linux-kernel - Re: [PATCH] lib: vsprintf: Avoid 32-bit truncation in vsscanf number parsing

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: <b200a554-be81-f8b0-28a4-39c6f3c6900f@opensource.cirrus.com>
Date:   Thu, 12 Nov 2020 15:46:46 +0000
From:   Richard Fitzgerald <rf@...nsource.cirrus.com>
To:     Steven Rostedt <rostedt@...dmis.org>
CC:     <pmladek@...e.com>, <sergey.senozhatsky@...il.com>,
        <linux-kernel@...r.kernel.org>, <patches@...nsource.cirrus.com>
Subject: Re: [PATCH] lib: vsprintf: Avoid 32-bit truncation in vsscanf number
 parsing

On 12/11/2020 15:35, Steven Rostedt wrote:
> On Thu, 12 Nov 2020 11:17:59 +0000
> Richard Fitzgerald <rf@...nsource.cirrus.com> wrote:
> 
>> Number conversion in vsscanf converts a whole string of digits and then
>> extracts the field width part from the converted value. The maximum run
>> of digits is limited by overflow. Conversion was using either
>> simple_strto[u]l or simple_strto[u]ll based on the 'L' qualifier. This
>> created a difference in truncation between builds where long is 32-bit
>> and builds where it is 64-bit. This especially affects parsing a run of
>> contiguous digits into separate fields - the maximum length of the run
>> is 16 digits if long is 64-bit but only 8 digits if long is 32-bits.
>> For example a conversion "%6x%6x" would convert both fields correctly if
>> long is 64-bit but not if long is 32-bit.
>>
>> It is undesirable for vsscanf to parse numbers differently depending on
>> the size of long on the target build.
>>
>> As simple_strto[u]l just calls simple_strto[u]ll anyway the conversion
>> is always 64-bit, and the result is manipulated as a u64, so this is an
>> avoidable behaviour difference between 32-bit and 64-bit builds. The
>> conversion can call simple_strto[u]ll directly and preserve the full
>> 64-bits that were parsed out of the string.
>>
>> Signed-off-by: Richard Fitzgerald <rf@...nsource.cirrus.com>
>> ---
>>   lib/vsprintf.c | 8 ++------
>>   1 file changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/lib/vsprintf.c b/lib/vsprintf.c
>> index 14c9a6af1b23..63b6cddfa7f7 100644
>> --- a/lib/vsprintf.c
>> +++ b/lib/vsprintf.c
>> @@ -3444,13 +3444,9 @@ int vsscanf(const char *buf, const char *fmt, va_list args)
>>   			break;
>>   
>>   		if (is_sign)
>> -			val.s = qualifier != 'L' ?
>> -				simple_strtol(str, &next, base) :
>> -				simple_strtoll(str, &next, base);
>> +			val.s = simple_strtoll(str, &next, base);
>>   		else
>> -			val.u = qualifier != 'L' ?
>> -				simple_strtoul(str, &next, base) :
>> -				simple_strtoull(str, &next, base);
>> +			val.u = simple_strtoull(str, &next, base);
>>   
>>   		if (field_width > 0 && next - str > field_width) {
>>   			if (base == 0)
> 
> It looks like this is fixing the symptom and not the disease. The real
> issue I see here is that vsscanf is not honoring the '6' of '%6x' here. It
> should only read the 6 characters then do the conversion, not the other
> way around! This looks to me that the design is of issue.
> 
> -- Steve
> 

See this thread from 2014 where the field width problem was raised and
explained:
http://lkml.iu.edu/hypermail/linux/kernel/1401.1/03443.html

and the reply from Linus Torvalds that was against fixing field width
handling:
http://lkml.iu.edu/hypermail/linux/kernel/1401.1/03488.html

which I assume is why the field handling wasn't unoptimized to be
strictly correct.

Nevertheless, I see no reason not to remove avoidable inconsistencies
from the current design.