On Fri, Mar 18, 2016 at 06:58:02PM +0100, David Sterba wrote: > On Fri, Mar 11, 2016 at 10:17:03PM +0000, Hugo Mills wrote: > > I know I promised this a while ago and didn't get round to it, but > > Henk's tinkering reminded me of it. I note specifically that the > > algorithm used to give the free space to plain old df gives incorrect > > results -- probably because it's not using the algorithm below. > > So the algorithm is supposed to implement btrfs_calc_avail_data_space, > the raid5/6 parts are missing from there, otherwise the structure of the > function follows the proposed algorithm. Well, the current code gets the free space estimation wrong even with RAID-1. It's not usually *very* wrong, but it's still not right, somewhere. > > There is an algorithm that (seems to) give the correct number of > > block groups which can be allocated, given a current allocation state > > of the FS. I haven't been able to prove it correct, but it seems to > > work. Here's some pseudocode for it: > > It does sound correct to me. Thanks for writing it up. Hugo. -- Hugo Mills | Great films about cricket: The Umpire Strikes Back hugo@... carfax.org.uk | http://carfax.org.uk/ | PGP: E2AB1DE4 |
Attachment:
signature.asc
Description: Digital signature
