xref: /freebsd-14-stable/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c (revision 2ec8b69480708185a273254e4e254140eb2ce633)
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 https://opensource.org/licenses/CDDL-1.0.
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) 2012, 2018 by Delphix. All rights reserved.
24  */
25 
26 /* Portions Copyright 2010 Robert Milkowski */
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/sysmacros.h>
31 #include <sys/kmem.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
34 #include <sys/vfs.h>
35 #include <sys/mntent.h>
36 #include <sys/cmn_err.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/zfs_vnops.h>
39 #include <sys/zfs_dir.h>
40 #include <sys/zil.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/dmu.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_deleg.h>
46 #include <sys/spa.h>
47 #include <sys/zap.h>
48 #include <sys/sa.h>
49 #include <sys/sa_impl.h>
50 #include <sys/policy.h>
51 #include <sys/atomic.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_ctldir.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/zfs_quota.h>
56 #include <sys/sunddi.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/objlist.h>
60 #include <sys/zpl.h>
61 #include <linux/vfs_compat.h>
62 #include "zfs_comutil.h"
63 
64 enum {
65 	TOKEN_RO,
66 	TOKEN_RW,
67 	TOKEN_SETUID,
68 	TOKEN_NOSETUID,
69 	TOKEN_EXEC,
70 	TOKEN_NOEXEC,
71 	TOKEN_DEVICES,
72 	TOKEN_NODEVICES,
73 	TOKEN_DIRXATTR,
74 	TOKEN_SAXATTR,
75 	TOKEN_XATTR,
76 	TOKEN_NOXATTR,
77 	TOKEN_ATIME,
78 	TOKEN_NOATIME,
79 	TOKEN_RELATIME,
80 	TOKEN_NORELATIME,
81 	TOKEN_NBMAND,
82 	TOKEN_NONBMAND,
83 	TOKEN_MNTPOINT,
84 	TOKEN_LAST,
85 };
86 
87 static const match_table_t zpl_tokens = {
88 	{ TOKEN_RO,		MNTOPT_RO },
89 	{ TOKEN_RW,		MNTOPT_RW },
90 	{ TOKEN_SETUID,		MNTOPT_SETUID },
91 	{ TOKEN_NOSETUID,	MNTOPT_NOSETUID },
92 	{ TOKEN_EXEC,		MNTOPT_EXEC },
93 	{ TOKEN_NOEXEC,		MNTOPT_NOEXEC },
94 	{ TOKEN_DEVICES,	MNTOPT_DEVICES },
95 	{ TOKEN_NODEVICES,	MNTOPT_NODEVICES },
96 	{ TOKEN_DIRXATTR,	MNTOPT_DIRXATTR },
97 	{ TOKEN_SAXATTR,	MNTOPT_SAXATTR },
98 	{ TOKEN_XATTR,		MNTOPT_XATTR },
99 	{ TOKEN_NOXATTR,	MNTOPT_NOXATTR },
100 	{ TOKEN_ATIME,		MNTOPT_ATIME },
101 	{ TOKEN_NOATIME,	MNTOPT_NOATIME },
102 	{ TOKEN_RELATIME,	MNTOPT_RELATIME },
103 	{ TOKEN_NORELATIME,	MNTOPT_NORELATIME },
104 	{ TOKEN_NBMAND,		MNTOPT_NBMAND },
105 	{ TOKEN_NONBMAND,	MNTOPT_NONBMAND },
106 	{ TOKEN_MNTPOINT,	MNTOPT_MNTPOINT "=%s" },
107 	{ TOKEN_LAST,		NULL },
108 };
109 
110 static void
zfsvfs_vfs_free(vfs_t * vfsp)111 zfsvfs_vfs_free(vfs_t *vfsp)
112 {
113 	if (vfsp != NULL) {
114 		if (vfsp->vfs_mntpoint != NULL)
115 			kmem_strfree(vfsp->vfs_mntpoint);
116 		mutex_destroy(&vfsp->vfs_mntpt_lock);
117 		kmem_free(vfsp, sizeof (vfs_t));
118 	}
119 }
120 
121 static int
zfsvfs_parse_option(char * option,int token,substring_t * args,vfs_t * vfsp)122 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
123 {
124 	switch (token) {
125 	case TOKEN_RO:
126 		vfsp->vfs_readonly = B_TRUE;
127 		vfsp->vfs_do_readonly = B_TRUE;
128 		break;
129 	case TOKEN_RW:
130 		vfsp->vfs_readonly = B_FALSE;
131 		vfsp->vfs_do_readonly = B_TRUE;
132 		break;
133 	case TOKEN_SETUID:
134 		vfsp->vfs_setuid = B_TRUE;
135 		vfsp->vfs_do_setuid = B_TRUE;
136 		break;
137 	case TOKEN_NOSETUID:
138 		vfsp->vfs_setuid = B_FALSE;
139 		vfsp->vfs_do_setuid = B_TRUE;
140 		break;
141 	case TOKEN_EXEC:
142 		vfsp->vfs_exec = B_TRUE;
143 		vfsp->vfs_do_exec = B_TRUE;
144 		break;
145 	case TOKEN_NOEXEC:
146 		vfsp->vfs_exec = B_FALSE;
147 		vfsp->vfs_do_exec = B_TRUE;
148 		break;
149 	case TOKEN_DEVICES:
150 		vfsp->vfs_devices = B_TRUE;
151 		vfsp->vfs_do_devices = B_TRUE;
152 		break;
153 	case TOKEN_NODEVICES:
154 		vfsp->vfs_devices = B_FALSE;
155 		vfsp->vfs_do_devices = B_TRUE;
156 		break;
157 	case TOKEN_DIRXATTR:
158 		vfsp->vfs_xattr = ZFS_XATTR_DIR;
159 		vfsp->vfs_do_xattr = B_TRUE;
160 		break;
161 	case TOKEN_SAXATTR:
162 		vfsp->vfs_xattr = ZFS_XATTR_SA;
163 		vfsp->vfs_do_xattr = B_TRUE;
164 		break;
165 	case TOKEN_XATTR:
166 		vfsp->vfs_xattr = ZFS_XATTR_DIR;
167 		vfsp->vfs_do_xattr = B_TRUE;
168 		break;
169 	case TOKEN_NOXATTR:
170 		vfsp->vfs_xattr = ZFS_XATTR_OFF;
171 		vfsp->vfs_do_xattr = B_TRUE;
172 		break;
173 	case TOKEN_ATIME:
174 		vfsp->vfs_atime = B_TRUE;
175 		vfsp->vfs_do_atime = B_TRUE;
176 		break;
177 	case TOKEN_NOATIME:
178 		vfsp->vfs_atime = B_FALSE;
179 		vfsp->vfs_do_atime = B_TRUE;
180 		break;
181 	case TOKEN_RELATIME:
182 		vfsp->vfs_relatime = B_TRUE;
183 		vfsp->vfs_do_relatime = B_TRUE;
184 		break;
185 	case TOKEN_NORELATIME:
186 		vfsp->vfs_relatime = B_FALSE;
187 		vfsp->vfs_do_relatime = B_TRUE;
188 		break;
189 	case TOKEN_NBMAND:
190 		vfsp->vfs_nbmand = B_TRUE;
191 		vfsp->vfs_do_nbmand = B_TRUE;
192 		break;
193 	case TOKEN_NONBMAND:
194 		vfsp->vfs_nbmand = B_FALSE;
195 		vfsp->vfs_do_nbmand = B_TRUE;
196 		break;
197 	case TOKEN_MNTPOINT:
198 		if (vfsp->vfs_mntpoint != NULL)
199 			kmem_strfree(vfsp->vfs_mntpoint);
200 		vfsp->vfs_mntpoint = match_strdup(&args[0]);
201 		if (vfsp->vfs_mntpoint == NULL)
202 			return (SET_ERROR(ENOMEM));
203 		break;
204 	default:
205 		break;
206 	}
207 
208 	return (0);
209 }
210 
211 /*
212  * Parse the raw mntopts and return a vfs_t describing the options.
