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) 2011 Pawel Jakub Dawidek <[email protected]>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
28 */
29
30 /* Portions Copyright 2010 Robert Milkowski */
31
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sysmacros.h>
37 #include <sys/kmem.h>
38 #include <sys/acl.h>
39 #include <sys/vnode.h>
40 #include <sys/vfs.h>
41 #include <sys/mntent.h>
42 #include <sys/mount.h>
43 #include <sys/cmn_err.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_vnops.h>
46 #include <sys/zfs_dir.h>
47 #include <sys/zil.h>
48 #include <sys/fs/zfs.h>
49 #include <sys/dmu.h>
50 #include <sys/dsl_prop.h>
51 #include <sys/dsl_dataset.h>
52 #include <sys/dsl_deleg.h>
53 #include <sys/spa.h>
54 #include <sys/zap.h>
55 #include <sys/sa.h>
56 #include <sys/sa_impl.h>
57 #include <sys/policy.h>
58 #include <sys/atomic.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/zfs_ctldir.h>
61 #include <sys/zfs_fuid.h>
62 #include <sys/sunddi.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/dsl_dir.h>
65 #include <sys/spa_boot.h>
66 #include <sys/jail.h>
67 #include <ufs/ufs/quota.h>
68 #include <sys/zfs_quota.h>
69
70 #include "zfs_comutil.h"
71
72 #ifndef MNTK_VMSETSIZE_BUG
73 #define MNTK_VMSETSIZE_BUG 0
74 #endif
75 #ifndef MNTK_NOMSYNC
76 #define MNTK_NOMSYNC 8
77 #endif
78
79 /* BEGIN CSTYLED */
80 struct mtx zfs_debug_mtx;
81 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
82
83 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
84
85 int zfs_super_owner;
86 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
87 "File system owner can perform privileged operation on his file systems");
88
89 int zfs_debug_level;
90 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0,
91 "Debug level");
92
93 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
94 static int zfs_version_acl = ZFS_ACL_VERSION;
95 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
96 "ZFS_ACL_VERSION");
97 static int zfs_version_spa = SPA_VERSION;
98 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
99 "SPA_VERSION");
100 static int zfs_version_zpl = ZPL_VERSION;
101 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
102 "ZPL_VERSION");
103 /* END CSTYLED */
104
105 static int zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg);
106 static int zfs_mount(vfs_t *vfsp);
107 static int zfs_umount(vfs_t *vfsp, int fflag);
108 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
109 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
110 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
111 static int zfs_sync(vfs_t *vfsp, int waitfor);
112 #if __FreeBSD_version >= 1300098
113 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp,
114 struct ucred **credanonp, int *numsecflavors, int *secflavors);
115 #else
116 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
117 struct ucred **credanonp, int *numsecflavors, int **secflavors);
118 #endif
119 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
120 static void zfs_freevfs(vfs_t *vfsp);
121
122 struct vfsops zfs_vfsops = {
123 .vfs_mount = zfs_mount,
124 .vfs_unmount = zfs_umount,
125 #if __FreeBSD_version >= 1300049
126 .vfs_root = vfs_cache_root,
127 .vfs_cachedroot = zfs_root,
128 #else
129 .vfs_root = zfs_root,
130 #endif
131 .vfs_statfs = zfs_statfs,
132 .vfs_vget = zfs_vget,
133 .vfs_sync = zfs_sync,
134 .vfs_checkexp = zfs_checkexp,
135 .vfs_fhtovp = zfs_fhtovp,
136 .vfs_quotactl = zfs_quotactl,
137 };
138
139 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
140
141 /*
142 * We need to keep a count of active fs's.
143 * This is necessary to prevent our module
144 * from being unloaded after a umount -f
145 */
146 static uint32_t zfs_active_fs_count = 0;
147
148 int
zfs_get_temporary_prop(dsl_dataset_t * ds,zfs_prop_t zfs_prop,uint64_t * val,char * setpoint)149 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
150 char *setpoint)
151 {
152 int error;
153 zfsvfs_t *zfvp;
154 vfs_t *vfsp;
155 objset_t *os;
156 uint64_t tmp = *val;
157
158 error = dmu_objset_from_ds(ds, &os);
159 if (error != 0)
160 return (error);
161
162 error = getzfsvfs_impl(os, &zfvp);
163 if (error != 0)
164 return (error);
165 if (zfvp == NULL)
166 return (ENOENT);
167 vfsp = zfvp->z_vfs;
168 switch (zfs_prop) {
169 case ZFS_PROP_ATIME:
170 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL))
171 tmp = 0;
172 if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL))
173 tmp = 1;
174 break;
175 case ZFS_PROP_DEVICES:
176 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
177 tmp = 0;
178 if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL))
179 tmp = 1;
180 break;
181 case ZFS_PROP_EXEC:
182 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL))
183 tmp = 0;
184 if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL))
185 tmp = 1;
186 break;
187 case ZFS_PROP_SETUID:
188 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL))
189 tmp = 0;
190 if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL))
191 tmp = 1;
192 break;
193 case ZFS_PROP_READONLY:
194 if (vfs_optionisset(vfsp, MNTOPT_RW, NULL))
195 tmp = 0;
196 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL))
197 tmp = 1;
198 break;
199 case ZFS_PROP_XATTR:
200 if (zfvp->z_flags & ZSB_XATTR)
201 tmp = zfvp->z_xattr;
202 break;
203 case ZFS_PROP_NBMAND:
204 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL))
205 tmp = 0;
206 if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL))
207 tmp = 1;
208 break;
209 default:
210 vfs_unbusy(vfsp);
211 return (ENOENT);
212 }
213
214 vfs_unbusy(vfsp);
215 if (tmp != *val) {
216 (void) strcpy(setpoint, "temporary");
217 *val = tmp;
218 }
219 return (0);
220 }
221
222 static int
zfs_getquota(zfsvfs_t * zfsvfs,uid_t id,int isgroup,struct dqblk64 * dqp)223 zfs_getquota(zfsvfs_t *zfsvfs, uid_t id, int isgroup, struct dqblk64 *dqp)
224 {
225 int error = 0;
226 char buf[32];
227 uint64_t usedobj, quotaobj;
228 uint64_t quota, used = 0;
229 timespec_t now;
230
231 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
232 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
233
234 if (quotaobj == 0 || zfsvfs->z_replay) {
235 error = ENOENT;
236 goto done;
237 }
238 (void) sprintf(buf, "%llx", (longlong_t)id);
239 if ((error = zap_lookup(zfsvfs->z_os, quotaobj,
240 buf, sizeof (quota), 1, "a)) != 0) {
241 dprintf("%s(%d): quotaobj lookup failed\n",
242 __FUNCTION__, __LINE__);
243 goto done;
244 }
245 /*
246 * quota(8) uses bsoftlimit as "quoota", and hardlimit as "limit".
247 * So we set them to be the same.
248 */
249 dqp->dqb_bsoftlimit = dqp->dqb_bhardlimit = btodb(quota);
250 error = zap_lookup(zfsvfs->z_os, usedobj, buf, sizeof (used), 1, &used);
251 if (error && error != ENOENT) {
252 dprintf("%s(%d): usedobj failed; %d\n",
253 __FUNCTION__, __LINE__, error);
254 goto done;
255 }
256 dqp->dqb_curblocks = btodb(used);
257 dqp->dqb_ihardlimit = dqp->dqb_isoftlimit = 0;
258 vfs_timestamp(&now);
259 /*
260 * Setting this to 0 causes FreeBSD quota(8) to print
261 * the number of days since the epoch, which isn't
262 * particularly useful.
263 */
264 dqp->dqb_btime = dqp->dqb_itime = now.tv_sec;
265 done:
266 return (error);
267 }
268
269 static int
zfs_quotactl(vfs_t * vfsp,int cmds,uid_t id,void * arg)270 zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg)
271 {
272 zfsvfs_t *zfsvfs = vfsp->vfs_data;
273 struct thread *td;
274 int cmd, type, error = 0;
275 int bitsize;
276 zfs_userquota_prop_t quota_type;
277 struct dqblk64 dqblk = { 0 };
278
279 td = curthread;
280 cmd = cmds >> SUBCMDSHIFT;
281 type = cmds & SUBCMDMASK;
282
283 ZFS_ENTER(zfsvfs);
284 if (id == -1) {
285 switch (type) {
286 case USRQUOTA:
287 id = td->td_ucred->cr_ruid;
288 break;
289 case GRPQUOTA:
290 id = td->td_ucred->cr_rgid;
291 break;
292 default:
293 error = EINVAL;
294 if (cmd == Q_QUOTAON || cmd == Q_QUOTAOFF)
295 vfs_unbusy(vfsp);
296 goto done;
297 }
298 }
299 /*
300 * Map BSD type to:
301 * ZFS_PROP_USERUSED,
302 * ZFS_PROP_USERQUOTA,
303 * ZFS_PROP_GROUPUSED,
304 * ZFS_PROP_GROUPQUOTA
305 */
306 switch (cmd) {
307 case Q_SETQUOTA:
308 case Q_SETQUOTA32:
309 if (type == USRQUOTA)
310 quota_type = ZFS_PROP_USERQUOTA;
311 else if (type == GRPQUOTA)
312 quota_type = ZFS_PROP_GROUPQUOTA;
313 else
314 error = EINVAL;
315 break;
316 case Q_GETQUOTA:
317 case Q_GETQUOTA32:
318 if (type == USRQUOTA)
319 quota_type = ZFS_PROP_USERUSED;
320 else if (type == GRPQUOTA)
321 quota_type = ZFS_PROP_GROUPUSED;
322 else
323 error = EINVAL;
324 break;
325 }
326
327 /*
328 * Depending on the cmd, we may need to get
329 * the ruid and domain (see fuidstr_to_sid?),
330 * the fuid (how?), or other information.
