xref: /illumos-gate/usr/src/cmd/zoneadm/zfs.c (revision 5f8171005a0c33f3c67f7da52d41c2362c3fd891)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*
28  * This file contains the functions used to support the ZFS integration
29  * with zones.  This includes validation (e.g. zonecfg dataset), cloning,
30  * file system creation and destruction.
31  */
32 
33 #include <stdio.h>
34 #include <errno.h>
35 #include <unistd.h>
36 #include <string.h>
37 #include <locale.h>
38 #include <libintl.h>
39 #include <sys/stat.h>
40 #include <sys/statvfs.h>
41 #include <libgen.h>
42 #include <libzonecfg.h>
43 #include <sys/mnttab.h>
44 #include <libzfs.h>
45 #include <sys/mntent.h>
46 #include <values.h>
47 #include <strings.h>
48 #include <assert.h>
49 
50 #include "zoneadm.h"
51 
52 libzfs_handle_t *g_zfs;
53 
54 typedef struct zfs_mount_data {
55 	char		*match_name;
56 	zfs_handle_t	*match_handle;
57 } zfs_mount_data_t;
58 
59 typedef struct zfs_snapshot_data {
60 	char	*match_name;	/* zonename@SUNWzone */
61 	int	len;		/* strlen of match_name */
62 	int	max;		/* highest digit appended to snap name */
63 	int	num;		/* number of snapshots to rename */
64 	int	cntr;		/* counter for renaming snapshots */
65 } zfs_snapshot_data_t;
66 
67 typedef struct clone_data {
68 	zfs_handle_t	*clone_zhp;	/* clone dataset to promote */
69 	time_t		origin_creation; /* snapshot creation time of clone */
70 	const char	*snapshot;	/* snapshot of dataset being demoted */
71 } clone_data_t;
72 
73 /*
74  * A ZFS file system iterator call-back function which is used to validate
75  * datasets imported into the zone.
76  */
77 /* ARGSUSED */
78 static int
79 check_zvol(zfs_handle_t *zhp, void *unused)
80 {
81 	int ret;
82 
83 	if (zfs_get_type(zhp) == ZFS_TYPE_VOLUME) {
84 		/*
85 		 * TRANSLATION_NOTE
86 		 * zfs and dataset are literals that should not be translated.
87 		 */
88 		(void) fprintf(stderr, gettext("cannot verify zfs dataset %s: "
89 		    "volumes cannot be specified as a zone dataset resource\n"),
90 		    zfs_get_name(zhp));
91 		ret = -1;
92 	} else {
93 		ret = zfs_iter_children(zhp, check_zvol, NULL);
94 	}
95 
96 	zfs_close(zhp);
97 
98 	return (ret);
99 }
100 
101 /*
102  * A ZFS file system iterator call-back function which returns the
103  * zfs_handle_t for a ZFS file system on the specified mount point.
104  */
105 static int
106 match_mountpoint(zfs_handle_t *zhp, void *data)
107 {
108 	int			res;
109 	zfs_mount_data_t	*cbp;
110 	char			mp[ZFS_MAXPROPLEN];
111 
112 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
113 		zfs_close(zhp);
114 		return (0);
115 	}
116 
117 	/* First check if the dataset is mounted. */
118 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTED, mp, sizeof (mp), NULL, NULL,
119 	    0, B_FALSE) != 0 || strcmp(mp, "no") == 0) {
120 		zfs_close(zhp);
121 		return (0);
122 	}
123 
124 	/* Now check mount point. */
125 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, mp, sizeof (mp), NULL, NULL,
126 	    0, B_FALSE) != 0) {
127 		zfs_close(zhp);
128 		return (0);
129 	}
130 
131 	cbp = (zfs_mount_data_t *)data;
132 
133 	if (strcmp(mp, "legacy") == 0) {
134 		/* If legacy, must look in mnttab for mountpoint. */
135 		FILE		*fp;
136 		struct mnttab	entry;
137 		const char	*nm;
138 
139 		nm = zfs_get_name(zhp);
140 		if ((fp = fopen(MNTTAB, "r")) == NULL) {
141 			zfs_close(zhp);
142 			return (0);
143 		}
144 
145 		while (getmntent(fp, &entry) == 0) {
146 			if (strcmp(nm, entry.mnt_special) == 0) {
147 				if (strcmp(entry.mnt_mountp, cbp->match_name)
148 				    == 0) {
149 					(void) fclose(fp);
150 					cbp->match_handle = zhp;
151 					return (1);
152 				}
153 				break;
154 			}
155 		}
156 		(void) fclose(fp);
157 
158 	} else if (strcmp(mp, cbp->match_name) == 0) {
159 		cbp->match_handle = zhp;
160 		return (1);
161 	}
162 
163 	/* Iterate over any nested datasets. */
164 	res = zfs_iter_filesystems(zhp, match_mountpoint, data);
165 	zfs_close(zhp);
166 	return (res);
167 }
168 
169 /*
170  * Get ZFS handle for the specified mount point.
171  */
172 static zfs_handle_t *
173 mount2zhandle(char *mountpoint)
174 {
175 	zfs_mount_data_t	cb;
176 
177 	cb.match_name = mountpoint;
178 	cb.match_handle = NULL;
179 	(void) zfs_iter_root(g_zfs, match_mountpoint, &cb);
180 	return (cb.match_handle);
181 }
182 
183 /*
184  * Check if there is already a file system (zfs or any other type) mounted on
185  * path.
186  */
187 static boolean_t
188 is_mountpnt(char *path)
189 {
190 	FILE		*fp;
191 	struct mnttab	entry;
192 
193 	if ((fp = fopen(MNTTAB, "r")) == NULL)
194 		return (B_FALSE);
195 
196 	while (getmntent(fp, &entry) == 0) {
197 		if (strcmp(path, entry.mnt_mountp) == 0) {
198 			(void) fclose(fp);
199 			return (B_TRUE);
200 		}
201 	}
202 
203 	(void) fclose(fp);
204 	return (B_FALSE);
205 }
206 
207 /*
208  * Run the brand's pre-snapshot hook before we take a ZFS snapshot of the zone.
209  */
210 static int
211 pre_snapshot(char *presnapbuf)
212 {
213 	int status;
214 
215 	/* No brand-specific handler */
216 	if (presnapbuf[0] == '\0')
217 		return (Z_OK);
218 
219 	/* Run the hook */
220 	status = do_subproc(presnapbuf);
221 	if ((status = subproc_status(gettext("brand-specific presnapshot"),
222 	    status, B_FALSE)) != ZONE_SUBPROC_OK)
223 		return (Z_ERR);
224 
225 	return (Z_OK);
226 }
227 
228 /*
229  * Run the brand's post-snapshot hook after we take a ZFS snapshot of the zone.
230  */
231 static int
232 post_snapshot(char *postsnapbuf)
233 {
234 	int status;
235 
236 	/* No brand-specific handler */
237 	if (postsnapbuf[0] == '\0')
238 		return (Z_OK);
239 
240 	/* Run the hook */
241 	status = do_subproc(postsnapbuf);
242 	if ((status = subproc_status(gettext("brand-specific postsnapshot"),
243 	    status, B_FALSE)) != ZONE_SUBPROC_OK)
244 		return (Z_ERR);
245 
246 	return (Z_OK);
247 }
248 
249 /*
250  * This is a ZFS snapshot iterator call-back function which returns the
251  * highest number of SUNWzone snapshots that have been taken.
252  */
253 static int
254 get_snap_max(zfs_handle_t *zhp, void *data)
255 {
256 	int			res;
257 	zfs_snapshot_data_t	*cbp;
258 
259 	if (zfs_get_type(zhp) != ZFS_TYPE_SNAPSHOT) {
260 		zfs_close(zhp);
261 		return (0);
262 	}
263 
264 	cbp = (zfs_snapshot_data_t *)data;
265 
266 	if (strncmp(zfs_get_name(zhp), cbp->match_name, cbp->len) == 0) {
267 		char	*nump;
268 		int	num;
269 
270 		cbp->num++;
271 		nump = (char *)(zfs_get_name(zhp) + cbp->len);
272 		num = atoi(nump);
273 		if (num > cbp->max)
274 			cbp->max = num;
275 	}
276 
277 	res = zfs_iter_snapshots(zhp, get_snap_max, data);
278 	zfs_close(zhp);
279 	return (res);
280 }
281 
282 /*
283  * Take a ZFS snapshot to be used for cloning the zone.
284  */
285 static int
286 take_snapshot(zfs_handle_t *zhp, char *snapshot_name, int snap_size,
287     char *presnapbuf, char *postsnapbuf)
288 {
289 	int			res;
290 	char			template[ZFS_MAXNAMELEN];
291 	zfs_snapshot_data_t	cb;
292 
293 	/*
294 	 * First we need to figure out the next available name for the
295 	 * zone snapshot.  Look through the list of zones snapshots for
296 	 * this file system to determine the maximum snapshot name.
297 	 */
298 	if (snprintf(template, sizeof (template), "%s@SUNWzone",
299 	    zfs_get_name(zhp)) >=  sizeof (template))
300 		return (Z_ERR);
301 
302 	cb.match_name = template;
303 	cb.len = strlen(template);
304 	cb.max = 0;
305 
306 	if (zfs_iter_snapshots(zhp, get_snap_max, &cb) != 0)
307 		return (Z_ERR);
308 
309 	cb.max++;
310 
311 	if (snprintf(snapshot_name, snap_size, "%s@SUNWzone%d",
312 	    zfs_get_name(zhp), cb.max) >= snap_size)
313 		return (Z_ERR);
314 
315 	if (pre_snapshot(presnapbuf) != Z_OK)
316 		return (Z_ERR);
317 	res = zfs_snapshot(g_zfs, snapshot_name, B_FALSE, NULL);
318 	if (post_snapshot(postsnapbuf) != Z_OK)
319 		return (Z_ERR);
320 
321 	if (res != 0)
322 		return (Z_ERR);
323 	return (Z_OK);
324 }
325 
326 /*
327  * We are using an explicit snapshot from some earlier point in time so
328  * we need to validate it.  Run the brand specific hook.
329  */
330 static int
331 validate_snapshot(char *snapshot_name, char *snap_path, char *validsnapbuf)
332 {
333 	int status;
334 	char cmdbuf[MAXPATHLEN];
335 
336 	/* No brand-specific handler */
337 	if (validsnapbuf[0] == '\0')
338 		return (Z_OK);
339 
340 	/* pass args - snapshot_name & snap_path */
341 	if (snprintf(cmdbuf, sizeof (cmdbuf), "%s %s %s", validsnapbuf,
342 	    snapshot_name, snap_path) >= sizeof (cmdbuf)) {
343 		zerror("Command line too long");
344 		return (Z_ERR);
345 	}
346 
347 	/* Run the hook */
348 	status = do_subproc(cmdbuf);
349 	if ((status = subproc_status(gettext("brand-specific validatesnapshot"),
350 	    status, B_FALSE)) != ZONE_SUBPROC_OK)
351 		return (Z_ERR);
352 
353 	return (Z_OK);
354 }
355 
356 /*
357  * Remove the sw inventory file from inside this zonepath that we picked up out
358  * of the snapshot.
359  */
360 static int
361 clean_out_clone()
362 {
363 	int err;
364 	zone_dochandle_t handle;
365 
366 	if ((handle = zonecfg_init_handle()) == NULL) {
367 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
368 		return (Z_ERR);
369 	}
370 
371 	if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) {
372 		errno = err;
373 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
374 		zonecfg_fini_handle(handle);
375 		return (Z_ERR);
376 	}
377 
378 	zonecfg_rm_detached(handle, B_FALSE);
379 	zonecfg_fini_handle(handle);
380 
381 	return (Z_OK);
382 }
383 
384 /*
385  * Make a ZFS clone on zonepath from snapshot_name.
386  */
387 static int
388 clone_snap(char *snapshot_name, char *zonepath)
389 {
390 	int		res = Z_OK;
391 	int		err;
392 	zfs_handle_t	*zhp;
393 	zfs_handle_t	*clone;
394 	nvlist_t	*props = NULL;
395 
396 	if ((zhp = zfs_open(g_zfs, snapshot_name, ZFS_TYPE_SNAPSHOT)) == NULL)
397 		return (Z_NO_ENTRY);
398 
399 	(void) printf(gettext("Cloning snapshot %s\n"), snapshot_name);
400 
401 	if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0 ||
402 	    nvlist_add_string(props, zfs_prop_to_name(ZFS_PROP_SHARENFS),
403 	    "off") != 0) {
404 		if (props != NULL)
405 			nvlist_free(props);
406 		(void) fprintf(stderr, gettext("could not create ZFS clone "
407 		    "%s: out of memory\n"), zonepath);
408 		return (Z_ERR);
409 	}
410 
411 	err = zfs_clone(zhp, zonepath, props);
412 	zfs_close(zhp);
413 
414 	nvlist_free(props);
415 
416 	if (err != 0)
417 		return (Z_ERR);
418 
419 	/* create the mountpoint if necessary */
420 	if ((clone = zfs_open(g_zfs, zonepath, ZFS_TYPE_DATASET)) == NULL)
421 		return (Z_ERR);
422 
423 	/*
424 	 * The clone has been created so we need to print a diagnostic
425 	 * message if one of the following steps fails for some reason.
426 	 */
427 	if (zfs_mount(clone, NULL, 0) != 0) {
428 		(void) fprintf(stderr, gettext("could not mount ZFS clone "
429 		    "%s\n"), zfs_get_name(clone));
430 		res = Z_ERR;
431 
432 	} else if (clean_out_clone() != Z_OK) {
433 		(void) fprintf(stderr, gettext("could not remove the "
434 		    "software inventory from ZFS clone %s\n"),
435 		    zfs_get_name(clone));
436 		res = Z_ERR;
437 	}
438 
439 	zfs_close(clone);
440 	return (res);
441 }
442 
443 /*
444  * This function takes a zonepath and attempts to determine what the ZFS
445  * file system name (not mountpoint) should be for that path.  We do not
446  * assume that zonepath is an existing directory or ZFS fs since we use
447  * this function as part of the process of creating a new ZFS fs or clone.
448  *
449  * The way this works is that we look at the parent directory of the zonepath
450  * to see if it is a ZFS fs.  If it is, we get the name of that ZFS fs and
451  * append the last component of the zonepath to generate the ZFS name for the
452  * zonepath.  This matches the algorithm that ZFS uses for automatically
453  * mounting a new fs after it is created.
454  *
455  * Although a ZFS fs can be mounted anywhere, we don't worry about handling
456  * all of the complexity that a user could possibly configure with arbitrary
457  * mounts since there is no way to generate a ZFS name from a random path in
458  * the file system.  We only try to handle the automatic mounts that ZFS does
459  * for each file system.  ZFS restricts this so that a new fs must be created
460  * in an existing parent ZFS fs.  It then automatically mounts the new fs
461  * directly under the mountpoint for the parent fs using the last component
462  * of the name as the mountpoint directory.
463  *
464  * For example:
465  *    Name			Mountpoint
466  *    space/eng/dev/test/zone1	/project1/eng/dev/test/zone1
467  *
468  * Return Z_OK if the path mapped to a ZFS file system name, otherwise return
469  * Z_ERR.
470  */
471 static int
472 path2name(char *zonepath, char *zfs_name, int len)
473 {
474 	int		res;
475 	char		*bnm, *dnm, *dname, *bname;
476 	zfs_handle_t	*zhp;
477 	struct stat	stbuf;
478 
479 	/*
480 	 * We need two tmp strings to handle paths directly in / (e.g. /foo)
481 	 * since dirname will overwrite the first char after "/" in this case.
482 	 */
483 	if ((bnm = strdup(zonepath)) == NULL)
484 		return (Z_ERR);
485 
486 	if ((dnm = strdup(zonepath)) == NULL) {
487 		free(bnm);
488 		return (Z_ERR);
489 	}
490 
491 	bname = basename(bnm);
492 	dname = dirname(dnm);
493 
494 	/*
495 	 * This is a quick test to save iterating over all of the zfs datasets
496 	 * on the system (which can be a lot).  If the parent dir is not in a
497 	 * ZFS fs, then we're done.
498 	 */
499 	if (stat(dname, &stbuf) != 0 || !S_ISDIR(stbuf.st_mode) ||
500 	    strcmp(stbuf.st_fstype, MNTTYPE_ZFS) != 0) {
501 		free(bnm);
502 		free(dnm);
503 		return (Z_ERR);
504 	}
505 
506 	/* See if the parent directory is its own ZFS dataset. */
507 	if ((zhp = mount2zhandle(dname)) == NULL) {
508 		/*
509 		 * The parent is not a ZFS dataset so we can't automatically
510 		 * create a dataset on the given path.
511 		 */
512 		free(bnm);
513 		free(dnm);
514 		return (Z_ERR);
515 	}
516 
517 	res = snprintf(zfs_name, len, "%s/%s", zfs_get_name(zhp), bname);
518 
519 	free(bnm);
520 	free(dnm);
521 	zfs_close(zhp);
522 	if (res >= len)
523 		return (Z_ERR);
524 
525 	return (Z_OK);
526 }
527 
528 /*
529  * A ZFS file system iterator call-back function used to determine if the
530  * file system has dependents (snapshots & clones).
531  */
532 /* ARGSUSED */
533 static int
534 has_dependent(zfs_handle_t *zhp, void *data)
535 {
536 	zfs_close(zhp);
537 	return (1);
538 }
539 
540 /*
541  * Given a snapshot name, get the file system path where the snapshot lives.
542  * A snapshot name is of the form fs_name@snap_name.  For example, snapshot
543  * pl/zones/z1@SUNWzone1 would have a path of
544  * /pl/zones/z1/.zfs/snapshot/SUNWzone1.
545  */
546 static int
547 snap2path(char *snap_name, char *path, int len)
548 {
549 	char		*p;
550 	zfs_handle_t	*zhp;
551 	char		mp[ZFS_MAXPROPLEN];
552 
553 	if ((p = strrchr(snap_name, '@')) == NULL)
554 		return (Z_ERR);
555 
556 	/* Get the file system name from the snap_name. */
557 	*p = '\0';
558 	zhp = zfs_open(g_zfs, snap_name, ZFS_TYPE_DATASET);
559 	*p = '@';
560 	if (zhp == NULL)
561 		return (Z_ERR);
562 
563 	/* Get the file system mount point. */
564 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, mp, sizeof (mp), NULL, NULL,
565 	    0, B_FALSE) != 0) {
566 		zfs_close(zhp);
567 		return (Z_ERR);
568 	}
569 	zfs_close(zhp);
570 
571 	p++;
572 	if (snprintf(path, len, "%s/.zfs/snapshot/%s", mp, p) >= len)
573 		return (Z_ERR);
574 
575 	return (Z_OK);
576 }
577 
578 /*
579  * This callback function is used to iterate through a snapshot's dependencies
580  * to find a filesystem that is a direct clone of the snapshot being iterated.
581  */
582 static int
583 get_direct_clone(zfs_handle_t *zhp, void *data)
584 {
585 	clone_data_t	*cd = data;
586 	char		origin[ZFS_MAXNAMELEN];
587 	char		ds_path[ZFS_MAXNAMELEN];
588 
589 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
590 		zfs_close(zhp);
591 		return (0);
592 	}
593 
594 	(void) strlcpy(ds_path, zfs_get_name(zhp), sizeof (ds_path));
595 
596 	/* Make sure this is a direct clone of the snapshot we're iterating. */
597 	if (zfs_prop_get(zhp, ZFS_PROP_ORIGIN, origin, sizeof (origin), NULL,
598 	    NULL, 0, B_FALSE) != 0 || strcmp(origin, cd->snapshot) != 0) {
599 		zfs_close(zhp);
600 		return (0);
601 	}
602 
603 	if (cd->clone_zhp != NULL)
604 		zfs_close(cd->clone_zhp);
605 
606 	cd->clone_zhp = zhp;
607 	return (1);
608 }
609 
610 /*
611  * A ZFS file system iterator call-back function used to determine the clone
612  * to promote.  This function finds the youngest (i.e. last one taken) snapshot
613  * that has a clone.  If found, it returns a reference to that clone in the
614  * callback data.
615  */
616 static int
617 find_clone(zfs_handle_t *zhp, void *data)
618 {
619 	clone_data_t	*cd = data;
620 	time_t		snap_creation;
621 	int		zret = 0;
622 
623 	/* If snapshot has no clones, skip it */
624 	if (zfs_prop_get_int(zhp, ZFS_PROP_NUMCLONES) == 0) {
625 		zfs_close(zhp);
626 		return (0);
627 	}
628 
629 	cd->snapshot = zfs_get_name(zhp);
630 
631 	/* Get the creation time of this snapshot */
632 	snap_creation = (time_t)zfs_prop_get_int(zhp, ZFS_PROP_CREATION);
633 
634 	/*
635 	 * If this snapshot's creation time is greater than (i.e. younger than)
636 	 * the current youngest snapshot found, iterate this snapshot to
637 	 * get the right clone.
638 	 */
639 	if (snap_creation >= cd->origin_creation) {
640 		/*
641 		 * Iterate the dependents of this snapshot to find a clone
642 		 * that's a direct dependent.
643 		 */
644 		if ((zret = zfs_iter_dependents(zhp, B_FALSE, get_direct_clone,
645 		    cd)) == -1) {
646 			zfs_close(zhp);
647 			return (1);
648 		} else if (zret == 1) {
649 			/*
650 			 * Found a clone, update the origin_creation time
651 			 * in the callback data.
652 			 */
653 			cd->origin_creation = snap_creation;
654 		}
655 	}
656 
657 	zfs_close(zhp);
658 	return (0);
659 }
660 
661 /*
662  * A ZFS file system iterator call-back function used to remove standalone
663  * snapshots.
664  */
665 /* ARGSUSED */
666 static int
667 rm_snap(zfs_handle_t *zhp, void *data)
668 {
669 	/* If snapshot has clones, something is wrong */
670 	if (zfs_prop_get_int(zhp, ZFS_PROP_NUMCLONES) != 0) {
671 		zfs_close(zhp);
672 		return (1);
673 	}
674 
675 	if (zfs_unmount(zhp, NULL, 0) == 0) {
676 		(void) zfs_destroy(zhp, B_FALSE);
677 	}
678 
679 	zfs_close(zhp);
680 	return (0);
681 }
682 
683 /*
684  * A ZFS snapshot iterator call-back function which renames snapshots.
685  */
686 static int
687 rename_snap(zfs_handle_t *zhp, void *data)
688 {
689 	int			res;
690 	zfs_snapshot_data_t	*cbp;
691 	char			template[ZFS_MAXNAMELEN];
692 
693 	cbp = (zfs_snapshot_data_t *)data;
694 
695 	/*
696 	 * When renaming snapshots with the iterator, the iterator can see
697 	 * the same snapshot after we've renamed up in the namespace.  To
698 	 * prevent this we check the count for the number of snapshots we have
699 	 * to rename and stop at that point.
