1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
25 */
26
27 /* Portions Copyright 2010 Robert Milkowski */
28
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
33 #include <sys/kmem.h>
34 #include <sys/pathname.h>
35 #include <sys/vnode.h>
36 #include <sys/vfs.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/mntent.h>
39 #include <sys/mount.h>
40 #include <sys/cmn_err.h>
41 #include "fs/fs_subr.h"
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_dir.h>
44 #include <sys/zil.h>
45 #include <sys/fs/zfs.h>
46 #include <sys/dmu.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_dataset.h>
49 #include <sys/dsl_deleg.h>
50 #include <sys/spa.h>
51 #include <sys/zap.h>
52 #include <sys/sa.h>
53 #include <sys/sa_impl.h>
54 #include <sys/varargs.h>
55 #include <sys/policy.h>
56 #include <sys/atomic.h>
57 #include <sys/mkdev.h>
58 #include <sys/modctl.h>
59 #include <sys/refstr.h>
60 #include <sys/zfs_ioctl.h>
61 #include <sys/zfs_ctldir.h>
62 #include <sys/zfs_fuid.h>
63 #include <sys/bootconf.h>
64 #include <sys/sunddi.h>
65 #include <sys/dnlc.h>
66 #include <sys/dmu_objset.h>
67 #include <sys/spa_boot.h>
68 #include "zfs_comutil.h"
69
70 int zfsfstype;
71 vfsops_t *zfs_vfsops = NULL;
72 static major_t zfs_major;
73 static minor_t zfs_minor;
74 static kmutex_t zfs_dev_mtx;
75
76 extern int sys_shutdown;
77
78 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
79 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
80 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
81 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
82 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
83 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
84 static void zfs_freevfs(vfs_t *vfsp);
85
86 static const fs_operation_def_t zfs_vfsops_template[] = {
87 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
88 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
89 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
90 VFSNAME_ROOT, { .vfs_root = zfs_root },
91 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
92 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
93 VFSNAME_VGET, { .vfs_vget = zfs_vget },
94 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
95 NULL, NULL
96 };
97
98 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
99 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
100 NULL, NULL
101 };
102
103 /*
104 * We need to keep a count of active fs's.
105 * This is necessary to prevent our module
106 * from being unloaded after a umount -f
107 */
108 static uint32_t zfs_active_fs_count = 0;
109
110 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
111 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
112 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
113 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
114
115 /*
116 * MO_DEFAULT is not used since the default value is determined
117 * by the equivalent property.
118 */
119 static mntopt_t mntopts[] = {
120 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
121 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
122 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
123 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
124 };
125
126 static mntopts_t zfs_mntopts = {
127 sizeof (mntopts) / sizeof (mntopt_t),
128 mntopts
129 };
130
131 /*ARGSUSED*/
132 int
133 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
134 {
135 /*
136 * Data integrity is job one. We don't want a compromised kernel
137 * writing to the storage pool, so we never sync during panic.
138 */
139 if (panicstr)
140 return (0);
141
142 /*
143 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
144 * to sync metadata, which they would otherwise cache indefinitely.
145 * Semantically, the only requirement is that the sync be initiated.
146 * The DMU syncs out txgs frequently, so there's nothing to do.
147 */
148 if (flag & SYNC_ATTR)
149 return (0);
150
151 if (vfsp != NULL) {
152 /*
153 * Sync a specific filesystem.
154 */
155 zfsvfs_t *zfsvfs = vfsp->vfs_data;
156 dsl_pool_t *dp;
157
158 ZFS_ENTER(zfsvfs);
159 dp = dmu_objset_pool(zfsvfs->z_os);
160
161 /*
162 * If the system is shutting down, then skip any
163 * filesystems which may exist on a suspended pool.
164 */
165 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
166 ZFS_EXIT(zfsvfs);
167 return (0);
168 }
169
170 if (zfsvfs->z_log != NULL)
171 zil_commit(zfsvfs->z_log, 0);
172
173 ZFS_EXIT(zfsvfs);
174 } else {
175 /*
176 * Sync all ZFS filesystems. This is what happens when you
177 * run sync(1M). Unlike other filesystems, ZFS honors the
178 * request by waiting for all pools to commit all dirty data.
179 */
180 spa_sync_allpools();
181 }
182
183 return (0);
184 }
185
186 static int
187 zfs_create_unique_device(dev_t *dev)
188 {
189 major_t new_major;
190
191 do {
192 ASSERT3U(zfs_minor, <=, MAXMIN32);
193 minor_t start = zfs_minor;
194 do {
195 mutex_enter(&zfs_dev_mtx);
196 if (zfs_minor >= MAXMIN32) {
197 /*
198 * If we're still using the real major
199 * keep out of /dev/zfs and /dev/zvol minor
200 * number space. If we're using a getudev()'ed
201 * major number, we can use all of its minors.
202 */
203 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
204 zfs_minor = ZFS_MIN_MINOR;
205 else
206 zfs_minor = 0;
207 } else {
208 zfs_minor++;
209 }
210 *dev = makedevice(zfs_major, zfs_minor);
211 mutex_exit(&zfs_dev_mtx);
212 } while (vfs_devismounted(*dev) && zfs_minor != start);
213 if (zfs_minor == start) {
214 /*
215 * We are using all ~262,000 minor numbers for the
216 * current major number. Create a new major number.
217 */
218 if ((new_major = getudev()) == (major_t)-1) {
219 cmn_err(CE_WARN,
220 "zfs_mount: Can't get unique major "
221 "device number.");
222 return (-1);
223 }
224 mutex_enter(&zfs_dev_mtx);
225 zfs_major = new_major;
226 zfs_minor = 0;
227
228 mutex_exit(&zfs_dev_mtx);
229 } else {
230 break;
231 }
232 /* CONSTANTCONDITION */
233 } while (1);
234
235 return (0);
236 }
237
238 static void
239 atime_changed_cb(void *arg, uint64_t newval)
240 {
241 zfsvfs_t *zfsvfs = arg;
242
243 if (newval == TRUE) {
244 zfsvfs->z_atime = TRUE;
245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
247 } else {
248 zfsvfs->z_atime = FALSE;
249 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
250 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
251 }
252 }
253
254 static void
255 xattr_changed_cb(void *arg, uint64_t newval)
256 {
257 zfsvfs_t *zfsvfs = arg;
258
259 if (newval == TRUE) {
260 /* XXX locking on vfs_flag? */
261 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
262 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
263 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
264 } else {
265 /* XXX locking on vfs_flag? */
266 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
267 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
268 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
269 }
270 }
271
272 static void
273 blksz_changed_cb(void *arg, uint64_t newval)
274 {
275 zfsvfs_t *zfsvfs = arg;
276 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
277 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
278 ASSERT(ISP2(newval));
279
280 zfsvfs->z_max_blksz = newval;
281 zfsvfs->z_vfs->vfs_bsize = newval;
282 }
283
284 static void
285 readonly_changed_cb(void *arg, uint64_t newval)
286 {
287 zfsvfs_t *zfsvfs = arg;
288
289 if (newval) {
290 /* XXX locking on vfs_flag? */
291 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
292 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
293 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
294 } else {
295 /* XXX locking on vfs_flag? */
296 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
299 }
300 }
301
302 static void
303 devices_changed_cb(void *arg, uint64_t newval)
304 {
305 zfsvfs_t *zfsvfs = arg;
306
307 if (newval == FALSE) {
308 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
309 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
310 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
311 } else {
312 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
313 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
314 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
315 }
316 }
317
318 static void
319 setuid_changed_cb(void *arg, uint64_t newval)
320 {
321 zfsvfs_t *zfsvfs = arg;
322
323 if (newval == FALSE) {
324 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
325 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
326 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
327 } else {
328 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
329 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
330 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
331 }
332 }
333
334 static void
335 exec_changed_cb(void *arg, uint64_t newval)
336 {
337 zfsvfs_t *zfsvfs = arg;
338
339 if (newval == FALSE) {
340 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
341 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
342 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
343 } else {
344 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
347 }
348 }
349
350 /*
351 * The nbmand mount option can be changed at mount time.
