]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/zfs_vfsops.c
Revert "Revert "Fix unlink/xattr deadlock""
[mirror_zfs.git] / module / zfs / zfs_vfsops.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 */
24
25 /* Portions Copyright 2010 Robert Milkowski */
26
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.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/vfs_opreg.h>
36 #include <sys/mntent.h>
37 #include <sys/mount.h>
38 #include <sys/cmn_err.h>
39 #include "fs/fs_subr.h"
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_vnops.h>
42 #include <sys/zfs_dir.h>
43 #include <sys/zil.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/dmu.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_dataset.h>
48 #include <sys/dsl_deleg.h>
49 #include <sys/spa.h>
50 #include <sys/zap.h>
51 #include <sys/sa.h>
52 #include <sys/sa_impl.h>
53 #include <sys/varargs.h>
54 #include <sys/policy.h>
55 #include <sys/atomic.h>
56 #include <sys/mkdev.h>
57 #include <sys/modctl.h>
58 #include <sys/refstr.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/zfs_ctldir.h>
61 #include <sys/zfs_fuid.h>
62 #include <sys/bootconf.h>
63 #include <sys/sunddi.h>
64 #include <sys/dnlc.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/spa_boot.h>
67 #include <sys/zpl.h>
68 #include "zfs_comutil.h"
69
70
71 /*ARGSUSED*/
72 int
73 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
74 {
75 zfs_sb_t *zsb = sb->s_fs_info;
76
77 /*
78 * Data integrity is job one. We don't want a compromised kernel
79 * writing to the storage pool, so we never sync during panic.
80 */
81 if (unlikely(oops_in_progress))
82 return (0);
83
84 /*
85 * Semantically, the only requirement is that the sync be initiated.
86 * The DMU syncs out txgs frequently, so there's nothing to do.
87 */
88 if (!wait)
89 return (0);
90
91 if (zsb != NULL) {
92 /*
93 * Sync a specific filesystem.
94 */
95 dsl_pool_t *dp;
96
97 ZFS_ENTER(zsb);
98 dp = dmu_objset_pool(zsb->z_os);
99
100 /*
101 * If the system is shutting down, then skip any
102 * filesystems which may exist on a suspended pool.
103 */
104 if (spa_suspended(dp->dp_spa)) {
105 ZFS_EXIT(zsb);
106 return (0);
107 }
108
109 if (zsb->z_log != NULL)
110 zil_commit(zsb->z_log, 0);
111
112 ZFS_EXIT(zsb);
113 } else {
114 /*
115 * Sync all ZFS filesystems. This is what happens when you
116 * run sync(1M). Unlike other filesystems, ZFS honors the
117 * request by waiting for all pools to commit all dirty data.
118 */
119 spa_sync_allpools();
120 }
121
122 return (0);
123 }
124 EXPORT_SYMBOL(zfs_sync);
125
126 boolean_t
127 zfs_is_readonly(zfs_sb_t *zsb)
128 {
129 return (!!(zsb->z_sb->s_flags & MS_RDONLY));
130 }
131 EXPORT_SYMBOL(zfs_is_readonly);
132
133 static void
134 atime_changed_cb(void *arg, uint64_t newval)
135 {
136 ((zfs_sb_t *)arg)->z_atime = newval;
137 }
138
139 static void
140 xattr_changed_cb(void *arg, uint64_t newval)
141 {
142 zfs_sb_t *zsb = arg;
143
144 if (newval == ZFS_XATTR_OFF) {
145 zsb->z_flags &= ~ZSB_XATTR;
146 } else {
147 zsb->z_flags |= ZSB_XATTR;
148
149 if (newval == ZFS_XATTR_SA)
150 zsb->z_xattr_sa = B_TRUE;
151 else
152 zsb->z_xattr_sa = B_FALSE;
153 }
154 }
155
156 static void
157 blksz_changed_cb(void *arg, uint64_t newval)
158 {
159 zfs_sb_t *zsb = arg;
160
161 if (newval < SPA_MINBLOCKSIZE ||
162 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
163 newval = SPA_MAXBLOCKSIZE;
164
165 zsb->z_max_blksz = newval;
166 }
167
168 static void
169 readonly_changed_cb(void *arg, uint64_t newval)
170 {
171 zfs_sb_t *zsb = arg;
172 struct super_block *sb = zsb->z_sb;
173
174 if (sb == NULL)
175 return;
176
177 if (newval)
178 sb->s_flags |= MS_RDONLY;
179 else
180 sb->s_flags &= ~MS_RDONLY;
181 }
182
183 static void
184 devices_changed_cb(void *arg, uint64_t newval)
185 {
186 }
187
188 static void
189 setuid_changed_cb(void *arg, uint64_t newval)
190 {
191 }
192
193 static void
194 exec_changed_cb(void *arg, uint64_t newval)
195 {
196 }
197
198 static void
199 nbmand_changed_cb(void *arg, uint64_t newval)
200 {
201 zfs_sb_t *zsb = arg;
202 struct super_block *sb = zsb->z_sb;
203
204 if (sb == NULL)
205 return;
206
207 if (newval == TRUE)
208 sb->s_flags |= MS_MANDLOCK;
209 else
210 sb->s_flags &= ~MS_MANDLOCK;
211 }
212
213 static void
214 snapdir_changed_cb(void *arg, uint64_t newval)
215 {
216 ((zfs_sb_t *)arg)->z_show_ctldir = newval;
217 }
218
219 static void
220 vscan_changed_cb(void *arg, uint64_t newval)
221 {
222 ((zfs_sb_t *)arg)->z_vscan = newval;
223 }
224
225 static void
226 acl_inherit_changed_cb(void *arg, uint64_t newval)
227 {
228 ((zfs_sb_t *)arg)->z_acl_inherit = newval;
229 }
230
231 int
232 zfs_register_callbacks(zfs_sb_t *zsb)
233 {
234 struct dsl_dataset *ds = NULL;
235 objset_t *os = zsb->z_os;
236 int error = 0;
237
238 if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os)))
239 readonly_changed_cb(zsb, B_TRUE);
240
241 /*
242 * Register property callbacks.
243 *
244 * It would probably be fine to just check for i/o error from
245 * the first prop_register(), but I guess I like to go
246 * overboard...
