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) 2013 by Delphix. All rights reserved.
24 */
25
26 /* Portions Copyright 2010 Robert Milkowski */
27
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
32 #include <sys/kmem.h>
33 #include <sys/pathname.h>
34 #include <sys/vnode.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/mntent.h>
38 #include <sys/mount.h>
39 #include <sys/cmn_err.h>
40 #include "fs/fs_subr.h"
41 #include <sys/zfs_znode.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 "zfs_comutil.h"
68
69 int zfsfstype;
70 vfsops_t *zfs_vfsops = NULL;
71 static major_t zfs_major;
72 static minor_t zfs_minor;
73 static kmutex_t zfs_dev_mtx;
74
75 extern int sys_shutdown;
76
77 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
78 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
79 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
80 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
81 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
82 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
83 static void zfs_freevfs(vfs_t *vfsp);
84
85 static const fs_operation_def_t zfs_vfsops_template[] = {
86 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
87 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
88 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
89 VFSNAME_ROOT, { .vfs_root = zfs_root },
90 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
91 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
92 VFSNAME_VGET, { .vfs_vget = zfs_vget },
93 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
94 NULL, NULL
95 };
96
97 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
98 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
99 NULL, NULL
100 };
101
102 /*
103 * We need to keep a count of active fs's.
104 * This is necessary to prevent our module
105 * from being unloaded after a umount -f
106 */
107 static uint32_t zfs_active_fs_count = 0;
108
109 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
110 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
111 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
112 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
113
114 /*
115 * MO_DEFAULT is not used since the default value is determined
116 * by the equivalent property.
117 */
118 static mntopt_t mntopts[] = {
119 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
120 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
121 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
122 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
123 };
124
125 static mntopts_t zfs_mntopts = {
126 sizeof (mntopts) / sizeof (mntopt_t),
127 mntopts
128 };
129
130 /*ARGSUSED*/
131 int
132 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
133 {
134 /*
135 * Data integrity is job one. We don't want a compromised kernel
136 * writing to the storage pool, so we never sync during panic.
137 */
138 if (panicstr)
139 return (0);
140
141 /*
142 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
143 * to sync metadata, which they would otherwise cache indefinitely.
144 * Semantically, the only requirement is that the sync be initiated.
145 * The DMU syncs out txgs frequently, so there's nothing to do.
146 */
147 if (flag & SYNC_ATTR)
148 return (0);
149
150 if (vfsp != NULL) {
151 /*
152 * Sync a specific filesystem.
153 */
154 zfsvfs_t *zfsvfs = vfsp->vfs_data;
155 dsl_pool_t *dp;
156
157 ZFS_ENTER(zfsvfs);
158 dp = dmu_objset_pool(zfsvfs->z_os);
159
160 /*
161 * If the system is shutting down, then skip any
162 * filesystems which may exist on a suspended pool.
163 */
164 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
165 ZFS_EXIT(zfsvfs);
166 return (0);
167 }
168
169 if (zfsvfs->z_log != NULL)
170 zil_commit(zfsvfs->z_log, 0);
171
172 ZFS_EXIT(zfsvfs);
173 } else {
174 /*
175 * Sync all ZFS filesystems. This is what happens when you
176 * run sync(1M). Unlike other filesystems, ZFS honors the
177 * request by waiting for all pools to commit all dirty data.
178 */
179 spa_sync_allpools();
180 }
181
182 return (0);
183 }
184
185 static int
186 zfs_create_unique_device(dev_t *dev)
187 {
188 major_t new_major;
189
190 do {
191 ASSERT3U(zfs_minor, <=, MAXMIN32);
192 minor_t start = zfs_minor;
193 do {
194 mutex_enter(&zfs_dev_mtx);
195 if (zfs_minor >= MAXMIN32) {
196 /*
197 * If we're still using the real major
198 * keep out of /dev/zfs and /dev/zvol minor
199 * number space. If we're using a getudev()'ed
200 * major number, we can use all of its minors.
201 */
202 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
203 zfs_minor = ZFS_MIN_MINOR;
204 else
205 zfs_minor = 0;
206 } else {
207 zfs_minor++;
208 }
209 *dev = makedevice(zfs_major, zfs_minor);
210 mutex_exit(&zfs_dev_mtx);
211 } while (vfs_devismounted(*dev) && zfs_minor != start);
212 if (zfs_minor == start) {
213 /*
214 * We are using all ~262,000 minor numbers for the
215 * current major number. Create a new major number.
216 */
217 if ((new_major = getudev()) == (major_t)-1) {
218 cmn_err(CE_WARN,
219 "zfs_mount: Can't get unique major "
220 "device number.");
221 return (-1);
222 }
223 mutex_enter(&zfs_dev_mtx);
224 zfs_major = new_major;
225 zfs_minor = 0;
226
227 mutex_exit(&zfs_dev_mtx);
228 } else {
229 break;
230 }
231 /* CONSTANTCONDITION */
232 } while (1);
233
234 return (0);
235 }
236
237 static void
238 atime_changed_cb(void *arg, uint64_t newval)
239 {
240 zfsvfs_t *zfsvfs = arg;
241
242 if (newval == TRUE) {
243 zfsvfs->z_atime = TRUE;
244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
246 } else {
247 zfsvfs->z_atime = FALSE;
248 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
249 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
250 }
251 }
252
253 static void
254 xattr_changed_cb(void *arg, uint64_t newval)
255 {
256 zfsvfs_t *zfsvfs = arg;
257
258 if (newval == TRUE) {
259 /* XXX locking on vfs_flag? */
260 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
261 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
262 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
263 } else {
264 /* XXX locking on vfs_flag? */
265 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
266 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
267 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
268 }
269 }
270
271 static void
272 blksz_changed_cb(void *arg, uint64_t newval)
273 {
274 zfsvfs_t *zfsvfs = arg;
275
276 if (newval < SPA_MINBLOCKSIZE ||
277 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
278 newval = SPA_MAXBLOCKSIZE;
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[MAXNAMELEN];
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 /*
566 * We may attempt to unregister some callbacks that are not
567 * registered, but this is OK; it will simply return ENOMSG,
568 * which we will ignore.
569 */
570 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
571 atime_changed_cb, zfsvfs);
572 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
573 xattr_changed_cb, zfsvfs);
574 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
575 blksz_changed_cb, zfsvfs);
576 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
577 readonly_changed_cb, zfsvfs);
578 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
579 devices_changed_cb, zfsvfs);
580 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
581 setuid_changed_cb, zfsvfs);
582 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
583 exec_changed_cb, zfsvfs);
584 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
585 snapdir_changed_cb, zfsvfs);
586 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
587 acl_mode_changed_cb, zfsvfs);
588 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
589 acl_inherit_changed_cb, zfsvfs);
590 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
591 vscan_changed_cb, zfsvfs);
592 return (error);
593 }
594
595 static int
596 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
597 uint64_t *userp, uint64_t *groupp)
598 {
599 /*
600 * Is it a valid type of object to track?
