xref: /freebsd/sys/contrib/openzfs/module/os/linux/zfs/zfs_vfsops.c (revision 87bf66d4a7488c496af110d4d05cc0273d49f82e)
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/zfeature.h>
61 #include <sys/zpl.h>
62 #include <linux/vfs_compat.h>
63 #include <linux/fs.h>
64 #include "zfs_comutil.h"
65 
66 enum {
67 	TOKEN_RO,
68 	TOKEN_RW,
69 	TOKEN_SETUID,
70 	TOKEN_NOSETUID,
71 	TOKEN_EXEC,
72 	TOKEN_NOEXEC,
73 	TOKEN_DEVICES,
74 	TOKEN_NODEVICES,
75 	TOKEN_DIRXATTR,
76 	TOKEN_SAXATTR,
77 	TOKEN_XATTR,
78 	TOKEN_NOXATTR,
79 	TOKEN_ATIME,
80 	TOKEN_NOATIME,
81 	TOKEN_RELATIME,
82 	TOKEN_NORELATIME,
83 	TOKEN_NBMAND,
84 	TOKEN_NONBMAND,
85 	TOKEN_MNTPOINT,
86 	TOKEN_LAST,
87 };
88 
89 static const match_table_t zpl_tokens = {
90 	{ TOKEN_RO,		MNTOPT_RO },
91 	{ TOKEN_RW,		MNTOPT_RW },
92 	{ TOKEN_SETUID,		MNTOPT_SETUID },
93 	{ TOKEN_NOSETUID,	MNTOPT_NOSETUID },
94 	{ TOKEN_EXEC,		MNTOPT_EXEC },
95 	{ TOKEN_NOEXEC,		MNTOPT_NOEXEC },
96 	{ TOKEN_DEVICES,	MNTOPT_DEVICES },
97 	{ TOKEN_NODEVICES,	MNTOPT_NODEVICES },
98 	{ TOKEN_DIRXATTR,	MNTOPT_DIRXATTR },
99 	{ TOKEN_SAXATTR,	MNTOPT_SAXATTR },
100 	{ TOKEN_XATTR,		MNTOPT_XATTR },
101 	{ TOKEN_NOXATTR,	MNTOPT_NOXATTR },
102 	{ TOKEN_ATIME,		MNTOPT_ATIME },
103 	{ TOKEN_NOATIME,	MNTOPT_NOATIME },
104 	{ TOKEN_RELATIME,	MNTOPT_RELATIME },
105 	{ TOKEN_NORELATIME,	MNTOPT_NORELATIME },
106 	{ TOKEN_NBMAND,		MNTOPT_NBMAND },
107 	{ TOKEN_NONBMAND,	MNTOPT_NONBMAND },
108 	{ TOKEN_MNTPOINT,	MNTOPT_MNTPOINT "=%s" },
109 	{ TOKEN_LAST,		NULL },
110 };
111 
112 static void
zfsvfs_vfs_free(vfs_t * vfsp)113 zfsvfs_vfs_free(vfs_t *vfsp)
114 {
115 	if (vfsp != NULL) {
116 		if (vfsp->vfs_mntpoint != NULL)
117 			kmem_strfree(vfsp->vfs_mntpoint);
118 		mutex_destroy(&vfsp->vfs_mntpt_lock);
119 		kmem_free(vfsp, sizeof (vfs_t));
120 	}
121 }
122 
123 static int
zfsvfs_parse_option(char * option,int token,substring_t * args,vfs_t * vfsp)124 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
125 {
126 	switch (token) {
127 	case TOKEN_RO:
128 		vfsp->vfs_readonly = B_TRUE;
129 		vfsp->vfs_do_readonly = B_TRUE;
130 		break;
131 	case TOKEN_RW:
132 		vfsp->vfs_readonly = B_FALSE;
133 		vfsp->vfs_do_readonly = B_TRUE;
134 		break;
135 	case TOKEN_SETUID:
136 		vfsp->vfs_setuid = B_TRUE;
137 		vfsp->vfs_do_setuid = B_TRUE;
138 		break;
139 	case TOKEN_NOSETUID:
140 		vfsp->vfs_setuid = B_FALSE;
141 		vfsp->vfs_do_setuid = B_TRUE;
142 		break;
143 	case TOKEN_EXEC:
144 		vfsp->vfs_exec = B_TRUE;
145 		vfsp->vfs_do_exec = B_TRUE;
146 		break;
147 	case TOKEN_NOEXEC:
148 		vfsp->vfs_exec = B_FALSE;
149 		vfsp->vfs_do_exec = B_TRUE;
150 		break;
151 	case TOKEN_DEVICES:
152 		vfsp->vfs_devices = B_TRUE;
153 		vfsp->vfs_do_devices = B_TRUE;
154 		break;
155 	case TOKEN_NODEVICES:
156 		vfsp->vfs_devices = B_FALSE;
157 		vfsp->vfs_do_devices = B_TRUE;
158 		break;
159 	case TOKEN_DIRXATTR:
160 		vfsp->vfs_xattr = ZFS_XATTR_DIR;
161 		vfsp->vfs_do_xattr = B_TRUE;
162 		break;
163 	case TOKEN_SAXATTR:
164 		vfsp->vfs_xattr = ZFS_XATTR_SA;
165 		vfsp->vfs_do_xattr = B_TRUE;
166 		break;
167 	case TOKEN_XATTR:
168 		vfsp->vfs_xattr = ZFS_XATTR_SA;
169 		vfsp->vfs_do_xattr = B_TRUE;
170 		break;
171 	case TOKEN_NOXATTR:
172 		vfsp->vfs_xattr = ZFS_XATTR_OFF;
173 		vfsp->vfs_do_xattr = B_TRUE;
174 		break;
175 	case TOKEN_ATIME:
176 		vfsp->vfs_atime = B_TRUE;
177 		vfsp->vfs_do_atime = B_TRUE;
178 		break;
179 	case TOKEN_NOATIME:
180 		vfsp->vfs_atime = B_FALSE;
181 		vfsp->vfs_do_atime = B_TRUE;
182 		break;
183 	case TOKEN_RELATIME:
184 		vfsp->vfs_relatime = B_TRUE;
185 		vfsp->vfs_do_relatime = B_TRUE;
186 		break;
187 	case TOKEN_NORELATIME:
188 		vfsp->vfs_relatime = B_FALSE;
189 		vfsp->vfs_do_relatime = B_TRUE;
190 		break;
191 	case TOKEN_NBMAND:
192 		vfsp->vfs_nbmand = B_TRUE;
193 		vfsp->vfs_do_nbmand = B_TRUE;
194 		break;
195 	case TOKEN_NONBMAND:
196 		vfsp->vfs_nbmand = B_FALSE;
197 		vfsp->vfs_do_nbmand = B_TRUE;
198 		break;
199 	case TOKEN_MNTPOINT:
200 		if (vfsp->vfs_mntpoint != NULL)
201 			kmem_strfree(vfsp->vfs_mntpoint);
202 		vfsp->vfs_mntpoint = match_strdup(&args[0]);
203 		if (vfsp->vfs_mntpoint == NULL)
204 			return (SET_ERROR(ENOMEM));
205 		break;
206 	default:
207 		break;
208 	}
209 
210 	return (0);
211 }
212 
213 /*
214  * Parse the raw mntopts and return a vfs_t describing the options.
215  */
216 static int
zfsvfs_parse_options(char * mntopts,vfs_t ** vfsp)217 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
218 {
219 	vfs_t *tmp_vfsp;
220 	int error;
221 
222 	tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
223 	mutex_init(&tmp_vfsp->vfs_mntpt_lock, NULL, MUTEX_DEFAULT, NULL);
224 
225 	if (mntopts != NULL) {
226 		substring_t args[MAX_OPT_ARGS];
227 		char *tmp_mntopts, *p, *t;
228 		int token;
229 
230 		tmp_mntopts = t = kmem_strdup(mntopts);
231 		if (tmp_mntopts == NULL)
232 			return (SET_ERROR(ENOMEM));
233 
234 		while ((p = strsep(&t, ",")) != NULL) {
235 			if (!*p)
236 				continue;
237 
238 			args[0].to = args[0].from = NULL;
239 			token = match_token(p, zpl_tokens, args);
240 			error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
241 			if (error) {
242 				kmem_strfree(tmp_mntopts);
243 				zfsvfs_vfs_free(tmp_vfsp);
244 				return (error);
245 			}
246 		}
247 
248 		kmem_strfree(tmp_mntopts);
249 	}
250 
251 	*vfsp = tmp_vfsp;
252 
253 	return (0);
254 }
255 
256 boolean_t
zfs_is_readonly(zfsvfs_t * zfsvfs)257 zfs_is_readonly(zfsvfs_t *zfsvfs)
258 {
259 	return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
260 }
261 
262 int
zfs_sync(struct super_block * sb,int wait,cred_t * cr)263 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
264 {
265 	(void) cr;
266 	zfsvfs_t *zfsvfs = sb->s_fs_info;
267 
268 	/*
269 	 * Semantically, the only requirement is that the sync be initiated.
