Bug 31306

	After using the Disk Label Editor to specify the characteristics
	of the fileesystems to be used and the partitions where they
	reside, sysinstall creates the device entries in /dev for the
	root and swap partitions, but none of the others.

	In cases where significant parts of the system are intended to be
	installed on a non-root filesystem, this tends to cause the install
	to fail.

Fix: 

I haven't actually done this code (yet), but....  Looking at install.c,
	rev. 1.308, starting at line 953:


    /* Now buzz through the rest of the partitions and mount them too */
    devs = deviceFind(NULL, DEVICE_TYPE_DISK);
    for (i = 0; devs[i]; i++) {
	if (!devs[i]->enabled)
	    continue;

	disk = (Disk *)devs[i]->private;
	if (!disk->chunks) {
	    msgConfirm("No chunk list found for %s!", disk->name);
	    return DITEM_FAILURE | DITEM_RESTORE;
	}
	if (RunningAsInit && root && (root->newfs || upgrade)) {
	    Mkdir("/mnt/dev");
	    if (!Fake)
		MakeDevDisk(disk, "/mnt/dev");
	}
	else if (!RunningAsInit && !Fake)
	    MakeDevDisk(disk, "/dev");

	for (c1 = disk->chunks->part; c1; c1 = c1->next) {
	    if (c1->type == freebsd) {
		for (c2 = c1->part; c2; c2 = c2->next) {
		    if (c2->type == part && c2->subtype != FS_SWAP && c2->private_data) {
			PartInfo *tmp = (PartInfo *)c2->private_data;

			/* Already did root */
			if (c2 == rootdev)
			    continue;

			if (tmp->newfs && (!upgrade || !msgNoYes("You are upgrading - are you SURE you want to newfs /dev/%s?", c2->name)))
			    command_shell_add(tmp->mountpoint, "%s %s/dev/%s", tmp->newfs_cmd, RunningAsInit ? "/mnt" : "", c2->name);
			else
			    command_shell_add(tmp->mountpoint, "fsck -y %s/dev/%s", RunningAsInit ? "/mnt" : "", c2->name);
			if (tmp->soft)
			    command_shell_add(tmp->mountpoint, "tunefs -n enable %s/dev/%s", RunningAsInit ? "/mnt" : "", c2->name);
			command_func_add(tmp->mountpoint, Mount, c2->name);
		    }
		    else if (c2->type == part && c2->subtype == FS_SWAP) {
			char fname[80];
			int i;

			if (c2 == swapdev)
			    continue;
			sprintf(fname, "%s/dev/%s", RunningAsInit ? "/mnt" : "", c2->name);
			i = (Fake || swapon(fname));
			if (!i) {
			    dialog_clear_norefresh();
			    msgNotify("Added %s as an additional swap device", fname);
			}
			else {
			    msgConfirm("Unable to add %s as a swap device: %s", fname, strerror(errno));
			}
		    }
		}
	    }
	    else if (c1->type == fat && c1->private_data && (root->newfs || upgrade)) {
		char name[FILENAME_MAX];

		sprintf(name, "%s/%s", RunningAsInit ? "/mnt" : "", ((PartInfo *)c1->private_data)->mountpoint);
		Mkdir(name);
	    }
	}
    }


	We see that there's a call to MakeDevDisk() for the root case, but
	not for the non-root case.  (Indeed:  there are a couple of them,
	for different situations.  And those 2 MakeDevDisk() calls are the
	only ones in /usr/src/usr.sbin/sysinstall/*.)  It's been quite a
	while since I actually programmed for a living, but that appears
	a tad suspect to me.  And note that the first things we try to do
	(in the non-root case) is a newfs command, which corresponds to
	the empirical evidence.
How-To-Repeat: 	Grab the kern.flp & msroot.flp images from a -CURRENT snapshot; copy
	them to appropriate media.  When you boot, it may be instructive to
	interrupt the boot sequence and specify "boot -v" (in order to get
	sysinstall debug information on vty2).

	Go through the install process.  At the Disk Label Editor, specify a
	modestly-sized root filesystem -- say, 160 MB or less.  Specify a
	reasonable (separate) filesystem for /usr, such as 2 GB.  You might
	specify a separate filesystem for /var, but the /usr one is the
	critical part, here.

	Select a distribution; for simplicity, I suggest "Developer" (not
	"X-Developer", as that requires a bit more interaction).  I declined
	the ports, as I was going to do that via CVSup anyhow.

	Affirm that you do, indeed, wish to Commit.  :-)

	A window is displayed, reading something on the order of

	Error mounting /mnt/dev/ad0s2f on /mnt/usr : No such file or directory

	Use Alt+F2 to see the debug messages.  Note that the first mention
	of ad0s2f we see is an attempt to do a "newfs".  See below for evidence
	that we should have seen that a mknod() call was issued; available
	evidence indicates that this was, in fact, not done.