Tidying up after Linux 16762

Consider a text editor, editing file "foo". When you write out your changes, it might do it this way:

1. write out a new file, "foo.tmp".

2. rename "foo" to "foo.bak".

3. rename "foo.tmp" to "foo".

Unless the editor knows about the metadata, you've now lost your metadata for foo.

Windows faced a similar problem when they added support for long file names. Basically, each file had both a long name and a short name. If an application that did not know about long names manipulated a file, you might lose the long name. The kludge they used was to have the system keep a cache of the long names buttociated with short names. When an application created a file using the short name APIs, Windows checked, and if there had recently been deleted or renamed a file with the same short name that had also had a long name, it gave that long name to the new file. I don't think people would want that kind of hack in Linux.

Pretty much every app that writes files ends up being a potential problem when you add a metadata system to a filesystem that didn't have it.

This problem also arises with access control lists. This is one of the reasons most practical uses of ACLs in Linux are on a per-directory basis, rather than a per-file basis.

-- --Tim Smith