]>
Commit | Line | Data |
---|---|---|
9f0a21e6 MM |
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 | |
1d3ba0bf | 9 | * or https://opensource.org/licenses/CDDL-1.0. |
9f0a21e6 MM |
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 <pawel@dawidek.net>. | |
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> | |
ab8c935e | 45 | #include <sys/zfs_vnops.h> |
9f0a21e6 MM |
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> | |
9f0a21e6 | 65 | #include <sys/jail.h> |
595d3ac2 | 66 | #include <sys/osd.h> |
9f0a21e6 MM |
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 | ||
9f0a21e6 MM |
79 | struct mtx zfs_debug_mtx; |
80 | MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF); | |
81 | ||
82 | SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system"); | |
83 | ||
84 | int zfs_super_owner; | |
85 | SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0, | |
7ada752a | 86 | "File system owners can perform privileged operation on file systems"); |
9f0a21e6 MM |
87 | |
88 | int zfs_debug_level; | |
89 | SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0, | |
90 | "Debug level"); | |
91 | ||
595d3ac2 AJ |
92 | struct zfs_jailparam { |
93 | int mount_snapshot; | |
94 | }; | |
95 | ||
96 | static struct zfs_jailparam zfs_jailparam0 = { | |
97 | .mount_snapshot = 0, | |
98 | }; | |
99 | ||
100 | static int zfs_jailparam_slot; | |
101 | ||
102 | SYSCTL_JAIL_PARAM_SYS_NODE(zfs, CTLFLAG_RW, "Jail ZFS parameters"); | |
103 | SYSCTL_JAIL_PARAM(_zfs, mount_snapshot, CTLTYPE_INT | CTLFLAG_RW, "I", | |
104 | "Allow mounting snapshots in the .zfs directory for unjailed datasets"); | |
105 | ||
9f0a21e6 MM |
106 | SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions"); |
107 | static int zfs_version_acl = ZFS_ACL_VERSION; | |
108 | SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0, | |
7ada752a | 109 | "ZFS_ACL_VERSION"); |
9f0a21e6 MM |
110 | static int zfs_version_spa = SPA_VERSION; |
111 | SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0, | |
7ada752a | 112 | "SPA_VERSION"); |
9f0a21e6 MM |
113 | static int zfs_version_zpl = ZPL_VERSION; |
114 | SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0, | |
7ada752a | 115 | "ZPL_VERSION"); |
9f0a21e6 | 116 | |
8dddb25d JH |
117 | #if __FreeBSD_version >= 1400018 |
118 | static int zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg, | |
119 | bool *mp_busy); | |
120 | #else | |
9f0a21e6 | 121 | static int zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg); |
8dddb25d | 122 | #endif |
9f0a21e6 MM |
123 | static int zfs_mount(vfs_t *vfsp); |
124 | static int zfs_umount(vfs_t *vfsp, int fflag); | |
125 | static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp); | |
126 | static int zfs_statfs(vfs_t *vfsp, struct statfs *statp); | |
127 | static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp); | |
128 | static int zfs_sync(vfs_t *vfsp, int waitfor); | |
2e6af52b RM |
129 | #if __FreeBSD_version >= 1300098 |
130 | static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp, | |
131 | struct ucred **credanonp, int *numsecflavors, int *secflavors); | |
132 | #else | |
9f0a21e6 MM |
133 | static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, |
134 | struct ucred **credanonp, int *numsecflavors, int **secflavors); | |
2e6af52b | 135 | #endif |
9f0a21e6 MM |
136 | static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp); |
137 | static void zfs_freevfs(vfs_t *vfsp); | |
138 | ||
139 | struct vfsops zfs_vfsops = { | |
140 | .vfs_mount = zfs_mount, | |
141 | .vfs_unmount = zfs_umount, | |
142 | #if __FreeBSD_version >= 1300049 | |
143 | .vfs_root = vfs_cache_root, | |
144 | .vfs_cachedroot = zfs_root, | |
145 | #else | |
146 | .vfs_root = zfs_root, | |
147 | #endif | |
148 | .vfs_statfs = zfs_statfs, | |
149 | .vfs_vget = zfs_vget, | |
150 | .vfs_sync = zfs_sync, | |
151 | .vfs_checkexp = zfs_checkexp, | |
152 | .vfs_fhtovp = zfs_fhtovp, | |
153 | .vfs_quotactl = zfs_quotactl, | |
154 | }; | |
155 | ||
67a1b037 PJD |
156 | #ifdef VFCF_CROSS_COPY_FILE_RANGE |
157 | VFS_SET(zfs_vfsops, zfs, | |
158 | VFCF_DELEGADMIN | VFCF_JAIL | VFCF_CROSS_COPY_FILE_RANGE); | |
159 | #else | |
160 | VFS_SET(zfs_vfsops, zfs, VFCF_DELEGADMIN | VFCF_JAIL); | |
161 | #endif | |
9f0a21e6 MM |
162 | |
163 | /* | |
164 | * We need to keep a count of active fs's. | |
165 | * This is necessary to prevent our module | |
166 | * from being unloaded after a umount -f | |
167 | */ | |
168 | static uint32_t zfs_active_fs_count = 0; | |
169 | ||
170 | int | |
171 | zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val, | |
172 | char *setpoint) | |
173 | { | |
174 | int error; | |
175 | zfsvfs_t *zfvp; | |
176 | vfs_t *vfsp; | |
177 | objset_t *os; | |
178 | uint64_t tmp = *val; | |
179 | ||
180 | error = dmu_objset_from_ds(ds, &os); | |
181 | if (error != 0) | |
182 | return (error); | |
183 | ||
184 | error = getzfsvfs_impl(os, &zfvp); | |
185 | if (error != 0) | |
186 | return (error); | |
187 | if (zfvp == NULL) | |
188 | return (ENOENT); | |
189 | vfsp = zfvp->z_vfs; | |
190 | switch (zfs_prop) { | |
191 | case ZFS_PROP_ATIME: | |
192 | if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) | |
193 | tmp = 0; | |
194 | if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) | |
195 | tmp = 1; | |
196 | break; | |
197 | case ZFS_PROP_DEVICES: | |
198 | if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) | |
199 | tmp = 0; | |
200 | if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) | |
201 | tmp = 1; | |
202 | break; | |
203 | case ZFS_PROP_EXEC: | |
204 | if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) | |
205 | tmp = 0; | |
206 | if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) | |
207 | tmp = 1; | |
208 | break; | |
209 | case ZFS_PROP_SETUID: | |
210 | if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) | |
211 | tmp = 0; | |
212 | if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) | |
213 | tmp = 1; | |
214 | break; | |
215 | case ZFS_PROP_READONLY: | |
216 | if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) | |
217 | tmp = 0; | |
218 | if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) | |
219 | tmp = 1; | |
220 | break; | |
221 | case ZFS_PROP_XATTR: | |
222 | if (zfvp->z_flags & ZSB_XATTR) | |
223 | tmp = zfvp->z_xattr; | |
224 | break; | |
225 | case ZFS_PROP_NBMAND: | |
226 | if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) | |
227 | tmp = 0; | |
228 | if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) | |
229 | tmp = 1; | |
230 | break; | |
231 | default: | |
232 | vfs_unbusy(vfsp); | |
233 | return (ENOENT); | |
234 | } | |
235 | ||
236 | vfs_unbusy(vfsp); | |
237 | if (tmp != *val) { | |
3a7d2a0c RY |
238 | if (setpoint) |
239 | (void) strcpy(setpoint, "temporary"); | |
9f0a21e6 MM |
240 | *val = tmp; |
241 | } | |
242 | return (0); | |
243 | } | |
244 | ||
245 | static int | |
246 | zfs_getquota(zfsvfs_t *zfsvfs, uid_t id, int isgroup, struct dqblk64 *dqp) | |
247 | { | |
248 | int error = 0; | |
249 | char buf[32]; | |
250 | uint64_t usedobj, quotaobj; | |
251 | uint64_t quota, used = 0; | |
252 | timespec_t now; | |
253 | ||
254 | usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; | |
255 | quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj; | |
256 | ||
257 | if (quotaobj == 0 || zfsvfs->z_replay) { | |
258 | error = ENOENT; | |
259 | goto done; | |
260 | } | |
261 | (void) sprintf(buf, "%llx", (longlong_t)id); | |
262 | if ((error = zap_lookup(zfsvfs->z_os, quotaobj, | |
263 | buf, sizeof (quota), 1, "a)) != 0) { | |
264 | dprintf("%s(%d): quotaobj lookup failed\n", | |
265 | __FUNCTION__, __LINE__); | |
266 | goto done; | |
267 | } | |
268 | /* | |
269 | * quota(8) uses bsoftlimit as "quoota", and hardlimit as "limit". | |
270 | * So we set them to be the same. | |
271 | */ | |
272 | dqp->dqb_bsoftlimit = dqp->dqb_bhardlimit = btodb(quota); | |
273 | error = zap_lookup(zfsvfs->z_os, usedobj, buf, sizeof (used), 1, &used); | |
274 | if (error && error != ENOENT) { | |
275 | dprintf("%s(%d): usedobj failed; %d\n", | |
276 | __FUNCTION__, __LINE__, error); | |
277 | goto done; | |
278 | } | |
279 | dqp->dqb_curblocks = btodb(used); | |
280 | dqp->dqb_ihardlimit = dqp->dqb_isoftlimit = 0; | |
281 | vfs_timestamp(&now); | |
282 | /* | |
283 | * Setting this to 0 causes FreeBSD quota(8) to print | |
284 | * the number of days since the epoch, which isn't | |
285 | * particularly useful. | |
286 | */ | |
287 | dqp->dqb_btime = dqp->dqb_itime = now.tv_sec; | |
288 | done: | |
289 | return (error); | |
290 | } | |
291 | ||
292 | static int | |
8dddb25d JH |
293 | #if __FreeBSD_version >= 1400018 |
294 | zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg, bool *mp_busy) | |
295 | #else | |
9f0a21e6 | 296 | zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg) |
8dddb25d | 297 | #endif |
9f0a21e6 MM |
298 | { |
299 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
300 | struct thread *td; | |
301 | int cmd, type, error = 0; | |
302 | int bitsize; | |
303 | zfs_userquota_prop_t quota_type; | |
304 | struct dqblk64 dqblk = { 0 }; | |
305 | ||
306 | td = curthread; | |
307 | cmd = cmds >> SUBCMDSHIFT; | |
308 | type = cmds & SUBCMDMASK; | |
309 | ||
768eaced CC |
310 | if ((error = zfs_enter(zfsvfs, FTAG)) != 0) |
311 | return (error); | |
9f0a21e6 MM |
312 | if (id == -1) { |
313 | switch (type) { | |
314 | case USRQUOTA: | |
315 | id = td->td_ucred->cr_ruid; | |
316 | break; | |
317 | case GRPQUOTA: | |
318 | id = td->td_ucred->cr_rgid; | |
319 | break; | |
320 | default: | |
321 | error = EINVAL; | |
8dddb25d | 322 | #if __FreeBSD_version < 1400018 |
9f0a21e6 MM |
323 | if (cmd == Q_QUOTAON || cmd == Q_QUOTAOFF) |
324 | vfs_unbusy(vfsp); | |
8dddb25d | 325 | #endif |
9f0a21e6 MM |
326 | goto done; |
327 | } | |
328 | } | |
329 | /* | |
330 | * Map BSD type to: | |
331 | * ZFS_PROP_USERUSED, | |
332 | * ZFS_PROP_USERQUOTA, | |
333 | * ZFS_PROP_GROUPUSED, | |
334 | * ZFS_PROP_GROUPQUOTA | |
335 | */ | |
336 | switch (cmd) { | |
337 | case Q_SETQUOTA: | |
338 | case Q_SETQUOTA32: | |
339 | if (type == USRQUOTA) | |
340 | quota_type = ZFS_PROP_USERQUOTA; | |
341 | else if (type == GRPQUOTA) | |
342 | quota_type = ZFS_PROP_GROUPQUOTA; | |
343 | else | |
344 | error = EINVAL; | |
345 | break; | |
346 | case Q_GETQUOTA: | |
347 | case Q_GETQUOTA32: | |
348 | if (type == USRQUOTA) | |
349 | quota_type = ZFS_PROP_USERUSED; | |
350 | else if (type == GRPQUOTA) | |
351 | quota_type = ZFS_PROP_GROUPUSED; | |
352 | else | |
353 | error = EINVAL; | |
354 | break; | |
355 | } | |
356 | ||
357 | /* | |
358 | * Depending on the cmd, we may need to get | |
359 | * the ruid and domain (see fuidstr_to_sid?), | |
360 | * the fuid (how?), or other information. | |
361 | * Create fuid using zfs_fuid_create(zfsvfs, id, | |
362 | * ZFS_OWNER or ZFS_GROUP, cr, &fuidp)? | |
363 | * I think I can use just the id? | |
364 | * | |
365 | * Look at zfs_id_overquota() to look up a quota. | |
366 | * zap_lookup(something, quotaobj, fuidstring, | |
367 | * sizeof (long long), 1, "a) | |
368 | * | |
369 | * See zfs_set_userquota() to set a quota. | |
370 | */ | |
371 | if ((uint32_t)type >= MAXQUOTAS) { | |
372 | error = EINVAL; | |
373 | goto done; | |
374 | } | |
375 | ||
376 | switch (cmd) { | |