601 */
602 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
603 return (SET_ERROR(ENOENT));
604
605 /*
606 * If we have a NULL data pointer
607 * then assume the id's aren't changing and
608 * return EEXIST to the dmu to let it know to
609 * use the same ids
610 */
611 if (data == NULL)
612 return (SET_ERROR(EEXIST));
613
614 if (bonustype == DMU_OT_ZNODE) {
615 znode_phys_t *znp = data;
616 *userp = znp->zp_uid;
617 *groupp = znp->zp_gid;
618 } else {
619 int hdrsize;
620 sa_hdr_phys_t *sap = data;
621 sa_hdr_phys_t sa = *sap;
622 boolean_t swap = B_FALSE;
623
624 ASSERT(bonustype == DMU_OT_SA);
625
626 if (sa.sa_magic == 0) {
627 /*
628 * This should only happen for newly created
629 * files that haven't had the znode data filled
630 * in yet.
631 */
632 *userp = 0;
633 *groupp = 0;
634 return (0);
635 }
636 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
637 sa.sa_magic = SA_MAGIC;
638 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
639 swap = B_TRUE;
640 } else {
641 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
642 }
643
644 hdrsize = sa_hdrsize(&sa);
645 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
646 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
647 SA_UID_OFFSET));
648 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
649 SA_GID_OFFSET));
650 if (swap) {
651 *userp = BSWAP_64(*userp);
652 *groupp = BSWAP_64(*groupp);
653 }
654 }
655 return (0);
656 }
657
658 static void
659 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
660 char *domainbuf, int buflen, uid_t *ridp)
661 {
662 uint64_t fuid;
663 const char *domain;
664
665 fuid = strtonum(fuidstr, NULL);
666
667 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
668 if (domain)
669 (void) strlcpy(domainbuf, domain, buflen);
670 else
671 domainbuf[0] = '\0';
672 *ridp = FUID_RID(fuid);
673 }
674
675 static uint64_t
676 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
677 {
678 switch (type) {
679 case ZFS_PROP_USERUSED:
680 return (DMU_USERUSED_OBJECT);
681 case ZFS_PROP_GROUPUSED:
682 return (DMU_GROUPUSED_OBJECT);
683 case ZFS_PROP_USERQUOTA:
684 return (zfsvfs->z_userquota_obj);
685 case ZFS_PROP_GROUPQUOTA:
686 return (zfsvfs->z_groupquota_obj);
687 }
688 return (0);
689 }
690
691 int
692 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
693 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
694 {
695 int error;
696 zap_cursor_t zc;
697 zap_attribute_t za;
698 zfs_useracct_t *buf = vbuf;
699 uint64_t obj;
700
701 if (!dmu_objset_userspace_present(zfsvfs->z_os))
702 return (SET_ERROR(ENOTSUP));
703
704 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
705 if (obj == 0) {
706 *bufsizep = 0;
707 return (0);
708 }
709
710 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
711 (error = zap_cursor_retrieve(&zc, &za)) == 0;
712 zap_cursor_advance(&zc)) {
713 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
714 *bufsizep)
715 break;
716
717 fuidstr_to_sid(zfsvfs, za.za_name,
718 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
719
720 buf->zu_space = za.za_first_integer;
721 buf++;
722 }
723 if (error == ENOENT)
724 error = 0;
725
726 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
727 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
728 *cookiep = zap_cursor_serialize(&zc);
729 zap_cursor_fini(&zc);
730 return (error);
731 }
732
733 /*
734 * buf must be big enough (eg, 32 bytes)
735 */
736 static int
737 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
738 char *buf, boolean_t addok)
739 {
740 uint64_t fuid;
741 int domainid = 0;
742
743 if (domain && domain[0]) {
744 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
745 if (domainid == -1)
746 return (SET_ERROR(ENOENT));
747 }
748 fuid = FUID_ENCODE(domainid, rid);
749 (void) sprintf(buf, "%llx", (longlong_t)fuid);
750 return (0);
751 }
752
753 int
754 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
755 const char *domain, uint64_t rid, uint64_t *valp)
756 {
757 char buf[32];
758 int err;
759 uint64_t obj;
760
761 *valp = 0;
762
763 if (!dmu_objset_userspace_present(zfsvfs->z_os))
764 return (SET_ERROR(ENOTSUP));
765
766 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
767 if (obj == 0)
768 return (0);
769
770 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
771 if (err)
772 return (err);
773
774 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
775 if (err == ENOENT)
776 err = 0;
777 return (err);
778 }
779
780 int
781 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
782 const char *domain, uint64_t rid, uint64_t quota)
783 {
784 char buf[32];
785 int err;
786 dmu_tx_t *tx;
787 uint64_t *objp;
788 boolean_t fuid_dirtied;
789
790 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
791 return (SET_ERROR(EINVAL));
792
793 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
794 return (SET_ERROR(ENOTSUP));
795
796 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
797 &zfsvfs->z_groupquota_obj;
798
799 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
800 if (err)
801 return (err);
802 fuid_dirtied = zfsvfs->z_fuid_dirty;
803
804 tx = dmu_tx_create(zfsvfs->z_os);
805 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
806 if (*objp == 0) {
807 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
808 zfs_userquota_prop_prefixes[type]);
809 }
810 if (fuid_dirtied)
811 zfs_fuid_txhold(zfsvfs, tx);
812 err = dmu_tx_assign(tx, TXG_WAIT);
813 if (err) {
814 dmu_tx_abort(tx);
815 return (err);
816 }
817
818 mutex_enter(&zfsvfs->z_lock);
819 if (*objp == 0) {
820 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
821 DMU_OT_NONE, 0, tx);
822 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
823 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
824 }
825 mutex_exit(&zfsvfs->z_lock);
826
827 if (quota == 0) {
828 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
829 if (err == ENOENT)
830 err = 0;
831 } else {
832 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
833 }
834 ASSERT(err == 0);
835 if (fuid_dirtied)
836 zfs_fuid_sync(zfsvfs, tx);
837 dmu_tx_commit(tx);
838 return (err);
839 }
840
841 boolean_t
842 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
843 {
844 char buf[32];
845 uint64_t used, quota, usedobj, quotaobj;
846 int err;
847
848 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
849 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
850
851 if (quotaobj == 0 || zfsvfs->z_replay)
852 return (B_FALSE);
853
854 (void) sprintf(buf, "%llx", (longlong_t)fuid);
855 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
856 if (err != 0)
857 return (B_FALSE);
858
859 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
860 if (err != 0)
861 return (B_FALSE);
862 return (used >= quota);
863 }
864
865 boolean_t
866 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
867 {
868 uint64_t fuid;
869 uint64_t quotaobj;
870
871 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
872
873 fuid = isgroup ? zp->z_gid : zp->z_uid;
874
875 if (quotaobj == 0 || zfsvfs->z_replay)
876 return (B_FALSE);
877
878 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
879 }
880
881 int
882 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
883 {
884 objset_t *os;
885 zfsvfs_t *zfsvfs;
886 uint64_t zval;
887 int i, error;
888 uint64_t sa_obj;
889
890 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
891
892 /*
893 * We claim to always be readonly so we can open snapshots;
894 * other ZPL code will prevent us from writing to snapshots.
