Fix BOOTP root FS mounts. Also cleanup vfs_getnewfsid() and collapse

    addaliasu() into addalias() (no operational change) and clarify comments
    relating to a trick that vclean() uses.

    The fix to BOOTP is yet another hack.  Actually, rootfsid handling
    is already a major hack.  The whole thing needs to be cleaned up.

Reviewed by:	David Greenman <dg@root.com>, Alan Cox <alc@cs.rice.edu>

5 files changed, 117 insertions, 44 deletions

diff --git a/sys/kern/vfs_conf.c b/sys/kern/vfs_conf.c
index 87f3e4a..7c11af7 100644
--- a/sys/kern/vfs_conf.c
+++ b/sys/kern/vfs_conf.c
@@ -136,7 +136,27 @@ vfs_mountrootfs(void *unused)
 	mp->mnt_flag		|= MNT_ROOTFS;
 
 	/*
-	 * Attempt the mount
+	 * If we have no idea what the device is because the VFS root mount
+	 * initialization code couldn't figure it out, take a guess by
+	 * assuming that vfs_getnewfsid() will be called when we try the
+	 * mount.  For the moment this is necessary for NFS-baesd BOOTP
+	 * boots.  Ultimately we would like to get rid of 'rootdev' entirely
+	 * and go with a linked list of possible roots and device-specific
+	 * auxillary data that we do not try to interpret ourselves.
+	 */
+	if (rootdev == NODEV && rootdevs[0] == NODEV)
+		rootdev = vfs_getrootfsid(mp);
+
+	/*
+	 * Attempt the mount.  This is rather messy due to many historical
+	 * layers.  Basically what it comes down to is that 'rootdev' is an
+	 * override to the rootdevs[] array.  The rootdevs[] array itself
+	 * cannot normally be accessed directly by other modules, but FFS
+	 * plays with it.  NFS, on the otherhand, has no clue what the 
+	 * device assignment for a mount will be until it actually does it.
+	 *
+	 * During the loop we set rootdev to rootdevs[i].  This is used
+	 * by FFS and a few other modules.  It is ignored by NFS.
 	 */
 	err = ENXIO;
 	orootdev = rootdev;
diff --git a/sys/kern/vfs_export.c b/sys/kern/vfs_export.c
index 757e4aa..6737312 100644
--- a/sys/kern/vfs_export.c
+++ b/sys/kern/vfs_export.c
@@ -333,36 +333,55 @@ vfs_getvfs(fsid)
 
 /*
  * Get a new unique fsid
+ *
+ * Keep in mind that several mounts may be running in parallel,
+ * so always increment mntid_base even if lower numbers are available.
  */
+
+static u_short mntid_base;
+
 void
 vfs_getnewfsid(mp)
 	struct mount *mp;
 {
-	static u_short xxxfs_mntid;
-
 	fsid_t tfsid;
 	int mtype;
 
 	simple_lock(&mntid_slock); 
+
 	mtype = mp->mnt_vfc->vfc_typenum;
-	mp->mnt_stat.f_fsid.val[0] = makeudev(255, mtype);
-	mp->mnt_stat.f_fsid.val[1] = mtype;
-	if (xxxfs_mntid == 0)
-		++xxxfs_mntid;
-	tfsid.val[0] = makeudev(255, mtype + (xxxfs_mntid << 16));
-	tfsid.val[1] = mtype;
-	if (mountlist.cqh_first != (void *)&mountlist) {
-		while (vfs_getvfs(&tfsid)) {
-			xxxfs_mntid++;
-			tfsid.val[0] = makeudev(255,
-			    mtype + (xxxfs_mntid << 16));
-		}
+	for (;;) {
+		tfsid.val[0] = makeudev(255, mtype + (mntid_base << 16));
+		tfsid.val[1] = mtype;
+		++mntid_base;
+		if (vfs_getvfs(&tfsid) == NULL)
+			break;
 	}
+
 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
+	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
+
 	simple_unlock(&mntid_slock);
 }
 
