xref: /netbsd-src/external/cddl/osnet/dist/uts/common/fs/zfs/zfs_vfsops.c (revision 45fbb8e465559ab4448f4c2267739db090b49691)
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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24  * All rights reserved.
25  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
26  * Copyright (c) 2014 Integros [integros.com]
27  */
28 
29 /* Portions Copyright 2010 Robert Milkowski */
30 
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sysmacros.h>
36 #include <sys/kmem.h>
37 #include <sys/acl.h>
38 #include <sys/vnode.h>
39 #include <sys/vfs.h>
40 #include <sys/mntent.h>
41 #include <sys/mount.h>
42 #include <sys/cmn_err.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_dir.h>
45 #include <sys/zil.h>
46 #include <sys/fs/zfs.h>
47 #include <sys/dmu.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/dsl_dataset.h>
50 #include <sys/dsl_deleg.h>
51 #include <sys/spa.h>
52 #include <sys/zap.h>
53 #include <sys/sa.h>
54 #include <sys/sa_impl.h>
55 #include <sys/varargs.h>
56 #include <sys/policy.h>
57 #include <sys/atomic.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/zfs_ctldir.h>
60 #include <sys/zfs_fuid.h>
61 #include <sys/sunddi.h>
62 #include <sys/dnlc.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
65 #include "zfs_comutil.h"
66 
67 #ifdef __FreeBSD_kernel__
68 
69 #include <sys/jail.h>
70 
71 struct mtx zfs_debug_mtx;
72 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
73 
74 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
75 
76 int zfs_super_owner;
77 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
78     "File system owner can perform privileged operation on his file systems");
79 
80 int zfs_debug_level;
81 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0,
82     "Debug level");
83 
84 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
85 static int zfs_version_acl = ZFS_ACL_VERSION;
86 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
87     "ZFS_ACL_VERSION");
88 static int zfs_version_spa = SPA_VERSION;
89 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
90     "SPA_VERSION");
91 static int zfs_version_zpl = ZPL_VERSION;
92 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
93     "ZPL_VERSION");
94 
95 static int zfs_mount(vfs_t *vfsp);
96 static int zfs_umount(vfs_t *vfsp, int fflag);
97 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
98 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
99 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
100 static int zfs_sync(vfs_t *vfsp, int waitfor);
101 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
102     struct ucred **credanonp, int *numsecflavors, int **secflavors);
103 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
104 static void zfs_objset_close(zfsvfs_t *zfsvfs);
105 static void zfs_freevfs(vfs_t *vfsp);
106 
107 struct vfsops zfs_vfsops = {
108 	.vfs_mount =		zfs_mount,
109 	.vfs_unmount =		zfs_umount,
110 	.vfs_root =		zfs_root,
111 	.vfs_statfs =		zfs_statfs,
112 	.vfs_vget =		zfs_vget,
113 	.vfs_sync =		zfs_sync,
114 	.vfs_checkexp =		zfs_checkexp,
115 	.vfs_fhtovp =		zfs_fhtovp,
116 };
117 
118 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
119 
120 #endif /* __FreeBSD_kernel__ */
121 
122 #ifdef __NetBSD__
123 
124 #include <sys/fstrans.h>
125 #include <sys/mkdev.h>
126 #include <miscfs/genfs/genfs.h>
127 
128 int zfs_debug_level;
129 kmutex_t zfs_debug_mtx;
130 
131 #define	DROP_GIANT()	/* nothing */
132 #define PICKUP_GIANT()	/* nothing */
133 #define vfs_stdsync(a, b) 0
134 
135 static int zfs_mount(vfs_t *vfsp, const char *path, void *data, size_t *data_len);
136 static int zfs_umount(vfs_t *vfsp, int fflag);
137 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
138 static int zfs_statvfs(vfs_t *vfsp, struct statvfs *statp);
139 static int zfs_netbsd_vptofh(vnode_t *vp, fid_t *fidp, size_t *fh_size);
140 static int zfs_netbsd_fhtovp(vfs_t *vfsp, fid_t *fidp, int lktype, vnode_t **vpp);
141 static int zfs_vget(vfs_t *vfsp, ino_t ino, int lktype, vnode_t **vpp);
142 static int zfs_sync(vfs_t *vfsp, int waitfor);
143 static int zfs_netbsd_sync(vfs_t *vfsp, int waitfor, cred_t *cr);
144 static void zfs_freevfs(vfs_t *vfsp);
145 
146 void zfs_init(void);
147 void zfs_fini(void);
148 
149 extern const struct vnodeopv_desc zfs_vnodeop_opv_desc;
150 extern const struct vnodeopv_desc zfs_specop_opv_desc;
151 extern const struct vnodeopv_desc zfs_fifoop_opv_desc;
152 extern const struct vnodeopv_desc zfs_sfsop_opv_desc;
153 
154 static const struct vnodeopv_desc * const zfs_vnodeop_descs[] = {
155 	&zfs_vnodeop_opv_desc,
156 	&zfs_specop_opv_desc,
157 	&zfs_fifoop_opv_desc,
158 	&zfs_sfsop_opv_desc,
159 	NULL,
160 };
161 
162 struct vfsops zfs_vfsops = {
163 	.vfs_name = MOUNT_ZFS,
164 	.vfs_min_mount_data = sizeof(struct zfs_args),
165 	.vfs_opv_descs = zfs_vnodeop_descs,
166 	.vfs_mount = zfs_mount,
167 	.vfs_unmount = zfs_umount,
168 	.vfs_root = zfs_root,
169 	.vfs_statvfs = zfs_statvfs,
170 	.vfs_sync = zfs_netbsd_sync,
171 	.vfs_vget = zfs_vget,
172 	.vfs_loadvnode = zfs_loadvnode,
173 	.vfs_newvnode = zfs_newvnode,
174 	.vfs_init = zfs_init,
175 	.vfs_done = zfs_fini,
176 	.vfs_start = (void *)nullop,
177 	.vfs_renamelock_enter = genfs_renamelock_enter,
178 	.vfs_renamelock_exit = genfs_renamelock_exit,
179 	.vfs_reinit = (void *)nullop,
180 	.vfs_vptofh = zfs_netbsd_vptofh,
181 	.vfs_fhtovp = zfs_netbsd_fhtovp,
182 	.vfs_quotactl = (void *)eopnotsupp,
183 	.vfs_extattrctl = (void *)eopnotsupp,
184 	.vfs_suspendctl = genfs_suspendctl,
185 	.vfs_snapshot = (void *)eopnotsupp,
186 	.vfs_fsync = (void *)eopnotsupp,
187 };
188 
189 static bool
zfs_sync_selector(void * cl,struct vnode * vp)190 zfs_sync_selector(void *cl, struct vnode *vp)
191 {
192 	znode_t *zp;
193 
194 	/*
195 	 * Skip the vnode/inode if inaccessible, is control node or if the
196 	 * atime is clean.
197 	 */
198 	if (zfsctl_is_node(vp))
199 		return false;
200 	zp = VTOZ(vp);
201 	return zp != NULL && vp->v_type != VNON && zp->z_atime_dirty != 0
202 	    && !zp->z_unlinked;
203 }
204 
205 static int
zfs_netbsd_sync(vfs_t * vfsp,int waitfor,cred_t * cr)206 zfs_netbsd_sync(vfs_t *vfsp, int waitfor, cred_t *cr)
207 {
208 	struct vnode_iterator *marker;
209 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
210 	vnode_t *vp;
211 
212 	/*
213 	 * On NetBSD, we need to push out atime updates.  Solaris does
214 	 * this during VOP_INACTIVE, but that does not work well with the
215 	 * BSD VFS, so we do it in batch here.
216 	 */
217 	vfs_vnode_iterator_init(vfsp, &marker);
218 	while ((vp = vfs_vnode_iterator_next(marker, zfs_sync_selector, NULL)))
219 	{
220 		znode_t *zp;
221 		dmu_buf_t *dbp;
222 		dmu_tx_t *tx;
223 		int error;
224 
225 		error = vn_lock(vp, LK_EXCLUSIVE);
226 		if (error) {
227 			VN_RELE(vp);
228 			continue;
229 		}
230 		ZFS_ENTER(zfsvfs);
231 		zp = VTOZ(vp);
232 		tx = dmu_tx_create(zfsvfs->z_os);
233 		dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE);
234 		zfs_sa_upgrade_txholds(tx, zp);
235 		error = dmu_tx_assign(tx, TXG_WAIT);
236 		if (error) {
237 			dmu_tx_abort(tx);
238 		} else {
239 			(void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs),
240 			    (void *)&zp->z_atime, sizeof (zp->z_atime), tx);
241 			zp->z_atime_dirty = 0;
242 			dmu_tx_commit(tx);
243 		}
244 		ZFS_EXIT(zfsvfs);
245 		vput(vp);
246 	}
247 	vfs_vnode_iterator_destroy(marker);
248 
249 	/*
250 	 * Then do the regular ZFS stuff.
251 	 */
252 	return zfs_sync(vfsp, waitfor);
253 }
254 
255 static int
zfs_netbsd_vptofh(vnode_t * vp,fid_t * fidp,size_t * fh_size)256 zfs_netbsd_vptofh(vnode_t *vp, fid_t *fidp, size_t *fh_size)
257 {
258 	znode_t		*zp;
259 	zfsvfs_t	*zfsvfs;
260 	uint32_t	gen;
261 	uint64_t	gen64;
262 	uint64_t	object;
263 	zfid_short_t	*zfid;
264 	int		size, i, error;
265 
266 	if (zfsctl_is_node(vp))
267 		return zfsctl_vptofh(vp, fidp, fh_size);
268 
269 	zp = VTOZ(vp);
270 	zfsvfs = zp->z_zfsvfs;
271 	object = zp->z_id;
272 
273 	ZFS_ENTER(zfsvfs);
274 	ZFS_VERIFY_ZP(zp);
275 
276 	if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs),
277 	    &gen64, sizeof (uint64_t))) != 0) {
278 		ZFS_EXIT(zfsvfs);
279 		return (error);
280 	}
281 
282 	gen = (uint32_t)gen64;
283 
284 	size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN;
285 
286 	if (*fh_size < size) {
287 		ZFS_EXIT(zfsvfs);
288 		*fh_size = size;
289 		return SET_ERROR(E2BIG);
290 	}
291 	*fh_size = size;
292 
293 	zfid = (zfid_short_t *)fidp;
294 
295 	zfid->zf_len = size;
296 
297 	for (i = 0; i < sizeof (zfid->zf_object); i++)
298 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
299 
300 	/* Must have a non-zero generation number to distinguish from .zfs */
301 	if (gen == 0)
302 		gen = 1;
303 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
304 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
305 
306 	if (size == LONG_FID_LEN) {
307 		uint64_t	objsetid = dmu_objset_id(zfsvfs->z_os);
308 		zfid_long_t	*zlfid;
309 
310 		zlfid = (zfid_long_t *)fidp;
311 
312 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
313 			zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
314 
315 		/* XXX - this should be the generation number for the objset */
316 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
317 			zlfid->zf_setgen[i] = 0;
318 	}
319 
320 	ZFS_EXIT(zfsvfs);
321 	return 0;
322 }
323 
324 static int
zfs_netbsd_fhtovp(vfs_t * vfsp,fid_t * fidp,int lktype,vnode_t ** vpp)325 zfs_netbsd_fhtovp(vfs_t *vfsp, fid_t *fidp, int lktype, vnode_t **vpp)
326 {
327 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
328 	znode_t		*zp;
329 	vnode_t		*dvp;
330 	uint64_t	object = 0;
331 	uint64_t	fid_gen = 0;
332 	uint64_t	gen_mask;
333 	uint64_t	zp_gen;
334 	int 		i, err;
335 
336 	*vpp = NULL;
337 
338 	ZFS_ENTER(zfsvfs);
339 
340 	if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
341 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
342 		uint64_t	objsetid = 0;
343 		uint64_t	setgen = 0;
344 
345 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
346 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
347 
348 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
349 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
350 
351 		ZFS_EXIT(zfsvfs);
352 
353 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
354 		if (err)
355 			return (SET_ERROR(EINVAL));
356 		ZFS_ENTER(zfsvfs);
357 	}
358 
359 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
360 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
361 
362 		for (i = 0; i < sizeof (zfid->zf_object); i++)
363 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
364 
365 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
366 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
367 	} else {
368 		ZFS_EXIT(zfsvfs);
369 		return (SET_ERROR(EINVAL));
370 	}
371 
372 	/* A zero fid_gen means we are in the .zfs control directories */
373 	if (fid_gen == 0 &&
374 	     (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
375 		ZFS_EXIT(zfsvfs);
376 		if (object == ZFSCTL_INO_ROOT)
377 			err = zfsctl_root(zfsvfs, vpp);
378 		else
379 			err = zfsctl_snapshot(zfsvfs, vpp);
380 		if (err)
381 			return err;
382 		err = vn_lock(*vpp, LK_EXCLUSIVE);
383 		if (err) {
384 			vrele(*vpp);
385 			*vpp = NULL;
386 			return err;
387 		}
388 		return 0;
389 	}
390 
391 	gen_mask = -1ULL >> (64 - 8 * i);
392 
393 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
394 	if (err = zfs_zget(zfsvfs, object, &zp)) {
395 		ZFS_EXIT(zfsvfs);
396 		return SET_ERROR(ESTALE);
397 	}
398 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
399 	    sizeof (uint64_t));
400 	zp_gen = zp_gen & gen_mask;
401 	if (zp_gen == 0)
402 		zp_gen = 1;
403 	if (zp->z_unlinked || zp_gen != fid_gen) {
404 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
405 		VN_RELE(ZTOV(zp));
406 		ZFS_EXIT(zfsvfs);
407 		return SET_ERROR(ESTALE);
408 	}
409 
410 	*vpp = ZTOV(zp);
411 	ZFS_EXIT(zfsvfs);
412 	err = vn_lock(*vpp, lktype);
413 	if (err) {
414 		vrele(*vpp);
415 		*vpp = NULL;
416 		return err;
417 	}
418 	return 0;
419 }
420 #endif /* __NetBSD__ */
421 
422 /*
423  * We need to keep a count of active fs's.
