]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/zfs_vfsops.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
OpenZFS 7431 - ZFS Channel Programs
[mirror_zfs.git] / module / zfs / zfs_vfsops.c
1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 */
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 enum {
72 TOKEN_RO,
73 TOKEN_RW,
74 TOKEN_SETUID,
75 TOKEN_NOSETUID,
76 TOKEN_EXEC,
77 TOKEN_NOEXEC,
78 TOKEN_DEVICES,
79 TOKEN_NODEVICES,
80 TOKEN_DIRXATTR,
81 TOKEN_SAXATTR,
82 TOKEN_XATTR,
83 TOKEN_NOXATTR,
84 TOKEN_ATIME,
85 TOKEN_NOATIME,
86 TOKEN_RELATIME,
87 TOKEN_NORELATIME,
88 TOKEN_NBMAND,
89 TOKEN_NONBMAND,
90 TOKEN_MNTPOINT,
91 TOKEN_LAST,
92 };
93
94 static const match_table_t zpl_tokens = {
95 { TOKEN_RO, MNTOPT_RO },
96 { TOKEN_RW, MNTOPT_RW },
97 { TOKEN_SETUID, MNTOPT_SETUID },
98 { TOKEN_NOSETUID, MNTOPT_NOSETUID },
99 { TOKEN_EXEC, MNTOPT_EXEC },
100 { TOKEN_NOEXEC, MNTOPT_NOEXEC },
101 { TOKEN_DEVICES, MNTOPT_DEVICES },
102 { TOKEN_NODEVICES, MNTOPT_NODEVICES },
103 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR },
104 { TOKEN_SAXATTR, MNTOPT_SAXATTR },
105 { TOKEN_XATTR, MNTOPT_XATTR },
106 { TOKEN_NOXATTR, MNTOPT_NOXATTR },
107 { TOKEN_ATIME, MNTOPT_ATIME },
108 { TOKEN_NOATIME, MNTOPT_NOATIME },
109 { TOKEN_RELATIME, MNTOPT_RELATIME },
110 { TOKEN_NORELATIME, MNTOPT_NORELATIME },
111 { TOKEN_NBMAND, MNTOPT_NBMAND },
112 { TOKEN_NONBMAND, MNTOPT_NONBMAND },
113 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" },
114 { TOKEN_LAST, NULL },
115 };
116
117 static void
118 zfsvfs_vfs_free(vfs_t *vfsp)
119 {
120 if (vfsp != NULL) {
121 if (vfsp->vfs_mntpoint != NULL)
122 strfree(vfsp->vfs_mntpoint);
123
124 kmem_free(vfsp, sizeof (vfs_t));
125 }
126 }
127
128 static int
129 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
130 {
131 switch (token) {
132 case TOKEN_RO:
133 vfsp->vfs_readonly = B_TRUE;
134 vfsp->vfs_do_readonly = B_TRUE;
135 break;
136 case TOKEN_RW:
137 vfsp->vfs_readonly = B_FALSE;
138 vfsp->vfs_do_readonly = B_TRUE;
139 break;
140 case TOKEN_SETUID:
141 vfsp->vfs_setuid = B_TRUE;
142 vfsp->vfs_do_setuid = B_TRUE;
143 break;
144 case TOKEN_NOSETUID:
145 vfsp->vfs_setuid = B_FALSE;
146 vfsp->vfs_do_setuid = B_TRUE;
147 break;
148 case TOKEN_EXEC:
149 vfsp->vfs_exec = B_TRUE;
150 vfsp->vfs_do_exec = B_TRUE;
151 break;
152 case TOKEN_NOEXEC:
153 vfsp->vfs_exec = B_FALSE;
154 vfsp->vfs_do_exec = B_TRUE;
155 break;
156 case TOKEN_DEVICES:
157 vfsp->vfs_devices = B_TRUE;
158 vfsp->vfs_do_devices = B_TRUE;
159 break;
160 case TOKEN_NODEVICES:
161 vfsp->vfs_devices = B_FALSE;
162 vfsp->vfs_do_devices = B_TRUE;
163 break;
164 case TOKEN_DIRXATTR:
165 vfsp->vfs_xattr = ZFS_XATTR_DIR;
166 vfsp->vfs_do_xattr = B_TRUE;
167 break;
168 case TOKEN_SAXATTR:
169 vfsp->vfs_xattr = ZFS_XATTR_SA;
170 vfsp->vfs_do_xattr = B_TRUE;
171 break;
172 case TOKEN_XATTR:
173 vfsp->vfs_xattr = ZFS_XATTR_DIR;
174 vfsp->vfs_do_xattr = B_TRUE;
175 break;
176 case TOKEN_NOXATTR:
177 vfsp->vfs_xattr = ZFS_XATTR_OFF;
178 vfsp->vfs_do_xattr = B_TRUE;
179 break;
180 case TOKEN_ATIME:
181 vfsp->vfs_atime = B_TRUE;
182 vfsp->vfs_do_atime = B_TRUE;
183 break;
184 case TOKEN_NOATIME:
185 vfsp->vfs_atime = B_FALSE;
186 vfsp->vfs_do_atime = B_TRUE;
187 break;
188 case TOKEN_RELATIME:
189 vfsp->vfs_relatime = B_TRUE;
190 vfsp->vfs_do_relatime = B_TRUE;
191 break;
192 case TOKEN_NORELATIME:
193 vfsp->vfs_relatime = B_FALSE;
194 vfsp->vfs_do_relatime = B_TRUE;
195 break;
196 case TOKEN_NBMAND:
197 vfsp->vfs_nbmand = B_TRUE;
198 vfsp->vfs_do_nbmand = B_TRUE;
199 break;
200 case TOKEN_NONBMAND:
201 vfsp->vfs_nbmand = B_FALSE;
202 vfsp->vfs_do_nbmand = B_TRUE;
203 break;
204 case TOKEN_MNTPOINT:
205 vfsp->vfs_mntpoint = match_strdup(&args[0]);
206 if (vfsp->vfs_mntpoint == NULL)
207 return (SET_ERROR(ENOMEM));
208
209 break;
210 default:
211 break;
212 }
213
214 return (0);
215 }
216
217 /*
218 * Parse the raw mntopts and return a vfs_t describing the options.
219 */
220 static int
221 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
222 {
223 vfs_t *tmp_vfsp;
224 int error;
225
226 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
227
228 if (mntopts != NULL) {
229 substring_t args[MAX_OPT_ARGS];
230 char *tmp_mntopts, *p, *t;
231 int token;
232
233 tmp_mntopts = t = strdup(mntopts);
234 if (tmp_mntopts == NULL)
235 return (SET_ERROR(ENOMEM));
236
237 while ((p = strsep(&t, ",")) != NULL) {
238 if (!*p)
239 continue;
240
241 args[0].to = args[0].from = NULL;
242 token = match_token(p, zpl_tokens, args);
243 error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
244 if (error) {
245 strfree(tmp_mntopts);
246 zfsvfs_vfs_free(tmp_vfsp);
247 return (error);
248 }
249 }
250
251 strfree(tmp_mntopts);
252 }
253
254 *vfsp = tmp_vfsp;
255
256 return (0);
257 }
258
259 boolean_t
260 zfs_is_readonly(zfsvfs_t *zfsvfs)
261 {
262 return (!!(zfsvfs->z_sb->s_flags & MS_RDONLY));
263 }
264
265 /*ARGSUSED*/
266 int
267 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
268 {
269 zfsvfs_t *zfsvfs = sb->s_fs_info;
270
271 /*
272 * Data integrity is job one. We don't want a compromised kernel
273 * writing to the storage pool, so we never sync during panic.
274 */
275 if (unlikely(oops_in_progress))
276 return (0);
277
278 /*
279 * Semantically, the only requirement is that the sync be initiated.
280 * The DMU syncs out txgs frequently, so there's nothing to do.
281 */
282 if (!wait)
283 return (0);
284
285 if (zfsvfs != NULL) {
286 /*
287 * Sync a specific filesystem.
