Miklos, This is the 2nd revision of patch set to optimize file handle decode. It is available on branch ovl-nfs-export on my github [1]. Patch 2 introduces a semantic change to exportfs_decode_fh() to allow callers to opt-out of reconnecting the dentry. I like this minimal semantic chance, but if you think I should factor-out a "_noreconnect" variant of exportfs_decode_fh(), I can do that. Also, in ovl_lower_fh_to_d(), if we get a disconnected directory dentry and we end up needing a connected dentry, we drop the reference to the disconnected dentry and call exportfs_decode_fh() again without opting-out of reconnect. I could also export exportfs_reconnect_path() for that case, but seems unnecessary to do that as the disconnected dentry will likely still be in cache anyway. To verify optimization, I ran all the overlay/exportfs tests with trace prints in index lookup and reconnect functions to verify that reconnect is not called when decoding a lower dir file handle of a deleted dir. Detailed analysis on this test results are in commit message of the last patch. Thanks, Amir. Changes since v1: - Discard hash inodes by file handle - Lookup in inode cache before lookup in index dir - Lookup underlying fs inode cache without reconnect lower dir [1] https://github.com/amir73il/linux/commits/ovl-nfs-export Amir Goldstein (3): ovl: disambiguate ovl_encode_fh() ovl: do not try to reconnect a disconnected origin dentry ovl: lookup in inode cache first when decoding lower file handle fs/exportfs/expfs.c | 9 +++++++ fs/overlayfs/copy_up.c | 6 ++--- fs/overlayfs/export.c | 70 +++++++++++++++++++++++++++--------------------- fs/overlayfs/namei.c | 20 +++++++------- fs/overlayfs/overlayfs.h | 7 ++--- 5 files changed, 66 insertions(+), 46 deletions(-) -- 2.7.4 -- To unsubscribe from this list: send the line "unsubscribe linux-unionfs" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
