On Fri, Sep 23, 2022 at 05:57:04PM +0200, Miklos Szeredi wrote: > On Fri, 23 Sept 2022 at 17:47, Christian Brauner <brauner@xxxxxxxxxx> wrote: > > > > On Fri, Sep 23, 2022 at 05:38:34PM +0200, Miklos Szeredi wrote: > > > It's basically just like when you copy a file betweed idmapped mounts: > > > > mount -o X-mount.idmap=<mapping1> --bind /source /source-idmapped > > mount -o X-mount.idmap=<mapping2> --bind /target /target-idmapped > > > > cp /source-idmapped/file1 /target-idmapped > > > > where you need to take the source and target idmappings into account. > > > > So basically like what ovl_set_attr() is doing just for acls. But I > > But then before this patch the copy-up behavior was incorrect, right? > > In that case this needs to be advertised as a bug fix, rather than a cleanup. No, it was correct. The testsuite I added would've yelled loudly if that were buggy. It's just that this was done in vfs_getxattr() before. But that is rather hacky and is one of the things we're getting rid off. If vfs_get_acl() did that it would return mount not filesystem values which isn't correct for the general case.
