⬢ DragonFlyBSD Kernel Audit
DF-4395 / verdict.json
← back to finding ↓ download raw 
 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
{
  "finding_id": "DF-4395",
  "status": "reproduced",
  "reproduced": 1,
  "impact": "corruption",
  "confidence": "certain",
  "verdict": "Heap OOB write in get_dcefclk_voltage_dependency_table (vega10_processpptables.c:695-755): on the pioneer SKU (PCI 0x6863 rev 0, last entry <90000) the destination kzalloc is sized min(ucNumEntries+1, 8) records (28 B each = 228 B max) but the copy loop at :740 iterates the raw firmware ucNumEntries — with a crafted VBIOS table of ucNumEntries>8 the loop writes (ucNumEntries-8)*28 = up to 6916 bytes of VBIOS-controlled records (arbitrary u32 ulClk + byte ucVddInd every 28 bytes) linearly past the kernel heap object. Reproduced on the audit guest with a faithful lifted-copy harness (DF-1179 precedent for the same driver family): guard smashed from ucNumEntries=9 onward, 6916 B past at 255; negative controls (different device id / revision != 0) and the one-line patched loop leave the guard intact. In-kernel path hardware-gated (no AMD GPU on the guest, run.log).",
  "exploit_chain": "",
  "evidence": "run.log (guard smashed at n=9: 5+ bytes; n=255: writes through byte 7144 of a 228-byte object, first OOB record clk=0x3f0; PATCHED control intact); dcefclk_oflow.c (verbatim replica of :718-750); fix.diff (git apply --check OK)",
  "kernel_refs": [
    "sys/dev/drm/amd/powerplay/hwmgr/vega10_processpptables.c:695",
    "sys/dev/drm/amd/powerplay/hwmgr/vega10_processpptables.c:721",
    "sys/dev/drm/amd/powerplay/hwmgr/vega10_processpptables.c:729",
    "sys/dev/drm/amd/powerplay/hwmgr/vega10_processpptables.c:740",
    "sys/dev/drm/amd/powerplay/hwmgr/vega10_pptable.h:181",
    "sys/dev/drm/amd/powerplay/hwmgr/hwmgr_ppt.h:31",
    "sys/dev/drm/amd/amdgpu/amdgpu_pm.c:1007"
  ],
  "poc_changes": "authored fresh this run (pass 2); no seed PoC existed",
  "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": 240,
  "guest_dirty": 0,
  "build_cmd": "./build.sh (guest cc -O2 -Wall dcefclk_oflow.c)",
  "run_cmd": "./run.sh",
  "code_hash": "fbf82496194a1c74950628d5f72b0ce5a8f104769b80ff15fb17d934dba6c72d",
  "notes": "reproduced via lifted-copy harness per DF-1179 convention (in-kernel path needs an AMD Vega10 GPU with device id 0x6863 rev 0 — absent on guest; hardware gate in run.log). Upstream-equivalent fix: the copy loop is bounded by num_entries in current Linux. cve_match suggestion: equivalent of the upstream num_entries loop fix.",
  "recommended_fix": "bound the dcefclk copy loop by num_entries (fix.diff, one line)",
  "fix_status": "not_testable",
  "fix_baseline_reproduced": 1,
  "fix_patched_reproduced": 0,
  "fix_verdict": "Patched kernel not runnable on the guest (no ASIC). fix.diff is git-apply-able; the harness PATCHED control (loop bounded by num_entries) shows the overflow is gone for the same crafted input, and the DF-4396 central fix (which subsumes this one-liner) syntax-validizes with 0 errors on the guest kernel toolchain (build.log).",
  "fix_evidence": "fix.diff; run.log PATCHED control; findings/poc/DF-4396/build.log (0 syntax errors)",
  "fix_kernel_uname": "DragonFly dfbsd 6.5-DEVELOPMENT #0 x86_64 (syntax validation only)"
}