Re: [PATCH 07/19] fs: Add freezing handling to mnt_want_write() / mnt_drop_write()

[Jan Kara <jack@suse.cz>]

On Thu 29-03-12 12:19:43, Szeredi Miklos wrote:
> On Thu, Mar 29, 2012 at 1:43 AM, Jan Kara <jack@suse.cz> wrote:
> > Most of places where we want freeze protection coincides with the places where
> > we also have remount-ro protection. So make mnt_want_write() and
> > mnt_drop_write() (and their _file alternative) prevent freezing as well.
> > For the few cases that are really interested only in remount-ro protection
> > provide new function variants.
> 
> The underlying mechanism of mnt_want_write() and sb_start_write() is
> basically the same:
> 
>  - increment per-CPU counter
>  - mb
>  - check flag
>  - retry loop for races
  Well, __mnt_want_write() has the special handling of MNT_WRITE_HOLD.
Otherwise yes, the locking logic is rather similar.

> It is scalable but still there is a non-zero overhead, and duplicating
> it doesn't seem to make a lot of sense.  Why not instead try to use a
> common abstraction for both?
  I'm not sure here. Are you concerned about performance - i.e. you would
like to merge the remount-ro counter and topmost anti-freeze counter? That
cannot work because in some cases you want remount-ro protection but not
freezing protection. In particular when you have file open for writing,
you want to refuse remount-ro but still allow freezing.

If you are concerned about code / source size, the implementations of the
protection mechanisms could be possibly merged but I'm not sure it's worth
the hassle since we'd save like 50 LOC. MNT_WRITE_HOLD is one difference,
how each of the implementation detects whether we have succeeded in taking
our lock is another difference...

								Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR