diff --git a/cmd/zfs/zfs_main.c b/cmd/zfs/zfs_main.c
index 6e0a6d5bc216..2b7fe93038b3 100644
--- a/cmd/zfs/zfs_main.c
+++ b/cmd/zfs/zfs_main.c
@@ -6059,7 +6059,12 @@ zfs_do_holds(int argc, char **argv)
 
 #define	CHECK_SPINNER 30
 #define	SPINNER_TIME 3		/* seconds */
-#define	MOUNT_TIME 5		/* seconds */
+#define	MOUNT_TIME 1		/* seconds */
+
+typedef struct get_all_state {
+	boolean_t	ga_verbose;
+	get_all_cb_t	*ga_cbp;
+} get_all_state_t;
 
 static int
 get_one_dataset(zfs_handle_t *zhp, void *data)
@@ -6068,10 +6073,10 @@ get_one_dataset(zfs_handle_t *zhp, void *data)
 	static int spinval = 0;
 	static int spincheck = 0;
 	static time_t last_spin_time = (time_t)0;
-	get_all_cb_t *cbp = data;
+	get_all_state_t *state = data;
 	zfs_type_t type = zfs_get_type(zhp);
 
-	if (cbp->cb_verbose) {
+	if (state->ga_verbose) {
 		if (--spincheck < 0) {
 			time_t now = time(NULL);
 			if (last_spin_time + SPINNER_TIME < now) {
@@ -6097,25 +6102,23 @@ get_one_dataset(zfs_handle_t *zhp, void *data)
 		zfs_close(zhp);
 		return (0);
 	}
-	libzfs_add_handle(cbp, zhp);
-	assert(cbp->cb_used <= cbp->cb_alloc);
+	libzfs_add_handle(state->ga_cbp, zhp);
+	assert(state->ga_cbp->cb_used <= state->ga_cbp->cb_alloc);
 
 	return (0);
 }
 
 static void
-get_all_datasets(zfs_handle_t ***dslist, size_t *count, boolean_t verbose)
+get_all_datasets(get_all_cb_t *cbp, boolean_t verbose)
 {
-	get_all_cb_t cb = { 0 };
-	cb.cb_verbose = verbose;
-	cb.cb_getone = get_one_dataset;
+	get_all_state_t state = {
+	    .ga_verbose = verbose,
+	    .ga_cbp = cbp
+	};
 
 	if (verbose)
 		set_progress_header(gettext("Reading ZFS config"));
-	(void) zfs_iter_root(g_zfs, get_one_dataset, &cb);
-
-	*dslist = cb.cb_handles;
-	*count = cb.cb_used;
+	(void) zfs_iter_root(g_zfs, get_one_dataset, &state);
 
 	if (verbose)
 		finish_progress(gettext("done."));
@@ -6126,8 +6129,19 @@ get_all_datasets(zfs_handle_t ***dslist, size_t *count, boolean_t verbose)
  * similar, we have a common function with an extra parameter to determine which
  * mode we are using.
  */
-#define	OP_SHARE	0x1
-#define	OP_MOUNT	0x2
+typedef enum { OP_SHARE, OP_MOUNT } share_mount_op_t;
+
+typedef struct share_mount_state {
+	share_mount_op_t	sm_op;
+	boolean_t	sm_verbose;
+	int	sm_flags;
+	char	*sm_options;
+	char	*sm_proto; /* only valid for OP_SHARE */
+	pthread_mutex_t	sm_lock; /* protects the remaining fields */
+	uint_t	sm_total; /* number of filesystems to process */
+	uint_t	sm_done; /* number of filesystems processed */
+	int	sm_status; /* -1 if any of the share/mount operations failed */
+} share_mount_state_t;
 
 /*
  * Share or mount a dataset.
@@ -6385,6 +6399,29 @@ report_mount_progress(int current, int total)
 		update_progress(info);
 }
 
+/*
+ * zfs_foreach_mountpoint() callback that mounts or shares one filesystem and
+ * updates the progress meter.
+ */
+static int
+share_mount_one_cb(zfs_handle_t *zhp, void *arg)
+{
+	share_mount_state_t *sms = arg;
+	int ret;
+
+	ret = share_mount_one(zhp, sms->sm_op, sms->sm_flags, sms->sm_proto,
+	    B_FALSE, sms->sm_options);
+
+	pthread_mutex_lock(&sms->sm_lock);
+	if (ret != 0)
+		sms->sm_status = ret;
+	sms->sm_done++;
+	if (sms->sm_verbose)
+		report_mount_progress(sms->sm_done, sms->sm_total);
+	pthread_mutex_unlock(&sms->sm_lock);
+	return (ret);
+}
+
 static void
 append_options(char *mntopts, char *newopts)
 {
@@ -6459,8 +6496,6 @@ share_mount(int op, int argc, char **argv)
 
 	/* check number of arguments */
 	if (do_all) {
-		zfs_handle_t **dslist = NULL;
-		size_t i, count = 0;
 		char *protocol = NULL;
 
 		if (op == OP_SHARE && argc > 0) {
@@ -6481,27 +6516,35 @@ share_mount(int op, int argc, char **argv)
 		}
 
 		start_progress_timer();
-		get_all_datasets(&dslist, &count, verbose);
+		get_all_cb_t cb = { 0 };
+		get_all_datasets(&cb, verbose);
 
-		if (count == 0) {
+		if (cb.cb_used == 0) {
 			if (options != NULL)
 				free(options);
 			return (0);
 		}
 
-		qsort(dslist, count, sizeof (void *), libzfs_dataset_cmp);
-
-		for (i = 0; i < count; i++) {
-			if (verbose)
-				report_mount_progress(i, count);
+		share_mount_state_t share_mount_state = { 0 };
+		share_mount_state.sm_op = op;
+		share_mount_state.sm_verbose = verbose;
+		share_mount_state.sm_flags = flags;
+		share_mount_state.sm_options = options;
+		share_mount_state.sm_proto = protocol;
+		share_mount_state.sm_total = cb.cb_used;
+		pthread_mutex_init(&share_mount_state.sm_lock, NULL);
 
