On Wed, Apr 14, 2021 at 1:56 PM Greg Kurz <groug@xxxxxxxx> wrote: > > On Mon, 12 Apr 2021 17:08:26 +0200 > Miklos Szeredi <miklos@xxxxxxxxxx> wrote: > > > On Mon, Apr 12, 2021 at 4:59 PM Vivek Goyal <vgoyal@xxxxxxxxxx> wrote: > > > > > > Hi Miklos, > > > > > > Robert Krawitz drew attention to the fact that fuse does not seem to > > > have a ->sync_fs implementation. That probably means that in case of > > > virtiofs, upon sync()/syncfs(), host cache will not be written back > > > to disk. And that's not something people expect. > > > > > > I read somewhere that fuse did not implement ->sync_fs because file > > > server might not be trusted and it could block sync(). > > > > > > In case of virtiofs, file server is trusted entity (w.r.t guest kernel), > > > so it probably should be ok to implement ->sync_fs atleast for virtiofs? > > > > Yes, that looks like a good idea. > > > > I've started looking into this. First observation is that implementing > ->sync_fs() is file server agnostic, so if we want this to only be used > by a trusted file server, we need to introduce such a notion in FUSE. > Not sure where though... in struct fuse_fs_context maybe ? Yep, makes sense. Thanks, Miklos
