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 2015 Nexenta Systems, Inc.  All rights reserved.
24  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25  * Copyright (c) 2014, 2017 by Delphix. All rights reserved.
26  * Copyright 2016 Igor Kozhukhov <ikozhukhov@gmail.com>
27  * Copyright 2017 Joyent, Inc.
28  * Copyright 2017 RackTop Systems.
29  * Copyright 2018 OmniOS Community Edition (OmniOSce) Association.
30  */
31 
32 /*
33  * Routines to manage ZFS mounts.  We separate all the nasty routines that have
34  * to deal with the OS.  The following functions are the main entry points --
35  * they are used by mount and unmount and when changing a filesystem's
36  * mountpoint.
37  *
38  *	zfs_is_mounted()
39  *	zfs_mount()
40  *	zfs_unmount()
41  *	zfs_unmountall()
42  *
43  * This file also contains the functions used to manage sharing filesystems via
44  * NFS and iSCSI:
45  *
46  *	zfs_is_shared()
47  *	zfs_share()
48  *	zfs_unshare()
49  *
50  *	zfs_is_shared_nfs()
51  *	zfs_is_shared_smb()
52  *	zfs_share_proto()
53  *	zfs_shareall();
54  *	zfs_unshare_nfs()
55  *	zfs_unshare_smb()
56  *	zfs_unshareall_nfs()
57  *	zfs_unshareall_smb()
58  *	zfs_unshareall()
59  *	zfs_unshareall_bypath()
60  *
61  * The following functions are available for pool consumers, and will
62  * mount/unmount and share/unshare all datasets within pool:
63  *
64  *	zpool_enable_datasets()
65  *	zpool_disable_datasets()
66  */
67 
68 #include <dirent.h>
69 #include <dlfcn.h>
70 #include <errno.h>
71 #include <fcntl.h>
72 #include <libgen.h>
73 #include <libintl.h>
74 #include <stdio.h>
75 #include <stdlib.h>
76 #include <strings.h>
77 #include <unistd.h>
78 #include <zone.h>
79 #include <sys/mntent.h>
80 #include <sys/mount.h>
81 #include <sys/stat.h>
82 #include <sys/statvfs.h>
83 #include <sys/dsl_crypt.h>
84 
85 #include <libzfs.h>
86 
87 #include "libzfs_impl.h"
88 #include "libzfs_taskq.h"
89 
90 #include <libshare.h>
91 #include <sys/systeminfo.h>
92 #define	MAXISALEN	257	/* based on sysinfo(2) man page */
93 
94 static int mount_tq_nthr = 512;	/* taskq threads for multi-threaded mounting */
95 
96 static void zfs_mount_task(void *);
97 static int zfs_share_proto(zfs_handle_t *, zfs_share_proto_t *);
98 zfs_share_type_t zfs_is_shared_proto(zfs_handle_t *, char **,
99     zfs_share_proto_t);
100 
101 /*
102  * The share protocols table must be in the same order as the zfs_share_proto_t
103  * enum in libzfs_impl.h
104  */
105 typedef struct {
106 	zfs_prop_t p_prop;
107 	char *p_name;
108 	int p_share_err;
109 	int p_unshare_err;
110 } proto_table_t;
111 
112 proto_table_t proto_table[PROTO_END] = {
113 	{ZFS_PROP_SHARENFS, "nfs", EZFS_SHARENFSFAILED, EZFS_UNSHARENFSFAILED},
114 	{ZFS_PROP_SHARESMB, "smb", EZFS_SHARESMBFAILED, EZFS_UNSHARESMBFAILED},
115 };
116 
117 zfs_share_proto_t nfs_only[] = {
118 	PROTO_NFS,
119 	PROTO_END
120 };
121 
122 zfs_share_proto_t smb_only[] = {
123 	PROTO_SMB,
124 	PROTO_END
125 };
126 zfs_share_proto_t share_all_proto[] = {
127 	PROTO_NFS,
128 	PROTO_SMB,
129 	PROTO_END
130 };
131 
132 /*
133  * Search the sharetab for the given mountpoint and protocol, returning
134  * a zfs_share_type_t value.
135  */
136 static zfs_share_type_t
137 is_shared(libzfs_handle_t *hdl, const char *mountpoint, zfs_share_proto_t proto)
138 {
139 	char buf[MAXPATHLEN], *tab;
140 	char *ptr;
141 
142 	if (hdl->libzfs_sharetab == NULL)
143 		return (SHARED_NOT_SHARED);
144 
145 	(void) fseek(hdl->libzfs_sharetab, 0, SEEK_SET);
146 
147 	while (fgets(buf, sizeof (buf), hdl->libzfs_sharetab) != NULL) {
148 
149 		/* the mountpoint is the first entry on each line */
150 		if ((tab = strchr(buf, '\t')) == NULL)
151 			continue;
152 
153 		*tab = '\0';
154 		if (strcmp(buf, mountpoint) == 0) {
155 			/*
156 			 * the protocol field is the third field
157 			 * skip over second field
158 			 */
159 			ptr = ++tab;
160 			if ((tab = strchr(ptr, '\t')) == NULL)
161 				continue;
162 			ptr = ++tab;
163 			if ((tab = strchr(ptr, '\t')) == NULL)
164 				continue;
165 			*tab = '\0';
166 			if (strcmp(ptr,
167 			    proto_table[proto].p_name) == 0) {
168 				switch (proto) {
169 				case PROTO_NFS:
170 					return (SHARED_NFS);
171 				case PROTO_SMB:
172 					return (SHARED_SMB);
173 				default:
174 					return (0);
175 				}
176 			}
177 		}
178 	}
179 
180 	return (SHARED_NOT_SHARED);
181 }
182 
183 static boolean_t
184 dir_is_empty_stat(const char *dirname)
185 {
186 	struct stat st;
187 
188 	/*
189 	 * We only want to return false if the given path is a non empty
190 	 * directory, all other errors are handled elsewhere.
191 	 */
192 	if (stat(dirname, &st) < 0 || !S_ISDIR(st.st_mode)) {
193 		return (B_TRUE);
194 	}
195 
196 	/*
197 	 * An empty directory will still have two entries in it, one
198 	 * entry for each of "." and "..".
199 	 */
200 	if (st.st_size > 2) {
201 		return (B_FALSE);
202 	}
203 
204 	return (B_TRUE);
205 }
206 
207 static boolean_t
208 dir_is_empty_readdir(const char *dirname)
209 {
210 	DIR *dirp;
211 	struct dirent64 *dp;
212 	int dirfd;
213 
214 	if ((dirfd = openat(AT_FDCWD, dirname,
215 	    O_RDONLY | O_NDELAY | O_LARGEFILE | O_CLOEXEC, 0)) < 0) {
216 		return (B_TRUE);
217 	}
218 
219 	if ((dirp = fdopendir(dirfd)) == NULL) {
220 		(void) close(dirfd);
221 		return (B_TRUE);
222 	}
223 
224 	while ((dp = readdir64(dirp)) != NULL) {
225 
226 		if (strcmp(dp->d_name, ".") == 0 ||
227 		    strcmp(dp->d_name, "..") == 0)
228 			continue;
229 
230 		(void) closedir(dirp);
231 		return (B_FALSE);
232 	}
233 
234 	(void) closedir(dirp);
235 	return (B_TRUE);
236 }
237 
238 /*
239  * Returns true if the specified directory is empty.  If we can't open the
240  * directory at all, return true so that the mount can fail with a more
241  * informative error message.
242  */
243 static boolean_t
244 dir_is_empty(const char *dirname)
245 {
246 	struct statvfs64 st;
247 
248 	/*
249 	 * If the statvfs call fails or the filesystem is not a ZFS
250 	 * filesystem, fall back to the slow path which uses readdir.
251 	 */
252 	if ((statvfs64(dirname, &st) != 0) ||
253 	    (strcmp(st.f_basetype, "zfs") != 0)) {
254 		return (dir_is_empty_readdir(dirname));
255 	}
256 
257 	/*
258 	 * At this point, we know the provided path is on a ZFS
259 	 * filesystem, so we can use stat instead of readdir to
260 	 * determine if the directory is empty or not. We try to avoid
261 	 * using readdir because that requires opening "dirname"; this
262 	 * open file descriptor can potentially end up in a child
263 	 * process if there's a concurrent fork, thus preventing the
264 	 * zfs_mount() from otherwise succeeding (the open file
265 	 * descriptor inherited by the child process will cause the
266 	 * parent's mount to fail with EBUSY). The performance
267 	 * implications of replacing the open, read, and close with a
268 	 * single stat is nice; but is not the main motivation for the
269 	 * added complexity.