270 	 * The DMU syncs out txgs frequently, so there's nothing to do.
271 	 */
272 	if (!wait)
273 		return (0);
274 
275 	if (zfsvfs != NULL) {
276 		/*
277 		 * Sync a specific filesystem.
278 		 */
279 		dsl_pool_t *dp;
280 		int error;
281 
282 		if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
283 			return (error);
284 		dp = dmu_objset_pool(zfsvfs->z_os);
285 
286 		/*
287 		 * If the system is shutting down, then skip any
288 		 * filesystems which may exist on a suspended pool.
289 		 */
290 		if (spa_suspended(dp->dp_spa)) {
291 			zfs_exit(zfsvfs, FTAG);
292 			return (0);
293 		}
294 
295 		if (zfsvfs->z_log != NULL)
296 			zil_commit(zfsvfs->z_log, 0);
297 
298 		zfs_exit(zfsvfs, FTAG);
299 	} else {
300 		/*
301 		 * Sync all ZFS filesystems.  This is what happens when you
302 		 * run sync(1).  Unlike other filesystems, ZFS honors the
303 		 * request by waiting for all pools to commit all dirty data.
304 		 */
305 		spa_sync_allpools();
306 	}
307 
308 	return (0);
309 }
310 
311 static void
atime_changed_cb(void * arg,uint64_t newval)312 atime_changed_cb(void *arg, uint64_t newval)
313 {
314 	zfsvfs_t *zfsvfs = arg;
315 	struct super_block *sb = zfsvfs->z_sb;
316 
317 	if (sb == NULL)
318 		return;
319 	/*
320 	 * Update SB_NOATIME bit in VFS super block.  Since atime update is
321 	 * determined by atime_needs_update(), atime_needs_update() needs to
322 	 * return false if atime is turned off, and not unconditionally return
323 	 * false if atime is turned on.
324 	 */
325 	if (newval)
326 		sb->s_flags &= ~SB_NOATIME;
327 	else
328 		sb->s_flags |= SB_NOATIME;
329 }
330 
331 static void
relatime_changed_cb(void * arg,uint64_t newval)332 relatime_changed_cb(void *arg, uint64_t newval)
333 {
334 	((zfsvfs_t *)arg)->z_relatime = newval;
335 }
336 
337 static void
xattr_changed_cb(void * arg,uint64_t newval)338 xattr_changed_cb(void *arg, uint64_t newval)
339 {
340 	zfsvfs_t *zfsvfs = arg;
341 
342 	if (newval == ZFS_XATTR_OFF) {
343 		zfsvfs->z_flags &= ~ZSB_XATTR;
344 	} else {
345 		zfsvfs->z_flags |= ZSB_XATTR;
346 
347 		if (newval == ZFS_XATTR_SA)
348 			zfsvfs->z_xattr_sa = B_TRUE;
349 		else
350 			zfsvfs->z_xattr_sa = B_FALSE;
351 	}
352 }
353 
354 static void
acltype_changed_cb(void * arg,uint64_t newval)355 acltype_changed_cb(void *arg, uint64_t newval)
356 {
357 	zfsvfs_t *zfsvfs = arg;
358 
359 	switch (newval) {
360 	case ZFS_ACLTYPE_NFSV4:
361 	case ZFS_ACLTYPE_OFF:
362 		zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
363 		zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
364 		break;
365 	case ZFS_ACLTYPE_POSIX:
366 #ifdef CONFIG_FS_POSIX_ACL
367 		zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX;
368 		zfsvfs->z_sb->s_flags |= SB_POSIXACL;
369 #else
370 		zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
371 		zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
372 #endif /* CONFIG_FS_POSIX_ACL */
373 		break;
374 	default:
375 		break;
376 	}
377 }
378 
379 static void
blksz_changed_cb(void * arg,uint64_t newval)380 blksz_changed_cb(void *arg, uint64_t newval)
381 {
382 	zfsvfs_t *zfsvfs = arg;
383 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
384 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
385 	ASSERT(ISP2(newval));
386 
387 	zfsvfs->z_max_blksz = newval;
388 }
389 
390 static void
readonly_changed_cb(void * arg,uint64_t newval)391 readonly_changed_cb(void *arg, uint64_t newval)
392 {
393 	zfsvfs_t *zfsvfs = arg;
394 	struct super_block *sb = zfsvfs->z_sb;
395 
396 	if (sb == NULL)
397 		return;
398 
399 	if (newval)
400 		sb->s_flags |= SB_RDONLY;
401 	else
402 		sb->s_flags &= ~SB_RDONLY;
403 }
404 
405 static void
devices_changed_cb(void * arg,uint64_t newval)406 devices_changed_cb(void *arg, uint64_t newval)
407 {
408 }
409 
410 static void
setuid_changed_cb(void * arg,uint64_t newval)411 setuid_changed_cb(void *arg, uint64_t newval)
412 {
413 }
414 
415 static void
exec_changed_cb(void * arg,uint64_t newval)416 exec_changed_cb(void *arg, uint64_t newval)
417 {
418 }
419 
420 static void
nbmand_changed_cb(void * arg,uint64_t newval)421 nbmand_changed_cb(void *arg, uint64_t newval)
422 {
423 	zfsvfs_t *zfsvfs = arg;
424 	struct super_block *sb = zfsvfs->z_sb;
425 
426 	if (sb == NULL)
427 		return;
428 
429 	if (newval == TRUE)
430 		sb->s_flags |= SB_MANDLOCK;
431 	else
432 		sb->s_flags &= ~SB_MANDLOCK;
433 }
434 
435 static void
snapdir_changed_cb(void * arg,uint64_t newval)436 snapdir_changed_cb(void *arg, uint64_t newval)
437 {
438 	((zfsvfs_t *)arg)->z_show_ctldir = newval;
439 }
440 
441 static void
acl_mode_changed_cb(void * arg,uint64_t newval)442 acl_mode_changed_cb(void *arg, uint64_t newval)
443 {
444 	zfsvfs_t *zfsvfs = arg;
445 
446 	zfsvfs->z_acl_mode = newval;
447 }
448 
449 static void
acl_inherit_changed_cb(void * arg,uint64_t newval)450 acl_inherit_changed_cb(void *arg, uint64_t newval)
451 {
452 	((zfsvfs_t *)arg)->z_acl_inherit = newval;
453 }
454 
455 static void
longname_changed_cb(void * arg,uint64_t newval)456 longname_changed_cb(void *arg, uint64_t newval)
457 {
458 	((zfsvfs_t *)arg)->z_longname = newval;
459 }
460 
461 static int
zfs_register_callbacks(vfs_t * vfsp)462 zfs_register_callbacks(vfs_t *vfsp)
463 {
464 	struct dsl_dataset *ds = NULL;
465 	objset_t *os = NULL;
466 	zfsvfs_t *zfsvfs = NULL;
467 	int error = 0;
468 
469 	ASSERT(vfsp);
470 	zfsvfs = vfsp->vfs_data;
471 	ASSERT(zfsvfs);
472 	os = zfsvfs->z_os;
473 
474 	/*
475 	 * The act of registering our callbacks will destroy any mount
476 	 * options we may have.  In order to enable temporary overrides
477 	 * of mount options, we stash away the current values and
478 	 * restore them after we register the callbacks.
479 	 */
480 	if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
481 		vfsp->vfs_do_readonly = B_TRUE;
482 		vfsp->vfs_readonly = B_TRUE;
483 	}
484 
485 	/*
486 	 * Register property callbacks.
487 	 *
488 	 * It would probably be fine to just check for i/o error from
489 	 * the first prop_register(), but I guess I like to go
490 	 * overboard...
491 	 */
492 	ds = dmu_objset_ds(os);
493 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
494 	error = dsl_prop_register(ds,
495 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
496 	error = error ? error : dsl_prop_register(ds,
497 	    zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
498 	error = error ? error : dsl_prop_register(ds,
499 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
500 	error = error ? error : dsl_prop_register(ds,
501 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
502 	error = error ? error : dsl_prop_register(ds,
503 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
504 	error = error ? error : dsl_prop_register(ds,
505 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
506 	error = error ? error : dsl_prop_register(ds,
507 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
508 	error = error ? error : dsl_prop_register(ds,
509 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
510 	error = error ? error : dsl_prop_register(ds,
511 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
512 	error = error ? error : dsl_prop_register(ds,
513 	    zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
514 	error = error ? error : dsl_prop_register(ds,
515 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
516 	error = error ? error : dsl_prop_register(ds,
517 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
518 	    zfsvfs);
519 	error = error ? error : dsl_prop_register(ds,
520 	    zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
521 	error = error ? error : dsl_prop_register(ds,
522 	    zfs_prop_to_name(ZFS_PROP_LONGNAME), longname_changed_cb, zfsvfs);
523 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
524 	if (error)
525 		goto unregister;
526 
527 	/*
528 	 * Invoke our callbacks to restore temporary mount options.
