InodePager

ABS /afs/cs.washington.edu/project/spin/yasushi/spin/user/vm/inodepager/src, spin/user/vm/vmcore/doc/MemoryObject.html. Inode pager manages memory objects that back files. Such memory objects are useful when you want to access files using usual CPU memory instructions.

Inode manager does not handle pageout properly; it just throws away all the pages.

Inode pager has a table that remembers what object it has created, and if the Create request is issued for the file for which the memory object already exists, the inode pager just returns the old one. The table entry is deleted as soon as the memory object gets isolated, i.e., the memory object is unmapped from all the address spaces.

Thus, if a process opens a file A , mmaps, does work, and terminates, the memory object for A is destroyed when the process terminates. When another process mmaps A, another memory object is created.

This table hack will be replaced by generic file caching mechanism in the future.

INTERFACE InodePager;
IMPORT MemoryObject;
IMPORT CacheObject;
IMPORT File;

PROCEDURE Create(file: File.T; cache: CacheObject.T := NIL;
name: TEXT := NIL): MemoryObject.T;

Create a memory object. If cache is NIL, then the default cache object is created. name is used just for debugging.

END InodePager.

Last updated: Mon Oct 14 14:23:28 PDT 1996

Yasushi Saito (yasushi@cs.washington.edu)