-			if (share_mount_one(dslist[i], op, flags, protocol,
-			    B_FALSE, options) != 0)
-				ret = 1;
-			zfs_close(dslist[i]);
-		}
+		/*
+		 * libshare isn't mt-safe, so only do the operation in parallel
+		 * if we're mounting.
+		 */
+		zfs_foreach_mountpoint(g_zfs, cb.cb_handles, cb.cb_used,
+		    share_mount_one_cb, &share_mount_state, op == OP_MOUNT);
+		ret = share_mount_state.sm_status;
 
-		free(dslist);
+		for (int i = 0; i < cb.cb_used; i++)
+			zfs_close(cb.cb_handles[i]);
+		free(cb.cb_handles);
 	} else if (argc == 0) {
 		struct mnttab entry;
 
diff --git a/include/libzfs.h b/include/libzfs.h
index d34658055da3..762d57bef4eb 100644
--- a/include/libzfs.h
+++ b/include/libzfs.h
@@ -573,12 +573,11 @@ typedef struct get_all_cb {
 	zfs_handle_t	**cb_handles;
 	size_t		cb_alloc;
 	size_t		cb_used;
-	boolean_t	cb_verbose;
-	int		(*cb_getone)(zfs_handle_t *, void *);
 } get_all_cb_t;
 
+void zfs_foreach_mountpoint(libzfs_handle_t *, zfs_handle_t **, size_t,
+    zfs_iter_f, void *, boolean_t);
 void libzfs_add_handle(get_all_cb_t *, zfs_handle_t *);
-int libzfs_dataset_cmp(const void *, const void *);
 
 /*
  * Functions to create and destroy datasets.
diff --git a/include/libzfs_impl.h b/include/libzfs_impl.h
index 568103f4be36..9a46b9f12960 100644
--- a/include/libzfs_impl.h
+++ b/include/libzfs_impl.h
@@ -21,7 +21,7 @@
 
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2011, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
  * Copyright (c) 2018 Datto Inc.
  */
 
@@ -60,6 +60,13 @@ struct libzfs_handle {
 	void *libzfs_sharehdl; /* libshare handle */
 	uint_t libzfs_shareflags;
 	boolean_t libzfs_mnttab_enable;
+	/*
+	 * We need a lock to handle the case where parallel mount
+	 * threads are populating the mnttab cache simultaneously. The
+	 * lock only protects the integrity of the avl tree, and does
+	 * not protect the contents of the mnttab entries themselves.
+	 */
+	pthread_mutex_t libzfs_mnttab_cache_lock;
 	avl_tree_t libzfs_mnttab_cache;
 	int libzfs_pool_iter;
 	char libzfs_chassis_id[256];
diff --git a/lib/libzfs/libzfs_dataset.c b/lib/libzfs/libzfs_dataset.c
index e79a936f94e7..237933c371f0 100644
--- a/lib/libzfs/libzfs_dataset.c
+++ b/lib/libzfs/libzfs_dataset.c
@@ -791,6 +791,7 @@ libzfs_mnttab_cache_compare(const void *arg1, const void *arg2)
 void
 libzfs_mnttab_init(libzfs_handle_t *hdl)
 {
+	pthread_mutex_init(&hdl->libzfs_mnttab_cache_lock, NULL);
 	assert(avl_numnodes(&hdl->libzfs_mnttab_cache) == 0);
 	avl_create(&hdl->libzfs_mnttab_cache, libzfs_mnttab_cache_compare,
 	    sizeof (mnttab_node_t), offsetof(mnttab_node_t, mtn_node));
@@ -849,6 +850,7 @@ libzfs_mnttab_fini(libzfs_handle_t *hdl)
 		free(mtn);
 	}
 	avl_destroy(&hdl->libzfs_mnttab_cache);
+	(void) pthread_mutex_destroy(&hdl->libzfs_mnttab_cache_lock);
 }
 
 void
@@ -863,7 +865,7 @@ libzfs_mnttab_find(libzfs_handle_t *hdl, const char *fsname,
 {
 	mnttab_node_t find;
 	mnttab_node_t *mtn;
-	int error;
+	int ret = ENOENT;
 
 	if (!hdl->libzfs_mnttab_enable) {
 		struct mnttab srch = { 0 };
@@ -883,17 +885,24 @@ libzfs_mnttab_find(libzfs_handle_t *hdl, const char *fsname,
 			return (ENOENT);
 	}
 
-	if (avl_numnodes(&hdl->libzfs_mnttab_cache) == 0)
-		if ((error = libzfs_mnttab_update(hdl)) != 0)
+	pthread_mutex_lock(&hdl->libzfs_mnttab_cache_lock);
+	if (avl_numnodes(&hdl->libzfs_mnttab_cache) == 0) {
+		int error;
+
+		if ((error = libzfs_mnttab_update(hdl)) != 0) {
+			pthread_mutex_unlock(&hdl->libzfs_mnttab_cache_lock);
 			return (error);
+		}
+	}
 
 	find.mtn_mt.mnt_special = (char *)fsname;
 	mtn = avl_find(&hdl->libzfs_mnttab_cache, &find, NULL);
 	if (mtn) {
 		*entry = mtn->mtn_mt;
-		return (0);
+		ret = 0;
 	}
-	return (ENOENT);
+	pthread_mutex_unlock(&hdl->libzfs_mnttab_cache_lock);
+	return (ret);
 }
 
 void
@@ -902,14 +911,23 @@ libzfs_mnttab_add(libzfs_handle_t *hdl, const char *special,
 {
 	mnttab_node_t *mtn;
 