529 	 */
530 	if (vfsp->vfs_do_readonly)
531 		readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
532 	if (vfsp->vfs_do_setuid)
533 		setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
534 	if (vfsp->vfs_do_exec)
535 		exec_changed_cb(zfsvfs, vfsp->vfs_exec);
536 	if (vfsp->vfs_do_devices)
537 		devices_changed_cb(zfsvfs, vfsp->vfs_devices);
538 	if (vfsp->vfs_do_xattr)
539 		xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
540 	if (vfsp->vfs_do_atime)
541 		atime_changed_cb(zfsvfs, vfsp->vfs_atime);
542 	if (vfsp->vfs_do_relatime)
543 		relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
544 	if (vfsp->vfs_do_nbmand)
545 		nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
546 
547 	return (0);
548 
549 unregister:
550 	dsl_prop_unregister_all(ds, zfsvfs);
551 	return (error);
552 }
553 
554 /*
555  * Takes a dataset, a property, a value and that value's setpoint as
556  * found in the ZAP. Checks if the property has been changed in the vfs.
557  * If so, val and setpoint will be overwritten with updated content.
558  * Otherwise, they are left unchanged.
559  */
560 int
zfs_get_temporary_prop(dsl_dataset_t * ds,zfs_prop_t zfs_prop,uint64_t * val,char * setpoint)561 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
562     char *setpoint)
563 {
564 	int error;
565 	zfsvfs_t *zfvp;
566 	vfs_t *vfsp;
567 	objset_t *os;
568 	uint64_t tmp = *val;
569 
570 	error = dmu_objset_from_ds(ds, &os);
571 	if (error != 0)
572 		return (error);
573 
574 	if (dmu_objset_type(os) != DMU_OST_ZFS)
575 		return (EINVAL);
576 
577 	mutex_enter(&os->os_user_ptr_lock);
578 	zfvp = dmu_objset_get_user(os);
579 	mutex_exit(&os->os_user_ptr_lock);
580 	if (zfvp == NULL)
581 		return (ESRCH);
582 
583 	vfsp = zfvp->z_vfs;
584 
585 	switch (zfs_prop) {
586 	case ZFS_PROP_ATIME:
587 		if (vfsp->vfs_do_atime)
588 			tmp = vfsp->vfs_atime;
589 		break;
590 	case ZFS_PROP_RELATIME:
591 		if (vfsp->vfs_do_relatime)
592 			tmp = vfsp->vfs_relatime;
593 		break;
594 	case ZFS_PROP_DEVICES:
595 		if (vfsp->vfs_do_devices)
596 			tmp = vfsp->vfs_devices;
597 		break;
598 	case ZFS_PROP_EXEC:
599 		if (vfsp->vfs_do_exec)
600 			tmp = vfsp->vfs_exec;
601 		break;
602 	case ZFS_PROP_SETUID:
603 		if (vfsp->vfs_do_setuid)
604 			tmp = vfsp->vfs_setuid;
605 		break;
606 	case ZFS_PROP_READONLY:
607 		if (vfsp->vfs_do_readonly)
608 			tmp = vfsp->vfs_readonly;
609 		break;
610 	case ZFS_PROP_XATTR:
611 		if (vfsp->vfs_do_xattr)
612 			tmp = vfsp->vfs_xattr;
613 		break;
614 	case ZFS_PROP_NBMAND:
615 		if (vfsp->vfs_do_nbmand)
616 			tmp = vfsp->vfs_nbmand;
617 		break;
618 	default:
619 		return (ENOENT);
620 	}
621 
622 	if (tmp != *val) {
623 		if (setpoint)
624 			(void) strcpy(setpoint, "temporary");
625 		*val = tmp;
626 	}
627 	return (0);
628 }
629 
630 /*
631  * Associate this zfsvfs with the given objset, which must be owned.
632  * This will cache a bunch of on-disk state from the objset in the
633  * zfsvfs.
634  */
635 static int
zfsvfs_init(zfsvfs_t * zfsvfs,objset_t * os)636 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
637 {
638 	int error;
639 	uint64_t val;
640 
641 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
642 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
643 	zfsvfs->z_os = os;
644 
645 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
646 	if (error != 0)
647 		return (error);
648 	if (zfsvfs->z_version >
649 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
650 		(void) printk("Can't mount a version %lld file system "
651 		    "on a version %lld pool\n. Pool must be upgraded to mount "
652 		    "this file system.\n", (u_longlong_t)zfsvfs->z_version,
653 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
654 		return (SET_ERROR(ENOTSUP));
655 	}
656 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
657 	if (error != 0)
658 		return (error);
659 	zfsvfs->z_norm = (int)val;
660 
661 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
662 	if (error != 0)
663 		return (error);
664 	zfsvfs->z_utf8 = (val != 0);
665 
666 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
667 	if (error != 0)
668 		return (error);
669 	zfsvfs->z_case = (uint_t)val;
670 
671 	if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
672 		return (error);
673 	zfsvfs->z_acl_type = (uint_t)val;
674 
675 	/*
676 	 * Fold case on file systems that are always or sometimes case
677 	 * insensitive.
678 	 */
679 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
680 	    zfsvfs->z_case == ZFS_CASE_MIXED)
681 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
682 
683 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
684 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
685 
686 	uint64_t sa_obj = 0;
687 	if (zfsvfs->z_use_sa) {
688 		/* should either have both of these objects or none */
689 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
690 		    &sa_obj);
691 		if (error != 0)
692 			return (error);
693 
694 		error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
695 		if ((error == 0) && (val == ZFS_XATTR_SA))
696 			zfsvfs->z_xattr_sa = B_TRUE;
697 	}
698 
699 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
700 	    &zfsvfs->z_root);
701 	if (error != 0)
702 		return (error);
703 	ASSERT(zfsvfs->z_root != 0);
704 
705 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
706 	    &zfsvfs->z_unlinkedobj);
707 	if (error != 0)
708 		return (error);
709 
710 	error = zap_lookup(os, MASTER_NODE_OBJ,
711 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
712 	    8, 1, &zfsvfs->z_userquota_obj);
713 	if (error == ENOENT)
714 		zfsvfs->z_userquota_obj = 0;
715 	else if (error != 0)
716 		return (error);
717 
718 	error = zap_lookup(os, MASTER_NODE_OBJ,
719 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
720 	    8, 1, &zfsvfs->z_groupquota_obj);
721 	if (error == ENOENT)
722 		zfsvfs->z_groupquota_obj = 0;
723 	else if (error != 0)
724 		return (error);
725 
726 	error = zap_lookup(os, MASTER_NODE_OBJ,
727 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
728 	    8, 1, &zfsvfs->z_projectquota_obj);
729 	if (error == ENOENT)
730 		zfsvfs->z_projectquota_obj = 0;
731 	else if (error != 0)
732 		return (error);
733 
734 	error = zap_lookup(os, MASTER_NODE_OBJ,
735 	    zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
736 	    8, 1, &zfsvfs->z_userobjquota_obj);
737 	if (error == ENOENT)
738 		zfsvfs->z_userobjquota_obj = 0;
739 	else if (error != 0)
740 		return (error);
741 
742 	error = zap_lookup(os, MASTER_NODE_OBJ,
743 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
744 	    8, 1, &zfsvfs->z_groupobjquota_obj);
745 	if (error == ENOENT)
746 		zfsvfs->z_groupobjquota_obj = 0;
747 	else if (error != 0)
748 		return (error);
749 
750 	error = zap_lookup(os, MASTER_NODE_OBJ,
751 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
752 	    8, 1, &zfsvfs->z_projectobjquota_obj);
753 	if (error == ENOENT)
754 		zfsvfs->z_projectobjquota_obj = 0;
755 	else if (error != 0)
756 		return (error);
757 
758 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
759 	    &zfsvfs->z_fuid_obj);
760 	if (error == ENOENT)
761 		zfsvfs->z_fuid_obj = 0;
762 	else if (error != 0)
763 		return (error);
764 
765 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
766 	    &zfsvfs->z_shares_dir);
767 	if (error == ENOENT)
768 		zfsvfs->z_shares_dir = 0;
769 	else if (error != 0)
770 		return (error);
771 
772 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
773 	    &zfsvfs->z_attr_table);
774 	if (error != 0)
775 		return (error);
776 
777 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
778 		sa_register_update_callback(os, zfs_sa_upgrade);
779 
780 	return (0);
781 }
782 
783 int
zfsvfs_create(const char * osname,boolean_t readonly,zfsvfs_t ** zfvp)784 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
785 {
786 	objset_t *os;
787 	zfsvfs_t *zfsvfs;
788 	int error;
789 	boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
790 
791 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
792 
793 	error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
794 	if (error != 0) {
795 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
796 		return (error);
797 	}
798 
799 	error = zfsvfs_create_impl(zfvp, zfsvfs, os);
800 
801 	return (error);
802 }
803 
804 
805 /*
806  * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
807  * on a failure.  Do not pass in a statically allocated zfsvfs.