895 */
896 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
897 if (error) {
898 kmem_free(zfsvfs, sizeof (zfsvfs_t));
899 return (error);
900 }
901
902 /*
903 * Initialize the zfs-specific filesystem structure.
904 * Should probably make this a kmem cache, shuffle fields,
905 * and just bzero up to z_hold_mtx[].
906 */
907 zfsvfs->z_vfs = NULL;
908 zfsvfs->z_parent = zfsvfs;
909 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
910 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
911 zfsvfs->z_os = os;
912
913 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
914 if (error) {
915 goto out;
916 } else if (zfsvfs->z_version >
917 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
918 (void) printf("Can't mount a version %lld file system "
919 "on a version %lld pool\n. Pool must be upgraded to mount "
920 "this file system.", (u_longlong_t)zfsvfs->z_version,
921 (u_longlong_t)spa_version(dmu_objset_spa(os)));
922 error = SET_ERROR(ENOTSUP);
923 goto out;
924 }
925 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
926 goto out;
927 zfsvfs->z_norm = (int)zval;
928
929 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
930 goto out;
931 zfsvfs->z_utf8 = (zval != 0);
932
933 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
934 goto out;
935 zfsvfs->z_case = (uint_t)zval;
936
937 /*
938 * Fold case on file systems that are always or sometimes case
939 * insensitive.
940 */
941 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
942 zfsvfs->z_case == ZFS_CASE_MIXED)
943 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
944
945 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
946 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
947
948 if (zfsvfs->z_use_sa) {
949 /* should either have both of these objects or none */
950 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
951 &sa_obj);
952 if (error)
953 return (error);
954 } else {
955 /*
956 * Pre SA versions file systems should never touch
957 * either the attribute registration or layout objects.
958 */
959 sa_obj = 0;
960 }
961
962 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
963 &zfsvfs->z_attr_table);
964 if (error)
965 goto out;
966
967 if (zfsvfs->z_version >= ZPL_VERSION_SA)
968 sa_register_update_callback(os, zfs_sa_upgrade);
969
970 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
971 &zfsvfs->z_root);
972 if (error)
973 goto out;
974 ASSERT(zfsvfs->z_root != 0);
975
976 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
977 &zfsvfs->z_unlinkedobj);
978 if (error)
979 goto out;
980
981 error = zap_lookup(os, MASTER_NODE_OBJ,
982 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
983 8, 1, &zfsvfs->z_userquota_obj);
984 if (error && error != ENOENT)
985 goto out;
986
987 error = zap_lookup(os, MASTER_NODE_OBJ,
988 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
989 8, 1, &zfsvfs->z_groupquota_obj);
990 if (error && error != ENOENT)
991 goto out;
992
993 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
994 &zfsvfs->z_fuid_obj);
995 if (error && error != ENOENT)
996 goto out;
997
998 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
999 &zfsvfs->z_shares_dir);
1000 if (error && error != ENOENT)
1001 goto out;
1002
1003 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1004 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1005 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1006 offsetof(znode_t, z_link_node));
1007 rrw_init(&zfsvfs->z_teardown_lock, B_FALSE);
1008 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1009 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1010 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1011 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1012
1013 *zfvp = zfsvfs;
1014 return (0);
1015
1016 out:
1017 dmu_objset_disown(os, zfsvfs);
1018 *zfvp = NULL;
1019 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1020 return (error);
1021 }
1022
1023 static int
1024 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1025 {
1026 int error;
1027
1028 error = zfs_register_callbacks(zfsvfs->z_vfs);
1029 if (error)
1030 return (error);
1031
1032 /*
1033 * Set the objset user_ptr to track its zfsvfs.
1034 */
1035 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1036 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1037 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1038
1039 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1040
1041 /*
1042 * If we are not mounting (ie: online recv), then we don't
1043 * have to worry about replaying the log as we blocked all
1044 * operations out since we closed the ZIL.
1045 */
1046 if (mounting) {
1047 boolean_t readonly;
1048
1049 /*
1050 * During replay we remove the read only flag to
1051 * allow replays to succeed.
1052 */
1053 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1054 if (readonly != 0)
1055 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1056 else
1057 zfs_unlinked_drain(zfsvfs);
1058
1059 /*
1060 * Parse and replay the intent log.
1061 *
1062 * Because of ziltest, this must be done after
1063 * zfs_unlinked_drain(). (Further note: ziltest
1064 * doesn't use readonly mounts, where
1065 * zfs_unlinked_drain() isn't called.) This is because
1066 * ziltest causes spa_sync() to think it's committed,
1067 * but actually it is not, so the intent log contains
1068 * many txg's worth of changes.
1069 *
1070 * In particular, if object N is in the unlinked set in
1071 * the last txg to actually sync, then it could be
1072 * actually freed in a later txg and then reallocated
1073 * in a yet later txg. This would write a "create
1074 * object N" record to the intent log. Normally, this
1075 * would be fine because the spa_sync() would have
1076 * written out the fact that object N is free, before
1077 * we could write the "create object N" intent log
1078 * record.
1079 *
1080 * But when we are in ziltest mode, we advance the "open
1081 * txg" without actually spa_sync()-ing the changes to
1082 * disk. So we would see that object N is still
1083 * allocated and in the unlinked set, and there is an
1084 * intent log record saying to allocate it.
1085 */
1086 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1087 if (zil_replay_disable) {
1088 zil_destroy(zfsvfs->z_log, B_FALSE);
1089 } else {
1090 zfsvfs->z_replay = B_TRUE;
1091 zil_replay(zfsvfs->z_os, zfsvfs,
1092 zfs_replay_vector);
1093 zfsvfs->z_replay = B_FALSE;
1094 }
1095 }
1096 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1097 }
1098
1099 return (0);
1100 }
1101
1102 void
1103 zfsvfs_free(zfsvfs_t *zfsvfs)
1104 {
1105 int i;
1106 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1107
1108 /*
1109 * This is a barrier to prevent the filesystem from going away in
1110 * zfs_znode_move() until we can safely ensure that the filesystem is
1111 * not unmounted. We consider the filesystem valid before the barrier
1112 * and invalid after the barrier.
