]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/zfs_vfsops.c
projects / mirror_zfs.git / blob
? search:


781708ba96a28171c0e5d5df38960af0f5790773
[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, 2018 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/sysmacros.h>
31 #include <sys/kmem.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
34 #include <sys/vfs.h>
35 #include <sys/mntent.h>
36 #include <sys/cmn_err.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/zfs_vnops.h>
39 #include <sys/zfs_dir.h>
40 #include <sys/zil.h>
41 #include <sys/fs/zfs.h>
42 #include <sys/dmu.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_deleg.h>
46 #include <sys/spa.h>
47 #include <sys/zap.h>
48 #include <sys/sa.h>
49 #include <sys/sa_impl.h>
50 #include <sys/policy.h>
51 #include <sys/atomic.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_ctldir.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/sunddi.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/spa_boot.h>
58 #include <sys/zpl.h>
59 #include <linux/vfs_compat.h>
60 #include "zfs_comutil.h"
61
62 enum {
63 TOKEN_RO,
64 TOKEN_RW,
65 TOKEN_SETUID,
66 TOKEN_NOSETUID,
67 TOKEN_EXEC,
68 TOKEN_NOEXEC,
69 TOKEN_DEVICES,
70 TOKEN_NODEVICES,
71 TOKEN_DIRXATTR,
72 TOKEN_SAXATTR,
73 TOKEN_XATTR,
74 TOKEN_NOXATTR,
75 TOKEN_ATIME,
76 TOKEN_NOATIME,
77 TOKEN_RELATIME,
78 TOKEN_NORELATIME,
79 TOKEN_NBMAND,
80 TOKEN_NONBMAND,
81 TOKEN_MNTPOINT,
82 TOKEN_LAST,
83 };
84
85 static const match_table_t zpl_tokens = {
86 { TOKEN_RO, MNTOPT_RO },
87 { TOKEN_RW, MNTOPT_RW },
88 { TOKEN_SETUID, MNTOPT_SETUID },
89 { TOKEN_NOSETUID, MNTOPT_NOSETUID },
90 { TOKEN_EXEC, MNTOPT_EXEC },
91 { TOKEN_NOEXEC, MNTOPT_NOEXEC },
92 { TOKEN_DEVICES, MNTOPT_DEVICES },
93 { TOKEN_NODEVICES, MNTOPT_NODEVICES },
94 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR },
95 { TOKEN_SAXATTR, MNTOPT_SAXATTR },
96 { TOKEN_XATTR, MNTOPT_XATTR },
97 { TOKEN_NOXATTR, MNTOPT_NOXATTR },
98 { TOKEN_ATIME, MNTOPT_ATIME },
99 { TOKEN_NOATIME, MNTOPT_NOATIME },
100 { TOKEN_RELATIME, MNTOPT_RELATIME },
101 { TOKEN_NORELATIME, MNTOPT_NORELATIME },
102 { TOKEN_NBMAND, MNTOPT_NBMAND },
103 { TOKEN_NONBMAND, MNTOPT_NONBMAND },
104 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" },
105 { TOKEN_LAST, NULL },
106 };
107
108 static void
109 zfsvfs_vfs_free(vfs_t *vfsp)
110 {
111 if (vfsp != NULL) {
112 if (vfsp->vfs_mntpoint != NULL)
113 strfree(vfsp->vfs_mntpoint);
114
115 kmem_free(vfsp, sizeof (vfs_t));
116 }
117 }
118
119 static int
120 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
121 {
122 switch (token) {
123 case TOKEN_RO:
124 vfsp->vfs_readonly = B_TRUE;
125 vfsp->vfs_do_readonly = B_TRUE;
126 break;
127 case TOKEN_RW:
128 vfsp->vfs_readonly = B_FALSE;
129 vfsp->vfs_do_readonly = B_TRUE;
130 break;
131 case TOKEN_SETUID:
132 vfsp->vfs_setuid = B_TRUE;
133 vfsp->vfs_do_setuid = B_TRUE;
134 break;
135 case TOKEN_NOSETUID:
136 vfsp->vfs_setuid = B_FALSE;
137 vfsp->vfs_do_setuid = B_TRUE;
138 break;
139 case TOKEN_EXEC:
140 vfsp->vfs_exec = B_TRUE;
141 vfsp->vfs_do_exec = B_TRUE;
142 break;
143 case TOKEN_NOEXEC:
144 vfsp->vfs_exec = B_FALSE;
145 vfsp->vfs_do_exec = B_TRUE;
146 break;
147 case TOKEN_DEVICES:
148 vfsp->vfs_devices = B_TRUE;
149 vfsp->vfs_do_devices = B_TRUE;
150 break;
151 case TOKEN_NODEVICES:
152 vfsp->vfs_devices = B_FALSE;
153 vfsp->vfs_do_devices = B_TRUE;
154 break;
155 case TOKEN_DIRXATTR:
156 vfsp->vfs_xattr = ZFS_XATTR_DIR;
157 vfsp->vfs_do_xattr = B_TRUE;
158 break;
159 case TOKEN_SAXATTR:
160 vfsp->vfs_xattr = ZFS_XATTR_SA;
161 vfsp->vfs_do_xattr = B_TRUE;
162 break;
163 case TOKEN_XATTR:
164 vfsp->vfs_xattr = ZFS_XATTR_DIR;
165 vfsp->vfs_do_xattr = B_TRUE;
166 break;
167 case TOKEN_NOXATTR:
168 vfsp->vfs_xattr = ZFS_XATTR_OFF;
169 vfsp->vfs_do_xattr = B_TRUE;
170 break;
171 case TOKEN_ATIME:
172 vfsp->vfs_atime = B_TRUE;
173 vfsp->vfs_do_atime = B_TRUE;
174 break;
175 case TOKEN_NOATIME:
176 vfsp->vfs_atime = B_FALSE;
177 vfsp->vfs_do_atime = B_TRUE;
178 break;
179 case TOKEN_RELATIME:
180 vfsp->vfs_relatime = B_TRUE;
181 vfsp->vfs_do_relatime = B_TRUE;
182 break;
183 case TOKEN_NORELATIME:
184 vfsp->vfs_relatime = B_FALSE;
185 vfsp->vfs_do_relatime = B_TRUE;
186 break;
187 case TOKEN_NBMAND:
188 vfsp->vfs_nbmand = B_TRUE;
189 vfsp->vfs_do_nbmand = B_TRUE;
190 break;
191 case TOKEN_NONBMAND:
192 vfsp->vfs_nbmand = B_FALSE;
193 vfsp->vfs_do_nbmand = B_TRUE;
194 break;
195 case TOKEN_MNTPOINT:
196 vfsp->vfs_mntpoint = match_strdup(&args[0]);
197 if (vfsp->vfs_mntpoint == NULL)
198 return (SET_ERROR(ENOMEM));
199
200 break;
201 default:
202 break;
203 }
204
205 return (0);
206 }
207
208 /*
209 * Parse the raw mntopts and return a vfs_t describing the options.