377 | case Q_GETQUOTASIZE: | |
378 | bitsize = 64; | |
379 | error = copyout(&bitsize, arg, sizeof (int)); | |
380 | break; | |
381 | case Q_QUOTAON: | |
382 | // As far as I can tell, you can't turn quotas on or off on zfs | |
383 | error = 0; | |
8dddb25d | 384 | #if __FreeBSD_version < 1400018 |
9f0a21e6 | 385 | vfs_unbusy(vfsp); |
8dddb25d | 386 | #endif |
9f0a21e6 MM |
387 | break; |
388 | case Q_QUOTAOFF: | |
389 | error = ENOTSUP; | |
8dddb25d | 390 | #if __FreeBSD_version < 1400018 |
9f0a21e6 | 391 | vfs_unbusy(vfsp); |
8dddb25d | 392 | #endif |
9f0a21e6 MM |
393 | break; |
394 | case Q_SETQUOTA: | |
7b0e3903 | 395 | error = copyin(arg, &dqblk, sizeof (dqblk)); |
9f0a21e6 MM |
396 | if (error == 0) |
397 | error = zfs_set_userquota(zfsvfs, quota_type, | |
398 | "", id, dbtob(dqblk.dqb_bhardlimit)); | |
399 | break; | |
400 | case Q_GETQUOTA: | |
401 | error = zfs_getquota(zfsvfs, id, type == GRPQUOTA, &dqblk); | |
402 | if (error == 0) | |
403 | error = copyout(&dqblk, arg, sizeof (dqblk)); | |
404 | break; | |
405 | default: | |
406 | error = EINVAL; | |
407 | break; | |
408 | } | |
409 | done: | |
768eaced | 410 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
411 | return (error); |
412 | } | |
413 | ||
414 | ||
415 | boolean_t | |
416 | zfs_is_readonly(zfsvfs_t *zfsvfs) | |
417 | { | |
418 | return (!!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY)); | |
419 | } | |
420 | ||
9f0a21e6 MM |
421 | static int |
422 | zfs_sync(vfs_t *vfsp, int waitfor) | |
423 | { | |
424 | ||
425 | /* | |
426 | * Data integrity is job one. We don't want a compromised kernel | |
427 | * writing to the storage pool, so we never sync during panic. | |
428 | */ | |
429 | if (panicstr) | |
430 | return (0); | |
431 | ||
432 | /* | |
433 | * Ignore the system syncher. ZFS already commits async data | |
434 | * at zfs_txg_timeout intervals. | |
435 | */ | |
436 | if (waitfor == MNT_LAZY) | |
437 | return (0); | |
438 | ||
439 | if (vfsp != NULL) { | |
440 | /* | |
441 | * Sync a specific filesystem. | |
442 | */ | |
443 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
444 | dsl_pool_t *dp; | |
445 | int error; | |
446 | ||
768eaced CC |
447 | if ((error = zfs_enter(zfsvfs, FTAG)) != 0) |
448 | return (error); | |
9f0a21e6 MM |
449 | dp = dmu_objset_pool(zfsvfs->z_os); |
450 | ||
451 | /* | |
452 | * If the system is shutting down, then skip any | |
453 | * filesystems which may exist on a suspended pool. | |
454 | */ | |
455 | if (rebooting && spa_suspended(dp->dp_spa)) { | |
768eaced | 456 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
457 | return (0); |
458 | } | |
459 | ||
460 | if (zfsvfs->z_log != NULL) | |
461 | zil_commit(zfsvfs->z_log, 0); | |
462 | ||
768eaced | 463 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
464 | } else { |
465 | /* | |
466 | * Sync all ZFS filesystems. This is what happens when you | |
76d04993 | 467 | * run sync(8). Unlike other filesystems, ZFS honors the |
9f0a21e6 MM |
468 | * request by waiting for all pools to commit all dirty data. |
469 | */ | |
470 | spa_sync_allpools(); | |
471 | } | |
472 | ||
473 | return (0); | |
474 | } | |
475 | ||
476 | static void | |
477 | atime_changed_cb(void *arg, uint64_t newval) | |
478 | { | |
479 | zfsvfs_t *zfsvfs = arg; | |
480 | ||
481 | if (newval == TRUE) { | |
482 | zfsvfs->z_atime = TRUE; | |
483 | zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME; | |
484 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME); | |
485 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0); | |
486 | } else { | |
487 | zfsvfs->z_atime = FALSE; | |
488 | zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME; | |
489 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME); | |
490 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0); | |
491 | } | |
492 | } | |
493 | ||
494 | static void | |
495 | xattr_changed_cb(void *arg, uint64_t newval) | |
496 | { | |
497 | zfsvfs_t *zfsvfs = arg; | |
498 | ||
499 | if (newval == ZFS_XATTR_OFF) { | |
500 | zfsvfs->z_flags &= ~ZSB_XATTR; | |
501 | } else { | |
502 | zfsvfs->z_flags |= ZSB_XATTR; | |
503 | ||
504 | if (newval == ZFS_XATTR_SA) | |
505 | zfsvfs->z_xattr_sa = B_TRUE; | |
506 | else | |
507 | zfsvfs->z_xattr_sa = B_FALSE; | |
508 | } | |
509 | } | |
510 | ||
511 | static void | |
512 | blksz_changed_cb(void *arg, uint64_t newval) | |
513 | { | |
514 | zfsvfs_t *zfsvfs = arg; | |
515 | ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os))); | |
516 | ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); | |
517 | ASSERT(ISP2(newval)); | |
518 | ||
519 | zfsvfs->z_max_blksz = newval; | |
520 | zfsvfs->z_vfs->mnt_stat.f_iosize = newval; | |
521 | } | |
522 | ||
523 | static void | |
524 | readonly_changed_cb(void *arg, uint64_t newval) | |
525 | { | |
526 | zfsvfs_t *zfsvfs = arg; | |
527 | ||
528 | if (newval) { | |
529 | /* XXX locking on vfs_flag? */ | |
530 | zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; | |
531 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW); | |
532 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0); | |
533 | } else { | |
534 | /* XXX locking on vfs_flag? */ | |
535 | zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; | |
536 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO); | |
537 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0); | |
538 | } | |
539 | } | |
540 | ||
541 | static void | |
542 | setuid_changed_cb(void *arg, uint64_t newval) | |
543 | { | |
544 | zfsvfs_t *zfsvfs = arg; | |
545 | ||
546 | if (newval == FALSE) { | |
547 | zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID; | |
548 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID); | |
549 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0); | |
550 | } else { | |
551 | zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID; | |
552 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID); | |
553 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0); | |
554 | } | |
555 | } | |
556 | ||
557 | static void | |
558 | exec_changed_cb(void *arg, uint64_t newval) | |
559 | { | |
560 | zfsvfs_t *zfsvfs = arg; | |
561 | ||
562 | if (newval == FALSE) { | |
563 | zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC; | |
564 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC); | |
565 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0); | |
566 | } else { | |
567 | zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC; | |
568 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC); | |
569 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0); | |
570 | } | |
571 | } | |
572 | ||
573 | /* | |
574 | * The nbmand mount option can be changed at mount time. | |
575 | * We can't allow it to be toggled on live file systems or incorrect | |
576 | * behavior may be seen from cifs clients | |
577 | * | |
578 | * This property isn't registered via dsl_prop_register(), but this callback | |
579 | * will be called when a file system is first mounted | |
580 | */ | |
581 | static void | |
582 | nbmand_changed_cb(void *arg, uint64_t newval) | |
583 | { | |
584 | zfsvfs_t *zfsvfs = arg; | |
585 | if (newval == FALSE) { | |
586 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND); | |
587 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0); | |
588 | } else { | |
589 | vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND); | |
590 | vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0); | |
591 | } | |
592 | } | |
593 | ||
594 | static void | |
595 | snapdir_changed_cb(void *arg, uint64_t newval) | |
596 | { | |
597 | zfsvfs_t *zfsvfs = arg; | |
598 | ||
599 | zfsvfs->z_show_ctldir = newval; | |
600 | } | |
601 | ||
9f0a21e6 MM |
602 | static void |
603 | acl_mode_changed_cb(void *arg, uint64_t newval) | |
604 | { | |
605 | zfsvfs_t *zfsvfs = arg; | |
606 | ||
607 | zfsvfs->z_acl_mode = newval; | |
608 | } | |
609 | ||
610 | static void | |
611 | acl_inherit_changed_cb(void *arg, uint64_t newval) | |
612 | { | |
613 | zfsvfs_t *zfsvfs = arg; | |
614 | ||
615 | zfsvfs->z_acl_inherit = newval; | |
616 | } | |
617 | ||
485b50bb RM |
618 | static void |
619 | acl_type_changed_cb(void *arg, uint64_t newval) | |
620 | { | |
621 | zfsvfs_t *zfsvfs = arg; | |
622 | ||
623 | zfsvfs->z_acl_type = newval; | |
624 | } | |
625 | ||
9f0a21e6 MM |
626 | static int |
627 | zfs_register_callbacks(vfs_t *vfsp) | |
628 | { | |
629 | struct dsl_dataset *ds = NULL; | |
630 | objset_t *os = NULL; | |
631 | zfsvfs_t *zfsvfs = NULL; | |
632 | uint64_t nbmand; | |
633 | boolean_t readonly = B_FALSE; | |
634 | boolean_t do_readonly = B_FALSE; | |
635 | boolean_t setuid = B_FALSE; | |
636 | boolean_t do_setuid = B_FALSE; | |
637 | boolean_t exec = B_FALSE; | |
638 | boolean_t do_exec = B_FALSE; | |
639 | boolean_t xattr = B_FALSE; | |
640 | boolean_t atime = B_FALSE; | |
641 | boolean_t do_atime = B_FALSE; | |
642 | boolean_t do_xattr = B_FALSE; | |
643 | int error = 0; | |
644 | ||
e4efb709 | 645 | ASSERT3P(vfsp, !=, NULL); |
9f0a21e6 | 646 | zfsvfs = vfsp->vfs_data; |
e4efb709 | 647 | ASSERT3P(zfsvfs, !=, NULL); |
9f0a21e6 MM |
648 | os = zfsvfs->z_os; |
649 | ||
650 | /* | |
651 | * This function can be called for a snapshot when we update snapshot's | |
652 | * mount point, which isn't really supported. | |
653 | */ | |
654 | if (dmu_objset_is_snapshot(os)) | |
655 | return (EOPNOTSUPP); | |
656 | ||
657 | /* | |
658 | * The act of registering our callbacks will destroy any mount | |
659 | * options we may have. In order to enable temporary overrides | |
660 | * of mount options, we stash away the current values and | |
661 | * restore them after we register the callbacks. | |
662 | */ | |
663 | if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) || | |
664 | !spa_writeable(dmu_objset_spa(os))) { | |
665 | readonly = B_TRUE; | |
666 | do_readonly = B_TRUE; | |
667 | } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) { | |
668 | readonly = B_FALSE; | |
669 | do_readonly = B_TRUE; | |
670 | } | |
671 | if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) { | |
672 | setuid = B_FALSE; | |
673 | do_setuid = B_TRUE; | |
674 | } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) { | |
675 | setuid = B_TRUE; | |
676 | do_setuid = B_TRUE; | |
677 | } | |
678 | if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) { | |
679 | exec = B_FALSE; | |
680 | do_exec = B_TRUE; | |
681 | } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) { | |
682 | exec = B_TRUE; | |
683 | do_exec = B_TRUE; | |
684 | } | |
685 | if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) { | |
686 | zfsvfs->z_xattr = xattr = ZFS_XATTR_OFF; | |
687 | do_xattr = B_TRUE; | |
688 | } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) { | |
689 | zfsvfs->z_xattr = xattr = ZFS_XATTR_DIR; | |
690 | do_xattr = B_TRUE; | |
691 | } else if (vfs_optionisset(vfsp, MNTOPT_DIRXATTR, NULL)) { | |
692 | zfsvfs->z_xattr = xattr = ZFS_XATTR_DIR; | |
693 | do_xattr = B_TRUE; | |
694 | } else if (vfs_optionisset(vfsp, MNTOPT_SAXATTR, NULL)) { | |
695 | zfsvfs->z_xattr = xattr = ZFS_XATTR_SA; | |
696 | do_xattr = B_TRUE; | |
697 | } | |
698 | if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) { | |
699 | atime = B_FALSE; | |
700 | do_atime = B_TRUE; | |
701 | } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) { | |
702 | atime = B_TRUE; | |
703 | do_atime = B_TRUE; | |
704 | } | |
705 | ||
706 | /* | |
707 | * We need to enter pool configuration here, so that we can use | |
708 | * dsl_prop_get_int_ds() to handle the special nbmand property below. | |
709 | * dsl_prop_get_integer() can not be used, because it has to acquire | |
710 | * spa_namespace_lock and we can not do that because we already hold | |
711 | * z_teardown_lock. The problem is that spa_write_cachefile() is called | |
712 | * with spa_namespace_lock held and the function calls ZFS vnode | |
713 | * operations to write the cache file and thus z_teardown_lock is | |
714 | * acquired after spa_namespace_lock. | |
715 | */ | |
716 | ds = dmu_objset_ds(os); | |
717 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
718 | ||
719 | /* | |
720 | * nbmand is a special property. It can only be changed at | |
721 | * mount time. | |
722 | * | |
723 | * This is weird, but it is documented to only be changeable | |
724 | * at mount time. | |
725 | */ | |
726 | if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) { | |
727 | nbmand = B_FALSE; | |
728 | } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) { | |
729 | nbmand = B_TRUE; | |
7b9a4230 | 730 | } else if ((error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand)) != 0) { |
9f0a21e6 MM |
731 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); |
732 | return (error); | |
733 | } | |
734 | ||
735 | /* | |
736 | * Register property callbacks. | |
737 | * | |
738 | * It would probably be fine to just check for i/o error from | |
739 | * the first prop_register(), but I guess I like to go | |
740 | * overboard... | |
741 | */ | |
742 | error = dsl_prop_register(ds, | |
743 | zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs); | |
744 | error = error ? error : dsl_prop_register(ds, | |
745 | zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs); | |
746 | error = error ? error : dsl_prop_register(ds, | |
747 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs); | |
748 | error = error ? error : dsl_prop_register(ds, | |
749 | zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs); | |
750 | error = error ? error : dsl_prop_register(ds, | |
751 | zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs); | |
752 | error = error ? error : dsl_prop_register(ds, | |
753 | zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs); | |
754 | error = error ? error : dsl_prop_register(ds, | |
755 | zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs); | |
485b50bb RM |
756 | error = error ? error : dsl_prop_register(ds, |
757 | zfs_prop_to_name(ZFS_PROP_ACLTYPE), acl_type_changed_cb, zfsvfs); | |
9f0a21e6 MM |
758 | error = error ? error : dsl_prop_register(ds, |
759 | zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs); | |
760 | error = error ? error : dsl_prop_register(ds, | |
761 | zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, | |
762 | zfsvfs); | |
9f0a21e6 MM |
763 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); |
764 | if (error) | |
765 | goto unregister; | |
766 | ||
767 | /* | |
768 | * Invoke our callbacks to restore temporary mount options. | |
769 | */ | |
770 | if (do_readonly) | |
771 | readonly_changed_cb(zfsvfs, readonly); | |
772 | if (do_setuid) | |
773 | setuid_changed_cb(zfsvfs, setuid); | |
774 | if (do_exec) | |
775 | exec_changed_cb(zfsvfs, exec); | |
776 | if (do_xattr) | |
777 | xattr_changed_cb(zfsvfs, xattr); | |
778 | if (do_atime) | |
779 | atime_changed_cb(zfsvfs, atime); | |
780 | ||
781 | nbmand_changed_cb(zfsvfs, nbmand); | |
782 | ||
783 | return (0); | |
784 | ||
785 | unregister: | |
786 | dsl_prop_unregister_all(ds, zfsvfs); | |
787 | return (error); | |
788 | } | |
789 | ||
790 | /* | |
791 | * Associate this zfsvfs with the given objset, which must be owned. | |
792 | * This will cache a bunch of on-disk state from the objset in the | |
793 | * zfsvfs. | |
794 | */ | |
795 | static int | |
796 | zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os) | |
797 | { | |
798 | int error; | |
799 | uint64_t val; | |
800 | ||
801 | zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; | |
802 | zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; | |
803 | zfsvfs->z_os = os; | |
804 | ||
805 | error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version); | |
806 | if (error != 0) | |
807 | return (error); | |
808 | if (zfsvfs->z_version > | |
809 | zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) { | |
810 | (void) printf("Can't mount a version %lld file system " | |
811 | "on a version %lld pool\n. Pool must be upgraded to mount " | |
812 | "this file system.", (u_longlong_t)zfsvfs->z_version, | |
813 | (u_longlong_t)spa_version(dmu_objset_spa(os))); | |
814 | return (SET_ERROR(ENOTSUP)); | |
815 | } | |
816 | error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val); | |
817 | if (error != 0) | |
818 | return (error); | |
819 | zfsvfs->z_norm = (int)val; | |
820 | ||
821 | error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val); | |
822 | if (error != 0) | |
823 | return (error); | |
824 | zfsvfs->z_utf8 = (val != 0); | |
825 | ||
826 | error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val); | |
827 | if (error != 0) | |
828 | return (error); | |
829 | zfsvfs->z_case = (uint_t)val; | |
830 | ||
485b50bb RM |
831 | error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val); |
832 | if (error != 0) | |
833 | return (error); | |
834 | zfsvfs->z_acl_type = (uint_t)val; | |
835 | ||
9f0a21e6 MM |
836 | /* |
837 | * Fold case on file systems that are always or sometimes case | |
838 | * insensitive. | |
839 | */ | |
840 | if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE || | |
841 | zfsvfs->z_case == ZFS_CASE_MIXED) | |
842 | zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER; | |
843 | ||
844 | zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); | |
845 | zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); | |
846 | ||
847 | uint64_t sa_obj = 0; | |
848 | if (zfsvfs->z_use_sa) { | |
849 | /* should either have both of these objects or none */ | |
850 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, | |
851 | &sa_obj); | |
852 | if (error != 0) | |
853 | return (error); | |
210231ed RM |
854 | |
855 | error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val); | |
856 | if (error == 0 && val == ZFS_XATTR_SA) | |
857 | zfsvfs->z_xattr_sa = B_TRUE; | |
9f0a21e6 MM |
858 | } |
859 | ||
860 | error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, | |
861 | &zfsvfs->z_attr_table); | |
862 | if (error != 0) | |
863 | return (error); | |
864 | ||
865 | if (zfsvfs->z_version >= ZPL_VERSION_SA) | |
866 | sa_register_update_callback(os, zfs_sa_upgrade); | |
867 | ||
868 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, | |
869 | &zfsvfs->z_root); | |
870 | if (error != 0) | |
871 | return (error); | |
e4efb709 | 872 | ASSERT3U(zfsvfs->z_root, !=, 0); |
9f0a21e6 MM |
873 | |
874 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, | |
875 | &zfsvfs->z_unlinkedobj); | |
876 | if (error != 0) | |
877 | return (error); | |
878 | ||
879 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
880 | zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], | |
881 | 8, 1, &zfsvfs->z_userquota_obj); | |
882 | if (error == ENOENT) | |
883 | zfsvfs->z_userquota_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_GROUPQUOTA], | |
889 | 8, 1, &zfsvfs->z_groupquota_obj); | |
890 | if (error == ENOENT) | |
891 | zfsvfs->z_groupquota_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_PROJECTQUOTA], | |
897 | 8, 1, &zfsvfs->z_projectquota_obj); | |
898 | if (error == ENOENT) | |
899 | zfsvfs->z_projectquota_obj = 0; | |
900 | else if (error != 0) | |
901 | return (error); | |
902 | ||
903 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
904 | zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA], | |
905 | 8, 1, &zfsvfs->z_userobjquota_obj); | |
906 | if (error == ENOENT) | |
907 | zfsvfs->z_userobjquota_obj = 0; | |
908 | else if (error != 0) | |
909 | return (error); | |
910 | ||
911 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
912 | zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA], | |
913 | 8, 1, &zfsvfs->z_groupobjquota_obj); | |
914 | if (error == ENOENT) | |
915 | zfsvfs->z_groupobjquota_obj = 0; | |
916 | else if (error != 0) | |
917 | return (error); | |
918 | ||
919 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
920 | zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA], | |
921 | 8, 1, &zfsvfs->z_projectobjquota_obj); | |
922 | if (error == ENOENT) | |
923 | zfsvfs->z_projectobjquota_obj = 0; | |
924 | else if (error != 0) | |
925 | return (error); | |
926 | ||
927 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, | |
928 | &zfsvfs->z_fuid_obj); | |
929 | if (error == ENOENT) | |
930 | zfsvfs->z_fuid_obj = 0; | |
931 | else if (error != 0) | |
932 | return (error); | |
933 | ||
934 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, | |
935 | &zfsvfs->z_shares_dir); | |
936 | if (error == ENOENT) | |
937 | zfsvfs->z_shares_dir = 0; | |
938 | else if (error != 0) | |
939 | return (error); | |
940 | ||
941 | /* | |
942 | * Only use the name cache if we are looking for a | |
943 | * name on a file system that does not require normalization | |
944 | * or case folding. We can also look there if we happen to be | |
945 | * on a non-normalizing, mixed sensitivity file system IF we | |
946 | * are looking for the exact name (which is always the case on | |
947 | * FreeBSD). | |
948 | */ | |
949 | zfsvfs->z_use_namecache = !zfsvfs->z_norm || | |
950 | ((zfsvfs->z_case == ZFS_CASE_MIXED) && | |
951 | !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER)); | |
952 | ||
953 | return (0); | |
954 | } | |
955 | ||
956 | taskq_t *zfsvfs_taskq; | |
957 | ||
958 | static void | |
959 | zfsvfs_task_unlinked_drain(void *context, int pending __unused) | |
960 | { | |
961 | ||
962 | zfs_unlinked_drain((zfsvfs_t *)context); | |
963 | } | |
964 | ||
965 | int | |
966 | zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp) | |
967 | { | |
968 | objset_t *os; | |
969 | zfsvfs_t *zfsvfs; | |
970 | int error; | |
971 | boolean_t ro = (readonly || (strchr(osname, '@') != NULL)); | |
972 | ||
973 | /* | |
974 | * XXX: Fix struct statfs so this isn't necessary! | |
975 | * | |
976 | * The 'osname' is used as the filesystem's special node, which means | |
977 | * it must fit in statfs.f_mntfromname, or else it can't be | |
978 | * enumerated, so libzfs_mnttab_find() returns NULL, which causes | |
979 | * 'zfs unmount' to think it's not mounted when it is. | |
980 | */ | |
981 | if (strlen(osname) >= MNAMELEN) | |
982 | return (SET_ERROR(ENAMETOOLONG)); | |
983 | ||
984 | zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP); | |
985 | ||
986 | error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, | |
987 | &os); | |
988 | if (error != 0) { | |
989 | kmem_free(zfsvfs, sizeof (zfsvfs_t)); | |
990 | return (error); | |
991 | } | |
992 | ||
993 | error = zfsvfs_create_impl(zfvp, zfsvfs, os); | |
994 | ||
995 | return (error); | |
996 | } | |
997 | ||
998 | ||
999 | int | |
1000 | zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os) | |
1001 | { | |
1002 | int error; | |
1003 | ||
1004 | zfsvfs->z_vfs = NULL; | |
1005 | zfsvfs->z_parent = zfsvfs; | |
1006 | ||
1007 | mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); | |
1008 | mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL); | |
1009 | list_create(&zfsvfs->z_all_znodes, sizeof (znode_t), | |
1010 | offsetof(znode_t, z_link_node)); | |
1011 | TASK_INIT(&zfsvfs->z_unlinked_drain_task, 0, | |
1012 | zfsvfs_task_unlinked_drain, zfsvfs); | |
5ebe425a | 1013 | ZFS_TEARDOWN_INIT(zfsvfs); |
9847f77f | 1014 | ZFS_TEARDOWN_INACTIVE_INIT(zfsvfs); |
9f0a21e6 MM |
1015 | rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL); |
1016 | for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++) | |
1017 | mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL); | |
1018 | ||
1019 | error = zfsvfs_init(zfsvfs, os); | |
1020 | if (error != 0) { | |
1021 | dmu_objset_disown(os, B_TRUE, zfsvfs); | |
1022 | *zfvp = NULL; | |
1023 | kmem_free(zfsvfs, sizeof (zfsvfs_t)); | |
1024 | return (error); | |
1025 | } | |
1026 | ||
1027 | *zfvp = zfsvfs; | |
1028 | return (0); | |
1029 | } | |
1030 | ||
1031 | static int | |
1032 | zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting) | |
1033 | { | |
1034 | int error; | |
1035 | ||
1036 | /* | |
1037 | * Check for a bad on-disk format version now since we | |
1038 | * lied about owning the dataset readonly before. | |
1039 | */ | |
1040 | if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) && | |
1041 | dmu_objset_incompatible_encryption_version(zfsvfs->z_os)) | |
1042 | return (SET_ERROR(EROFS)); | |
1043 | ||
1044 | error = zfs_register_callbacks(zfsvfs->z_vfs); | |
1045 | if (error) | |
1046 | return (error); | |
1047 | ||
9f0a21e6 MM |
1048 | /* |
1049 | * If we are not mounting (ie: online recv), then we don't | |
1050 | * have to worry about replaying the log as we blocked all | |
1051 | * operations out since we closed the ZIL. | |
1052 | */ | |
1053 | if (mounting) { | |
1054 | boolean_t readonly; | |
1055 | ||
4547fc4e | 1056 | ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL); |
fb087146 AH |
1057 | error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os); |
1058 | if (error) | |
1059 | return (error); | |
1060 | zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, | |
1061 | &zfsvfs->z_kstat.dk_zil_sums); | |
4547fc4e | 1062 | |
9f0a21e6 MM |
1063 | /* |
1064 | * During replay we remove the read only flag to | |
1065 | * allow replays to succeed. | |
1066 | */ | |
1067 | readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY; | |
1068 | if (readonly != 0) { | |
1069 | zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY; | |
1070 | } else { | |
1071 | dsl_dir_t *dd; | |
4547fc4e AJ |
1072 | zap_stats_t zs; |
1073 | ||
1074 | if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj, | |
1075 | &zs) == 0) { | |
1076 | dataset_kstats_update_nunlinks_kstat( | |
1077 | &zfsvfs->z_kstat, zs.zs_num_entries); | |
1078 | dprintf_ds(zfsvfs->z_os->os_dsl_dataset, | |
1079 | "num_entries in unlinked set: %llu", | |
8e739b2c | 1080 | (u_longlong_t)zs.zs_num_entries); |
4547fc4e | 1081 | } |
9f0a21e6 MM |
1082 | |
1083 | zfs_unlinked_drain(zfsvfs); | |
1084 | dd = zfsvfs->z_os->os_dsl_dataset->ds_dir; | |
1085 | dd->dd_activity_cancelled = B_FALSE; | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * Parse and replay the intent log. | |
1090 | * | |
1091 | * Because of ziltest, this must be done after | |
1092 | * zfs_unlinked_drain(). (Further note: ziltest | |
1093 | * doesn't use readonly mounts, where | |
1094 | * zfs_unlinked_drain() isn't called.) This is because | |
1095 | * ziltest causes spa_sync() to think it's committed, | |
1096 | * but actually it is not, so the intent log contains | |
1097 | * many txg's worth of changes. | |
1098 | * | |
1099 | * In particular, if object N is in the unlinked set in | |
1100 | * the last txg to actually sync, then it could be | |
1101 | * actually freed in a later txg and then reallocated | |
1102 | * in a yet later txg. This would write a "create | |
1103 | * object N" record to the intent log. Normally, this | |
1104 | * would be fine because the spa_sync() would have | |
1105 | * written out the fact that object N is free, before | |
1106 | * we could write the "create object N" intent log | |
1107 | * record. | |
1108 | * | |
1109 | * But when we are in ziltest mode, we advance the "open | |
1110 | * txg" without actually spa_sync()-ing the changes to | |
1111 | * disk. So we would see that object N is still | |
1112 | * allocated and in the unlinked set, and there is an | |
1113 | * intent log record saying to allocate it. | |
1114 | */ | |
1115 | if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) { | |
1116 | if (zil_replay_disable) { | |
1117 | zil_destroy(zfsvfs->z_log, B_FALSE); | |
1118 | } else { | |
1119 | boolean_t use_nc = zfsvfs->z_use_namecache; | |
1120 | zfsvfs->z_use_namecache = B_FALSE; | |
1121 | zfsvfs->z_replay = B_TRUE; | |
1122 | zil_replay(zfsvfs->z_os, zfsvfs, | |
1123 | zfs_replay_vector); | |
1124 | zfsvfs->z_replay = B_FALSE; | |
1125 | zfsvfs->z_use_namecache = use_nc; | |
1126 | } | |
1127 | } | |
1128 | ||
1129 | /* restore readonly bit */ | |
1130 | if (readonly != 0) | |
1131 | zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY; | |
fb087146 AH |
1132 | } else { |
1133 | ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL); | |
1134 | zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data, | |
1135 | &zfsvfs->z_kstat.dk_zil_sums); | |
9f0a21e6 MM |
1136 | } |
1137 | ||
1138 | /* | |
1139 | * Set the objset user_ptr to track its zfsvfs. | |
1140 | */ | |
1141 | mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); | |
1142 | dmu_objset_set_user(zfsvfs->z_os, zfsvfs); | |
1143 | mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); | |
1144 | ||
1145 | return (0); | |
1146 | } | |
1147 | ||
9f0a21e6 MM |
1148 | void |
1149 | zfsvfs_free(zfsvfs_t *zfsvfs) | |
1150 | { | |
1151 | int i; | |
1152 | ||
9f0a21e6 MM |
1153 | zfs_fuid_destroy(zfsvfs); |
1154 | ||
1155 | mutex_destroy(&zfsvfs->z_znodes_lock); | |
1156 | mutex_destroy(&zfsvfs->z_lock); | |
9f0a21e6 | 1157 | list_destroy(&zfsvfs->z_all_znodes); |
5ebe425a | 1158 | ZFS_TEARDOWN_DESTROY(zfsvfs); |
9847f77f | 1159 | ZFS_TEARDOWN_INACTIVE_DESTROY(zfsvfs); |
9f0a21e6 MM |
1160 | rw_destroy(&zfsvfs->z_fuid_lock); |
1161 | for (i = 0; i != ZFS_OBJ_MTX_SZ; i++) | |
1162 | mutex_destroy(&zfsvfs->z_hold_mtx[i]); | |
4547fc4e | 1163 | dataset_kstats_destroy(&zfsvfs->z_kstat); |
9f0a21e6 MM |
1164 | kmem_free(zfsvfs, sizeof (zfsvfs_t)); |
1165 | } | |
1166 | ||
1167 | static void | |
1168 | zfs_set_fuid_feature(zfsvfs_t *zfsvfs) | |
1169 | { | |
1170 | zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os); | |
9f0a21e6 MM |
1171 | zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os); |
1172 | } | |
1173 | ||
1174 | static int | |
1175 | zfs_domount(vfs_t *vfsp, char *osname) | |
1176 | { | |
1177 | uint64_t recordsize, fsid_guid; | |
1178 | int error = 0; | |
1179 | zfsvfs_t *zfsvfs; | |
1180 | ||
e4efb709 RM |
1181 | ASSERT3P(vfsp, !=, NULL); |
1182 | ASSERT3P(osname, !=, NULL); | |
9f0a21e6 MM |
1183 | |
1184 | error = zfsvfs_create(osname, vfsp->mnt_flag & MNT_RDONLY, &zfsvfs); | |
1185 | if (error) | |
1186 | return (error); | |
1187 | zfsvfs->z_vfs = vfsp; | |
1188 | ||
1189 | if ((error = dsl_prop_get_integer(osname, | |
1190 | "recordsize", &recordsize, NULL))) | |
1191 | goto out; | |
1192 | zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE; | |
1193 | zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize; | |
1194 | ||
1195 | vfsp->vfs_data = zfsvfs; | |
1196 | vfsp->mnt_flag |= MNT_LOCAL; | |
1197 | vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED; | |
1198 | vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES; | |
1199 | vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED; | |
1200 | /* | |
1201 | * This can cause a loss of coherence between ARC and page cache | |
1202 | * on ZoF - unclear if the problem is in FreeBSD or ZoF | |
1203 | */ | |
1204 | vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */ | |
1205 | vfsp->mnt_kern_flag |= MNTK_NOMSYNC; | |
1206 | vfsp->mnt_kern_flag |= MNTK_VMSETSIZE_BUG; | |
1207 | ||
1b376d17 MM |
1208 | #if defined(_KERNEL) && !defined(KMEM_DEBUG) |
1209 | vfsp->mnt_kern_flag |= MNTK_FPLOOKUP; | |
1210 | #endif | |
9f0a21e6 MM |
1211 | /* |
1212 | * The fsid is 64 bits, composed of an 8-bit fs type, which | |
1213 | * separates our fsid from any other filesystem types, and a | |
1214 | * 56-bit objset unique ID. The objset unique ID is unique to | |
1215 | * all objsets open on this system, provided by unique_create(). | |
1216 | * The 8-bit fs type must be put in the low bits of fsid[1] | |
1217 | * because that's where other Solaris filesystems put it. | |
1218 | */ | |
1219 | fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os); | |
e4efb709 | 1220 | ASSERT3U((fsid_guid & ~((1ULL << 56) - 1)), ==, 0); |
9f0a21e6 | 1221 | vfsp->vfs_fsid.val[0] = fsid_guid; |
e4efb709 | 1222 | vfsp->vfs_fsid.val[1] = ((fsid_guid >> 32) << 8) | |
9f0a21e6 MM |
1223 | (vfsp->mnt_vfc->vfc_typenum & 0xFF); |
1224 | ||
1225 | /* | |
1226 | * Set features for file system. | |
1227 | */ | |
1228 | zfs_set_fuid_feature(zfsvfs); | |
9f0a21e6 MM |
1229 | |
1230 | if (dmu_objset_is_snapshot(zfsvfs->z_os)) { | |
1231 | uint64_t pval; | |
1232 | ||
1233 | atime_changed_cb(zfsvfs, B_FALSE); | |
1234 | readonly_changed_cb(zfsvfs, B_TRUE); | |
1235 | if ((error = dsl_prop_get_integer(osname, | |
1236 | "xattr", &pval, NULL))) | |
1237 | goto out; | |
1238 | xattr_changed_cb(zfsvfs, pval); | |
485b50bb RM |
1239 | if ((error = dsl_prop_get_integer(osname, |
1240 | "acltype", &pval, NULL))) | |
1241 | goto out; | |
1242 | acl_type_changed_cb(zfsvfs, pval); | |
9f0a21e6 MM |
1243 | zfsvfs->z_issnap = B_TRUE; |
1244 | zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED; | |
1245 | ||
1246 | mutex_enter(&zfsvfs->z_os->os_user_ptr_lock); | |
1247 | dmu_objset_set_user(zfsvfs->z_os, zfsvfs); | |
1248 | mutex_exit(&zfsvfs->z_os->os_user_ptr_lock); | |
1249 | } else { | |
1250 | if ((error = zfsvfs_setup(zfsvfs, B_TRUE))) | |
1251 | goto out; | |
1252 | } | |
1253 | ||
1254 | vfs_mountedfrom(vfsp, osname); | |
1255 | ||
1256 | if (!zfsvfs->z_issnap) | |
1257 | zfsctl_create(zfsvfs); | |
1258 | out: | |
1259 | if (error) { | |
1260 | dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs); | |
1261 | zfsvfs_free(zfsvfs); | |
1262 | } else { | |
1263 | atomic_inc_32(&zfs_active_fs_count); | |
1264 | } | |
1265 | ||
1266 | return (error); | |
1267 | } | |
1268 | ||
65c7cc49 | 1269 | static void |
9f0a21e6 MM |
1270 | zfs_unregister_callbacks(zfsvfs_t *zfsvfs) |
1271 | { | |
1272 | objset_t *os = zfsvfs->z_os; | |
1273 | ||
1274 | if (!dmu_objset_is_snapshot(os)) | |
1275 | dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs); | |
1276 | } | |
1277 | ||
9f0a21e6 MM |
1278 | static int |
1279 | getpoolname(const char *osname, char *poolname) | |
1280 | { | |
1281 | char *p; | |
1282 | ||
1283 | p = strchr(osname, '/'); | |
1284 | if (p == NULL) { | |
1285 | if (strlen(osname) >= MAXNAMELEN) | |
1286 | return (ENAMETOOLONG); | |