270 	 */
271 	return (dir_is_empty_stat(dirname));
272 }
273 
274 /*
275  * Checks to see if the mount is active.  If the filesystem is mounted, we fill
276  * in 'where' with the current mountpoint, and return 1.  Otherwise, we return
277  * 0.
278  */
279 boolean_t
280 is_mounted(libzfs_handle_t *zfs_hdl, const char *special, char **where)
281 {
282 	struct mnttab entry;
283 
284 	if (libzfs_mnttab_find(zfs_hdl, special, &entry) != 0)
285 		return (B_FALSE);
286 
287 	if (where != NULL)
288 		*where = zfs_strdup(zfs_hdl, entry.mnt_mountp);
289 
290 	return (B_TRUE);
291 }
292 
293 boolean_t
294 zfs_is_mounted(zfs_handle_t *zhp, char **where)
295 {
296 	return (is_mounted(zhp->zfs_hdl, zfs_get_name(zhp), where));
297 }
298 
299 /*
300  * Returns true if the given dataset is mountable, false otherwise.  Returns the
301  * mountpoint in 'buf'.
302  */
303 static boolean_t
304 zfs_is_mountable(zfs_handle_t *zhp, char *buf, size_t buflen,
305     zprop_source_t *source)
306 {
307 	char sourceloc[MAXNAMELEN];
308 	zprop_source_t sourcetype;
309 
310 	if (!zfs_prop_valid_for_type(ZFS_PROP_MOUNTPOINT, zhp->zfs_type))
311 		return (B_FALSE);
312 
313 	verify(zfs_prop_get(zhp, ZFS_PROP_MOUNTPOINT, buf, buflen,
314 	    &sourcetype, sourceloc, sizeof (sourceloc), B_FALSE) == 0);
315 
316 	if (strcmp(buf, ZFS_MOUNTPOINT_NONE) == 0 ||
317 	    strcmp(buf, ZFS_MOUNTPOINT_LEGACY) == 0)
318 		return (B_FALSE);
319 
320 	if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_OFF)
321 		return (B_FALSE);
322 
323 	if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED) &&
324 	    getzoneid() == GLOBAL_ZONEID)
325 		return (B_FALSE);
326 
327 	if (source)
328 		*source = sourcetype;
329 
330 	return (B_TRUE);
331 }
332 
333 /*
334  * Mount the given filesystem.
335  */
336 int
337 zfs_mount(zfs_handle_t *zhp, const char *options, int flags)
338 {
339 	struct stat buf;
340 	char mountpoint[ZFS_MAXPROPLEN];
341 	char mntopts[MNT_LINE_MAX];
342 	libzfs_handle_t *hdl = zhp->zfs_hdl;
343 	uint64_t keystatus;
344 	int rc;
345 
346 	if (options == NULL)
347 		mntopts[0] = '\0';
348 	else
349 		(void) strlcpy(mntopts, options, sizeof (mntopts));
350 
351 	/*
352 	 * If the pool is imported read-only then all mounts must be read-only
353 	 */
354 	if (zpool_get_prop_int(zhp->zpool_hdl, ZPOOL_PROP_READONLY, NULL))
355 		flags |= MS_RDONLY;
356 
357 	if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL))
358 		return (0);
359 
360 	/*
361 	 * If the filesystem is encrypted the key must be loaded  in order to
362 	 * mount. If the key isn't loaded, the MS_CRYPT flag decides whether
363 	 * or not we attempt to load the keys. Note: we must call
364 	 * zfs_refresh_properties() here since some callers of this function
365 	 * (most notably zpool_enable_datasets()) may implicitly load our key
366 	 * by loading the parent's key first.
367 	 */
368 	if (zfs_prop_get_int(zhp, ZFS_PROP_ENCRYPTION) != ZIO_CRYPT_OFF) {
369 		zfs_refresh_properties(zhp);
370 		keystatus = zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS);
371 
372 		/*
373 		 * If the key is unavailable and MS_CRYPT is set give the
374 		 * user a chance to enter the key. Otherwise just fail
375 		 * immediately.
376 		 */
377 		if (keystatus == ZFS_KEYSTATUS_UNAVAILABLE) {
378 			if (flags & MS_CRYPT) {
379 				rc = zfs_crypto_load_key(zhp, B_FALSE, NULL);
380 				if (rc != 0)
381 					return (rc);
382 			} else {
383 				zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
384 				    "encryption key not loaded"));
385 				return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
386 				    dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
387 				    mountpoint));
388 			}
389 		}
390 
391 	}
392 
393 	/* Create the directory if it doesn't already exist */
394 	if (lstat(mountpoint, &buf) != 0) {
395 		if (mkdirp(mountpoint, 0755) != 0) {
396 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
397 			    "failed to create mountpoint"));
398 			return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
399 			    dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
400 			    mountpoint));
401 		}
402 	}
403 
404 	/*
405 	 * Determine if the mountpoint is empty.  If so, refuse to perform the
406 	 * mount.  We don't perform this check if MS_OVERLAY is specified, which
407 	 * would defeat the point.  We also avoid this check if 'remount' is
408 	 * specified.
409 	 */
410 	if ((flags & MS_OVERLAY) == 0 &&
411 	    strstr(mntopts, MNTOPT_REMOUNT) == NULL &&
412 	    !dir_is_empty(mountpoint)) {
413 		zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
414 		    "directory is not empty"));
415 		return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
416 		    dgettext(TEXT_DOMAIN, "cannot mount '%s'"), mountpoint));
417 	}
418 
419 	/* perform the mount */
420 	if (mount(zfs_get_name(zhp), mountpoint, MS_OPTIONSTR | flags,
421 	    MNTTYPE_ZFS, NULL, 0, mntopts, sizeof (mntopts)) != 0) {
422 		/*
423 		 * Generic errors are nasty, but there are just way too many
424 		 * from mount(), and they're well-understood.  We pick a few
425 		 * common ones to improve upon.
426 		 */
427 		if (errno == EBUSY) {
428 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
429 			    "mountpoint or dataset is busy"));
430 		} else if (errno == EPERM) {
431 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN,
432 			    "Insufficient privileges"));
433 		} else if (errno == ENOTSUP) {
434 			char buf[256];
435 			int spa_version;
436 
437 			VERIFY(zfs_spa_version(zhp, &spa_version) == 0);
438 			(void) snprintf(buf, sizeof (buf),
439 			    dgettext(TEXT_DOMAIN, "Can't mount a version %lld "
440 			    "file system on a version %d pool. Pool must be"
441 			    " upgraded to mount this file system."),
442 			    (u_longlong_t)zfs_prop_get_int(zhp,
443 			    ZFS_PROP_VERSION), spa_version);
444 			zfs_error_aux(hdl, dgettext(TEXT_DOMAIN, buf));
445 		} else {
446 			zfs_error_aux(hdl, strerror(errno));
447 		}
448 		return (zfs_error_fmt(hdl, EZFS_MOUNTFAILED,
449 		    dgettext(TEXT_DOMAIN, "cannot mount '%s'"),
450 		    zhp->zfs_name));
451 	}
452 
453 	/* add the mounted entry into our cache */
454 	libzfs_mnttab_add(hdl, zfs_get_name(zhp), mountpoint,
455 	    mntopts);
456 	return (0);
457 }
458 
459 /*
460  * Unmount a single filesystem.
461  */
462 static int
463 unmount_one(libzfs_handle_t *hdl, const char *mountpoint, int flags)
464 {
465 	if (umount2(mountpoint, flags) != 0) {
466 		zfs_error_aux(hdl, strerror(errno));
467 		return (zfs_error_fmt(hdl, EZFS_UMOUNTFAILED,
468 		    dgettext(TEXT_DOMAIN, "cannot unmount '%s'"),
469 		    mountpoint));
470 	}
471 
472 	return (0);
473 }
474 
475 /*
476  * Unmount the given filesystem.
477  */
478 int
479 zfs_unmount(zfs_handle_t *zhp, const char *mountpoint, int flags)
480 {
481 	libzfs_handle_t *hdl = zhp->zfs_hdl;
482 	struct mnttab entry;
483 	char *mntpt = NULL;
484 
485 	/* check to see if we need to unmount the filesystem */
486 	if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) &&
487 	    libzfs_mnttab_find(hdl, zhp->zfs_name, &entry) == 0)) {
488 		/*
489 		 * mountpoint may have come from a call to
490 		 * getmnt/getmntany if it isn't NULL. If it is NULL,
491 		 * we know it comes from libzfs_mnttab_find which can
492 		 * then get freed later. We strdup it to play it safe.
493 		 */
494 		if (mountpoint == NULL)
495 			mntpt = zfs_strdup(hdl, entry.mnt_mountp);
496 		else
497 			mntpt = zfs_strdup(hdl, mountpoint);
498 
499 		/*
500 		 * Unshare and unmount the filesystem
501 		 */
502 		if (zfs_unshare_proto(zhp, mntpt, share_all_proto) != 0)
503 			return (-1);
504 
505 		if (unmount_one(hdl, mntpt, flags) != 0) {
506 			free(mntpt);
507 			(void) zfs_shareall(zhp);
508 			return (-1);
509 		}
510 		libzfs_mnttab_remove(hdl, zhp->zfs_name);
511 		free(mntpt);
512 	}
513 
514 	return (0);
515 }
516 
517 /*
518  * Unmount this filesystem and any children inheriting the mountpoint property.
519  * To do this, just act like we're changing the mountpoint property, but don't
520  * remount the filesystems afterwards.
521  */
522 int
523 zfs_unmountall(zfs_handle_t *zhp, int flags)
524 {
525 	prop_changelist_t *clp;
526 	int ret;
527 
528 	clp = changelist_gather(zhp, ZFS_PROP_MOUNTPOINT, 0, flags);
529 	if (clp == NULL)
530 		return (-1);
531 
532 	ret = changelist_prefix(clp);
533 	changelist_free(clp);
534 
535 	return (ret);
536 }
537 
538 boolean_t
539 zfs_is_shared(zfs_handle_t *zhp)
540 {
541 	zfs_share_type_t rc = 0;
542 	zfs_share_proto_t *curr_proto;
543 
544 	if (ZFS_IS_VOLUME(zhp))
545 		return (B_FALSE);
546 
547 	for (curr_proto = share_all_proto; *curr_proto != PROTO_END;
548 	    curr_proto++)
549 		rc |= zfs_is_shared_proto(zhp, NULL, *curr_proto);
550 
551 	return (rc ? B_TRUE : B_FALSE);
552 }
553 
554 int
555 zfs_share(zfs_handle_t *zhp)
556 {
557 	assert(!ZFS_IS_VOLUME(zhp));
558 	return (zfs_share_proto(zhp, share_all_proto));
559 }
560 
561 int
562 zfs_unshare(zfs_handle_t *zhp)
563 {
564 	assert(!ZFS_IS_VOLUME(zhp));
565 	return (zfs_unshareall(zhp));
566 }
567 
568 /*
569  * Check to see if the filesystem is currently shared.
570  */
571 zfs_share_type_t
572 zfs_is_shared_proto(zfs_handle_t *zhp, char **where, zfs_share_proto_t proto)
573 {
574 	char *mountpoint;
575 	zfs_share_type_t rc;
576 
577 	if (!zfs_is_mounted(zhp, &mountpoint))
578 		return (SHARED_NOT_SHARED);
579 
580 	if ((rc = is_shared(zhp->zfs_hdl, mountpoint, proto))
581 	    != SHARED_NOT_SHARED) {
582 		if (where != NULL)
583 			*where = mountpoint;
584 		else
585 			free(mountpoint);
586 		return (rc);
587 	} else {
588 		free(mountpoint);
589 		return (SHARED_NOT_SHARED);
590 	}
591 }
592 
593 boolean_t
594 zfs_is_shared_nfs(zfs_handle_t *zhp, char **where)
595 {
596 	return (zfs_is_shared_proto(zhp, where,
597 	    PROTO_NFS) != SHARED_NOT_SHARED);
598 }
599 
600 boolean_t
601 zfs_is_shared_smb(zfs_handle_t *zhp, char **where)
602 {
603 	return (zfs_is_shared_proto(zhp, where,
604 	    PROTO_SMB) != SHARED_NOT_SHARED);
605 }
606 
607 /*
608  * Make sure things will work if libshare isn't installed by using
609  * wrapper functions that check to see that the pointers to functions
610  * initialized in _zfs_init_libshare() are actually present.
611  */
612 
613 static sa_handle_t (*_sa_init)(int);
614 static sa_handle_t (*_sa_init_arg)(int, void *);
615 static int (*_sa_service)(sa_handle_t);
616 static void (*_sa_fini)(sa_handle_t);
617 static sa_share_t (*_sa_find_share)(sa_handle_t, char *);
618 static int (*_sa_enable_share)(sa_share_t, char *);
619 static int (*_sa_disable_share)(sa_share_t, char *);
620 static char *(*_sa_errorstr)(int);
621 static int (*_sa_parse_legacy_options)(sa_group_t, char *, char *);
622 static boolean_t (*_sa_needs_refresh)(sa_handle_t *);
623 static libzfs_handle_t *(*_sa_get_zfs_handle)(sa_handle_t);
624 static int (*_sa_zfs_process_share)(sa_handle_t, sa_group_t, sa_share_t,
625     char *, char *, zprop_source_t, char *, char *, char *);
626 static void (*_sa_update_sharetab_ts)(sa_handle_t);
627 
628 /*
629  * _zfs_init_libshare()
630  *
631  * Find the libshare.so.1 entry points that we use here and save the
632  * values to be used later. This is triggered by the runtime loader.
633  * Make sure the correct ISA version is loaded.
634  */
635 
636 #pragma init(_zfs_init_libshare)
637 static void
638 _zfs_init_libshare(void)
639 {
640 	void *libshare;
641 	char path[MAXPATHLEN];
642 	char isa[MAXISALEN];
643 
644 #if defined(_LP64)
645 	if (sysinfo(SI_ARCHITECTURE_64, isa, MAXISALEN) == -1)
646 		isa[0] = '\0';
647 #else
648 	isa[0] = '\0';
649 #endif
650 	(void) snprintf(path, MAXPATHLEN,
651 	    "/usr/lib/%s/libshare.so.1", isa);
652 
653 	if ((libshare = dlopen(path, RTLD_LAZY | RTLD_GLOBAL)) != NULL) {
654 		_sa_init = (sa_handle_t (*)(int))dlsym(libshare, "sa_init");
655 		_sa_init_arg = (sa_handle_t (*)(int, void *))dlsym(libshare,
656 		    "sa_init_arg");
657 		_sa_fini = (void (*)(sa_handle_t))dlsym(libshare, "sa_fini");
658 		_sa_service = (int (*)(sa_handle_t))dlsym(libshare,
659 		    "sa_service");
660 		_sa_find_share = (sa_share_t (*)(sa_handle_t, char *))
661 		    dlsym(libshare, "sa_find_share");
662 		_sa_enable_share = (int (*)(sa_share_t, char *))dlsym(libshare,
663 		    "sa_enable_share");
664 		_sa_disable_share = (int (*)(sa_share_t, char *))dlsym(libshare,
665 		    "sa_disable_share");
666 		_sa_errorstr = (char *(*)(int))dlsym(libshare, "sa_errorstr");
667 		_sa_parse_legacy_options = (int (*)(sa_group_t, char *, char *))
668 		    dlsym(libshare, "sa_parse_legacy_options");
669 		_sa_needs_refresh = (boolean_t (*)(sa_handle_t *))
670 		    dlsym(libshare, "sa_needs_refresh");
671 		_sa_get_zfs_handle = (libzfs_handle_t *(*)(sa_handle_t))
672 		    dlsym(libshare, "sa_get_zfs_handle");
673 		_sa_zfs_process_share = (int (*)(sa_handle_t, sa_group_t,
674 		    sa_share_t, char *, char *, zprop_source_t, char *,
675 		    char *, char *))dlsym(libshare, "sa_zfs_process_share");
676 		_sa_update_sharetab_ts = (void (*)(sa_handle_t))
677 		    dlsym(libshare, "sa_update_sharetab_ts");
678 		if (_sa_init == NULL || _sa_init_arg == NULL ||
679 		    _sa_fini == NULL || _sa_find_share == NULL ||
680 		    _sa_enable_share == NULL || _sa_disable_share == NULL ||
681 		    _sa_errorstr == NULL || _sa_parse_legacy_options == NULL ||
682 		    _sa_needs_refresh == NULL || _sa_get_zfs_handle == NULL ||
683 		    _sa_zfs_process_share == NULL || _sa_service == NULL ||
684 		    _sa_update_sharetab_ts == NULL) {
685 			_sa_init = NULL;
686 			_sa_init_arg = NULL;
687 			_sa_service = NULL;
688 			_sa_fini = NULL;
689 			_sa_disable_share = NULL;
690 			_sa_enable_share = NULL;
691 			_sa_errorstr = NULL;
692 			_sa_parse_legacy_options = NULL;
693 			(void) dlclose(libshare);
694 			_sa_needs_refresh = NULL;
695 			_sa_get_zfs_handle = NULL;
696 			_sa_zfs_process_share = NULL;
697 			_sa_update_sharetab_ts = NULL;
698 		}
699 	}
700 }
701 
702 /*
703  * zfs_init_libshare(zhandle, service)
704  *
705  * Initialize the libshare API if it hasn't already been initialized.
