Re: FreeBSD nfsroot - will it work with NFSv4.2?

On Thu, Feb 19, 2026 at 3:04 PM Dan Shelton <[email protected]> wrote:
>
> On Fri, 20 Feb 2026 at 00:01, Rick Macklem <[email protected]> wrote:
> >
> > On Thu, Feb 19, 2026 at 2:40 PM Dan Shelton <[email protected]> wrote:
> > >
> > > On Thu, 19 Feb 2026 at 23:30, Rick Macklem <[email protected]> wrote:
> > > >
> > > > On Thu, Feb 19, 2026 at 2:23 PM Dan Shelton <[email protected]> wrote:
> > > > >
> > > > > Hello,
> > > > >
> > > > > Will a FreeBSD 15.0 diskless setup with nfsroot work with NFSv4.2?
> > > > Nope. On FreeBSD a diskless NFS root fs must be NFSv3.
> > >
> > > Which is a bit of a problem since our IT guidelines prohibit the use of NFSv3.
> > > Also, we assume that NFSv4.2 will perform much better, and we need ACL support.
> > >
> > > > (There are various bootstrap issues, such as a standalone NFS client
> > > > that does not do NFSv4.
> > >
> > > Do you have a full list?
> > Well, fixing the uid/gid mapping issue is the main one.
> >
> > To be honest (I've wondered why for many years) FreeBSD has two
> > different ways that an NFS root fs can be booted.
> > - One of them (which seems to be favored by the x86 crowd) uses
> >   the afor mentioned NFSv3 standalone client.
> > - The other doesn't use the NFSv3 standalone client and uses bootp.
> >   (It seems to be favored by the non-x86 folk, like arm.)
> >
> > If only using the second one would be acceptable for NFSv4 root
> > file systems, then it probably isn't that hard to do (hardwire a few
> > mappings, so it works until nfsuserd is running) and fix up any
> > bootstrap issues (FreeBSD servers generate the same file handles for NFSv3
> > and NFSv4, but the client would need to handle any NFSv4 file handle
> > of up to 128bytes.)
> >
> > I don't currently have a diskless setup to play with, but I will
> > put it onto my to-do list. (The list never gets empty and often
> > gets priority items, so I cannot say when I'll get around to it.)
> >
> > Of course if someone like yourself was to do it and come up
> > with a patch, that would speed up the process.
> >
> > >
> > > > There is also the handling of
> > > > uid/gid<-->user/group name
> > > > mappings by nfsuserd, which cannot be running until it is booted.
> > > > Yes, there are ways to handle the mappings:
> > > > - Do it like Linux and insist that it be uid/gid numbers in the strings.
> > > > or
> > > > - Hardwire enough mappings in the kernel to get the system going so
> > > >   that nfsuserd can run.
> > >
> > > Can such a list be built in, or passed as a boot argument string?
> > >
> > > >
> > > > Since there has not been any demand for NFSv4 diskless root file
> > > > systems, it has not been done.
> > >
> > > We have demand, if we get a contract for that cluster setup... which
> > > requires that this actually works in a demonstrator.
> > >
> > > I recall that there was this topic when Opensolaris.org was active,
> > > but I never followed the SUN bugster RFE to implement that feature.
> > > Everyone was just bloody crazy that ZFS will take over the world, so
> > > cluster setups with nfsroot were treated like the orphaned stepchild
> > > which also has rabies and scabies.
> > I do know some folk use diskless NFS root file systems, but I have no
> > idea how many there are? You are about the 2nd person to ask about
> > NFSv4 root file systems.
>
> NFSv4.2 root filesystem
The issues apply to all minor versions of NFSv4 (uid/gid mapping, root
file handle size and how to bootstrap it).

rick

>
> Dan
> --
> Dan Shelton - Cluster Specialist Win/Lin/Bsd
>