{
  "finding_id": "DF-4051",
  "status": "inconclusive",
  "reproduced": 0,
  "impact": "dos",
  "confidence": "certain",
  "verdict": "Mechanism PROVEN on the guest kernel by nfereplica.ko, a deterministic replica of if_nfe.cs exact callout/serializer protocol (non-MP callout_init if_nfe.c:636; nfe_tick blocking lwkt_serialize_enter :2187 with unconditional under-lock self re-arm :2190; nfe_stops synchronous callout_stop under the serializer :1593): the nfe_stop-shaped stop wedged permanently with the stopper in D2/costp (kern_timeout.c:920 sync wait of callout_stop) against the ticks softclock thread in slize (lwkt_serialize_enter) - console ends at the AB-BA marker, two samples confirm permanence, observing shell froze, guest could not shut down cleanly. nfe(4) runtime itself is hardware-gated (guest has no nForce MCP NIC; QEMU does not emulate forcedeth), so per DF-4025/DF-4046 family convention the persisted status is inconclusive. Local DoS: any ifconfig down/up/MTU/capchange/detach racing the 1 Hz tick permanently wedges two kernel threads and one CPUs callout subsystem.",
  "exploit_chain": "",
  "evidence": "run.log (ps: stopper pid 974 D2 costp vs softclock 3 slize; early-run softclock-4 starvation variant), serial_console.txt (console ends at IF THIS IS THE LAST nfereplica LINE, AB-BA DEADLOCK IS PROVEN after nfe_tick parked ... (fired=201 ran=200) - entering callout_stop()), build.log (-Werror clean .ko), VERDICT.md (full path:line trace)",
  "kernel_refs": [
    "sys/dev/netif/nfe/if_nfe.c:636",
    "sys/dev/netif/nfe/if_nfe.c:1591",
    "sys/dev/netif/nfe/if_nfe.c:1593",
    "sys/dev/netif/nfe/if_nfe.c:2187",
    "sys/dev/netif/nfe/if_nfe.c:2190",
    "sys/dev/netif/nfe/if_nfe.c:1446",
    "sys/dev/netif/nfe/if_nfe.c:669",
    "sys/kern/kern_timeout.c:920",
    "sys/kern/kern_timeout.c:1091"
  ],
  "poc_changes": "v1 used a tsleep-based park-wait in the replicas stop sysctl; its timeout is delivered by the same softclock the tick parks, so the stopper starved in its own aid (D4/nfetkwt, still a permanent wedge) - v2 switched the park-wait to a busy DELAY to isolate the true block point at callout_stop/costp. The if_nfe.c protocol itself is modeled verbatim.",
  "attempts": 2,
  "guest_uname": "DragonFly dfbsd 6.5-DEVELOPMENT DragonFly 6.5-DEVELOPMENT #0: Thu Jul  2 06:02:54 UTC 2026 root@dfbsd:/usr/obj/usr/src/sys/X86_64_GENERIC x86_64",
  "runtime_sec": 180,
  "guest_dirty": 0,
  "build_cmd": "ssh -F dfbsd-qemu/config dfbsd cd /root/nfereplica && make (bsd.kmod.mk, -Werror)",
  "run_cmd": "kldload /root/nfereplica/nfereplica.ko; sysctl kern.nfereplica.arm=1; sleep 2; (sysctl kern.nfereplica.stop=1 &); sleep 6; dmesg | grep nfereplica | tail; ps -o pid,stat,wchan,command -ax | grep -E stop=1|softclock",
  "code_hash": "e73a1a00fb0445223438372f2b967432e5dd1982a203ce4e21f5bf53bbefdf8b",
  "notes": "nfe(4) instance of the proven DF-3715(em)/DF-4025(re)/DF-4046(stge) AB-BA family. Guest reset with-src after the wedge (recovered healthy, env.txt); no AMD/NVIDIA hardware on guest.",
  "recommended_fix": "nfe_tick: lwkt_serialize_try + 1-tick retry instead of blocking lwkt_serialize_enter (same shape as the DF-3715/DF-4046 family fix); see fix.diff",
  "fix_status": "not_testable",
  "fix_baseline_reproduced": 0,
  "fix_patched_reproduced": 0,
  "fix_verdict": "In-driver fix validation impossible on this guest (no nForce hardware; QEMU does not emulate forcedeth). fix.diff applies the lwkt_serialize_try + retry shape whose negative control unloaded cleanly in the DF-3715/DF-4046 family verifications.",
  "fix_evidence": "findings/poc/DF-4051/fix.diff; family negative-control precedent in findings/poc/DF-4046/VERDICT.md"
}