Re: [bcachefs] Mounting the same device multiple times now works

[Kent Overstreet <kent.overstreet@gmail.com>]

On Wed, Aug 12, 2015 at 12:36:02AM -0700, Ming Lin wrote:
> On Tue, 2015-08-11 at 17:38 -0800, Kent Overstreet wrote:
> > On Tue, Aug 11, 2015 at 10:59:42AM -0700, Ming Lin wrote:
> > > On Tue, 2015-08-11 at 00:34 -0800, Kent Overstreet wrote:
> > > > Like in other filesystems, you can now do
> > > > 
> > > > mount /dev/sda1 /foo
> > > > mount /dev/sda1 /bar
> > > > 
> > > > What this really means is that if the udev hook has registered the bcache device
> > > > via the old style sysfs interface for caching, you'll still be able to mount the
> > > > device like normal.
> > > 
> > > Thanks!
> > > 
> > > BTW, could you share the bcachefs long term development plans and short
> > > term high priority tasks?
> > 
> > Right now, my main priority is stabilizing it and making it usable - fixing all
> > the things that people are going to trip over when they try it out. I want to
> > get it somewhat more polished before I announce on linux-fsdevel.
> > 
> > The big thing I'm working on at the moment is debugging atomic renames, I had to
> > back that code out at the last minute.
> > 
> > > I'd like to pick something to work on. Probably the debugfs tool you
> > > mentioned or other things?
> > 
> > One thing that would actually be really helpful is more benchmarking vs.
> > ext4/xfs - identifying areas where there's still more performance work to do.
> > 
> > The debugfs tool is definitely important but that's a longer term thing. If
> > you're feeling productive like that feel free to keep going with it.
> > 
> > Really though right now the main thing we need is more testing and debugging.
> 
> I have run xfstests. Here is the results for generic tests.
> http://minggr.net/pub/bcachefs/xfstests.log
> 
> One thing is fallocate not supported yet. Is it known?
> 
> root@bee:~# mount |grep bcache
> /dev/sda on /mnt/test type bcache (rw,relatime)
> 
> root@bee:~# fallocate -l 4096 /mnt/test/hello.c
> fallocate: /mnt/test/hello.c: fallocate failed: Operation not supported

I do have code for fpunch/fcollapse in another branch - but it's not entirely
working yet. Fallocate itself is going to be more involved, we'll need some sort
of persistent reservations which are going to be a pain... should probably look
and find out what semantics btrfs implements first.