288 */
289 dsl_pool_t *dp;
290
291 ZFS_ENTER(zfsvfs);
292 dp = dmu_objset_pool(zfsvfs->z_os);
293
294 /*
295 * If the system is shutting down, then skip any
296 * filesystems which may exist on a suspended pool.
297 */
298 if (spa_suspended(dp->dp_spa)) {
299 ZFS_EXIT(zfsvfs);
300 return (0);
301 }
302
303 if (zfsvfs->z_log != NULL)
304 zil_commit(zfsvfs->z_log, 0);
305
306 ZFS_EXIT(zfsvfs);
307 } else {
308 /*
309 * Sync all ZFS filesystems. This is what happens when you
310 * run sync(1M). Unlike other filesystems, ZFS honors the
311 * request by waiting for all pools to commit all dirty data.
312 */
313 spa_sync_allpools();
314 }
315
316 return (0);
317 }
318
319 static void
320 atime_changed_cb(void *arg, uint64_t newval)
321 {
322 ((zfsvfs_t *)arg)->z_atime = newval;
323 }
324
325 static void
326 relatime_changed_cb(void *arg, uint64_t newval)
327 {
328 ((zfsvfs_t *)arg)->z_relatime = newval;
329 }
330
331 static void
332 xattr_changed_cb(void *arg, uint64_t newval)
333 {
334 zfsvfs_t *zfsvfs = arg;
335
336 if (newval == ZFS_XATTR_OFF) {
337 zfsvfs->z_flags &= ~ZSB_XATTR;
338 } else {
339 zfsvfs->z_flags |= ZSB_XATTR;
340
341 if (newval == ZFS_XATTR_SA)
342 zfsvfs->z_xattr_sa = B_TRUE;
343 else
344 zfsvfs->z_xattr_sa = B_FALSE;
345 }
346 }
347
348 static void
349 acltype_changed_cb(void *arg, uint64_t newval)
350 {
351 zfsvfs_t *zfsvfs = arg;
352
353 switch (newval) {
354 case ZFS_ACLTYPE_OFF:
355 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
356 zfsvfs->z_sb->s_flags &= ~MS_POSIXACL;
357 break;
358 case ZFS_ACLTYPE_POSIXACL:
359 #ifdef CONFIG_FS_POSIX_ACL
360 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIXACL;
361 zfsvfs->z_sb->s_flags |= MS_POSIXACL;
362 #else
363 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
364 zfsvfs->z_sb->s_flags &= ~MS_POSIXACL;
365 #endif /* CONFIG_FS_POSIX_ACL */
366 break;
367 default:
368 break;
369 }
370 }
371
372 static void
373 blksz_changed_cb(void *arg, uint64_t newval)
374 {
375 zfsvfs_t *zfsvfs = arg;
376 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
377 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
378 ASSERT(ISP2(newval));
379
380 zfsvfs->z_max_blksz = newval;
381 }
382
383 static void
384 readonly_changed_cb(void *arg, uint64_t newval)
385 {
386 zfsvfs_t *zfsvfs = arg;
387 struct super_block *sb = zfsvfs->z_sb;
388
389 if (sb == NULL)
390 return;
391
392 if (newval)
393 sb->s_flags |= MS_RDONLY;
394 else
395 sb->s_flags &= ~MS_RDONLY;
396 }
397
398 static void
399 devices_changed_cb(void *arg, uint64_t newval)
400 {
401 }
402
403 static void
404 setuid_changed_cb(void *arg, uint64_t newval)
405 {
406 }
407
408 static void
409 exec_changed_cb(void *arg, uint64_t newval)
410 {
411 }
412
413 static void
414 nbmand_changed_cb(void *arg, uint64_t newval)
415 {
416 zfsvfs_t *zfsvfs = arg;
417 struct super_block *sb = zfsvfs->z_sb;
418
419 if (sb == NULL)
420 return;
421
422 if (newval == TRUE)
423 sb->s_flags |= MS_MANDLOCK;
424 else
425 sb->s_flags &= ~MS_MANDLOCK;
426 }
427
428 static void
429 snapdir_changed_cb(void *arg, uint64_t newval)
430 {
431 ((zfsvfs_t *)arg)->z_show_ctldir = newval;
432 }
433
434 static void
435 vscan_changed_cb(void *arg, uint64_t newval)
436 {
437 ((zfsvfs_t *)arg)->z_vscan = newval;
438 }
439
440 static void
441 acl_inherit_changed_cb(void *arg, uint64_t newval)
442 {
443 ((zfsvfs_t *)arg)->z_acl_inherit = newval;
444 }
445
446 static int
447 zfs_register_callbacks(vfs_t *vfsp)
448 {
449 struct dsl_dataset *ds = NULL;
450 objset_t *os = NULL;
451 zfsvfs_t *zfsvfs = NULL;
452 int error = 0;
453
454 ASSERT(vfsp);
455 zfsvfs = vfsp->vfs_data;
456 ASSERT(zfsvfs);
457 os = zfsvfs->z_os;
458
459 /*
460 * The act of registering our callbacks will destroy any mount
461 * options we may have. In order to enable temporary overrides
462 * of mount options, we stash away the current values and
463 * restore them after we register the callbacks.
464 */
465 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
466 vfsp->vfs_do_readonly = B_TRUE;
467 vfsp->vfs_readonly = B_TRUE;
468 }
469
470 /*
471 * Register property callbacks.
472 *
473 * It would probably be fine to just check for i/o error from
474 * the first prop_register(), but I guess I like to go
475 * overboard...
476 */
477 ds = dmu_objset_ds(os);
478 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
479 error = dsl_prop_register(ds,
480 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
481 error = error ? error : dsl_prop_register(ds,
482 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
483 error = error ? error : dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
485 error = error ? error : dsl_prop_register(ds,
486 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
487 error = error ? error : dsl_prop_register(ds,
488 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
489 error = error ? error : dsl_prop_register(ds,
490 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
491 error = error ? error : dsl_prop_register(ds,
492 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
493 error = error ? error : dsl_prop_register(ds,
494 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
495 error = error ? error : dsl_prop_register(ds,
496 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
497 error = error ? error : dsl_prop_register(ds,
498 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
499 error = error ? error : dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
501 zfsvfs);
502 error = error ? error : dsl_prop_register(ds,
503 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
504 error = error ? error : dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
506 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
507 if (error)
508 goto unregister;
509
510 /*
511 * Invoke our callbacks to restore temporary mount options.
512 */
513 if (vfsp->vfs_do_readonly)
514 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
515 if (vfsp->vfs_do_setuid)
516 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
517 if (vfsp->vfs_do_exec)
518 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
519 if (vfsp->vfs_do_devices)
520 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
521 if (vfsp->vfs_do_xattr)
522 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
523 if (vfsp->vfs_do_atime)
524 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
525 if (vfsp->vfs_do_relatime)
526 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
527 if (vfsp->vfs_do_nbmand)
528 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
529
530 return (0);
531
532 unregister:
533 dsl_prop_unregister_all(ds, zfsvfs);
534 return (error);
535 }
536
537 static int
538 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
539 uint64_t *userp, uint64_t *groupp)
540 {
541 /*
542 * Is it a valid type of object to track?
543 */
544 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
545 return (SET_ERROR(ENOENT));
546
547 /*
548 * If we have a NULL data pointer
549 * then assume the id's aren't changing and
550 * return EEXIST to the dmu to let it know to
551 * use the same ids
552 */
553 if (data == NULL)
554 return (SET_ERROR(EEXIST));
555
556 if (bonustype == DMU_OT_ZNODE) {
557 znode_phys_t *znp = data;
558 *userp = znp->zp_uid;
559 *groupp = znp->zp_gid;
560 } else {
561 int hdrsize;
562 sa_hdr_phys_t *sap = data;
563 sa_hdr_phys_t sa = *sap;
564 boolean_t swap = B_FALSE;
565
566 ASSERT(bonustype == DMU_OT_SA);
567
568 if (sa.sa_magic == 0) {
569 /*
570 * This should only happen for newly created
571 * files that haven't had the znode data filled
572 * in yet.
