[9fans] Another namespace question

[9fans] Another namespace question

[Keith Nash]

This is a question about why namespaces were designed in the way that they were.

The command
bind path1 path2
where both path1 and path2 are directories, makes path2 a union directory consisting of path1 alone.  So far, so good.

Why was it decided that this command also does the following:
(a) any union directory previously bound to path2 is automatically unmounted, and is found to have disappeared if the new union directory at path2 is unmounted (this is not the case for Lunix mount); *but*...
(b) ...any union directory previously bound to a subdirectory of path2 is not unmounted.

A possible reason for (a) is to clean up the namespace by removing binds that are invisible; but then why (b)?

Keith.

Re: [9fans] Another namespace question

[Russ Cox]

It's an optimization that caught the common case,
which is mounting something new on /n/foo and
never looking back.  You're probably right that the
implicit unmount is more of a bug than a feature.

Russ

Re: [9fans] Another namespace question

[David Presotto]

[-- Attachment #1: Type: text/plain, Size: 345 bytes --]

The easy answer is that it was easier to implement.  The per-process
mount table knows nothing about the hierarchical structure of
the bindings.  When you bind something, it is to a file, not to a
path.  If you later bind to something higher in the tree, we
don't know enough to change the lower binding.

I do think that (b) is confusing.

[-- Attachment #2: Type: message/rfc822, Size: 1880 bytes --]

From: Keith Nash <kjn9@citizenearth.com>
To: 9fans@cse.psu.edu
Subject: [9fans] Another namespace question
Date: 3 Mar 2003 17:41:33 -0000
Message-ID: <20030303174133.15491.qmail@mail.dirac.net>

This is a question about why namespaces were designed in the way that they were.

The command
bind path1 path2
where both path1 and path2 are directories, makes path2 a union directory consisting of path1 alone.  So far, so good.

Why was it decided that this command also does the following:
(a) any union directory previously bound to path2 is automatically unmounted, and is found to have disappeared if the new union directory at path2 is unmounted (this is not the case for Lunix mount); *but*...
(b) ...any union directory previously bound to a subdirectory of path2 is not unmounted.

A possible reason for (a) is to clean up the namespace by removing binds that are invisible; but then why (b)?

Keith.