Miklos, This is v2 of the ino patches. v1 is here [1]. I reabsed to overlayfs-next and addressed your comments on the ino collision patch. The branch passes overlay xfstests including the new tests 07[01] that I wrote to test this series. Note that i_ino uses the private atomic counter not only for xino overflow case, but also for non-samefs with xino disabled, but it is only used for directory inodes. I don't think that should cause any performance regressions and the kernel gets rid of a potentially massive abuser of the global get_next_ino() pool. Thanks, Amir. Changes since v1: - Cleanup patches merged - Use private atomic counter for non-persistent ino - Don't abuse i_generation - Keep fsid an index to fs array (fewer magic shifts) - Added xino_mode fix patch (for v5.6), which includes disabling xino on 32bit kernel [1] https://lore.kernel.org/linux-unionfs/20200101175814.14144-1-amir73il@xxxxxxxxx/ Amir Goldstein (5): ovl: fix some xino configurations ovl: use a private non-persistent ino pool ovl: avoid possible inode number collisions with xino=on ovl: enable xino automatically in more cases ovl: document xino expected behavior Documentation/filesystems/overlayfs.rst | 38 +++++++++++++++- fs/overlayfs/Kconfig | 1 + fs/overlayfs/inode.c | 58 ++++++++++++++++++------- fs/overlayfs/overlayfs.h | 16 +++++++ fs/overlayfs/ovl_entry.h | 2 + fs/overlayfs/readdir.c | 25 ++++++++--- fs/overlayfs/super.c | 35 ++++++++------- 7 files changed, 136 insertions(+), 39 deletions(-) -- 2.17.1
