Thoughts on Multi-Symlink Concept

[Brooks Davis]

On Mon, Feb 24, 2014 at 06:31:08PM +0000, Robert Watson wrote:
> On Sat, 22 Feb 2014, Jordan Hubbard wrote:
> 
> >> Yes, please can we get these ....
> >>
> >> Apollo Domain systems had those, and they were great. Set SYSTYPE to BSD4 
> >> and get the BSD tree and all that came with it, or SYSV to get the other 
> >> stuff.
> >
> > Yep, I loved these things on Domain/OS!  We system admin types used them to 
> > do all kinds of clever (and useful) things.
> >
> > Looks like FreeBSD has actually *had* an implementation for 6 years now.  I 
> > don?t necessarily agree with the architectural decision to create a 
> > different namespace and command (varsym) to manipulate it - it was really 
> > nice just having it be a part of the standard environ(7) - but hey, any 
> > implementation is better than no implementation.  Whatever happened to 
> > https://wiki.freebsd.org/200808DevSummit?action=AttachFile&do=get&target=variant-symlinks-for-freebsd.pdf 
> > ?
> 
> Some care is required here: at least one of the past implementations floating 
> around had the neat property that user-defined symlink expansions occurred 
> before system-defined ones, even for setuid binaries.  This allowed trivial 
> replacement of libraries out from under a binary, making rooting boxes easy. 
> I'm actually a fan of variant symlinks as well, having used them in AFS -- I'd 
> just prefer we aim for a model that minimises inconvenient rooting of boxes. 
> (I'm not passing judgement on this particular patch, mind you.)  I believe 
> Brooks Davis did the last serious pass at variant symlinks and might opine 
> further on the topic.

The version I wrote worked find last time I ported it forward (20 months
or so ago).  There's a copy in svn that's the latest.  The thing that's
held me back from committing it is lack to time to do some solid macro
and micro benchmarks to determine what the performance impact is when
it's compiled in, but disabled.  It should probably just be updated,
reviewed by a VFS person and committed.

I believe I've successfully avoided the pitfalls Robert describes.

-- Brooks
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 326 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20140224/deb239bc/attachment.sig>