247 */
248 ds = dmu_objset_ds(os);
249 error = dsl_prop_register(ds,
250 "atime", atime_changed_cb, zsb);
251 error = error ? error : dsl_prop_register(ds,
252 "xattr", xattr_changed_cb, zsb);
253 error = error ? error : dsl_prop_register(ds,
254 "recordsize", blksz_changed_cb, zsb);
255 error = error ? error : dsl_prop_register(ds,
256 "readonly", readonly_changed_cb, zsb);
257 error = error ? error : dsl_prop_register(ds,
258 "devices", devices_changed_cb, zsb);
259 error = error ? error : dsl_prop_register(ds,
260 "setuid", setuid_changed_cb, zsb);
261 error = error ? error : dsl_prop_register(ds,
262 "exec", exec_changed_cb, zsb);
263 error = error ? error : dsl_prop_register(ds,
264 "snapdir", snapdir_changed_cb, zsb);
265 error = error ? error : dsl_prop_register(ds,
266 "aclinherit", acl_inherit_changed_cb, zsb);
267 error = error ? error : dsl_prop_register(ds,
268 "vscan", vscan_changed_cb, zsb);
269 error = error ? error : dsl_prop_register(ds,
270 "nbmand", nbmand_changed_cb, zsb);
271 if (error)
272 goto unregister;
273
274 return (0);
275
276 unregister:
277 /*
278 * We may attempt to unregister some callbacks that are not
279 * registered, but this is OK; it will simply return ENOMSG,
280 * which we will ignore.
281 */
282 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zsb);
283 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zsb);
284 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zsb);
285 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zsb);
286 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zsb);
287 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zsb);
288 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zsb);
289 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zsb);
290 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
291 zsb);
292 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zsb);
293 (void) dsl_prop_unregister(ds, "nbmand", nbmand_changed_cb, zsb);
294
295 return (error);
296 }
297 EXPORT_SYMBOL(zfs_register_callbacks);
298
299 static int
300 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
301 uint64_t *userp, uint64_t *groupp)
302 {
303 int error = 0;
304
305 /*
306 * Is it a valid type of object to track?
307 */
308 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
309 return (ENOENT);
310
311 /*
312 * If we have a NULL data pointer
313 * then assume the id's aren't changing and
314 * return EEXIST to the dmu to let it know to
315 * use the same ids
316 */
317 if (data == NULL)
318 return (EEXIST);
319
320 if (bonustype == DMU_OT_ZNODE) {
321 znode_phys_t *znp = data;
322 *userp = znp->zp_uid;
323 *groupp = znp->zp_gid;
324 } else {
325 int hdrsize;
326 sa_hdr_phys_t *sap = data;
327 sa_hdr_phys_t sa = *sap;
328 boolean_t swap = B_FALSE;
329
330 ASSERT(bonustype == DMU_OT_SA);
331
332 if (sa.sa_magic == 0) {
333 /*
334 * This should only happen for newly created
335 * files that haven't had the znode data filled
336 * in yet.
337 */
338 *userp = 0;
339 *groupp = 0;
340 return (0);
341 }
342 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
343 sa.sa_magic = SA_MAGIC;
344 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
345 swap = B_TRUE;
346 } else {
347 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
348 }
349
350 hdrsize = sa_hdrsize(&sa);
351 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
352 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
353 SA_UID_OFFSET));
354 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
355 SA_GID_OFFSET));
356 if (swap) {
357 *userp = BSWAP_64(*userp);
358 *groupp = BSWAP_64(*groupp);
359 }
360 }
361 return (error);
362 }
363
364 static void
365 fuidstr_to_sid(zfs_sb_t *zsb, const char *fuidstr,
366 char *domainbuf, int buflen, uid_t *ridp)
367 {
368 uint64_t fuid;
369 const char *domain;
370
371 fuid = strtonum(fuidstr, NULL);
372
373 domain = zfs_fuid_find_by_idx(zsb, FUID_INDEX(fuid));
374 if (domain)
375 (void) strlcpy(domainbuf, domain, buflen);
376 else
377 domainbuf[0] = '\0';
378 *ridp = FUID_RID(fuid);
379 }
380
381 static uint64_t
382 zfs_userquota_prop_to_obj(zfs_sb_t *zsb, zfs_userquota_prop_t type)
383 {
384 switch (type) {
385 case ZFS_PROP_USERUSED:
386 return (DMU_USERUSED_OBJECT);
387 case ZFS_PROP_GROUPUSED:
388 return (DMU_GROUPUSED_OBJECT);
389 case ZFS_PROP_USERQUOTA:
390 return (zsb->z_userquota_obj);
391 case ZFS_PROP_GROUPQUOTA:
392 return (zsb->z_groupquota_obj);
393 default:
394 return (ENOTSUP);
395 }
396 return (0);
397 }
398
399 int
400 zfs_userspace_many(zfs_sb_t *zsb, zfs_userquota_prop_t type,
401 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
402 {
403 int error;
404 zap_cursor_t zc;
405 zap_attribute_t za;
406 zfs_useracct_t *buf = vbuf;
407 uint64_t obj;
408
409 if (!dmu_objset_userspace_present(zsb->z_os))
410 return (ENOTSUP);
411
412 obj = zfs_userquota_prop_to_obj(zsb, type);
413 if (obj == 0) {
414 *bufsizep = 0;
415 return (0);
416 }
417
418 for (zap_cursor_init_serialized(&zc, zsb->z_os, obj, *cookiep);
419 (error = zap_cursor_retrieve(&zc, &za)) == 0;
420 zap_cursor_advance(&zc)) {
421 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
422 *bufsizep)
423 break;
424
425 fuidstr_to_sid(zsb, za.za_name,
426 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
427
428 buf->zu_space = za.za_first_integer;
429 buf++;
430 }
431 if (error == ENOENT)
432 error = 0;
433
434 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
435 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
436 *cookiep = zap_cursor_serialize(&zc);
437 zap_cursor_fini(&zc);
438 return (error);
439 }
440 EXPORT_SYMBOL(zfs_userspace_many);
