⬢ DragonFlyBSD Kernel Audit
DF-3667 / 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
{
  "finding_id": "DF-3667",
  "status": "inconclusive",
  "reproduced": 0,
  "impact": "none",
  "confidence": "certain",
  "verdict": "Source-level certain, runtime-inconclusive: ix_rxeof (sys/dev/netif/ix/if_ix.c:2974,3031,3040,3044) publishes the NIC-DMAd 16-bit descriptor length as m_len/m_pkthdr.len with no bound against the posted cluster size (rx_mbuf_sz, MCLBYTES=2048 or MJUMPAGESIZE=4096, if_ix.c:2635-2638,4142). A malicious/faulty NIC writing wb.upper.length=0xFFFF yields an mbuf claiming 65535 bytes in a 2048-byte cluster -> up to 63,487 bytes of adjacent kernel heap read by the network stack (info leak; same family as DF-3604/3639/3663). The audit QEMU guest has no ixgbe PCI device (env.txt), so the kernel path cannot execute there; a userspace harness (df3667_sim.c, run.log) proves the bound math. fix.diff compiles clean (-Werror, build_patched.log) but runtime fix validation is impossible without hardware.",
  "exploit_chain": "",
  "evidence": "VERDICT.md (path:line walkthrough); run.log (harness: desc len=0xffff -> OOB span 63487 for 2K cluster); build_baseline.log vs build_patched.log (module builds with the fix under -Werror); env.txt (device absent, driver built into kernel)",
  "kernel_refs": [
    "sys/dev/netif/ix/if_ix.c:2974",
    "sys/dev/netif/ix/if_ix.c:3031",
    "sys/dev/netif/ix/if_ix.c:3040",
    "sys/dev/netif/ix/if_ix.c:3044",
    "sys/dev/netif/ix/if_ix.c:2635",
    "sys/dev/netif/ix/if_ix.c:4142",
    "sys/dev/netif/ix/if_ix.h:101"
  ],
  "poc_changes": "No seed existed (pass-2 new finding). Harness written from scratch; first fix.diff hunk placed the guard before eop is computed and the -Werror module build flagged maybe-uninitialized -> moved after the --count line.",
  "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": 480,
  "guest_dirty": 0,
  "build_cmd": "cd /usr/src/sys/dev/netif/ix && make obj && make   (guest /usr/src md5-identical to audit tree)",
  "run_cmd": "cc -O2 -Wall -o df3667_sim df3667_sim.c && ./df3667_sim",
  "code_hash": "78b6a39b301b32cd0a0841156dd4c15b0cac0a717e80e85dd4722d7e4c2ef5cd",
  "notes": "Guest kernel has ix(4) compiled in (X86_64_GENERIC:217) but no Intel 10GbE PCI device: attach/rxeof unreachable at runtime. Guest source restored to pristine after fix compile validation; guest left clean.",
  "recommended_fix": "In ix_rxeof, discard the frame when le16toh(cur->wb.upper.length) > rxr->rx_mbuf_sz (see fix.diff).",
  "fix_status": "not_testable",
  "fix_baseline_reproduced": 0,
  "fix_patched_reproduced": 0,
  "fix_verdict": "fix.diff compile-validated on the guest against md5-identical source (if_ix.ko built with -Werror, zero errors); runtime before/after comparison impossible without ixgbe hardware.",
  "fix_evidence": "build_baseline.log and build_patched.log (both clean); fix.diff applies with patch -p1 from /usr/src (verified PATCHED + build)."
}