808  */
809 int
zfsvfs_create_impl(zfsvfs_t ** zfvp,zfsvfs_t * zfsvfs,objset_t * os)810 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
811 {
812 	int error;
813 
814 	zfsvfs->z_vfs = NULL;
815 	zfsvfs->z_sb = NULL;
816 	zfsvfs->z_parent = zfsvfs;
817 
818 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
819 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
820 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
821 	    offsetof(znode_t, z_link_node));
822 	ZFS_TEARDOWN_INIT(zfsvfs);
823 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
824 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
825 
826 	int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
827 	    ZFS_OBJ_MTX_MAX);
828 	zfsvfs->z_hold_size = size;
829 	zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
830 	    KM_SLEEP);
831 	zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
832 	for (int i = 0; i != size; i++) {
833 		avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
834 		    sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
835 		mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
836 	}
837 
838 	error = zfsvfs_init(zfsvfs, os);
839 	if (error != 0) {
840 		dmu_objset_disown(os, B_TRUE, zfsvfs);
841 		*zfvp = NULL;
842 		zfsvfs_free(zfsvfs);
843 		return (error);
844 	}
845 
846 	zfsvfs->z_drain_task = TASKQID_INVALID;
847 	zfsvfs->z_draining = B_FALSE;
848 	zfsvfs->z_drain_cancel = B_TRUE;
849 
850 	*zfvp = zfsvfs;
851 	return (0);
852 }
853 
854 static int
zfsvfs_setup(zfsvfs_t * zfsvfs,boolean_t mounting)855 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
856 {
857 	int error;
858 	boolean_t readonly = zfs_is_readonly(zfsvfs);
859 
860 	error = zfs_register_callbacks(zfsvfs->z_vfs);
861 	if (error)
862 		return (error);
863 
864 	/*
865 	 * If we are not mounting (ie: online recv), then we don't
866 	 * have to worry about replaying the log as we blocked all
867 	 * operations out since we closed the ZIL.
868 	 */
869 	if (mounting) {
870 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
871 		error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
872 		if (error)
873 			return (error);
874 		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
875 		    &zfsvfs->z_kstat.dk_zil_sums);
876 
877 		/*
878 		 * During replay we remove the read only flag to
879 		 * allow replays to succeed.
880 		 */
881 		if (readonly != 0) {
882 			readonly_changed_cb(zfsvfs, B_FALSE);
883 		} else {
884 			zap_stats_t zs;
885 			if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
886 			    &zs) == 0) {
887 				dataset_kstats_update_nunlinks_kstat(
888 				    &zfsvfs->z_kstat, zs.zs_num_entries);
889 				dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
890 				    "num_entries in unlinked set: %llu",
891 				    zs.zs_num_entries);
892 			}
893 			zfs_unlinked_drain(zfsvfs);
894 			dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
895 			dd->dd_activity_cancelled = B_FALSE;
896 		}
897 
898 		/*
899 		 * Parse and replay the intent log.
900 		 *
901 		 * Because of ziltest, this must be done after
902 		 * zfs_unlinked_drain().  (Further note: ziltest
903 		 * doesn't use readonly mounts, where
904 		 * zfs_unlinked_drain() isn't called.)  This is because
905 		 * ziltest causes spa_sync() to think it's committed,
906 		 * but actually it is not, so the intent log contains
907 		 * many txg's worth of changes.
908 		 *
909 		 * In particular, if object N is in the unlinked set in
910 		 * the last txg to actually sync, then it could be
911 		 * actually freed in a later txg and then reallocated
912 		 * in a yet later txg.  This would write a "create
913 		 * object N" record to the intent log.  Normally, this
914 		 * would be fine because the spa_sync() would have
915 		 * written out the fact that object N is free, before
916 		 * we could write the "create object N" intent log
917 		 * record.
918 		 *
919 		 * But when we are in ziltest mode, we advance the "open
920 		 * txg" without actually spa_sync()-ing the changes to
921 		 * disk.  So we would see that object N is still
922 		 * allocated and in the unlinked set, and there is an
923 		 * intent log record saying to allocate it.
924 		 */
925 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
926 			if (zil_replay_disable) {
927 				zil_destroy(zfsvfs->z_log, B_FALSE);
928 			} else {
929 				zfsvfs->z_replay = B_TRUE;
930 				zil_replay(zfsvfs->z_os, zfsvfs,
931 				    zfs_replay_vector);
932 				zfsvfs->z_replay = B_FALSE;
933 			}
934 		}
935 
936 		/* restore readonly bit */
937 		if (readonly != 0)
938 			readonly_changed_cb(zfsvfs, B_TRUE);
939 	} else {
940 		ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
941 		zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
942 		    &zfsvfs->z_kstat.dk_zil_sums);
943 	}
944 
945 	/*
946 	 * Set the objset user_ptr to track its zfsvfs.
947 	 */
948 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
949 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
950 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
951 
952 	return (0);
953 }
954 
955 void
zfsvfs_free(zfsvfs_t * zfsvfs)956 zfsvfs_free(zfsvfs_t *zfsvfs)
957 {
958 	int i, size = zfsvfs->z_hold_size;
959 
960 	zfs_fuid_destroy(zfsvfs);
961 
962 	mutex_destroy(&zfsvfs->z_znodes_lock);
963 	mutex_destroy(&zfsvfs->z_lock);
964 	list_destroy(&zfsvfs->z_all_znodes);
965 	ZFS_TEARDOWN_DESTROY(zfsvfs);
966 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
967 	rw_destroy(&zfsvfs->z_fuid_lock);
968 	for (i = 0; i != size; i++) {
969 		avl_destroy(&zfsvfs->z_hold_trees[i]);
970 		mutex_destroy(&zfsvfs->z_hold_locks[i]);
971 	}
972 	vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
973 	vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
974 	zfsvfs_vfs_free(zfsvfs->z_vfs);
975 	dataset_kstats_destroy(&zfsvfs->z_kstat);
976 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
977 }
978 
979 static void
zfs_set_fuid_feature(zfsvfs_t * zfsvfs)980 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
981 {
982 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
983 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
984 }
985 
986 static void
zfs_unregister_callbacks(zfsvfs_t * zfsvfs)987 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
988 {
989 	objset_t *os = zfsvfs->z_os;
990 
991 	if (!dmu_objset_is_snapshot(os))
992 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
993 }
994 
995 #ifdef HAVE_MLSLABEL
996 /*
997  * Check that the hex label string is appropriate for the dataset being
998  * mounted into the global_zone proper.
999  *
1000  * Return an error if the hex label string is not default or
1001  * admin_low/admin_high.  For admin_low labels, the corresponding
1002  * dataset must be readonly.
1003  */
1004 int
zfs_check_global_label(const char * dsname,const char * hexsl)1005 zfs_check_global_label(const char *dsname, const char *hexsl)
1006 {
1007 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1008 		return (0);
1009 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1010 		return (0);
1011 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1012 		/* must be readonly */
1013 		uint64_t rdonly;
1014 
1015 		if (dsl_prop_get_integer(dsname,
1016 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1017 			return (SET_ERROR(EACCES));
1018 		return (rdonly ? 0 : SET_ERROR(EACCES));
1019 	}
1020 	return (SET_ERROR(EACCES));
1021 }
1022 #endif /* HAVE_MLSLABEL */
1023 
1024 static int
zfs_statfs_project(zfsvfs_t * zfsvfs,znode_t * zp,struct kstatfs * statp,uint32_t bshift)1025 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1026     uint32_t bshift)
1027 {
1028 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1029 	uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1030 	uint64_t quota;
1031 	uint64_t used;
1032 	int err;
1033 
1034 	strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1035 	err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset,
1036 	    sizeof (buf) - offset, B_FALSE);
1037 	if (err)
1038 		return (err);
1039 
1040 	if (zfsvfs->z_projectquota_obj == 0)
1041 		goto objs;
1042 
1043 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1044 	    buf + offset, 8, 1, &quota);
1045 	if (err == ENOENT)
1046 		goto objs;
1047 	else if (err)
1048 		return (err);
1049 
1050 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1051 	    buf + offset, 8, 1, &used);
1052 	if (unlikely(err == ENOENT)) {
1053 		uint32_t blksize;
1054 		u_longlong_t nblocks;
1055 
1056 		/*
1057 		 * Quota accounting is async, so it is possible race case.
1058 		 * There is at least one object with the given project ID.