573 */
574 *userp = 0;
575 *groupp = 0;
576 return (0);
577 }
578 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
579 sa.sa_magic = SA_MAGIC;
580 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
581 swap = B_TRUE;
582 } else {
583 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
584 }
585
586 hdrsize = sa_hdrsize(&sa);
587 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
588 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
589 SA_UID_OFFSET));
590 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
591 SA_GID_OFFSET));
592 if (swap) {
593 *userp = BSWAP_64(*userp);
594 *groupp = BSWAP_64(*groupp);
595 }
596 }
597 return (0);
598 }
599
600 static void
601 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
602 char *domainbuf, int buflen, uid_t *ridp)
603 {
604 uint64_t fuid;
605 const char *domain;
606
607 fuid = zfs_strtonum(fuidstr, NULL);
608
609 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
610 if (domain)
611 (void) strlcpy(domainbuf, domain, buflen);
612 else
613 domainbuf[0] = '\0';
614 *ridp = FUID_RID(fuid);
615 }
616
617 static uint64_t
618 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
619 {
620 switch (type) {
621 case ZFS_PROP_USERUSED:
622 case ZFS_PROP_USEROBJUSED:
623 return (DMU_USERUSED_OBJECT);
624 case ZFS_PROP_GROUPUSED:
625 case ZFS_PROP_GROUPOBJUSED:
626 return (DMU_GROUPUSED_OBJECT);
627 case ZFS_PROP_USERQUOTA:
628 return (zfsvfs->z_userquota_obj);
629 case ZFS_PROP_GROUPQUOTA:
630 return (zfsvfs->z_groupquota_obj);
631 case ZFS_PROP_USEROBJQUOTA:
632 return (zfsvfs->z_userobjquota_obj);
633 case ZFS_PROP_GROUPOBJQUOTA:
634 return (zfsvfs->z_groupobjquota_obj);
635 default:
636 return (ZFS_NO_OBJECT);
637 }
638 }
639
640 int
641 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
642 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
643 {
644 int error;
645 zap_cursor_t zc;
646 zap_attribute_t za;
647 zfs_useracct_t *buf = vbuf;
648 uint64_t obj;
649 int offset = 0;
650
651 if (!dmu_objset_userspace_present(zfsvfs->z_os))
652 return (SET_ERROR(ENOTSUP));
653
654 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
655 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA) &&
656 !dmu_objset_userobjspace_present(zfsvfs->z_os))
657 return (SET_ERROR(ENOTSUP));
658
659 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
660 if (obj == ZFS_NO_OBJECT) {
661 *bufsizep = 0;
662 return (0);
663 }
664
665 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED)
666 offset = DMU_OBJACCT_PREFIX_LEN;
667
668 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
669 (error = zap_cursor_retrieve(&zc, &za)) == 0;
670 zap_cursor_advance(&zc)) {
671 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
672 *bufsizep)
673 break;
674
675 /*
676 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
677 * when dealing with block quota and vice versa.
678 */
679 if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
680 DMU_OBJACCT_PREFIX_LEN) == 0))
681 continue;
682
683 fuidstr_to_sid(zfsvfs, za.za_name + offset,
684 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
685
686 buf->zu_space = za.za_first_integer;
687 buf++;
688 }
689 if (error == ENOENT)
690 error = 0;
691
692 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
693 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
694 *cookiep = zap_cursor_serialize(&zc);
695 zap_cursor_fini(&zc);
696 return (error);
697 }
698
699 /*
700 * buf must be big enough (eg, 32 bytes)
701 */
702 static int
703 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
704 char *buf, boolean_t addok)
705 {
706 uint64_t fuid;
707 int domainid = 0;
708
709 if (domain && domain[0]) {
710 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
711 if (domainid == -1)
712 return (SET_ERROR(ENOENT));
713 }
714 fuid = FUID_ENCODE(domainid, rid);
715 (void) sprintf(buf, "%llx", (longlong_t)fuid);
716 return (0);
717 }
718
719 int
720 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
721 const char *domain, uint64_t rid, uint64_t *valp)
722 {
723 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
724 int offset = 0;
725 int err;
726 uint64_t obj;
727
728 *valp = 0;
729
730 if (!dmu_objset_userspace_present(zfsvfs->z_os))
731 return (SET_ERROR(ENOTSUP));
732
733 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
734 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA) &&
735 !dmu_objset_userobjspace_present(zfsvfs->z_os))
736 return (SET_ERROR(ENOTSUP));
737
738 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
739 if (obj == ZFS_NO_OBJECT)
740 return (0);
741
742 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED) {
743 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
744 offset = DMU_OBJACCT_PREFIX_LEN;
745 }
746
747 err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
748 if (err)
749 return (err);
750
751 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
752 if (err == ENOENT)
753 err = 0;
754 return (err);
755 }
756
757 int
758 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
759 const char *domain, uint64_t rid, uint64_t quota)
760 {
761 char buf[32];
762 int err;
763 dmu_tx_t *tx;
764 uint64_t *objp;
765 boolean_t fuid_dirtied;
766
767 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
768 return (SET_ERROR(ENOTSUP));
769
770 switch (type) {
771 case ZFS_PROP_USERQUOTA:
772 objp = &zfsvfs->z_userquota_obj;
773 break;
774 case ZFS_PROP_GROUPQUOTA:
775 objp = &zfsvfs->z_groupquota_obj;
776 break;
777 case ZFS_PROP_USEROBJQUOTA:
778 objp = &zfsvfs->z_userobjquota_obj;
779 break;
780 case ZFS_PROP_GROUPOBJQUOTA:
781 objp = &zfsvfs->z_groupobjquota_obj;
782 break;
783 default:
784 return (SET_ERROR(EINVAL));
785 }
786
787 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
788 if (err)
789 return (err);
790 fuid_dirtied = zfsvfs->z_fuid_dirty;
791
792 tx = dmu_tx_create(zfsvfs->z_os);
793 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
794 if (*objp == 0) {
795 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
796 zfs_userquota_prop_prefixes[type]);
797 }
798 if (fuid_dirtied)
799 zfs_fuid_txhold(zfsvfs, tx);
800 err = dmu_tx_assign(tx, TXG_WAIT);
801 if (err) {
802 dmu_tx_abort(tx);
803 return (err);
804 }
805
806 mutex_enter(&zfsvfs->z_lock);
807 if (*objp == 0) {
808 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
809 DMU_OT_NONE, 0, tx);
810 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
811 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
812 }
813 mutex_exit(&zfsvfs->z_lock);
814
815 if (quota == 0) {
816 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
817 if (err == ENOENT)
818 err = 0;
819 } else {
820 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
821 }
822 ASSERT(err == 0);
823 if (fuid_dirtied)
824 zfs_fuid_sync(zfsvfs, tx);
825 dmu_tx_commit(tx);
826 return (err);
827 }
828
829 boolean_t
830 zfs_fuid_overobjquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
831 {
832 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
833 uint64_t used, quota, usedobj, quotaobj;
834 int err;
835
836 if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
837 if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os)) {
838 dsl_pool_config_enter(
839 dmu_objset_pool(zfsvfs->z_os), FTAG);
840 dmu_objset_userobjspace_upgrade(zfsvfs->z_os);
841 dsl_pool_config_exit(
842 dmu_objset_pool(zfsvfs->z_os), FTAG);
843 }
844 return (B_FALSE);
845 }
846
847 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
848 quotaobj = isgroup ? zfsvfs->z_groupobjquota_obj :
849 zfsvfs->z_userobjquota_obj;
850 if (quotaobj == 0 || zfsvfs->z_replay)
851 return (B_FALSE);
852
853 (void) sprintf(buf, "%llx", (longlong_t)fuid);
854 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
855 if (err != 0)
856 return (B_FALSE);
857
858 (void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)fuid);
859 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
860 if (err != 0)
861 return (B_FALSE);
862 return (used >= quota);
863 }
864
865 boolean_t
866 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
867 {
868 char buf[20];
869 uint64_t used, quota, usedobj, quotaobj;
870 int err;
871
872 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
873 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
874
875 if (quotaobj == 0 || zfsvfs->z_replay)
876 return (B_FALSE);
877
878 (void) sprintf(buf, "%llx", (longlong_t)fuid);
879 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
880 if (err != 0)
881 return (B_FALSE);
882
883 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
884 if (err != 0)
885 return (B_FALSE);
886 return (used >= quota);
887 }
888
889 boolean_t
890 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
891 {
892 uint64_t fuid;
893 uint64_t quotaobj;
894 struct inode *ip = ZTOI(zp);
895
896 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
897
898 fuid = isgroup ? KGID_TO_SGID(ip->i_gid) : KUID_TO_SUID(ip->i_uid);
899
900 if (quotaobj == 0 || zfsvfs->z_replay)
901 return (B_FALSE);
902
903 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
904 }
905
906 /*
907 * Associate this zfsvfs with the given objset, which must be owned.
