Re: [PATCH] First casuality of hlist poisoning in 2.5.70

[Linus Torvalds]

On Wed, 11 Jun 2003, Trond Myklebust wrote:
>
> AFAICS one should not rehash the dentry until after the d_move(). Does
> that make sense?

Yeah, it does seem that rehashing before actually calling d_move() means 
that there is a small window where another process might now come in, and 
use the dcache (without getting the semaphore) to see the old value of the 
target dentry, even though the low-level filesystem has already move the 
new dentry value over the target. Ie the window would be between

		i_op->rename(...)
	d_rehash(new_dentry)
	... race here ...
	d_move(old_dentry, new_dentry)

That might confuse a filesystem that expected that the target was deleted 
an no longer reachable by anybody.

Al? What do you think?

		Linus