-	if (avl_numnodes(&hdl->libzfs_mnttab_cache) == 0)
-		return;
-	mtn = zfs_alloc(hdl, sizeof (mnttab_node_t));
-	mtn->mtn_mt.mnt_special = zfs_strdup(hdl, special);
-	mtn->mtn_mt.mnt_mountp = zfs_strdup(hdl, mountp);
-	mtn->mtn_mt.mnt_fstype = zfs_strdup(hdl, MNTTYPE_ZFS);
-	mtn->mtn_mt.mnt_mntopts = zfs_strdup(hdl, mntopts);
-	avl_add(&hdl->libzfs_mnttab_cache, mtn);
+	pthread_mutex_lock(&hdl->libzfs_mnttab_cache_lock);
+	if (avl_numnodes(&hdl->libzfs_mnttab_cache) != 0) {
+		mtn = zfs_alloc(hdl, sizeof (mnttab_node_t));
+		mtn->mtn_mt.mnt_special = zfs_strdup(hdl, special);
+		mtn->mtn_mt.mnt_mountp = zfs_strdup(hdl, mountp);
+		mtn->mtn_mt.mnt_fstype = zfs_strdup(hdl, MNTTYPE_ZFS);
+		mtn->mtn_mt.mnt_mntopts = zfs_strdup(hdl, mntopts);
+		/*
+		 * Another thread may have already added this entry
+		 * via libzfs_mnttab_update. If so we should skip it.
+		 */
+		if (avl_find(&hdl->libzfs_mnttab_cache, mtn, NULL) != NULL)
+			free(mtn);
+		else
+			avl_add(&hdl->libzfs_mnttab_cache, mtn);
+	}
+	pthread_mutex_unlock(&hdl->libzfs_mnttab_cache_lock);
 }
 
 void
@@ -918,6 +936,7 @@ libzfs_mnttab_remove(libzfs_handle_t *hdl, const char *fsname)
 	mnttab_node_t find;
 	mnttab_node_t *ret;
 
+	pthread_mutex_lock(&hdl->libzfs_mnttab_cache_lock);
 	find.mtn_mt.mnt_special = (char *)fsname;
 	if ((ret = avl_find(&hdl->libzfs_mnttab_cache, (void *)&find, NULL))
 	    != NULL) {
@@ -928,6 +947,7 @@ libzfs_mnttab_remove(libzfs_handle_t *hdl, const char *fsname)
 		free(ret->mtn_mt.mnt_mntopts);
 		free(ret);
 	}
+	pthread_mutex_unlock(&hdl->libzfs_mnttab_cache_lock);
 }
 
 int
diff --git a/lib/libzfs/libzfs_mount.c b/lib/libzfs/libzfs_mount.c
index 23e45d0d3ec5..ef18bafab2ae 100644
--- a/lib/libzfs/libzfs_mount.c
+++ b/lib/libzfs/libzfs_mount.c
@@ -22,7 +22,7 @@
 /*
  * Copyright 2015 Nexenta Systems, Inc.  All rights reserved.
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
- * Copyright (c) 2014, 2015 by Delphix. All rights reserved.
+ * Copyright (c) 2014, 2017 by Delphix. All rights reserved.
  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
  * Copyright 2017 RackTop Systems.
  * Copyright (c) 2018 Datto Inc.
@@ -34,25 +34,25 @@
  * they are used by mount and unmount and when changing a filesystem's
  * mountpoint.
  *
- * 	zfs_is_mounted()
- * 	zfs_mount()
- * 	zfs_unmount()
- * 	zfs_unmountall()
+ *	zfs_is_mounted()
+ *	zfs_mount()
+ *	zfs_unmount()
+ *	zfs_unmountall()
  *
  * This file also contains the functions used to manage sharing filesystems via
  * NFS and iSCSI:
  *
- * 	zfs_is_shared()
- * 	zfs_share()
- * 	zfs_unshare()
+ *	zfs_is_shared()
+ *	zfs_share()
+ *	zfs_unshare()
  *
- * 	zfs_is_shared_nfs()
- * 	zfs_is_shared_smb()
- * 	zfs_share_proto()
- * 	zfs_shareall();
- * 	zfs_unshare_nfs()
- * 	zfs_unshare_smb()
- * 	zfs_unshareall_nfs()
+ *	zfs_is_shared_nfs()
+ *	zfs_is_shared_smb()
+ *	zfs_share_proto()
+ *	zfs_shareall();
+ *	zfs_unshare_nfs()
+ *	zfs_unshare_smb()
+ *	zfs_unshareall_nfs()
  *	zfs_unshareall_smb()
  *	zfs_unshareall()
  *	zfs_unshareall_bypath()
@@ -60,8 +60,8 @@
  * The following functions are available for pool consumers, and will
  * mount/unmount and share/unshare all datasets within pool:
  *
- * 	zpool_enable_datasets()
- * 	zpool_disable_datasets()
+ *	zpool_enable_datasets()
+ *	zpool_disable_datasets()
  */
 
 #include <dirent.h>
@@ -84,11 +84,15 @@
 #include <libzfs.h>
 
 #include "libzfs_impl.h"
+#include <thread_pool.h>
 
 #include <libshare.h>
 #include <sys/systeminfo.h>
 #define	MAXISALEN	257	/* based on sysinfo(2) man page */
 
+static int mount_tp_nthr = 512;	/* tpool threads for multi-threaded mounting */
+
+static void zfs_mount_task(void *);
 static int zfs_share_proto(zfs_handle_t *, zfs_share_proto_t *);
 zfs_share_type_t zfs_is_shared_proto(zfs_handle_t *, char **,
     zfs_share_proto_t);
@@ -1146,25 +1150,32 @@ remove_mountpoint(zfs_handle_t *zhp)
 	}
 }
 