706  * In all cases it returns 0 if it succeeded and an error if not. The
707  * service value is which part(s) of the API to initialize and is a
708  * direct map to the libshare sa_init(service) interface.
709  */
710 static int
711 zfs_init_libshare_impl(libzfs_handle_t *zhandle, int service, void *arg)
712 {
713 	/*
714 	 * libshare is either not installed or we're in a branded zone. The
715 	 * rest of the wrapper functions around the libshare calls already
716 	 * handle NULL function pointers, but we don't want the callers of
717 	 * zfs_init_libshare() to fail prematurely if libshare is not available.
718 	 */
719 	if (_sa_init == NULL)
720 		return (SA_OK);
721 
722 	/*
723 	 * Attempt to refresh libshare. This is necessary if there was a cache
724 	 * miss for a new ZFS dataset that was just created, or if state of the
725 	 * sharetab file has changed since libshare was last initialized. We
726 	 * want to make sure so check timestamps to see if a different process
727 	 * has updated any of the configuration. If there was some non-ZFS
728 	 * change, we need to re-initialize the internal cache.
729 	 */
730 	if (_sa_needs_refresh != NULL &&
731 	    _sa_needs_refresh(zhandle->libzfs_sharehdl)) {
732 		zfs_uninit_libshare(zhandle);
733 		zhandle->libzfs_sharehdl = _sa_init_arg(service, arg);
734 	}
735 
736 	if (zhandle && zhandle->libzfs_sharehdl == NULL)
737 		zhandle->libzfs_sharehdl = _sa_init_arg(service, arg);
738 
739 	if (zhandle->libzfs_sharehdl == NULL)
740 		return (SA_NO_MEMORY);
741 
742 	return (SA_OK);
743 }
744 int
745 zfs_init_libshare(libzfs_handle_t *zhandle, int service)
746 {
747 	return (zfs_init_libshare_impl(zhandle, service, NULL));
748 }
749 
750 int
751 zfs_init_libshare_arg(libzfs_handle_t *zhandle, int service, void *arg)
752 {
753 	return (zfs_init_libshare_impl(zhandle, service, arg));
754 }
755 
756 
757 /*
758  * zfs_uninit_libshare(zhandle)
759  *
760  * Uninitialize the libshare API if it hasn't already been
761  * uninitialized. It is OK to call multiple times.
762  */
763 void
764 zfs_uninit_libshare(libzfs_handle_t *zhandle)
765 {
766 	if (zhandle != NULL && zhandle->libzfs_sharehdl != NULL) {
767 		if (_sa_fini != NULL)
768 			_sa_fini(zhandle->libzfs_sharehdl);
769 		zhandle->libzfs_sharehdl = NULL;
770 	}
771 }
772 
773 /*
774  * zfs_parse_options(options, proto)
775  *
776  * Call the legacy parse interface to get the protocol specific
777  * options using the NULL arg to indicate that this is a "parse" only.
778  */
779 int
780 zfs_parse_options(char *options, zfs_share_proto_t proto)
781 {
782 	if (_sa_parse_legacy_options != NULL) {
783 		return (_sa_parse_legacy_options(NULL, options,
784 		    proto_table[proto].p_name));
785 	}
786 	return (SA_CONFIG_ERR);
787 }
788 
789 /*
790  * zfs_sa_find_share(handle, path)
791  *
792  * wrapper around sa_find_share to find a share path in the
793  * configuration.
794  */
795 static sa_share_t
796 zfs_sa_find_share(sa_handle_t handle, char *path)
797 {
798 	if (_sa_find_share != NULL)
799 		return (_sa_find_share(handle, path));
800 	return (NULL);
801 }
802 
803 /*
804  * zfs_sa_enable_share(share, proto)
805  *
806  * Wrapper for sa_enable_share which enables a share for a specified
807  * protocol.
808  */
809 static int
810 zfs_sa_enable_share(sa_share_t share, char *proto)
811 {
812 	if (_sa_enable_share != NULL)
813 		return (_sa_enable_share(share, proto));
814 	return (SA_CONFIG_ERR);
815 }
816 
817 /*
818  * zfs_sa_disable_share(share, proto)
819  *
820  * Wrapper for sa_enable_share which disables a share for a specified
821  * protocol.
822  */
823 static int
824 zfs_sa_disable_share(sa_share_t share, char *proto)
825 {
826 	if (_sa_disable_share != NULL)
827 		return (_sa_disable_share(share, proto));
828 	return (SA_CONFIG_ERR);
829 }
830 
831 /*
832  * Share the given filesystem according to the options in the specified
833  * protocol specific properties (sharenfs, sharesmb).  We rely
834  * on "libshare" to the dirty work for us.
835  */
836 static int
837 zfs_share_proto(zfs_handle_t *zhp, zfs_share_proto_t *proto)
838 {
839 	char mountpoint[ZFS_MAXPROPLEN];
840 	char shareopts[ZFS_MAXPROPLEN];
841 	char sourcestr[ZFS_MAXPROPLEN];
842 	libzfs_handle_t *hdl = zhp->zfs_hdl;
843 	sa_share_t share;
844 	zfs_share_proto_t *curr_proto;
845 	zprop_source_t sourcetype;
846 	int service = SA_INIT_ONE_SHARE_FROM_HANDLE;
847 	int ret;
848 
849 	if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint), NULL))
850 		return (0);
851 
852 	/*
853 	 * Function may be called in a loop from higher up stack, with libshare
854 	 * initialized for multiple shares (SA_INIT_SHARE_API_SELECTIVE).
855 	 * zfs_init_libshare_arg will refresh the handle's cache if necessary.
856 	 * In this case we do not want to switch to per share initialization.
857 	 * Specify SA_INIT_SHARE_API to do full refresh, if refresh required.
858 	 */
859 	if ((hdl->libzfs_sharehdl != NULL) && (_sa_service != NULL) &&
860 	    (_sa_service(hdl->libzfs_sharehdl) ==
861 	    SA_INIT_SHARE_API_SELECTIVE)) {
862 		service = SA_INIT_SHARE_API;
863 	}
864 
865 	for (curr_proto = proto; *curr_proto != PROTO_END; curr_proto++) {
866 		/*
867 		 * Return success if there are no share options.
868 		 */
869 		if (zfs_prop_get(zhp, proto_table[*curr_proto].p_prop,
870 		    shareopts, sizeof (shareopts), &sourcetype, sourcestr,
871 		    ZFS_MAXPROPLEN, B_FALSE) != 0 ||
872 		    strcmp(shareopts, "off") == 0)
873 			continue;
874 		ret = zfs_init_libshare_arg(hdl, service, zhp);
875 		if (ret != SA_OK) {
876 			(void) zfs_error_fmt(hdl, EZFS_SHARENFSFAILED,
877 			    dgettext(TEXT_DOMAIN, "cannot share '%s': %s"),
878 			    zfs_get_name(zhp), _sa_errorstr != NULL ?
879 			    _sa_errorstr(ret) : "");
880 			return (-1);
881 		}
882 
883 		/*
884 		 * If the 'zoned' property is set, then zfs_is_mountable()
885 		 * will have already bailed out if we are in the global zone.
886 		 * But local zones cannot be NFS servers, so we ignore it for
887 		 * local zones as well.
888 		 */
889 		if (zfs_prop_get_int(zhp, ZFS_PROP_ZONED))
890 			continue;
891 
892 		share = zfs_sa_find_share(hdl->libzfs_sharehdl, mountpoint);
893 		if (share == NULL) {
894 			/*
895 			 * This may be a new file system that was just
896 			 * created so isn't in the internal cache
897 			 * (second time through). Rather than
898 			 * reloading the entire configuration, we can
899 			 * assume ZFS has done the checking and it is
900 			 * safe to add this to the internal
901 			 * configuration.