 /*
+ * Get what should become the root fsid.
+ *
+ * This is somewhat of a hack.  If the rootdev is not known we
+ * assume that vfs_getnewfsid() will be called momentarily to
+ * allocate it, and we return what vfs_getnewfsid() will return.
+ */
+
+dev_t
+vfs_getrootfsid(struct mount *mp)
+{
+	int mtype;
+
+	mtype = mp->mnt_vfc->vfc_typenum;
+	return(makedev(255, mtype + (mntid_base << 16)));
+}
+
+/*
  * Knob to control the precision of file timestamps:
  *
  *   0 = seconds only; nanoseconds zeroed.
@@ -1303,11 +1322,7 @@ addaliasu(nvp, nvp_rdev)
 
 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
 		panic("addaliasu on non-special vnode");
-
-	nvp->v_rdev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
-	simple_lock(&spechash_slock);
-	SLIST_INSERT_HEAD(&nvp->v_rdev->si_hlist, nvp, v_specnext);
-	simple_unlock(&spechash_slock);
+	addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
 }
 
 void
@@ -1650,8 +1665,9 @@ vclean(vp, flags, p)
 	if ((obj = vp->v_object) != NULL) {
 		if (obj->ref_count == 0) {
 			/*
-			 * This is a normal way of shutting down the object/vnode
-			 * association.
+			 * vclean() may be called twice.  The first time removes the
+			 * primary reference to the object, the second time goes
+			 * one further and is a special-case to terminate the object.
 			 */
 			vm_object_terminate(obj);
 		} else {
diff --git a/sys/kern/vfs_mount.c b/sys/kern/vfs_mount.c
index 87f3e4a..7c11af7 100644
--- a/sys/kern/vfs_mount.c
+++ b/sys/kern/vfs_mount.c
@@ -136,7 +136,27 @@ vfs_mountrootfs(void *unused)
 	mp->mnt_flag		|= MNT_ROOTFS;
 
 	/*
-	 * Attempt the mount
+	 * If we have no idea what the device is because the VFS root mount
+	 * initialization code couldn't figure it out, take a guess by
+	 * assuming that vfs_getnewfsid() will be called when we try the
+	 * mount.  For the moment this is necessary for NFS-baesd BOOTP
+	 * boots.  Ultimately we would like to get rid of 'rootdev' entirely
+	 * and go with a linked list of possible roots and device-specific
+	 * auxillary data that we do not try to interpret ourselves.
+	 */
+	if (rootdev == NODEV && rootdevs[0] == NODEV)
+		rootdev = vfs_getrootfsid(mp);
+
+	/*
+	 * Attempt the mount.  This is rather messy due to many historical
+	 * layers.  Basically what it comes down to is that 'rootdev' is an
+	 * override to the rootdevs[] array.  The rootdevs[] array itself
+	 * cannot normally be accessed directly by other modules, but FFS
+	 * plays with it.  NFS, on the otherhand, has no clue what the 
+	 * device assignment for a mount will be until it actually does it.
+	 *
+	 * During the loop we set rootdev to rootdevs[i].  This is used
+	 * by FFS and a few other modules.  It is ignored by NFS.
 	 */
 	err = ENXIO;
 	orootdev = rootdev;
diff --git a/sys/kern/vfs_subr.c b/sys/kern/vfs_subr.c
index 757e4aa..6737312 100644
--- a/sys/kern/vfs_subr.c
+++ b/sys/kern/vfs_subr.c
@@ -333,36 +333,55 @@ vfs_getvfs(fsid)
 