1287 | (void) strcpy(poolname, osname); | |
1288 | } else { | |
1289 | if (p - osname >= MAXNAMELEN) | |
1290 | return (ENAMETOOLONG); | |
7584fbe8 | 1291 | (void) strlcpy(poolname, osname, p - osname + 1); |
9f0a21e6 MM |
1292 | } |
1293 | return (0); | |
1294 | } | |
1295 | ||
e464f7c7 MZ |
1296 | static void |
1297 | fetch_osname_options(char *name, bool *checkpointrewind) | |
1298 | { | |
1299 | ||
1300 | if (name[0] == '!') { | |
1301 | *checkpointrewind = true; | |
1302 | memmove(name, name + 1, strlen(name)); | |
1303 | } else { | |
1304 | *checkpointrewind = false; | |
1305 | } | |
1306 | } | |
1307 | ||
9f0a21e6 MM |
1308 | static int |
1309 | zfs_mount(vfs_t *vfsp) | |
1310 | { | |
1311 | kthread_t *td = curthread; | |
1312 | vnode_t *mvp = vfsp->mnt_vnodecovered; | |
1313 | cred_t *cr = td->td_ucred; | |
1314 | char *osname; | |
1315 | int error = 0; | |
1316 | int canwrite; | |
595d3ac2 | 1317 | bool checkpointrewind, isctlsnap = false; |
9f0a21e6 MM |
1318 | |
1319 | if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL)) | |
1320 | return (SET_ERROR(EINVAL)); | |
1321 | ||
1322 | /* | |
1323 | * If full-owner-access is enabled and delegated administration is | |
1324 | * turned on, we must set nosuid. | |
1325 | */ | |
1326 | if (zfs_super_owner && | |
1327 | dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) { | |
1328 | secpolicy_fs_mount_clearopts(cr, vfsp); | |
1329 | } | |
1330 | ||
e464f7c7 | 1331 | fetch_osname_options(osname, &checkpointrewind); |
d27a0028 AJ |
1332 | isctlsnap = (mvp != NULL && zfsctl_is_node(mvp) && |
1333 | strchr(osname, '@') != NULL); | |
e464f7c7 | 1334 | |
9f0a21e6 MM |
1335 | /* |
1336 | * Check for mount privilege? | |
1337 | * | |
1338 | * If we don't have privilege then see if | |
1339 | * we have local permission to allow it | |
1340 | */ | |
1341 | error = secpolicy_fs_mount(cr, mvp, vfsp); | |
595d3ac2 AJ |
1342 | if (error && isctlsnap) { |
1343 | secpolicy_fs_mount_clearopts(cr, vfsp); | |
1344 | } else if (error) { | |
9f0a21e6 MM |
1345 | if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0) |
1346 | goto out; | |
1347 | ||
1348 | if (!(vfsp->vfs_flag & MS_REMOUNT)) { | |
1349 | vattr_t vattr; | |
1350 | ||
1351 | /* | |
1352 | * Make sure user is the owner of the mount point | |
1353 | * or has sufficient privileges. | |
1354 | */ | |
1355 | ||
1356 | vattr.va_mask = AT_UID; | |
1357 | ||
1358 | vn_lock(mvp, LK_SHARED | LK_RETRY); | |
1359 | if (VOP_GETATTR(mvp, &vattr, cr)) { | |
1360 | VOP_UNLOCK1(mvp); | |
1361 | goto out; | |
1362 | } | |
1363 | ||
1364 | if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 && | |
1365 | VOP_ACCESS(mvp, VWRITE, cr, td) != 0) { | |
1366 | VOP_UNLOCK1(mvp); | |
1367 | goto out; | |
1368 | } | |
1369 | VOP_UNLOCK1(mvp); | |
1370 | } | |
1371 | ||
1372 | secpolicy_fs_mount_clearopts(cr, vfsp); | |
1373 | } | |
1374 | ||
1375 | /* | |
1376 | * Refuse to mount a filesystem if we are in a local zone and the | |
1377 | * dataset is not visible. | |
1378 | */ | |
1379 | if (!INGLOBALZONE(curproc) && | |
1380 | (!zone_dataset_visible(osname, &canwrite) || !canwrite)) { | |
595d3ac2 AJ |
1381 | boolean_t mount_snapshot = B_FALSE; |
1382 | ||
1383 | /* | |
1384 | * Snapshots may be mounted in .zfs for unjailed datasets | |
1385 | * if allowed by the jail param zfs.mount_snapshot. | |
1386 | */ | |
1387 | if (isctlsnap) { | |
1388 | struct prison *pr; | |
1389 | struct zfs_jailparam *zjp; | |
1390 | ||
1391 | pr = curthread->td_ucred->cr_prison; | |
1392 | mtx_lock(&pr->pr_mtx); | |
1393 | zjp = osd_jail_get(pr, zfs_jailparam_slot); | |
1394 | mtx_unlock(&pr->pr_mtx); | |
1395 | if (zjp && zjp->mount_snapshot) | |
1396 | mount_snapshot = B_TRUE; | |
1397 | } | |
1398 | if (!mount_snapshot) { | |
1399 | error = SET_ERROR(EPERM); | |
1400 | goto out; | |
1401 | } | |
9f0a21e6 MM |
1402 | } |
1403 | ||
9f0a21e6 MM |
1404 | vfsp->vfs_flag |= MNT_NFS4ACLS; |
1405 | ||
1406 | /* | |
1407 | * When doing a remount, we simply refresh our temporary properties | |
1408 | * according to those options set in the current VFS options. | |
1409 | */ | |
1410 | if (vfsp->vfs_flag & MS_REMOUNT) { | |
1411 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1412 | ||
1413 | /* | |
1414 | * Refresh mount options with z_teardown_lock blocking I/O while | |
1415 | * the filesystem is in an inconsistent state. | |
1416 | * The lock also serializes this code with filesystem | |
1417 | * manipulations between entry to zfs_suspend_fs() and return | |
1418 | * from zfs_resume_fs(). | |
1419 | */ | |
5ebe425a | 1420 | ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); |
9f0a21e6 MM |
1421 | zfs_unregister_callbacks(zfsvfs); |
1422 | error = zfs_register_callbacks(vfsp); | |
5ebe425a | 1423 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
1424 | goto out; |
1425 | } | |
1426 | ||
1427 | /* Initial root mount: try hard to import the requested root pool. */ | |
1428 | if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 && | |
1429 | (vfsp->vfs_flag & MNT_UPDATE) == 0) { | |
1430 | char pname[MAXNAMELEN]; | |
1431 | ||
1432 | error = getpoolname(osname, pname); | |
1433 | if (error == 0) | |
e464f7c7 | 1434 | error = spa_import_rootpool(pname, checkpointrewind); |
9f0a21e6 MM |
1435 | if (error) |
1436 | goto out; | |
1437 | } | |
1438 | DROP_GIANT(); | |
1439 | error = zfs_domount(vfsp, osname); | |
1440 | PICKUP_GIANT(); | |
1441 | ||
1442 | out: | |
1443 | return (error); | |
1444 | } | |
1445 | ||
1446 | static int | |
1447 | zfs_statfs(vfs_t *vfsp, struct statfs *statp) | |
1448 | { | |
1449 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1450 | uint64_t refdbytes, availbytes, usedobjs, availobjs; | |
768eaced | 1451 | int error; |
9f0a21e6 MM |
1452 | |
1453 | statp->f_version = STATFS_VERSION; | |
1454 | ||
768eaced CC |
1455 | if ((error = zfs_enter(zfsvfs, FTAG)) != 0) |
1456 | return (error); | |
9f0a21e6 MM |
1457 | |
1458 | dmu_objset_space(zfsvfs->z_os, | |
1459 | &refdbytes, &availbytes, &usedobjs, &availobjs); | |
1460 | ||
1461 | /* | |
1462 | * The underlying storage pool actually uses multiple block sizes. | |
1463 | * We report the fragsize as the smallest block size we support, | |
1464 | * and we report our blocksize as the filesystem's maximum blocksize. | |
1465 | */ | |
1466 | statp->f_bsize = SPA_MINBLOCKSIZE; | |
1467 | statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize; | |
1468 | ||
1469 | /* | |
1470 | * The following report "total" blocks of various kinds in the | |
1471 | * file system, but reported in terms of f_frsize - the | |
1472 | * "fragment" size. | |
1473 | */ | |
1474 | ||
1475 | statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT; | |
1476 | statp->f_bfree = availbytes / statp->f_bsize; | |
1477 | statp->f_bavail = statp->f_bfree; /* no root reservation */ | |
1478 | ||
1479 | /* | |
1480 | * statvfs() should really be called statufs(), because it assumes | |
1481 | * static metadata. ZFS doesn't preallocate files, so the best | |
1482 | * we can do is report the max that could possibly fit in f_files, | |
1483 | * and that minus the number actually used in f_ffree. | |
1484 | * For f_ffree, report the smaller of the number of object available | |
1485 | * and the number of blocks (each object will take at least a block). | |
1486 | */ | |
1487 | statp->f_ffree = MIN(availobjs, statp->f_bfree); | |
1488 | statp->f_files = statp->f_ffree + usedobjs; | |
1489 | ||
1490 | /* | |
1491 | * We're a zfs filesystem. | |
1492 | */ | |
1493 | strlcpy(statp->f_fstypename, "zfs", | |
1494 | sizeof (statp->f_fstypename)); | |
1495 | ||
1496 | strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname, | |
1497 | sizeof (statp->f_mntfromname)); | |
1498 | strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname, | |
1499 | sizeof (statp->f_mntonname)); | |
1500 | ||
1501 | statp->f_namemax = MAXNAMELEN - 1; | |
1502 | ||
768eaced | 1503 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1504 | return (0); |
1505 | } | |
1506 | ||
1507 | static int | |
1508 | zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp) | |
1509 | { | |
1510 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1511 | znode_t *rootzp; | |
1512 | int error; | |
1513 | ||
768eaced CC |
1514 | if ((error = zfs_enter(zfsvfs, FTAG)) != 0) |
1515 | return (error); | |
9f0a21e6 MM |
1516 | |
1517 | error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp); | |
1518 | if (error == 0) | |
1519 | *vpp = ZTOV(rootzp); | |
1520 | ||
768eaced | 1521 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1522 | |
1523 | if (error == 0) { | |
1524 | error = vn_lock(*vpp, flags); | |
1525 | if (error != 0) { | |
1526 | VN_RELE(*vpp); | |
1527 | *vpp = NULL; | |
1528 | } | |
1529 | } | |
1530 | return (error); | |
1531 | } | |
1532 | ||
1533 | /* | |
1534 | * Teardown the zfsvfs::z_os. | |
1535 | * | |
1536 | * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock' | |
1537 | * and 'z_teardown_inactive_lock' held. | |
1538 | */ | |
1539 | static int | |
1540 | zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting) | |
1541 | { | |
1542 | znode_t *zp; | |
1543 | dsl_dir_t *dd; | |
1544 | ||
1545 | /* | |
1546 | * If someone has not already unmounted this file system, | |
1547 | * drain the zrele_taskq to ensure all active references to the | |
1548 | * zfsvfs_t have been handled only then can it be safely destroyed. | |
1549 | */ | |
1550 | if (zfsvfs->z_os) { | |
1551 | /* | |
1552 | * If we're unmounting we have to wait for the list to | |
1553 | * drain completely. | |
1554 | * | |
1555 | * If we're not unmounting there's no guarantee the list | |
1556 | * will drain completely, but zreles run from the taskq | |
1557 | * may add the parents of dir-based xattrs to the taskq | |
1558 | * so we want to wait for these. | |
1559 | * | |
ee720ad7 MG |
1560 | * We can safely check z_all_znodes for being empty because the |
1561 | * VFS has already blocked operations which add to it. | |
9f0a21e6 MM |
1562 | */ |
1563 | int round = 0; | |
ee720ad7 | 1564 | while (!list_is_empty(&zfsvfs->z_all_znodes)) { |
9f0a21e6 MM |
1565 | taskq_wait_outstanding(dsl_pool_zrele_taskq( |
1566 | dmu_objset_pool(zfsvfs->z_os)), 0); | |
1567 | if (++round > 1 && !unmounting) | |
1568 | break; | |
1569 | } | |
1570 | } | |
5ebe425a | 1571 | ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); |
9f0a21e6 MM |
1572 | |
1573 | if (!unmounting) { | |
1574 | /* | |
1575 | * We purge the parent filesystem's vfsp as the parent | |
1576 | * filesystem and all of its snapshots have their vnode's | |
1577 | * v_vfsp set to the parent's filesystem's vfsp. Note, | |
1578 | * 'z_parent' is self referential for non-snapshots. | |
1579 | */ | |
1580 | #ifdef FREEBSD_NAMECACHE | |
f6bb7c02 MG |
1581 | #if __FreeBSD_version >= 1300117 |
1582 | cache_purgevfs(zfsvfs->z_parent->z_vfs); | |
1583 | #else | |
9f0a21e6 | 1584 | cache_purgevfs(zfsvfs->z_parent->z_vfs, true); |
f6bb7c02 | 1585 | #endif |
9f0a21e6 MM |
1586 | #endif |
1587 | } | |
1588 | ||
1589 | /* | |
1590 | * Close the zil. NB: Can't close the zil while zfs_inactive | |
1591 | * threads are blocked as zil_close can call zfs_inactive. | |
1592 | */ | |
1593 | if (zfsvfs->z_log) { | |
1594 | zil_close(zfsvfs->z_log); | |