352 * We can't allow it to be toggled on live file systems or incorrect
353 * behavior may be seen from cifs clients
354 *
355 * This property isn't registered via dsl_prop_register(), but this callback
356 * will be called when a file system is first mounted
357 */
358 static void
359 nbmand_changed_cb(void *arg, uint64_t newval)
360 {
361 zfsvfs_t *zfsvfs = arg;
362 if (newval == FALSE) {
363 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
364 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
365 } else {
366 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
367 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
368 }
369 }
370
371 static void
372 snapdir_changed_cb(void *arg, uint64_t newval)
373 {
374 zfsvfs_t *zfsvfs = arg;
375
376 zfsvfs->z_show_ctldir = newval;
377 }
378
379 static void
380 vscan_changed_cb(void *arg, uint64_t newval)
381 {
382 zfsvfs_t *zfsvfs = arg;
383
384 zfsvfs->z_vscan = newval;
385 }
386
387 static void
388 acl_mode_changed_cb(void *arg, uint64_t newval)
389 {
390 zfsvfs_t *zfsvfs = arg;
391
392 zfsvfs->z_acl_mode = newval;
393 }
394
395 static void
396 acl_inherit_changed_cb(void *arg, uint64_t newval)
397 {
398 zfsvfs_t *zfsvfs = arg;
399
400 zfsvfs->z_acl_inherit = newval;
401 }
402
403 static int
404 zfs_register_callbacks(vfs_t *vfsp)
405 {
406 struct dsl_dataset *ds = NULL;
407 objset_t *os = NULL;
408 zfsvfs_t *zfsvfs = NULL;
409 uint64_t nbmand;
410 boolean_t readonly = B_FALSE;
411 boolean_t do_readonly = B_FALSE;
412 boolean_t setuid = B_FALSE;
413 boolean_t do_setuid = B_FALSE;
414 boolean_t exec = B_FALSE;
415 boolean_t do_exec = B_FALSE;
416 boolean_t devices = B_FALSE;
417 boolean_t do_devices = B_FALSE;
418 boolean_t xattr = B_FALSE;
419 boolean_t do_xattr = B_FALSE;
420 boolean_t atime = B_FALSE;
421 boolean_t do_atime = B_FALSE;
422 int error = 0;
423
424 ASSERT(vfsp);
425 zfsvfs = vfsp->vfs_data;
426 ASSERT(zfsvfs);
427 os = zfsvfs->z_os;
428
429 /*
430 * The act of registering our callbacks will destroy any mount
431 * options we may have. In order to enable temporary overrides
432 * of mount options, we stash away the current values and
433 * restore them after we register the callbacks.
434 */
435 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
436 !spa_writeable(dmu_objset_spa(os))) {
437 readonly = B_TRUE;
438 do_readonly = B_TRUE;
439 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
440 readonly = B_FALSE;
441 do_readonly = B_TRUE;
442 }
443 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
444 devices = B_FALSE;
445 setuid = B_FALSE;
446 do_devices = B_TRUE;
447 do_setuid = B_TRUE;
448 } else {
449 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
450 devices = B_FALSE;
451 do_devices = B_TRUE;
452 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
453 devices = B_TRUE;
454 do_devices = B_TRUE;
455 }
456
457 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
458 setuid = B_FALSE;
459 do_setuid = B_TRUE;
460 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
461 setuid = B_TRUE;
462 do_setuid = B_TRUE;
463 }
464 }
465 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
466 exec = B_FALSE;
467 do_exec = B_TRUE;
468 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
469 exec = B_TRUE;
470 do_exec = B_TRUE;
471 }
472 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
473 xattr = B_FALSE;
474 do_xattr = B_TRUE;
475 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
476 xattr = B_TRUE;
477 do_xattr = B_TRUE;
478 }
479 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
480 atime = B_FALSE;
481 do_atime = B_TRUE;
482 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
483 atime = B_TRUE;
484 do_atime = B_TRUE;
485 }
486
487 /*
488 * nbmand is a special property. It can only be changed at
489 * mount time.
490 *
491 * This is weird, but it is documented to only be changeable
492 * at mount time.
493 */
494 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
495 nbmand = B_FALSE;
496 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
497 nbmand = B_TRUE;
498 } else {
499 char osname[ZFS_MAX_DATASET_NAME_LEN];
500
501 dmu_objset_name(os, osname);
502 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
503 NULL)) {
504 return (error);
505 }
506 }
507
508 /*
509 * Register property callbacks.
510 *
511 * It would probably be fine to just check for i/o error from
512 * the first prop_register(), but I guess I like to go
513 * overboard...
514 */
515 ds = dmu_objset_ds(os);
516 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
517 error = dsl_prop_register(ds,
518 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
519 error = error ? error : dsl_prop_register(ds,
520 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
521 error = error ? error : dsl_prop_register(ds,
522 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
523 error = error ? error : dsl_prop_register(ds,
524 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
525 error = error ? error : dsl_prop_register(ds,
526 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
527 error = error ? error : dsl_prop_register(ds,
528 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
529 error = error ? error : dsl_prop_register(ds,
530 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
531 error = error ? error : dsl_prop_register(ds,
532 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
533 error = error ? error : dsl_prop_register(ds,
534 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
535 error = error ? error : dsl_prop_register(ds,
536 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
537 zfsvfs);
538 error = error ? error : dsl_prop_register(ds,
539 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
540 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
541 if (error)
542 goto unregister;
543
544 /*
545 * Invoke our callbacks to restore temporary mount options.
546 */
547 if (do_readonly)
548 readonly_changed_cb(zfsvfs, readonly);
549 if (do_setuid)
550 setuid_changed_cb(zfsvfs, setuid);
551 if (do_exec)
552 exec_changed_cb(zfsvfs, exec);
553 if (do_devices)
554 devices_changed_cb(zfsvfs, devices);
555 if (do_xattr)
556 xattr_changed_cb(zfsvfs, xattr);
557 if (do_atime)
558 atime_changed_cb(zfsvfs, atime);
559
560 nbmand_changed_cb(zfsvfs, nbmand);
561
562 return (0);
563
564 unregister:
565 dsl_prop_unregister_all(ds, zfsvfs);
566 return (error);
567 }
568
569 static int
570 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
571 uint64_t *userp, uint64_t *groupp)
572 {
573 /*
574 * Is it a valid type of object to track?
575 */
576 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
577 return (SET_ERROR(ENOENT));
578
579 /*
580 * If we have a NULL data pointer
581 * then assume the id's aren't changing and
582 * return EEXIST to the dmu to let it know to
583 * use the same ids
584 */
585 if (data == NULL)
586 return (SET_ERROR(EEXIST));
587
588 if (bonustype == DMU_OT_ZNODE) {
589 znode_phys_t *znp = data;
590 *userp = znp->zp_uid;
591 *groupp = znp->zp_gid;
592 } else {
593 int hdrsize;
594 sa_hdr_phys_t *sap = data;
595 sa_hdr_phys_t sa = *sap;
596 boolean_t swap = B_FALSE;
597
598 ASSERT(bonustype == DMU_OT_SA);
599
600 if (sa.sa_magic == 0) {
601 /*
602 * This should only happen for newly created
603 * files that haven't had the znode data filled
604 * in yet.