210 */
211 static int
212 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
213 {
214 vfs_t *tmp_vfsp;
215 int error;
216
217 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
218
219 if (mntopts != NULL) {
220 substring_t args[MAX_OPT_ARGS];
221 char *tmp_mntopts, *p, *t;
222 int token;
223
224 tmp_mntopts = t = strdup(mntopts);
225 if (tmp_mntopts == NULL)
226 return (SET_ERROR(ENOMEM));
227
228 while ((p = strsep(&t, ",")) != NULL) {
229 if (!*p)
230 continue;
231
232 args[0].to = args[0].from = NULL;
233 token = match_token(p, zpl_tokens, args);
234 error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
235 if (error) {
236 strfree(tmp_mntopts);
237 zfsvfs_vfs_free(tmp_vfsp);
238 return (error);
239 }
240 }
241
242 strfree(tmp_mntopts);
243 }
244
245 *vfsp = tmp_vfsp;
246
247 return (0);
248 }
249
250 boolean_t
251 zfs_is_readonly(zfsvfs_t *zfsvfs)
252 {
253 return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
254 }
255
256 /*ARGSUSED*/
257 int
258 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
259 {
260 zfsvfs_t *zfsvfs = sb->s_fs_info;
261
262 /*
263 * Semantically, the only requirement is that the sync be initiated.
264 * The DMU syncs out txgs frequently, so there's nothing to do.
265 */
266 if (!wait)
267 return (0);
268
269 if (zfsvfs != NULL) {
270 /*
271 * Sync a specific filesystem.
272 */
273 dsl_pool_t *dp;
274
275 ZFS_ENTER(zfsvfs);
276 dp = dmu_objset_pool(zfsvfs->z_os);
277
278 /*
279 * If the system is shutting down, then skip any
280 * filesystems which may exist on a suspended pool.
281 */
282 if (spa_suspended(dp->dp_spa)) {
283 ZFS_EXIT(zfsvfs);
284 return (0);
285 }
286
287 if (zfsvfs->z_log != NULL)
288 zil_commit(zfsvfs->z_log, 0);
289
290 ZFS_EXIT(zfsvfs);
291 } else {
292 /*
293 * Sync all ZFS filesystems. This is what happens when you
294 * run sync(1M). Unlike other filesystems, ZFS honors the
295 * request by waiting for all pools to commit all dirty data.
296 */
297 spa_sync_allpools();
298 }
299
300 return (0);
301 }
302
303 static void
304 atime_changed_cb(void *arg, uint64_t newval)
305 {
306 ((zfsvfs_t *)arg)->z_atime = newval;
307 }
308
309 static void
310 relatime_changed_cb(void *arg, uint64_t newval)
311 {
312 ((zfsvfs_t *)arg)->z_relatime = newval;
313 }
314
315 static void
316 xattr_changed_cb(void *arg, uint64_t newval)
317 {
318 zfsvfs_t *zfsvfs = arg;
319
320 if (newval == ZFS_XATTR_OFF) {
321 zfsvfs->z_flags &= ~ZSB_XATTR;
322 } else {
323 zfsvfs->z_flags |= ZSB_XATTR;
324
325 if (newval == ZFS_XATTR_SA)
326 zfsvfs->z_xattr_sa = B_TRUE;
327 else
328 zfsvfs->z_xattr_sa = B_FALSE;
329 }
330 }
331
332 static void
333 acltype_changed_cb(void *arg, uint64_t newval)
334 {
335 zfsvfs_t *zfsvfs = arg;
336
337 switch (newval) {
338 case ZFS_ACLTYPE_OFF:
339 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
340 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
341 break;
342 case ZFS_ACLTYPE_POSIXACL:
343 #ifdef CONFIG_FS_POSIX_ACL
344 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIXACL;
345 zfsvfs->z_sb->s_flags |= SB_POSIXACL;
346 #else
347 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
348 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
349 #endif /* CONFIG_FS_POSIX_ACL */
350 break;
351 default:
352 break;
353 }
354 }
355
356 static void
357 blksz_changed_cb(void *arg, uint64_t newval)
358 {
359 zfsvfs_t *zfsvfs = arg;
360 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
361 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
362 ASSERT(ISP2(newval));
363
364 zfsvfs->z_max_blksz = newval;
365 }
366
367 static void
368 readonly_changed_cb(void *arg, uint64_t newval)
369 {
370 zfsvfs_t *zfsvfs = arg;
371 struct super_block *sb = zfsvfs->z_sb;
372
373 if (sb == NULL)
374 return;
375
376 if (newval)
377 sb->s_flags |= SB_RDONLY;
378 else
379 sb->s_flags &= ~SB_RDONLY;
380 }
381
382 static void
383 devices_changed_cb(void *arg, uint64_t newval)
384 {
385 }
386
387 static void
388 setuid_changed_cb(void *arg, uint64_t newval)
389 {
390 }
391
392 static void
393 exec_changed_cb(void *arg, uint64_t newval)
394 {
395 }
396
397 static void
398 nbmand_changed_cb(void *arg, uint64_t newval)
399 {
400 zfsvfs_t *zfsvfs = arg;
401 struct super_block *sb = zfsvfs->z_sb;
402
403 if (sb == NULL)
404 return;
405
406 if (newval == TRUE)
407 sb->s_flags |= SB_MANDLOCK;
408 else
409 sb->s_flags &= ~SB_MANDLOCK;
410 }
411
412 static void
413 snapdir_changed_cb(void *arg, uint64_t newval)
414 {
415 ((zfsvfs_t *)arg)->z_show_ctldir = newval;
416 }
417
418 static void
419 vscan_changed_cb(void *arg, uint64_t newval)
420 {
421 ((zfsvfs_t *)arg)->z_vscan = newval;
422 }
423
424 static void
425 acl_inherit_changed_cb(void *arg, uint64_t newval)
426 {
427 ((zfsvfs_t *)arg)->z_acl_inherit = newval;
428 }
429
430 static int
431 zfs_register_callbacks(vfs_t *vfsp)
432 {
433 struct dsl_dataset *ds = NULL;
434 objset_t *os = NULL;
435 zfsvfs_t *zfsvfs = NULL;
436 int error = 0;
437
438 ASSERT(vfsp);
439 zfsvfs = vfsp->vfs_data;
440 ASSERT(zfsvfs);
441 os = zfsvfs->z_os;
442
443 /*
444 * The act of registering our callbacks will destroy any mount
445 * options we may have. In order to enable temporary overrides
446 * of mount options, we stash away the current values and
447 * restore them after we register the callbacks.
448 */
449 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
450 vfsp->vfs_do_readonly = B_TRUE;
451 vfsp->vfs_readonly = B_TRUE;
452 }
453
454 /*
455 * Register property callbacks.
456 *
457 * It would probably be fine to just check for i/o error from
458 * the first prop_register(), but I guess I like to go
459 * overboard...
460 */
461 ds = dmu_objset_ds(os);
462 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
463 error = dsl_prop_register(ds,
464 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
465 error = error ? error : dsl_prop_register(ds,
466 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
467 error = error ? error : dsl_prop_register(ds,
468 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
469 error = error ? error : dsl_prop_register(ds,
470 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
471 error = error ? error : dsl_prop_register(ds,
472 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
473 error = error ? error : dsl_prop_register(ds,
474 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
475 error = error ? error : dsl_prop_register(ds,
476 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
477 error = error ? error : dsl_prop_register(ds,
478 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
479 error = error ? error : dsl_prop_register(ds,
480 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
481 error = error ? error : dsl_prop_register(ds,
482 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
483 error = error ? error : dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
485 zfsvfs);
486 error = error ? error : dsl_prop_register(ds,
487 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
488 error = error ? error : dsl_prop_register(ds,
489 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
490 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
491 if (error)
492 goto unregister;
493
494 /*
495 * Invoke our callbacks to restore temporary mount options.