424  * This is necessary to prevent our module
425  * from being unloaded after a umount -f
426  */
427 static uint32_t	zfs_active_fs_count = 0;
428 
429 /*ARGSUSED*/
430 static int
zfs_sync(vfs_t * vfsp,int waitfor)431 zfs_sync(vfs_t *vfsp, int waitfor)
432 {
433         /*
434 	 * Data integrity is job one.  We don't want a compromised kernel
435 	 * writing to the storage pool, so we never sync during panic.
436 	 */
437 	if (panicstr)
438 		return (0);
439 
440 	/*
441 	 * Ignore the system syncher.  ZFS already commits async data
442 	 * at zfs_txg_timeout intervals.
443 	 */
444 	if (waitfor == MNT_LAZY)
445 		return (0);
446 
447 	if (vfsp != NULL) {
448 		/*
449 		 * Sync a specific filesystem.
450 		 */
451 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
452 		dsl_pool_t *dp;
453 		int error;
454 
455 		error = vfs_stdsync(vfsp, waitfor);
456 		if (error != 0)
457 			return (error);
458 
459 		ZFS_ENTER(zfsvfs);
460 		dp = dmu_objset_pool(zfsvfs->z_os);
461 
462 		/*
463 		 * If the system is shutting down, then skip any
464 		 * filesystems which may exist on a suspended pool.
465 		 */
466 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
467 			ZFS_EXIT(zfsvfs);
468 			return (0);
469 		}
470 
471 		if (zfsvfs->z_log != NULL)
472 			zil_commit(zfsvfs->z_log, 0);
473 
474 		ZFS_EXIT(zfsvfs);
475 	} else {
476 		/*
477 		 * Sync all ZFS filesystems.  This is what happens when you
478 		 * run sync(1M).  Unlike other filesystems, ZFS honors the
479 		 * request by waiting for all pools to commit all dirty data.
480 		 */
481 		spa_sync_allpools();
482 	}
483 
484 	return (0);
485 }
486 
487 #ifdef illumos
488 static int
zfs_create_unique_device(dev_t * dev)489 zfs_create_unique_device(dev_t *dev)
490 {
491 	major_t new_major;
492 
493 	do {
494 		ASSERT3U(zfs_minor, <=, MAXMIN32);
495 		minor_t start = zfs_minor;
496 		do {
497 			mutex_enter(&zfs_dev_mtx);
498 			if (zfs_minor >= MAXMIN32) {
499 				/*
500 				 * If we're still using the real major
501 				 * keep out of /dev/zfs and /dev/zvol minor
502 				 * number space.  If we're using a getudev()'ed
503 				 * major number, we can use all of its minors.
504 				 */
505 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
506 					zfs_minor = ZFS_MIN_MINOR;
507 				else
508 					zfs_minor = 0;
509 			} else {
510 				zfs_minor++;
511 			}
512 			*dev = makedevice(zfs_major, zfs_minor);
513 			mutex_exit(&zfs_dev_mtx);
514 		} while (vfs_devismounted(*dev) && zfs_minor != start);
515 #ifdef illumos
516 		if (zfs_minor == start) {
517 			/*
518 			 * We are using all ~262,000 minor numbers for the
519 			 * current major number.  Create a new major number.
520 			 */
521 			if ((new_major = getudev()) == (major_t)-1) {
522 				cmn_err(CE_WARN,
523 				    "zfs_mount: Can't get unique major "
524 				    "device number.");
525 				return (-1);
526 			}
527 			mutex_enter(&zfs_dev_mtx);
528 			zfs_major = new_major;
529 			zfs_minor = 0;
530 
531 			mutex_exit(&zfs_dev_mtx);
532 		} else {
533 			break;
534 		}
535 		/* CONSTANTCONDITION */
536 #endif
537 	} while (1);
538 
539 	return (0);
540 }
541 #endif	/* illumos */
542 
543 
544 static void
atime_changed_cb(void * arg,uint64_t newval)545 atime_changed_cb(void *arg, uint64_t newval)
546 {
547 	zfsvfs_t *zfsvfs = arg;
548 
549 	if (newval == TRUE) {
550 		zfsvfs->z_atime = TRUE;
551 		zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
552 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
553 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
554 	} else {
555 		zfsvfs->z_atime = FALSE;
556 		zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
557 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
558 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
559 	}
560 }
561 
562 static void
xattr_changed_cb(void * arg,uint64_t newval)563 xattr_changed_cb(void *arg, uint64_t newval)
564 {
565 	zfsvfs_t *zfsvfs = arg;
566 
567 	if (newval == TRUE) {
568 		/* XXX locking on vfs_flag? */
569 #ifdef TODO
570 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
571 #endif
572 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
573 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
574 	} else {
575 		/* XXX locking on vfs_flag? */
576 #ifdef TODO
577 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
578 #endif
579 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
580 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
581 	}
582 }
583 
584 static void
blksz_changed_cb(void * arg,uint64_t newval)585 blksz_changed_cb(void *arg, uint64_t newval)
586 {
587 	zfsvfs_t *zfsvfs = arg;
588 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
589 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
590 	ASSERT(ISP2(newval));
591 
592 	zfsvfs->z_max_blksz = newval;
593 	zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
594 }
595 
596 static void
readonly_changed_cb(void * arg,uint64_t newval)597 readonly_changed_cb(void *arg, uint64_t newval)
598 {
599 	zfsvfs_t *zfsvfs = arg;
600 
601 	if (newval) {
602 		/* XXX locking on vfs_flag? */
603 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
604 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
605 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
606 	} else {
607 		/* XXX locking on vfs_flag? */
608 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
609 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
610 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
611 	}
612 }
613 
614 static void
setuid_changed_cb(void * arg,uint64_t newval)615 setuid_changed_cb(void *arg, uint64_t newval)
616 {
617 	zfsvfs_t *zfsvfs = arg;
618 
619 	if (newval == FALSE) {
620 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
621 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
622 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
623 	} else {
624 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
625 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
626 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
627 	}
628 }
629 
630 static void
exec_changed_cb(void * arg,uint64_t newval)631 exec_changed_cb(void *arg, uint64_t newval)
632 {
633 	zfsvfs_t *zfsvfs = arg;
634 
635 	if (newval == FALSE) {
636 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
637 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
638 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
639 	} else {
640 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
641 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
642 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
643 	}
644 }
645 
646 /*
647  * The nbmand mount option can be changed at mount time.
648  * We can't allow it to be toggled on live file systems or incorrect
649  * behavior may be seen from cifs clients
650  *
651  * This property isn't registered via dsl_prop_register(), but this callback
652  * will be called when a file system is first mounted
653  */
654 static void
nbmand_changed_cb(void * arg,uint64_t newval)655 nbmand_changed_cb(void *arg, uint64_t newval)
656 {
657 	zfsvfs_t *zfsvfs = arg;
658 	if (newval == FALSE) {
659 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
660 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
661 	} else {
662 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
663 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
664 	}
665 }
666 
667 static void
snapdir_changed_cb(void * arg,uint64_t newval)668 snapdir_changed_cb(void *arg, uint64_t newval)
669 {
670 	zfsvfs_t *zfsvfs = arg;
671 
672 	zfsvfs->z_show_ctldir = newval;
673 }
674 
675 static void
vscan_changed_cb(void * arg,uint64_t newval)676 vscan_changed_cb(void *arg, uint64_t newval)
677 {
678 	zfsvfs_t *zfsvfs = arg;
679 
680 	zfsvfs->z_vscan = newval;
681 }
682 
683 static void
acl_mode_changed_cb(void * arg,uint64_t newval)684 acl_mode_changed_cb(void *arg, uint64_t newval)
685 {
686 	zfsvfs_t *zfsvfs = arg;
687 
688 	zfsvfs->z_acl_mode = newval;
689 }
690 
691 static void
acl_inherit_changed_cb(void * arg,uint64_t newval)692 acl_inherit_changed_cb(void *arg, uint64_t newval)
693 {
694 	zfsvfs_t *zfsvfs = arg;
695 
696 	zfsvfs->z_acl_inherit = newval;
697 }
698 
699 static int
zfs_register_callbacks(vfs_t * vfsp)700 zfs_register_callbacks(vfs_t *vfsp)
701 {
702 	struct dsl_dataset *ds = NULL;
703 	objset_t *os = NULL;
704 	zfsvfs_t *zfsvfs = NULL;
705 	uint64_t nbmand;
706 	boolean_t readonly = B_FALSE;
707 	boolean_t do_readonly = B_FALSE;
708 	boolean_t setuid = B_FALSE;
709 	boolean_t do_setuid = B_FALSE;
710 	boolean_t exec = B_FALSE;
711 	boolean_t do_exec = B_FALSE;
712 #ifdef illumos
713 	boolean_t devices = B_FALSE;
714 	boolean_t do_devices = B_FALSE;
715 #endif
716 	boolean_t xattr = B_FALSE;
717 	boolean_t do_xattr = B_FALSE;
718 	boolean_t atime = B_FALSE;
719 	boolean_t do_atime = B_FALSE;
720 	int error = 0;
721 
722 	ASSERT(vfsp);
723 	zfsvfs = vfsp->vfs_data;
724 	ASSERT(zfsvfs);
725 	os = zfsvfs->z_os;
726 
727 	/*
728 	 * This function can be called for a snapshot when we update snapshot's
729 	 * mount point, which isn't really supported.
730 	 */
731 	if (dmu_objset_is_snapshot(os))
732 		return (EOPNOTSUPP);
733 
734 	/*
735 	 * The act of registering our callbacks will destroy any mount
736 	 * options we may have.  In order to enable temporary overrides
737 	 * of mount options, we stash away the current values and
738 	 * restore them after we register the callbacks.
