CVE-2026-35217

NanoMQ contains a protocol-semantics flaw in its MQTT v5 SUBSCRIBE handling: if a subscription entry is missing the final 1-byte Subscription Options field, the broker may still accept the malformed packet and install the subscription into internal broker state. Under a specific packet-length construction, the same parser flaw also causes a 1-byte out-of-bounds read that crosses the real heap allocation boundary and is detected by ASAN as a heap-buffer-overflow.

If the consumed byte happens to look acceptable, NanoMQ may continue and append the malformed subscription entry into its internal subinfol state. In that case, a SUBSCRIBE packet that should be rejected by MQTT rules is instead treated as a successful subscription. Whether ASAN reports the bug does not depend on MQTT's logical remain boundary; it depends on whether the read crosses the real heap allocation boundary of the underlying message buffer. In other words, these are not two unrelated issues. They are two manifestations of the same parsing defect: by default, it appears as a semantic vulnerability, and under suitable input conditions, it also becomes a verifiable out-of-bounds read vulnerability.

NanoMQ contains a protocol-semantics flaw in its MQTT v5 SUBSCRIBE handling: if a subscription entry is missing the final 1-byte Subscription Options field, the broker may still accept the malformed packet and install the subscription into internal broker state. Under a specific packet-length construction, the same parser flaw also causes a 1-byte out-of-bounds read that crosses the real heap allocation boundary and is detected by ASAN as a heap-buffer-overflow.

If the consumed byte happens to look acceptable, NanoMQ may continue and append the malformed subscription entry into its internal subinfol state. In that case, a SUBSCRIBE packet that should be rejected by MQTT rules is instead treated as a successful subscription. Whether ASAN reports the bug does not depend on MQTT's logical remain boundary; it depends on whether the read crosses the real heap allocation boundary of the underlying message buffer. In other words, these are not two unrelated issues. They are two manifestations of the same parsing defect: by default, it appears as a semantic vulnerability, and under suitable input conditions, it also becomes a verifiable out-of-bounds read vulnerability.

NanoMQ contains a protocol-semantics flaw in its MQTT v5 SUBSCRIBE handling: if a subscription entry is missing the final 1-byte Subscription Options field, the broker may still accept the malformed packet and install the subscription into internal broker state. Under a specific packet-length construction, the same parser flaw also causes a 1-byte out-of-bounds read that crosses the real heap allocation boundary and is detected by ASAN as a heap-buffer-overflow.

If the consumed byte happens to look acceptable, NanoMQ may continue and append the malformed subscription entry into its internal subinfol state. In that case, a SUBSCRIBE packet that should be rejected by MQTT rules is instead treated as a successful subscription. Whether ASAN reports the bug does not depend on MQTT's logical remain boundary; it depends on whether the read crosses the real heap allocation boundary of the underlying message buffer. In other words, these are not two unrelated issues. They are two manifestations of the same parsing defect: by default, it appears as a semantic vulnerability, and under suitable input conditions, it also becomes a verifiable out-of-bounds read vulnerability.

NanoMQ contains a protocol-semantics flaw in its MQTT v5 SUBSCRIBE handling: if a subscription entry is missing the final 1-byte Subscription Options field, the broker may still accept the malformed packet and install the subscription into internal broker state. Under a specific packet-length construction, the same parser flaw also causes a 1-byte out-of-bounds read that crosses the real heap allocation boundary and is detected by ASAN as a heap-buffer-overflow.

If the consumed byte happens to look acceptable, NanoMQ may continue and append the malformed subscription entry into its internal subinfol state. In that case, a SUBSCRIBE packet that should be rejected by MQTT rules is instead treated as a successful subscription. Whether ASAN reports the bug does not depend on MQTT's logical remain boundary; it depends on whether the read crosses the real heap allocation boundary of the underlying message buffer. In other words, these are not two unrelated issues. They are two manifestations of the same parsing defect: by default, it appears as a semantic vulnerability, and under suitable input conditions, it also becomes a verifiable out-of-bounds read vulnerability.