Re: [PATCH/RFC] VFS: don't keep disconnected dentries on d_anon

From: Al Viro
Date: Wed Dec 20 2017 - 18:16:59 EST


On Thu, Dec 21, 2017 at 09:45:40AM +1100, NeilBrown wrote:

> -c/ Helper routines to allocate anonymous dentries, and to help attach
> + prefix. If the refcount on a dentry with this flag set
> + becomes zero, the dentry is immediately discarded, rather than being
> + kept in the dcache. If a dentry that is not already in the dcache
> + is repeatedly accessed by filehandle (as NFSD might do), an new dentry
> + will be a allocated for each access, and discarded at the end of
> + the access. As there is no parent, children, or name in the dentry
^^^^^^^^
That part is where I have a problem with it. Consider nfsd failing to
reconnect a growing subtree with the root. It has managed to get to
some point, but then failed to get the parent for some reason (IO error, OOM,
anything). Now we have a non-trivial subtree; its root does have children,
but it's not connected to anything. It has been created by d_obtain_alias();
in __d_obtain_alias() IS_ROOT() had been true, and so was 'disconnected'
argument.

The question is not whether they carry any valuable information - it's
whether we are guaranteed that they won't need pruning on umount. And
they will - the invariant we maintain is that all descendents will have
DCACHE_DISCONNECTED set until the sucker is reconnected to root. That's
why they won't stick around - nothing in such subtree will be retained in
dcache once the refcount hits 0.

I believe that the actual changes are OK, but your explanation above is
wrong and the logics there is convoluted enough, so this needs to be
written accurately.

BTW, I would like comments from Lustre folks - the situation with dcache
in there is rather unusual.