700 	 */
701 	if (cbp->cntr >= cbp->num) {
702 		zfs_close(zhp);
703 		return (0);
704 	}
705 
706 	if (zfs_get_type(zhp) != ZFS_TYPE_SNAPSHOT) {
707 		zfs_close(zhp);
708 		return (0);
709 	}
710 
711 	/* Only rename the snapshots we automatically generate when we clone. */
712 	if (strncmp(zfs_get_name(zhp), cbp->match_name, cbp->len) != 0) {
713 		zfs_close(zhp);
714 		return (0);
715 	}
716 
717 	(void) snprintf(template, sizeof (template), "%s%d", cbp->match_name,
718 	    cbp->max++);
719 
720 	res = (zfs_rename(zhp, template, B_FALSE) != 0);
721 	if (res != 0)
722 		(void) fprintf(stderr, gettext("failed to rename snapshot %s "
723 		    "to %s: %s\n"), zfs_get_name(zhp), template,
724 		    libzfs_error_description(g_zfs));
725 
726 	cbp->cntr++;
727 
728 	zfs_close(zhp);
729 	return (res);
730 }
731 
732 /*
733  * Rename the source dataset's snapshots that are automatically generated when
734  * we clone a zone so that there won't be a name collision when we promote the
735  * cloned dataset.  Once the snapshots have been renamed, then promote the
736  * clone.
737  *
738  * The snapshot rename process gets the highest number on the snapshot names
739  * (the format is zonename@SUNWzoneXX where XX are digits) on both the source
740  * and clone datasets, then renames the source dataset snapshots starting at
741  * the next number.
742  */
743 static int
744 promote_clone(zfs_handle_t *src_zhp, zfs_handle_t *cln_zhp)
745 {
746 	zfs_snapshot_data_t	sd;
747 	char			nm[ZFS_MAXNAMELEN];
748 	char			template[ZFS_MAXNAMELEN];
749 
750 	(void) strlcpy(nm, zfs_get_name(cln_zhp), sizeof (nm));
751 	/*
752 	 * Start by getting the clone's snapshot max which we use
753 	 * during the rename of the original dataset's snapshots.
754 	 */
755 	(void) snprintf(template, sizeof (template), "%s@SUNWzone", nm);
756 	sd.match_name = template;
757 	sd.len = strlen(template);
758 	sd.max = 0;
759 
760 	if (zfs_iter_snapshots(cln_zhp, get_snap_max, &sd) != 0)
761 		return (Z_ERR);
762 
763 	/*
764 	 * Now make sure the source's snapshot max is at least as high as
765 	 * the clone's snapshot max.
766 	 */
767 	(void) snprintf(template, sizeof (template), "%s@SUNWzone",
768 	    zfs_get_name(src_zhp));
769 	sd.match_name = template;
770 	sd.len = strlen(template);
771 	sd.num = 0;
772 
773 	if (zfs_iter_snapshots(src_zhp, get_snap_max, &sd) != 0)
774 		return (Z_ERR);
775 
776 	/*
777 	 * Now rename the source dataset's snapshots so there's no
778 	 * conflict when we promote the clone.
779 	 */
780 	sd.max++;
781 	sd.cntr = 0;
782 	if (zfs_iter_snapshots(src_zhp, rename_snap, &sd) != 0)
783 		return (Z_ERR);
784 
785 	/* close and reopen the clone dataset to get the latest info */
786 	zfs_close(cln_zhp);
787 	if ((cln_zhp = zfs_open(g_zfs, nm, ZFS_TYPE_FILESYSTEM)) == NULL)
788 		return (Z_ERR);
789 
790 	if (zfs_promote(cln_zhp) != 0) {
791 		(void) fprintf(stderr, gettext("failed to promote %s: %s\n"),
792 		    nm, libzfs_error_description(g_zfs));
793 		return (Z_ERR);
794 	}
795 
796 	zfs_close(cln_zhp);
797 	return (Z_OK);
798 }
799 
800 /*
801  * Promote the youngest clone.  That clone will then become the origin of all
802  * of the other clones that were hanging off of the source dataset.
803  */
804 int
805 promote_all_clones(zfs_handle_t *zhp)
806 {
807 	clone_data_t	cd;
808 	char		nm[ZFS_MAXNAMELEN];
809 
810 	cd.clone_zhp = NULL;
811 	cd.origin_creation = 0;
812 	cd.snapshot = NULL;
813 
814 	if (zfs_iter_snapshots(zhp, find_clone, &cd) != 0) {
815 		zfs_close(zhp);
816 		return (Z_ERR);
817 	}
818 
819 	/* Nothing to promote. */
820 	if (cd.clone_zhp == NULL)
821 		return (Z_OK);
822 
823 	/* Found the youngest clone to promote.  Promote it. */
824 	if (promote_clone(zhp, cd.clone_zhp) != 0) {
825 		zfs_close(cd.clone_zhp);
826 		zfs_close(zhp);
827 		return (Z_ERR);
828 	}
829 
830 	/* close and reopen the main dataset to get the latest info */
831 	(void) strlcpy(nm, zfs_get_name(zhp), sizeof (nm));
832 	zfs_close(zhp);
833 	if ((zhp = zfs_open(g_zfs, nm, ZFS_TYPE_FILESYSTEM)) == NULL)
834 		return (Z_ERR);
835 
836 	return (Z_OK);
837 }
838 
839 /*
840  * Clone a pre-existing ZFS snapshot, either by making a direct ZFS clone, if
841  * possible, or by copying the data from the snapshot to the zonepath.
842  */
843 int
844 clone_snapshot_zfs(char *snap_name, char *zonepath, char *validatesnap)
845 {
846 	int	err = Z_OK;
847 	char	clone_name[MAXPATHLEN];
848 	char	snap_path[MAXPATHLEN];
849 
850 	if (snap2path(snap_name, snap_path, sizeof (snap_path)) != Z_OK) {
851 		(void) fprintf(stderr, gettext("unable to find path for %s.\n"),
852 		    snap_name);
853 		return (Z_ERR);
854 	}
855 
856 	if (validate_snapshot(snap_name, snap_path, validatesnap) != Z_OK)
857 		return (Z_NO_ENTRY);
858 
859 	/*
860 	 * The zonepath cannot be ZFS cloned, try to copy the data from
861 	 * within the snapshot to the zonepath.
862 	 */
863 	if (path2name(zonepath, clone_name, sizeof (clone_name)) != Z_OK) {
864 		if ((err = clone_copy(snap_path, zonepath)) == Z_OK)
865 			if (clean_out_clone() != Z_OK)
866 				(void) fprintf(stderr,
867 				    gettext("could not remove the "
868 				    "software inventory from %s\n"), zonepath);
869 
870 		return (err);
871 	}
872 
873 	if ((err = clone_snap(snap_name, clone_name)) != Z_OK) {
874 		if (err != Z_NO_ENTRY) {
875 			/*
876 			 * Cloning the snapshot failed.  Fall back to trying
877 			 * to install the zone by copying from the snapshot.
878 			 */
879 			if ((err = clone_copy(snap_path, zonepath)) == Z_OK)
880 				if (clean_out_clone() != Z_OK)
881 					(void) fprintf(stderr,
882 					    gettext("could not remove the "
883 					    "software inventory from %s\n"),
884 					    zonepath);
885 		} else {
886 			/*
887 			 * The snapshot is unusable for some reason so restore
888 			 * the zone state to configured since we were unable to
889 			 * actually do anything about getting the zone
890 			 * installed.
891 			 */
892 			int tmp;
893 
894 			if ((tmp = zone_set_state(target_zone,
895 			    ZONE_STATE_CONFIGURED)) != Z_OK) {
896 				errno = tmp;
897 				zperror2(target_zone,
898 				    gettext("could not set state"));
899 			}
900 		}
901 	}
902 
903 	return (err);
904 }
905 
906 /*
907  * Attempt to clone a source_zone to a target zonepath by using a ZFS clone.
908  */
909 int
910 clone_zfs(char *source_zonepath, char *zonepath, char *presnapbuf,
911     char *postsnapbuf)
912 {
913 	zfs_handle_t	*zhp;
914 	char		clone_name[MAXPATHLEN];
915 	char		snap_name[MAXPATHLEN];
916 
917 	/*
918 	 * Try to get a zfs handle for the source_zonepath.  If this fails
919 	 * the source_zonepath is not ZFS so return an error.
920 	 */
921 	if ((zhp = mount2zhandle(source_zonepath)) == NULL)
922 		return (Z_ERR);
923 
924 	/*
925 	 * Check if there is a file system already mounted on zonepath.  If so,
926 	 * we can't clone to the path so we should fall back to copying.