605 */
606 *userp = 0;
607 *groupp = 0;
608 return (0);
609 }
610 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
611 sa.sa_magic = SA_MAGIC;
612 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
613 swap = B_TRUE;
614 } else {
615 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
616 }
617
618 hdrsize = sa_hdrsize(&sa);
619 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
620 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
621 SA_UID_OFFSET));
622 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
623 SA_GID_OFFSET));
624 if (swap) {
625 *userp = BSWAP_64(*userp);
626 *groupp = BSWAP_64(*groupp);
627 }
628 }
629 return (0);
630 }
631
632 static void
633 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
634 char *domainbuf, int buflen, uid_t *ridp)
635 {
636 uint64_t fuid;
637 const char *domain;
638
639 fuid = strtonum(fuidstr, NULL);
640
641 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
642 if (domain)
643 (void) strlcpy(domainbuf, domain, buflen);
644 else
645 domainbuf[0] = '\0';
646 *ridp = FUID_RID(fuid);
647 }
648
649 static uint64_t
650 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
651 {
652 switch (type) {
653 case ZFS_PROP_USERUSED:
654 return (DMU_USERUSED_OBJECT);
655 case ZFS_PROP_GROUPUSED:
656 return (DMU_GROUPUSED_OBJECT);
657 case ZFS_PROP_USERQUOTA:
658 return (zfsvfs->z_userquota_obj);
659 case ZFS_PROP_GROUPQUOTA:
660 return (zfsvfs->z_groupquota_obj);
661 }
662 return (0);
663 }
664
665 int
666 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
667 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
668 {
669 int error;
670 zap_cursor_t zc;
671 zap_attribute_t za;
672 zfs_useracct_t *buf = vbuf;
673 uint64_t obj;
674
675 if (!dmu_objset_userspace_present(zfsvfs->z_os))
676 return (SET_ERROR(ENOTSUP));
677
678 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
679 if (obj == 0) {
680 *bufsizep = 0;
681 return (0);
682 }
683
684 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
685 (error = zap_cursor_retrieve(&zc, &za)) == 0;
686 zap_cursor_advance(&zc)) {
687 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
688 *bufsizep)
689 break;
690
691 fuidstr_to_sid(zfsvfs, za.za_name,
692 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
693
694 buf->zu_space = za.za_first_integer;
695 buf++;
696 }
697 if (error == ENOENT)
698 error = 0;
699
700 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
701 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
702 *cookiep = zap_cursor_serialize(&zc);
703 zap_cursor_fini(&zc);
704 return (error);
705 }
706
707 /*
708 * buf must be big enough (eg, 32 bytes)
709 */
710 static int
711 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
712 char *buf, boolean_t addok)
713 {
714 uint64_t fuid;
715 int domainid = 0;
716
717 if (domain && domain[0]) {
718 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
719 if (domainid == -1)
720 return (SET_ERROR(ENOENT));
721 }
722 fuid = FUID_ENCODE(domainid, rid);
723 (void) sprintf(buf, "%llx", (longlong_t)fuid);
724 return (0);
725 }
726
727 int
728 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
729 const char *domain, uint64_t rid, uint64_t *valp)
730 {
731 char buf[32];
732 int err;
733 uint64_t obj;
734
735 *valp = 0;
736
737 if (!dmu_objset_userspace_present(zfsvfs->z_os))
738 return (SET_ERROR(ENOTSUP));
739
740 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
741 if (obj == 0)
742 return (0);
743
744 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
745 if (err)
746 return (err);
747
748 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
749 if (err == ENOENT)
750 err = 0;
751 return (err);
752 }
753
754 int
755 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
756 const char *domain, uint64_t rid, uint64_t quota)
757 {
758 char buf[32];
759 int err;
760 dmu_tx_t *tx;
761 uint64_t *objp;
762 boolean_t fuid_dirtied;
763
764 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
765 return (SET_ERROR(EINVAL));
766
767 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
768 return (SET_ERROR(ENOTSUP));
769
770 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
771 &zfsvfs->z_groupquota_obj;
772
773 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
774 if (err)
775 return (err);
776 fuid_dirtied = zfsvfs->z_fuid_dirty;
777
778 tx = dmu_tx_create(zfsvfs->z_os);
779 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
780 if (*objp == 0) {
781 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
782 zfs_userquota_prop_prefixes[type]);
783 }
784 if (fuid_dirtied)
785 zfs_fuid_txhold(zfsvfs, tx);
786 err = dmu_tx_assign(tx, TXG_WAIT);
787 if (err) {
788 dmu_tx_abort(tx);
789 return (err);
790 }
791
792 mutex_enter(&zfsvfs->z_lock);
793 if (*objp == 0) {
794 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
795 DMU_OT_NONE, 0, tx);
796 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
797 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
798 }
799 mutex_exit(&zfsvfs->z_lock);
800
801 if (quota == 0) {
802 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
803 if (err == ENOENT)
804 err = 0;
805 } else {
806 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
807 }
808 ASSERT(err == 0);
809 if (fuid_dirtied)
810 zfs_fuid_sync(zfsvfs, tx);
811 dmu_tx_commit(tx);
812 return (err);
813 }
814
815 boolean_t
816 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
817 {
818 char buf[32];
819 uint64_t used, quota, usedobj, quotaobj;
820 int err;
821
822 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
823 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
824
825 if (quotaobj == 0 || zfsvfs->z_replay)
826 return (B_FALSE);
827
828 (void) sprintf(buf, "%llx", (longlong_t)fuid);
829 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
830 if (err != 0)
831 return (B_FALSE);
832
833 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
834 if (err != 0)
835 return (B_FALSE);
836 return (used >= quota);
837 }
838
839 boolean_t
840 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
841 {
842 uint64_t fuid;
843 uint64_t quotaobj;
844
845 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
846
847 fuid = isgroup ? zp->z_gid : zp->z_uid;
848
849 if (quotaobj == 0 || zfsvfs->z_replay)
850 return (B_FALSE);
851
852 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
853 }
854
855 /*
856 * Associate this zfsvfs with the given objset, which must be owned.
857 * This will cache a bunch of on-disk state from the objset in the
858 * zfsvfs.
859 */
860 static int
861 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
862 {
863 int error;
864 uint64_t val;
865
866 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
867 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
868 zfsvfs->z_os = os;
869
870 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
871 if (error != 0)
872 return (error);
873 if (zfsvfs->z_version >
874 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
875 (void) printf("Can't mount a version %lld file system "
876 "on a version %lld pool\n. Pool must be upgraded to mount "
877 "this file system.", (u_longlong_t)zfsvfs->z_version,
878 (u_longlong_t)spa_version(dmu_objset_spa(os)));
879 return (SET_ERROR(ENOTSUP));
880 }
881 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
882 if (error != 0)
883 return (error);
884 zfsvfs->z_norm = (int)val;
885
886 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
887 if (error != 0)
888 return (error);
889 zfsvfs->z_utf8 = (val != 0);
890
891 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
892 if (error != 0)
893 return (error);
894 zfsvfs->z_case = (uint_t)val;
895
896 /*
897 * Fold case on file systems that are always or sometimes case
898 * insensitive.
