MFC r264145:

Add property and sysctl to control how ZVOLs are exposed to OS.

New ZFS property volmode and sysctl vfs.zfs.vol.mode allow switching ZVOL
between three modes:
 geom -- existing fully functional behavior (default);
 dev -- exposing volumes only as raw disk device file in devfs;
 none -- not exposing volumes outside ZFS.

The "dev" mode is less functional (can't be partitioned, mounted, etc),
but it is faster, and in some scenarios with untrusted consumers safer.
It can be useful for NAS, VM block storages, etc.
The "none" mode may be convenient for backup servers, etc. that don't
need direct data access.

Due to the way ZVOL is integrated with main ZFS code, those property
and sysctl are checked only during pool import and volume creation.

4 files changed, 403 insertions, 118 deletions

diff --git a/cddl/contrib/opensolaris/cmd/zfs/zfs.8 b/cddl/contrib/opensolaris/cmd/zfs/zfs.8
index 8fac48a..d14ce13 100644
--- a/cddl/contrib/opensolaris/cmd/zfs/zfs.8
+++ b/cddl/contrib/opensolaris/cmd/zfs/zfs.8
@@ -30,7 +30,7 @@
 .\"
 .\" $FreeBSD$
 .\"
-.Dd March 20, 2014
+.Dd April 5, 2014
 .Dt ZFS 8
 .Os
 .Sh NAME
@@ -1306,6 +1306,38 @@ Consequently, writes to a sparse volume can fail with
 when the pool is low on space. For a sparse volume, changes to
 .Sy volsize
 are not reflected in the reservation.
+.It Sy volmode Ns = Ns Cm default | geom | dev | none
+This property specifies how volumes should be exposed to the OS.
+Setting it to
+.Sy geom
+exposes volumes as
+.Xr geom 4
+providers, providing maximal functionality.
+Setting it to
+.Sy dev
+exposes volumes only as cdev device in devfs.
+Such volumes can be accessed only as raw disk device files, i.e. they
+can not be partitioned, mounted, participate in RAIDs, etc, but they
+are faster, and in some use scenarios with untrusted consumer, such as
+NAS or VM storage, can be more safe.
+Volumes with property set to
+.Sy none
+are not exposed outside ZFS, but can be snapshoted, cloned, replicated, etc,
+that can be suitable for backup purposes.
+Value
+.Sy default
+means that volumes exposition is controlled by system-wide sysctl/tunable
+.Va vfs.zfs.vol.mode ,
+where
+.Sy geom ,
+.Sy dev
+and
+.Sy none
+are encoded as 1, 2 and 3 respectively.
+The default values is
+.Sy geom .
+This property can be changed any time, but so far it is processed only
+during volume creation and pool import.
 .It Sy vscan Ns = Ns Cm off | on
 The
 .Sy vscan
diff --git a/sys/cddl/contrib/opensolaris/common/zfs/zfs_prop.c b/sys/cddl/contrib/opensolaris/common/zfs/zfs_prop.c
index 8403732..fde51a7 100644
--- a/sys/cddl/contrib/opensolaris/common/zfs/zfs_prop.c
+++ b/sys/cddl/contrib/opensolaris/common/zfs/zfs_prop.c
@@ -198,6 +198,14 @@ zfs_prop_init(void)
 		{ NULL }
 	};
 
+	static zprop_index_t volmode_table[] = {
+		{ "default",	ZFS_VOLMODE_DEFAULT },
+		{ "geom",	ZFS_VOLMODE_GEOM },
+		{ "dev",	ZFS_VOLMODE_DEV },
+		{ "none",	ZFS_VOLMODE_NONE },
+		{ NULL }
+	};
+
 	/* inherit index properties */
 	zprop_register_index(ZFS_PROP_SYNC, "sync", ZFS_SYNC_STANDARD,
 	    PROP_INHERIT, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME,
@@ -242,6 +250,10 @@ zfs_prop_init(void)
 	zprop_register_index(ZFS_PROP_LOGBIAS, "logbias", ZFS_LOGBIAS_LATENCY,
 	    PROP_INHERIT, ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME,
 	    "latency | throughput", "LOGBIAS", logbias_table);
+	zprop_register_index(ZFS_PROP_VOLMODE, "volmode",
+	    ZFS_VOLMODE_DEFAULT, PROP_INHERIT,
+	    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_SNAPSHOT | ZFS_TYPE_VOLUME,
+	    "default | geom | dev | none", "VOLMODE", volmode_table);
 