213  */
214 static int
zfsvfs_parse_options(char * mntopts,vfs_t ** vfsp)215 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
216 {
217 	vfs_t *tmp_vfsp;
218 	int error;
219 
220 	tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
221 	mutex_init(&tmp_vfsp->vfs_mntpt_lock, NULL, MUTEX_DEFAULT, NULL);
222 
223 	if (mntopts != NULL) {
224 		substring_t args[MAX_OPT_ARGS];
225 		char *tmp_mntopts, *p, *t;
226 		int token;
227 
228 		tmp_mntopts = t = kmem_strdup(mntopts);
229 		if (tmp_mntopts == NULL)
230 			return (SET_ERROR(ENOMEM));
231 
232 		while ((p = strsep(&t, ",")) != NULL) {
233 			if (!*p)
234 				continue;
235 
236 			args[0].to = args[0].from = NULL;
237 			token = match_token(p, zpl_tokens, args);
238 			error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
239 			if (error) {
240 				kmem_strfree(tmp_mntopts);
241 				zfsvfs_vfs_free(tmp_vfsp);
242 				return (error);
243 			}
244 		}
245 
246 		kmem_strfree(tmp_mntopts);
247 	}
248 
249 	*vfsp = tmp_vfsp;
250 
251 	return (0);
252 }
253 
254 boolean_t
zfs_is_readonly(zfsvfs_t * zfsvfs)255 zfs_is_readonly(zfsvfs_t *zfsvfs)
256 {
257 	return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
258 }
259 
260 int
zfs_sync(struct super_block * sb,int wait,cred_t * cr)261 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
262 {
263 	(void) cr;
264 	zfsvfs_t *zfsvfs = sb->s_fs_info;
265 
266 	/*
267 	 * Semantically, the only requirement is that the sync be initiated.
268 	 * The DMU syncs out txgs frequently, so there's nothing to do.
269 	 */
270 	if (!wait)
271 		return (0);
272 
273 	if (zfsvfs != NULL) {
274 		/*
275 		 * Sync a specific filesystem.
276 		 */
277 		dsl_pool_t *dp;
278 		int error;
279 
280 		if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
281 			return (error);
282 		dp = dmu_objset_pool(zfsvfs->z_os);
283 
284 		/*
285 		 * If the system is shutting down, then skip any
286 		 * filesystems which may exist on a suspended pool.
287 		 */
288 		if (spa_suspended(dp->dp_spa)) {
289 			zfs_exit(zfsvfs, FTAG);
290 			return (0);
291 		}
292 
293 		if (zfsvfs->z_log != NULL)
294 			zil_commit(zfsvfs->z_log, 0);
295 
296 		zfs_exit(zfsvfs, FTAG);
297 	} else {
298 		/*
299 		 * Sync all ZFS filesystems.  This is what happens when you
300 		 * run sync(1).  Unlike other filesystems, ZFS honors the
301 		 * request by waiting for all pools to commit all dirty data.
302 		 */
303 		spa_sync_allpools();
304 	}
305 
306 	return (0);
307 }
308 
309 static void
atime_changed_cb(void * arg,uint64_t newval)310 atime_changed_cb(void *arg, uint64_t newval)
311 {
312 	zfsvfs_t *zfsvfs = arg;
313 	struct super_block *sb = zfsvfs->z_sb;
314 
315 	if (sb == NULL)
316 		return;
317 	/*
318 	 * Update SB_NOATIME bit in VFS super block.  Since atime update is
319 	 * determined by atime_needs_update(), atime_needs_update() needs to
320 	 * return false if atime is turned off, and not unconditionally return
321 	 * false if atime is turned on.
322 	 */
323 	if (newval)
324 		sb->s_flags &= ~SB_NOATIME;
325 	else
326 		sb->s_flags |= SB_NOATIME;
327 }
328 
329 static void
relatime_changed_cb(void * arg,uint64_t newval)330 relatime_changed_cb(void *arg, uint64_t newval)
331 {
332 	((zfsvfs_t *)arg)->z_relatime = newval;
333 }
334 
335 static void
xattr_changed_cb(void * arg,uint64_t newval)336 xattr_changed_cb(void *arg, uint64_t newval)
337 {
338 	zfsvfs_t *zfsvfs = arg;
339 
340 	if (newval == ZFS_XATTR_OFF) {
341 		zfsvfs->z_flags &= ~ZSB_XATTR;
342 	} else {
343 		zfsvfs->z_flags |= ZSB_XATTR;
344 
345 		if (newval == ZFS_XATTR_SA)
346 			zfsvfs->z_xattr_sa = B_TRUE;
347 		else
348 			zfsvfs->z_xattr_sa = B_FALSE;
349 	}
350 }
351 
352 static void
acltype_changed_cb(void * arg,uint64_t newval)353 acltype_changed_cb(void *arg, uint64_t newval)
354 {
355 	zfsvfs_t *zfsvfs = arg;
356 
357 	switch (newval) {
358 	case ZFS_ACLTYPE_NFSV4:
359 	case ZFS_ACLTYPE_OFF:
360 		zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
361 		zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
362 		break;
363 	case ZFS_ACLTYPE_POSIX:
364 #ifdef CONFIG_FS_POSIX_ACL
365 		zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX;
366 		zfsvfs->z_sb->s_flags |= SB_POSIXACL;
367 #else
368 		zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
369 		zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
370 #endif /* CONFIG_FS_POSIX_ACL */
371 		break;
372 	default:
373 		break;
374 	}
375 }
376 
377 static void
blksz_changed_cb(void * arg,uint64_t newval)378 blksz_changed_cb(void *arg, uint64_t newval)
379 {
380 	zfsvfs_t *zfsvfs = arg;
381 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
382 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
383 	ASSERT(ISP2(newval));
384 
385 	zfsvfs->z_max_blksz = newval;
386 }
387 
388 static void
readonly_changed_cb(void * arg,uint64_t newval)389 readonly_changed_cb(void *arg, uint64_t newval)
390 {
391 	zfsvfs_t *zfsvfs = arg;
392 	struct super_block *sb = zfsvfs->z_sb;
393 
394 	if (sb == NULL)
395 		return;
396 
397 	if (newval)
398 		sb->s_flags |= SB_RDONLY;
399 	else
400 		sb->s_flags &= ~SB_RDONLY;
401 }
402 
403 static void
devices_changed_cb(void * arg,uint64_t newval)404 devices_changed_cb(void *arg, uint64_t newval)
405 {
406 }
407 
408 static void
setuid_changed_cb(void * arg,uint64_t newval)409 setuid_changed_cb(void *arg, uint64_t newval)
410 {
411 }
412 
413 static void
exec_changed_cb(void * arg,uint64_t newval)414 exec_changed_cb(void *arg, uint64_t newval)
415 {
416 }
417 
418 static void
nbmand_changed_cb(void * arg,uint64_t newval)419 nbmand_changed_cb(void *arg, uint64_t newval)
420 {
421 	zfsvfs_t *zfsvfs = arg;
422 	struct super_block *sb = zfsvfs->z_sb;
423 
424 	if (sb == NULL)
425 		return;
426 
427 	if (newval == TRUE)
428 		sb->s_flags |= SB_MANDLOCK;
429 	else
430 		sb->s_flags &= ~SB_MANDLOCK;
431 }
432 
433 static void
snapdir_changed_cb(void * arg,uint64_t newval)434 snapdir_changed_cb(void *arg, uint64_t newval)
435 {
436 	((zfsvfs_t *)arg)->z_show_ctldir = newval;
437 }
438 
439 static void
acl_mode_changed_cb(void * arg,uint64_t newval)440 acl_mode_changed_cb(void *arg, uint64_t newval)
441 {
442 	zfsvfs_t *zfsvfs = arg;
443 
444 	zfsvfs->z_acl_mode = newval;
445 }
446 
447 static void
acl_inherit_changed_cb(void * arg,uint64_t newval)448 acl_inherit_changed_cb(void *arg, uint64_t newval)
449 {
450 	((zfsvfs_t *)arg)->z_acl_inherit = newval;
451 }
452 
453 static int
zfs_register_callbacks(vfs_t * vfsp)454 zfs_register_callbacks(vfs_t *vfsp)
455 {
456 	struct dsl_dataset *ds = NULL;
457 	objset_t *os = NULL;
458 	zfsvfs_t *zfsvfs = NULL;
459 	int error = 0;
460 
461 	ASSERT(vfsp);
462 	zfsvfs = vfsp->vfs_data;
463 	ASSERT(zfsvfs);
464 	os = zfsvfs->z_os;
465 
466 	/*
467 	 * The act of registering our callbacks will destroy any mount
468 	 * options we may have.  In order to enable temporary overrides
469 	 * of mount options, we stash away the current values and
470 	 * restore them after we register the callbacks.
471 	 */
472 	if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
473 		vfsp->vfs_do_readonly = B_TRUE;
474 		vfsp->vfs_readonly = B_TRUE;
475 	}
476 
477 	/*
478 	 * Register property callbacks.
479 	 *
480 	 * It would probably be fine to just check for i/o error from
481 	 * the first prop_register(), but I guess I like to go
482 	 * overboard...
483 	 */
484 	ds = dmu_objset_ds(os);
485 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
486 	error = dsl_prop_register(ds,
487 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
488 	error = error ? error : dsl_prop_register(ds,
489 	    zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
490 	error = error ? error : dsl_prop_register(ds,
491 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
492 	error = error ? error : dsl_prop_register(ds,
493 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
494 	error = error ? error : dsl_prop_register(ds,
495 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
496 	error = error ? error : dsl_prop_register(ds,
497 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
498 	error = error ? error : dsl_prop_register(ds,
499 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
500 	error = error ? error : dsl_prop_register(ds,
501 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
502 	error = error ? error : dsl_prop_register(ds,
503 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
504 	error = error ? error : dsl_prop_register(ds,
505 	    zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
506 	error = error ? error : dsl_prop_register(ds,
507 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
508 	error = error ? error : dsl_prop_register(ds,
509 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
510 	    zfsvfs);
511 	error = error ? error : dsl_prop_register(ds,
512 	    zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
513 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
514 	if (error)
515 		goto unregister;
516 
517 	/*
518 	 * Invoke our callbacks to restore temporary mount options.