927 	 */
928 	if (is_mountpnt(zonepath)) {
929 		zfs_close(zhp);
930 		(void) fprintf(stderr,
931 		    gettext("A file system is already mounted on %s,\n"
932 		    "preventing use of a ZFS clone.\n"), zonepath);
933 		return (Z_ERR);
934 	}
935 
936 	/*
937 	 * Instead of using path2name to get the clone name from the zonepath,
938 	 * we could generate a name from the source zone ZFS name.  However,
939 	 * this would mean we would create the clone under the ZFS fs of the
940 	 * source instead of what the zonepath says.  For example,
941 	 *
942 	 * source_zonepath		zonepath
943 	 * /pl/zones/dev/z1		/pl/zones/deploy/z2
944 	 *
945 	 * We don't want the clone to be under "dev", we want it under
946 	 * "deploy", so that we can leverage the normal attribute inheritance
947 	 * that ZFS provides in the fs hierarchy.
948 	 */
949 	if (path2name(zonepath, clone_name, sizeof (clone_name)) != Z_OK) {
950 		zfs_close(zhp);
951 		return (Z_ERR);
952 	}
953 
954 	if (take_snapshot(zhp, snap_name, sizeof (snap_name), presnapbuf,
955 	    postsnapbuf) != Z_OK) {
956 		zfs_close(zhp);
957 		return (Z_ERR);
958 	}
959 	zfs_close(zhp);
960 
961 	if (clone_snap(snap_name, clone_name) != Z_OK) {
962 		/* Clean up the snapshot we just took. */
963 		if ((zhp = zfs_open(g_zfs, snap_name, ZFS_TYPE_SNAPSHOT))
964 		    != NULL) {
965 			if (zfs_unmount(zhp, NULL, 0) == 0)
966 				(void) zfs_destroy(zhp, B_FALSE);
967 			zfs_close(zhp);
968 		}
969 
970 		return (Z_ERR);
971 	}
972 
973 	(void) printf(gettext("Instead of copying, a ZFS clone has been "
974 	    "created for this zone.\n"));
975 
976 	return (Z_OK);
977 }
978 
979 /*
980  * Attempt to create a ZFS file system for the specified zonepath.
981  * We either will successfully create a ZFS file system and get it mounted
982  * on the zonepath or we don't.  The caller doesn't care since a regular
983  * directory is used for the zonepath if no ZFS file system is mounted there.
984  */
985 void
986 create_zfs_zonepath(char *zonepath)
987 {
988 	zfs_handle_t	*zhp;
989 	char		zfs_name[MAXPATHLEN];
990 	nvlist_t	*props = NULL;
991 
992 	if (path2name(zonepath, zfs_name, sizeof (zfs_name)) != Z_OK)
993 		return;
994 
995 	/* Check if the dataset already exists. */
996 	if ((zhp = zfs_open(g_zfs, zfs_name, ZFS_TYPE_DATASET)) != NULL) {
997 		zfs_close(zhp);
998 		return;
999 	}
1000 
1001 	if (nvlist_alloc(&props, NV_UNIQUE_NAME, 0) != 0 ||
1002 	    nvlist_add_string(props, zfs_prop_to_name(ZFS_PROP_SHARENFS),
1003 	    "off") != 0) {
1004 		if (props != NULL)
1005 			nvlist_free(props);
1006 		(void) fprintf(stderr, gettext("cannot create ZFS dataset %s: "
1007 		    "out of memory\n"), zfs_name);
1008 	}
1009 
1010 	if (zfs_create(g_zfs, zfs_name, ZFS_TYPE_FILESYSTEM, props) != 0 ||
1011 	    (zhp = zfs_open(g_zfs, zfs_name, ZFS_TYPE_DATASET)) == NULL) {
1012 		(void) fprintf(stderr, gettext("cannot create ZFS dataset %s: "
1013 		    "%s\n"), zfs_name, libzfs_error_description(g_zfs));
1014 		nvlist_free(props);
1015 		return;
1016 	}
1017 
1018 	nvlist_free(props);
1019 
1020 	if (zfs_mount(zhp, NULL, 0) != 0) {
1021 		(void) fprintf(stderr, gettext("cannot mount ZFS dataset %s: "
1022 		    "%s\n"), zfs_name, libzfs_error_description(g_zfs));
1023 		(void) zfs_destroy(zhp, B_FALSE);
1024 	} else {
1025 		if (chmod(zonepath, S_IRWXU) != 0) {
1026 			(void) fprintf(stderr, gettext("file system %s "
1027 			    "successfully created, but chmod %o failed: %s\n"),
1028 			    zfs_name, S_IRWXU, strerror(errno));
1029 			(void) destroy_zfs(zonepath);
1030 		} else {
1031 			(void) printf(gettext("A ZFS file system has been "
1032 			    "created for this zone.\n"));
1033 		}
1034 	}
1035 
1036 	zfs_close(zhp);
1037 }
1038 
1039 /*
1040  * If the zonepath is a ZFS file system, attempt to destroy it.  We return Z_OK
1041  * if we were able to zfs_destroy the zonepath, otherwise we return Z_ERR
1042  * which means the caller should clean up the zonepath in the traditional
1043  * way.
1044  */
1045 int
1046 destroy_zfs(char *zonepath)
1047 {
1048 	zfs_handle_t	*zhp;
1049 	boolean_t	is_clone = B_FALSE;
1050 	char		origin[ZFS_MAXPROPLEN];
1051 
1052 	if ((zhp = mount2zhandle(zonepath)) == NULL)
1053 		return (Z_ERR);
1054 
1055 	if (promote_all_clones(zhp) != 0)
1056 		return (Z_ERR);
1057 
1058 	/* Now cleanup any snapshots remaining. */
1059 	if (zfs_iter_snapshots(zhp, rm_snap, NULL) != 0) {
1060 		zfs_close(zhp);
1061 		return (Z_ERR);
1062 	}
1063 
1064 	/*
1065 	 * We can't destroy the file system if it has still has dependents.
1066 	 * There shouldn't be any at this point, but we'll double check.
1067 	 */
1068 	if (zfs_iter_dependents(zhp, B_TRUE, has_dependent, NULL) != 0) {
1069 		(void) fprintf(stderr, gettext("zfs destroy %s failed: the "
1070 		    "dataset still has dependents\n"), zfs_get_name(zhp));
1071 		zfs_close(zhp);
1072 		return (Z_ERR);
1073 	}
1074 
1075 	/*
1076 	 * This might be a clone.  Try to get the snapshot so we can attempt
1077 	 * to destroy that as well.
1078 	 */
1079 	if (zfs_prop_get(zhp, ZFS_PROP_ORIGIN, origin, sizeof (origin), NULL,
1080 	    NULL, 0, B_FALSE) == 0)
1081 		is_clone = B_TRUE;
1082 
1083 	if (zfs_unmount(zhp, NULL, 0) != 0) {
1084 		(void) fprintf(stderr, gettext("zfs unmount %s failed: %s\n"),
1085 		    zfs_get_name(zhp), libzfs_error_description(g_zfs));
1086 		zfs_close(zhp);
1087 		return (Z_ERR);
1088 	}
1089 
1090 	if (zfs_destroy(zhp, B_FALSE) != 0) {
1091 		/*
1092 		 * If the destroy fails for some reason, try to remount
1093 		 * the file system so that we can use "rm -rf" to clean up
1094 		 * instead.
1095 		 */
1096 		(void) fprintf(stderr, gettext("zfs destroy %s failed: %s\n"),
1097 		    zfs_get_name(zhp), libzfs_error_description(g_zfs));
1098 		(void) zfs_mount(zhp, NULL, 0);
1099 		zfs_close(zhp);
1100 		return (Z_ERR);
1101 	}
1102 
1103 	/*
1104 	 * If the zone has ever been moved then the mountpoint dir will not be
1105 	 * cleaned up by the zfs_destroy().  To handle this case try to clean
1106 	 * it up now but don't worry if it fails, that will be normal.
