linux-kernel - Re: [PATCH] proc: Fix uninitialized byte read in get_mm

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

Message-ID: <4f729521-7c34-251b-3939-dfcdc96aa50a@ispras.ru>
Date:   Sat, 13 Jul 2019 01:29:50 +0300
From:   Alexey Izbyshev <izbyshev@...ras.ru>
To:     Jakub Jankowski <shasta@...corp.com>,
        Alexey Dobriyan <adobriyan@...il.com>,
        Oleg Nesterov <oleg@...hat.com>
Cc:     linux-kernel@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        security@...nel.org
Subject: Re: [PATCH] proc: Fix uninitialized byte read in get_mm_cmdline()

On 7/13/19 12:17 AM, Jakub Jankowski wrote:
> On 2019-07-12, Alexey Izbyshev wrote:
> 
>> On 7/12/19 8:46 PM, Alexey Dobriyan wrote:
>>> The proper fix to all /proc/*/cmdline problems is to revert
>>>
>>>     f5b65348fd77839b50e79bc0a5e536832ea52d8d
>>>     proc: fix missing final NUL in get_mm_cmdline() rewrite
>>>
>>>     5ab8271899658042fabc5ae7e6a99066a210bc0e
>>>     fs/proc: simplify and clarify get_mm_cmdline() function
>>>
>> Should this be interpreted as an actual suggestion to revert the patches,
>> fix the conflicts, test and submit them, or is this more like thinking out
>> loud? In the former case, will it be OK for long term branches?
>>
>> get_mm_cmdline() does seem easier to read for me before 5ab8271899658042.
>> But it also has different semantics in corner cases, for example:
>>
>> - If there is no NUL at arg_end-1, it reads only the first string in
>> the combined arg/env block, and doesn't terminate it with NUL.
>>
>> - If there is any problem with access_remote_vm() or copy_to_user(),
>> it returns -EFAULT even if some data were copied to userspace.
>>
>> On the other hand, 5ab8271899658042 was merged not too long ago (about a year),
>> so it's possible that the current semantics isn't heavily relied upon.
> 
> I posted this (corner?) case ~3 months ago, unfortunately it wasn't picked up by anyone: https://lkml.org/lkml/2019/4/5/825
> You can treat it as another datapoint in this discussion.
> 
Thanks, this is interesting. Perl explicitly relies on special treatment of
non-NUL at arg_end-1 in pre-5ab8271899658042: on argv0 replace, it fills
everything after the new argv0 in the combined argv/env block with spaces [1,2].

While personally I don't approve what Perl does here, 5ab8271899658042
did change the user-visible behavior in this case.

[1] https://perl5.git.perl.org/perl.git/blob/86b50d930caa:/mg.c#l2733
[2] https://perl5.git.perl.org/perl.git/blob/86b50d930caa:/perl.c#l1698

Alexey
> 
> Regards,
>  Jakub
>