GHSA-7qmm-29vg-7hf8: Out-of-bounds read in the NetX Duo SNMP OID parser via a multi-byte BER length

GHSA ID

GHSA‑7qmm‑29vg‑7hf8

CVE ID

CVE‑2026‑102718

Severity

High

CWE

CWE-125, CWE-130

Published

2026‑10‑13

Last updated

2026‑10‑13

Affected packages

Package Vulnerable versions Patched versions

NetX Duo

⇐ 6.5.2.202603

6.5.2.202603a

Description

Summary

_nx_snmp_utility_object_id_get reads a declared length from an incoming OID and never checks it against the bytes actually present when that length uses either BER multi-byte form. The parse loop then runs for the declared count, reading past the end of the packet buffer.

Details

The single-byte form is checked: the function compares the declared total against the remaining buffer and rejects it when it does not fit. Neither multi-byte form is. The two-byte form verifies only that five bytes are present before reading its two length bytes, which allows a declared length of up to 65535; the one-byte form performs no check at all beyond the function’s four-byte entry test, allowing up to 255. In both cases the declared value is used directly as the loop count.

The sibling function _nx_snmp_utility_octet_get, in the same file, applies a single check covering every length form. That check is the one absent here.

Each byte read past the buffer is decoded as an OID component and appended to the agent’s internal OID string, where it persists in nx_snmp_agent_current_octet_string until the next request overwrites it. The response path echoes the original request rather than the parsed OID, so the over-read bytes are not returned to the sender.

A second, smaller over-read sits in the same function. Where an OID component uses multi-byte encoding, the parser scans forward for the end of the component and only afterwards rejects a run longer than three bytes, so up to three bytes beyond the buffer are read before the bound is applied.

Impact

On a target with memory protection the over-read faults and takes down the SNMP agent thread, costing the device its network management, and the request can be repeated against each restart. Without memory protection the read succeeds silently and the agent’s stored OID string is left holding adjacent memory, which can disturb subsequent legitimate requests and may reach a log or other side channel depending on what the application does with it.

Affected configurations

Any build running the NetX Duo SNMP agent. The defect is in the common OID parsing path and is not behind a build option.

One crafted SNMP request reaches it. Reaching the OID parser requires passing community-string validation, but the nx_snmp_agent_username_process callback is optional: an application that does not register it performs no validation at all. Where it is registered, SNMPv1 and v2c community strings travel in cleartext and are frequently left at public.

Releases up to and including 6.5.2.202603 are affected.

Severity

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, 8.7, High. CWE-125 and CWE-130.

Confidentiality is scored none because the response path echoes the original request rather than the parsed OID, so nothing read past the buffer is returned to the sender. CWE-130 is recorded alongside CWE-125 because the root cause is a declared length accepted without reconciliation against the data actually present.

Patches

Fixed in NetX Duo 6.5.2.202603a.

Both multi-byte length forms now apply the same remaining-bytes check the sibling octet parser already uses, and the forward scan over a multi-byte OID component is bounded as it iterates rather than after it finishes. No public API changes, so applications need no source change.

Workarounds

None within the agent. A device that does not run the SNMP agent is unaffected, and restricting which hosts may reach the SNMP port limits who can send the request. Registering a community-string validation callback with a non-default community string raises the bar but does not remove the defect.

Credits

Reported by @D7EAD and @adawn0106.

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