1113 */
1114 rw_enter(&zfsvfs_lock, RW_READER);
1115 rw_exit(&zfsvfs_lock);
1116
1117 zfs_fuid_destroy(zfsvfs);
1118
1119 mutex_destroy(&zfsvfs->z_znodes_lock);
1120 mutex_destroy(&zfsvfs->z_lock);
1121 list_destroy(&zfsvfs->z_all_znodes);
1122 rrw_destroy(&zfsvfs->z_teardown_lock);
1123 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1124 rw_destroy(&zfsvfs->z_fuid_lock);
1125 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1126 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1127 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1128 }
1129
1130 static void
1131 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1132 {
1133 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1134 if (zfsvfs->z_vfs) {
1135 if (zfsvfs->z_use_fuids) {
1136 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1137 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1138 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1139 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1140 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1141 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1142 } else {
1143 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1144 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1145 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1146 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1147 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1148 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1149 }
1150 }
1151 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1152 }
1153
1154 static int
1155 zfs_domount(vfs_t *vfsp, char *osname)
1156 {
1157 dev_t mount_dev;
1158 uint64_t recordsize, fsid_guid;
1159 int error = 0;
1160 zfsvfs_t *zfsvfs;
1161
1162 ASSERT(vfsp);
1163 ASSERT(osname);
1164
1165 error = zfsvfs_create(osname, &zfsvfs);
1166 if (error)
1167 return (error);
1168 zfsvfs->z_vfs = vfsp;
1169
1170 /* Initialize the generic filesystem structure. */
1171 vfsp->vfs_bcount = 0;
1172 vfsp->vfs_data = NULL;
1173
1174 if (zfs_create_unique_device(&mount_dev) == -1) {
1175 error = SET_ERROR(ENODEV);
1176 goto out;
1177 }
1178 ASSERT(vfs_devismounted(mount_dev) == 0);
1179
1180 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1181 NULL))
1182 goto out;
1183
1184 vfsp->vfs_dev = mount_dev;
1185 vfsp->vfs_fstype = zfsfstype;
1186 vfsp->vfs_bsize = recordsize;
1187 vfsp->vfs_flag |= VFS_NOTRUNC;
1188 vfsp->vfs_data = zfsvfs;
1189
1190 /*
1191 * The fsid is 64 bits, composed of an 8-bit fs type, which
1192 * separates our fsid from any other filesystem types, and a
1193 * 56-bit objset unique ID. The objset unique ID is unique to
1194 * all objsets open on this system, provided by unique_create().
1195 * The 8-bit fs type must be put in the low bits of fsid[1]
1196 * because that's where other Solaris filesystems put it.
1197 */
1198 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1199 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1200 vfsp->vfs_fsid.val[0] = fsid_guid;
1201 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1202 zfsfstype & 0xFF;
1203
1204 /*
1205 * Set features for file system.
1206 */
1207 zfs_set_fuid_feature(zfsvfs);
1208 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1209 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1210 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1211 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1212 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1213 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1214 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1215 }
1216 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1217
1218 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1219 uint64_t pval;
1220
1221 atime_changed_cb(zfsvfs, B_FALSE);
1222 readonly_changed_cb(zfsvfs, B_TRUE);
1223 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1224 goto out;
1225 xattr_changed_cb(zfsvfs, pval);
1226 zfsvfs->z_issnap = B_TRUE;
1227 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1228
1229 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1230 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1231 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1232 } else {
1233 error = zfsvfs_setup(zfsvfs, B_TRUE);
1234 }
1235
1236 if (!zfsvfs->z_issnap)
1237 zfsctl_create(zfsvfs);
1238 out:
1239 if (error) {
1240 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1241 zfsvfs_free(zfsvfs);
1242 } else {
1243 atomic_add_32(&zfs_active_fs_count, 1);
1244 }
1245
1246 return (error);
1247 }
1248
1249 void
1250 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1251 {
1252 objset_t *os = zfsvfs->z_os;
1253 struct dsl_dataset *ds;
1254
1255 /*
1256 * Unregister properties.
1257 */
1258 if (!dmu_objset_is_snapshot(os)) {
1259 ds = dmu_objset_ds(os);
1260 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1261 zfsvfs) == 0);
1262
1263 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1264 zfsvfs) == 0);
1265
1266 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1267 zfsvfs) == 0);
1268
1269 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1270 zfsvfs) == 0);
1271
1272 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1273 zfsvfs) == 0);
1274
1275 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1276 zfsvfs) == 0);
1277
1278 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1279 zfsvfs) == 0);
1280
1281 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1282 zfsvfs) == 0);
1283
1284 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1285 zfsvfs) == 0);
1286
1287 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1288 acl_inherit_changed_cb, zfsvfs) == 0);
1289
1290 VERIFY(dsl_prop_unregister(ds, "vscan",
1291 vscan_changed_cb, zfsvfs) == 0);
1292 }
1293 }
1294
1295 /*
1296 * Convert a decimal digit string to a uint64_t integer.
1297 */
1298 static int
1299 str_to_uint64(char *str, uint64_t *objnum)
1300 {
1301 uint64_t num = 0;
1302
1303 while (*str) {
1304 if (*str < '0' || *str > '9')
1305 return (SET_ERROR(EINVAL));
1306
1307 num = num*10 + *str++ - '0';
1308 }
1309
1310 *objnum = num;
1311 return (0);
1312 }
1313
1314 /*
1315 * The boot path passed from the boot loader is in the form of
1316 * "rootpool-name/root-filesystem-object-number'. Convert this
1317 * string to a dataset name: "rootpool-name/root-filesystem-name".
1318 */
1319 static int
1320 zfs_parse_bootfs(char *bpath, char *outpath)
1321 {
1322 char *slashp;
1323 uint64_t objnum;
1324 int error;
1325
1326 if (*bpath == 0 || *bpath == '/')
1327 return (SET_ERROR(EINVAL));
1328
1329 (void) strcpy(outpath, bpath);
1330
1331 slashp = strchr(bpath, '/');
1332
1333 /* if no '/', just return the pool name */
1334 if (slashp == NULL) {
1335 return (0);
1336 }
1337
1338 /* if not a number, just return the root dataset name */
1339 if (str_to_uint64(slashp+1, &objnum)) {
1340 return (0);
1341 }
1342
1343 *slashp = '\0';
1344 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1345 *slashp = '/';
1346
1347 return (error);
1348 }
1349
1350 /*
1351 * zfs_check_global_label:
1352 * Check that the hex label string is appropriate for the dataset
1353 * being mounted into the global_zone proper.
1354 *
1355 * Return an error if the hex label string is not default or
1356 * admin_low/admin_high. For admin_low labels, the corresponding
1357 * dataset must be readonly.