331 * Create fuid using zfs_fuid_create(zfsvfs, id,
332 * ZFS_OWNER or ZFS_GROUP, cr, &fuidp)?
333 * I think I can use just the id?
334 *
335 * Look at zfs_id_overquota() to look up a quota.
336 * zap_lookup(something, quotaobj, fuidstring,
337 * sizeof (long long), 1, "a)
338 *
339 * See zfs_set_userquota() to set a quota.
340 */
341 if ((uint32_t)type >= MAXQUOTAS) {
342 error = EINVAL;
343 goto done;
344 }
345
346 switch (cmd) {
347 case Q_GETQUOTASIZE:
348 bitsize = 64;
349 error = copyout(&bitsize, arg, sizeof (int));
350 break;
351 case Q_QUOTAON:
352 // As far as I can tell, you can't turn quotas on or off on zfs
353 error = 0;
354 vfs_unbusy(vfsp);
355 break;
356 case Q_QUOTAOFF:
357 error = ENOTSUP;
358 vfs_unbusy(vfsp);
359 break;
360 case Q_SETQUOTA:
361 error = copyin(arg, &dqblk, sizeof (dqblk));
362 if (error == 0)
363 error = zfs_set_userquota(zfsvfs, quota_type,
364 "", id, dbtob(dqblk.dqb_bhardlimit));
365 break;
366 case Q_GETQUOTA:
367 error = zfs_getquota(zfsvfs, id, type == GRPQUOTA, &dqblk);
368 if (error == 0)
369 error = copyout(&dqblk, arg, sizeof (dqblk));
370 break;
371 default:
372 error = EINVAL;
373 break;
374 }
375 done:
376 ZFS_EXIT(zfsvfs);
377 return (error);
378 }
379
380
381 boolean_t
zfs_is_readonly(zfsvfs_t * zfsvfs)382 zfs_is_readonly(zfsvfs_t *zfsvfs)
383 {
384 return (!!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY));
385 }
386
387 /*ARGSUSED*/
388 static int
zfs_sync(vfs_t * vfsp,int waitfor)389 zfs_sync(vfs_t *vfsp, int waitfor)
390 {
391
392 /*
393 * Data integrity is job one. We don't want a compromised kernel
394 * writing to the storage pool, so we never sync during panic.
395 */
396 if (panicstr)
397 return (0);
398
399 /*
400 * Ignore the system syncher. ZFS already commits async data
401 * at zfs_txg_timeout intervals.
402 */
403 if (waitfor == MNT_LAZY)
404 return (0);
405
406 if (vfsp != NULL) {
407 /*
408 * Sync a specific filesystem.
409 */
410 zfsvfs_t *zfsvfs = vfsp->vfs_data;
411 dsl_pool_t *dp;
412 int error;
413
414 error = vfs_stdsync(vfsp, waitfor);
415 if (error != 0)
416 return (error);
417
418 ZFS_ENTER(zfsvfs);
419 dp = dmu_objset_pool(zfsvfs->z_os);
420
421 /*
422 * If the system is shutting down, then skip any
423 * filesystems which may exist on a suspended pool.
424 */
425 if (rebooting && spa_suspended(dp->dp_spa)) {
426 ZFS_EXIT(zfsvfs);
427 return (0);
428 }
429
430 if (zfsvfs->z_log != NULL)
431 zil_commit(zfsvfs->z_log, 0);
432
433 ZFS_EXIT(zfsvfs);
434 } else {
435 /*
436 * Sync all ZFS filesystems. This is what happens when you
437 * run sync(8). Unlike other filesystems, ZFS honors the
438 * request by waiting for all pools to commit all dirty data.
439 */
440 spa_sync_allpools();
441 }
442
443 return (0);
444 }
445
446 static void
atime_changed_cb(void * arg,uint64_t newval)447 atime_changed_cb(void *arg, uint64_t newval)
448 {
449 zfsvfs_t *zfsvfs = arg;
450
451 if (newval == TRUE) {
452 zfsvfs->z_atime = TRUE;
453 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
454 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
455 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
456 } else {
457 zfsvfs->z_atime = FALSE;
458 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
459 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
460 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
461 }
462 }
463
464 static void
xattr_changed_cb(void * arg,uint64_t newval)465 xattr_changed_cb(void *arg, uint64_t newval)
466 {
467 zfsvfs_t *zfsvfs = arg;
468
469 if (newval == ZFS_XATTR_OFF) {
470 zfsvfs->z_flags &= ~ZSB_XATTR;
471 } else {
472 zfsvfs->z_flags |= ZSB_XATTR;
473
474 if (newval == ZFS_XATTR_SA)
475 zfsvfs->z_xattr_sa = B_TRUE;
476 else
477 zfsvfs->z_xattr_sa = B_FALSE;
478 }
479 }
480
481 static void
blksz_changed_cb(void * arg,uint64_t newval)482 blksz_changed_cb(void *arg, uint64_t newval)
483 {
484 zfsvfs_t *zfsvfs = arg;
485 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
486 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
487 ASSERT(ISP2(newval));
488
489 zfsvfs->z_max_blksz = newval;
490 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
491 }
492
493 static void
readonly_changed_cb(void * arg,uint64_t newval)494 readonly_changed_cb(void *arg, uint64_t newval)
495 {
496 zfsvfs_t *zfsvfs = arg;
497
498 if (newval) {
499 /* XXX locking on vfs_flag? */
500 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
501 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
502 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
503 } else {
504 /* XXX locking on vfs_flag? */
505 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
506 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
507 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
508 }
509 }
510
511 static void
setuid_changed_cb(void * arg,uint64_t newval)512 setuid_changed_cb(void *arg, uint64_t newval)
513 {
514 zfsvfs_t *zfsvfs = arg;
515
516 if (newval == FALSE) {
517 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
518 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
519 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
520 } else {
521 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
522 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
523 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
524 }
525 }
526
527 static void
exec_changed_cb(void * arg,uint64_t newval)528 exec_changed_cb(void *arg, uint64_t newval)
529 {
530 zfsvfs_t *zfsvfs = arg;
531
532 if (newval == FALSE) {
533 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
534 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
535 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
536 } else {
537 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
538 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
539 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
540 }
541 }
542
543 /*
544 * The nbmand mount option can be changed at mount time.
545 * We can't allow it to be toggled on live file systems or incorrect
546 * behavior may be seen from cifs clients
547 *
548 * This property isn't registered via dsl_prop_register(), but this callback
549 * will be called when a file system is first mounted
550 */
551 static void
nbmand_changed_cb(void * arg,uint64_t newval)552 nbmand_changed_cb(void *arg, uint64_t newval)
553 {
554 zfsvfs_t *zfsvfs = arg;
555 if (newval == FALSE) {
556 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
557 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
558 } else {
559 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
560 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
561 }
562 }
563
564 static void
snapdir_changed_cb(void * arg,uint64_t newval)565 snapdir_changed_cb(void *arg, uint64_t newval)
566 {
567 zfsvfs_t *zfsvfs = arg;
568
569 zfsvfs->z_show_ctldir = newval;
570 }
571
572 static void
vscan_changed_cb(void * arg,uint64_t newval)573 vscan_changed_cb(void *arg, uint64_t newval)
574 {
575 zfsvfs_t *zfsvfs = arg;
576
577 zfsvfs->z_vscan = newval;
578 }
579
580 static void
acl_mode_changed_cb(void * arg,uint64_t newval)581 acl_mode_changed_cb(void *arg, uint64_t newval)
582 {
583 zfsvfs_t *zfsvfs = arg;
584
585 zfsvfs->z_acl_mode = newval;
586 }
587
588 static void
acl_inherit_changed_cb(void * arg,uint64_t newval)589 acl_inherit_changed_cb(void *arg, uint64_t newval)
590 {
591 zfsvfs_t *zfsvfs = arg;
592
593 zfsvfs->z_acl_inherit = newval;
594 }
595
596 static void
acl_type_changed_cb(void * arg,uint64_t newval)597 acl_type_changed_cb(void *arg, uint64_t newval)
598 {
599 zfsvfs_t *zfsvfs = arg;
600
601 zfsvfs->z_acl_type = newval;
602 }
603
604 static int
zfs_register_callbacks(vfs_t * vfsp)605 zfs_register_callbacks(vfs_t *vfsp)
606 {
607 struct dsl_dataset *ds = NULL;
608 objset_t *os = NULL;
609 zfsvfs_t *zfsvfs = NULL;
610 uint64_t nbmand;
611 boolean_t readonly = B_FALSE;
612 boolean_t do_readonly = B_FALSE;
613 boolean_t setuid = B_FALSE;
614 boolean_t do_setuid = B_FALSE;
615 boolean_t exec = B_FALSE;
616 boolean_t do_exec = B_FALSE;
617 boolean_t xattr = B_FALSE;
618 boolean_t atime = B_FALSE;
619 boolean_t do_atime = B_FALSE;
620 boolean_t do_xattr = B_FALSE;
621 int error = 0;
622
623 ASSERT(vfsp);
624 zfsvfs = vfsp->vfs_data;
625 ASSERT(zfsvfs);
626 os = zfsvfs->z_os;
627
628 /*
629 * This function can be called for a snapshot when we update snapshot's
630 * mount point, which isn't really supported.
