linux-kernel - Re: [PATCH] x86: Implement arch_prctl(ARCH_VSYSCALL

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Thu, 16 Dec 2021 19:31:19 +0100
From:   Florian Weimer <fweimer@...hat.com>
To:     "Andy Lutomirski" <luto@...nel.org>
Cc:     linux-arch@...r.kernel.org,
        "Linux API" <linux-api@...r.kernel.org>,
        linux-x86_64@...r.kernel.org, kernel-hardening@...ts.openwall.com,
        linux-mm@...ck.org, "the arch/x86 maintainers" <x86@...nel.org>,
        musl@...ts.openwall.com,
        "Dave Hansen via Libc-alpha" <libc-alpha@...rceware.org>,
        "Linux Kernel Mailing List" <linux-kernel@...r.kernel.org>,
        "Dave Hansen" <dave.hansen@...el.com>,
        "Kees Cook" <keescook@...omium.org>
Subject: Re: [PATCH] x86: Implement arch_prctl(ARCH_VSYSCALL_LOCKOUT) to
 disable vsyscall

* Andy Lutomirski:

> This could possibly be much more generic: have a mask of legacy
> features to disable and a separate mask of lock bits.

Is that really necessary?  Adding additional ARCH_* constants does not
seem to be particularly onerous and helps with detection of kernel
support.

>> I can turn this into a toggle, and we could probably default our builds
>> to vsyscalls=xonly.  Given the userspace ABI impact, we'd still have to
>> upstream the toggle.  Do you see a chance of a patch a long these lines
>> going in at all, given that it's an incomplete solution for
>> vsyscall=emulate?
>
> There is basically no reason for anyone to use vsyscall=emulate any
> more.  I'm aware of exactly one use case, and it's quite bizarre and
> involves instrumenting an outdated binary with an outdated
> instrumentation tool.  If either one is recent (last few years),
> vsyscall=xonly is fine.

Yeah, we plan to stick to vsyscall=xonly.  This means that the toggle is
easier to implement, of course.

>> Hmm.  But only for vsyscall=xonly, right?  With vsyscall=emulate,
>> reading at those addresses will still succeed.
>
> IMO if vsyscall is disabled for a process, reads and executes should
> both fail.  This is trivial in xonly mode.

Right, I'll document this as a glitch for now.

I've got a v2 (with the toggle rather than pure lockout) and will sent
it out shortly.

Thanks,
Florian