441
442 /*
443 * buf must be big enough (eg, 32 bytes)
444 */
445 static int
446 id_to_fuidstr(zfs_sb_t *zsb, const char *domain, uid_t rid,
447 char *buf, boolean_t addok)
448 {
449 uint64_t fuid;
450 int domainid = 0;
451
452 if (domain && domain[0]) {
453 domainid = zfs_fuid_find_by_domain(zsb, domain, NULL, addok);
454 if (domainid == -1)
455 return (ENOENT);
456 }
457 fuid = FUID_ENCODE(domainid, rid);
458 (void) sprintf(buf, "%llx", (longlong_t)fuid);
459 return (0);
460 }
461
462 int
463 zfs_userspace_one(zfs_sb_t *zsb, zfs_userquota_prop_t type,
464 const char *domain, uint64_t rid, uint64_t *valp)
465 {
466 char buf[32];
467 int err;
468 uint64_t obj;
469
470 *valp = 0;
471
472 if (!dmu_objset_userspace_present(zsb->z_os))
473 return (ENOTSUP);
474
475 obj = zfs_userquota_prop_to_obj(zsb, type);
476 if (obj == 0)
477 return (0);
478
479 err = id_to_fuidstr(zsb, domain, rid, buf, B_FALSE);
480 if (err)
481 return (err);
482
483 err = zap_lookup(zsb->z_os, obj, buf, 8, 1, valp);
484 if (err == ENOENT)
485 err = 0;
486 return (err);
487 }
488 EXPORT_SYMBOL(zfs_userspace_one);
489
490 int
491 zfs_set_userquota(zfs_sb_t *zsb, zfs_userquota_prop_t type,
492 const char *domain, uint64_t rid, uint64_t quota)
493 {
494 char buf[32];
495 int err;
496 dmu_tx_t *tx;
497 uint64_t *objp;
498 boolean_t fuid_dirtied;
499
500 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
501 return (EINVAL);
502
503 if (zsb->z_version < ZPL_VERSION_USERSPACE)
504 return (ENOTSUP);
505
506 objp = (type == ZFS_PROP_USERQUOTA) ? &zsb->z_userquota_obj :
507 &zsb->z_groupquota_obj;
508
509 err = id_to_fuidstr(zsb, domain, rid, buf, B_TRUE);
510 if (err)
511 return (err);
512 fuid_dirtied = zsb->z_fuid_dirty;
513
514 tx = dmu_tx_create(zsb->z_os);
515 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
516 if (*objp == 0) {
517 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
518 zfs_userquota_prop_prefixes[type]);
519 }
520 if (fuid_dirtied)
521 zfs_fuid_txhold(zsb, tx);
522 err = dmu_tx_assign(tx, TXG_WAIT);
523 if (err) {
524 dmu_tx_abort(tx);
525 return (err);
526 }
527
528 mutex_enter(&zsb->z_lock);
529 if (*objp == 0) {
530 *objp = zap_create(zsb->z_os, DMU_OT_USERGROUP_QUOTA,
531 DMU_OT_NONE, 0, tx);
532 VERIFY(0 == zap_add(zsb->z_os, MASTER_NODE_OBJ,
533 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
534 }
535 mutex_exit(&zsb->z_lock);
536
537 if (quota == 0) {
538 err = zap_remove(zsb->z_os, *objp, buf, tx);
539 if (err == ENOENT)
540 err = 0;
541 } else {
542 err = zap_update(zsb->z_os, *objp, buf, 8, 1, &quota, tx);
543 }
544 ASSERT(err == 0);
545 if (fuid_dirtied)
546 zfs_fuid_sync(zsb, tx);
547 dmu_tx_commit(tx);
548 return (err);
549 }
550 EXPORT_SYMBOL(zfs_set_userquota);
551
552 boolean_t
553 zfs_fuid_overquota(zfs_sb_t *zsb, boolean_t isgroup, uint64_t fuid)
554 {
555 char buf[32];
556 uint64_t used, quota, usedobj, quotaobj;
557 int err;
558
559 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
560 quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;
561
562 if (quotaobj == 0 || zsb->z_replay)
563 return (B_FALSE);
564
565 (void) sprintf(buf, "%llx", (longlong_t)fuid);
566 err = zap_lookup(zsb->z_os, quotaobj, buf, 8, 1, &quota);
567 if (err != 0)
568 return (B_FALSE);
569
570 err = zap_lookup(zsb->z_os, usedobj, buf, 8, 1, &used);
571 if (err != 0)
572 return (B_FALSE);
573 return (used >= quota);
574 }
575 EXPORT_SYMBOL(zfs_fuid_overquota);
576
577 boolean_t
578 zfs_owner_overquota(zfs_sb_t *zsb, znode_t *zp, boolean_t isgroup)
579 {
580 uint64_t fuid;
581 uint64_t quotaobj;
582
583 quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;
584
585 fuid = isgroup ? zp->z_gid : zp->z_uid;
586
587 if (quotaobj == 0 || zsb->z_replay)
588 return (B_FALSE);
589
590 return (zfs_fuid_overquota(zsb, isgroup, fuid));
591 }
592 EXPORT_SYMBOL(zfs_owner_overquota);
593
594 int
595 zfs_sb_create(const char *osname, zfs_sb_t **zsbp)
596 {
597 objset_t *os;
598 zfs_sb_t *zsb;
599 uint64_t zval;
600 int i, error;
601 uint64_t sa_obj;
602
603 zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP | KM_NODEBUG);
604
605 /*
606 * We claim to always be readonly so we can open snapshots;
607 * other ZPL code will prevent us from writing to snapshots.
608 */
609 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zsb, &os);
610 if (error) {
611 kmem_free(zsb, sizeof (zfs_sb_t));
612 return (error);
613 }
614
615 /*
616 * Initialize the zfs-specific filesystem structure.
617 * Should probably make this a kmem cache, shuffle fields,
618 * and just bzero up to z_hold_mtx[].
619 */
620 zsb->z_sb = NULL;
621 zsb->z_parent = zsb;
622 zsb->z_max_blksz = SPA_MAXBLOCKSIZE;
623 zsb->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
624 zsb->z_os = os;
625
626 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zsb->z_version);
627 if (error) {
628 goto out;
629 } else if (zsb->z_version >
630 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
631 (void) printk("Can't mount a version %lld file system "
632 "on a version %lld pool\n. Pool must be upgraded to mount "
633 "this file system.", (u_longlong_t)zsb->z_version,
634 (u_longlong_t)spa_version(dmu_objset_spa(os)));
635 error = ENOTSUP;
636 goto out;
637 }
638 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
639 goto out;
640 zsb->z_norm = (int)zval;
641
642 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
643 goto out;
644 zsb->z_utf8 = (zval != 0);
645
646 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
647 goto out;
648 zsb->z_case = (uint_t)zval;
649
650 /*
651 * Fold case on file systems that are always or sometimes case
652 * insensitive.