899 */
900 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
901 zfsvfs->z_case == ZFS_CASE_MIXED)
902 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
903
904 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
905 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
906
907 uint64_t sa_obj = 0;
908 if (zfsvfs->z_use_sa) {
909 /* should either have both of these objects or none */
910 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
911 &sa_obj);
912 if (error != 0)
913 return (error);
914 }
915
916 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
917 &zfsvfs->z_attr_table);
918 if (error != 0)
919 return (error);
920
921 if (zfsvfs->z_version >= ZPL_VERSION_SA)
922 sa_register_update_callback(os, zfs_sa_upgrade);
923
924 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
925 &zfsvfs->z_root);
926 if (error != 0)
927 return (error);
928 ASSERT(zfsvfs->z_root != 0);
929
930 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
931 &zfsvfs->z_unlinkedobj);
932 if (error != 0)
933 return (error);
934
935 error = zap_lookup(os, MASTER_NODE_OBJ,
936 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
937 8, 1, &zfsvfs->z_userquota_obj);
938 if (error == ENOENT)
939 zfsvfs->z_userquota_obj = 0;
940 else if (error != 0)
941 return (error);
942
943 error = zap_lookup(os, MASTER_NODE_OBJ,
944 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
945 8, 1, &zfsvfs->z_groupquota_obj);
946 if (error == ENOENT)
947 zfsvfs->z_groupquota_obj = 0;
948 else if (error != 0)
949 return (error);
950
951 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
952 &zfsvfs->z_fuid_obj);
953 if (error == ENOENT)
954 zfsvfs->z_fuid_obj = 0;
955 else if (error != 0)
956 return (error);
957
958 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
959 &zfsvfs->z_shares_dir);
960 if (error == ENOENT)
961 zfsvfs->z_shares_dir = 0;
962 else if (error != 0)
963 return (error);
964
965 return (0);
966 }
967
968 int
969 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
970 {
971 objset_t *os;
972 zfsvfs_t *zfsvfs;
973 int error;
974
975 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
976
977 /*
978 * We claim to always be readonly so we can open snapshots;
979 * other ZPL code will prevent us from writing to snapshots.
980 */
981 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
982 if (error) {
983 kmem_free(zfsvfs, sizeof (zfsvfs_t));
984 return (error);
985 }
986
987 zfsvfs->z_vfs = NULL;
988 zfsvfs->z_parent = zfsvfs;
989
990 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
991 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
992 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
993 offsetof(znode_t, z_link_node));
994 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
995 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
996 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
997 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
998 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
999
1000 error = zfsvfs_init(zfsvfs, os);
1001 if (error != 0) {
1002 dmu_objset_disown(os, zfsvfs);
1003 *zfvp = NULL;
1004 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1005 return (error);
1006 }
1007
1008 *zfvp = zfsvfs;
1009 return (0);
1010 }
1011
1012 static int
1013 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1014 {
1015 int error;
1016
1017 error = zfs_register_callbacks(zfsvfs->z_vfs);
1018 if (error)
1019 return (error);
1020
1021 /*
1022 * Set the objset user_ptr to track its zfsvfs.
1023 */
1024 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1025 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1026 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1027
1028 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1029
1030 /*
1031 * If we are not mounting (ie: online recv), then we don't
1032 * have to worry about replaying the log as we blocked all
1033 * operations out since we closed the ZIL.
1034 */
1035 if (mounting) {
1036 boolean_t readonly;
1037
1038 /*
1039 * During replay we remove the read only flag to
1040 * allow replays to succeed.
1041 */
1042 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1043 if (readonly != 0)
1044 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1045 else
1046 zfs_unlinked_drain(zfsvfs);
1047
1048 /*
1049 * Parse and replay the intent log.
1050 *
1051 * Because of ziltest, this must be done after
1052 * zfs_unlinked_drain(). (Further note: ziltest
1053 * doesn't use readonly mounts, where
1054 * zfs_unlinked_drain() isn't called.) This is because
1055 * ziltest causes spa_sync() to think it's committed,
1056 * but actually it is not, so the intent log contains
1057 * many txg's worth of changes.
1058 *
1059 * In particular, if object N is in the unlinked set in
1060 * the last txg to actually sync, then it could be
1061 * actually freed in a later txg and then reallocated
1062 * in a yet later txg. This would write a "create
1063 * object N" record to the intent log. Normally, this
1064 * would be fine because the spa_sync() would have
1065 * written out the fact that object N is free, before
1066 * we could write the "create object N" intent log
1067 * record.
1068 *
1069 * But when we are in ziltest mode, we advance the "open
1070 * txg" without actually spa_sync()-ing the changes to
1071 * disk. So we would see that object N is still
1072 * allocated and in the unlinked set, and there is an
1073 * intent log record saying to allocate it.
1074 */
1075 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1076 if (zil_replay_disable) {
1077 zil_destroy(zfsvfs->z_log, B_FALSE);
1078 } else {
1079 zfsvfs->z_replay = B_TRUE;
1080 zil_replay(zfsvfs->z_os, zfsvfs,
1081 zfs_replay_vector);
1082 zfsvfs->z_replay = B_FALSE;
1083 }
1084 }
1085 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1086 }
1087
1088 return (0);
1089 }
1090
1091 void
1092 zfsvfs_free(zfsvfs_t *zfsvfs)
1093 {
1094 int i;
1095 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1096
1097 /*
1098 * This is a barrier to prevent the filesystem from going away in
1099 * zfs_znode_move() until we can safely ensure that the filesystem is
1100 * not unmounted. We consider the filesystem valid before the barrier
1101 * and invalid after the barrier.
1102 */
1103 rw_enter(&zfsvfs_lock, RW_READER);
1104 rw_exit(&zfsvfs_lock);
1105
1106 zfs_fuid_destroy(zfsvfs);
1107
1108 mutex_destroy(&zfsvfs->z_znodes_lock);
1109 mutex_destroy(&zfsvfs->z_lock);
1110 list_destroy(&zfsvfs->z_all_znodes);
1111 rrm_destroy(&zfsvfs->z_teardown_lock);
1112 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1113 rw_destroy(&zfsvfs->z_fuid_lock);
1114 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1115 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1116 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1117 }
1118
1119 static void
1120 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1121 {
1122 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1123 if (zfsvfs->z_vfs) {
1124 if (zfsvfs->z_use_fuids) {
1125 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1126 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1127 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1128 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1129 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1130 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1131 } else {
1132 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1133 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1134 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1135 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1136 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1137 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1138 }
1139 }
1140 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1141 }
1142
1143 static int
1144 zfs_domount(vfs_t *vfsp, char *osname)
1145 {
1146 dev_t mount_dev;
1147 uint64_t recordsize, fsid_guid;
1148 int error = 0;
1149 zfsvfs_t *zfsvfs;
1150
1151 ASSERT(vfsp);
1152 ASSERT(osname);
1153
1154 error = zfsvfs_create(osname, &zfsvfs);
1155 if (error)
1156 return (error);
1157 zfsvfs->z_vfs = vfsp;
1158
1159 /* Initialize the generic filesystem structure. */
1160 vfsp->vfs_bcount = 0;
1161 vfsp->vfs_data = NULL;
1162
1163 if (zfs_create_unique_device(&mount_dev) == -1) {
1164 error = SET_ERROR(ENODEV);
1165 goto out;
1166 }
1167 ASSERT(vfs_devismounted(mount_dev) == 0);
1168
1169 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1170 NULL))
1171 goto out;
1172
1173 vfsp->vfs_dev = mount_dev;
1174 vfsp->vfs_fstype = zfsfstype;
1175 vfsp->vfs_bsize = recordsize;
1176 vfsp->vfs_flag |= VFS_NOTRUNC;
1177 vfsp->vfs_data = zfsvfs;
1178
1179 /*
1180 * The fsid is 64 bits, composed of an 8-bit fs type, which
1181 * separates our fsid from any other filesystem types, and a
1182 * 56-bit objset unique ID. The objset unique ID is unique to
1183 * all objsets open on this system, provided by unique_create().