1059 		 */
1060 		sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1061 		if (unlikely(zp->z_blksz == 0))
1062 			blksize = zfsvfs->z_max_blksz;
1063 
1064 		used = blksize * nblocks;
1065 	} else if (err) {
1066 		return (err);
1067 	}
1068 
1069 	statp->f_blocks = quota >> bshift;
1070 	statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1071 	statp->f_bavail = statp->f_bfree;
1072 
1073 objs:
1074 	if (zfsvfs->z_projectobjquota_obj == 0)
1075 		return (0);
1076 
1077 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1078 	    buf + offset, 8, 1, &quota);
1079 	if (err == ENOENT)
1080 		return (0);
1081 	else if (err)
1082 		return (err);
1083 
1084 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1085 	    buf, 8, 1, &used);
1086 	if (unlikely(err == ENOENT)) {
1087 		/*
1088 		 * Quota accounting is async, so it is possible race case.
1089 		 * There is at least one object with the given project ID.
1090 		 */
1091 		used = 1;
1092 	} else if (err) {
1093 		return (err);
1094 	}
1095 
1096 	statp->f_files = quota;
1097 	statp->f_ffree = (quota > used) ? (quota - used) : 0;
1098 
1099 	return (0);
1100 }
1101 
1102 int
zfs_statvfs(struct inode * ip,struct kstatfs * statp)1103 zfs_statvfs(struct inode *ip, struct kstatfs *statp)
1104 {
1105 	zfsvfs_t *zfsvfs = ITOZSB(ip);
1106 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1107 	int err = 0;
1108 
1109 	if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1110 		return (err);
1111 
1112 	dmu_objset_space(zfsvfs->z_os,
1113 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1114 
1115 	uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1116 	/*
1117 	 * The underlying storage pool actually uses multiple block
1118 	 * size.  Under Solaris frsize (fragment size) is reported as
1119 	 * the smallest block size we support, and bsize (block size)
1120 	 * as the filesystem's maximum block size.  Unfortunately,
1121 	 * under Linux the fragment size and block size are often used
1122 	 * interchangeably.  Thus we are forced to report both of them
1123 	 * as the filesystem's maximum block size.
1124 	 */
1125 	statp->f_frsize = zfsvfs->z_max_blksz;
1126 	statp->f_bsize = zfsvfs->z_max_blksz;
1127 	uint32_t bshift = fls(statp->f_bsize) - 1;
1128 
1129 	/*
1130 	 * The following report "total" blocks of various kinds in
1131 	 * the file system, but reported in terms of f_bsize - the
1132 	 * "preferred" size.
1133 	 */
1134 
1135 	/* Round up so we never have a filesystem using 0 blocks. */
1136 	refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1137 	statp->f_blocks = (refdbytes + availbytes) >> bshift;
1138 	statp->f_bfree = availbytes >> bshift;
1139 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1140 
1141 	/*
1142 	 * statvfs() should really be called statufs(), because it assumes
1143 	 * static metadata.  ZFS doesn't preallocate files, so the best
1144 	 * we can do is report the max that could possibly fit in f_files,
1145 	 * and that minus the number actually used in f_ffree.
1146 	 * For f_ffree, report the smaller of the number of objects available
1147 	 * and the number of blocks (each object will take at least a block).
1148 	 */
1149 	statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1150 	statp->f_files = statp->f_ffree + usedobjs;
1151 	statp->f_fsid.val[0] = (uint32_t)fsid;
1152 	statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1153 	statp->f_type = ZFS_SUPER_MAGIC;
1154 	statp->f_namelen =
1155 	    zfsvfs->z_longname ? (ZAP_MAXNAMELEN_NEW - 1) : (MAXNAMELEN - 1);
1156 
1157 	/*
1158 	 * We have all of 40 characters to stuff a string here.
1159 	 * Is there anything useful we could/should provide?
1160 	 */
1161 	memset(statp->f_spare, 0, sizeof (statp->f_spare));
1162 
1163 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1164 	    dmu_objset_projectquota_present(zfsvfs->z_os)) {
1165 		znode_t *zp = ITOZ(ip);
1166 
1167 		if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1168 		    zpl_is_valid_projid(zp->z_projid))
1169 			err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1170 	}
1171 
1172 	zfs_exit(zfsvfs, FTAG);
1173 	return (err);
1174 }
1175 
1176 static int
zfs_root(zfsvfs_t * zfsvfs,struct inode ** ipp)1177 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1178 {
1179 	znode_t *rootzp;
1180 	int error;
1181 
1182 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1183 		return (error);
1184 
1185 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1186 	if (error == 0)
1187 		*ipp = ZTOI(rootzp);
1188 
1189 	zfs_exit(zfsvfs, FTAG);
1190 	return (error);
1191 }
1192 
1193 /*
1194  * The ARC has requested that the filesystem drop entries from the dentry
1195  * and inode caches.  This can occur when the ARC needs to free meta data
1196  * blocks but can't because they are all pinned by entries in these caches.
1197  */
1198 #if defined(HAVE_SUPER_BLOCK_S_SHRINK)
1199 #define	S_SHRINK(sb)	(&(sb)->s_shrink)
1200 #elif defined(HAVE_SUPER_BLOCK_S_SHRINK_PTR)
1201 #define	S_SHRINK(sb)	((sb)->s_shrink)
1202 #endif
1203 
1204 int
zfs_prune(struct super_block * sb,unsigned long nr_to_scan,int * objects)1205 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1206 {
1207 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1208 	int error = 0;
1209 	struct shrinker *shrinker = S_SHRINK(sb);
1210 	struct shrink_control sc = {
1211 		.nr_to_scan = nr_to_scan,
1212 		.gfp_mask = GFP_KERNEL,
1213 	};
1214 
1215 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1216 		return (error);
1217 
1218 #ifdef SHRINKER_NUMA_AWARE
1219 	if (shrinker->flags & SHRINKER_NUMA_AWARE) {
1220 		long tc = 1;
1221 		for_each_online_node(sc.nid) {
1222 			long c = shrinker->count_objects(shrinker, &sc);
1223 			if (c  == 0 || c == SHRINK_EMPTY)
1224 				continue;
1225 			tc += c;
1226 		}
1227 		*objects = 0;
1228 		for_each_online_node(sc.nid) {
1229 			long c = shrinker->count_objects(shrinker, &sc);
1230 			if (c  == 0 || c == SHRINK_EMPTY)
1231 				continue;
1232 			if (c > tc)
1233 				tc = c;
1234 			sc.nr_to_scan = mult_frac(nr_to_scan, c, tc) + 1;
1235 			*objects += (*shrinker->scan_objects)(shrinker, &sc);
1236 		}
1237 	} else {
1238 			*objects = (*shrinker->scan_objects)(shrinker, &sc);
1239 	}
1240 #else
1241 	*objects = (*shrinker->scan_objects)(shrinker, &sc);
1242 #endif
1243 
1244 	zfs_exit(zfsvfs, FTAG);
1245 
1246 	dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1247 	    "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1248 	    nr_to_scan, *objects, error);
1249 
1250 	return (error);
1251 }
1252 
1253 /*
1254  * Teardown the zfsvfs_t.
1255  *
1256  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1257  * and 'z_teardown_inactive_lock' held.
1258  */
1259 static int
zfsvfs_teardown(zfsvfs_t * zfsvfs,boolean_t unmounting)1260 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1261 {
1262 	znode_t	*zp;
1263 
1264 	zfs_unlinked_drain_stop_wait(zfsvfs);
1265 
1266 	/*
1267 	 * If someone has not already unmounted this file system,
1268 	 * drain the zrele_taskq to ensure all active references to the
1269 	 * zfsvfs_t have been handled only then can it be safely destroyed.
1270 	 */
1271 	if (zfsvfs->z_os) {
1272 		/*
1273 		 * If we're unmounting we have to wait for the list to
1274 		 * drain completely.
1275 		 *
1276 		 * If we're not unmounting there's no guarantee the list
1277 		 * will drain completely, but iputs run from the taskq
1278 		 * may add the parents of dir-based xattrs to the taskq
1279 		 * so we want to wait for these.
1280 		 *
1281 		 * We can safely check z_all_znodes for being empty because the
1282 		 * VFS has already blocked operations which add to it.
1283 		 */
1284 		int round = 0;
1285 		while (!list_is_empty(&zfsvfs->z_all_znodes)) {
1286 			taskq_wait_outstanding(dsl_pool_zrele_taskq(
1287 			    dmu_objset_pool(zfsvfs->z_os)), 0);
1288 			if (++round > 1 && !unmounting)
1289 				break;
1290 		}
1291 	}
1292 
1293 	ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1294 
1295 	if (!unmounting) {
1296 		/*
1297 		 * We purge the parent filesystem's super block as the
1298 		 * parent filesystem and all of its snapshots have their
1299 		 * inode's super block set to the parent's filesystem's
1300 		 * super block.  Note,  'z_parent' is self referential
1301 		 * for non-snapshots.
1302 		 */
1303 		shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1304 	}
1305 
1306 	/*
1307 	 * Close the zil. NB: Can't close the zil while zfs_inactive
1308 	 * threads are blocked as zil_close can call zfs_inactive.