519 	 */
520 	if (vfsp->vfs_do_readonly)
521 		readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
522 	if (vfsp->vfs_do_setuid)
523 		setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
524 	if (vfsp->vfs_do_exec)
525 		exec_changed_cb(zfsvfs, vfsp->vfs_exec);
526 	if (vfsp->vfs_do_devices)
527 		devices_changed_cb(zfsvfs, vfsp->vfs_devices);
528 	if (vfsp->vfs_do_xattr)
529 		xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
530 	if (vfsp->vfs_do_atime)
531 		atime_changed_cb(zfsvfs, vfsp->vfs_atime);
532 	if (vfsp->vfs_do_relatime)
533 		relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
534 	if (vfsp->vfs_do_nbmand)
535 		nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
536 
537 	return (0);
538 
539 unregister:
540 	dsl_prop_unregister_all(ds, zfsvfs);
541 	return (error);
542 }
543 
544 /*
545  * Takes a dataset, a property, a value and that value's setpoint as
546  * found in the ZAP. Checks if the property has been changed in the vfs.
547  * If so, val and setpoint will be overwritten with updated content.
548  * Otherwise, they are left unchanged.
549  */
550 int
zfs_get_temporary_prop(dsl_dataset_t * ds,zfs_prop_t zfs_prop,uint64_t * val,char * setpoint)551 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
552     char *setpoint)
553 {
554 	int error;
555 	zfsvfs_t *zfvp;
556 	vfs_t *vfsp;
557 	objset_t *os;
558 	uint64_t tmp = *val;
559 
560 	error = dmu_objset_from_ds(ds, &os);
561 	if (error != 0)
562 		return (error);
563 
564 	if (dmu_objset_type(os) != DMU_OST_ZFS)
565 		return (EINVAL);
566 
567 	mutex_enter(&os->os_user_ptr_lock);
568 	zfvp = dmu_objset_get_user(os);
569 	mutex_exit(&os->os_user_ptr_lock);
570 	if (zfvp == NULL)
571 		return (ESRCH);
572 
573 	vfsp = zfvp->z_vfs;
574 
575 	switch (zfs_prop) {
576 	case ZFS_PROP_ATIME:
577 		if (vfsp->vfs_do_atime)
578 			tmp = vfsp->vfs_atime;
579 		break;
580 	case ZFS_PROP_RELATIME:
581 		if (vfsp->vfs_do_relatime)
582 			tmp = vfsp->vfs_relatime;
583 		break;
584 	case ZFS_PROP_DEVICES:
585 		if (vfsp->vfs_do_devices)
586 			tmp = vfsp->vfs_devices;
587 		break;
588 	case ZFS_PROP_EXEC:
589 		if (vfsp->vfs_do_exec)
590 			tmp = vfsp->vfs_exec;
591 		break;
592 	case ZFS_PROP_SETUID:
593 		if (vfsp->vfs_do_setuid)
594 			tmp = vfsp->vfs_setuid;
595 		break;
596 	case ZFS_PROP_READONLY:
597 		if (vfsp->vfs_do_readonly)
598 			tmp = vfsp->vfs_readonly;
599 		break;
600 	case ZFS_PROP_XATTR:
601 		if (vfsp->vfs_do_xattr)
602 			tmp = vfsp->vfs_xattr;
603 		break;
604 	case ZFS_PROP_NBMAND:
605 		if (vfsp->vfs_do_nbmand)
606 			tmp = vfsp->vfs_nbmand;
607 		break;
608 	default:
609 		return (ENOENT);
610 	}
611 
612 	if (tmp != *val) {
613 		if (setpoint)
614 			(void) strcpy(setpoint, "temporary");
615 		*val = tmp;
616 	}
617 	return (0);
618 }
619 
620 /*
621  * Associate this zfsvfs with the given objset, which must be owned.
622  * This will cache a bunch of on-disk state from the objset in the
623  * zfsvfs.
624  */
625 static int
zfsvfs_init(zfsvfs_t * zfsvfs,objset_t * os)626 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
627 {
628 	int error;
629 	uint64_t val;
630 
631 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
632 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
633 	zfsvfs->z_os = os;
634 
635 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
636 	if (error != 0)
637 		return (error);
638 	if (zfsvfs->z_version >
639 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
640 		(void) printk("Can't mount a version %lld file system "
641 		    "on a version %lld pool\n. Pool must be upgraded to mount "
642 		    "this file system.\n", (u_longlong_t)zfsvfs->z_version,
643 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
644 		return (SET_ERROR(ENOTSUP));
645 	}
646 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
647 	if (error != 0)
648 		return (error);
649 	zfsvfs->z_norm = (int)val;
650 
651 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
652 	if (error != 0)
653 		return (error);
654 	zfsvfs->z_utf8 = (val != 0);
655 
656 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
657 	if (error != 0)
658 		return (error);
659 	zfsvfs->z_case = (uint_t)val;
660 
661 	if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
662 		return (error);
663 	zfsvfs->z_acl_type = (uint_t)val;
664 
665 	/*
666 	 * Fold case on file systems that are always or sometimes case
667 	 * insensitive.
668 	 */
669 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
670 	    zfsvfs->z_case == ZFS_CASE_MIXED)
671 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
672 
673 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
674 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
675 
676 	uint64_t sa_obj = 0;
677 	if (zfsvfs->z_use_sa) {
678 		/* should either have both of these objects or none */
679 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
680 		    &sa_obj);
681 		if (error != 0)
682 			return (error);
683 
684 		error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
685 		if ((error == 0) && (val == ZFS_XATTR_SA))
686 			zfsvfs->z_xattr_sa = B_TRUE;
687 	}
688 
689 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
690 	    &zfsvfs->z_root);
691 	if (error != 0)
692 		return (error);
693 	ASSERT(zfsvfs->z_root != 0);
694 
695 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
696 	    &zfsvfs->z_unlinkedobj);
697 	if (error != 0)
698 		return (error);
699 
700 	error = zap_lookup(os, MASTER_NODE_OBJ,
701 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
702 	    8, 1, &zfsvfs->z_userquota_obj);
703 	if (error == ENOENT)
704 		zfsvfs->z_userquota_obj = 0;
705 	else if (error != 0)
706 		return (error);
707 
708 	error = zap_lookup(os, MASTER_NODE_OBJ,
709 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
710 	    8, 1, &zfsvfs->z_groupquota_obj);
711 	if (error == ENOENT)
712 		zfsvfs->z_groupquota_obj = 0;
713 	else if (error != 0)
714 		return (error);
715 
716 	error = zap_lookup(os, MASTER_NODE_OBJ,
717 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
718 	    8, 1, &zfsvfs->z_projectquota_obj);
719 	if (error == ENOENT)
720 		zfsvfs->z_projectquota_obj = 0;
721 	else if (error != 0)
722 		return (error);
723 
724 	error = zap_lookup(os, MASTER_NODE_OBJ,
725 	    zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
726 	    8, 1, &zfsvfs->z_userobjquota_obj);
727 	if (error == ENOENT)
728 		zfsvfs->z_userobjquota_obj = 0;
729 	else if (error != 0)
730 		return (error);
731 
732 	error = zap_lookup(os, MASTER_NODE_OBJ,
733 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
734 	    8, 1, &zfsvfs->z_groupobjquota_obj);
735 	if (error == ENOENT)
736 		zfsvfs->z_groupobjquota_obj = 0;
737 	else if (error != 0)
738 		return (error);
739 
740 	error = zap_lookup(os, MASTER_NODE_OBJ,
741 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
742 	    8, 1, &zfsvfs->z_projectobjquota_obj);
743 	if (error == ENOENT)
744 		zfsvfs->z_projectobjquota_obj = 0;
745 	else if (error != 0)
746 		return (error);
747 
748 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
749 	    &zfsvfs->z_fuid_obj);
750 	if (error == ENOENT)
751 		zfsvfs->z_fuid_obj = 0;
752 	else if (error != 0)
753 		return (error);
754 
755 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
756 	    &zfsvfs->z_shares_dir);
757 	if (error == ENOENT)
758 		zfsvfs->z_shares_dir = 0;
759 	else if (error != 0)
760 		return (error);
761 
762 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
763 	    &zfsvfs->z_attr_table);
764 	if (error != 0)
765 		return (error);
766 
767 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
768 		sa_register_update_callback(os, zfs_sa_upgrade);
769 
770 	return (0);
771 }
772 
773 int
zfsvfs_create(const char * osname,boolean_t readonly,zfsvfs_t ** zfvp)774 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
775 {
776 	objset_t *os;
777 	zfsvfs_t *zfsvfs;
778 	int error;
779 	boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
780 
781 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
782 
783 	error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
784 	if (error != 0) {
785 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
786 		return (error);
787 	}
788 
789 	error = zfsvfs_create_impl(zfvp, zfsvfs, os);
790 
791 	return (error);
792 }
793 
794 
795 /*
796  * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
797  * on a failure.  Do not pass in a statically allocated zfsvfs.
