o bugfix - instantiate-revalidate race Caching d_inode should be after write_locking the aufs dentry. Otherwise a race condition happen between instantiate and revalidate on the same dentry and aufs_d_revalidate() makes the dentry invalid incorrectly. But there is nothing wrong happens except a small overhead. This problem was discoverd during a test on linux-3.17-rc1 which prohibits aliased dentries on a single directory inode explicitly. - possible bugfix, temporary d_inode d_inode is protected by d_lock, so lock/unlock it in au_cpup_or_link(). Actually there is nothing wrong happened without d_lock since the aufs dentry is already write-locked. But lock/unlock should be here theoretically.
J. R. Okajima ---------------------------------------- - aufs3-linux.git#aufs3.9..aufs3.16 branch aufs: bugfix, instantiate-revalidate race aufs: possible bugfix, temporary d_inode aufs: tiny, cosmetic cleanup aufs: debug, au_dbg_mtx to serialize the messages aufs: fhsm and br_del, allow the root dir only 1/2 aufs: fhsm and br_del, allow the root dir only 2/2 aufs: MAINTAINERS aufs: tiny, cosmetic level optimization - aufs3-linux.git#aufs3.x-rcN branch Addition to above, aufs: for linux-3.17, support for d_materialise_unique() - aufs3-standalone.git ditto - aufs-util.git none ------------------------------------------------------------------------------ Slashdot TV. Video for Nerds. Stuff that matters. http://tv.slashdot.org/