 /*
  * Get a new unique fsid
+ *
+ * Keep in mind that several mounts may be running in parallel,
+ * so always increment mntid_base even if lower numbers are available.
  */
+
+static u_short mntid_base;
+
 void
 vfs_getnewfsid(mp)
 	struct mount *mp;
 {
-	static u_short xxxfs_mntid;
-
 	fsid_t tfsid;
 	int mtype;
 
 	simple_lock(&mntid_slock); 
+
 	mtype = mp->mnt_vfc->vfc_typenum;
-	mp->mnt_stat.f_fsid.val[0] = makeudev(255, mtype);
-	mp->mnt_stat.f_fsid.val[1] = mtype;
-	if (xxxfs_mntid == 0)
-		++xxxfs_mntid;
-	tfsid.val[0] = makeudev(255, mtype + (xxxfs_mntid << 16));
-	tfsid.val[1] = mtype;
-	if (mountlist.cqh_first != (void *)&mountlist) {
-		while (vfs_getvfs(&tfsid)) {
-			xxxfs_mntid++;
-			tfsid.val[0] = makeudev(255,
-			    mtype + (xxxfs_mntid << 16));
-		}
+	for (;;) {
+		tfsid.val[0] = makeudev(255, mtype + (mntid_base << 16));
+		tfsid.val[1] = mtype;
+		++mntid_base;
+		if (vfs_getvfs(&tfsid) == NULL)
+			break;
 	}
+
 	mp->mnt_stat.f_fsid.val[0] = tfsid.val[0];
+	mp->mnt_stat.f_fsid.val[1] = tfsid.val[1];
+
 	simple_unlock(&mntid_slock);
 }
 
 /*
+ * Get what should become the root fsid.
+ *
+ * This is somewhat of a hack.  If the rootdev is not known we
+ * assume that vfs_getnewfsid() will be called momentarily to
+ * allocate it, and we return what vfs_getnewfsid() will return.
+ */
+
+dev_t
+vfs_getrootfsid(struct mount *mp)
+{
+	int mtype;
+
+	mtype = mp->mnt_vfc->vfc_typenum;
+	return(makedev(255, mtype + (mntid_base << 16)));
+}
+
+/*
  * Knob to control the precision of file timestamps:
  *
  *   0 = seconds only; nanoseconds zeroed.
@@ -1303,11 +1322,7 @@ addaliasu(nvp, nvp_rdev)
 
 	if (nvp->v_type != VBLK && nvp->v_type != VCHR)
 		panic("addaliasu on non-special vnode");
-
-	nvp->v_rdev = udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0);
-	simple_lock(&spechash_slock);
-	SLIST_INSERT_HEAD(&nvp->v_rdev->si_hlist, nvp, v_specnext);
-	simple_unlock(&spechash_slock);
+	addalias(nvp, udev2dev(nvp_rdev, nvp->v_type == VBLK ? 1 : 0));
 }
 
 void
@@ -1650,8 +1665,9 @@ vclean(vp, flags, p)
 	if ((obj = vp->v_object) != NULL) {
 		if (obj->ref_count == 0) {
 			/*
-			 * This is a normal way of shutting down the object/vnode
-			 * association.
+			 * vclean() may be called twice.  The first time removes the
+			 * primary reference to the object, the second time goes
+			 * one further and is a special-case to terminate the object.
 			 */
 			vm_object_terminate(obj);
 		} else {
diff --git a/sys/sys/mount.h b/sys/sys/mount.h
index bb4b490..2da7a1e 100644
--- a/sys/sys/mount.h
+++ b/sys/sys/mount.h
@@ -391,6 +391,7 @@ struct	netcred *vfs_export_lookup	    /* lookup host in fs export list */
 	  __P((struct mount *, struct netexport *, struct sockaddr *));
 int	vfs_allocate_syncvnode __P((struct mount *));
 void	vfs_getnewfsid __P((struct mount *));
+dev_t	vfs_getrootfsid __P((struct mount *));
 struct	mount *vfs_getvfs __P((fsid_t *));      /* return vfs given fsid */
 int	vfs_modevent __P((module_t, int, void *));
 int	vfs_mountedon __P((struct vnode *));    /* is a vfs mounted on vp */