]>
Commit | Line | Data |
---|---|---|
87d546d8 TG |
1 | /* |
2 | * CDDL HEADER START | |
3 | * | |
4 | * The contents of this file are subject to the terms of the | |
5 | * Common Development and Distribution License (the "License"). | |
6 | * You may not use this file except in compliance with the License. | |
7 | * | |
8 | * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE | |
9 | * or http://www.opensolaris.org/os/licensing. | |
10 | * See the License for the specific language governing permissions | |
11 | * and limitations under the License. | |
12 | * | |
13 | * When distributing Covered Code, include this CDDL HEADER in each | |
14 | * file and include the License file at usr/src/OPENSOLARIS.LICENSE. | |
15 | * If applicable, add the following below this CDDL HEADER, with the | |
16 | * fields enclosed by brackets "[]" replaced with your own identifying | |
17 | * information: Portions Copyright [yyyy] [name of copyright owner] | |
18 | * | |
19 | * CDDL HEADER END | |
20 | */ | |
21 | /* | |
22 | * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. | |
23 | * Copyright (c) 2013 by Delphix. All rights reserved. | |
24 | */ | |
25 | ||
26 | /* Portions Copyright 2010 Robert Milkowski */ | |
27 | ||
28 | #include <sys/types.h> | |
29 | #include <sys/param.h> | |
30 | #include <sys/systm.h> | |
31 | #include <sys/sysmacros.h> | |
32 | #include <sys/kmem.h> | |
33 | #include <sys/pathname.h> | |
34 | #include <sys/vnode.h> | |
35 | #include <sys/vfs.h> | |
36 | #include <sys/vfs_opreg.h> | |
37 | #include <sys/mntent.h> | |
38 | #include <sys/mount.h> | |
39 | #include <sys/cmn_err.h> | |
40 | #include "fs/fs_subr.h" | |
41 | #include <sys/zfs_znode.h> | |
42 | #include <sys/zfs_vnops.h> | |
43 | #include <sys/zfs_dir.h> | |
44 | #include <sys/zil.h> | |
45 | #include <sys/fs/zfs.h> | |
46 | #include <sys/dmu.h> | |
47 | #include <sys/dsl_prop.h> | |
48 | #include <sys/dsl_dataset.h> | |
49 | #include <sys/dsl_deleg.h> | |
50 | #include <sys/spa.h> | |
51 | #include <sys/zap.h> | |
52 | #include <sys/sa.h> | |
53 | #include <sys/sa_impl.h> | |
54 | #include <sys/varargs.h> | |
55 | #include <sys/policy.h> | |
56 | #include <sys/atomic.h> | |
57 | #include <sys/mkdev.h> | |
58 | #include <sys/modctl.h> | |
59 | #include <sys/refstr.h> | |
60 | #include <sys/zfs_ioctl.h> | |
61 | #include <sys/zfs_ctldir.h> | |
62 | #include <sys/zfs_fuid.h> | |
63 | #include <sys/bootconf.h> | |
64 | #include <sys/sunddi.h> | |
65 | #include <sys/dnlc.h> | |
66 | #include <sys/dmu_objset.h> | |
67 | #include <sys/spa_boot.h> | |
68 | #include <sys/zpl.h> | |
69 | #include "zfs_comutil.h" | |
70 | ||
71 | /*ARGSUSED*/ | |
72 | int | |
73 | zfs_sync(struct super_block *sb, int wait, cred_t *cr) | |
74 | { | |
75 | zfs_sb_t *zsb = sb->s_fs_info; | |
76 | ||
77 | /* | |
78 | * Data integrity is job one. We don't want a compromised kernel | |
79 | * writing to the storage pool, so we never sync during panic. | |
80 | */ | |
81 | if (unlikely(oops_in_progress)) | |
82 | return (0); | |
83 | ||
84 | /* | |
85 | * Semantically, the only requirement is that the sync be initiated. | |
86 | * The DMU syncs out txgs frequently, so there's nothing to do. | |
87 | */ | |
88 | if (!wait) | |
89 | return (0); | |
90 | ||
91 | if (zsb != NULL) { | |
92 | /* | |
93 | * Sync a specific filesystem. | |
94 | */ | |
95 | dsl_pool_t *dp; | |
96 | ||
97 | ZFS_ENTER(zsb); | |
98 | dp = dmu_objset_pool(zsb->z_os); | |
99 | ||
100 | /* | |
101 | * If the system is shutting down, then skip any | |
102 | * filesystems which may exist on a suspended pool. | |
103 | */ | |
104 | if (spa_suspended(dp->dp_spa)) { | |
105 | ZFS_EXIT(zsb); | |
106 | return (0); | |
107 | } | |
108 | ||
109 | if (zsb->z_log != NULL) | |
110 | zil_commit(zsb->z_log, 0); | |
111 | ||
112 | ZFS_EXIT(zsb); | |
113 | } else { | |
114 | /* | |
115 | * Sync all ZFS filesystems. This is what happens when you | |
116 | * run sync(1M). Unlike other filesystems, ZFS honors the | |
117 | * request by waiting for all pools to commit all dirty data. | |
118 | */ | |
119 | spa_sync_allpools(); | |
120 | } | |
121 | ||
122 | return (0); | |
123 | } | |
124 | EXPORT_SYMBOL(zfs_sync); | |
125 | ||
126 | boolean_t | |
127 | zfs_is_readonly(zfs_sb_t *zsb) | |
128 | { | |
129 | return (!!(zsb->z_sb->s_flags & MS_RDONLY)); | |
130 | } | |
131 | EXPORT_SYMBOL(zfs_is_readonly); | |
132 | ||
133 | static void | |
134 | atime_changed_cb(void *arg, uint64_t newval) | |
135 | { | |
136 | ((zfs_sb_t *)arg)->z_atime = newval; | |
137 | } | |
138 | ||
139 | static void | |
140 | relatime_changed_cb(void *arg, uint64_t newval) | |
141 | { | |
142 | ((zfs_sb_t *)arg)->z_relatime = newval; | |
143 | } | |
144 | ||
145 | static void | |
146 | xattr_changed_cb(void *arg, uint64_t newval) | |
147 | { | |
148 | zfs_sb_t *zsb = arg; | |
149 | ||
150 | if (newval == ZFS_XATTR_OFF) { | |
151 | zsb->z_flags &= ~ZSB_XATTR; | |
152 | } else { | |
153 | zsb->z_flags |= ZSB_XATTR; | |
154 | ||
155 | if (newval == ZFS_XATTR_SA) | |
156 | zsb->z_xattr_sa = B_TRUE; | |
157 | else | |
158 | zsb->z_xattr_sa = B_FALSE; | |
159 | } | |
160 | } | |
161 | ||
162 | static void | |
163 | acltype_changed_cb(void *arg, uint64_t newval) | |
164 | { | |
165 | zfs_sb_t *zsb = arg; | |
166 | ||
167 | switch (newval) { | |
168 | case ZFS_ACLTYPE_OFF: | |
169 | zsb->z_acl_type = ZFS_ACLTYPE_OFF; | |
170 | zsb->z_sb->s_flags &= ~MS_POSIXACL; | |
171 | break; | |
172 | case ZFS_ACLTYPE_POSIXACL: | |
173 | #ifdef CONFIG_FS_POSIX_ACL | |
174 | zsb->z_acl_type = ZFS_ACLTYPE_POSIXACL; | |
175 | zsb->z_sb->s_flags |= MS_POSIXACL; | |
176 | #else | |
177 | zsb->z_acl_type = ZFS_ACLTYPE_OFF; | |
178 | zsb->z_sb->s_flags &= ~MS_POSIXACL; | |
179 | #endif /* CONFIG_FS_POSIX_ACL */ | |
180 | break; | |
181 | default: | |
182 | break; | |
183 | } | |
184 | } | |
185 | ||
186 | static void | |
187 | blksz_changed_cb(void *arg, uint64_t newval) | |
188 | { | |
189 | zfs_sb_t *zsb = arg; | |
190 | ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zsb->z_os))); | |
191 | ASSERT3U(newval, >=, SPA_MINBLOCKSIZE); | |
192 | ASSERT(ISP2(newval)); | |
193 | ||
194 | zsb->z_max_blksz = newval; | |
195 | } | |
196 | ||
197 | static void | |
198 | readonly_changed_cb(void *arg, uint64_t newval) | |
199 | { | |
200 | zfs_sb_t *zsb = arg; | |
201 | struct super_block *sb = zsb->z_sb; | |
202 | ||
203 | if (sb == NULL) | |
204 | return; | |
205 | ||
206 | if (newval) | |
207 | sb->s_flags |= MS_RDONLY; | |
208 | else | |
209 | sb->s_flags &= ~MS_RDONLY; | |
210 | } | |
211 | ||
212 | static void | |
213 | devices_changed_cb(void *arg, uint64_t newval) | |
214 | { | |
215 | } | |
216 | ||
217 | static void | |
218 | setuid_changed_cb(void *arg, uint64_t newval) | |
219 | { | |
220 | } | |
221 | ||
222 | static void | |
223 | exec_changed_cb(void *arg, uint64_t newval) | |
224 | { | |
225 | } | |
226 | ||
227 | static void | |
228 | nbmand_changed_cb(void *arg, uint64_t newval) | |
229 | { | |
230 | zfs_sb_t *zsb = arg; | |
231 | struct super_block *sb = zsb->z_sb; | |
232 | ||
233 | if (sb == NULL) | |
234 | return; | |
235 | ||
236 | if (newval == TRUE) | |
237 | sb->s_flags |= MS_MANDLOCK; | |
238 | else | |
239 | sb->s_flags &= ~MS_MANDLOCK; | |
240 | } | |
241 | ||
242 | static void | |
243 | snapdir_changed_cb(void *arg, uint64_t newval) | |
244 | { | |
245 | ((zfs_sb_t *)arg)->z_show_ctldir = newval; | |
246 | } | |
247 | ||
248 | static void | |
249 | vscan_changed_cb(void *arg, uint64_t newval) | |
250 | { | |
251 | ((zfs_sb_t *)arg)->z_vscan = newval; | |
252 | } | |
253 | ||
254 | static void | |
255 | acl_inherit_changed_cb(void *arg, uint64_t newval) | |
256 | { | |
257 | ((zfs_sb_t *)arg)->z_acl_inherit = newval; | |
258 | } | |
259 | ||
260 | int | |
261 | zfs_register_callbacks(zfs_sb_t *zsb) | |
262 | { | |
263 | struct dsl_dataset *ds = NULL; | |
264 | objset_t *os = zsb->z_os; | |
265 | zfs_mntopts_t *zmo = zsb->z_mntopts; | |
266 | int error = 0; | |
267 | ||
268 | ASSERT(zsb); | |
269 | ASSERT(zmo); | |
270 | ||
