On Tue, May 16, 2017 at 08:47:43AM +0200, Richard Weinberger wrote: > Eric, > > Am 16.05.2017 um 01:25 schrieb Eric Biggers: > > On Mon, May 15, 2017 at 09:51:03PM +0200, Richard Weinberger wrote: > >>> > >>> The test is repeatedly creating and removing a directory "dir" while lookups are > >>> being done in it. It seems the problem is that many dentries are being created > >>> for "dir", and they pin many different inodes, all at the same time. This > >>> actually happens for ext4 too; it just doesn't cause an observable error. > >>> > >>> I doubt it's the right solution to make fscrypt_d_revalidate() look at > >>> ->i_nlink, since ->d_revalidate() is meant to validate the filename, not the > >>> inode. I think there is probably a VFS bug that is causing the dentries to not > >>> be freed. > >> > >> Not sure. Al? :-) > >> > > > > I can reproduce this on an unencrypted directory after updating path_init() in > > fs/namei.c to always clear LOOKUP_RCU, so that all path lookups are done in > > ref-walk mode. So I think fscrypt_d_revalidate() was only relevant because it > > causes all path lookups to drop out of rcu-walk mode. > > On ext4 or UBIFS? > > Thanks, > //richard Both; the inode "leak" isn't filesystem-specific, beyond the fact that UBIFS apparently limits the number of inodes on its orphan list while ext4 does not. (How do I know it happens on ext4 then? /proc/slabinfo shows that lots of ext4 inodes have been allocated, and after an unclean shutdown and remounting, it's reported that 50,000+ orphan inodes were deleted.) There are lots of cases in which path lookups switch from rcu-walk mode into ref-walk mode, so the fact that it was being caused by ->d_revalidate() in this specific situation isn't really important, IMO. I think the actual fix would be something along the lines of making vfs_rmdir() unhash any negative child dentries, so that they get "killed" by dput() later. Eric
