zfsvfs_create(): Refuse to mount datasets whose names are too long.

This is checked for in the zfs_snapshot_004_neg STF/ATF test (currently
still in projects/zfsd rather than head).

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c:
- zfsvfs_create(): Check whether the objset name fits into
  statfs.f_mntfromname, and return ENAMETOOLONG if not.  Although
  the filesystem can be unmounted via the umount(8) command, any
  interface that relies on iterating on statfs (e.g. libzfs) will
  fail to find the filesystem by its objset name, and thus assume
  it's not mounted.  This causes "zfs unmount", "zfs destroy",
  etc. to fail on these filesystems, whether or not -f is passed.

MFC after:	1 month
Sponsored by:	Spectra Logic
MFSpectraBSD:	974872 on 2013/08/09

sys/cddl/contrib/opensolaris/uts/common/fs/zfs/zfs_vfsops.c

@@ -870,6 +870,17 @@ zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
 	int i, error;
 	uint64_t sa_obj;
 
+	/*
+	 * XXX: Fix struct statfs so this isn't necessary!
+	 *
+	 * The 'osname' is used as the filesystem's special node, which means
+	 * it must fit in statfs.f_mntfromname, or else it can't be
+	 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
+	 * 'zfs unmount' to think it's not mounted when it is.
+	 */
+	if (strlen(osname) >= MNAMELEN)
+		return (SET_ERROR(ENAMETOOLONG));
+
 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
 
 	/*