jec@iuvax.cs.indiana.edu (James E. Conley) (12/09/88)
/* Written 3:34 am Nov 18, 1988 by sahayman@iuvax in iuvax:cs.works.apollo */ /* ---------- "How to make Achilles crash." ---------- */ After a number of experiments and a lot of standing around waiting for the machine to reboot, it appears that Achilles will always crash if... a) /etc/tcpd is running b) Both interfaces are configured (don't just restart tcpd, you have to re-ifconfig the interfaces too.) c) You do a wbak (the utility that writes backup tapes) which includes //achilles/sys/drivers/ecmb. A simple wbak -f 1 -owner root -full //achilles/sys/drivers/ecmb is sufficient. Sometimes achilles will crash when backing up this directory, sometimes tcpd will merely segmentation fault. I'm not sure which file in the directory is causing the problem. I'm losing patience ... This suggests to me that perhaps something in the driver is depending on some file in /sys/drivers/ecmb being left alone; wbak comes traipsing along, backs up one of the files and tcpd says hey, what's going on, somebody else is reading one of my files, I better cause the machine to crash. This has probably got something to do with why the machine always crashes when doing a full backup. Eventually it gets around to /sys/drivers/ecmb, and blammo, down goes the machine. /* End of text from iuvax:cs.works.apollo */
sahayman@iuvax.cs.indiana.edu (Steve Hayman) (12/09/88)
James E. Conley writes about a way we found to crash a DSP-90
or at least cause tcpd to segmentation fault, by using "wbak".
If anyone's interested, it can be narrowed down further.
"wbak -f 1 -stdout /sys/drivers/ecmb/ether_int.lib >/dev/null"
will cause our tcpd to die, with a segmentation fault.
We ran tcpd with the debugging option and it didn't
produce any obvious information.
We've let Apollo know about this problem, and they are working
on trying to reproduce it.
Steve "Another Indiana U. Workstation Guy, but One Without
A Too-Long Signature File" Hayman