failure re-entering
hammer2_autodmsg's LNK_CONN|CREATE|REPLY block via the state's icmd,
kern_dmsg.c:1050-1058; no panics, no wedges).

Post-run kernel state: kern.openfiles: 93; vmstat -m "file" zone at rest
(99 allocs, no growth from the 200 successful reconnects).

Read of the leak branch: NOT ENTERED (RECLUSTER_EINVAL=0) — see VERDICT.md.