xref: /illumos-gate/usr/src/cmd/zoneadm/zfs.c (revision 3f2f09c1)
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 2006 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 /*
30  * This file contains the functions used to support the ZFS integration
31  * with zones.  This includes validation (e.g. zonecfg dataset), cloning,
32  * file system creation and destruction.
33  */
34 
35 #include <stdio.h>
36 #include <errno.h>
37 #include <unistd.h>
38 #include <string.h>
39 #include <locale.h>
40 #include <libintl.h>
41 #include <sys/stat.h>
42 #include <sys/statvfs.h>
43 #include <libgen.h>
44 #include <libzonecfg.h>
45 #include <sys/mnttab.h>
46 #include <libzfs.h>
47 
48 #include "zoneadm.h"
49 
50 libzfs_handle_t *g_zfs;
51 
52 typedef struct zfs_mount_data {
53 	char		*match_name;
54 	zfs_handle_t	*match_handle;
55 } zfs_mount_data_t;
56 
57 typedef struct zfs_snapshot_data {
58 	char	*match_name;
59 	int	len;
60 	int	max;
61 } zfs_snapshot_data_t;
62 
63 /*
64  * A ZFS file system iterator call-back function which is used to validate
65  * datasets imported into the zone.
66  */
67 /* ARGSUSED */
68 static int
69 check_zvol(zfs_handle_t *zhp, void *unused)
70 {
71 	int ret;
72 
73 	if (zfs_get_type(zhp) == ZFS_TYPE_VOLUME) {
74 		/*
75 		 * TRANSLATION_NOTE
76 		 * zfs and dataset are literals that should not be translated.
77 		 */
78 		(void) fprintf(stderr, gettext("cannot verify zfs dataset %s: "
79 		    "volumes cannot be specified as a zone dataset resource\n"),
80 		    zfs_get_name(zhp));
81 		ret = -1;
82 	} else {
83 		ret = zfs_iter_children(zhp, check_zvol, NULL);
84 	}
85 
86 	zfs_close(zhp);
87 
88 	return (ret);
89 }
90 
91 /*
92  * A ZFS file system iterator call-back function which returns the
93  * zfs_handle_t for a ZFS file system on the specified mount point.
94  */
95 static int
96 match_mountpoint(zfs_handle_t *zhp, void *data)
97 {
98 	int			res;
99 	zfs_mount_data_t	*cbp;
100 	char			mp[ZFS_MAXPROPLEN];
101 
102 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
103 		zfs_close(zhp);
104 		return (0);
105 	}
106 
107 	cbp = (zfs_mount_data_t *)data;
108 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, mp, sizeof (mp), NULL, NULL,
109 	    0, B_FALSE) == 0 && strcmp(mp, cbp->match_name) == 0) {
110 		cbp->match_handle = zhp;
111 		return (1);
112 	}
113 
114 	res = zfs_iter_filesystems(zhp, match_mountpoint, data);
115 	zfs_close(zhp);
116 	return (res);
117 }
118 
119 /*
120  * Get ZFS handle for the specified mount point.
121  */
122 static zfs_handle_t *
123 mount2zhandle(char *mountpoint)
124 {
125 	zfs_mount_data_t	cb;
126 
127 	cb.match_name = mountpoint;
128 	cb.match_handle = NULL;
129 	(void) zfs_iter_root(g_zfs, match_mountpoint, &cb);
130 	return (cb.match_handle);
131 }
132 
133 /*
134  * Check if there is already a file system (zfs or any other type) mounted on
135  * path.
136  */
137 static boolean_t
138 is_mountpnt(char *path)
139 {
140 	FILE		*fp;
141 	struct mnttab	entry;
142 
143 	if ((fp = fopen("/etc/mnttab", "r")) == NULL)
144 		return (B_FALSE);
145 
146 	while (getmntent(fp, &entry) == 0) {
147 		if (strcmp(path, entry.mnt_mountp) == 0) {
148 			(void) fclose(fp);
149 			return (B_TRUE);
150 		}
151 	}
152 
153 	(void) fclose(fp);
154 	return (B_FALSE);
155 }
156 
157 /*
158  * Perform any necessary housekeeping tasks we need to do before we take
159  * a ZFS snapshot of the zone.  What this really entails is that we are
160  * taking a sw inventory of the source zone, like we do when we detach,
161  * so that there is the XML manifest in the snapshot.  We use that to
162  * validate the snapshot if it is the source of a clone at some later time.