1358 */
1359 int
1360 zfs_check_global_label(const char *dsname, const char *hexsl)
1361 {
1362 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1363 return (0);
1364 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1365 return (0);
1366 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1367 /* must be readonly */
1368 uint64_t rdonly;
1369
1370 if (dsl_prop_get_integer(dsname,
1371 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1372 return (SET_ERROR(EACCES));
1373 return (rdonly ? 0 : EACCES);
1374 }
1375 return (SET_ERROR(EACCES));
1376 }
1377
1378 /*
1379 * zfs_mount_label_policy:
1380 * Determine whether the mount is allowed according to MAC check.
1381 * by comparing (where appropriate) label of the dataset against
1382 * the label of the zone being mounted into. If the dataset has
1383 * no label, create one.
1384 *
1385 * Returns:
1386 * 0 : access allowed
1387 * >0 : error code, such as EACCES
1388 */
1389 static int
1390 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1391 {
1392 int error, retv;
1393 zone_t *mntzone = NULL;
1394 ts_label_t *mnt_tsl;
1395 bslabel_t *mnt_sl;
1396 bslabel_t ds_sl;
1397 char ds_hexsl[MAXNAMELEN];
1398
1399 retv = EACCES; /* assume the worst */
1400
1401 /*
1402 * Start by getting the dataset label if it exists.
1403 */
1404 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1405 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1406 if (error)
1407 return (SET_ERROR(EACCES));
1408
1409 /*
1410 * If labeling is NOT enabled, then disallow the mount of datasets
1411 * which have a non-default label already. No other label checks
1412 * are needed.
1413 */
1414 if (!is_system_labeled()) {
1415 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1416 return (0);
1417 return (SET_ERROR(EACCES));
1418 }
1419
1420 /*
1421 * Get the label of the mountpoint. If mounting into the global
1422 * zone (i.e. mountpoint is not within an active zone and the
1423 * zoned property is off), the label must be default or
1424 * admin_low/admin_high only; no other checks are needed.
1425 */
1426 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1427 if (mntzone->zone_id == GLOBAL_ZONEID) {
1428 uint64_t zoned;
1429
1430 zone_rele(mntzone);
1431
1432 if (dsl_prop_get_integer(osname,
1433 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1434 return (SET_ERROR(EACCES));
1435 if (!zoned)
1436 return (zfs_check_global_label(osname, ds_hexsl));
1437 else
1438 /*
1439 * This is the case of a zone dataset being mounted
1440 * initially, before the zone has been fully created;
1441 * allow this mount into global zone.
1442 */
1443 return (0);
1444 }
1445
1446 mnt_tsl = mntzone->zone_slabel;
1447 ASSERT(mnt_tsl != NULL);
1448 label_hold(mnt_tsl);
1449 mnt_sl = label2bslabel(mnt_tsl);
1450
1451 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1452 /*
1453 * The dataset doesn't have a real label, so fabricate one.
1454 */
1455 char *str = NULL;
1456
1457 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1458 dsl_prop_set_string(osname,
1459 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1460 ZPROP_SRC_LOCAL, str) == 0)
1461 retv = 0;
1462 if (str != NULL)
1463 kmem_free(str, strlen(str) + 1);
1464 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1465 /*
1466 * Now compare labels to complete the MAC check. If the
1467 * labels are equal then allow access. If the mountpoint
1468 * label dominates the dataset label, allow readonly access.
1469 * Otherwise, access is denied.
1470 */
1471 if (blequal(mnt_sl, &ds_sl))
1472 retv = 0;
1473 else if (bldominates(mnt_sl, &ds_sl)) {
1474 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1475 retv = 0;
1476 }
1477 }
1478
1479 label_rele(mnt_tsl);
1480 zone_rele(mntzone);
1481 return (retv);
1482 }
1483
1484 static int
1485 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1486 {
1487 int error = 0;
1488 static int zfsrootdone = 0;
1489 zfsvfs_t *zfsvfs = NULL;
1490 znode_t *zp = NULL;
1491 vnode_t *vp = NULL;
1492 char *zfs_bootfs;
1493 char *zfs_devid;
1494
1495 ASSERT(vfsp);
1496
1497 /*
1498 * The filesystem that we mount as root is defined in the
1499 * boot property "zfs-bootfs" with a format of
1500 * "poolname/root-dataset-objnum".
1501 */
1502 if (why == ROOT_INIT) {
1503 if (zfsrootdone++)
1504 return (SET_ERROR(EBUSY));
1505 /*
1506 * the process of doing a spa_load will require the
1507 * clock to be set before we could (for example) do
1508 * something better by looking at the timestamp on
1509 * an uberblock, so just set it to -1.
1510 */
1511 clkset(-1);
1512
1513 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1514 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1515 "bootfs name");
1516 return (SET_ERROR(EINVAL));
1517 }
1518 zfs_devid = spa_get_bootprop("diskdevid");
1519 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1520 if (zfs_devid)
1521 spa_free_bootprop(zfs_devid);
1522 if (error) {
1523 spa_free_bootprop(zfs_bootfs);
1524 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1525 error);
1526 return (error);
1527 }
1528 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1529 spa_free_bootprop(zfs_bootfs);
1530 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1531 error);
1532 return (error);
1533 }
1534
1535 spa_free_bootprop(zfs_bootfs);
1536
1537 if (error = vfs_lock(vfsp))
1538 return (error);
1539
1540 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1541 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1542 goto out;
1543 }
1544
1545 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1546 ASSERT(zfsvfs);
1547 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1548 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1549 goto out;
1550 }
1551
1552 vp = ZTOV(zp);
1553 mutex_enter(&vp->v_lock);
1554 vp->v_flag |= VROOT;
1555 mutex_exit(&vp->v_lock);
1556 rootvp = vp;
1557
1558 /*
1559 * Leave rootvp held. The root file system is never unmounted.
1560 */
1561
1562 vfs_add((struct vnode *)0, vfsp,
1563 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1564 out:
1565 vfs_unlock(vfsp);
1566 return (error);
1567 } else if (why == ROOT_REMOUNT) {
1568 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1569 vfsp->vfs_flag |= VFS_REMOUNT;
1570
1571 /* refresh mount options */
1572 zfs_unregister_callbacks(vfsp->vfs_data);
1573 return (zfs_register_callbacks(vfsp));
1574
1575 } else if (why == ROOT_UNMOUNT) {
1576 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1577 (void) zfs_sync(vfsp, 0, 0);
1578 return (0);
1579 }
1580
1581 /*
1582 * if "why" is equal to anything else other than ROOT_INIT,
1583 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1584 */
1585 return (SET_ERROR(ENOTSUP));
1586 }
1587
1588 /*ARGSUSED*/
1589 static int
1590 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1591 {
1592 char *osname;
1593 pathname_t spn;
1594 int error = 0;
1595 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1596 UIO_SYSSPACE : UIO_USERSPACE;
1597 int canwrite;
1598
1599 if (mvp->v_type != VDIR)
1600 return (SET_ERROR(ENOTDIR));
1601
1602 mutex_enter(&mvp->v_lock);
1603 if ((uap->flags & MS_REMOUNT) == 0 &&
1604 (uap->flags & MS_OVERLAY) == 0 &&
1605 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1606 mutex_exit(&mvp->v_lock);
1607 return (SET_ERROR(EBUSY));
1608 }
1609 mutex_exit(&mvp->v_lock);
1610
1611 /*
1612 * ZFS does not support passing unparsed data in via MS_DATA.
