roland wrote: > what is simply want is that an nfs share behaves just as another ordinary > filesystem share.
Your sense of "ordinary filesystem" does not match reality; see below. > if i share a directory via samba, it doesn`t matter how the mount strcuture > looks like, i.e. it simply exports a directory structure. it`s irrelevant if > it`s a single tree or a tree which is built from 2314 different filesystems , > i.e. which has 2314 mountpoints inside. the user can just see (and reach) > dirs and files. > > same with ftp. > same with http. > same with rsync. > same with webdav. Your comparisons are all with applications that deal with filenames, and would have to work harder to notice when filesystems are crossed. NFS has almost always been implemented inside the kernel, and services in the kernel cannot help but notice when one filesystem stops and another begins. Our in-kernel NFS service has always represented separate filesystems separately, and our in-kernel CIFS service now does the same. I realize that this is not always convenient, but it's been reality for 20 years of NFS, and was also the case for CIFS until Samba. Your client needs the smarts to handle this, and we can now with NFSv4 and mirror mounts. Perhaps because of the influence of Samba, the Linux folks put an option in their server to cover up the fact that nested filesystems were separate; that is a bad idea for reasons already discussed, and you won't find that on most NFS servers. And we won't implement that option on Solaris. Rob T
