BugTraq summary, week of 26 oct 1998.


sshd fiasco
Last week a site called "rootshell" was broken into, and had its web page replaced.  The maintainers of rootshell determined that access was probably made through ssh, and that therefore there was a security hole in the sshd they were running (1.2.26 on Linux).  Within a few days, IBM was writing a draft of an advisory for sshd, including a patch that would fix the hole.  A few days later, however, the author of ssh got in contact with IBM and they discovered that the patch was ineffective, and that there also probably was no hole in sshd.  IBM was never able to reproduce the bug after recompiling sshd from scratch.  The rootshell website put out a little zine claiming that sshd is at fault, but currently the facts seem to support the notion that sshd 1.2.26 is fine.  Many people started poring over the code as a result of the hype, though, and a few small bugs (believed to not be exploitable) have been found and fixed.  End result: sshd 1.2.26 is probably safe.

users can control tapes on solaris
The /dev/rmt directory on Solaris has devices with permissions letting users manipulate the tape drives.  Some of the amusing exploits for this involved ruining your backups, or restoring from a backup into a user's directory where they can read secret files (like /etc/shadow).

windows may export its system dir
If you share a printer under Windows (95 or 98), your Windows system directory (usually \windows\system) will be exported read-only, so anyone can see what's in it (but can't change it).  Microsoft later defended this, saying it is used to share printer drivers, and that you should run NT in environments where security is an issue.  You can minimize the damage by turning off "File and Printer sharing for Microsoft networks" under TCP/IP properties, after which your system directory is only exported to your local LAN.

sendmail can be blocked on linux
Linux's network stack behaves slightly differently than an old-style BSD stack, mostly for performance reasons.  During an initial connection, Linux may detect an error and notify the application using error codes that exist in all Unixes but aren't typically returned during a passive connect attempt.  (Linux will sometimes notify an application of a connection before the 3-way TCP handshake is actually complete -- if the 3-way handshake fails to complete, it will then return an error to the application.)  Sendmail "copes" with unknown error codes by logging an error message and putting itself to sleep for 5 seconds.  A malicious person can therefore cause your sendmail to spend most of its time asleep by sending a SYN followed by RST every few seconds.

irix autofsd exploit explained
The autofsd problem reported last week on IRIX is apparently caused by executable maps.  If a client makes an autofs request to a map which is an executable file, autofsd will attempt to execute that file, appending the client's key.  Unfortunately it uses popen() to do this, which executes a shell.  A client can make an autofs request which contains a semicolon (;) followed by a command to execute.

hpux sharedx goes south
"Certain messages" sent to HPUX's SharedX port can cause it to enter an infinite CPU spin.  HPUX released patches without giving any further information.

password protection on mac disks can be avoided
There's a utility for the Mac called FWB Hard Disk Toolkit 2.5.  It's a disk driver that you can install which will let you password-protect volumes on the disk.  Most Mac formatting utilities will let you replace the FWB disk driver, but they will generally also corrupt the data on the drive.  If you replace the FWB driver with La Cie Silverlining, though, the drive can be read and the password is no longer needed to access it.  This is kind of obscure and maybe not very important, but I wanted to give equal time to Mac people...

two new nestea victims
USR Netserver 8/16 V.34 running OS 2.0.14 can be knocked down by the "nestea" IP attack, and has to be restarted.  FreeBSD 3.0-release was also vulnerable for a while, although they fixed the problem rapidly (again).  I can't figure out FreeBSD's naming scheme, so I'm not sure what to tell anyone who wants to upgrade to a "safe" version.  Apparently both the broken and fixed releases are just called "FreeBSD 3.0-release".

browser bugs

buffer overflows