1595 | zfsvfs->z_log = NULL; | |
1596 | } | |
1597 | ||
9847f77f | 1598 | ZFS_TEARDOWN_INACTIVE_ENTER_WRITE(zfsvfs); |
9f0a21e6 MM |
1599 | |
1600 | /* | |
1601 | * If we are not unmounting (ie: online recv) and someone already | |
1602 | * unmounted this file system while we were doing the switcheroo, | |
1603 | * or a reopen of z_os failed then just bail out now. | |
1604 | */ | |
1605 | if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) { | |
9847f77f | 1606 | ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs); |
5ebe425a | 1607 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
1608 | return (SET_ERROR(EIO)); |
1609 | } | |
1610 | ||
1611 | /* | |
1612 | * At this point there are no vops active, and any new vops will | |
1613 | * fail with EIO since we have z_teardown_lock for writer (only | |
dd4bc569 | 1614 | * relevant for forced unmount). |
9f0a21e6 MM |
1615 | * |
1616 | * Release all holds on dbufs. | |
1617 | */ | |
1618 | mutex_enter(&zfsvfs->z_znodes_lock); | |
1619 | for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL; | |
e4efb709 RM |
1620 | zp = list_next(&zfsvfs->z_all_znodes, zp)) { |
1621 | if (zp->z_sa_hdl != NULL) { | |
9f0a21e6 MM |
1622 | zfs_znode_dmu_fini(zp); |
1623 | } | |
e4efb709 | 1624 | } |
9f0a21e6 MM |
1625 | mutex_exit(&zfsvfs->z_znodes_lock); |
1626 | ||
1627 | /* | |
1628 | * If we are unmounting, set the unmounted flag and let new vops | |
1629 | * unblock. zfs_inactive will have the unmounted behavior, and all | |
1630 | * other vops will fail with EIO. | |
1631 | */ | |
1632 | if (unmounting) { | |
1633 | zfsvfs->z_unmounted = B_TRUE; | |
9847f77f | 1634 | ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs); |
5ebe425a | 1635 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
1636 | } |
1637 | ||
1638 | /* | |
1639 | * z_os will be NULL if there was an error in attempting to reopen | |
1640 | * zfsvfs, so just return as the properties had already been | |
1641 | * unregistered and cached data had been evicted before. | |
1642 | */ | |
1643 | if (zfsvfs->z_os == NULL) | |
1644 | return (0); | |
1645 | ||
1646 | /* | |
1647 | * Unregister properties. | |
1648 | */ | |
1649 | zfs_unregister_callbacks(zfsvfs); | |
1650 | ||
1651 | /* | |
1652 | * Evict cached data | |
1653 | */ | |
1654 | if (!zfs_is_readonly(zfsvfs)) | |
1655 | txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0); | |
1656 | dmu_objset_evict_dbufs(zfsvfs->z_os); | |
1657 | dd = zfsvfs->z_os->os_dsl_dataset->ds_dir; | |
1658 | dsl_dir_cancel_waiters(dd); | |
1659 | ||
1660 | return (0); | |
1661 | } | |
1662 | ||
9f0a21e6 MM |
1663 | static int |
1664 | zfs_umount(vfs_t *vfsp, int fflag) | |
1665 | { | |
1666 | kthread_t *td = curthread; | |
1667 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1668 | objset_t *os; | |
1669 | cred_t *cr = td->td_ucred; | |
1670 | int ret; | |
1671 | ||
1672 | ret = secpolicy_fs_unmount(cr, vfsp); | |
1673 | if (ret) { | |
1674 | if (dsl_deleg_access((char *)vfsp->vfs_resource, | |
1675 | ZFS_DELEG_PERM_MOUNT, cr)) | |
1676 | return (ret); | |
1677 | } | |
1678 | ||
1679 | /* | |
1680 | * Unmount any snapshots mounted under .zfs before unmounting the | |
1681 | * dataset itself. | |
1682 | */ | |
1683 | if (zfsvfs->z_ctldir != NULL) { | |
1684 | if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) | |
1685 | return (ret); | |
1686 | } | |
1687 | ||
1688 | if (fflag & MS_FORCE) { | |
1689 | /* | |
1690 | * Mark file system as unmounted before calling | |
1691 | * vflush(FORCECLOSE). This way we ensure no future vnops | |
1692 | * will be called and risk operating on DOOMED vnodes. | |
1693 | */ | |
5ebe425a | 1694 | ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG); |
9f0a21e6 | 1695 | zfsvfs->z_unmounted = B_TRUE; |
5ebe425a | 1696 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
1697 | } |
1698 | ||
1699 | /* | |
1700 | * Flush all the files. | |
1701 | */ | |
1702 | ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td); | |
1703 | if (ret != 0) | |
1704 | return (ret); | |
1705 | while (taskqueue_cancel(zfsvfs_taskq->tq_queue, | |
1706 | &zfsvfs->z_unlinked_drain_task, NULL) != 0) | |
1707 | taskqueue_drain(zfsvfs_taskq->tq_queue, | |
1708 | &zfsvfs->z_unlinked_drain_task); | |
1709 | ||
e4efb709 | 1710 | VERIFY0(zfsvfs_teardown(zfsvfs, B_TRUE)); |
9f0a21e6 MM |
1711 | os = zfsvfs->z_os; |
1712 | ||
1713 | /* | |
1714 | * z_os will be NULL if there was an error in | |
1715 | * attempting to reopen zfsvfs. | |
1716 | */ | |
1717 | if (os != NULL) { | |
1718 | /* | |
1719 | * Unset the objset user_ptr. | |
1720 | */ | |
1721 | mutex_enter(&os->os_user_ptr_lock); | |
1722 | dmu_objset_set_user(os, NULL); | |
1723 | mutex_exit(&os->os_user_ptr_lock); | |
1724 | ||
1725 | /* | |
1726 | * Finally release the objset | |
1727 | */ | |
1728 | dmu_objset_disown(os, B_TRUE, zfsvfs); | |
1729 | } | |
1730 | ||
1731 | /* | |
1732 | * We can now safely destroy the '.zfs' directory node. | |
1733 | */ | |
1734 | if (zfsvfs->z_ctldir != NULL) | |
1735 | zfsctl_destroy(zfsvfs); | |
1736 | zfs_freevfs(vfsp); | |
1737 | ||
1738 | return (0); | |
1739 | } | |
1740 | ||
1741 | static int | |
1742 | zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp) | |
1743 | { | |
1744 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1745 | znode_t *zp; | |
1746 | int err; | |
1747 | ||
1748 | /* | |
1749 | * zfs_zget() can't operate on virtual entries like .zfs/ or | |
1750 | * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP. | |
1751 | * This will make NFS to switch to LOOKUP instead of using VGET. | |
1752 | */ | |
1753 | if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR || | |
1754 | (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir)) | |
1755 | return (EOPNOTSUPP); | |
1756 | ||
768eaced CC |
1757 | if ((err = zfs_enter(zfsvfs, FTAG)) != 0) |
1758 | return (err); | |
9f0a21e6 MM |
1759 | err = zfs_zget(zfsvfs, ino, &zp); |
1760 | if (err == 0 && zp->z_unlinked) { | |
1761 | vrele(ZTOV(zp)); | |
1762 | err = EINVAL; | |
1763 | } | |
1764 | if (err == 0) | |
1765 | *vpp = ZTOV(zp); | |
768eaced | 1766 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1767 | if (err == 0) { |
1768 | err = vn_lock(*vpp, flags); | |
1769 | if (err != 0) | |
1770 | vrele(*vpp); | |
1771 | } | |
1772 | if (err != 0) | |
1773 | *vpp = NULL; | |
1774 | return (err); | |
1775 | } | |
1776 | ||
1c08fa8b | 1777 | static int |
2e6af52b | 1778 | #if __FreeBSD_version >= 1300098 |
1c08fa8b RM |
1779 | zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp, |
1780 | struct ucred **credanonp, int *numsecflavors, int *secflavors) | |
2e6af52b | 1781 | #else |
9f0a21e6 MM |
1782 | zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp, |
1783 | struct ucred **credanonp, int *numsecflavors, int **secflavors) | |
2e6af52b | 1784 | #endif |
9f0a21e6 MM |
1785 | { |
1786 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1787 | ||
1788 | /* | |
1789 | * If this is regular file system vfsp is the same as | |
1790 | * zfsvfs->z_parent->z_vfs, but if it is snapshot, | |
1791 | * zfsvfs->z_parent->z_vfs represents parent file system | |
1792 | * which we have to use here, because only this file system | |
1793 | * has mnt_export configured. | |
1794 | */ | |
1795 | return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp, | |
1796 | credanonp, numsecflavors, secflavors)); | |
1797 | } | |
1798 | ||
c70bb2f6 AZ |
1799 | _Static_assert(sizeof (struct fid) >= SHORT_FID_LEN, |
1800 | "struct fid bigger than SHORT_FID_LEN"); | |
1801 | _Static_assert(sizeof (struct fid) >= LONG_FID_LEN, | |
1802 | "struct fid bigger than LONG_FID_LEN"); | |
9f0a21e6 MM |
1803 | |
1804 | static int | |
1805 | zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp) | |
1806 | { | |
1807 | struct componentname cn; | |
1808 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
1809 | znode_t *zp; | |
1810 | vnode_t *dvp; | |
1811 | uint64_t object = 0; | |
1812 | uint64_t fid_gen = 0; | |
43dbf881 | 1813 | uint64_t setgen = 0; |
9f0a21e6 MM |
1814 | uint64_t gen_mask; |
1815 | uint64_t zp_gen; | |
1816 | int i, err; | |
1817 | ||
1818 | *vpp = NULL; | |
1819 | ||
768eaced CC |
1820 | if ((err = zfs_enter(zfsvfs, FTAG)) != 0) |
1821 | return (err); | |
9f0a21e6 MM |
1822 | |
1823 | /* | |
1824 | * On FreeBSD we can get snapshot's mount point or its parent file | |
1825 | * system mount point depending if snapshot is already mounted or not. | |
1826 | */ | |
1827 | if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) { | |
1828 | zfid_long_t *zlfid = (zfid_long_t *)fidp; | |
1829 | uint64_t objsetid = 0; | |
9f0a21e6 MM |
1830 | |
1831 | for (i = 0; i < sizeof (zlfid->zf_setid); i++) | |
1832 | objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); | |
1833 | ||
1834 | for (i = 0; i < sizeof (zlfid->zf_setgen); i++) | |
1835 | setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); | |
1836 | ||
768eaced | 1837 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1838 | |
1839 | err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs); | |
1840 | if (err) | |
1841 | return (SET_ERROR(EINVAL)); | |
768eaced CC |
1842 | if ((err = zfs_enter(zfsvfs, FTAG)) != 0) |
1843 | return (err); | |
9f0a21e6 MM |
1844 | } |
1845 | ||
1846 | if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { | |
1847 | zfid_short_t *zfid = (zfid_short_t *)fidp; | |
1848 | ||
1849 | for (i = 0; i < sizeof (zfid->zf_object); i++) | |
1850 | object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); | |
1851 | ||
1852 | for (i = 0; i < sizeof (zfid->zf_gen); i++) | |
1853 | fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); | |
1854 | } else { | |
768eaced | 1855 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1856 | return (SET_ERROR(EINVAL)); |
1857 | } | |
1858 | ||
ed566bf1 MJ |
1859 | if (fidp->fid_len == LONG_FID_LEN && setgen != 0) { |
1860 | zfs_exit(zfsvfs, FTAG); | |
43dbf881 AZ |
1861 | dprintf("snapdir fid: fid_gen (%llu) and setgen (%llu)\n", |
1862 | (u_longlong_t)fid_gen, (u_longlong_t)setgen); | |
1863 | return (SET_ERROR(EINVAL)); | |
1864 | } | |
1865 | ||
9f0a21e6 MM |
1866 | /* |
1867 | * A zero fid_gen means we are in .zfs or the .zfs/snapshot | |
1868 | * directory tree. If the object == zfsvfs->z_shares_dir, then | |
1869 | * we are in the .zfs/shares directory tree. | |
1870 | */ | |
1871 | if ((fid_gen == 0 && | |
1872 | (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) || | |
1873 | (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) { | |
768eaced | 1874 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1875 | VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp)); |
1876 | if (object == ZFSCTL_INO_SNAPDIR) { | |
1877 | cn.cn_nameptr = "snapshot"; | |
1878 | cn.cn_namelen = strlen(cn.cn_nameptr); | |
1879 | cn.cn_nameiop = LOOKUP; | |
1880 | cn.cn_flags = ISLASTCN | LOCKLEAF; | |
1881 | cn.cn_lkflags = flags; | |
1882 | VERIFY0(VOP_LOOKUP(dvp, vpp, &cn)); | |
1883 | vput(dvp); | |
1884 | } else if (object == zfsvfs->z_shares_dir) { | |
1885 | /* | |
1886 | * XXX This branch must not be taken, | |
1887 | * if it is, then the lookup below will | |
1888 | * explode. | |
1889 | */ | |
1890 | cn.cn_nameptr = "shares"; | |
1891 | cn.cn_namelen = strlen(cn.cn_nameptr); | |