739 	 */
740 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
741 	    !spa_writeable(dmu_objset_spa(os))) {
742 		readonly = B_TRUE;
743 		do_readonly = B_TRUE;
744 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
745 		readonly = B_FALSE;
746 		do_readonly = B_TRUE;
747 	}
748 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
749 		setuid = B_FALSE;
750 		do_setuid = B_TRUE;
751 	} else {
752 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
753 			setuid = B_FALSE;
754 			do_setuid = B_TRUE;
755 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
756 			setuid = B_TRUE;
757 			do_setuid = B_TRUE;
758 		}
759 	}
760 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
761 		exec = B_FALSE;
762 		do_exec = B_TRUE;
763 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
764 		exec = B_TRUE;
765 		do_exec = B_TRUE;
766 	}
767 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
768 		xattr = B_FALSE;
769 		do_xattr = B_TRUE;
770 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
771 		xattr = B_TRUE;
772 		do_xattr = B_TRUE;
773 	}
774 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
775 		atime = B_FALSE;
776 		do_atime = B_TRUE;
777 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
778 		atime = B_TRUE;
779 		do_atime = B_TRUE;
780 	}
781 
782 	/*
783 	 * We need to enter pool configuration here, so that we can use
784 	 * dsl_prop_get_int_ds() to handle the special nbmand property below.
785 	 * dsl_prop_get_integer() can not be used, because it has to acquire
786 	 * spa_namespace_lock and we can not do that because we already hold
787 	 * z_teardown_lock.  The problem is that spa_config_sync() is called
788 	 * with spa_namespace_lock held and the function calls ZFS vnode
789 	 * operations to write the cache file and thus z_teardown_lock is
790 	 * acquired after spa_namespace_lock.
791 	 */
792 	ds = dmu_objset_ds(os);
793 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
794 
795 	/*
796 	 * nbmand is a special property.  It can only be changed at
797 	 * mount time.
798 	 *
799 	 * This is weird, but it is documented to only be changeable
800 	 * at mount time.
801 	 */
802 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
803 		nbmand = B_FALSE;
804 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
805 		nbmand = B_TRUE;
806 	} else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
807 		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
808 		return (error);
809 	}
810 
811 	/*
812 	 * Register property callbacks.
813 	 *
814 	 * It would probably be fine to just check for i/o error from
815 	 * the first prop_register(), but I guess I like to go
816 	 * overboard...
817 	 */
818 	error = dsl_prop_register(ds,
819 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
820 	error = error ? error : dsl_prop_register(ds,
821 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
822 	error = error ? error : dsl_prop_register(ds,
823 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
824 	error = error ? error : dsl_prop_register(ds,
825 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
826 #ifdef illumos
827 	error = error ? error : dsl_prop_register(ds,
828 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
829 #endif
830 	error = error ? error : dsl_prop_register(ds,
831 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
832 	error = error ? error : dsl_prop_register(ds,
833 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
834 	error = error ? error : dsl_prop_register(ds,
835 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
836 	error = error ? error : dsl_prop_register(ds,
837 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
838 	error = error ? error : dsl_prop_register(ds,
839 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
840 	    zfsvfs);
841 	error = error ? error : dsl_prop_register(ds,
842 	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
843 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
844 	if (error)
845 		goto unregister;
846 
847 	/*
848 	 * Invoke our callbacks to restore temporary mount options.
849 	 */
850 	if (do_readonly)
851 		readonly_changed_cb(zfsvfs, readonly);
852 	if (do_setuid)
853 		setuid_changed_cb(zfsvfs, setuid);
854 	if (do_exec)
855 		exec_changed_cb(zfsvfs, exec);
856 	if (do_xattr)
857 		xattr_changed_cb(zfsvfs, xattr);
858 	if (do_atime)
859 		atime_changed_cb(zfsvfs, atime);
860 
861 	nbmand_changed_cb(zfsvfs, nbmand);
862 
863 	return (0);
864 
865 unregister:
866 	dsl_prop_unregister_all(ds, zfsvfs);
867 	return (error);
868 }
869 
870 static int
zfs_space_delta_cb(dmu_object_type_t bonustype,void * data,uint64_t * userp,uint64_t * groupp)871 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
872     uint64_t *userp, uint64_t *groupp)
873 {
874 	/*
875 	 * Is it a valid type of object to track?
876 	 */
877 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
878 		return (SET_ERROR(ENOENT));
879 
880 	/*
881 	 * If we have a NULL data pointer
882 	 * then assume the id's aren't changing and
883 	 * return EEXIST to the dmu to let it know to
884 	 * use the same ids
885 	 */
886 	if (data == NULL)
887 		return (SET_ERROR(EEXIST));
888 
889 	if (bonustype == DMU_OT_ZNODE) {
890 		znode_phys_t *znp = data;
891 		*userp = znp->zp_uid;
892 		*groupp = znp->zp_gid;
893 	} else {
894 		int hdrsize;
895 		sa_hdr_phys_t *sap = data;
896 		sa_hdr_phys_t sa = *sap;
897 		boolean_t swap = B_FALSE;
898 
899 		ASSERT(bonustype == DMU_OT_SA);
900 
901 		if (sa.sa_magic == 0) {
902 			/*
903 			 * This should only happen for newly created
904 			 * files that haven't had the znode data filled
905 			 * in yet.
906 			 */
907 			*userp = 0;
908 			*groupp = 0;
909 			return (0);
910 		}
911 		if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
912 			sa.sa_magic = SA_MAGIC;
913 			sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
914 			swap = B_TRUE;
915 		} else {
916 			VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
917 		}
918 
919 		hdrsize = sa_hdrsize(&sa);
920 		VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
921 		*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
922 		    SA_UID_OFFSET));
923 		*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
924 		    SA_GID_OFFSET));
925 		if (swap) {
926 			*userp = BSWAP_64(*userp);
927 			*groupp = BSWAP_64(*groupp);
928 		}
929 	}
930 	return (0);
931 }
932 
933 static void
fuidstr_to_sid(zfsvfs_t * zfsvfs,const char * fuidstr,char * domainbuf,int buflen,uid_t * ridp)934 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
935     char *domainbuf, int buflen, uid_t *ridp)
936 {
937 	uint64_t fuid;
938 	const char *domain;
939 
940 	fuid = strtonum(fuidstr, NULL);
941 
942 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
943 	if (domain)
944 		(void) strlcpy(domainbuf, domain, buflen);
945 	else
946 		domainbuf[0] = '\0';
947 	*ridp = FUID_RID(fuid);
948 }
949 
950 static uint64_t
zfs_userquota_prop_to_obj(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type)951 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
952 {
953 	switch (type) {
954 	case ZFS_PROP_USERUSED:
955 		return (DMU_USERUSED_OBJECT);
956 	case ZFS_PROP_GROUPUSED:
957 		return (DMU_GROUPUSED_OBJECT);
958 	case ZFS_PROP_USERQUOTA:
959 		return (zfsvfs->z_userquota_obj);
960 	case ZFS_PROP_GROUPQUOTA:
961 		return (zfsvfs->z_groupquota_obj);
962 	}
963 	return (0);
964 }
965 
966 int
zfs_userspace_many(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,uint64_t * cookiep,void * vbuf,uint64_t * bufsizep)967 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
968     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
969 {
970 	int error;
971 	zap_cursor_t zc;
972 	zap_attribute_t za;
973 	zfs_useracct_t *buf = vbuf;
974 	uint64_t obj;
975 
976 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
977 		return (SET_ERROR(ENOTSUP));
978 
979 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
980 	if (obj == 0) {
981 		*bufsizep = 0;
982 		return (0);
983 	}
984 
985 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
986 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
987 	    zap_cursor_advance(&zc)) {
988 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
989 		    *bufsizep)
990 			break;
991 
992 		fuidstr_to_sid(zfsvfs, za.za_name,
993 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
994 
995 		buf->zu_space = za.za_first_integer;
996 		buf++;
997 	}
998 	if (error == ENOENT)
999 		error = 0;
1000 
1001 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
1002 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
1003 	*cookiep = zap_cursor_serialize(&zc);
1004 	zap_cursor_fini(&zc);
1005 	return (error);
1006 }
1007 
1008 /*
1009  * buf must be big enough (eg, 32 bytes)
1010  */
1011 static int
id_to_fuidstr(zfsvfs_t * zfsvfs,const char * domain,uid_t rid,char * buf,boolean_t addok)1012 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
1013     char *buf, boolean_t addok)
1014 {
1015 	uint64_t fuid;
1016 	int domainid = 0;
1017 
1018 	if (domain && domain[0]) {
1019 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
1020 		if (domainid == -1)
1021 			return (SET_ERROR(ENOENT));
1022 	}
1023 	fuid = FUID_ENCODE(domainid, rid);
1024 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
1025 	return (0);
1026 }
1027 
1028 int
zfs_userspace_one(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,const char * domain,uint64_t rid,uint64_t * valp)1029 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
1030     const char *domain, uint64_t rid, uint64_t *valp)
1031 {
1032 	char buf[32];
1033 	int err;
1034 	uint64_t obj;
1035 
1036 	*valp = 0;
1037 
1038 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
1039 		return (SET_ERROR(ENOTSUP));
1040 
1041 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
1042 	if (obj == 0)
1043 		return (0);
1044 
1045 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
1046 	if (err)
1047 		return (err);
1048 
1049 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
1050 	if (err == ENOENT)
1051 		err = 0;
1052 	return (err);
1053 }
1054 
1055 int
zfs_set_userquota(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,const char * domain,uint64_t rid,uint64_t quota)1056 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
1057     const char *domain, uint64_t rid, uint64_t quota)
1058 {
1059 	char buf[32];
1060 	int err;
1061 	dmu_tx_t *tx;
1062 	uint64_t *objp;
1063 	boolean_t fuid_dirtied;
1064 
1065 	if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
1066 		return (SET_ERROR(EINVAL));
1067 
1068 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
1069 		return (SET_ERROR(ENOTSUP));
1070 
1071 	objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
1072 	    &zfsvfs->z_groupquota_obj;
1073 
1074 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
1075 	if (err)
1076 		return (err);
1077 	fuid_dirtied = zfsvfs->z_fuid_dirty;
1078 
1079 	tx = dmu_tx_create(zfsvfs->z_os);
1080 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
1081 	if (*objp == 0) {
1082 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1083 		    zfs_userquota_prop_prefixes[type]);
1084 	}
1085 	if (fuid_dirtied)
1086 		zfs_fuid_txhold(zfsvfs, tx);
1087 	err = dmu_tx_assign(tx, TXG_WAIT);
1088 	if (err) {
1089 		dmu_tx_abort(tx);
1090 		return (err);
1091 	}
1092 
1093 	mutex_enter(&zfsvfs->z_lock);
1094 	if (*objp == 0) {
1095 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
1096 		    DMU_OT_NONE, 0, tx);
1097 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
1098 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
1099 	}
1100 	mutex_exit(&zfsvfs->z_lock);
1101 
1102 	if (quota == 0) {
1103 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
1104 		if (err == ENOENT)
1105 			err = 0;
1106 	} else {
1107 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
1108 	}
1109 	ASSERT(err == 0);
1110 	if (fuid_dirtied)
1111 		zfs_fuid_sync(zfsvfs, tx);
1112 	dmu_tx_commit(tx);
1113 	return (err);
1114 }
1115 
1116 boolean_t
zfs_fuid_overquota(zfsvfs_t * zfsvfs,boolean_t isgroup,uint64_t fuid)1117 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
1118 {
1119 	char buf[32];
1120 	uint64_t used, quota, usedobj, quotaobj;
1121 	int err;
1122 
1123 	usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
1124 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
1125 
1126 	if (quotaobj == 0 || zfsvfs->z_replay)
1127 		return (B_FALSE);
1128 
1129 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
1130 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
1131 	if (err != 0)
1132 		return (B_FALSE);
1133 
1134 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
1135 	if (err != 0)
1136 		return (B_FALSE);
1137 	return (used >= quota);
1138 }
1139 
1140 boolean_t
zfs_owner_overquota(zfsvfs_t * zfsvfs,znode_t * zp,boolean_t isgroup)1141 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
1142 {
1143 	uint64_t fuid;
1144 	uint64_t quotaobj;
1145 
1146 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
1147 
1148 	fuid = isgroup ? zp->z_gid : zp->z_uid;
1149 
1150 	if (quotaobj == 0 || zfsvfs->z_replay)
1151 		return (B_FALSE);
1152 
1153 	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
1154 }
1155 
1156 /*
1157  * Associate this zfsvfs with the given objset, which must be owned.
