Hello, Josef.

On Wed, Aug 10, 2016 at 05:16:03PM -0400, Josef Bacik wrote:

quoted

It bothers me a bit that sb's can actually be off bdi->sb_list while
sb_list_lock is released.  Can we make this explicit?  e.g. keep
separate bdi sb list for sb's pending metadata writeout (like b_dirty)
or by just walking the list in a smart way?

Yeah I wasn't super thrilled with this either, but since I'm only using it
for writeback I figured it was ok for now.  I suppose I could make it less
sb_list and just make it dirty_sb_list, and only add if the super says it
has dirty metadata.  Does that sound better?  Then us being off of the list
during writeback isn't that big of a deal.

Yeah, that feels more logical to me.

quoted

I can't find where sb's are actually added to the list.  Is that
supposed to happen from specific filesystems?  Also, it might make
sense to split up additions of sb_list and stat into separate patches.

I was going to be lazy and only add it if we cared about writing out
metadata, and then we could expand it to everybody if somebody else had a
usecase.  But I think maybe the b_sb_dirty list idea is a better fit overall
so I can just do that so it makes more sense.  I can split up these patches
as well, thanks,

b_sb_dirty list sounds good and we wouldn't have to worry too much
about lifetime issues either as the list would be linked iff the sb is
dirty.  It might still make sense to ensure that the list is unlinked
when sb is putting the bdi tho.

Thanks.

-- 
tejun