Is this line' indentation too large?

I am unsure about 32bit counter there, but do not have sufficient arguments right now. 4G vnodes should be enough.

Why do you think that this number is interesting to user?

I'm not sure how useful it is. I figured we could fit them both in the 64 bit reserved slot.

May be, export only vnodelistsize, in 64bit counter?

I do not object against lazy size, but it is some minor implementation detail. Basically it is list of vnodes where e.g. inactivation or atime processing was deferred due to not suitable context.

In fact, I think another counter would be (much) more useful, namely the active vnode count. It roughly is the count of vnodes which back open files, and also it is the counter which prevents non-forced unmounts. Giving this info to user is IMO indeed valuable.

You would need to calculate it, code should be like this:

count = 0;
MNT_ILOCK(mp);
TAILQ_FOREACH(vp, &mp->mnt_vnodelist, v_nmntvnodes) {
   if (vrefcount(vp) > 0) /* racy but does not matter */
      count++;
}
MNT_IUNLOCK(mp);

I couldn't find vrefcount(), but it the ddb hook I saw:

TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
        if (vp->v_type != VMARKER && vp->v_holdcnt > 0) {
                vn_printf(vp, "vnode ");
                if (db_pager_quit)
                        break;
        }
}

so I tweaked your code snippet to:

count = 0;
MNT_ILOCK(mp);
TAILQ_FOREACH(vp, &mp->mnt_nvnodelist, v_nmntvnodes) {
        //if (vrefcount(vp) > 0) /* racy but does not matter */
        /* racy but does not matter */
        if (vp->v_type != VMARKER && vp->v_holdcnt > 0)
                count++;
}
MNT_IUNLOCK(mp);
sbp->f_avnodecount = count;

Then output like this (without the mount update from lazy to active:
192.168.35.1:/data/home/ambrisko/netboot on / (nfs, fsid 01ff003a3a000000, vnodes: count 547 lazy 531)
devfs on /dev (devfs, fsid 00ff007171000000, vnodes: count 27 lazy 16)
procfs on /proc (procfs, local, fsid 02ff000202000000, vnodes: count 2 lazy 1)
linprocfs on /compat/linux/proc (linprocfs, local, fsid 03ff00b5b5000000, vnodes: count 2 lazy 1)
linsysfs on /compat/linux/sys (linsysfs, local, fsid 04ff008a8a000000, vnodes: count 2 lazy 1)
fdescfs on /dev/fd (fdescfs, fsid 05ff005959000000, vnodes: count 3 lazy 2)
tmpfs on /tmp (tmpfs, local, fsid 06ff008787000000, vnodes: count 6 lazy 2)
tmpfs on /var/tmp (tmpfs, local, fsid 07ff008787000000, vnodes: count 2 lazy 1)
tmpfs on /var/run (tmpfs, local, fsid 08ff008787000000, vnodes: count 8 lazy 5)
devfs on /compat/linux/dev (devfs, fsid 09ff007171000000, vnodes: count 4 lazy 4)
fdescfs on /compat/linux/dev/fd (fdescfs, fsid 0aff005959000000, vnodes: count 2 lazy 2)
tmpfs on /compat/linux/dev/shm (tmpfs, local, fsid 0bff008787000000, vnodes: count 2 lazy 1)
192.168.31.1:/data on /data (nfs, fsid 0cff003a3a000000, vnodes: count 2 lazy 2)
192.168.35.1:/data/home/ambrisko/netboot/usr/local on /usr/local (nfs, fsid 0dff003a3a000000, vnodes: count 73 lazy 72)
192.168.35.1:/data/home/ambrisko/netboot/var on /var (nfs, fsid 0eff003a3a000000, vnodes: count 82 lazy 49)

Sorry, it is vrefcnt(), I wrote the snippet from memory. See sys/vnode.h, line 1052.
If you still want to use the vnode field directly, it is v_usecount > 0 that idenitifies active vnodes, not v_holdcnt. The check for VMARKER is not strongly necessary.

Also please replace 'lazy' name with 'active' in the userspace printf.

Updated the review with the new code. FYI, I was running the old mount binary with a POC of the kernel change. I have updated mount to reflect the change. Thanks.