 	/* inherit index (boolean) properties */
 	zprop_register_index(ZFS_PROP_ATIME, "atime", 1, PROP_INHERIT,
diff --git a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c
index 5d14597..fb3789f 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c
+++ b/sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zvol.c
@@ -66,6 +66,7 @@
 #include <sys/spa.h>
 #include <sys/spa_impl.h>
 #include <sys/zio.h>
+#include <sys/disk.h>
 #include <sys/dmu_traverse.h>
 #include <sys/dnode.h>
 #include <sys/dsl_dataset.h>
@@ -75,6 +76,7 @@
 #include <sys/sunddi.h>
 #include <sys/dirent.h>
 #include <sys/policy.h>
+#include <sys/queue.h>
 #include <sys/fs/zfs.h>
 #include <sys/zfs_ioctl.h>
 #include <sys/zil.h>
@@ -114,6 +116,13 @@ static char *zvol_tag = "zvol_tag";
  */
 static uint32_t zvol_minors;
 
+SYSCTL_DECL(_vfs_zfs);
+SYSCTL_NODE(_vfs_zfs, OID_AUTO, vol, CTLFLAG_RW, 0, "ZFS VOLUME");
+static int	volmode = ZFS_VOLMODE_GEOM;
+TUNABLE_INT("vfs.zfs.vol.mode", &volmode);
+SYSCTL_INT(_vfs_zfs_vol, OID_AUTO, mode, CTLFLAG_RWTUN, &volmode, 0,
+    "Expose as GEOM providers (1), device files (2) or neither");
+
 typedef struct zvol_extent {
 	list_node_t	ze_node;
 	dva_t		ze_dva;		/* dva associated with this extent */
@@ -124,9 +133,11 @@ typedef struct zvol_extent {
  * The in-core state of each volume.
  */
 typedef struct zvol_state {
+	LIST_ENTRY(zvol_state)	zv_links;
 	char		zv_name[MAXPATHLEN]; /* pool/dd name */
 	uint64_t	zv_volsize;	/* amount of space we advertise */
 	uint64_t	zv_volblocksize; /* volume block size */
+	struct cdev	*zv_dev;	/* non-GEOM device */
 	struct g_provider *zv_provider;	/* GEOM provider */
 	uint8_t		zv_min_bs;	/* minimum addressable block shift */
 	uint8_t		zv_flags;	/* readonly, dumpified, etc. */
@@ -137,10 +148,13 @@ typedef struct zvol_state {
 	znode_t		zv_znode;	/* for range locking */
 	dmu_buf_t	*zv_dbuf;	/* bonus handle */
 	int		zv_state;
+	int		zv_volmode;	/* Provide GEOM or cdev */
 	struct bio_queue_head zv_queue;
 	struct mtx	zv_queue_mtx;	/* zv_queue mutex */
 } zvol_state_t;
 
+static LIST_HEAD(, zvol_state) all_zvols;
+
 /*
  * zvol specific flags
  */
@@ -154,6 +168,25 @@ typedef struct zvol_state {
  */
 int zvol_maxphys = DMU_MAX_ACCESS/2;
 
+static d_open_t		zvol_d_open;
+static d_close_t	zvol_d_close;
+static d_read_t		zvol_read;
+static d_write_t	zvol_write;
+static d_ioctl_t	zvol_d_ioctl;
+static d_strategy_t	zvol_strategy;
+
+static struct cdevsw zvol_cdevsw = {
+	.d_version =	D_VERSION,
+	.d_open =	zvol_d_open,
+	.d_close =	zvol_d_close,
+	.d_read =	zvol_read,
+	.d_write =	zvol_write,
+	.d_ioctl =	zvol_d_ioctl,
+	.d_strategy =	zvol_strategy,
+	.d_name =	"zvol",
+	.d_flags =	D_DISK | D_TRACKCLOSE,
+};
+
 extern int zfs_set_prop_nvlist(const char *, zprop_source_t,
     nvlist_t *, nvlist_t *);
 static void zvol_log_truncate(zvol_state_t *zv, dmu_tx_t *tx, uint64_t off,
@@ -164,7 +197,6 @@ static int zvol_dumpify(zvol_state_t *zv);
 static int zvol_dump_fini(zvol_state_t *zv);
 static int zvol_dump_init(zvol_state_t *zv, boolean_t resize);
 
