Re: 2.1.50 VFS: Busy inodes after unmount. Self-destruct in 5 seconds.

Linus Torvalds (torvalds@transmeta.com)
Sat, 16 Aug 1997 22:31:24 -0700 (PDT)

On 17 Aug 1997, Andi Kleen wrote:
>
> And when I stressed the dcache a little bit (I started a gcc compilation
> and continued my interactive work) the system quickly managed to corrupt
> a commonly accessed dcache entry: everytime a program tried to access
> or open it it crashed with a kernel null pointer reference in do_follow_link().

This isn't actually a NULL pointer dereference despite what the oops says:
it actually dereferences a pointer that is -1 (0xffffffff), but there's a
structure offset of 0x50 that makes it wrap around and look like it
dereferences a NULL pointer.

I've seen this bug reported by two or three people now, and I can't for
teh life of me see how it could happen: the "inode" pointer in the dentry
is totally corrupted (0xffffffff is definitely a bad inode pointer).

Could people who see the 0x4f "NULL pointer dereference" please tell me
what filesystems they have compiled in, and try to find if there is
something that brings on the problem. Also, I'd suggest trying the pre-51
patch from ftp.kernel.org, as it does fix a few known problems.

Linus