496 */
497 if (vfsp->vfs_do_readonly)
498 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
499 if (vfsp->vfs_do_setuid)
500 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
501 if (vfsp->vfs_do_exec)
502 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
503 if (vfsp->vfs_do_devices)
504 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
505 if (vfsp->vfs_do_xattr)
506 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
507 if (vfsp->vfs_do_atime)
508 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
509 if (vfsp->vfs_do_relatime)
510 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
511 if (vfsp->vfs_do_nbmand)
512 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
513
514 return (0);
515
516 unregister:
517 dsl_prop_unregister_all(ds, zfsvfs);
518 return (error);
519 }
520
521 static int
522 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
523 uint64_t *userp, uint64_t *groupp, uint64_t *projectp)
524 {
525 sa_hdr_phys_t sa;
526 sa_hdr_phys_t *sap = data;
527 uint64_t flags;
528 int hdrsize;
529 boolean_t swap = B_FALSE;
530
531 /*
532 * Is it a valid type of object to track?
533 */
534 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
535 return (SET_ERROR(ENOENT));
536
537 /*
538 * If we have a NULL data pointer
539 * then assume the id's aren't changing and
540 * return EEXIST to the dmu to let it know to
541 * use the same ids
542 */
543 if (data == NULL)
544 return (SET_ERROR(EEXIST));
545
546 if (bonustype == DMU_OT_ZNODE) {
547 znode_phys_t *znp = data;
548 *userp = znp->zp_uid;
549 *groupp = znp->zp_gid;
550 *projectp = ZFS_DEFAULT_PROJID;
551 return (0);
552 }
553
554 if (sap->sa_magic == 0) {
555 /*
556 * This should only happen for newly created files
557 * that haven't had the znode data filled in yet.
558 */
559 *userp = 0;
560 *groupp = 0;
561 *projectp = ZFS_DEFAULT_PROJID;
562 return (0);
563 }
564
565 sa = *sap;
566 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
567 sa.sa_magic = SA_MAGIC;
568 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
569 swap = B_TRUE;
570 } else {
571 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
572 }
573
574 hdrsize = sa_hdrsize(&sa);
575 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
576
577 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET));
578 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET));
579 flags = *((uint64_t *)((uintptr_t)data + hdrsize + SA_FLAGS_OFFSET));
580 if (swap)
581 flags = BSWAP_64(flags);
582
583 if (flags & ZFS_PROJID)
584 *projectp = *((uint64_t *)((uintptr_t)data + hdrsize +
585 SA_PROJID_OFFSET));
586 else
587 *projectp = ZFS_DEFAULT_PROJID;
588
589 if (swap) {
590 *userp = BSWAP_64(*userp);
591 *groupp = BSWAP_64(*groupp);
592 *projectp = BSWAP_64(*projectp);
593 }
594 return (0);
595 }
596
597 static void
598 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
599 char *domainbuf, int buflen, uid_t *ridp)
600 {
601 uint64_t fuid;
602 const char *domain;
603
604 fuid = zfs_strtonum(fuidstr, NULL);
605
606 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
607 if (domain)
608 (void) strlcpy(domainbuf, domain, buflen);
609 else
610 domainbuf[0] = '\0';
611 *ridp = FUID_RID(fuid);
612 }
613
614 static uint64_t
615 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
616 {
617 switch (type) {
618 case ZFS_PROP_USERUSED:
619 case ZFS_PROP_USEROBJUSED:
620 return (DMU_USERUSED_OBJECT);
621 case ZFS_PROP_GROUPUSED:
622 case ZFS_PROP_GROUPOBJUSED:
623 return (DMU_GROUPUSED_OBJECT);
624 case ZFS_PROP_PROJECTUSED:
625 case ZFS_PROP_PROJECTOBJUSED:
626 return (DMU_PROJECTUSED_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 case ZFS_PROP_PROJECTQUOTA:
636 return (zfsvfs->z_projectquota_obj);
637 case ZFS_PROP_PROJECTOBJQUOTA:
638 return (zfsvfs->z_projectobjquota_obj);
639 default:
640 return (ZFS_NO_OBJECT);
641 }
642 }
643
644 int
645 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
646 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
647 {
648 int error;
649 zap_cursor_t zc;
650 zap_attribute_t za;
651 zfs_useracct_t *buf = vbuf;
652 uint64_t obj;
653 int offset = 0;
654
655 if (!dmu_objset_userspace_present(zfsvfs->z_os))
656 return (SET_ERROR(ENOTSUP));
657
658 if ((type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
659 type == ZFS_PROP_PROJECTOBJQUOTA ||
660 type == ZFS_PROP_PROJECTOBJUSED) &&
661 !dmu_objset_projectquota_present(zfsvfs->z_os))
662 return (SET_ERROR(ENOTSUP));
663
664 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
665 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
666 type == ZFS_PROP_PROJECTOBJUSED ||
667 type == ZFS_PROP_PROJECTOBJQUOTA) &&
668 !dmu_objset_userobjspace_present(zfsvfs->z_os))
669 return (SET_ERROR(ENOTSUP));
670
671 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
672 if (obj == ZFS_NO_OBJECT) {
673 *bufsizep = 0;
674 return (0);
675 }
676
677 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
678 type == ZFS_PROP_PROJECTOBJUSED)
679 offset = DMU_OBJACCT_PREFIX_LEN;
680
681 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
682 (error = zap_cursor_retrieve(&zc, &za)) == 0;
683 zap_cursor_advance(&zc)) {
684 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
685 *bufsizep)
686 break;
687
688 /*
689 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
690 * when dealing with block quota and vice versa.
