Add support for devid and phys_path keys in vdev disk labels

This is foundational work for ZED.

Updates a leaf vdev's persistent device strings on Linux platform

* only applies for a dedicated leaf vdev (aka whole disk)
* updated during pool create|add|attach|import
* used for matching device matching during auto-{online,expand,replace}
* stored in a leaf disk config label (i.e. alongside 'path' NVP)
* can opt-out using env var ZFS_VDEV_DEVID_OPT_OUT=YES

Some examples:

    path: '/dev/sdb1'
    devid: 'scsi-350000394a8ca4fbc-part1'
    phys_path: 'pci-0000:04:00.0-sas-0x50000394a8ca4fbf-lun-0'

    path: '/dev/mapper/mpatha'
    devid: 'dm-uuid-mpath-35000c5006304de3f'

Signed-off-by: Don Brady <don.brady@intel.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #2856
Closes #3978
Closes #4416

cmd/zpool/zpool_vdev.c

@@ -22,6 +22,7 @@
 /*
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2013 by Delphix. All rights reserved.
+ * Copyright (c) 2016 Intel Corporation.
  */
 
 /*
@@ -447,7 +448,7 @@ check_disk(const char *path, blkid_cache cache, int force,
 		(void) close(fd);
 
 		if (force) {
-			/* Partitions will no be created using the backup */
+			/* Partitions will now be created using the backup */
 			return (0);
 		} else {
 			vdev_error(gettext("%s contains a corrupt primary "
@@ -1178,6 +1179,12 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 		    &wholedisk));
 
 		if (!wholedisk) {
+			/*
+			 * Update device id string for mpath nodes (Linux only)
+			 */
+			if (is_mpath_whole_disk(path))
+				update_vdev_config_dev_strs(nv);
+
 			(void) zero_label(path);
 			return (0);
 		}
@@ -1249,6 +1256,12 @@ make_disks(zpool_handle_t *zhp, nvlist_t *nv)
 		 */
 		verify(nvlist_add_string(nv, ZPOOL_CONFIG_PATH, udevpath) == 0);
 
+		/*
+		 * Update device id strings for whole disks (Linux only)
+		 */
+		if (wholedisk)
+			update_vdev_config_dev_strs(nv);
+
 		return (0);
 	}
 

config/user-libudev.m4

@@ -0,0 +1,13 @@
+dnl #
+dnl # Check for libudev - needed for vdev auto-online and auto-replace
+dnl #
+AC_DEFUN([ZFS_AC_CONFIG_USER_LIBUDEV], [
+	LIBUDEV=
+
+	AC_CHECK_HEADER([libudev.h], [AC_SUBST([LIBUDEV], ["-ludev"])
+	    AC_DEFINE([HAVE_LIBUDEV], 1, [Define if you have libudev])], [])
+
+	AC_SEARCH_LIBS(i[udev_device_get_is_initialized], [udev], [
+	    AC_DEFINE([HAVE_LIBUDEV_UDEV_DEVICE_GET_IS_INITIALIZED], 1, [
+	    Define if udev_device_get_is_initialized is available])], [])
+])

config/user.m4

@@ -12,6 +12,7 @@ AC_DEFUN([ZFS_AC_CONFIG_USER], [
 	ZFS_AC_CONFIG_USER_LIBUUID
 	ZFS_AC_CONFIG_USER_LIBBLKID
 	ZFS_AC_CONFIG_USER_LIBATTR
+	ZFS_AC_CONFIG_USER_LIBUDEV
 	ZFS_AC_CONFIG_USER_FRAME_LARGER_THAN
 	ZFS_AC_CONFIG_USER_RUNSTATEDIR
 

include/libzfs.h

@@ -25,6 +25,7 @@
  * Copyright (c) 2012, Joyent, Inc. All rights reserved.
  * Copyright (c) 2013 Steven Hartland. All rights reserved.
  * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
+ * Copyright (c) 2016, Intel Corporation.
  */
 
 #ifndef	_LIBZFS_H
@@ -801,6 +802,12 @@ extern boolean_t libzfs_fru_compare(libzfs_handle_t *, const char *,
 extern boolean_t libzfs_fru_notself(libzfs_handle_t *, const char *);
 extern int zpool_fru_set(zpool_handle_t *, uint64_t, const char *);
 
