home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!munnari.oz.au!uniwa!DIALix!zeus!zeus!not-for-mail
- From: peter@zeus.DIALix.oz.au (Peter Wemm)
- Newsgroups: comp.unix.sysv386
- Subject: Re: perl 4.035 induces kernel panic in SVR4
- Date: 13 Sep 1992 10:57:00 +0800
- Organization: Karam Pty. Ltd., Perth, Western Australia.
- Lines: 90
- Message-ID: <18ualsINN2iv@zeus.dialix.oz.au>
- References: <1992Sep7.002515.15426@husc3.harvard.edu> <jrh.715877574@mustang> <1992Sep9.180635.27817@crd.ge.com> <1992Sep10.204557.15522@husc3.harvard.edu> <1992Sep11.145204.13475@crd.ge.com>
- NNTP-Posting-Host: zeus.dialix.oz.au
-
- davidsen@ariel.crd.GE.COM (william E Davidsen) writes:
- >In article <1992Sep10.204557.15522@husc3.harvard.edu>, maziere1@husc8.harvard.edu (David Mazieres) writes:
- >|Thanks for all the help I've gotten. In fact, my problem was simply that I
- >|was including libucb before libc. I changed the library flags to -lc -lucb,
- >|and the new version tested fine. When I posted I figured I'd probably messed
- >|up the flags somehow-the really disturbing thing is that a non-root user can
- >|bring the whole system down. I know the ucb library has tons of bugs in it,
- >|but should even the most bizarely warped and misused system calls panic the
- >|kernel? (i.e. not should as in is this a good thing, but should as in have
- >|other people been able to crash their SVR4 kernels reproducibly)
-
- > I used to crash Dell regularly with a temp sensitive motherboard ;-)
- >Other than that I have never had a panic with Dell (or Intel, when I was
- >doing beta on that). It's been totally resistant to software causing
- >panics, one of the reasons I like it. I've never corrupted the
- >filesystems, either, no that I think of it.
-
- There is a program that is periodically posted called "crashme". It is
- designed to do it's worst (or best) in an attempt to crash the kernel.
- There have been *SERIOUS* side effects every time time I have run it,
- varying from instant panics, to trashing write protected files.
-
- It sets up all signal handlers to point to some randomly generated trash
- and then attempts to run the "garbage" as code. This induces a recursivly
- called (and bombing) signal handler. When the child finally died, the parent
- noted the fact in a log file, and launched another one (or something like
- that).
-
- Some of the thinks I have seen.. (running as a "standard" user..)
-
- * For a while, it would instantly panic the kernel. This changed after I
- reconfigured the kernel some time later for a new driver.
- * Extreme performance degradation..
- * core dumps (there shouldn't be any - the resource limit for core files
- was set to zero)
- * core dumps to WRITE-PROTECTED core files!!!
- * core dumps to WRITE-PROTECTED and ROOT-OWNED core files!!!
- * even core dumps in write-protected (and even root owned (like /etc))
- directories!!!
- * corrupted executable!!! - even when the executable is write protected and
- foreign owned...
-
- This is what I mean....
-
- Before:
-
- dr-xr-xr-x 2 root root 512 Mar 13 10:23 .
- ---------- 1 root root 0 Mar 13 10:22 core
- ---x--x--x 1 root root 12560 Mar 13 10:23 crashme
-
- After..
-
- dr-xr-xr-x 2 root root 512 Mar 13 10:26 .
- ---------- 1 root root 512 Mar 13 10:26 core
- ---x--x--x 1 root root 4096 Mar 13 10:26 crashme
-
- This is with "crashme" run as userid 103 (me).
- If I start with no core file, one i created, just the same..
- The executable is now corrupt... It resembles nothing like the original.
-
- This was a fair while ago, but I still have the source that did this...
- I saved a directory listing before I deleted them, rebooted and fsck'ed.. :-)
-
- This is on Dell 2.1 BTW (I Think.. It might have been 2.0.1 back then.
- I am not sure and I don't want to risk finding out...)
-
- Imagine the consequences of typing this as a normal user...
-
- cd /tmp (or somewhere on the root filesystem you can write to)
- ln /etc/passwd core
- crashme
-
- Or how about
- cd /tmp
- ln /sbin/sh core
- crashme
-
- This was on a UFS root filesystem, and at the time a "pristine" kernel.
- (No 387, and a 386-33, 16M ram, AHA-1542b)
-
- -Peter
-
- PS: The author of the program claimed that a suprising number systems he
- tried never made it past an hour of "stress testing".
- The known "bombable" cases listed in the source code include:
- Most Suns, DecStation 3100,5000, IBM RT, RS/6000, Iris4d 3.3.2, and Nixdorf.
- --
- Peter Wemm : peter@zeus.dialix.oz.au If it's broke, fix it (The MS-DOS way)
- Work phone: +61-9-479-1855 If it aint broke, don't touch it (The Unix way)
- Fax: +61-9-479-1134 If we can't fix it, it ain't broke (Maintainer's Motto)
-