1892 | cn.cn_nameiop = LOOKUP; | |
1893 | cn.cn_flags = ISLASTCN; | |
1894 | cn.cn_lkflags = flags; | |
1895 | VERIFY0(VOP_LOOKUP(dvp, vpp, &cn)); | |
1896 | vput(dvp); | |
1897 | } else { | |
1898 | *vpp = dvp; | |
1899 | } | |
1900 | return (err); | |
1901 | } | |
1902 | ||
1903 | gen_mask = -1ULL >> (64 - 8 * i); | |
1904 | ||
8e739b2c RE |
1905 | dprintf("getting %llu [%llu mask %llx]\n", (u_longlong_t)object, |
1906 | (u_longlong_t)fid_gen, | |
1907 | (u_longlong_t)gen_mask); | |
9f0a21e6 | 1908 | if ((err = zfs_zget(zfsvfs, object, &zp))) { |
768eaced | 1909 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1910 | return (err); |
1911 | } | |
1912 | (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen, | |
1913 | sizeof (uint64_t)); | |
1914 | zp_gen = zp_gen & gen_mask; | |
1915 | if (zp_gen == 0) | |
1916 | zp_gen = 1; | |
1917 | if (zp->z_unlinked || zp_gen != fid_gen) { | |
8e739b2c RE |
1918 | dprintf("znode gen (%llu) != fid gen (%llu)\n", |
1919 | (u_longlong_t)zp_gen, (u_longlong_t)fid_gen); | |
9f0a21e6 | 1920 | vrele(ZTOV(zp)); |
768eaced | 1921 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1922 | return (SET_ERROR(EINVAL)); |
1923 | } | |
1924 | ||
1925 | *vpp = ZTOV(zp); | |
768eaced | 1926 | zfs_exit(zfsvfs, FTAG); |
9f0a21e6 MM |
1927 | err = vn_lock(*vpp, flags); |
1928 | if (err == 0) | |
1929 | vnode_create_vobject(*vpp, zp->z_size, curthread); | |
1930 | else | |
1931 | *vpp = NULL; | |
1932 | return (err); | |
1933 | } | |
1934 | ||
1935 | /* | |
1936 | * Block out VOPs and close zfsvfs_t::z_os | |
1937 | * | |
1938 | * Note, if successful, then we return with the 'z_teardown_lock' and | |
1939 | * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying | |
1940 | * dataset and objset intact so that they can be atomically handed off during | |
1941 | * a subsequent rollback or recv operation and the resume thereafter. | |
1942 | */ | |
1943 | int | |
1944 | zfs_suspend_fs(zfsvfs_t *zfsvfs) | |
1945 | { | |
1946 | int error; | |
1947 | ||
1948 | if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0) | |
1949 | return (error); | |
1950 | ||
1951 | return (0); | |
1952 | } | |
1953 | ||
1954 | /* | |
1955 | * Rebuild SA and release VOPs. Note that ownership of the underlying dataset | |
1956 | * is an invariant across any of the operations that can be performed while the | |
1957 | * filesystem was suspended. Whether it succeeded or failed, the preconditions | |
1958 | * are the same: the relevant objset and associated dataset are owned by | |
1959 | * zfsvfs, held, and long held on entry. | |
1960 | */ | |
1961 | int | |
1962 | zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) | |
1963 | { | |
1964 | int err; | |
1965 | znode_t *zp; | |
1966 | ||
5ebe425a | 1967 | ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); |
9847f77f | 1968 | ASSERT(ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zfsvfs)); |
9f0a21e6 MM |
1969 | |
1970 | /* | |
1971 | * We already own this, so just update the objset_t, as the one we | |
1972 | * had before may have been evicted. | |
1973 | */ | |
1974 | objset_t *os; | |
1975 | VERIFY3P(ds->ds_owner, ==, zfsvfs); | |
1976 | VERIFY(dsl_dataset_long_held(ds)); | |
1977 | dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); | |
1978 | dsl_pool_config_enter(dp, FTAG); | |
1979 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
1980 | dsl_pool_config_exit(dp, FTAG); | |
1981 | ||
1982 | err = zfsvfs_init(zfsvfs, os); | |
1983 | if (err != 0) | |
1984 | goto bail; | |
1985 | ||
1986 | ds->ds_dir->dd_activity_cancelled = B_FALSE; | |
e4efb709 | 1987 | VERIFY0(zfsvfs_setup(zfsvfs, B_FALSE)); |
9f0a21e6 MM |
1988 | |
1989 | zfs_set_fuid_feature(zfsvfs); | |
1990 | ||
1991 | /* | |
1992 | * Attempt to re-establish all the active znodes with | |
1993 | * their dbufs. If a zfs_rezget() fails, then we'll let | |
768eaced | 1994 | * any potential callers discover that via zfs_enter_verify_zp |
9f0a21e6 MM |
1995 | * when they try to use their znode. |
1996 | */ | |
1997 | mutex_enter(&zfsvfs->z_znodes_lock); | |
1998 | for (zp = list_head(&zfsvfs->z_all_znodes); zp; | |
1999 | zp = list_next(&zfsvfs->z_all_znodes, zp)) { | |
2000 | (void) zfs_rezget(zp); | |
2001 | } | |
2002 | mutex_exit(&zfsvfs->z_znodes_lock); | |
2003 | ||
2004 | bail: | |
2005 | /* release the VOPs */ | |
9847f77f | 2006 | ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs); |
5ebe425a | 2007 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
2008 | |
2009 | if (err) { | |
2010 | /* | |
2011 | * Since we couldn't setup the sa framework, try to force | |
2012 | * unmount this file system. | |
2013 | */ | |
2014 | if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) { | |
2015 | vfs_ref(zfsvfs->z_vfs); | |
2016 | (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread); | |
2017 | } | |
2018 | } | |
2019 | return (err); | |
2020 | } | |
2021 | ||
2022 | static void | |
2023 | zfs_freevfs(vfs_t *vfsp) | |
2024 | { | |
2025 | zfsvfs_t *zfsvfs = vfsp->vfs_data; | |
2026 | ||
2027 | zfsvfs_free(zfsvfs); | |
2028 | ||
2029 | atomic_dec_32(&zfs_active_fs_count); | |
2030 | } | |
2031 | ||
2032 | #ifdef __i386__ | |
2033 | static int desiredvnodes_backup; | |
47ed79ff MM |
2034 | #include <sys/vmmeter.h> |
2035 | ||
2036 | ||
2037 | #include <vm/vm_page.h> | |
2038 | #include <vm/vm_object.h> | |
2039 | #include <vm/vm_kern.h> | |
2040 | #include <vm/vm_map.h> | |
9f0a21e6 MM |
2041 | #endif |
2042 | ||
2043 | static void | |
2044 | zfs_vnodes_adjust(void) | |
2045 | { | |
2046 | #ifdef __i386__ | |
2047 | int newdesiredvnodes; | |
2048 | ||
2049 | desiredvnodes_backup = desiredvnodes; | |
2050 | ||
2051 | /* | |
2052 | * We calculate newdesiredvnodes the same way it is done in | |
2053 | * vntblinit(). If it is equal to desiredvnodes, it means that | |
2054 | * it wasn't tuned by the administrator and we can tune it down. | |
2055 | */ | |
2056 | newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 * | |
2057 | vm_kmem_size / (5 * (sizeof (struct vm_object) + | |
2058 | sizeof (struct vnode)))); | |
2059 | if (newdesiredvnodes == desiredvnodes) | |
2060 | desiredvnodes = (3 * newdesiredvnodes) / 4; | |
2061 | #endif | |
2062 | } | |
2063 | ||
2064 | static void | |
2065 | zfs_vnodes_adjust_back(void) | |
2066 | { | |
2067 | ||
2068 | #ifdef __i386__ | |
2069 | desiredvnodes = desiredvnodes_backup; | |
2070 | #endif | |
2071 | } | |
2072 | ||
799e09f7 AM |
2073 | #if __FreeBSD_version >= 1300139 |
2074 | static struct sx zfs_vnlru_lock; | |
2075 | static struct vnode *zfs_vnlru_marker; | |
2076 | #endif | |
2077 | static arc_prune_t *zfs_prune; | |
2078 | ||
2079 | static void | |
2080 | zfs_prune_task(uint64_t nr_to_scan, void *arg __unused) | |
2081 | { | |
2082 | if (nr_to_scan > INT_MAX) | |
2083 | nr_to_scan = INT_MAX; | |
2084 | #if __FreeBSD_version >= 1300139 | |
2085 | sx_xlock(&zfs_vnlru_lock); | |
2086 | vnlru_free_vfsops(nr_to_scan, &zfs_vfsops, zfs_vnlru_marker); | |
2087 | sx_xunlock(&zfs_vnlru_lock); | |
2088 | #else | |
2089 | vnlru_free(nr_to_scan, &zfs_vfsops); | |
2090 | #endif | |
2091 | } | |
2092 | ||
9f0a21e6 MM |
2093 | void |
2094 | zfs_init(void) | |
2095 | { | |
2096 | ||
2097 | printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n"); | |
2098 | ||
2099 | /* | |
2100 | * Initialize .zfs directory structures | |
2101 | */ | |
2102 | zfsctl_init(); | |
2103 | ||
2104 | /* | |
2105 | * Initialize znode cache, vnode ops, etc... | |
2106 | */ | |
2107 | zfs_znode_init(); | |
2108 | ||
2109 | /* | |
2110 | * Reduce number of vnodes. Originally number of vnodes is calculated | |
2111 | * with UFS inode in mind. We reduce it here, because it's too big for | |
2112 | * ZFS/i386. | |
2113 | */ | |
2114 | zfs_vnodes_adjust(); | |
2115 | ||
7bcb7f08 | 2116 | dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info); |
9f0a21e6 MM |
2117 | |
2118 | zfsvfs_taskq = taskq_create("zfsvfs", 1, minclsyspri, 0, 0, 0); | |
799e09f7 AM |
2119 | |
2120 | #if __FreeBSD_version >= 1300139 | |
2121 | zfs_vnlru_marker = vnlru_alloc_marker(); | |
2122 | sx_init(&zfs_vnlru_lock, "zfs vnlru lock"); | |
2123 | #endif | |
2124 | zfs_prune = arc_add_prune_callback(zfs_prune_task, NULL); | |
9f0a21e6 MM |
2125 | } |
2126 | ||
2127 | void | |
2128 | zfs_fini(void) | |
2129 | { | |
799e09f7 AM |
2130 | arc_remove_prune_callback(zfs_prune); |
2131 | #if __FreeBSD_version >= 1300139 | |
2132 | vnlru_free_marker(zfs_vnlru_marker); | |
2133 | sx_destroy(&zfs_vnlru_lock); | |
2134 | #endif | |
2135 | ||
9f0a21e6 MM |
2136 | taskq_destroy(zfsvfs_taskq); |
2137 | zfsctl_fini(); | |
2138 | zfs_znode_fini(); | |
2139 | zfs_vnodes_adjust_back(); | |
2140 | } | |
2141 | ||
2142 | int | |
2143 | zfs_busy(void) | |
2144 | { | |
2145 | return (zfs_active_fs_count != 0); | |
2146 | } | |
2147 | ||
2148 | /* | |
2149 | * Release VOPs and unmount a suspended filesystem. | |
2150 | */ | |
2151 | int | |
2152 | zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds) | |
2153 | { | |
5ebe425a | 2154 | ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs)); |
9847f77f | 2155 | ASSERT(ZFS_TEARDOWN_INACTIVE_WRITE_HELD(zfsvfs)); |
9f0a21e6 MM |
2156 | |
2157 | /* | |
2158 | * We already own this, so just hold and rele it to update the | |
2159 | * objset_t, as the one we had before may have been evicted. | |
2160 | */ | |
2161 | objset_t *os; | |
2162 | VERIFY3P(ds->ds_owner, ==, zfsvfs); | |
2163 | VERIFY(dsl_dataset_long_held(ds)); | |
2164 | dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds)); | |
2165 | dsl_pool_config_enter(dp, FTAG); | |
2166 | VERIFY0(dmu_objset_from_ds(ds, &os)); | |
2167 | dsl_pool_config_exit(dp, FTAG); | |
2168 | zfsvfs->z_os = os; | |
2169 | ||
2170 | /* release the VOPs */ | |
9847f77f | 2171 | ZFS_TEARDOWN_INACTIVE_EXIT_WRITE(zfsvfs); |
5ebe425a | 2172 | ZFS_TEARDOWN_EXIT(zfsvfs, FTAG); |
9f0a21e6 MM |
2173 | |
2174 | /* | |
2175 | * Try to force unmount this file system. | |
2176 | */ | |
2177 | (void) zfs_umount(zfsvfs->z_vfs, 0); | |
2178 | zfsvfs->z_unmounted = B_TRUE; | |
2179 | return (0); | |
2180 | } | |
2181 | ||
2182 | int | |
2183 | zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers) | |
2184 | { | |
2185 | int error; | |
2186 | objset_t *os = zfsvfs->z_os; | |
2187 | dmu_tx_t *tx; | |
2188 | ||
2189 | if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) | |
2190 | return (SET_ERROR(EINVAL)); | |
2191 | ||
2192 | if (newvers < zfsvfs->z_version) | |
2193 | return (SET_ERROR(EINVAL)); | |
2194 | ||
2195 | if (zfs_spa_version_map(newvers) > | |
2196 | spa_version(dmu_objset_spa(zfsvfs->z_os))) | |
2197 | return (SET_ERROR(ENOTSUP)); | |
2198 | ||
2199 | tx = dmu_tx_create(os); | |
2200 | dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); | |
2201 | if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { | |
2202 | dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, | |
2203 | ZFS_SA_ATTRS); | |
2204 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); | |
2205 | } | |
2206 | error = dmu_tx_assign(tx, TXG_WAIT); | |
2207 | if (error) { | |
2208 | dmu_tx_abort(tx); | |
2209 | return (error); | |
2210 | } | |
2211 | ||
2212 | error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, | |
2213 | 8, 1, &newvers, tx); | |
2214 | ||
