Re: [RFC PATCH] Introduce filesystem type tracking

["Tom Spink" <tspink@gmail.com>]

2008/5/20 Al Viro <viro@zeniv.linux.org.uk>:
> On Tue, May 20, 2008 at 02:06:42PM +0100, Tom Spink wrote:
> [snip]
>
>> I'm just adding people to CC here, but also I had a couple of thoughts
>> after reviewing my own code.
>>
>> I see that do_kern_mount is encapsulated with the BKL, but would it be
>> wise to introduce a lock (e.g. a mutex) now for reading and updating
>> nr_mounts (and hence calling ->init), rather than wait for the BKL
>> removal to come round here?
>>
>> Also, have I got all the cases where a filesystem is unmounted,
>> because I now see umount_tree, and am wondering if decrementing the
>> nr_mounts field should be done in here, in the loop of vfsmounts... or
>> is it sufficient to leave it at the end of do_umount?

Hi Al,

> No, you have not and no, doing that anywhere near that layer is hopeless.
>
>        a) Instances of filesystem can easily outlive all vfsmounts,
> let alone their attachment to namespaces.

I see!  Whoops...

>        b) What should happen if init is done in the middle of exit?

Okay, I guess *some* sort of locking is in order. :)

>        c) Why do we need to bother, anyway?

Well, just for the reason I mentioned, I saw the posting about XFS
starting threads (when compiled into the kernel), but there's no use
of an XFS filesystem at all - there was a suggestion that something
like this be tried, so I thought I'd give it a go.

Thanks for replying!

-- 
Regards,
Tom Spink