Re: [PATCH 4/6] kill-the-BKL/reiserfs: release the write lock inside get_neighbors()

[Frederic Weisbecker <fweisbec@gmail.com>]

On Fri, May 01, 2009 at 07:51:35AM +0200, Ingo Molnar wrote:
> 
> * Frederic Weisbecker <fweisbec@gmail.com> wrote:
> 
> > get_neighbors() is used to get the left and/or right blocks 
> > against a given one in order to balance a tree.
> > 
> > sb_bread() is used to read the buffer of these neighors blocks and 
> > while it waits for this operation, it might sleep.
> > 
> > The bkl was released at this point, and then we can also release 
> > the write lock before calling sb_bread().
> > 
> > This is safe because if the filesystem is changed after this lock 
> > release, the function returns REPEAT_SEARCH (aka SCHEDULE_OCCURRED 
> > in the function header comments) in order to repeat the neighbhor 
> > research.
> > 
> > [ Impact: release the reiserfs write lock when it is not needed ]
> 
> This should also be safe because under the BKL we _already_ dropped 
> the lock when sb_bread() blocked (which it really would in the 
> normal case).
> 
> There's one special case to consider though: sb_read() maps to 
> __bread() which can return without sleeping if the bh is already 
> uptodate. So if the filesystem _knows_ that the bh is already 
> uptodate and holds a reference to it (this is common pattern in 
> filesystems), it can have a locking assumption on that.
> 
> No such assumption seems to be present here though.
> 
> 	Ingo


Yeah, fortunately it doesn't base its check on the state of the
buffer but on the tree number of rebalancing.

But I have to remember this pattern, it could be present elsewhere
in reiserfs.

Thanks.