On Thu, Jun 25, 2015 at 08:46:44AM -0400, Brian Foster wrote: > So as noted above, were this treated as a typical metadata block, the > corresponding buffer would be logged and thus everything would be > ordered appropriately with respect to the log. It is not a logged buffer > in this case, so we drop XFS_BMAPI_METADATA flag from the allocation > which would cause any previous frees of this block to be flushed to the > ail. Yes. > I don't see anything explicit that "handles" the case of being written > before the allocation transaction is committed. By "handles," do you > simply mean the aforementioned log force? In other words, we're > implicitly saying it's fine that the block is written out of order from > the allocation, so long as the block is marked free on-disk. Yes and yes. Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx _______________________________________________ xfs mailing list xfs@xxxxxxxxxxx http://oss.sgi.com/mailman/listinfo/xfs
