[Fwd: Xinetd 2.3.10 Memory Leaks]
Christopher Nehren
apeiron at comcast.net
Fri Apr 18 12:50:21 PDT 2003
I just submitted a PR for this (haven't even gotten the confirmation
email), but since not everyone tracks the GNATS CVS distribution, I
figured that I'd send it here as well.
-----Forwarded Message-----
> From: Steve Grubb <linux_4ever at yahoo.com>
> To: bugtraq at securityfocus.com
> Subject: Xinetd 2.3.10 Memory Leaks
> Date: 18 Apr 2003 16:18:36 +0000
>
>
>
> BACKGROUND
>
> -----------
>
>
>
> Xinetd is a popular inetd replacement. Shortly after the 2.3.9 release in
>
> September 2002, it was realized that xinetd was leaking file descriptors.
>
> That problem turned out to be that file descriptors were not always being
>
> closed whenever a connection was rejected. 2.3.10 was released with this
>
> fixup among others in January.
>
>
>
> Sometime in February, a machine that I admin was hit by an ftp worm. It
>
> created > 5000 connections in 1 second. Xinetd promptly keeled over.
>
> Xinetd had been running for over a month with no downtime. The machine has
>
> next to no ftp traffic and only from 2 sources, so it was configured to be
>
> run via xinetd rejecting connections via tcp_wrappers. The machine had
>
> weathered worm attacks in the past, so this puzzled me.
>
>
>
>
>
> TESTING
>
> -------
>
>
>
> Eventually, I started looking at xinetd with valgrind. I used the
>
> following commandline:
>
>
>
> valgrind --leak-check=yes --leak-resolution=med --num-callers=8 \
>
> --logfile-fd=9 /usr/sbin/xinetd -d -pidfile /var/run/xinetd.pid \
>
> -stayalive 9> out.txt
>
>
>
> Depending on your setup, you may need to use something higher than 9.
>
> Xinetd was tested on connections that succeed and connections that are
>
> rejected due to configuration settings. The easiest way to test this is to
>
> use the following setup for chargen:
>
>
>
> service chargen
>
> {
>
> type = INTERNAL
>
> user = root
>
> protocol = tcp
>
> wait = no
>
> access_times = 2:00-3:00
>
> # only_from = 192.168.1.3/24
>
> # no_access = 192.168.1.3/24
>
> }
>
>
>
> The point is to set it up in a way that the connection is guaranteed to be
>
> rejected. Then do a:
>
>
>
> telnet localhost chargen
>
> After a couple seconds "ctl-] quit"
>
> Then, /etc/rc.d/init.d/xinetd stop
>
>
>
> Valgrind reports the following:
>
>
>
> ==18939== 144 bytes in 1 blocks are definitely lost in loss record 36 of 45
>
> ==18939== at 0x40160DB8: malloc (vg_clientfuncs.c:103)
>
> ==18939== by 0x804FE22: (within /usr/sbin/xinetd)
>
> ==18939== by 0x805A496: (within /usr/sbin/xinetd)
>
> ==18939== by 0x8053611: (within /usr/sbin/xinetd)
>
> ==18939== by 0x805340D: (within /usr/sbin/xinetd)
>
> ==18939== by 0x40294A46: __libc_start_main (in /lib/libc-2.3.2.so)
>
> ==18939== by 0x804A310: (within /usr/sbin/xinetd)
>
> ==18939==
>
>
>
>
>
> THE PROBLEM
>
> -----------
>
>
>
> Using objdump -S /usr/sbin/xinetd, the block of code in question comes
>
> from service.c:
>
>
>
> void svc_request( struct service *sp )
>
> {
>
> connection_s *cp ;
>
> status_e ret_code;
>
>
>
>
>
> cp = conn_new( sp ) ;
>
> if ( cp == CONN_NULL )
>
> return ;
>
> if (sp->svc_not_generic)
>
> ret_code = spec_service_handler(sp, cp);
>
> else
>
> ret_code = svc_generic_handler(sp, cp);
>
>
>
>
>
> if ( ret_code != OK )
>
> {
>
> if ( SVC_LOGS_USERID_ON_FAILURE( sp ) )
>
> if( spec_service_handler( LOG_SERVICE( ps ), cp ) == FAILED ) {
>
> conn_free( cp, 1 );
>
> return;
>
> }
>
> CONN_CLOSE(cp);
>
> }
>
> }
>
>
>
> The above code has several problems. One background piece of information
>
> is that the sigchld handler in xinetd (child_exit->server_end->
>
> svc_postmortem) normally frees the connection's data. If the ret_code is
>
> not OK, the connection was only closed. This is little more than close(cp-
>
> >co_descriptor); This does not free cp since sigchld will not be called.
>
> It was only if the log service call failed that the connection was freed.
>
>
>
> The above code also did not take into account ret_code == OK if the
>
> service was no_wait or special. In both of those cases, the sigchld
>
> handler is not invoked so the memory pointed to by cp is lost when the
>
> call returns.
>
>
>
>
>
> CONSEQUENCES
>
> ------------
>
>
>
> The memory area pointed to by cp is 144 bytes. Since the variable goes out
>
> of scope, it is permanently lost with no way of finding it again. The
>
> memory losses are cumulative, too. It would take little more than
>
>
>
> while true; do telnet localhost chargen < /dev/null; done;
>
>
>
> to DOS the services provided by xinetd if you could identify a machine
>
> that uses xinetd to reject connections. Xinetd provides a rich set of
>
> options for rejecting connections, this includes: tcp_wrappers, only_from,
>
> no_access, sensors, access_times, cps, load_avg, etc.
>
>
>
> It should also be noted that if you DO NOT have any statements in the
>
> xinetd.conf file that would cause xinetd to reject a connection, then you
>
> are free from this problem.
>
>
>
>
>
> SOLUTION
>
> --------
>
>
>
> Xinetd 2.3.11 fixes the memory leaks as well as other problems discovered
>
> since 2.3.10 was released. All users of xinetd 2.3.10 are strongly urged
>
> to upgrade ASAP to avoid DOS conditions. Anyone running 2.3.9 is also
>
> strongly urged to upgrade since they are leaking file descriptors.
>
>
>
> Your xinetd version can be determined by typing "xinetd -version" (that's
>
> version with 1 dash).
>
>
>
> The new tarball is: www.xinetd.org/xinetd-2.3.11.tar.gz
>
>
>
> This problem has been assigned CAN-2003-0211 to track the bug.
>
>
>
> This bug was also reported here:
>
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=88537
>
>
>
> If you are affected, see if your vendor has an updated xinetd for you.
>
>
>
> -Steve Grubb
>
>
>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 187 bytes
Desc: This is a digitally signed message part
Url : http://lists.freebsd.org/pipermail/freebsd-security/attachments/20030418/fbb371a5/attachment.bin
More information about the freebsd-security
mailing list