On Tue, Jul 14, 2020 at 02:46:49PM -0700, Darrick J. Wong wrote: > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > Repair has this strange behavior during phase 4 where it will zero the > parent pointer entry of a shortform directory if the pointer is > obviously invalid. Unfortunately, this causes the inode fork verifiers > to fail, so change it to reset bad pointers (ondisk) to the root > directory. If repair crashes, a subsequent run will notice the > incorrect parent pointer and either fix the dir or move it to > lost+found. > > Note that we maintain the practice of setting the *incore* parent to > NULLFSINO so that phase 7 knows that it has to fix the directory. I think we probably want to take Brians series instead, which does a few more things in the area.
