Re: [PATCH] more autofs4 fixes for 2.3.99-pre5

[Linus Torvalds]

On Thu, 13 Apr 2000, Jeremy Fitzhardinge wrote:
>
> Here's another set of fixes which incorporates the patch this morning
> (ie, its against pristine 2.3.99pre5).

Hmm.. I like the fact that it removes a lot of code.

However, I also wonder what keeps the dentries around? I don't see why a
currently unused dentry would notbe thrown away by the VFS layer, leaving
an inode free'd that now is no longer reachable because the inode isn't
hashed.

Basically, as far as I can tell the VFS layer may garbage-collect the
dentries when memory gets low. I didn't look all that closely at it, but
this is the reason why ramfs does the extra "dget()" when creating a new
positive dentry and does an extra "dput()" when deleting a dentry.

Note that this may be hard to see in practice - especially as most autofs
dentries end up being mount-points (and then the _mount_ will keep the
dentry active). It looks like a bug none-the-less.

Maybe I missed something.

		Linus