Well, I didn't get any replies to my question, so I kept poking around. Since the problem seems to appear only after the server has been running for a while, seemingly can appear in any part of my system where taint checking matters, and produces nonsensical results, I wondered if perhaps something within perl's taint checking mechanism itself was getting corrupted.
Looking back through my notes, I remembered that when I had installed Taint-0.09 a number of tests had failed during "make test". (my bad decision to use it like that) Reading the "BUGS" section of the doc for that module put fear in my heart about taint checking (although the doc is 10 years old), so I stopped using it in my code. The server has been running for about a week now, and the problem hasn't reappeared. I guess all my new development tickled a problem that was already there. Let that be a lesson to me. ;-) Charlie Katz On Wednesday 07 March 2007 12:27 pm, Charlie Katz wrote: > Hi all, > > The site I develop (Apache 2.2.3, mod_perl 2.0.2 [perl 5.8.5], Mason 1.33) > runs with taint checking ("PerlSwitches -wT -I/www"). It's been working > fine for many months now, with my scripts happily untainting variables as > required. > > In the last couple of weeks, all of a sudden I am seeing occasional and > sporadic "Insecure dependency in XXX while running setgid" errors all > around the site. Seemingly important things I've observed about the > errors: > > -they seem to start after the server's been running for a day or two; > restarting it makes them go away for a while > > -inconsistent: after an occurence (which returns 500 to the client), simply > hitting reload in the browser gets the same request answered successfully > > -not process dependent: the reload is successful whether the request hits > the same Apache child that previously had the error, or a different child > > -nonsensical: one of the places I found it occurring is in a sysopen() > using a variable which was explicitly untainted in the preceding two lines > of code > > -not limited to any particular script; when they happen, they can be > anywhere in my code that taint checking matters > > > I've been doing a lot of development lately (in particular adding a CDBI > based system), but these errors are occurring in scripts that haven't been > touched in over a year. > > After some investigation, all I've learned is that perl definitely does > think the variables are tainted (duh!). I'm afraid I have little idea of > what to do next. Any suggested courses of inquiry I could take up would be > greatly appreciated. > > Regards, > Charlie -- Charlie Katz Harvard-Smithsonian Center for Astrophysics [EMAIL PROTECTED]