163  */
164 static int
165 pre_snapshot(char *source_zone)
166 {
167 	int err;
168 	zone_dochandle_t handle;
169 
170 	if ((handle = zonecfg_init_handle()) == NULL) {
171 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
172 		return (Z_ERR);
173 	}
174 
175 	if ((err = zonecfg_get_handle(source_zone, handle)) != Z_OK) {
176 		errno = err;
177 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
178 		zonecfg_fini_handle(handle);
179 		return (Z_ERR);
180 	}
181 
182 	if ((err = zonecfg_get_detach_info(handle, B_TRUE)) != Z_OK) {
183 		errno = err;
184 		zperror(gettext("getting the software version information "
185 		    "failed"), B_TRUE);
186 		zonecfg_fini_handle(handle);
187 		return (Z_ERR);
188 	}
189 
190 	if ((err = zonecfg_detach_save(handle, 0)) != Z_OK) {
191 		errno = err;
192 		zperror(gettext("saving the software version manifest failed"),
193 		    B_TRUE);
194 		zonecfg_fini_handle(handle);
195 		return (Z_ERR);
196 	}
197 
198 	zonecfg_fini_handle(handle);
199 	return (Z_OK);
200 }
201 
202 /*
203  * Perform any necessary housekeeping tasks we need to do after we take
204  * a ZFS snapshot of the zone.  What this really entails is removing the
205  * sw inventory XML file from the zone.  It is still in the snapshot where
206  * we want it, but we don't want it in the source zone itself.
207  */
208 static int
209 post_snapshot(char *source_zone)
210 {
211 	int err;
212 	zone_dochandle_t handle;
213 
214 	if ((handle = zonecfg_init_handle()) == NULL) {
215 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
216 		return (Z_ERR);
217 	}
218 
219 	if ((err = zonecfg_get_handle(source_zone, handle)) != Z_OK) {
220 		errno = err;
221 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
222 		zonecfg_fini_handle(handle);
223 		return (Z_ERR);
224 	}
225 
226 	zonecfg_rm_detached(handle, B_FALSE);
227 	zonecfg_fini_handle(handle);
228 
229 	return (Z_OK);
230 }
231 
232 /*
233  * This is a ZFS snapshot iterator call-back function which returns the
234  * highest number of SUNWzone snapshots that have been taken.
235  */
236 static int
237 get_snap_max(zfs_handle_t *zhp, void *data)
238 {
239 	int			res;
240 	zfs_snapshot_data_t	*cbp;
241 
242 	if (zfs_get_type(zhp) != ZFS_TYPE_SNAPSHOT) {
243 		zfs_close(zhp);
244 		return (0);
245 	}
246 
247 	cbp = (zfs_snapshot_data_t *)data;
248 
249 	if (strncmp(zfs_get_name(zhp), cbp->match_name, cbp->len) == 0) {
250 		char	*nump;
251 		int	num;
252 
253 		nump = (char *)(zfs_get_name(zhp) + cbp->len);
254 		num = atoi(nump);
255 		if (num > cbp->max)
256 			cbp->max = num;
257 	}
258 
259 	res = zfs_iter_snapshots(zhp, get_snap_max, data);
260 	zfs_close(zhp);
261 	return (res);
262 }
263 
264 /*
265  * Take a ZFS snapshot to be used for cloning the zone.
266  */
267 static int
268 take_snapshot(char *source_zone, zfs_handle_t *zhp, char *snapshot_name,
269     int snap_size)
270 {
271 	int			res;
272 	char			template[ZFS_MAXNAMELEN];
273 	zfs_snapshot_data_t	cb;
274 
275 	/*
276 	 * First we need to figure out the next available name for the
277 	 * zone snapshot.  Look through the list of zones snapshots for
278 	 * this file system to determine the maximum snapshot name.
