Re: cvs commit: src/sbin/mount Makefile mount.c
On Fri, Feb 02, 2007 at 07:25:19PM -0500, Craig Rodrigues wrote:
> On Sat, Feb 03, 2007 at 12:59:43AM +0100, Pawel Jakub Dawidek wrote:
> > Should we also restart mountd(8) on successful umount(8)?
> Your fix to mount(8) is a good one, since the existing behavior
> of mount(8) is to SIGHUP mountd, based on the pid
> in /var/run/mountd.pid.
> I don't like how these sneaky behaviors with non-obvious side-effects
> sneak into existing utilities like mount(8).....but that's what we have
> A sneaky side effect of the existing behavior, is that when you
> SIGHUP mountd, it deletes all existing NFS exports, re-reads
> /etc/exports, and then re-creates all the NFS exports.
> Some people find this very annoying, especially when you
> are already using an NFS export ( http://www.freebsd.org/cgi/query-pr.cgi?pr=kern/9619 ).
> I am looking at some patches from Andrey Simonenko which fix this
> problem, but they are quite intrusive changes to userland and kernel,
> so it is slow going on my part.
> So for now, I would recommend leaving umount alone, to avoid adding
> new side effects, even though based on existing behavior, it seems logical
> to add it.
> However, if this behavior is necessary for ZFS, then I won't oppose
> modifying umount to SIGHUP mountd. :)
> It seems a lot of cruft has accumulated in a simple utility like
> mount(8) over the years!
I also was surprised when I found what mount(8) is doing with mountd(8).
I don't like this behaviour, but well.
What I think we should do on SIGHUP is to create new exports list from
the given files, and compare all entries with the entries we got
already. Based on that comparsion we can only remove entires which
disappeared or changed and add entries which were added or changed.
Will it be hard to implement something like this or am I missing
Pawel Jakub Dawidek http://www.wheel.pl
FreeBSD committer Am I Evil? Yes, I Am!
Received on Sat Feb 03 2007 - 00:53:15 UTC
- application/pgp-signature attachment: stored