[RHSA-1999:028-01] Buffer overflow in libtermcap tgetent()
Michal Zalewski
> A buffer overflow existed in libtermcap's tgetent() function,
> which could cause the user to execute arbitrary code if they
> were able to supply their own termcap file.
>
> Under Red Hat Linux 5.2 and 4.2, this could lead to local users
> gaining root privileges, as xterm (as well as other possibly
> setuid programs) are linked against libtermcap. Under Red Hat
> Linux 6.0, xterm is not setuid root.
>
> Thanks go to Kevin Vajk and the Linux Security Audit team for
> noting and providing a fix for this vulnerability.
So, here I am.
Well, as this vunerability become well-known, I have nothing to loose,
enjoy: most of terminfo-based programs will accept TERM variable set to
eg. '../../../tmp/x'. All we have to do is to provide 'our own termcap
file', set TERM, then execute vunerable program w/terminfo support. In
fact, in.telnetd daemon shipped eg. with RH 6.0 /as well as with many
other recent distributions based on terminfo entries/, is vunerable... And
TERM variable can be passed using telnet ENVIRON option during protocol
negotiation before login procedure... Guess what?;) Almost remote root
(well, all you have to do locally is puting /tmp/x).
_______________________________________________________________________
Michal Zalewski [lca...@ids.pl] [link / marchew] [dione.ids.pl SYSADM]
[Marchew Industries] ! [http://lcamtuf.na.export.pl] bash$ :(){ :|:&};:
[voice phone: +48 (0) 22 813 25 86] ? [cellular phone: (0) 501 4000 69]
Iterowac jest rzecza ludzka, wykonywac rekursywnie - boska [P. Deutsch]
Michal Zalewski
> [...] most of terminfo-based programs will accept TERM variable set to
> eg. '../../../tmp/x'. All we have to do is to provide 'our own termcap
> file', set TERM, then execute vunerable program w/terminfo support. In
> fact, in.telnetd daemon shipped eg. with RH 6.0 /as well as with many
> other recent distributions based on terminfo entries/, is vunerable...
Oh, haven't said, for clearance... I'm talking about terminfo support and
tgetent() function implemented in libncurses, which is buggy as well,
while ncurses allows '../' tricks.
Michal Zalewski
> And as Chris Evans pointed out on linux-security, libncurses on RedHat
> is built with -DPURE_TERMINFO, which keeps it from using the buggy
> buffer code in libtermcap.
...not quite true - we're able to cause at least several SEGVs in ncurses'
tgetent() function by putting junk into terminfo files. Simply, try some
brute-force algorithms. I don't want to discuss about possible
consequences of this bug, as we haven't checked carefully terminfo format,
nor parser code.
Martin Schulze
> Well, as this vunerability become well-known, I have nothing to loose,
> eg. '../../../tmp/x'. All we have to do is to provide 'our own termcap
> file', set TERM, then execute vunerable program w/terminfo support. In
> fact, in.telnetd daemon shipped eg. with RH 6.0 /as well as with many
> TERM variable can be passed using telnet ENVIRON option during protocol
> negotiation before login procedure... Guess what?;) Almost remote root
> (well, all you have to do locally is puting /tmp/x).
Are you referring to terminfo or termcap? They are designed differently,
refer to different files and use different code.
Regards,
Joey
--
GNU does not eliminate all the world's problems, only some of them.
-- The GNU Manifesto
Tymm Twillman
has in.telnetd linked with libncurses, *NOT* libtermcap:
$ ldd /usr/sbin/in.telnetd
libncurses.so.4 => /usr/lib/libncurses.so.4 (0x40019000)
libutil.so.1 => /lib/libutil.so.1 (0x40056000)
libc.so.6 => /lib/libc.so.6 (0x40059000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)
ncurses ignores the buffer parameter to tgetent() that is usable for
exploits.
Note that this doesn't mean everything is safe; there are still
exploitable programs linked with libtermcap. But in.telnetd as delivered
with RH6.0 is fine in this respect.
-Tymm
On Sun, 4 Jul 1999, Michal Zalewski wrote:
> On Tue, 17 Aug 1999, Bill Nottingham wrote:
>
> > A buffer overflow existed in libtermcap's tgetent() function,
> > which could cause the user to execute arbitrary code if they
> > were able to supply their own termcap file.
> >
> > Under Red Hat Linux 5.2 and 4.2, this could lead to local users
> > gaining root privileges, as xterm (as well as other possibly
> > setuid programs) are linked against libtermcap. Under Red Hat
> > Linux 6.0, xterm is not setuid root.
> >
> > Thanks go to Kevin Vajk and the Linux Security Audit team for
> > noting and providing a fix for this vulnerability.
>
> So, here I am.