1613 * Users should use the MS_OPTIONSTR interface; this means
1614 * that all option parsing is already done and the options struct
1615 * can be interrogated.
1616 */
1617 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1618 return (SET_ERROR(EINVAL));
1619
1620 /*
1621 * Get the objset name (the "special" mount argument).
1622 */
1623 if (error = pn_get(uap->spec, fromspace, &spn))
1624 return (error);
1625
1626 osname = spn.pn_path;
1627
1628 /*
1629 * Check for mount privilege?
1630 *
1631 * If we don't have privilege then see if
1632 * we have local permission to allow it
1633 */
1634 error = secpolicy_fs_mount(cr, mvp, vfsp);
1635 if (error) {
1636 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1637 vattr_t vattr;
1638
1639 /*
1640 * Make sure user is the owner of the mount point
1641 * or has sufficient privileges.
1642 */
1643
1644 vattr.va_mask = AT_UID;
1645
1646 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1647 goto out;
1648 }
1649
1650 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1651 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1652 goto out;
1653 }
1654 secpolicy_fs_mount_clearopts(cr, vfsp);
1655 } else {
1656 goto out;
1657 }
1658 }
1659
1660 /*
1661 * Refuse to mount a filesystem if we are in a local zone and the
1662 * dataset is not visible.
1663 */
1664 if (!INGLOBALZONE(curproc) &&
1665 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1666 error = SET_ERROR(EPERM);
1667 goto out;
1668 }
1669
1670 error = zfs_mount_label_policy(vfsp, osname);
1671 if (error)
1672 goto out;
1673
1674 /*
1675 * When doing a remount, we simply refresh our temporary properties
1676 * according to those options set in the current VFS options.
1677 */
1678 if (uap->flags & MS_REMOUNT) {
1679 /* refresh mount options */
1680 zfs_unregister_callbacks(vfsp->vfs_data);
1681 error = zfs_register_callbacks(vfsp);
1682 goto out;
1683 }
1684
1685 error = zfs_domount(vfsp, osname);
1686
1687 /*
1688 * Add an extra VFS_HOLD on our parent vfs so that it can't
1689 * disappear due to a forced unmount.
1690 */
1691 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1692 VFS_HOLD(mvp->v_vfsp);
1693
1694 out:
1695 pn_free(&spn);
1696 return (error);
1697 }
1698
1699 static int
1700 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1701 {
1702 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1703 dev32_t d32;
1704 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1705
1706 ZFS_ENTER(zfsvfs);
1707
1708 dmu_objset_space(zfsvfs->z_os,
1709 &refdbytes, &availbytes, &usedobjs, &availobjs);
1710
1711 /*
1712 * The underlying storage pool actually uses multiple block sizes.
1713 * We report the fragsize as the smallest block size we support,
1714 * and we report our blocksize as the filesystem's maximum blocksize.
1715 */
1716 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1717 statp->f_bsize = zfsvfs->z_max_blksz;
1718
1719 /*
1720 * The following report "total" blocks of various kinds in the
1721 * file system, but reported in terms of f_frsize - the
1722 * "fragment" size.
1723 */
1724
1725 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1726 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1727 statp->f_bavail = statp->f_bfree; /* no root reservation */
1728
1729 /*
1730 * statvfs() should really be called statufs(), because it assumes
1731 * static metadata. ZFS doesn't preallocate files, so the best
1732 * we can do is report the max that could possibly fit in f_files,
1733 * and that minus the number actually used in f_ffree.
1734 * For f_ffree, report the smaller of the number of object available
1735 * and the number of blocks (each object will take at least a block).
1736 */
1737 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1738 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1739 statp->f_files = statp->f_ffree + usedobjs;
1740
1741 (void) cmpldev(&d32, vfsp->vfs_dev);
1742 statp->f_fsid = d32;
1743
1744 /*
1745 * We're a zfs filesystem.
1746 */
1747 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1748
1749 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1750
1751 statp->f_namemax = ZFS_MAXNAMELEN;
1752
1753 /*
1754 * We have all of 32 characters to stuff a string here.
1755 * Is there anything useful we could/should provide?
1756 */
1757 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1758
1759 ZFS_EXIT(zfsvfs);
1760 return (0);
1761 }
1762
1763 static int
1764 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1765 {
1766 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1767 znode_t *rootzp;
1768 int error;
1769
1770 ZFS_ENTER(zfsvfs);
1771
1772 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1773 if (error == 0)
1774 *vpp = ZTOV(rootzp);
1775
1776 ZFS_EXIT(zfsvfs);
1777 return (error);
1778 }
1779
1780 /*
1781 * Teardown the zfsvfs::z_os.
1782 *
1783 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1784 * and 'z_teardown_inactive_lock' held.
1785 */
1786 static int
1787 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1788 {
1789 znode_t *zp;
1790
1791 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1792
1793 if (!unmounting) {
1794 /*
1795 * We purge the parent filesystem's vfsp as the parent
1796 * filesystem and all of its snapshots have their vnode's
1797 * v_vfsp set to the parent's filesystem's vfsp. Note,
1798 * 'z_parent' is self referential for non-snapshots.
1799 */
1800 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1801 }
1802
1803 /*
1804 * Close the zil. NB: Can't close the zil while zfs_inactive
1805 * threads are blocked as zil_close can call zfs_inactive.
1806 */
1807 if (zfsvfs->z_log) {
1808 zil_close(zfsvfs->z_log);
1809 zfsvfs->z_log = NULL;
1810 }
1811
1812 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1813
1814 /*
1815 * If we are not unmounting (ie: online recv) and someone already
1816 * unmounted this file system while we were doing the switcheroo,
1817 * or a reopen of z_os failed then just bail out now.
1818 */
1819 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1820 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1821 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1822 return (SET_ERROR(EIO));
1823 }
1824
1825 /*
1826 * At this point there are no vops active, and any new vops will
1827 * fail with EIO since we have z_teardown_lock for writer (only
1828 * relavent for forced unmount).
1829 *
1830 * Release all holds on dbufs.