1309 	 */
1310 	if (zfsvfs->z_log) {
1311 		zil_close(zfsvfs->z_log);
1312 		zfsvfs->z_log = NULL;
1313 	}
1314 
1315 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1316 
1317 	/*
1318 	 * If we are not unmounting (ie: online recv) and someone already
1319 	 * unmounted this file system while we were doing the switcheroo,
1320 	 * or a reopen of z_os failed then just bail out now.
1321 	 */
1322 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1323 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1324 		ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1325 		return (SET_ERROR(EIO));
1326 	}
1327 
1328 	/*
1329 	 * At this point there are no VFS ops active, and any new VFS ops
1330 	 * will fail with EIO since we have z_teardown_lock for writer (only
1331 	 * relevant for forced unmount).
1332 	 *
1333 	 * Release all holds on dbufs. We also grab an extra reference to all
1334 	 * the remaining inodes so that the kernel does not attempt to free
1335 	 * any inodes of a suspended fs. This can cause deadlocks since the
1336 	 * zfs_resume_fs() process may involve starting threads, which might
1337 	 * attempt to free unreferenced inodes to free up memory for the new
1338 	 * thread.
1339 	 */
1340 	if (!unmounting) {
1341 		mutex_enter(&zfsvfs->z_znodes_lock);
1342 		for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1343 		    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1344 			if (zp->z_sa_hdl)
1345 				zfs_znode_dmu_fini(zp);
1346 			if (igrab(ZTOI(zp)) != NULL)
1347 				zp->z_suspended = B_TRUE;
1348 
1349 		}
1350 		mutex_exit(&zfsvfs->z_znodes_lock);
1351 	}
1352 
1353 	/*
1354 	 * If we are unmounting, set the unmounted flag and let new VFS ops
1355 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
1356 	 * other VFS ops will fail with EIO.
1357 	 */
1358 	if (unmounting) {
1359 		zfsvfs->z_unmounted = B_TRUE;
1360 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1361 		ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1362 	}
1363 
1364 	/*
1365 	 * z_os will be NULL if there was an error in attempting to reopen
1366 	 * zfsvfs, so just return as the properties had already been
1367 	 *
1368 	 * unregistered and cached data had been evicted before.
1369 	 */
1370 	if (zfsvfs->z_os == NULL)
1371 		return (0);
1372 
1373 	/*
1374 	 * Unregister properties.
1375 	 */
1376 	zfs_unregister_callbacks(zfsvfs);
1377 
1378 	/*
1379 	 * Evict cached data. We must write out any dirty data before
1380 	 * disowning the dataset.
1381 	 */
1382 	objset_t *os = zfsvfs->z_os;
1383 	boolean_t os_dirty = B_FALSE;
1384 	for (int t = 0; t < TXG_SIZE; t++) {
1385 		if (dmu_objset_is_dirty(os, t)) {
1386 			os_dirty = B_TRUE;
1387 			break;
1388 		}
1389 	}
1390 	if (!zfs_is_readonly(zfsvfs) && os_dirty) {
1391 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1392 	}
1393 	dmu_objset_evict_dbufs(zfsvfs->z_os);
1394 	dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1395 	dsl_dir_cancel_waiters(dd);
1396 
1397 	return (0);
1398 }
1399 
1400 static atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1401 
1402 int
zfs_domount(struct super_block * sb,zfs_mnt_t * zm,int silent)1403 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1404 {
1405 	const char *osname = zm->mnt_osname;
1406 	struct inode *root_inode = NULL;
1407 	uint64_t recordsize;
1408 	int error = 0;
1409 	zfsvfs_t *zfsvfs = NULL;
1410 	vfs_t *vfs = NULL;
1411 	int canwrite;
1412 	int dataset_visible_zone;
1413 
1414 	ASSERT(zm);
1415 	ASSERT(osname);
1416 
1417 	dataset_visible_zone = zone_dataset_visible(osname, &canwrite);
1418 
1419 	/*
1420 	 * Refuse to mount a filesystem if we are in a namespace and the
1421 	 * dataset is not visible or writable in that namespace.
1422 	 */
1423 	if (!INGLOBALZONE(curproc) &&
1424 	    (!dataset_visible_zone || !canwrite)) {
1425 		return (SET_ERROR(EPERM));
1426 	}
1427 
1428 	error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1429 	if (error)
1430 		return (error);
1431 
1432 	/*
1433 	 * If a non-writable filesystem is being mounted without the
1434 	 * read-only flag, pretend it was set, as done for snapshots.
1435 	 */
1436 	if (!canwrite)
1437 		vfs->vfs_readonly = B_TRUE;
1438 
1439 	error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1440 	if (error) {
1441 		zfsvfs_vfs_free(vfs);
1442 		goto out;
1443 	}
1444 
1445 	if ((error = dsl_prop_get_integer(osname, "recordsize",
1446 	    &recordsize, NULL))) {
1447 		zfsvfs_vfs_free(vfs);
1448 		goto out;
1449 	}
1450 
1451 	vfs->vfs_data = zfsvfs;
1452 	zfsvfs->z_vfs = vfs;
1453 	zfsvfs->z_sb = sb;
1454 	sb->s_fs_info = zfsvfs;
1455 	sb->s_magic = ZFS_SUPER_MAGIC;
1456 	sb->s_maxbytes = MAX_LFS_FILESIZE;
1457 	sb->s_time_gran = 1;
1458 	sb->s_blocksize = recordsize;
1459 	sb->s_blocksize_bits = ilog2(recordsize);
1460 
1461 	error = -super_setup_bdi_name(sb, "%.28s-%ld", "zfs",
1462 	    atomic_long_inc_return(&zfs_bdi_seq));
1463 	if (error)
1464 		goto out;
1465 
1466 	sb->s_bdi->ra_pages = 0;
1467 
1468 	/* Set callback operations for the file system. */
1469 	sb->s_op = &zpl_super_operations;
1470 	sb->s_xattr = zpl_xattr_handlers;
1471 	sb->s_export_op = &zpl_export_operations;
1472 
1473 	/* Set features for file system. */
1474 	zfs_set_fuid_feature(zfsvfs);
1475 
1476 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1477 		uint64_t pval;
1478 
1479 		atime_changed_cb(zfsvfs, B_FALSE);
1480 		readonly_changed_cb(zfsvfs, B_TRUE);
1481 		if ((error = dsl_prop_get_integer(osname,
1482 		    "xattr", &pval, NULL)))
1483 			goto out;
1484 		xattr_changed_cb(zfsvfs, pval);
1485 		if ((error = dsl_prop_get_integer(osname,
1486 		    "acltype", &pval, NULL)))
1487 			goto out;
1488 		acltype_changed_cb(zfsvfs, pval);
1489 		zfsvfs->z_issnap = B_TRUE;
1490 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1491 		zfsvfs->z_snap_defer_time = jiffies;
1492 
1493 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1494 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1495 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1496 	} else {
1497 		if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1498 			goto out;
1499 	}
1500 
1501 	/* Allocate a root inode for the filesystem. */
1502 	error = zfs_root(zfsvfs, &root_inode);
1503 	if (error) {
1504 		(void) zfs_umount(sb);
1505 		zfsvfs = NULL; /* avoid double-free; first in zfs_umount */
1506 		goto out;
1507 	}
1508 
1509 	/* Allocate a root dentry for the filesystem */
1510 	sb->s_root = d_make_root(root_inode);
1511 	if (sb->s_root == NULL) {
1512 		(void) zfs_umount(sb);
1513 		zfsvfs = NULL; /* avoid double-free; first in zfs_umount */
1514 		error = SET_ERROR(ENOMEM);
1515 		goto out;
1516 	}
1517 
1518 	if (!zfsvfs->z_issnap)
1519 		zfsctl_create(zfsvfs);
1520 
1521 	zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1522 out:
1523 	if (error) {
1524 		if (zfsvfs != NULL) {
1525 			dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1526 			zfsvfs_free(zfsvfs);
1527 		}
1528 		/*
1529 		 * make sure we don't have dangling sb->s_fs_info which
1530 		 * zfs_preumount will use.
1531 		 */
1532 		sb->s_fs_info = NULL;
1533 	}
1534 
1535 	return (error);
1536 }
1537 
1538 /*
1539  * Called when an unmount is requested and certain sanity checks have
1540  * already passed.  At this point no dentries or inodes have been reclaimed
1541  * from their respective caches.  We drop the extra reference on the .zfs
1542  * control directory to allow everything to be reclaimed.  All snapshots
1543  * must already have been unmounted to reach this point.
1544  */
1545 void
zfs_preumount(struct super_block * sb)1546 zfs_preumount(struct super_block *sb)
1547 {
1548 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1549 
1550 	/* zfsvfs is NULL when zfs_domount fails during mount */
1551 	if (zfsvfs) {
1552 		zfs_unlinked_drain_stop_wait(zfsvfs);
1553 		zfsctl_destroy(sb->s_fs_info);
1554 		/*
1555 		 * Wait for zrele_async before entering evict_inodes in
1556 		 * generic_shutdown_super. The reason we must finish before
1557 		 * evict_inodes is when lazytime is on, or when zfs_purgedir
1558 		 * calls zfs_zget, zrele would bump i_count from 0 to 1. This
1559 		 * would race with the i_count check in evict_inodes. This means
1560 		 * it could destroy the inode while we are still using it.
