{
  "finding_id": "DF-2704",
  "status": "untested",
  "reproduced": 0,
  "impact": "none",
  "confidence": "speculative",
  "verdict": "vop_cache_operate_ap (sys/kern/vfs_vopops.c:1781-1782) loads and dispatches through mp->mnt_vn_journal_ops with no token or refcount; journal_detach (sys/kern/vfs_jops.c:247-252) via vfs_rm_vnodeops (sys/kern/vfs_init.c:143-162) NULLs and kfrees that same struct vop_ops, also unsynchronized. A racing VOP thread can VOCALL through freed M_VNODEOP memory (use-after-free function dispatch). The free side is reachable only by root (mountctl gated by SYSCAP_RESTRICTEDROOT, sys/kern/vfs_syscalls.c:1277-1283), and no in-tree workload detaches journals under I/O load, so this is a genuine protocol flaw with low practical exploitability. Phase V skipped per contract (Low/speculative/root-gated); would need a root mountctl storm plus scheduler injection to demo.",
  "exploit_chain": "",
  "evidence": [
    "sys/kern/vfs_vopops.c:1774-1786 unlocked mnt_vn_journal_ops read + VOCALL",
    "sys/kern/vfs_jops.c:247-252 journal_detach -> vfs_rm_vnodeops",
    "sys/kern/vfs_init.c:143-162 vfs_rm_vnodeops: *ops_pp=NULL then kfree, no lock",
    "sys/kern/vfs_syscalls.c:1277-1283 mountctl root gate",
    "findings/poc/DF-2704/VERDICT.md full protocol analysis"
  ],
  "kernel_refs": [
    "sys/kern/vfs_vopops.c:1774",
    "sys/kern/vfs_vopops.c:1781",
    "sys/kern/vfs_jops.c:247",
    "sys/kern/vfs_init.c:143",
    "sys/kern/vfs_mount.c:209",
    "sys/kern/vfs_syscalls.c:1277"
  ],
  "poc_changes": "no PoC built: race window is a single load-to-call sequence and the free side needs root; not reproducible honestly without scheduler injection",
  "attempts": 0,
  "guest_uname": "",
  "runtime_sec": 0,
  "guest_dirty": 0,
  "build_cmd": "",
  "run_cmd": "",
  "code_hash": "",
  "notes": "Pass-2 lock-protocol depth result. If promoted for verification later: root loop of mountctl install/remove journal on tmpfs while unprivileged writer hammers the mount. Fix direction: hold mnt_token (or refcount vop_ops) across the journal_ops read in vop_cache_operate_ap and across the swap+free in vfs_rm_vnodeops.",
  "recommended_fix": "Serialize the journal-ops swap with the dispatch layer (mnt_token around the mnt_vn_journal_ops read in vop_cache_operate_ap and the NULL+ kfree in vfs_rm_vnodeops), or refcount struct vop_ops before freeing."
}