Bug#1059393: openssh: CVE-2023-51767

[Salvatore Bonaccorso]

Source: openssh
Version: 1:9.6p1-2
Severity: important
Tags: security upstream
X-Debbugs-Cc: car...@debian.org, Debian Security Team <te...@security.debian.org>

Hi Colin,

The following vulnerability was published for openssh. This is for now
just to track the issue as pointed out by a current paper. Apparently
openssh and sudo got CVEs assigned for their respective issues.

CVE-2023-51767[0]:
| OpenSSH through 9.6, when common types of DRAM are used, might allow
| row hammer attacks (for authentication bypass) because the integer
| value of authenticated in mm_answer_authpassword does not resist
| flips of a single bit. NOTE: this is applicable to a certain threat
| model of attacker-victim co-location in which the attacker has user
| privileges.


If you fix the vulnerability please also make sure to include the
CVE (Common Vulnerabilities & Exposures) id in your changelog entry.

For further information see:

[0] https://security-tracker.debian.org/tracker/CVE-2023-51767
https://www.cve.org/CVERecord?id=CVE-2023-51767
[1] https://arxiv.org/abs/2309.02545

Please adjust the affected versions in the BTS as needed.

Regards,
Salvatore

[Boud Roukema]

hi openssh maintainers,

There's a proposed mitigation for CVE-2023-51767 with ABACuS:

https://arxiv.org/abs/2310.09977

https://github.com/CMU-SAFARI/ABACuS

Something on this should probably be added to the "Notes" at
https://security-tracker.debian.org/tracker/CVE-2023-51767 .

Disclaimer: I just saw the citation - I have no expertise in checking
the validity of the exploit or the mitigation.

Cheers
Boud

PS: Conspiracy theory (numerology): this bug number is 105000 +
101*93, while the ArXiv ID after YYMM is 101*97. Common to
both is 101*p where p is a prime and p < 100 . ;)

[Boud Roukema]

On Tue, 26 Dec 2023, Boud Roukema wrote:

> PS: Conspiracy theory (numerology): this bug number is 105000 +
> 101*93, while the ArXiv ID after YYMM is 101*97. Common to
> both is 101*p where p is a prime and p < 100 . ;)

Correction: 101*93 = 9393, but 11*907 = 9977. So much for the conspiracy ;).
Boud

[Colin Watson]

On Tue, Dec 26, 2023 at 12:03:36PM +0100, Boud Roukema wrote:
> There's a proposed mitigation for CVE-2023-51767 with ABACuS:
>
> https://arxiv.org/abs/2310.09977
>
> https://github.com/CMU-SAFARI/ABACuS

This is a proposal for redesigned memory controllers. It isn't an
actionable mitigation at the level of OpenSSH.

--
Colin Watson (he/him) [cjwa...@debian.org]

[Boud Roukema]

hi Colin, all,

On Sat, 30 Dec 2023, Colin Watson wrote:

> This is a proposal for redesigned memory controllers. It isn't an
> actionable mitigation at the level of OpenSSH.

You're right: the end of the paragraph "Key Mechanism" on page 2 and
Section 4.1 "ABACuS's Hardware Design" make that clear. It's a
(proposed) hardware solution.

Cheers
Boud