653 */
654 if (zsb->z_case == ZFS_CASE_INSENSITIVE ||
655 zsb->z_case == ZFS_CASE_MIXED)
656 zsb->z_norm |= U8_TEXTPREP_TOUPPER;
657
658 zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
659 zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);
660
661 if (zsb->z_use_sa) {
662 /* should either have both of these objects or none */
663 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
664 &sa_obj);
665 if (error)
666 goto out;
667
668 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &zval);
669 if ((error == 0) && (zval == ZFS_XATTR_SA))
670 zsb->z_xattr_sa = B_TRUE;
671 } else {
672 /*
673 * Pre SA versions file systems should never touch
674 * either the attribute registration or layout objects.
675 */
676 sa_obj = 0;
677 }
678
679 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
680 &zsb->z_attr_table);
681 if (error)
682 goto out;
683
684 if (zsb->z_version >= ZPL_VERSION_SA)
685 sa_register_update_callback(os, zfs_sa_upgrade);
686
687 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
688 &zsb->z_root);
689 if (error)
690 goto out;
691 ASSERT(zsb->z_root != 0);
692
693 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
694 &zsb->z_unlinkedobj);
695 if (error)
696 goto out;
697
698 error = zap_lookup(os, MASTER_NODE_OBJ,
699 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
700 8, 1, &zsb->z_userquota_obj);
701 if (error && error != ENOENT)
702 goto out;
703
704 error = zap_lookup(os, MASTER_NODE_OBJ,
705 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
706 8, 1, &zsb->z_groupquota_obj);
707 if (error && error != ENOENT)
708 goto out;
709
710 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
711 &zsb->z_fuid_obj);
712 if (error && error != ENOENT)
713 goto out;
714
715 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
716 &zsb->z_shares_dir);
717 if (error && error != ENOENT)
718 goto out;
719
720 mutex_init(&zsb->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
721 mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL);
722 list_create(&zsb->z_all_znodes, sizeof (znode_t),
723 offsetof(znode_t, z_link_node));
724 rrw_init(&zsb->z_teardown_lock);
725 rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
726 rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
727 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
728 mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
729
730 avl_create(&zsb->z_ctldir_snaps, snapentry_compare,
731 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
732 mutex_init(&zsb->z_ctldir_lock, NULL, MUTEX_DEFAULT, NULL);
733
734 *zsbp = zsb;
735 return (0);
736
737 out:
738 dmu_objset_disown(os, zsb);
739 *zsbp = NULL;
740 kmem_free(zsb, sizeof (zfs_sb_t));
741 return (error);
742 }
743 EXPORT_SYMBOL(zfs_sb_create);
744
745 int
746 zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting)
747 {
748 int error;
749
750 error = zfs_register_callbacks(zsb);
751 if (error)
752 return (error);
753
754 /*
755 * Set the objset user_ptr to track its zsb.
756 */
757 mutex_enter(&zsb->z_os->os_user_ptr_lock);
758 dmu_objset_set_user(zsb->z_os, zsb);
759 mutex_exit(&zsb->z_os->os_user_ptr_lock);
760
761 zsb->z_log = zil_open(zsb->z_os, zfs_get_data);
762
763 /*
764 * If we are not mounting (ie: online recv), then we don't
765 * have to worry about replaying the log as we blocked all
766 * operations out since we closed the ZIL.
767 */
768 if (mounting) {
769 boolean_t readonly;
770
771 /*
772 * During replay we remove the read only flag to
773 * allow replays to succeed.
774 */
775 readonly = zfs_is_readonly(zsb);
776 if (readonly != 0)
777 readonly_changed_cb(zsb, B_FALSE);
778 else
779 zfs_unlinked_drain(zsb);
780
781 /*
782 * Parse and replay the intent log.
783 *
784 * Because of ziltest, this must be done after
785 * zfs_unlinked_drain(). (Further note: ziltest
786 * doesn't use readonly mounts, where
787 * zfs_unlinked_drain() isn't called.) This is because
788 * ziltest causes spa_sync() to think it's committed,
789 * but actually it is not, so the intent log contains
790 * many txg's worth of changes.
791 *
792 * In particular, if object N is in the unlinked set in
793 * the last txg to actually sync, then it could be
794 * actually freed in a later txg and then reallocated
795 * in a yet later txg. This would write a "create
796 * object N" record to the intent log. Normally, this
797 * would be fine because the spa_sync() would have
798 * written out the fact that object N is free, before
799 * we could write the "create object N" intent log
800 * record.
801 *
802 * But when we are in ziltest mode, we advance the "open
803 * txg" without actually spa_sync()-ing the changes to
804 * disk. So we would see that object N is still
805 * allocated and in the unlinked set, and there is an
806 * intent log record saying to allocate it.
807 */
808 if (spa_writeable(dmu_objset_spa(zsb->z_os))) {
809 if (zil_replay_disable) {
810 zil_destroy(zsb->z_log, B_FALSE);
811 } else {
812 zsb->z_replay = B_TRUE;
813 zil_replay(zsb->z_os, zsb,
814 zfs_replay_vector);
815 zsb->z_replay = B_FALSE;
816 }
817 }
818
819 /* restore readonly bit */
820 if (readonly != 0)
821 readonly_changed_cb(zsb, B_TRUE);
822 }
823
824 return (0);
825 }
826 EXPORT_SYMBOL(zfs_sb_setup);
827
828 void
829 zfs_sb_free(zfs_sb_t *zsb)
830 {
831 int i;
832
833 zfs_fuid_destroy(zsb);
834
835 mutex_destroy(&zsb->z_znodes_lock);
836 mutex_destroy(&zsb->z_lock);
837 list_destroy(&zsb->z_all_znodes);
838 rrw_destroy(&zsb->z_teardown_lock);
839 rw_destroy(&zsb->z_teardown_inactive_lock);
840 rw_destroy(&zsb->z_fuid_lock);
841 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
842 mutex_destroy(&zsb->z_hold_mtx[i]);
843 mutex_destroy(&zsb->z_ctldir_lock);
844 avl_destroy(&zsb->z_ctldir_snaps);
845 kmem_free(zsb, sizeof (zfs_sb_t));
846 }
847 EXPORT_SYMBOL(zfs_sb_free);
848
849 static void
850 zfs_set_fuid_feature(zfs_sb_t *zsb)
851 {
852 zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
853 zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);
854 }
855
856 void
857 zfs_unregister_callbacks(zfs_sb_t *zsb)
858 {
859 objset_t *os = zsb->z_os;
860 struct dsl_dataset *ds;
861
862 /*
863 * Unregister properties.