2215 | if (error) { | |
2216 | dmu_tx_commit(tx); | |
2217 | return (error); | |
2218 | } | |
2219 | ||
2220 | if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) { | |
2221 | uint64_t sa_obj; | |
2222 | ||
2223 | ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=, | |
2224 | SPA_VERSION_SA); | |
2225 | sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, | |
2226 | DMU_OT_NONE, 0, tx); | |
2227 | ||
2228 | error = zap_add(os, MASTER_NODE_OBJ, | |
2229 | ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); | |
2230 | ASSERT0(error); | |
2231 | ||
e4efb709 | 2232 | VERIFY0(sa_set_sa_object(os, sa_obj)); |
9f0a21e6 MM |
2233 | sa_register_update_callback(os, zfs_sa_upgrade); |
2234 | } | |
2235 | ||
2236 | spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, | |
47ed79ff MM |
2237 | "from %ju to %ju", (uintmax_t)zfsvfs->z_version, |
2238 | (uintmax_t)newvers); | |
9f0a21e6 MM |
2239 | dmu_tx_commit(tx); |
2240 | ||
2241 | zfsvfs->z_version = newvers; | |
2242 | os->os_version = newvers; | |
2243 | ||
2244 | zfs_set_fuid_feature(zfsvfs); | |
2245 | ||
2246 | return (0); | |
2247 | } | |
2248 | ||
9f0a21e6 | 2249 | /* |
dd4bc569 | 2250 | * Return true if the corresponding vfs's unmounted flag is set. |
9f0a21e6 MM |
2251 | * Otherwise return false. |
2252 | * If this function returns true we know VFS unmount has been initiated. | |
2253 | */ | |
2254 | boolean_t | |
2255 | zfs_get_vfs_flag_unmounted(objset_t *os) | |
2256 | { | |
2257 | zfsvfs_t *zfvp; | |
2258 | boolean_t unmounted = B_FALSE; | |
2259 | ||
e4efb709 | 2260 | ASSERT3U(dmu_objset_type(os), ==, DMU_OST_ZFS); |
9f0a21e6 MM |
2261 | |
2262 | mutex_enter(&os->os_user_ptr_lock); | |
2263 | zfvp = dmu_objset_get_user(os); | |
2264 | if (zfvp != NULL && zfvp->z_vfs != NULL && | |
2265 | (zfvp->z_vfs->mnt_kern_flag & MNTK_UNMOUNT)) | |
2266 | unmounted = B_TRUE; | |
2267 | mutex_exit(&os->os_user_ptr_lock); | |
2268 | ||
2269 | return (unmounted); | |
2270 | } | |
2271 | ||
2272 | #ifdef _KERNEL | |
2273 | void | |
2274 | zfsvfs_update_fromname(const char *oldname, const char *newname) | |
2275 | { | |
2276 | char tmpbuf[MAXPATHLEN]; | |
2277 | struct mount *mp; | |
2278 | char *fromname; | |
2279 | size_t oldlen; | |
2280 | ||
2281 | oldlen = strlen(oldname); | |
2282 | ||
2283 | mtx_lock(&mountlist_mtx); | |
2284 | TAILQ_FOREACH(mp, &mountlist, mnt_list) { | |
2285 | fromname = mp->mnt_stat.f_mntfromname; | |
2286 | if (strcmp(fromname, oldname) == 0) { | |
2287 | (void) strlcpy(fromname, newname, | |
2288 | sizeof (mp->mnt_stat.f_mntfromname)); | |
2289 | continue; | |
2290 | } | |
2291 | if (strncmp(fromname, oldname, oldlen) == 0 && | |
2292 | (fromname[oldlen] == '/' || fromname[oldlen] == '@')) { | |
2293 | (void) snprintf(tmpbuf, sizeof (tmpbuf), "%s%s", | |
2294 | newname, fromname + oldlen); | |
2295 | (void) strlcpy(fromname, tmpbuf, | |
2296 | sizeof (mp->mnt_stat.f_mntfromname)); | |
2297 | continue; | |
2298 | } | |
2299 | } | |
2300 | mtx_unlock(&mountlist_mtx); | |
2301 | } | |
2302 | #endif | |
595d3ac2 AJ |
2303 | |
2304 | /* | |
2305 | * Find a prison with ZFS info. | |
2306 | * Return the ZFS info and the (locked) prison. | |
2307 | */ | |
2308 | static struct zfs_jailparam * | |
2309 | zfs_jailparam_find(struct prison *spr, struct prison **prp) | |
2310 | { | |
2311 | struct prison *pr; | |
2312 | struct zfs_jailparam *zjp; | |
2313 | ||
2314 | for (pr = spr; ; pr = pr->pr_parent) { | |
2315 | mtx_lock(&pr->pr_mtx); | |
2316 | if (pr == &prison0) { | |
2317 | zjp = &zfs_jailparam0; | |
2318 | break; | |
2319 | } | |
2320 | zjp = osd_jail_get(pr, zfs_jailparam_slot); | |
2321 | if (zjp != NULL) | |
2322 | break; | |
2323 | mtx_unlock(&pr->pr_mtx); | |
2324 | } | |
2325 | *prp = pr; | |
2326 | ||
2327 | return (zjp); | |
2328 | } | |
2329 | ||
2330 | /* | |
2331 | * Ensure a prison has its own ZFS info. If zjpp is non-null, point it to the | |
2332 | * ZFS info and lock the prison. | |
2333 | */ | |
2334 | static void | |
2335 | zfs_jailparam_alloc(struct prison *pr, struct zfs_jailparam **zjpp) | |
2336 | { | |
2337 | struct prison *ppr; | |
2338 | struct zfs_jailparam *zjp, *nzjp; | |
2339 | void **rsv; | |
2340 | ||
2341 | /* If this prison already has ZFS info, return that. */ | |
2342 | zjp = zfs_jailparam_find(pr, &ppr); | |
2343 | if (ppr == pr) | |
2344 | goto done; | |
2345 | ||
2346 | /* | |
2347 | * Allocate a new info record. Then check again, in case something | |
2348 | * changed during the allocation. | |
2349 | */ | |
2350 | mtx_unlock(&ppr->pr_mtx); | |
2351 | nzjp = malloc(sizeof (struct zfs_jailparam), M_PRISON, M_WAITOK); | |
2352 | rsv = osd_reserve(zfs_jailparam_slot); | |
2353 | zjp = zfs_jailparam_find(pr, &ppr); | |
2354 | if (ppr == pr) { | |
2355 | free(nzjp, M_PRISON); | |
2356 | osd_free_reserved(rsv); | |
2357 | goto done; | |
2358 | } | |
2359 | /* Inherit the initial values from the ancestor. */ | |
2360 | mtx_lock(&pr->pr_mtx); | |
2361 | (void) osd_jail_set_reserved(pr, zfs_jailparam_slot, rsv, nzjp); | |
2362 | (void) memcpy(nzjp, zjp, sizeof (*zjp)); | |
2363 | zjp = nzjp; | |
2364 | mtx_unlock(&ppr->pr_mtx); | |
2365 | done: | |
2366 | if (zjpp != NULL) | |
2367 | *zjpp = zjp; | |
2368 | else | |
2369 | mtx_unlock(&pr->pr_mtx); | |
2370 | } | |
2371 | ||
2372 | /* | |
2373 | * Jail OSD methods for ZFS VFS info. | |
2374 | */ | |
2375 | static int | |
2376 | zfs_jailparam_create(void *obj, void *data) | |
2377 | { | |
2378 | struct prison *pr = obj; | |
2379 | struct vfsoptlist *opts = data; | |
2380 | int jsys; | |
2381 | ||
2382 | if (vfs_copyopt(opts, "zfs", &jsys, sizeof (jsys)) == 0 && | |
2383 | jsys == JAIL_SYS_INHERIT) | |
2384 | return (0); | |
2385 | /* | |
2386 | * Inherit a prison's initial values from its parent | |
2387 | * (different from JAIL_SYS_INHERIT which also inherits changes). | |
2388 | */ | |
2389 | zfs_jailparam_alloc(pr, NULL); | |
2390 | return (0); | |
2391 | } | |
2392 | ||
2393 | static int | |
2394 | zfs_jailparam_get(void *obj, void *data) | |
2395 | { | |
2396 | struct prison *ppr, *pr = obj; | |
2397 | struct vfsoptlist *opts = data; | |
2398 | struct zfs_jailparam *zjp; | |
2399 | int jsys, error; | |
2400 | ||
2401 | zjp = zfs_jailparam_find(pr, &ppr); | |
2402 | jsys = (ppr == pr) ? JAIL_SYS_NEW : JAIL_SYS_INHERIT; | |
2403 | error = vfs_setopt(opts, "zfs", &jsys, sizeof (jsys)); | |
2404 | if (error != 0 && error != ENOENT) | |
2405 | goto done; | |
2406 | if (jsys == JAIL_SYS_NEW) { | |
2407 | error = vfs_setopt(opts, "zfs.mount_snapshot", | |
2408 | &zjp->mount_snapshot, sizeof (zjp->mount_snapshot)); | |
2409 | if (error != 0 && error != ENOENT) | |
2410 | goto done; | |
2411 | } else { | |
2412 | /* | |
2413 | * If this prison is inheriting its ZFS info, report | |
2414 | * empty/zero parameters. | |
2415 | */ | |
2416 | static int mount_snapshot = 0; | |
2417 | ||
2418 | error = vfs_setopt(opts, "zfs.mount_snapshot", | |
2419 | &mount_snapshot, sizeof (mount_snapshot)); | |
2420 | if (error != 0 && error != ENOENT) | |
2421 | goto done; | |
2422 | } | |
2423 | error = 0; | |
2424 | done: | |
2425 | mtx_unlock(&ppr->pr_mtx); | |
2426 | return (error); | |
2427 | } | |
2428 | ||
2429 | static int | |
2430 | zfs_jailparam_set(void *obj, void *data) | |
2431 | { | |
2432 | struct prison *pr = obj; | |
2433 | struct prison *ppr; | |
2434 | struct vfsoptlist *opts = data; | |
2435 | int error, jsys, mount_snapshot; | |
2436 | ||
2437 | /* Set the parameters, which should be correct. */ | |
2438 | error = vfs_copyopt(opts, "zfs", &jsys, sizeof (jsys)); | |
2439 | if (error == ENOENT) | |
2440 | jsys = -1; | |
2441 | error = vfs_copyopt(opts, "zfs.mount_snapshot", &mount_snapshot, | |
2442 | sizeof (mount_snapshot)); | |
2443 | if (error == ENOENT) | |
2444 | mount_snapshot = -1; | |
2445 | else | |
2446 | jsys = JAIL_SYS_NEW; | |
1d56c6d0 AJ |
2447 | switch (jsys) { |
2448 | case JAIL_SYS_NEW: | |
2449 | { | |
595d3ac2 AJ |
2450 | /* "zfs=new" or "zfs.*": the prison gets its own ZFS info. */ |
2451 | struct zfs_jailparam *zjp; | |
2452 | ||
2453 | /* | |
2454 | * A child jail cannot have more permissions than its parent | |
2455 | */ | |
2456 | if (pr->pr_parent != &prison0) { | |
2457 | zjp = zfs_jailparam_find(pr->pr_parent, &ppr); | |
2458 | mtx_unlock(&ppr->pr_mtx); | |
2459 | if (zjp->mount_snapshot < mount_snapshot) { | |
2460 | return (EPERM); | |
2461 | } | |
2462 | } | |
2463 | zfs_jailparam_alloc(pr, &zjp); | |
2464 | if (mount_snapshot != -1) | |
2465 | zjp->mount_snapshot = mount_snapshot; | |
2466 | mtx_unlock(&pr->pr_mtx); | |
1d56c6d0 AJ |
2467 | break; |
2468 | } | |
2469 | case JAIL_SYS_INHERIT: | |
595d3ac2 AJ |
2470 | /* "zfs=inherit": inherit the parent's ZFS info. */ |
2471 | mtx_lock(&pr->pr_mtx); | |
2472 | osd_jail_del(pr, zfs_jailparam_slot); | |
2473 | mtx_unlock(&pr->pr_mtx); | |
1d56c6d0 AJ |
2474 | break; |
2475 | case -1: | |
2476 | /* | |
2477 | * If the setting being changed is not ZFS related | |
2478 | * then do nothing. | |
2479 | */ | |
2480 | break; | |
595d3ac2 | 2481 | } |
1d56c6d0 | 2482 | |
595d3ac2 AJ |
2483 | return (0); |
2484 | } | |
2485 | ||
2486 | static int | |
2487 | zfs_jailparam_check(void *obj __unused, void *data) | |
2488 | { | |
2489 | struct vfsoptlist *opts = data; | |
2490 | int error, jsys, mount_snapshot; | |
2491 | ||
2492 | /* Check that the parameters are correct. */ | |
2493 | error = vfs_copyopt(opts, "zfs", &jsys, sizeof (jsys)); | |
2494 | if (error != ENOENT) { | |
2495 | if (error != 0) | |
2496 | return (error); | |
2497 | if (jsys != JAIL_SYS_NEW && jsys != JAIL_SYS_INHERIT) | |
2498 | return (EINVAL); | |
2499 | } | |
2500 | error = vfs_copyopt(opts, "zfs.mount_snapshot", &mount_snapshot, | |
2501 | sizeof (mount_snapshot)); | |
2502 | if (error != ENOENT) { | |
2503 | if (error != 0) | |
2504 | return (error); | |
2505 | if (mount_snapshot != 0 && mount_snapshot != 1) | |
2506 | return (EINVAL); | |
2507 | } | |
2508 | return (0); | |
2509 | } | |
2510 | ||
2511 | static void | |
2512 | zfs_jailparam_destroy(void *data) | |
2513 | { | |
2514 | ||
2515 | free(data, M_PRISON); | |
2516 | } | |
2517 | ||
2518 | static void | |
2519 | zfs_jailparam_sysinit(void *arg __unused) | |
2520 | { | |
2521 | struct prison *pr; | |
2522 | osd_method_t methods[PR_MAXMETHOD] = { | |
2523 | [PR_METHOD_CREATE] = zfs_jailparam_create, | |
2524 | [PR_METHOD_GET] = zfs_jailparam_get, | |
2525 | [PR_METHOD_SET] = zfs_jailparam_set, | |
2526 | [PR_METHOD_CHECK] = zfs_jailparam_check, | |
2527 | }; | |
2528 | ||
2529 | zfs_jailparam_slot = osd_jail_register(zfs_jailparam_destroy, methods); | |
2530 | /* Copy the defaults to any existing prisons. */ | |
2531 | sx_slock(&allprison_lock); | |
2532 | TAILQ_FOREACH(pr, &allprison, pr_list) | |
2533 | zfs_jailparam_alloc(pr, NULL); | |
2534 | sx_sunlock(&allprison_lock); | |
2535 | } | |
2536 | ||
2537 | static void | |
2538 | zfs_jailparam_sysuninit(void *arg __unused) | |
2539 | { | |
2540 | ||
2541 | osd_jail_deregister(zfs_jailparam_slot); | |
2542 | } | |
2543 | ||
2544 | SYSINIT(zfs_jailparam_sysinit, SI_SUB_DRIVERS, SI_ORDER_ANY, | |
2545 | zfs_jailparam_sysinit, NULL); | |
2546 | SYSUNINIT(zfs_jailparam_sysuninit, SI_SUB_DRIVERS, SI_ORDER_ANY, | |
2547 | zfs_jailparam_sysuninit, NULL); |