798  */
799 int
zfsvfs_create_impl(zfsvfs_t ** zfvp,zfsvfs_t * zfsvfs,objset_t * os)800 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
801 {
802 	int error;
803 
804 	zfsvfs->z_vfs = NULL;
805 	zfsvfs->z_sb = NULL;
806 	zfsvfs->z_parent = zfsvfs;
807 
808 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
809 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
810 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
811 	    offsetof(znode_t, z_link_node));
812 	ZFS_TEARDOWN_INIT(zfsvfs);
813 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
814 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
815 
816 	int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
817 	    ZFS_OBJ_MTX_MAX);
818 	zfsvfs->z_hold_size = size;
819 	zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
820 	    KM_SLEEP);
821 	zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
822 	for (int i = 0; i != size; i++) {
823 		avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
824 		    sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
825 		mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
826 	}
827 
828 	error = zfsvfs_init(zfsvfs, os);
829 	if (error != 0) {
830 		dmu_objset_disown(os, B_TRUE, zfsvfs);
831 		*zfvp = NULL;
832 		zfsvfs_free(zfsvfs);
833 		return (error);
834 	}
835 
836 	zfsvfs->z_drain_task = TASKQID_INVALID;
837 	zfsvfs->z_draining = B_FALSE;
838 	zfsvfs->z_drain_cancel = B_TRUE;
839 
840 	*zfvp = zfsvfs;
841 	return (0);
842 }
843 
844 static int
zfsvfs_setup(zfsvfs_t * zfsvfs,boolean_t mounting)845 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
846 {
847 	int error;
848 	boolean_t readonly = zfs_is_readonly(zfsvfs);
849 
850 	error = zfs_register_callbacks(zfsvfs->z_vfs);
851 	if (error)
852 		return (error);
853 
854 	/*
855 	 * If we are not mounting (ie: online recv), then we don't
856 	 * have to worry about replaying the log as we blocked all
857 	 * operations out since we closed the ZIL.
858 	 */
859 	if (mounting) {
860 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
861 		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
862 		if (error)
863 			return (error);
864 		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
865 		    &zfsvfs->z_kstat.dk_zil_sums);
866 
867 		/*
868 		 * During replay we remove the read only flag to
869 		 * allow replays to succeed.
870 		 */
871 		if (readonly != 0) {
872 			readonly_changed_cb(zfsvfs, B_FALSE);
873 		} else {
874 			zap_stats_t zs;
875 			if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
876 			    &zs) == 0) {
877 				dataset_kstats_update_nunlinks_kstat(
878 				    &zfsvfs->z_kstat, zs.zs_num_entries);
879 				dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
880 				    "num_entries in unlinked set: %llu",
881 				    zs.zs_num_entries);
882 			}
883 			zfs_unlinked_drain(zfsvfs);
884 			dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
885 			dd->dd_activity_cancelled = B_FALSE;
886 		}
887 
888 		/*
889 		 * Parse and replay the intent log.
890 		 *
891 		 * Because of ziltest, this must be done after
892 		 * zfs_unlinked_drain().  (Further note: ziltest
893 		 * doesn't use readonly mounts, where
894 		 * zfs_unlinked_drain() isn't called.)  This is because
895 		 * ziltest causes spa_sync() to think it's committed,
896 		 * but actually it is not, so the intent log contains
897 		 * many txg's worth of changes.
898 		 *
899 		 * In particular, if object N is in the unlinked set in
900 		 * the last txg to actually sync, then it could be
901 		 * actually freed in a later txg and then reallocated
902 		 * in a yet later txg.  This would write a "create
903 		 * object N" record to the intent log.  Normally, this
904 		 * would be fine because the spa_sync() would have
905 		 * written out the fact that object N is free, before
906 		 * we could write the "create object N" intent log
907 		 * record.
908 		 *
909 		 * But when we are in ziltest mode, we advance the "open
910 		 * txg" without actually spa_sync()-ing the changes to
911 		 * disk.  So we would see that object N is still
912 		 * allocated and in the unlinked set, and there is an
913 		 * intent log record saying to allocate it.
914 		 */
915 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
916 			if (zil_replay_disable) {
917 				zil_destroy(zfsvfs->z_log, B_FALSE);
918 			} else {
919 				zfsvfs->z_replay = B_TRUE;
920 				zil_replay(zfsvfs->z_os, zfsvfs,
921 				    zfs_replay_vector);
922 				zfsvfs->z_replay = B_FALSE;
923 			}
924 		}
925 
926 		/* restore readonly bit */
927 		if (readonly != 0)
928 			readonly_changed_cb(zfsvfs, B_TRUE);
929 	} else {
930 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
931 		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
932 		    &zfsvfs->z_kstat.dk_zil_sums);
933 	}
934 
935 	/*
936 	 * Set the objset user_ptr to track its zfsvfs.
937 	 */
938 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
939 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
940 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
941 
942 	return (0);
943 }
944 
945 void
zfsvfs_free(zfsvfs_t * zfsvfs)946 zfsvfs_free(zfsvfs_t *zfsvfs)
947 {
948 	int i, size = zfsvfs->z_hold_size;
949 
950 	zfs_fuid_destroy(zfsvfs);
951 
952 	mutex_destroy(&zfsvfs->z_znodes_lock);
953 	mutex_destroy(&zfsvfs->z_lock);
954 	list_destroy(&zfsvfs->z_all_znodes);
955 	ZFS_TEARDOWN_DESTROY(zfsvfs);
956 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
957 	rw_destroy(&zfsvfs->z_fuid_lock);
958 	for (i = 0; i != size; i++) {
959 		avl_destroy(&zfsvfs->z_hold_trees[i]);
960 		mutex_destroy(&zfsvfs->z_hold_locks[i]);
961 	}
962 	vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
963 	vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
964 	zfsvfs_vfs_free(zfsvfs->z_vfs);
965 	dataset_kstats_destroy(&zfsvfs->z_kstat);
966 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
967 }
968 
969 static void
zfs_set_fuid_feature(zfsvfs_t * zfsvfs)970 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
971 {
972 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
973 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
974 }
975 
976 static void
zfs_unregister_callbacks(zfsvfs_t * zfsvfs)977 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
978 {
979 	objset_t *os = zfsvfs->z_os;
980 
981 	if (!dmu_objset_is_snapshot(os))
982 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
983 }
984 
985 #ifdef HAVE_MLSLABEL
986 /*
987  * Check that the hex label string is appropriate for the dataset being
988  * mounted into the global_zone proper.
989  *
990  * Return an error if the hex label string is not default or
991  * admin_low/admin_high.  For admin_low labels, the corresponding
992  * dataset must be readonly.
993  */
994 int
zfs_check_global_label(const char * dsname,const char * hexsl)995 zfs_check_global_label(const char *dsname, const char *hexsl)
996 {
997 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
998 		return (0);
999 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1000 		return (0);
1001 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1002 		/* must be readonly */
1003 		uint64_t rdonly;
1004 
1005 		if (dsl_prop_get_integer(dsname,
1006 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1007 			return (SET_ERROR(EACCES));
1008 		return (rdonly ? 0 : SET_ERROR(EACCES));
1009 	}
1010 	return (SET_ERROR(EACCES));
1011 }
1012 #endif /* HAVE_MLSLABEL */
1013 
1014 static int
zfs_statfs_project(zfsvfs_t * zfsvfs,znode_t * zp,struct kstatfs * statp,uint32_t bshift)1015 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1016     uint32_t bshift)
1017 {
1018 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1019 	uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1020 	uint64_t quota;
1021 	uint64_t used;
1022 	int err;
1023 
1024 	strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1025 	err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset,
1026 	    sizeof (buf) - offset, B_FALSE);
1027 	if (err)
1028 		return (err);
1029 
1030 	if (zfsvfs->z_projectquota_obj == 0)
1031 		goto objs;
1032 
1033 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1034 	    buf + offset, 8, 1, &quota);
1035 	if (err == ENOENT)
1036 		goto objs;
1037 	else if (err)
1038 		return (err);
1039 
1040 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1041 	    buf + offset, 8, 1, &used);
1042 	if (unlikely(err == ENOENT)) {
1043 		uint32_t blksize;
1044 		u_longlong_t nblocks;
1045 
1046 		/*
1047 		 * Quota accounting is async, so it is possible race case.
