Jeff Layton <jlayton@xxxxxxxxxx> wrote: > This seems like something useful, but I wonder if it'll need some sort > of serialization vs. concurrent updates. Quite possibly, though that may be something that we have to delegate to the network filesystem. kafs, for instance, performs local serialisation of StoreData RPCs to any given inode because the server will exclusively lock the remote vnode around the write-to-disk and callback break (ie. change notification) phases. This does not preclude, however, another client making a change whilst the local lock is held. Of course, in such a case, we're into conflict resolution and may end up invalidating the local copy. > Can we assume that netfs itself will never call netfs_resize_file? Probably. Depends on how truncation gets handled. > Ceph already has some pretty complicated size tracking, since it can get > async notifications of size changes from the MDS. We'll have to consider > how to integrate this with what it does. Probably this will replace one > (or more?) of its fields. ceph's i_reported_size maybe? cifs has server_eof. > We may need to offer up some guidance wrt locking. i_lock may be a good fit. I wonder if it's worth at some point moving i_lock to being a seqlock so that various values ordinarily protected by it are accessible using read_seqbegin(). David -- Linux-cachefs mailing list Linux-cachefs@xxxxxxxxxx https://listman.redhat.com/mailman/listinfo/linux-cachefs