271 | /* | |
272 | * The act of registering our callbacks will destroy any mount | |
273 | * options we may have. In order to enable temporary overrides | |
274 | * of mount options, we stash away the current values and | |
275 | * restore them after we register the callbacks. | |
276 | */ | |
277 | if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os))) { | |
278 | zmo->z_do_readonly = B_TRUE; | |
279 | zmo->z_readonly = B_TRUE; | |
280 | } | |
281 | ||
282 | /* | |
283 | * Register property callbacks. | |
284 | * | |
285 | * It would probably be fine to just check for i/o error from | |
286 | * the first prop_register(), but I guess I like to go | |
287 | * overboard... | |
288 | */ | |
289 | ds = dmu_objset_ds(os); | |
290 | dsl_pool_config_enter(dmu_objset_pool(os), FTAG); | |
291 | error = dsl_prop_register(ds, | |
292 | zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb); | |
293 | error = error ? error : dsl_prop_register(ds, | |
294 | zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zsb); | |
295 | error = error ? error : dsl_prop_register(ds, | |
296 | zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb); | |
297 | error = error ? error : dsl_prop_register(ds, | |
298 | zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb); | |
299 | error = error ? error : dsl_prop_register(ds, | |
300 | zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb); | |
301 | error = error ? error : dsl_prop_register(ds, | |
302 | zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb); | |
303 | error = error ? error : dsl_prop_register(ds, | |
304 | zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb); | |
305 | error = error ? error : dsl_prop_register(ds, | |
306 | zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb); | |
307 | error = error ? error : dsl_prop_register(ds, | |
308 | zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb); | |
309 | error = error ? error : dsl_prop_register(ds, | |
310 | zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zsb); | |
311 | error = error ? error : dsl_prop_register(ds, | |
312 | zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zsb); | |
313 | error = error ? error : dsl_prop_register(ds, | |
314 | zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb); | |
315 | error = error ? error : dsl_prop_register(ds, | |
316 | zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb); | |
317 | dsl_pool_config_exit(dmu_objset_pool(os), FTAG); | |
318 | if (error) | |
319 | goto unregister; | |
320 | ||
321 | /* | |
322 | * Invoke our callbacks to restore temporary mount options. | |
323 | */ | |
324 | if (zmo->z_do_readonly) | |
325 | readonly_changed_cb(zsb, zmo->z_readonly); | |
326 | if (zmo->z_do_setuid) | |
327 | setuid_changed_cb(zsb, zmo->z_setuid); | |
328 | if (zmo->z_do_exec) | |
329 | exec_changed_cb(zsb, zmo->z_exec); | |
330 | if (zmo->z_do_devices) | |
331 | devices_changed_cb(zsb, zmo->z_devices); | |
332 | if (zmo->z_do_xattr) | |
333 | xattr_changed_cb(zsb, zmo->z_xattr); | |
334 | if (zmo->z_do_atime) | |
335 | atime_changed_cb(zsb, zmo->z_atime); | |
336 | if (zmo->z_do_relatime) | |
337 | relatime_changed_cb(zsb, zmo->z_relatime); | |
338 | if (zmo->z_do_nbmand) | |
339 | nbmand_changed_cb(zsb, zmo->z_nbmand); | |
340 | ||
341 | return (0); | |
342 | ||
343 | unregister: | |
344 | /* | |
345 | * We may attempt to unregister some callbacks that are not | |
346 | * registered, but this is OK; it will simply return ENOMSG, | |
347 | * which we will ignore. | |
348 | */ | |
349 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME), | |
350 | atime_changed_cb, zsb); | |
351 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RELATIME), | |
352 | relatime_changed_cb, zsb); | |
353 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR), | |
354 | xattr_changed_cb, zsb); | |
355 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE), | |
356 | blksz_changed_cb, zsb); | |
357 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY), | |
358 | readonly_changed_cb, zsb); | |
359 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES), | |
360 | devices_changed_cb, zsb); | |
361 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID), | |
362 | setuid_changed_cb, zsb); | |
363 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC), | |
364 | exec_changed_cb, zsb); | |
365 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR), | |
366 | snapdir_changed_cb, zsb); | |
367 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLTYPE), | |
368 | acltype_changed_cb, zsb); | |
369 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT), | |
370 | acl_inherit_changed_cb, zsb); | |
371 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN), | |
372 | vscan_changed_cb, zsb); | |
373 | (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND), | |
374 | nbmand_changed_cb, zsb); | |
375 | ||
376 | return (error); | |
377 | } | |
378 | EXPORT_SYMBOL(zfs_register_callbacks); | |
379 | ||
380 | static int | |
381 | zfs_space_delta_cb(dmu_object_type_t bonustype, void *data, | |
382 | uint64_t *userp, uint64_t *groupp) | |
383 | { | |
384 | /* | |
385 | * Is it a valid type of object to track? | |
386 | */ | |
387 | if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA) | |
388 | return (SET_ERROR(ENOENT)); | |
389 | ||
390 | /* | |
391 | * If we have a NULL data pointer | |
392 | * then assume the id's aren't changing and | |
393 | * return EEXIST to the dmu to let it know to | |
394 | * use the same ids | |
395 | */ | |
396 | if (data == NULL) | |
397 | return (SET_ERROR(EEXIST)); | |
398 | ||
399 | if (bonustype == DMU_OT_ZNODE) { | |
400 | znode_phys_t *znp = data; | |
401 | *userp = znp->zp_uid; | |
402 | *groupp = znp->zp_gid; | |
403 | } else { | |
404 | int hdrsize; | |
405 | sa_hdr_phys_t *sap = data; | |
406 | sa_hdr_phys_t sa = *sap; | |
407 | boolean_t swap = B_FALSE; | |
408 | ||
409 | ASSERT(bonustype == DMU_OT_SA); | |
410 | ||
411 | if (sa.sa_magic == 0) { | |
412 | /* | |
413 | * This should only happen for newly created | |
414 | * files that haven't had the znode data filled | |
415 | * in yet. | |
416 | */ | |
417 | *userp = 0; | |
418 | *groupp = 0; | |
419 | return (0); | |
420 | } | |
421 | if (sa.sa_magic == BSWAP_32(SA_MAGIC)) { | |
422 | sa.sa_magic = SA_MAGIC; | |
423 | sa.sa_layout_info = BSWAP_16(sa.sa_layout_info); | |
424 | swap = B_TRUE; | |
425 | } else { | |
426 | VERIFY3U(sa.sa_magic, ==, SA_MAGIC); | |
427 | } | |
428 | ||
429 | hdrsize = sa_hdrsize(&sa); | |
430 | VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t)); | |
431 | *userp = *((uint64_t *)((uintptr_t)data + hdrsize + | |
432 | SA_UID_OFFSET)); | |
433 | *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + | |
434 | SA_GID_OFFSET)); | |
435 | if (swap) { | |
436 | *userp = BSWAP_64(*userp); | |
437 | *groupp = BSWAP_64(*groupp); | |
438 | } | |
439 | } | |
440 | return (0); | |
441 | } | |
442 | ||
443 | static void | |
444 | fuidstr_to_sid(zfs_sb_t *zsb, const char *fuidstr, | |
445 | char *domainbuf, int buflen, uid_t *ridp) | |
446 | { | |
447 | uint64_t fuid; | |
448 | const char *domain; | |
449 | ||
450 | fuid = strtonum(fuidstr, NULL); | |
451 | ||
452 | domain = zfs_fuid_find_by_idx(zsb, FUID_INDEX(fuid)); | |
453 | if (domain) | |
454 | (void) strlcpy(domainbuf, domain, buflen); | |
455 | else | |
456 | domainbuf[0] = '\0'; | |
457 | *ridp = FUID_RID(fuid); | |
458 | } | |
459 | ||
460 | static uint64_t | |
461 | zfs_userquota_prop_to_obj(zfs_sb_t *zsb, zfs_userquota_prop_t type) | |
462 | { | |
463 | switch (type) { | |
464 | case ZFS_PROP_USERUSED: | |
465 | return (DMU_USERUSED_OBJECT); | |
466 | case ZFS_PROP_GROUPUSED: | |
467 | return (DMU_GROUPUSED_OBJECT); | |
468 | case ZFS_PROP_USERQUOTA: | |
469 | return (zsb->z_userquota_obj); | |
470 | case ZFS_PROP_GROUPQUOTA: | |
471 | return (zsb->z_groupquota_obj); | |
472 | default: | |
473 | return (SET_ERROR(ENOTSUP)); | |
474 | } | |
475 | return (0); | |
476 | } | |
477 | ||
478 | int | |
479 | zfs_userspace_many(zfs_sb_t *zsb, zfs_userquota_prop_t type, | |