-static zvol_state_t *zvol_geom_create(const char *name);
 static void zvol_geom_run(zvol_state_t *zv);
 static void zvol_geom_destroy(zvol_state_t *zv);
 static int zvol_geom_access(struct g_provider *pp, int acr, int acw, int ace);
@@ -186,14 +218,16 @@ zvol_size_changed(zvol_state_t *zv)
 	spec_size_invalidate(dev, VBLK);
 	spec_size_invalidate(dev, VCHR);
 #else	/* !sun */
-	struct g_provider *pp;
+	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
+		struct g_provider *pp;
 
-	pp = zv->zv_provider;
-	if (pp == NULL)
-		return;
-	g_topology_lock();
-	g_resize_provider(pp, zv->zv_volsize);
-	g_topology_unlock();
+		pp = zv->zv_provider;
+		if (pp == NULL)
+			return;
+		g_topology_lock();
+		g_resize_provider(pp, zv->zv_volsize);
+		g_topology_unlock();
+	}
 #endif	/* !sun */
 }
 
@@ -250,26 +284,16 @@ zvol_get_stats(objset_t *os, nvlist_t *nv)
 static zvol_state_t *
 zvol_minor_lookup(const char *name)
 {
-	struct g_provider *pp;
-	struct g_geom *gp;
-	zvol_state_t *zv = NULL;
+	zvol_state_t *zv;
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
 
-	g_topology_lock();
-	LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) {
-		pp = LIST_FIRST(&gp->provider);
-		if (pp == NULL)
-			continue;
-		zv = pp->private;
-		if (zv == NULL)
-			continue;
+	LIST_FOREACH(zv, &all_zvols, zv_links) {
 		if (strcmp(zv->zv_name, name) == 0)
 			break;
 	}
-	g_topology_unlock();
 
-	return (gp != NULL ? zv : NULL);
+	return (zv);
 }
 
 /* extent mapping arg */
@@ -507,8 +531,11 @@ zvol_create_minor(const char *name)
 	zfs_soft_state_t *zs;
 	zvol_state_t *zv;
 	objset_t *os;
+	struct cdev *dev;
+	struct g_provider *pp;
+	struct g_geom *gp;
 	dmu_object_info_t doi;
-	uint64_t volsize;
+	uint64_t volsize, mode;
 	int error;
 
 	ZFS_LOG(1, "Creating ZVOL %s...", name);
@@ -569,20 +596,51 @@ zvol_create_minor(const char *name)
 	zv = zs->zss_data = kmem_zalloc(sizeof (zvol_state_t), KM_SLEEP);
 #else	/* !sun */
 
+	zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
+	zv->zv_state = 0;
 	error = zap_lookup(os, ZVOL_ZAP_OBJ, "size", 8, 1, &volsize);
 	if (error) {
-		ASSERT(error == 0);
+		kmem_free(zv, sizeof(*zv));
 		dmu_objset_disown(os, zvol_tag);
 		mutex_exit(&spa_namespace_lock);
 		return (error);
 	}
+	error = dsl_prop_get_integer(name,
+	    zfs_prop_to_name(ZFS_PROP_VOLMODE), &mode, NULL);
+	if (error != 0 || mode == ZFS_VOLMODE_DEFAULT)
+		mode = volmode;
 
