On Thu, Mar 14, 2019 at 12:41 PM aszlig <aszlig@nix.build> wrote: > > On Thu, Mar 14, 2019 at 09:47:23AM +0200, Amir Goldstein wrote: > > Overlayfs is not expected to modify the lower layer. > > That is pretty clear, I mean that's the whole point of overlayfs, right? :-) > > > OTOH, I can't really think anything that should break horribly if > > we allow overlayfs to update atime on a writable lower layer?? > > It wasn't using O_NOATIME prior to Linux 4.19 (which is where this was starting > to break our tooling), but the bisected commit I mentioned initially > (a6518f73e60e5044656d1ba587e7463479a9381a) was (implicitly) introducing > O_NOATIME, so to the contrary I'd say it would actually un-break networking > file systems with overlayfs. > Hmm, pre 4.19 was different, but it's true, I think that lower open files did not have O_NOATIME AFAIK. OTOH, ovl_path_open() from ovl_copy_up_data() does open lower file with O_NOATIME, so I am a bit surprised you did not see the same issue there. I suppose the answer will require another dive... Thanks, Amir.