480 | uint64_t *cookiep, void *vbuf, uint64_t *bufsizep) | |
481 | { | |
482 | int error; | |
483 | zap_cursor_t zc; | |
484 | zap_attribute_t za; | |
485 | zfs_useracct_t *buf = vbuf; | |
486 | uint64_t obj; | |
487 | ||
488 | if (!dmu_objset_userspace_present(zsb->z_os)) | |
489 | return (SET_ERROR(ENOTSUP)); | |
490 | ||
491 | obj = zfs_userquota_prop_to_obj(zsb, type); | |
492 | if (obj == 0) { | |
493 | *bufsizep = 0; | |
494 | return (0); | |
495 | } | |
496 | ||
497 | for (zap_cursor_init_serialized(&zc, zsb->z_os, obj, *cookiep); | |
498 | (error = zap_cursor_retrieve(&zc, &za)) == 0; | |
499 | zap_cursor_advance(&zc)) { | |
500 | if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) > | |
501 | *bufsizep) | |
502 | break; | |
503 | ||
504 | fuidstr_to_sid(zsb, za.za_name, | |
505 | buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid); | |
506 | ||
507 | buf->zu_space = za.za_first_integer; | |
508 | buf++; | |
509 | } | |
510 | if (error == ENOENT) | |
511 | error = 0; | |
512 | ||
513 | ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep); | |
514 | *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf; | |
515 | *cookiep = zap_cursor_serialize(&zc); | |
516 | zap_cursor_fini(&zc); | |
517 | return (error); | |
518 | } | |
519 | EXPORT_SYMBOL(zfs_userspace_many); | |
520 | ||
521 | /* | |
522 | * buf must be big enough (eg, 32 bytes) | |
523 | */ | |
524 | static int | |
525 | id_to_fuidstr(zfs_sb_t *zsb, const char *domain, uid_t rid, | |
526 | char *buf, boolean_t addok) | |
527 | { | |
528 | uint64_t fuid; | |
529 | int domainid = 0; | |
530 | ||
531 | if (domain && domain[0]) { | |
532 | domainid = zfs_fuid_find_by_domain(zsb, domain, NULL, addok); | |
533 | if (domainid == -1) | |
534 | return (SET_ERROR(ENOENT)); | |
535 | } | |
536 | fuid = FUID_ENCODE(domainid, rid); | |
537 | (void) sprintf(buf, "%llx", (longlong_t)fuid); | |
538 | return (0); | |
539 | } | |
540 | ||
541 | int | |
542 | zfs_userspace_one(zfs_sb_t *zsb, zfs_userquota_prop_t type, | |
543 | const char *domain, uint64_t rid, uint64_t *valp) | |
544 | { | |
545 | char buf[32]; | |
546 | int err; | |
547 | uint64_t obj; | |
548 | ||
549 | *valp = 0; | |
550 | ||
551 | if (!dmu_objset_userspace_present(zsb->z_os)) | |
552 | return (SET_ERROR(ENOTSUP)); | |
553 | ||
554 | obj = zfs_userquota_prop_to_obj(zsb, type); | |
555 | if (obj == 0) | |
556 | return (0); | |
557 | ||
558 | err = id_to_fuidstr(zsb, domain, rid, buf, B_FALSE); | |
559 | if (err) | |
560 | return (err); | |
561 | ||
562 | err = zap_lookup(zsb->z_os, obj, buf, 8, 1, valp); | |
563 | if (err == ENOENT) | |
564 | err = 0; | |
565 | return (err); | |
566 | } | |
567 | EXPORT_SYMBOL(zfs_userspace_one); | |
568 | ||
569 | int | |
570 | zfs_set_userquota(zfs_sb_t *zsb, zfs_userquota_prop_t type, | |
571 | const char *domain, uint64_t rid, uint64_t quota) | |
572 | { | |
573 | char buf[32]; | |
574 | int err; | |
575 | dmu_tx_t *tx; | |
576 | uint64_t *objp; | |
577 | boolean_t fuid_dirtied; | |
578 | ||
579 | if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA) | |
580 | return (SET_ERROR(EINVAL)); | |
581 | ||
582 | if (zsb->z_version < ZPL_VERSION_USERSPACE) | |
583 | return (SET_ERROR(ENOTSUP)); | |
584 | ||
585 | objp = (type == ZFS_PROP_USERQUOTA) ? &zsb->z_userquota_obj : | |
586 | &zsb->z_groupquota_obj; | |
587 | ||
588 | err = id_to_fuidstr(zsb, domain, rid, buf, B_TRUE); | |
589 | if (err) | |
590 | return (err); | |
591 | fuid_dirtied = zsb->z_fuid_dirty; | |
592 | ||
593 | tx = dmu_tx_create(zsb->z_os); | |
594 | dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL); | |
595 | if (*objp == 0) { | |
596 | dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, | |
597 | zfs_userquota_prop_prefixes[type]); | |
598 | } | |
599 | if (fuid_dirtied) | |
600 | zfs_fuid_txhold(zsb, tx); | |
601 | err = dmu_tx_assign(tx, TXG_WAIT); | |
602 | if (err) { | |
603 | dmu_tx_abort(tx); | |
604 | return (err); | |
605 | } | |
606 | ||
607 | mutex_enter(&zsb->z_lock); | |
608 | if (*objp == 0) { | |
609 | *objp = zap_create(zsb->z_os, DMU_OT_USERGROUP_QUOTA, | |
610 | DMU_OT_NONE, 0, tx); | |
611 | VERIFY(0 == zap_add(zsb->z_os, MASTER_NODE_OBJ, | |
612 | zfs_userquota_prop_prefixes[type], 8, 1, objp, tx)); | |
613 | } | |
614 | mutex_exit(&zsb->z_lock); | |
615 | ||
616 | if (quota == 0) { | |
617 | err = zap_remove(zsb->z_os, *objp, buf, tx); | |
618 | if (err == ENOENT) | |
619 | err = 0; | |
620 | } else { | |
621 | err = zap_update(zsb->z_os, *objp, buf, 8, 1, "a, tx); | |
622 | } | |
623 | ASSERT(err == 0); | |
624 | if (fuid_dirtied) | |
625 | zfs_fuid_sync(zsb, tx); | |
626 | dmu_tx_commit(tx); | |
627 | return (err); | |
628 | } | |
629 | EXPORT_SYMBOL(zfs_set_userquota); | |
630 | ||
631 | boolean_t | |
632 | zfs_fuid_overquota(zfs_sb_t *zsb, boolean_t isgroup, uint64_t fuid) | |
633 | { | |
634 | char buf[32]; | |
635 | uint64_t used, quota, usedobj, quotaobj; | |
636 | int err; | |
637 | ||
638 | usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT; | |
639 | quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj; | |
640 | ||
641 | if (quotaobj == 0 || zsb->z_replay) | |
642 | return (B_FALSE); | |
643 | ||
644 | (void) sprintf(buf, "%llx", (longlong_t)fuid); | |
645 | err = zap_lookup(zsb->z_os, quotaobj, buf, 8, 1, "a); | |
646 | if (err != 0) | |
647 | return (B_FALSE); | |
648 | ||
649 | err = zap_lookup(zsb->z_os, usedobj, buf, 8, 1, &used); | |
650 | if (err != 0) | |
651 | return (B_FALSE); | |
652 | return (used >= quota); | |
653 | } | |
654 | EXPORT_SYMBOL(zfs_fuid_overquota); | |
655 | ||
656 | boolean_t | |
657 | zfs_owner_overquota(zfs_sb_t *zsb, znode_t *zp, boolean_t isgroup) | |
658 | { | |
659 | uint64_t fuid; | |
660 | uint64_t quotaobj; | |
661 | ||
662 | quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj; | |
663 | ||
664 | fuid = isgroup ? zp->z_gid : zp->z_uid; | |
665 | ||
666 | if (quotaobj == 0 || zsb->z_replay) | |
667 | return (B_FALSE); | |
668 | ||
669 | return (zfs_fuid_overquota(zsb, isgroup, fuid)); | |
670 | } | |
671 | EXPORT_SYMBOL(zfs_owner_overquota); | |
672 | ||
673 | zfs_mntopts_t * | |
674 | zfs_mntopts_alloc(void) | |
675 | { | |
676 | return (kmem_zalloc(sizeof (zfs_mntopts_t), KM_SLEEP)); | |
677 | } | |
678 | ||
679 | void | |
680 | zfs_mntopts_free(zfs_mntopts_t *zmo) | |
681 | { | |
682 | if (zmo->z_osname) | |
683 | strfree(zmo->z_osname); | |
684 | ||
685 | if (zmo->z_mntpoint) | |
686 | strfree(zmo->z_mntpoint); | |
687 | ||
688 | kmem_free(zmo, sizeof (zfs_mntopts_t)); | |
689 | } | |
690 | ||
691 | int | |
692 | zfs_sb_create(const char *osname, zfs_mntopts_t *zmo, zfs_sb_t **zsbp) | |
693 | { | |
694 | objset_t *os; | |
695 | zfs_sb_t *zsb; | |
696 | uint64_t zval; | |
697 | int i, size, error; | |
698 | uint64_t sa_obj; | |
699 | ||
700 | zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP); | |
701 | ||
702 | /* | |
703 | * Optional temporary mount options, free'd in zfs_sb_free(). | |
704 | */ | |
705 | zsb->z_mntopts = (zmo ? zmo : zfs_mntopts_alloc()); | |
706 | ||
707 | /* | |
708 | * We claim to always be readonly so we can open snapshots; | |
709 | * other ZPL code will prevent us from writing to snapshots. | |
710 | */ | |
711 | error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zsb, &os); | |
712 | if (error) | |
713 | goto out_zmo; | |
714 | ||
715 | /* | |
716 | * Initialize the zfs-specific filesystem structure. | |
717 | * Should probably make this a kmem cache, shuffle fields. | |
718 | */ | |
719 | zsb->z_sb = NULL; | |
720 | zsb->z_parent = zsb; | |
721 | zsb->z_max_blksz = SPA_OLD_MAXBLOCKSIZE; | |
722 | zsb->z_show_ctldir = ZFS_SNAPDIR_VISIBLE; | |
723 | zsb->z_os = os; | |
724 | ||
725 | error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zsb->z_version); | |
726 | if (error) { | |
727 | goto out; | |
728 | } else if (zsb->z_version > ZPL_VERSION) { | |
729 | error = SET_ERROR(ENOTSUP); | |
730 | goto out; | |
731 | } | |
732 | if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0) | |
733 | goto out; | |