 	DROP_GIANT();
-	g_topology_lock();
-	zv = zvol_geom_create(name);
 	zv->zv_volsize = volsize;
-	zv->zv_provider->mediasize = zv->zv_volsize;
-
+	zv->zv_volmode = mode;
+	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
+		g_topology_lock();
+		gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
+		gp->start = zvol_geom_start;
+		gp->access = zvol_geom_access;
+		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
+		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
+		pp->sectorsize = DEV_BSIZE;
+		pp->mediasize = zv->zv_volsize;
+		pp->private = zv;
+
+		zv->zv_provider = pp;
+		bioq_init(&zv->zv_queue);
+		mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
+	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
+		if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
+		    &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
+		    0640, "%s/%s", ZVOL_DRIVER, name) != 0) {
+			kmem_free(zv, sizeof(*zv));
+			dmu_objset_disown(os, FTAG);
+			mutex_exit(&spa_namespace_lock);
+			return (SET_ERROR(ENXIO));
+		}
+		zv->zv_dev = dev;
+		dev->si_iosize_max = MAXPHYS;
+		dev->si_drv2 = zv;
+	}
+	LIST_INSERT_HEAD(&all_zvols, zv, zv_links);
 #endif	/* !sun */
 
 	(void) strlcpy(zv->zv_name, name, MAXPATHLEN);
@@ -613,10 +671,13 @@ zvol_create_minor(const char *name)
 
 	mutex_exit(&spa_namespace_lock);
 
-	zvol_geom_run(zv);
-
-	g_topology_unlock();
+#ifndef sun
+	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
+		zvol_geom_run(zv);
+		g_topology_unlock();
+	}
 	PICKUP_GIANT();
+#endif
 
 	ZFS_LOG(1, "ZVOL %s created.", name);
 
@@ -642,12 +703,20 @@ zvol_remove_zv(zvol_state_t *zv)
 #ifdef sun
 	(void) snprintf(nmbuf, sizeof (nmbuf), "%u,raw", minor);
 	ddi_remove_minor_node(zfs_dip, nmbuf);
+#else
+	LIST_REMOVE(zv, zv_links);
+	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
+		g_topology_lock();
+		zvol_geom_destroy(zv);
+		g_topology_unlock();
+	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV)
+		destroy_dev(zv->zv_dev);
 #endif	/* sun */
 
 	avl_destroy(&zv->zv_znode.z_range_avl);
 	mutex_destroy(&zv->zv_znode.z_range_lock);
 
-	zvol_geom_destroy(zv);
+	kmem_free(zv, sizeof(*zv));
 
 	zvol_minors--;
 	return (0);
@@ -664,9 +733,7 @@ zvol_remove_minor(const char *name)
 		mutex_exit(&spa_namespace_lock);
 		return (SET_ERROR(ENXIO));
 	}
-	g_topology_lock();
 	rc = zvol_remove_zv(zv);
-	g_topology_unlock();
 	mutex_exit(&spa_namespace_lock);
 	return (rc);
 }
@@ -802,24 +869,15 @@ zvol_update_volsize(objset_t *os, uint64_t volsize)
 void
 zvol_remove_minors(const char *name)
 {
-	struct g_geom *gp, *gptmp;
-	struct g_provider *pp;
-	zvol_state_t *zv;
+	zvol_state_t *zv, *tzv;
 	size_t namelen;
 
 	namelen = strlen(name);
 
 	DROP_GIANT();
 	mutex_enter(&spa_namespace_lock);
-	g_topology_lock();
 
-	LIST_FOREACH_SAFE(gp, &zfs_zvol_class.geom, geom, gptmp) {
-		pp = LIST_FIRST(&gp->provider);
-		if (pp == NULL)
-			continue;
-		zv = pp->private;
-		if (zv == NULL)
-			continue;
+	LIST_FOREACH_SAFE(zv, &all_zvols, zv_links, tzv) {
 		if (strcmp(zv->zv_name, name) == 0 ||
 		    (strncmp(zv->zv_name, name, namelen) == 0 &&
 		     zv->zv_name[namelen] == '/')) {
@@ -827,7 +885,6 @@ zvol_remove_minors(const char *name)
 		}
 	}
 
-	g_topology_unlock();
 	mutex_exit(&spa_namespace_lock);
 	PICKUP_GIANT();
 }
@@ -1286,28 +1343,46 @@ zvol_dumpio(zvol_state_t *zv, void *addr, uint64_t offset, uint64_t size,
 }
 #endif	/* sun */
 
