Re: [PATCH v2 00/16] nfsd/sunrpc: add support for a workqueue-based nfsd

[Al Viro <viro@ZenIV.linux.org.uk>]

On Fri, Dec 12, 2014 at 03:02:06AM +0000, Al Viro wrote:
> On Thu, Dec 11, 2014 at 06:29:37PM -0800, Linus Torvalds wrote:
> > On Thu, Dec 11, 2014 at 6:12 PM, Al Viro <viro@zeniv.linux.org.uk> wrote:
> > >
> > > Linus, do you see any problems with the following patch (against the mainline)?
> > 
> > Not concpetually, but create_kthread() uses CLONE_FS, and I don't
> > think it's just umask that things like nfsd want to avoid sharing.
> > What about all the *other* fields?
> > 
> > Just as an example: even if all the threads actually end up all having
> > the same global root, what about contention on 'fs->lock'?
> > 
> > I have *not* looked at the details, and maybe there's some reason I'm
> > completely off, but it worries me.
> 
> Umm...  I would be very surprised if it turned out to be a problem.
> nfsd really doesn't give a fuck about its cwd and root - not in the
> thread side of things.  And (un)exporting is (a) not on a hot path
> and (b) not done from a kernel thread anyway.  fh_to_dentry and friends
> doesn't care about root/cwd, etc.
> 
> I don't see anything that could cause that kind of issues.

PS: I haven't checked if lustre is trying to avoid that kind of contention;
it might be, but I would consider having those threads resolve _any_ kind
of pathnames from root or cwd as serious bug - after all, we are not guaranteed
that filesystem in question is reachable from the namespace PID 1 is running
it.