902 			 */
903 			if (_sa_zfs_process_share(hdl->libzfs_sharehdl,
904 			    NULL, NULL, mountpoint,
905 			    proto_table[*curr_proto].p_name, sourcetype,
906 			    shareopts, sourcestr, zhp->zfs_name) != SA_OK) {
907 				(void) zfs_error_fmt(hdl,
908 				    proto_table[*curr_proto].p_share_err,
909 				    dgettext(TEXT_DOMAIN, "cannot share '%s'"),
910 				    zfs_get_name(zhp));
911 				return (-1);
912 			}
913 			share = zfs_sa_find_share(hdl->libzfs_sharehdl,
914 			    mountpoint);
915 		}
916 		if (share != NULL) {
917 			int err;
918 			err = zfs_sa_enable_share(share,
919 			    proto_table[*curr_proto].p_name);
920 			if (err != SA_OK) {
921 				(void) zfs_error_fmt(hdl,
922 				    proto_table[*curr_proto].p_share_err,
923 				    dgettext(TEXT_DOMAIN, "cannot share '%s'"),
924 				    zfs_get_name(zhp));
925 				return (-1);
926 			}
927 		} else {
928 			(void) zfs_error_fmt(hdl,
929 			    proto_table[*curr_proto].p_share_err,
930 			    dgettext(TEXT_DOMAIN, "cannot share '%s'"),
931 			    zfs_get_name(zhp));
932 			return (-1);
933 		}
934 
935 	}
936 	return (0);
937 }
938 
939 
940 int
941 zfs_share_nfs(zfs_handle_t *zhp)
942 {
943 	return (zfs_share_proto(zhp, nfs_only));
944 }
945 
946 int
947 zfs_share_smb(zfs_handle_t *zhp)
948 {
949 	return (zfs_share_proto(zhp, smb_only));
950 }
951 
952 int
953 zfs_shareall(zfs_handle_t *zhp)
954 {
955 	return (zfs_share_proto(zhp, share_all_proto));
956 }
957 
958 /*
959  * Unshare a filesystem by mountpoint.
960  */
961 static int
962 unshare_one(libzfs_handle_t *hdl, const char *name, const char *mountpoint,
963     zfs_share_proto_t proto)
964 {
965 	sa_share_t share;
966 	int err;
967 	char *mntpt;
968 	int service = SA_INIT_ONE_SHARE_FROM_NAME;
969 
970 	/*
971 	 * Mountpoint could get trashed if libshare calls getmntany
972 	 * which it does during API initialization, so strdup the
973 	 * value.
974 	 */
975 	mntpt = zfs_strdup(hdl, mountpoint);
976 
977 	/*
978 	 * Function may be called in a loop from higher up stack, with libshare
979 	 * initialized for multiple shares (SA_INIT_SHARE_API_SELECTIVE).
980 	 * zfs_init_libshare_arg will refresh the handle's cache if necessary.
981 	 * In this case we do not want to switch to per share initialization.
982 	 * Specify SA_INIT_SHARE_API to do full refresh, if refresh required.
983 	 */
984 	if ((hdl->libzfs_sharehdl != NULL) && (_sa_service != NULL) &&
985 	    (_sa_service(hdl->libzfs_sharehdl) ==
986 	    SA_INIT_SHARE_API_SELECTIVE)) {
987 		service = SA_INIT_SHARE_API;
988 	}
989 
990 	err = zfs_init_libshare_arg(hdl, service, (void *)name);
991 	if (err != SA_OK) {
992 		free(mntpt);	/* don't need the copy anymore */
993 		return (zfs_error_fmt(hdl, proto_table[proto].p_unshare_err,
994 		    dgettext(TEXT_DOMAIN, "cannot unshare '%s': %s"),
995 		    name, _sa_errorstr(err)));
996 	}
997 
998 	share = zfs_sa_find_share(hdl->libzfs_sharehdl, mntpt);
999 	free(mntpt);	/* don't need the copy anymore */
1000 
1001 	if (share != NULL) {
1002 		err = zfs_sa_disable_share(share, proto_table[proto].p_name);
1003 		if (err != SA_OK) {
1004 			return (zfs_error_fmt(hdl,
1005 			    proto_table[proto].p_unshare_err,
1006 			    dgettext(TEXT_DOMAIN, "cannot unshare '%s': %s"),
1007 			    name, _sa_errorstr(err)));
1008 		}
1009 	} else {
1010 		return (zfs_error_fmt(hdl, proto_table[proto].p_unshare_err,
1011 		    dgettext(TEXT_DOMAIN, "cannot unshare '%s': not found"),
1012 		    name));
1013 	}
1014 	return (0);
1015 }
1016 
1017 /*
1018  * Unshare the given filesystem.
1019  */
1020 int
1021 zfs_unshare_proto(zfs_handle_t *zhp, const char *mountpoint,
1022     zfs_share_proto_t *proto)
1023 {
1024 	libzfs_handle_t *hdl = zhp->zfs_hdl;
1025 	struct mnttab entry;
1026 	char *mntpt = NULL;
1027 
1028 	/* check to see if need to unmount the filesystem */
1029 	rewind(zhp->zfs_hdl->libzfs_mnttab);
1030 	if (mountpoint != NULL)
1031 		mountpoint = mntpt = zfs_strdup(hdl, mountpoint);
1032 
1033 	if (mountpoint != NULL || ((zfs_get_type(zhp) == ZFS_TYPE_FILESYSTEM) &&
1034 	    libzfs_mnttab_find(hdl, zfs_get_name(zhp), &entry) == 0)) {
1035 		zfs_share_proto_t *curr_proto;
1036 
1037 		if (mountpoint == NULL)
1038 			mntpt = zfs_strdup(zhp->zfs_hdl, entry.mnt_mountp);
1039 
1040 		for (curr_proto = proto; *curr_proto != PROTO_END;
1041 		    curr_proto++) {
1042 
1043 			if (is_shared(hdl, mntpt, *curr_proto) &&
1044 			    unshare_one(hdl, zhp->zfs_name,
1045 			    mntpt, *curr_proto) != 0) {
1046 				if (mntpt != NULL)
1047 					free(mntpt);
1048 				return (-1);
1049 			}
1050 		}
1051 	}
1052 	if (mntpt != NULL)
1053 		free(mntpt);
1054 
1055 	return (0);
1056 }
1057 
1058 int
1059 zfs_unshare_nfs(zfs_handle_t *zhp, const char *mountpoint)
1060 {
1061 	return (zfs_unshare_proto(zhp, mountpoint, nfs_only));
1062 }
1063 
1064 int
1065 zfs_unshare_smb(zfs_handle_t *zhp, const char *mountpoint)
1066 {
1067 	return (zfs_unshare_proto(zhp, mountpoint, smb_only));
1068 }
1069 
1070 /*
1071  * Same as zfs_unmountall(), but for NFS and SMB unshares.
1072  */
1073 int
1074 zfs_unshareall_proto(zfs_handle_t *zhp, zfs_share_proto_t *proto)
1075 {
1076 	prop_changelist_t *clp;
1077 	int ret;
1078 
1079 	clp = changelist_gather(zhp, ZFS_PROP_SHARENFS, 0, 0);
1080 	if (clp == NULL)
1081 		return (-1);
1082 
1083 	ret = changelist_unshare(clp, proto);
1084 	changelist_free(clp);
1085 
1086 	return (ret);
1087 }
1088 
1089 int
1090 zfs_unshareall_nfs(zfs_handle_t *zhp)
1091 {
1092 	return (zfs_unshareall_proto(zhp, nfs_only));
1093 }
1094 
1095 int
1096 zfs_unshareall_smb(zfs_handle_t *zhp)
1097 {
1098 	return (zfs_unshareall_proto(zhp, smb_only));
1099 }
1100 
1101 int
1102 zfs_unshareall(zfs_handle_t *zhp)
1103 {
1104 	return (zfs_unshareall_proto(zhp, share_all_proto));
1105 }
1106 
1107 int
1108 zfs_unshareall_bypath(zfs_handle_t *zhp, const char *mountpoint)
1109 {
1110 	return (zfs_unshare_proto(zhp, mountpoint, share_all_proto));
1111 }
1112 
1113 /*
1114  * Remove the mountpoint associated with the current dataset, if necessary.
1115  * We only remove the underlying directory if:
1116  *
1117  *	- The mountpoint is not 'none' or 'legacy'
1118  *	- The mountpoint is non-empty
1119  *	- The mountpoint is the default or inherited
1120  *	- The 'zoned' property is set, or we're in a local zone
1121  *
1122  * Any other directories we leave alone.