-int
+void
 zvol_strategy(struct bio *bp)
 {
-	zvol_state_t *zv = bp->bio_to->private;
+	zvol_state_t *zv;
 	uint64_t off, volsize;
 	size_t resid;
 	char *addr;
 	objset_t *os;
 	rl_t *rl;
 	int error = 0;
-	boolean_t doread = (bp->bio_cmd == BIO_READ);
+	boolean_t doread = 0;
 	boolean_t is_dumpified;
 	boolean_t sync;
 
+	if (bp->bio_to)
+		zv = bp->bio_to->private;
+	else
+		zv = bp->bio_dev->si_drv2;
+
 	if (zv == NULL) {
-		g_io_deliver(bp, ENXIO);
-		return (0);
+		error = ENXIO;
+		goto out;
 	}
 
 	if (bp->bio_cmd != BIO_READ && (zv->zv_flags & ZVOL_RDONLY)) {
-		g_io_deliver(bp, EROFS);
-		return (0);
+		error = EROFS;
+		goto out;
+	}
+
+	switch (bp->bio_cmd) {
+	case BIO_FLUSH:
+		goto sync;
+	case BIO_READ:
+		doread = 1;
+	case BIO_WRITE:
+	case BIO_DELETE:
+		break;
+	default:
+		error = EOPNOTSUPP;
+		goto out;
 	}
 
 	off = bp->bio_offset;
@@ -1320,8 +1395,8 @@ zvol_strategy(struct bio *bp)
 	resid = bp->bio_length;
 
 	if (resid > 0 && (off < 0 || off >= volsize)) {
-		g_io_deliver(bp, EIO);
-		return (0);
+		error = EIO;
+		goto out;
 	}
 
 #ifdef illumos
@@ -1393,14 +1468,18 @@ unlock:
 	zfs_range_unlock(rl);
 
 	bp->bio_completed = bp->bio_length - resid;
-	if (bp->bio_completed < bp->bio_length)
-		bp->bio_error = (off > volsize ? EINVAL : error);
+	if (bp->bio_completed < bp->bio_length && off > volsize)
+		error = EINVAL;
 
-	if (sync)
+	if (sync) {
+sync:
 		zil_commit(zv->zv_zilog, ZVOL_OBJ);
-	g_io_deliver(bp, 0);
-
-	return (0);
+	}
+out:
+	if (bp->bio_to)
+		g_io_deliver(bp, error);
+	else
+		biofinish(bp, NULL, error);
 }
 
 #ifdef sun
@@ -1459,25 +1538,36 @@ int
 zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 {
 	minor_t minor = getminor(dev);
+#else
+int
+zvol_read(struct cdev *dev, struct uio *uio, int ioflag)
+{
+#endif
 	zvol_state_t *zv;
 	uint64_t volsize;
 	rl_t *rl;
 	int error = 0;
 
+#ifdef sun
 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
 	if (zv == NULL)
 		return (SET_ERROR(ENXIO));
+#else
+	zv = dev->si_drv2;
+#endif
 
 	volsize = zv->zv_volsize;
 	if (uio->uio_resid > 0 &&
-	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
+	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
 		return (SET_ERROR(EIO));
 
+#ifdef illumos
 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
 		error = physio(zvol_strategy, NULL, dev, B_READ,
 		    zvol_minphys, uio);
 		return (error);
 	}
+#endif
 
 	rl = zfs_range_lock(&zv->zv_znode, uio->uio_loffset, uio->uio_resid,
 	    RL_READER);
@@ -1500,31 +1590,43 @@ zvol_read(dev_t dev, uio_t *uio, cred_t *cr)
 	return (error);
 }
 
+#ifdef sun
 /*ARGSUSED*/
 int
 zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 {
 	minor_t minor = getminor(dev);
+#else
+int
+zvol_write(struct cdev *dev, struct uio *uio, int ioflag)
+{
+#endif
 	zvol_state_t *zv;
 	uint64_t volsize;
 	rl_t *rl;
 	int error = 0;
 	boolean_t sync;
 
+#ifdef sun
 	zv = zfsdev_get_soft_state(minor, ZSST_ZVOL);
 	if (zv == NULL)
 		return (SET_ERROR(ENXIO));
+#else
+	zv = dev->si_drv2;
+#endif
 
 	volsize = zv->zv_volsize;
 	if (uio->uio_resid > 0 &&
-	    (uio->uio_loffset < 0 || uio->uio_loffset >= volsize))
+	    (uio->uio_loffset < 0 || uio->uio_loffset > volsize))
 		return (SET_ERROR(EIO));
 