631 */
632 if (dmu_objset_is_snapshot(os))
633 return (EOPNOTSUPP);
634
635 /*
636 * The act of registering our callbacks will destroy any mount
637 * options we may have. In order to enable temporary overrides
638 * of mount options, we stash away the current values and
639 * restore them after we register the callbacks.
640 */
641 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
642 !spa_writeable(dmu_objset_spa(os))) {
643 readonly = B_TRUE;
644 do_readonly = B_TRUE;
645 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
646 readonly = B_FALSE;
647 do_readonly = B_TRUE;
648 }
649 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
650 setuid = B_FALSE;
651 do_setuid = B_TRUE;
652 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
653 setuid = B_TRUE;
654 do_setuid = B_TRUE;
655 }
656 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
657 exec = B_FALSE;
658 do_exec = B_TRUE;
659 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
660 exec = B_TRUE;
661 do_exec = B_TRUE;
662 }
663 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
664 zfsvfs->z_xattr = xattr = ZFS_XATTR_OFF;
665 do_xattr = B_TRUE;
666 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
667 zfsvfs->z_xattr = xattr = ZFS_XATTR_DIR;
668 do_xattr = B_TRUE;
669 } else if (vfs_optionisset(vfsp, MNTOPT_DIRXATTR, NULL)) {
670 zfsvfs->z_xattr = xattr = ZFS_XATTR_DIR;
671 do_xattr = B_TRUE;
672 } else if (vfs_optionisset(vfsp, MNTOPT_SAXATTR, NULL)) {
673 zfsvfs->z_xattr = xattr = ZFS_XATTR_SA;
674 do_xattr = B_TRUE;
675 }
676 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
677 atime = B_FALSE;
678 do_atime = B_TRUE;
679 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
680 atime = B_TRUE;
681 do_atime = B_TRUE;
682 }
683
684 /*
685 * We need to enter pool configuration here, so that we can use
686 * dsl_prop_get_int_ds() to handle the special nbmand property below.
687 * dsl_prop_get_integer() can not be used, because it has to acquire
688 * spa_namespace_lock and we can not do that because we already hold
689 * z_teardown_lock. The problem is that spa_write_cachefile() is called
690 * with spa_namespace_lock held and the function calls ZFS vnode
691 * operations to write the cache file and thus z_teardown_lock is
692 * acquired after spa_namespace_lock.
693 */
694 ds = dmu_objset_ds(os);
695 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
696
697 /*
698 * nbmand is a special property. It can only be changed at
699 * mount time.
700 *
701 * This is weird, but it is documented to only be changeable
702 * at mount time.
703 */
704 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
705 nbmand = B_FALSE;
706 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
707 nbmand = B_TRUE;
708 } else if ((error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0)) {
709 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
710 return (error);
711 }
712
713 /*
714 * Register property callbacks.
715 *
716 * It would probably be fine to just check for i/o error from
717 * the first prop_register(), but I guess I like to go
718 * overboard...
719 */
720 error = dsl_prop_register(ds,
721 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
722 error = error ? error : dsl_prop_register(ds,
723 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
724 error = error ? error : dsl_prop_register(ds,
725 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
726 error = error ? error : dsl_prop_register(ds,
727 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
728 error = error ? error : dsl_prop_register(ds,
729 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
730 error = error ? error : dsl_prop_register(ds,
731 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
732 error = error ? error : dsl_prop_register(ds,
733 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
734 error = error ? error : dsl_prop_register(ds,
735 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acl_type_changed_cb, zfsvfs);
736 error = error ? error : dsl_prop_register(ds,
737 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
738 error = error ? error : dsl_prop_register(ds,
739 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
740 zfsvfs);
741 error = error ? error : dsl_prop_register(ds,
742 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
743 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
744 if (error)
745 goto unregister;
746
747 /*
748 * Invoke our callbacks to restore temporary mount options.
749 */
750 if (do_readonly)
751 readonly_changed_cb(zfsvfs, readonly);
752 if (do_setuid)
753 setuid_changed_cb(zfsvfs, setuid);
754 if (do_exec)
755 exec_changed_cb(zfsvfs, exec);
756 if (do_xattr)
757 xattr_changed_cb(zfsvfs, xattr);
758 if (do_atime)
759 atime_changed_cb(zfsvfs, atime);
760
761 nbmand_changed_cb(zfsvfs, nbmand);
762
763 return (0);
764
765 unregister:
766 dsl_prop_unregister_all(ds, zfsvfs);
767 return (error);
768 }
769
770 /*
771 * Associate this zfsvfs with the given objset, which must be owned.
772 * This will cache a bunch of on-disk state from the objset in the
773 * zfsvfs.
774 */
775 static int
zfsvfs_init(zfsvfs_t * zfsvfs,objset_t * os)776 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
777 {
778 int error;
779 uint64_t val;
780
781 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
782 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
783 zfsvfs->z_os = os;
784
785 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
786 if (error != 0)
787 return (error);
788 if (zfsvfs->z_version >
789 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
790 (void) printf("Can't mount a version %lld file system "
791 "on a version %lld pool\n. Pool must be upgraded to mount "
792 "this file system.", (u_longlong_t)zfsvfs->z_version,
793 (u_longlong_t)spa_version(dmu_objset_spa(os)));
794 return (SET_ERROR(ENOTSUP));
795 }
796 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
797 if (error != 0)
798 return (error);
799 zfsvfs->z_norm = (int)val;
800
801 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
802 if (error != 0)
803 return (error);
804 zfsvfs->z_utf8 = (val != 0);
805
806 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
807 if (error != 0)
808 return (error);
809 zfsvfs->z_case = (uint_t)val;
810
811 error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val);
812 if (error != 0)
813 return (error);
814 zfsvfs->z_acl_type = (uint_t)val;
815
816 /*
817 * Fold case on file systems that are always or sometimes case
818 * insensitive.
819 */
820 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
821 zfsvfs->z_case == ZFS_CASE_MIXED)
822 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
823
824 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
825 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
826
827 uint64_t sa_obj = 0;
828 if (zfsvfs->z_use_sa) {
829 /* should either have both of these objects or none */
830 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
831 &sa_obj);
832 if (error != 0)
833 return (error);
834 }
835
836 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
837 &zfsvfs->z_attr_table);
838 if (error != 0)
839 return (error);
840
841 if (zfsvfs->z_version >= ZPL_VERSION_SA)
842 sa_register_update_callback(os, zfs_sa_upgrade);
843
844 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
845 &zfsvfs->z_root);
846 if (error != 0)
847 return (error);
848 ASSERT(zfsvfs->z_root != 0);
849
850 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
851 &zfsvfs->z_unlinkedobj);
852 if (error != 0)
853 return (error);
854
855 error = zap_lookup(os, MASTER_NODE_OBJ,
856 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
857 8, 1, &zfsvfs->z_userquota_obj);
858 if (error == ENOENT)
859 zfsvfs->z_userquota_obj = 0;
860 else if (error != 0)
861 return (error);
862
863 error = zap_lookup(os, MASTER_NODE_OBJ,
864 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
865 8, 1, &zfsvfs->z_groupquota_obj);
866 if (error == ENOENT)
867 zfsvfs->z_groupquota_obj = 0;
868 else if (error != 0)
869 return (error);
870
871 error = zap_lookup(os, MASTER_NODE_OBJ,
872 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
873 8, 1, &zfsvfs->z_projectquota_obj);
874 if (error == ENOENT)
875 zfsvfs->z_projectquota_obj = 0;
876 else if (error != 0)
877 return (error);
878
879 error = zap_lookup(os, MASTER_NODE_OBJ,
880 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
881 8, 1, &zfsvfs->z_userobjquota_obj);
882 if (error == ENOENT)
883 zfsvfs->z_userobjquota_obj = 0;
884 else if (error != 0)
885 return (error);
886
887 error = zap_lookup(os, MASTER_NODE_OBJ,
888 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
889 8, 1, &zfsvfs->z_groupobjquota_obj);
890 if (error == ENOENT)
891 zfsvfs->z_groupobjquota_obj = 0;
892 else if (error != 0)
893 return (error);
894
895 error = zap_lookup(os, MASTER_NODE_OBJ,
896 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
897 8, 1, &zfsvfs->z_projectobjquota_obj);
898 if (error == ENOENT)
899 zfsvfs->z_projectobjquota_obj = 0;
900 else if (error != 0)
901 return (error);
902
903 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
904 &zfsvfs->z_fuid_obj);
905 if (error == ENOENT)
906 zfsvfs->z_fuid_obj = 0;
907 else if (error != 0)
908 return (error);
909
910 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
911 &zfsvfs->z_shares_dir);
912 if (error == ENOENT)
913 zfsvfs->z_shares_dir = 0;
914 else if (error != 0)
915 return (error);
916
917 /*
918 * Only use the name cache if we are looking for a
919 * name on a file system that does not require normalization
920 * or case folding. We can also look there if we happen to be
921 * on a non-normalizing, mixed sensitivity file system IF we
922 * are looking for the exact name (which is always the case on
923 * FreeBSD).