908 * This will cache a bunch of on-disk state from the objset in the
909 * zfsvfs.
910 */
911 static int
912 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
913 {
914 int error;
915 uint64_t val;
916
917 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
918 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
919 zfsvfs->z_os = os;
920
921 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
922 if (error != 0)
923 return (error);
924 if (zfsvfs->z_version >
925 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
926 (void) printk("Can't mount a version %lld file system "
927 "on a version %lld pool\n. Pool must be upgraded to mount "
928 "this file system.", (u_longlong_t)zfsvfs->z_version,
929 (u_longlong_t)spa_version(dmu_objset_spa(os)));
930 return (SET_ERROR(ENOTSUP));
931 }
932 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
933 if (error != 0)
934 return (error);
935 zfsvfs->z_norm = (int)val;
936
937 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
938 if (error != 0)
939 return (error);
940 zfsvfs->z_utf8 = (val != 0);
941
942 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
943 if (error != 0)
944 return (error);
945 zfsvfs->z_case = (uint_t)val;
946
947 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
948 return (error);
949 zfsvfs->z_acl_type = (uint_t)val;
950
951 /*
952 * Fold case on file systems that are always or sometimes case
953 * insensitive.
954 */
955 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
956 zfsvfs->z_case == ZFS_CASE_MIXED)
957 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
958
959 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
960 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
961
962 uint64_t sa_obj = 0;
963 if (zfsvfs->z_use_sa) {
964 /* should either have both of these objects or none */
965 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
966 &sa_obj);
967 if (error != 0)
968 return (error);
969
970 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
971 if ((error == 0) && (val == ZFS_XATTR_SA))
972 zfsvfs->z_xattr_sa = B_TRUE;
973 }
974
975 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
976 &zfsvfs->z_attr_table);
977 if (error != 0)
978 return (error);
979
980 if (zfsvfs->z_version >= ZPL_VERSION_SA)
981 sa_register_update_callback(os, zfs_sa_upgrade);
982
983 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
984 &zfsvfs->z_root);
985 if (error != 0)
986 return (error);
987 ASSERT(zfsvfs->z_root != 0);
988
989 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
990 &zfsvfs->z_unlinkedobj);
991 if (error != 0)
992 return (error);
993
994 error = zap_lookup(os, MASTER_NODE_OBJ,
995 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
996 8, 1, &zfsvfs->z_userquota_obj);
997 if (error == ENOENT)
998 zfsvfs->z_userquota_obj = 0;
999 else if (error != 0)
1000 return (error);
1001
1002 error = zap_lookup(os, MASTER_NODE_OBJ,
1003 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1004 8, 1, &zfsvfs->z_groupquota_obj);
1005 if (error == ENOENT)
1006 zfsvfs->z_groupquota_obj = 0;
1007 else if (error != 0)
1008 return (error);
1009
1010 error = zap_lookup(os, MASTER_NODE_OBJ,
1011 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1012 8, 1, &zfsvfs->z_userobjquota_obj);
1013 if (error == ENOENT)
1014 zfsvfs->z_userobjquota_obj = 0;
1015 else if (error != 0)
1016 return (error);
1017
1018 error = zap_lookup(os, MASTER_NODE_OBJ,
1019 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1020 8, 1, &zfsvfs->z_groupobjquota_obj);
1021 if (error == ENOENT)
1022 zfsvfs->z_groupobjquota_obj = 0;
1023 else if (error != 0)
1024 return (error);
1025
1026 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1027 &zfsvfs->z_fuid_obj);
1028 if (error == ENOENT)
1029 zfsvfs->z_fuid_obj = 0;
1030 else if (error != 0)
1031 return (error);
1032
1033 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1034 &zfsvfs->z_shares_dir);
1035 if (error == ENOENT)
1036 zfsvfs->z_shares_dir = 0;
1037 else if (error != 0)
1038 return (error);
1039
1040 return (0);
1041 }
1042
1043 int
1044 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
1045 {
1046 objset_t *os;
1047 zfsvfs_t *zfsvfs;
1048 int error;
1049
1050 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1051
1052 /*
1053 * We claim to always be readonly so we can open snapshots;
1054 * other ZPL code will prevent us from writing to snapshots.
1055 */
1056
1057 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, B_TRUE,
1058 zfsvfs, &os);
1059 if (error != 0) {
1060 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1061 return (error);
1062 }
1063
1064 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1065 if (error != 0) {
1066 dmu_objset_disown(os, B_TRUE, zfsvfs);
1067 }
1068 return (error);
1069 }
1070
1071 int
1072 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1073 {
1074 int error;
1075
1076 zfsvfs->z_vfs = NULL;
1077 zfsvfs->z_sb = NULL;
1078 zfsvfs->z_parent = zfsvfs;
1079
1080 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1081 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1082 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1083 offsetof(znode_t, z_link_node));
1084 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1085 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1086 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1087
1088 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
1089 ZFS_OBJ_MTX_MAX);
1090 zfsvfs->z_hold_size = size;
1091 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1092 KM_SLEEP);
1093 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1094 for (int i = 0; i != size; i++) {
1095 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1096 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1097 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1098 }
1099
1100 error = zfsvfs_init(zfsvfs, os);
1101 if (error != 0) {
1102 *zfvp = NULL;
1103 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1104 return (error);
1105 }
1106
1107 *zfvp = zfsvfs;
1108 return (0);
1109 }
1110
1111 static int
1112 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1113 {
1114 int error;
1115 boolean_t readonly = zfs_is_readonly(zfsvfs);
1116
1117 /*
1118 * Check for a bad on-disk format version now since we
1119 * lied about owning the dataset readonly before.
1120 */
1121 if (!readonly &&
1122 dmu_objset_incompatible_encryption_version(zfsvfs->z_os))
1123 return (SET_ERROR(EROFS));
1124
1125 error = zfs_register_callbacks(zfsvfs->z_vfs);
1126 if (error)
1127 return (error);
1128
1129 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1130
1131 /*
1132 * If we are not mounting (ie: online recv), then we don't
1133 * have to worry about replaying the log as we blocked all
1134 * operations out since we closed the ZIL.
1135 */
1136 if (mounting) {
1137 /*
1138 * During replay we remove the read only flag to
1139 * allow replays to succeed.
1140 */
1141 if (readonly != 0)
1142 readonly_changed_cb(zfsvfs, B_FALSE);
1143 else
1144 zfs_unlinked_drain(zfsvfs);
1145
1146 /*
1147 * Parse and replay the intent log.
1148 *
1149 * Because of ziltest, this must be done after
1150 * zfs_unlinked_drain(). (Further note: ziltest
1151 * doesn't use readonly mounts, where
1152 * zfs_unlinked_drain() isn't called.) This is because
1153 * ziltest causes spa_sync() to think it's committed,
1154 * but actually it is not, so the intent log contains
1155 * many txg's worth of changes.
1156 *
1157 * In particular, if object N is in the unlinked set in
1158 * the last txg to actually sync, then it could be
1159 * actually freed in a later txg and then reallocated
1160 * in a yet later txg. This would write a "create
1161 * object N" record to the intent log. Normally, this
1162 * would be fine because the spa_sync() would have
1163 * written out the fact that object N is free, before
1164 * we could write the "create object N" intent log
1165 * record.