1158  * This will cache a bunch of on-disk state from the objset in the
1159  * zfsvfs.
1160  */
1161 static int
zfsvfs_init(zfsvfs_t * zfsvfs,objset_t * os)1162 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1163 {
1164 	int error;
1165 	uint64_t val;
1166 
1167 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1168 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1169 	zfsvfs->z_os = os;
1170 
1171 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1172 	if (error != 0)
1173 		return (error);
1174 	if (zfsvfs->z_version >
1175 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1176 		(void) printf("Can't mount a version %lld file system "
1177 		    "on a version %lld pool\n. Pool must be upgraded to mount "
1178 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
1179 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
1180 		return (SET_ERROR(ENOTSUP));
1181 	}
1182 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1183 	if (error != 0)
1184 		return (error);
1185 	zfsvfs->z_norm = (int)val;
1186 
1187 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1188 	if (error != 0)
1189 		return (error);
1190 	zfsvfs->z_utf8 = (val != 0);
1191 
1192 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1193 	if (error != 0)
1194 		return (error);
1195 	zfsvfs->z_case = (uint_t)val;
1196 
1197 	/*
1198 	 * Fold case on file systems that are always or sometimes case
1199 	 * insensitive.
1200 	 */
1201 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1202 	    zfsvfs->z_case == ZFS_CASE_MIXED)
1203 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1204 
1205 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1206 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1207 
1208 	uint64_t sa_obj = 0;
1209 	if (zfsvfs->z_use_sa) {
1210 		/* should either have both of these objects or none */
1211 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1212 		    &sa_obj);
1213 		if (error != 0)
1214 			return (error);
1215 	}
1216 
1217 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1218 	    &zfsvfs->z_attr_table);
1219 	if (error != 0)
1220 		return (error);
1221 
1222 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
1223 		sa_register_update_callback(os, zfs_sa_upgrade);
1224 
1225 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1226 	    &zfsvfs->z_root);
1227 	if (error != 0)
1228 		return (error);
1229 	ASSERT(zfsvfs->z_root != 0);
1230 
1231 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1232 	    &zfsvfs->z_unlinkedobj);
1233 	if (error != 0)
1234 		return (error);
1235 
1236 	error = zap_lookup(os, MASTER_NODE_OBJ,
1237 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1238 	    8, 1, &zfsvfs->z_userquota_obj);
1239 	if (error == ENOENT)
1240 		zfsvfs->z_userquota_obj = 0;
1241 	else if (error != 0)
1242 		return (error);
1243 
1244 	error = zap_lookup(os, MASTER_NODE_OBJ,
1245 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1246 	    8, 1, &zfsvfs->z_groupquota_obj);
1247 	if (error == ENOENT)
1248 		zfsvfs->z_groupquota_obj = 0;
1249 	else if (error != 0)
1250 		return (error);
1251 
1252 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1253 	    &zfsvfs->z_fuid_obj);
1254 	if (error == ENOENT)
1255 		zfsvfs->z_fuid_obj = 0;
1256 	else if (error != 0)
1257 		return (error);
1258 
1259 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1260 	    &zfsvfs->z_shares_dir);
1261 	if (error == ENOENT)
1262 		zfsvfs->z_shares_dir = 0;
1263 	else if (error != 0)
1264 		return (error);
1265 
1266 	/*
1267 	 * Only use the name cache if we are looking for a
1268 	 * name on a file system that does not require normalization
1269 	 * or case folding.  We can also look there if we happen to be
1270 	 * on a non-normalizing, mixed sensitivity file system IF we
1271 	 * are looking for the exact name (which is always the case on
1272 	 * FreeBSD).
1273 	 */
1274 	zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
1275 	    ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
1276 	    !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
1277 
1278 	return (0);
1279 }
1280 
1281 int
zfsvfs_create(const char * osname,zfsvfs_t ** zfvp)1282 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
1283 {
1284 	objset_t *os;
1285 	zfsvfs_t *zfsvfs;
1286 	int error;
1287 
1288 	/*
1289 	 * XXX: Fix struct statfs so this isn't necessary!
1290 	 *
1291 	 * The 'osname' is used as the filesystem's special node, which means
1292 	 * it must fit in statfs.f_mntfromname, or else it can't be
1293 	 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
1294 	 * 'zfs unmount' to think it's not mounted when it is.
1295 	 */
1296 	if (strlen(osname) >= MNAMELEN)
1297 		return (SET_ERROR(ENAMETOOLONG));
1298 
1299 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1300 
1301 	/*
1302 	 * We claim to always be readonly so we can open snapshots;
1303 	 * other ZPL code will prevent us from writing to snapshots.
1304 	 */
1305 	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1306 	if (error) {
1307 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1308 		return (error);
1309 	}
1310 
1311 	zfsvfs->z_vfs = NULL;
1312 	zfsvfs->z_parent = zfsvfs;
1313 
1314 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1315 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1316 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1317 	    offsetof(znode_t, z_link_node));
1318 #ifdef DIAGNOSTIC
1319 	rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1320 #else
1321 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1322 #endif
1323 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1324 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1325 	for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1326 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1327 
1328 	error = zfsvfs_init(zfsvfs, os);
1329 	if (error != 0) {
1330 		dmu_objset_disown(os, zfsvfs);
1331 		*zfvp = NULL;
1332 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1333 		return (error);
1334 	}
1335 
1336 	*zfvp = zfsvfs;
1337 	return (0);
1338 }
1339 
1340 static int
zfsvfs_setup(zfsvfs_t * zfsvfs,boolean_t mounting)1341 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1342 {
1343 	int error;
1344 
1345 	error = zfs_register_callbacks(zfsvfs->z_vfs);
1346 	if (error)
1347 		return (error);
1348 
1349 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1350 
1351 	/*
1352 	 * If we are not mounting (ie: online recv), then we don't
1353 	 * have to worry about replaying the log as we blocked all
1354 	 * operations out since we closed the ZIL.
1355 	 */
1356 	if (mounting) {
1357 		boolean_t readonly;
1358 
1359 		/*
1360 		 * During replay we remove the read only flag to
1361 		 * allow replays to succeed.
1362 		 */
1363 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1364 		if (readonly != 0)
1365 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1366 		else
1367 			zfs_unlinked_drain(zfsvfs);
1368 
1369 		/*
1370 		 * Parse and replay the intent log.
1371 		 *
1372 		 * Because of ziltest, this must be done after
1373 		 * zfs_unlinked_drain().  (Further note: ziltest
1374 		 * doesn't use readonly mounts, where
1375 		 * zfs_unlinked_drain() isn't called.)  This is because
1376 		 * ziltest causes spa_sync() to think it's committed,
1377 		 * but actually it is not, so the intent log contains
1378 		 * many txg's worth of changes.
1379 		 *
1380 		 * In particular, if object N is in the unlinked set in
1381 		 * the last txg to actually sync, then it could be
1382 		 * actually freed in a later txg and then reallocated
1383 		 * in a yet later txg.  This would write a "create
1384 		 * object N" record to the intent log.  Normally, this
1385 		 * would be fine because the spa_sync() would have
1386 		 * written out the fact that object N is free, before
1387 		 * we could write the "create object N" intent log
1388 		 * record.
1389 		 *
1390 		 * But when we are in ziltest mode, we advance the "open
1391 		 * txg" without actually spa_sync()-ing the changes to
1392 		 * disk.  So we would see that object N is still
1393 		 * allocated and in the unlinked set, and there is an
1394 		 * intent log record saying to allocate it.
1395 		 */
1396 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1397 			if (zil_replay_disable) {
1398 				zil_destroy(zfsvfs->z_log, B_FALSE);
1399 			} else {
1400 				zfsvfs->z_replay = B_TRUE;
1401 				zil_replay(zfsvfs->z_os, zfsvfs,
1402 				    zfs_replay_vector);
1403 				zfsvfs->z_replay = B_FALSE;
1404 			}
1405 		}
1406 		zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1407 	}
1408 
1409 	/*
1410 	 * Set the objset user_ptr to track its zfsvfs.
1411 	 */
1412 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1413 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1414 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1415 
1416 	return (0);
1417 }
1418 
1419 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1420 
1421 void
zfsvfs_free(zfsvfs_t * zfsvfs)1422 zfsvfs_free(zfsvfs_t *zfsvfs)
1423 {
1424 	int i;
1425 
1426 	/*
1427 	 * This is a barrier to prevent the filesystem from going away in
1428 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1429 	 * not unmounted. We consider the filesystem valid before the barrier
1430 	 * and invalid after the barrier.
1431 	 */
1432 	rw_enter(&zfsvfs_lock, RW_READER);
1433 	rw_exit(&zfsvfs_lock);
1434 
1435 	zfs_fuid_destroy(zfsvfs);
1436 
1437 	mutex_destroy(&zfsvfs->z_znodes_lock);
1438 	mutex_destroy(&zfsvfs->z_lock);
1439 	list_destroy(&zfsvfs->z_all_znodes);
1440 	rrm_destroy(&zfsvfs->z_teardown_lock);
1441 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1442 	rw_destroy(&zfsvfs->z_fuid_lock);
1443 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1444 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1445 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1446 }
1447 
1448 static void
zfs_set_fuid_feature(zfsvfs_t * zfsvfs)1449 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1450 {
1451 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1452 	if (zfsvfs->z_vfs) {
1453 		if (zfsvfs->z_use_fuids) {
1454 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1455 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1456 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1457 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1458 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1459 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1460 		} else {
1461 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1462 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1463 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1464 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1465 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1466 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1467 		}
1468 	}
1469 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1470 }
1471 
1472 #ifdef __NetBSD__
1473 int
1474 #else
1475 static int
1476 #endif
zfs_domount(vfs_t * vfsp,char * osname)1477 zfs_domount(vfs_t *vfsp, char *osname)
1478 {
1479 	uint64_t recordsize, fsid_guid;
1480 	int error = 0;
1481 	zfsvfs_t *zfsvfs;
1482 	vnode_t *vp;
1483 
1484 	ASSERT(vfsp);
1485 	ASSERT(osname);
1486 
1487 	error = zfsvfs_create(osname, &zfsvfs);
1488 	if (error)
1489 		return (error);
1490 	zfsvfs->z_vfs = vfsp;
1491 
1492 #ifdef illumos
1493 	/* Initialize the generic filesystem structure. */
1494 	vfsp->vfs_bcount = 0;
1495 	vfsp->vfs_data = NULL;
1496 
1497 	if (zfs_create_unique_device(&mount_dev) == -1) {
1498 		error = SET_ERROR(ENODEV);
1499 		goto out;
1500 	}
1501 	ASSERT(vfs_devismounted(mount_dev) == 0);
1502 #endif
1503 
1504 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1505 	    NULL))
1506 		goto out;
1507 	zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1508 	zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1509 
1510 	vfsp->vfs_data = zfsvfs;
1511 #ifdef __FreeBSD_kernel__
1512 	vfsp->mnt_flag |= MNT_LOCAL;
1513 	vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1514 	vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1515 	vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1516 	vfsp->mnt_kern_flag |= MNTK_NO_IOPF;	/* vn_io_fault can be used */
1517 #endif
1518 #ifdef __NetBSD__
1519 	vfsp->mnt_flag |= MNT_LOCAL;
1520 	vfsp->mnt_iflag |= IMNT_MPSAFE | IMNT_NCLOOKUP;
1521 #endif
1522 
1523 	/*
1524 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1525 	 * separates our fsid from any other filesystem types, and a
1526 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1527 	 * all objsets open on this system, provided by unique_create().
1528 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1529 	 * because that's where other Solaris filesystems put it.
1530 	 */
1531 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1532 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1533 #ifdef __FreeBSD_kernel__
1534 	vfsp->vfs_fsid.val[0] = fsid_guid;
1535 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1536 	    vfsp->mnt_vfc->vfc_typenum & 0xFF;
1537 #endif
1538 #ifdef __NetBSD__
1539 	vfsp->mnt_stat.f_fsidx.__fsid_val[0] = fsid_guid;
1540 	vfsp->mnt_stat.f_fsidx.__fsid_val[1] = ((fsid_guid>>32) << 8) |
1541 	    makefstype(vfsp->mnt_op->vfs_name) & 0xFF;
1542 	vfsp->mnt_stat.f_fsid = fsid_guid;
1543 #endif
1544 
1545 	/*
1546 	 * Set features for file system.
