On Mon, 19 Jul 1999, Ronald G. Minnich wrote: > Great idea. I liked it so much I bought the company -- er, I mean, I wrote > something like this. It's private name spaces for Linux and FreeBSD (among > others) and it allows you to mount things from remote file servers into > your name space.
I forgot to make this clear: it's a full-fledged VFS on linux supporting things like mmap etc. It is indistinguishable to any process as being anything other than a file system -- because it is a file system. It's just that mounts into the file system are controlled by your process, not by /etc/fstab. Different groups of processes can have totally different sets of mount points. In fact one test is to have a login that does a chroot to a totally empty tree. You then mount things like /usr/bin, etc. into your private tree and at that point all programs run quite normally. If you want ou even get translucent file system behaviour for free, since the file system supports union mounts and the Plan 9 semantics are that the "top" mount is where files get created. A user can thus have read-only mounts from a server or servers somewhere, mount local disk on top, and voila: all new files are local. This would also support booting from a CDROM and running even though the local disk is either nonexistant or empty of any system files. Stacking comes for free too, of course. ron p.s. I looked at the "VOP upcall" idea but there's lots of interesting issues as regards parameters in the VOP structures being handed out to a user process and then back to the kernel. I'm not sure it's a good idea. To Unsubscribe: send mail to majord...@freebsd.org with "unsubscribe freebsd-hackers" in the body of the message