279 	 */
280 	if (snprintf(template, sizeof (template), "%s@SUNWzone",
281 	    zfs_get_name(zhp)) >=  sizeof (template))
282 		return (Z_ERR);
283 
284 	cb.match_name = template;
285 	cb.len = strlen(template);
286 	cb.max = 0;
287 
288 	if (zfs_iter_snapshots(zhp, get_snap_max, &cb) != 0)
289 		return (Z_ERR);
290 
291 	cb.max++;
292 
293 	if (snprintf(snapshot_name, snap_size, "%s@SUNWzone%d",
294 	    zfs_get_name(zhp), cb.max) >= snap_size)
295 		return (Z_ERR);
296 
297 	if (pre_snapshot(source_zone) != Z_OK)
298 		return (Z_ERR);
299 	res = zfs_snapshot(g_zfs, snapshot_name, B_FALSE);
300 	if (post_snapshot(source_zone) != Z_OK)
301 		return (Z_ERR);
302 
303 	if (res != 0)
304 		return (Z_ERR);
305 	return (Z_OK);
306 }
307 
308 /*
309  * We are using an explicit snapshot from some earlier point in time so
310  * we need to validate it.  This involves checking the sw inventory that
311  * we took when we made the snapshot to verify that the current sw config
312  * on the host is still valid to run a zone made from this snapshot.
313  */
314 static int
315 validate_snapshot(char *snapshot_name, char *snap_path)
316 {
317 	int err;
318 	zone_dochandle_t handle;
319 	zone_dochandle_t athandle = NULL;
320 
321 	if ((handle = zonecfg_init_handle()) == NULL) {
322 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
323 		return (Z_ERR);
324 	}
325 
326 	if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) {
327 		errno = err;
328 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
329 		zonecfg_fini_handle(handle);
330 		return (Z_ERR);
331 	}
332 
333 	if ((athandle = zonecfg_init_handle()) == NULL) {
334 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
335 		goto done;
336 	}
337 
338 	if ((err = zonecfg_get_attach_handle(snap_path, target_zone, B_TRUE,
339 	    athandle)) != Z_OK) {
340 		if (err == Z_NO_ZONE)
341 			(void) fprintf(stderr, gettext("snapshot %s was not "
342 			    "taken\n\tby a 'zoneadm clone' command.  It can "
343 			    "not be used to clone zones.\n"), snapshot_name);
344 		else
345 			(void) fprintf(stderr, gettext("snapshot %s is "
346 			    "out-dated\n\tIt can no longer be used to clone "
347 			    "zones on this system.\n"), snapshot_name);
348 		goto done;
349 	}
350 
351 	/* Get the detach information for the locally defined zone. */
352 	if ((err = zonecfg_get_detach_info(handle, B_FALSE)) != Z_OK) {
353 		errno = err;
354 		zperror(gettext("getting the attach information failed"),
355 		    B_TRUE);
356 		goto done;
357 	}
358 
359 	if ((err = sw_cmp(handle, athandle, SW_CMP_SILENT)) != Z_OK)
360 		(void) fprintf(stderr, gettext("snapshot %s is out-dated\n\t"
361 		    "It can no longer be used to clone zones on this "
362 		    "system.\n"), snapshot_name);
363 
364 done:
365 	zonecfg_fini_handle(handle);
366 	if (athandle != NULL)
367 		zonecfg_fini_handle(athandle);
368 
369 	return (err);
370 }
371 
372 /*
373  * Remove the sw inventory file from inside this zonepath that we picked up out
374  * of the snapshot.
375  */
376 static int
377 clean_out_clone()
378 {
379 	int err;
380 	zone_dochandle_t handle;
381 
382 	if ((handle = zonecfg_init_handle()) == NULL) {
383 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
384 		return (Z_ERR);
385 	}
386 
387 	if ((err = zonecfg_get_handle(target_zone, handle)) != Z_OK) {
388 		errno = err;
389 		zperror(cmd_to_str(CMD_CLONE), B_TRUE);
390 		zonecfg_fini_handle(handle);
391 		return (Z_ERR);
392 	}
393 
394 	zonecfg_rm_detached(handle, B_FALSE);
395 	zonecfg_fini_handle(handle);
396 
397 	return (Z_OK);
398 }
399 
400 /*
401  * Make a ZFS clone on zonepath from snapshot_name.
402  */
403 static int
404 clone_snap(char *snapshot_name, char *zonepath)
405 {
406 	int		res = Z_OK;
407 	int		err;
408 	zfs_handle_t	*zhp;
409 	zfs_handle_t	*clone;
410 
411 	if ((zhp = zfs_open(g_zfs, snapshot_name, ZFS_TYPE_SNAPSHOT)) == NULL)
412 		return (Z_NO_ENTRY);
413 
414 	(void) printf(gettext("Cloning snapshot %s\n"), snapshot_name);
415 
416 	err = zfs_clone(zhp, zonepath);
417 	zfs_close(zhp);
418 	if (err != 0)
419 		return (Z_ERR);
420 
421 	/* create the mountpoint if necessary */
422 	if ((clone = zfs_open(g_zfs, zonepath, ZFS_TYPE_ANY)) == NULL)
423 		return (Z_ERR);
424 
425 	/*
426 	 * The clone has been created so we need to print a diagnostic
427 	 * message if one of the following steps fails for some reason.