1547 	 */
1548 	zfs_set_fuid_feature(zfsvfs);
1549 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1550 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1551 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1552 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1553 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1554 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1555 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1556 	}
1557 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1558 
1559 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1560 		uint64_t pval;
1561 
1562 		atime_changed_cb(zfsvfs, B_FALSE);
1563 		readonly_changed_cb(zfsvfs, B_TRUE);
1564 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1565 			goto out;
1566 		xattr_changed_cb(zfsvfs, pval);
1567 		zfsvfs->z_issnap = B_TRUE;
1568 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1569 
1570 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1571 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1572 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1573 	} else {
1574 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1575 	}
1576 
1577 #ifdef __FreeBSD_kernel__
1578 	vfs_mountedfrom(vfsp, osname);
1579 #endif
1580 #ifdef __NetBSD__
1581 	set_statvfs_info("on-name", UIO_SYSSPACE, osname, UIO_SYSSPACE, "zfs", vfsp, curlwp);
1582 #endif
1583 
1584 	if (!zfsvfs->z_issnap)
1585 		zfsctl_create(zfsvfs);
1586 out:
1587 	if (error) {
1588 		dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1589 		zfsvfs_free(zfsvfs);
1590 	} else {
1591 		atomic_inc_32(&zfs_active_fs_count);
1592 	}
1593 
1594 	return (error);
1595 }
1596 
1597 void
zfs_unregister_callbacks(zfsvfs_t * zfsvfs)1598 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1599 {
1600 	objset_t *os = zfsvfs->z_os;
1601 
1602 	if (!dmu_objset_is_snapshot(os))
1603 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1604 }
1605 
1606 #ifdef SECLABEL
1607 /*
1608  * Convert a decimal digit string to a uint64_t integer.
1609  */
1610 static int
str_to_uint64(char * str,uint64_t * objnum)1611 str_to_uint64(char *str, uint64_t *objnum)
1612 {
1613 	uint64_t num = 0;
1614 
1615 	while (*str) {
1616 		if (*str < '0' || *str > '9')
1617 			return (SET_ERROR(EINVAL));
1618 
1619 		num = num*10 + *str++ - '0';
1620 	}
1621 
1622 	*objnum = num;
1623 	return (0);
1624 }
1625 
1626 /*
1627  * The boot path passed from the boot loader is in the form of
1628  * "rootpool-name/root-filesystem-object-number'. Convert this
1629  * string to a dataset name: "rootpool-name/root-filesystem-name".
1630  */
1631 static int
zfs_parse_bootfs(char * bpath,char * outpath)1632 zfs_parse_bootfs(char *bpath, char *outpath)
1633 {
1634 	char *slashp;
1635 	uint64_t objnum;
1636 	int error;
1637 
1638 	if (*bpath == 0 || *bpath == '/')
1639 		return (SET_ERROR(EINVAL));
1640 
1641 	(void) strcpy(outpath, bpath);
1642 
1643 	slashp = strchr(bpath, '/');
1644 
1645 	/* if no '/', just return the pool name */
1646 	if (slashp == NULL) {
1647 		return (0);
1648 	}
1649 
1650 	/* if not a number, just return the root dataset name */
1651 	if (str_to_uint64(slashp+1, &objnum)) {
1652 		return (0);
1653 	}
1654 
1655 	*slashp = '\0';
1656 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1657 	*slashp = '/';
1658 
1659 	return (error);
1660 }
1661 
1662 /*
1663  * Check that the hex label string is appropriate for the dataset being
1664  * mounted into the global_zone proper.
1665  *
1666  * Return an error if the hex label string is not default or
1667  * admin_low/admin_high.  For admin_low labels, the corresponding
1668  * dataset must be readonly.
1669  */
1670 int
zfs_check_global_label(const char * dsname,const char * hexsl)1671 zfs_check_global_label(const char *dsname, const char *hexsl)
1672 {
1673 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1674 		return (0);
1675 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1676 		return (0);
1677 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1678 		/* must be readonly */
1679 		uint64_t rdonly;
1680 
1681 		if (dsl_prop_get_integer(dsname,
1682 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1683 			return (SET_ERROR(EACCES));
1684 		return (rdonly ? 0 : EACCES);
1685 	}
1686 	return (SET_ERROR(EACCES));
1687 }
1688 
1689 /*
1690  * Determine whether the mount is allowed according to MAC check.
1691  * by comparing (where appropriate) label of the dataset against
1692  * the label of the zone being mounted into.  If the dataset has
1693  * no label, create one.
1694  *
1695  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1696  */
1697 static int
zfs_mount_label_policy(vfs_t * vfsp,char * osname)1698 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1699 {
1700 	int		error, retv;
1701 	zone_t		*mntzone = NULL;
1702 	ts_label_t	*mnt_tsl;
1703 	bslabel_t	*mnt_sl;
1704 	bslabel_t	ds_sl;
1705 	char		ds_hexsl[MAXNAMELEN];
1706 
1707 	retv = EACCES;				/* assume the worst */
1708 
1709 	/*
1710 	 * Start by getting the dataset label if it exists.
1711 	 */
1712 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1713 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1714 	if (error)
1715 		return (SET_ERROR(EACCES));
1716 
1717 	/*
1718 	 * If labeling is NOT enabled, then disallow the mount of datasets
1719 	 * which have a non-default label already.  No other label checks
1720 	 * are needed.
1721 	 */
1722 	if (!is_system_labeled()) {
1723 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1724 			return (0);
1725 		return (SET_ERROR(EACCES));
1726 	}
1727 
1728 	/*
1729 	 * Get the label of the mountpoint.  If mounting into the global
1730 	 * zone (i.e. mountpoint is not within an active zone and the
1731 	 * zoned property is off), the label must be default or
1732 	 * admin_low/admin_high only; no other checks are needed.
1733 	 */
1734 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1735 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1736 		uint64_t zoned;
1737 
1738 		zone_rele(mntzone);
1739 
1740 		if (dsl_prop_get_integer(osname,
1741 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1742 			return (SET_ERROR(EACCES));
1743 		if (!zoned)
1744 			return (zfs_check_global_label(osname, ds_hexsl));
1745 		else
1746 			/*
1747 			 * This is the case of a zone dataset being mounted
1748 			 * initially, before the zone has been fully created;
1749 			 * allow this mount into global zone.
1750 			 */
1751 			return (0);
1752 	}
1753 
1754 	mnt_tsl = mntzone->zone_slabel;
1755 	ASSERT(mnt_tsl != NULL);
1756 	label_hold(mnt_tsl);
1757 	mnt_sl = label2bslabel(mnt_tsl);
1758 
1759 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1760 		/*
1761 		 * The dataset doesn't have a real label, so fabricate one.
1762 		 */
1763 		char *str = NULL;
1764 
1765 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1766 		    dsl_prop_set_string(osname,
1767 		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1768 		    ZPROP_SRC_LOCAL, str) == 0)
1769 			retv = 0;
1770 		if (str != NULL)
1771 			kmem_free(str, strlen(str) + 1);
1772 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1773 		/*
1774 		 * Now compare labels to complete the MAC check.  If the
1775 		 * labels are equal then allow access.  If the mountpoint
1776 		 * label dominates the dataset label, allow readonly access.
1777 		 * Otherwise, access is denied.
1778 		 */
1779 		if (blequal(mnt_sl, &ds_sl))
1780 			retv = 0;
1781 		else if (bldominates(mnt_sl, &ds_sl)) {
1782 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1783 			retv = 0;
1784 		}
1785 	}
1786 
1787 	label_rele(mnt_tsl);
1788 	zone_rele(mntzone);
1789 	return (retv);
1790 }
1791 #endif	/* SECLABEL */
1792 
1793 #ifdef OPENSOLARIS_MOUNTROOT
1794 static int
zfs_mountroot(vfs_t * vfsp,enum whymountroot why)1795 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1796 {
1797 	int error = 0;
1798 	static int zfsrootdone = 0;
1799 	zfsvfs_t *zfsvfs = NULL;
1800 	znode_t *zp = NULL;
1801 	vnode_t *vp = NULL;
1802 	char *zfs_bootfs;
1803 	char *zfs_devid;
1804 
1805 	ASSERT(vfsp);
1806 
1807 	/*
1808 	 * The filesystem that we mount as root is defined in the
1809 	 * boot property "zfs-bootfs" with a format of
1810 	 * "poolname/root-dataset-objnum".
1811 	 */
1812 	if (why == ROOT_INIT) {
1813 		if (zfsrootdone++)
1814 			return (SET_ERROR(EBUSY));
1815 		/*
1816 		 * the process of doing a spa_load will require the
1817 		 * clock to be set before we could (for example) do
1818 		 * something better by looking at the timestamp on
1819 		 * an uberblock, so just set it to -1.
1820 		 */
1821 		clkset(-1);
1822 
1823 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1824 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1825 			    "bootfs name");
1826 			return (SET_ERROR(EINVAL));
1827 		}
1828 		zfs_devid = spa_get_bootprop("diskdevid");
1829 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1830 		if (zfs_devid)
1831 			spa_free_bootprop(zfs_devid);
1832 		if (error) {
1833 			spa_free_bootprop(zfs_bootfs);
1834 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1835 			    error);
1836 			return (error);
1837 		}
1838 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1839 			spa_free_bootprop(zfs_bootfs);
1840 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1841 			    error);
1842 			return (error);
1843 		}
1844 
1845 		spa_free_bootprop(zfs_bootfs);
1846 
1847 		if (error = vfs_lock(vfsp))
1848 			return (error);
1849 
1850 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1851 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1852 			goto out;
1853 		}
1854 
1855 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1856 		ASSERT(zfsvfs);
1857 		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1858 			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1859 			goto out;
1860 		}
1861 
1862 		vp = ZTOV(zp);
1863 		mutex_enter(&vp->v_lock);
1864 		vp->v_flag |= VROOT;
1865 		mutex_exit(&vp->v_lock);
1866 		rootvp = vp;
1867 
1868 		/*
1869 		 * Leave rootvp held.  The root file system is never unmounted.
1870 		 */
1871 
1872 		vfs_add((struct vnode *)0, vfsp,
1873 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1874 out:
1875 		vfs_unlock(vfsp);
1876 		return (error);
1877 	} else if (why == ROOT_REMOUNT) {
1878 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1879 		vfsp->vfs_flag |= VFS_REMOUNT;
1880 
1881 		/* refresh mount options */
1882 		zfs_unregister_callbacks(vfsp->vfs_data);
1883 		return (zfs_register_callbacks(vfsp));
1884 
1885 	} else if (why == ROOT_UNMOUNT) {
1886 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1887 		(void) zfs_sync(vfsp, 0, 0);
1888 		return (0);
1889 	}
1890 
1891 	/*
1892 	 * if "why" is equal to anything else other than ROOT_INIT,
1893 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1894 	 */
1895 	return (SET_ERROR(ENOTSUP));
1896 }
1897 #endif	/* OPENSOLARIS_MOUNTROOT */
1898 
1899 static int
getpoolname(const char * osname,char * poolname)1900 getpoolname(const char *osname, char *poolname)
1901 {
1902 	char *p;
1903 
1904 	p = strchr(osname, '/');
1905 	if (p == NULL) {
1906 		if (strlen(osname) >= MAXNAMELEN)
1907 			return (ENAMETOOLONG);
1908 		(void) strcpy(poolname, osname);
1909 	} else {
1910 		if (p - osname >= MAXNAMELEN)
1911 			return (ENAMETOOLONG);
1912 		(void) strncpy(poolname, osname, p - osname);
1913 		poolname[p - osname] = '\0';
1914 	}
1915 	return (0);
1916 }
1917 
1918 /*ARGSUSED*/
1919 #ifdef illumos
1920 static int
zfs_mount(vfs_t * vfsp,vnode_t * mvp,struct mounta * uap,cred_t * cr)1921 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1922 #endif
1923 #ifdef __FreeBSD_kernel__
1924 static int
1925 zfs_mount(vfs_t *vfsp)
1926 #endif
1927 #ifdef __NetBSD__
1928 static int
1929 zfs_mount(vfs_t *vfsp, const char *path, void *data, size_t *data_len)
1930 #endif
1931 {
1932 	vnode_t		*mvp = vfsp->mnt_vnodecovered;
1933 	char		*osname;
1934 	int		error = 0;
1935 	int		canwrite;
1936 
1937 #ifdef illumos
1938 	if (mvp->v_type != VDIR)
1939 		return (SET_ERROR(ENOTDIR));
1940 
1941 	mutex_enter(&mvp->v_lock);
1942 	if ((uap->flags & MS_REMOUNT) == 0 &&
1943 	    (uap->flags & MS_OVERLAY) == 0 &&
1944 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1945 		mutex_exit(&mvp->v_lock);
1946 		return (SET_ERROR(EBUSY));
1947 	}
1948 	mutex_exit(&mvp->v_lock);
1949 
1950 	/*
1951 	 * ZFS does not support passing unparsed data in via MS_DATA.