+/*
+ * Add the given zfs handle to the cb_handles array, dynamically reallocating
+ * the array if it is out of space.
+ */
 void
 libzfs_add_handle(get_all_cb_t *cbp, zfs_handle_t *zhp)
 {
 	if (cbp->cb_alloc == cbp->cb_used) {
 		size_t newsz;
-		void *ptr;
+		zfs_handle_t **newhandles;
 
-		newsz = cbp->cb_alloc ? cbp->cb_alloc * 2 : 64;
-		ptr = zfs_realloc(zhp->zfs_hdl,
-		    cbp->cb_handles, cbp->cb_alloc * sizeof (void *),
-		    newsz * sizeof (void *));
-		cbp->cb_handles = ptr;
+		newsz = cbp->cb_alloc != 0 ? cbp->cb_alloc * 2 : 64;
+		newhandles = zfs_realloc(zhp->zfs_hdl,
+		    cbp->cb_handles, cbp->cb_alloc * sizeof (zfs_handle_t *),
+		    newsz * sizeof (zfs_handle_t *));
+		cbp->cb_handles = newhandles;
 		cbp->cb_alloc = newsz;
 	}
 	cbp->cb_handles[cbp->cb_used++] = zhp;
 }
 
+/*
+ * Recursive helper function used during file system enumeration
+ */
 static int
-mount_cb(zfs_handle_t *zhp, void *data)
+zfs_iter_cb(zfs_handle_t *zhp, void *data)
 {
 	get_all_cb_t *cbp = data;
 
@@ -1196,112 +1207,351 @@ mount_cb(zfs_handle_t *zhp, void *data)
 	}
 
 	libzfs_add_handle(cbp, zhp);
-	if (zfs_iter_filesystems(zhp, mount_cb, cbp) != 0) {
+	if (zfs_iter_filesystems(zhp, zfs_iter_cb, cbp) != 0) {
 		zfs_close(zhp);
 		return (-1);
 	}
 	return (0);
 }
 
+/*
+ * Sort comparator that compares two mountpoint paths. We sort these paths so
+ * that subdirectories immediately follow their parents. This means that we
+ * effectively treat the '/' character as the lowest value non-nul char. An
+ * example sorted list using this comparator would look like:
+ *
+ * /foo
+ * /foo/bar
+ * /foo/bar/baz
+ * /foo/baz
+ * /foo.bar
+ *
+ * The mounting code depends on this ordering to deterministically iterate
+ * over filesystems in order to spawn parallel mount tasks.
+ */
 int
-libzfs_dataset_cmp(const void *a, const void *b)
+mountpoint_cmp(const void *arga, const void *argb)
 {
-	zfs_handle_t **za = (zfs_handle_t **)a;
-	zfs_handle_t **zb = (zfs_handle_t **)b;
+	zfs_handle_t *const *zap = arga;
+	zfs_handle_t *za = *zap;
+	zfs_handle_t *const *zbp = argb;
+	zfs_handle_t *zb = *zbp;
 	char mounta[MAXPATHLEN];
 	char mountb[MAXPATHLEN];
+	const char *a = mounta;
+	const char *b = mountb;
 	boolean_t gota, gotb;
 
-	if ((gota = (zfs_get_type(*za) == ZFS_TYPE_FILESYSTEM)) != 0)
-		verify(zfs_prop_get(*za, ZFS_PROP_MOUNTPOINT, mounta,
+	gota = (zfs_get_type(za) == ZFS_TYPE_FILESYSTEM);
+	if (gota) {
+		verify(zfs_prop_get(za, ZFS_PROP_MOUNTPOINT, mounta,
 		    sizeof (mounta), NULL, NULL, 0, B_FALSE) == 0);
-	if ((gotb = (zfs_get_type(*zb) == ZFS_TYPE_FILESYSTEM)) != 0)
-		verify(zfs_prop_get(*zb, ZFS_PROP_MOUNTPOINT, mountb,
+	}
+	gotb = (zfs_get_type(zb) == ZFS_TYPE_FILESYSTEM);
+	if (gotb) {
+		verify(zfs_prop_get(zb, ZFS_PROP_MOUNTPOINT, mountb,
 		    sizeof (mountb), NULL, NULL, 0, B_FALSE) == 0);
+	}
 
-	if (gota && gotb)
-		return (strcmp(mounta, mountb));
+	if (gota && gotb) {
+		while (*a != '\0' && (*a == *b)) {
+			a++;
+			b++;
+		}
+		if (*a == *b)
+			return (0);
+		if (*a == '\0')
+			return (-1);
+		if (*b == '\0')
+			return (1);
+		if (*a == '/')
+			return (-1);
+		if (*b == '/')
+			return (1);
+		return (*a < *b ? -1 : *a > *b);
+	}
 
 	if (gota)
 		return (-1);
 	if (gotb)
 		return (1);
 
-	return (strcmp(zfs_get_name(*za), zfs_get_name(*zb)));
+	/*
+	 * If neither filesystem has a mountpoint, revert to sorting by
+	 * dataset name.
+	 */
+	return (strcmp(zfs_get_name(za), zfs_get_name(zb)));
 }
 
