Hi all, Al Viro posted a proposal to cleanup overlayfs handling of temporary cloned real file references [1]. This is a proposal in the opposite direction to get rid of the temporary cloned file references, because they are inefficient and cause for ugly subtle code. Al, I think that with these changes, overlayfs no longer has value in using the proposed fderr struct? FWIW, struct backing_file has no dedicated memcache pool - before Christian's diet to struct file, struct backing_file was 248 bytes on x86 and now it is 200 bytes, so the addition of 8 more bytes to strucy backing_file changes nothing wrt memory usage. Miklos, The implementation of ovl_real_file() helper is roughly based on ovl_real_dir_file(). do you see any problems with this approach or any races not handled? Thanks, Amir. [1] https://lore.kernel.org/linux-fsdevel/20241003234534.GM4017910@ZenIV/ Amir Goldstein (4): ovl: do not open non-data lower file for fsync ovl: stash upper real file in backing_file struct ovl: convert ovl_real_fdget_meta() callers to ovl_real_file_meta() ovl: convert ovl_real_fdget() callers to ovl_real_file() fs/file_table.c | 7 ++ fs/internal.h | 6 ++ fs/overlayfs/file.c | 230 +++++++++++++++++++++++--------------------- 3 files changed, 133 insertions(+), 110 deletions(-) -- 2.34.1