428 	 */
429 	if (zfs_mount(clone, NULL, 0) != 0) {
430 		(void) fprintf(stderr, gettext("could not mount ZFS clone "
431 		    "%s\n"), zfs_get_name(clone));
432 		res = Z_ERR;
433 
434 	} else {
435 		if (zfs_prop_set(clone, ZFS_PROP_SHARENFS, "off") != 0) {
436 			/* we won't consider this a failure */
437 			(void) fprintf(stderr, gettext("could not turn off the "
438 			    "'sharenfs' property on ZFS clone %s\n"),
439 			    zfs_get_name(clone));
440 		}
441 
442 		if (clean_out_clone() != Z_OK) {
443 			(void) fprintf(stderr, gettext("could not remove the "
444 			    "software inventory from ZFS clone %s\n"),
445 			    zfs_get_name(clone));
446 			res = Z_ERR;
447 		}
448 	}
449 
450 	zfs_close(clone);
451 	return (res);
452 }
453 
454 /*
455  * This function takes a zonepath and attempts to determine what the ZFS
456  * file system name (not mountpoint) should be for that path.  We do not
457  * assume that zonepath is an existing directory or ZFS fs since we use
458  * this function as part of the process of creating a new ZFS fs or clone.
459  *
460  * The way this works is that we look at the parent directory of the zonepath
461  * to see if it is a ZFS fs.  If it is, we get the name of that ZFS fs and
462  * append the last component of the zonepath to generate the ZFS name for the
463  * zonepath.  This matches the algorithm that ZFS uses for automatically
464  * mounting a new fs after it is created.
465  *
466  * Although a ZFS fs can be mounted anywhere, we don't worry about handling
467  * all of the complexity that a user could possibly configure with arbitrary
468  * mounts since there is no way to generate a ZFS name from a random path in
469  * the file system.  We only try to handle the automatic mounts that ZFS does
470  * for each file system.  ZFS restricts this so that a new fs must be created
471  * in an existing parent ZFS fs.  It then automatically mounts the new fs
472  * directly under the mountpoint for the parent fs using the last component
473  * of the name as the mountpoint directory.
474  *
475  * For example:
476  *    Name			Mountpoint
477  *    space/eng/dev/test/zone1	/project1/eng/dev/test/zone1
478  *
479  * Return Z_OK if the path mapped to a ZFS file system name, otherwise return
480  * Z_ERR.
481  */
482 static int
483 path2name(char *zonepath, char *zfs_name, int len)
484 {
485 	int		res;
486 	char		*p;
487 	zfs_handle_t	*zhp;
488 
489 	if ((p = strrchr(zonepath, '/')) == NULL)
490 		return (Z_ERR);
491 
492 	/*
493 	 * If the parent directory is not its own ZFS fs, then we can't
494 	 * automatically create a new ZFS fs at the 'zonepath' mountpoint
495 	 * so return an error.
496 	 */
497 	*p = '\0';
498 	zhp = mount2zhandle(zonepath);
499 	*p = '/';
500 	if (zhp == NULL)
501 		return (Z_ERR);
502 
503 	res = snprintf(zfs_name, len, "%s/%s", zfs_get_name(zhp), p + 1);
504 
505 	zfs_close(zhp);
506 	if (res >= len)
507 		return (Z_ERR);
508 
509 	return (Z_OK);
510 }
511 
512 /*
513  * A ZFS file system iterator call-back function used to determine if the
514  * file system has dependents (snapshots & clones).
515  */
516 /* ARGSUSED */
517 static int
518 has_dependent(zfs_handle_t *zhp, void *data)
519 {
520 	zfs_close(zhp);
521 	return (1);
522 }
523 
524 /*
525  * Given a snapshot name, get the file system path where the snapshot lives.
526  * A snapshot name is of the form fs_name@snap_name.  For example, snapshot
527  * pl/zones/z1@SUNWzone1 would have a path of
528  * /pl/zones/z1/.zfs/snapshot/SUNWzone1.