1166 *
1167 * But when we are in ziltest mode, we advance the "open
1168 * txg" without actually spa_sync()-ing the changes to
1169 * disk. So we would see that object N is still
1170 * allocated and in the unlinked set, and there is an
1171 * intent log record saying to allocate it.
1172 */
1173 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1174 if (zil_replay_disable) {
1175 zil_destroy(zfsvfs->z_log, B_FALSE);
1176 } else {
1177 zfsvfs->z_replay = B_TRUE;
1178 zil_replay(zfsvfs->z_os, zfsvfs,
1179 zfs_replay_vector);
1180 zfsvfs->z_replay = B_FALSE;
1181 }
1182 }
1183
1184 /* restore readonly bit */
1185 if (readonly != 0)
1186 readonly_changed_cb(zfsvfs, B_TRUE);
1187 }
1188
1189 /*
1190 * Set the objset user_ptr to track its zfsvfs.
1191 */
1192 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1193 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1194 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1195
1196 return (0);
1197 }
1198
1199 void
1200 zfsvfs_free(zfsvfs_t *zfsvfs)
1201 {
1202 int i, size = zfsvfs->z_hold_size;
1203
1204 zfs_fuid_destroy(zfsvfs);
1205
1206 mutex_destroy(&zfsvfs->z_znodes_lock);
1207 mutex_destroy(&zfsvfs->z_lock);
1208 list_destroy(&zfsvfs->z_all_znodes);
1209 rrm_destroy(&zfsvfs->z_teardown_lock);
1210 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1211 rw_destroy(&zfsvfs->z_fuid_lock);
1212 for (i = 0; i != size; i++) {
1213 avl_destroy(&zfsvfs->z_hold_trees[i]);
1214 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1215 }
1216 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1217 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1218 zfsvfs_vfs_free(zfsvfs->z_vfs);
1219 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1220 }
1221
1222 static void
1223 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1224 {
1225 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1226 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1227 }
1228
1229 void
1230 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1231 {
1232 objset_t *os = zfsvfs->z_os;
1233
1234 if (!dmu_objset_is_snapshot(os))
1235 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1236 }
1237
1238 #ifdef HAVE_MLSLABEL
1239 /*
1240 * Check that the hex label string is appropriate for the dataset being
1241 * mounted into the global_zone proper.
1242 *
1243 * Return an error if the hex label string is not default or
1244 * admin_low/admin_high. For admin_low labels, the corresponding
1245 * dataset must be readonly.
1246 */
1247 int
1248 zfs_check_global_label(const char *dsname, const char *hexsl)
1249 {
1250 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1251 return (0);
1252 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1253 return (0);
1254 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1255 /* must be readonly */
1256 uint64_t rdonly;
1257
1258 if (dsl_prop_get_integer(dsname,
1259 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1260 return (SET_ERROR(EACCES));
1261 return (rdonly ? 0 : EACCES);
1262 }
1263 return (SET_ERROR(EACCES));
1264 }
1265 #endif /* HAVE_MLSLABEL */
1266
1267 int
1268 zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
1269 {
1270 zfsvfs_t *zfsvfs = dentry->d_sb->s_fs_info;
1271 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1272 uint64_t fsid;
1273 uint32_t bshift;
1274
1275 ZFS_ENTER(zfsvfs);
1276
1277 dmu_objset_space(zfsvfs->z_os,
1278 &refdbytes, &availbytes, &usedobjs, &availobjs);
1279
1280 fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1281 /*
1282 * The underlying storage pool actually uses multiple block
1283 * size. Under Solaris frsize (fragment size) is reported as
1284 * the smallest block size we support, and bsize (block size)
1285 * as the filesystem's maximum block size. Unfortunately,
1286 * under Linux the fragment size and block size are often used
1287 * interchangeably. Thus we are forced to report both of them
1288 * as the filesystem's maximum block size.
1289 */
1290 statp->f_frsize = zfsvfs->z_max_blksz;
1291 statp->f_bsize = zfsvfs->z_max_blksz;
1292 bshift = fls(statp->f_bsize) - 1;
1293
1294 /*
1295 * The following report "total" blocks of various kinds in
1296 * the file system, but reported in terms of f_bsize - the
1297 * "preferred" size.
1298 */
1299
1300 statp->f_blocks = (refdbytes + availbytes) >> bshift;
1301 statp->f_bfree = availbytes >> bshift;
1302 statp->f_bavail = statp->f_bfree; /* no root reservation */
1303
1304 /*
1305 * statvfs() should really be called statufs(), because it assumes
1306 * static metadata. ZFS doesn't preallocate files, so the best
1307 * we can do is report the max that could possibly fit in f_files,
1308 * and that minus the number actually used in f_ffree.
1309 * For f_ffree, report the smaller of the number of object available
1310 * and the number of blocks (each object will take at least a block).
1311 */
1312 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1313 statp->f_files = statp->f_ffree + usedobjs;
1314 statp->f_fsid.val[0] = (uint32_t)fsid;
1315 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1316 statp->f_type = ZFS_SUPER_MAGIC;
1317 statp->f_namelen = MAXNAMELEN - 1;
1318
1319 /*
1320 * We have all of 40 characters to stuff a string here.
1321 * Is there anything useful we could/should provide?
1322 */
1323 bzero(statp->f_spare, sizeof (statp->f_spare));
1324
1325 ZFS_EXIT(zfsvfs);
1326 return (0);
1327 }
1328
1329 int
1330 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1331 {
1332 znode_t *rootzp;
1333 int error;
1334
1335 ZFS_ENTER(zfsvfs);
1336
1337 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1338 if (error == 0)
1339 *ipp = ZTOI(rootzp);
1340
1341 ZFS_EXIT(zfsvfs);
1342 return (error);
1343 }
1344
1345 #ifdef HAVE_D_PRUNE_ALIASES
1346 /*
1347 * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1348 * To accommodate this we must improvise and manually walk the list of znodes
1349 * attempting to prune dentries in order to be able to drop the inodes.
1350 *
1351 * To avoid scanning the same znodes multiple times they are always rotated
1352 * to the end of the z_all_znodes list. New znodes are inserted at the
1353 * end of the list so we're always scanning the oldest znodes first.
1354 */
1355 static int
1356 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1357 {
1358 znode_t **zp_array, *zp;
1359 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1360 int objects = 0;
1361 int i = 0, j = 0;
1362
1363 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1364
1365 mutex_enter(&zfsvfs->z_znodes_lock);
1366 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1367
1368 if ((i++ > nr_to_scan) || (j >= max_array))
1369 break;
1370
1371 ASSERT(list_link_active(&zp->z_link_node));
1372 list_remove(&zfsvfs->z_all_znodes, zp);
1373 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1374
1375 /* Skip active znodes and .zfs entries */
1376 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1377 continue;
1378
1379 if (igrab(ZTOI(zp)) == NULL)
1380 continue;
1381
1382 zp_array[j] = zp;
1383 j++;
1384 }
1385 mutex_exit(&zfsvfs->z_znodes_lock);
1386
1387 for (i = 0; i < j; i++) {
1388 zp = zp_array[i];
1389
1390 ASSERT3P(zp, !=, NULL);
1391 d_prune_aliases(ZTOI(zp));
1392
1393 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1394 objects++;
1395
1396 iput(ZTOI(zp));
1397 }
1398
1399 kmem_free(zp_array, max_array * sizeof (znode_t *));
1400
1401 return (objects);
1402 }
1403 #endif /* HAVE_D_PRUNE_ALIASES */
1404
1405 /*
1406 * The ARC has requested that the filesystem drop entries from the dentry
1407 * and inode caches. This can occur when the ARC needs to free meta data
1408 * blocks but can't because they are all pinned by entries in these caches.