924 */
925 zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
926 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
927 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
928
929 return (0);
930 }
931
932 taskq_t *zfsvfs_taskq;
933
934 static void
zfsvfs_task_unlinked_drain(void * context,int pending __unused)935 zfsvfs_task_unlinked_drain(void *context, int pending __unused)
936 {
937
938 zfs_unlinked_drain((zfsvfs_t *)context);
939 }
940
941 int
zfsvfs_create(const char * osname,boolean_t readonly,zfsvfs_t ** zfvp)942 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
943 {
944 objset_t *os;
945 zfsvfs_t *zfsvfs;
946 int error;
947 boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
948
949 /*
950 * XXX: Fix struct statfs so this isn't necessary!
951 *
952 * The 'osname' is used as the filesystem's special node, which means
953 * it must fit in statfs.f_mntfromname, or else it can't be
954 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
955 * 'zfs unmount' to think it's not mounted when it is.
956 */
957 if (strlen(osname) >= MNAMELEN)
958 return (SET_ERROR(ENAMETOOLONG));
959
960 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
961
962 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs,
963 &os);
964 if (error != 0) {
965 kmem_free(zfsvfs, sizeof (zfsvfs_t));
966 return (error);
967 }
968
969 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
970
971 return (error);
972 }
973
974
975 int
zfsvfs_create_impl(zfsvfs_t ** zfvp,zfsvfs_t * zfsvfs,objset_t * os)976 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
977 {
978 int error;
979
980 zfsvfs->z_vfs = NULL;
981 zfsvfs->z_parent = zfsvfs;
982
983 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
984 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
985 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
986 offsetof(znode_t, z_link_node));
987 TASK_INIT(&zfsvfs->z_unlinked_drain_task, 0,
988 zfsvfs_task_unlinked_drain, zfsvfs);
989 ZFS_TEARDOWN_INIT(zfsvfs);
990 ZFS_TEARDOWN_INACTIVE_INIT(zfsvfs);
991 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
992 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
993 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
994
995 error = zfsvfs_init(zfsvfs, os);
996 if (error != 0) {
997 dmu_objset_disown(os, B_TRUE, zfsvfs);
998 *zfvp = NULL;
999 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1000 return (error);
1001 }
1002
1003 *zfvp = zfsvfs;
1004 return (0);
1005 }
1006
1007 static int
zfsvfs_setup(zfsvfs_t * zfsvfs,boolean_t mounting)1008 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1009 {
1010 int error;
1011
1012 /*
1013 * Check for a bad on-disk format version now since we
1014 * lied about owning the dataset readonly before.
1015 */
1016 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) &&
1017 dmu_objset_incompatible_encryption_version(zfsvfs->z_os))
1018 return (SET_ERROR(EROFS));
1019
1020 error = zfs_register_callbacks(zfsvfs->z_vfs);
1021 if (error)
1022 return (error);
1023
1024 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1025
1026 /*
1027 * If we are not mounting (ie: online recv), then we don't
1028 * have to worry about replaying the log as we blocked all
1029 * operations out since we closed the ZIL.
1030 */
1031 if (mounting) {
1032 boolean_t readonly;
1033
1034 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
1035 dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
1036
1037 /*
1038 * During replay we remove the read only flag to
1039 * allow replays to succeed.
1040 */
1041 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1042 if (readonly != 0) {
1043 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1044 } else {
1045 dsl_dir_t *dd;
1046 zap_stats_t zs;
1047
1048 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
1049 &zs) == 0) {
1050 dataset_kstats_update_nunlinks_kstat(
1051 &zfsvfs->z_kstat, zs.zs_num_entries);
1052 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1053 "num_entries in unlinked set: %llu",
1054 zs.zs_num_entries);
1055 }
1056
1057 zfs_unlinked_drain(zfsvfs);
1058 dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
1059 dd->dd_activity_cancelled = B_FALSE;
1060 }
1061
1062 /*
1063 * Parse and replay the intent log.
1064 *
1065 * Because of ziltest, this must be done after
1066 * zfs_unlinked_drain(). (Further note: ziltest
1067 * doesn't use readonly mounts, where
1068 * zfs_unlinked_drain() isn't called.) This is because
1069 * ziltest causes spa_sync() to think it's committed,
1070 * but actually it is not, so the intent log contains
1071 * many txg's worth of changes.
1072 *
1073 * In particular, if object N is in the unlinked set in
1074 * the last txg to actually sync, then it could be
1075 * actually freed in a later txg and then reallocated
1076 * in a yet later txg. This would write a "create
1077 * object N" record to the intent log. Normally, this
1078 * would be fine because the spa_sync() would have
1079 * written out the fact that object N is free, before
1080 * we could write the "create object N" intent log
1081 * record.
1082 *
1083 * But when we are in ziltest mode, we advance the "open
1084 * txg" without actually spa_sync()-ing the changes to
1085 * disk. So we would see that object N is still
1086 * allocated and in the unlinked set, and there is an
1087 * intent log record saying to allocate it.
1088 */
1089 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1090 if (zil_replay_disable) {
1091 zil_destroy(zfsvfs->z_log, B_FALSE);
1092 } else {
1093 boolean_t use_nc = zfsvfs->z_use_namecache;
1094 zfsvfs->z_use_namecache = B_FALSE;
1095 zfsvfs->z_replay = B_TRUE;
1096 zil_replay(zfsvfs->z_os, zfsvfs,
1097 zfs_replay_vector);
1098 zfsvfs->z_replay = B_FALSE;
1099 zfsvfs->z_use_namecache = use_nc;
1100 }
1101 }
1102
1103 /* restore readonly bit */
1104 if (readonly != 0)
1105 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
1106 }
1107
1108 /*
1109 * Set the objset user_ptr to track its zfsvfs.
1110 */
1111 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1112 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1113 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1114
1115 return (0);
1116 }
1117
1118 void
zfsvfs_free(zfsvfs_t * zfsvfs)1119 zfsvfs_free(zfsvfs_t *zfsvfs)
1120 {
1121 int i;
1122
1123 zfs_fuid_destroy(zfsvfs);
1124
1125 mutex_destroy(&zfsvfs->z_znodes_lock);
1126 mutex_destroy(&zfsvfs->z_lock);
1127 ASSERT(zfsvfs->z_nr_znodes == 0);
1128 list_destroy(&zfsvfs->z_all_znodes);
1129 ZFS_TEARDOWN_DESTROY(zfsvfs);
1130 ZFS_TEARDOWN_INACTIVE_DESTROY(zfsvfs);
1131 rw_destroy(&zfsvfs->z_fuid_lock);
1132 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1133 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1134 dataset_kstats_destroy(&zfsvfs->z_kstat);
1135 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1136 }
1137
1138 static void
zfs_set_fuid_feature(zfsvfs_t * zfsvfs)1139 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1140 {
1141 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1142 if (zfsvfs->z_vfs) {
1143 if (zfsvfs->z_use_fuids) {
1144 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1145 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1146 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1147 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1148 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1149 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1150 } else {
1151 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1152 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1153 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1154 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1155 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1156 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1157 }
1158 }
1159 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1160 }
1161
1162 static int
zfs_domount(vfs_t * vfsp,char * osname)1163 zfs_domount(vfs_t *vfsp, char *osname)
1164 {
1165 uint64_t recordsize, fsid_guid;
1166 int error = 0;
1167 zfsvfs_t *zfsvfs;
1168
1169 ASSERT(vfsp);
1170 ASSERT(osname);
1171
1172 error = zfsvfs_create(osname, vfsp->mnt_flag & MNT_RDONLY, &zfsvfs);
1173 if (error)
1174 return (error);
1175 zfsvfs->z_vfs = vfsp;
1176
1177 if ((error = dsl_prop_get_integer(osname,
1178 "recordsize", &recordsize, NULL)))
1179 goto out;
1180 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1181 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1182
1183 vfsp->vfs_data = zfsvfs;
1184 vfsp->mnt_flag |= MNT_LOCAL;
1185 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1186 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1187 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1188 /*
1189 * This can cause a loss of coherence between ARC and page cache
1190 * on ZoF - unclear if the problem is in FreeBSD or ZoF
1191 */
1192 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1193 vfsp->mnt_kern_flag |= MNTK_NOMSYNC;
1194 vfsp->mnt_kern_flag |= MNTK_VMSETSIZE_BUG;
1195
1196 #if defined(_KERNEL) && !defined(KMEM_DEBUG)
1197 vfsp->mnt_kern_flag |= MNTK_FPLOOKUP;
1198 #endif
1199 /*
1200 * The fsid is 64 bits, composed of an 8-bit fs type, which
1201 * separates our fsid from any other filesystem types, and a
1202 * 56-bit objset unique ID. The objset unique ID is unique to
1203 * all objsets open on this system, provided by unique_create().