1831 */
1832 mutex_enter(&zfsvfs->z_znodes_lock);
1833 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1834 zp = list_next(&zfsvfs->z_all_znodes, zp))
1835 if (zp->z_sa_hdl) {
1836 ASSERT(ZTOV(zp)->v_count > 0);
1837 zfs_znode_dmu_fini(zp);
1838 }
1839 mutex_exit(&zfsvfs->z_znodes_lock);
1840
1841 /*
1842 * If we are unmounting, set the unmounted flag and let new vops
1843 * unblock. zfs_inactive will have the unmounted behavior, and all
1844 * other vops will fail with EIO.
1845 */
1846 if (unmounting) {
1847 zfsvfs->z_unmounted = B_TRUE;
1848 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1849 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1850 }
1851
1852 /*
1853 * z_os will be NULL if there was an error in attempting to reopen
1854 * zfsvfs, so just return as the properties had already been
1855 * unregistered and cached data had been evicted before.
1856 */
1857 if (zfsvfs->z_os == NULL)
1858 return (0);
1859
1860 /*
1861 * Unregister properties.
1862 */
1863 zfs_unregister_callbacks(zfsvfs);
1864
1865 /*
1866 * Evict cached data
1867 */
1868 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1869 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1870 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1871 dmu_objset_evict_dbufs(zfsvfs->z_os);
1872
1873 return (0);
1874 }
1875
1876 /*ARGSUSED*/
1877 static int
1878 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1879 {
1880 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1881 objset_t *os;
1882 int ret;
1883
1884 ret = secpolicy_fs_unmount(cr, vfsp);
1885 if (ret) {
1886 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1887 ZFS_DELEG_PERM_MOUNT, cr))
1888 return (ret);
1889 }
1890
1891 /*
1892 * We purge the parent filesystem's vfsp as the parent filesystem
1893 * and all of its snapshots have their vnode's v_vfsp set to the
1894 * parent's filesystem's vfsp. Note, 'z_parent' is self
1895 * referential for non-snapshots.
1896 */
1897 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1898
1899 /*
1900 * Unmount any snapshots mounted under .zfs before unmounting the
1901 * dataset itself.
1902 */
1903 if (zfsvfs->z_ctldir != NULL &&
1904 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1905 return (ret);
1906 }
1907
1908 if (!(fflag & MS_FORCE)) {
1909 /*
1910 * Check the number of active vnodes in the file system.
1911 * Our count is maintained in the vfs structure, but the
1912 * number is off by 1 to indicate a hold on the vfs
1913 * structure itself.
1914 *
1915 * The '.zfs' directory maintains a reference of its
1916 * own, and any active references underneath are
1917 * reflected in the vnode count.
1918 */
1919 if (zfsvfs->z_ctldir == NULL) {
1920 if (vfsp->vfs_count > 1)
1921 return (SET_ERROR(EBUSY));
1922 } else {
1923 if (vfsp->vfs_count > 2 ||
1924 zfsvfs->z_ctldir->v_count > 1)
1925 return (SET_ERROR(EBUSY));
1926 }
1927 }
1928
1929 vfsp->vfs_flag |= VFS_UNMOUNTED;
1930
1931 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1932 os = zfsvfs->z_os;
1933
1934 /*
1935 * z_os will be NULL if there was an error in
1936 * attempting to reopen zfsvfs.
1937 */
1938 if (os != NULL) {
1939 /*
1940 * Unset the objset user_ptr.
1941 */
1942 mutex_enter(&os->os_user_ptr_lock);
1943 dmu_objset_set_user(os, NULL);
1944 mutex_exit(&os->os_user_ptr_lock);
1945
1946 /*
1947 * Finally release the objset
1948 */
1949 dmu_objset_disown(os, zfsvfs);
1950 }
1951
1952 /*
1953 * We can now safely destroy the '.zfs' directory node.
1954 */
1955 if (zfsvfs->z_ctldir != NULL)
1956 zfsctl_destroy(zfsvfs);
1957
1958 return (0);
1959 }
1960
1961 static int
1962 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1963 {
1964 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1965 znode_t *zp;
1966 uint64_t object = 0;
1967 uint64_t fid_gen = 0;
1968 uint64_t gen_mask;
1969 uint64_t zp_gen;
1970 int i, err;
1971
1972 *vpp = NULL;
1973
1974 ZFS_ENTER(zfsvfs);
1975
1976 if (fidp->fid_len == LONG_FID_LEN) {
1977 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1978 uint64_t objsetid = 0;
1979 uint64_t setgen = 0;
1980
1981 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1982 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1983
1984 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1985 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1986
1987 ZFS_EXIT(zfsvfs);
1988
1989 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1990 if (err)
1991 return (SET_ERROR(EINVAL));
1992 ZFS_ENTER(zfsvfs);
1993 }
1994
1995 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1996 zfid_short_t *zfid = (zfid_short_t *)fidp;
1997
1998 for (i = 0; i < sizeof (zfid->zf_object); i++)
1999 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2000
2001 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2002 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2003 } else {
2004 ZFS_EXIT(zfsvfs);
2005 return (SET_ERROR(EINVAL));
2006 }
2007
2008 /* A zero fid_gen means we are in the .zfs control directories */
2009 if (fid_gen == 0 &&
2010 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2011 *vpp = zfsvfs->z_ctldir;
2012 ASSERT(*vpp != NULL);
2013 if (object == ZFSCTL_INO_SNAPDIR) {
2014 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2015 0, NULL, NULL, NULL, NULL, NULL) == 0);
2016 } else {
2017 VN_HOLD(*vpp);
2018 }
2019 ZFS_EXIT(zfsvfs);
2020 return (0);
2021 }
2022
2023 gen_mask = -1ULL >> (64 - 8 * i);
2024
2025 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2026 if (err = zfs_zget(zfsvfs, object, &zp)) {
2027 ZFS_EXIT(zfsvfs);
2028 return (err);
2029 }
2030 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2031 sizeof (uint64_t));
2032 zp_gen = zp_gen & gen_mask;
2033 if (zp_gen == 0)
2034 zp_gen = 1;
2035 if (zp->z_unlinked || zp_gen != fid_gen) {
2036 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2037 VN_RELE(ZTOV(zp));
2038 ZFS_EXIT(zfsvfs);
2039 return (SET_ERROR(EINVAL));
2040 }
2041
2042 *vpp = ZTOV(zp);
2043 ZFS_EXIT(zfsvfs);
2044 return (0);
2045 }
2046
2047 /*
2048 * Block out VOPs and close zfsvfs_t::z_os
2049 *
2050 * Note, if successful, then we return with the 'z_teardown_lock' and
2051 * 'z_teardown_inactive_lock' write held.
2052 */
2053 int
2054 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2055 {
2056 int error;
2057
2058 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2059 return (error);
2060 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2061
2062 return (0);
2063 }
2064
2065 /*
2066 * Reopen zfsvfs_t::z_os and release VOPs.