+#ifdef illumos
 	if (zv->zv_flags & ZVOL_DUMPIFIED) {
 		error = physio(zvol_strategy, NULL, dev, B_WRITE,
 		    zvol_minphys, uio);
 		return (error);
 	}
+#endif
 
 	sync = !(zv->zv_flags & ZVOL_WCE) ||
 	    (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS);
@@ -1559,6 +1661,7 @@ zvol_write(dev_t dev, uio_t *uio, cred_t *cr)
 	return (error);
 }
 
+#ifdef sun
 int
 zvol_getefi(void *arg, int flag, uint64_t vs, uint8_t bs)
 {
@@ -2232,31 +2335,6 @@ zvol_dump_fini(zvol_state_t *zv)
 }
 #endif	/* sun */
 
-static zvol_state_t *
-zvol_geom_create(const char *name)
-{
-	struct g_provider *pp;
-	struct g_geom *gp;
-	zvol_state_t *zv;
-
-	gp = g_new_geomf(&zfs_zvol_class, "zfs::zvol::%s", name);
-	gp->start = zvol_geom_start;
-	gp->access = zvol_geom_access;
-	pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, name);
-	pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
-	pp->sectorsize = DEV_BSIZE;
-
-	zv = kmem_zalloc(sizeof(*zv), KM_SLEEP);
-	zv->zv_provider = pp;
-	zv->zv_state = 0;
-	bioq_init(&zv->zv_queue);
-	mtx_init(&zv->zv_queue_mtx, "zvol", NULL, MTX_DEF);
-
-	pp->private = zv;
-
-	return (zv);
-}
-
 static void
 zvol_geom_run(zvol_state_t *zv)
 {
@@ -2287,8 +2365,6 @@ zvol_geom_destroy(zvol_state_t *zv)
 	zv->zv_provider = NULL;
 	pp->private = NULL;
 	g_wither_geom(pp->geom, ENXIO);
-
-	kmem_free(zv, sizeof(*zv));
 }
 
 static int
@@ -2537,30 +2613,47 @@ zvol_create_minors(const char *name)
 }
 
 static void
-zvol_rename_minor(struct g_geom *gp, const char *newname)
+zvol_rename_minor(zvol_state_t *zv, const char *newname)
 {
+	struct g_geom *gp;
 	struct g_provider *pp;
-	zvol_state_t *zv;
+	struct cdev *dev;
 
 	ASSERT(MUTEX_HELD(&spa_namespace_lock));
-	g_topology_assert();
 
-	pp = LIST_FIRST(&gp->provider);
-	ASSERT(pp != NULL);
-	zv = pp->private;
-	ASSERT(zv != NULL);
+	if (zv->zv_volmode == ZFS_VOLMODE_GEOM) {
+		g_topology_lock();
+		pp = zv->zv_provider;
+		ASSERT(pp != NULL);
+		gp = pp->geom;
+		ASSERT(gp != NULL);
 
-	zv->zv_provider = NULL;
-	g_wither_provider(pp, ENXIO);
-
-	pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
-	pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
-	pp->sectorsize = DEV_BSIZE;
-	pp->mediasize = zv->zv_volsize;
-	pp->private = zv;
-	zv->zv_provider = pp;
+		zv->zv_provider = NULL;
+		g_wither_provider(pp, ENXIO);
+
+		pp = g_new_providerf(gp, "%s/%s", ZVOL_DRIVER, newname);
+		pp->flags |= G_PF_DIRECT_RECEIVE | G_PF_DIRECT_SEND;
+		pp->sectorsize = DEV_BSIZE;
+		pp->mediasize = zv->zv_volsize;
+		pp->private = zv;
+		zv->zv_provider = pp;
+		g_error_provider(pp, 0);
+		g_topology_unlock();
+	} else if (zv->zv_volmode == ZFS_VOLMODE_DEV) {
+		dev = zv->zv_dev;
+		ASSERT(dev != NULL);
+		zv->zv_dev = NULL;
+		destroy_dev(dev);
+
+		if (make_dev_p(MAKEDEV_CHECKNAME | MAKEDEV_WAITOK,
+		    &dev, &zvol_cdevsw, NULL, UID_ROOT, GID_OPERATOR,
+		    0640, "%s/%s", ZVOL_DRIVER, newname) == 0) {
+			zv->zv_dev = dev;
+			dev->si_iosize_max = MAXPHYS;
+			dev->si_drv2 = zv;
+		}
+	}
 	strlcpy(zv->zv_name, newname, sizeof(zv->zv_name));
-	g_error_provider(pp, 0);
 }
 