1952 	 * Users should use the MS_OPTIONSTR interface; this means
1953 	 * that all option parsing is already done and the options struct
1954 	 * can be interrogated.
1955 	 */
1956 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1957 		return (SET_ERROR(EINVAL));
1958 #endif /* illumos */
1959 
1960 #ifdef __FreeBSD_kernel__
1961 	kthread_t	*td = curthread;
1962 	cred_t		*cr = td->td_ucred;
1963 
1964 	if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1965 		return (SET_ERROR(EPERM));
1966 
1967 	if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1968 		return (SET_ERROR(EINVAL));
1969 
1970 	/*
1971 	 * If full-owner-access is enabled and delegated administration is
1972 	 * turned on, we must set nosuid.
1973 	 */
1974 	if (zfs_super_owner &&
1975 	    dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1976 		secpolicy_fs_mount_clearopts(cr, vfsp);
1977 	}
1978 
1979 #endif /* __FreeBSD_kernel__ */
1980 
1981 #ifdef __NetBSD__
1982 	cred_t		*cr = CRED();
1983 	struct mounta	*uap = data;
1984 
1985 	if (uap == NULL)
1986 		return (SET_ERROR(EINVAL));
1987 
1988 	if (*data_len < sizeof *uap)
1989 		return (SET_ERROR(EINVAL));
1990 
1991 	if (mvp->v_type != VDIR)
1992 		return (SET_ERROR(ENOTDIR));
1993 
1994 	mutex_enter(mvp->v_interlock);
1995 	if ((uap->flags & MS_REMOUNT) == 0 &&
1996 	    (uap->flags & MS_OVERLAY) == 0 &&
1997 	    (vrefcnt(mvp) != 1 || (mvp->v_flag & VROOT))) {
1998 		mutex_exit(mvp->v_interlock);
1999 		return (SET_ERROR(EBUSY));
2000 	}
2001 	mutex_exit(mvp->v_interlock);
2002 
2003 	osname = PNBUF_GET();
2004 	strlcpy(osname, uap->fspec, strlen(uap->fspec) + 1);
2005 #endif /* __NetBSD__ */
2006 
2007 	/*
2008 	 * Check for mount privilege?
2009 	 *
2010 	 * If we don't have privilege then see if
2011 	 * we have local permission to allow it
2012 	 */
2013 	error = secpolicy_fs_mount(cr, mvp, vfsp);
2014 	if (error) {
2015 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
2016 			goto out;
2017 
2018 		if (!(vfsp->vfs_flag & MS_REMOUNT)) {
2019 			vattr_t		vattr;
2020 
2021 			/*
2022 			 * Make sure user is the owner of the mount point
2023 			 * or has sufficient privileges.
2024 			 */
2025 
2026 			vattr.va_mask = AT_UID;
2027 
2028 #ifdef __FreeBSD_kernel__
2029 			vn_lock(mvp, LK_SHARED | LK_RETRY);
2030 			if (VOP_GETATTR(mvp, &vattr, cr)) {
2031 				VOP_UNLOCK(mvp, 0);
2032 				goto out;
2033 			}
2034 
2035 			if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
2036 			    VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
2037 				VOP_UNLOCK(mvp, 0);
2038 				goto out;
2039 			}
2040 			VOP_UNLOCK(mvp, 0);
2041 #endif
2042 #ifdef __NetBSD__
2043 			vn_lock(mvp, LK_SHARED | LK_RETRY);
2044 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
2045 				VOP_UNLOCK(mvp, 0);
2046 				goto out;
2047 			}
2048 
2049 			if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
2050 			    VOP_ACCESS(mvp, VWRITE, cr) != 0) {
2051 				VOP_UNLOCK(mvp, 0);
2052 				goto out;
2053 			}
2054 			VOP_UNLOCK(mvp, 0);
2055 #endif
2056 		}
2057 
2058 		secpolicy_fs_mount_clearopts(cr, vfsp);
2059 	}
2060 
2061 	/*
2062 	 * Refuse to mount a filesystem if we are in a local zone and the
2063 	 * dataset is not visible.
2064 	 */
2065 	if (!INGLOBALZONE(curthread) &&
2066 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
2067 		error = SET_ERROR(EPERM);
2068 		goto out;
2069 	}
2070 
2071 #ifdef SECLABEL
2072 	error = zfs_mount_label_policy(vfsp, osname);
2073 	if (error)
2074 		goto out;
2075 #endif
2076 
2077 #ifdef __FreeBSD_kernel__
2078 	vfsp->vfs_flag |= MNT_NFS4ACLS;
2079 #endif
2080 #ifdef __NetBSD__
2081 	vfsp->mnt_iflag |= IMNT_MPSAFE | IMNT_NCLOOKUP;
2082 #endif
2083 
2084 	/*
2085 	 * When doing a remount, we simply refresh our temporary properties
2086 	 * according to those options set in the current VFS options.
2087 	 */
2088 	if (vfsp->vfs_flag & MS_REMOUNT) {
2089 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
2090 
2091 		/*
2092 		 * Refresh mount options with z_teardown_lock blocking I/O while
2093 		 * the filesystem is in an inconsistent state.
2094 		 * The lock also serializes this code with filesystem
2095 		 * manipulations between entry to zfs_suspend_fs() and return
2096 		 * from zfs_resume_fs().
2097 		 */
2098 		rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2099 		zfs_unregister_callbacks(zfsvfs);
2100 		error = zfs_register_callbacks(vfsp);
2101 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2102 		goto out;
2103 	}
2104 
2105 #ifdef __FreeBSD_kernel__
2106 	/* Initial root mount: try hard to import the requested root pool. */
2107 	if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
2108 	    (vfsp->vfs_flag & MNT_UPDATE) == 0) {
2109 		char pname[MAXNAMELEN];
2110 
2111 		error = getpoolname(osname, pname);
2112 		if (error == 0)
2113 			error = spa_import_rootpool(pname);
2114 		if (error)
2115 			goto out;
2116 	}
2117 #endif
2118 
2119 	DROP_GIANT();
2120 	error = zfs_domount(vfsp, osname);
2121 	PICKUP_GIANT();
2122 
2123 #ifdef illumos
2124 	/*
2125 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
2126 	 * disappear due to a forced unmount.
2127 	 */
2128 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
2129 		VFS_HOLD(mvp->v_vfsp);
2130 #endif
2131 
2132 #ifdef __NetBSD__
2133 	/* setup zfs mount info */
2134 	strlcpy(vfsp->mnt_stat.f_mntfromname, osname,
2135 	    sizeof(vfsp->mnt_stat.f_mntfromname));
2136 	set_statvfs_info(path, UIO_USERSPACE, vfsp->mnt_stat.f_mntfromname,
2137 	    UIO_SYSSPACE, vfsp->mnt_op->vfs_name, vfsp, curlwp);
2138 #endif
2139 
2140 out:
2141 	return (error);
2142 }
2143 
2144 #ifdef __FreeBSD_kernel__
2145 static int
zfs_statfs(vfs_t * vfsp,struct statfs * statp)2146 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
2147 #endif
2148 #ifdef __NetBSD__
2149 static int
2150 zfs_statvfs(vfs_t *vfsp, struct statvfs *statp)
2151 #endif
2152 {
2153 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2154 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
2155 
2156 #ifdef __FreeBSD_kernel__
2157 	statp->f_version = STATFS_VERSION;
2158 #endif
2159 
2160 	ZFS_ENTER(zfsvfs);
2161 
2162 	dmu_objset_space(zfsvfs->z_os,
2163 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
2164 
2165 	/*
2166 	 * The underlying storage pool actually uses multiple block sizes.
2167 	 * We report the fragsize as the smallest block size we support,
2168 	 * and we report our blocksize as the filesystem's maximum blocksize.
2169 	 */
2170 	statp->f_bsize = SPA_MINBLOCKSIZE;
2171 #ifdef __NetBSD__
2172 	statp->f_frsize = SPA_MINBLOCKSIZE;
2173 #endif
2174 	statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
2175 
2176 	/*
2177 	 * The following report "total" blocks of various kinds in the
2178 	 * file system, but reported in terms of f_frsize - the
2179 	 * "fragment" size.
2180 	 */
2181 
2182 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
2183 	statp->f_bfree = availbytes / statp->f_bsize;
2184 	statp->f_bavail = statp->f_bfree; /* no root reservation */
2185 	statp->f_bresvd = 0;
2186 
2187 	/*
2188 	 * statvfs() should really be called statufs(), because it assumes
2189 	 * static metadata.  ZFS doesn't preallocate files, so the best
2190 	 * we can do is report the max that could possibly fit in f_files,
2191 	 * and that minus the number actually used in f_ffree.
2192 	 * For f_ffree, report the smaller of the number of object available
2193 	 * and the number of blocks (each object will take at least a block).
2194 	 */
2195 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
2196 #ifndef __FreeBSD__
2197 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
2198 #endif
2199 	statp->f_files = statp->f_ffree + usedobjs;
2200 	statp->f_fresvd = 0;
2201 
2202 #ifdef __FreeBSD__
2203 	(void) cmpldev(&d32, vfsp->vfs_dev);
2204 	statp->f_fsid = d32;
2205 #endif
2206 #ifdef __NetBSD__
2207 	statp->f_fsid = vfsp->mnt_stat.f_fsid;
2208 	statp->f_fsidx = vfsp->mnt_stat.f_fsidx;
2209 #endif
2210 
2211 	/*
2212 	 * We're a zfs filesystem.
2213 	 */
2214 	(void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
2215 
2216 	strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
2217 	    sizeof(statp->f_mntfromname));
2218 	strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
2219 	    sizeof(statp->f_mntonname));
2220 
2221 #ifdef __FreeBSD_kernel__
2222 	statp->f_namemax = MAXNAMELEN - 1;
2223 #endif
2224 #ifdef __NetBSD__
2225 	statp->f_namemax = ZFS_MAXNAMELEN;
2226 #endif
2227 
2228 	ZFS_EXIT(zfsvfs);
2229 	return (0);
2230 }
2231 
2232 static int
zfs_root(vfs_t * vfsp,int flags,vnode_t ** vpp)2233 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
2234 {
2235 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2236 	znode_t *rootzp;
2237 	int error;
2238 
2239 	ZFS_ENTER(zfsvfs);
2240 
2241 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
2242 	if (error == 0)
2243 		*vpp = ZTOV(rootzp);
2244 
2245 	ZFS_EXIT(zfsvfs);
2246 
2247 	if (error == 0) {
2248 		error = vn_lock(*vpp, flags);
2249 		if (error != 0) {
2250 			VN_RELE(*vpp);
2251 			*vpp = NULL;
2252 		}
2253 	}
2254 	return (error);
2255 }
2256 
2257 /*
2258  * Teardown the zfsvfs::z_os.
2259  *
2260  * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
2261  * and 'z_teardown_inactive_lock' held.
