Bug in expire locking

[Sparry]

There is a major problem with locking in expire. The code is roughly as
follows, with bits merged from expire.c and funcs.c.
The problem is that a general log error and exit routine is being called
from the code which sets up the lock. This general routine removes the lock.

I have shown the code for System 5. I have a cron job that applies harder
and harder expires when the disk is getting full. If inews is unbatching at
the time that the expire is called, then the expire fails, but removes the
lock. No extra disk space is available, so a harder expire is started. It
finds no lock, so the entire history database is trashed, often along with
'active'. I then need 20mins of CPU time on a badly overloaded machine to
rebuild it.

The fix is to create a static that is set when dolock() sets the lock, and
make rmlock() only delete the lock if it is set.

The rest of this article is just extracts of code.

#ifdef SCCSID
static char	*SccsId = "@(#)expire.c	2.57	11/30/87";
#endif /* SCCSID */

main()
{
	.
	.
	dolock();
	.
	.
}

dolock()
{
	int i = 0;
	sprintf(afline,"%s.lock", ACTIVE);
	while (LINK(ACTIVE, afline) < 0 && errno == EEXIST) {
		if (i++ > 5) {
			xerror("Can't get lock for expire");
		}
		sleep(i*2);
	}
}

/* VARARGS1 */
xerror(message, arg1, arg2, arg3)
char *message;
long arg1, arg2, arg3;
{
	char buffer[LBUFLEN];

	fflush(stdout);
	sprintf(buffer, message, arg1, arg2, arg3);
	logerr(buffer);
	xxit(1);
}

xxit(i)
{
	if (i) UNLINK(ARTFILE);
	rmlock();
	exit(i);
}

rmlock()
{
	sprintf(bfr, "%s.lock", ACTIVE);
	UNLINK(bfr);
}
-- 
Mr I. W. J.  Sparry	Phone:	+44 225 826983
University of Bath	JANET:	cc_is at UK.AC.BATH.UX63
Bath BA2 7AY		UUCP:	uunet!mcvax!ukc!bath63!cc_is (bath63.UUCP)
England			ARPA:	cc_is%ux63.bath.ac.uk at ucl-cs.arpa