 /*
- * Mount and share all datasets within the given pool.  This assumes that no
- * datasets within the pool are currently mounted.  Because users can create
- * complicated nested hierarchies of mountpoints, we first gather all the
- * datasets and mountpoints within the pool, and sort them by mountpoint.  Once
- * we have the list of all filesystems, we iterate over them in order and mount
- * and/or share each one.
+ * Return true if path2 is a child of path1.
  */
-#pragma weak zpool_mount_datasets = zpool_enable_datasets
-int
-zpool_enable_datasets(zpool_handle_t *zhp, const char *mntopts, int flags)
+static boolean_t
+libzfs_path_contains(const char *path1, const char *path2)
 {
-	get_all_cb_t cb = { 0 };
-	libzfs_handle_t *hdl = zhp->zpool_hdl;
-	zfs_handle_t *zfsp;
-	int i, ret = -1;
-	int *good;
+	return (strstr(path2, path1) == path2 && path2[strlen(path1)] == '/');
+}
+
+/*
+ * Given a mountpoint specified by idx in the handles array, find the first
+ * non-descendent of that mountpoint and return its index. Descendant paths
+ * start with the parent's path. This function relies on the ordering
+ * enforced by mountpoint_cmp().
+ */
+static int
+non_descendant_idx(zfs_handle_t **handles, size_t num_handles, int idx)
+{
+	char parent[ZFS_MAXPROPLEN];
+	char child[ZFS_MAXPROPLEN];
+	int i;
+
+	verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, parent,
+	    sizeof (parent), NULL, NULL, 0, B_FALSE) == 0);
+
+	for (i = idx + 1; i < num_handles; i++) {
+		verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT, child,
+		    sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
+		if (!libzfs_path_contains(parent, child))
+			break;
+	}
+	return (i);
+}
+
+typedef struct mnt_param {
+	libzfs_handle_t	*mnt_hdl;
+	tpool_t		*mnt_tp;
+	zfs_handle_t	**mnt_zhps; /* filesystems to mount */
+	size_t		mnt_num_handles;
+	int		mnt_idx;	/* Index of selected entry to mount */
+	zfs_iter_f	mnt_func;
+	void		*mnt_data;
+} mnt_param_t;
+
+/*
+ * Allocate and populate the parameter struct for mount function, and
+ * schedule mounting of the entry selected by idx.
+ */
+static void
+zfs_dispatch_mount(libzfs_handle_t *hdl, zfs_handle_t **handles,
+    size_t num_handles, int idx, zfs_iter_f func, void *data, tpool_t *tp)
+{
+	mnt_param_t *mnt_param = zfs_alloc(hdl, sizeof (mnt_param_t));
+
+	mnt_param->mnt_hdl = hdl;
+	mnt_param->mnt_tp = tp;
+	mnt_param->mnt_zhps = handles;
+	mnt_param->mnt_num_handles = num_handles;
+	mnt_param->mnt_idx = idx;
+	mnt_param->mnt_func = func;
+	mnt_param->mnt_data = data;
+
+	(void) tpool_dispatch(tp, zfs_mount_task, (void*)mnt_param);
+}
+
+/*
+ * This is the structure used to keep state of mounting or sharing operations
+ * during a call to zpool_enable_datasets().
+ */
+typedef struct mount_state {
+	/*
+	 * ms_mntstatus is set to -1 if any mount fails. While multiple threads
+	 * could update this variable concurrently, no synchronization is
+	 * needed as it's only ever set to -1.
+	 */
+	int		ms_mntstatus;
+	int		ms_mntflags;
+	const char	*ms_mntopts;
+} mount_state_t;
+
+static int
+zfs_mount_one(zfs_handle_t *zhp, void *arg)
+{
+	mount_state_t *ms = arg;
+	int ret = 0;
 
 	/*
-	 * Gather all non-snap datasets within the pool.
+	 * don't attempt to mount encrypted datasets with
+	 * unloaded keys
 	 */
-	if ((zfsp = zfs_open(hdl, zhp->zpool_name, ZFS_TYPE_DATASET)) == NULL)
-		goto out;
+	if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
+	    ZFS_KEYSTATUS_UNAVAILABLE)
+		return (0);
+
+	if (zfs_mount(zhp, ms->ms_mntopts, ms->ms_mntflags) != 0)
+		ret = ms->ms_mntstatus = -1;
+	return (ret);
+}
+
+static int
+zfs_share_one(zfs_handle_t *zhp, void *arg)
+{
+	mount_state_t *ms = arg;
+	int ret = 0;
+
+	if (zfs_share(zhp) != 0)
+		ret = ms->ms_mntstatus = -1;
+	return (ret);
+}
+
+/*
+ * Thread pool function to mount one file system. On completion, it finds and
+ * schedules its children to be mounted. This depends on the sorting done in
+ * zfs_foreach_mountpoint(). Note that the degenerate case (chain of entries
+ * each descending from the previous) will have no parallelism since we always
+ * have to wait for the parent to finish mounting before we can schedule
+ * its children.
+ */
+static void
+zfs_mount_task(void *arg)
+{
+	mnt_param_t *mp = arg;
+	int idx = mp->mnt_idx;
+	zfs_handle_t **handles = mp->mnt_zhps;
+	size_t num_handles = mp->mnt_num_handles;
+	char mountpoint[ZFS_MAXPROPLEN];
+
+	verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, mountpoint,
+	    sizeof (mountpoint), NULL, NULL, 0, B_FALSE) == 0);
+
+	if (mp->mnt_func(handles[idx], mp->mnt_data) != 0)
+		return;
 
-	libzfs_add_handle(&cb, zfsp);
-	if (zfs_iter_filesystems(zfsp, mount_cb, &cb) != 0)
-		goto out;
 	/*
-	 * Sort the datasets by mountpoint.
+	 * We dispatch tasks to mount filesystems with mountpoints underneath
+	 * this one. We do this by dispatching the next filesystem with a
+	 * descendant mountpoint of the one we just mounted, then skip all of
+	 * its descendants, dispatch the next descendant mountpoint, and so on.
+	 * The non_descendant_idx() function skips over filesystems that are
+	 * descendants of the filesystem we just dispatched.
 	 */
-	qsort(cb.cb_handles, cb.cb_used, sizeof (void *),
-	    libzfs_dataset_cmp);
+	for (int i = idx + 1; i < num_handles;
+	    i = non_descendant_idx(handles, num_handles, i)) {
+		char child[ZFS_MAXPROPLEN];
+		verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT,
+		    child, sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
+
+		if (!libzfs_path_contains(mountpoint, child))
+			break; /* not a descendant, return */
+		zfs_dispatch_mount(mp->mnt_hdl, handles, num_handles, i,
+		    mp->mnt_func, mp->mnt_data, mp->mnt_tp);
+	}
+	free(mp);
+}
 