 void
@@ -2578,28 +2671,169 @@ zvol_rename_minors(const char *oldname, const char *newname)
 
 	DROP_GIANT();
 	mutex_enter(&spa_namespace_lock);
-	g_topology_lock();
 
-	LIST_FOREACH(gp, &zfs_zvol_class.geom, geom) {
-		pp = LIST_FIRST(&gp->provider);
-		if (pp == NULL)
-			continue;
-		zv = pp->private;
-		if (zv == NULL)
-			continue;
+	LIST_FOREACH(zv, &all_zvols, zv_links) {
 		if (strcmp(zv->zv_name, oldname) == 0) {
-			zvol_rename_minor(gp, newname);
+			zvol_rename_minor(zv, newname);
 		} else if (strncmp(zv->zv_name, oldname, oldnamelen) == 0 &&
 		    (zv->zv_name[oldnamelen] == '/' ||
 		     zv->zv_name[oldnamelen] == '@')) {
 			snprintf(name, sizeof(name), "%s%c%s", newname,
 			    zv->zv_name[oldnamelen],
 			    zv->zv_name + oldnamelen + 1);
-			zvol_rename_minor(gp, name);
+			zvol_rename_minor(zv, name);
 		}
 	}
 
-	g_topology_unlock();
 	mutex_exit(&spa_namespace_lock);
 	PICKUP_GIANT();
 }
+
+static int
+zvol_d_open(struct cdev *dev, int flags, int fmt, struct thread *td)
+{
+	zvol_state_t *zv;
+	int err = 0;
+
+	mutex_enter(&spa_namespace_lock);
+	zv = dev->si_drv2;
+	if (zv == NULL) {
+		mutex_exit(&spa_namespace_lock);
+		return(ENXIO);		/* zvol_create_minor() not done yet */
+	}
+
+	if (zv->zv_total_opens == 0)
+		err = zvol_first_open(zv);
+	if (err) {
+		mutex_exit(&spa_namespace_lock);
+		return (err);
+	}
+	if ((flags & FWRITE) && (zv->zv_flags & ZVOL_RDONLY)) {
+		err = SET_ERROR(EROFS);
+		goto out;
+	}
+	if (zv->zv_flags & ZVOL_EXCL) {
+		err = SET_ERROR(EBUSY);
+		goto out;
+	}
+#ifdef FEXCL
+	if (flags & FEXCL) {
+		if (zv->zv_total_opens != 0) {
+			err = SET_ERROR(EBUSY);
+			goto out;
+		}
+		zv->zv_flags |= ZVOL_EXCL;
+	}
+#endif
+
+	zv->zv_total_opens++;
+	mutex_exit(&spa_namespace_lock);
+	return (err);
+out:
+	if (zv->zv_total_opens == 0)
+		zvol_last_close(zv);
+	mutex_exit(&spa_namespace_lock);
+	return (err);
+}
+
+static int
+zvol_d_close(struct cdev *dev, int flags, int fmt, struct thread *td)
+{
+	zvol_state_t *zv;
+	int err = 0;
+
+	mutex_enter(&spa_namespace_lock);
+	zv = dev->si_drv2;
+	if (zv == NULL) {
+		mutex_exit(&spa_namespace_lock);
+		return(ENXIO);
+	}
+
+	if (zv->zv_flags & ZVOL_EXCL) {
+		ASSERT(zv->zv_total_opens == 1);
+		zv->zv_flags &= ~ZVOL_EXCL;
+	}
+
+	/*
+	 * If the open count is zero, this is a spurious close.
+	 * That indicates a bug in the kernel / DDI framework.
+	 */
+	ASSERT(zv->zv_total_opens != 0);
+
+	/*
+	 * You may get multiple opens, but only one close.
