{
  "finding_id": "DF-4329",
  "status": "reproduced",
  "reproduced": 1,
  "impact": "panic",
  "confidence": "certain",
  "verdict": "REPRODUCED on the live guest kernel (sequenced protocol replica, no GPU needed): every layer of the i915 request-object RCU lifetime protocol is a no-op on DFly - synchronize_rcu() is an EMPTY MACRO, cond_synchronize_rcu is cpu_mfence, get_state_synchronize_rcu is 0 (rcupdate.h:94-96), SLAB_TYPESAFE_BY_RCU is ignored (kmem_cache==generic kmalloc(M_DRM)), and dma_fence_get_rcu is an unconditional kref_get. Demonstrated: freed request chunk instantly reused by an unrelated allocation (same address), refcount resurrection 0->1, and the __i915_gem_active_get_rcu reader chasing the planted engine marker 0x5151515151515151 -> kernel page fault at exactly 0x515151515169 (run_replica+0x350). On Intel-GPU systems this is an unprivileged-reader UAF (GEM_BUSY) with attacker-influenced pointer deref + foreign refcount corruption; impact classed panic (corruption primitive; hw-gated for full chain).",
  "exploit_chain": "unpriv GL client: (1) execbuf churn allocates/retires requests; (2) i915_gem_busy_ioctl reader takes active->request locklessly; (3) request retires -> kmem_cache_free -> chunk instantly reused by attacker-influenced drm allocation; (4) reader chases rq->engine from reused contents -> chosen-address kernel fault (demonstrated) / get_rcu resurrects or strays a refcount on foreign memory -> UAF substrate",
  "evidence": "run.log (mode1: reuse+resurrection+chase-target; mode2: panic), panic.txt (fault virtual address = 0x515151515169), i915_rcu_harness.c, fix.diff",
  "kernel_refs": [
    "sys/dev/drm/i915/i915_request.c:732",
    "sys/dev/drm/i915/i915_request.c:741",
    "sys/dev/drm/i915/i915_request.c:761",
    "sys/dev/drm/i915/i915_request.h:514",
    "sys/dev/drm/i915/i915_request.h:318",
    "sys/dev/drm/include/linux/rcupdate.h:94",
    "sys/dev/drm/include/linux/slab.h:103",
    "sys/dev/drm/include/linux/dma-fence.h:98"
  ],
  "poc_changes": "authored from scratch (none seeded): sequenced replica of __i915_gem_active_get_rcu under the ports rcu_read_lock (crit_enter), publishing via plain store (== rcu_assign_pointer shim), freeing via kfree (== kmem_cache_free shim), reusing via same-zone kmalloc",
  "attempts": 3,
  "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": 1500,
  "guest_dirty": 0,
  "build_cmd": "cd /root/kmod/i915rcu && make SYSDIR=/usr/src/sys",
  "run_cmd": "kldload i915_rcu_harness.ko && sysctl kern.i915rcu.run=1 && sysctl kern.i915rcu.run=2",
  "code_hash": "da229cb6595f4a6ea97c03bf551d14d127805a07f323c968fe4a313d14ebc528",
  "notes": "panic run reset afterwards; distinct from DF-1881 (reservation readers) - this is the request-slab protocol + allocator/grace-period layers; guest has no Intel GPU so full uid0 chain not exercisable",
  "recommended_fix": "implement the shim: map synchronize_rcu/cond_synchronize_rcu/get_state_synchronize_rcu to DragonFlys native RCU and make dma_fence_get_rcu use kref_get_unless_zero",
  "fix_status": "not_testable",
  "fix_baseline_reproduced": 1,
  "fix_patched_reproduced": 0,
  "fix_verdict": "fix edits drm shim headers compiled into the drm/i915 module set; guest has no Intel GPU to load i915.ko and re-running the replica would validate the replica, not the driver path",
  "fix_evidence": "fix.diff authored (rcupdate.h native-RCU mapping + dma-fence.h unless-zero)"
}