1204 * The 8-bit fs type must be put in the low bits of fsid[1]
1205 * because that's where other Solaris filesystems put it.
1206 */
1207 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1208 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1209 vfsp->vfs_fsid.val[0] = fsid_guid;
1210 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1211 (vfsp->mnt_vfc->vfc_typenum & 0xFF);
1212
1213 /*
1214 * Set features for file system.
1215 */
1216 zfs_set_fuid_feature(zfsvfs);
1217 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1218 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1219 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1220 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1221 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1222 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1223 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1224 }
1225 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1226
1227 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1228 uint64_t pval;
1229
1230 atime_changed_cb(zfsvfs, B_FALSE);
1231 readonly_changed_cb(zfsvfs, B_TRUE);
1232 if ((error = dsl_prop_get_integer(osname,
1233 "xattr", &pval, NULL)))
1234 goto out;
1235 xattr_changed_cb(zfsvfs, pval);
1236 if ((error = dsl_prop_get_integer(osname,
1237 "acltype", &pval, NULL)))
1238 goto out;
1239 acl_type_changed_cb(zfsvfs, pval);
1240 zfsvfs->z_issnap = B_TRUE;
1241 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1242
1243 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1244 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1245 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1246 } else {
1247 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1248 goto out;
1249 }
1250
1251 vfs_mountedfrom(vfsp, osname);
1252
1253 if (!zfsvfs->z_issnap)
1254 zfsctl_create(zfsvfs);
1255 out:
1256 if (error) {
1257 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1258 zfsvfs_free(zfsvfs);
1259 } else {
1260 atomic_inc_32(&zfs_active_fs_count);
1261 }
1262
1263 return (error);
1264 }
1265
1266 static void
zfs_unregister_callbacks(zfsvfs_t * zfsvfs)1267 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1268 {
1269 objset_t *os = zfsvfs->z_os;
1270
1271 if (!dmu_objset_is_snapshot(os))
1272 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1273 }
1274
1275 static int
getpoolname(const char * osname,char * poolname)1276 getpoolname(const char *osname, char *poolname)
1277 {
1278 char *p;
1279
1280 p = strchr(osname, '/');
1281 if (p == NULL) {
1282 if (strlen(osname) >= MAXNAMELEN)
1283 return (ENAMETOOLONG);
1284 (void) strcpy(poolname, osname);
1285 } else {
1286 if (p - osname >= MAXNAMELEN)
1287 return (ENAMETOOLONG);
1288 (void) strncpy(poolname, osname, p - osname);
1289 poolname[p - osname] = '\0';
1290 }
1291 return (0);
1292 }
1293
1294 static void
fetch_osname_options(char * name,bool * checkpointrewind)1295 fetch_osname_options(char *name, bool *checkpointrewind)
1296 {
1297
1298 if (name[0] == '!') {
1299 *checkpointrewind = true;
1300 memmove(name, name + 1, strlen(name));
1301 } else {
1302 *checkpointrewind = false;
1303 }
1304 }
1305
1306 /*ARGSUSED*/
1307 static int
zfs_mount(vfs_t * vfsp)1308 zfs_mount(vfs_t *vfsp)
1309 {
1310 kthread_t *td = curthread;
1311 vnode_t *mvp = vfsp->mnt_vnodecovered;
1312 cred_t *cr = td->td_ucred;
1313 char *osname;
1314 int error = 0;
1315 int canwrite;
1316 bool checkpointrewind;
1317
1318 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1319 return (SET_ERROR(EINVAL));
1320
1321 /*
1322 * If full-owner-access is enabled and delegated administration is
1323 * turned on, we must set nosuid.
1324 */
1325 if (zfs_super_owner &&
1326 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1327 secpolicy_fs_mount_clearopts(cr, vfsp);
1328 }
1329
1330 fetch_osname_options(osname, &checkpointrewind);
1331
1332 /*
1333 * Check for mount privilege?
1334 *
1335 * If we don't have privilege then see if
1336 * we have local permission to allow it
1337 */
1338 error = secpolicy_fs_mount(cr, mvp, vfsp);
1339 if (error) {
1340 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1341 goto out;
1342
1343 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1344 vattr_t vattr;
1345
1346 /*
1347 * Make sure user is the owner of the mount point
1348 * or has sufficient privileges.
1349 */
1350
1351 vattr.va_mask = AT_UID;
1352
1353 vn_lock(mvp, LK_SHARED | LK_RETRY);
1354 if (VOP_GETATTR(mvp, &vattr, cr)) {
1355 VOP_UNLOCK1(mvp);
1356 goto out;
1357 }
1358
1359 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1360 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1361 VOP_UNLOCK1(mvp);
1362 goto out;
1363 }
1364 VOP_UNLOCK1(mvp);
1365 }
1366
1367 secpolicy_fs_mount_clearopts(cr, vfsp);
1368 }
1369
1370 /*
1371 * Refuse to mount a filesystem if we are in a local zone and the
1372 * dataset is not visible.
1373 */
1374 if (!INGLOBALZONE(curproc) &&
1375 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1376 error = SET_ERROR(EPERM);
1377 goto out;
1378 }
1379
1380 vfsp->vfs_flag |= MNT_NFS4ACLS;
1381
1382 /*
1383 * When doing a remount, we simply refresh our temporary properties
1384 * according to those options set in the current VFS options.
1385 */
1386 if (vfsp->vfs_flag & MS_REMOUNT) {
1387 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1388
1389 /*
1390 * Refresh mount options with z_teardown_lock blocking I/O while
1391 * the filesystem is in an inconsistent state.
1392 * The lock also serializes this code with filesystem
1393 * manipulations between entry to zfs_suspend_fs() and return
1394 * from zfs_resume_fs().
1395 */
1396 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1397 zfs_unregister_callbacks(zfsvfs);
1398 error = zfs_register_callbacks(vfsp);
1399 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1400 goto out;
1401 }
1402
1403 /* Initial root mount: try hard to import the requested root pool. */
1404 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1405 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1406 char pname[MAXNAMELEN];
1407
1408 error = getpoolname(osname, pname);
1409 if (error == 0)
1410 error = spa_import_rootpool(pname, checkpointrewind);
1411 if (error)
1412 goto out;
1413 }
1414 DROP_GIANT();
1415 error = zfs_domount(vfsp, osname);
1416 PICKUP_GIANT();
1417
1418 out:
1419 return (error);
1420 }
1421
1422 static int
zfs_statfs(vfs_t * vfsp,struct statfs * statp)1423 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1424 {
1425 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1426 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1427
1428 statp->f_version = STATFS_VERSION;
1429
1430 ZFS_ENTER(zfsvfs);
1431
1432 dmu_objset_space(zfsvfs->z_os,
1433 &refdbytes, &availbytes, &usedobjs, &availobjs);
1434
1435 /*
1436 * The underlying storage pool actually uses multiple block sizes.
1437 * We report the fragsize as the smallest block size we support,
1438 * and we report our blocksize as the filesystem's maximum blocksize.
1439 */
1440 statp->f_bsize = SPA_MINBLOCKSIZE;
1441 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1442
1443 /*
1444 * The following report "total" blocks of various kinds in the
1445 * file system, but reported in terms of f_frsize - the
1446 * "fragment" size.
1447 */
1448
1449 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1450 statp->f_bfree = availbytes / statp->f_bsize;
1451 statp->f_bavail = statp->f_bfree; /* no root reservation */
1452
1453 /*
1454 * statvfs() should really be called statufs(), because it assumes
1455 * static metadata. ZFS doesn't preallocate files, so the best
1456 * we can do is report the max that could possibly fit in f_files,
1457 * and that minus the number actually used in f_ffree.