+/*
+ * Issue the func callback for each ZFS handle contained in the handles
+ * array. This function is used to mount all datasets, and so this function
+ * guarantees that filesystems for parent mountpoints are called before their
+ * children. As such, before issuing any callbacks, we first sort the array
+ * of handles by mountpoint.
+ *
+ * Callbacks are issued in one of two ways:
+ *
+ * 1. Sequentially: If the parallel argument is B_FALSE or the ZFS_SERIAL_MOUNT
+ *    environment variable is set, then we issue callbacks sequentially.
+ *
+ * 2. In parallel: If the parallel argument is B_TRUE and the ZFS_SERIAL_MOUNT
+ *    environment variable is not set, then we use a tpool to dispatch threads
+ *    to mount filesystems in parallel. This function dispatches tasks to mount
+ *    the filesystems at the top-level mountpoints, and these tasks in turn
+ *    are responsible for recursively mounting filesystems in their children
+ *    mountpoints.
+ */
+void
+zfs_foreach_mountpoint(libzfs_handle_t *hdl, zfs_handle_t **handles,
+    size_t num_handles, zfs_iter_f func, void *data, boolean_t parallel)
+{
 	/*
-	 * And mount all the datasets, keeping track of which ones
-	 * succeeded or failed.
+	 * The ZFS_SERIAL_MOUNT environment variable is an undocumented
+	 * variable that can be used as a convenience to do a/b comparison
+	 * of serial vs. parallel mounting.
 	 */
-	if ((good = zfs_alloc(zhp->zpool_hdl,
-	    cb.cb_used * sizeof (int))) == NULL)
-		goto out;
+	boolean_t serial_mount = !parallel ||
+	    (getenv("ZFS_SERIAL_MOUNT") != NULL);
 
-	ret = 0;
-	for (i = 0; i < cb.cb_used; i++) {
-		/*
-		 * don't attempt to mount encrypted datasets with
-		 * unloaded keys
-		 */
-		if (zfs_prop_get_int(cb.cb_handles[i], ZFS_PROP_KEYSTATUS) ==
-		    ZFS_KEYSTATUS_UNAVAILABLE)
-			continue;
+	/*
+	 * Sort the datasets by mountpoint. See mountpoint_cmp for details
+	 * of how these are sorted.
+	 */
+	qsort(handles, num_handles, sizeof (zfs_handle_t *), mountpoint_cmp);
 
-		if (zfs_mount(cb.cb_handles[i], mntopts, flags) != 0)
-			ret = -1;
-		else
-			good[i] = 1;
+	if (serial_mount) {
+		for (int i = 0; i < num_handles; i++) {
+			func(handles[i], data);
+		}
+		return;
 	}
 
 	/*
-	 * Then share all the ones that need to be shared. This needs
-	 * to be a separate pass in order to avoid excessive reloading
-	 * of the configuration. Good should never be NULL since
-	 * zfs_alloc is supposed to exit if memory isn't available.
+	 * Issue the callback function for each dataset using a parallel
+	 * algorithm that uses a thread pool to manage threads.
+	 */
+	tpool_t *tp = tpool_create(1, mount_tp_nthr, 0, NULL);
+
+	/*
+	 * There may be multiple "top level" mountpoints outside of the pool's
+	 * root mountpoint, e.g.: /foo /bar. Dispatch a mount task for each of
+	 * these.
 	 */
-	for (i = 0; i < cb.cb_used; i++) {
-		if (good[i] && zfs_share(cb.cb_handles[i]) != 0)
-			ret = -1;
+	for (int i = 0; i < num_handles;
+	    i = non_descendant_idx(handles, num_handles, i)) {
+		zfs_dispatch_mount(hdl, handles, num_handles, i, func, data,
+		    tp);
 	}
 
-	free(good);
+	tpool_wait(tp);	/* wait for all scheduled mounts to complete */
+	tpool_destroy(tp);
+}
+
+/*
+ * Mount and share all datasets within the given pool.  This assumes that no
+ * datasets within the pool are currently mounted.
+ */
+#pragma weak zpool_mount_datasets = zpool_enable_datasets
+int
+zpool_enable_datasets(zpool_handle_t *zhp, const char *mntopts, int flags)
+{
+	get_all_cb_t cb = { 0 };
+	mount_state_t ms = { 0 };
+	zfs_handle_t *zfsp;
+	int ret = 0;
+
+	if ((zfsp = zfs_open(zhp->zpool_hdl, zhp->zpool_name,
+	    ZFS_TYPE_DATASET)) == NULL)
+		goto out;
+
+	/*
+	 * Gather all non-snapshot datasets within the pool. Start by adding
+	 * the root filesystem for this pool to the list, and then iterate
+	 * over all child filesystems.
+	 */
+	libzfs_add_handle(&cb, zfsp);
+	if (zfs_iter_filesystems(zfsp, zfs_iter_cb, &cb) != 0)
+		goto out;
+
+	/*
+	 * Mount all filesystems
+	 */
+	ms.ms_mntopts = mntopts;
+	ms.ms_mntflags = flags;
+	zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
+	    zfs_mount_one, &ms, B_TRUE);
+	if (ms.ms_mntstatus != 0)
+		ret = ms.ms_mntstatus;
+
+	/*
+	 * Share all filesystems that need to be shared. This needs to be
+	 * a separate pass because libshare is not mt-safe, and so we need
+	 * to share serially.
+	 */
+	ms.ms_mntstatus = 0;
+	zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
+	    zfs_share_one, &ms, B_FALSE);
+	if (ms.ms_mntstatus != 0)
+		ret = ms.ms_mntstatus;
 