734 | zsb->z_norm = (int)zval; | |
735 | ||
736 | if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0) | |
737 | goto out; | |
738 | zsb->z_utf8 = (zval != 0); | |
739 | ||
740 | if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0) | |
741 | goto out; | |
742 | zsb->z_case = (uint_t)zval; | |
743 | ||
744 | if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &zval)) != 0) | |
745 | goto out; | |
746 | zsb->z_acl_type = (uint_t)zval; | |
747 | ||
748 | /* | |
749 | * Fold case on file systems that are always or sometimes case | |
750 | * insensitive. | |
751 | */ | |
752 | if (zsb->z_case == ZFS_CASE_INSENSITIVE || | |
753 | zsb->z_case == ZFS_CASE_MIXED) | |
754 | zsb->z_norm |= U8_TEXTPREP_TOUPPER; | |
755 | ||
756 | zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os); | |
757 | zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os); | |
758 | ||
759 | if (zsb->z_use_sa) { | |
760 | /* should either have both of these objects or none */ | |
761 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1, | |
762 | &sa_obj); | |
763 | if (error) | |
764 | goto out; | |
765 | ||
766 | error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &zval); | |
767 | if ((error == 0) && (zval == ZFS_XATTR_SA)) | |
768 | zsb->z_xattr_sa = B_TRUE; | |
769 | } else { | |
770 | /* | |
771 | * Pre SA versions file systems should never touch | |
772 | * either the attribute registration or layout objects. | |
773 | */ | |
774 | sa_obj = 0; | |
775 | } | |
776 | ||
777 | error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END, | |
778 | &zsb->z_attr_table); | |
779 | if (error) | |
780 | goto out; | |
781 | ||
782 | if (zsb->z_version >= ZPL_VERSION_SA) | |
783 | sa_register_update_callback(os, zfs_sa_upgrade); | |
784 | ||
785 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, | |
786 | &zsb->z_root); | |
787 | if (error) | |
788 | goto out; | |
789 | ASSERT(zsb->z_root != 0); | |
790 | ||
791 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1, | |
792 | &zsb->z_unlinkedobj); | |
793 | if (error) | |
794 | goto out; | |
795 | ||
796 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
797 | zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA], | |
798 | 8, 1, &zsb->z_userquota_obj); | |
799 | if (error && error != ENOENT) | |
800 | goto out; | |
801 | ||
802 | error = zap_lookup(os, MASTER_NODE_OBJ, | |
803 | zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA], | |
804 | 8, 1, &zsb->z_groupquota_obj); | |
805 | if (error && error != ENOENT) | |
806 | goto out; | |
807 | ||
808 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1, | |
809 | &zsb->z_fuid_obj); | |
810 | if (error && error != ENOENT) | |
811 | goto out; | |
812 | ||
813 | error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1, | |
814 | &zsb->z_shares_dir); | |
815 | if (error && error != ENOENT) | |
816 | goto out; | |
817 | ||
818 | mutex_init(&zsb->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL); | |
819 | mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL); | |
820 | list_create(&zsb->z_all_znodes, sizeof (znode_t), | |
821 | offsetof(znode_t, z_link_node)); | |
822 | rrm_init(&zsb->z_teardown_lock, B_FALSE); | |
823 | rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL); | |
824 | rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL); | |
825 | ||
826 | size = MIN(1 << (highbit64(zfs_object_mutex_size)-1), ZFS_OBJ_MTX_MAX); | |
827 | zsb->z_hold_size = size; | |
828 | zsb->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size, KM_SLEEP); | |
829 | zsb->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP); | |
830 | for (i = 0; i != size; i++) { | |
831 | avl_create(&zsb->z_hold_trees[i], zfs_znode_hold_compare, | |
832 | sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node)); | |
833 | mutex_init(&zsb->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL); | |
834 | } | |
835 | ||
836 | *zsbp = zsb; | |
837 | return (0); | |
838 | ||
839 | out: | |
840 | dmu_objset_disown(os, zsb); | |
841 | out_zmo: | |
842 | *zsbp = NULL; | |
843 | zfs_mntopts_free(zsb->z_mntopts); | |
844 | kmem_free(zsb, sizeof (zfs_sb_t)); | |
845 | return (error); | |
846 | } | |
847 | EXPORT_SYMBOL(zfs_sb_create); | |
848 | ||
849 | int | |
850 | zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting) | |
851 | { | |
852 | int error; | |
853 | ||
854 | error = zfs_register_callbacks(zsb); | |
855 | if (error) | |
856 | return (error); | |
857 | ||
858 | /* | |
859 | * Set the objset user_ptr to track its zsb. | |
860 | */ | |
861 | mutex_enter(&zsb->z_os->os_user_ptr_lock); | |
862 | dmu_objset_set_user(zsb->z_os, zsb); | |
863 | mutex_exit(&zsb->z_os->os_user_ptr_lock); | |
864 | ||
865 | zsb->z_log = zil_open(zsb->z_os, zfs_get_data); | |
866 | ||
867 | /* | |
868 | * If we are not mounting (ie: online recv), then we don't | |
869 | * have to worry about replaying the log as we blocked all | |
870 | * operations out since we closed the ZIL. | |
871 | */ | |
872 | if (mounting) { | |
873 | boolean_t readonly; | |
874 | ||
875 | /* | |
876 | * During replay we remove the read only flag to | |
877 | * allow replays to succeed. | |
878 | */ | |
879 | readonly = zfs_is_readonly(zsb); | |
880 | if (readonly != 0) | |
881 | readonly_changed_cb(zsb, B_FALSE); | |
882 | else | |
883 | zfs_unlinked_drain(zsb); | |
884 | ||
885 | /* | |
886 | * Parse and replay the intent log. | |
887 | * | |
888 | * Because of ziltest, this must be done after | |
889 | * zfs_unlinked_drain(). (Further note: ziltest | |
890 | * doesn't use readonly mounts, where | |
891 | * zfs_unlinked_drain() isn't called.) This is because | |
892 | * ziltest causes spa_sync() to think it's committed, | |
893 | * but actually it is not, so the intent log contains | |
894 | * many txg's worth of changes. | |
895 | * | |
896 | * In particular, if object N is in the unlinked set in | |
897 | * the last txg to actually sync, then it could be | |
898 | * actually freed in a later txg and then reallocated | |
899 | * in a yet later txg. This would write a "create | |
900 | * object N" record to the intent log. Normally, this | |
901 | * would be fine because the spa_sync() would have | |
902 | * written out the fact that object N is free, before | |
903 | * we could write the "create object N" intent log | |
904 | * record. | |
905 | * | |
906 | * But when we are in ziltest mode, we advance the "open | |
907 | * txg" without actually spa_sync()-ing the changes to | |
908 | * disk. So we would see that object N is still | |
909 | * allocated and in the unlinked set, and there is an | |
910 | * intent log record saying to allocate it. | |
911 | */ | |
912 | if (spa_writeable(dmu_objset_spa(zsb->z_os))) { | |
913 | if (zil_replay_disable) { | |
914 | zil_destroy(zsb->z_log, B_FALSE); | |
915 | } else { | |
916 | zsb->z_replay = B_TRUE; | |
917 | zil_replay(zsb->z_os, zsb, | |
918 | zfs_replay_vector); | |
919 | zsb->z_replay = B_FALSE; | |
920 | } | |
921 | } | |
922 | ||
923 | /* restore readonly bit */ | |
924 | if (readonly != 0) | |
925 | readonly_changed_cb(zsb, B_TRUE); | |
926 | } | |
927 | ||
928 | return (0); | |
929 | } | |
930 | EXPORT_SYMBOL(zfs_sb_setup); | |
931 | ||
932 | void | |
933 | zfs_sb_free(zfs_sb_t *zsb) | |
934 | { | |
935 | int i, size = zsb->z_hold_size; | |
936 | ||
937 | zfs_fuid_destroy(zsb); | |
938 | ||
939 | mutex_destroy(&zsb->z_znodes_lock); | |
940 | mutex_destroy(&zsb->z_lock); | |
941 | list_destroy(&zsb->z_all_znodes); | |
942 | rrm_destroy(&zsb->z_teardown_lock); | |
943 | rw_destroy(&zsb->z_teardown_inactive_lock); | |
944 | rw_destroy(&zsb->z_fuid_lock); | |
945 | for (i = 0; i != size; i++) { | |
946 | avl_destroy(&zsb->z_hold_trees[i]); | |
947 | mutex_destroy(&zsb->z_hold_locks[i]); | |
948 | } | |
949 | vmem_free(zsb->z_hold_trees, sizeof (avl_tree_t) * size); | |
950 | vmem_free(zsb->z_hold_locks, sizeof (kmutex_t) * size); | |
951 | zfs_mntopts_free(zsb->z_mntopts); | |
952 | kmem_free(zsb, sizeof (zfs_sb_t)); | |
953 | } | |
954 | EXPORT_SYMBOL(zfs_sb_free); | |
955 | ||