1107 	 */
1108 	(void) rmdir(zonepath);
1109 
1110 	(void) printf(gettext("The ZFS file system for this zone has been "
1111 	    "destroyed.\n"));
1112 
1113 	if (is_clone) {
1114 		zfs_handle_t	*ohp;
1115 
1116 		/*
1117 		 * Try to clean up the snapshot that the clone was taken from.
1118 		 */
1119 		if ((ohp = zfs_open(g_zfs, origin,
1120 		    ZFS_TYPE_SNAPSHOT)) != NULL) {
1121 			if (zfs_iter_dependents(ohp, B_TRUE, has_dependent,
1122 			    NULL) == 0 && zfs_unmount(ohp, NULL, 0) == 0)
1123 				(void) zfs_destroy(ohp, B_FALSE);
1124 			zfs_close(ohp);
1125 		}
1126 	}
1127 
1128 	zfs_close(zhp);
1129 	return (Z_OK);
1130 }
1131 
1132 /*
1133  * Return true if the path is its own zfs file system.  We determine this
1134  * by stat-ing the path to see if it is zfs and stat-ing the parent to see
1135  * if it is a different fs.
1136  */
1137 boolean_t
1138 is_zonepath_zfs(char *zonepath)
1139 {
1140 	int res;
1141 	char *path;
1142 	char *parent;
1143 	struct statvfs64 buf1, buf2;
1144 
1145 	if (statvfs64(zonepath, &buf1) != 0)
1146 		return (B_FALSE);
1147 
1148 	if (strcmp(buf1.f_basetype, "zfs") != 0)
1149 		return (B_FALSE);
1150 
1151 	if ((path = strdup(zonepath)) == NULL)
1152 		return (B_FALSE);
1153 
1154 	parent = dirname(path);
1155 	res = statvfs64(parent, &buf2);
1156 	free(path);
1157 
1158 	if (res != 0)
1159 		return (B_FALSE);
1160 
1161 	if (buf1.f_fsid == buf2.f_fsid)
1162 		return (B_FALSE);
1163 
1164 	return (B_TRUE);
1165 }
1166 
1167 /*
1168  * Implement the fast move of a ZFS file system by simply updating the
1169  * mountpoint.  Since it is file system already, we don't have the
1170  * issue of cross-file system copying.
1171  */
1172 int
1173 move_zfs(char *zonepath, char *new_zonepath)
1174 {
1175 	int		ret = Z_ERR;
1176 	zfs_handle_t	*zhp;
1177 
1178 	if ((zhp = mount2zhandle(zonepath)) == NULL)
1179 		return (Z_ERR);
1180 
1181 	if (zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1182 	    new_zonepath) == 0) {
1183 		/*
1184 		 * Clean up the old mount point.  We ignore any failure since
1185 		 * the zone is already successfully mounted on the new path.
1186 		 */
1187 		(void) rmdir(zonepath);
1188 		ret = Z_OK;
1189 	}
1190 
1191 	zfs_close(zhp);
1192 
1193 	return (ret);
1194 }
1195 
1196 /*
1197  * Validate that the given dataset exists on the system, and that neither it nor
1198  * its children are zvols.
1199  *
1200  * Note that we don't do anything with the 'zoned' property here.  All
1201  * management is done in zoneadmd when the zone is actually rebooted.  This
1202  * allows us to automatically set the zoned property even when a zone is
1203  * rebooted by the administrator.
1204  */
1205 int
1206 verify_datasets(zone_dochandle_t handle)
1207 {
1208 	int return_code = Z_OK;
1209 	struct zone_dstab dstab;
1210 	zfs_handle_t *zhp;
1211 	char propbuf[ZFS_MAXPROPLEN];
1212 	char source[ZFS_MAXNAMELEN];
1213 	zprop_source_t srctype;
1214 
1215 	if (zonecfg_setdsent(handle) != Z_OK) {
1216 		/*
1217 		 * TRANSLATION_NOTE
1218 		 * zfs and dataset are literals that should not be translated.
1219 		 */
1220 		(void) fprintf(stderr, gettext("could not verify zfs datasets: "
1221 		    "unable to enumerate datasets\n"));
1222 		return (Z_ERR);
1223 	}
1224 
1225 	while (zonecfg_getdsent(handle, &dstab) == Z_OK) {
1226 
1227 		if ((zhp = zfs_open(g_zfs, dstab.zone_dataset_name,
1228 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
1229 			(void) fprintf(stderr, gettext("could not verify zfs "
1230 			    "dataset %s: %s\n"), dstab.zone_dataset_name,
1231 			    libzfs_error_description(g_zfs));
1232 			return_code = Z_ERR;
1233 			continue;
1234 		}
1235 
1236 		if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, propbuf,
1237 		    sizeof (propbuf), &srctype, source,
1238 		    sizeof (source), 0) == 0 &&
1239 		    (srctype == ZPROP_SRC_INHERITED)) {
1240 			(void) fprintf(stderr, gettext("could not verify zfs "
1241 			    "dataset %s: mountpoint cannot be inherited\n"),
1242 			    dstab.zone_dataset_name);
1243 			return_code = Z_ERR;
1244 			zfs_close(zhp);
1245 			continue;
1246 		}
1247 
1248 		if (zfs_get_type(zhp) == ZFS_TYPE_VOLUME) {
1249 			(void) fprintf(stderr, gettext("cannot verify zfs "
1250 			    "dataset %s: volumes cannot be specified as a "
1251 			    "zone dataset resource\n"),
1252 			    dstab.zone_dataset_name);
1253 			return_code = Z_ERR;
1254 		}
1255 
1256 		if (zfs_iter_children(zhp, check_zvol, NULL) != 0)
1257 			return_code = Z_ERR;
1258 
1259 		zfs_close(zhp);
1260 	}
1261 	(void) zonecfg_enddsent(handle);
1262 
1263 	return (return_code);
1264 }
1265 
1266 /*
1267  * Verify that the ZFS dataset exists, and its mountpoint
1268  * property is set to "legacy".
1269  */
1270 int
1271 verify_fs_zfs(struct zone_fstab *fstab)
1272 {
1273 	zfs_handle_t *zhp;
1274 	char propbuf[ZFS_MAXPROPLEN];
1275 
1276 	if ((zhp = zfs_open(g_zfs, fstab->zone_fs_special,
1277 	    ZFS_TYPE_DATASET)) == NULL) {
1278 		(void) fprintf(stderr, gettext("could not verify fs %s: "
1279 		    "could not access zfs dataset '%s'\n"),
1280 		    fstab->zone_fs_dir, fstab->zone_fs_special);
1281 		return (Z_ERR);
1282 	}
1283 
1284 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
1285 		(void) fprintf(stderr, gettext("cannot verify fs %s: "
1286 		    "'%s' is not a file system\n"),
1287 		    fstab->zone_fs_dir, fstab->zone_fs_special);
1288 		zfs_close(zhp);
1289 		return (Z_ERR);
1290 	}
1291 
1292 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, propbuf, sizeof (propbuf),
1293 	    NULL, NULL, 0, 0) != 0 || strcmp(propbuf, "legacy") != 0) {
1294 		(void) fprintf(stderr, gettext("could not verify fs %s: "
1295 		    "zfs '%s' mountpoint is not \"legacy\"\n"),
1296 		    fstab->zone_fs_dir, fstab->zone_fs_special);
1297 		zfs_close(zhp);
1298 		return (Z_ERR);
1299 	}
1300 
1301 	zfs_close(zhp);
1302 	return (Z_OK);
1303 }
1304 
1305 /*
1306  * Destroy the specified mnttab structure that was created by mnttab_dup().
1307  * NOTE: The structure's mnt_time field isn't freed.