2067 */
2068 int
2069 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2070 {
2071 int err;
2072
2073 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2074 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2075
2076 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2077 &zfsvfs->z_os);
2078 if (err) {
2079 zfsvfs->z_os = NULL;
2080 } else {
2081 znode_t *zp;
2082 uint64_t sa_obj = 0;
2083
2084 /*
2085 * Make sure version hasn't changed
2086 */
2087
2088 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2089 &zfsvfs->z_version);
2090
2091 if (err)
2092 goto bail;
2093
2094 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2095 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2096
2097 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2098 goto bail;
2099
2100 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2101 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2102 goto bail;
2103
2104 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2105 sa_register_update_callback(zfsvfs->z_os,
2106 zfs_sa_upgrade);
2107
2108 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2109
2110 zfs_set_fuid_feature(zfsvfs);
2111
2112 /*
2113 * Attempt to re-establish all the active znodes with
2114 * their dbufs. If a zfs_rezget() fails, then we'll let
2115 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2116 * when they try to use their znode.
2117 */
2118 mutex_enter(&zfsvfs->z_znodes_lock);
2119 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2120 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2121 (void) zfs_rezget(zp);
2122 }
2123 mutex_exit(&zfsvfs->z_znodes_lock);
2124 }
2125
2126 bail:
2127 /* release the VOPs */
2128 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2129 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2130
2131 if (err) {
2132 /*
2133 * Since we couldn't reopen zfsvfs::z_os, or
2134 * setup the sa framework force unmount this file system.
2135 */
2136 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2137 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2138 }
2139 return (err);
2140 }
2141
2142 static void
2143 zfs_freevfs(vfs_t *vfsp)
2144 {
2145 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2146
2147 /*
2148 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2149 * from zfs_mount(). Release it here. If we came through
2150 * zfs_mountroot() instead, we didn't grab an extra hold, so
2151 * skip the VFS_RELE for rootvfs.
2152 */
2153 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2154 VFS_RELE(zfsvfs->z_parent->z_vfs);
2155
2156 zfsvfs_free(zfsvfs);
2157
2158 atomic_add_32(&zfs_active_fs_count, -1);
2159 }
2160
2161 /*
2162 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2163 * so we can't safely do any non-idempotent initialization here.
2164 * Leave that to zfs_init() and zfs_fini(), which are called
2165 * from the module's _init() and _fini() entry points.
2166 */
2167 /*ARGSUSED*/
2168 static int
2169 zfs_vfsinit(int fstype, char *name)
2170 {
2171 int error;
2172
2173 zfsfstype = fstype;
2174
2175 /*
2176 * Setup vfsops and vnodeops tables.
2177 */
2178 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2179 if (error != 0) {
2180 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2181 }
2182
2183 error = zfs_create_op_tables();
2184 if (error) {
2185 zfs_remove_op_tables();
2186 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2187 (void) vfs_freevfsops_by_type(zfsfstype);
2188 return (error);
2189 }
2190
2191 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2192
2193 /*
2194 * Unique major number for all zfs mounts.
2195 * If we run out of 32-bit minors, we'll getudev() another major.
2196 */
2197 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2198 zfs_minor = ZFS_MIN_MINOR;
2199
2200 return (0);
2201 }
2202
2203 void
2204 zfs_init(void)
2205 {
2206 /*
2207 * Initialize .zfs directory structures
2208 */
2209 zfsctl_init();
2210
2211 /*
2212 * Initialize znode cache, vnode ops, etc...
2213 */
2214 zfs_znode_init();
2215
2216 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2217 }
2218
2219 void
2220 zfs_fini(void)
2221 {
2222 zfsctl_fini();
2223 zfs_znode_fini();
2224 }
2225
2226 int
2227 zfs_busy(void)
2228 {
2229 return (zfs_active_fs_count != 0);
2230 }
2231
2232 int
2233 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2234 {
2235 int error;
2236 objset_t *os = zfsvfs->z_os;
2237 dmu_tx_t *tx;
2238
2239 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2240 return (SET_ERROR(EINVAL));
2241
2242 if (newvers < zfsvfs->z_version)
2243 return (SET_ERROR(EINVAL));
2244
2245 if (zfs_spa_version_map(newvers) >
2246 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2247 return (SET_ERROR(ENOTSUP));
2248
2249 tx = dmu_tx_create(os);
2250 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2251 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2252 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2253 ZFS_SA_ATTRS);
2254 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2255 }
2256 error = dmu_tx_assign(tx, TXG_WAIT);
2257 if (error) {
2258 dmu_tx_abort(tx);
2259 return (error);
2260 }
2261
2262 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2263 8, 1, &newvers, tx);
2264
2265 if (error) {
2266 dmu_tx_commit(tx);
2267 return (error);
2268 }
2269
2270 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2271 uint64_t sa_obj;
2272
2273 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2274 SPA_VERSION_SA);
2275 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2276 DMU_OT_NONE, 0, tx);
2277
2278 error = zap_add(os, MASTER_NODE_OBJ,
2279 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2280 ASSERT0(error);
2281
2282 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2283 sa_register_update_callback(os, zfs_sa_upgrade);
2284 }
2285
2286 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2287 "from %llu to %llu", zfsvfs->z_version, newvers);
2288
2289 dmu_tx_commit(tx);
2290
2291 zfsvfs->z_version = newvers;
2292
2293 zfs_set_fuid_feature(zfsvfs);
2294
2295 return (0);
2296 }
2297
2298 /*
2299 * Read a property stored within the master node.
2300 */
2301 int
2302 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2303 {
2304 const char *pname;
2305 int error = ENOENT;
2306
2307 /*
2308 * Look up the file system's value for the property. For the
2309 * version property, we look up a slightly different string.
2310 */
2311 if (prop == ZFS_PROP_VERSION)
2312 pname = ZPL_VERSION_STR;
2313 else
2314 pname = zfs_prop_to_name(prop);
2315
2316 if (os != NULL)
2317 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2318
2319 if (error == ENOENT) {
2320 /* No value set, use the default value */
2321 switch (prop) {
2322 case ZFS_PROP_VERSION:
2323 *value = ZPL_VERSION;
2324 break;
2325 case ZFS_PROP_NORMALIZE:
2326 case ZFS_PROP_UTF8ONLY:
2327 *value = 0;
2328 break;
2329 case ZFS_PROP_CASE:
2330 *value = ZFS_CASE_SENSITIVE;
2331 break;
2332 default:
2333 return (error);
2334 }
2335 error = 0;
2336 }
2337 return (error);
2338 }
2339
2340 static vfsdef_t vfw = {
2341 VFSDEF_VERSION,
2342 MNTTYPE_ZFS,
2343 zfs_vfsinit,
2344 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2345 VSW_XID|VSW_ZMOUNT,
2346 &zfs_mntopts
2347 };
2348
2349 struct modlfs zfs_modlfs = {
2350 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2351 };