+/*
+ * Support for Linux libudev derived persistent device strings
+ */
+extern boolean_t is_mpath_whole_disk(const char *);
+extern void update_vdev_config_dev_strs(nvlist_t *);
+
 #ifdef	__cplusplus
 }
 #endif

lib/libspl/include/devid.h

@@ -28,82 +28,5 @@
 #define	_LIBSPL_DEVID_H
 
 #include <sys/types.h>
-#include <stdlib.h>
-
-typedef int ddi_devid_t;
-
-typedef struct devid_nmlist {
-	char *devname;
-	dev_t dev;
-} devid_nmlist_t;
-
-static inline
-int
-devid_str_decode(
-    char *devidstr,
-    ddi_devid_t *retdevid,
-    char **retminor_name)
-{
-	abort();
-}
-
-static inline
-int
-devid_deviceid_to_nmlist(
-    char *search_path,
-    ddi_devid_t devid,
-    char *minor_name,
-    devid_nmlist_t **retlist)
-{
-	abort();
-}
-
-static inline
-void
-devid_str_free(char *str)
-{
-	abort();
-}
-
-static inline
-void
-devid_free(ddi_devid_t devid)
-{
-	abort();
-}
-
-static inline
-void
-devid_free_nmlist(devid_nmlist_t *list)
-{
-	abort();
-}
-
-static inline
-int
-devid_get(
-    int fd,
-    ddi_devid_t *retdevid)
-{
-	return (-1);
-}
-
-static inline
-int
-devid_get_minor_name(
-    int fd,
-    char **retminor_name)
-{
-	abort();
-}
-
-static inline
-char *
-devid_str_encode(
-    ddi_devid_t devid,
-    char *minor_name)
-{
-	abort();
-}
 
 #endif

lib/libzfs/Makefile.am

@@ -36,7 +36,7 @@ libzfs_la_LIBADD = \
 	$(top_builddir)/lib/libnvpair/libnvpair.la \
 	$(top_builddir)/lib/libzpool/libzpool.la
 
-libzfs_la_LIBADD += -lm $(LIBBLKID)
+libzfs_la_LIBADD += -lm $(LIBBLKID) $(LIBUDEV)
 libzfs_la_LDFLAGS = -version-info 2:0:0
 
 EXTRA_DIST = $(libzfs_pc_DATA) $(USER_C)

lib/libzfs/libzfs_import.c

@@ -23,6 +23,7 @@
  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  * Copyright (c) 2012 by Delphix. All rights reserved.
  * Copyright 2015 RackTop Systems.
+ * Copyright (c) 2016, Intel Corporation.
  */
 
 /*
@@ -46,6 +47,10 @@
 #include <dirent.h>
 #include <errno.h>
 #include <libintl.h>
+#ifdef HAVE_LIBUDEV
+#include <libudev.h>
+#include <sched.h>
+#endif
 #include <stddef.h>
 #include <stdlib.h>
 #include <string.h>
@@ -94,30 +99,328 @@ typedef struct pool_list {
 	name_entry_t		*names;
 } pool_list_t;
 
-static char *
+/*
-get_devid(const char *path)
+ * Linux persistent device strings for vdev labels
+ *
+ * based on libudev for consistency with libudev disk add/remove events
+ */
+#ifdef HAVE_LIBUDEV
+
+#define	DEV_BYID_PATH	"/dev/disk/by-id/"
+
+typedef struct vdev_dev_strs {
+	char	vds_devid[128];
+	char	vds_devphys[128];
+} vdev_dev_strs_t;
+
+/*
+ * Obtain the persistent device id string (describes what)
+ *
+ * used by ZED auto-{online,expand,replace}
+ */
+static int
+udev_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
 {
-	int fd;
+	struct udev_list_entry *entry;
-	ddi_devid_t devid;
+	const char *bus;
-	char *minor, *ret;
+	char devbyid[MAXPATHLEN];
 
-	if ((fd = open(path, O_RDONLY)) < 0)
+	/* The bus based by-id path is preferred */
-		return (NULL);
+	bus = udev_device_get_property_value(dev, "ID_BUS");
 
-	minor = NULL;
+	if (bus == NULL) {
-	ret = NULL;
+		const char *dm_uuid;
-	if (devid_get(fd, &devid) == 0) {
+
-		if (devid_get_minor_name(fd, &minor) == 0)
+		/*
-			ret = devid_str_encode(devid, minor);
+		 * For multipath nodes use the persistent uuid based identifier
-		if (minor != NULL)
+		 *
-			devid_str_free(minor);
+		 * Example: /dev/disk/by-id/dm-uuid-mpath-35000c5006304de3f
-		devid_free(devid);
+		 */
+		dm_uuid = udev_device_get_property_value(dev, "DM_UUID");
+		if (dm_uuid != NULL) {
+			(void) snprintf(bufptr, buflen, "dm-uuid-%s", dm_uuid);
+			return (0);
 		}
-	(void) close(fd);
+		return (ENODATA);
+	}
+
+	/*
+	 * locate the bus specific by-id link
+	 */
+	(void) snprintf(devbyid, sizeof (devbyid), "%s%s-", DEV_BYID_PATH, bus);
+	entry = udev_device_get_devlinks_list_entry(dev);
+	while (entry != NULL) {
+		const char *name;
+
+		name = udev_list_entry_get_name(entry);
+		if (strncmp(name, devbyid, strlen(devbyid)) == 0) {
+			name += strlen(DEV_BYID_PATH);
+			(void) strlcpy(bufptr, name, buflen);
+			return (0);
+		}
+		entry = udev_list_entry_get_next(entry);
+	}
+
+	return (ENODATA);
+}
+
+/*
+ * Obtain the persistent physical location string (describes where)
+ *
+ * used by ZED auto-{online,expand,replace}
+ */
+static int
+udev_device_get_physical(struct udev_device *dev, char *bufptr, size_t buflen)
+{
+	const char *physpath, *value;
+
+	/*
+	 * Skip indirect multipath device nodes
+	 */
+	value = udev_device_get_property_value(dev, "DM_MULTIPATH_DEVICE_PATH");
+	if (value != NULL && strcmp(value, "1") == 0)
+		return (ENODATA);  /* skip physical for multipath nodes */
+
+	physpath = udev_device_get_property_value(dev, "ID_PATH");
+	if (physpath != NULL && physpath[0] != '\0') {
+		(void) strlcpy(bufptr, physpath, buflen);
+		return (0);
+	}
+
+	return (ENODATA);
+}
+
+/*
+ * A disk is considered a multipath whole disk when:
+ *	DEVNAME key value has "dm-"
+ *	DM_NAME key value has "mpath" prefix
+ *	DM_UUID key exists
+ *	ID_PART_TABLE_TYPE key does not exist or is not gpt
+ */
+static boolean_t
+udev_mpath_whole_disk(struct udev_device *dev)
+{
+	const char *devname, *mapname, *type, *uuid;
+
+	devname = udev_device_get_property_value(dev, "DEVNAME");
+	mapname = udev_device_get_property_value(dev, "DM_NAME");
+	type = udev_device_get_property_value(dev, "ID_PART_TABLE_TYPE");
+	uuid = udev_device_get_property_value(dev, "DM_UUID");
+
+	if ((devname != NULL && strncmp(devname, "/dev/dm-", 8) == 0) &&
+	    (mapname != NULL && strncmp(mapname, "mpath", 5) == 0) &&
+	    ((type == NULL) || (strcmp(type, "gpt") != 0)) &&
+	    (uuid != NULL)) {
+		return (B_TRUE);
+	}
+
+	return (B_FALSE);
+}
+
+/*
+ * Check if a disk is effectively a multipath whole disk
+ */
+boolean_t
+is_mpath_whole_disk(const char *path)
+{
+	struct udev *udev;
+	struct udev_device *dev = NULL;
+	char nodepath[MAXPATHLEN];
+	char *sysname;
+	boolean_t wholedisk = B_FALSE;
+
+	if (realpath(path, nodepath) == NULL)
+		return (B_FALSE);
+	sysname = strrchr(nodepath, '/') + 1;
+	if (strncmp(sysname, "dm-", 3) != 0)
+		return (B_FALSE);
+	if ((udev = udev_new()) == NULL)
+		return (B_FALSE);
+	if ((dev = udev_device_new_from_subsystem_sysname(udev, "block",
+	    sysname)) == NULL) {
+		udev_device_unref(dev);
+		return (B_FALSE);
+	}
+
+	wholedisk = udev_mpath_whole_disk(dev);
+
+	udev_device_unref(dev);
+	return (wholedisk);
+}
+
+static int
+udev_device_is_ready(struct udev_device *dev)
+{
+#ifdef HAVE_LIBUDEV_UDEV_DEVICE_GET_IS_INITIALIZED
+	return (udev_device_get_is_initialized(dev));
+#else
+	/* wait for DEVLINKS property to be initialized */
+	return (udev_device_get_property_value(dev, "DEVLINKS") != NULL);
+#endif
+}
+
+/*
+ * Encode the persistent devices strings
+ * used for the vdev disk label
+ */
+static int
+encode_device_strings(const char *path, vdev_dev_strs_t *ds,
+    boolean_t wholedisk)
+{
+	struct udev *udev;
+	struct udev_device *dev = NULL;
+	char nodepath[MAXPATHLEN];
+	char *sysname;
+	int ret = ENODEV;
+	hrtime_t start;
+
+	if ((udev = udev_new()) == NULL)
+		return (ENXIO);
+
+	/* resolve path to a runtime device node instance */
+	if (realpath(path, nodepath) == NULL)
+		goto no_dev;
+
+	sysname = strrchr(nodepath, '/') + 1;
+
+	/*
+	 * Wait up to 3 seconds for udev to set up the device node context
+	 */
+	start = gethrtime();
+	do {
+		dev = udev_device_new_from_subsystem_sysname(udev, "block",
+		    sysname);
+		if (dev == NULL)
+			goto no_dev;
+		if (udev_device_is_ready(dev))
+			break;  /* udev ready */
+
+		udev_device_unref(dev);
+		dev = NULL;
+
+		if (NSEC2MSEC(gethrtime() - start) < 10)
+			(void) sched_yield();	/* yield/busy wait up to 10ms */
+		else
+			(void) usleep(10 * MILLISEC);
+
+	} while (NSEC2MSEC(gethrtime() - start) < (3 * MILLISEC));
+
+	if (dev == NULL)
+		goto no_dev;
+
+	/*
+	 * Only whole disks require extra device strings
+	 */
+	if (!wholedisk && !udev_mpath_whole_disk(dev))
+		goto no_dev;
+
+	ret = udev_device_get_devid(dev, ds->vds_devid, sizeof (ds->vds_devid));
+	if (ret != 0)
+		goto no_dev_ref;
+
+	/* physical location string (optional) */
+	if (udev_device_get_physical(dev, ds->vds_devphys,
+	    sizeof (ds->vds_devphys)) != 0) {
+		ds->vds_devphys[0] = '\0'; /* empty string --> not available */
+	}
+
+no_dev_ref:
+	udev_device_unref(dev);
+no_dev:
+	udev_unref(udev);
 
 	return (ret);
 }
 
+/*
+ * Update a leaf vdev's persistent device strings (Linux only)
+ *
+ * - only applies for a dedicated leaf vdev (aka whole disk)
+ * - updated during pool create|add|attach|import
+ * - used for matching device matching during auto-{online,expand,replace}
+ * - stored in a leaf disk config label (i.e. alongside 'path' NVP)
+ * - these strings are currently not used in kernel (i.e. for vdev_disk_open)
+ *
+ * single device node example:
+ * 	devid:		'scsi-MG03SCA300_350000494a8cb3d67-part1'
+ * 	phys_path:	'pci-0000:04:00.0-sas-0x50000394a8cb3d67-lun-0'
+ *
+ * multipath device node example:
+ * 	devid:		'dm-uuid-mpath-35000c5006304de3f'
+ */
+void
+update_vdev_config_dev_strs(nvlist_t *nv)
+{
+	vdev_dev_strs_t vds;
+	char *env, *type, *path;
+	uint64_t wholedisk = 0;
+
+	/*
+	 * For the benefit of legacy ZFS implementations, allow
+	 * for opting out of devid strings in the vdev label.
+	 *
+	 * example use:
+	 *	env ZFS_VDEV_DEVID_OPT_OUT=YES zpool import dozer
+	 *
+	 * explanation:
+	 * Older ZFS on Linux implementations had issues when attempting to
+	 * display pool config VDEV names if a "devid" NVP value is present
+	 * in the pool's config.
+	 *
+	 * For example, a pool that originated on illumos platform would
+	 * have a devid value in the config and "zpool status" would fail
+	 * when listing the config.
+	 *
+	 * A pool can be stripped of any "devid" values on import or
+	 * prevented from adding them on zpool create|add by setting
+	 * ZFS_VDEV_DEVID_OPT_OUT.
+	 */
+	env = getenv("ZFS_VDEV_DEVID_OPT_OUT");
+	if (env && (strtoul(env, NULL, 0) > 0 ||
+	    !strncasecmp(env, "YES", 3) || !strncasecmp(env, "ON", 2))) {
+		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
+		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_PHYS_PATH);
+		return;
+	}
+
+	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_TYPE, &type) != 0 ||
+	    strcmp(type, VDEV_TYPE_DISK) != 0) {
+		return;
+	}
+	if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) != 0)
+		return;
+	(void) nvlist_lookup_uint64(nv, ZPOOL_CONFIG_WHOLE_DISK, &wholedisk);
+
+	/*
+	 * Update device string values in config nvlist
+	 */
+	if (encode_device_strings(path, &vds, (boolean_t)wholedisk) == 0) {
+		(void) nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, vds.vds_devid);
+		if (vds.vds_devphys[0] != '\0') {
+			(void) nvlist_add_string(nv, ZPOOL_CONFIG_PHYS_PATH,
+			    vds.vds_devphys);
+		}
+	} else {
+		/* clear out any stale entries */
+		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
+		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_PHYS_PATH);
+	}
+}
+#else
+
+boolean_t
+is_mpath_whole_disk(const char *path)
+{
+	return (B_FALSE);
+}
+
+void
+update_vdev_config_dev_strs(nvlist_t *nv)
+{
+}
+
+#endif /* HAVE_LIBUDEV */
+
 
 /*
  * Go through and fix up any path and/or devid information for the given vdev
@@ -130,7 +433,7 @@ fix_paths(nvlist_t *nv, name_entry_t *names)
 	uint_t c, children;
 	uint64_t guid;
 	name_entry_t *ne, *best;
-	char *path, *devid;
+	char *path;
 
 	if (nvlist_lookup_nvlist_array(nv, ZPOOL_CONFIG_CHILDREN,
 	    &child, &children) == 0) {
@@ -197,15 +500,8 @@ fix_paths(nvlist_t *nv, name_entry_t *names)
 	if (nvlist_add_string(nv, ZPOOL_CONFIG_PATH, best->ne_name) != 0)
 		return (-1);
 
-	if ((devid = get_devid(best->ne_name)) == NULL) {
+	/* Linux only - update ZPOOL_CONFIG_DEVID and ZPOOL_CONFIG_PHYS_PATH */
-		(void) nvlist_remove_all(nv, ZPOOL_CONFIG_DEVID);
+	update_vdev_config_dev_strs(nv);
-	} else {
-		if (nvlist_add_string(nv, ZPOOL_CONFIG_DEVID, devid) != 0) {
-			devid_str_free(devid);
-			return (-1);
-		}
-		devid_str_free(devid);
-	}
 
 	return (0);
 }

lib/libzfs/libzfs_pool.c

@@ -3312,6 +3312,7 @@ zpool_reopen(zpool_handle_t *zhp)
 	return (zpool_standard_error(hdl, errno, msg));
 }
 
+#if defined(__sun__) || defined(__sun)
 /*
  * Convert from a devid string to a path.
  */
@@ -3389,6 +3390,7 @@ set_path(zpool_handle_t *zhp, nvlist_t *nv, const char *path)
 
 	(void) ioctl(zhp->zpool_hdl->libzfs_fd, ZFS_IOC_VDEV_SETPATH, &zc);
 }
+#endif /* sun */
 
 /*
  * Remove partition suffix from a vdev path.  Partition suffixes may take three
@@ -3443,12 +3445,10 @@ char *
 zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
     int name_flags)
 {
-	char *path, *devid, *type, *env;
+	char *path, *type, *env;
 	uint64_t value;
 	char buf[PATH_BUF_LEN];
 	char tmpbuf[PATH_BUF_LEN];
-	vdev_stat_t *vs;
-	uint_t vsc;
 
 	env = getenv("ZPOOL_VDEV_NAME_PATH");
 	if (env && (strtoul(env, NULL, 0) > 0 ||
@@ -3471,6 +3471,15 @@ zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
 		(void) snprintf(buf, sizeof (buf), "%llu", (u_longlong_t)value);
 		path = buf;
 	} else if (nvlist_lookup_string(nv, ZPOOL_CONFIG_PATH, &path) == 0) {
+#if defined(__sun__) || defined(__sun)
+		/*
+		 * Live VDEV path updates to a kernel VDEV during a
+		 * zpool_vdev_name lookup are not supported on Linux.
+		 */
+		char *devid;
+		vdev_stat_t *vs;
+		uint_t vsc;
+
 		/*
 		 * If the device is dead (faulted, offline, etc) then don't
 		 * bother opening it.  Otherwise we may be forcing the user to
@@ -3508,6 +3517,7 @@ zpool_vdev_name(libzfs_handle_t *hdl, zpool_handle_t *zhp, nvlist_t *nv,
 			if (newdevid)
 				devid_str_free(newdevid);
 		}
+#endif /* sun */
 
 		if (name_flags & VDEV_NAME_FOLLOW_LINKS) {
 			char *rp = realpath(path, NULL);

man/man8/zpool.8

@@ -2603,6 +2603,17 @@ Cause \fBzpool\fR subcommands to follow links for vdev names by default.  This b
 .B "ZPOOL_VDEV_NAME_PATH"
 Cause \fBzpool\fR subcommands to output full vdev path names by default.  This
 behavior is identical to the \fBzpool status -p\fR command line option.
+.TP
+.B "ZFS_VDEV_DEVID_OPT_OUT"
+Older ZFS on Linux implementations had issues when attempting to display pool
+config VDEV names if a "devid" NVP value is present in the pool's config.
+
+For example, a pool that originated on illumos platform would have a devid
+value in the config and \fBzpool status\fR would fail when listing the config.
+This would also be true for future Linux based pools.
+
+A pool can be stripped of any "devid" values on import or prevented from adding
+them on \fBzpool create\fR or \fBzpool add\fR by setting ZFS_VDEV_DEVID_OPT_OUT.
 
 .SH SEE ALSO
 .sp

tests/zfs-tests/cmd/devname2devid/Makefile.am

@@ -1,7 +1,7 @@
 include $(top_srcdir)/config/Rules.am
 
-pkgdatadir = $(datadir)/@PACKAGE@/zfs-tests/bin
+pkgexecdir = $(datadir)/@PACKAGE@/zfs-tests/bin
 
-# DISABLED: Not supported under Linux
+pkgexec_PROGRAMS = devname2devid
-# pkgdata_PROGRAMS = devname2devid
+devname2devid_SOURCES = devname2devid.c
-# devname2devid_SOURCES = devname2devid.c
+devname2devid_LDADD = -ludev

tests/zfs-tests/cmd/devname2devid/devname2devid.c

@@ -22,27 +22,85 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ *
+ * Copyright (c) 2016, Intel Corporation.
  */
 
 #include <sys/types.h>
+#include <sys/param.h>
 #include <sys/stat.h>
-#include <devid.h>
+#include <libudev.h>
 #include <errno.h>
 #include <stdio.h>
 #include <stdlib.h>
+#include <string.h>
 #include <fcntl.h>
 
+/*
+ * Linux persistent device strings for vdev labels
+ *
+ * based on udev_device_get_devid() at zfs/lib/libzfs/libzfs_import.c
+ */
+
+#define	DEV_BYID_PATH	"/dev/disk/by-id/"
+
+static int
+udev_device_get_devid(struct udev_device *dev, char *bufptr, size_t buflen)
+{
+	struct udev_list_entry *entry;
+	const char *bus;
+	char devbyid[MAXPATHLEN];
+
+	/* The bus based by-id path is preferred */
+	bus = udev_device_get_property_value(dev, "ID_BUS");
+
+	if (bus == NULL) {
+		const char *dm_uuid;
+
+		/*
+		 * For multipath nodes use the persistent uuid based identifier
+		 *
+		 * Example: 'dm-uuid-mpath-35000c5006304de3f'
+		 */
+		dm_uuid = udev_device_get_property_value(dev, "DM_UUID");
+		if (dm_uuid != NULL) {
+			(void) snprintf(bufptr, buflen, "dm-uuid-%s", dm_uuid);
+			return (0);
+		}
+		return (ENODATA);
+	}
+
+	/*
+	 * locate the bus specific by-id link
+	 *
+	 * Example: 'scsi-MG03SCA300_350000494a8cb3d67-part1'
+	 */
+	(void) snprintf(devbyid, sizeof (devbyid), "%s%s-", DEV_BYID_PATH, bus);
+	entry = udev_device_get_devlinks_list_entry(dev);
+	while (entry != NULL) {
+		const char *name;
+
+		name = udev_list_entry_get_name(entry);
+		if (strncmp(name, devbyid, strlen(devbyid)) == 0) {
+			name += strlen(DEV_BYID_PATH);
+			(void) stpncpy(bufptr, name, buflen);
+			return (0);
+		}
+		entry = udev_list_entry_get_next(entry);
+	}
+
+	return (ENODATA);
+}
+
 /*
  * Usage: devname2devid <devicepath>
  *
  * Examples:
- *	# ./devname2devid /dev/dsk/c1t4d0s0
+ * # ./devname2devid /dev/sda1
- *	devid id1,sd@SSEAGATE_ST318404LSUN18G_3BT2G0Z300002146G4CR/a
+ * devid scsi-350000394a8caede4-part1
- *	# ./devname2devid /dev/dsk/c1t4d0
+ *
- *	devid id1,sd@SSEAGATE_ST318404LSUN18G_3BT2G0Z300002146G4CR/wd
+ * # ./devname2devid /dev/dm-1
- *	# ./devname2devid /dev/dsk/c1t4d0s1
+ * devid: 'dm-uuid-mpath-35000c5006304de3f'
- *	devid id1,sd@SSEAGATE_ST318404LSUN18G_3BT2G0Z300002146G4CR/b
- *	#
  *
  * This program accepts a disk or disk slice path and prints a
  * device id.
@@ -55,66 +113,47 @@
 int
 main(int argc, char *argv[])
 {
-	int		fd;
+	struct udev *udev;
-	ddi_devid_t	devid;
+	struct udev_device *dev = NULL;
-	char		*minor_name, *devidstr, *device;
+	char devid[128], nodepath[MAXPATHLEN];
-#ifdef DEBUG
+	char *device, *sysname;
-	devid_nmlist_t  *list = NULL;
+	int ret;
-	char		*search_path;
-	int		i;
-#endif
 
 	if (argc == 1) {
-		(void) printf("%s <devicepath> [search path]\n",
+		(void) printf("%s <devicepath> [search path]\n", argv[0]);
-		    argv[0]);
 		exit(1);
 	}
 	device = argv[1];
 
-	if ((fd = open(device, O_RDONLY|O_NDELAY)) < 0) {
+	if ((udev = udev_new()) == NULL) {
-		perror(device);
+		perror("udev_new");
-		exit(1);
-	}
-	if (devid_get(fd, &devid) != 0) {
-		perror("devid_get");
-		exit(1);
-	}
-	if (devid_get_minor_name(fd, &minor_name) != 0) {
-		perror("devid_get_minor_name");
-		exit(1);
-	}
-	if ((devidstr = devid_str_encode(devid, minor_name)) == 0) {
-		perror("devid_str_encode");
 		exit(1);
 	}
 
-	(void) printf("devid %s\n", devidstr);
+	/* resolve path to a runtime device node instance */
-
+	if (realpath(device, nodepath) == NULL) {
-	devid_str_free(devidstr);
+		perror("realpath");
-
+		exit(1);
-#ifdef DEBUG
-	if (argc == 3) {
-		search_path = argv[2];
-	} else {
-		search_path = "/dev/rdsk";
 	}
+	sysname = strrchr(nodepath, '/') + 1;
 
-	if (devid_deviceid_to_nmlist(search_path, devid, DEVID_MINOR_NAME_ALL,
+	if ((dev = udev_device_new_from_subsystem_sysname(udev, "block",
-	    &list)) {
+	    sysname)) == NULL) {
-		perror("devid_deviceid_to_nmlist");
+		perror(sysname);
 		exit(1);
 	}
 
-	/* loop through list and process device names and numbers */
+	if ((ret = udev_device_get_devid(dev, devid, sizeof (devid))) != 0) {
-	for (i = 0; list[i].devname != NULL; i++) {
+		udev_device_unref(dev);
-		(void) printf("devname: %s %p\n", list[i].devname, list[i].dev);
+		errno = ret;
+		perror(sysname);
+		exit(1);
 	}
-	devid_free_nmlist(list);
 
-#endif /* DEBUG */
+	(void) printf("devid %s\n", devid);
 
-	devid_str_free(minor_name);
+	udev_device_unref(dev);
-	devid_free(devid);
+	udev_unref(udev);
 
 	return (0);
 }