bug in uucp locking mechanism

[lund%ucla-locus at sri-unix.UUCP]

From:            Laurence Lundblade <lund at ucla-locus>

	There is a window in the locking code used by uucp.
The difficulty is in ulockf(). First it calls onelock() which attempts to
create the lock file, then if that fails, it attempts to stat() the file to see
if it is old enough to delete. If the stat fails it recognizes this
and attempts to remove the lockfile without regard for someone else
who created the lockfile immediately follwing the failure of the stat.
The fix is simply to move the removal of the lockfile so it is
only executed if the stat() succeeds and the file is found to be old enough 
to delete. 

In the routine ulockf() in the file ulockf.c

	if (onelock(pid, tempfile, file) == -1) {
		/* lock file exists */
		/* get status to check age of the lock file */
		ret = stat(file, &stbuf);
		if (ret != -1) {
			time(&ptime);
			if ((ptime - stbuf.st_ctime) < atime) {
				/* file not old enough to delete */
				return(FAIL);
#ifdef BUGFIX
			/* * * * * * * * * * * * 
			 * repair window in LOCKING that occurs if onelock()
			 * fails,stat() fails then someone else locks 
			 * immediately after stat fails. The original code
			 * will remove lock if stat fails disregarding the
			 * posibility that someone locked after stat failed.
			 */
			}
			else {
				ret = unlink(file);
				ret = onelock(pid, tempfile, file);
#endif BUGFIX
			}
		}
#ifndef BUGFIX
		ret = unlink(file);
		ret = onelock(pid, tempfile, file);
#endif BUGFIX
		if (ret != 0) 
			return(FAIL);
	}


	This was discovered when the sequence file got munged frequently
running with a modified unix kernal and modified gename() routine in uucp,
however it could happen to anyone.

			....Larry

UUCP: ucbvax!ucla-va!lund
ARPA: lund at ucla-locus