1184 * The 8-bit fs type must be put in the low bits of fsid[1]
1185 * because that's where other Solaris filesystems put it.
1186 */
1187 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1188 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1189 vfsp->vfs_fsid.val[0] = fsid_guid;
1190 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1191 zfsfstype & 0xFF;
1192
1193 /*
1194 * Set features for file system.
1195 */
1196 zfs_set_fuid_feature(zfsvfs);
1197 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1198 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1199 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1200 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1201 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1202 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1203 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1204 }
1205 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1206
1207 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1208 uint64_t pval;
1209
1210 atime_changed_cb(zfsvfs, B_FALSE);
1211 readonly_changed_cb(zfsvfs, B_TRUE);
1212 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1213 goto out;
1214 xattr_changed_cb(zfsvfs, pval);
1215 zfsvfs->z_issnap = B_TRUE;
1216 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1217
1218 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1219 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1220 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1221 } else {
1222 error = zfsvfs_setup(zfsvfs, B_TRUE);
1223 }
1224
1225 if (!zfsvfs->z_issnap)
1226 zfsctl_create(zfsvfs);
1227 out:
1228 if (error) {
1229 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1230 zfsvfs_free(zfsvfs);
1231 } else {
1232 atomic_inc_32(&zfs_active_fs_count);
1233 }
1234
1235 return (error);
1236 }
1237
1238 void
1239 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1240 {
1241 objset_t *os = zfsvfs->z_os;
1242
1243 if (!dmu_objset_is_snapshot(os))
1244 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1245 }
1246
1247 /*
1248 * Convert a decimal digit string to a uint64_t integer.
1249 */
1250 static int
1251 str_to_uint64(char *str, uint64_t *objnum)
1252 {
1253 uint64_t num = 0;
1254
1255 while (*str) {
1256 if (*str < '0' || *str > '9')
1257 return (SET_ERROR(EINVAL));
1258
1259 num = num*10 + *str++ - '0';
1260 }
1261
1262 *objnum = num;
1263 return (0);
1264 }
1265
1266 /*
1267 * The boot path passed from the boot loader is in the form of
1268 * "rootpool-name/root-filesystem-object-number'. Convert this
1269 * string to a dataset name: "rootpool-name/root-filesystem-name".
1270 */
1271 static int
1272 zfs_parse_bootfs(char *bpath, char *outpath)
1273 {
1274 char *slashp;
1275 uint64_t objnum;
1276 int error;
1277
1278 if (*bpath == 0 || *bpath == '/')
1279 return (SET_ERROR(EINVAL));
1280
1281 (void) strcpy(outpath, bpath);
1282
1283 slashp = strchr(bpath, '/');
1284
1285 /* if no '/', just return the pool name */
1286 if (slashp == NULL) {
1287 return (0);
1288 }
1289
1290 /* if not a number, just return the root dataset name */
1291 if (str_to_uint64(slashp+1, &objnum)) {
1292 return (0);
1293 }
1294
1295 *slashp = '\0';
1296 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1297 *slashp = '/';
1298
1299 return (error);
1300 }
1301
1302 /*
1303 * Check that the hex label string is appropriate for the dataset being
1304 * mounted into the global_zone proper.
1305 *
1306 * Return an error if the hex label string is not default or
1307 * admin_low/admin_high. For admin_low labels, the corresponding
1308 * dataset must be readonly.
1309 */
1310 int
1311 zfs_check_global_label(const char *dsname, const char *hexsl)
1312 {
1313 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1314 return (0);
1315 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1316 return (0);
1317 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1318 /* must be readonly */
1319 uint64_t rdonly;
1320
1321 if (dsl_prop_get_integer(dsname,
1322 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1323 return (SET_ERROR(EACCES));
1324 return (rdonly ? 0 : EACCES);
1325 }
1326 return (SET_ERROR(EACCES));
1327 }
1328
1329 /*
1330 * Determine whether the mount is allowed according to MAC check.
1331 * by comparing (where appropriate) label of the dataset against
1332 * the label of the zone being mounted into. If the dataset has
1333 * no label, create one.
1334 *
1335 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1336 */
1337 static int
1338 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1339 {
1340 int error, retv;
1341 zone_t *mntzone = NULL;
1342 ts_label_t *mnt_tsl;
1343 bslabel_t *mnt_sl;
1344 bslabel_t ds_sl;
1345 char ds_hexsl[MAXNAMELEN];
1346
1347 retv = EACCES; /* assume the worst */
1348
1349 /*
1350 * Start by getting the dataset label if it exists.
1351 */
1352 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1353 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1354 if (error)
1355 return (SET_ERROR(EACCES));
1356
1357 /*
1358 * If labeling is NOT enabled, then disallow the mount of datasets
1359 * which have a non-default label already. No other label checks
1360 * are needed.
1361 */
1362 if (!is_system_labeled()) {
1363 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1364 return (0);
1365 return (SET_ERROR(EACCES));
1366 }
1367
1368 /*
1369 * Get the label of the mountpoint. If mounting into the global
1370 * zone (i.e. mountpoint is not within an active zone and the
1371 * zoned property is off), the label must be default or
1372 * admin_low/admin_high only; no other checks are needed.
1373 */
1374 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1375 if (mntzone->zone_id == GLOBAL_ZONEID) {
1376 uint64_t zoned;
1377
1378 zone_rele(mntzone);
1379
1380 if (dsl_prop_get_integer(osname,
1381 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1382 return (SET_ERROR(EACCES));
1383 if (!zoned)
1384 return (zfs_check_global_label(osname, ds_hexsl));
1385 else
1386 /*
1387 * This is the case of a zone dataset being mounted
1388 * initially, before the zone has been fully created;
1389 * allow this mount into global zone.
1390 */
1391 return (0);
1392 }
1393
1394 mnt_tsl = mntzone->zone_slabel;
1395 ASSERT(mnt_tsl != NULL);
1396 label_hold(mnt_tsl);
1397 mnt_sl = label2bslabel(mnt_tsl);
1398
1399 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1400 /*
1401 * The dataset doesn't have a real label, so fabricate one.
1402 */
1403 char *str = NULL;
1404
1405 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1406 dsl_prop_set_string(osname,
1407 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1408 ZPROP_SRC_LOCAL, str) == 0)
1409 retv = 0;
1410 if (str != NULL)
1411 kmem_free(str, strlen(str) + 1);
1412 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1413 /*
1414 * Now compare labels to complete the MAC check. If the
1415 * labels are equal then allow access. If the mountpoint
1416 * label dominates the dataset label, allow readonly access.
1417 * Otherwise, access is denied.
1418 */
1419 if (blequal(mnt_sl, &ds_sl))
1420 retv = 0;
1421 else if (bldominates(mnt_sl, &ds_sl)) {
1422 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1423 retv = 0;
1424 }
1425 }
1426
1427 label_rele(mnt_tsl);
1428 zone_rele(mntzone);
1429 return (retv);
1430 }
1431
1432 static int
1433 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1434 {
1435 int error = 0;
1436 static int zfsrootdone = 0;
1437 zfsvfs_t *zfsvfs = NULL;
1438 znode_t *zp = NULL;
1439 vnode_t *vp = NULL;
1440 char *zfs_bootfs;
1441 char *zfs_devid;
1442
1443 ASSERT(vfsp);
1444
1445 /*
1446 * The filesystem that we mount as root is defined in the
1447 * boot property "zfs-bootfs" with a format of
1448 * "poolname/root-dataset-objnum".