864 */
865 if (!dmu_objset_is_snapshot(os)) {
866 ds = dmu_objset_ds(os);
867 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
868 zsb) == 0);
869
870 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
871 zsb) == 0);
872
873 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
874 zsb) == 0);
875
876 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
877 zsb) == 0);
878
879 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
880 zsb) == 0);
881
882 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
883 zsb) == 0);
884
885 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
886 zsb) == 0);
887
888 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
889 zsb) == 0);
890
891 VERIFY(dsl_prop_unregister(ds, "aclinherit",
892 acl_inherit_changed_cb, zsb) == 0);
893
894 VERIFY(dsl_prop_unregister(ds, "vscan",
895 vscan_changed_cb, zsb) == 0);
896
897 VERIFY(dsl_prop_unregister(ds, "nbmand",
898 nbmand_changed_cb, zsb) == 0);
899 }
900 }
901 EXPORT_SYMBOL(zfs_unregister_callbacks);
902
903 #ifdef HAVE_MLSLABEL
904 /*
905 * zfs_check_global_label:
906 * Check that the hex label string is appropriate for the dataset
907 * being mounted into the global_zone proper.
908 *
909 * Return an error if the hex label string is not default or
910 * admin_low/admin_high. For admin_low labels, the corresponding
911 * dataset must be readonly.
912 */
913 int
914 zfs_check_global_label(const char *dsname, const char *hexsl)
915 {
916 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
917 return (0);
918 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
919 return (0);
920 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
921 /* must be readonly */
922 uint64_t rdonly;
923
924 if (dsl_prop_get_integer(dsname,
925 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
926 return (EACCES);
927 return (rdonly ? 0 : EACCES);
928 }
929 return (EACCES);
930 }
931 EXPORT_SYMBOL(zfs_check_global_label);
932 #endif /* HAVE_MLSLABEL */
933
934 int
935 zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
936 {
937 zfs_sb_t *zsb = dentry->d_sb->s_fs_info;
938 uint64_t refdbytes, availbytes, usedobjs, availobjs;
939 uint64_t fsid;
940 uint32_t bshift;
941
942 ZFS_ENTER(zsb);
943
944 dmu_objset_space(zsb->z_os,
945 &refdbytes, &availbytes, &usedobjs, &availobjs);
946
947 fsid = dmu_objset_fsid_guid(zsb->z_os);
948 /*
949 * The underlying storage pool actually uses multiple block
950 * size. Under Solaris frsize (fragment size) is reported as
951 * the smallest block size we support, and bsize (block size)
952 * as the filesystem's maximum block size. Unfortunately,
953 * under Linux the fragment size and block size are often used
954 * interchangeably. Thus we are forced to report both of them
955 * as the filesystem's maximum block size.
956 */
957 statp->f_frsize = zsb->z_max_blksz;
958 statp->f_bsize = zsb->z_max_blksz;
959 bshift = fls(statp->f_bsize) - 1;
960
961 /*
962 * The following report "total" blocks of various kinds in
963 * the file system, but reported in terms of f_bsize - the
964 * "preferred" size.
965 */
966
967 statp->f_blocks = (refdbytes + availbytes) >> bshift;
968 statp->f_bfree = availbytes >> bshift;
969 statp->f_bavail = statp->f_bfree; /* no root reservation */
970
971 /*
972 * statvfs() should really be called statufs(), because it assumes
973 * static metadata. ZFS doesn't preallocate files, so the best
974 * we can do is report the max that could possibly fit in f_files,
975 * and that minus the number actually used in f_ffree.
976 * For f_ffree, report the smaller of the number of object available
977 * and the number of blocks (each object will take at least a block).
978 */
979 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
980 statp->f_files = statp->f_ffree + usedobjs;
981 statp->f_fsid.val[0] = (uint32_t)fsid;
982 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
983 statp->f_type = ZFS_SUPER_MAGIC;
984 statp->f_namelen = ZFS_MAXNAMELEN;
985
986 /*
987 * We have all of 40 characters to stuff a string here.
988 * Is there anything useful we could/should provide?
989 */
990 bzero(statp->f_spare, sizeof (statp->f_spare));
991
992 ZFS_EXIT(zsb);
993 return (0);
994 }
995 EXPORT_SYMBOL(zfs_statvfs);
996
997 int
998 zfs_root(zfs_sb_t *zsb, struct inode **ipp)
999 {
1000 znode_t *rootzp;
1001 int error;
1002
1003 ZFS_ENTER(zsb);
1004
1005 error = zfs_zget(zsb, zsb->z_root, &rootzp);
1006 if (error == 0)
1007 *ipp = ZTOI(rootzp);
1008
1009 ZFS_EXIT(zsb);
1010 return (error);
1011 }
1012 EXPORT_SYMBOL(zfs_root);
1013
1014 #ifdef HAVE_SHRINK
1015 int
1016 zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1017 {
1018 zfs_sb_t *zsb = sb->s_fs_info;
1019 struct shrinker *shrinker = &sb->s_shrink;
1020 struct shrink_control sc = {
1021 .nr_to_scan = nr_to_scan,
1022 .gfp_mask = GFP_KERNEL,
1023 };
1024
1025 ZFS_ENTER(zsb);
1026 *objects = (*shrinker->shrink)(shrinker, &sc);
1027 ZFS_EXIT(zsb);
1028
1029 return (0);
1030 }
1031 EXPORT_SYMBOL(zfs_sb_prune);
1032 #endif /* HAVE_SHRINK */
1033
1034 /*
1035 * Teardown the zfs_sb_t.
1036 *
1037 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1038 * and 'z_teardown_inactive_lock' held.