956 | static void | |
957 | zfs_set_fuid_feature(zfs_sb_t *zsb) | |
958 | { | |
959 | zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os); | |
960 | zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os); | |
961 | } | |
962 | ||
963 | void | |
964 | zfs_unregister_callbacks(zfs_sb_t *zsb) | |
965 | { | |
966 | objset_t *os = zsb->z_os; | |
967 | struct dsl_dataset *ds; | |
968 | ||
969 | /* | |
970 | * Unregister properties. | |
971 | */ | |
972 | if (!dmu_objset_is_snapshot(os)) { | |
973 | ds = dmu_objset_ds(os); | |
974 | VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb, | |
975 | zsb) == 0); | |
976 | ||
977 | VERIFY(dsl_prop_unregister(ds, "relatime", relatime_changed_cb, | |
978 | zsb) == 0); | |
979 | ||
980 | VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb, | |
981 | zsb) == 0); | |
982 | ||
983 | VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, | |
984 | zsb) == 0); | |
985 | ||
986 | VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb, | |
987 | zsb) == 0); | |
988 | ||
989 | VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb, | |
990 | zsb) == 0); | |
991 | ||
992 | VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb, | |
993 | zsb) == 0); | |
994 | ||
995 | VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb, | |
996 | zsb) == 0); | |
997 | ||
998 | VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, | |
999 | zsb) == 0); | |
1000 | ||
1001 | VERIFY(dsl_prop_unregister(ds, "acltype", acltype_changed_cb, | |
1002 | zsb) == 0); | |
1003 | ||
1004 | VERIFY(dsl_prop_unregister(ds, "aclinherit", | |
1005 | acl_inherit_changed_cb, zsb) == 0); | |
1006 | ||
1007 | VERIFY(dsl_prop_unregister(ds, "vscan", | |
1008 | vscan_changed_cb, zsb) == 0); | |
1009 | ||
1010 | VERIFY(dsl_prop_unregister(ds, "nbmand", | |
1011 | nbmand_changed_cb, zsb) == 0); | |
1012 | } | |
1013 | } | |
1014 | EXPORT_SYMBOL(zfs_unregister_callbacks); | |
1015 | ||
1016 | #ifdef HAVE_MLSLABEL | |
1017 | /* | |
1018 | * Check that the hex label string is appropriate for the dataset being | |
1019 | * mounted into the global_zone proper. | |
1020 | * | |
1021 | * Return an error if the hex label string is not default or | |
1022 | * admin_low/admin_high. For admin_low labels, the corresponding | |
1023 | * dataset must be readonly. | |
1024 | */ | |
1025 | int | |
1026 | zfs_check_global_label(const char *dsname, const char *hexsl) | |
1027 | { | |
1028 | if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0) | |
1029 | return (0); | |
1030 | if (strcasecmp(hexsl, ADMIN_HIGH) == 0) | |
1031 | return (0); | |
1032 | if (strcasecmp(hexsl, ADMIN_LOW) == 0) { | |
1033 | /* must be readonly */ | |
1034 | uint64_t rdonly; | |
1035 | ||
1036 | if (dsl_prop_get_integer(dsname, | |
1037 | zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL)) | |
1038 | return (SET_ERROR(EACCES)); | |
1039 | return (rdonly ? 0 : EACCES); | |
1040 | } | |
1041 | return (SET_ERROR(EACCES)); | |
1042 | } | |
1043 | EXPORT_SYMBOL(zfs_check_global_label); | |
1044 | #endif /* HAVE_MLSLABEL */ | |
1045 | ||
1046 | int | |
1047 | zfs_statvfs(struct dentry *dentry, struct kstatfs *statp) | |
1048 | { | |
1049 | zfs_sb_t *zsb = dentry->d_sb->s_fs_info; | |
1050 | uint64_t refdbytes, availbytes, usedobjs, availobjs; | |
1051 | uint64_t fsid; | |
1052 | uint32_t bshift; | |
1053 | ||
1054 | ZFS_ENTER(zsb); | |
1055 | ||
1056 | dmu_objset_space(zsb->z_os, | |
1057 | &refdbytes, &availbytes, &usedobjs, &availobjs); | |
1058 | ||
1059 | fsid = dmu_objset_fsid_guid(zsb->z_os); | |
1060 | /* | |
1061 | * The underlying storage pool actually uses multiple block | |
1062 | * size. Under Solaris frsize (fragment size) is reported as | |
1063 | * the smallest block size we support, and bsize (block size) | |
1064 | * as the filesystem's maximum block size. Unfortunately, | |
1065 | * under Linux the fragment size and block size are often used | |
1066 | * interchangeably. Thus we are forced to report both of them | |
1067 | * as the filesystem's maximum block size. | |
1068 | */ | |
1069 | statp->f_frsize = zsb->z_max_blksz; | |
1070 | statp->f_bsize = zsb->z_max_blksz; | |
1071 | bshift = fls(statp->f_bsize) - 1; | |
1072 | ||
1073 | /* | |
1074 | * The following report "total" blocks of various kinds in | |
1075 | * the file system, but reported in terms of f_bsize - the | |
1076 | * "preferred" size. | |
1077 | */ | |
1078 | ||
1079 | statp->f_blocks = (refdbytes + availbytes) >> bshift; | |
1080 | statp->f_bfree = availbytes >> bshift; | |
1081 | statp->f_bavail = statp->f_bfree; /* no root reservation */ | |
1082 | ||
1083 | /* | |
1084 | * statvfs() should really be called statufs(), because it assumes | |
1085 | * static metadata. ZFS doesn't preallocate files, so the best | |
1086 | * we can do is report the max that could possibly fit in f_files, | |
1087 | * and that minus the number actually used in f_ffree. | |
1088 | * For f_ffree, report the smaller of the number of object available | |
1089 | * and the number of blocks (each object will take at least a block). | |
1090 | */ | |
1091 | statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT); | |
1092 | statp->f_files = statp->f_ffree + usedobjs; | |
1093 | statp->f_fsid.val[0] = (uint32_t)fsid; | |
1094 | statp->f_fsid.val[1] = (uint32_t)(fsid >> 32); | |
1095 | statp->f_type = ZFS_SUPER_MAGIC; | |
1096 | statp->f_namelen = ZFS_MAXNAMELEN; | |
1097 | ||
1098 | /* | |
1099 | * We have all of 40 characters to stuff a string here. | |
1100 | * Is there anything useful we could/should provide? | |
1101 | */ | |
1102 | bzero(statp->f_spare, sizeof (statp->f_spare)); | |
1103 | ||
1104 | ZFS_EXIT(zsb); | |
1105 | return (0); | |
1106 | } | |
1107 | EXPORT_SYMBOL(zfs_statvfs); | |
1108 | ||
1109 | int | |
1110 | zfs_root(zfs_sb_t *zsb, struct inode **ipp) | |
1111 | { | |
1112 | znode_t *rootzp; | |
1113 | int error; | |
1114 | ||
1115 | ZFS_ENTER(zsb); | |
1116 | ||
1117 | error = zfs_zget(zsb, zsb->z_root, &rootzp); | |
1118 | if (error == 0) | |
1119 | *ipp = ZTOI(rootzp); | |
1120 | ||
1121 | ZFS_EXIT(zsb); | |
1122 | return (error); | |
1123 | } | |
1124 | EXPORT_SYMBOL(zfs_root); | |
1125 | ||
1126 | #ifdef HAVE_D_PRUNE_ALIASES | |
1127 | /* | |
1128 | * Linux kernels older than 3.1 do not support a per-filesystem shrinker. | |
1129 | * To accommodate this we must improvise and manually walk the list of znodes | |
1130 | * attempting to prune dentries in order to be able to drop the inodes. | |
1131 | * | |
1132 | * To avoid scanning the same znodes multiple times they are always rotated | |
1133 | * to the end of the z_all_znodes list. New znodes are inserted at the | |
1134 | * end of the list so we're always scanning the oldest znodes first. | |
1135 | */ | |
1136 | static int | |
1137 | zfs_sb_prune_aliases(zfs_sb_t *zsb, unsigned long nr_to_scan) | |
1138 | { | |
1139 | znode_t **zp_array, *zp; | |
1140 | int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *)); | |
1141 | int objects = 0; | |
1142 | int i = 0, j = 0; | |
1143 | ||
1144 | zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP); | |
1145 | ||
1146 | mutex_enter(&zsb->z_znodes_lock); | |
1147 | while ((zp = list_head(&zsb->z_all_znodes)) != NULL) { | |
1148 | ||
1149 | if ((i++ > nr_to_scan) || (j >= max_array)) | |
1150 | break; | |
1151 | ||
1152 | ASSERT(list_link_active(&zp->z_link_node)); | |
1153 | list_remove(&zsb->z_all_znodes, zp); | |
1154 | list_insert_tail(&zsb->z_all_znodes, zp); | |
1155 | ||
1156 | /* Skip active znodes and .zfs entries */ | |
1157 | if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir) | |
1158 | continue; | |
1159 | ||
1160 | if (igrab(ZTOI(zp)) == NULL) | |
1161 | continue; | |
1162 | ||
1163 | zp_array[j] = zp; | |
1164 | j++; | |
1165 | } | |
1166 | mutex_exit(&zsb->z_znodes_lock); | |
1167 | ||
1168 | for (i = 0; i < j; i++) { | |
1169 | zp = zp_array[i]; | |
1170 | ||
1171 | ASSERT3P(zp, !=, NULL); | |
1172 | d_prune_aliases(ZTOI(zp)); | |
1173 | ||
1174 | if (atomic_read(&ZTOI(zp)->i_count) == 1) | |
1175 | objects++; | |