1449 */
1450 if (why == ROOT_INIT) {
1451 if (zfsrootdone++)
1452 return (SET_ERROR(EBUSY));
1453 /*
1454 * the process of doing a spa_load will require the
1455 * clock to be set before we could (for example) do
1456 * something better by looking at the timestamp on
1457 * an uberblock, so just set it to -1.
1458 */
1459 clkset(-1);
1460
1461 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1462 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1463 "bootfs name");
1464 return (SET_ERROR(EINVAL));
1465 }
1466 zfs_devid = spa_get_bootprop("diskdevid");
1467 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1468 if (zfs_devid)
1469 spa_free_bootprop(zfs_devid);
1470 if (error) {
1471 spa_free_bootprop(zfs_bootfs);
1472 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1473 error);
1474 return (error);
1475 }
1476 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1477 spa_free_bootprop(zfs_bootfs);
1478 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1479 error);
1480 return (error);
1481 }
1482
1483 spa_free_bootprop(zfs_bootfs);
1484
1485 if (error = vfs_lock(vfsp))
1486 return (error);
1487
1488 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1489 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1490 goto out;
1491 }
1492
1493 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1494 ASSERT(zfsvfs);
1495 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1496 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1497 goto out;
1498 }
1499
1500 vp = ZTOV(zp);
1501 mutex_enter(&vp->v_lock);
1502 vp->v_flag |= VROOT;
1503 mutex_exit(&vp->v_lock);
1504 rootvp = vp;
1505
1506 /*
1507 * Leave rootvp held. The root file system is never unmounted.
1508 */
1509
1510 vfs_add((struct vnode *)0, vfsp,
1511 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1512 out:
1513 vfs_unlock(vfsp);
1514 return (error);
1515 } else if (why == ROOT_REMOUNT) {
1516 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1517 vfsp->vfs_flag |= VFS_REMOUNT;
1518
1519 /* refresh mount options */
1520 zfs_unregister_callbacks(vfsp->vfs_data);
1521 return (zfs_register_callbacks(vfsp));
1522
1523 } else if (why == ROOT_UNMOUNT) {
1524 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1525 (void) zfs_sync(vfsp, 0, 0);
1526 return (0);
1527 }
1528
1529 /*
1530 * if "why" is equal to anything else other than ROOT_INIT,
1531 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1532 */
1533 return (SET_ERROR(ENOTSUP));
1534 }
1535
1536 /*ARGSUSED*/
1537 static int
1538 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1539 {
1540 char *osname;
1541 pathname_t spn;
1542 int error = 0;
1543 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1544 UIO_SYSSPACE : UIO_USERSPACE;
1545 int canwrite;
1546
1547 if (mvp->v_type != VDIR)
1548 return (SET_ERROR(ENOTDIR));
1549
1550 mutex_enter(&mvp->v_lock);
1551 if ((uap->flags & MS_REMOUNT) == 0 &&
1552 (uap->flags & MS_OVERLAY) == 0 &&
1553 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1554 mutex_exit(&mvp->v_lock);
1555 return (SET_ERROR(EBUSY));
1556 }
1557 mutex_exit(&mvp->v_lock);
1558
1559 /*
1560 * ZFS does not support passing unparsed data in via MS_DATA.
1561 * Users should use the MS_OPTIONSTR interface; this means
1562 * that all option parsing is already done and the options struct
1563 * can be interrogated.
1564 */
1565 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1566 return (SET_ERROR(EINVAL));
1567
1568 /*
1569 * Get the objset name (the "special" mount argument).
1570 */
1571 if (error = pn_get(uap->spec, fromspace, &spn))
1572 return (error);
1573
1574 osname = spn.pn_path;
1575
1576 /*
1577 * Check for mount privilege?
1578 *
1579 * If we don't have privilege then see if
1580 * we have local permission to allow it
1581 */
1582 error = secpolicy_fs_mount(cr, mvp, vfsp);
1583 if (error) {
1584 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1585 vattr_t vattr;
1586
1587 /*
1588 * Make sure user is the owner of the mount point
1589 * or has sufficient privileges.
1590 */
1591
1592 vattr.va_mask = AT_UID;
1593
1594 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1595 goto out;
1596 }
1597
1598 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1599 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1600 goto out;
1601 }
1602 secpolicy_fs_mount_clearopts(cr, vfsp);
1603 } else {
1604 goto out;
1605 }
1606 }
1607
1608 /*
1609 * Refuse to mount a filesystem if we are in a local zone and the
1610 * dataset is not visible.
1611 */
1612 if (!INGLOBALZONE(curproc) &&
1613 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1614 error = SET_ERROR(EPERM);
1615 goto out;
1616 }
1617
1618 error = zfs_mount_label_policy(vfsp, osname);
1619 if (error)
1620 goto out;
1621
1622 /*
1623 * When doing a remount, we simply refresh our temporary properties
1624 * according to those options set in the current VFS options.
1625 */
1626 if (uap->flags & MS_REMOUNT) {
1627 /* refresh mount options */
1628 zfs_unregister_callbacks(vfsp->vfs_data);
1629 error = zfs_register_callbacks(vfsp);
1630 goto out;
1631 }
1632
1633 error = zfs_domount(vfsp, osname);
1634
1635 /*
1636 * Add an extra VFS_HOLD on our parent vfs so that it can't
1637 * disappear due to a forced unmount.
1638 */
1639 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1640 VFS_HOLD(mvp->v_vfsp);
1641
1642 out:
1643 pn_free(&spn);
1644 return (error);
1645 }
1646
1647 static int
1648 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1649 {
1650 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1651 dev32_t d32;
1652 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1653
1654 ZFS_ENTER(zfsvfs);
1655
1656 dmu_objset_space(zfsvfs->z_os,
1657 &refdbytes, &availbytes, &usedobjs, &availobjs);
1658
1659 /*
1660 * The underlying storage pool actually uses multiple block sizes.
1661 * We report the fragsize as the smallest block size we support,
1662 * and we report our blocksize as the filesystem's maximum blocksize.
1663 */
1664 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1665 statp->f_bsize = zfsvfs->z_max_blksz;
1666
1667 /*
1668 * The following report "total" blocks of various kinds in the
1669 * file system, but reported in terms of f_frsize - the
1670 * "fragment" size.
1671 */
1672
1673 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1674 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1675 statp->f_bavail = statp->f_bfree; /* no root reservation */
1676
1677 /*
1678 * statvfs() should really be called statufs(), because it assumes
1679 * static metadata. ZFS doesn't preallocate files, so the best
1680 * we can do is report the max that could possibly fit in f_files,
1681 * and that minus the number actually used in f_ffree.
1682 * For f_ffree, report the smaller of the number of object available
1683 * and the number of blocks (each object will take at least a block).
1684 */
1685 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1686 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1687 statp->f_files = statp->f_ffree + usedobjs;
1688
1689 (void) cmpldev(&d32, vfsp->vfs_dev);
1690 statp->f_fsid = d32;
1691
1692 /*
1693 * We're a zfs filesystem.
1694 */
1695 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1696
1697 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1698
1699 statp->f_namemax = MAXNAMELEN - 1;
1700
1701 /*
1702 * We have all of 32 characters to stuff a string here.
1703 * Is there anything useful we could/should provide?
1704 */
1705 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1706
1707 ZFS_EXIT(zfsvfs);
1708 return (0);
1709 }
1710
1711 static int
1712 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1713 {
1714 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1715 znode_t *rootzp;
1716 int error;
1717
1718 ZFS_ENTER(zfsvfs);
1719
1720 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1721 if (error == 0)
1722 *vpp = ZTOV(rootzp);
1723
1724 ZFS_EXIT(zfsvfs);
1725 return (error);
1726 }
1727
1728 /*
1729 * Teardown the zfsvfs::z_os.
