Check out the new USENIX Web site. next up previous
Next: New vnode lock location Up: NetBSD vfs changes Previous: Layering works

Most file systems do real locking

 To better support the vnode locking and chaining described in the next section, all file systems (except for UNIONFS and NFS) were changed to do vnode locking. Previously only file systems with on-disk storage actually did vnode locking, while the others merely returned success. As of NetBSD 1.5, the NFS file system is the only leaf file system which does not do real vnode locking. This defect remains as lock release and reacquisition during RPC calls to the NFS server has not been implemented. As mentioned above, the UNION file system has not been updated with these locking changes.

While it is not likely that one would want to layer a file system above many of the file systems which gained true locking (such as the KERNFS layer), this change makes it easier to impliment layered file systems. Lock management is one of the keys of getting layered file systems to work. By being able to rely on all leaf vnodes doing locking, the layered locking becomes much easier. Additionally the changes to add this support were not difficult.


next up previous
Next: New vnode lock location Up: NetBSD vfs changes Previous: Layering works
Bill Studenmund
2000-04-24