1176 | ||
1177 | iput(ZTOI(zp)); | |
1178 | } | |
1179 | ||
1180 | kmem_free(zp_array, max_array * sizeof (znode_t *)); | |
1181 | ||
1182 | return (objects); | |
1183 | } | |
1184 | #endif /* HAVE_D_PRUNE_ALIASES */ | |
1185 | ||
1186 | /* | |
1187 | * The ARC has requested that the filesystem drop entries from the dentry | |
1188 | * and inode caches. This can occur when the ARC needs to free meta data | |
1189 | * blocks but can't because they are all pinned by entries in these caches. | |
1190 | */ | |
1191 | int | |
1192 | zfs_sb_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects) | |
1193 | { | |
1194 | zfs_sb_t *zsb = sb->s_fs_info; | |
1195 | int error = 0; | |
1196 | #if defined(HAVE_SHRINK) || defined(HAVE_SPLIT_SHRINKER_CALLBACK) | |
1197 | struct shrinker *shrinker = &sb->s_shrink; | |
1198 | struct shrink_control sc = { | |
1199 | .nr_to_scan = nr_to_scan, | |
1200 | .gfp_mask = GFP_KERNEL, | |
1201 | }; | |
1202 | #endif | |
1203 | ||
1204 | ZFS_ENTER(zsb); | |
1205 | ||
1206 | #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \ | |
1207 | defined(SHRINK_CONTROL_HAS_NID) && \ | |
1208 | defined(SHRINKER_NUMA_AWARE) | |
1209 | if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) { | |
1210 | *objects = 0; | |
1211 | for_each_online_node(sc.nid) | |
1212 | *objects += (*shrinker->scan_objects)(shrinker, &sc); | |
1213 | } else { | |
1214 | *objects = (*shrinker->scan_objects)(shrinker, &sc); | |
1215 | } | |
1216 | ||
1217 | #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK) | |
1218 | *objects = (*shrinker->scan_objects)(shrinker, &sc); | |
1219 | #elif defined(HAVE_SHRINK) | |
1220 | *objects = (*shrinker->shrink)(shrinker, &sc); | |
1221 | #elif defined(HAVE_D_PRUNE_ALIASES) | |
1222 | #define D_PRUNE_ALIASES_IS_DEFAULT | |
1223 | *objects = zfs_sb_prune_aliases(zsb, nr_to_scan); | |
1224 | #else | |
1225 | #error "No available dentry and inode cache pruning mechanism." | |
1226 | #endif | |
1227 | ||
1228 | #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT) | |
1229 | #undef D_PRUNE_ALIASES_IS_DEFAULT | |
1230 | /* | |
1231 | * Fall back to zfs_sb_prune_aliases if the kernel's per-superblock | |
1232 | * shrinker couldn't free anything, possibly due to the inodes being | |
1233 | * allocated in a different memcg. | |
1234 | */ | |
1235 | if (*objects == 0) | |
1236 | *objects = zfs_sb_prune_aliases(zsb, nr_to_scan); | |
1237 | #endif | |
1238 | ||
1239 | ZFS_EXIT(zsb); | |
1240 | ||
1241 | dprintf_ds(zsb->z_os->os_dsl_dataset, | |
1242 | "pruning, nr_to_scan=%lu objects=%d error=%d\n", | |
1243 | nr_to_scan, *objects, error); | |
1244 | ||
1245 | return (error); | |
1246 | } | |
1247 | EXPORT_SYMBOL(zfs_sb_prune); | |
1248 | ||
1249 | /* | |
1250 | * Teardown the zfs_sb_t. | |
1251 | * | |
1252 | * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock' | |
1253 | * and 'z_teardown_inactive_lock' held. | |
1254 | */ | |
1255 | int | |
1256 | zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting) | |
1257 | { | |
1258 | znode_t *zp; | |
1259 | ||
1260 | /* | |
1261 | * If someone has not already unmounted this file system, | |
1262 | * drain the iput_taskq to ensure all active references to the | |
1263 | * zfs_sb_t have been handled only then can it be safely destroyed. | |
1264 | */ | |
1265 | if (zsb->z_os) { | |
1266 | /* | |
1267 | * If we're unmounting we have to wait for the list to | |
1268 | * drain completely. | |
1269 | * | |
1270 | * If we're not unmounting there's no guarantee the list | |
1271 | * will drain completely, but iputs run from the taskq | |
1272 | * may add the parents of dir-based xattrs to the taskq | |
1273 | * so we want to wait for these. | |
1274 | * | |
1275 | * We can safely read z_nr_znodes without locking because the | |
1276 | * VFS has already blocked operations which add to the | |
1277 | * z_all_znodes list and thus increment z_nr_znodes. | |
1278 | */ | |
1279 | int round = 0; | |
1280 | while (zsb->z_nr_znodes > 0) { | |
1281 | taskq_wait_outstanding(dsl_pool_iput_taskq( | |
1282 | dmu_objset_pool(zsb->z_os)), 0); | |
1283 | if (++round > 1 && !unmounting) | |
1284 | break; | |
1285 | } | |
1286 | } | |
1287 | ||
1288 | rrm_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG); | |
1289 | ||
1290 | if (!unmounting) { | |
1291 | /* | |
1292 | * We purge the parent filesystem's super block as the | |
1293 | * parent filesystem and all of its snapshots have their | |
1294 | * inode's super block set to the parent's filesystem's | |
1295 | * super block. Note, 'z_parent' is self referential | |
1296 | * for non-snapshots. | |
1297 | */ | |
1298 | shrink_dcache_sb(zsb->z_parent->z_sb); | |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * Close the zil. NB: Can't close the zil while zfs_inactive | |
1303 | * threads are blocked as zil_close can call zfs_inactive. | |
1304 | */ | |
1305 | if (zsb->z_log) { | |
1306 | zil_close(zsb->z_log); | |
1307 | zsb->z_log = NULL; | |
1308 | } | |
1309 | ||
1310 | rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER); | |
1311 | ||
1312 | /* | |
1313 | * If we are not unmounting (ie: online recv) and someone already | |
1314 | * unmounted this file system while we were doing the switcheroo, | |
1315 | * or a reopen of z_os failed then just bail out now. | |
1316 | */ | |
1317 | if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) { | |
1318 | rw_exit(&zsb->z_teardown_inactive_lock); | |
1319 | rrm_exit(&zsb->z_teardown_lock, FTAG); | |
1320 | return (SET_ERROR(EIO)); | |
1321 | } | |
1322 | ||
1323 | /* | |
1324 | * At this point there are no VFS ops active, and any new VFS ops | |
1325 | * will fail with EIO since we have z_teardown_lock for writer (only | |
1326 | * relevant for forced unmount). | |
1327 | * | |
1328 | * Release all holds on dbufs. | |
1329 | */ | |
1330 | if (!unmounting) { | |
1331 | mutex_enter(&zsb->z_znodes_lock); | |
1332 | for (zp = list_head(&zsb->z_all_znodes); zp != NULL; | |
1333 | zp = list_next(&zsb->z_all_znodes, zp)) { | |
1334 | if (zp->z_sa_hdl) | |
1335 | zfs_znode_dmu_fini(zp); | |
1336 | } | |
1337 | mutex_exit(&zsb->z_znodes_lock); | |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * If we are unmounting, set the unmounted flag and let new VFS ops | |
1342 | * unblock. zfs_inactive will have the unmounted behavior, and all | |
1343 | * other VFS ops will fail with EIO. | |
1344 | */ | |
1345 | if (unmounting) { | |
1346 | zsb->z_unmounted = B_TRUE; | |
1347 | rrm_exit(&zsb->z_teardown_lock, FTAG); | |
1348 | rw_exit(&zsb->z_teardown_inactive_lock); | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * z_os will be NULL if there was an error in attempting to reopen | |
1353 | * zsb, so just return as the properties had already been | |
1354 | * | |
1355 | * unregistered and cached data had been evicted before. | |
1356 | */ | |
1357 | if (zsb->z_os == NULL) | |
1358 | return (0); | |
1359 | ||
1360 | /* | |
1361 | * Unregister properties. | |
1362 | */ | |
1363 | zfs_unregister_callbacks(zsb); | |
1364 | ||
1365 | /* | |
1366 | * Evict cached data | |
1367 | */ | |
1368 | if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) && | |
1369 | !zfs_is_readonly(zsb)) | |
1370 | txg_wait_synced(dmu_objset_pool(zsb->z_os), 0); | |
1371 | dmu_objset_evict_dbufs(zsb->z_os); | |
1372 | ||
1373 | return (0); | |
1374 | } | |
1375 | EXPORT_SYMBOL(zfs_sb_teardown); | |
1376 | ||
1377 | #if !defined(HAVE_2ARGS_BDI_SETUP_AND_REGISTER) && \ | |
1378 | !defined(HAVE_3ARGS_BDI_SETUP_AND_REGISTER) | |
1379 | atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0); | |
1380 | #endif | |
1381 | ||
1382 | int | |
1383 | zfs_domount(struct super_block *sb, zfs_mntopts_t *zmo, int silent) | |
1384 | { | |
1385 | const char *osname = zmo->z_osname; | |
1386 | zfs_sb_t *zsb; | |
1387 | struct inode *root_inode; | |
1388 | uint64_t recordsize; | |
1389 | int error; | |
1390 | ||
1391 | error = zfs_sb_create(osname, zmo, &zsb); | |
1392 | if (error) | |
1393 | return (error); | |
1394 | ||
1395 | if ((error = dsl_prop_get_integer(osname, "recordsize", | |