1048 		 * There is at least one object with the given project ID.
1049 		 */
1050 		sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1051 		if (unlikely(zp->z_blksz == 0))
1052 			blksize = zfsvfs->z_max_blksz;
1053 
1054 		used = blksize * nblocks;
1055 	} else if (err) {
1056 		return (err);
1057 	}
1058 
1059 	statp->f_blocks = quota >> bshift;
1060 	statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1061 	statp->f_bavail = statp->f_bfree;
1062 
1063 objs:
1064 	if (zfsvfs->z_projectobjquota_obj == 0)
1065 		return (0);
1066 
1067 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1068 	    buf + offset, 8, 1, &quota);
1069 	if (err == ENOENT)
1070 		return (0);
1071 	else if (err)
1072 		return (err);
1073 
1074 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1075 	    buf, 8, 1, &used);
1076 	if (unlikely(err == ENOENT)) {
1077 		/*
1078 		 * Quota accounting is async, so it is possible race case.
1079 		 * There is at least one object with the given project ID.
1080 		 */
1081 		used = 1;
1082 	} else if (err) {
1083 		return (err);
1084 	}
1085 
1086 	statp->f_files = quota;
1087 	statp->f_ffree = (quota > used) ? (quota - used) : 0;
1088 
1089 	return (0);
1090 }
1091 
1092 int
zfs_statvfs(struct inode * ip,struct kstatfs * statp)1093 zfs_statvfs(struct inode *ip, struct kstatfs *statp)
1094 {
1095 	zfsvfs_t *zfsvfs = ITOZSB(ip);
1096 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1097 	int err = 0;
1098 
1099 	if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1100 		return (err);
1101 
1102 	dmu_objset_space(zfsvfs->z_os,
1103 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1104 
1105 	uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1106 	/*
1107 	 * The underlying storage pool actually uses multiple block
1108 	 * size.  Under Solaris frsize (fragment size) is reported as
1109 	 * the smallest block size we support, and bsize (block size)
1110 	 * as the filesystem's maximum block size.  Unfortunately,
1111 	 * under Linux the fragment size and block size are often used
1112 	 * interchangeably.  Thus we are forced to report both of them
1113 	 * as the filesystem's maximum block size.
1114 	 */
1115 	statp->f_frsize = zfsvfs->z_max_blksz;
1116 	statp->f_bsize = zfsvfs->z_max_blksz;
1117 	uint32_t bshift = fls(statp->f_bsize) - 1;
1118 
1119 	/*
1120 	 * The following report "total" blocks of various kinds in
1121 	 * the file system, but reported in terms of f_bsize - the
1122 	 * "preferred" size.
1123 	 */
1124 
1125 	/* Round up so we never have a filesystem using 0 blocks. */
1126 	refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1127 	statp->f_blocks = (refdbytes + availbytes) >> bshift;
1128 	statp->f_bfree = availbytes >> bshift;
1129 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1130 
1131 	/*
1132 	 * statvfs() should really be called statufs(), because it assumes
1133 	 * static metadata.  ZFS doesn't preallocate files, so the best
1134 	 * we can do is report the max that could possibly fit in f_files,
1135 	 * and that minus the number actually used in f_ffree.
1136 	 * For f_ffree, report the smaller of the number of objects available
1137 	 * and the number of blocks (each object will take at least a block).
1138 	 */
1139 	statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1140 	statp->f_files = statp->f_ffree + usedobjs;
1141 	statp->f_fsid.val[0] = (uint32_t)fsid;
1142 	statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1143 	statp->f_type = ZFS_SUPER_MAGIC;
1144 	statp->f_namelen = MAXNAMELEN - 1;
1145 
1146 	/*
1147 	 * We have all of 40 characters to stuff a string here.
1148 	 * Is there anything useful we could/should provide?
1149 	 */
1150 	memset(statp->f_spare, 0, sizeof (statp->f_spare));
1151 
1152 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1153 	    dmu_objset_projectquota_present(zfsvfs->z_os)) {
1154 		znode_t *zp = ITOZ(ip);
1155 
1156 		if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1157 		    zpl_is_valid_projid(zp->z_projid))
1158 			err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1159 	}
1160 
1161 	zfs_exit(zfsvfs, FTAG);
1162 	return (err);
1163 }
1164 
1165 static int
zfs_root(zfsvfs_t * zfsvfs,struct inode ** ipp)1166 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1167 {
1168 	znode_t *rootzp;
1169 	int error;
1170 
1171 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1172 		return (error);
1173 
1174 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1175 	if (error == 0)
1176 		*ipp = ZTOI(rootzp);
1177 
1178 	zfs_exit(zfsvfs, FTAG);
1179 	return (error);
1180 }
1181 
1182 /*
1183  * The ARC has requested that the filesystem drop entries from the dentry
1184  * and inode caches.  This can occur when the ARC needs to free meta data
1185  * blocks but can't because they are all pinned by entries in these caches.
1186  */
1187 #if defined(HAVE_SUPER_BLOCK_S_SHRINK)
1188 #define	S_SHRINK(sb)	(&(sb)->s_shrink)
1189 #elif defined(HAVE_SUPER_BLOCK_S_SHRINK_PTR)
1190 #define	S_SHRINK(sb)	((sb)->s_shrink)
1191 #endif
1192 
1193 int
zfs_prune(struct super_block * sb,unsigned long nr_to_scan,int * objects)1194 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1195 {
1196 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1197 	int error = 0;
1198 	struct shrinker *shrinker = S_SHRINK(sb);
1199 	struct shrink_control sc = {
1200 		.nr_to_scan = nr_to_scan,
1201 		.gfp_mask = GFP_KERNEL,
1202 	};
1203 
1204 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1205 		return (error);
1206 
1207 #ifdef SHRINKER_NUMA_AWARE
1208 	if (shrinker->flags & SHRINKER_NUMA_AWARE) {
1209 		*objects = 0;
1210 		for_each_online_node(sc.nid) {
1211 			*objects += (*shrinker->scan_objects)(shrinker, &sc);
1212 			/*
1213 			 * reset sc.nr_to_scan, modified by
1214 			 * scan_objects == super_cache_scan
1215 			 */
1216 			sc.nr_to_scan = nr_to_scan;
1217 		}
1218 	} else {
1219 			*objects = (*shrinker->scan_objects)(shrinker, &sc);
1220 	}
1221 #else
1222 	*objects = (*shrinker->scan_objects)(shrinker, &sc);
1223 #endif
1224 
1225 	zfs_exit(zfsvfs, FTAG);
1226 
1227 	dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1228 	    "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1229 	    nr_to_scan, *objects, error);
1230 
1231 	return (error);
1232 }
1233 
1234 /*
1235  * Teardown the zfsvfs_t.
1236  *
1237  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1238  * and 'z_teardown_inactive_lock' held.
1239  */
1240 static int
zfsvfs_teardown(zfsvfs_t * zfsvfs,boolean_t unmounting)1241 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1242 {
1243 	znode_t	*zp;
1244 
1245 	zfs_unlinked_drain_stop_wait(zfsvfs);
1246 
1247 	/*
1248 	 * If someone has not already unmounted this file system,
1249 	 * drain the zrele_taskq to ensure all active references to the
1250 	 * zfsvfs_t have been handled only then can it be safely destroyed.
1251 	 */
1252 	if (zfsvfs->z_os) {
1253 		/*
1254 		 * If we're unmounting we have to wait for the list to
1255 		 * drain completely.
1256 		 *
1257 		 * If we're not unmounting there's no guarantee the list
1258 		 * will drain completely, but iputs run from the taskq
1259 		 * may add the parents of dir-based xattrs to the taskq
1260 		 * so we want to wait for these.
1261 		 *
1262 		 * We can safely check z_all_znodes for being empty because the
1263 		 * VFS has already blocked operations which add to it.
1264 		 */
1265 		int round = 0;
1266 		while (!list_is_empty(&zfsvfs->z_all_znodes)) {
1267 			taskq_wait_outstanding(dsl_pool_zrele_taskq(
1268 			    dmu_objset_pool(zfsvfs->z_os)), 0);
1269 			if (++round > 1 && !unmounting)
1270 				break;
1271 		}
1272 	}
1273 
1274 	ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1275 
1276 	if (!unmounting) {
1277 		/*
1278 		 * We purge the parent filesystem's super block as the
1279 		 * parent filesystem and all of its snapshots have their
1280 		 * inode's super block set to the parent's filesystem's
1281 		 * super block.  Note,  'z_parent' is self referential
1282 		 * for non-snapshots.
1283 		 */
1284 		shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1285 	}
1286 
1287 	/*
1288 	 * Close the zil. NB: Can't close the zil while zfs_inactive
1289 	 * threads are blocked as zil_close can call zfs_inactive.
