Re: autofs: kernel-assisted automounter

H. Peter Anvin (hpa@transmeta.com)
Mon, 7 Apr 1997 23:14:07 -0700 (PDT)


> MdI> What is the difference between autofs and amd?
>
> Also, how does autofs compare with the Solaris automounter? (The latter is
> actually pretty nice, because it's invisible and is also coincidentally
> called autofs.)

Hardly a coincidence; I stole ideas and naming conventions fairly
liberally from Solaris, since in this particular case I think Sun did
something fairly reasonable in Solaris. I *do* think the Solaris
automounter is overfeatured, but that is another issue. The Linux
automounter daemon will probably only support a subset of the Solaris
features, unless some poor sucker actually writes a parser module for
me -- I won't have that kind of time available (sorry).

(The next version of the autofs tools will have the lookup and
parse/mount modules split; I plan to include a few more kinds of
lookup maps.)

> I vaguely recall problems with kernel-level automounters in general really
> messing things up on a hanging mount, and that multithreading within the
> automounter was the answer. How does Linux autofs handle this?

I have no idea what you are talking about here. The only thing I can
think about is if you try to do name lookups or mounts in the main
loop of the automount daemon you will lose; however, my daemon forks a
process to do any dirty (= potentially slow) work on a mount point,
avoiding that problem. Using *kernel* threads would be significantly
nicer, but I do not plan to use threads until kernel-based threading
libraries are closer to universally available.

-hpa