1458 * For f_ffree, report the smaller of the number of object available
1459 * and the number of blocks (each object will take at least a block).
1460 */
1461 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1462 statp->f_files = statp->f_ffree + usedobjs;
1463
1464 /*
1465 * We're a zfs filesystem.
1466 */
1467 strlcpy(statp->f_fstypename, "zfs",
1468 sizeof (statp->f_fstypename));
1469
1470 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1471 sizeof (statp->f_mntfromname));
1472 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1473 sizeof (statp->f_mntonname));
1474
1475 statp->f_namemax = MAXNAMELEN - 1;
1476
1477 ZFS_EXIT(zfsvfs);
1478 return (0);
1479 }
1480
1481 static int
zfs_root(vfs_t * vfsp,int flags,vnode_t ** vpp)1482 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1483 {
1484 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1485 znode_t *rootzp;
1486 int error;
1487
1488 ZFS_ENTER(zfsvfs);
1489
1490 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1491 if (error == 0)
1492 *vpp = ZTOV(rootzp);
1493
1494 ZFS_EXIT(zfsvfs);
1495
1496 if (error == 0) {
1497 error = vn_lock(*vpp, flags);
1498 if (error != 0) {
1499 VN_RELE(*vpp);
1500 *vpp = NULL;
1501 }
1502 }
1503 return (error);
1504 }
1505
1506 /*
1507 * Teardown the zfsvfs::z_os.
1508 *
1509 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1510 * and 'z_teardown_inactive_lock' held.
1511 */
1512 static int
zfsvfs_teardown(zfsvfs_t * zfsvfs,boolean_t unmounting)1513 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1514 {
1515 znode_t *zp;
1516 dsl_dir_t *dd;
1517
1518 /*
1519 * If someone has not already unmounted this file system,
1520 * drain the zrele_taskq to ensure all active references to the
1521 * zfsvfs_t have been handled only then can it be safely destroyed.
1522 */
1523 if (zfsvfs->z_os) {
1524 /*
1525 * If we're unmounting we have to wait for the list to
1526 * drain completely.
1527 *
1528 * If we're not unmounting there's no guarantee the list
1529 * will drain completely, but zreles run from the taskq
1530 * may add the parents of dir-based xattrs to the taskq
1531 * so we want to wait for these.
1532 *
1533 * We can safely read z_nr_znodes without locking because the
1534 * VFS has already blocked operations which add to the
1535 * z_all_znodes list and thus increment z_nr_znodes.
1536 */
1537 int round = 0;
1538 while (zfsvfs->z_nr_znodes > 0) {
1539 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1540 dmu_objset_pool(zfsvfs->z_os)), 0);
1541 if (++round > 1 && !unmounting)
1542 break;
1543 }
1544 }
1545 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1546
1547 if (!unmounting) {
1548 /*
1549 * We purge the parent filesystem's vfsp as the parent
1550 * filesystem and all of its snapshots have their vnode's
1551 * v_vfsp set to the parent's filesystem's vfsp. Note,
1552 * 'z_parent' is self referential for non-snapshots.
1553 */
1554 #ifdef FREEBSD_NAMECACHE
1555 #if __FreeBSD_version >= 1300117
1556 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1557 #else
1558 cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
1559 #endif
1560 #endif
1561 }
1562
1563 /*
1564 * Close the zil. NB: Can't close the zil while zfs_inactive
1565 * threads are blocked as zil_close can call zfs_inactive.
1566 */
1567 if (zfsvfs->z_log) {
1568 zil_close(zfsvfs->z_log);
1569 zfsvfs->z_log = NULL;
1570 }
1571
1572 ZFS_TEARDOWN_INACTIVE_ENTER_WRITE(zfsvfs);
1573
1574 /*
1575 * If we are not unmounting (ie: online recv) and someone already
1576 * unmounted this file system while we were doing the switcheroo,
1577 * or a reopen of z_os failed then just bail out now.
1578 */
1579 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1580 ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs);
1581 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1582 return (SET_ERROR(EIO));
1583 }
1584
1585 /*
1586 * At this point there are no vops active, and any new vops will
1587 * fail with EIO since we have z_teardown_lock for writer (only
1588 * relevant for forced unmount).
1589 *
1590 * Release all holds on dbufs.
1591 */
1592 mutex_enter(&zfsvfs->z_znodes_lock);
1593 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1594 zp = list_next(&zfsvfs->z_all_znodes, zp))
1595 if (zp->z_sa_hdl) {
1596 ASSERT(ZTOV(zp)->v_count >= 0);
1597 zfs_znode_dmu_fini(zp);
1598 }
1599 mutex_exit(&zfsvfs->z_znodes_lock);
1600
1601 /*
1602 * If we are unmounting, set the unmounted flag and let new vops
1603 * unblock. zfs_inactive will have the unmounted behavior, and all
1604 * other vops will fail with EIO.
1605 */
1606 if (unmounting) {
1607 zfsvfs->z_unmounted = B_TRUE;
1608 ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs);
1609 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1610 }
1611
1612 /*
1613 * z_os will be NULL if there was an error in attempting to reopen
1614 * zfsvfs, so just return as the properties had already been
1615 * unregistered and cached data had been evicted before.
1616 */
1617 if (zfsvfs->z_os == NULL)
1618 return (0);
1619
1620 /*
1621 * Unregister properties.
1622 */
1623 zfs_unregister_callbacks(zfsvfs);
1624
1625 /*
1626 * Evict cached data
1627 */
1628 if (!zfs_is_readonly(zfsvfs))
1629 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1630 dmu_objset_evict_dbufs(zfsvfs->z_os);
1631 dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
1632 dsl_dir_cancel_waiters(dd);
1633
1634 return (0);
1635 }
1636
1637 /*ARGSUSED*/
1638 static int
zfs_umount(vfs_t * vfsp,int fflag)1639 zfs_umount(vfs_t *vfsp, int fflag)
1640 {
1641 kthread_t *td = curthread;
1642 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1643 objset_t *os;
1644 cred_t *cr = td->td_ucred;
1645 int ret;
1646
1647 ret = secpolicy_fs_unmount(cr, vfsp);
1648 if (ret) {
1649 if (dsl_deleg_access((char *)vfsp->vfs_resource,
1650 ZFS_DELEG_PERM_MOUNT, cr))
1651 return (ret);
1652 }
1653
1654 /*
1655 * Unmount any snapshots mounted under .zfs before unmounting the
1656 * dataset itself.
1657 */
1658 if (zfsvfs->z_ctldir != NULL) {
1659 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1660 return (ret);
1661 }
1662
1663 if (fflag & MS_FORCE) {
1664 /*
1665 * Mark file system as unmounted before calling
1666 * vflush(FORCECLOSE). This way we ensure no future vnops
1667 * will be called and risk operating on DOOMED vnodes.
1668 */
1669 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1670 zfsvfs->z_unmounted = B_TRUE;
1671 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1672 }
1673
1674 /*
1675 * Flush all the files.
1676 */
1677 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1678 if (ret != 0)
1679 return (ret);
1680 while (taskqueue_cancel(zfsvfs_taskq->tq_queue,
1681 &zfsvfs->z_unlinked_drain_task, NULL) != 0)
1682 taskqueue_drain(zfsvfs_taskq->tq_queue,
1683 &zfsvfs->z_unlinked_drain_task);
1684
1685 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1686 os = zfsvfs->z_os;
1687
1688 /*
1689 * z_os will be NULL if there was an error in
1690 * attempting to reopen zfsvfs.
1691 */
1692 if (os != NULL) {
1693 /*
1694 * Unset the objset user_ptr.
1695 */
1696 mutex_enter(&os->os_user_ptr_lock);
1697 dmu_objset_set_user(os, NULL);
1698 mutex_exit(&os->os_user_ptr_lock);
1699
1700 /*
1701 * Finally release the objset
1702 */
1703 dmu_objset_disown(os, B_TRUE, zfsvfs);
1704 }
1705
1706 /*
1707 * We can now safely destroy the '.zfs' directory node.
1708 */
1709 if (zfsvfs->z_ctldir != NULL)
1710 zfsctl_destroy(zfsvfs);
1711 zfs_freevfs(vfsp);
1712
1713 return (0);
1714 }
1715
1716 static int
zfs_vget(vfs_t * vfsp,ino_t ino,int flags,vnode_t ** vpp)1717 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
1718 {
1719 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1720 znode_t *zp;
1721 int err;
1722
1723 /*
1724 * zfs_zget() can't operate on virtual entries like .zfs/ or
1725 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
1726 * This will make NFS to switch to LOOKUP instead of using VGET.