529  */
530 static int
531 snap2path(char *snap_name, char *path, int len)
532 {
533 	char		*p;
534 	zfs_handle_t	*zhp;
535 	char		mp[ZFS_MAXPROPLEN];
536 
537 	if ((p = strrchr(snap_name, '@')) == NULL)
538 		return (Z_ERR);
539 
540 	/* Get the file system name from the snap_name. */
541 	*p = '\0';
542 	zhp = zfs_open(g_zfs, snap_name, ZFS_TYPE_ANY);
543 	*p = '@';
544 	if (zhp == NULL)
545 		return (Z_ERR);
546 
547 	/* Get the file system mount point. */
548 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, mp, sizeof (mp), NULL, NULL,
549 	    0, B_FALSE) != 0) {
550 		zfs_close(zhp);
551 		return (Z_ERR);
552 	}
553 	zfs_close(zhp);
554 
555 	p++;
556 	if (snprintf(path, len, "%s/.zfs/snapshot/%s", mp, p) >= len)
557 		return (Z_ERR);
558 
559 	return (Z_OK);
560 }
561 
562 /*
563  * Clone a pre-existing ZFS snapshot, either by making a direct ZFS clone, if
564  * possible, or by copying the data from the snapshot to the zonepath.
565  */
566 int
567 clone_snapshot_zfs(char *snap_name, char *zonepath)
568 {
569 	int	err = Z_OK;
570 	char	clone_name[MAXPATHLEN];
571 	char	snap_path[MAXPATHLEN];
572 
573 	if (snap2path(snap_name, snap_path, sizeof (snap_path)) != Z_OK) {
574 		(void) fprintf(stderr, gettext("unable to find path for %s.\n"),
575 		    snap_name);
576 		return (Z_ERR);
577 	}
578 
579 	if (validate_snapshot(snap_name, snap_path) != Z_OK)
580 		return (Z_NO_ENTRY);
581 
582 	/*
583 	 * The zonepath cannot be ZFS cloned, try to copy the data from
584 	 * within the snapshot to the zonepath.
585 	 */
586 	if (path2name(zonepath, clone_name, sizeof (clone_name)) != Z_OK) {
587 		if ((err = clone_copy(snap_path, zonepath)) == Z_OK)
588 			if (clean_out_clone() != Z_OK)
589 				(void) fprintf(stderr,
590 				    gettext("could not remove the "
591 				    "software inventory from %s\n"), zonepath);
592 
593 		return (err);
594 	}
595 
596 	if ((err = clone_snap(snap_name, clone_name)) != Z_OK) {
597 		if (err != Z_NO_ENTRY) {
598 			/*
599 			 * Cloning the snapshot failed.  Fall back to trying
600 			 * to install the zone by copying from the snapshot.
601 			 */
602 			if ((err = clone_copy(snap_path, zonepath)) == Z_OK)
603 				if (clean_out_clone() != Z_OK)
604 					(void) fprintf(stderr,
605 					    gettext("could not remove the "
606 					    "software inventory from %s\n"),
607 					    zonepath);
608 		} else {
609 			/*
610 			 * The snapshot is unusable for some reason so restore
611 			 * the zone state to configured since we were unable to
612 			 * actually do anything about getting the zone
613 			 * installed.
614 			 */
615 			int tmp;
616 
617 			if ((tmp = zone_set_state(target_zone,
618 			    ZONE_STATE_CONFIGURED)) != Z_OK) {
619 				errno = tmp;
620 				zperror2(target_zone,
621 				    gettext("could not set state"));
622 			}
623 		}
624 	}
625 
626 	return (err);
627 }
628 
629 /*
630  * Attempt to clone a source_zone to a target zonepath by using a ZFS clone.
631  */
632 int
633 clone_zfs(char *source_zone, char *source_zonepath, char *zonepath)
634 {
635 	zfs_handle_t	*zhp;
636 	char		clone_name[MAXPATHLEN];
637 	char		snap_name[MAXPATHLEN];
638 
639 	/*
640 	 * Try to get a zfs handle for the source_zonepath.  If this fails
641 	 * the source_zonepath is not ZFS so return an error.
642 	 */
643 	if ((zhp = mount2zhandle(source_zonepath)) == NULL)
644 		return (Z_ERR);
645 
646 	/*
647 	 * Check if there is a file system already mounted on zonepath.  If so,
648 	 * we can't clone to the path so we should fall back to copying.