>
> Well, as this vunerability become well-known, I have nothing to loose,
> enjoy: most of terminfo-based programs will accept TERM variable set to
> eg. '../../../tmp/x'. All we have to do is to provide 'our own termcap
> file', set TERM, then execute vunerable program w/terminfo support. In
> fact, in.telnetd daemon shipped eg. with RH 6.0 /as well as with many
> other recent distributions based on terminfo entries/, is vunerable... And
> TERM variable can be passed using telnet ENVIRON option during protocol
> negotiation before login procedure... Guess what?;) Almost remote root
> (well, all you have to do locally is puting /tmp/x).
>
Tymm Twillman
built with -DPURE_TERMINFO, which keeps it from using the buggy buffer
code in libtermcap.
-Tymm
On Sun, 4 Jul 1999, Michal Zalewski wrote:
> On Sun, 4 Jul 1999, Michal Zalewski wrote:
>
> > [...] most of terminfo-based programs will accept TERM variable set to
> > eg. '../../../tmp/x'. All we have to do is to provide 'our own termcap
> > file', set TERM, then execute vunerable program w/terminfo support. In
> > fact, in.telnetd daemon shipped eg. with RH 6.0 /as well as with many
> > other recent distributions based on terminfo entries/, is vunerable...
>
> Oh, haven't said, for clearance... I'm talking about terminfo support and
> tgetent() function implemented in libncurses, which is buggy as well,
> while ncurses allows '../' tricks.
>
Olaf Kirch
> Oh, haven't said, for clearance... I'm talking about terminfo support and
> tgetent() function implemented in libncurses, which is buggy as well,
> while ncurses allows '../' tricks.
Do you have any more information about this problem? As far as I can remember,
ncurses doesn't do much parsing with a terminfo file, so there's little
harm that can be done here. Or do you have a demonstrable exploit?
Olaf
--
Olaf Kirch | --- o --- Nous sommes du soleil we love when we play
ok...@monad.swb.de | / | \ sol.dhoop.naytheet.ah kin.ir.samse.qurax
ok...@caldera.de +-------------------- Why Not?! -----------------------
UNIX, n.: Spanish manufacturer of fire extinguishers.
Kurt Wall
> [blah blah]
>
[Linux opens files with real not effective UID]
> The problem with telnetd is that you can pass a terminal name that indicates
> 'use a local file'. Now the ncurses library then goes 'ok leading slash
> all well and good', Im not suid uid==euid, lets open it as root and read a
> few bytes. You can't do much with it - you can rewind the machines tape
> drive for example however. Also if your termcap parser has bugs you can
> hit those.
This is fixed in the latest (pre-)release of ncurses-5.0. From the release
notes posted to bug-ncurses mailing list (as of last night) from da man
hissef:
990821 pre-release
+ updated configure macros CF_MAKEFLAGS, CF_CHECK_ERRNO
+ minor corrections to beterm terminfo entry.
+ modify lib_setup.c to reject values of $TERM which have a '/' in them.
So, version 5.0 will no longer accept $TERM that has a slash in it at all,
much less a leading one. I haven't looked closely at the source code, but a
similar change to the 4.2 sources, the version most distributions are using
now, should address this at least where tgetent() is concerned.
> It is a very nice example of why saying "lets ignore XYZ variable" is not
> security but a quick fix for emergencies. If you don't fix the code it
> will get you..
Yep...
Kurt
--
Life's too short to dance with ugly women.
Carlo M. Arenas Belon
> > The problem with telnetd is that you can pass a terminal name that indicates
> > 'use a local file'. Now the ncurses library then goes 'ok leading slash
> > all well and good', Im not suid uid==euid, lets open it as root and read a
> > few bytes. You can't do much with it - you can rewind the machines tape
> > drive for example however. Also if your termcap parser has bugs you can
> > hit those.
>
> This is fixed in the latest (pre-)release of ncurses-5.0. From the release
> notes posted to bug-ncurses mailing list (as of last night) from da man
> hissef:
>
> 990821 pre-release
> + updated configure macros CF_MAKEFLAGS, CF_CHECK_ERRNO
> + minor corrections to beterm terminfo entry.
> + modify lib_setup.c to reject values of $TERM which have a '/' in them.
>
> So, version 5.0 will no longer accept $TERM that has a slash in it at all,
> much less a leading one. I haven't looked closely at the source code, but a
> similar change to the 4.2 sources, the version most distributions are using
> now, should address this at least where tgetent() is concerned.
from a COL2.2 system
ldd /usr/sbin/in.telnetd
libncurses.so.4 => /lib/libncurses.so.4 (0x40018000)
libc.so.6 => /lib/libc.so.6 (0x4005a000)
/lib/ld-linux.so.2 => /lib/ld-linux.so.2 (0x40000000)
i think someone should fix the last post on Caldera's homepage regarding
this vulnerability
http://www.calderasystems.com/news/security/CSSA-1999:020.0.txt
.... Olaf? ;)
> > It is a very nice example of why saying "lets ignore XYZ variable" is not
> > security but a quick fix for emergencies. If you don't fix the code it
> > will get you..
>
> Yep...
wise words
Carlo