1730 *
1731 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1732 * and 'z_teardown_inactive_lock' held.
1733 */
1734 static int
1735 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1736 {
1737 znode_t *zp;
1738
1739 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1740
1741 if (!unmounting) {
1742 /*
1743 * We purge the parent filesystem's vfsp as the parent
1744 * filesystem and all of its snapshots have their vnode's
1745 * v_vfsp set to the parent's filesystem's vfsp. Note,
1746 * 'z_parent' is self referential for non-snapshots.
1747 */
1748 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1749 }
1750
1751 /*
1752 * Close the zil. NB: Can't close the zil while zfs_inactive
1753 * threads are blocked as zil_close can call zfs_inactive.
1754 */
1755 if (zfsvfs->z_log) {
1756 zil_close(zfsvfs->z_log);
1757 zfsvfs->z_log = NULL;
1758 }
1759
1760 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1761
1762 /*
1763 * If we are not unmounting (ie: online recv) and someone already
1764 * unmounted this file system while we were doing the switcheroo,
1765 * or a reopen of z_os failed then just bail out now.
1766 */
1767 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1768 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1769 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1770 return (SET_ERROR(EIO));
1771 }
1772
1773 /*
1774 * At this point there are no vops active, and any new vops will
1775 * fail with EIO since we have z_teardown_lock for writer (only
1776 * relavent for forced unmount).
1777 *
1778 * Release all holds on dbufs.
1779 */
1780 mutex_enter(&zfsvfs->z_znodes_lock);
1781 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1782 zp = list_next(&zfsvfs->z_all_znodes, zp))
1783 if (zp->z_sa_hdl) {
1784 ASSERT(ZTOV(zp)->v_count > 0);
1785 zfs_znode_dmu_fini(zp);
1786 }
1787 mutex_exit(&zfsvfs->z_znodes_lock);
1788
1789 /*
1790 * If we are unmounting, set the unmounted flag and let new vops
1791 * unblock. zfs_inactive will have the unmounted behavior, and all
1792 * other vops will fail with EIO.
1793 */
1794 if (unmounting) {
1795 zfsvfs->z_unmounted = B_TRUE;
1796 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1797 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1798 }
1799
1800 /*
1801 * z_os will be NULL if there was an error in attempting to reopen
1802 * zfsvfs, so just return as the properties had already been
1803 * unregistered and cached data had been evicted before.
1804 */
1805 if (zfsvfs->z_os == NULL)
1806 return (0);
1807
1808 /*
1809 * Unregister properties.
1810 */
1811 zfs_unregister_callbacks(zfsvfs);
1812
1813 /*
1814 * Evict cached data
1815 */
1816 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1817 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1818 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1819 dmu_objset_evict_dbufs(zfsvfs->z_os);
1820
1821 return (0);
1822 }
1823
1824 /*ARGSUSED*/
1825 static int
1826 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1827 {
1828 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1829 objset_t *os;
1830 int ret;
1831
1832 ret = secpolicy_fs_unmount(cr, vfsp);
1833 if (ret) {
1834 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1835 ZFS_DELEG_PERM_MOUNT, cr))
1836 return (ret);
1837 }
1838
1839 /*
1840 * We purge the parent filesystem's vfsp as the parent filesystem
1841 * and all of its snapshots have their vnode's v_vfsp set to the
1842 * parent's filesystem's vfsp. Note, 'z_parent' is self
1843 * referential for non-snapshots.
1844 */
1845 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1846
1847 /*
1848 * Unmount any snapshots mounted under .zfs before unmounting the
1849 * dataset itself.
1850 */
1851 if (zfsvfs->z_ctldir != NULL &&
1852 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1853 return (ret);
1854 }
1855
1856 if (!(fflag & MS_FORCE)) {
1857 /*
1858 * Check the number of active vnodes in the file system.
1859 * Our count is maintained in the vfs structure, but the
1860 * number is off by 1 to indicate a hold on the vfs
1861 * structure itself.
1862 *
1863 * The '.zfs' directory maintains a reference of its
1864 * own, and any active references underneath are
1865 * reflected in the vnode count.
1866 */
1867 if (zfsvfs->z_ctldir == NULL) {
1868 if (vfsp->vfs_count > 1)
1869 return (SET_ERROR(EBUSY));
1870 } else {
1871 if (vfsp->vfs_count > 2 ||
1872 zfsvfs->z_ctldir->v_count > 1)
1873 return (SET_ERROR(EBUSY));
1874 }
1875 }
1876
1877 vfsp->vfs_flag |= VFS_UNMOUNTED;
1878
1879 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1880 os = zfsvfs->z_os;
1881
1882 /*
1883 * z_os will be NULL if there was an error in
1884 * attempting to reopen zfsvfs.
1885 */
1886 if (os != NULL) {
1887 /*
1888 * Unset the objset user_ptr.
1889 */
1890 mutex_enter(&os->os_user_ptr_lock);
1891 dmu_objset_set_user(os, NULL);
1892 mutex_exit(&os->os_user_ptr_lock);
1893
1894 /*
1895 * Finally release the objset
1896 */
1897 dmu_objset_disown(os, zfsvfs);
1898 }
1899
1900 /*
1901 * We can now safely destroy the '.zfs' directory node.
1902 */
1903 if (zfsvfs->z_ctldir != NULL)
1904 zfsctl_destroy(zfsvfs);
1905
1906 return (0);
1907 }
1908
1909 static int
1910 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1911 {
1912 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1913 znode_t *zp;
1914 uint64_t object = 0;
1915 uint64_t fid_gen = 0;
1916 uint64_t gen_mask;
1917 uint64_t zp_gen;
1918 int i, err;
1919
1920 *vpp = NULL;
1921
1922 ZFS_ENTER(zfsvfs);
1923
1924 if (fidp->fid_len == LONG_FID_LEN) {
1925 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1926 uint64_t objsetid = 0;
1927 uint64_t setgen = 0;
1928
1929 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1930 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1931
1932 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1933 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1934
1935 ZFS_EXIT(zfsvfs);
1936
1937 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1938 if (err)
1939 return (SET_ERROR(EINVAL));
1940 ZFS_ENTER(zfsvfs);
1941 }
1942
1943 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1944 zfid_short_t *zfid = (zfid_short_t *)fidp;
1945
1946 for (i = 0; i < sizeof (zfid->zf_object); i++)
1947 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1948
1949 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1950 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1951 } else {
1952 ZFS_EXIT(zfsvfs);
1953 return (SET_ERROR(EINVAL));
1954 }
1955
1956 /* A zero fid_gen means we are in the .zfs control directories */
1957 if (fid_gen == 0 &&
1958 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1959 *vpp = zfsvfs->z_ctldir;
1960 ASSERT(*vpp != NULL);
1961 if (object == ZFSCTL_INO_SNAPDIR) {
1962 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1963 0, NULL, NULL, NULL, NULL, NULL) == 0);
1964 } else {
1965 VN_HOLD(*vpp);
1966 }
1967 ZFS_EXIT(zfsvfs);
1968 return (0);
1969 }
1970
1971 gen_mask = -1ULL >> (64 - 8 * i);
1972
1973 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1974 if (err = zfs_zget(zfsvfs, object, &zp)) {
1975 ZFS_EXIT(zfsvfs);
1976 return (err);
1977 }
1978 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1979 sizeof (uint64_t));
1980 zp_gen = zp_gen & gen_mask;
1981 if (zp_gen == 0)
1982 zp_gen = 1;
1983 if (zp->z_unlinked || zp_gen != fid_gen) {
1984 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1985 VN_RELE(ZTOV(zp));
1986 ZFS_EXIT(zfsvfs);
1987 return (SET_ERROR(EINVAL));
1988 }
1989
1990 *vpp = ZTOV(zp);
1991 ZFS_EXIT(zfsvfs);
1992 return (0);
1993 }
1994
1995 /*
1996 * Block out VOPs and close zfsvfs_t::z_os
1997 *
1998 * Note, if successful, then we return with the 'z_teardown_lock' and
1999 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2000 * dataset and objset intact so that they can be atomically handed off during
2001 * a subsequent rollback or recv operation and the resume thereafter.
