CVE-2023-45802

Published: 23 October 2023

When a HTTP/2 stream was reset (RST frame) by a client, there was a time window were the request's memory resources were not reclaimed immediately. Instead, de-allocation was deferred to connection close. A client could send new requests and resets, keeping the connection busy and open and causing the memory footprint to keep on growing. On connection close, all resources were reclaimed, but the process might run out of memory before that. This was found by the reporter during testing of CVE-2023-44487 (HTTP/2 Rapid Reset Exploit) with their own test client. During "normal" HTTP/2 use, the probability to hit this bug is very low. The kept memory would not become noticeable before the connection closes or times out. Users are recommended to upgrade to version 2.4.58, which fixes the issue.

Notes

backporting this to jammy and earlier will likely require
backporting the whole 2.0.10 version of the http/2 module as
it refactors how connections and streams are handled:
https://github.com/apache/httpd/commit/9767274b884a110e9244f59f50bd31ff1cae2933

Status

Severity score breakdown

References

• https://www.openwall.com/lists/oss-security/2023/10/19/6
• https://httpd.apache.org/security/vulnerabilities_24.html#CVE-2023-45802
• https://github.com/icing/blog/blob/main/h2-rapid-reset.md#cve-2023-45802
• https://ubuntu.com/security/notices/USN-6506-1
• https://www.cve.org/CVERecord?id=CVE-2023-45802
• NVD
• Launchpad
• Debian