649 	 */
650 	if (is_mountpnt(zonepath)) {
651 		zfs_close(zhp);
652 		(void) fprintf(stderr,
653 		    gettext("A file system is already mounted on %s,\n"
654 		    "preventing use of a ZFS clone.\n"), zonepath);
655 		return (Z_ERR);
656 	}
657 
658 	/*
659 	 * Instead of using path2name to get the clone name from the zonepath,
660 	 * we could generate a name from the source zone ZFS name.  However,
661 	 * this would mean we would create the clone under the ZFS fs of the
662 	 * source instead of what the zonepath says.  For example,
663 	 *
664 	 * source_zonepath		zonepath
665 	 * /pl/zones/dev/z1		/pl/zones/deploy/z2
666 	 *
667 	 * We don't want the clone to be under "dev", we want it under
668 	 * "deploy", so that we can leverage the normal attribute inheritance
669 	 * that ZFS provides in the fs hierarchy.
670 	 */
671 	if (path2name(zonepath, clone_name, sizeof (clone_name)) != Z_OK) {
672 		zfs_close(zhp);
673 		return (Z_ERR);
674 	}
675 
676 	if (take_snapshot(source_zone, zhp, snap_name, sizeof (snap_name))
677 	    != Z_OK) {
678 		zfs_close(zhp);
679 		return (Z_ERR);
680 	}
681 	zfs_close(zhp);
682 
683 	if (clone_snap(snap_name, clone_name) != Z_OK)
684 		return (Z_ERR);
685 
686 	(void) printf(gettext("Instead of copying, a ZFS clone has been "
687 	    "created for this zone.\n"));
688 
689 	return (Z_OK);
690 }
691 
692 /*
693  * Attempt to create a ZFS file system for the specified zonepath.
694  * We either will successfully create a ZFS file system and get it mounted
695  * on the zonepath or we don't.  The caller doesn't care since a regular
696  * directory is used for the zonepath if no ZFS file system is mounted there.
697  */
698 void
699 create_zfs_zonepath(char *zonepath)
700 {
701 	zfs_handle_t	*zhp;
702 	char		zfs_name[MAXPATHLEN];
703 
704 	if (path2name(zonepath, zfs_name, sizeof (zfs_name)) != Z_OK)
705 		return;
706 
707 	if (zfs_create(g_zfs, zfs_name, ZFS_TYPE_FILESYSTEM, NULL, NULL) != 0 ||
708 	    (zhp = zfs_open(g_zfs, zfs_name, ZFS_TYPE_ANY)) == NULL) {
709 		(void) fprintf(stderr, gettext("cannot create ZFS dataset %s: "
710 		    "%s\n"), zfs_name, libzfs_error_description(g_zfs));
711 		return;
712 	}
713 
714 	if (zfs_mount(zhp, NULL, 0) != 0) {
715 		(void) fprintf(stderr, gettext("cannot mount ZFS dataset %s: "
716 		    "%s\n"), zfs_name, libzfs_error_description(g_zfs));
717 		(void) zfs_destroy(zhp);
718 	} else if (zfs_prop_set(zhp, ZFS_PROP_SHARENFS, "off") != 0) {
719 		(void) fprintf(stderr, gettext("file system %s successfully "
720 		    "created,\nbut could not turn off the 'sharenfs' "
721 		    "property\n"), zfs_name);
722 	} else {
723 		if (chmod(zonepath, S_IRWXU) != 0) {
724 			(void) fprintf(stderr, gettext("file system %s "
725 			    "successfully created, but chmod %o failed: %s\n"),
726 			    zfs_name, S_IRWXU, strerror(errno));
727 			(void) destroy_zfs(zonepath);
728 		} else {
729 			(void) printf(gettext("A ZFS file system has been "
730 			    "created for this zone.\n"));
731 		}
732 	}
733 
734 	zfs_close(zhp);
735 }
736 
737 /*
738  * If the zonepath is a ZFS file system, attempt to destroy it.  We return Z_OK
739  * if we were able to zfs_destroy the zonepath, otherwise we return Z_ERR
740  * which means the caller should clean up the zonepath in the traditional
741  * way.
742  */
743 int
744 destroy_zfs(char *zonepath)
745 {
746 	zfs_handle_t	*zhp;
747 	boolean_t	is_clone = B_FALSE;
748 	char		origin[ZFS_MAXPROPLEN];
749 
750 	if ((zhp = mount2zhandle(zonepath)) == NULL)
751 		return (Z_ERR);
752 
753 	/*
754 	 * We can't destroy the file system if it has dependents.