1396 | &recordsize, NULL))) | |
1397 | goto out; | |
1398 | ||
1399 | zsb->z_sb = sb; | |
1400 | sb->s_fs_info = zsb; | |
1401 | sb->s_magic = ZFS_SUPER_MAGIC; | |
1402 | sb->s_maxbytes = MAX_LFS_FILESIZE; | |
1403 | sb->s_time_gran = 1; | |
1404 | sb->s_blocksize = recordsize; | |
1405 | sb->s_blocksize_bits = ilog2(recordsize); | |
87d546d8 | 1406 | |
df9d7621 | 1407 | error = -zpl_bdi_setup(sb, "zfs"); |
87d546d8 TG |
1408 | if (error) |
1409 | goto out; | |
1410 | ||
df9d7621 CIK |
1411 | sb->s_bdi->ra_pages = 0; |
1412 | ||
87d546d8 TG |
1413 | /* Set callback operations for the file system. */ |
1414 | sb->s_op = &zpl_super_operations; | |
1415 | sb->s_xattr = zpl_xattr_handlers; | |
1416 | sb->s_export_op = &zpl_export_operations; | |
1417 | #ifdef HAVE_S_D_OP | |
1418 | sb->s_d_op = &zpl_dentry_operations; | |
1419 | #endif /* HAVE_S_D_OP */ | |
1420 | ||
1421 | /* Set features for file system. */ | |
1422 | zfs_set_fuid_feature(zsb); | |
1423 | ||
1424 | if (dmu_objset_is_snapshot(zsb->z_os)) { | |
1425 | uint64_t pval; | |
1426 | ||
1427 | atime_changed_cb(zsb, B_FALSE); | |
1428 | readonly_changed_cb(zsb, B_TRUE); | |
1429 | if ((error = dsl_prop_get_integer(osname, | |
1430 | "xattr", &pval, NULL))) | |
1431 | goto out; | |
1432 | xattr_changed_cb(zsb, pval); | |
1433 | if ((error = dsl_prop_get_integer(osname, | |
1434 | "acltype", &pval, NULL))) | |
1435 | goto out; | |
1436 | acltype_changed_cb(zsb, pval); | |
1437 | zsb->z_issnap = B_TRUE; | |
1438 | zsb->z_os->os_sync = ZFS_SYNC_DISABLED; | |
1439 | zsb->z_snap_defer_time = jiffies; | |
1440 | ||
1441 | mutex_enter(&zsb->z_os->os_user_ptr_lock); | |
1442 | dmu_objset_set_user(zsb->z_os, zsb); | |
1443 | mutex_exit(&zsb->z_os->os_user_ptr_lock); | |
1444 | } else { | |
1445 | error = zfs_sb_setup(zsb, B_TRUE); | |
1446 | } | |
1447 | ||
1448 | /* Allocate a root inode for the filesystem. */ | |
1449 | error = zfs_root(zsb, &root_inode); | |
1450 | if (error) { | |
1451 | (void) zfs_umount(sb); | |
1452 | goto out; | |
1453 | } | |
1454 | ||
1455 | /* Allocate a root dentry for the filesystem */ | |
1456 | sb->s_root = d_make_root(root_inode); | |
1457 | if (sb->s_root == NULL) { | |
1458 | (void) zfs_umount(sb); | |
1459 | error = SET_ERROR(ENOMEM); | |
1460 | goto out; | |
1461 | } | |
1462 | ||
1463 | if (!zsb->z_issnap) | |
1464 | zfsctl_create(zsb); | |
1465 | ||
1466 | zsb->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb); | |
1467 | out: | |
1468 | if (error) { | |
1469 | dmu_objset_disown(zsb->z_os, zsb); | |
1470 | zfs_sb_free(zsb); | |
1471 | } | |
1472 | ||
1473 | return (error); | |
1474 | } | |
1475 | EXPORT_SYMBOL(zfs_domount); | |
1476 | ||
1477 | /* | |
1478 | * Called when an unmount is requested and certain sanity checks have | |
1479 | * already passed. At this point no dentries or inodes have been reclaimed | |
1480 | * from their respective caches. We drop the extra reference on the .zfs | |
1481 | * control directory to allow everything to be reclaimed. All snapshots | |
1482 | * must already have been unmounted to reach this point. | |
1483 | */ | |
1484 | void | |
1485 | zfs_preumount(struct super_block *sb) | |
1486 | { | |
1487 | zfs_sb_t *zsb = sb->s_fs_info; | |
1488 | ||
1489 | if (zsb) | |
1490 | zfsctl_destroy(sb->s_fs_info); | |
1491 | } | |
1492 | EXPORT_SYMBOL(zfs_preumount); | |
1493 | ||
1494 | /* | |
1495 | * Called once all other unmount released tear down has occurred. | |
1496 | * It is our responsibility to release any remaining infrastructure. | |
1497 | */ | |
1498 | /*ARGSUSED*/ | |
1499 | int | |
1500 | zfs_umount(struct super_block *sb) | |
1501 | { | |
1502 | zfs_sb_t *zsb = sb->s_fs_info; | |
1503 | objset_t *os; | |
1504 | ||
1505 | arc_remove_prune_callback(zsb->z_arc_prune); | |
1506 | VERIFY(zfs_sb_teardown(zsb, B_TRUE) == 0); | |
1507 | os = zsb->z_os; | |
df9d7621 | 1508 | zpl_bdi_destroy(sb); |
87d546d8 TG |
1509 | |
1510 | /* | |
1511 | * z_os will be NULL if there was an error in | |
1512 | * attempting to reopen zsb. | |
1513 | */ | |
1514 | if (os != NULL) { | |
1515 | /* | |
1516 | * Unset the objset user_ptr. | |
1517 | */ | |
1518 | mutex_enter(&os->os_user_ptr_lock); | |
1519 | dmu_objset_set_user(os, NULL); | |
1520 | mutex_exit(&os->os_user_ptr_lock); | |
1521 | ||
1522 | /* | |
1523 | * Finally release the objset | |
1524 | */ | |
1525 | dmu_objset_disown(os, zsb); | |
1526 | } | |
1527 | ||
1528 | zfs_sb_free(zsb); | |
1529 | return (0); | |
1530 | } | |
1531 | EXPORT_SYMBOL(zfs_umount); | |
1532 | ||
1533 | int | |
1534 | zfs_remount(struct super_block *sb, int *flags, zfs_mntopts_t *zmo) | |
1535 | { | |
1536 | zfs_sb_t *zsb = sb->s_fs_info; | |
1537 | int error; | |
1538 | ||
1539 | zfs_unregister_callbacks(zsb); | |
1540 | error = zfs_register_callbacks(zsb); | |
1541 | ||
1542 | return (error); | |
1543 | } | |
1544 | EXPORT_SYMBOL(zfs_remount); | |
1545 | ||
1546 | int | |
1547 | zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp) | |
1548 | { | |
1549 | zfs_sb_t *zsb = sb->s_fs_info; | |
1550 | znode_t *zp; | |
1551 | uint64_t object = 0; | |
1552 | uint64_t fid_gen = 0; | |
1553 | uint64_t gen_mask; | |
1554 | uint64_t zp_gen; | |
1555 | int i, err; | |
1556 | ||
1557 | *ipp = NULL; | |
1558 | ||
1559 | ZFS_ENTER(zsb); | |
1560 | ||
1561 | if (fidp->fid_len == LONG_FID_LEN) { | |
1562 | zfid_long_t *zlfid = (zfid_long_t *)fidp; | |
1563 | uint64_t objsetid = 0; | |
1564 | uint64_t setgen = 0; | |
1565 | ||
1566 | for (i = 0; i < sizeof (zlfid->zf_setid); i++) | |
1567 | objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i); | |
1568 | ||
1569 | for (i = 0; i < sizeof (zlfid->zf_setgen); i++) | |
1570 | setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i); | |
1571 | ||
1572 | ZFS_EXIT(zsb); | |
1573 | ||
1574 | err = zfsctl_lookup_objset(sb, objsetid, &zsb); | |
1575 | if (err) | |
1576 | return (SET_ERROR(EINVAL)); | |
1577 | ||
1578 | ZFS_ENTER(zsb); | |
1579 | } | |
1580 | ||
1581 | if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) { | |
1582 | zfid_short_t *zfid = (zfid_short_t *)fidp; | |
1583 | ||
1584 | for (i = 0; i < sizeof (zfid->zf_object); i++) | |
1585 | object |= ((uint64_t)zfid->zf_object[i]) << (8 * i); | |
1586 | ||
1587 | for (i = 0; i < sizeof (zfid->zf_gen); i++) | |
1588 | fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i); | |
1589 | } else { | |
1590 | ZFS_EXIT(zsb); | |
1591 | return (SET_ERROR(EINVAL)); | |
1592 | } | |
1593 | ||
1594 | /* A zero fid_gen means we are in the .zfs control directories */ | |
1595 | if (fid_gen == 0 && | |
1596 | (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) { | |
1597 | *ipp = zsb->z_ctldir; | |
1598 | ASSERT(*ipp != NULL); | |
1599 | if (object == ZFSCTL_INO_SNAPDIR) { | |
1600 | VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, | |
1601 | 0, kcred, NULL, NULL) == 0); | |
1602 | } else { | |
1603 | igrab(*ipp); | |
1604 | } | |
1605 | ZFS_EXIT(zsb); | |
1606 | return (0); | |
1607 | } | |
1608 | ||
1609 | gen_mask = -1ULL >> (64 - 8 * i); | |
1610 | ||
1611 | dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask); | |
1612 | if ((err = zfs_zget(zsb, object, &zp))) { | |
1613 | ZFS_EXIT(zsb); | |
1614 | return (err); | |
1615 | } | |
1616 | ||
1617 | /* Don't export xattr stuff */ | |
1618 | if (zp->z_pflags & ZFS_XATTR) { | |
1619 | iput(ZTOI(zp)); | |
1620 | ZFS_EXIT(zsb); | |
1621 | return (SET_ERROR(ENOENT)); | |
1622 | } | |
1623 | ||
1624 | (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb), &zp_gen, | |
1625 | sizeof (uint64_t)); | |
1626 | zp_gen = zp_gen & gen_mask; | |
1627 | if (zp_gen == 0) | |
1628 | zp_gen = 1; | |
1629 | if ((fid_gen == 0) && (zsb->z_root == object)) | |
1630 | fid_gen = zp_gen; | |
1631 | if (zp->z_unlinked || zp_gen != fid_gen) { | |
1632 | dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen, | |
1633 | fid_gen); | |
1634 | iput(ZTOI(zp)); | |
1635 | ZFS_EXIT(zsb); | |
1636 | return (SET_ERROR(ENOENT)); | |
1637 | } | |
1638 | ||
1639 | *ipp = ZTOI(zp); | |