2262  */
2263 static int
zfsvfs_teardown(zfsvfs_t * zfsvfs,boolean_t unmounting)2264 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2265 {
2266 	znode_t	*zp;
2267 
2268 	rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2269 
2270 	if (!unmounting) {
2271 		/*
2272 		 * We purge the parent filesystem's vfsp as the parent
2273 		 * filesystem and all of its snapshots have their vnode's
2274 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
2275 		 * 'z_parent' is self referential for non-snapshots.
2276 		 */
2277 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2278 #ifdef FREEBSD_NAMECACHE
2279 		cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
2280 #endif
2281 	}
2282 
2283 	/*
2284 	 * Close the zil. NB: Can't close the zil while zfs_inactive
2285 	 * threads are blocked as zil_close can call zfs_inactive.
2286 	 */
2287 	if (zfsvfs->z_log) {
2288 		zil_close(zfsvfs->z_log);
2289 		zfsvfs->z_log = NULL;
2290 	}
2291 
2292 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
2293 
2294 	/*
2295 	 * If we are not unmounting (ie: online recv) and someone already
2296 	 * unmounted this file system while we were doing the switcheroo,
2297 	 * or a reopen of z_os failed then just bail out now.
2298 	 */
2299 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2300 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2301 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2302 		return (SET_ERROR(EIO));
2303 	}
2304 
2305 	/*
2306 	 * At this point there are no vops active, and any new vops will
2307 	 * fail with EIO since we have z_teardown_lock for writer (only
2308 	 * relavent for forced unmount).
2309 	 *
2310 	 * Release all holds on dbufs.
2311 	 */
2312 	mutex_enter(&zfsvfs->z_znodes_lock);
2313 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2314 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
2315 		if (zp->z_sa_hdl) {
2316 #ifdef __NetBSD__
2317 			ASSERT(vrefcnt(ZTOV(zp)) >= 0);
2318 #else
2319 			ASSERT(ZTOV(zp)->v_count >= 0);
2320 #endif
2321 			zfs_znode_dmu_fini(zp);
2322 		}
2323 	mutex_exit(&zfsvfs->z_znodes_lock);
2324 
2325 	/*
2326 	 * If we are unmounting, set the unmounted flag and let new vops
2327 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
2328 	 * other vops will fail with EIO.
2329 	 */
2330 	if (unmounting) {
2331 		zfsvfs->z_unmounted = B_TRUE;
2332 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2333 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2334 	}
2335 
2336 	/*
2337 	 * z_os will be NULL if there was an error in attempting to reopen
2338 	 * zfsvfs, so just return as the properties had already been
2339 	 * unregistered and cached data had been evicted before.
2340 	 */
2341 	if (zfsvfs->z_os == NULL)
2342 		return (0);
2343 
2344 	/*
2345 	 * Unregister properties.
2346 	 */
2347 	zfs_unregister_callbacks(zfsvfs);
2348 
2349 	/*
2350 	 * Evict cached data
2351 	 */
2352 	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2353 	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2354 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2355 	dmu_objset_evict_dbufs(zfsvfs->z_os);
2356 
2357 	return (0);
2358 }
2359 
2360 /*ARGSUSED*/
2361 static int
zfs_umount(vfs_t * vfsp,int fflag)2362 zfs_umount(vfs_t *vfsp, int fflag)
2363 {
2364 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2365 	objset_t *os;
2366 	int ret;
2367 #ifdef __FreeBSD_kernel__
2368 	kthread_t *td = curthread;
2369 	cred_t *cr = td->td_ucred;
2370 #endif
2371 #ifdef __NetBSD__
2372 	cred_t *cr = CRED();
2373 #endif
2374 
2375 	ret = secpolicy_fs_unmount(cr, vfsp);
2376 	if (ret) {
2377 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2378 		    ZFS_DELEG_PERM_MOUNT, cr))
2379 			return (ret);
2380 	}
2381 
2382 	/*
2383 	 * We purge the parent filesystem's vfsp as the parent filesystem
2384 	 * and all of its snapshots have their vnode's v_vfsp set to the
2385 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
2386 	 * referential for non-snapshots.
2387 	 */
2388 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2389 
2390 	/*
2391 	 * Unmount any snapshots mounted under .zfs before unmounting the
2392 	 * dataset itself.
2393 	 */
2394 	if (zfsvfs->z_ctldir != NULL) {
2395 		if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
2396 			return (ret);
2397 	}
2398 
2399 	if (fflag & MS_FORCE) {
2400 		/*
2401 		 * Mark file system as unmounted before calling
2402 		 * vflush(FORCECLOSE). This way we ensure no future vnops
2403 		 * will be called and risk operating on DOOMED vnodes.
2404 		 */
2405 		rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2406 		zfsvfs->z_unmounted = B_TRUE;
2407 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2408 	}
2409 
2410 	/*
2411 	 * Flush all the files.
2412 	 */
2413 #ifdef __FreeBSD_kernel__
2414 	ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
2415 #endif
2416 #ifdef __NetBSD__
2417 	ret = vflush(vfsp, NULL, (fflag & MS_FORCE) ? FORCECLOSE : 0);
2418 #endif
2419 	if (ret != 0)
2420 		return (ret);
2421 
2422 #ifdef illumos
2423 	if (!(fflag & MS_FORCE)) {
2424 		/*
2425 		 * Check the number of active vnodes in the file system.
2426 		 * Our count is maintained in the vfs structure, but the
2427 		 * number is off by 1 to indicate a hold on the vfs
2428 		 * structure itself.
2429 		 *
2430 		 * The '.zfs' directory maintains a reference of its
2431 		 * own, and any active references underneath are
2432 		 * reflected in the vnode count.
2433 		 */
2434 		if (zfsvfs->z_ctldir == NULL) {
2435 			if (vfsp->vfs_count > 1)
2436 				return (SET_ERROR(EBUSY));
2437 		} else {
2438 			if (vfsp->vfs_count > 2 ||
2439 			    zfsvfs->z_ctldir->v_count > 1)
2440 				return (SET_ERROR(EBUSY));
2441 		}
2442 	}
2443 #endif
2444 
2445 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2446 	os = zfsvfs->z_os;
2447 
2448 	/*
2449 	 * z_os will be NULL if there was an error in
2450 	 * attempting to reopen zfsvfs.
2451 	 */
2452 	if (os != NULL) {
2453 		/*
2454 		 * Unset the objset user_ptr.
2455 		 */
2456 		mutex_enter(&os->os_user_ptr_lock);
2457 		dmu_objset_set_user(os, NULL);
2458 		mutex_exit(&os->os_user_ptr_lock);
2459 
2460 		/*
2461 		 * Finally release the objset
2462 		 */
2463 		dmu_objset_disown(os, zfsvfs);
2464 	}
2465 
2466 	/*
2467 	 * We can now safely destroy the '.zfs' directory node.
2468 	 */
2469 	if (zfsvfs->z_ctldir != NULL)
2470 		zfsctl_destroy(zfsvfs);
2471 	zfs_freevfs(vfsp);
2472 
2473 	return (0);
2474 }
2475 
2476 static int
zfs_vget(vfs_t * vfsp,ino_t ino,int flags,vnode_t ** vpp)2477 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2478 {
2479 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
2480 	znode_t		*zp;
2481 	int 		err;
2482 
2483 	/*
2484 	 * zfs_zget() can't operate on virtual entries like .zfs/ or
2485 	 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2486 	 * This will make NFS to switch to LOOKUP instead of using VGET.
2487 	 */
2488 	if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2489 	    (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2490 		return (EOPNOTSUPP);
2491 
2492 	ZFS_ENTER(zfsvfs);
2493 	err = zfs_zget(zfsvfs, ino, &zp);
2494 	if (err == 0 && zp->z_unlinked) {
2495 		VN_RELE(ZTOV(zp));
2496 		err = EINVAL;
2497 	}
2498 	if (err == 0)
2499 		*vpp = ZTOV(zp);
2500 	ZFS_EXIT(zfsvfs);
2501 	if (err == 0)
2502 		err = vn_lock(*vpp, flags);
2503 	if (err != 0)
2504 		*vpp = NULL;
2505 
2506 	return (err);
2507 }
2508 
2509 #ifdef __FreeBSD_kernel__
2510 static int
zfs_checkexp(vfs_t * vfsp,struct sockaddr * nam,int * extflagsp,struct ucred ** credanonp,int * numsecflavors,int ** secflavors)2511 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2512     struct ucred **credanonp, int *numsecflavors, int **secflavors)
2513 {
2514 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2515 
2516 	/*
2517 	 * If this is regular file system vfsp is the same as
2518 	 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2519 	 * zfsvfs->z_parent->z_vfs represents parent file system
2520 	 * which we have to use here, because only this file system
2521 	 * has mnt_export configured.
2522 	 */
2523 	return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2524 	    credanonp, numsecflavors, secflavors));
2525 }
2526 
2527 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2528 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2529 #endif
2530 
2531 #ifdef __FreeBSD_kernel__
2532 static int
zfs_fhtovp(vfs_t * vfsp,fid_t * fidp,int flags,vnode_t ** vpp)2533 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2534 {
2535 	struct componentname cn;
2536 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
2537 	znode_t		*zp;
2538 	vnode_t		*dvp;
2539 	uint64_t	object = 0;
2540 	uint64_t	fid_gen = 0;
2541 	uint64_t	gen_mask;
2542 	uint64_t	zp_gen;
2543 	int 		i, err;
2544 
2545 	*vpp = NULL;
2546 
2547 	ZFS_ENTER(zfsvfs);
2548 
2549 	/*
2550 	 * On FreeBSD we can get snapshot's mount point or its parent file
2551 	 * system mount point depending if snapshot is already mounted or not.
2552 	 */
2553 	if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2554 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
2555 		uint64_t	objsetid = 0;
2556 		uint64_t	setgen = 0;
2557 
2558 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2559 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2560 
2561 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2562 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2563 
2564 		ZFS_EXIT(zfsvfs);
2565 
2566 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2567 		if (err)
2568 			return (SET_ERROR(EINVAL));
2569 		ZFS_ENTER(zfsvfs);
2570 	}
2571 
2572 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2573 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
2574 
2575 		for (i = 0; i < sizeof (zfid->zf_object); i++)
2576 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2577 
2578 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
2579 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2580 	} else {
2581 		ZFS_EXIT(zfsvfs);
2582 		return (SET_ERROR(EINVAL));
2583 	}
2584 
2585 	/*
2586 	 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2587 	 * directory tree. If the object == zfsvfs->z_shares_dir, then
2588 	 * we are in the .zfs/shares directory tree.
2589 	 */
2590 	if ((fid_gen == 0 &&
2591 	     (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2592 	    (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2593 		ZFS_EXIT(zfsvfs);
2594 		VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2595 		if (object == ZFSCTL_INO_SNAPDIR) {
2596 			cn.cn_nameptr = "snapshot";
2597 			cn.cn_namelen = strlen(cn.cn_nameptr);
2598 			cn.cn_nameiop = LOOKUP;
2599 			cn.cn_flags = ISLASTCN | LOCKLEAF;
2600 			cn.cn_lkflags = flags;
2601 			VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2602 			vput(dvp);
2603 		} else if (object == zfsvfs->z_shares_dir) {
2604 			/*
2605 			 * XXX This branch must not be taken,
2606 			 * if it is, then the lookup below will
2607 			 * explode.
2608 			 */
2609 			cn.cn_nameptr = "shares";
2610 			cn.cn_namelen = strlen(cn.cn_nameptr);
2611 			cn.cn_nameiop = LOOKUP;
2612 			cn.cn_flags = ISLASTCN;
2613 			cn.cn_lkflags = flags;
2614 			VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2615 			vput(dvp);
2616 		} else {
2617 			*vpp = dvp;
2618 		}
2619 		return (err);
2620 	}
2621 
2622 	gen_mask = -1ULL >> (64 - 8 * i);
2623 
2624 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2625 	if (err = zfs_zget(zfsvfs, object, &zp)) {
2626 		ZFS_EXIT(zfsvfs);
2627 		return (err);
2628 	}
2629 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2630 	    sizeof (uint64_t));
2631 	zp_gen = zp_gen & gen_mask;
2632 	if (zp_gen == 0)
2633 		zp_gen = 1;
2634 	if (zp->z_unlinked || zp_gen != fid_gen) {
2635 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2636 		VN_RELE(ZTOV(zp));
2637 		ZFS_EXIT(zfsvfs);
2638 		return (SET_ERROR(EINVAL));
2639 	}
2640 
2641 	*vpp = ZTOV(zp);
2642 	ZFS_EXIT(zfsvfs);
2643 	err = vn_lock(*vpp, flags);
2644 	if (err == 0)
2645 		vnode_create_vobject(*vpp, zp->z_size, curthread);
2646 	else
2647 		*vpp = NULL;
2648 	return (err);
2649 }
2650 #endif /* __FreeBSD_kernel__ */
2651 
2652 /*
2653  * Block out VOPs and close zfsvfs_t::z_os
2654  *
2655  * Note, if successful, then we return with the 'z_teardown_lock' and
2656  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2657  * dataset and objset intact so that they can be atomically handed off during
2658  * a subsequent rollback or recv operation and the resume thereafter.