755 	 */
756 	if (zfs_iter_dependents(zhp, has_dependent, NULL) != 0 ||
757 	    zfs_unmount(zhp, NULL, 0) != 0) {
758 		zfs_close(zhp);
759 		return (Z_ERR);
760 	}
761 
762 	/*
763 	 * This might be a clone.  Try to get the snapshot so we can attempt
764 	 * to destroy that as well.
765 	 */
766 	if (zfs_prop_get(zhp, ZFS_PROP_ORIGIN, origin, sizeof (origin), NULL,
767 	    NULL, 0, B_FALSE) == 0)
768 		is_clone = B_TRUE;
769 
770 	if (zfs_destroy(zhp) != 0) {
771 		/*
772 		 * If the destroy fails for some reason, try to remount
773 		 * the file system so that we can use "rm -rf" to clean up
774 		 * instead.
775 		 */
776 		(void) zfs_mount(zhp, NULL, 0);
777 		zfs_close(zhp);
778 		return (Z_ERR);
779 	}
780 
781 	(void) printf(gettext("The ZFS file system for this zone has been "
782 	    "destroyed.\n"));
783 
784 	if (is_clone) {
785 		zfs_handle_t	*ohp;
786 
787 		/*
788 		 * Try to clean up the snapshot that the clone was taken from.
789 		 */
790 		if ((ohp = zfs_open(g_zfs, origin,
791 		    ZFS_TYPE_SNAPSHOT)) != NULL) {
792 			if (zfs_iter_dependents(ohp, has_dependent, NULL)
793 			    == 0 && zfs_unmount(ohp, NULL, 0) == 0)
794 				(void) zfs_destroy(ohp);
795 			zfs_close(ohp);
796 		}
797 	}
798 
799 	zfs_close(zhp);
800 	return (Z_OK);
801 }
802 
803 /*
804  * Return true if the path is its own zfs file system.  We determine this
805  * by stat-ing the path to see if it is zfs and stat-ing the parent to see
806  * if it is a different fs.
807  */
808 boolean_t
809 is_zonepath_zfs(char *zonepath)
810 {
811 	int res;
812 	char *path;
813 	char *parent;
814 	struct statvfs64 buf1, buf2;
815 
816 	if (statvfs64(zonepath, &buf1) != 0)
817 		return (B_FALSE);
818 
819 	if (strcmp(buf1.f_basetype, "zfs") != 0)
820 		return (B_FALSE);
821 
822 	if ((path = strdup(zonepath)) == NULL)
823 		return (B_FALSE);
824 
825 	parent = dirname(path);
826 	res = statvfs64(parent, &buf2);
827 	free(path);
828 
829 	if (res != 0)
830 		return (B_FALSE);
831 
832 	if (buf1.f_fsid == buf2.f_fsid)
833 		return (B_FALSE);
834 
835 	return (B_TRUE);
836 }
837 
838 /*
839  * Implement the fast move of a ZFS file system by simply updating the
840  * mountpoint.  Since it is file system already, we don't have the
841  * issue of cross-file system copying.
842  */
843 int
844 move_zfs(char *zonepath, char *new_zonepath)
845 {
846 	int		ret = Z_ERR;
847 	zfs_handle_t	*zhp;
848 
849 	if ((zhp = mount2zhandle(zonepath)) == NULL)
850 		return (Z_ERR);
851 
852 	if (zfs_prop_set(zhp, ZFS_PROP_MOUNTPOINT, new_zonepath) == 0) {
853 		/*
854 		 * Clean up the old mount point.  We ignore any failure since
855 		 * the zone is already successfully mounted on the new path.
856 		 */
857 		(void) rmdir(zonepath);
858 		ret = Z_OK;
859 	}
860 
861 	zfs_close(zhp);
862 
863 	return (ret);
864 }
865 
866 /*
867  * Validate that the given dataset exists on the system, and that neither it nor
868  * its children are zvols.
869  *
870  * Note that we don't do anything with the 'zoned' property here.  All
871  * management is done in zoneadmd when the zone is actually rebooted.  This
872  * allows us to automatically set the zoned property even when a zone is
873  * rebooted by the administrator.
