Please report security vulnerabilities through GitHub's private vulnerability reporting:

https://github.com/osrg/gobgp/security/advisories/new

Do not open a public issue or report to third-party CVE numbering authorities (CNAs). CVEs for this project are managed by the maintainers through GitHub Security Advisories.

Please include a reproducer that runs against a released version, and state the version you tested.

We assign a CVE when input from an untrusted source alone breaks memory safety or availability of gobgpd. Untrusted sources are the BGP session, the RPKI-RTR connection, and BFD packets.

Examples in scope:

• a panic or a fatal error
• a read or a write outside a buffer
• an unbounded allocation that leads to OOM
• a deadlock or a livelock that stops the daemon

Any other impact is out of scope, even when a peer can trigger it over the wire. A wrong routing decision, a violation of an RFC, and a wrong message sent to a peer are ordinary bugs.

The following are ordinary bugs. Please open a GitHub issue or send a pull request for them. We fix them, but we do not assign a CVE.

• Anything that requires access to the gRPC management API. A caller of that API can already add peers, add and delete routes, and stop the daemon. We treat every caller as trusted.
• Anything in the gobgp command. It is a client of that API. It also parses local files that the operator chooses, such as the MRT file given to gobgp mrt inject. Those files are trusted input.
• Anything that requires another program to call the Go packages of this project, such as pkg/packet/bgp, in a way gobgpd does not. Such a program must pass valid input, for example a body that matches its header, or a message that it builds itself. When a peer can reach the same bug through gobgpd, the Scope section applies.
• Anything that requires control of zebra or of the ZAPI connection to it. zebra runs under the same administration as gobgpd, so we treat it as trusted.
• Anything that requires a non-default configuration that grants the attacker the ability in question.
• Protocol conformance bugs and wrong routing decisions. See Scope.

You may use AI tools to find an issue and to write the report. Please tell us if you did. The reporter is responsible for the content of the report, not the tool. Check every claim before you send it.

Only the latest release is supported with security updates.