1290 	 */
1291 	if (zfsvfs->z_log) {
1292 		zil_close(zfsvfs->z_log);
1293 		zfsvfs->z_log = NULL;
1294 	}
1295 
1296 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1297 
1298 	/*
1299 	 * If we are not unmounting (ie: online recv) and someone already
1300 	 * unmounted this file system while we were doing the switcheroo,
1301 	 * or a reopen of z_os failed then just bail out now.
1302 	 */
1303 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1304 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1305 		ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1306 		return (SET_ERROR(EIO));
1307 	}
1308 
1309 	/*
1310 	 * At this point there are no VFS ops active, and any new VFS ops
1311 	 * will fail with EIO since we have z_teardown_lock for writer (only
1312 	 * relevant for forced unmount).
1313 	 *
1314 	 * Release all holds on dbufs. We also grab an extra reference to all
1315 	 * the remaining inodes so that the kernel does not attempt to free
1316 	 * any inodes of a suspended fs. This can cause deadlocks since the
1317 	 * zfs_resume_fs() process may involve starting threads, which might
1318 	 * attempt to free unreferenced inodes to free up memory for the new
1319 	 * thread.
1320 	 */
1321 	if (!unmounting) {
1322 		mutex_enter(&zfsvfs->z_znodes_lock);
1323 		for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1324 		    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1325 			if (zp->z_sa_hdl)
1326 				zfs_znode_dmu_fini(zp);
1327 			if (igrab(ZTOI(zp)) != NULL)
1328 				zp->z_suspended = B_TRUE;
1329 
1330 		}
1331 		mutex_exit(&zfsvfs->z_znodes_lock);
1332 	}
1333 
1334 	/*
1335 	 * If we are unmounting, set the unmounted flag and let new VFS ops
1336 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
1337 	 * other VFS ops will fail with EIO.
1338 	 */
1339 	if (unmounting) {
1340 		zfsvfs->z_unmounted = B_TRUE;
1341 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1342 		ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1343 	}
1344 
1345 	/*
1346 	 * z_os will be NULL if there was an error in attempting to reopen
1347 	 * zfsvfs, so just return as the properties had already been
1348 	 *
1349 	 * unregistered and cached data had been evicted before.
1350 	 */
1351 	if (zfsvfs->z_os == NULL)
1352 		return (0);
1353 
1354 	/*
1355 	 * Unregister properties.
1356 	 */
1357 	zfs_unregister_callbacks(zfsvfs);
1358 
1359 	/*
1360 	 * Evict cached data. We must write out any dirty data before
1361 	 * disowning the dataset.
1362 	 */
1363 	objset_t *os = zfsvfs->z_os;
1364 	boolean_t os_dirty = B_FALSE;
1365 	for (int t = 0; t < TXG_SIZE; t++) {
1366 		if (dmu_objset_is_dirty(os, t)) {
1367 			os_dirty = B_TRUE;
1368 			break;
1369 		}
1370 	}
1371 	if (!zfs_is_readonly(zfsvfs) && os_dirty) {
1372 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1373 	}
1374 	dmu_objset_evict_dbufs(zfsvfs->z_os);
1375 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1376 	dsl_dir_cancel_waiters(dd);
1377 
1378 	return (0);
1379 }
1380 
1381 static atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1382 
1383 int
zfs_domount(struct super_block * sb,zfs_mnt_t * zm,int silent)1384 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1385 {
1386 	const char *osname = zm->mnt_osname;
1387 	struct inode *root_inode = NULL;
1388 	uint64_t recordsize;
1389 	int error = 0;
1390 	zfsvfs_t *zfsvfs = NULL;
1391 	vfs_t *vfs = NULL;
1392 	int canwrite;
1393 	int dataset_visible_zone;
1394 
1395 	ASSERT(zm);
1396 	ASSERT(osname);
1397 
1398 	dataset_visible_zone = zone_dataset_visible(osname, &canwrite);
1399 
1400 	/*
1401 	 * Refuse to mount a filesystem if we are in a namespace and the
1402 	 * dataset is not visible or writable in that namespace.
1403 	 */
1404 	if (!INGLOBALZONE(curproc) &&
1405 	    (!dataset_visible_zone || !canwrite)) {
1406 		return (SET_ERROR(EPERM));
1407 	}
1408 
1409 	error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1410 	if (error)
1411 		return (error);
1412 
1413 	/*
1414 	 * If a non-writable filesystem is being mounted without the
1415 	 * read-only flag, pretend it was set, as done for snapshots.
1416 	 */
1417 	if (!canwrite)
1418 		vfs->vfs_readonly = B_TRUE;
1419 
1420 	error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1421 	if (error) {
1422 		zfsvfs_vfs_free(vfs);
1423 		goto out;
1424 	}
1425 
1426 	if ((error = dsl_prop_get_integer(osname, "recordsize",
1427 	    &recordsize, NULL))) {
1428 		zfsvfs_vfs_free(vfs);
1429 		goto out;
1430 	}
1431 
1432 	vfs->vfs_data = zfsvfs;
1433 	zfsvfs->z_vfs = vfs;
1434 	zfsvfs->z_sb = sb;
1435 	sb->s_fs_info = zfsvfs;
1436 	sb->s_magic = ZFS_SUPER_MAGIC;
1437 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1438 	sb->s_time_gran = 1;
1439 	sb->s_blocksize = recordsize;
1440 	sb->s_blocksize_bits = ilog2(recordsize);
1441 
1442 	error = -super_setup_bdi_name(sb, "%.28s-%ld", "zfs",
1443 	    atomic_long_inc_return(&zfs_bdi_seq));
1444 	if (error)
1445 		goto out;
1446 
1447 	sb->s_bdi->ra_pages = 0;
1448 
1449 	/* Set callback operations for the file system. */
1450 	sb->s_op = &zpl_super_operations;
1451 	sb->s_xattr = zpl_xattr_handlers;
1452 	sb->s_export_op = &zpl_export_operations;
1453 
1454 	/* Set features for file system. */
1455 	zfs_set_fuid_feature(zfsvfs);
1456 
1457 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1458 		uint64_t pval;
1459 
1460 		atime_changed_cb(zfsvfs, B_FALSE);
1461 		readonly_changed_cb(zfsvfs, B_TRUE);
1462 		if ((error = dsl_prop_get_integer(osname,
1463 		    "xattr", &pval, NULL)))
1464 			goto out;
1465 		xattr_changed_cb(zfsvfs, pval);
1466 		if ((error = dsl_prop_get_integer(osname,
1467 		    "acltype", &pval, NULL)))
1468 			goto out;
1469 		acltype_changed_cb(zfsvfs, pval);
1470 		zfsvfs->z_issnap = B_TRUE;
1471 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1472 		zfsvfs->z_snap_defer_time = jiffies;
1473 
1474 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1475 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1476 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1477 	} else {
1478 		if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1479 			goto out;
1480 	}
1481 
1482 	/* Allocate a root inode for the filesystem. */
1483 	error = zfs_root(zfsvfs, &root_inode);
1484 	if (error) {
1485 		(void) zfs_umount(sb);
1486 		zfsvfs = NULL; /* avoid double-free; first in zfs_umount */
1487 		goto out;
1488 	}
1489 
1490 	/* Allocate a root dentry for the filesystem */
1491 	sb->s_root = d_make_root(root_inode);
1492 	if (sb->s_root == NULL) {
1493 		(void) zfs_umount(sb);
1494 		zfsvfs = NULL; /* avoid double-free; first in zfs_umount */
1495 		error = SET_ERROR(ENOMEM);
1496 		goto out;
1497 	}
1498 
1499 	if (!zfsvfs->z_issnap)
1500 		zfsctl_create(zfsvfs);
1501 
1502 	zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1503 out:
1504 	if (error) {
1505 		if (zfsvfs != NULL) {
1506 			dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1507 			zfsvfs_free(zfsvfs);
1508 		}
1509 		/*
1510 		 * make sure we don't have dangling sb->s_fs_info which
1511 		 * zfs_preumount will use.
1512 		 */
1513 		sb->s_fs_info = NULL;
1514 	}
1515 
1516 	return (error);
1517 }
1518 
1519 /*
1520  * Called when an unmount is requested and certain sanity checks have
1521  * already passed.  At this point no dentries or inodes have been reclaimed
1522  * from their respective caches.  We drop the extra reference on the .zfs
1523  * control directory to allow everything to be reclaimed.  All snapshots
1524  * must already have been unmounted to reach this point.
1525  */
1526 void
zfs_preumount(struct super_block * sb)1527 zfs_preumount(struct super_block *sb)
1528 {
1529 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1530 
1531 	/* zfsvfs is NULL when zfs_domount fails during mount */
1532 	if (zfsvfs) {
1533 		zfs_unlinked_drain_stop_wait(zfsvfs);
1534 		zfsctl_destroy(sb->s_fs_info);
1535 		/*
1536 		 * Wait for zrele_async before entering evict_inodes in
1537 		 * generic_shutdown_super. The reason we must finish before
1538 		 * evict_inodes is when lazytime is on, or when zfs_purgedir
1539 		 * calls zfs_zget, zrele would bump i_count from 0 to 1. This
1540 		 * would race with the i_count check in evict_inodes. This means
1541 		 * it could destroy the inode while we are still using it.