1123  */
1124 void
1125 remove_mountpoint(zfs_handle_t *zhp)
1126 {
1127 	char mountpoint[ZFS_MAXPROPLEN];
1128 	zprop_source_t source;
1129 
1130 	if (!zfs_is_mountable(zhp, mountpoint, sizeof (mountpoint),
1131 	    &source))
1132 		return;
1133 
1134 	if (source == ZPROP_SRC_DEFAULT ||
1135 	    source == ZPROP_SRC_INHERITED) {
1136 		/*
1137 		 * Try to remove the directory, silently ignoring any errors.
1138 		 * The filesystem may have since been removed or moved around,
1139 		 * and this error isn't really useful to the administrator in
1140 		 * any way.
1141 		 */
1142 		(void) rmdir(mountpoint);
1143 	}
1144 }
1145 
1146 /*
1147  * Add the given zfs handle to the cb_handles array, dynamically reallocating
1148  * the array if it is out of space.
1149  */
1150 void
1151 libzfs_add_handle(get_all_cb_t *cbp, zfs_handle_t *zhp)
1152 {
1153 	if (cbp->cb_alloc == cbp->cb_used) {
1154 		size_t newsz;
1155 		zfs_handle_t **newhandles;
1156 
1157 		newsz = cbp->cb_alloc != 0 ? cbp->cb_alloc * 2 : 64;
1158 		newhandles = zfs_realloc(zhp->zfs_hdl,
1159 		    cbp->cb_handles, cbp->cb_alloc * sizeof (zfs_handle_t *),
1160 		    newsz * sizeof (zfs_handle_t *));
1161 		cbp->cb_handles = newhandles;
1162 		cbp->cb_alloc = newsz;
1163 	}
1164 	cbp->cb_handles[cbp->cb_used++] = zhp;
1165 }
1166 
1167 /*
1168  * Recursive helper function used during file system enumeration
1169  */
1170 static int
1171 zfs_iter_cb(zfs_handle_t *zhp, void *data)
1172 {
1173 	get_all_cb_t *cbp = data;
1174 
1175 	if (!(zfs_get_type(zhp) & ZFS_TYPE_FILESYSTEM)) {
1176 		zfs_close(zhp);
1177 		return (0);
1178 	}
1179 
1180 	if (zfs_prop_get_int(zhp, ZFS_PROP_CANMOUNT) == ZFS_CANMOUNT_NOAUTO) {
1181 		zfs_close(zhp);
1182 		return (0);
1183 	}
1184 
1185 	if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
1186 	    ZFS_KEYSTATUS_UNAVAILABLE) {
1187 		zfs_close(zhp);
1188 		return (0);
1189 	}
1190 
1191 	/*
1192 	 * If this filesystem is inconsistent and has a receive resume
1193 	 * token, we can not mount it.
1194 	 */
1195 	if (zfs_prop_get_int(zhp, ZFS_PROP_INCONSISTENT) &&
1196 	    zfs_prop_get(zhp, ZFS_PROP_RECEIVE_RESUME_TOKEN,
1197 	    NULL, 0, NULL, NULL, 0, B_TRUE) == 0) {
1198 		zfs_close(zhp);
1199 		return (0);
1200 	}
1201 
1202 	libzfs_add_handle(cbp, zhp);
1203 	if (zfs_iter_filesystems(zhp, zfs_iter_cb, cbp) != 0) {
1204 		zfs_close(zhp);
1205 		return (-1);
1206 	}
1207 	return (0);
1208 }
1209 
1210 /*
1211  * Sort comparator that compares two mountpoint paths. We sort these paths so
1212  * that subdirectories immediately follow their parents. This means that we
1213  * effectively treat the '/' character as the lowest value non-nul char.
1214  * Since filesystems from non-global zones can have the same mountpoint
1215  * as other filesystems, the comparator sorts global zone filesystems to
1216  * the top of the list. This means that the global zone will traverse the
1217  * filesystem list in the correct order and can stop when it sees the
1218  * first zoned filesystem. In a non-global zone, only the delegated
1219  * filesystems are seen.
1220  *
1221  * An example sorted list using this comparator would look like:
1222  *
1223  * /foo
1224  * /foo/bar
1225  * /foo/bar/baz
1226  * /foo/baz
1227  * /foo.bar
1228  * /foo (NGZ1)
1229  * /foo (NGZ2)
1230  *
1231  * The mounting code depends on this ordering to deterministically iterate
1232  * over filesystems in order to spawn parallel mount tasks.
1233  */
1234 static int
1235 mountpoint_cmp(const void *arga, const void *argb)
1236 {
1237 	zfs_handle_t *const *zap = arga;
1238 	zfs_handle_t *za = *zap;
1239 	zfs_handle_t *const *zbp = argb;
1240 	zfs_handle_t *zb = *zbp;
1241 	char mounta[MAXPATHLEN];
1242 	char mountb[MAXPATHLEN];
1243 	const char *a = mounta;
1244 	const char *b = mountb;
1245 	boolean_t gota, gotb;
1246 	uint64_t zoneda, zonedb;
1247 
1248 	zoneda = zfs_prop_get_int(za, ZFS_PROP_ZONED);
1249 	zonedb = zfs_prop_get_int(zb, ZFS_PROP_ZONED);
1250 	if (zoneda && !zonedb)
1251 		return (1);
1252 	if (!zoneda && zonedb)
1253 		return (-1);
1254 
1255 	gota = (zfs_get_type(za) == ZFS_TYPE_FILESYSTEM);
1256 	if (gota) {
1257 		verify(zfs_prop_get(za, ZFS_PROP_MOUNTPOINT, mounta,
1258 		    sizeof (mounta), NULL, NULL, 0, B_FALSE) == 0);
1259 	}
1260 	gotb = (zfs_get_type(zb) == ZFS_TYPE_FILESYSTEM);
1261 	if (gotb) {
1262 		verify(zfs_prop_get(zb, ZFS_PROP_MOUNTPOINT, mountb,
1263 		    sizeof (mountb), NULL, NULL, 0, B_FALSE) == 0);
1264 	}
1265 
1266 	if (gota && gotb) {
1267 		while (*a != '\0' && (*a == *b)) {
1268 			a++;
1269 			b++;
1270 		}
1271 		if (*a == *b)
1272 			return (0);
1273 		if (*a == '\0')
1274 			return (-1);
1275 		if (*b == '\0')
1276 			return (1);
1277 		if (*a == '/')
1278 			return (-1);
1279 		if (*b == '/')
1280 			return (1);
1281 		return (*a < *b ? -1 : *a > *b);
1282 	}
1283 
1284 	if (gota)
1285 		return (-1);
1286 	if (gotb)
1287 		return (1);
1288 
1289 	/*
1290 	 * If neither filesystem has a mountpoint, revert to sorting by
1291 	 * dataset name.
1292 	 */
1293 	return (strcmp(zfs_get_name(za), zfs_get_name(zb)));
1294 }
1295 
1296 /*
1297  * Return true if path2 is a child of path1.
1298  */
1299 static boolean_t
1300 libzfs_path_contains(const char *path1, const char *path2)
1301 {
1302 	return (strstr(path2, path1) == path2 && path2[strlen(path1)] == '/');
1303 }
1304 
1305 /*
1306  * Given a mountpoint specified by idx in the handles array, find the first
1307  * non-descendent of that mountpoint and return its index. Descendant paths
1308  * start with the parent's path. This function relies on the ordering
1309  * enforced by mountpoint_cmp().
1310  */
1311 static int
1312 non_descendant_idx(zfs_handle_t **handles, size_t num_handles, int idx)
1313 {
1314 	char parent[ZFS_MAXPROPLEN];
1315 	char child[ZFS_MAXPROPLEN];
1316 	int i;
1317 
1318 	verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, parent,
1319 	    sizeof (parent), NULL, NULL, 0, B_FALSE) == 0);
1320 
1321 	for (i = idx + 1; i < num_handles; i++) {
1322 		verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT, child,
1323 		    sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
1324 		if (!libzfs_path_contains(parent, child))
1325 			break;
1326 	}
1327 	return (i);
1328 }
1329 
1330 typedef struct mnt_param {
1331 	libzfs_handle_t	*mnt_hdl;
1332 	zfs_taskq_t	*mnt_tq;
1333 	zfs_handle_t	**mnt_zhps; /* filesystems to mount */
1334 	size_t		mnt_num_handles;
1335 	int		mnt_idx;	/* Index of selected entry to mount */
1336 	zfs_iter_f	mnt_func;
1337 	void		*mnt_data;
1338 } mnt_param_t;
1339 
1340 /*
1341  * Allocate and populate the parameter struct for mount function, and
1342  * schedule mounting of the entry selected by idx.