691 */
692 if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
693 DMU_OBJACCT_PREFIX_LEN) == 0))
694 continue;
695
696 fuidstr_to_sid(zfsvfs, za.za_name + offset,
697 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
698
699 buf->zu_space = za.za_first_integer;
700 buf++;
701 }
702 if (error == ENOENT)
703 error = 0;
704
705 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
706 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
707 *cookiep = zap_cursor_serialize(&zc);
708 zap_cursor_fini(&zc);
709 return (error);
710 }
711
712 /*
713 * buf must be big enough (eg, 32 bytes)
714 */
715 static int
716 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
717 char *buf, boolean_t addok)
718 {
719 uint64_t fuid;
720 int domainid = 0;
721
722 if (domain && domain[0]) {
723 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
724 if (domainid == -1)
725 return (SET_ERROR(ENOENT));
726 }
727 fuid = FUID_ENCODE(domainid, rid);
728 (void) sprintf(buf, "%llx", (longlong_t)fuid);
729 return (0);
730 }
731
732 int
733 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
734 const char *domain, uint64_t rid, uint64_t *valp)
735 {
736 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
737 int offset = 0;
738 int err;
739 uint64_t obj;
740
741 *valp = 0;
742
743 if (!dmu_objset_userspace_present(zfsvfs->z_os))
744 return (SET_ERROR(ENOTSUP));
745
746 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
747 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
748 type == ZFS_PROP_PROJECTOBJUSED ||
749 type == ZFS_PROP_PROJECTOBJQUOTA) &&
750 !dmu_objset_userobjspace_present(zfsvfs->z_os))
751 return (SET_ERROR(ENOTSUP));
752
753 if (type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
754 type == ZFS_PROP_PROJECTOBJQUOTA ||
755 type == ZFS_PROP_PROJECTOBJUSED) {
756 if (!dmu_objset_projectquota_present(zfsvfs->z_os))
757 return (SET_ERROR(ENOTSUP));
758 if (!zpl_is_valid_projid(rid))
759 return (SET_ERROR(EINVAL));
760 }
761
762 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
763 if (obj == ZFS_NO_OBJECT)
764 return (0);
765
766 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
767 type == ZFS_PROP_PROJECTOBJUSED) {
768 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
769 offset = DMU_OBJACCT_PREFIX_LEN;
770 }
771
772 err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
773 if (err)
774 return (err);
775
776 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
777 if (err == ENOENT)
778 err = 0;
779 return (err);
780 }
781
782 int
783 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
784 const char *domain, uint64_t rid, uint64_t quota)
785 {
786 char buf[32];
787 int err;
788 dmu_tx_t *tx;
789 uint64_t *objp;
790 boolean_t fuid_dirtied;
791
792 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
793 return (SET_ERROR(ENOTSUP));
794
795 switch (type) {
796 case ZFS_PROP_USERQUOTA:
797 objp = &zfsvfs->z_userquota_obj;
798 break;
799 case ZFS_PROP_GROUPQUOTA:
800 objp = &zfsvfs->z_groupquota_obj;
801 break;
802 case ZFS_PROP_USEROBJQUOTA:
803 objp = &zfsvfs->z_userobjquota_obj;
804 break;
805 case ZFS_PROP_GROUPOBJQUOTA:
806 objp = &zfsvfs->z_groupobjquota_obj;
807 break;
808 case ZFS_PROP_PROJECTQUOTA:
809 if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
810 return (SET_ERROR(ENOTSUP));
811 if (!zpl_is_valid_projid(rid))
812 return (SET_ERROR(EINVAL));
813
814 objp = &zfsvfs->z_projectquota_obj;
815 break;
816 case ZFS_PROP_PROJECTOBJQUOTA:
817 if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
818 return (SET_ERROR(ENOTSUP));
819 if (!zpl_is_valid_projid(rid))
820 return (SET_ERROR(EINVAL));
821
822 objp = &zfsvfs->z_projectobjquota_obj;
823 break;
824 default:
825 return (SET_ERROR(EINVAL));
826 }
827
828 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
829 if (err)
830 return (err);
831 fuid_dirtied = zfsvfs->z_fuid_dirty;
832
833 tx = dmu_tx_create(zfsvfs->z_os);
834 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
835 if (*objp == 0) {
836 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
837 zfs_userquota_prop_prefixes[type]);
838 }
839 if (fuid_dirtied)
840 zfs_fuid_txhold(zfsvfs, tx);
841 err = dmu_tx_assign(tx, TXG_WAIT);
842 if (err) {
843 dmu_tx_abort(tx);
844 return (err);
845 }
846
847 mutex_enter(&zfsvfs->z_lock);
848 if (*objp == 0) {
849 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
850 DMU_OT_NONE, 0, tx);
851 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
852 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
853 }
854 mutex_exit(&zfsvfs->z_lock);
855
856 if (quota == 0) {
857 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
858 if (err == ENOENT)
859 err = 0;
860 } else {
861 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
862 }
863 ASSERT(err == 0);
864 if (fuid_dirtied)
865 zfs_fuid_sync(zfsvfs, tx);
866 dmu_tx_commit(tx);
867 return (err);
868 }
869
870 boolean_t
871 zfs_id_overobjquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
872 {
873 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
874 uint64_t used, quota, quotaobj;
875 int err;
876
877 if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
878 if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os)) {
879 dsl_pool_config_enter(
880 dmu_objset_pool(zfsvfs->z_os), FTAG);
881 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
882 dsl_pool_config_exit(
883 dmu_objset_pool(zfsvfs->z_os), FTAG);
884 }
885 return (B_FALSE);
886 }
887
888 if (usedobj == DMU_PROJECTUSED_OBJECT) {
889 if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
890 if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
891 dsl_pool_config_enter(
892 dmu_objset_pool(zfsvfs->z_os), FTAG);
893 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
894 dsl_pool_config_exit(
895 dmu_objset_pool(zfsvfs->z_os), FTAG);
896 }
897 return (B_FALSE);
898 }
899 quotaobj = zfsvfs->z_projectobjquota_obj;
900 } else if (usedobj == DMU_USERUSED_OBJECT) {
901 quotaobj = zfsvfs->z_userobjquota_obj;
902 } else if (usedobj == DMU_GROUPUSED_OBJECT) {
903 quotaobj = zfsvfs->z_groupobjquota_obj;
904 } else {
905 return (B_FALSE);
906 }
907 if (quotaobj == 0 || zfsvfs->z_replay)
908 return (B_FALSE);
909
910 (void) sprintf(buf, "%llx", (longlong_t)id);
911 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
912 if (err != 0)
913 return (B_FALSE);
914
915 (void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)id);
916 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
917 if (err != 0)
918 return (B_FALSE);
919 return (used >= quota);
920 }
921
922 boolean_t
923 zfs_id_overblockquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
924 {
925 char buf[20];
926 uint64_t used, quota, quotaobj;
927 int err;
928
929 if (usedobj == DMU_PROJECTUSED_OBJECT) {
930 if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
931 if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
932 dsl_pool_config_enter(
933 dmu_objset_pool(zfsvfs->z_os), FTAG);
934 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
935 dsl_pool_config_exit(
936 dmu_objset_pool(zfsvfs->z_os), FTAG);
937 }
938 return (B_FALSE);
939 }
940 quotaobj = zfsvfs->z_projectquota_obj;
941 } else if (usedobj == DMU_USERUSED_OBJECT) {
942 quotaobj = zfsvfs->z_userquota_obj;
943 } else if (usedobj == DMU_GROUPUSED_OBJECT) {
944 quotaobj = zfsvfs->z_groupquota_obj;
945 } else {
946 return (B_FALSE);
947 }
948 if (quotaobj == 0 || zfsvfs->z_replay)
949 return (B_FALSE);
950
951 (void) sprintf(buf, "%llx", (longlong_t)id);
952 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
953 if (err != 0)
954 return (B_FALSE);
955
956 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
957 if (err != 0)
958 return (B_FALSE);
959 return (used >= quota);
960 }
961
962 boolean_t
963 zfs_id_overquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
964 {
965 return (zfs_id_overblockquota(zfsvfs, usedobj, id) ||
966 zfs_id_overobjquota(zfsvfs, usedobj, id));
967 }
968
969 /*
970 * Associate this zfsvfs with the given objset, which must be owned.
971 * This will cache a bunch of on-disk state from the objset in the
972 * zfsvfs.
973 */
974 static int
975 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
976 {
977 int error;
978 uint64_t val;
979
980 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
981 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
982 zfsvfs->z_os = os;
983
984 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
985 if (error != 0)
986 return (error);
987 if (zfsvfs->z_version >
988 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
989 (void) printk("Can't mount a version %lld file system "
990 "on a version %lld pool\n. Pool must be upgraded to mount "
991 "this file system.", (u_longlong_t)zfsvfs->z_version,
992 (u_longlong_t)spa_version(dmu_objset_spa(os)));
993 return (SET_ERROR(ENOTSUP));
994 }
995 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
996 if (error != 0)
997 return (error);
998 zfsvfs->z_norm = (int)val;
999
1000 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1001 if (error != 0)
1002 return (error);
1003 zfsvfs->z_utf8 = (val != 0);
1004
1005 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1006 if (error != 0)
1007 return (error);
1008 zfsvfs->z_case = (uint_t)val;
1009
1010 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
1011 return (error);
1012 zfsvfs->z_acl_type = (uint_t)val;
1013
1014 /*
1015 * Fold case on file systems that are always or sometimes case
1016 * insensitive.