1727 */
1728 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
1729 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
1730 return (EOPNOTSUPP);
1731
1732 ZFS_ENTER(zfsvfs);
1733 err = zfs_zget(zfsvfs, ino, &zp);
1734 if (err == 0 && zp->z_unlinked) {
1735 vrele(ZTOV(zp));
1736 err = EINVAL;
1737 }
1738 if (err == 0)
1739 *vpp = ZTOV(zp);
1740 ZFS_EXIT(zfsvfs);
1741 if (err == 0) {
1742 err = vn_lock(*vpp, flags);
1743 if (err != 0)
1744 vrele(*vpp);
1745 }
1746 if (err != 0)
1747 *vpp = NULL;
1748 return (err);
1749 }
1750
1751 static int
1752 #if __FreeBSD_version >= 1300098
zfs_checkexp(vfs_t * vfsp,struct sockaddr * nam,uint64_t * extflagsp,struct ucred ** credanonp,int * numsecflavors,int * secflavors)1753 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp,
1754 struct ucred **credanonp, int *numsecflavors, int *secflavors)
1755 #else
1756 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
1757 struct ucred **credanonp, int *numsecflavors, int **secflavors)
1758 #endif
1759 {
1760 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1761
1762 /*
1763 * If this is regular file system vfsp is the same as
1764 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
1765 * zfsvfs->z_parent->z_vfs represents parent file system
1766 * which we have to use here, because only this file system
1767 * has mnt_export configured.
1768 */
1769 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
1770 credanonp, numsecflavors, secflavors));
1771 }
1772
1773 CTASSERT(SHORT_FID_LEN <= sizeof (struct fid));
1774 CTASSERT(LONG_FID_LEN <= sizeof (struct fid));
1775
1776 static int
zfs_fhtovp(vfs_t * vfsp,fid_t * fidp,int flags,vnode_t ** vpp)1777 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
1778 {
1779 struct componentname cn;
1780 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1781 znode_t *zp;
1782 vnode_t *dvp;
1783 uint64_t object = 0;
1784 uint64_t fid_gen = 0;
1785 uint64_t gen_mask;
1786 uint64_t zp_gen;
1787 int i, err;
1788
1789 *vpp = NULL;
1790
1791 ZFS_ENTER(zfsvfs);
1792
1793 /*
1794 * On FreeBSD we can get snapshot's mount point or its parent file
1795 * system mount point depending if snapshot is already mounted or not.
1796 */
1797 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
1798 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1799 uint64_t objsetid = 0;
1800 uint64_t setgen = 0;
1801
1802 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1803 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1804
1805 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1806 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1807
1808 ZFS_EXIT(zfsvfs);
1809
1810 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1811 if (err)
1812 return (SET_ERROR(EINVAL));
1813 ZFS_ENTER(zfsvfs);
1814 }
1815
1816 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1817 zfid_short_t *zfid = (zfid_short_t *)fidp;
1818
1819 for (i = 0; i < sizeof (zfid->zf_object); i++)
1820 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1821
1822 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1823 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1824 } else {
1825 ZFS_EXIT(zfsvfs);
1826 return (SET_ERROR(EINVAL));
1827 }
1828
1829 /*
1830 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
1831 * directory tree. If the object == zfsvfs->z_shares_dir, then
1832 * we are in the .zfs/shares directory tree.
1833 */
1834 if ((fid_gen == 0 &&
1835 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
1836 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
1837 ZFS_EXIT(zfsvfs);
1838 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
1839 if (object == ZFSCTL_INO_SNAPDIR) {
1840 cn.cn_nameptr = "snapshot";
1841 cn.cn_namelen = strlen(cn.cn_nameptr);
1842 cn.cn_nameiop = LOOKUP;
1843 cn.cn_flags = ISLASTCN | LOCKLEAF;
1844 cn.cn_lkflags = flags;
1845 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
1846 vput(dvp);
1847 } else if (object == zfsvfs->z_shares_dir) {
1848 /*
1849 * XXX This branch must not be taken,
1850 * if it is, then the lookup below will
1851 * explode.
1852 */
1853 cn.cn_nameptr = "shares";
1854 cn.cn_namelen = strlen(cn.cn_nameptr);
1855 cn.cn_nameiop = LOOKUP;
1856 cn.cn_flags = ISLASTCN;
1857 cn.cn_lkflags = flags;
1858 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
1859 vput(dvp);
1860 } else {
1861 *vpp = dvp;
1862 }
1863 return (err);
1864 }
1865
1866 gen_mask = -1ULL >> (64 - 8 * i);
1867
1868 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1869 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1870 ZFS_EXIT(zfsvfs);
1871 return (err);
1872 }
1873 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1874 sizeof (uint64_t));
1875 zp_gen = zp_gen & gen_mask;
1876 if (zp_gen == 0)
1877 zp_gen = 1;
1878 if (zp->z_unlinked || zp_gen != fid_gen) {
1879 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1880 vrele(ZTOV(zp));
1881 ZFS_EXIT(zfsvfs);
1882 return (SET_ERROR(EINVAL));
1883 }
1884
1885 *vpp = ZTOV(zp);
1886 ZFS_EXIT(zfsvfs);
1887 err = vn_lock(*vpp, flags);
1888 if (err == 0)
1889 vnode_create_vobject(*vpp, zp->z_size, curthread);
1890 else
1891 *vpp = NULL;
1892 return (err);
1893 }
1894
1895 /*
1896 * Block out VOPs and close zfsvfs_t::z_os
1897 *
1898 * Note, if successful, then we return with the 'z_teardown_lock' and
1899 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
1900 * dataset and objset intact so that they can be atomically handed off during
1901 * a subsequent rollback or recv operation and the resume thereafter.
1902 */
1903 int
zfs_suspend_fs(zfsvfs_t * zfsvfs)1904 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1905 {
1906 int error;
1907
1908 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1909 return (error);
1910
1911 return (0);
1912 }
1913
1914 /*
1915 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
1916 * is an invariant across any of the operations that can be performed while the
1917 * filesystem was suspended. Whether it succeeded or failed, the preconditions
1918 * are the same: the relevant objset and associated dataset are owned by
1919 * zfsvfs, held, and long held on entry.
1920 */
1921 int
zfs_resume_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)1922 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1923 {
1924 int err;
1925 znode_t *zp;
1926
1927 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1928 ASSERT(ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zfsvfs));
1929
1930 /*
1931 * We already own this, so just update the objset_t, as the one we
1932 * had before may have been evicted.
1933 */
1934 objset_t *os;
1935 VERIFY3P(ds->ds_owner, ==, zfsvfs);
1936 VERIFY(dsl_dataset_long_held(ds));
1937 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1938 dsl_pool_config_enter(dp, FTAG);
1939 VERIFY0(dmu_objset_from_ds(ds, &os));
1940 dsl_pool_config_exit(dp, FTAG);
1941
1942 err = zfsvfs_init(zfsvfs, os);
1943 if (err != 0)
1944 goto bail;
1945
1946 ds->ds_dir->dd_activity_cancelled = B_FALSE;
1947 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1948
1949 zfs_set_fuid_feature(zfsvfs);
1950
1951 /*
1952 * Attempt to re-establish all the active znodes with
1953 * their dbufs. If a zfs_rezget() fails, then we'll let
1954 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1955 * when they try to use their znode.
1956 */
1957 mutex_enter(&zfsvfs->z_znodes_lock);
1958 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1959 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1960 (void) zfs_rezget(zp);
1961 }
1962 mutex_exit(&zfsvfs->z_znodes_lock);
1963
1964 bail:
1965 /* release the VOPs */
1966 ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs);
1967 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1968
1969 if (err) {
1970 /*
1971 * Since we couldn't setup the sa framework, try to force
1972 * unmount this file system.
1973 */
1974 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
1975 vfs_ref(zfsvfs->z_vfs);
1976 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
1977 }
1978 }
1979 return (err);
1980 }
1981
1982 static void
zfs_freevfs(vfs_t * vfsp)1983 zfs_freevfs(vfs_t *vfsp)
1984 {
1985 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1986
1987 zfsvfs_free(zfsvfs);
1988
1989 atomic_dec_32(&zfs_active_fs_count);
1990 }
1991
1992 #ifdef __i386__
1993 static int desiredvnodes_backup;
1994 #include <sys/vmmeter.h>
1995
1996
1997 #include <vm/vm_page.h>
1998 #include <vm/vm_object.h>
1999 #include <vm/vm_kern.h>
2000 #include <vm/vm_map.h>
2001 #endif
2002
2003 static void
zfs_vnodes_adjust(void)2004 zfs_vnodes_adjust(void)
2005 {
2006 #ifdef __i386__
2007 int newdesiredvnodes;
2008
2009 desiredvnodes_backup = desiredvnodes;
2010
2011 /*
2012 * We calculate newdesiredvnodes the same way it is done in
2013 * vntblinit(). If it is equal to desiredvnodes, it means that
2014 * it wasn't tuned by the administrator and we can tune it down.