1561 		 *
1562 		 * We wait for two passes. xattr directories in the first pass
1563 		 * may add xattr entries in zfs_purgedir, so in the second pass
1564 		 * we wait for them. We don't use taskq_wait here because it is
1565 		 * a pool wide taskq. Other mounted filesystems can constantly
1566 		 * do zrele_async and there's no guarantee when taskq will be
1567 		 * empty.
1568 		 */
1569 		taskq_wait_outstanding(dsl_pool_zrele_taskq(
1570 		    dmu_objset_pool(zfsvfs->z_os)), 0);
1571 		taskq_wait_outstanding(dsl_pool_zrele_taskq(
1572 		    dmu_objset_pool(zfsvfs->z_os)), 0);
1573 	}
1574 }
1575 
1576 /*
1577  * Called once all other unmount released tear down has occurred.
1578  * It is our responsibility to release any remaining infrastructure.
1579  */
1580 int
zfs_umount(struct super_block * sb)1581 zfs_umount(struct super_block *sb)
1582 {
1583 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1584 	objset_t *os;
1585 
1586 	if (zfsvfs->z_arc_prune != NULL)
1587 		arc_remove_prune_callback(zfsvfs->z_arc_prune);
1588 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1589 	os = zfsvfs->z_os;
1590 
1591 	/*
1592 	 * z_os will be NULL if there was an error in
1593 	 * attempting to reopen zfsvfs.
1594 	 */
1595 	if (os != NULL) {
1596 		/*
1597 		 * Unset the objset user_ptr.
1598 		 */
1599 		mutex_enter(&os->os_user_ptr_lock);
1600 		dmu_objset_set_user(os, NULL);
1601 		mutex_exit(&os->os_user_ptr_lock);
1602 
1603 		/*
1604 		 * Finally release the objset
1605 		 */
1606 		dmu_objset_disown(os, B_TRUE, zfsvfs);
1607 	}
1608 
1609 	zfsvfs_free(zfsvfs);
1610 	sb->s_fs_info = NULL;
1611 	return (0);
1612 }
1613 
1614 int
zfs_remount(struct super_block * sb,int * flags,zfs_mnt_t * zm)1615 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1616 {
1617 	zfsvfs_t *zfsvfs = sb->s_fs_info;
1618 	vfs_t *vfsp;
1619 	boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1620 	int error;
1621 
1622 	if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1623 	    !(*flags & SB_RDONLY)) {
1624 		*flags |= SB_RDONLY;
1625 		return (EROFS);
1626 	}
1627 
1628 	error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1629 	if (error)
1630 		return (error);
1631 
1632 	if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1633 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1634 
1635 	zfs_unregister_callbacks(zfsvfs);
1636 	zfsvfs_vfs_free(zfsvfs->z_vfs);
1637 
1638 	vfsp->vfs_data = zfsvfs;
1639 	zfsvfs->z_vfs = vfsp;
1640 	if (!issnap)
1641 		(void) zfs_register_callbacks(vfsp);
1642 
1643 	return (error);
1644 }
1645 
1646 int
zfs_vget(struct super_block * sb,struct inode ** ipp,fid_t * fidp)1647 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1648 {
1649 	zfsvfs_t	*zfsvfs = sb->s_fs_info;
1650 	znode_t		*zp;
1651 	uint64_t	object = 0;
1652 	uint64_t	fid_gen = 0;
1653 	uint64_t	gen_mask;
1654 	uint64_t	zp_gen;
1655 	int		i, err;
1656 
1657 	*ipp = NULL;
1658 
1659 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1660 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
1661 
1662 		for (i = 0; i < sizeof (zfid->zf_object); i++)
1663 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1664 
1665 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
1666 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1667 	} else {
1668 		return (SET_ERROR(EINVAL));
1669 	}
1670 
1671 	/* LONG_FID_LEN means snapdirs */
1672 	if (fidp->fid_len == LONG_FID_LEN) {
1673 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
1674 		uint64_t	objsetid = 0;
1675 		uint64_t	setgen = 0;
1676 
1677 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1678 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1679 
1680 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1681 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1682 
1683 		if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1684 			dprintf("snapdir fid: objsetid (%llu) != "
1685 			    "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1686 			    objsetid, ZFSCTL_INO_SNAPDIRS, object);
1687 
1688 			return (SET_ERROR(EINVAL));
1689 		}
1690 
1691 		if (fid_gen > 1 || setgen != 0) {
1692 			dprintf("snapdir fid: fid_gen (%llu) and setgen "
1693 			    "(%llu)\n", fid_gen, setgen);
1694 			return (SET_ERROR(EINVAL));
1695 		}
1696 
1697 		return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1698 	}
1699 
1700 	if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1701 		return (err);
1702 	/* A zero fid_gen means we are in the .zfs control directories */
1703 	if (fid_gen == 0 &&
1704 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1705 		*ipp = zfsvfs->z_ctldir;
1706 		ASSERT(*ipp != NULL);
1707 
1708 		if (zfsvfs->z_show_ctldir == ZFS_SNAPDIR_DISABLED) {
1709 			return (SET_ERROR(ENOENT));
1710 		}
1711 
1712 		if (object == ZFSCTL_INO_SNAPDIR) {
1713 			VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1714 			    0, kcred, NULL, NULL) == 0);
1715 		} else {
1716 			/*
1717 			 * Must have an existing ref, so igrab()
1718 			 * cannot return NULL
1719 			 */
1720 			VERIFY3P(igrab(*ipp), !=, NULL);
1721 		}
1722 		zfs_exit(zfsvfs, FTAG);
1723 		return (0);
1724 	}
1725 
1726 	gen_mask = -1ULL >> (64 - 8 * i);
1727 
1728 	dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1729 	if ((err = zfs_zget(zfsvfs, object, &zp))) {
1730 		zfs_exit(zfsvfs, FTAG);
1731 		return (err);
1732 	}
1733 
1734 	/* Don't export xattr stuff */
1735 	if (zp->z_pflags & ZFS_XATTR) {
1736 		zrele(zp);
1737 		zfs_exit(zfsvfs, FTAG);
1738 		return (SET_ERROR(ENOENT));
1739 	}
1740 
1741 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1742 	    sizeof (uint64_t));
1743 	zp_gen = zp_gen & gen_mask;
1744 	if (zp_gen == 0)
1745 		zp_gen = 1;
1746 	if ((fid_gen == 0) && (zfsvfs->z_root == object))
1747 		fid_gen = zp_gen;
1748 	if (zp->z_unlinked || zp_gen != fid_gen) {
1749 		dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1750 		    fid_gen);
1751 		zrele(zp);
1752 		zfs_exit(zfsvfs, FTAG);
1753 		return (SET_ERROR(ENOENT));
1754 	}
1755 
1756 	*ipp = ZTOI(zp);
1757 	if (*ipp)
1758 		zfs_znode_update_vfs(ITOZ(*ipp));
1759 
1760 	zfs_exit(zfsvfs, FTAG);
1761 	return (0);
1762 }
1763 
1764 /*
1765  * Block out VFS ops and close zfsvfs_t
1766  *
1767  * Note, if successful, then we return with the 'z_teardown_lock' and
1768  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
1769  * dataset and objset intact so that they can be atomically handed off during
1770  * a subsequent rollback or recv operation and the resume thereafter.
1771  */
1772 int
zfs_suspend_fs(zfsvfs_t * zfsvfs)1773 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1774 {
1775 	int error;
1776 
1777 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1778 		return (error);
1779 
1780 	return (0);
1781 }
1782 
1783 /*
1784  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
1785  * is an invariant across any of the operations that can be performed while the
1786  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
1787  * are the same: the relevant objset and associated dataset are owned by
1788  * zfsvfs, held, and long held on entry.
1789  */
1790 int
zfs_resume_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)1791 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1792 {
1793 	int err, err2;
1794 	znode_t *zp;
1795 
1796 	ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1797 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1798 
1799 	/*
1800 	 * We already own this, so just update the objset_t, as the one we
1801 	 * had before may have been evicted.
1802 	 */
1803 	objset_t *os;
1804 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
1805 	VERIFY(dsl_dataset_long_held(ds));
1806 	dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1807 	dsl_pool_config_enter(dp, FTAG);
1808 	VERIFY0(dmu_objset_from_ds(ds, &os));
1809 	dsl_pool_config_exit(dp, FTAG);
1810 
1811 	err = zfsvfs_init(zfsvfs, os);
1812 	if (err != 0)
1813 		goto bail;
1814 
1815 	ds->ds_dir->dd_activity_cancelled = B_FALSE;
1816 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1817 
1818 	zfs_set_fuid_feature(zfsvfs);
1819 	zfsvfs->z_rollback_time = jiffies;
1820 
1821 	/*
1822 	 * Attempt to re-establish all the active inodes with their
1823 	 * dbufs.  If a zfs_rezget() fails, then we unhash the inode
1824 	 * and mark it stale.  This prevents a collision if a new
1825 	 * inode/object is created which must use the same inode
1826 	 * number.  The stale inode will be be released when the
1827 	 * VFS prunes the dentry holding the remaining references
1828 	 * on the stale inode.