1409 */
1410 int
1411 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1412 {
1413 zfsvfs_t *zfsvfs = sb->s_fs_info;
1414 int error = 0;
1415 #if defined(HAVE_SHRINK) || defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1416 struct shrinker *shrinker = &sb->s_shrink;
1417 struct shrink_control sc = {
1418 .nr_to_scan = nr_to_scan,
1419 .gfp_mask = GFP_KERNEL,
1420 };
1421 #endif
1422
1423 ZFS_ENTER(zfsvfs);
1424
1425 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1426 defined(SHRINK_CONTROL_HAS_NID) && \
1427 defined(SHRINKER_NUMA_AWARE)
1428 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1429 *objects = 0;
1430 for_each_online_node(sc.nid) {
1431 *objects += (*shrinker->scan_objects)(shrinker, &sc);
1432 }
1433 } else {
1434 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1435 }
1436
1437 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1438 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1439 #elif defined(HAVE_SHRINK)
1440 *objects = (*shrinker->shrink)(shrinker, &sc);
1441 #elif defined(HAVE_D_PRUNE_ALIASES)
1442 #define D_PRUNE_ALIASES_IS_DEFAULT
1443 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1444 #else
1445 #error "No available dentry and inode cache pruning mechanism."
1446 #endif
1447
1448 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1449 #undef D_PRUNE_ALIASES_IS_DEFAULT
1450 /*
1451 * Fall back to zfs_prune_aliases if the kernel's per-superblock
1452 * shrinker couldn't free anything, possibly due to the inodes being
1453 * allocated in a different memcg.
1454 */
1455 if (*objects == 0)
1456 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1457 #endif
1458
1459 ZFS_EXIT(zfsvfs);
1460
1461 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1462 "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1463 nr_to_scan, *objects, error);
1464
1465 return (error);
1466 }
1467
1468 /*
1469 * Teardown the zfsvfs_t.
1470 *
1471 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1472 * and 'z_teardown_inactive_lock' held.
1473 */
1474 static int
1475 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1476 {
1477 znode_t *zp;
1478
1479 /*
1480 * If someone has not already unmounted this file system,
1481 * drain the iput_taskq to ensure all active references to the
1482 * zfsvfs_t have been handled only then can it be safely destroyed.
1483 */
1484 if (zfsvfs->z_os) {
1485 /*
1486 * If we're unmounting we have to wait for the list to
1487 * drain completely.
1488 *
1489 * If we're not unmounting there's no guarantee the list
1490 * will drain completely, but iputs run from the taskq
1491 * may add the parents of dir-based xattrs to the taskq
1492 * so we want to wait for these.
1493 *
1494 * We can safely read z_nr_znodes without locking because the
1495 * VFS has already blocked operations which add to the
1496 * z_all_znodes list and thus increment z_nr_znodes.
1497 */
1498 int round = 0;
1499 while (zfsvfs->z_nr_znodes > 0) {
1500 taskq_wait_outstanding(dsl_pool_iput_taskq(
1501 dmu_objset_pool(zfsvfs->z_os)), 0);
1502 if (++round > 1 && !unmounting)
1503 break;
1504 }
1505 }
1506
1507 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1508
1509 if (!unmounting) {
1510 /*
1511 * We purge the parent filesystem's super block as the
1512 * parent filesystem and all of its snapshots have their
1513 * inode's super block set to the parent's filesystem's
1514 * super block. Note, 'z_parent' is self referential
1515 * for non-snapshots.
1516 */
1517 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1518 }
1519
1520 /*
1521 * Close the zil. NB: Can't close the zil while zfs_inactive
1522 * threads are blocked as zil_close can call zfs_inactive.
1523 */
1524 if (zfsvfs->z_log) {
1525 zil_close(zfsvfs->z_log);
1526 zfsvfs->z_log = NULL;
1527 }
1528
1529 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1530
1531 /*
1532 * If we are not unmounting (ie: online recv) and someone already
1533 * unmounted this file system while we were doing the switcheroo,
1534 * or a reopen of z_os failed then just bail out now.
1535 */
1536 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1537 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1538 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1539 return (SET_ERROR(EIO));
1540 }
1541
1542 /*
1543 * At this point there are no VFS ops active, and any new VFS ops
1544 * will fail with EIO since we have z_teardown_lock for writer (only
1545 * relevant for forced unmount).
1546 *
1547 * Release all holds on dbufs.
1548 */
1549 if (!unmounting) {
1550 mutex_enter(&zfsvfs->z_znodes_lock);
1551 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1552 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1553 if (zp->z_sa_hdl)
1554 zfs_znode_dmu_fini(zp);
1555 }
1556 mutex_exit(&zfsvfs->z_znodes_lock);
1557 }
1558
1559 /*
1560 * If we are unmounting, set the unmounted flag and let new VFS ops
1561 * unblock. zfs_inactive will have the unmounted behavior, and all
1562 * other VFS ops will fail with EIO.
1563 */
1564 if (unmounting) {
1565 zfsvfs->z_unmounted = B_TRUE;
1566 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1567 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1568 }
1569
1570 /*
1571 * z_os will be NULL if there was an error in attempting to reopen
1572 * zfsvfs, so just return as the properties had already been
1573 *
1574 * unregistered and cached data had been evicted before.
1575 */
1576 if (zfsvfs->z_os == NULL)
1577 return (0);
1578
1579 /*
1580 * Unregister properties.
1581 */
1582 zfs_unregister_callbacks(zfsvfs);
1583
1584 /*
1585 * Evict cached data
1586 */
1587 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1588 !zfs_is_readonly(zfsvfs))
1589 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1590 dmu_objset_evict_dbufs(zfsvfs->z_os);
1591
1592 return (0);
1593 }
1594
1595 #if !defined(HAVE_2ARGS_BDI_SETUP_AND_REGISTER) && \
1596 !defined(HAVE_3ARGS_BDI_SETUP_AND_REGISTER)
1597 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1598 #endif
1599
1600 int
1601 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1602 {
1603 const char *osname = zm->mnt_osname;
1604 struct inode *root_inode;
1605 uint64_t recordsize;
1606 int error = 0;
1607 zfsvfs_t *zfsvfs;
1608
1609 ASSERT(zm);
1610 ASSERT(osname);
1611
1612 error = zfsvfs_create(osname, &zfsvfs);
1613 if (error)
1614 return (error);
1615
1616 error = zfsvfs_parse_options(zm->mnt_data, &zfsvfs->z_vfs);
1617 if (error)
1618 goto out;
1619
1620 if ((error = dsl_prop_get_integer(osname, "recordsize",
1621 &recordsize, NULL)))
1622 goto out;
1623
1624 zfsvfs->z_vfs->vfs_data = zfsvfs;
1625 zfsvfs->z_sb = sb;
1626 sb->s_fs_info = zfsvfs;
1627 sb->s_magic = ZFS_SUPER_MAGIC;
1628 sb->s_maxbytes = MAX_LFS_FILESIZE;
1629 sb->s_time_gran = 1;
1630 sb->s_blocksize = recordsize;
1631 sb->s_blocksize_bits = ilog2(recordsize);
1632
1633 error = -zpl_bdi_setup(sb, "zfs");
1634 if (error)
1635 goto out;
1636
1637 sb->s_bdi->ra_pages = 0;
1638
1639 /* Set callback operations for the file system. */
1640 sb->s_op = &zpl_super_operations;
1641 sb->s_xattr = zpl_xattr_handlers;
1642 sb->s_export_op = &zpl_export_operations;
1643 #ifdef HAVE_S_D_OP
1644 sb->s_d_op = &zpl_dentry_operations;
1645 #endif /* HAVE_S_D_OP */
1646
1647 /* Set features for file system. */
1648 zfs_set_fuid_feature(zfsvfs);
1649
1650 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1651 uint64_t pval;
1652
1653 atime_changed_cb(zfsvfs, B_FALSE);
1654 readonly_changed_cb(zfsvfs, B_TRUE);
1655 if ((error = dsl_prop_get_integer(osname,
1656 "xattr", &pval, NULL)))
1657 goto out;
1658 xattr_changed_cb(zfsvfs, pval);
1659 if ((error = dsl_prop_get_integer(osname,
1660 "acltype", &pval, NULL)))
1661 goto out;
1662 acltype_changed_cb(zfsvfs, pval);
1663 zfsvfs->z_issnap = B_TRUE;
1664 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1665 zfsvfs->z_snap_defer_time = jiffies;
1666
1667 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1668 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1669 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1670 } else {
1671 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1672 goto out;
1673 }
1674
1675 /* Allocate a root inode for the filesystem. */
1676 error = zfs_root(zfsvfs, &root_inode);
1677 if (error) {
1678 (void) zfs_umount(sb);
1679 goto out;
1680 }
1681
1682 /* Allocate a root dentry for the filesystem */
1683 sb->s_root = d_make_root(root_inode);
1684 if (sb->s_root == NULL) {
1685 (void) zfs_umount(sb);
1686 error = SET_ERROR(ENOMEM);
1687 goto out;
1688 }
1689
1690 if (!zfsvfs->z_issnap)
1691 zfsctl_create(zfsvfs);
1692
1693 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1694 out:
1695 if (error) {
1696 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1697 zfsvfs_free(zfsvfs);
1698 /*
1699 * make sure we don't have dangling sb->s_fs_info which
1700 * zfs_preumount will use.