1308  */
1309 static void
1310 mnttab_destroy(struct mnttab *tabp)
1311 {
1312 	assert(tabp != NULL);
1313 
1314 	free(tabp->mnt_mountp);
1315 	free(tabp->mnt_special);
1316 	free(tabp->mnt_fstype);
1317 	free(tabp->mnt_mntopts);
1318 	free(tabp);
1319 }
1320 
1321 /*
1322  * Duplicate the specified mnttab structure.  The mnt_mountp and mnt_time
1323  * fields aren't duplicated.  This function returns a pointer to the new mnttab
1324  * structure or NULL if an error occurred.  If an error occurs, then this
1325  * function sets errno to reflect the error.  mnttab structures created by
1326  * this function should be destroyed via mnttab_destroy().
1327  */
1328 static struct mnttab *
1329 mnttab_dup(const struct mnttab *srcp)
1330 {
1331 	struct mnttab *retval;
1332 
1333 	assert(srcp != NULL);
1334 
1335 	retval = (struct mnttab *)calloc(1, sizeof (*retval));
1336 	if (retval == NULL) {
1337 		errno = ENOMEM;
1338 		return (NULL);
1339 	}
1340 	if (srcp->mnt_special != NULL) {
1341 		retval->mnt_special = strdup(srcp->mnt_special);
1342 		if (retval->mnt_special == NULL)
1343 			goto err;
1344 	}
1345 	if (srcp->mnt_fstype != NULL) {
1346 		retval->mnt_fstype = strdup(srcp->mnt_fstype);
1347 		if (retval->mnt_fstype == NULL)
1348 			goto err;
1349 	}
1350 	retval->mnt_mntopts = (char *)malloc(MAX_MNTOPT_STR * sizeof (char));
1351 	if (retval->mnt_mntopts == NULL)
1352 		goto err;
1353 	if (srcp->mnt_mntopts != NULL) {
1354 		if (strlcpy(retval->mnt_mntopts, srcp->mnt_mntopts,
1355 		    MAX_MNTOPT_STR * sizeof (char)) >= MAX_MNTOPT_STR *
1356 		    sizeof (char)) {
1357 			mnttab_destroy(retval);
1358 			errno = EOVERFLOW; /* similar to mount(2) behavior */
1359 			return (NULL);
1360 		}
1361 	} else {
1362 		retval->mnt_mntopts[0] = '\0';
1363 	}
1364 	return (retval);
1365 
1366 err:
1367 	mnttab_destroy(retval);
1368 	errno = ENOMEM;
1369 	return (NULL);
1370 }
1371 
1372 /*
1373  * Determine whether the specified ZFS dataset's mountpoint property is set
1374  * to "legacy".  If the specified dataset does not have a legacy mountpoint,
1375  * then the string pointer to which the mountpoint argument points is assigned
1376  * a dynamically-allocated string containing the dataset's mountpoint
1377  * property.  If the dataset's mountpoint property is "legacy" or a libzfs
1378  * error occurs, then the string pointer to which the mountpoint argument
1379  * points isn't modified.
1380  *
1381  * This function returns B_TRUE if it doesn't encounter any fatal errors.
1382  * It returns B_FALSE if it encounters a fatal error and sets errno to the
1383  * appropriate error code.
1384  */
1385 static boolean_t
1386 get_zfs_non_legacy_mountpoint(const char *dataset_name, char **mountpoint)
1387 {
1388 	zfs_handle_t *zhp;
1389 	char propbuf[ZFS_MAXPROPLEN];
1390 
1391 	assert(dataset_name != NULL);
1392 	assert(mountpoint != NULL);
1393 
1394 	if ((zhp = zfs_open(g_zfs, dataset_name, ZFS_TYPE_DATASET)) == NULL) {
1395 		errno = EINVAL;
1396 		return (B_FALSE);
1397 	}
1398 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, propbuf, sizeof (propbuf),
1399 	    NULL, NULL, 0, 0) != 0) {
1400 		zfs_close(zhp);
1401 		errno = EINVAL;
1402 		return (B_FALSE);
1403 	}
1404 	zfs_close(zhp);
1405 	if (strcmp(propbuf, "legacy") != 0) {
1406 		if ((*mountpoint = strdup(propbuf)) == NULL) {
1407 			errno = ENOMEM;
1408 			return (B_FALSE);
1409 		}
1410 	}
1411 	return (B_TRUE);
1412 }
1413 
1414 
1415 /*
1416  * This zonecfg_find_mounts() callback records information about mounts of
1417  * interest in a zonepath.  It also tallies the number of zone
1418  * root overlay mounts and the number of unexpected mounts found.
1419  * This function outputs errors using zerror() if it finds unexpected
1420  * mounts.  cookiep should point to an initialized zone_mounts_t structure.
1421  *
1422  * This function returns zero on success and a nonzero value on failure.
1423  */
1424 static int
1425 zone_mounts_cb(const struct mnttab *mountp, void *cookiep)
1426 {
1427 	zone_mounts_t *mounts;
1428 	const char *zone_mount_dir;
1429 
1430 	assert(mountp != NULL);
1431 	assert(cookiep != NULL);
1432 
1433 	mounts = (zone_mounts_t *)cookiep;
1434 	zone_mount_dir = mountp->mnt_mountp + mounts->zonepath_len;
1435 	if (strcmp(zone_mount_dir, "/root") == 0) {
1436 		/*
1437 		 * Check for an overlay mount.  If we already detected a /root
1438 		 * mount, then the current mount must be an overlay mount.
1439 		 */
1440 		if (mounts->root_mnttab != NULL) {
1441 			mounts->num_root_overlay_mounts++;
1442 			return (0);
1443 		}
1444 
1445 		/*
1446 		 * Store the root mount's mnttab information in the
1447 		 * zone_mounts_t structure for future use.
1448 		 */
1449 		if ((mounts->root_mnttab = mnttab_dup(mountp)) == NULL) {
1450 			zperror(cmd_to_str(CMD_MOVE), B_FALSE);
1451 			return (-1);
1452 		}
1453 
1454 		/*
1455 		 * Determine if the filesystem is a ZFS filesystem with a
1456 		 * non-legacy mountpoint.  If it is, then set the root
1457 		 * filesystem's mnttab's mnt_mountp field to a non-NULL
1458 		 * value, which will serve as a flag to indicate this special
1459 		 * condition.
1460 		 */
1461 		if (strcmp(mountp->mnt_fstype, MNTTYPE_ZFS) == 0 &&
1462 		    get_zfs_non_legacy_mountpoint(mountp->mnt_special,
1463 		    &mounts->root_mnttab->mnt_mountp) != B_TRUE) {
1464 			zperror(cmd_to_str(CMD_MOVE), B_FALSE);
1465 			return (-1);
1466 		}
1467 	} else {
1468 		/*
1469 		 * An unexpected mount was found.  Notify the user.
1470 		 */
1471 		if (mounts->num_unexpected_mounts == 0)
1472 			zerror(gettext("These file systems are mounted on "
1473 			    "subdirectories of %s.\n"), mounts->zonepath);
1474 		mounts->num_unexpected_mounts++;
1475 		(void) zfm_print(mountp, NULL);
1476 	}
1477 	return (0);
1478 }
1479 
1480 /*
1481  * Initialize the specified zone_mounts_t structure for the given zonepath.
1482  * If this function succeeds, it returns zero and the specified zone_mounts_t
1483  * structure contains information about mounts in the specified zonepath.
1484  * The function returns a nonzero value if it fails.  The zone_mounts_t
1485  * structure doesn't need be destroyed via zone_mounts_destroy() if this
1486  * function fails.
1487  */
1488 int
1489 zone_mounts_init(zone_mounts_t *mounts, const char *zonepath)
1490 {
1491 	assert(mounts != NULL);
1492 	assert(zonepath != NULL);
1493 
1494 	bzero(mounts, sizeof (*mounts));
1495 	if ((mounts->zonepath = strdup(zonepath)) == NULL) {
1496 		zerror(gettext("the process ran out of memory while checking "
1497 		    "for mounts in zonepath %s."), zonepath);
1498 		return (-1);
1499 	}
1500 	mounts->zonepath_len = strlen(zonepath);
1501 	if (zonecfg_find_mounts((char *)zonepath, zone_mounts_cb, mounts) ==
1502 	    -1) {
1503 		zerror(gettext("an error occurred while checking for mounts "
1504 		    "in zonepath %s."), zonepath);
1505 		zone_mounts_destroy(mounts);
1506 		return (-1);
1507 	}
1508 	return (0);
1509 }
1510 
1511 /*
1512  * Destroy the memory used by the specified zone_mounts_t structure's fields.