1017 */
1018 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1019 zfsvfs->z_case == ZFS_CASE_MIXED)
1020 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1021
1022 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1023 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1024
1025 uint64_t sa_obj = 0;
1026 if (zfsvfs->z_use_sa) {
1027 /* should either have both of these objects or none */
1028 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1029 &sa_obj);
1030 if (error != 0)
1031 return (error);
1032
1033 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
1034 if ((error == 0) && (val == ZFS_XATTR_SA))
1035 zfsvfs->z_xattr_sa = B_TRUE;
1036 }
1037
1038 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1039 &zfsvfs->z_root);
1040 if (error != 0)
1041 return (error);
1042 ASSERT(zfsvfs->z_root != 0);
1043
1044 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1045 &zfsvfs->z_unlinkedobj);
1046 if (error != 0)
1047 return (error);
1048
1049 error = zap_lookup(os, MASTER_NODE_OBJ,
1050 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1051 8, 1, &zfsvfs->z_userquota_obj);
1052 if (error == ENOENT)
1053 zfsvfs->z_userquota_obj = 0;
1054 else if (error != 0)
1055 return (error);
1056
1057 error = zap_lookup(os, MASTER_NODE_OBJ,
1058 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1059 8, 1, &zfsvfs->z_groupquota_obj);
1060 if (error == ENOENT)
1061 zfsvfs->z_groupquota_obj = 0;
1062 else if (error != 0)
1063 return (error);
1064
1065 error = zap_lookup(os, MASTER_NODE_OBJ,
1066 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
1067 8, 1, &zfsvfs->z_projectquota_obj);
1068 if (error == ENOENT)
1069 zfsvfs->z_projectquota_obj = 0;
1070 else if (error != 0)
1071 return (error);
1072
1073 error = zap_lookup(os, MASTER_NODE_OBJ,
1074 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1075 8, 1, &zfsvfs->z_userobjquota_obj);
1076 if (error == ENOENT)
1077 zfsvfs->z_userobjquota_obj = 0;
1078 else if (error != 0)
1079 return (error);
1080
1081 error = zap_lookup(os, MASTER_NODE_OBJ,
1082 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1083 8, 1, &zfsvfs->z_groupobjquota_obj);
1084 if (error == ENOENT)
1085 zfsvfs->z_groupobjquota_obj = 0;
1086 else if (error != 0)
1087 return (error);
1088
1089 error = zap_lookup(os, MASTER_NODE_OBJ,
1090 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
1091 8, 1, &zfsvfs->z_projectobjquota_obj);
1092 if (error == ENOENT)
1093 zfsvfs->z_projectobjquota_obj = 0;
1094 else if (error != 0)
1095 return (error);
1096
1097 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1098 &zfsvfs->z_fuid_obj);
1099 if (error == ENOENT)
1100 zfsvfs->z_fuid_obj = 0;
1101 else if (error != 0)
1102 return (error);
1103
1104 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1105 &zfsvfs->z_shares_dir);
1106 if (error == ENOENT)
1107 zfsvfs->z_shares_dir = 0;
1108 else if (error != 0)
1109 return (error);
1110
1111 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1112 &zfsvfs->z_attr_table);
1113 if (error != 0)
1114 return (error);
1115
1116 if (zfsvfs->z_version >= ZPL_VERSION_SA)
1117 sa_register_update_callback(os, zfs_sa_upgrade);
1118
1119 return (0);
1120 }
1121
1122 int
1123 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
1124 {
1125 objset_t *os;
1126 zfsvfs_t *zfsvfs;
1127 int error;
1128 boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
1129
1130 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1131
1132 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
1133 if (error != 0) {
1134 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1135 return (error);
1136 }
1137
1138 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1139 if (error != 0) {
1140 dmu_objset_disown(os, B_TRUE, zfsvfs);
1141 }
1142 return (error);
1143 }
1144
1145
1146 /*
1147 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
1148 * on a failure. Do not pass in a statically allocated zfsvfs.
1149 */
1150 int
1151 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1152 {
1153 int error;
1154
1155 zfsvfs->z_vfs = NULL;
1156 zfsvfs->z_sb = NULL;
1157 zfsvfs->z_parent = zfsvfs;
1158
1159 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1160 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1161 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1162 offsetof(znode_t, z_link_node));
1163 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1164 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1165 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1166
1167 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
1168 ZFS_OBJ_MTX_MAX);
1169 zfsvfs->z_hold_size = size;
1170 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1171 KM_SLEEP);
1172 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1173 for (int i = 0; i != size; i++) {
1174 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1175 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1176 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1177 }
1178
1179 error = zfsvfs_init(zfsvfs, os);
1180 if (error != 0) {
1181 *zfvp = NULL;
1182 zfsvfs_free(zfsvfs);
1183 return (error);
1184 }
1185
1186 zfsvfs->z_drain_task = TASKQID_INVALID;
1187 zfsvfs->z_draining = B_FALSE;
1188 zfsvfs->z_drain_cancel = B_TRUE;
1189
1190 *zfvp = zfsvfs;
1191 return (0);
1192 }
1193
1194 static int
1195 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1196 {
1197 int error;
1198 boolean_t readonly = zfs_is_readonly(zfsvfs);
1199
1200 error = zfs_register_callbacks(zfsvfs->z_vfs);
1201 if (error)
1202 return (error);
1203
1204 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1205
1206 /*
1207 * If we are not mounting (ie: online recv), then we don't
1208 * have to worry about replaying the log as we blocked all
1209 * operations out since we closed the ZIL.
1210 */
1211 if (mounting) {
1212 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
1213 dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
1214
1215 /*
1216 * During replay we remove the read only flag to
1217 * allow replays to succeed.
1218 */
1219 if (readonly != 0) {
1220 readonly_changed_cb(zfsvfs, B_FALSE);
1221 } else {
1222 zap_stats_t zs;
1223 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
1224 &zs) == 0) {
1225 dataset_kstats_update_nunlinks_kstat(
1226 &zfsvfs->z_kstat, zs.zs_num_entries);
1227 }
1228 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1229 "num_entries in unlinked set: %llu",
1230 zs.zs_num_entries);
1231 zfs_unlinked_drain(zfsvfs);
1232 }
1233
1234 /*
1235 * Parse and replay the intent log.
1236 *
1237 * Because of ziltest, this must be done after
1238 * zfs_unlinked_drain(). (Further note: ziltest
1239 * doesn't use readonly mounts, where
1240 * zfs_unlinked_drain() isn't called.) This is because
1241 * ziltest causes spa_sync() to think it's committed,
1242 * but actually it is not, so the intent log contains
1243 * many txg's worth of changes.
1244 *
1245 * In particular, if object N is in the unlinked set in
1246 * the last txg to actually sync, then it could be
1247 * actually freed in a later txg and then reallocated
1248 * in a yet later txg. This would write a "create
1249 * object N" record to the intent log. Normally, this
1250 * would be fine because the spa_sync() would have
1251 * written out the fact that object N is free, before
1252 * we could write the "create object N" intent log
1253 * record.
1254 *
1255 * But when we are in ziltest mode, we advance the "open
1256 * txg" without actually spa_sync()-ing the changes to
1257 * disk. So we would see that object N is still
1258 * allocated and in the unlinked set, and there is an
1259 * intent log record saying to allocate it.
1260 */
1261 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1262 if (zil_replay_disable) {
1263 zil_destroy(zfsvfs->z_log, B_FALSE);
1264 } else {
1265 zfsvfs->z_replay = B_TRUE;
1266 zil_replay(zfsvfs->z_os, zfsvfs,
1267 zfs_replay_vector);
1268 zfsvfs->z_replay = B_FALSE;
1269 }
1270 }
1271
1272 /* restore readonly bit */
1273 if (readonly != 0)
1274 readonly_changed_cb(zfsvfs, B_TRUE);
1275 }
1276
1277 /*
1278 * Set the objset user_ptr to track its zfsvfs.