1701 */
1702 sb->s_fs_info = NULL;
1703 }
1704
1705 return (error);
1706 }
1707
1708 /*
1709 * Called when an unmount is requested and certain sanity checks have
1710 * already passed. At this point no dentries or inodes have been reclaimed
1711 * from their respective caches. We drop the extra reference on the .zfs
1712 * control directory to allow everything to be reclaimed. All snapshots
1713 * must already have been unmounted to reach this point.
1714 */
1715 void
1716 zfs_preumount(struct super_block *sb)
1717 {
1718 zfsvfs_t *zfsvfs = sb->s_fs_info;
1719
1720 /* zfsvfs is NULL when zfs_domount fails during mount */
1721 if (zfsvfs) {
1722 zfsctl_destroy(sb->s_fs_info);
1723 /*
1724 * Wait for iput_async before entering evict_inodes in
1725 * generic_shutdown_super. The reason we must finish before
1726 * evict_inodes is when lazytime is on, or when zfs_purgedir
1727 * calls zfs_zget, iput would bump i_count from 0 to 1. This
1728 * would race with the i_count check in evict_inodes. This means
1729 * it could destroy the inode while we are still using it.
1730 *
1731 * We wait for two passes. xattr directories in the first pass
1732 * may add xattr entries in zfs_purgedir, so in the second pass
1733 * we wait for them. We don't use taskq_wait here because it is
1734 * a pool wide taskq. Other mounted filesystems can constantly
1735 * do iput_async and there's no guarantee when taskq will be
1736 * empty.
1737 */
1738 taskq_wait_outstanding(dsl_pool_iput_taskq(
1739 dmu_objset_pool(zfsvfs->z_os)), 0);
1740 taskq_wait_outstanding(dsl_pool_iput_taskq(
1741 dmu_objset_pool(zfsvfs->z_os)), 0);
1742 }
1743 }
1744
1745 /*
1746 * Called once all other unmount released tear down has occurred.
1747 * It is our responsibility to release any remaining infrastructure.
1748 */
1749 /*ARGSUSED*/
1750 int
1751 zfs_umount(struct super_block *sb)
1752 {
1753 zfsvfs_t *zfsvfs = sb->s_fs_info;
1754 objset_t *os;
1755
1756 if (zfsvfs->z_arc_prune != NULL)
1757 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1758 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1759 os = zfsvfs->z_os;
1760 zpl_bdi_destroy(sb);
1761
1762 /*
1763 * z_os will be NULL if there was an error in
1764 * attempting to reopen zfsvfs.
1765 */
1766 if (os != NULL) {
1767 /*
1768 * Unset the objset user_ptr.
1769 */
1770 mutex_enter(&os->os_user_ptr_lock);
1771 dmu_objset_set_user(os, NULL);
1772 mutex_exit(&os->os_user_ptr_lock);
1773
1774 /*
1775 * Finally release the objset
1776 */
1777 dmu_objset_disown(os, B_TRUE, zfsvfs);
1778 }
1779
1780 zfsvfs_free(zfsvfs);
1781 return (0);
1782 }
1783
1784 int
1785 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1786 {
1787 zfsvfs_t *zfsvfs = sb->s_fs_info;
1788 vfs_t *vfsp;
1789 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1790 int error;
1791
1792 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1793 !(*flags & MS_RDONLY)) {
1794 *flags |= MS_RDONLY;
1795 return (EROFS);
1796 }
1797
1798 error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1799 if (error)
1800 return (error);
1801
1802 zfs_unregister_callbacks(zfsvfs);
1803 zfsvfs_vfs_free(zfsvfs->z_vfs);
1804
1805 vfsp->vfs_data = zfsvfs;
1806 zfsvfs->z_vfs = vfsp;
1807 if (!issnap)
1808 (void) zfs_register_callbacks(vfsp);
1809
1810 return (error);
1811 }
1812
1813 int
1814 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1815 {
1816 zfsvfs_t *zfsvfs = sb->s_fs_info;
1817 znode_t *zp;
1818 uint64_t object = 0;
1819 uint64_t fid_gen = 0;
1820 uint64_t gen_mask;
1821 uint64_t zp_gen;
1822 int i, err;
1823
1824 *ipp = NULL;
1825
1826 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1827 zfid_short_t *zfid = (zfid_short_t *)fidp;
1828
1829 for (i = 0; i < sizeof (zfid->zf_object); i++)
1830 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1831
1832 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1833 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1834 } else {
1835 return (SET_ERROR(EINVAL));
1836 }
1837
1838 /* LONG_FID_LEN means snapdirs */
1839 if (fidp->fid_len == LONG_FID_LEN) {
1840 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1841 uint64_t objsetid = 0;
1842 uint64_t setgen = 0;
1843
1844 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1845 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1846
1847 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1848 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1849
1850 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1851 dprintf("snapdir fid: objsetid (%llu) != "
1852 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1853 objsetid, ZFSCTL_INO_SNAPDIRS, object);
1854
1855 return (SET_ERROR(EINVAL));
1856 }
1857
1858 if (fid_gen > 1 || setgen != 0) {
1859 dprintf("snapdir fid: fid_gen (%llu) and setgen "
1860 "(%llu)\n", fid_gen, setgen);
1861 return (SET_ERROR(EINVAL));
1862 }
1863
1864 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1865 }
1866
1867 ZFS_ENTER(zfsvfs);
1868 /* A zero fid_gen means we are in the .zfs control directories */
1869 if (fid_gen == 0 &&
1870 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1871 *ipp = zfsvfs->z_ctldir;
1872 ASSERT(*ipp != NULL);
1873 if (object == ZFSCTL_INO_SNAPDIR) {
1874 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1875 0, kcred, NULL, NULL) == 0);
1876 } else {
1877 igrab(*ipp);
1878 }
1879 ZFS_EXIT(zfsvfs);
1880 return (0);
1881 }
1882
1883 gen_mask = -1ULL >> (64 - 8 * i);
1884
1885 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1886 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1887 ZFS_EXIT(zfsvfs);
1888 return (err);
1889 }
1890
1891 /* Don't export xattr stuff */
1892 if (zp->z_pflags & ZFS_XATTR) {
1893 iput(ZTOI(zp));
1894 ZFS_EXIT(zfsvfs);
1895 return (SET_ERROR(ENOENT));
1896 }
1897
1898 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1899 sizeof (uint64_t));
1900 zp_gen = zp_gen & gen_mask;
1901 if (zp_gen == 0)
1902 zp_gen = 1;
1903 if ((fid_gen == 0) && (zfsvfs->z_root == object))
1904 fid_gen = zp_gen;
1905 if (zp->z_unlinked || zp_gen != fid_gen) {
1906 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1907 fid_gen);
1908 iput(ZTOI(zp));
1909 ZFS_EXIT(zfsvfs);
1910 return (SET_ERROR(ENOENT));
1911 }
1912
1913 *ipp = ZTOI(zp);
1914 if (*ipp)
1915 zfs_inode_update(ITOZ(*ipp));
1916
1917 ZFS_EXIT(zfsvfs);
1918 return (0);
1919 }
1920
1921 /*
1922 * Block out VFS ops and close zfsvfs_t
1923 *
1924 * Note, if successful, then we return with the 'z_teardown_lock' and
1925 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
1926 * dataset and objset intact so that they can be atomically handed off during
1927 * a subsequent rollback or recv operation and the resume thereafter.
