[TUHS] Re: Patches to improve 6th Edition

[Brian S Walden]

Yes there are other bugs you will want to fix as well
if you are using v6 heavily and not on slow hardware
(e.g. a simulator).  In the inode allocation route
(ialloc()) in /usr/sys/ken/alloc.c there is a bug in
the placement of the loop: tag.  You just have to move
it up two line to above the while() loop. See the diff below.
This rare Ken Thompson bug was found in 1988 on an Amdahl UTS machine
that the hardware finally became fast enough that it caused us panics.
This bug existed from (at least) the Fifth up until the Eighth
editions from 1127 (BTL research). The Ninth edition had much
rewrites in the kernel and the algorthim was changed at that point.
Since it was in UTS, it must have been in the porting base of
System V Release 2. This code is the basis for the System V
filesystem, so it probably didn't exist in BSD releases
since they used their fast filesystem (cylinder groups, et al.)
which became UFS. It's amazing thet with the amount of eyes on it
(via Lions Commentary, etc.) it was not spotted earlier.

It would be good to get a collection of all "unreleased" bug fixes
that others may have.


*** alloc.c.orig        Fri Dec 27 23:23:30 2002
--- alloc.c     Fri Dec 27 23:23:30 2002
***************
*** 163,171 ****
        int i, j, k, ino;
  
        fp = getfs(dev);
        while(fp->s_ilock)
                sleep(&fp->s_ilock, PINOD);
- loop:
        if(fp->s_ninode > 0) {
                ino = fp->s_inode[--fp->s_ninode];
                ip = iget(dev, ino);
--- 163,171 ----
        int i, j, k, ino;
  
        fp = getfs(dev);
+ loop:
        while(fp->s_ilock)
                sleep(&fp->s_ilock, PINOD);
        if(fp->s_ninode > 0) {
                ino = fp->s_inode[--fp->s_ninode];
                ip = iget(dev, ino);


In article by Mirian Crzig Lennox:
> This reminds me that I fixed a couple of "bugs" (some outright bugs,
> and some mere behaviours which I didn't quite like) in playing with my
> V6/simh environment, and I'm wondering if there is a place I should
> submit them in case others might be interested.