On Sun, Dec 01, 2019 at 11:00:32AM +0200, Alex Lyakas wrote: > We can definitely adhere to the recommended behavior of setting > sunit/swidth=1 during mkfs, provided the repair still works after > mounting with different sunit/swidth. FWIW, sunit/sw=1fsb doesn't work reliably either, if you grow the sunit by a large enough value that it changes the alignment of the root inode.... Essentially, there is no magic bullet here. We need to ensure that the kernel doesn't do something that breaks assumptions about root inode location, and we need to fix xfs_repair not to trash the root inode without first checking that the current root inode pointer is invalid... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx