xref: /titanic_50/usr/src/lib/libzfs/common/libzfs_dataset.c (revision 551bc2a66868b5cb5be6b70ab9f55515e77a39a9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <assert.h>
30 #include <ctype.h>
31 #include <errno.h>
32 #include <libdevinfo.h>
33 #include <libintl.h>
34 #include <math.h>
35 #include <stdio.h>
36 #include <stdlib.h>
37 #include <strings.h>
38 #include <unistd.h>
39 #include <stddef.h>
40 #include <zone.h>
41 #include <fcntl.h>
42 #include <sys/mntent.h>
43 #include <sys/mnttab.h>
44 #include <sys/mount.h>
45 #include <sys/avl.h>
46 #include <priv.h>
47 #include <pwd.h>
48 #include <grp.h>
49 #include <stddef.h>
50 #include <ucred.h>
51 
52 #include <sys/spa.h>
53 #include <sys/zap.h>
54 #include <libzfs.h>
55 
56 #include "zfs_namecheck.h"
57 #include "zfs_prop.h"
58 #include "libzfs_impl.h"
59 #include "zfs_deleg.h"
60 
61 static int zvol_create_link_common(libzfs_handle_t *, const char *, int);
62 
63 /*
64  * Given a single type (not a mask of types), return the type in a human
65  * readable form.
66  */
67 const char *
68 zfs_type_to_name(zfs_type_t type)
69 {
70 	switch (type) {
71 	case ZFS_TYPE_FILESYSTEM:
72 		return (dgettext(TEXT_DOMAIN, "filesystem"));
73 	case ZFS_TYPE_SNAPSHOT:
74 		return (dgettext(TEXT_DOMAIN, "snapshot"));
75 	case ZFS_TYPE_VOLUME:
76 		return (dgettext(TEXT_DOMAIN, "volume"));
77 	}
78 
79 	return (NULL);
80 }
81 
82 /*
83  * Given a path and mask of ZFS types, return a string describing this dataset.
84  * This is used when we fail to open a dataset and we cannot get an exact type.
85  * We guess what the type would have been based on the path and the mask of
86  * acceptable types.
87  */
88 static const char *
89 path_to_str(const char *path, int types)
90 {
91 	/*
92 	 * When given a single type, always report the exact type.
93 	 */
94 	if (types == ZFS_TYPE_SNAPSHOT)
95 		return (dgettext(TEXT_DOMAIN, "snapshot"));
96 	if (types == ZFS_TYPE_FILESYSTEM)
97 		return (dgettext(TEXT_DOMAIN, "filesystem"));
98 	if (types == ZFS_TYPE_VOLUME)
99 		return (dgettext(TEXT_DOMAIN, "volume"));
100 
101 	/*
102 	 * The user is requesting more than one type of dataset.  If this is the
103 	 * case, consult the path itself.  If we're looking for a snapshot, and
104 	 * a '@' is found, then report it as "snapshot".  Otherwise, remove the
105 	 * snapshot attribute and try again.
106 	 */
107 	if (types & ZFS_TYPE_SNAPSHOT) {
108 		if (strchr(path, '@') != NULL)
109 			return (dgettext(TEXT_DOMAIN, "snapshot"));
110 		return (path_to_str(path, types & ~ZFS_TYPE_SNAPSHOT));
111 	}
112 
113 
114 	/*
115 	 * The user has requested either filesystems or volumes.
116 	 * We have no way of knowing a priori what type this would be, so always
117 	 * report it as "filesystem" or "volume", our two primitive types.
118 	 */
119 	if (types & ZFS_TYPE_FILESYSTEM)
120 		return (dgettext(TEXT_DOMAIN, "filesystem"));
121 
122 	assert(types & ZFS_TYPE_VOLUME);
123 	return (dgettext(TEXT_DOMAIN, "volume"));
124 }
125 
126 /*
127  * Validate a ZFS path.  This is used even before trying to open the dataset, to
128  * provide a more meaningful error message.  We place a more useful message in
129  * 'buf' detailing exactly why the name was not valid.
130  */
131 static int
132 zfs_validate_name(libzfs_handle_t *hdl, const char *path, int type,
133     boolean_t modifying)
134 {
135 	namecheck_err_t why;
136 	char what;
137 
138 	if (dataset_namecheck(path, &why, &what) != 0) {
139 		if (hdl != NULL) {
140 			switch (why) {
141 			case NAME_ERR_TOOLONG:
142 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
143 				    "name is too long"));
144 				break;
145 
146 			case NAME_ERR_LEADING_SLASH:
147 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
148 				    "leading slash in name"));
149 				break;
150 
151 			case NAME_ERR_EMPTY_COMPONENT:
152 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
153 				    "empty component in name"));
154 				break;
155 
156 			case NAME_ERR_TRAILING_SLASH:
157 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
158 				    "trailing slash in name"));
159 				break;
160 
161 			case NAME_ERR_INVALCHAR:
162 				zfs_error_aux(hdl,
163 				    dgettext(TEXT_DOMAIN, "invalid character "
164 				    "'%c' in name"), what);
165 				break;
166 
167 			case NAME_ERR_MULTIPLE_AT:
168 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
169 				    "multiple '@' delimiters in name"));
170 				break;
171 
172 			case NAME_ERR_NOLETTER:
173 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
174 				    "pool doesn't begin with a letter"));
175 				break;
176 
177 			case NAME_ERR_RESERVED:
178 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
179 				    "name is reserved"));
180 				break;
181 
182 			case NAME_ERR_DISKLIKE:
183 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
184 				    "reserved disk name"));
185 				break;
186 			}
187 		}
188 
189 		return (0);
190 	}
191 
192 	if (!(type & ZFS_TYPE_SNAPSHOT) && strchr(path, '@') != NULL) {
193 		if (hdl != NULL)
194 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
195 			    "snapshot delimiter '@' in filesystem name"));
196 		return (0);
197 	}
198 
199 	if (type == ZFS_TYPE_SNAPSHOT && strchr(path, '@') == NULL) {
200 		if (hdl != NULL)
201 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
202 			    "missing '@' delimiter in snapshot name"));
203 		return (0);
204 	}
205 
206 	if (modifying && strchr(path, '%') != NULL) {
207 		if (hdl != NULL)
208 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
209 			    "invalid character %c in name"), '%');
210 		return (0);
211 	}
212 
213 	return (-1);
214 }
215 
216 int
217 zfs_name_valid(const char *name, zfs_type_t type)
218 {
219 	return (zfs_validate_name(NULL, name, type, B_FALSE));
220 }
221 
222 /*
223  * This function takes the raw DSL properties, and filters out the user-defined
224  * properties into a separate nvlist.
225  */
226 static nvlist_t *
227 process_user_props(zfs_handle_t *zhp, nvlist_t *props)
228 {
229 	libzfs_handle_t *hdl = zhp->zfs_hdl;
230 	nvpair_t *elem;
231 	nvlist_t *propval;
232 	nvlist_t *nvl;
233 
234 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0) {
235 		(void) no_memory(hdl);
236 		return (NULL);
237 	}
238 
239 	elem = NULL;
240 	while ((elem = nvlist_next_nvpair(props, elem)) != NULL) {
241 		if (!zfs_prop_user(nvpair_name(elem)))
242 			continue;
243 
244 		verify(nvpair_value_nvlist(elem, &propval) == 0);
245 		if (nvlist_add_nvlist(nvl, nvpair_name(elem), propval) != 0) {
246 			nvlist_free(nvl);
247 			(void) no_memory(hdl);
248 			return (NULL);
249 		}
250 	}
251 
252 	return (nvl);
253 }
254 
255 /*
256  * Utility function to gather stats (objset and zpl) for the given object.
257  */
258 static int
259 get_stats(zfs_handle_t *zhp)
260 {
261 	zfs_cmd_t zc = { 0 };
262 	libzfs_handle_t *hdl = zhp->zfs_hdl;
263 	nvlist_t *allprops, *userprops;
264 
265 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
266 
267 	if (zcmd_alloc_dst_nvlist(hdl, &zc, 0) != 0)
268 		return (-1);
269 
270 	while (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0) {
271 		if (errno == ENOMEM) {
272 			if (zcmd_expand_dst_nvlist(hdl, &zc) != 0) {
273 				zcmd_free_nvlists(&zc);
274 				return (-1);
275 			}
276 		} else {
277 			zcmd_free_nvlists(&zc);
278 			return (-1);
279 		}
280 	}
281 
282 	zhp->zfs_dmustats = zc.zc_objset_stats; /* structure assignment */
283 
284 	(void) strlcpy(zhp->zfs_root, zc.zc_value, sizeof (zhp->zfs_root));
285 
286 	if (zcmd_read_dst_nvlist(hdl, &zc, &allprops) != 0) {
287 		zcmd_free_nvlists(&zc);
288 		return (-1);
289 	}
290 
291 	zcmd_free_nvlists(&zc);
292 
293 	if ((userprops = process_user_props(zhp, allprops)) == NULL) {
294 		nvlist_free(allprops);
295 		return (-1);
296 	}
297 
298 	nvlist_free(zhp->zfs_props);
299 	nvlist_free(zhp->zfs_user_props);
300 
301 	zhp->zfs_props = allprops;
302 	zhp->zfs_user_props = userprops;
303 
304 	return (0);
305 }
306 
307 /*
308  * Refresh the properties currently stored in the handle.
309  */
310 void
311 zfs_refresh_properties(zfs_handle_t *zhp)
312 {
313 	(void) get_stats(zhp);
314 }
315 
316 /*
317  * Makes a handle from the given dataset name.  Used by zfs_open() and
318  * zfs_iter_* to create child handles on the fly.
319  */
320 zfs_handle_t *
321 make_dataset_handle(libzfs_handle_t *hdl, const char *path)
322 {
323 	zfs_handle_t *zhp = calloc(sizeof (zfs_handle_t), 1);
324 	char *logstr;
325 
326 	if (zhp == NULL)
327 		return (NULL);
328 
329 	zhp->zfs_hdl = hdl;
330 
331 	/*
332 	 * Preserve history log string.
333 	 * any changes performed here will be
334 	 * logged as an internal event.
335 	 */
336 	logstr = zhp->zfs_hdl->libzfs_log_str;
337 	zhp->zfs_hdl->libzfs_log_str = NULL;
338 top:
339 	(void) strlcpy(zhp->zfs_name, path, sizeof (zhp->zfs_name));
340 
341 	if (get_stats(zhp) != 0) {
342 		zhp->zfs_hdl->libzfs_log_str = logstr;
343 		free(zhp);
344 		return (NULL);
345 	}
346 
347 	if (zhp->zfs_dmustats.dds_inconsistent) {
348 		zfs_cmd_t zc = { 0 };
349 
350 		/*
351 		 * If it is dds_inconsistent, then we've caught it in
352 		 * the middle of a 'zfs receive' or 'zfs destroy', and
353 		 * it is inconsistent from the ZPL's point of view, so
354 		 * can't be mounted.  However, it could also be that we
355 		 * have crashed in the middle of one of those
356 		 * operations, in which case we need to get rid of the
357 		 * inconsistent state.  We do that by either rolling
358 		 * back to the previous snapshot (which will fail if
359 		 * there is none), or destroying the filesystem.  Note
360 		 * that if we are still in the middle of an active
361 		 * 'receive' or 'destroy', then the rollback and destroy
362 		 * will fail with EBUSY and we will drive on as usual.
363 		 */
364 
365 		(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
366 
367 		if (zhp->zfs_dmustats.dds_type == DMU_OST_ZVOL) {
368 			(void) zvol_remove_link(hdl, zhp->zfs_name);
369 			zc.zc_objset_type = DMU_OST_ZVOL;
370 		} else {
371 			zc.zc_objset_type = DMU_OST_ZFS;
372 		}
373 
374 		/*
375 		 * If we can successfully destroy it, pretend that it
376 		 * never existed.
377 		 */
378 		if (ioctl(hdl->libzfs_fd, ZFS_IOC_DESTROY, &zc) == 0) {
379 			zhp->zfs_hdl->libzfs_log_str = logstr;
380 			free(zhp);
381 			errno = ENOENT;
382 			return (NULL);
383 		}
384 		/* If we can successfully roll it back, reget the stats */
385 		if (ioctl(hdl->libzfs_fd, ZFS_IOC_ROLLBACK, &zc) == 0)
386 			goto top;
387 	}
388 
389 	/*
390 	 * We've managed to open the dataset and gather statistics.  Determine
391 	 * the high-level type.
392 	 */
393 	if (zhp->zfs_dmustats.dds_type == DMU_OST_ZVOL)
394 		zhp->zfs_head_type = ZFS_TYPE_VOLUME;
395 	else if (zhp->zfs_dmustats.dds_type == DMU_OST_ZFS)
396 		zhp->zfs_head_type = ZFS_TYPE_FILESYSTEM;
397 	else
398 		abort();
399 
400 	if (zhp->zfs_dmustats.dds_is_snapshot)
401 		zhp->zfs_type = ZFS_TYPE_SNAPSHOT;
402 	else if (zhp->zfs_dmustats.dds_type == DMU_OST_ZVOL)
403 		zhp->zfs_type = ZFS_TYPE_VOLUME;
404 	else if (zhp->zfs_dmustats.dds_type == DMU_OST_ZFS)
405 		zhp->zfs_type = ZFS_TYPE_FILESYSTEM;
406 	else
407 		abort();	/* we should never see any other types */
408 
409 	zhp->zfs_hdl->libzfs_log_str = logstr;
410 	return (zhp);
411 }
412 
413 /*
414  * Opens the given snapshot, filesystem, or volume.   The 'types'
415  * argument is a mask of acceptable types.  The function will print an
416  * appropriate error message and return NULL if it can't be opened.
417  */
418 zfs_handle_t *
419 zfs_open(libzfs_handle_t *hdl, const char *path, int types)
420 {
421 	zfs_handle_t *zhp;
422 	char errbuf[1024];
423 
424 	(void) snprintf(errbuf, sizeof (errbuf),
425 	    dgettext(TEXT_DOMAIN, "cannot open '%s'"), path);
426 
427 	/*
428 	 * Validate the name before we even try to open it.
429 	 */
430 	if (!zfs_validate_name(hdl, path, ZFS_TYPE_DATASET, B_FALSE)) {
431 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
432 		    "invalid dataset name"));
433 		(void) zfs_error(hdl, EZFS_INVALIDNAME, errbuf);
434 		return (NULL);
435 	}
436 
437 	/*
438 	 * Try to get stats for the dataset, which will tell us if it exists.
439 	 */
440 	errno = 0;
441 	if ((zhp = make_dataset_handle(hdl, path)) == NULL) {
442 		(void) zfs_standard_error(hdl, errno, errbuf);
443 		return (NULL);
444 	}
445 
446 	if (!(types & zhp->zfs_type)) {
447 		(void) zfs_error(hdl, EZFS_BADTYPE, errbuf);
448 		zfs_close(zhp);
449 		return (NULL);
450 	}
451 
452 	return (zhp);
453 }
454 
455 /*
456  * Release a ZFS handle.  Nothing to do but free the associated memory.
457  */
458 void
459 zfs_close(zfs_handle_t *zhp)
460 {
461 	if (zhp->zfs_mntopts)
462 		free(zhp->zfs_mntopts);
463 	nvlist_free(zhp->zfs_props);
464 	nvlist_free(zhp->zfs_user_props);
465 	free(zhp);
466 }
467 
468 int
469 zfs_spa_version(zfs_handle_t *zhp, int *spa_version)
470 {
471 	char *pool_name;
472 	zpool_handle_t *zpool_handle;
473 	char *p;
474 
475 	pool_name = zfs_alloc(zhp->zfs_hdl, MAXPATHLEN);
476 	if (zfs_prop_get(zhp, ZFS_PROP_NAME, pool_name,
477 	    MAXPATHLEN, NULL, NULL, 0, B_FALSE) != 0) {
478 		free(pool_name);
479 		return (-1);
480 	}
481 
482 	if (p = strchr(pool_name, '/'))
483 		*p = '\0';
484 	zpool_handle = zpool_open(zhp->zfs_hdl, pool_name);
485 	free(pool_name);
486 	if (zpool_handle == NULL)
487 		return (-1);
488 
489 	*spa_version = zpool_get_prop_int(zpool_handle,
490 	    ZPOOL_PROP_VERSION, NULL);
491 	zpool_close(zpool_handle);
492 	return (0);
493 }
494 
495 /*
496  * The choice of reservation property depends on the SPA version.
497  */
498 static int
499 zfs_which_resv_prop(zfs_handle_t *zhp, zfs_prop_t *resv_prop)
500 {
501 	int spa_version;
502 
503 	if (zfs_spa_version(zhp, &spa_version) < 0)
504 		return (-1);
505 
506 	if (spa_version >= SPA_VERSION_REFRESERVATION)
507 		*resv_prop = ZFS_PROP_REFRESERVATION;
508 	else
509 		*resv_prop = ZFS_PROP_RESERVATION;
510 
511 	return (0);
512 }
513 
514 /*
515  * Given an nvlist of properties to set, validates that they are correct, and
516  * parses any numeric properties (index, boolean, etc) if they are specified as
517  * strings.
518  */
519 static nvlist_t *
520 zfs_validate_properties(libzfs_handle_t *hdl, zfs_type_t type, nvlist_t *nvl,
521     uint64_t zoned, zfs_handle_t *zhp, const char *errbuf)
522 {
523 	nvpair_t *elem;
524 	uint64_t intval;
525 	char *strval;
526 	zfs_prop_t prop;
527 	nvlist_t *ret;
528 	int chosen_normal = -1;
529 	int chosen_utf = -1;
530 
531 	if (type == ZFS_TYPE_SNAPSHOT) {
532 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
533 		    "snapshot properties cannot be modified"));
534 		(void) zfs_error(hdl, EZFS_PROPTYPE, errbuf);
535 		return (NULL);
536 	}
537 
538 	if (nvlist_alloc(&ret, NV_UNIQUE_NAME, 0) != 0) {
539 		(void) no_memory(hdl);
540 		return (NULL);
541 	}
542 
543 	elem = NULL;
544 	while ((elem = nvlist_next_nvpair(nvl, elem)) != NULL) {
545 		const char *propname = nvpair_name(elem);
546 
547 		/*
548 		 * Make sure this property is valid and applies to this type.
549 		 */
550 		if ((prop = zfs_name_to_prop(propname)) == ZPROP_INVAL) {
551 			if (!zfs_prop_user(propname)) {
552 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
553 				    "invalid property '%s'"), propname);
554 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
555 				goto error;
556 			}
557 
558 			/*
559 			 * If this is a user property, make sure it's a
560 			 * string, and that it's less than ZAP_MAXNAMELEN.
561 			 */
562 			if (nvpair_type(elem) != DATA_TYPE_STRING) {
563 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
564 				    "'%s' must be a string"), propname);
565 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
566 				goto error;
567 			}
568 
569 			if (strlen(nvpair_name(elem)) >= ZAP_MAXNAMELEN) {
570 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
571 				    "property name '%s' is too long"),
572 				    propname);
573 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
574 				goto error;
575 			}
576 
577 			(void) nvpair_value_string(elem, &strval);
578 			if (nvlist_add_string(ret, propname, strval) != 0) {
579 				(void) no_memory(hdl);
580 				goto error;
581 			}
582 			continue;
583 		}
584 
585 		if (!zfs_prop_valid_for_type(prop, type)) {
586 			zfs_error_aux(hdl,
587 			    dgettext(TEXT_DOMAIN, "'%s' does not "
588 			    "apply to datasets of this type"), propname);
589 			(void) zfs_error(hdl, EZFS_PROPTYPE, errbuf);
590 			goto error;
591 		}
592 
593 		if (zfs_prop_readonly(prop) &&
594 		    (!zfs_prop_setonce(prop) || zhp != NULL)) {
595 			zfs_error_aux(hdl,
596 			    dgettext(TEXT_DOMAIN, "'%s' is readonly"),
597 			    propname);
598 			(void) zfs_error(hdl, EZFS_PROPREADONLY, errbuf);
599 			goto error;
600 		}
601 
602 		if (zprop_parse_value(hdl, elem, prop, type, ret,
603 		    &strval, &intval, errbuf) != 0)
604 			goto error;
605 
606 		/*
607 		 * Perform some additional checks for specific properties.
608 		 */
609 		switch (prop) {
610 		case ZFS_PROP_VERSION:
611 		{
612 			int version;
613 
614 			if (zhp == NULL)
615 				break;
616 			version = zfs_prop_get_int(zhp, ZFS_PROP_VERSION);
617 			if (intval < version) {
618 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
619 				    "Can not downgrade; already at version %u"),
620 				    version);
621 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
622 				goto error;
623 			}
624 			break;
625 		}
626 
627 		case ZFS_PROP_RECORDSIZE:
628 		case ZFS_PROP_VOLBLOCKSIZE:
629 			/* must be power of two within SPA_{MIN,MAX}BLOCKSIZE */
630 			if (intval < SPA_MINBLOCKSIZE ||
631 			    intval > SPA_MAXBLOCKSIZE || !ISP2(intval)) {
632 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
633 				    "'%s' must be power of 2 from %u "
634 				    "to %uk"), propname,
635 				    (uint_t)SPA_MINBLOCKSIZE,
636 				    (uint_t)SPA_MAXBLOCKSIZE >> 10);
637 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
638 				goto error;
639 			}
640 			break;
641 
642 		case ZFS_PROP_SHAREISCSI:
643 			if (strcmp(strval, "off") != 0 &&
644 			    strcmp(strval, "on") != 0 &&
645 			    strcmp(strval, "type=disk") != 0) {
646 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
647 				    "'%s' must be 'on', 'off', or 'type=disk'"),
648 				    propname);
649 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
650 				goto error;
651 			}
652 
653 			break;
654 
655 		case ZFS_PROP_MOUNTPOINT:
656 		{
657 			namecheck_err_t why;
658 
659 			if (strcmp(strval, ZFS_MOUNTPOINT_NONE) == 0 ||
660 			    strcmp(strval, ZFS_MOUNTPOINT_LEGACY) == 0)
661 				break;
662 
663 			if (mountpoint_namecheck(strval, &why)) {
664 				switch (why) {
665 				case NAME_ERR_LEADING_SLASH:
666 					zfs_error_aux(hdl,
667 					    dgettext(TEXT_DOMAIN,
668 					    "'%s' must be an absolute path, "
669 					    "'none', or 'legacy'"), propname);
670 					break;
671 				case NAME_ERR_TOOLONG:
672 					zfs_error_aux(hdl,
673 					    dgettext(TEXT_DOMAIN,
674 					    "component of '%s' is too long"),
675 					    propname);
676 					break;
677 				}
678 				(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
679 				goto error;
680 			}
681 		}
682 
683 			/*FALLTHRU*/
684 
685 		case ZFS_PROP_SHARESMB:
686 		case ZFS_PROP_SHARENFS:
687 			/*
688 			 * For the mountpoint and sharenfs or sharesmb
689 			 * properties, check if it can be set in a
690 			 * global/non-global zone based on
691 			 * the zoned property value:
692 			 *
693 			 *		global zone	    non-global zone
694 			 * --------------------------------------------------
695 			 * zoned=on	mountpoint (no)	    mountpoint (yes)
696 			 *		sharenfs (no)	    sharenfs (no)
697 			 *		sharesmb (no)	    sharesmb (no)
698 			 *
699 			 * zoned=off	mountpoint (yes)	N/A
700 			 *		sharenfs (yes)
701 			 *		sharesmb (yes)
702 			 */
703 			if (zoned) {
704 				if (getzoneid() == GLOBAL_ZONEID) {
705 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
706 					    "'%s' cannot be set on "
707 					    "dataset in a non-global zone"),
708 					    propname);
709 					(void) zfs_error(hdl, EZFS_ZONED,
710 					    errbuf);
711 					goto error;
712 				} else if (prop == ZFS_PROP_SHARENFS ||
713 				    prop == ZFS_PROP_SHARESMB) {
714 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
715 					    "'%s' cannot be set in "
716 					    "a non-global zone"), propname);
717 					(void) zfs_error(hdl, EZFS_ZONED,
718 					    errbuf);
719 					goto error;
720 				}
721 			} else if (getzoneid() != GLOBAL_ZONEID) {
722 				/*
723 				 * If zoned property is 'off', this must be in
724 				 * a globle zone. If not, something is wrong.
725 				 */
726 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
727 				    "'%s' cannot be set while dataset "
728 				    "'zoned' property is set"), propname);
729 				(void) zfs_error(hdl, EZFS_ZONED, errbuf);
730 				goto error;
731 			}
732 
733 			/*
734 			 * At this point, it is legitimate to set the
735 			 * property. Now we want to make sure that the
736 			 * property value is valid if it is sharenfs.
737 			 */
738 			if ((prop == ZFS_PROP_SHARENFS ||
739 			    prop == ZFS_PROP_SHARESMB) &&
740 			    strcmp(strval, "on") != 0 &&
741 			    strcmp(strval, "off") != 0) {
742 				zfs_share_proto_t proto;
743 
744 				if (prop == ZFS_PROP_SHARESMB)
745 					proto = PROTO_SMB;
746 				else
747 					proto = PROTO_NFS;
748 
749 				/*
750 				 * Must be an valid sharing protocol
751 				 * option string so init the libshare
752 				 * in order to enable the parser and
753 				 * then parse the options. We use the
754 				 * control API since we don't care about
755 				 * the current configuration and don't
756 				 * want the overhead of loading it
757 				 * until we actually do something.
758 				 */
759 
760 				if (zfs_init_libshare(hdl,
761 				    SA_INIT_CONTROL_API) != SA_OK) {
762 					/*
763 					 * An error occurred so we can't do
764 					 * anything
765 					 */
766 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
767 					    "'%s' cannot be set: problem "
768 					    "in share initialization"),
769 					    propname);
770 					(void) zfs_error(hdl, EZFS_BADPROP,
771 					    errbuf);
772 					goto error;
773 				}
774 
775 				if (zfs_parse_options(strval, proto) != SA_OK) {
776 					/*
777 					 * There was an error in parsing so
778 					 * deal with it by issuing an error
779 					 * message and leaving after
780 					 * uninitializing the the libshare
781 					 * interface.
782 					 */
783 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
784 					    "'%s' cannot be set to invalid "
785 					    "options"), propname);
786 					(void) zfs_error(hdl, EZFS_BADPROP,
787 					    errbuf);
788 					zfs_uninit_libshare(hdl);
789 					goto error;
790 				}
791 				zfs_uninit_libshare(hdl);
792 			}
793 
794 			break;
795 		case ZFS_PROP_UTF8ONLY:
796 			chosen_utf = (int)intval;
797 			break;
798 		case ZFS_PROP_NORMALIZE:
799 			chosen_normal = (int)intval;
800 			break;
801 		}
802 
803 		/*
804 		 * For changes to existing volumes, we have some additional
805 		 * checks to enforce.
806 		 */
807 		if (type == ZFS_TYPE_VOLUME && zhp != NULL) {
808 			uint64_t volsize = zfs_prop_get_int(zhp,
809 			    ZFS_PROP_VOLSIZE);
810 			uint64_t blocksize = zfs_prop_get_int(zhp,
811 			    ZFS_PROP_VOLBLOCKSIZE);
812 			char buf[64];
813 
814 			switch (prop) {
815 			case ZFS_PROP_RESERVATION:
816 			case ZFS_PROP_REFRESERVATION:
817 				if (intval > volsize) {
818 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
819 					    "'%s' is greater than current "
820 					    "volume size"), propname);
821 					(void) zfs_error(hdl, EZFS_BADPROP,
822 					    errbuf);
823 					goto error;
824 				}
825 				break;
826 
827 			case ZFS_PROP_VOLSIZE:
828 				if (intval % blocksize != 0) {
829 					zfs_nicenum(blocksize, buf,
830 					    sizeof (buf));
831 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
832 					    "'%s' must be a multiple of "
833 					    "volume block size (%s)"),
834 					    propname, buf);
835 					(void) zfs_error(hdl, EZFS_BADPROP,
836 					    errbuf);
837 					goto error;
838 				}
839 
840 				if (intval == 0) {
841 					zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
842 					    "'%s' cannot be zero"),
843 					    propname);
844 					(void) zfs_error(hdl, EZFS_BADPROP,
845 					    errbuf);
846 					goto error;
847 				}
848 				break;
849 			}
850 		}
851 	}
852 
853 	/*
854 	 * If normalization was chosen, but no UTF8 choice was made,
855 	 * enforce rejection of non-UTF8 names.
856 	 *
857 	 * If normalization was chosen, but rejecting non-UTF8 names
858 	 * was explicitly not chosen, it is an error.
859 	 */
860 	if (chosen_normal > 0 && chosen_utf < 0) {
861 		if (nvlist_add_uint64(ret,
862 		    zfs_prop_to_name(ZFS_PROP_UTF8ONLY), 1) != 0) {
863 			(void) no_memory(hdl);
864 			goto error;
865 		}
866 	} else if (chosen_normal > 0 && chosen_utf == 0) {
867 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
868 		    "'%s' must be set 'on' if normalization chosen"),
869 		    zfs_prop_to_name(ZFS_PROP_UTF8ONLY));
870 		(void) zfs_error(hdl, EZFS_BADPROP, errbuf);
871 		goto error;
872 	}
873 
874 	/*
875 	 * If this is an existing volume, and someone is setting the volsize,
876 	 * make sure that it matches the reservation, or add it if necessary.
877 	 */
878 	if (zhp != NULL && type == ZFS_TYPE_VOLUME &&
879 	    nvlist_lookup_uint64(ret, zfs_prop_to_name(ZFS_PROP_VOLSIZE),
880 	    &intval) == 0) {
881 		uint64_t old_volsize = zfs_prop_get_int(zhp,
882 		    ZFS_PROP_VOLSIZE);
883 		uint64_t old_reservation;
884 		uint64_t new_reservation;
885 		zfs_prop_t resv_prop;
886 
887 		if (zfs_which_resv_prop(zhp, &resv_prop) < 0)
888 			goto error;
889 		old_reservation = zfs_prop_get_int(zhp, resv_prop);
890 
891 		if (old_volsize == old_reservation &&
892 		    nvlist_lookup_uint64(ret, zfs_prop_to_name(resv_prop),
893 		    &new_reservation) != 0) {
894 			if (nvlist_add_uint64(ret,
895 			    zfs_prop_to_name(resv_prop), intval) != 0) {
896 				(void) no_memory(hdl);
897 				goto error;
898 			}
899 		}
900 	}
901 	return (ret);
902 
903 error:
904 	nvlist_free(ret);
905 	return (NULL);
906 }
907 
908 static int
909 zfs_get_perm_who(const char *who, zfs_deleg_who_type_t *who_type,
910     uint64_t *ret_who)
911 {
912 	struct passwd *pwd;
913 	struct group *grp;
914 	uid_t id;
915 
916 	if (*who_type == ZFS_DELEG_EVERYONE || *who_type == ZFS_DELEG_CREATE ||
917 	    *who_type == ZFS_DELEG_NAMED_SET) {
918 		*ret_who = -1;
919 		return (0);
920 	}
921 	if (who == NULL && !(*who_type == ZFS_DELEG_EVERYONE))
922 		return (EZFS_BADWHO);
923 
924 	if (*who_type == ZFS_DELEG_WHO_UNKNOWN &&
925 	    strcmp(who, "everyone") == 0) {
926 		*ret_who = -1;
927 		*who_type = ZFS_DELEG_EVERYONE;
928 		return (0);
929 	}
930 
931 	pwd = getpwnam(who);
932 	grp = getgrnam(who);
933 
934 	if ((*who_type == ZFS_DELEG_USER) && pwd) {
935 		*ret_who = pwd->pw_uid;
936 	} else if ((*who_type == ZFS_DELEG_GROUP) && grp) {
937 		*ret_who = grp->gr_gid;
938 	} else if (pwd) {
939 		*ret_who = pwd->pw_uid;
940 		*who_type = ZFS_DELEG_USER;
941 	} else if (grp) {
942 		*ret_who = grp->gr_gid;
943 		*who_type = ZFS_DELEG_GROUP;
944 	} else {
945 		char *end;
946 
947 		id = strtol(who, &end, 10);
948 		if (errno != 0 || *end != '\0') {
949 			return (EZFS_BADWHO);
950 		} else {
951 			*ret_who = id;
952 			if (*who_type == ZFS_DELEG_WHO_UNKNOWN)
953 				*who_type = ZFS_DELEG_USER;
954 		}
955 	}
956 
957 	return (0);
958 }
959 
960 static void
961 zfs_perms_add_to_nvlist(nvlist_t *who_nvp, char *name, nvlist_t *perms_nvp)
962 {
963 	if (perms_nvp != NULL) {
964 		verify(nvlist_add_nvlist(who_nvp,
965 		    name, perms_nvp) == 0);
966 	} else {
967 		verify(nvlist_add_boolean(who_nvp, name) == 0);
968 	}
969 }
970 
971 static void
972 helper(zfs_deleg_who_type_t who_type, uint64_t whoid, char *whostr,
973     zfs_deleg_inherit_t inherit, nvlist_t *who_nvp, nvlist_t *perms_nvp,
974     nvlist_t *sets_nvp)
975 {
976 	boolean_t do_perms, do_sets;
977 	char name[ZFS_MAX_DELEG_NAME];
978 
979 	do_perms = (nvlist_next_nvpair(perms_nvp, NULL) != NULL);
980 	do_sets = (nvlist_next_nvpair(sets_nvp, NULL) != NULL);
981 
982 	if (!do_perms && !do_sets)
983 		do_perms = do_sets = B_TRUE;
984 
985 	if (do_perms) {
986 		zfs_deleg_whokey(name, who_type, inherit,
987 		    (who_type == ZFS_DELEG_NAMED_SET) ?
988 		    whostr : (void *)&whoid);
989 		zfs_perms_add_to_nvlist(who_nvp, name, perms_nvp);
990 	}
991 	if (do_sets) {
992 		zfs_deleg_whokey(name, toupper(who_type), inherit,
993 		    (who_type == ZFS_DELEG_NAMED_SET) ?
994 		    whostr : (void *)&whoid);
995 		zfs_perms_add_to_nvlist(who_nvp, name, sets_nvp);
996 	}
997 }
998 
999 static void
1000 zfs_perms_add_who_nvlist(nvlist_t *who_nvp, uint64_t whoid, void *whostr,
1001     nvlist_t *perms_nvp, nvlist_t *sets_nvp,
1002     zfs_deleg_who_type_t who_type, zfs_deleg_inherit_t inherit)
1003 {
1004 	if (who_type == ZFS_DELEG_NAMED_SET || who_type == ZFS_DELEG_CREATE) {
1005 		helper(who_type, whoid, whostr, 0,
1006 		    who_nvp, perms_nvp, sets_nvp);
1007 	} else {
1008 		if (inherit & ZFS_DELEG_PERM_LOCAL) {
1009 			helper(who_type, whoid, whostr, ZFS_DELEG_LOCAL,
1010 			    who_nvp, perms_nvp, sets_nvp);
1011 		}
1012 		if (inherit & ZFS_DELEG_PERM_DESCENDENT) {
1013 			helper(who_type, whoid, whostr, ZFS_DELEG_DESCENDENT,
1014 			    who_nvp, perms_nvp, sets_nvp);
1015 		}
1016 	}
1017 }
1018 
1019 /*
1020  * Construct nvlist to pass down to kernel for setting/removing permissions.
1021  *
1022  * The nvlist is constructed as a series of nvpairs with an optional embedded
1023  * nvlist of permissions to remove or set.  The topmost nvpairs are the actual
1024  * base attribute named stored in the dsl.
1025  * Arguments:
1026  *
1027  * whostr:   is a comma separated list of users, groups, or a single set name.
1028  *           whostr may be null for everyone or create perms.
1029  * who_type: is the type of entry in whostr.  Typically this will be
1030  *           ZFS_DELEG_WHO_UNKNOWN.
1031  * perms:    common separated list of permissions.  May be null if user
1032  *           is requested to remove permissions by who.
1033  * inherit:  Specifies the inheritance of the permissions.  Will be either
1034  *           ZFS_DELEG_PERM_LOCAL and/or  ZFS_DELEG_PERM_DESCENDENT.
1035  * nvp       The constructed nvlist to pass to zfs_perm_set().
1036  *           The output nvp will look something like this.
1037  *              ul$1234 -> {create ; destroy }
1038  *              Ul$1234 -> { @myset }
1039  *              s-$@myset - { snapshot; checksum; compression }
1040  */
1041 int
1042 zfs_build_perms(zfs_handle_t *zhp, char *whostr, char *perms,
1043     zfs_deleg_who_type_t who_type, zfs_deleg_inherit_t inherit, nvlist_t **nvp)
1044 {
1045 	nvlist_t *who_nvp;
1046 	nvlist_t *perms_nvp = NULL;
1047 	nvlist_t *sets_nvp = NULL;
1048 	char errbuf[1024];
1049 	char *who_tok, *perm;
1050 	int error;
1051 
1052 	*nvp = NULL;
1053 
1054 	if (perms) {
1055 		if ((error = nvlist_alloc(&perms_nvp,
1056 		    NV_UNIQUE_NAME, 0)) != 0) {
1057 			return (1);
1058 		}
1059 		if ((error = nvlist_alloc(&sets_nvp,
1060 		    NV_UNIQUE_NAME, 0)) != 0) {
1061 			nvlist_free(perms_nvp);
1062 			return (1);
1063 		}
1064 	}
1065 
1066 	if ((error = nvlist_alloc(&who_nvp, NV_UNIQUE_NAME, 0)) != 0) {
1067 		if (perms_nvp)
1068 			nvlist_free(perms_nvp);
1069 		if (sets_nvp)
1070 			nvlist_free(sets_nvp);
1071 		return (1);
1072 	}
1073 
1074 	if (who_type == ZFS_DELEG_NAMED_SET) {
1075 		namecheck_err_t why;
1076 		char what;
1077 
1078 		if ((error = permset_namecheck(whostr, &why, &what)) != 0) {
1079 			nvlist_free(who_nvp);
1080 			if (perms_nvp)
1081 				nvlist_free(perms_nvp);
1082 			if (sets_nvp)
1083 				nvlist_free(sets_nvp);
1084 
1085 			switch (why) {
1086 			case NAME_ERR_NO_AT:
1087 				zfs_error_aux(zhp->zfs_hdl,
1088 				    dgettext(TEXT_DOMAIN,
1089 				    "set definition must begin with an '@' "
1090 				    "character"));
1091 			}
1092 			return (zfs_error(zhp->zfs_hdl,
1093 			    EZFS_BADPERMSET, whostr));
1094 		}
1095 	}
1096 
1097 	/*
1098 	 * Build up nvlist(s) of permissions.  Two nvlists are maintained.
1099 	 * The first nvlist perms_nvp will have normal permissions and the
1100 	 * other sets_nvp will have only permssion set names in it.
1101 	 */
1102 	for (perm = strtok(perms, ","); perm; perm = strtok(NULL, ",")) {
1103 		const char *perm_canonical = zfs_deleg_canonicalize_perm(perm);
1104 
1105 		if (perm_canonical) {
1106 			verify(nvlist_add_boolean(perms_nvp,
1107 			    perm_canonical) == 0);
1108 		} else if (perm[0] == '@') {
1109 			verify(nvlist_add_boolean(sets_nvp, perm) == 0);
1110 		} else {
1111 			nvlist_free(who_nvp);
1112 			nvlist_free(perms_nvp);
1113 			nvlist_free(sets_nvp);
1114 			return (zfs_error(zhp->zfs_hdl, EZFS_BADPERM, perm));
1115 		}
1116 	}
1117 
1118 	if (whostr && who_type != ZFS_DELEG_CREATE) {
1119 		who_tok = strtok(whostr, ",");
1120 		if (who_tok == NULL) {
1121 			nvlist_free(who_nvp);
1122 			if (perms_nvp)
1123 				nvlist_free(perms_nvp);
1124 			if (sets_nvp)
1125 				nvlist_free(sets_nvp);
1126 			(void) snprintf(errbuf, sizeof (errbuf),
1127 			    dgettext(TEXT_DOMAIN, "Who string is NULL"),
1128 			    whostr);
1129 			return (zfs_error(zhp->zfs_hdl, EZFS_BADWHO, errbuf));
1130 		}
1131 	}
1132 
1133 	/*
1134 	 * Now create the nvlist(s)
1135 	 */
1136 	do {
1137 		uint64_t who_id;
1138 
1139 		error = zfs_get_perm_who(who_tok, &who_type,
1140 		    &who_id);
1141 		if (error) {
1142 			nvlist_free(who_nvp);
1143 			if (perms_nvp)
1144 				nvlist_free(perms_nvp);
1145 			if (sets_nvp)
1146 				nvlist_free(sets_nvp);
1147 			(void) snprintf(errbuf, sizeof (errbuf),
1148 			    dgettext(TEXT_DOMAIN,
1149 			    "Unable to determine uid/gid for "
1150 			    "%s "), who_tok);
1151 			return (zfs_error(zhp->zfs_hdl, EZFS_BADWHO, errbuf));
1152 		}
1153 
1154 		/*
1155 		 * add entries for both local and descendent when required
1156 		 */
1157 		zfs_perms_add_who_nvlist(who_nvp, who_id, who_tok,
1158 		    perms_nvp, sets_nvp, who_type, inherit);
1159 
1160 	} while (who_tok = strtok(NULL, ","));
1161 	*nvp = who_nvp;
1162 	return (0);
1163 }
1164 
1165 static int
1166 zfs_perm_set_common(zfs_handle_t *zhp, nvlist_t *nvp, boolean_t unset)
1167 {
1168 	zfs_cmd_t zc = { 0 };
1169 	int error;
1170 	char errbuf[1024];
1171 
1172 	(void) snprintf(errbuf, sizeof (errbuf),
1173 	    dgettext(TEXT_DOMAIN, "Cannot update 'allows' for '%s'"),
1174 	    zhp->zfs_name);
1175 
1176 	if (zcmd_write_src_nvlist(zhp->zfs_hdl, &zc, nvp))
1177 		return (-1);
1178 
1179 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1180 	zc.zc_perm_action = unset;
1181 
1182 	error = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SET_FSACL, &zc);
1183 	if (error && errno == ENOTSUP) {
1184 		(void) snprintf(errbuf, sizeof (errbuf),
1185 		    gettext("Pool must be upgraded to use 'allow/unallow'"));
1186 		zcmd_free_nvlists(&zc);
1187 		return (zfs_error(zhp->zfs_hdl, EZFS_BADVERSION, errbuf));
1188 	} else if (error) {
1189 		return (zfs_standard_error(zhp->zfs_hdl, errno, errbuf));
1190 	}
1191 	zcmd_free_nvlists(&zc);
1192 
1193 	return (error);
1194 }
1195 
1196 int
1197 zfs_perm_set(zfs_handle_t *zhp, nvlist_t *nvp)
1198 {
1199 	return (zfs_perm_set_common(zhp, nvp, B_FALSE));
1200 }
1201 
1202 int
1203 zfs_perm_remove(zfs_handle_t *zhp, nvlist_t *perms)
1204 {
1205 	return (zfs_perm_set_common(zhp, perms, B_TRUE));
1206 }
1207 
1208 static int
1209 perm_compare(const void *arg1, const void *arg2)
1210 {
1211 	const zfs_perm_node_t *node1 = arg1;
1212 	const zfs_perm_node_t *node2 = arg2;
1213 	int ret;
1214 
1215 	ret = strcmp(node1->z_pname, node2->z_pname);
1216 
1217 	if (ret > 0)
1218 		return (1);
1219 	if (ret < 0)
1220 		return (-1);
1221 	else
1222 		return (0);
1223 }
1224 
1225 static void
1226 zfs_destroy_perm_tree(avl_tree_t *tree)
1227 {
1228 	zfs_perm_node_t *permnode;
1229 	void *cookie = NULL;
1230 
1231 	while ((permnode = avl_destroy_nodes(tree,  &cookie)) != NULL)
1232 		free(permnode);
1233 	avl_destroy(tree);
1234 }
1235 
1236 static void
1237 zfs_destroy_tree(avl_tree_t *tree)
1238 {
1239 	zfs_allow_node_t *allownode;
1240 	void *cookie = NULL;
1241 
1242 	while ((allownode = avl_destroy_nodes(tree, &cookie)) != NULL) {
1243 		zfs_destroy_perm_tree(&allownode->z_localdescend);
1244 		zfs_destroy_perm_tree(&allownode->z_local);
1245 		zfs_destroy_perm_tree(&allownode->z_descend);
1246 		free(allownode);
1247 	}
1248 	avl_destroy(tree);
1249 }
1250 
1251 void
1252 zfs_free_allows(zfs_allow_t *allow)
1253 {
1254 	zfs_allow_t *allownext;
1255 	zfs_allow_t *freeallow;
1256 
1257 	allownext = allow;
1258 	while (allownext) {
1259 		zfs_destroy_tree(&allownext->z_sets);
1260 		zfs_destroy_tree(&allownext->z_crperms);
1261 		zfs_destroy_tree(&allownext->z_user);
1262 		zfs_destroy_tree(&allownext->z_group);
1263 		zfs_destroy_tree(&allownext->z_everyone);
1264 		freeallow = allownext;
1265 		allownext = allownext->z_next;
1266 		free(freeallow);
1267 	}
1268 }
1269 
1270 static zfs_allow_t *
1271 zfs_alloc_perm_tree(zfs_handle_t *zhp, zfs_allow_t *prev, char *setpoint)
1272 {
1273 	zfs_allow_t *ptree;
1274 
1275 	if ((ptree = zfs_alloc(zhp->zfs_hdl,
1276 	    sizeof (zfs_allow_t))) == NULL) {
1277 		return (NULL);
1278 	}
1279 
1280 	(void) strlcpy(ptree->z_setpoint, setpoint, sizeof (ptree->z_setpoint));
1281 	avl_create(&ptree->z_sets,
1282 	    perm_compare, sizeof (zfs_allow_node_t),
1283 	    offsetof(zfs_allow_node_t, z_node));
1284 	avl_create(&ptree->z_crperms,
1285 	    perm_compare, sizeof (zfs_allow_node_t),
1286 	    offsetof(zfs_allow_node_t, z_node));
1287 	avl_create(&ptree->z_user,
1288 	    perm_compare, sizeof (zfs_allow_node_t),
1289 	    offsetof(zfs_allow_node_t, z_node));
1290 	avl_create(&ptree->z_group,
1291 	    perm_compare, sizeof (zfs_allow_node_t),
1292 	    offsetof(zfs_allow_node_t, z_node));
1293 	avl_create(&ptree->z_everyone,
1294 	    perm_compare, sizeof (zfs_allow_node_t),
1295 	    offsetof(zfs_allow_node_t, z_node));
1296 
1297 	if (prev)
1298 		prev->z_next = ptree;
1299 	ptree->z_next = NULL;
1300 	return (ptree);
1301 }
1302 
1303 /*
1304  * Add permissions to the appropriate AVL permission tree.
1305  * The appropriate tree may not be the requested tree.
1306  * For example if ld indicates a local permission, but
1307  * same permission also exists as a descendent permission
1308  * then the permission will be removed from the descendent
1309  * tree and add the the local+descendent tree.
1310  */
1311 static int
1312 zfs_coalesce_perm(zfs_handle_t *zhp, zfs_allow_node_t *allownode,
1313     char *perm, char ld)
1314 {
1315 	zfs_perm_node_t pnode, *permnode, *permnode2;
1316 	zfs_perm_node_t *newnode;
1317 	avl_index_t where, where2;
1318 	avl_tree_t *tree, *altree;
1319 
1320 	(void) strlcpy(pnode.z_pname, perm, sizeof (pnode.z_pname));
1321 
1322 	if (ld == ZFS_DELEG_NA) {
1323 		tree =  &allownode->z_localdescend;
1324 		altree = &allownode->z_descend;
1325 	} else if (ld == ZFS_DELEG_LOCAL) {
1326 		tree = &allownode->z_local;
1327 		altree = &allownode->z_descend;
1328 	} else {
1329 		tree = &allownode->z_descend;
1330 		altree = &allownode->z_local;
1331 	}
1332 	permnode = avl_find(tree, &pnode, &where);
1333 	permnode2 = avl_find(altree, &pnode, &where2);
1334 
1335 	if (permnode2) {
1336 		avl_remove(altree, permnode2);
1337 		free(permnode2);
1338 		if (permnode == NULL) {
1339 			tree =  &allownode->z_localdescend;
1340 		}
1341 	}
1342 
1343 	/*
1344 	 * Now insert new permission in either requested location
1345 	 * local/descendent or into ld when perm will exist in both.
1346 	 */
1347 	if (permnode == NULL) {
1348 		if ((newnode = zfs_alloc(zhp->zfs_hdl,
1349 		    sizeof (zfs_perm_node_t))) == NULL) {
1350 			return (-1);
1351 		}
1352 		*newnode = pnode;
1353 		avl_add(tree, newnode);
1354 	}
1355 	return (0);
1356 }
1357 
1358 /*
1359  * Uggh, this is going to be a bit complicated.
1360  * we have an nvlist coming out of the kernel that
1361  * will indicate where the permission is set and then
1362  * it will contain allow of the various "who's", and what
1363  * their permissions are.  To further complicate this
1364  * we will then have to coalesce the local,descendent
1365  * and local+descendent permissions where appropriate.
1366  * The kernel only knows about a permission as being local
1367  * or descendent, but not both.
1368  *
1369  * In order to make this easier for zfs_main to deal with
1370  * a series of AVL trees will be used to maintain
1371  * all of this, primarily for sorting purposes as well
1372  * as the ability to quickly locate a specific entry.
1373  *
1374  * What we end up with are tree's for sets, create perms,
1375  * user, groups and everyone.  With each of those trees
1376  * we have subtrees for local, descendent and local+descendent
1377  * permissions.
1378  */
1379 int
1380 zfs_perm_get(zfs_handle_t *zhp, zfs_allow_t **zfs_perms)
1381 {
1382 	zfs_cmd_t zc = { 0 };
1383 	int error;
1384 	nvlist_t *nvlist;
1385 	nvlist_t *permnv, *sourcenv;
1386 	nvpair_t *who_pair, *source_pair;
1387 	nvpair_t *perm_pair;
1388 	char errbuf[1024];
1389 	zfs_allow_t *zallowp, *newallowp;
1390 	char  ld;
1391 	char *nvpname;
1392 	uid_t	uid;
1393 	gid_t	gid;
1394 	avl_tree_t *tree;
1395 	avl_index_t where;
1396 
1397 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1398 
1399 	if (zcmd_alloc_dst_nvlist(zhp->zfs_hdl, &zc, 0) != 0)
1400 		return (-1);
1401 
1402 	while (ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_GET_FSACL, &zc) != 0) {
1403 		if (errno == ENOMEM) {
1404 			if (zcmd_expand_dst_nvlist(zhp->zfs_hdl, &zc) != 0) {
1405 				zcmd_free_nvlists(&zc);
1406 				return (-1);
1407 			}
1408 		} else if (errno == ENOTSUP) {
1409 			zcmd_free_nvlists(&zc);
1410 			(void) snprintf(errbuf, sizeof (errbuf),
1411 			    gettext("Pool must be upgraded to use 'allow'"));
1412 			return (zfs_error(zhp->zfs_hdl,
1413 			    EZFS_BADVERSION, errbuf));
1414 		} else {
1415 			zcmd_free_nvlists(&zc);
1416 			return (-1);
1417 		}
1418 	}
1419 
1420 	if (zcmd_read_dst_nvlist(zhp->zfs_hdl, &zc, &nvlist) != 0) {
1421 		zcmd_free_nvlists(&zc);
1422 		return (-1);
1423 	}
1424 
1425 	zcmd_free_nvlists(&zc);
1426 
1427 	source_pair = nvlist_next_nvpair(nvlist, NULL);
1428 
1429 	if (source_pair == NULL) {
1430 		*zfs_perms = NULL;
1431 		return (0);
1432 	}
1433 
1434 	*zfs_perms = zfs_alloc_perm_tree(zhp, NULL, nvpair_name(source_pair));
1435 	if (*zfs_perms == NULL) {
1436 		return (0);
1437 	}
1438 
1439 	zallowp = *zfs_perms;
1440 
1441 	for (;;) {
1442 		struct passwd *pwd;
1443 		struct group *grp;
1444 		zfs_allow_node_t *allownode;
1445 		zfs_allow_node_t  findallownode;
1446 		zfs_allow_node_t *newallownode;
1447 
1448 		(void) strlcpy(zallowp->z_setpoint,
1449 		    nvpair_name(source_pair),
1450 		    sizeof (zallowp->z_setpoint));
1451 
1452 		if ((error = nvpair_value_nvlist(source_pair, &sourcenv)) != 0)
1453 			goto abort;
1454 
1455 		/*
1456 		 * Make sure nvlist is composed correctly
1457 		 */
1458 		if (zfs_deleg_verify_nvlist(sourcenv)) {
1459 			goto abort;
1460 		}
1461 
1462 		who_pair = nvlist_next_nvpair(sourcenv, NULL);
1463 		if (who_pair == NULL) {
1464 			goto abort;
1465 		}
1466 
1467 		do {
1468 			error = nvpair_value_nvlist(who_pair, &permnv);
1469 			if (error) {
1470 				goto abort;
1471 			}
1472 
1473 			/*
1474 			 * First build up the key to use
1475 			 * for looking up in the various
1476 			 * who trees.
1477 			 */
1478 			ld = nvpair_name(who_pair)[1];
1479 			nvpname = nvpair_name(who_pair);
1480 			switch (nvpair_name(who_pair)[0]) {
1481 			case ZFS_DELEG_USER:
1482 			case ZFS_DELEG_USER_SETS:
1483 				tree = &zallowp->z_user;
1484 				uid = atol(&nvpname[3]);
1485 				pwd = getpwuid(uid);
1486 				(void) snprintf(findallownode.z_key,
1487 				    sizeof (findallownode.z_key), "user %s",
1488 				    (pwd) ? pwd->pw_name :
1489 				    &nvpair_name(who_pair)[3]);
1490 				break;
1491 			case ZFS_DELEG_GROUP:
1492 			case ZFS_DELEG_GROUP_SETS:
1493 				tree = &zallowp->z_group;
1494 				gid = atol(&nvpname[3]);
1495 				grp = getgrgid(gid);
1496 				(void) snprintf(findallownode.z_key,
1497 				    sizeof (findallownode.z_key), "group %s",
1498 				    (grp) ? grp->gr_name :
1499 				    &nvpair_name(who_pair)[3]);
1500 				break;
1501 			case ZFS_DELEG_CREATE:
1502 			case ZFS_DELEG_CREATE_SETS:
1503 				tree = &zallowp->z_crperms;
1504 				(void) strlcpy(findallownode.z_key, "",
1505 				    sizeof (findallownode.z_key));
1506 				break;
1507 			case ZFS_DELEG_EVERYONE:
1508 			case ZFS_DELEG_EVERYONE_SETS:
1509 				(void) snprintf(findallownode.z_key,
1510 				    sizeof (findallownode.z_key), "everyone");
1511 				tree = &zallowp->z_everyone;
1512 				break;
1513 			case ZFS_DELEG_NAMED_SET:
1514 			case ZFS_DELEG_NAMED_SET_SETS:
1515 				(void) snprintf(findallownode.z_key,
1516 				    sizeof (findallownode.z_key), "%s",
1517 				    &nvpair_name(who_pair)[3]);
1518 				tree = &zallowp->z_sets;
1519 				break;
1520 			}
1521 
1522 			/*
1523 			 * Place who in tree
1524 			 */
1525 			allownode = avl_find(tree, &findallownode, &where);
1526 			if (allownode == NULL) {
1527 				if ((newallownode = zfs_alloc(zhp->zfs_hdl,
1528 				    sizeof (zfs_allow_node_t))) == NULL) {
1529 					goto abort;
1530 				}
1531 				avl_create(&newallownode->z_localdescend,
1532 				    perm_compare,
1533 				    sizeof (zfs_perm_node_t),
1534 				    offsetof(zfs_perm_node_t, z_node));
1535 				avl_create(&newallownode->z_local,
1536 				    perm_compare,
1537 				    sizeof (zfs_perm_node_t),
1538 				    offsetof(zfs_perm_node_t, z_node));
1539 				avl_create(&newallownode->z_descend,
1540 				    perm_compare,
1541 				    sizeof (zfs_perm_node_t),
1542 				    offsetof(zfs_perm_node_t, z_node));
1543 				(void) strlcpy(newallownode->z_key,
1544 				    findallownode.z_key,
1545 				    sizeof (findallownode.z_key));
1546 				avl_insert(tree, newallownode, where);
1547 				allownode = newallownode;
1548 			}
1549 
1550 			/*
1551 			 * Now iterate over the permissions and
1552 			 * place them in the appropriate local,
1553 			 * descendent or local+descendent tree.
1554 			 *
1555 			 * The permissions are added to the tree
1556 			 * via zfs_coalesce_perm().
1557 			 */
1558 			perm_pair = nvlist_next_nvpair(permnv, NULL);
1559 			if (perm_pair == NULL)
1560 				goto abort;
1561 			do {
1562 				if (zfs_coalesce_perm(zhp, allownode,
1563 				    nvpair_name(perm_pair), ld) != 0)
1564 					goto abort;
1565 			} while (perm_pair = nvlist_next_nvpair(permnv,
1566 			    perm_pair));
1567 		} while (who_pair = nvlist_next_nvpair(sourcenv, who_pair));
1568 
1569 		source_pair = nvlist_next_nvpair(nvlist, source_pair);
1570 		if (source_pair == NULL)
1571 			break;
1572 
1573 		/*
1574 		 * allocate another node from the link list of
1575 		 * zfs_allow_t structures
1576 		 */
1577 		newallowp = zfs_alloc_perm_tree(zhp, zallowp,
1578 		    nvpair_name(source_pair));
1579 		if (newallowp == NULL) {
1580 			goto abort;
1581 		}
1582 		zallowp = newallowp;
1583 	}
1584 	nvlist_free(nvlist);
1585 	return (0);
1586 abort:
1587 	zfs_free_allows(*zfs_perms);
1588 	nvlist_free(nvlist);
1589 	return (-1);
1590 }
1591 
1592 /*
1593  * Given a property name and value, set the property for the given dataset.
1594  */
1595 int
1596 zfs_prop_set(zfs_handle_t *zhp, const char *propname, const char *propval)
1597 {
1598 	zfs_cmd_t zc = { 0 };
1599 	int ret = -1;
1600 	prop_changelist_t *cl = NULL;
1601 	char errbuf[1024];
1602 	libzfs_handle_t *hdl = zhp->zfs_hdl;
1603 	nvlist_t *nvl = NULL, *realprops;
1604 	zfs_prop_t prop;
1605 
1606 	(void) snprintf(errbuf, sizeof (errbuf),
1607 	    dgettext(TEXT_DOMAIN, "cannot set property for '%s'"),
1608 	    zhp->zfs_name);
1609 
1610 	if (nvlist_alloc(&nvl, NV_UNIQUE_NAME, 0) != 0 ||
1611 	    nvlist_add_string(nvl, propname, propval) != 0) {
1612 		(void) no_memory(hdl);
1613 		goto error;
1614 	}
1615 
1616 	if ((realprops = zfs_validate_properties(hdl, zhp->zfs_type, nvl,
1617 	    zfs_prop_get_int(zhp, ZFS_PROP_ZONED), zhp, errbuf)) == NULL)
1618 		goto error;
1619 
1620 	nvlist_free(nvl);
1621 	nvl = realprops;
1622 
1623 	prop = zfs_name_to_prop(propname);
1624 
1625 	if ((cl = changelist_gather(zhp, prop, 0)) == NULL)
1626 		goto error;
1627 
1628 	if (prop == ZFS_PROP_MOUNTPOINT && changelist_haszonedchild(cl)) {
1629 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1630 		    "child dataset with inherited mountpoint is used "
1631 		    "in a non-global zone"));
1632 		ret = zfs_error(hdl, EZFS_ZONED, errbuf);
1633 		goto error;
1634 	}
1635 
1636 	if ((ret = changelist_prefix(cl)) != 0)
1637 		goto error;
1638 
1639 	/*
1640 	 * Execute the corresponding ioctl() to set this property.
1641 	 */
1642 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1643 
1644 	if (zcmd_write_src_nvlist(hdl, &zc, nvl) != 0)
1645 		goto error;
1646 
1647 	ret = zfs_ioctl(hdl, ZFS_IOC_SET_PROP, &zc);
1648 	if (ret != 0) {
1649 		switch (errno) {
1650 
1651 		case ENOSPC:
1652 			/*
1653 			 * For quotas and reservations, ENOSPC indicates
1654 			 * something different; setting a quota or reservation
1655 			 * doesn't use any disk space.
1656 			 */
1657 			switch (prop) {
1658 			case ZFS_PROP_QUOTA:
1659 			case ZFS_PROP_REFQUOTA:
1660 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1661 				    "size is less than current used or "
1662 				    "reserved space"));
1663 				(void) zfs_error(hdl, EZFS_PROPSPACE, errbuf);
1664 				break;
1665 
1666 			case ZFS_PROP_RESERVATION:
1667 			case ZFS_PROP_REFRESERVATION:
1668 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1669 				    "size is greater than available space"));
1670 				(void) zfs_error(hdl, EZFS_PROPSPACE, errbuf);
1671 				break;
1672 
1673 			default:
1674 				(void) zfs_standard_error(hdl, errno, errbuf);
1675 				break;
1676 			}
1677 			break;
1678 
1679 		case EBUSY:
1680 			if (prop == ZFS_PROP_VOLBLOCKSIZE)
1681 				(void) zfs_error(hdl, EZFS_VOLHASDATA, errbuf);
1682 			else
1683 				(void) zfs_standard_error(hdl, EBUSY, errbuf);
1684 			break;
1685 
1686 		case EROFS:
1687 			(void) zfs_error(hdl, EZFS_DSREADONLY, errbuf);
1688 			break;
1689 
1690 		case ENOTSUP:
1691 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1692 			    "pool must be upgraded to set this "
1693 			    "property or value"));
1694 			(void) zfs_error(hdl, EZFS_BADVERSION, errbuf);
1695 			break;
1696 
1697 		case EOVERFLOW:
1698 			/*
1699 			 * This platform can't address a volume this big.
1700 			 */
1701 #ifdef _ILP32
1702 			if (prop == ZFS_PROP_VOLSIZE) {
1703 				(void) zfs_error(hdl, EZFS_VOLTOOBIG, errbuf);
1704 				break;
1705 			}
1706 #endif
1707 			/* FALLTHROUGH */
1708 		default:
1709 			(void) zfs_standard_error(hdl, errno, errbuf);
1710 		}
1711 	} else {
1712 		/*
1713 		 * Refresh the statistics so the new property value
1714 		 * is reflected.
1715 		 */
1716 		if ((ret = changelist_postfix(cl)) == 0)
1717 			(void) get_stats(zhp);
1718 	}
1719 
1720 error:
1721 	nvlist_free(nvl);
1722 	zcmd_free_nvlists(&zc);
1723 	if (cl)
1724 		changelist_free(cl);
1725 	return (ret);
1726 }
1727 
1728 /*
1729  * Given a property, inherit the value from the parent dataset.
1730  */
1731 int
1732 zfs_prop_inherit(zfs_handle_t *zhp, const char *propname)
1733 {
1734 	zfs_cmd_t zc = { 0 };
1735 	int ret;
1736 	prop_changelist_t *cl;
1737 	libzfs_handle_t *hdl = zhp->zfs_hdl;
1738 	char errbuf[1024];
1739 	zfs_prop_t prop;
1740 
1741 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
1742 	    "cannot inherit %s for '%s'"), propname, zhp->zfs_name);
1743 
1744 	if ((prop = zfs_name_to_prop(propname)) == ZPROP_INVAL) {
1745 		/*
1746 		 * For user properties, the amount of work we have to do is very
1747 		 * small, so just do it here.
1748 		 */
1749 		if (!zfs_prop_user(propname)) {
1750 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1751 			    "invalid property"));
1752 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
1753 		}
1754 
1755 		(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1756 		(void) strlcpy(zc.zc_value, propname, sizeof (zc.zc_value));
1757 
1758 		if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_INHERIT_PROP, &zc) != 0)
1759 			return (zfs_standard_error(hdl, errno, errbuf));
1760 
1761 		return (0);
1762 	}
1763 
1764 	/*
1765 	 * Verify that this property is inheritable.
1766 	 */
1767 	if (zfs_prop_readonly(prop))
1768 		return (zfs_error(hdl, EZFS_PROPREADONLY, errbuf));
1769 
1770 	if (!zfs_prop_inheritable(prop))
1771 		return (zfs_error(hdl, EZFS_PROPNONINHERIT, errbuf));
1772 
1773 	/*
1774 	 * Check to see if the value applies to this type
1775 	 */
1776 	if (!zfs_prop_valid_for_type(prop, zhp->zfs_type))
1777 		return (zfs_error(hdl, EZFS_PROPTYPE, errbuf));
1778 
1779 	/*
1780 	 * Normalize the name, to get rid of shorthand abbrevations.
1781 	 */
1782 	propname = zfs_prop_to_name(prop);
1783 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
1784 	(void) strlcpy(zc.zc_value, propname, sizeof (zc.zc_value));
1785 
1786 	if (prop == ZFS_PROP_MOUNTPOINT && getzoneid() == GLOBAL_ZONEID &&
1787 	    zfs_prop_get_int(zhp, ZFS_PROP_ZONED)) {
1788 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1789 		    "dataset is used in a non-global zone"));
1790 		return (zfs_error(hdl, EZFS_ZONED, errbuf));
1791 	}
1792 
1793 	/*
1794 	 * Determine datasets which will be affected by this change, if any.
1795 	 */
1796 	if ((cl = changelist_gather(zhp, prop, 0)) == NULL)
1797 		return (-1);
1798 
1799 	if (prop == ZFS_PROP_MOUNTPOINT && changelist_haszonedchild(cl)) {
1800 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
1801 		    "child dataset with inherited mountpoint is used "
1802 		    "in a non-global zone"));
1803 		ret = zfs_error(hdl, EZFS_ZONED, errbuf);
1804 		goto error;
1805 	}
1806 
1807 	if ((ret = changelist_prefix(cl)) != 0)
1808 		goto error;
1809 
1810 	if ((ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_INHERIT_PROP, &zc)) != 0) {
1811 		return (zfs_standard_error(hdl, errno, errbuf));
1812 	} else {
1813 
1814 		if ((ret = changelist_postfix(cl)) != 0)
1815 			goto error;
1816 
1817 		/*
1818 		 * Refresh the statistics so the new property is reflected.
1819 		 */
1820 		(void) get_stats(zhp);
1821 	}
1822 
1823 error:
1824 	changelist_free(cl);
1825 	return (ret);
1826 }
1827 
1828 /*
1829  * True DSL properties are stored in an nvlist.  The following two functions
1830  * extract them appropriately.
1831  */
1832 static uint64_t
1833 getprop_uint64(zfs_handle_t *zhp, zfs_prop_t prop, char **source)
1834 {
1835 	nvlist_t *nv;
1836 	uint64_t value;
1837 
1838 	*source = NULL;
1839 	if (nvlist_lookup_nvlist(zhp->zfs_props,
1840 	    zfs_prop_to_name(prop), &nv) == 0) {
1841 		verify(nvlist_lookup_uint64(nv, ZPROP_VALUE, &value) == 0);
1842 		(void) nvlist_lookup_string(nv, ZPROP_SOURCE, source);
1843 	} else {
1844 		value = zfs_prop_default_numeric(prop);
1845 		*source = "";
1846 	}
1847 
1848 	return (value);
1849 }
1850 
1851 static char *
1852 getprop_string(zfs_handle_t *zhp, zfs_prop_t prop, char **source)
1853 {
1854 	nvlist_t *nv;
1855 	char *value;
1856 
1857 	*source = NULL;
1858 	if (nvlist_lookup_nvlist(zhp->zfs_props,
1859 	    zfs_prop_to_name(prop), &nv) == 0) {
1860 		verify(nvlist_lookup_string(nv, ZPROP_VALUE, &value) == 0);
1861 		(void) nvlist_lookup_string(nv, ZPROP_SOURCE, source);
1862 	} else {
1863 		if ((value = (char *)zfs_prop_default_string(prop)) == NULL)
1864 			value = "";
1865 		*source = "";
1866 	}
1867 
1868 	return (value);
1869 }
1870 
1871 /*
1872  * Internal function for getting a numeric property.  Both zfs_prop_get() and
1873  * zfs_prop_get_int() are built using this interface.
1874  *
1875  * Certain properties can be overridden using 'mount -o'.  In this case, scan
1876  * the contents of the /etc/mnttab entry, searching for the appropriate options.
1877  * If they differ from the on-disk values, report the current values and mark
1878  * the source "temporary".
1879  */
1880 static int
1881 get_numeric_property(zfs_handle_t *zhp, zfs_prop_t prop, zprop_source_t *src,
1882     char **source, uint64_t *val)
1883 {
1884 	zfs_cmd_t zc = { 0 };
1885 	nvlist_t *zplprops = NULL;
1886 	struct mnttab mnt;
1887 	char *mntopt_on = NULL;
1888 	char *mntopt_off = NULL;
1889 
1890 	*source = NULL;
1891 
1892 	switch (prop) {
1893 	case ZFS_PROP_ATIME:
1894 		mntopt_on = MNTOPT_ATIME;
1895 		mntopt_off = MNTOPT_NOATIME;
1896 		break;
1897 
1898 	case ZFS_PROP_DEVICES:
1899 		mntopt_on = MNTOPT_DEVICES;
1900 		mntopt_off = MNTOPT_NODEVICES;
1901 		break;
1902 
1903 	case ZFS_PROP_EXEC:
1904 		mntopt_on = MNTOPT_EXEC;
1905 		mntopt_off = MNTOPT_NOEXEC;
1906 		break;
1907 
1908 	case ZFS_PROP_READONLY:
1909 		mntopt_on = MNTOPT_RO;
1910 		mntopt_off = MNTOPT_RW;
1911 		break;
1912 
1913 	case ZFS_PROP_SETUID:
1914 		mntopt_on = MNTOPT_SETUID;
1915 		mntopt_off = MNTOPT_NOSETUID;
1916 		break;
1917 
1918 	case ZFS_PROP_XATTR:
1919 		mntopt_on = MNTOPT_XATTR;
1920 		mntopt_off = MNTOPT_NOXATTR;
1921 		break;
1922 
1923 	case ZFS_PROP_NBMAND:
1924 		mntopt_on = MNTOPT_NBMAND;
1925 		mntopt_off = MNTOPT_NONBMAND;
1926 		break;
1927 	}
1928 
1929 	/*
1930 	 * Because looking up the mount options is potentially expensive
1931 	 * (iterating over all of /etc/mnttab), we defer its calculation until
1932 	 * we're looking up a property which requires its presence.
1933 	 */
1934 	if (!zhp->zfs_mntcheck &&
1935 	    (mntopt_on != NULL || prop == ZFS_PROP_MOUNTED)) {
1936 		struct mnttab entry, search = { 0 };
1937 		FILE *mnttab = zhp->zfs_hdl->libzfs_mnttab;
1938 
1939 		search.mnt_special = (char *)zhp->zfs_name;
1940 		search.mnt_fstype = MNTTYPE_ZFS;
1941 		rewind(mnttab);
1942 
1943 		if (getmntany(mnttab, &entry, &search) == 0) {
1944 			zhp->zfs_mntopts = zfs_strdup(zhp->zfs_hdl,
1945 			    entry.mnt_mntopts);
1946 			if (zhp->zfs_mntopts == NULL)
1947 				return (-1);
1948 		}
1949 
1950 		zhp->zfs_mntcheck = B_TRUE;
1951 	}
1952 
1953 	if (zhp->zfs_mntopts == NULL)
1954 		mnt.mnt_mntopts = "";
1955 	else
1956 		mnt.mnt_mntopts = zhp->zfs_mntopts;
1957 
1958 	switch (prop) {
1959 	case ZFS_PROP_ATIME:
1960 	case ZFS_PROP_DEVICES:
1961 	case ZFS_PROP_EXEC:
1962 	case ZFS_PROP_READONLY:
1963 	case ZFS_PROP_SETUID:
1964 	case ZFS_PROP_XATTR:
1965 	case ZFS_PROP_NBMAND:
1966 		*val = getprop_uint64(zhp, prop, source);
1967 
1968 		if (hasmntopt(&mnt, mntopt_on) && !*val) {
1969 			*val = B_TRUE;
1970 			if (src)
1971 				*src = ZPROP_SRC_TEMPORARY;
1972 		} else if (hasmntopt(&mnt, mntopt_off) && *val) {
1973 			*val = B_FALSE;
1974 			if (src)
1975 				*src = ZPROP_SRC_TEMPORARY;
1976 		}
1977 		break;
1978 
1979 	case ZFS_PROP_CANMOUNT:
1980 		*val = getprop_uint64(zhp, prop, source);
1981 		if (*val == 0)
1982 			*source = zhp->zfs_name;
1983 		else
1984 			*source = "";	/* default */
1985 		break;
1986 
1987 	case ZFS_PROP_QUOTA:
1988 	case ZFS_PROP_REFQUOTA:
1989 	case ZFS_PROP_RESERVATION:
1990 	case ZFS_PROP_REFRESERVATION:
1991 		*val = getprop_uint64(zhp, prop, source);
1992 		if (*val == 0)
1993 			*source = "";	/* default */
1994 		else
1995 			*source = zhp->zfs_name;
1996 		break;
1997 
1998 	case ZFS_PROP_MOUNTED:
1999 		*val = (zhp->zfs_mntopts != NULL);
2000 		break;
2001 
2002 	case ZFS_PROP_NUMCLONES:
2003 		*val = zhp->zfs_dmustats.dds_num_clones;
2004 		break;
2005 
2006 	case ZFS_PROP_VERSION:
2007 	case ZFS_PROP_NORMALIZE:
2008 	case ZFS_PROP_UTF8ONLY:
2009 	case ZFS_PROP_CASE:
2010 		if (!zfs_prop_valid_for_type(prop, zhp->zfs_head_type) ||
2011 		    zcmd_alloc_dst_nvlist(zhp->zfs_hdl, &zc, 0) != 0)
2012 			return (-1);
2013 		(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
2014 		if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_OBJSET_ZPLPROPS, &zc)) {
2015 			zcmd_free_nvlists(&zc);
2016 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
2017 			    "unable to get %s property"),
2018 			    zfs_prop_to_name(prop));
2019 			return (zfs_error(zhp->zfs_hdl, EZFS_BADVERSION,
2020 			    dgettext(TEXT_DOMAIN, "internal error")));
2021 		}
2022 		if (zcmd_read_dst_nvlist(zhp->zfs_hdl, &zc, &zplprops) != 0 ||
2023 		    nvlist_lookup_uint64(zplprops, zfs_prop_to_name(prop),
2024 		    val) != 0) {
2025 			zcmd_free_nvlists(&zc);
2026 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
2027 			    "unable to get %s property"),
2028 			    zfs_prop_to_name(prop));
2029 			return (zfs_error(zhp->zfs_hdl, EZFS_NOMEM,
2030 			    dgettext(TEXT_DOMAIN, "internal error")));
2031 		}
2032 		if (zplprops)
2033 			nvlist_free(zplprops);
2034 		zcmd_free_nvlists(&zc);
2035 		break;
2036 
2037 	default:
2038 		switch (zfs_prop_get_type(prop)) {
2039 		case PROP_TYPE_NUMBER:
2040 		case PROP_TYPE_INDEX:
2041 			*val = getprop_uint64(zhp, prop, source);
2042 			break;
2043 
2044 		case PROP_TYPE_STRING:
2045 		default:
2046 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
2047 			    "cannot get non-numeric property"));
2048 			return (zfs_error(zhp->zfs_hdl, EZFS_BADPROP,
2049 			    dgettext(TEXT_DOMAIN, "internal error")));
2050 		}
2051 	}
2052 
2053 	return (0);
2054 }
2055 
2056 /*
2057  * Calculate the source type, given the raw source string.
2058  */
2059 static void
2060 get_source(zfs_handle_t *zhp, zprop_source_t *srctype, char *source,
2061     char *statbuf, size_t statlen)
2062 {
2063 	if (statbuf == NULL || *srctype == ZPROP_SRC_TEMPORARY)
2064 		return;
2065 
2066 	if (source == NULL) {
2067 		*srctype = ZPROP_SRC_NONE;
2068 	} else if (source[0] == '\0') {
2069 		*srctype = ZPROP_SRC_DEFAULT;
2070 	} else {
2071 		if (strcmp(source, zhp->zfs_name) == 0) {
2072 			*srctype = ZPROP_SRC_LOCAL;
2073 		} else {
2074 			(void) strlcpy(statbuf, source, statlen);
2075 			*srctype = ZPROP_SRC_INHERITED;
2076 		}
2077 	}
2078 
2079 }
2080 
2081 /*
2082  * Retrieve a property from the given object.  If 'literal' is specified, then
2083  * numbers are left as exact values.  Otherwise, numbers are converted to a
2084  * human-readable form.
2085  *
2086  * Returns 0 on success, or -1 on error.
2087  */
2088 int
2089 zfs_prop_get(zfs_handle_t *zhp, zfs_prop_t prop, char *propbuf, size_t proplen,
2090     zprop_source_t *src, char *statbuf, size_t statlen, boolean_t literal)
2091 {
2092 	char *source = NULL;
2093 	uint64_t val;
2094 	char *str;
2095 	const char *root;
2096 	const char *strval;
2097 
2098 	/*
2099 	 * Check to see if this property applies to our object
2100 	 */
2101 	if (!zfs_prop_valid_for_type(prop, zhp->zfs_type))
2102 		return (-1);
2103 
2104 	if (src)
2105 		*src = ZPROP_SRC_NONE;
2106 
2107 	switch (prop) {
2108 	case ZFS_PROP_CREATION:
2109 		/*
2110 		 * 'creation' is a time_t stored in the statistics.  We convert
2111 		 * this into a string unless 'literal' is specified.
2112 		 */
2113 		{
2114 			val = getprop_uint64(zhp, prop, &source);
2115 			time_t time = (time_t)val;
2116 			struct tm t;
2117 
2118 			if (literal ||
2119 			    localtime_r(&time, &t) == NULL ||
2120 			    strftime(propbuf, proplen, "%a %b %e %k:%M %Y",
2121 			    &t) == 0)
2122 				(void) snprintf(propbuf, proplen, "%llu", val);
2123 		}
2124 		break;
2125 
2126 	case ZFS_PROP_MOUNTPOINT:
2127 		/*
2128 		 * Getting the precise mountpoint can be tricky.
2129 		 *
2130 		 *  - for 'none' or 'legacy', return those values.
2131 		 *  - for default mountpoints, construct it as /zfs/<dataset>
2132 		 *  - for inherited mountpoints, we want to take everything
2133 		 *    after our ancestor and append it to the inherited value.
2134 		 *
2135 		 * If the pool has an alternate root, we want to prepend that
2136 		 * root to any values we return.
2137 		 */
2138 		root = zhp->zfs_root;
2139 		str = getprop_string(zhp, prop, &source);
2140 
2141 		if (str[0] == '\0') {
2142 			(void) snprintf(propbuf, proplen, "%s/zfs/%s",
2143 			    root, zhp->zfs_name);
2144 		} else if (str[0] == '/') {
2145 			const char *relpath = zhp->zfs_name + strlen(source);
2146 
2147 			if (relpath[0] == '/')
2148 				relpath++;
2149 			if (str[1] == '\0')
2150 				str++;
2151 
2152 			if (relpath[0] == '\0')
2153 				(void) snprintf(propbuf, proplen, "%s%s",
2154 				    root, str);
2155 			else
2156 				(void) snprintf(propbuf, proplen, "%s%s%s%s",
2157 				    root, str, relpath[0] == '@' ? "" : "/",
2158 				    relpath);
2159 		} else {
2160 			/* 'legacy' or 'none' */
2161 			(void) strlcpy(propbuf, str, proplen);
2162 		}
2163 
2164 		break;
2165 
2166 	case ZFS_PROP_ORIGIN:
2167 		(void) strlcpy(propbuf, getprop_string(zhp, prop, &source),
2168 		    proplen);
2169 		/*
2170 		 * If there is no parent at all, return failure to indicate that
2171 		 * it doesn't apply to this dataset.
2172 		 */
2173 		if (propbuf[0] == '\0')
2174 			return (-1);
2175 		break;
2176 
2177 	case ZFS_PROP_QUOTA:
2178 	case ZFS_PROP_REFQUOTA:
2179 	case ZFS_PROP_RESERVATION:
2180 	case ZFS_PROP_REFRESERVATION:
2181 
2182 		if (get_numeric_property(zhp, prop, src, &source, &val) != 0)
2183 			return (-1);
2184 
2185 		/*
2186 		 * If quota or reservation is 0, we translate this into 'none'
2187 		 * (unless literal is set), and indicate that it's the default
2188 		 * value.  Otherwise, we print the number nicely and indicate
2189 		 * that its set locally.
2190 		 */
2191 		if (val == 0) {
2192 			if (literal)
2193 				(void) strlcpy(propbuf, "0", proplen);
2194 			else
2195 				(void) strlcpy(propbuf, "none", proplen);
2196 		} else {
2197 			if (literal)
2198 				(void) snprintf(propbuf, proplen, "%llu",
2199 				    (u_longlong_t)val);
2200 			else
2201 				zfs_nicenum(val, propbuf, proplen);
2202 		}
2203 		break;
2204 
2205 	case ZFS_PROP_COMPRESSRATIO:
2206 		if (get_numeric_property(zhp, prop, src, &source, &val) != 0)
2207 			return (-1);
2208 		(void) snprintf(propbuf, proplen, "%lld.%02lldx", (longlong_t)
2209 		    val / 100, (longlong_t)val % 100);
2210 		break;
2211 
2212 	case ZFS_PROP_TYPE:
2213 		switch (zhp->zfs_type) {
2214 		case ZFS_TYPE_FILESYSTEM:
2215 			str = "filesystem";
2216 			break;
2217 		case ZFS_TYPE_VOLUME:
2218 			str = "volume";
2219 			break;
2220 		case ZFS_TYPE_SNAPSHOT:
2221 			str = "snapshot";
2222 			break;
2223 		default:
2224 			abort();
2225 		}
2226 		(void) snprintf(propbuf, proplen, "%s", str);
2227 		break;
2228 
2229 	case ZFS_PROP_MOUNTED:
2230 		/*
2231 		 * The 'mounted' property is a pseudo-property that described
2232 		 * whether the filesystem is currently mounted.  Even though
2233 		 * it's a boolean value, the typical values of "on" and "off"
2234 		 * don't make sense, so we translate to "yes" and "no".
2235 		 */
2236 		if (get_numeric_property(zhp, ZFS_PROP_MOUNTED,
2237 		    src, &source, &val) != 0)
2238 			return (-1);
2239 		if (val)
2240 			(void) strlcpy(propbuf, "yes", proplen);
2241 		else
2242 			(void) strlcpy(propbuf, "no", proplen);
2243 		break;
2244 
2245 	case ZFS_PROP_NAME:
2246 		/*
2247 		 * The 'name' property is a pseudo-property derived from the
2248 		 * dataset name.  It is presented as a real property to simplify
2249 		 * consumers.
2250 		 */
2251 		(void) strlcpy(propbuf, zhp->zfs_name, proplen);
2252 		break;
2253 
2254 	default:
2255 		switch (zfs_prop_get_type(prop)) {
2256 		case PROP_TYPE_NUMBER:
2257 			if (get_numeric_property(zhp, prop, src,
2258 			    &source, &val) != 0)
2259 				return (-1);
2260 			if (literal)
2261 				(void) snprintf(propbuf, proplen, "%llu",
2262 				    (u_longlong_t)val);
2263 			else
2264 				zfs_nicenum(val, propbuf, proplen);
2265 			break;
2266 
2267 		case PROP_TYPE_STRING:
2268 			(void) strlcpy(propbuf,
2269 			    getprop_string(zhp, prop, &source), proplen);
2270 			break;
2271 
2272 		case PROP_TYPE_INDEX:
2273 			if (get_numeric_property(zhp, prop, src,
2274 			    &source, &val) != 0)
2275 				return (-1);
2276 			if (zfs_prop_index_to_string(prop, val, &strval) != 0)
2277 				return (-1);
2278 			(void) strlcpy(propbuf, strval, proplen);
2279 			break;
2280 
2281 		default:
2282 			abort();
2283 		}
2284 	}
2285 
2286 	get_source(zhp, src, source, statbuf, statlen);
2287 
2288 	return (0);
2289 }
2290 
2291 /*
2292  * Utility function to get the given numeric property.  Does no validation that
2293  * the given property is the appropriate type; should only be used with
2294  * hard-coded property types.
2295  */
2296 uint64_t
2297 zfs_prop_get_int(zfs_handle_t *zhp, zfs_prop_t prop)
2298 {
2299 	char *source;
2300 	uint64_t val;
2301 
2302 	(void) get_numeric_property(zhp, prop, NULL, &source, &val);
2303 
2304 	return (val);
2305 }
2306 
2307 int
2308 zfs_prop_set_int(zfs_handle_t *zhp, zfs_prop_t prop, uint64_t val)
2309 {
2310 	char buf[64];
2311 
2312 	zfs_nicenum(val, buf, sizeof (buf));
2313 	return (zfs_prop_set(zhp, zfs_prop_to_name(prop), buf));
2314 }
2315 
2316 /*
2317  * Similar to zfs_prop_get(), but returns the value as an integer.
2318  */
2319 int
2320 zfs_prop_get_numeric(zfs_handle_t *zhp, zfs_prop_t prop, uint64_t *value,
2321     zprop_source_t *src, char *statbuf, size_t statlen)
2322 {
2323 	char *source;
2324 
2325 	/*
2326 	 * Check to see if this property applies to our object
2327 	 */
2328 	if (!zfs_prop_valid_for_type(prop, zhp->zfs_type)) {
2329 		return (zfs_error_fmt(zhp->zfs_hdl, EZFS_PROPTYPE,
2330 		    dgettext(TEXT_DOMAIN, "cannot get property '%s'"),
2331 		    zfs_prop_to_name(prop)));
2332 	}
2333 
2334 	if (src)
2335 		*src = ZPROP_SRC_NONE;
2336 
2337 	if (get_numeric_property(zhp, prop, src, &source, value) != 0)
2338 		return (-1);
2339 
2340 	get_source(zhp, src, source, statbuf, statlen);
2341 
2342 	return (0);
2343 }
2344 
2345 /*
2346  * Returns the name of the given zfs handle.
2347  */
2348 const char *
2349 zfs_get_name(const zfs_handle_t *zhp)
2350 {
2351 	return (zhp->zfs_name);
2352 }
2353 
2354 /*
2355  * Returns the type of the given zfs handle.
2356  */
2357 zfs_type_t
2358 zfs_get_type(const zfs_handle_t *zhp)
2359 {
2360 	return (zhp->zfs_type);
2361 }
2362 
2363 /*
2364  * Iterate over all child filesystems
2365  */
2366 int
2367 zfs_iter_filesystems(zfs_handle_t *zhp, zfs_iter_f func, void *data)
2368 {
2369 	zfs_cmd_t zc = { 0 };
2370 	zfs_handle_t *nzhp;
2371 	int ret;
2372 
2373 	if (zhp->zfs_type != ZFS_TYPE_FILESYSTEM)
2374 		return (0);
2375 
2376 	for ((void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
2377 	    ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_DATASET_LIST_NEXT, &zc) == 0;
2378 	    (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name))) {
2379 		/*
2380 		 * Ignore private dataset names.
2381 		 */
2382 		if (dataset_name_hidden(zc.zc_name))
2383 			continue;
2384 
2385 		/*
2386 		 * Silently ignore errors, as the only plausible explanation is
2387 		 * that the pool has since been removed.
2388 		 */
2389 		if ((nzhp = make_dataset_handle(zhp->zfs_hdl,
2390 		    zc.zc_name)) == NULL)
2391 			continue;
2392 
2393 		if ((ret = func(nzhp, data)) != 0)
2394 			return (ret);
2395 	}
2396 
2397 	/*
2398 	 * An errno value of ESRCH indicates normal completion.  If ENOENT is
2399 	 * returned, then the underlying dataset has been removed since we
2400 	 * obtained the handle.
2401 	 */
2402 	if (errno != ESRCH && errno != ENOENT)
2403 		return (zfs_standard_error(zhp->zfs_hdl, errno,
2404 		    dgettext(TEXT_DOMAIN, "cannot iterate filesystems")));
2405 
2406 	return (0);
2407 }
2408 
2409 /*
2410  * Iterate over all snapshots
2411  */
2412 int
2413 zfs_iter_snapshots(zfs_handle_t *zhp, zfs_iter_f func, void *data)
2414 {
2415 	zfs_cmd_t zc = { 0 };
2416 	zfs_handle_t *nzhp;
2417 	int ret;
2418 
2419 	if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT)
2420 		return (0);
2421 
2422 	for ((void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
2423 	    ioctl(zhp->zfs_hdl->libzfs_fd, ZFS_IOC_SNAPSHOT_LIST_NEXT,
2424 	    &zc) == 0;
2425 	    (void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name))) {
2426 
2427 		if ((nzhp = make_dataset_handle(zhp->zfs_hdl,
2428 		    zc.zc_name)) == NULL)
2429 			continue;
2430 
2431 		if ((ret = func(nzhp, data)) != 0)
2432 			return (ret);
2433 	}
2434 
2435 	/*
2436 	 * An errno value of ESRCH indicates normal completion.  If ENOENT is
2437 	 * returned, then the underlying dataset has been removed since we
2438 	 * obtained the handle.  Silently ignore this case, and return success.
2439 	 */
2440 	if (errno != ESRCH && errno != ENOENT)
2441 		return (zfs_standard_error(zhp->zfs_hdl, errno,
2442 		    dgettext(TEXT_DOMAIN, "cannot iterate filesystems")));
2443 
2444 	return (0);
2445 }
2446 
2447 /*
2448  * Iterate over all children, snapshots and filesystems
2449  */
2450 int
2451 zfs_iter_children(zfs_handle_t *zhp, zfs_iter_f func, void *data)
2452 {
2453 	int ret;
2454 
2455 	if ((ret = zfs_iter_filesystems(zhp, func, data)) != 0)
2456 		return (ret);
2457 
2458 	return (zfs_iter_snapshots(zhp, func, data));
2459 }
2460 
2461 /*
2462  * Given a complete name, return just the portion that refers to the parent.
2463  * Can return NULL if this is a pool.
2464  */
2465 static int
2466 parent_name(const char *path, char *buf, size_t buflen)
2467 {
2468 	char *loc;
2469 
2470 	if ((loc = strrchr(path, '/')) == NULL)
2471 		return (-1);
2472 
2473 	(void) strncpy(buf, path, MIN(buflen, loc - path));
2474 	buf[loc - path] = '\0';
2475 
2476 	return (0);
2477 }
2478 
2479 /*
2480  * If accept_ancestor is false, then check to make sure that the given path has
2481  * a parent, and that it exists.  If accept_ancestor is true, then find the
2482  * closest existing ancestor for the given path.  In prefixlen return the
2483  * length of already existing prefix of the given path.  We also fetch the
2484  * 'zoned' property, which is used to validate property settings when creating
2485  * new datasets.
2486  */
2487 static int
2488 check_parents(libzfs_handle_t *hdl, const char *path, uint64_t *zoned,
2489     boolean_t accept_ancestor, int *prefixlen)
2490 {
2491 	zfs_cmd_t zc = { 0 };
2492 	char parent[ZFS_MAXNAMELEN];
2493 	char *slash;
2494 	zfs_handle_t *zhp;
2495 	char errbuf[1024];
2496 
2497 	(void) snprintf(errbuf, sizeof (errbuf), "cannot create '%s'",
2498 	    path);
2499 
2500 	/* get parent, and check to see if this is just a pool */
2501 	if (parent_name(path, parent, sizeof (parent)) != 0) {
2502 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2503 		    "missing dataset name"));
2504 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2505 	}
2506 
2507 	/* check to see if the pool exists */
2508 	if ((slash = strchr(parent, '/')) == NULL)
2509 		slash = parent + strlen(parent);
2510 	(void) strncpy(zc.zc_name, parent, slash - parent);
2511 	zc.zc_name[slash - parent] = '\0';
2512 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_OBJSET_STATS, &zc) != 0 &&
2513 	    errno == ENOENT) {
2514 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2515 		    "no such pool '%s'"), zc.zc_name);
2516 		return (zfs_error(hdl, EZFS_NOENT, errbuf));
2517 	}
2518 
2519 	/* check to see if the parent dataset exists */
2520 	while ((zhp = make_dataset_handle(hdl, parent)) == NULL) {
2521 		if (errno == ENOENT && accept_ancestor) {
2522 			/*
2523 			 * Go deeper to find an ancestor, give up on top level.
2524 			 */
2525 			if (parent_name(parent, parent, sizeof (parent)) != 0) {
2526 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2527 				    "no such pool '%s'"), zc.zc_name);
2528 				return (zfs_error(hdl, EZFS_NOENT, errbuf));
2529 			}
2530 		} else if (errno == ENOENT) {
2531 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2532 			    "parent does not exist"));
2533 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
2534 		} else
2535 			return (zfs_standard_error(hdl, errno, errbuf));
2536 	}
2537 
2538 	*zoned = zfs_prop_get_int(zhp, ZFS_PROP_ZONED);
2539 	/* we are in a non-global zone, but parent is in the global zone */
2540 	if (getzoneid() != GLOBAL_ZONEID && !(*zoned)) {
2541 		(void) zfs_standard_error(hdl, EPERM, errbuf);
2542 		zfs_close(zhp);
2543 		return (-1);
2544 	}
2545 
2546 	/* make sure parent is a filesystem */
2547 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
2548 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2549 		    "parent is not a filesystem"));
2550 		(void) zfs_error(hdl, EZFS_BADTYPE, errbuf);
2551 		zfs_close(zhp);
2552 		return (-1);
2553 	}
2554 
2555 	zfs_close(zhp);
2556 	if (prefixlen != NULL)
2557 		*prefixlen = strlen(parent);
2558 	return (0);
2559 }
2560 
2561 /*
2562  * Finds whether the dataset of the given type(s) exists.
2563  */
2564 boolean_t
2565 zfs_dataset_exists(libzfs_handle_t *hdl, const char *path, zfs_type_t types)
2566 {
2567 	zfs_handle_t *zhp;
2568 
2569 	if (!zfs_validate_name(hdl, path, types, B_FALSE))
2570 		return (B_FALSE);
2571 
2572 	/*
2573 	 * Try to get stats for the dataset, which will tell us if it exists.
2574 	 */
2575 	if ((zhp = make_dataset_handle(hdl, path)) != NULL) {
2576 		int ds_type = zhp->zfs_type;
2577 
2578 		zfs_close(zhp);
2579 		if (types & ds_type)
2580 			return (B_TRUE);
2581 	}
2582 	return (B_FALSE);
2583 }
2584 
2585 /*
2586  * Given a path to 'target', create all the ancestors between
2587  * the prefixlen portion of the path, and the target itself.
2588  * Fail if the initial prefixlen-ancestor does not already exist.
2589  */
2590 int
2591 create_parents(libzfs_handle_t *hdl, char *target, int prefixlen)
2592 {
2593 	zfs_handle_t *h;
2594 	char *cp;
2595 	const char *opname;
2596 
2597 	/* make sure prefix exists */
2598 	cp = target + prefixlen;
2599 	if (*cp != '/') {
2600 		assert(strchr(cp, '/') == NULL);
2601 		h = zfs_open(hdl, target, ZFS_TYPE_FILESYSTEM);
2602 	} else {
2603 		*cp = '\0';
2604 		h = zfs_open(hdl, target, ZFS_TYPE_FILESYSTEM);
2605 		*cp = '/';
2606 	}
2607 	if (h == NULL)
2608 		return (-1);
2609 	zfs_close(h);
2610 
2611 	/*
2612 	 * Attempt to create, mount, and share any ancestor filesystems,
2613 	 * up to the prefixlen-long one.
2614 	 */
2615 	for (cp = target + prefixlen + 1;
2616 	    cp = strchr(cp, '/'); *cp = '/', cp++) {
2617 		char *logstr;
2618 
2619 		*cp = '\0';
2620 
2621 		h = make_dataset_handle(hdl, target);
2622 		if (h) {
2623 			/* it already exists, nothing to do here */
2624 			zfs_close(h);
2625 			continue;
2626 		}
2627 
2628 		logstr = hdl->libzfs_log_str;
2629 		hdl->libzfs_log_str = NULL;
2630 		if (zfs_create(hdl, target, ZFS_TYPE_FILESYSTEM,
2631 		    NULL) != 0) {
2632 			hdl->libzfs_log_str = logstr;
2633 			opname = dgettext(TEXT_DOMAIN, "create");
2634 			goto ancestorerr;
2635 		}
2636 
2637 		hdl->libzfs_log_str = logstr;
2638 		h = zfs_open(hdl, target, ZFS_TYPE_FILESYSTEM);
2639 		if (h == NULL) {
2640 			opname = dgettext(TEXT_DOMAIN, "open");
2641 			goto ancestorerr;
2642 		}
2643 
2644 		if (zfs_mount(h, NULL, 0) != 0) {
2645 			opname = dgettext(TEXT_DOMAIN, "mount");
2646 			goto ancestorerr;
2647 		}
2648 
2649 		if (zfs_share(h) != 0) {
2650 			opname = dgettext(TEXT_DOMAIN, "share");
2651 			goto ancestorerr;
2652 		}
2653 
2654 		zfs_close(h);
2655 	}
2656 
2657 	return (0);
2658 
2659 ancestorerr:
2660 	zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2661 	    "failed to %s ancestor '%s'"), opname, target);
2662 	return (-1);
2663 }
2664 
2665 /*
2666  * Creates non-existing ancestors of the given path.
2667  */
2668 int
2669 zfs_create_ancestors(libzfs_handle_t *hdl, const char *path)
2670 {
2671 	int prefix;
2672 	uint64_t zoned;
2673 	char *path_copy;
2674 	int rc;
2675 
2676 	if (check_parents(hdl, path, &zoned, B_TRUE, &prefix) != 0)
2677 		return (-1);
2678 
2679 	if ((path_copy = strdup(path)) != NULL) {
2680 		rc = create_parents(hdl, path_copy, prefix);
2681 		free(path_copy);
2682 	}
2683 	if (path_copy == NULL || rc != 0)
2684 		return (-1);
2685 
2686 	return (0);
2687 }
2688 
2689 /*
2690  * Create a new filesystem or volume.
2691  */
2692 int
2693 zfs_create(libzfs_handle_t *hdl, const char *path, zfs_type_t type,
2694     nvlist_t *props)
2695 {
2696 	zfs_cmd_t zc = { 0 };
2697 	int ret;
2698 	uint64_t size = 0;
2699 	uint64_t blocksize = zfs_prop_default_numeric(ZFS_PROP_VOLBLOCKSIZE);
2700 	char errbuf[1024];
2701 	uint64_t zoned;
2702 
2703 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2704 	    "cannot create '%s'"), path);
2705 
2706 	/* validate the path, taking care to note the extended error message */
2707 	if (!zfs_validate_name(hdl, path, type, B_TRUE))
2708 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
2709 
2710 	/* validate parents exist */
2711 	if (check_parents(hdl, path, &zoned, B_FALSE, NULL) != 0)
2712 		return (-1);
2713 
2714 	/*
2715 	 * The failure modes when creating a dataset of a different type over
2716 	 * one that already exists is a little strange.  In particular, if you
2717 	 * try to create a dataset on top of an existing dataset, the ioctl()
2718 	 * will return ENOENT, not EEXIST.  To prevent this from happening, we
2719 	 * first try to see if the dataset exists.
2720 	 */
2721 	(void) strlcpy(zc.zc_name, path, sizeof (zc.zc_name));
2722 	if (zfs_dataset_exists(hdl, zc.zc_name, ZFS_TYPE_DATASET)) {
2723 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2724 		    "dataset already exists"));
2725 		return (zfs_error(hdl, EZFS_EXISTS, errbuf));
2726 	}
2727 
2728 	if (type == ZFS_TYPE_VOLUME)
2729 		zc.zc_objset_type = DMU_OST_ZVOL;
2730 	else
2731 		zc.zc_objset_type = DMU_OST_ZFS;
2732 
2733 	if (props && (props = zfs_validate_properties(hdl, type, props,
2734 	    zoned, NULL, errbuf)) == 0)
2735 		return (-1);
2736 
2737 	if (type == ZFS_TYPE_VOLUME) {
2738 		/*
2739 		 * If we are creating a volume, the size and block size must
2740 		 * satisfy a few restraints.  First, the blocksize must be a
2741 		 * valid block size between SPA_{MIN,MAX}BLOCKSIZE.  Second, the
2742 		 * volsize must be a multiple of the block size, and cannot be
2743 		 * zero.
2744 		 */
2745 		if (props == NULL || nvlist_lookup_uint64(props,
2746 		    zfs_prop_to_name(ZFS_PROP_VOLSIZE), &size) != 0) {
2747 			nvlist_free(props);
2748 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2749 			    "missing volume size"));
2750 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
2751 		}
2752 
2753 		if ((ret = nvlist_lookup_uint64(props,
2754 		    zfs_prop_to_name(ZFS_PROP_VOLBLOCKSIZE),
2755 		    &blocksize)) != 0) {
2756 			if (ret == ENOENT) {
2757 				blocksize = zfs_prop_default_numeric(
2758 				    ZFS_PROP_VOLBLOCKSIZE);
2759 			} else {
2760 				nvlist_free(props);
2761 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2762 				    "missing volume block size"));
2763 				return (zfs_error(hdl, EZFS_BADPROP, errbuf));
2764 			}
2765 		}
2766 
2767 		if (size == 0) {
2768 			nvlist_free(props);
2769 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2770 			    "volume size cannot be zero"));
2771 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
2772 		}
2773 
2774 		if (size % blocksize != 0) {
2775 			nvlist_free(props);
2776 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2777 			    "volume size must be a multiple of volume block "
2778 			    "size"));
2779 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
2780 		}
2781 	}
2782 
2783 	if (props && zcmd_write_src_nvlist(hdl, &zc, props) != 0)
2784 		return (-1);
2785 	nvlist_free(props);
2786 
2787 	/* create the dataset */
2788 	ret = zfs_ioctl(hdl, ZFS_IOC_CREATE, &zc);
2789 
2790 	if (ret == 0 && type == ZFS_TYPE_VOLUME) {
2791 		ret = zvol_create_link(hdl, path);
2792 		if (ret) {
2793 			(void) zfs_standard_error(hdl, errno,
2794 			    dgettext(TEXT_DOMAIN,
2795 			    "Volume successfully created, but device links "
2796 			    "were not created"));
2797 			zcmd_free_nvlists(&zc);
2798 			return (-1);
2799 		}
2800 	}
2801 
2802 	zcmd_free_nvlists(&zc);
2803 
2804 	/* check for failure */
2805 	if (ret != 0) {
2806 		char parent[ZFS_MAXNAMELEN];
2807 		(void) parent_name(path, parent, sizeof (parent));
2808 
2809 		switch (errno) {
2810 		case ENOENT:
2811 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2812 			    "no such parent '%s'"), parent);
2813 			return (zfs_error(hdl, EZFS_NOENT, errbuf));
2814 
2815 		case EINVAL:
2816 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2817 			    "parent '%s' is not a filesystem"), parent);
2818 			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
2819 
2820 		case EDOM:
2821 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2822 			    "volume block size must be power of 2 from "
2823 			    "%u to %uk"),
2824 			    (uint_t)SPA_MINBLOCKSIZE,
2825 			    (uint_t)SPA_MAXBLOCKSIZE >> 10);
2826 
2827 			return (zfs_error(hdl, EZFS_BADPROP, errbuf));
2828 
2829 		case ENOTSUP:
2830 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
2831 			    "pool must be upgraded to set this "
2832 			    "property or value"));
2833 			return (zfs_error(hdl, EZFS_BADVERSION, errbuf));
2834 
2835 #ifdef _ILP32
2836 		case EOVERFLOW:
2837 			/*
2838 			 * This platform can't address a volume this big.
2839 			 */
2840 			if (type == ZFS_TYPE_VOLUME)
2841 				return (zfs_error(hdl, EZFS_VOLTOOBIG,
2842 				    errbuf));
2843 #endif
2844 			/* FALLTHROUGH */
2845 		default:
2846 			return (zfs_standard_error(hdl, errno, errbuf));
2847 		}
2848 	}
2849 
2850 	return (0);
2851 }
2852 
2853 /*
2854  * Destroys the given dataset.  The caller must make sure that the filesystem
2855  * isn't mounted, and that there are no active dependents.
2856  */
2857 int
2858 zfs_destroy(zfs_handle_t *zhp)
2859 {
2860 	zfs_cmd_t zc = { 0 };
2861 
2862 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
2863 
2864 	if (ZFS_IS_VOLUME(zhp)) {
2865 		/*
2866 		 * If user doesn't have permissions to unshare volume, then
2867 		 * abort the request.  This would only happen for a
2868 		 * non-privileged user.
2869 		 */
2870 		if (zfs_unshare_iscsi(zhp) != 0) {
2871 			return (-1);
2872 		}
2873 
2874 		if (zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name) != 0)
2875 			return (-1);
2876 
2877 		zc.zc_objset_type = DMU_OST_ZVOL;
2878 	} else {
2879 		zc.zc_objset_type = DMU_OST_ZFS;
2880 	}
2881 
2882 	if (zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_DESTROY, &zc) != 0) {
2883 		return (zfs_standard_error_fmt(zhp->zfs_hdl, errno,
2884 		    dgettext(TEXT_DOMAIN, "cannot destroy '%s'"),
2885 		    zhp->zfs_name));
2886 	}
2887 
2888 	remove_mountpoint(zhp);
2889 
2890 	return (0);
2891 }
2892 
2893 struct destroydata {
2894 	char *snapname;
2895 	boolean_t gotone;
2896 	boolean_t closezhp;
2897 };
2898 
2899 static int
2900 zfs_remove_link_cb(zfs_handle_t *zhp, void *arg)
2901 {
2902 	struct destroydata *dd = arg;
2903 	zfs_handle_t *szhp;
2904 	char name[ZFS_MAXNAMELEN];
2905 	boolean_t closezhp = dd->closezhp;
2906 	int rv;
2907 
2908 	(void) strlcpy(name, zhp->zfs_name, sizeof (name));
2909 	(void) strlcat(name, "@", sizeof (name));
2910 	(void) strlcat(name, dd->snapname, sizeof (name));
2911 
2912 	szhp = make_dataset_handle(zhp->zfs_hdl, name);
2913 	if (szhp) {
2914 		dd->gotone = B_TRUE;
2915 		zfs_close(szhp);
2916 	}
2917 
2918 	if (zhp->zfs_type == ZFS_TYPE_VOLUME) {
2919 		(void) zvol_remove_link(zhp->zfs_hdl, name);
2920 		/*
2921 		 * NB: this is simply a best-effort.  We don't want to
2922 		 * return an error, because then we wouldn't visit all
2923 		 * the volumes.
2924 		 */
2925 	}
2926 
2927 	dd->closezhp = B_TRUE;
2928 	rv = zfs_iter_filesystems(zhp, zfs_remove_link_cb, arg);
2929 	if (closezhp)
2930 		zfs_close(zhp);
2931 	return (rv);
2932 }
2933 
2934 /*
2935  * Destroys all snapshots with the given name in zhp & descendants.
2936  */
2937 int
2938 zfs_destroy_snaps(zfs_handle_t *zhp, char *snapname)
2939 {
2940 	zfs_cmd_t zc = { 0 };
2941 	int ret;
2942 	struct destroydata dd = { 0 };
2943 
2944 	dd.snapname = snapname;
2945 	(void) zfs_remove_link_cb(zhp, &dd);
2946 
2947 	if (!dd.gotone) {
2948 		return (zfs_standard_error_fmt(zhp->zfs_hdl, ENOENT,
2949 		    dgettext(TEXT_DOMAIN, "cannot destroy '%s@%s'"),
2950 		    zhp->zfs_name, snapname));
2951 	}
2952 
2953 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
2954 	(void) strlcpy(zc.zc_value, snapname, sizeof (zc.zc_value));
2955 
2956 	ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_DESTROY_SNAPS, &zc);
2957 	if (ret != 0) {
2958 		char errbuf[1024];
2959 
2960 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2961 		    "cannot destroy '%s@%s'"), zc.zc_name, snapname);
2962 
2963 		switch (errno) {
2964 		case EEXIST:
2965 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
2966 			    "snapshot is cloned"));
2967 			return (zfs_error(zhp->zfs_hdl, EZFS_EXISTS, errbuf));
2968 
2969 		default:
2970 			return (zfs_standard_error(zhp->zfs_hdl, errno,
2971 			    errbuf));
2972 		}
2973 	}
2974 
2975 	return (0);
2976 }
2977 
2978 /*
2979  * Clones the given dataset.  The target must be of the same type as the source.
2980  */
2981 int
2982 zfs_clone(zfs_handle_t *zhp, const char *target, nvlist_t *props)
2983 {
2984 	zfs_cmd_t zc = { 0 };
2985 	char parent[ZFS_MAXNAMELEN];
2986 	int ret;
2987 	char errbuf[1024];
2988 	libzfs_handle_t *hdl = zhp->zfs_hdl;
2989 	zfs_type_t type;
2990 	uint64_t zoned;
2991 
2992 	assert(zhp->zfs_type == ZFS_TYPE_SNAPSHOT);
2993 
2994 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
2995 	    "cannot create '%s'"), target);
2996 
2997 	/* validate the target name */
2998 	if (!zfs_validate_name(hdl, target, ZFS_TYPE_FILESYSTEM, B_TRUE))
2999 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3000 
3001 	/* validate parents exist */
3002 	if (check_parents(hdl, target, &zoned, B_FALSE, NULL) != 0)
3003 		return (-1);
3004 
3005 	(void) parent_name(target, parent, sizeof (parent));
3006 
3007 	/* do the clone */
3008 	if (ZFS_IS_VOLUME(zhp)) {
3009 		zc.zc_objset_type = DMU_OST_ZVOL;
3010 		type = ZFS_TYPE_VOLUME;
3011 	} else {
3012 		zc.zc_objset_type = DMU_OST_ZFS;
3013 		type = ZFS_TYPE_FILESYSTEM;
3014 	}
3015 
3016 	if (props) {
3017 		if ((props = zfs_validate_properties(hdl, type, props,
3018 		    zoned, zhp, errbuf)) == NULL)
3019 			return (-1);
3020 
3021 		if (zcmd_write_src_nvlist(hdl, &zc, props) != 0) {
3022 			nvlist_free(props);
3023 			return (-1);
3024 		}
3025 
3026 		nvlist_free(props);
3027 	}
3028 
3029 	(void) strlcpy(zc.zc_name, target, sizeof (zc.zc_name));
3030 	(void) strlcpy(zc.zc_value, zhp->zfs_name, sizeof (zc.zc_value));
3031 	ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_CREATE, &zc);
3032 
3033 	zcmd_free_nvlists(&zc);
3034 
3035 	if (ret != 0) {
3036 		switch (errno) {
3037 
3038 		case ENOENT:
3039 			/*
3040 			 * The parent doesn't exist.  We should have caught this
3041 			 * above, but there may a race condition that has since
3042 			 * destroyed the parent.
3043 			 *
3044 			 * At this point, we don't know whether it's the source
3045 			 * that doesn't exist anymore, or whether the target
3046 			 * dataset doesn't exist.
3047 			 */
3048 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
3049 			    "no such parent '%s'"), parent);
3050 			return (zfs_error(zhp->zfs_hdl, EZFS_NOENT, errbuf));
3051 
3052 		case EXDEV:
3053 			zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
3054 			    "source and target pools differ"));
3055 			return (zfs_error(zhp->zfs_hdl, EZFS_CROSSTARGET,
3056 			    errbuf));
3057 
3058 		default:
3059 			return (zfs_standard_error(zhp->zfs_hdl, errno,
3060 			    errbuf));
3061 		}
3062 	} else if (ZFS_IS_VOLUME(zhp)) {
3063 		ret = zvol_create_link(zhp->zfs_hdl, target);
3064 	}
3065 
3066 	return (ret);
3067 }
3068 
3069 typedef struct promote_data {
3070 	char cb_mountpoint[MAXPATHLEN];
3071 	const char *cb_target;
3072 	const char *cb_errbuf;
3073 	uint64_t cb_pivot_txg;
3074 } promote_data_t;
3075 
3076 static int
3077 promote_snap_cb(zfs_handle_t *zhp, void *data)
3078 {
3079 	promote_data_t *pd = data;
3080 	zfs_handle_t *szhp;
3081 	char snapname[MAXPATHLEN];
3082 	int rv = 0;
3083 
3084 	/* We don't care about snapshots after the pivot point */
3085 	if (zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG) > pd->cb_pivot_txg) {
3086 		zfs_close(zhp);
3087 		return (0);
3088 	}
3089 
3090 	/* Remove the device link if it's a zvol. */
3091 	if (ZFS_IS_VOLUME(zhp))
3092 		(void) zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name);
3093 
3094 	/* Check for conflicting names */
3095 	(void) strlcpy(snapname, pd->cb_target, sizeof (snapname));
3096 	(void) strlcat(snapname, strchr(zhp->zfs_name, '@'), sizeof (snapname));
3097 	szhp = make_dataset_handle(zhp->zfs_hdl, snapname);
3098 	if (szhp != NULL) {
3099 		zfs_close(szhp);
3100 		zfs_error_aux(zhp->zfs_hdl, dgettext(TEXT_DOMAIN,
3101 		    "snapshot name '%s' from origin \n"
3102 		    "conflicts with '%s' from target"),
3103 		    zhp->zfs_name, snapname);
3104 		rv = zfs_error(zhp->zfs_hdl, EZFS_EXISTS, pd->cb_errbuf);
3105 	}
3106 	zfs_close(zhp);
3107 	return (rv);
3108 }
3109 
3110 static int
3111 promote_snap_done_cb(zfs_handle_t *zhp, void *data)
3112 {
3113 	promote_data_t *pd = data;
3114 
3115 	/* We don't care about snapshots after the pivot point */
3116 	if (zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG) <= pd->cb_pivot_txg) {
3117 		/* Create the device link if it's a zvol. */
3118 		if (ZFS_IS_VOLUME(zhp))
3119 			(void) zvol_create_link(zhp->zfs_hdl, zhp->zfs_name);
3120 	}
3121 
3122 	zfs_close(zhp);
3123 	return (0);
3124 }
3125 
3126 /*
3127  * Promotes the given clone fs to be the clone parent.
3128  */
3129 int
3130 zfs_promote(zfs_handle_t *zhp)
3131 {
3132 	libzfs_handle_t *hdl = zhp->zfs_hdl;
3133 	zfs_cmd_t zc = { 0 };
3134 	char parent[MAXPATHLEN];
3135 	char *cp;
3136 	int ret;
3137 	zfs_handle_t *pzhp;
3138 	promote_data_t pd;
3139 	char errbuf[1024];
3140 
3141 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3142 	    "cannot promote '%s'"), zhp->zfs_name);
3143 
3144 	if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) {
3145 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3146 		    "snapshots can not be promoted"));
3147 		return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
3148 	}
3149 
3150 	(void) strlcpy(parent, zhp->zfs_dmustats.dds_origin, sizeof (parent));
3151 	if (parent[0] == '\0') {
3152 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3153 		    "not a cloned filesystem"));
3154 		return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
3155 	}
3156 	cp = strchr(parent, '@');
3157 	*cp = '\0';
3158 
3159 	/* Walk the snapshots we will be moving */
3160 	pzhp = zfs_open(hdl, zhp->zfs_dmustats.dds_origin, ZFS_TYPE_SNAPSHOT);
3161 	if (pzhp == NULL)
3162 		return (-1);
3163 	pd.cb_pivot_txg = zfs_prop_get_int(pzhp, ZFS_PROP_CREATETXG);
3164 	zfs_close(pzhp);
3165 	pd.cb_target = zhp->zfs_name;
3166 	pd.cb_errbuf = errbuf;
3167 	pzhp = zfs_open(hdl, parent, ZFS_TYPE_DATASET);
3168 	if (pzhp == NULL)
3169 		return (-1);
3170 	(void) zfs_prop_get(pzhp, ZFS_PROP_MOUNTPOINT, pd.cb_mountpoint,
3171 	    sizeof (pd.cb_mountpoint), NULL, NULL, 0, FALSE);
3172 	ret = zfs_iter_snapshots(pzhp, promote_snap_cb, &pd);
3173 	if (ret != 0) {
3174 		zfs_close(pzhp);
3175 		return (-1);
3176 	}
3177 
3178 	/* issue the ioctl */
3179 	(void) strlcpy(zc.zc_value, zhp->zfs_dmustats.dds_origin,
3180 	    sizeof (zc.zc_value));
3181 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
3182 	ret = zfs_ioctl(hdl, ZFS_IOC_PROMOTE, &zc);
3183 
3184 	if (ret != 0) {
3185 		int save_errno = errno;
3186 
3187 		(void) zfs_iter_snapshots(pzhp, promote_snap_done_cb, &pd);
3188 		zfs_close(pzhp);
3189 
3190 		switch (save_errno) {
3191 		case EEXIST:
3192 			/*
3193 			 * There is a conflicting snapshot name.  We
3194 			 * should have caught this above, but they could
3195 			 * have renamed something in the mean time.
3196 			 */
3197 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3198 			    "conflicting snapshot name from parent '%s'"),
3199 			    parent);
3200 			return (zfs_error(hdl, EZFS_EXISTS, errbuf));
3201 
3202 		default:
3203 			return (zfs_standard_error(hdl, save_errno, errbuf));
3204 		}
3205 	} else {
3206 		(void) zfs_iter_snapshots(zhp, promote_snap_done_cb, &pd);
3207 	}
3208 
3209 	zfs_close(pzhp);
3210 	return (ret);
3211 }
3212 
3213 struct createdata {
3214 	const char *cd_snapname;
3215 	int cd_ifexists;
3216 };
3217 
3218 static int
3219 zfs_create_link_cb(zfs_handle_t *zhp, void *arg)
3220 {
3221 	struct createdata *cd = arg;
3222 	int ret;
3223 
3224 	if (zhp->zfs_type == ZFS_TYPE_VOLUME) {
3225 		char name[MAXPATHLEN];
3226 
3227 		(void) strlcpy(name, zhp->zfs_name, sizeof (name));
3228 		(void) strlcat(name, "@", sizeof (name));
3229 		(void) strlcat(name, cd->cd_snapname, sizeof (name));
3230 		(void) zvol_create_link_common(zhp->zfs_hdl, name,
3231 		    cd->cd_ifexists);
3232 		/*
3233 		 * NB: this is simply a best-effort.  We don't want to
3234 		 * return an error, because then we wouldn't visit all
3235 		 * the volumes.
3236 		 */
3237 	}
3238 
3239 	ret = zfs_iter_filesystems(zhp, zfs_create_link_cb, cd);
3240 
3241 	zfs_close(zhp);
3242 
3243 	return (ret);
3244 }
3245 
3246 /*
3247  * Takes a snapshot of the given dataset.
3248  */
3249 int
3250 zfs_snapshot(libzfs_handle_t *hdl, const char *path, boolean_t recursive)
3251 {
3252 	const char *delim;
3253 	char *parent;
3254 	zfs_handle_t *zhp;
3255 	zfs_cmd_t zc = { 0 };
3256 	int ret;
3257 	char errbuf[1024];
3258 
3259 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3260 	    "cannot snapshot '%s'"), path);
3261 
3262 	/* validate the target name */
3263 	if (!zfs_validate_name(hdl, path, ZFS_TYPE_SNAPSHOT, B_TRUE))
3264 		return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3265 
3266 	/* make sure the parent exists and is of the appropriate type */
3267 	delim = strchr(path, '@');
3268 	if ((parent = zfs_alloc(hdl, delim - path + 1)) == NULL)
3269 		return (-1);
3270 	(void) strncpy(parent, path, delim - path);
3271 	parent[delim - path] = '\0';
3272 
3273 	if ((zhp = zfs_open(hdl, parent, ZFS_TYPE_FILESYSTEM |
3274 	    ZFS_TYPE_VOLUME)) == NULL) {
3275 		free(parent);
3276 		return (-1);
3277 	}
3278 
3279 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
3280 	(void) strlcpy(zc.zc_value, delim+1, sizeof (zc.zc_value));
3281 	if (ZFS_IS_VOLUME(zhp))
3282 		zc.zc_objset_type = DMU_OST_ZVOL;
3283 	else
3284 		zc.zc_objset_type = DMU_OST_ZFS;
3285 	zc.zc_cookie = recursive;
3286 	ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_SNAPSHOT, &zc);
3287 
3288 	/*
3289 	 * if it was recursive, the one that actually failed will be in
3290 	 * zc.zc_name.
3291 	 */
3292 	if (ret != 0)
3293 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3294 		    "cannot create snapshot '%s@%s'"), zc.zc_name, zc.zc_value);
3295 
3296 	if (ret == 0 && recursive) {
3297 		struct createdata cd;
3298 
3299 		cd.cd_snapname = delim + 1;
3300 		cd.cd_ifexists = B_FALSE;
3301 		(void) zfs_iter_filesystems(zhp, zfs_create_link_cb, &cd);
3302 	}
3303 	if (ret == 0 && zhp->zfs_type == ZFS_TYPE_VOLUME) {
3304 		ret = zvol_create_link(zhp->zfs_hdl, path);
3305 		if (ret != 0) {
3306 			(void) zfs_standard_error(hdl, errno,
3307 			    dgettext(TEXT_DOMAIN,
3308 			    "Volume successfully snapshotted, but device links "
3309 			    "were not created"));
3310 			free(parent);
3311 			zfs_close(zhp);
3312 			return (-1);
3313 		}
3314 	}
3315 
3316 	if (ret != 0)
3317 		(void) zfs_standard_error(hdl, errno, errbuf);
3318 
3319 	free(parent);
3320 	zfs_close(zhp);
3321 
3322 	return (ret);
3323 }
3324 
3325 /*
3326  * Destroy any more recent snapshots.  We invoke this callback on any dependents
3327  * of the snapshot first.  If the 'cb_dependent' member is non-zero, then this
3328  * is a dependent and we should just destroy it without checking the transaction
3329  * group.
3330  */
3331 typedef struct rollback_data {
3332 	const char	*cb_target;		/* the snapshot */
3333 	uint64_t	cb_create;		/* creation time reference */
3334 	int		cb_error;
3335 	boolean_t	cb_dependent;
3336 } rollback_data_t;
3337 
3338 static int
3339 rollback_destroy(zfs_handle_t *zhp, void *data)
3340 {
3341 	rollback_data_t *cbp = data;
3342 
3343 	if (!cbp->cb_dependent) {
3344 		if (strcmp(zhp->zfs_name, cbp->cb_target) != 0 &&
3345 		    zfs_get_type(zhp) == ZFS_TYPE_SNAPSHOT &&
3346 		    zfs_prop_get_int(zhp, ZFS_PROP_CREATETXG) >
3347 		    cbp->cb_create) {
3348 			char *logstr;
3349 
3350 			cbp->cb_dependent = B_TRUE;
3351 			cbp->cb_error |= zfs_iter_dependents(zhp, B_FALSE,
3352 			    rollback_destroy, cbp);
3353 			cbp->cb_dependent = B_FALSE;
3354 
3355 			logstr = zhp->zfs_hdl->libzfs_log_str;
3356 			zhp->zfs_hdl->libzfs_log_str = NULL;
3357 			cbp->cb_error |= zfs_destroy(zhp);
3358 			zhp->zfs_hdl->libzfs_log_str = logstr;
3359 		}
3360 	} else {
3361 		cbp->cb_error |= zfs_destroy(zhp);
3362 	}
3363 
3364 	zfs_close(zhp);
3365 	return (0);
3366 }
3367 
3368 /*
3369  * Given a dataset, rollback to a specific snapshot, discarding any
3370  * data changes since then and making it the active dataset.
3371  *
3372  * Any snapshots more recent than the target are destroyed, along with
3373  * their dependents.
3374  */
3375 int
3376 zfs_rollback(zfs_handle_t *zhp, zfs_handle_t *snap)
3377 {
3378 	rollback_data_t cb = { 0 };
3379 	int err;
3380 	zfs_cmd_t zc = { 0 };
3381 	boolean_t restore_resv = 0;
3382 	uint64_t old_volsize, new_volsize;
3383 	zfs_prop_t resv_prop;
3384 
3385 	assert(zhp->zfs_type == ZFS_TYPE_FILESYSTEM ||
3386 	    zhp->zfs_type == ZFS_TYPE_VOLUME);
3387 
3388 	/*
3389 	 * Destroy all recent snapshots and its dependends.
3390 	 */
3391 	cb.cb_target = snap->zfs_name;
3392 	cb.cb_create = zfs_prop_get_int(snap, ZFS_PROP_CREATETXG);
3393 	(void) zfs_iter_children(zhp, rollback_destroy, &cb);
3394 
3395 	if (cb.cb_error != 0)
3396 		return (cb.cb_error);
3397 
3398 	/*
3399 	 * Now that we have verified that the snapshot is the latest,
3400 	 * rollback to the given snapshot.
3401 	 */
3402 
3403 	if (zhp->zfs_type == ZFS_TYPE_VOLUME) {
3404 		if (zvol_remove_link(zhp->zfs_hdl, zhp->zfs_name) != 0)
3405 			return (-1);
3406 		if (zfs_which_resv_prop(zhp, &resv_prop) < 0)
3407 			return (-1);
3408 		old_volsize = zfs_prop_get_int(zhp, ZFS_PROP_VOLSIZE);
3409 		restore_resv =
3410 		    (old_volsize == zfs_prop_get_int(zhp, resv_prop));
3411 	}
3412 
3413 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
3414 
3415 	if (ZFS_IS_VOLUME(zhp))
3416 		zc.zc_objset_type = DMU_OST_ZVOL;
3417 	else
3418 		zc.zc_objset_type = DMU_OST_ZFS;
3419 
3420 	/*
3421 	 * We rely on zfs_iter_children() to verify that there are no
3422 	 * newer snapshots for the given dataset.  Therefore, we can
3423 	 * simply pass the name on to the ioctl() call.  There is still
3424 	 * an unlikely race condition where the user has taken a
3425 	 * snapshot since we verified that this was the most recent.
3426 	 *
3427 	 */
3428 	if ((err = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_ROLLBACK, &zc)) != 0) {
3429 		(void) zfs_standard_error_fmt(zhp->zfs_hdl, errno,
3430 		    dgettext(TEXT_DOMAIN, "cannot rollback '%s'"),
3431 		    zhp->zfs_name);
3432 		return (err);
3433 	}
3434 
3435 	/*
3436 	 * For volumes, if the pre-rollback volsize matched the pre-
3437 	 * rollback reservation and the volsize has changed then set
3438 	 * the reservation property to the post-rollback volsize.
3439 	 * Make a new handle since the rollback closed the dataset.
3440 	 */
3441 	if ((zhp->zfs_type == ZFS_TYPE_VOLUME) &&
3442 	    (zhp = make_dataset_handle(zhp->zfs_hdl, zhp->zfs_name))) {
3443 		if (err = zvol_create_link(zhp->zfs_hdl, zhp->zfs_name)) {
3444 			zfs_close(zhp);
3445 			return (err);
3446 		}
3447 		if (restore_resv) {
3448 			new_volsize = zfs_prop_get_int(zhp, ZFS_PROP_VOLSIZE);
3449 			if (old_volsize != new_volsize)
3450 				err = zfs_prop_set_int(zhp, resv_prop,
3451 				    new_volsize);
3452 		}
3453 		zfs_close(zhp);
3454 	}
3455 	return (err);
3456 }
3457 
3458 /*
3459  * Iterate over all dependents for a given dataset.  This includes both
3460  * hierarchical dependents (children) and data dependents (snapshots and
3461  * clones).  The bulk of the processing occurs in get_dependents() in
3462  * libzfs_graph.c.
3463  */
3464 int
3465 zfs_iter_dependents(zfs_handle_t *zhp, boolean_t allowrecursion,
3466     zfs_iter_f func, void *data)
3467 {
3468 	char **dependents;
3469 	size_t count;
3470 	int i;
3471 	zfs_handle_t *child;
3472 	int ret = 0;
3473 
3474 	if (get_dependents(zhp->zfs_hdl, allowrecursion, zhp->zfs_name,
3475 	    &dependents, &count) != 0)
3476 		return (-1);
3477 
3478 	for (i = 0; i < count; i++) {
3479 		if ((child = make_dataset_handle(zhp->zfs_hdl,
3480 		    dependents[i])) == NULL)
3481 			continue;
3482 
3483 		if ((ret = func(child, data)) != 0)
3484 			break;
3485 	}
3486 
3487 	for (i = 0; i < count; i++)
3488 		free(dependents[i]);
3489 	free(dependents);
3490 
3491 	return (ret);
3492 }
3493 
3494 /*
3495  * Renames the given dataset.
3496  */
3497 int
3498 zfs_rename(zfs_handle_t *zhp, const char *target, boolean_t recursive)
3499 {
3500 	int ret;
3501 	zfs_cmd_t zc = { 0 };
3502 	char *delim;
3503 	prop_changelist_t *cl = NULL;
3504 	zfs_handle_t *zhrp = NULL;
3505 	char *parentname = NULL;
3506 	char parent[ZFS_MAXNAMELEN];
3507 	libzfs_handle_t *hdl = zhp->zfs_hdl;
3508 	char errbuf[1024];
3509 
3510 	/* if we have the same exact name, just return success */
3511 	if (strcmp(zhp->zfs_name, target) == 0)
3512 		return (0);
3513 
3514 	(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3515 	    "cannot rename to '%s'"), target);
3516 
3517 	/*
3518 	 * Make sure the target name is valid
3519 	 */
3520 	if (zhp->zfs_type == ZFS_TYPE_SNAPSHOT) {
3521 		if ((strchr(target, '@') == NULL) ||
3522 		    *target == '@') {
3523 			/*
3524 			 * Snapshot target name is abbreviated,
3525 			 * reconstruct full dataset name
3526 			 */
3527 			(void) strlcpy(parent, zhp->zfs_name,
3528 			    sizeof (parent));
3529 			delim = strchr(parent, '@');
3530 			if (strchr(target, '@') == NULL)
3531 				*(++delim) = '\0';
3532 			else
3533 				*delim = '\0';
3534 			(void) strlcat(parent, target, sizeof (parent));
3535 			target = parent;
3536 		} else {
3537 			/*
3538 			 * Make sure we're renaming within the same dataset.
3539 			 */
3540 			delim = strchr(target, '@');
3541 			if (strncmp(zhp->zfs_name, target, delim - target)
3542 			    != 0 || zhp->zfs_name[delim - target] != '@') {
3543 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3544 				    "snapshots must be part of same "
3545 				    "dataset"));
3546 				return (zfs_error(hdl, EZFS_CROSSTARGET,
3547 				    errbuf));
3548 			}
3549 		}
3550 		if (!zfs_validate_name(hdl, target, zhp->zfs_type, B_TRUE))
3551 			return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3552 	} else {
3553 		if (recursive) {
3554 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3555 			    "recursive rename must be a snapshot"));
3556 			return (zfs_error(hdl, EZFS_BADTYPE, errbuf));
3557 		}
3558 
3559 		if (!zfs_validate_name(hdl, target, zhp->zfs_type, B_TRUE))
3560 			return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3561 		uint64_t unused;
3562 
3563 		/* validate parents */
3564 		if (check_parents(hdl, target, &unused, B_FALSE, NULL) != 0)
3565 			return (-1);
3566 
3567 		(void) parent_name(target, parent, sizeof (parent));
3568 
3569 		/* make sure we're in the same pool */
3570 		verify((delim = strchr(target, '/')) != NULL);
3571 		if (strncmp(zhp->zfs_name, target, delim - target) != 0 ||
3572 		    zhp->zfs_name[delim - target] != '/') {
3573 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3574 			    "datasets must be within same pool"));
3575 			return (zfs_error(hdl, EZFS_CROSSTARGET, errbuf));
3576 		}
3577 
3578 		/* new name cannot be a child of the current dataset name */
3579 		if (strncmp(parent, zhp->zfs_name,
3580 		    strlen(zhp->zfs_name)) == 0) {
3581 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3582 			    "New dataset name cannot be a descendent of "
3583 			    "current dataset name"));
3584 			return (zfs_error(hdl, EZFS_INVALIDNAME, errbuf));
3585 		}
3586 	}
3587 
3588 	(void) snprintf(errbuf, sizeof (errbuf),
3589 	    dgettext(TEXT_DOMAIN, "cannot rename '%s'"), zhp->zfs_name);
3590 
3591 	if (getzoneid() == GLOBAL_ZONEID &&
3592 	    zfs_prop_get_int(zhp, ZFS_PROP_ZONED)) {
3593 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3594 		    "dataset is used in a non-global zone"));
3595 		return (zfs_error(hdl, EZFS_ZONED, errbuf));
3596 	}
3597 
3598 	if (recursive) {
3599 		struct destroydata dd;
3600 
3601 		parentname = zfs_strdup(zhp->zfs_hdl, zhp->zfs_name);
3602 		if (parentname == NULL) {
3603 			ret = -1;
3604 			goto error;
3605 		}
3606 		delim = strchr(parentname, '@');
3607 		*delim = '\0';
3608 		zhrp = zfs_open(zhp->zfs_hdl, parentname, ZFS_TYPE_DATASET);
3609 		if (zhrp == NULL) {
3610 			ret = -1;
3611 			goto error;
3612 		}
3613 
3614 		dd.snapname = delim + 1;
3615 		dd.gotone = B_FALSE;
3616 		dd.closezhp = B_TRUE;
3617 
3618 		/* We remove any zvol links prior to renaming them */
3619 		ret = zfs_iter_filesystems(zhrp, zfs_remove_link_cb, &dd);
3620 		if (ret) {
3621 			goto error;
3622 		}
3623 	} else {
3624 		if ((cl = changelist_gather(zhp, ZFS_PROP_NAME, 0)) == NULL)
3625 			return (-1);
3626 
3627 		if (changelist_haszonedchild(cl)) {
3628 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3629 			    "child dataset with inherited mountpoint is used "
3630 			    "in a non-global zone"));
3631 			(void) zfs_error(hdl, EZFS_ZONED, errbuf);
3632 			goto error;
3633 		}
3634 
3635 		if ((ret = changelist_prefix(cl)) != 0)
3636 			goto error;
3637 	}
3638 
3639 	if (ZFS_IS_VOLUME(zhp))
3640 		zc.zc_objset_type = DMU_OST_ZVOL;
3641 	else
3642 		zc.zc_objset_type = DMU_OST_ZFS;
3643 
3644 	(void) strlcpy(zc.zc_name, zhp->zfs_name, sizeof (zc.zc_name));
3645 	(void) strlcpy(zc.zc_value, target, sizeof (zc.zc_value));
3646 
3647 	zc.zc_cookie = recursive;
3648 
3649 	if ((ret = zfs_ioctl(zhp->zfs_hdl, ZFS_IOC_RENAME, &zc)) != 0) {
3650 		/*
3651 		 * if it was recursive, the one that actually failed will
3652 		 * be in zc.zc_name
3653 		 */
3654 		(void) snprintf(errbuf, sizeof (errbuf), dgettext(TEXT_DOMAIN,
3655 		    "cannot rename '%s'"), zc.zc_name);
3656 
3657 		if (recursive && errno == EEXIST) {
3658 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
3659 			    "a child dataset already has a snapshot "
3660 			    "with the new name"));
3661 			(void) zfs_error(hdl, EZFS_EXISTS, errbuf);
3662 		} else {
3663 			(void) zfs_standard_error(zhp->zfs_hdl, errno, errbuf);
3664 		}
3665 
3666 		/*
3667 		 * On failure, we still want to remount any filesystems that
3668 		 * were previously mounted, so we don't alter the system state.
3669 		 */
3670 		if (recursive) {
3671 			struct createdata cd;
3672 
3673 			/* only create links for datasets that had existed */
3674 			cd.cd_snapname = delim + 1;
3675 			cd.cd_ifexists = B_TRUE;
3676 			(void) zfs_iter_filesystems(zhrp, zfs_create_link_cb,
3677 			    &cd);
3678 		} else {
3679 			(void) changelist_postfix(cl);
3680 		}
3681 	} else {
3682 		if (recursive) {
3683 			struct createdata cd;
3684 
3685 			/* only create links for datasets that had existed */
3686 			cd.cd_snapname = strchr(target, '@') + 1;
3687 			cd.cd_ifexists = B_TRUE;
3688 			ret = zfs_iter_filesystems(zhrp, zfs_create_link_cb,
3689 			    &cd);
3690 		} else {
3691 			changelist_rename(cl, zfs_get_name(zhp), target);
3692 			ret = changelist_postfix(cl);
3693 		}
3694 	}
3695 
3696 error:
3697 	if (parentname) {
3698 		free(parentname);
3699 	}
3700 	if (zhrp) {
3701 		zfs_close(zhrp);
3702 	}
3703 	if (cl) {
3704 		changelist_free(cl);
3705 	}
3706 	return (ret);
3707 }
3708 
3709 /*
3710  * Given a zvol dataset, issue the ioctl to create the appropriate minor node,
3711  * poke devfsadm to create the /dev link, and then wait for the link to appear.
3712  */
3713 int
3714 zvol_create_link(libzfs_handle_t *hdl, const char *dataset)
3715 {
3716 	return (zvol_create_link_common(hdl, dataset, B_FALSE));
3717 }
3718 
3719 static int
3720 zvol_create_link_common(libzfs_handle_t *hdl, const char *dataset, int ifexists)
3721 {
3722 	zfs_cmd_t zc = { 0 };
3723 	di_devlink_handle_t dhdl;
3724 	priv_set_t *priv_effective;
3725 	int privileged;
3726 
3727 	(void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
3728 
3729 	/*
3730 	 * Issue the appropriate ioctl.
3731 	 */
3732 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_CREATE_MINOR, &zc) != 0) {
3733 		switch (errno) {
3734 		case EEXIST:
3735 			/*
3736 			 * Silently ignore the case where the link already
3737 			 * exists.  This allows 'zfs volinit' to be run multiple
3738 			 * times without errors.
3739 			 */
3740 			return (0);
3741 
3742 		case ENOENT:
3743 			/*
3744 			 * Dataset does not exist in the kernel.  If we
3745 			 * don't care (see zfs_rename), then ignore the
3746 			 * error quietly.
3747 			 */
3748 			if (ifexists) {
3749 				return (0);
3750 			}
3751 
3752 			/* FALLTHROUGH */
3753 
3754 		default:
3755 			return (zfs_standard_error_fmt(hdl, errno,
3756 			    dgettext(TEXT_DOMAIN, "cannot create device links "
3757 			    "for '%s'"), dataset));
3758 		}
3759 	}
3760 
3761 	/*
3762 	 * If privileged call devfsadm and wait for the links to
3763 	 * magically appear.
3764 	 * Otherwise, print out an informational message.
3765 	 */
3766 
3767 	priv_effective = priv_allocset();
3768 	(void) getppriv(PRIV_EFFECTIVE, priv_effective);
3769 	privileged = (priv_isfullset(priv_effective) == B_TRUE);
3770 	priv_freeset(priv_effective);
3771 
3772 	if (privileged) {
3773 		if ((dhdl = di_devlink_init(ZFS_DRIVER,
3774 		    DI_MAKE_LINK)) == NULL) {
3775 			zfs_error_aux(hdl, strerror(errno));
3776 			(void) zfs_standard_error_fmt(hdl, EZFS_DEVLINKS,
3777 			    dgettext(TEXT_DOMAIN, "cannot create device links "
3778 			    "for '%s'"), dataset);
3779 			(void) ioctl(hdl->libzfs_fd, ZFS_IOC_REMOVE_MINOR, &zc);
3780 			return (-1);
3781 		} else {
3782 			(void) di_devlink_fini(&dhdl);
3783 		}
3784 	} else {
3785 		char pathname[MAXPATHLEN];
3786 		struct stat64 statbuf;
3787 		int i;
3788 
3789 #define	MAX_WAIT	10
3790 
3791 		/*
3792 		 * This is the poor mans way of waiting for the link
3793 		 * to show up.  If after 10 seconds we still don't
3794 		 * have it, then print out a message.
3795 		 */
3796 		(void) snprintf(pathname, sizeof (pathname), "/dev/zvol/dsk/%s",
3797 		    dataset);
3798 
3799 		for (i = 0; i != MAX_WAIT; i++) {
3800 			if (stat64(pathname, &statbuf) == 0)
3801 				break;
3802 			(void) sleep(1);
3803 		}
3804 		if (i == MAX_WAIT)
3805 			(void) printf(gettext("%s may not be immediately "
3806 			    "available\n"), pathname);
3807 	}
3808 
3809 	return (0);
3810 }
3811 
3812 /*
3813  * Remove a minor node for the given zvol and the associated /dev links.
3814  */
3815 int
3816 zvol_remove_link(libzfs_handle_t *hdl, const char *dataset)
3817 {
3818 	zfs_cmd_t zc = { 0 };
3819 
3820 	(void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
3821 
3822 	if (ioctl(hdl->libzfs_fd, ZFS_IOC_REMOVE_MINOR, &zc) != 0) {
3823 		switch (errno) {
3824 		case ENXIO:
3825 			/*
3826 			 * Silently ignore the case where the link no longer
3827 			 * exists, so that 'zfs volfini' can be run multiple
3828 			 * times without errors.
3829 			 */
3830 			return (0);
3831 
3832 		default:
3833 			return (zfs_standard_error_fmt(hdl, errno,
3834 			    dgettext(TEXT_DOMAIN, "cannot remove device "
3835 			    "links for '%s'"), dataset));
3836 		}
3837 	}
3838 
3839 	return (0);
3840 }
3841 
3842 nvlist_t *
3843 zfs_get_user_props(zfs_handle_t *zhp)
3844 {
3845 	return (zhp->zfs_user_props);
3846 }
3847 
3848 /*
3849  * This function is used by 'zfs list' to determine the exact set of columns to
3850  * display, and their maximum widths.  This does two main things:
3851  *
3852  *      - If this is a list of all properties, then expand the list to include
3853  *        all native properties, and set a flag so that for each dataset we look
3854  *        for new unique user properties and add them to the list.
3855  *
3856  *      - For non fixed-width properties, keep track of the maximum width seen
3857  *        so that we can size the column appropriately.
3858  */
3859 int
3860 zfs_expand_proplist(zfs_handle_t *zhp, zprop_list_t **plp)
3861 {
3862 	libzfs_handle_t *hdl = zhp->zfs_hdl;
3863 	zprop_list_t *entry;
3864 	zprop_list_t **last, **start;
3865 	nvlist_t *userprops, *propval;
3866 	nvpair_t *elem;
3867 	char *strval;
3868 	char buf[ZFS_MAXPROPLEN];
3869 
3870 	if (zprop_expand_list(hdl, plp, ZFS_TYPE_DATASET) != 0)
3871 		return (-1);
3872 
3873 	userprops = zfs_get_user_props(zhp);
3874 
3875 	entry = *plp;
3876 	if (entry->pl_all && nvlist_next_nvpair(userprops, NULL) != NULL) {
3877 		/*
3878 		 * Go through and add any user properties as necessary.  We
3879 		 * start by incrementing our list pointer to the first
3880 		 * non-native property.
3881 		 */
3882 		start = plp;
3883 		while (*start != NULL) {
3884 			if ((*start)->pl_prop == ZPROP_INVAL)
3885 				break;
3886 			start = &(*start)->pl_next;
3887 		}
3888 
3889 		elem = NULL;
3890 		while ((elem = nvlist_next_nvpair(userprops, elem)) != NULL) {
3891 			/*
3892 			 * See if we've already found this property in our list.
3893 			 */
3894 			for (last = start; *last != NULL;
3895 			    last = &(*last)->pl_next) {
3896 				if (strcmp((*last)->pl_user_prop,
3897 				    nvpair_name(elem)) == 0)
3898 					break;
3899 			}
3900 
3901 			if (*last == NULL) {
3902 				if ((entry = zfs_alloc(hdl,
3903 				    sizeof (zprop_list_t))) == NULL ||
3904 				    ((entry->pl_user_prop = zfs_strdup(hdl,
3905 				    nvpair_name(elem)))) == NULL) {
3906 					free(entry);
3907 					return (-1);
3908 				}
3909 
3910 				entry->pl_prop = ZPROP_INVAL;
3911 				entry->pl_width = strlen(nvpair_name(elem));
3912 				entry->pl_all = B_TRUE;
3913 				*last = entry;
3914 			}
3915 		}
3916 	}
3917 
3918 	/*
3919 	 * Now go through and check the width of any non-fixed columns
3920 	 */
3921 	for (entry = *plp; entry != NULL; entry = entry->pl_next) {
3922 		if (entry->pl_fixed)
3923 			continue;
3924 
3925 		if (entry->pl_prop != ZPROP_INVAL) {
3926 			if (zfs_prop_get(zhp, entry->pl_prop,
3927 			    buf, sizeof (buf), NULL, NULL, 0, B_FALSE) == 0) {
3928 				if (strlen(buf) > entry->pl_width)
3929 					entry->pl_width = strlen(buf);
3930 			}
3931 		} else if (nvlist_lookup_nvlist(userprops,
3932 		    entry->pl_user_prop, &propval)  == 0) {
3933 			verify(nvlist_lookup_string(propval,
3934 			    ZPROP_VALUE, &strval) == 0);
3935 			if (strlen(strval) > entry->pl_width)
3936 				entry->pl_width = strlen(strval);
3937 		}
3938 	}
3939 
3940 	return (0);
3941 }
3942 
3943 int
3944 zfs_iscsi_perm_check(libzfs_handle_t *hdl, char *dataset, ucred_t *cred)
3945 {
3946 	zfs_cmd_t zc = { 0 };
3947 	nvlist_t *nvp;
3948 	gid_t gid;
3949 	uid_t uid;
3950 	const gid_t *groups;
3951 	int group_cnt;
3952 	int error;
3953 
3954 	if (nvlist_alloc(&nvp, NV_UNIQUE_NAME, 0) != 0)
3955 		return (no_memory(hdl));
3956 
3957 	uid = ucred_geteuid(cred);
3958 	gid = ucred_getegid(cred);
3959 	group_cnt = ucred_getgroups(cred, &groups);
3960 
3961 	if (uid == (uid_t)-1 || gid == (uid_t)-1 || group_cnt == (uid_t)-1)
3962 		return (1);
3963 
3964 	if (nvlist_add_uint32(nvp, ZFS_DELEG_PERM_UID, uid) != 0) {
3965 		nvlist_free(nvp);
3966 		return (1);
3967 	}
3968 
3969 	if (nvlist_add_uint32(nvp, ZFS_DELEG_PERM_GID, gid) != 0) {
3970 		nvlist_free(nvp);
3971 		return (1);
3972 	}
3973 
3974 	if (nvlist_add_uint32_array(nvp,
3975 	    ZFS_DELEG_PERM_GROUPS, (uint32_t *)groups, group_cnt) != 0) {
3976 		nvlist_free(nvp);
3977 		return (1);
3978 	}
3979 	(void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
3980 
3981 	if (zcmd_write_src_nvlist(hdl, &zc, nvp))
3982 		return (-1);
3983 
3984 	error = ioctl(hdl->libzfs_fd, ZFS_IOC_ISCSI_PERM_CHECK, &zc);
3985 	nvlist_free(nvp);
3986 	return (error);
3987 }
3988 
3989 int
3990 zfs_deleg_share_nfs(libzfs_handle_t *hdl, char *dataset, char *path,
3991     void *export, void *sharetab, int sharemax, zfs_share_op_t operation)
3992 {
3993 	zfs_cmd_t zc = { 0 };
3994 	int error;
3995 
3996 	(void) strlcpy(zc.zc_name, dataset, sizeof (zc.zc_name));
3997 	(void) strlcpy(zc.zc_value, path, sizeof (zc.zc_value));
3998 	zc.zc_share.z_sharedata = (uint64_t)(uintptr_t)sharetab;
3999 	zc.zc_share.z_exportdata = (uint64_t)(uintptr_t)export;
4000 	zc.zc_share.z_sharetype = operation;
4001 	zc.zc_share.z_sharemax = sharemax;
4002 
4003 	error = ioctl(hdl->libzfs_fd, ZFS_IOC_SHARE, &zc);
4004 	return (error);
4005 }
4006