On Tue, 4 Jun 2024 at 11:31, Peter-Jan Gootzen <pgoot...@nvidia.com> wrote: > > On Tue, 2024-06-04 at 11:18 +0200, Miklos Szeredi wrote: > > External email: Use caution opening links or attachments > > > > > > On Tue, 4 Jun 2024 at 11:08, Peter-Jan Gootzen <pgoot...@nvidia.com> > > wrote: > > > > > Option 2 is detectable if fuse_init_out.minor < CANON_ARCH_MINOR, > > > not > > > sure yet what we could do with that knowledge (maybe useful in error > > > logging?). > > > > Using the version for feature detection breaks if a feature is > > backported. So this method has been deprecated and not used on new > > features. > Oh that is very good to know. So for new features, feature detection is > only done through the flags? > > If so, then in this case (and correct me if I'm wrong), > if the client doesn't set the FUSE_CANON_ARCH flag, the server/device > should not read the arch_id.
Since reserved fields are zeroed, it's possible to check for arch_id being zero (meaning the client has unknown arch). So if the client sets the arch but doesn't set FUSE_CANON_ARCH, it would mean that it does not support translation for this particular architecture. The server can still check to see if the arches match, continue if so, and error out otherwise. > As this is in some sense a bug-fix for certain systems, would this new > feature qualify for backporting? Certainly. Thanks, Miklos