1513  * This function doesn't free the memory occupied by the structure itself
1514  * (i.e., it doesn't free the parameter).
1515  */
1516 void
1517 zone_mounts_destroy(zone_mounts_t *mounts)
1518 {
1519 	assert(mounts != NULL);
1520 
1521 	free(mounts->zonepath);
1522 	if (mounts->root_mnttab != NULL)
1523 		mnttab_destroy(mounts->root_mnttab);
1524 }
1525 
1526 /*
1527  * Mount a moving zone's root filesystem (if it had a root filesystem mount
1528  * prior to the move) using the specified zonepath.  mounts should refer to
1529  * the zone_mounts_t structure describing the zone's mount information.
1530  *
1531  * This function returns zero if the mount succeeds and a nonzero value
1532  * if it doesn't.
1533  */
1534 int
1535 zone_mount_rootfs(zone_mounts_t *mounts, const char *zonepath)
1536 {
1537 	char zoneroot[MAXPATHLEN];
1538 	struct mnttab *mtab;
1539 	int flags;
1540 
1541 	assert(mounts != NULL);
1542 	assert(zonepath != NULL);
1543 
1544 	/*
1545 	 * If there isn't a root filesystem, then don't do anything.
1546 	 */
1547 	mtab = mounts->root_mnttab;
1548 	if (mtab == NULL)
1549 		return (0);
1550 
1551 	/*
1552 	 * Determine the root filesystem's new mountpoint.
1553 	 */
1554 	if (snprintf(zoneroot, sizeof (zoneroot), "%s/root", zonepath) >=
1555 	    sizeof (zoneroot)) {
1556 		zerror(gettext("Zonepath %s is too long.\n"), zonepath);
1557 		return (-1);
1558 	}
1559 
1560 	/*
1561 	 * If the root filesystem is a non-legacy ZFS filesystem (i.e., if it's
1562 	 * mnt_mountp field is non-NULL), then make the filesystem's new
1563 	 * mount point its mountpoint property and mount the filesystem.
1564 	 */
1565 	if (mtab->mnt_mountp != NULL) {
1566 		zfs_handle_t *zhp;
1567 
1568 		if ((zhp = zfs_open(g_zfs, mtab->mnt_special,
1569 		    ZFS_TYPE_DATASET)) == NULL) {
1570 			zerror(gettext("could not get ZFS handle for the zone's"
1571 			    " root filesystem"));
1572 			return (-1);
1573 		}
1574 		if (zfs_prop_set(zhp, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1575 		    zoneroot) != 0) {
1576 			zerror(gettext("could not modify zone's root "
1577 			    "filesystem's mountpoint property"));
1578 			zfs_close(zhp);
1579 			return (-1);
1580 		}
1581 		if (zfs_mount(zhp, mtab->mnt_mntopts, 0) != 0) {
1582 			zerror(gettext("unable to mount zone root %s: %s"),
1583 			    zoneroot, libzfs_error_description(g_zfs));
1584 			if (zfs_prop_set(zhp,
1585 			    zfs_prop_to_name(ZFS_PROP_MOUNTPOINT),
1586 			    mtab->mnt_mountp) != 0)
1587 				zerror(gettext("unable to restore zone's root "
1588 				    "filesystem's mountpoint property"));
1589 			zfs_close(zhp);
1590 			return (-1);
1591 		}
1592 		zfs_close(zhp);
1593 		return (0);
1594 	}
1595 
1596 	/*
1597 	 * The root filesystem is either a legacy-mounted ZFS filesystem or
1598 	 * a non-ZFS filesystem.  Use mount(2) to mount the root filesystem.
1599 	 */
1600 	if (mtab->mnt_mntopts != NULL)
1601 		flags = MS_OPTIONSTR;
1602 	else
1603 		flags = 0;
1604 	if (mount(mtab->mnt_special, zoneroot, flags, mtab->mnt_fstype, NULL, 0,
1605 	    mtab->mnt_mntopts, MAX_MNTOPT_STR * sizeof (char)) != 0) {
1606 		flags = errno;
1607 		zerror(gettext("unable to mount zone root %s: %s"), zoneroot,
1608 		    strerror(flags));
1609 		return (-1);
1610 	}
1611 	return (0);
1612 }
1613 
1614 /*
1615  * Unmount a moving zone's root filesystem (if such a mount exists) using the
1616  * specified zonepath.  mounts should refer to the zone_mounts_t structure
1617  * describing the zone's mount information.  If force is B_TRUE, then if the
1618  * unmount fails, then the function will try to forcibly unmount the zone's root
1619  * filesystem.
1620  *
1621  * This function returns zero if the unmount (forced or otherwise) succeeds;
1622  * otherwise, it returns a nonzero value.
1623  */
1624 int
1625 zone_unmount_rootfs(zone_mounts_t *mounts, const char *zonepath,
1626     boolean_t force)
1627 {
1628 	char zoneroot[MAXPATHLEN];
1629 	struct mnttab *mtab;
1630 	int err;
1631 
1632 	assert(mounts != NULL);
1633 	assert(zonepath != NULL);
1634 
1635 	/*
1636 	 * If there isn't a root filesystem, then don't do anything.
1637 	 */
1638 	mtab = mounts->root_mnttab;
1639 	if (mtab == NULL)
1640 		return (0);
1641 
1642 	/*
1643 	 * Determine the root filesystem's mountpoint.
1644 	 */
1645 	if (snprintf(zoneroot, sizeof (zoneroot), "%s/root", zonepath) >=
1646 	    sizeof (zoneroot)) {
1647 		zerror(gettext("Zonepath %s is too long.\n"), zonepath);
1648 		return (-1);
1649 	}
1650 
1651 	/*
1652 	 * If the root filesystem is a non-legacy ZFS fileystem, then unmount
1653 	 * the filesystem via libzfs.
1654 	 */
1655 	if (mtab->mnt_mountp != NULL) {
1656 		zfs_handle_t *zhp;
1657 
1658 		if ((zhp = zfs_open(g_zfs, mtab->mnt_special,
1659 		    ZFS_TYPE_DATASET)) == NULL) {
1660 			zerror(gettext("could not get ZFS handle for the zone's"
1661 			    " root filesystem"));
1662 			return (-1);
1663 		}
1664 		if (zfs_unmount(zhp, zoneroot, 0) != 0) {
1665 			if (force && zfs_unmount(zhp, zoneroot, MS_FORCE) ==
1666 			    0) {
1667 				zfs_close(zhp);
1668 				return (0);
1669 			}
1670 			zerror(gettext("unable to unmount zone root %s: %s"),
1671 			    zoneroot, libzfs_error_description(g_zfs));
1672 			zfs_close(zhp);
1673 			return (-1);
1674 		}
1675 		zfs_close(zhp);
1676 		return (0);
1677 	}
1678 
1679 	/*
1680 	 * Use umount(2) to unmount the root filesystem.  If this fails, then
1681 	 * forcibly unmount it if the force flag is set.
1682 	 */
1683 	if (umount(zoneroot) != 0) {
1684 		if (force && umount2(zoneroot, MS_FORCE) == 0)
1685 			return (0);
1686 		err = errno;
1687 		zerror(gettext("unable to unmount zone root %s: %s"), zoneroot,
1688 		    strerror(err));
1689 		return (-1);
1690 	}
1691 	return (0);
1692 }
1693 
1694 int
1695 init_zfs(void)
1696 {
1697 	if ((g_zfs = libzfs_init()) == NULL) {
1698 		(void) fprintf(stderr, gettext("failed to initialize ZFS "
1699 		    "library\n"));
1700 		return (Z_ERR);
1701 	}
1702 
1703 	return (Z_OK);
1704 }
1705