1343  */
1344 static void
1345 zfs_dispatch_mount(libzfs_handle_t *hdl, zfs_handle_t **handles,
1346     size_t num_handles, int idx, zfs_iter_f func, void *data, zfs_taskq_t *tq)
1347 {
1348 	mnt_param_t *mnt_param = zfs_alloc(hdl, sizeof (mnt_param_t));
1349 
1350 	mnt_param->mnt_hdl = hdl;
1351 	mnt_param->mnt_tq = tq;
1352 	mnt_param->mnt_zhps = handles;
1353 	mnt_param->mnt_num_handles = num_handles;
1354 	mnt_param->mnt_idx = idx;
1355 	mnt_param->mnt_func = func;
1356 	mnt_param->mnt_data = data;
1357 
1358 	(void) zfs_taskq_dispatch(tq, zfs_mount_task, (void*)mnt_param,
1359 	    ZFS_TQ_SLEEP);
1360 }
1361 
1362 /*
1363  * This is the structure used to keep state of mounting or sharing operations
1364  * during a call to zpool_enable_datasets().
1365  */
1366 typedef struct mount_state {
1367 	/*
1368 	 * ms_mntstatus is set to -1 if any mount fails. While multiple threads
1369 	 * could update this variable concurrently, no synchronization is
1370 	 * needed as it's only ever set to -1.
1371 	 */
1372 	int		ms_mntstatus;
1373 	int		ms_mntflags;
1374 	const char	*ms_mntopts;
1375 } mount_state_t;
1376 
1377 static int
1378 zfs_mount_one(zfs_handle_t *zhp, void *arg)
1379 {
1380 	mount_state_t *ms = arg;
1381 	int ret = 0;
1382 
1383 	if (zfs_prop_get_int(zhp, ZFS_PROP_KEYSTATUS) ==
1384 	    ZFS_KEYSTATUS_UNAVAILABLE)
1385 		return (0);
1386 
1387 	if (zfs_mount(zhp, ms->ms_mntopts, ms->ms_mntflags) != 0)
1388 		ret = ms->ms_mntstatus = -1;
1389 	return (ret);
1390 }
1391 
1392 static int
1393 zfs_share_one(zfs_handle_t *zhp, void *arg)
1394 {
1395 	mount_state_t *ms = arg;
1396 	int ret = 0;
1397 
1398 	if (zfs_share(zhp) != 0)
1399 		ret = ms->ms_mntstatus = -1;
1400 	return (ret);
1401 }
1402 
1403 /*
1404  * Task queue function to mount one file system. On completion, it finds and
1405  * schedules its children to be mounted. This depends on the sorting done in
1406  * zfs_foreach_mountpoint(). Note that the degenerate case (chain of entries
1407  * each descending from the previous) will have no parallelism since we always
1408  * have to wait for the parent to finish mounting before we can schedule
1409  * its children.
1410  */
1411 static void
1412 zfs_mount_task(void *arg)
1413 {
1414 	mnt_param_t *mp = arg;
1415 	int idx = mp->mnt_idx;
1416 	zfs_handle_t **handles = mp->mnt_zhps;
1417 	size_t num_handles = mp->mnt_num_handles;
1418 	char mountpoint[ZFS_MAXPROPLEN];
1419 
1420 	verify(zfs_prop_get(handles[idx], ZFS_PROP_MOUNTPOINT, mountpoint,
1421 	    sizeof (mountpoint), NULL, NULL, 0, B_FALSE) == 0);
1422 
1423 	if (mp->mnt_func(handles[idx], mp->mnt_data) != 0)
1424 		return;
1425 
1426 	/*
1427 	 * We dispatch tasks to mount filesystems with mountpoints underneath
1428 	 * this one. We do this by dispatching the next filesystem with a
1429 	 * descendant mountpoint of the one we just mounted, then skip all of
1430 	 * its descendants, dispatch the next descendant mountpoint, and so on.
1431 	 * The non_descendant_idx() function skips over filesystems that are
1432 	 * descendants of the filesystem we just dispatched.
1433 	 */
1434 	for (int i = idx + 1; i < num_handles;
1435 	    i = non_descendant_idx(handles, num_handles, i)) {
1436 		char child[ZFS_MAXPROPLEN];
1437 		verify(zfs_prop_get(handles[i], ZFS_PROP_MOUNTPOINT,
1438 		    child, sizeof (child), NULL, NULL, 0, B_FALSE) == 0);
1439 
1440 		if (!libzfs_path_contains(mountpoint, child))
1441 			break; /* not a descendant, return */
1442 		zfs_dispatch_mount(mp->mnt_hdl, handles, num_handles, i,
1443 		    mp->mnt_func, mp->mnt_data, mp->mnt_tq);
1444 	}
1445 	free(mp);
1446 }
1447 
1448 /*
1449  * Issue the func callback for each ZFS handle contained in the handles
1450  * array. This function is used to mount all datasets, and so this function
1451  * guarantees that filesystems for parent mountpoints are called before their
1452  * children. As such, before issuing any callbacks, we first sort the array
1453  * of handles by mountpoint.
1454  *
1455  * Callbacks are issued in one of two ways:
1456  *
1457  * 1. Sequentially: If the parallel argument is B_FALSE or the ZFS_SERIAL_MOUNT
1458  *    environment variable is set, then we issue callbacks sequentially.
1459  *
1460  * 2. In parallel: If the parallel argument is B_TRUE and the ZFS_SERIAL_MOUNT
1461  *    environment variable is not set, then we use a taskq to dispatch threads
1462  *    to mount filesystems is parallel. This function dispatches tasks to mount
1463  *    the filesystems at the top-level mountpoints, and these tasks in turn
1464  *    are responsible for recursively mounting filesystems in their children
1465  *    mountpoints.
1466  */
1467 void
1468 zfs_foreach_mountpoint(libzfs_handle_t *hdl, zfs_handle_t **handles,
1469     size_t num_handles, zfs_iter_f func, void *data, boolean_t parallel)
1470 {
1471 	zoneid_t zoneid = getzoneid();
1472 
1473 	/*
1474 	 * The ZFS_SERIAL_MOUNT environment variable is an undocumented
1475 	 * variable that can be used as a convenience to do a/b comparison
1476 	 * of serial vs. parallel mounting.
1477 	 */
1478 	boolean_t serial_mount = !parallel ||
1479 	    (getenv("ZFS_SERIAL_MOUNT") != NULL);
1480 
1481 	/*
1482 	 * Sort the datasets by mountpoint. See mountpoint_cmp for details
1483 	 * of how these are sorted.
1484 	 */
1485 	qsort(handles, num_handles, sizeof (zfs_handle_t *), mountpoint_cmp);
1486 
1487 	if (serial_mount) {
1488 		for (int i = 0; i < num_handles; i++) {
1489 			func(handles[i], data);
1490 		}
1491 		return;
1492 	}
1493 
1494 	/*
1495 	 * Issue the callback function for each dataset using a parallel
1496 	 * algorithm that uses a taskq to manage threads.
1497 	 */
1498 	zfs_taskq_t *tq = zfs_taskq_create("mount_taskq", mount_tq_nthr, 0,
1499 	    mount_tq_nthr, mount_tq_nthr, ZFS_TASKQ_PREPOPULATE);
1500 
1501 	/*
1502 	 * There may be multiple "top level" mountpoints outside of the pool's
1503 	 * root mountpoint, e.g.: /foo /bar. Dispatch a mount task for each of
1504 	 * these.
1505 	 */
1506 	for (int i = 0; i < num_handles;
1507 	    i = non_descendant_idx(handles, num_handles, i)) {
1508 		/*
1509 		 * Since the mountpoints have been sorted so that the zoned
1510 		 * filesystems are at the end, a zoned filesystem seen from
1511 		 * the global zone means that we're done.
1512 		 */
1513 		if (zoneid == GLOBAL_ZONEID &&
1514 		    zfs_prop_get_int(handles[i], ZFS_PROP_ZONED))
1515 			break;
1516 		zfs_dispatch_mount(hdl, handles, num_handles, i, func, data,
1517 		    tq);
1518 	}
1519 
1520 	zfs_taskq_wait(tq); /* wait for all scheduled mounts to complete */
1521 	zfs_taskq_destroy(tq);
1522 }
1523 
1524 /*
1525  * Mount and share all datasets within the given pool.  This assumes that no
1526  * datasets within the pool are currently mounted.
