🚀 go-pugleaf

[ I post this here because we have no official way of notifying AT&T (nor,
  I suspect, enough incentive to do so or their attention if we did), and
  several machines running 3.1-derived code has this bug so this also
  notifies all the vendors of such machines.  A bug report has been
  called in to SGI. ]

OS version:	System V 3.1

Description: closedir() is supposed to deallocate the DIR * returned by
	opendir(), and close the filedescriptor whose value is stored
	in the DIR * structure.  This is done by first freeing the DIR *
	structure and then trying to get the value of the file descriptor
	stored within it.  The semantics of free doesn't guarantee data
	integrity of freed memory.

Symptom: an application repeatedly scanning a directory (i.e. opendir()
	readdir() closedir() in a loop) will run out of file descriptors
	when linked with a version of free that scribbles on the memory
	it frees (in order to catch these kinds of problems).

Workaround: instead of a normal closedir(), use the sequence:

	close(dirp->dd_fd);
	(void) closedir(dirp);	/* will return failure indication */

Severity: Embarrassing...
	Should be fixed in future releases (if someone can check on
	V.4 and let us and/or AT&T know...).

Fix:	in lib/libc/port/gen/closedir.c

RCS file: RCS/closedir.c,v
retrieving revision 1.1
diff -b -c -r1.1 closedir.c
*** /tmp/,RCSt1a09053	Mon Oct 23 22:01:05 1989
--- closedir.c	Mon Oct 23 22:01:06 1989
***************
*** 35,41 ****
  closedir( dirp )
  register DIR	*dirp;		/* stream from opendir() */
  {
  	free( dirp->dd_buf );
  	free( (char *)dirp );
! 	return(close( dirp->dd_fd ));
  }
--- 35,43 ----
  closedir( dirp )
  register DIR	*dirp;		/* stream from opendir() */
  {
+ 	int fd = dirp->dd_fd;
+ 
  	free( dirp->dd_buf );
  	free( (char *)dirp );
! 	return(close( fd ));
  }