2659  */
2660 int
zfs_suspend_fs(zfsvfs_t * zfsvfs)2661 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2662 {
2663 	int error;
2664 
2665 #ifdef __NetBSD__
2666 	if ((error = vfs_suspend(zfsvfs->z_vfs, 0)) != 0)
2667 		return error;
2668 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) {
2669 		vfs_resume(zfsvfs->z_vfs);
2670 		return (error);
2671 	}
2672 #else
2673 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2674 		return (error);
2675 #endif
2676 
2677 	return (0);
2678 }
2679 
2680 /*
2681  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2682  * is an invariant across any of the operations that can be performed while the
2683  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2684  * are the same: the relevant objset and associated dataset are owned by
2685  * zfsvfs, held, and long held on entry.
2686  */
2687 #ifdef __NetBSD__
2688 static bool
zfs_resume_selector(void * cl,struct vnode * vp)2689 zfs_resume_selector(void *cl, struct vnode *vp)
2690 {
2691 
2692 	if (zfsctl_is_node(vp))
2693 		return false;
2694 	return (VTOZ(vp)->z_sa_hdl == NULL);
2695 }
2696 #endif
2697 int
zfs_resume_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)2698 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2699 {
2700 	int err;
2701 	znode_t *zp;
2702 
2703 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2704 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2705 
2706 	/*
2707 	 * We already own this, so just update the objset_t, as the one we
2708 	 * had before may have been evicted.
2709 	 */
2710 	objset_t *os;
2711 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
2712 	VERIFY(dsl_dataset_long_held(ds));
2713 	VERIFY0(dmu_objset_from_ds(ds, &os));
2714 
2715 	err = zfsvfs_init(zfsvfs, os);
2716 	if (err != 0)
2717 		goto bail;
2718 
2719 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2720 
2721 	zfs_set_fuid_feature(zfsvfs);
2722 
2723 	/*
2724 	 * Attempt to re-establish all the active znodes with
2725 	 * their dbufs.  If a zfs_rezget() fails, then we'll let
2726 	 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2727 	 * when they try to use their znode.
2728 	 */
2729 	mutex_enter(&zfsvfs->z_znodes_lock);
2730 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2731 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2732 		(void) zfs_rezget(zp);
2733 	}
2734 	mutex_exit(&zfsvfs->z_znodes_lock);
2735 
2736 bail:
2737 	/* release the VOPs */
2738 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2739 	rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2740 #ifdef __NetBSD__
2741 	struct vnode_iterator *marker;
2742 	vnode_t *vp;
2743 
2744 	vfs_vnode_iterator_init(zfsvfs->z_vfs, &marker);
2745 	while ((vp = vfs_vnode_iterator_next(marker,
2746 	    zfs_resume_selector, NULL))) {
2747 		vgone(vp);
2748 	}
2749 	vfs_vnode_iterator_destroy(marker);
2750 	vfs_resume(zfsvfs->z_vfs);
2751 #endif
2752 
2753 	if (err) {
2754 		/*
2755 		 * Since we couldn't setup the sa framework, try to force
2756 		 * unmount this file system.
2757 		 */
2758 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2759 			vfs_ref(zfsvfs->z_vfs);
2760 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2761 		}
2762 	}
2763 	return (err);
2764 }
2765 
2766 static void
zfs_freevfs(vfs_t * vfsp)2767 zfs_freevfs(vfs_t *vfsp)
2768 {
2769 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2770 
2771 #ifdef illumos
2772 	/*
2773 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2774 	 * from zfs_mount().  Release it here.  If we came through
2775 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2776 	 * skip the VFS_RELE for rootvfs.
2777 	 */
2778 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2779 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2780 #endif
2781 
2782 	zfsvfs_free(zfsvfs);
2783 
2784 	atomic_dec_32(&zfs_active_fs_count);
2785 }
2786 
2787 #ifdef __FreeBSD_kernel__
2788 #ifdef __i386__
2789 static int desiredvnodes_backup;
2790 #endif
2791 
2792 static void
zfs_vnodes_adjust(void)2793 zfs_vnodes_adjust(void)
2794 {
2795 #ifdef __i386__
2796 	int newdesiredvnodes;
2797 
2798 	desiredvnodes_backup = desiredvnodes;
2799 
2800 	/*
2801 	 * We calculate newdesiredvnodes the same way it is done in
2802 	 * vntblinit(). If it is equal to desiredvnodes, it means that
2803 	 * it wasn't tuned by the administrator and we can tune it down.
2804 	 */
2805 	newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2806 	    vm_kmem_size / (5 * (sizeof(struct vm_object) +
2807 	    sizeof(struct vnode))));
2808 	if (newdesiredvnodes == desiredvnodes)
2809 		desiredvnodes = (3 * newdesiredvnodes) / 4;
2810 #endif
2811 }
2812 
2813 static void
zfs_vnodes_adjust_back(void)2814 zfs_vnodes_adjust_back(void)
2815 {
2816 
2817 #ifdef __i386__
2818 	desiredvnodes = desiredvnodes_backup;
2819 #endif
2820 }
2821 #endif /* __FreeBSD_kernel__ */
2822 
2823 #ifdef __NetBSD__
2824 static void
zfs_vnodes_adjust(void)2825 zfs_vnodes_adjust(void)
2826 {
2827 }
2828 
2829 static void
zfs_vnodes_adjust_back(void)2830 zfs_vnodes_adjust_back(void)
2831 {
2832 }
2833 #endif
2834 
2835 void
zfs_init(void)2836 zfs_init(void)
2837 {
2838 
2839 	printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2840 
2841 	/*
2842 	 * Initialize .zfs directory structures
2843 	 */
2844 	zfsctl_init();
2845 
2846 	/*
2847 	 * Initialize znode cache, vnode ops, etc...
2848 	 */
2849 	zfs_znode_init();
2850 
2851 	/*
2852 	 * Reduce number of vnodes. Originally number of vnodes is calculated
2853 	 * with UFS inode in mind. We reduce it here, because it's too big for
2854 	 * ZFS/i386.
2855 	 */
2856 	zfs_vnodes_adjust();
2857 
2858 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2859 }
2860 
2861 void
zfs_fini(void)2862 zfs_fini(void)
2863 {
2864 	zfsctl_fini();
2865 	zfs_znode_fini();
2866 	zfs_vnodes_adjust_back();
2867 }
2868 
2869 int
zfs_busy(void)2870 zfs_busy(void)
2871 {
2872 	return (zfs_active_fs_count != 0);
2873 }
2874 
2875 int
zfs_set_version(zfsvfs_t * zfsvfs,uint64_t newvers)2876 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2877 {
2878 	int error;
2879 	objset_t *os = zfsvfs->z_os;
2880 	dmu_tx_t *tx;
2881 
2882 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2883 		return (SET_ERROR(EINVAL));
2884 
2885 	if (newvers < zfsvfs->z_version)
2886 		return (SET_ERROR(EINVAL));
2887 
2888 	if (zfs_spa_version_map(newvers) >
2889 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2890 		return (SET_ERROR(ENOTSUP));
2891 
2892 	tx = dmu_tx_create(os);
2893 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2894 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2895 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2896 		    ZFS_SA_ATTRS);
2897 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2898 	}
2899 	error = dmu_tx_assign(tx, TXG_WAIT);
2900 	if (error) {
2901 		dmu_tx_abort(tx);
2902 		return (error);
2903 	}
2904 
2905 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2906 	    8, 1, &newvers, tx);
2907 
2908 	if (error) {
2909 		dmu_tx_commit(tx);
2910 		return (error);
2911 	}
2912 
2913 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2914 		uint64_t sa_obj;
2915 
2916 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2917 		    SPA_VERSION_SA);
2918 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2919 		    DMU_OT_NONE, 0, tx);
2920 
2921 		error = zap_add(os, MASTER_NODE_OBJ,
2922 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2923 		ASSERT0(error);
2924 
2925 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2926 		sa_register_update_callback(os, zfs_sa_upgrade);
2927 	}
2928 
2929 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2930 	    "from %llu to %llu", zfsvfs->z_version, newvers);
2931 
2932 	dmu_tx_commit(tx);
2933 
2934 	zfsvfs->z_version = newvers;
2935 
2936 	zfs_set_fuid_feature(zfsvfs);
2937 
2938 	return (0);
2939 }
2940 
2941 /*
2942  * Read a property stored within the master node.
2943  */
2944 int
zfs_get_zplprop(objset_t * os,zfs_prop_t prop,uint64_t * value)2945 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2946 {
2947 	const char *pname;
2948 	int error = ENOENT;
2949 
2950 	/*
2951 	 * Look up the file system's value for the property.  For the
2952 	 * version property, we look up a slightly different string.
2953 	 */
2954 	if (prop == ZFS_PROP_VERSION)
2955 		pname = ZPL_VERSION_STR;
2956 	else
2957 		pname = zfs_prop_to_name(prop);
2958 
2959 	if (os != NULL)
2960 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2961 
2962 	if (error == ENOENT) {
2963 		/* No value set, use the default value */
2964 		switch (prop) {
2965 		case ZFS_PROP_VERSION:
2966 			*value = ZPL_VERSION;
2967 			break;
2968 		case ZFS_PROP_NORMALIZE:
2969 		case ZFS_PROP_UTF8ONLY:
2970 			*value = 0;
2971 			break;
2972 		case ZFS_PROP_CASE:
2973 			*value = ZFS_CASE_SENSITIVE;
2974 			break;
2975 		default:
2976 			return (error);
2977 		}
2978 		error = 0;
2979 	}
2980 	return (error);
2981 }
2982 
2983 #if defined(__FreeBSD_kernel__) || defined(__NetBSD__)
2984 #ifdef _KERNEL
2985 void
zfsvfs_update_fromname(const char * oldname,const char * newname)2986 zfsvfs_update_fromname(const char *oldname, const char *newname)
2987 {
2988 	char tmpbuf[MAXPATHLEN];
2989 	struct mount *mp;
2990 	char *fromname;
2991 	size_t oldlen;
2992 
2993 	oldlen = strlen(oldname);
2994 
2995 #ifdef __NetBSD__
2996 	mount_iterator_t *iter;
2997 	mountlist_iterator_init(&iter);
2998 	while ((mp = mountlist_iterator_next(iter)) != NULL) {
2999 #else
3000 	mtx_lock(&mountlist_mtx);
3001 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
3002 #endif
3003 		fromname = mp->mnt_stat.f_mntfromname;
3004 		if (strcmp(fromname, oldname) == 0) {
3005 			(void)strlcpy(fromname, newname,
3006 			    sizeof(mp->mnt_stat.f_mntfromname));
3007 			continue;
3008 		}
3009 		if (strncmp(fromname, oldname, oldlen) == 0 &&
3010 		    (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
3011 			(void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
3012 			    newname, fromname + oldlen);
3013 			(void)strlcpy(fromname, tmpbuf,
3014 			    sizeof(mp->mnt_stat.f_mntfromname));
3015 			continue;
3016 		}
3017 	}
3018 #ifdef __NetBSD__
3019 	mountlist_iterator_destroy(iter);
3020 #else
3021 	mtx_unlock(&mountlist_mtx);
3022 #endif
3023 }
3024 #endif
3025 #endif
3026