874  */
875 int
876 verify_datasets(zone_dochandle_t handle)
877 {
878 	int return_code = Z_OK;
879 	struct zone_dstab dstab;
880 	zfs_handle_t *zhp;
881 	char propbuf[ZFS_MAXPROPLEN];
882 	char source[ZFS_MAXNAMELEN];
883 	zfs_source_t srctype;
884 
885 	if (zonecfg_setdsent(handle) != Z_OK) {
886 		/*
887 		 * TRANSLATION_NOTE
888 		 * zfs and dataset are literals that should not be translated.
889 		 */
890 		(void) fprintf(stderr, gettext("could not verify zfs datasets: "
891 		    "unable to enumerate datasets\n"));
892 		return (Z_ERR);
893 	}
894 
895 	while (zonecfg_getdsent(handle, &dstab) == Z_OK) {
896 
897 		if ((zhp = zfs_open(g_zfs, dstab.zone_dataset_name,
898 		    ZFS_TYPE_FILESYSTEM | ZFS_TYPE_VOLUME)) == NULL) {
899 			(void) fprintf(stderr, gettext("could not verify zfs "
900 			    "dataset %s: %s\n"), dstab.zone_dataset_name,
901 			    libzfs_error_description(g_zfs));
902 			return_code = Z_ERR;
903 			continue;
904 		}
905 
906 		if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, propbuf,
907 		    sizeof (propbuf), &srctype, source,
908 		    sizeof (source), 0) == 0 &&
909 		    (srctype == ZFS_SRC_INHERITED)) {
910 			(void) fprintf(stderr, gettext("could not verify zfs "
911 			    "dataset %s: mountpoint cannot be inherited\n"),
912 			    dstab.zone_dataset_name);
913 			return_code = Z_ERR;
914 			zfs_close(zhp);
915 			continue;
916 		}
917 
918 		if (zfs_get_type(zhp) == ZFS_TYPE_VOLUME) {
919 			(void) fprintf(stderr, gettext("cannot verify zfs "
920 			    "dataset %s: volumes cannot be specified as a "
921 			    "zone dataset resource\n"),
922 			    dstab.zone_dataset_name);
923 			return_code = Z_ERR;
924 		}
925 
926 		if (zfs_iter_children(zhp, check_zvol, NULL) != 0)
927 			return_code = Z_ERR;
928 
929 		zfs_close(zhp);
930 	}
931 	(void) zonecfg_enddsent(handle);
932 
933 	return (return_code);
934 }
935 
936 /*
937  * Verify that the ZFS dataset exists, and its mountpoint
938  * property is set to "legacy".
939  */
940 int
941 verify_fs_zfs(struct zone_fstab *fstab)
942 {
943 	zfs_handle_t *zhp;
944 	char propbuf[ZFS_MAXPROPLEN];
945 
946 	if ((zhp = zfs_open(g_zfs, fstab->zone_fs_special,
947 	    ZFS_TYPE_ANY)) == NULL) {
948 		(void) fprintf(stderr, gettext("could not verify fs %s: "
949 		    "could not access zfs dataset '%s'\n"),
950 		    fstab->zone_fs_dir, fstab->zone_fs_special);
951 		return (Z_ERR);
952 	}
953 
954 	if (zfs_get_type(zhp) != ZFS_TYPE_FILESYSTEM) {
955 		(void) fprintf(stderr, gettext("cannot verify fs %s: "
956 		    "'%s' is not a file system\n"),
957 		    fstab->zone_fs_dir, fstab->zone_fs_special);
958 		zfs_close(zhp);
959 		return (Z_ERR);
960 	}
961 
962 	if (zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, propbuf, sizeof (propbuf),
963 	    NULL, NULL, 0, 0) != 0 || strcmp(propbuf, "legacy") != 0) {
964 		(void) fprintf(stderr, gettext("could not verify fs %s: "
965 		    "zfs '%s' mountpoint is not \"legacy\"\n"),
966 		    fstab->zone_fs_dir, fstab->zone_fs_special);
967 		zfs_close(zhp);
968 		return (Z_ERR);
969 	}
970 
971 	zfs_close(zhp);
972 	return (Z_OK);
973 }
974 
975 int
976 init_zfs(void)
977 {
978 	if ((g_zfs = libzfs_init()) == NULL) {
979 		(void) fprintf(stderr, gettext("failed to initialize ZFS "
980 		    "library\n"));
981 		return (Z_ERR);
982 	}
983 
984 	return (Z_OK);
985 }
986