Btrfs has always had a dummy inode that we used to allocate pages for our metadata. This has allowed us to take advantage of balance_dirty_pages() since our dirty metadata is unbounded otherwise. This has worked fine for years, but now we want to add sub pagesize blocksize support. The easiest way to do this would be to just kmalloc() our pages, since we already do all of our buffer management ourselves anyway. But in order to switch over to that we need to kill the metadata inode and allow us to make our own allocations for metadata. Enter these two patches. The first one is fairly straightforward. balance_dirty_pages() works on a per-bdi case, the only reason we pass mapping around is so we can check and see if the fs we are working against has writeback cgroups enabled. So this change is just changing function arguments and has no behavior change at all. The second patch is where I'd like a little more attention. I've added some helpers to deal with the dirty metadata page accounting. Basically this just adds some page stats that balance_dirty_pages() can take into account when deciding whether to kick the background writeback. Then I've added a sb callback to handle writing back dirty metadata, and I've added a list to the bdi for the sb's attached to that bdi. My plan is to just add/remove the btrfs sb to it when we mount/umount. Any suggestions are welcome, this is relatively self contained and doesn't complicate things too much so I think works well, but I may not be thinking of other cases. Thanks, Josef -- To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