1279 */
1280 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1281 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1282 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1283
1284 return (0);
1285 }
1286
1287 void
1288 zfsvfs_free(zfsvfs_t *zfsvfs)
1289 {
1290 int i, size = zfsvfs->z_hold_size;
1291
1292 zfs_fuid_destroy(zfsvfs);
1293
1294 mutex_destroy(&zfsvfs->z_znodes_lock);
1295 mutex_destroy(&zfsvfs->z_lock);
1296 list_destroy(&zfsvfs->z_all_znodes);
1297 rrm_destroy(&zfsvfs->z_teardown_lock);
1298 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1299 rw_destroy(&zfsvfs->z_fuid_lock);
1300 for (i = 0; i != size; i++) {
1301 avl_destroy(&zfsvfs->z_hold_trees[i]);
1302 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1303 }
1304 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1305 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1306 zfsvfs_vfs_free(zfsvfs->z_vfs);
1307 dataset_kstats_destroy(&zfsvfs->z_kstat);
1308 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1309 }
1310
1311 static void
1312 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1313 {
1314 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1315 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1316 }
1317
1318 void
1319 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1320 {
1321 objset_t *os = zfsvfs->z_os;
1322
1323 if (!dmu_objset_is_snapshot(os))
1324 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1325 }
1326
1327 #ifdef HAVE_MLSLABEL
1328 /*
1329 * Check that the hex label string is appropriate for the dataset being
1330 * mounted into the global_zone proper.
1331 *
1332 * Return an error if the hex label string is not default or
1333 * admin_low/admin_high. For admin_low labels, the corresponding
1334 * dataset must be readonly.
1335 */
1336 int
1337 zfs_check_global_label(const char *dsname, const char *hexsl)
1338 {
1339 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1340 return (0);
1341 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1342 return (0);
1343 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1344 /* must be readonly */
1345 uint64_t rdonly;
1346
1347 if (dsl_prop_get_integer(dsname,
1348 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1349 return (SET_ERROR(EACCES));
1350 return (rdonly ? 0 : EACCES);
1351 }
1352 return (SET_ERROR(EACCES));
1353 }
1354 #endif /* HAVE_MLSLABEL */
1355
1356 static int
1357 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1358 uint32_t bshift)
1359 {
1360 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1361 uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1362 uint64_t quota;
1363 uint64_t used;
1364 int err;
1365
1366 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1367 err = id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, B_FALSE);
1368 if (err)
1369 return (err);
1370
1371 if (zfsvfs->z_projectquota_obj == 0)
1372 goto objs;
1373
1374 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1375 buf + offset, 8, 1, &quota);
1376 if (err == ENOENT)
1377 goto objs;
1378 else if (err)
1379 return (err);
1380
1381 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1382 buf + offset, 8, 1, &used);
1383 if (unlikely(err == ENOENT)) {
1384 uint32_t blksize;
1385 u_longlong_t nblocks;
1386
1387 /*
1388 * Quota accounting is async, so it is possible race case.
1389 * There is at least one object with the given project ID.
1390 */
1391 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1392 if (unlikely(zp->z_blksz == 0))
1393 blksize = zfsvfs->z_max_blksz;
1394
1395 used = blksize * nblocks;
1396 } else if (err) {
1397 return (err);
1398 }
1399
1400 statp->f_blocks = quota >> bshift;
1401 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1402 statp->f_bavail = statp->f_bfree;
1403
1404 objs:
1405 if (zfsvfs->z_projectobjquota_obj == 0)
1406 return (0);
1407
1408 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1409 buf + offset, 8, 1, &quota);
1410 if (err == ENOENT)
1411 return (0);
1412 else if (err)
1413 return (err);
1414
1415 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1416 buf, 8, 1, &used);
1417 if (unlikely(err == ENOENT)) {
1418 /*
1419 * Quota accounting is async, so it is possible race case.
1420 * There is at least one object with the given project ID.
1421 */
1422 used = 1;
1423 } else if (err) {
1424 return (err);
1425 }
1426
1427 statp->f_files = quota;
1428 statp->f_ffree = (quota > used) ? (quota - used) : 0;
1429
1430 return (0);
1431 }
1432
1433 int
1434 zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
1435 {
1436 zfsvfs_t *zfsvfs = dentry->d_sb->s_fs_info;
1437 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1438 int err = 0;
1439
1440 ZFS_ENTER(zfsvfs);
1441
1442 dmu_objset_space(zfsvfs->z_os,
1443 &refdbytes, &availbytes, &usedobjs, &availobjs);
1444
1445 uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1446 /*
1447 * The underlying storage pool actually uses multiple block
1448 * size. Under Solaris frsize (fragment size) is reported as
1449 * the smallest block size we support, and bsize (block size)
1450 * as the filesystem's maximum block size. Unfortunately,
1451 * under Linux the fragment size and block size are often used
1452 * interchangeably. Thus we are forced to report both of them
1453 * as the filesystem's maximum block size.
1454 */
1455 statp->f_frsize = zfsvfs->z_max_blksz;
1456 statp->f_bsize = zfsvfs->z_max_blksz;
1457 uint32_t bshift = fls(statp->f_bsize) - 1;
1458
1459 /*
1460 * The following report "total" blocks of various kinds in
1461 * the file system, but reported in terms of f_bsize - the
1462 * "preferred" size.
1463 */
1464
1465 /* Round up so we never have a filesytem using 0 blocks. */
1466 refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1467 statp->f_blocks = (refdbytes + availbytes) >> bshift;
1468 statp->f_bfree = availbytes >> bshift;
1469 statp->f_bavail = statp->f_bfree; /* no root reservation */
1470
1471 /*
1472 * statvfs() should really be called statufs(), because it assumes
1473 * static metadata. ZFS doesn't preallocate files, so the best
1474 * we can do is report the max that could possibly fit in f_files,
1475 * and that minus the number actually used in f_ffree.
1476 * For f_ffree, report the smaller of the number of objects available
1477 * and the number of blocks (each object will take at least a block).
1478 */
1479 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1480 statp->f_files = statp->f_ffree + usedobjs;
1481 statp->f_fsid.val[0] = (uint32_t)fsid;
1482 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1483 statp->f_type = ZFS_SUPER_MAGIC;
1484 statp->f_namelen = MAXNAMELEN - 1;
1485
1486 /*
1487 * We have all of 40 characters to stuff a string here.
1488 * Is there anything useful we could/should provide?
1489 */
1490 bzero(statp->f_spare, sizeof (statp->f_spare));
1491
1492 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1493 dmu_objset_projectquota_present(zfsvfs->z_os)) {
1494 znode_t *zp = ITOZ(dentry->d_inode);
1495
1496 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1497 zpl_is_valid_projid(zp->z_projid))
1498 err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1499 }
1500
1501 ZFS_EXIT(zfsvfs);
1502 return (err);
1503 }
1504
1505 int
1506 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1507 {
1508 znode_t *rootzp;
1509 int error;
1510
1511 ZFS_ENTER(zfsvfs);
1512
1513 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1514 if (error == 0)
1515 *ipp = ZTOI(rootzp);
1516
1517 ZFS_EXIT(zfsvfs);
1518 return (error);
1519 }
1520
1521 #ifdef HAVE_D_PRUNE_ALIASES
1522 /*
1523 * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1524 * To accommodate this we must improvise and manually walk the list of znodes
1525 * attempting to prune dentries in order to be able to drop the inodes.