2015 */
2016 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2017 vm_kmem_size / (5 * (sizeof (struct vm_object) +
2018 sizeof (struct vnode))));
2019 if (newdesiredvnodes == desiredvnodes)
2020 desiredvnodes = (3 * newdesiredvnodes) / 4;
2021 #endif
2022 }
2023
2024 static void
zfs_vnodes_adjust_back(void)2025 zfs_vnodes_adjust_back(void)
2026 {
2027
2028 #ifdef __i386__
2029 desiredvnodes = desiredvnodes_backup;
2030 #endif
2031 }
2032
2033 void
zfs_init(void)2034 zfs_init(void)
2035 {
2036
2037 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2038
2039 /*
2040 * Initialize .zfs directory structures
2041 */
2042 zfsctl_init();
2043
2044 /*
2045 * Initialize znode cache, vnode ops, etc...
2046 */
2047 zfs_znode_init();
2048
2049 /*
2050 * Reduce number of vnodes. Originally number of vnodes is calculated
2051 * with UFS inode in mind. We reduce it here, because it's too big for
2052 * ZFS/i386.
2053 */
2054 zfs_vnodes_adjust();
2055
2056 dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2057
2058 zfsvfs_taskq = taskq_create("zfsvfs", 1, minclsyspri, 0, 0, 0);
2059 }
2060
2061 void
zfs_fini(void)2062 zfs_fini(void)
2063 {
2064 taskq_destroy(zfsvfs_taskq);
2065 zfsctl_fini();
2066 zfs_znode_fini();
2067 zfs_vnodes_adjust_back();
2068 }
2069
2070 int
zfs_busy(void)2071 zfs_busy(void)
2072 {
2073 return (zfs_active_fs_count != 0);
2074 }
2075
2076 /*
2077 * Release VOPs and unmount a suspended filesystem.
2078 */
2079 int
zfs_end_fs(zfsvfs_t * zfsvfs,dsl_dataset_t * ds)2080 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2081 {
2082 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
2083 ASSERT(ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zfsvfs));
2084
2085 /*
2086 * We already own this, so just hold and rele it to update the
2087 * objset_t, as the one we had before may have been evicted.
2088 */
2089 objset_t *os;
2090 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2091 VERIFY(dsl_dataset_long_held(ds));
2092 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
2093 dsl_pool_config_enter(dp, FTAG);
2094 VERIFY0(dmu_objset_from_ds(ds, &os));
2095 dsl_pool_config_exit(dp, FTAG);
2096 zfsvfs->z_os = os;
2097
2098 /* release the VOPs */
2099 ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs);
2100 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
2101
2102 /*
2103 * Try to force unmount this file system.
2104 */
2105 (void) zfs_umount(zfsvfs->z_vfs, 0);
2106 zfsvfs->z_unmounted = B_TRUE;
2107 return (0);
2108 }
2109
2110 int
zfs_set_version(zfsvfs_t * zfsvfs,uint64_t newvers)2111 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2112 {
2113 int error;
2114 objset_t *os = zfsvfs->z_os;
2115 dmu_tx_t *tx;
2116
2117 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2118 return (SET_ERROR(EINVAL));
2119
2120 if (newvers < zfsvfs->z_version)
2121 return (SET_ERROR(EINVAL));
2122
2123 if (zfs_spa_version_map(newvers) >
2124 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2125 return (SET_ERROR(ENOTSUP));
2126
2127 tx = dmu_tx_create(os);
2128 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2129 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2130 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2131 ZFS_SA_ATTRS);
2132 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2133 }
2134 error = dmu_tx_assign(tx, TXG_WAIT);
2135 if (error) {
2136 dmu_tx_abort(tx);
2137 return (error);
2138 }
2139
2140 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2141 8, 1, &newvers, tx);
2142
2143 if (error) {
2144 dmu_tx_commit(tx);
2145 return (error);
2146 }
2147
2148 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2149 uint64_t sa_obj;
2150
2151 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2152 SPA_VERSION_SA);
2153 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2154 DMU_OT_NONE, 0, tx);
2155
2156 error = zap_add(os, MASTER_NODE_OBJ,
2157 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2158 ASSERT0(error);
2159
2160 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2161 sa_register_update_callback(os, zfs_sa_upgrade);
2162 }
2163
2164 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2165 "from %ju to %ju", (uintmax_t)zfsvfs->z_version,
2166 (uintmax_t)newvers);
2167 dmu_tx_commit(tx);
2168
2169 zfsvfs->z_version = newvers;
2170 os->os_version = newvers;
2171
2172 zfs_set_fuid_feature(zfsvfs);
2173
2174 return (0);
2175 }
2176
2177 /*
2178 * Read a property stored within the master node.
2179 */
2180 int
zfs_get_zplprop(objset_t * os,zfs_prop_t prop,uint64_t * value)2181 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2182 {
2183 uint64_t *cached_copy = NULL;
2184
2185 /*
2186 * Figure out where in the objset_t the cached copy would live, if it
2187 * is available for the requested property.
2188 */
2189 if (os != NULL) {
2190 switch (prop) {
2191 case ZFS_PROP_VERSION:
2192 cached_copy = &os->os_version;
2193 break;
2194 case ZFS_PROP_NORMALIZE:
2195 cached_copy = &os->os_normalization;
2196 break;
2197 case ZFS_PROP_UTF8ONLY:
2198 cached_copy = &os->os_utf8only;
2199 break;
2200 case ZFS_PROP_CASE:
2201 cached_copy = &os->os_casesensitivity;
2202 break;
2203 default:
2204 break;
2205 }
2206 }
2207 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2208 *value = *cached_copy;
2209 return (0);
2210 }
2211
2212 /*
2213 * If the property wasn't cached, look up the file system's value for
2214 * the property. For the version property, we look up a slightly
2215 * different string.
2216 */
2217 const char *pname;
2218 int error = ENOENT;
2219 if (prop == ZFS_PROP_VERSION) {
2220 pname = ZPL_VERSION_STR;
2221 } else {
2222 pname = zfs_prop_to_name(prop);
2223 }
2224
2225 if (os != NULL) {
2226 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2227 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2228 }
2229
2230 if (error == ENOENT) {
2231 /* No value set, use the default value */
2232 switch (prop) {
2233 case ZFS_PROP_VERSION:
2234 *value = ZPL_VERSION;
2235 break;
2236 case ZFS_PROP_NORMALIZE:
2237 case ZFS_PROP_UTF8ONLY:
2238 *value = 0;
2239 break;
2240 case ZFS_PROP_CASE:
2241 *value = ZFS_CASE_SENSITIVE;
2242 break;
2243 case ZFS_PROP_ACLTYPE:
2244 *value = ZFS_ACLTYPE_NFSV4;
2245 break;
2246 default:
2247 return (error);
2248 }
2249 error = 0;
2250 }
2251
2252 /*
2253 * If one of the methods for getting the property value above worked,
2254 * copy it into the objset_t's cache.
2255 */
2256 if (error == 0 && cached_copy != NULL) {
2257 *cached_copy = *value;
2258 }
2259
2260 return (error);
2261 }
2262
2263 /*
2264 * Return true if the corresponding vfs's unmounted flag is set.
2265 * Otherwise return false.
2266 * If this function returns true we know VFS unmount has been initiated.
2267 */
2268 boolean_t
zfs_get_vfs_flag_unmounted(objset_t * os)2269 zfs_get_vfs_flag_unmounted(objset_t *os)
2270 {
2271 zfsvfs_t *zfvp;
2272 boolean_t unmounted = B_FALSE;
2273
2274 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2275
2276 mutex_enter(&os->os_user_ptr_lock);
2277 zfvp = dmu_objset_get_user(os);
2278 if (zfvp != NULL && zfvp->z_vfs != NULL &&
2279 (zfvp->z_vfs->mnt_kern_flag & MNTK_UNMOUNT))
2280 unmounted = B_TRUE;
2281 mutex_exit(&os->os_user_ptr_lock);
2282
2283 return (unmounted);
2284 }
2285
2286 #ifdef _KERNEL
2287 void
zfsvfs_update_fromname(const char * oldname,const char * newname)2288 zfsvfs_update_fromname(const char *oldname, const char *newname)
2289 {
2290 char tmpbuf[MAXPATHLEN];
2291 struct mount *mp;
2292 char *fromname;
2293 size_t oldlen;
2294
2295 oldlen = strlen(oldname);
2296
2297 mtx_lock(&mountlist_mtx);
2298 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2299 fromname = mp->mnt_stat.f_mntfromname;
2300 if (strcmp(fromname, oldname) == 0) {
2301 (void) strlcpy(fromname, newname,
2302 sizeof (mp->mnt_stat.f_mntfromname));
2303 continue;
2304 }
2305 if (strncmp(fromname, oldname, oldlen) == 0 &&
2306 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2307 (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s%s",
2308 newname, fromname + oldlen);
2309 (void) strlcpy(fromname, tmpbuf,
2310 sizeof (mp->mnt_stat.f_mntfromname));
2311 continue;
2312 }
2313 }
2314 mtx_unlock(&mountlist_mtx);
2315 }
2316 #endif
2317