DF-1400 — wi_write_wep LUCENT WEP-key stack overflow (PoC)

Summary

wi_write_wep LUCENT branch (sys/dev/netif/wi/if_wi.c:1766-1769) does memcpy(wkey[i].wi_keydat, wk_key, keylen) with keylen up to IEEE80211_KEYBUF_SIZE=16 into wi_keydat[14], overflowing by 1-2 bytes per slot; on the last slot (i==3) the overflow runs off the wkey[4] stack array into the frame. Triggered on a WiFi state transition (AUTH/RUN) calling wi_write_wep. Requires SYSCAP_NONET_WIFI.

Reachability

wi(4) is a loadable module (if_wi.ko), not in GENERIC, attaching only to Prism/Lucent WiFi cards. The QEMU guest has no WiFi hardware, so the bug is not live-reachable here. Proven deterministically with a byte-exact harness replicating struct wi_key (if_wavelan_ieee.h:290-292) and the wkey[4] stack array.

Build / run

./build.sh && ./run.sh

Expected output (bug present)

slot 3: wrote 16 bytes into 14-byte wi_keydat -> 2 byte(s) past end [runs OFF the wkey[4] array into the frame]
frame region after wkey[4] corrupted by last slot: YES
OVERFLOW CONFIRMED: keylen=16 > wi_keydat[14] overflows by 2 byte(s) per slot ...

On a fixed driver the copy is bounded to sizeof(wi_keydat) = 14.

Fix

Clamp keylen to sizeof(wkey[i].wi_keydat) before the memcpy (fix.diff). Validated to compile into a rebuilt if_wi.ko (-Werror, rc=0).

Files

• wi_overflow.c — byte-exact harness.
• build.sh / run.sh / run.log / env.txt / fix.diff / VERDICT.md / manifest.json.

DF-1400 — wi_write_wep LUCENT WEP-key stack overflow

Verdict: REPRODUCED (primitive proven via source trace + byte-exact harness). Fix compiles.

wi_write_wep (LUCENT firmware branch) copies a WEP key of up to IEEE80211_KEYBUF_SIZE=16 bytes into a 14-byte wi_keydat field with no bounds check, overflowing by 1-2 bytes per key slot; on the last slot (i==3) the overflow runs off the wkey[4] stack array into the surrounding frame. Confirmed real by source tracing + a byte-exact harness. Not live-reachable on the QEMU guest because wi(4) attaches only to Prism/Lucent WiFi hardware (absent in QEMU) and the code is a loadable module. The fix compiles cleanly (-Werror).

Mechanism (trigger → primitive → effect)

sys/dev/netif/wi/if_wi.c:1764-1769 (LUCENT branch):

struct wi_key wkey[IEEE80211_WEP_NKID];                 /* wkey[4] on stack */
...
memset(wkey, 0, sizeof(wkey));
for (i = 0; i < IEEE80211_WEP_NKID; i++) {
    keylen = vap->iv_nw_keys[i].wk_keylen;              /* up to 16 */
    wkey[i].wi_keylen = htole16(keylen);
    memcpy(wkey[i].wi_keydat, vap->iv_nw_keys[i].wk_key, keylen);   /* SINK */
}

Destination (sys/dev/netif/wi/if_wavelan_ieee.h:290-292):

struct wi_key {
    u_int16_t  wi_keylen;
    u_int8_t   wi_keydat[14];      /* <-- 14-byte destination */
};

• IEEE80211_KEYBUF_SIZE = 16 (sys/netproto/802_11/ieee80211_crypto.h:34), IEEE80211_WEP_NKID = 4 (ieee80211.h:1272).
• wk_keylen is bounded by the ioctl path to IEEE80211_KEYBUF_SIZE (16); the WEP cipher accepts >=5 with no upper bound, so 15-16 reaches here.
• memcpy of 16 bytes into wi_keydat[14] overflows by 2 bytes. For i==3 (the last slot), the overflow runs off the wkey[4] array into the stack frame.

Primitive

• Class: stack buffer overflow, 1-2 bytes per slot (keylen 15-16), attacker-controlled key material content; on the last slot the overflow escapes the wkey[4] array into the surrounding frame.
• Trigger: WiFi state transition (AUTH/RUN) that calls wi_write_wep.
• Effect: stack corruption in the LUCENT WEP-key write path → panic or potential control-flow corruption on real Prism/Lucent WiFi hardware.

Reachability / threat model

• wi is a loadable module (if_wi.ko), NOT in X86_64_GENERIC. It attaches only to Prism/Lucent IEEE 802.11 WiFi cards. The QEMU guest has no WiFi hardware, so wi_write_wep never runs live here.
• Realistic threat: a host with Prism/Lucent WiFi joining a crafted/rogue AP, or a malicious local user triggering a WEP key set on attached WiFi that transitions to AUTH/RUN. Requires SYSCAP_NONET_WIFI.
• Validated the primitive deterministically with the byte-exact harness (wi_overflow.c): keylen=16 into wi_keydat[14] overflows by 2 bytes per slot; the last slot (i==3) corrupts the frame canary.

Harness proof (run.log)

wi_key.wi_keydat capacity: 14 bytes
keylen (wk_keylen, up to IEEE80211_KEYBUF_SIZE): 16
slot 3: wrote 16 bytes into 14-byte wi_keydat -> 2 byte(s) past end [runs OFF the wkey[4] array into the frame]
frame region after wkey[4] corrupted by last slot: YES
OVERFLOW CONFIRMED: keylen=16 > wi_keydat[14] overflows by 2 byte(s) per slot; last slot overruns wkey[4] into the stack frame.

Fix validation

fix.diff clamps keylen to sizeof(wkey[i].wi_keydat) (=14) before the memcpy. The fix was applied to in-guest /usr/src and the if_wi.ko module rebuilt cleanly (cc ... -Werror, RC=0). Runtime re-test is not possible on this guest (no WiFi HW), so fix_status = not_testable for runtime, with the diff verified to apply + compile and the harness logic confirming the clamp bounds the copy to 14 bytes.

Kernel references

• sys/dev/netif/wi/if_wi.c:1766-1769 (sink: unchecked keylen → memcpy)
• sys/dev/netif/wi/if_wavelan_ieee.h:290-292 (struct wi_key, 14-byte wi_keydat)
• sys/netproto/802_11/ieee80211_crypto.h:34 (IEEE80211_KEYBUF_SIZE=16)
• sys/netproto/802_11/ieee80211.h:1272 (IEEE80211_WEP_NKID=4)

Fix status (batch validation)

Applied with the 36 sibling fix.diffs of this batch in findings/poc/_batch41/all_fixes.patch; single in-guest make -j6 nativekernel KERNCONF=X86_64_GENERIC completed RC=0 with zero errors under -Werror (full log: findings/poc/_batch41/fix_build8.log). Behavior A/B intentionally skipped for this Info-tier item.