1542 		 *
1543 		 * We wait for two passes. xattr directories in the first pass
1544 		 * may add xattr entries in zfs_purgedir, so in the second pass
1545 		 * we wait for them. We don't use taskq_wait here because it is
1546 		 * a pool wide taskq. Other mounted filesystems can constantly
1547 		 * do zrele_async and there's no guarantee when taskq will be
1548 		 * empty.
1549 		 */
1550 		taskq_wait_outstanding(dsl_pool_zrele_taskq(
1551 		    dmu_objset_pool(zfsvfs->z_os)), 0);
1552 		taskq_wait_outstanding(dsl_pool_zrele_taskq(
1553 		    dmu_objset_pool(zfsvfs->z_os)), 0);
1554 	}
1555 }
1556 
1557 /*
1558  * Called once all other unmount released tear down has occurred.
1559  * It is our responsibility to release any remaining infrastructure.
1560  */
1561 int
zfs_umount(struct super_block * sb)1562 zfs_umount(struct super_block *sb)
1563 {
1564 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1565 	objset_t *os;
1566 
1567 	if (zfsvfs->z_arc_prune != NULL)
1568 		arc_remove_prune_callback(zfsvfs->z_arc_prune);
1569 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1570 	os = zfsvfs->z_os;
1571 
1572 	/*
1573 	 * z_os will be NULL if there was an error in
1574 	 * attempting to reopen zfsvfs.
1575 	 */
1576 	if (os != NULL) {
1577 		/*
1578 		 * Unset the objset user_ptr.
1579 		 */
1580 		mutex_enter(&os->os_user_ptr_lock);
1581 		dmu_objset_set_user(os, NULL);
1582 		mutex_exit(&os->os_user_ptr_lock);
1583 
1584 		/*
1585 		 * Finally release the objset
1586 		 */
1587 		dmu_objset_disown(os, B_TRUE, zfsvfs);
1588 	}
1589 
1590 	zfsvfs_free(zfsvfs);
1591 	sb->s_fs_info = NULL;
1592 	return (0);
1593 }
1594 
1595 int
zfs_remount(struct super_block * sb,int * flags,zfs_mnt_t * zm)1596 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1597 {
1598 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1599 	vfs_t *vfsp;
1600 	boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1601 	int error;
1602 
1603 	if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1604 	    !(*flags & SB_RDONLY)) {
1605 		*flags |= SB_RDONLY;
1606 		return (EROFS);
1607 	}
1608 
1609 	error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1610 	if (error)
1611 		return (error);
1612 
1613 	if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1614 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1615 
1616 	zfs_unregister_callbacks(zfsvfs);
1617 	zfsvfs_vfs_free(zfsvfs->z_vfs);
1618 
1619 	vfsp->vfs_data = zfsvfs;
1620 	zfsvfs->z_vfs = vfsp;
1621 	if (!issnap)
1622 		(void) zfs_register_callbacks(vfsp);
1623 
1624 	return (error);
1625 }
1626 
1627 int
zfs_vget(struct super_block * sb,struct inode ** ipp,fid_t * fidp)1628 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1629 {
1630 	zfsvfs_t	*zfsvfs = sb->s_fs_info;
1631 	znode_t		*zp;
1632 	uint64_t	object = 0;
1633 	uint64_t	fid_gen = 0;
1634 	uint64_t	gen_mask;
1635 	uint64_t	zp_gen;
1636 	int		i, err;
1637 
1638 	*ipp = NULL;
1639 
1640 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1641 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
1642 
1643 		for (i = 0; i < sizeof (zfid->zf_object); i++)
1644 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1645 
1646 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
1647 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1648 	} else {
1649 		return (SET_ERROR(EINVAL));
1650 	}
1651 
1652 	/* LONG_FID_LEN means snapdirs */
1653 	if (fidp->fid_len == LONG_FID_LEN) {
1654 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
1655 		uint64_t	objsetid = 0;
1656 		uint64_t	setgen = 0;
1657 
1658 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1659 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1660 
1661 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1662 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1663 
1664 		if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1665 			dprintf("snapdir fid: objsetid (%llu) != "
1666 			    "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1667 			    objsetid, ZFSCTL_INO_SNAPDIRS, object);
1668 
1669 			return (SET_ERROR(EINVAL));
1670 		}
1671 
1672 		if (fid_gen > 1 || setgen != 0) {
1673 			dprintf("snapdir fid: fid_gen (%llu) and setgen "
1674 			    "(%llu)\n", fid_gen, setgen);
1675 			return (SET_ERROR(EINVAL));
1676 		}
1677 
1678 		return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1679 	}
1680 
1681 	if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1682 		return (err);
1683 	/* A zero fid_gen means we are in the .zfs control directories */
1684 	if (fid_gen == 0 &&
1685 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1686 		*ipp = zfsvfs->z_ctldir;
1687 		ASSERT(*ipp != NULL);
1688 		if (object == ZFSCTL_INO_SNAPDIR) {
1689 			VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1690 			    0, kcred, NULL, NULL) == 0);
1691 		} else {
1692 			/*
1693 			 * Must have an existing ref, so igrab()
1694 			 * cannot return NULL
1695 			 */
1696 			VERIFY3P(igrab(*ipp), !=, NULL);
1697 		}
1698 		zfs_exit(zfsvfs, FTAG);
1699 		return (0);
1700 	}
1701 
1702 	gen_mask = -1ULL >> (64 - 8 * i);
1703 
1704 	dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1705 	if ((err = zfs_zget(zfsvfs, object, &zp))) {
1706 		zfs_exit(zfsvfs, FTAG);
1707 		return (err);
1708 	}
1709 
1710 	/* Don't export xattr stuff */
1711 	if (zp->z_pflags & ZFS_XATTR) {
1712 		zrele(zp);
1713 		zfs_exit(zfsvfs, FTAG);
1714 		return (SET_ERROR(ENOENT));
1715 	}
1716 
1717 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1718 	    sizeof (uint64_t));
1719 	zp_gen = zp_gen & gen_mask;
1720 	if (zp_gen == 0)
1721 		zp_gen = 1;
1722 	if ((fid_gen == 0) && (zfsvfs->z_root == object))
1723 		fid_gen = zp_gen;
1724 	if (zp->z_unlinked || zp_gen != fid_gen) {
1725 		dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1726 		    fid_gen);
1727 		zrele(zp);
1728 		zfs_exit(zfsvfs, FTAG);
1729 		return (SET_ERROR(ENOENT));
1730 	}
1731 
1732 	*ipp = ZTOI(zp);
1733 	if (*ipp)
1734 		zfs_znode_update_vfs(ITOZ(*ipp));
1735 
1736 	zfs_exit(zfsvfs, FTAG);
1737 	return (0);
1738 }
1739 
1740 /*
1741  * Block out VFS ops and close zfsvfs_t
1742  *
1743  * Note, if successful, then we return with the 'z_teardown_lock' and
1744  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
1745  * dataset and objset intact so that they can be atomically handed off during
1746  * a subsequent rollback or recv operation and the resume thereafter.
1747  */
1748 int
zfs_suspend_fs(zfsvfs_t * zfsvfs)1749 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1750 {
1751 	int error;
1752 
1753 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1754 		return (error);
1755 
1756 	return (0);
1757 }
1758 
1759 /*
1760  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
1761  * is an invariant across any of the operations that can be performed while the
1762  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
1763  * are the same: the relevant objset and associated dataset are owned by
1764  * zfsvfs, held, and long held on entry.
1765  */
1766 int
zfs_resume_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)1767 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1768 {
1769 	int err, err2;
1770 	znode_t *zp;
1771 
1772 	ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1773 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1774 
1775 	/*
1776 	 * We already own this, so just update the objset_t, as the one we
1777 	 * had before may have been evicted.
1778 	 */
1779 	objset_t *os;
1780 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
1781 	VERIFY(dsl_dataset_long_held(ds));
1782 	dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1783 	dsl_pool_config_enter(dp, FTAG);
1784 	VERIFY0(dmu_objset_from_ds(ds, &os));
1785 	dsl_pool_config_exit(dp, FTAG);
1786 
1787 	err = zfsvfs_init(zfsvfs, os);
1788 	if (err != 0)
1789 		goto bail;
1790 
1791 	ds->ds_dir->dd_activity_cancelled = B_FALSE;
1792 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1793 
1794 	zfs_set_fuid_feature(zfsvfs);
1795 	zfsvfs->z_rollback_time = jiffies;
1796 
1797 	/*
1798 	 * Attempt to re-establish all the active inodes with their
1799 	 * dbufs.  If a zfs_rezget() fails, then we unhash the inode
1800 	 * and mark it stale.  This prevents a collision if a new
1801 	 * inode/object is created which must use the same inode
1802 	 * number.  The stale inode will be be released when the
1803 	 * VFS prunes the dentry holding the remaining references
1804 	 * on the stale inode.
