GHSA-xjgr-2wmf-f78j: NULL pointer dereference in ICMPv6 Packet Too Big processing when the destination table is full

GHSA ID

GHSA‑xjgr‑2wmf‑f78j

CVE ID

CVE‑2026‑114408

Severity

High

CWE

CWE-252, CWE-476

Published

2026‑11‑01

Last updated

2026‑11‑01

Affected packages

Package Vulnerable versions Patched versions

NetX Duo

>= 6.4.5.202504, ⇐ 6.5.2.202603a

6.5.2.202603b

Description

Summary

_nx_icmpv6_process_packet_too_big adds the destination quoted in an ICMPv6 Packet Too Big message to the destination table, then reads the entry the add returns without checking that the add succeeded. When the table is full, the add fails and returns a NULL entry, and the function dereferences it. A sender that fills the table with Packet Too Big messages quoting new destinations crashes the IP thread with the next one.

Details

_nx_icmpv6_dest_table_add returns the new or updated destination entry through its third argument. It returns NX_NOT_SUCCESSFUL with that entry set to NX_NULL in two cases: the destination table has no free entry, or no neighbour cache entry can be added for the next hop. The Packet Too Big handler stores the status but does not test it before reading dest_entry_ptr → nx_ipv6_destination_entry_nd_entry.

Every input to the add comes from the message. The original destination is taken from the IPv6 header the message quotes, so the sender chooses it, and a message quoting a destination not yet in the table adds an entry. NX_IPV6_DESTINATION_TABLE_SIZE messages fill the table (8 by default), and the next one reaches the dereference. The handler accepts any reported MTU greater than zero and no larger than the interface MTU. It does not check the source address or the hop limit.

The unchecked read was introduced with the fix for CVE-2025-55102 (GHSA-f3rx-xrwm-q2rf), which added the block that reads the entry. The other callers of _nx_icmpv6_dest_table_add in the core stack either ignore the returned entry or test the status before using it.

Impact

The dereference runs in the IP helper thread, which processes every received packet.

Measured on the ThreadX Linux port: the process receives SIGSEGV on the first Packet Too Big after the table is full, which is the 9th message in a build with the default table of 8 and the 33rd with a table of 32. The messages came from a link-local sender on the same link.

On a target where the memory near address 0 is readable, the fault does not occur, and this was not measured. The function then reads a word at the offset of nx_ipv6_destination_entry_nd_entry from address 0. If that word is non-zero, it is used as a neighbour cache entry pointer. If the state field it points at reads as "created", the function sends a Neighbor Solicitation built from that memory and writes two fields through the pointer.

Affected configurations

Builds with FEATURE_NX_IPV6 and NX_ENABLE_IPV6_PATH_MTU_DISCOVERY. Path MTU discovery is off by default, and builds without it do not compile the vulnerable code.

Affected versions: [6.4.5.202504, 6.5.2.202603a]. Releases up to and including 6.4.4.202503a predate the CVE-2025-55102 fix and are not.

Severity

CVSS-B 8.7 (High): CVSS:4.0/AV:N/AC:L/AT:N/PR:N/UI:N/VC:N/VI:N/VA:H/SC:N/SI:N/SA:N. CWE-476 and CWE-252.

Attack requirements are scored none. Filling the table needs only messages that quote distinct destinations, and the handler checks neither the source address nor the hop limit. CWE-252 is recorded alongside CWE-476 because the root cause is a status returned by the add and never checked.

Patches

Fixed in NetX Duo 6.5.2.202603b. The handler now uses the returned entry only when the add succeeded. When it fails, the message is dropped as before, and the IP thread goes on processing received packets. No public API changes, so applications need no source change.

Proof, measured on dev: the regression test netx_icmpv6_packet_too_big_table_full_test fills the table, sends one more Packet Too Big, and then checks that the router’s Neighbor Advertisement is processed. Without the change it fails with SIGSEGV in 9 of 9 runs across three build configurations. With it, it passes in 60 of 60.

Workarounds

  • Building without NX_ENABLE_IPV6_PATH_MTU_DISCOVERY removes the vulnerable code.

  • A larger NX_IPV6_DESTINATION_TABLE_SIZE raises the number of messages needed but does not remove the defect.

  • Dropping ICMPv6 Packet Too Big messages before they reach the device removes the trigger, but it also disables path MTU discovery.

Credits

Found by @fdesbiens during coverage work for the Eclipse ThreadX safety certification.

The analysis of this defect and the fix were produced with AI assistance under human review, in accordance with the Eclipse Foundation’s Generative AI Usage Guidelines.