Software Engineering Institute

A remotely exploitable heap overflow exists in the cachefsd program shipped and installed by default with Sun Solaris 2.5.1, 2.6, 7, and 8 (SPARC and Intel architectures). A remote attacker can send a crafted RPC request to the cachefsd program to remotely exploit the vulnerability.

Logs of exploitation attempts may resemble the following:

May 16 22:46:08 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Segmentation Fault - core dumped
May 16 22:46:21 victim-host last message repeated 7 times
May 16 22:46:22 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Bus Error - core dumped
May 16 22:46:24 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Segmentation Fault - core dumped
May 16 22:46:56 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Bus Error - core dumped
May 16 22:46:59 victim-host last message repeated 1 time
May 16 22:47:02 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Segmentation Fault - core dumped
May 16 22:47:07 victim-host last message repeated 3 times
May 16 22:47:09 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Hangup
May 16 22:47:11 victim-host inetd[600]: /usr/lib/fs/cachefs/cachefsd: Segmentation Fault - core dumped


Sun Microsystems has released a Sun Alert Notification which addresses this issue as well as the issue described in VU#161931.

According to the Sun Alert Notification, failed attempts to exploit this vulnerability will leave core dumps in the root directory. The presence of the core file does not preclude the success of subsequent attacks. Additionally, if the file /etc/cachefstab exists, it may contain unusual entries.

This issue is also being referenced as CAN-2002-0033:

http://cve.mitre.org/cgi-bin/cvename.cgi?name=CAN-2002-0033

A remote attacker can execute code with the privileges of the cachefsd process, typically root.

The CERT/CC is currently unaware of patches for this problem.

According to a Sun Alert Notification a workaround is as follows:

Comment out cachefsd in /etc/inetd.conf as shown below:

#100235/1 tli rpc/tcp wait root /usr/lib/fs/cachefs/cachefsd cachefsd

Once the line is commented out either:

- reboot, or
- send a HUP signal to inetd(1M) and kill existing cachefsd processes, for example,
on Solaris 2.5.1 and 2.6 do the following:
$ kill -HUP <PID of inetd>
$ kill <PIDs of any cachefsd processes>

Solaris 7 and 8 do the following:
$ pkill -HUP inetd
$ pkill cachefsd

The possible side effects of the workaround are:

- for systems not using cachefs:

There is no impact.

- for systems using cachefs:

Only a "disconnected" operation is known to be affected by
disabling cachefsd. This feature is rarely used outside of AutoClient.

Mounts and unmounts should still succeed though an error message
may be seen, "mount -F cachefs: cachefsd is not running".

There is no performance impact.

- for systems using AutoClient:

The impact is unknown. Again, only "disconnected" mode is likely
to be affected.