1829 	 */
1830 	mutex_enter(&zfsvfs->z_znodes_lock);
1831 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1832 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1833 		err2 = zfs_rezget(zp);
1834 		if (err2) {
1835 			zpl_d_drop_aliases(ZTOI(zp));
1836 			remove_inode_hash(ZTOI(zp));
1837 		}
1838 
1839 		/* see comment in zfs_suspend_fs() */
1840 		if (zp->z_suspended) {
1841 			zfs_zrele_async(zp);
1842 			zp->z_suspended = B_FALSE;
1843 		}
1844 	}
1845 	mutex_exit(&zfsvfs->z_znodes_lock);
1846 
1847 	if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
1848 		/*
1849 		 * zfs_suspend_fs() could have interrupted freeing
1850 		 * of dnodes. We need to restart this freeing so
1851 		 * that we don't "leak" the space.
1852 		 */
1853 		zfs_unlinked_drain(zfsvfs);
1854 	}
1855 
1856 	/*
1857 	 * Most of the time zfs_suspend_fs is used for changing the contents
1858 	 * of the underlying dataset. ZFS rollback and receive operations
1859 	 * might create files for which negative dentries are present in
1860 	 * the cache. Since walking the dcache would require a lot of GPL-only
1861 	 * code duplication, it's much easier on these rather rare occasions
1862 	 * just to flush the whole dcache for the given dataset/filesystem.
1863 	 */
1864 	shrink_dcache_sb(zfsvfs->z_sb);
1865 
1866 bail:
1867 	if (err != 0)
1868 		zfsvfs->z_unmounted = B_TRUE;
1869 
1870 	/* release the VFS ops */
1871 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
1872 	ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1873 
1874 	if (err != 0) {
1875 		/*
1876 		 * Since we couldn't setup the sa framework, try to force
1877 		 * unmount this file system.
1878 		 */
1879 		if (zfsvfs->z_os)
1880 			(void) zfs_umount(zfsvfs->z_sb);
1881 	}
1882 	return (err);
1883 }
1884 
1885 /*
1886  * Release VOPs and unmount a suspended filesystem.
1887  */
1888 int
zfs_end_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)1889 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1890 {
1891 	ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1892 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1893 
1894 	/*
1895 	 * We already own this, so just hold and rele it to update the
1896 	 * objset_t, as the one we had before may have been evicted.
1897 	 */
1898 	objset_t *os;
1899 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
1900 	VERIFY(dsl_dataset_long_held(ds));
1901 	dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1902 	dsl_pool_config_enter(dp, FTAG);
1903 	VERIFY0(dmu_objset_from_ds(ds, &os));
1904 	dsl_pool_config_exit(dp, FTAG);
1905 	zfsvfs->z_os = os;
1906 
1907 	/* release the VOPs */
1908 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
1909 	ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1910 
1911 	/*
1912 	 * Try to force unmount this file system.
1913 	 */
1914 	(void) zfs_umount(zfsvfs->z_sb);
1915 	zfsvfs->z_unmounted = B_TRUE;
1916 	return (0);
1917 }
1918 
1919 /*
1920  * Automounted snapshots rely on periodic revalidation
1921  * to defer snapshots from being automatically unmounted.
1922  */
1923 
1924 inline void
zfs_exit_fs(zfsvfs_t * zfsvfs)1925 zfs_exit_fs(zfsvfs_t *zfsvfs)
1926 {
1927 	if (!zfsvfs->z_issnap)
1928 		return;
1929 
1930 	if (time_after(jiffies, zfsvfs->z_snap_defer_time +
1931 	    MAX(zfs_expire_snapshot * HZ / 2, HZ))) {
1932 		zfsvfs->z_snap_defer_time = jiffies;
1933 		zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa,
1934 		    dmu_objset_id(zfsvfs->z_os),
1935 		    zfs_expire_snapshot);
1936 	}
1937 }
1938 
1939 int
zfs_set_version(zfsvfs_t * zfsvfs,uint64_t newvers)1940 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1941 {
1942 	int error;
1943 	objset_t *os = zfsvfs->z_os;
1944 	dmu_tx_t *tx;
1945 
1946 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1947 		return (SET_ERROR(EINVAL));
1948 
1949 	if (newvers < zfsvfs->z_version)
1950 		return (SET_ERROR(EINVAL));
1951 
1952 	if (zfs_spa_version_map(newvers) >
1953 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
1954 		return (SET_ERROR(ENOTSUP));
1955 
1956 	tx = dmu_tx_create(os);
1957 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1958 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1959 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1960 		    ZFS_SA_ATTRS);
1961 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1962 	}
1963 	error = dmu_tx_assign(tx, TXG_WAIT);
1964 	if (error) {
1965 		dmu_tx_abort(tx);
1966 		return (error);
1967 	}
1968 
1969 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1970 	    8, 1, &newvers, tx);
1971 
1972 	if (error) {
1973 		dmu_tx_commit(tx);
1974 		return (error);
1975 	}
1976 
1977 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1978 		uint64_t sa_obj;
1979 
1980 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1981 		    SPA_VERSION_SA);
1982 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1983 		    DMU_OT_NONE, 0, tx);
1984 
1985 		error = zap_add(os, MASTER_NODE_OBJ,
1986 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1987 		ASSERT0(error);
1988 
1989 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
1990 		sa_register_update_callback(os, zfs_sa_upgrade);
1991 	}
1992 
1993 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
1994 	    "from %llu to %llu", zfsvfs->z_version, newvers);
1995 
1996 	dmu_tx_commit(tx);
1997 
1998 	zfsvfs->z_version = newvers;
1999 	os->os_version = newvers;
2000 
2001 	zfs_set_fuid_feature(zfsvfs);
2002 
2003 	return (0);
2004 }
2005 
2006 /*
2007  * Return true if the corresponding vfs's unmounted flag is set.
2008  * Otherwise return false.
2009  * If this function returns true we know VFS unmount has been initiated.
2010  */
2011 boolean_t
zfs_get_vfs_flag_unmounted(objset_t * os)2012 zfs_get_vfs_flag_unmounted(objset_t *os)
2013 {
2014 	zfsvfs_t *zfvp;
2015 	boolean_t unmounted = B_FALSE;
2016 
2017 	ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2018 
2019 	mutex_enter(&os->os_user_ptr_lock);
2020 	zfvp = dmu_objset_get_user(os);
2021 	if (zfvp != NULL && zfvp->z_unmounted)
2022 		unmounted = B_TRUE;
2023 	mutex_exit(&os->os_user_ptr_lock);
2024 
2025 	return (unmounted);
2026 }
2027 
2028 void
zfsvfs_update_fromname(const char * oldname,const char * newname)2029 zfsvfs_update_fromname(const char *oldname, const char *newname)
2030 {
2031 	/*
2032 	 * We don't need to do anything here, the devname is always current by
2033 	 * virtue of zfsvfs->z_sb->s_op->show_devname.
2034 	 */
2035 	(void) oldname, (void) newname;
2036 }
2037 
2038 void
zfs_init(void)2039 zfs_init(void)
2040 {
2041 	zfsctl_init();
2042 	zfs_znode_init();
2043 	dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2044 	register_filesystem(&zpl_fs_type);
2045 }
2046 
2047 void
zfs_fini(void)2048 zfs_fini(void)
2049 {
2050 	/*
2051 	 * we don't use outstanding because zpl_posix_acl_free might add more.
2052 	 */
2053 	taskq_wait(system_delay_taskq);
2054 	taskq_wait(system_taskq);
2055 	unregister_filesystem(&zpl_fs_type);
2056 	zfs_znode_fini();
2057 	zfsctl_fini();
2058 }
2059 
2060 #if defined(_KERNEL)
2061 EXPORT_SYMBOL(zfs_suspend_fs);
2062 EXPORT_SYMBOL(zfs_resume_fs);
2063 EXPORT_SYMBOL(zfs_set_version);
2064 EXPORT_SYMBOL(zfsvfs_create);
2065 EXPORT_SYMBOL(zfsvfs_free);
2066 EXPORT_SYMBOL(zfs_is_readonly);
2067 EXPORT_SYMBOL(zfs_domount);
2068 EXPORT_SYMBOL(zfs_preumount);
2069 EXPORT_SYMBOL(zfs_umount);
2070 EXPORT_SYMBOL(zfs_remount);
2071 EXPORT_SYMBOL(zfs_statvfs);
2072 EXPORT_SYMBOL(zfs_vget);
2073 EXPORT_SYMBOL(zfs_prune);
2074 #endif
2075