On Wed, Oct 3, 2018 at 7:17 PM Jan Stancek <jstancek@xxxxxxxxxx> wrote: > [...] > > > Would it make sense to enable this for 'all_filesystems = 1'? > > > > I don't know. Not sure which tests are good candidates for that. > > But anyway it's a different change. > > [...] > > Isn't the reason for your series to test on different fs as well? :-) > Well, yes and no. Yes, overlayfs is a filesystem driver. No, I don't think overlayfs was what the author of all_filesystems had in mind, because setting .dev_fs_type = "overlay" is not enough to describe the setup. That sort of setup would need to be described something like this: .dev_fs_type = "foo", .mntpoint = MNTPOINT, .overlay_mount = OVL_MNT, So overlayfs does not belong on the all_filesystems axis IMO, it belongs on a tangent axis of the test matrix, e.g.: .all_filesystems = 1, .needs_overlay = 1, .mntpoint = MNTPOINT, .overlay_mount = OVL_MNT, I personally question the value of such feature, but putting something like setup_overlay() in a library is probably not a bad idea. A similar pattern already exists in inotify07/inotify08, however in those tests as well as in readahead02, a feature of .need_overlay = 1 if existed would not have been sufficient. Thanks, Amir.