1039 */
1040 int
1041 zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
1042 {
1043 znode_t *zp;
1044
1045 rrw_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);
1046
1047 if (!unmounting) {
1048 /*
1049 * We purge the parent filesystem's super block as the
1050 * parent filesystem and all of its snapshots have their
1051 * inode's super block set to the parent's filesystem's
1052 * super block. Note, 'z_parent' is self referential
1053 * for non-snapshots.
1054 */
1055 shrink_dcache_sb(zsb->z_parent->z_sb);
1056 }
1057
1058 /*
1059 * Drain the iput_taskq to ensure all active references to the
1060 * zfs_sb_t have been handled only then can it be safely destroyed.
1061 */
1062 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os)));
1063
1064 /*
1065 * Close the zil. NB: Can't close the zil while zfs_inactive
1066 * threads are blocked as zil_close can call zfs_inactive.
1067 */
1068 if (zsb->z_log) {
1069 zil_close(zsb->z_log);
1070 zsb->z_log = NULL;
1071 }
1072
1073 rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER);
1074
1075 /*
1076 * If we are not unmounting (ie: online recv) and someone already
1077 * unmounted this file system while we were doing the switcheroo,
1078 * or a reopen of z_os failed then just bail out now.
1079 */
1080 if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
1081 rw_exit(&zsb->z_teardown_inactive_lock);
1082 rrw_exit(&zsb->z_teardown_lock, FTAG);
1083 return (EIO);
1084 }
1085
1086 /*
1087 * At this point there are no VFS ops active, and any new VFS ops
1088 * will fail with EIO since we have z_teardown_lock for writer (only
1089 * relevant for forced unmount).
1090 *
1091 * Release all holds on dbufs.
1092 */
1093 mutex_enter(&zsb->z_znodes_lock);
1094 for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
1095 zp = list_next(&zsb->z_all_znodes, zp)) {
1096 if (zp->z_sa_hdl) {
1097 ASSERT(atomic_read(&ZTOI(zp)->i_count) > 0);
1098 zfs_znode_dmu_fini(zp);
1099 }
1100 }
1101 mutex_exit(&zsb->z_znodes_lock);
1102
1103 /*
1104 * If we are unmounting, set the unmounted flag and let new VFS ops
1105 * unblock. zfs_inactive will have the unmounted behavior, and all
1106 * other VFS ops will fail with EIO.
1107 */
1108 if (unmounting) {
1109 zsb->z_unmounted = B_TRUE;
1110 rrw_exit(&zsb->z_teardown_lock, FTAG);
1111 rw_exit(&zsb->z_teardown_inactive_lock);
1112 }
1113
1114 /*
1115 * z_os will be NULL if there was an error in attempting to reopen
1116 * zsb, so just return as the properties had already been
1117 *
1118 * unregistered and cached data had been evicted before.
1119 */
1120 if (zsb->z_os == NULL)
1121 return (0);
1122
1123 /*
1124 * Unregister properties.
1125 */
1126 zfs_unregister_callbacks(zsb);
1127
1128 /*
1129 * Evict cached data
1130 */
1131 if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) &&
1132 !zfs_is_readonly(zsb))
1133 txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
1134 (void) dmu_objset_evict_dbufs(zsb->z_os);
1135
1136 return (0);
1137 }
1138 EXPORT_SYMBOL(zfs_sb_teardown);
1139
1140 #if defined(HAVE_BDI) && !defined(HAVE_BDI_SETUP_AND_REGISTER)
1141 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1142 #endif /* HAVE_BDI && !HAVE_BDI_SETUP_AND_REGISTER */
1143
1144 int
1145 zfs_domount(struct super_block *sb, void *data, int silent)
1146 {
1147 zpl_mount_data_t *zmd = data;
1148 const char *osname = zmd->z_osname;
1149 zfs_sb_t *zsb;
1150 struct inode *root_inode;
1151 uint64_t recordsize;
1152 int error;
1153
1154 error = zfs_sb_create(osname, &zsb);
1155 if (error)
1156 return (error);
1157
1158 if ((error = dsl_prop_get_integer(osname, "recordsize",
1159 &recordsize, NULL)))
1160 goto out;
1161
1162 zsb->z_sb = sb;
1163 sb->s_fs_info = zsb;
1164 sb->s_magic = ZFS_SUPER_MAGIC;
1165 sb->s_maxbytes = MAX_LFS_FILESIZE;
1166 sb->s_time_gran = 1;
1167 sb->s_blocksize = recordsize;
1168 sb->s_blocksize_bits = ilog2(recordsize);
1169
1170 #ifdef HAVE_BDI
1171 /*
1172 * 2.6.32 API change,
1173 * Added backing_device_info (BDI) per super block interfaces. A BDI
1174 * must be configured when using a non-device backed filesystem for
1175 * proper writeback. This is not required for older pdflush kernels.
1176 *
1177 * NOTE: Linux read-ahead is disabled in favor of zfs read-ahead.
1178 */
1179 zsb->z_bdi.ra_pages = 0;
1180 sb->s_bdi = &zsb->z_bdi;
1181
1182 error = -bdi_setup_and_register(&zsb->z_bdi, "zfs", BDI_CAP_MAP_COPY);
1183 if (error)
1184 goto out;
1185 #endif /* HAVE_BDI */
1186
1187 /* Set callback operations for the file system. */
1188 sb->s_op = &zpl_super_operations;
1189 sb->s_xattr = zpl_xattr_handlers;
1190 sb->s_export_op = &zpl_export_operations;
1191
1192 /* Set features for file system. */
1193 zfs_set_fuid_feature(zsb);
1194
1195 if (dmu_objset_is_snapshot(zsb->z_os)) {
1196 uint64_t pval;
1197
1198 atime_changed_cb(zsb, B_FALSE);
1199 readonly_changed_cb(zsb, B_TRUE);
1200 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1201 goto out;
1202 xattr_changed_cb(zsb, pval);
1203 zsb->z_issnap = B_TRUE;
1204 zsb->z_os->os_sync = ZFS_SYNC_DISABLED;
1205
1206 mutex_enter(&zsb->z_os->os_user_ptr_lock);
1207 dmu_objset_set_user(zsb->z_os, zsb);
1208 mutex_exit(&zsb->z_os->os_user_ptr_lock);
1209 } else {
1210 error = zfs_sb_setup(zsb, B_TRUE);
1211 }
1212
1213 /* Allocate a root inode for the filesystem. */
1214 error = zfs_root(zsb, &root_inode);
1215 if (error) {
1216 (void) zfs_umount(sb);
1217 goto out;
1218 }
1219
1220 /* Allocate a root dentry for the filesystem */
1221 sb->s_root = d_make_root(root_inode);
1222 if (sb->s_root == NULL) {
1223 (void) zfs_umount(sb);
1224 error = ENOMEM;
1225 goto out;
1226 }
1227
1228 if (!zsb->z_issnap)
1229 zfsctl_create(zsb);
1230 out:
1231 if (error) {
1232 dmu_objset_disown(zsb->z_os, zsb);
1233 zfs_sb_free(zsb);
1234 }
1235
1236 return (error);
1237 }
1238 EXPORT_SYMBOL(zfs_domount);
1239
1240 /*
1241 * Called when an unmount is requested and certain sanity checks have
1242 * already passed. At this point no dentries or inodes have been reclaimed
1243 * from their respective caches. We drop the extra reference on the .zfs
1244 * control directory to allow everything to be reclaimed. All snapshots
1245 * must already have been unmounted to reach this point.