1527  */
1528 #pragma weak zpool_mount_datasets = zpool_enable_datasets
1529 int
1530 zpool_enable_datasets(zpool_handle_t *zhp, const char *mntopts, int flags)
1531 {
1532 	get_all_cb_t cb = { 0 };
1533 	mount_state_t ms = { 0 };
1534 	zfs_handle_t *zfsp;
1535 	sa_init_selective_arg_t sharearg;
1536 	int ret = 0;
1537 
1538 	if ((zfsp = zfs_open(zhp->zpool_hdl, zhp->zpool_name,
1539 	    ZFS_TYPE_DATASET)) == NULL)
1540 		goto out;
1541 
1542 
1543 	/*
1544 	 * Gather all non-snapshot datasets within the pool. Start by adding
1545 	 * the root filesystem for this pool to the list, and then iterate
1546 	 * over all child filesystems.
1547 	 */
1548 	libzfs_add_handle(&cb, zfsp);
1549 	if (zfs_iter_filesystems(zfsp, zfs_iter_cb, &cb) != 0)
1550 		goto out;
1551 
1552 	ms.ms_mntopts = mntopts;
1553 	ms.ms_mntflags = flags;
1554 	zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
1555 	    zfs_mount_one, &ms, B_TRUE);
1556 	if (ms.ms_mntstatus != 0)
1557 		ret = ms.ms_mntstatus;
1558 
1559 	/*
1560 	 * Initialize libshare SA_INIT_SHARE_API_SELECTIVE here
1561 	 * to avoid unnecessary load/unload of the libshare API
1562 	 * per shared dataset downstream.
1563 	 */
1564 	sharearg.zhandle_arr = cb.cb_handles;
1565 	sharearg.zhandle_len = cb.cb_used;
1566 	if ((ret = zfs_init_libshare_arg(zhp->zpool_hdl,
1567 	    SA_INIT_SHARE_API_SELECTIVE, &sharearg)) != 0)
1568 		goto out;
1569 
1570 	ms.ms_mntstatus = 0;
1571 	zfs_foreach_mountpoint(zhp->zpool_hdl, cb.cb_handles, cb.cb_used,
1572 	    zfs_share_one, &ms, B_FALSE);
1573 	if (ms.ms_mntstatus != 0)
1574 		ret = ms.ms_mntstatus;
1575 
1576 out:
1577 	for (int i = 0; i < cb.cb_used; i++)
1578 		zfs_close(cb.cb_handles[i]);
1579 	free(cb.cb_handles);
1580 
1581 	return (ret);
1582 }
1583 
1584 static int
1585 mountpoint_compare(const void *a, const void *b)
1586 {
1587 	const char *mounta = *((char **)a);
1588 	const char *mountb = *((char **)b);
1589 
1590 	return (strcmp(mountb, mounta));
1591 }
1592 
1593 /* alias for 2002/240 */
1594 #pragma weak zpool_unmount_datasets = zpool_disable_datasets
1595 /*
1596  * Unshare and unmount all datasets within the given pool.  We don't want to
1597  * rely on traversing the DSL to discover the filesystems within the pool,
1598  * because this may be expensive (if not all of them are mounted), and can fail
1599  * arbitrarily (on I/O error, for example).  Instead, we walk /etc/mnttab and
1600  * gather all the filesystems that are currently mounted.
1601  */
1602 int
1603 zpool_disable_datasets(zpool_handle_t *zhp, boolean_t force)
1604 {
1605 	int used, alloc;
1606 	struct mnttab entry;
1607 	size_t namelen;
1608 	char **mountpoints = NULL;
1609 	zfs_handle_t **datasets = NULL;
1610 	libzfs_handle_t *hdl = zhp->zpool_hdl;
1611 	int i;
1612 	int ret = -1;
1613 	int flags = (force ? MS_FORCE : 0);
1614 	sa_init_selective_arg_t sharearg;
1615 
1616 	namelen = strlen(zhp->zpool_name);
1617 
1618 	rewind(hdl->libzfs_mnttab);
1619 	used = alloc = 0;
1620 	while (getmntent(hdl->libzfs_mnttab, &entry) == 0) {
1621 		/*
1622 		 * Ignore non-ZFS entries.
1623 		 */
1624 		if (entry.mnt_fstype == NULL ||
1625 		    strcmp(entry.mnt_fstype, MNTTYPE_ZFS) != 0)
1626 			continue;
1627 
1628 		/*
1629 		 * Ignore filesystems not within this pool.
1630 		 */
1631 		if (entry.mnt_mountp == NULL ||
1632 		    strncmp(entry.mnt_special, zhp->zpool_name, namelen) != 0 ||
1633 		    (entry.mnt_special[namelen] != '/' &&
1634 		    entry.mnt_special[namelen] != '\0'))
1635 			continue;
1636 
1637 		/*
1638 		 * At this point we've found a filesystem within our pool.  Add
1639 		 * it to our growing list.
1640 		 */
1641 		if (used == alloc) {
1642 			if (alloc == 0) {
1643 				if ((mountpoints = zfs_alloc(hdl,
1644 				    8 * sizeof (void *))) == NULL)
1645 					goto out;
1646 
1647 				if ((datasets = zfs_alloc(hdl,
1648 				    8 * sizeof (void *))) == NULL)
1649 					goto out;
1650 
1651 				alloc = 8;
1652 			} else {
1653 				void *ptr;
1654 
1655 				if ((ptr = zfs_realloc(hdl, mountpoints,
1656 				    alloc * sizeof (void *),
1657 				    alloc * 2 * sizeof (void *))) == NULL)
1658 					goto out;
1659 				mountpoints = ptr;
1660 
1661 				if ((ptr = zfs_realloc(hdl, datasets,
1662 				    alloc * sizeof (void *),
1663 				    alloc * 2 * sizeof (void *))) == NULL)
1664 					goto out;
1665 				datasets = ptr;
1666 
1667 				alloc *= 2;
1668 			}
1669 		}
1670 
1671 		if ((mountpoints[used] = zfs_strdup(hdl,
1672 		    entry.mnt_mountp)) == NULL)
1673 			goto out;
1674 
1675 		/*
1676 		 * This is allowed to fail, in case there is some I/O error.  It
1677 		 * is only used to determine if we need to remove the underlying
1678 		 * mountpoint, so failure is not fatal.
1679 		 */
1680 		datasets[used] = make_dataset_handle(hdl, entry.mnt_special);
1681 
1682 		used++;
1683 	}
1684 
1685 	/*
1686 	 * At this point, we have the entire list of filesystems, so sort it by
1687 	 * mountpoint.
1688 	 */
1689 	sharearg.zhandle_arr = datasets;
1690 	sharearg.zhandle_len = used;
1691 	ret = zfs_init_libshare_arg(hdl, SA_INIT_SHARE_API_SELECTIVE,
1692 	    &sharearg);
1693 	if (ret != 0)
1694 		goto out;
1695 	qsort(mountpoints, used, sizeof (char *), mountpoint_compare);
1696 
1697 	/*
1698 	 * Walk through and first unshare everything.
1699 	 */
1700 	for (i = 0; i < used; i++) {
1701 		zfs_share_proto_t *curr_proto;
1702 		for (curr_proto = share_all_proto; *curr_proto != PROTO_END;
1703 		    curr_proto++) {
1704 			if (is_shared(hdl, mountpoints[i], *curr_proto) &&
1705 			    unshare_one(hdl, mountpoints[i],
1706 			    mountpoints[i], *curr_proto) != 0)
1707 				goto out;
1708 		}
1709 	}
1710 
1711 	/*
1712 	 * Now unmount everything, removing the underlying directories as
1713 	 * appropriate.
1714 	 */
1715 	for (i = 0; i < used; i++) {
1716 		if (unmount_one(hdl, mountpoints[i], flags) != 0)
1717 			goto out;
1718 	}
1719 
1720 	for (i = 0; i < used; i++) {
1721 		if (datasets[i])
1722 			remove_mountpoint(datasets[i]);
1723 	}
1724 
1725 	ret = 0;
1726 out:
1727 	for (i = 0; i < used; i++) {
1728 		if (datasets[i])
1729 			zfs_close(datasets[i]);
1730 		free(mountpoints[i]);
1731 	}
1732 	free(datasets);
1733 	free(mountpoints);
1734 
1735 	return (ret);
1736 }
1737