⬢ DragonFlyBSD Kernel Audit
DF-4231 / 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
36
{
  "finding_id": "DF-4231",
  "status": "inconclusive",
  "reproduced": 0,
  "impact": "none",
  "confidence": "certain",
  "verdict": "Source-level certain: txp_rx_reclaim() (sys/dev/netif/txp/if_txp.c:676) stamps the firmware-fed 16-bit rxd->rx_len (volatile, inside the DMA-shared rx ring, if_txpreg.h:263) into the receive cluster mbuf with no bound, then m_copydata(m, 0, rx_len, mtod(mnew)) at :704 (__STRICT_ALIGNMENT is always defined, if_txp.c:85) copies rx_len bytes between two 2048-byte clusters in one bcopy (m_copydata trusts m->m_len, uipc_mbuf.c:1671/1690) -- for rx_len>2048 both an OOB heap read of the source cluster and an equal OOB heap write past the destination cluster, up to 63531 bytes, contents = adjacent cluster-zone heap. No KASSERT fires (driver itself stamped m_len). Runtime honestly inconclusive: driver verified present in the running kernel (kldstat -v: 161 pci/if_txp) but the guest has no 3Com 3c990 (pciconf: NO_3COM_PCI_DEVICE) and QEMU has no Typhoon model, so the DMA-fed precondition cannot be staged.",
  "exploit_chain": "On real/passthrough hardware: firmware writes rx_desc{flags=DATA, rx_len=65535} -> intr -> txp_rx_reclaim -> 63KB bcopy past mnews cluster -> adjacent mbuf-cluster-zone objects overwritten with heap-sourced content -> groom a victim cluster-adjacent object (e.g. in-flight mbuf m_data/m_ext) -> kernel R/W -> uid=0 for the device owner; faulty-firmware case = crash generator. Same class as audit-accepted DF-3604 (bce) family.",
  "evidence": "run.log (driver-in-kernel + NO_3COM_PCI_DEVICE + md5 parity), VERDICT.md (path:line trace + m_copydata mechanics), trigger.c (annotated precondition probe), fix.diff",
  "kernel_refs": [
    "sys/dev/netif/txp/if_txp.c:676",
    "sys/dev/netif/txp/if_txp.c:688",
    "sys/dev/netif/txp/if_txp.c:693",
    "sys/dev/netif/txp/if_txp.c:704",
    "sys/dev/netif/txp/if_txp.c:85",
    "sys/dev/netif/txp/if_txpreg.h:263",
    "sys/dev/netif/txp/if_txpreg.h:631",
    "sys/kern/uipc_mbuf.c:1671",
    "sys/kern/uipc_mbuf.c:1690"
  ],
  "poc_changes": "No seed existed (pass-2 new finding); trigger authored as an annotated hardware-precondition probe per the DF-3604 convention since no syscall path can forge DMA state.",
  "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": 140,
  "guest_dirty": 0,
  "build_cmd": "cc -o trig_4231 t4231.c (guest)",
  "run_cmd": "./trig_4231 && kldstat -v | grep if_txp && (pciconf -l | grep -i 10b7 || echo NO_3COM_PCI_DEVICE)",
  "code_hash": "f30b1c6ed94a205c31d1b5ac6fbe4e9e3936fb694994cfc3fad281295aec9cd5",
  "notes": "Pass-2 depth: DF-1461 (known) covers only txp_rxbuf_reclaims kfree error path -- different function, no overlap (DB queried). fw download bounds (if_txp.c:553-567) verified sound vs the compiled-in tc990image; tx walks are cnt-bounded; rsp_numdesc is u_int8 so the kmalloc cannot overflow.",
  "recommended_fix": "In txp_rx_reclaim, drop frames with rx_len outside [ETHER_MIN_LEN, MCLBYTES] before stamping m_len (see fix.diff).",
  "fix_status": "not_testable",
  "fix_baseline_reproduced": 0,
  "fix_patched_reproduced": 0,
  "fix_verdict": "fix.diff uses the files own error idiom (ierrors + goto next, cf. if_txp.c:664-668) so the buffer slot keeps its cluster for the next refill. Not runtime-validable: no 3c990 on the guest, same reason the PoC is inconclusive.",
  "fix_evidence": "fix.diff; run.log absence proof"
}