1246 */
1247 void
1248 zfs_preumount(struct super_block *sb)
1249 {
1250 zfs_sb_t *zsb = sb->s_fs_info;
1251
1252 if (zsb != NULL && zsb->z_ctldir != NULL)
1253 zfsctl_destroy(zsb);
1254 }
1255 EXPORT_SYMBOL(zfs_preumount);
1256
1257 /*
1258 * Called once all other unmount released tear down has occurred.
1259 * It is our responsibility to release any remaining infrastructure.
1260 */
1261 /*ARGSUSED*/
1262 int
1263 zfs_umount(struct super_block *sb)
1264 {
1265 zfs_sb_t *zsb = sb->s_fs_info;
1266 objset_t *os;
1267
1268 VERIFY(zfs_sb_teardown(zsb, B_TRUE) == 0);
1269 os = zsb->z_os;
1270
1271 #ifdef HAVE_BDI
1272 bdi_destroy(sb->s_bdi);
1273 #endif /* HAVE_BDI */
1274
1275 /*
1276 * z_os will be NULL if there was an error in
1277 * attempting to reopen zsb.
1278 */
1279 if (os != NULL) {
1280 /*
1281 * Unset the objset user_ptr.
1282 */
1283 mutex_enter(&os->os_user_ptr_lock);
1284 dmu_objset_set_user(os, NULL);
1285 mutex_exit(&os->os_user_ptr_lock);
1286
1287 /*
1288 * Finally release the objset
1289 */
1290 dmu_objset_disown(os, zsb);
1291 }
1292
1293 zfs_sb_free(zsb);
1294 return (0);
1295 }
1296 EXPORT_SYMBOL(zfs_umount);
1297
1298 int
1299 zfs_remount(struct super_block *sb, int *flags, char *data)
1300 {
1301 /*
1302 * All namespace flags (MNT_*) and super block flags (MS_*) will
1303 * be handled by the Linux VFS. Only handle custom options here.
1304 */
1305 return (0);
1306 }
1307 EXPORT_SYMBOL(zfs_remount);
1308
1309 int
1310 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1311 {
1312 zfs_sb_t *zsb = sb->s_fs_info;
1313 znode_t *zp;
1314 uint64_t object = 0;
1315 uint64_t fid_gen = 0;
1316 uint64_t gen_mask;
1317 uint64_t zp_gen;
1318 int i, err;
1319
1320 *ipp = NULL;
1321
1322 ZFS_ENTER(zsb);
1323
1324 if (fidp->fid_len == LONG_FID_LEN) {
1325 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1326 uint64_t objsetid = 0;
1327 uint64_t setgen = 0;
1328
1329 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1330 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1331
1332 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1333 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1334
1335 ZFS_EXIT(zsb);
1336
1337 err = zfsctl_lookup_objset(sb, objsetid, &zsb);
1338 if (err)
1339 return (EINVAL);
1340
1341 ZFS_ENTER(zsb);
1342 }
1343
1344 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1345 zfid_short_t *zfid = (zfid_short_t *)fidp;
1346
1347 for (i = 0; i < sizeof (zfid->zf_object); i++)
1348 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1349
1350 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1351 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1352 } else {
1353 ZFS_EXIT(zsb);
1354 return (EINVAL);
1355 }
1356
1357 /* A zero fid_gen means we are in the .zfs control directories */
1358 if (fid_gen == 0 &&
1359 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1360 *ipp = zsb->z_ctldir;
1361 ASSERT(*ipp != NULL);
1362 if (object == ZFSCTL_INO_SNAPDIR) {
1363 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1364 0, kcred, NULL, NULL) == 0);
1365 } else {
1366 igrab(*ipp);
1367 }
1368 ZFS_EXIT(zsb);
1369 return (0);
1370 }
1371
1372 gen_mask = -1ULL >> (64 - 8 * i);
1373
1374 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1375 if ((err = zfs_zget(zsb, object, &zp))) {
1376 ZFS_EXIT(zsb);
1377 return (err);
1378 }
1379 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb), &zp_gen,
1380 sizeof (uint64_t));
1381 zp_gen = zp_gen & gen_mask;
1382 if (zp_gen == 0)
1383 zp_gen = 1;
1384 if (zp->z_unlinked || zp_gen != fid_gen) {
1385 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1386 iput(ZTOI(zp));
1387 ZFS_EXIT(zsb);
1388 return (EINVAL);
1389 }
1390
1391 *ipp = ZTOI(zp);
1392 if (*ipp)
1393 zfs_inode_update(ITOZ(*ipp));
1394
1395 ZFS_EXIT(zsb);
1396 return (0);
1397 }
1398 EXPORT_SYMBOL(zfs_vget);
1399
1400 /*
1401 * Block out VFS ops and close zfs_sb_t
1402 *
1403 * Note, if successful, then we return with the 'z_teardown_lock' and
1404 * 'z_teardown_inactive_lock' write held.