1928 */
1929 int
1930 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1931 {
1932 int error;
1933
1934 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1935 return (error);
1936
1937 return (0);
1938 }
1939
1940 /*
1941 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
1942 * is an invariant across any of the operations that can be performed while the
1943 * filesystem was suspended. Whether it succeeded or failed, the preconditions
1944 * are the same: the relevant objset and associated dataset are owned by
1945 * zfsvfs, held, and long held on entry.
1946 */
1947 int
1948 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1949 {
1950 int err, err2;
1951 znode_t *zp;
1952
1953 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
1954 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1955
1956 /*
1957 * We already own this, so just update the objset_t, as the one we
1958 * had before may have been evicted.
1959 */
1960 objset_t *os;
1961 VERIFY3P(ds->ds_owner, ==, zfsvfs);
1962 VERIFY(dsl_dataset_long_held(ds));
1963 VERIFY0(dmu_objset_from_ds(ds, &os));
1964
1965 err = zfsvfs_init(zfsvfs, os);
1966 if (err != 0)
1967 goto bail;
1968
1969 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1970
1971 zfs_set_fuid_feature(zfsvfs);
1972 zfsvfs->z_rollback_time = jiffies;
1973
1974 /*
1975 * Attempt to re-establish all the active inodes with their
1976 * dbufs. If a zfs_rezget() fails, then we unhash the inode
1977 * and mark it stale. This prevents a collision if a new
1978 * inode/object is created which must use the same inode
1979 * number. The stale inode will be be released when the
1980 * VFS prunes the dentry holding the remaining references
1981 * on the stale inode.
1982 */
1983 mutex_enter(&zfsvfs->z_znodes_lock);
1984 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1985 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1986 err2 = zfs_rezget(zp);
1987 if (err2) {
1988 remove_inode_hash(ZTOI(zp));
1989 zp->z_is_stale = B_TRUE;
1990 }
1991 }
1992 mutex_exit(&zfsvfs->z_znodes_lock);
1993
1994 bail:
1995 /* release the VFS ops */
1996 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1997 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1998
1999 if (err) {
2000 /*
2001 * Since we couldn't setup the sa framework, try to force
2002 * unmount this file system.
2003 */
2004 if (zfsvfs->z_os)
2005 (void) zfs_umount(zfsvfs->z_sb);
2006 }
2007 return (err);
2008 }
2009
2010 int
2011 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2012 {
2013 int error;
2014 objset_t *os = zfsvfs->z_os;
2015 dmu_tx_t *tx;
2016
2017 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2018 return (SET_ERROR(EINVAL));
2019
2020 if (newvers < zfsvfs->z_version)
2021 return (SET_ERROR(EINVAL));
2022
2023 if (zfs_spa_version_map(newvers) >
2024 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2025 return (SET_ERROR(ENOTSUP));
2026
2027 tx = dmu_tx_create(os);
2028 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2029 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2030 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2031 ZFS_SA_ATTRS);
2032 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2033 }
2034 error = dmu_tx_assign(tx, TXG_WAIT);
2035 if (error) {
2036 dmu_tx_abort(tx);
2037 return (error);
2038 }
2039
2040 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2041 8, 1, &newvers, tx);
2042
2043 if (error) {
2044 dmu_tx_commit(tx);
2045 return (error);
2046 }
2047
2048 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2049 uint64_t sa_obj;
2050
2051 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2052 SPA_VERSION_SA);
2053 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2054 DMU_OT_NONE, 0, tx);
2055
2056 error = zap_add(os, MASTER_NODE_OBJ,
2057 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2058 ASSERT0(error);
2059
2060 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2061 sa_register_update_callback(os, zfs_sa_upgrade);
2062 }
2063
2064 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2065 "from %llu to %llu", zfsvfs->z_version, newvers);
2066
2067 dmu_tx_commit(tx);
2068
2069 zfsvfs->z_version = newvers;
2070
2071 zfs_set_fuid_feature(zfsvfs);
2072
2073 return (0);
2074 }
2075
2076 /*
2077 * Read a property stored within the master node.
2078 */
2079 int
2080 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2081 {
2082 const char *pname;
2083 int error = SET_ERROR(ENOENT);
2084
2085 /*
2086 * Look up the file system's value for the property. For the
2087 * version property, we look up a slightly different string.
2088 */
2089 if (prop == ZFS_PROP_VERSION)
2090 pname = ZPL_VERSION_STR;
2091 else
2092 pname = zfs_prop_to_name(prop);
2093
2094 if (os != NULL) {
2095 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2096 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2097 }
2098
2099 if (error == ENOENT) {
2100 /* No value set, use the default value */
2101 switch (prop) {
2102 case ZFS_PROP_VERSION:
2103 *value = ZPL_VERSION;
2104 break;
2105 case ZFS_PROP_NORMALIZE:
2106 case ZFS_PROP_UTF8ONLY:
2107 *value = 0;
2108 break;
2109 case ZFS_PROP_CASE:
2110 *value = ZFS_CASE_SENSITIVE;
2111 break;
2112 case ZFS_PROP_ACLTYPE:
2113 *value = ZFS_ACLTYPE_OFF;
2114 break;
2115 default:
2116 return (error);
2117 }
2118 error = 0;
2119 }
2120 return (error);
2121 }
2122
2123 /*
2124 * Return true if the coresponding vfs's unmounted flag is set.
2125 * Otherwise return false.
2126 * If this function returns true we know VFS unmount has been initiated.
2127 */
2128 boolean_t
2129 zfs_get_vfs_flag_unmounted(objset_t *os)
2130 {
2131 zfsvfs_t *zfvp;
2132 boolean_t unmounted = B_FALSE;
2133
2134 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2135
2136 mutex_enter(&os->os_user_ptr_lock);
2137 zfvp = dmu_objset_get_user(os);
2138 if (zfvp != NULL && zfvp->z_unmounted)
2139 unmounted = B_TRUE;
2140 mutex_exit(&os->os_user_ptr_lock);
2141
2142 return (unmounted);
2143 }
2144
2145 void
2146 zfs_init(void)
2147 {
2148 zfsctl_init();
2149 zfs_znode_init();
2150 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2151 register_filesystem(&zpl_fs_type);
2152 }
2153
2154 void
2155 zfs_fini(void)
2156 {
2157 /*
2158 * we don't use outstanding because zpl_posix_acl_free might add more.
2159 */
2160 taskq_wait(system_delay_taskq);
2161 taskq_wait(system_taskq);
2162 unregister_filesystem(&zpl_fs_type);
2163 zfs_znode_fini();
2164 zfsctl_fini();
2165 }
2166
2167 #if defined(_KERNEL) && defined(HAVE_SPL)
2168 EXPORT_SYMBOL(zfs_suspend_fs);
2169 EXPORT_SYMBOL(zfs_resume_fs);
2170 EXPORT_SYMBOL(zfs_userspace_one);
2171 EXPORT_SYMBOL(zfs_userspace_many);
2172 EXPORT_SYMBOL(zfs_set_userquota);
2173 EXPORT_SYMBOL(zfs_owner_overquota);
2174 EXPORT_SYMBOL(zfs_fuid_overquota);
2175 EXPORT_SYMBOL(zfs_fuid_overobjquota);
2176 EXPORT_SYMBOL(zfs_set_version);
2177 EXPORT_SYMBOL(zfsvfs_create);
2178 EXPORT_SYMBOL(zfsvfs_free);
2179 EXPORT_SYMBOL(zfs_is_readonly);
2180 EXPORT_SYMBOL(zfs_domount);
2181 EXPORT_SYMBOL(zfs_preumount);
2182 EXPORT_SYMBOL(zfs_umount);
2183 EXPORT_SYMBOL(zfs_remount);
2184 EXPORT_SYMBOL(zfs_statvfs);
2185 EXPORT_SYMBOL(zfs_vget);
2186 EXPORT_SYMBOL(zfs_prune);
2187 #endif
ZFS On Linux mirror for PVE