 out:
-	for (i = 0; i < cb.cb_used; i++)
+	for (int i = 0; i < cb.cb_used; i++)
 		zfs_close(cb.cb_handles[i]);
 	free(cb.cb_handles);
 
diff --git a/tests/runfiles/linux.run b/tests/runfiles/linux.run
index e5826dd7a2b1..225e9bc8b53d 100644
--- a/tests/runfiles/linux.run
+++ b/tests/runfiles/linux.run
@@ -181,7 +181,7 @@ tests = ['zfs_mount_001_pos', 'zfs_mount_002_pos', 'zfs_mount_003_pos',
     'zfs_mount_007_pos', 'zfs_mount_008_pos', 'zfs_mount_009_neg',
     'zfs_mount_010_neg', 'zfs_mount_011_neg', 'zfs_mount_012_neg',
     'zfs_mount_all_001_pos', 'zfs_mount_encrypted', 'zfs_mount_remount',
-    'zfs_multi_mount']
+    'zfs_multi_mount', 'zfs_mount_all_fail', 'zfs_mount_all_mountpoints']
 tags = ['functional', 'cli_root', 'zfs_mount']
 
 [tests/functional/cli_root/zfs_program]
diff --git a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/Makefile.am b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/Makefile.am
index 05cbd8a77baa..b2de98934b74 100644
--- a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/Makefile.am
+++ b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/Makefile.am
@@ -14,8 +14,10 @@ dist_pkgdata_SCRIPTS = \
 	zfs_mount_010_neg.ksh \
 	zfs_mount_011_neg.ksh \
 	zfs_mount_012_neg.ksh \
-	zfs_mount_encrypted.ksh \
 	zfs_mount_all_001_pos.ksh \
+	zfs_mount_all_fail.ksh \
+	zfs_mount_all_mountpoints.ksh \
+	zfs_mount_encrypted.ksh \
 	zfs_mount_remount.ksh \
 	zfs_multi_mount.ksh
 
diff --git a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib
index 2a7cf6ef735b..2afb9a547b50 100644
--- a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib
+++ b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib
@@ -25,7 +25,7 @@
 #
 
 #
-# Copyright (c) 2016 by Delphix. All rights reserved.
+# Copyright (c) 2017 by Delphix. All rights reserved.
 #
 
 . $STF_SUITE/include/libtest.shlib
@@ -84,14 +84,12 @@ function setup_filesystem #disklist #pool #fs #mntpoint #type #vdev
 	fi
 
 	case "$type" in
-		'ctr')	log_must zfs create $pool/$fs
-			log_must zfs set mountpoint=$mntpoint $pool/$fs
+		'ctr')	log_must zfs create -o mountpoint=$mntpoint $pool/$fs
 			;;
 		'vol')	log_must zfs create -V $VOLSIZE $pool/$fs
 			block_device_wait
 			;;
-		*)	log_must zfs create $pool/$fs
-			log_must zfs set mountpoint=$mntpoint $pool/$fs
+		*)	log_must zfs create -o mountpoint=$mntpoint $pool/$fs
 			;;
 	esac
 
