Re: [PATCH 12/13] sysfs: Propagate renames to the vfs on demand

[ebiederm@xmission.com (Eric W. Biederman)]

Miklos Szeredi <miklos@szeredi.hu> writes:

> On Fri, 06 Nov 2009, ebiederm@xmission.com (Eric W. Biederman wrote:
>> Tejun Heo <tj@kernel.org> writes:
>> 
>> > Hello,
>> >
>> > Eric W. Biederman wrote:
>> >> It isn't what I want but it is what the VFS requires.  If let the vfs
>> >> continue on it's delusional state we will leak the vfs mount and
>> >> everything mounted on top of it, with no way to remove the mounts.
>
> "umount -l" on the whole thing will clear any submounts up too.
>
>> >
>> > This is caused by not having any way to prevent deletion on
>> > directories with submounts, right?  How does other distributed
>> > filesystems deal with directories with submounts going away underneath
>> > it?
>> 
>> NFS does exactly the same thing I am doing.
>
> Yes, this is a problem for NFS too.  You cannot tell the NFS server
> "this directory is mounted on some client, don't let anything happen
> to it!".  Basically the remaining choices are:
>
>  a) let the old path leading up to the mount still be accessible, even
>  though it doesn't exist anymore on the server (or has been replaced
>  with something different)
>
>  b) automatically dissolve any submounts if the path disappeard on the
>  server
>
> I think Al was arguing in favor of b), while Linus said that mounts
> must never just disappear, so a) is better.  I don't think an
> agreement was reached.

I haven't seen that conversation.  I do know it is non-intutive and if
you attempt to delete what is a mount point in another mount namespace
and it won't go away.  (What we do for non-distributed filesystems).
So I would favor mount points dissolving if we had the infrastructure.

Regardless the goal for now is to simply catch up with other distributed
filesystems.

Eric