BugTraq summary, week of 28 dec 1998.

oracle 8 is skittish
Oracle 8 listens on ports 1521 and 1526.  If you connect to either port and enter a few gibberish commands (for example, "kill" followed by "oracle"), one of Oracle's processes will spin up to 100% CPU on at least Linux, NT, Solaris, and HPUX.  On NT it will require a reboot to get rid of the process.  Somewhat mind-boggling that nobody at Oracle ever thought to try that.

sshd2 could open root ports
sshd2 (the SSH daemon's newer cousin) versions 2.0.11 and older will allow non-root users to request remote forwarding from privileged ports (the first 1024 ports, reserved for services).  This is fixed in more recent sshd2 releases.

linux random devices are too intense
In the (development) Linux 2.1 kernels, a large read from /dev/random or /dev/urandom (devices used to give out cryptographically safe random numbers) could cause the kernel to be so busy computing numbers that the system is unresponsive.  A patch was posted and this should be fixed by kernel 2.2.

netware bluescreen and more nmap deaths
Windows 95/98 machines running Novel Intranetware Client v3.0.0.0 (the meaning of all those zeros at Novell is left to the reader) can be bluescreened by doing the rapid "half-open then close" attack (such as via nmap).  The affected port is 427.  Later, another guy reported that Hylafax 4.0 (hfaxd) can be dropped this way too.

icq gui problem
This isn't much of a security hole, but is somewhat interesting because it's different, so I'll report it.  ICQ (a Windows chat program) has a way to send files from one user to another.  If the filename is too long, ICQ will only display the first part of it -- and if it contains an embedded linefeed, ICQ will only show the part prior to the linefeed.  So it's possible to send someone a file named "robot.jpg<LF>.exe" and have it appear to be downloading a jpeg.  When the user double-clicks on the jpeg (presumably to display it), they'll execute a possibly-harmful program.  This depends on the user not paying attention to the filename when double-clicking on it (since Windows will display the full filename), but it's easy to imagine this happening.  Here, the security hole is entirely related to the GUI!

datalynx suguard stupidity
Mudge ripped apart a setuid-root "security checker" program (called DataLynx SuGuard) that does things like execute the first 'ps' it finds in the path, dump to statically-named /tmp files, etc.  Pretty sad.

buffer overflows