1526 *
1527 * To avoid scanning the same znodes multiple times they are always rotated
1528 * to the end of the z_all_znodes list. New znodes are inserted at the
1529 * end of the list so we're always scanning the oldest znodes first.
1530 */
1531 static int
1532 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1533 {
1534 znode_t **zp_array, *zp;
1535 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1536 int objects = 0;
1537 int i = 0, j = 0;
1538
1539 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1540
1541 mutex_enter(&zfsvfs->z_znodes_lock);
1542 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1543
1544 if ((i++ > nr_to_scan) || (j >= max_array))
1545 break;
1546
1547 ASSERT(list_link_active(&zp->z_link_node));
1548 list_remove(&zfsvfs->z_all_znodes, zp);
1549 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1550
1551 /* Skip active znodes and .zfs entries */
1552 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1553 continue;
1554
1555 if (igrab(ZTOI(zp)) == NULL)
1556 continue;
1557
1558 zp_array[j] = zp;
1559 j++;
1560 }
1561 mutex_exit(&zfsvfs->z_znodes_lock);
1562
1563 for (i = 0; i < j; i++) {
1564 zp = zp_array[i];
1565
1566 ASSERT3P(zp, !=, NULL);
1567 d_prune_aliases(ZTOI(zp));
1568
1569 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1570 objects++;
1571
1572 iput(ZTOI(zp));
1573 }
1574
1575 kmem_free(zp_array, max_array * sizeof (znode_t *));
1576
1577 return (objects);
1578 }
1579 #endif /* HAVE_D_PRUNE_ALIASES */
1580
1581 /*
1582 * The ARC has requested that the filesystem drop entries from the dentry
1583 * and inode caches. This can occur when the ARC needs to free meta data
1584 * blocks but can't because they are all pinned by entries in these caches.
1585 */
1586 int
1587 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1588 {
1589 zfsvfs_t *zfsvfs = sb->s_fs_info;
1590 int error = 0;
1591 #if defined(HAVE_SHRINK) || defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1592 struct shrinker *shrinker = &sb->s_shrink;
1593 struct shrink_control sc = {
1594 .nr_to_scan = nr_to_scan,
1595 .gfp_mask = GFP_KERNEL,
1596 };
1597 #endif
1598
1599 ZFS_ENTER(zfsvfs);
1600
1601 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1602 defined(SHRINK_CONTROL_HAS_NID) && \
1603 defined(SHRINKER_NUMA_AWARE)
1604 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1605 *objects = 0;
1606 for_each_online_node(sc.nid) {
1607 *objects += (*shrinker->scan_objects)(shrinker, &sc);
1608 }
1609 } else {
1610 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1611 }
1612
1613 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1614 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1615 #elif defined(HAVE_SHRINK)
1616 *objects = (*shrinker->shrink)(shrinker, &sc);
1617 #elif defined(HAVE_D_PRUNE_ALIASES)
1618 #define D_PRUNE_ALIASES_IS_DEFAULT
1619 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1620 #else
1621 #error "No available dentry and inode cache pruning mechanism."
1622 #endif
1623
1624 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1625 #undef D_PRUNE_ALIASES_IS_DEFAULT
1626 /*
1627 * Fall back to zfs_prune_aliases if the kernel's per-superblock
1628 * shrinker couldn't free anything, possibly due to the inodes being
1629 * allocated in a different memcg.
1630 */
1631 if (*objects == 0)
1632 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1633 #endif
1634
1635 ZFS_EXIT(zfsvfs);
1636
1637 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1638 "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1639 nr_to_scan, *objects, error);
1640
1641 return (error);
1642 }
1643
1644 /*
1645 * Teardown the zfsvfs_t.
1646 *
1647 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1648 * and 'z_teardown_inactive_lock' held.
1649 */
1650 static int
1651 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1652 {
1653 znode_t *zp;
1654
1655 zfs_unlinked_drain_stop_wait(zfsvfs);
1656
1657 /*
1658 * If someone has not already unmounted this file system,
1659 * drain the iput_taskq to ensure all active references to the
1660 * zfsvfs_t have been handled only then can it be safely destroyed.
1661 */
1662 if (zfsvfs->z_os) {
1663 /*
1664 * If we're unmounting we have to wait for the list to
1665 * drain completely.
1666 *
1667 * If we're not unmounting there's no guarantee the list
1668 * will drain completely, but iputs run from the taskq
1669 * may add the parents of dir-based xattrs to the taskq
1670 * so we want to wait for these.
1671 *
1672 * We can safely read z_nr_znodes without locking because the
1673 * VFS has already blocked operations which add to the
1674 * z_all_znodes list and thus increment z_nr_znodes.
1675 */
1676 int round = 0;
1677 while (zfsvfs->z_nr_znodes > 0) {
1678 taskq_wait_outstanding(dsl_pool_iput_taskq(
1679 dmu_objset_pool(zfsvfs->z_os)), 0);
1680 if (++round > 1 && !unmounting)
1681 break;
1682 }
1683 }
1684
1685 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1686
1687 if (!unmounting) {
1688 /*
1689 * We purge the parent filesystem's super block as the
1690 * parent filesystem and all of its snapshots have their
1691 * inode's super block set to the parent's filesystem's
1692 * super block. Note, 'z_parent' is self referential
1693 * for non-snapshots.
1694 */
1695 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1696 }
1697
1698 /*
1699 * Close the zil. NB: Can't close the zil while zfs_inactive
1700 * threads are blocked as zil_close can call zfs_inactive.
1701 */
1702 if (zfsvfs->z_log) {
1703 zil_close(zfsvfs->z_log);
1704 zfsvfs->z_log = NULL;
1705 }
1706
1707 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1708
1709 /*
1710 * If we are not unmounting (ie: online recv) and someone already
1711 * unmounted this file system while we were doing the switcheroo,
1712 * or a reopen of z_os failed then just bail out now.
1713 */
1714 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1715 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1716 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1717 return (SET_ERROR(EIO));
1718 }
1719
1720 /*
1721 * At this point there are no VFS ops active, and any new VFS ops
1722 * will fail with EIO since we have z_teardown_lock for writer (only
1723 * relevant for forced unmount).
1724 *
1725 * Release all holds on dbufs.
1726 */
1727 if (!unmounting) {
1728 mutex_enter(&zfsvfs->z_znodes_lock);
1729 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1730 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1731 if (zp->z_sa_hdl)
1732 zfs_znode_dmu_fini(zp);
1733 }
1734 mutex_exit(&zfsvfs->z_znodes_lock);
1735 }
1736
1737 /*
1738 * If we are unmounting, set the unmounted flag and let new VFS ops
1739 * unblock. zfs_inactive will have the unmounted behavior, and all
1740 * other VFS ops will fail with EIO.
1741 */
1742 if (unmounting) {
1743 zfsvfs->z_unmounted = B_TRUE;
1744 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1745 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1746 }
1747
1748 /*
1749 * z_os will be NULL if there was an error in attempting to reopen
1750 * zfsvfs, so just return as the properties had already been
1751 *
1752 * unregistered and cached data had been evicted before.
1753 */
1754 if (zfsvfs->z_os == NULL)
1755 return (0);
1756
1757 /*
1758 * Unregister properties.
1759 */
1760 zfs_unregister_callbacks(zfsvfs);
1761
1762 /*
1763 * Evict cached data. We must write out any dirty data before
1764 * disowning the dataset.
