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