2002 */
2003 int
2004 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2005 {
2006 int error;
2007
2008 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2009 return (error);
2010
2011 return (0);
2012 }
2013
2014 /*
2015 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2016 * is an invariant across any of the operations that can be performed while the
2017 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2018 * are the same: the relevant objset and associated dataset are owned by
2019 * zfsvfs, held, and long held on entry.
2020 */
2021 int
2022 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2023 {
2024 int err;
2025 znode_t *zp;
2026
2027 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2028 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2029
2030 /*
2031 * We already own this, so just hold and rele it to update the
2032 * objset_t, as the one we had before may have been evicted.
2033 */
2034 objset_t *os;
2035 VERIFY0(dmu_objset_hold(osname, zfsvfs, &os));
2036 VERIFY3P(os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2037 VERIFY(dsl_dataset_long_held(os->os_dsl_dataset));
2038 dmu_objset_rele(os, zfsvfs);
2039
2040 err = zfsvfs_init(zfsvfs, os);
2041 if (err != 0)
2042 goto bail;
2043
2044 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2045
2046 zfs_set_fuid_feature(zfsvfs);
2047
2048 /*
2049 * Attempt to re-establish all the active znodes with
2050 * their dbufs. If a zfs_rezget() fails, then we'll let
2051 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2052 * when they try to use their znode.
2053 */
2054 mutex_enter(&zfsvfs->z_znodes_lock);
2055 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2056 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2057 (void) zfs_rezget(zp);
2058 }
2059 mutex_exit(&zfsvfs->z_znodes_lock);
2060
2061 bail:
2062 /* release the VOPs */
2063 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2064 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2065
2066 if (err) {
2067 /*
2068 * Since we couldn't setup the sa framework, try to force
2069 * unmount this file system.
2070 */
2071 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2072 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2073 }
2074 return (err);
2075 }
2076
2077 static void
2078 zfs_freevfs(vfs_t *vfsp)
2079 {
2080 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2081
2082 /*
2083 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2084 * from zfs_mount(). Release it here. If we came through
2085 * zfs_mountroot() instead, we didn't grab an extra hold, so
2086 * skip the VFS_RELE for rootvfs.
2087 */
2088 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2089 VFS_RELE(zfsvfs->z_parent->z_vfs);
2090
2091 zfsvfs_free(zfsvfs);
2092
2093 atomic_dec_32(&zfs_active_fs_count);
2094 }
2095
2096 /*
2097 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2098 * so we can't safely do any non-idempotent initialization here.
2099 * Leave that to zfs_init() and zfs_fini(), which are called
2100 * from the module's _init() and _fini() entry points.
2101 */
2102 /*ARGSUSED*/
2103 static int
2104 zfs_vfsinit(int fstype, char *name)
2105 {
2106 int error;
2107
2108 zfsfstype = fstype;
2109
2110 /*
2111 * Setup vfsops and vnodeops tables.
2112 */
2113 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2114 if (error != 0) {
2115 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2116 }
2117
2118 error = zfs_create_op_tables();
2119 if (error) {
2120 zfs_remove_op_tables();
2121 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2122 (void) vfs_freevfsops_by_type(zfsfstype);
2123 return (error);
2124 }
2125
2126 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2127
2128 /*
2129 * Unique major number for all zfs mounts.
2130 * If we run out of 32-bit minors, we'll getudev() another major.
2131 */
2132 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2133 zfs_minor = ZFS_MIN_MINOR;
2134
2135 return (0);
2136 }
2137
2138 void
2139 zfs_init(void)
2140 {
2141 /*
2142 * Initialize .zfs directory structures
2143 */
2144 zfsctl_init();
2145
2146 /*
2147 * Initialize znode cache, vnode ops, etc...
2148 */
2149 zfs_znode_init();
2150
2151 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2152 }
2153
2154 void
2155 zfs_fini(void)
2156 {
2157 zfsctl_fini();
2158 zfs_znode_fini();
2159 }
2160
2161 int
2162 zfs_busy(void)
2163 {
2164 return (zfs_active_fs_count != 0);
2165 }
2166
2167 int
2168 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2169 {
2170 int error;
2171 objset_t *os = zfsvfs->z_os;
2172 dmu_tx_t *tx;
2173
2174 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2175 return (SET_ERROR(EINVAL));
2176
2177 if (newvers < zfsvfs->z_version)
2178 return (SET_ERROR(EINVAL));
2179
2180 if (zfs_spa_version_map(newvers) >
2181 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2182 return (SET_ERROR(ENOTSUP));
2183
2184 tx = dmu_tx_create(os);
2185 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2186 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2187 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2188 ZFS_SA_ATTRS);
2189 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2190 }
2191 error = dmu_tx_assign(tx, TXG_WAIT);
2192 if (error) {
2193 dmu_tx_abort(tx);
2194 return (error);
2195 }
2196
2197 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2198 8, 1, &newvers, tx);
2199
2200 if (error) {
2201 dmu_tx_commit(tx);
2202 return (error);
2203 }
2204
2205 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2206 uint64_t sa_obj;
2207
2208 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2209 SPA_VERSION_SA);
2210 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2211 DMU_OT_NONE, 0, tx);
2212
2213 error = zap_add(os, MASTER_NODE_OBJ,
2214 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2215 ASSERT0(error);
2216
2217 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2218 sa_register_update_callback(os, zfs_sa_upgrade);
2219 }
2220
2221 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2222 "from %llu to %llu", zfsvfs->z_version, newvers);
2223
2224 dmu_tx_commit(tx);
2225
2226 zfsvfs->z_version = newvers;
2227
2228 zfs_set_fuid_feature(zfsvfs);
2229
2230 return (0);
2231 }
2232
2233 /*
2234 * Read a property stored within the master node.
2235 */
2236 int
2237 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2238 {
2239 const char *pname;
2240 int error = ENOENT;
2241
2242 /*
2243 * Look up the file system's value for the property. For the
2244 * version property, we look up a slightly different string.
2245 */
2246 if (prop == ZFS_PROP_VERSION)
2247 pname = ZPL_VERSION_STR;
2248 else
2249 pname = zfs_prop_to_name(prop);
2250
2251 if (os != NULL)
2252 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2253
2254 if (error == ENOENT) {
2255 /* No value set, use the default value */
2256 switch (prop) {
2257 case ZFS_PROP_VERSION:
2258 *value = ZPL_VERSION;
2259 break;
2260 case ZFS_PROP_NORMALIZE:
2261 case ZFS_PROP_UTF8ONLY:
2262 *value = 0;
2263 break;
2264 case ZFS_PROP_CASE:
2265 *value = ZFS_CASE_SENSITIVE;
2266 break;
2267 default:
2268 return (error);
2269 }
2270 error = 0;
2271 }
2272 return (error);
2273 }
2274
2275 static vfsdef_t vfw = {
2276 VFSDEF_VERSION,
2277 MNTTYPE_ZFS,
2278 zfs_vfsinit,
2279 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2280 VSW_XID|VSW_ZMOUNT,
2281 &zfs_mntopts
2282 };
2283
2284 struct modlfs zfs_modlfs = {
2285 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2286 };