+	 */
+	zv->zv_total_opens--;
+
+	if (zv->zv_total_opens == 0)
+		zvol_last_close(zv);
+
+	mutex_exit(&spa_namespace_lock);
+	return (0);
+}
+
+static int
+zvol_d_ioctl(struct cdev *dev, u_long cmd, caddr_t data, int fflag, struct thread *td)
+{
+	zvol_state_t *zv;
+	rl_t *rl;
+	off_t offset, length, chunk;
+	int i, error;
+	u_int u;
+
+	zv = dev->si_drv2;
+
+	error = 0;
+	KASSERT(zv->zv_total_opens > 0,
+	    ("Device with zero access count in zvol_d_ioctl"));
+
+	i = IOCPARM_LEN(cmd);
+	switch (cmd) {
+	case DIOCGSECTORSIZE:
+		*(u_int *)data = DEV_BSIZE;
+		break;
+	case DIOCGMEDIASIZE:
+		*(off_t *)data = zv->zv_volsize;
+		break;
+	case DIOCGFLUSH:
+		zil_commit(zv->zv_zilog, ZVOL_OBJ);
+		break;
+	case DIOCGDELETE:
+		offset = ((off_t *)data)[0];
+		length = ((off_t *)data)[1];
+		if ((offset % DEV_BSIZE) != 0 || (length % DEV_BSIZE) != 0 ||
+		    offset < 0 || offset >= zv->zv_volsize ||
+		    length <= 0) {
+			printf("%s: offset=%jd length=%jd\n", __func__, offset,
+			    length);
+			error = EINVAL;
+			break;
+		}
+
+		rl = zfs_range_lock(&zv->zv_znode, offset, length, RL_WRITER);
+		dmu_tx_t *tx = dmu_tx_create(zv->zv_objset);
+		error = dmu_tx_assign(tx, TXG_WAIT);
+		if (error != 0) {
+			dmu_tx_abort(tx);
+		} else {
+			zvol_log_truncate(zv, tx, offset, length, B_TRUE);
+			dmu_tx_commit(tx);
+			error = dmu_free_long_range(zv->zv_objset, ZVOL_OBJ,
+			    offset, length);
+		}
+		zfs_range_unlock(rl);
+		if (zv->zv_objset->os_sync == ZFS_SYNC_ALWAYS)
+			zil_commit(zv->zv_zilog, ZVOL_OBJ);
+		break;
+	case DIOCGSTRIPESIZE:
+		*(off_t *)data = zv->zv_volblocksize;
+		break;
+	case DIOCGSTRIPEOFFSET:
+		*(off_t *)data = 0;
+		break;
+	default:
+		error = ENOIOCTL;
+	}
+
+	return (error);
+}
diff --git a/sys/cddl/contrib/opensolaris/uts/common/sys/fs/zfs.h b/sys/cddl/contrib/opensolaris/uts/common/sys/fs/zfs.h
index 4745066..e9d7420 100644
--- a/sys/cddl/contrib/opensolaris/uts/common/sys/fs/zfs.h
+++ b/sys/cddl/contrib/opensolaris/uts/common/sys/fs/zfs.h
@@ -145,6 +145,7 @@ typedef enum {
 	ZFS_PROP_LOGICALUSED,
 	ZFS_PROP_LOGICALREFERENCED,
 	ZFS_PROP_INCONSISTENT,		/* not exposed to the user */
+	ZFS_PROP_VOLMODE,
 	ZFS_NUM_PROPS
 } zfs_prop_t;
 
@@ -337,6 +338,12 @@ typedef enum {
 	ZFS_SYNC_DISABLED = 2
 } zfs_sync_type_t;
 
+typedef enum {
+	ZFS_VOLMODE_DEFAULT = 0,
+	ZFS_VOLMODE_GEOM = 1,
+	ZFS_VOLMODE_DEV = 2,
+	ZFS_VOLMODE_NONE = 3
+} zfs_volmode_t;
 
 /*
  * On-disk version number.