Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!watmath!clyde!rutgers!mit-eddie!apollo!rees
From: rees@apollo.UUCP
Newsgroups: news.software.b
Subject: Active file locking bug
Message-ID: <339e9797.982f@apollo.uucp>
Date: Thu, 12-Mar-87 09:30:00 EST
Article-I.D.: apollo.339e9797.982f
Posted: Thu Mar 12 09:30:00 1987
Date-Received: Fri, 13-Mar-87 20:56:49 EST
Organization: Apollo Computer, Chelmsford, Mass.
Lines: 13

With 1800 nodes all running news out of the same lib and spool, we tend to
uncover a lot of bugs in the news locking code.

The newgroup code in control.c does not lock the active file when it adds
a new newsgroup to the file.  The fix is to lock() before the loop that
reads the file, and unlock() on every exit from c_newgroup.

We use read-intend-write locks to improve concurrency, which standard Unix
doesn't have, so I can't just pass on our diffs.  I only have to lock()
while the write is actually taking place.

Without this fix, you get mass confusion if two newgroups come in at the
same time.