diff --git a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_fail.ksh b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_fail.ksh
new file mode 100755
index 000000000000..d7fcd20afa74
--- /dev/null
+++ b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_fail.ksh
@@ -0,0 +1,96 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2017 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib
+
+# DESCRIPTION:
+#       Verify that if 'zfs mount -a' fails to mount one filesystem,
+#       the command fails with a non-zero error code, but all other
+#       filesystems are mounted.
+#
+# STRATEGY:
+#       1. Create zfs filesystems
+#       2. Unmount a leaf filesystem
+#       3. Create a file in the above filesystem's mountpoint
+#       4. Verify that 'zfs mount -a' fails to mount the above
+#       5. Verify that all other filesystems were mounted
+#
+
+verify_runnable "both"
+
+typeset -a filesystems
+typeset path=${TEST_BASE_DIR%%/}/testroot$$/$TESTPOOL
+typeset fscount=10
+
+function setup_all
+{
+	# Create $fscount filesystems at the top level of $path
+	for ((i=0; i<$fscount; i++)); do
+		setup_filesystem "$DISKS" "$TESTPOOL" $i "$path/$i" ctr
+	done
+
+	zfs list -r $TESTPOOL
+
+	return 0
+}
+
+function cleanup_all
+{
+	export __ZFS_POOL_RESTRICT="$TESTPOOL"
+	log_must zfs $unmountall
+	unset __ZFS_POOL_RESTRICT
+
+	[[ -d ${TEST_BASE_DIR%%/}/testroot$$ ]] && \
+		rm -rf ${TEST_BASE_DIR%%/}/testroot$$
+}
+
+log_onexit cleanup_all
+
+log_must setup_all
+
+#
+# Unmount all of the above so that we can create the stray file
+# in one of the mountpoint directories.
+#
+export __ZFS_POOL_RESTRICT="$TESTPOOL"
+log_must zfs $unmountall
+unset __ZFS_POOL_RESTRICT
+
+# All of our filesystems should be unmounted at this point
+for ((i=0; i<$fscount; i++)); do
+	log_mustnot mounted "$TESTPOOL/$i"
+done
+
+# Create a stray file in one filesystem's mountpoint
+touch $path/0/strayfile
+
+# Verify that zfs mount -a fails
+export __ZFS_POOL_RESTRICT="$TESTPOOL"
+log_mustnot zfs $mountall
+unset __ZFS_POOL_RESTRICT
+
+# All filesystems except for "0" should be mounted
+log_mustnot mounted "$TESTPOOL/0"
+for ((i=1; i<$fscount; i++)); do
+	log_must mounted "$TESTPOOL/$i"
+done
+
+log_pass "'zfs $mountall' failed as expected."
diff --git a/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_mountpoints.ksh b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_mountpoints.ksh
new file mode 100755
index 000000000000..3e6a24bbcda3
--- /dev/null
+++ b/tests/zfs-tests/tests/functional/cli_root/zfs_mount/zfs_mount_all_mountpoints.ksh
@@ -0,0 +1,162 @@
+#!/bin/ksh -p
+#
+# CDDL HEADER START
+#
+# This file and its contents are supplied under the terms of the
+# Common Development and Distribution License ("CDDL"), version 1.0.
+# You may only use this file in accordance with the terms of version
+# 1.0 of the CDDL.
+#
+# A full copy of the text of the CDDL should have accompanied this
+# source.  A copy of the CDDL is also available via the Internet at
+# http://www.illumos.org/license/CDDL.
+#
+# CDDL HEADER END
+#
+
+#
+# Copyright (c) 2017 by Delphix. All rights reserved.
+#
+
+. $STF_SUITE/include/libtest.shlib
+. $STF_SUITE/tests/functional/cli_root/zfs_mount/zfs_mount.kshlib
+
+# DESCRIPTION:
+#       Verify that 'zfs mount -a' succeeds given a set of filesystems
+#       whose mountpoints have a parent/child relationship which is
+#       counter to the filesystem parent/child relationship.
+#
+# STRATEGY:
+#       1. Create zfs filesystems within the given pool.
+#       2. Unmount all the filesystems.
+#       3. Verify that 'zfs mount -a' command succeed,
+#	   and all available ZFS filesystems are mounted.
+#	4. Verify that 'zfs mount' is identical with 'df -F zfs'
+#
+
+verify_runnable "both"
+
+typeset -a filesystems
+
+function setup_all
+{
+	typeset path=${TEST_BASE_DIR%%/}/testroot$$/$TESTPOOL
+	typeset fscount=10
+
+	#
+	# Generate an array of filesystem names that represent a deep
+	# hierarchy as such:
+	#
+	# 0
+	# 0/1
+	# 0/1/2
+	# 0/1/2/3
+	# 0/1/2/3/4
+	# ...
+	#
+	fs=0
+	for ((i=0; i<$fscount; i++)); do
+		if [[ $i -gt 0 ]]; then
+			fs=$fs/$i
+		fi
+		filesystems+=($fs)
+	done
+
+	# Create all of the above filesystems
+	for ((i=0; i<$fscount; i++)); do
+		fs=${filesystems[$i]}
+		setup_filesystem "$DISKS" "$TESTPOOL" "$fs" "$path/$i" ctr
+	done
+
+	zfs list -r $TESTPOOL
+
+	#
+	# Unmount all of the above so that we can setup our convoluted
+	# mount paths.
+	#
+	export __ZFS_POOL_RESTRICT="$TESTPOOL"
+	log_must zfs $unmountall
+	unset __ZFS_POOL_RESTRICT
+
+	#
+	# Configure the mount paths so that each mountpoint is contained
+	# in a child filesystem. We should end up with something like the
+	# following structure (modulo the number of filesystems):
+	#
+	# NAME                       MOUNTPOINT
+	# testpool                   /testpool
+	# testpool/0                 /testroot25416/testpool/0/1/2/3/4/5/6
+	# testpool/0/1               /testroot25416/testpool/0/1/2/3/4/5
+	# testpool/0/1/2             /testroot25416/testpool/0/1/2/3/4
+	# testpool/0/1/2/3           /testroot25416/testpool/0/1/2/3
+	# testpool/0/1/2/3/4         /testroot25416/testpool/0/1/2
+	# testpool/0/1/2/3/4/5       /testroot25416/testpool/0/1
+	# testpool/0/1/2/3/4/5/6     /testroot25416/testpool/0
+	#
+	for ((i=0; i<$fscount; i++)); do
+		fs=$TESTPOOL/${filesystems[$(($fscount - $i - 1))]}
+		mnt=$path/${filesystems[$i]}
+		zfs set mountpoint=$mnt $fs
+	done
+
+	zfs list -r $TESTPOOL
+
+	return 0
+}
+
+function cleanup_all
+{
+	export __ZFS_POOL_RESTRICT="$TESTPOOL"
+	log_must zfs $unmountall
+	unset __ZFS_POOL_RESTRICT
+
+	for fs in ${filesystems[@]}; do
+		cleanup_filesystem "$TESTPOOL" "$fs"
+	done
+	[[ -d ${TEST_BASE_DIR%%/}/testroot$$ ]] && \
+		rm -rf ${TEST_BASE_DIR%%/}/testroot$$
+}
+
+#
+# This function takes a single true/false argument. If true it will verify that
+# all file systems are mounted. If false it will verify that they are not
+# mounted.
+#
+function verify_all
+{
+	if $1; then
+		logfunc=log_must
+	else
+		logfunc=log_mustnot
+	fi
+
+	for fs in ${filesystems[@]}; do
+		$logfunc mounted "$TESTPOOL/$fs"
+	done
+
+	return 0
+}
+
+log_onexit cleanup_all
+
+log_must setup_all
+
+export __ZFS_POOL_RESTRICT="$TESTPOOL"
+log_must zfs $unmountall
+unset __ZFS_POOL_RESTRICT
+
+verify_all false
+
+export __ZFS_POOL_RESTRICT="$TESTPOOL"
+log_must zfs $mountall
+unset __ZFS_POOL_RESTRICT
+
+verify_all true
+
+log_note "Verify that 'zfs $mountcmd' will display " \
+	"all ZFS filesystems currently mounted."
+
+verify_mount_display
+
+log_pass "'zfs $mountall' succeeds as root, " \
+	"and all available ZFS filesystems are mounted."