1805 	 */
1806 	mutex_enter(&zfsvfs->z_znodes_lock);
1807 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1808 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1809 		err2 = zfs_rezget(zp);
1810 		if (err2) {
1811 			zpl_d_drop_aliases(ZTOI(zp));
1812 			remove_inode_hash(ZTOI(zp));
1813 		}
1814 
1815 		/* see comment in zfs_suspend_fs() */
1816 		if (zp->z_suspended) {
1817 			zfs_zrele_async(zp);
1818 			zp->z_suspended = B_FALSE;
1819 		}
1820 	}
1821 	mutex_exit(&zfsvfs->z_znodes_lock);
1822 
1823 	if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
1824 		/*
1825 		 * zfs_suspend_fs() could have interrupted freeing
1826 		 * of dnodes. We need to restart this freeing so
1827 		 * that we don't "leak" the space.
1828 		 */
1829 		zfs_unlinked_drain(zfsvfs);
1830 	}
1831 
1832 	/*
1833 	 * Most of the time zfs_suspend_fs is used for changing the contents
1834 	 * of the underlying dataset. ZFS rollback and receive operations
1835 	 * might create files for which negative dentries are present in
1836 	 * the cache. Since walking the dcache would require a lot of GPL-only
1837 	 * code duplication, it's much easier on these rather rare occasions
1838 	 * just to flush the whole dcache for the given dataset/filesystem.
1839 	 */
1840 	shrink_dcache_sb(zfsvfs->z_sb);
1841 
1842 bail:
1843 	if (err != 0)
1844 		zfsvfs->z_unmounted = B_TRUE;
1845 
1846 	/* release the VFS ops */
1847 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
1848 	ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1849 
1850 	if (err != 0) {
1851 		/*
1852 		 * Since we couldn't setup the sa framework, try to force
1853 		 * unmount this file system.
1854 		 */
1855 		if (zfsvfs->z_os)
1856 			(void) zfs_umount(zfsvfs->z_sb);
1857 	}
1858 	return (err);
1859 }
1860 
1861 /*
1862  * Release VOPs and unmount a suspended filesystem.
1863  */
1864 int
zfs_end_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)1865 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1866 {
1867 	ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1868 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1869 
1870 	/*
1871 	 * We already own this, so just hold and rele it to update the
1872 	 * objset_t, as the one we had before may have been evicted.
1873 	 */
1874 	objset_t *os;
1875 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
1876 	VERIFY(dsl_dataset_long_held(ds));
1877 	dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1878 	dsl_pool_config_enter(dp, FTAG);
1879 	VERIFY0(dmu_objset_from_ds(ds, &os));
1880 	dsl_pool_config_exit(dp, FTAG);
1881 	zfsvfs->z_os = os;
1882 
1883 	/* release the VOPs */
1884 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
1885 	ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1886 
1887 	/*
1888 	 * Try to force unmount this file system.
1889 	 */
1890 	(void) zfs_umount(zfsvfs->z_sb);
1891 	zfsvfs->z_unmounted = B_TRUE;
1892 	return (0);
1893 }
1894 
1895 /*
1896  * Automounted snapshots rely on periodic revalidation
1897  * to defer snapshots from being automatically unmounted.
1898  */
1899 
1900 inline void
zfs_exit_fs(zfsvfs_t * zfsvfs)1901 zfs_exit_fs(zfsvfs_t *zfsvfs)
1902 {
1903 	if (!zfsvfs->z_issnap)
1904 		return;
1905 
1906 	if (time_after(jiffies, zfsvfs->z_snap_defer_time +
1907 	    MAX(zfs_expire_snapshot * HZ / 2, HZ))) {
1908 		zfsvfs->z_snap_defer_time = jiffies;
1909 		zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa,
1910 		    dmu_objset_id(zfsvfs->z_os),
1911 		    zfs_expire_snapshot);
1912 	}
1913 }
1914 
1915 int
zfs_set_version(zfsvfs_t * zfsvfs,uint64_t newvers)1916 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1917 {
1918 	int error;
1919 	objset_t *os = zfsvfs->z_os;
1920 	dmu_tx_t *tx;
1921 
1922 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1923 		return (SET_ERROR(EINVAL));
1924 
1925 	if (newvers < zfsvfs->z_version)
1926 		return (SET_ERROR(EINVAL));
1927 
1928 	if (zfs_spa_version_map(newvers) >
1929 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
1930 		return (SET_ERROR(ENOTSUP));
1931 
1932 	tx = dmu_tx_create(os);
1933 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1934 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1935 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1936 		    ZFS_SA_ATTRS);
1937 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1938 	}
1939 	error = dmu_tx_assign(tx, TXG_WAIT);
1940 	if (error) {
1941 		dmu_tx_abort(tx);
1942 		return (error);
1943 	}
1944 
1945 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1946 	    8, 1, &newvers, tx);
1947 
1948 	if (error) {
1949 		dmu_tx_commit(tx);
1950 		return (error);
1951 	}
1952 
1953 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1954 		uint64_t sa_obj;
1955 
1956 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1957 		    SPA_VERSION_SA);
1958 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1959 		    DMU_OT_NONE, 0, tx);
1960 
1961 		error = zap_add(os, MASTER_NODE_OBJ,
1962 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1963 		ASSERT0(error);
1964 
1965 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
1966 		sa_register_update_callback(os, zfs_sa_upgrade);
1967 	}
1968 
1969 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
1970 	    "from %llu to %llu", zfsvfs->z_version, newvers);
1971 
1972 	dmu_tx_commit(tx);
1973 
1974 	zfsvfs->z_version = newvers;
1975 	os->os_version = newvers;
1976 
1977 	zfs_set_fuid_feature(zfsvfs);
1978 
1979 	return (0);
1980 }
1981 
1982 /*
1983  * Return true if the corresponding vfs's unmounted flag is set.
1984  * Otherwise return false.
1985  * If this function returns true we know VFS unmount has been initiated.
1986  */
1987 boolean_t
zfs_get_vfs_flag_unmounted(objset_t * os)1988 zfs_get_vfs_flag_unmounted(objset_t *os)
1989 {
1990 	zfsvfs_t *zfvp;
1991 	boolean_t unmounted = B_FALSE;
1992 
1993 	ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
1994 
1995 	mutex_enter(&os->os_user_ptr_lock);
1996 	zfvp = dmu_objset_get_user(os);
1997 	if (zfvp != NULL && zfvp->z_unmounted)
1998 		unmounted = B_TRUE;
1999 	mutex_exit(&os->os_user_ptr_lock);
2000 
2001 	return (unmounted);
2002 }
2003 
2004 void
zfsvfs_update_fromname(const char * oldname,const char * newname)2005 zfsvfs_update_fromname(const char *oldname, const char *newname)
2006 {
2007 	/*
2008 	 * We don't need to do anything here, the devname is always current by
2009 	 * virtue of zfsvfs->z_sb->s_op->show_devname.
2010 	 */
2011 	(void) oldname, (void) newname;
2012 }
2013 
2014 void
zfs_init(void)2015 zfs_init(void)
2016 {
2017 	zfsctl_init();
2018 	zfs_znode_init();
2019 	dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2020 	register_filesystem(&zpl_fs_type);
2021 }
2022 
2023 void
zfs_fini(void)2024 zfs_fini(void)
2025 {
2026 	/*
2027 	 * we don't use outstanding because zpl_posix_acl_free might add more.
2028 	 */
2029 	taskq_wait(system_delay_taskq);
2030 	taskq_wait(system_taskq);
2031 	unregister_filesystem(&zpl_fs_type);
2032 	zfs_znode_fini();
2033 	zfsctl_fini();
2034 }
2035 
2036 #if defined(_KERNEL)
2037 EXPORT_SYMBOL(zfs_suspend_fs);
2038 EXPORT_SYMBOL(zfs_resume_fs);
2039 EXPORT_SYMBOL(zfs_set_version);
2040 EXPORT_SYMBOL(zfsvfs_create);
2041 EXPORT_SYMBOL(zfsvfs_free);
2042 EXPORT_SYMBOL(zfs_is_readonly);
2043 EXPORT_SYMBOL(zfs_domount);
2044 EXPORT_SYMBOL(zfs_preumount);
2045 EXPORT_SYMBOL(zfs_umount);
2046 EXPORT_SYMBOL(zfs_remount);
2047 EXPORT_SYMBOL(zfs_statvfs);
2048 EXPORT_SYMBOL(zfs_vget);
2049 EXPORT_SYMBOL(zfs_prune);
2050 #endif
2051