1640 | if (*ipp) | |
1641 | zfs_inode_update(ITOZ(*ipp)); | |
1642 | ||
1643 | ZFS_EXIT(zsb); | |
1644 | return (0); | |
1645 | } | |
1646 | EXPORT_SYMBOL(zfs_vget); | |
1647 | ||
1648 | /* | |
1649 | * Block out VFS ops and close zfs_sb_t | |
1650 | * | |
1651 | * Note, if successful, then we return with the 'z_teardown_lock' and | |
1652 | * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying | |
1653 | * dataset and objset intact so that they can be atomically handed off during | |
1654 | * a subsequent rollback or recv operation and the resume thereafter. | |
1655 | */ | |
1656 | int | |
1657 | zfs_suspend_fs(zfs_sb_t *zsb) | |
1658 | { | |
1659 | int error; | |
1660 | ||
1661 | if ((error = zfs_sb_teardown(zsb, B_FALSE)) != 0) | |
1662 | return (error); | |
1663 | ||
1664 | return (0); | |
1665 | } | |
1666 | EXPORT_SYMBOL(zfs_suspend_fs); | |
1667 | ||
1668 | /* | |
1669 | * Reopen zfs_sb_t and release VFS ops. | |
1670 | */ | |
1671 | int | |
1672 | zfs_resume_fs(zfs_sb_t *zsb, const char *osname) | |
1673 | { | |
1674 | int err, err2; | |
1675 | znode_t *zp; | |
1676 | uint64_t sa_obj = 0; | |
1677 | ||
1678 | ASSERT(RRM_WRITE_HELD(&zsb->z_teardown_lock)); | |
1679 | ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock)); | |
1680 | ||
1681 | /* | |
1682 | * We already own this, so just hold and rele it to update the | |
1683 | * objset_t, as the one we had before may have been evicted. | |
1684 | */ | |
1685 | VERIFY0(dmu_objset_hold(osname, zsb, &zsb->z_os)); | |
1686 | VERIFY3P(zsb->z_os->os_dsl_dataset->ds_owner, ==, zsb); | |
1687 | VERIFY(dsl_dataset_long_held(zsb->z_os->os_dsl_dataset)); | |
1688 | dmu_objset_rele(zsb->z_os, zsb); | |
1689 | ||
1690 | /* | |
1691 | * Make sure version hasn't changed | |
1692 | */ | |
1693 | ||
1694 | err = zfs_get_zplprop(zsb->z_os, ZFS_PROP_VERSION, | |
1695 | &zsb->z_version); | |
1696 | ||
1697 | if (err) | |
1698 | goto bail; | |
1699 | ||
1700 | err = zap_lookup(zsb->z_os, MASTER_NODE_OBJ, | |
1701 | ZFS_SA_ATTRS, 8, 1, &sa_obj); | |
1702 | ||
1703 | if (err && zsb->z_version >= ZPL_VERSION_SA) | |
1704 | goto bail; | |
1705 | ||
1706 | if ((err = sa_setup(zsb->z_os, sa_obj, | |
1707 | zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0) | |
1708 | goto bail; | |
1709 | ||
1710 | if (zsb->z_version >= ZPL_VERSION_SA) | |
1711 | sa_register_update_callback(zsb->z_os, | |
1712 | zfs_sa_upgrade); | |
1713 | ||
1714 | VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0); | |
1715 | ||
1716 | zfs_set_fuid_feature(zsb); | |
1717 | zsb->z_rollback_time = jiffies; | |
1718 | ||
1719 | /* | |
1720 | * Attempt to re-establish all the active inodes with their | |
1721 | * dbufs. If a zfs_rezget() fails, then we unhash the inode | |
1722 | * and mark it stale. This prevents a collision if a new | |
1723 | * inode/object is created which must use the same inode | |
1724 | * number. The stale inode will be be released when the | |
1725 | * VFS prunes the dentry holding the remaining references | |
1726 | * on the stale inode. | |
1727 | */ | |
1728 | mutex_enter(&zsb->z_znodes_lock); | |
1729 | for (zp = list_head(&zsb->z_all_znodes); zp; | |
1730 | zp = list_next(&zsb->z_all_znodes, zp)) { | |
1731 | err2 = zfs_rezget(zp); | |
1732 | if (err2) { | |
1733 | remove_inode_hash(ZTOI(zp)); | |
1734 | zp->z_is_stale = B_TRUE; | |
1735 | } | |
1736 | } | |
1737 | mutex_exit(&zsb->z_znodes_lock); | |
1738 | ||
1739 | bail: | |
1740 | /* release the VFS ops */ | |
1741 | rw_exit(&zsb->z_teardown_inactive_lock); | |
1742 | rrm_exit(&zsb->z_teardown_lock, FTAG); | |
1743 | ||
1744 | if (err) { | |
1745 | /* | |
1746 | * Since we couldn't setup the sa framework, try to force | |
1747 | * unmount this file system. | |
1748 | */ | |
1749 | if (zsb->z_os) | |
1750 | (void) zfs_umount(zsb->z_sb); | |
1751 | } | |
1752 | return (err); | |
1753 | } | |
1754 | EXPORT_SYMBOL(zfs_resume_fs); | |
1755 | ||
1756 | int | |
1757 | zfs_set_version(zfs_sb_t *zsb, uint64_t newvers) | |
1758 | { | |
1759 | int error; | |
1760 | objset_t *os = zsb->z_os; | |
1761 | dmu_tx_t *tx; | |
1762 | ||
1763 | if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION) | |
1764 | return (SET_ERROR(EINVAL)); | |
1765 | ||
1766 | if (newvers < zsb->z_version) | |
1767 | return (SET_ERROR(EINVAL)); | |
1768 | ||
1769 | if (zfs_spa_version_map(newvers) > | |
1770 | spa_version(dmu_objset_spa(zsb->z_os))) | |
1771 | return (SET_ERROR(ENOTSUP)); | |
1772 | ||
1773 | tx = dmu_tx_create(os); | |
1774 | dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR); | |
1775 | if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) { | |
1776 | dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE, | |
1777 | ZFS_SA_ATTRS); | |
1778 | dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); | |
1779 | } | |
1780 | error = dmu_tx_assign(tx, TXG_WAIT); | |
1781 | if (error) { | |
1782 | dmu_tx_abort(tx); | |
1783 | return (error); | |
1784 | } | |
1785 | ||
1786 | error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR, | |
1787 | 8, 1, &newvers, tx); | |
1788 | ||
1789 | if (error) { | |
1790 | dmu_tx_commit(tx); | |
1791 | return (error); | |
1792 | } | |
1793 | ||
1794 | if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) { | |
1795 | uint64_t sa_obj; | |
1796 | ||
1797 | ASSERT3U(spa_version(dmu_objset_spa(zsb->z_os)), >=, | |
1798 | SPA_VERSION_SA); | |
1799 | sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE, | |
1800 | DMU_OT_NONE, 0, tx); | |
1801 | ||
1802 | error = zap_add(os, MASTER_NODE_OBJ, | |
1803 | ZFS_SA_ATTRS, 8, 1, &sa_obj, tx); | |
1804 | ASSERT0(error); | |
1805 | ||
1806 | VERIFY(0 == sa_set_sa_object(os, sa_obj)); | |
1807 | sa_register_update_callback(os, zfs_sa_upgrade); | |
1808 | } | |
1809 | ||
1810 | spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx, | |
1811 | "from %llu to %llu", zsb->z_version, newvers); | |
1812 | ||
1813 | dmu_tx_commit(tx); | |
1814 | ||
1815 | zsb->z_version = newvers; | |
1816 | ||
1817 | zfs_set_fuid_feature(zsb); | |
1818 | ||
1819 | return (0); | |
1820 | } | |
1821 | EXPORT_SYMBOL(zfs_set_version); | |
1822 | ||
1823 | /* | |
1824 | * Read a property stored within the master node. | |
1825 | */ | |
1826 | int | |
1827 | zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value) | |
1828 | { | |
1829 | const char *pname; | |
1830 | int error = SET_ERROR(ENOENT); | |
1831 | ||
1832 | /* | |
1833 | * Look up the file system's value for the property. For the | |
1834 | * version property, we look up a slightly different string. | |
1835 | */ | |
1836 | if (prop == ZFS_PROP_VERSION) | |
1837 | pname = ZPL_VERSION_STR; | |
1838 | else | |
1839 | pname = zfs_prop_to_name(prop); | |
1840 | ||
1841 | if (os != NULL) | |
1842 | error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value); | |
1843 | ||
1844 | if (error == ENOENT) { | |
1845 | /* No value set, use the default value */ | |
1846 | switch (prop) { | |
1847 | case ZFS_PROP_VERSION: | |
1848 | *value = ZPL_VERSION; | |
1849 | break; | |
1850 | case ZFS_PROP_NORMALIZE: | |
1851 | case ZFS_PROP_UTF8ONLY: | |
1852 | *value = 0; | |
1853 | break; | |
1854 | case ZFS_PROP_CASE: | |
1855 | *value = ZFS_CASE_SENSITIVE; | |
1856 | break; | |
1857 | case ZFS_PROP_ACLTYPE: | |
1858 | *value = ZFS_ACLTYPE_OFF; | |
1859 | break; | |
1860 | default: | |
1861 | return (error); | |
1862 | } | |
1863 | error = 0; | |
1864 | } | |
1865 | return (error); | |
1866 | } | |
1867 | EXPORT_SYMBOL(zfs_get_zplprop); | |
1868 | ||
1869 | void | |
1870 | zfs_init(void) | |
1871 | { | |
1872 | zfsctl_init(); | |
1873 | zfs_znode_init(); | |
1874 | dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb); | |
1875 | register_filesystem(&zpl_fs_type); | |
1876 | } | |
1877 | ||
1878 | void | |
1879 | zfs_fini(void) | |
1880 | { | |
1881 | /* | |
1882 | * we don't use outstanding because zpl_posix_acl_free might add more. | |
1883 | */ | |
1884 | taskq_wait(system_taskq); | |
1885 | unregister_filesystem(&zpl_fs_type); | |
1886 | zfs_znode_fini(); | |
1887 | zfsctl_fini(); | |
1888 | } |