Coda File System

more race conditions (in venus 5.3.15) ?

From: Ivan Popov <pin_at_math.chalmers.se>
Date: Sun, 25 Nov 2001 19:22:16 +0100 (MET)
Hello,

unpacking openssl source tar archive gave me a directory
openssl-0.9.6b/include/openssl/
that should be populated by symlinks but some of them (5 to 100%)
(depending on I don't know what, but higher load *seems* to make it worse)
become files with 0 length and 0 access bits.

I see that tar makes the following:

open("name", O_CREAT|O_WRONLY...., 0)
unlink("name")
symlink("....","name")

It looks like it is the unlink() that does not happen for some reason.

The effect observes reliably on a 486DX66 client.

It does not occur on a P166 client.

(would upgrade to 5.3.17 fix it? cannot test at once... but as I didn't
see any similar complains on the list, I may be the first one to report
the problem, then it is probably not fixed yet)

A 486 client seems to be a good testbed for race condition testing.
Or should I rather shut up and upgrade my hardware?
I am afraid the race will show up anyway, once in a million... when you
are running your life's most important application... :)

Best regards,
--
Ivan
Received on 2001-11-25 13:22:37