next up previous contents
Next: 9.2.12 Applicfs Up: 9.2 In-Core File Systems Previous: 9.2.10 Hlfs

9.2.11 Automountfs

A file system that would perform automounter functions much like Amd  [Pendry91] does, but in the kernel. It can therefore avoid locking and work much faster. There is only one problem: Amd as it stands knows about the underlying types of file systems that it automounts. If Automountfs will have to know the same, it will violate the symmetry principle of stackable file systems. One solution is to move only part of the automounter code into the kernel, and keep the mount-specific code outside the kernel. This is exactly what Sun had done with Autofs [Callaghan93]: most of the code was moved into the kernel, but the actual mounting is initiated by a user-level daemon called automountd. Autofs talks to this daemon using RPCs initiated from the kernel. My file systems would be able to make use of kernel based RPCs to communicated with user-level (or remote) servers.

Erez Zadok