linux-kernel - Re: [PATCH v2 RESEND] m68k/kernel: array out of bound access in process_uboot

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]

Message-ID: <9775e266-5fee-b0e9-7fa3-b602ec4b7796@gmail.com>
Date:   Tue, 18 Jan 2022 10:18:16 +0800
From:   Hangyu Hua <hbh25y@...il.com>
To:     Greg Ungerer <gerg@...ux-m68k.org>, geert@...ux-m68k.org
Cc:     schwab@...ux-m68k.org, linux-m68k@...ts.linux-m68k.org,
        linux-kernel@...r.kernel.org
Subject: Re: [PATCH v2 RESEND] m68k/kernel: array out of bound access in
 process_uboot_commandline

Hi Greg,

On 2022/1/17 下午12:03, Greg Ungerer wrote:
> Hi Hangyu,
> 
> On 13/1/22 11:58 am, Hangyu Hua wrote:
>> When the size of commandp >= size, array out of bound write occurs 
>> because
>> len == 0.
>>
>> Signed-off-by: Hangyu Hua <hbh25y@...il.com>
>> ---
>>   arch/m68k/kernel/uboot.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/m68k/kernel/uboot.c b/arch/m68k/kernel/uboot.c
>> index 928dbd33fc4a..63eaf3c3ddcd 100644
>> --- a/arch/m68k/kernel/uboot.c
>> +++ b/arch/m68k/kernel/uboot.c
>> @@ -101,5 +101,6 @@ __init void process_uboot_commandline(char 
>> *commandp, int size)
>>       }
>>       parse_uboot_commandline(commandp, len);
>> -    commandp[len - 1] = 0;
>> +    if (len > 0)
>> +        commandp[len - 1] = 0;
>>   }
>>
> 
> I am not convinced this is wrong for the reason you think it is.
> Looking at the code in its entirety:
> 
> __init void process_uboot_commandline(char *commandp, int size)
> {
>          int len, n;
> 
>          n = strnlen(commandp, size);
>          commandp += n;
>          len = size - n;
>          if (len) {
>                  /* Add the whitespace separator */
>                  *commandp++ = ' ';
>                  len--;
>          }
> 
>          parse_uboot_commandline(commandp, len);
>          commandp[len - 1] = 0;
> }
> 
> 
> "commandp" is moved based on the return of the strnlen(). So in the
> case of commandp actually being full of valid characters (so n == size,
> and thus len == 0) the commandp technically points outside of its
> real size at that point. But "command[[len - 1]" would actually be
> pointing to the last char in the original commandp array (so the original
> commandp[size - 1]). Well at least if you are happy with the use of
> negative array indexes.
> 

You mean this is a friendly out of bound beacause "command[[len - 1]" 
pointing to the last char in the original commandp array. I used to 
think command[[len - 1] = 0 may be a zero-terminated for command. You 
can see my discussion with Andreas Schwab and my patch v1 in

https://lore.kernel.org/all/CAOo-nLJG71QqqD0-cJDyH0rY2VTx1eO9nHVQ5MCe8J0iiME_vw@mail.gmail.com/

But this still be a out of bound write because "commandp" is a macro 
definition with a fixed size.

> Clearly this could be structured better. There is no point in calling
> parse_uboot_commandline() if len == 0, or even if len == 1, since you
> cannot add anymore to the command line, it is full.
> 
I think it is no point too. But the caller (setup_arch()) don't check 
the size of "commandp" before call parse_uboot_commandline(). Instead we 
do this in parse_uboot_commandline(). So it may be better to move these 
checks to the caller ?

> Regards
> Greg

Thanks for your reply
Hangyu Hua