Re: Anybody working on removing sendmail from base?

[Christopher Vance <vance@xxxxxxxxxx>]

On Sun, Sep 28, 2003 at 10:06:58AM -0700, Matthew Dillon wrote:
>    of this.    Variant symlinks is something we probably should implement.

Does this need at least one of the following?

1) make symlinks vary only over stuff already known to the kernel
(Pyramid OSx only did it over universe [att vs usb], and Compaq Tru64
over member number within cluster; I think Convex OS did something
similar to Pyramid)

2) ensure that user space can't fiddle with things (even accidentally)
to make it hard for the kernel to find and use the relevant
environment variables

3) delegate at least some part of name lookup to userspace code within
the requesting process (didn't Apollo Aegis do this with many
syscallish things?) even if it's just "look for relevant
metacharacters within path, and expand if found, before invoking the
syscall with these all resolved"

4) kernel-managed file namespaces (potentially) separate per process
(Plan 9 rfork)

--
Christopher Vance