Comment 68 for bug 769927

(In reply to comment #11)
> Created attachment 501306 [details]
> info on root dentry, post-oops
>
> Poking around with crash gives me this info on the root dentry, post-oops. The
> d_flags are:
>
> d_flags = 0xc004 = DCACHE_DISCONNECTED|DCACHE_OP_REVALIDATE|DCACHE_OP_DELETE
>
> ...nothing terribly exciting there. d_parent points to itself (as expected),
> and all of the list_heads seem to be empty.

It's about time for a complete stab in the dark!

It's odd that the dentry above hasn't been __d_drop()ed ....

Which call of shrink_dcache_for_umount_subtree() does this oops on,
the first call or the second in shrink_dcache_for_umount()? The second
call is the one that walks the anonymous dentry list so that should be
the one. And the first thing shrink_dcache_for_umount_subtree() does
is __d_drop() the dentry. If it is the first call then how does the
anonymous dentry come to be under s_root?

And why does shrink_dcache_for_umount() decrement the count for the
s_root dentry but not for dentrys found on the sb->s_anon list, David?