Re: [lustre-devel] [PATCH 030/124] staging: lustre: llite: Replace write mutex with range lock

[Jan Kara]

On Mon 19-09-16 09:25:48, Dilger, Andreas wrote:
> On Sep 19, 2016, at 09:28, Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx> wrote:
> > 
> > On Sun, Sep 18, 2016 at 04:37:29PM -0400, James Simmons wrote:
> >> + * Range lock is used to allow multiple threads writing a single shared
> >> + * file given each thread is writing to a non-overlapping portion of the
> >> + * file.
> >> + *
> >> + * Refer to the possible upstream kernel version of range lock by
> >> + * Jan Kara <jack@xxxxxxx>: https://lkml.org/lkml/2013/1/31/480
> >> + *
> >> + * This file could later replaced by the upstream kernel version.
> > 
> > It doesn't look like range_lock ever got accepted in the kernel tree,
> > any idea what happened to it?  Having a per-filesystem lock type seems
> > odd to me...
> 
> I've added Jan and linux-fsdevel to the CC list to see what interest
> there is in the range locking implementaion.  At the time we added this
> to Lustre it appeared that this was moving nicely torward landing, but
> it seems to have stalled.
> 
> I think the range locking implementation is fairly generic, and if there
> are other users in the kernel it could easily be pulled out of the staging
> dir into vfs/.  I'm not against it going into vfs/ directly either, but
> not sure whether that is acceptable if the only user is in staging.

Yeah, so the problem with my range_lock implementation (and your one looks
fairly similar from a quick look) is that it is fairly heavyweight. That
may be OK for an inode_lock replacement but it needs a careful
benchmarking. Davidlohr was looking into making the implementation more
efficient (he wanted to use it for mmap_sem) but I think it got preempted
by other work.

								Honza
-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR