Re: [PATCH - RFC] VFS: disable new delegations during delegation-breaking operations

[Al Viro <viro@zeniv.linux.org.uk>]

On Thu, Sep 26, 2024 at 08:42:06AM +1000, NeilBrown wrote:

> I don't think so.
> The old delegated_inode_new will be carried in to vfs_rename() and
> passed to try_break_deleg() which will notice that it is not-NULL and
> will "do the right thing".
> 
> Both _old and _new are initialised to zero at the start of
> do_renameat2(), Both are passed to break_deleg_wait() on the last time
> through the retry_deleg loop which will drop the references - or will
> preserve the reference if it isn't the last time - and both are only set
> by try_break_deleg() which is careful to check if a prior value exists.
> So I think there are no leaks.

Yecchhhh...  What happens if break_deleg() in there returns e.g. -ENOMEM
when try_break_deleg() finds a matching inode?

I'm not even saying it won't work, but it's way too brittle for my taste ;-/