1405 */
1406 int
1407 zfs_suspend_fs(zfs_sb_t *zsb)
1408 {
1409 int error;
1410
1411 if ((error = zfs_sb_teardown(zsb, B_FALSE)) != 0)
1412 return (error);
1413
1414 dmu_objset_disown(zsb->z_os, zsb);
1415
1416 return (0);
1417 }
1418 EXPORT_SYMBOL(zfs_suspend_fs);
1419
1420 /*
1421 * Reopen zfs_sb_t and release VFS ops.
1422 */
1423 int
1424 zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
1425 {
1426 int err, err2;
1427
1428 ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
1429 ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));
1430
1431 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zsb, &zsb->z_os);
1432 if (err) {
1433 zsb->z_os = NULL;
1434 } else {
1435 znode_t *zp;
1436 uint64_t sa_obj = 0;
1437
1438 err2 = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
1439 ZFS_SA_ATTRS, 8, 1, &sa_obj);
1440
1441 if ((err || err2) && zsb->z_version >= ZPL_VERSION_SA)
1442 goto bail;
1443
1444
1445 if ((err = sa_setup(zsb->z_os, sa_obj,
1446 zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
1447 goto bail;
1448
1449 VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
1450 zsb->z_rollback_time = jiffies;
1451
1452 /*
1453 * Attempt to re-establish all the active inodes with their
1454 * dbufs. If a zfs_rezget() fails, then we unhash the inode
1455 * and mark it stale. This prevents a collision if a new
1456 * inode/object is created which must use the same inode
1457 * number. The stale inode will be be released when the
1458 * VFS prunes the dentry holding the remaining references
1459 * on the stale inode.
1460 */
1461 mutex_enter(&zsb->z_znodes_lock);
1462 for (zp = list_head(&zsb->z_all_znodes); zp;
1463 zp = list_next(&zsb->z_all_znodes, zp)) {
1464 err2 = zfs_rezget(zp);
1465 if (err2) {
1466 remove_inode_hash(ZTOI(zp));
1467 zp->z_is_stale = B_TRUE;
1468 }
1469 }
1470 mutex_exit(&zsb->z_znodes_lock);
1471 }
1472
1473 bail:
1474 /* release the VFS ops */
1475 rw_exit(&zsb->z_teardown_inactive_lock);
1476 rrw_exit(&zsb->z_teardown_lock, FTAG);
1477
1478 if (err) {
1479 /*
1480 * Since we couldn't reopen zfs_sb_t, force
1481 * unmount this file system.
1482 */
1483 (void) zfs_umount(zsb->z_sb);
1484 }
1485 return (err);
1486 }
1487 EXPORT_SYMBOL(zfs_resume_fs);
1488
1489 int
1490 zfs_set_version(zfs_sb_t *zsb, uint64_t newvers)
1491 {
1492 int error;
1493 objset_t *os = zsb->z_os;
1494 dmu_tx_t *tx;
1495
1496 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1497 return (EINVAL);
1498
1499 if (newvers < zsb->z_version)
1500 return (EINVAL);
1501
1502 if (zfs_spa_version_map(newvers) >
1503 spa_version(dmu_objset_spa(zsb->z_os)))
1504 return (ENOTSUP);
1505
1506 tx = dmu_tx_create(os);
1507 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1508 if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
1509 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1510 ZFS_SA_ATTRS);
1511 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1512 }
1513 error = dmu_tx_assign(tx, TXG_WAIT);
1514 if (error) {
1515 dmu_tx_abort(tx);
1516 return (error);
1517 }
1518
1519 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1520 8, 1, &newvers, tx);
1521
1522 if (error) {
1523 dmu_tx_commit(tx);
1524 return (error);
1525 }
1526
1527 if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
1528 uint64_t sa_obj;
1529
1530 ASSERT3U(spa_version(dmu_objset_spa(zsb->z_os)), >=,
1531 SPA_VERSION_SA);
1532 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1533 DMU_OT_NONE, 0, tx);
1534
1535 error = zap_add(os, MASTER_NODE_OBJ,
1536 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1537 ASSERT3U(error, ==, 0);
1538
1539 VERIFY(0 == sa_set_sa_object(os, sa_obj));
1540 sa_register_update_callback(os, zfs_sa_upgrade);
1541 }
1542
1543 spa_history_log_internal(LOG_DS_UPGRADE,
1544 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
1545 zsb->z_version, newvers, dmu_objset_id(os));
1546
1547 dmu_tx_commit(tx);
1548
1549 zsb->z_version = newvers;
1550
1551 if (zsb->z_version >= ZPL_VERSION_FUID)
1552 zfs_set_fuid_feature(zsb);
1553
1554 return (0);
1555 }
1556 EXPORT_SYMBOL(zfs_set_version);
1557
1558 /*
1559 * Read a property stored within the master node.
1560 */
1561 int
1562 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
1563 {
1564 const char *pname;
1565 int error = ENOENT;
1566
1567 /*
1568 * Look up the file system's value for the property. For the
1569 * version property, we look up a slightly different string.
1570 */
1571 if (prop == ZFS_PROP_VERSION)
1572 pname = ZPL_VERSION_STR;
1573 else
1574 pname = zfs_prop_to_name(prop);
1575
1576 if (os != NULL)
1577 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
1578
1579 if (error == ENOENT) {
1580 /* No value set, use the default value */
1581 switch (prop) {
1582 case ZFS_PROP_VERSION:
1583 *value = ZPL_VERSION;
1584 break;
1585 case ZFS_PROP_NORMALIZE:
1586 case ZFS_PROP_UTF8ONLY:
1587 *value = 0;
1588 break;
1589 case ZFS_PROP_CASE:
1590 *value = ZFS_CASE_SENSITIVE;
1591 break;
1592 default:
1593 return (error);
1594 }
1595 error = 0;
1596 }
1597 return (error);
1598 }
1599 EXPORT_SYMBOL(zfs_get_zplprop);
1600
1601 void
1602 zfs_init(void)
1603 {
1604 zfsctl_init();
1605 zfs_znode_init();
1606 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
1607 register_filesystem(&zpl_fs_type);
1608 (void) arc_add_prune_callback(zpl_prune_sbs, NULL);
1609 }
1610
1611 void
1612 zfs_fini(void)
1613 {
1614 unregister_filesystem(&zpl_fs_type);
1615 zfs_znode_fini();
1616 zfsctl_fini();
1617 }