linux-kernel - Re: Re: [RESEND PATCH v3] sysctl: simplify the min/max boundary check

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: <4rrwkbj5sh4anblrxzhehcir2z2w5qhrdxfu4gc4irfg4ubb7q@hjt3e6agz42i>
Date: Thu, 31 Oct 2024 10:39:39 +0100
From: Joel Granados <joel.granados@...nel.org>
To: Wen Yang <wen.yang@...ux.dev>
Cc: "Eric W . Biederman" <ebiederm@...ssion.com>, 
	Luis Chamberlain <mcgrof@...nel.org>, Kees Cook <keescook@...omium.org>, 
	Joel Granados <j.granados@...sung.com>, Christian Brauner <brauner@...nel.org>, 
	Dave Young <dyoung@...hat.com>, linux-kernel@...r.kernel.org
Subject: Re: Re: [RESEND PATCH v3] sysctl: simplify the min/max boundary check

On Wed, Oct 30, 2024 at 12:26:17AM +0800, Wen Yang wrote:
> 
> 
> On 2024/10/23 03:12, Joel Granados wrote:
> > On Thu, Sep 05, 2024 at 09:48:18PM +0800, Wen Yang wrote:
...

> >> @@ -936,10 +921,10 @@ static int do_proc_douintvec_minmax_conv(unsigned long *lvalp,
> >>   int proc_douintvec_minmax(const struct ctl_table *table, int write,
> >>   			  void *buffer, size_t *lenp, loff_t *ppos)
> >>   {
> >> -	struct do_proc_douintvec_minmax_conv_param param = {
> >> -		.min = (unsigned int *) table->extra1,
> >> -		.max = (unsigned int *) table->extra2,
> >> -	};
> >> +	struct proc_minmax_conv_param param;
> >> +
> >> +	param.min = (table->extra1) ? *(unsigned int *) table->extra1 : 0;
> >> +	param.max = (table->extra2) ? *(unsigned int *) table->extra2 : UINT_MAX;
> > This is one of the cases where there is potential issues. Here, if the
> >  value of table->extra{1,2}'s value is greater than when
> > the maximum value of a signed long, then the value assigned would be
> > incorrect. Note that the problem does not go away if you remove the
> > "unsigned" qualifier; it remains if table->extra{1,2} are originally
> > unsigned.
> > 
> 
> I set up a CentOS 7.9 32-bit VM on Virtuanbox:
> # uname  -a
> Linux osboxes.org 3.10.0-1160.2.2.el7.centos.plus.i686 #1 SMP Mon Oct 26 
> 11:56:29 UTC 2020 i686 i686 i386 GNU/Linux
> 
> And the following test code:
> 
> #include <stdio.h>
> #include <stdlib.h>
> 
> int main()
> {
> 	unsigned int i = 4294967294;
> 	long j = i;
> 
> 	printf("original hex(i) = 0x%x\n", i);
> 	printf("unsigned int(i) = %lu\n", i);
> 	printf("---------------------\n");
> 	printf("hex(j) = 0x%x\n", j);
> 	printf("long(j) = %ld\n", j);
> 	printf("unsigned long(j) = %lu\n", j);
> 	printf("int(j) = %d\n", j);
> 	printf("unsigned int(j) = %lu\n", j);
> 	return 0;
> }
> 
> 
> ./a.out
> 
> original hex(i) = 0xfffffffe
> unsigned int(i) = 4294967294
> ---------------------
> hex(j) = 0xfffffffe
> long(j) = -2
This ^^^^^ is exactly what I expected. Thx for the test!

When you transfer that to your patch, it means that for certain cases
[1] the value resulting from the interpretation of param.{min,max}
(signed long) is going to be different than the value resulting from the
interpretation of table-extra{1,2} (unsigned int).

Here is another way of thinking about it:
We are avoiding bugs where a developer thinks they are handling longs,
when in reality they are handling unsinged ints; The result of
subtracting 1 from (-2) is very different from subtracting 1 from
4294967294.

> unsigned long(j) = 4294967294
> int(j) = -2
> unsigned int(j) = 4294967294
> 
> 
> The original hexadecimal values are the same, using unsigned int, int, 
> unsigned long, or long is just interpreted in different ways.
Exactly. Hex remains the same but the interpretation changes. And it is
there where pain lies.

Please re-work the patch without merging everything into
do_proc_douintvec_minmax_conv_param

> 
> We also ensure consistency in numerical writing and type conversion in 
> the patch. For example, in proc_rointvec_jiffies, convert to int; And in 
> proc_rouintvec_minmax, it is converted to unsigned int.
> 
> 
> --
> Best wishes,
> Wen
> 
> 
> > I'm not sure if there are more, but just having one of these things
> > around make me uncomfortable. Please re-work the patch in order to
> > remove this issue in order to continue review.
> > 
> > best
> > 
Best

[1] This is one case that I can imagine
  1. On 32 bit architectures
  2. Where UINT_MAX > LONG_MAX
  3. The value being assigned is greater LONG_MAX

-- 

Joel Granados