1765 */
1766 if (!zfs_is_readonly(zfsvfs))
1767 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1768 dmu_objset_evict_dbufs(zfsvfs->z_os);
1769
1770 return (0);
1771 }
1772
1773 #if !defined(HAVE_2ARGS_BDI_SETUP_AND_REGISTER) && \
1774 !defined(HAVE_3ARGS_BDI_SETUP_AND_REGISTER)
1775 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1776 #endif
1777
1778 int
1779 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1780 {
1781 const char *osname = zm->mnt_osname;
1782 struct inode *root_inode;
1783 uint64_t recordsize;
1784 int error = 0;
1785 zfsvfs_t *zfsvfs = NULL;
1786 vfs_t *vfs = NULL;
1787
1788 ASSERT(zm);
1789 ASSERT(osname);
1790
1791 error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1792 if (error)
1793 return (error);
1794
1795 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1796 if (error) {
1797 zfsvfs_vfs_free(vfs);
1798 goto out;
1799 }
1800
1801 if ((error = dsl_prop_get_integer(osname, "recordsize",
1802 &recordsize, NULL))) {
1803 zfsvfs_vfs_free(vfs);
1804 goto out;
1805 }
1806
1807 vfs->vfs_data = zfsvfs;
1808 zfsvfs->z_vfs = vfs;
1809 zfsvfs->z_sb = sb;
1810 sb->s_fs_info = zfsvfs;
1811 sb->s_magic = ZFS_SUPER_MAGIC;
1812 sb->s_maxbytes = MAX_LFS_FILESIZE;
1813 sb->s_time_gran = 1;
1814 sb->s_blocksize = recordsize;
1815 sb->s_blocksize_bits = ilog2(recordsize);
1816
1817 error = -zpl_bdi_setup(sb, "zfs");
1818 if (error)
1819 goto out;
1820
1821 sb->s_bdi->ra_pages = 0;
1822
1823 /* Set callback operations for the file system. */
1824 sb->s_op = &zpl_super_operations;
1825 sb->s_xattr = zpl_xattr_handlers;
1826 sb->s_export_op = &zpl_export_operations;
1827 #ifdef HAVE_S_D_OP
1828 sb->s_d_op = &zpl_dentry_operations;
1829 #endif /* HAVE_S_D_OP */
1830
1831 /* Set features for file system. */
1832 zfs_set_fuid_feature(zfsvfs);
1833
1834 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1835 uint64_t pval;
1836
1837 atime_changed_cb(zfsvfs, B_FALSE);
1838 readonly_changed_cb(zfsvfs, B_TRUE);
1839 if ((error = dsl_prop_get_integer(osname,
1840 "xattr", &pval, NULL)))
1841 goto out;
1842 xattr_changed_cb(zfsvfs, pval);
1843 if ((error = dsl_prop_get_integer(osname,
1844 "acltype", &pval, NULL)))
1845 goto out;
1846 acltype_changed_cb(zfsvfs, pval);
1847 zfsvfs->z_issnap = B_TRUE;
1848 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1849 zfsvfs->z_snap_defer_time = jiffies;
1850
1851 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1852 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1853 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1854 } else {
1855 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1856 goto out;
1857 }
1858
1859 /* Allocate a root inode for the filesystem. */
1860 error = zfs_root(zfsvfs, &root_inode);
1861 if (error) {
1862 (void) zfs_umount(sb);
1863 goto out;
1864 }
1865
1866 /* Allocate a root dentry for the filesystem */
1867 sb->s_root = d_make_root(root_inode);
1868 if (sb->s_root == NULL) {
1869 (void) zfs_umount(sb);
1870 error = SET_ERROR(ENOMEM);
1871 goto out;
1872 }
1873
1874 if (!zfsvfs->z_issnap)
1875 zfsctl_create(zfsvfs);
1876
1877 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1878 out:
1879 if (error) {
1880 if (zfsvfs != NULL) {
1881 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1882 zfsvfs_free(zfsvfs);
1883 }
1884 /*
1885 * make sure we don't have dangling sb->s_fs_info which
1886 * zfs_preumount will use.
1887 */
1888 sb->s_fs_info = NULL;
1889 }
1890
1891 return (error);
1892 }
1893
1894 /*
1895 * Called when an unmount is requested and certain sanity checks have
1896 * already passed. At this point no dentries or inodes have been reclaimed
1897 * from their respective caches. We drop the extra reference on the .zfs
1898 * control directory to allow everything to be reclaimed. All snapshots
1899 * must already have been unmounted to reach this point.
1900 */
1901 void
1902 zfs_preumount(struct super_block *sb)
1903 {
1904 zfsvfs_t *zfsvfs = sb->s_fs_info;
1905
1906 /* zfsvfs is NULL when zfs_domount fails during mount */
1907 if (zfsvfs) {
1908 zfs_unlinked_drain_stop_wait(zfsvfs);
1909 zfsctl_destroy(sb->s_fs_info);
1910 /*
1911 * Wait for iput_async before entering evict_inodes in
1912 * generic_shutdown_super. The reason we must finish before
1913 * evict_inodes is when lazytime is on, or when zfs_purgedir
1914 * calls zfs_zget, iput would bump i_count from 0 to 1. This
1915 * would race with the i_count check in evict_inodes. This means
1916 * it could destroy the inode while we are still using it.
1917 *
1918 * We wait for two passes. xattr directories in the first pass
1919 * may add xattr entries in zfs_purgedir, so in the second pass
1920 * we wait for them. We don't use taskq_wait here because it is
1921 * a pool wide taskq. Other mounted filesystems can constantly
1922 * do iput_async and there's no guarantee when taskq will be
1923 * empty.
1924 */
1925 taskq_wait_outstanding(dsl_pool_iput_taskq(
1926 dmu_objset_pool(zfsvfs->z_os)), 0);
1927 taskq_wait_outstanding(dsl_pool_iput_taskq(
1928 dmu_objset_pool(zfsvfs->z_os)), 0);
1929 }
1930 }
1931
1932 /*
1933 * Called once all other unmount released tear down has occurred.
1934 * It is our responsibility to release any remaining infrastructure.
1935 */
1936 /*ARGSUSED*/
1937 int
1938 zfs_umount(struct super_block *sb)
1939 {
1940 zfsvfs_t *zfsvfs = sb->s_fs_info;
1941 objset_t *os;
1942
1943 if (zfsvfs->z_arc_prune != NULL)
1944 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1945 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1946 os = zfsvfs->z_os;
1947 zpl_bdi_destroy(sb);
1948
1949 /*
1950 * z_os will be NULL if there was an error in
1951 * attempting to reopen zfsvfs.
1952 */
1953 if (os != NULL) {
1954 /*
1955 * Unset the objset user_ptr.
1956 */
1957 mutex_enter(&os->os_user_ptr_lock);
1958 dmu_objset_set_user(os, NULL);
1959 mutex_exit(&os->os_user_ptr_lock);
1960
1961 /*
1962 * Finally release the objset
1963 */
1964 dmu_objset_disown(os, B_TRUE, zfsvfs);
1965 }
1966
1967 zfsvfs_free(zfsvfs);
1968 return (0);
1969 }
1970
1971 int
1972 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1973 {
1974 zfsvfs_t *zfsvfs = sb->s_fs_info;
1975 vfs_t *vfsp;
1976 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1977 int error;
1978
1979 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1980 !(*flags & SB_RDONLY)) {
1981 *flags |= SB_RDONLY;
1982 return (EROFS);
1983 }
1984
1985 error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1986 if (error)
1987 return (error);
1988
1989 if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1990 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1991
1992 zfs_unregister_callbacks(zfsvfs);
1993 zfsvfs_vfs_free(zfsvfs->z_vfs);
1994
1995 vfsp->vfs_data = zfsvfs;
1996 zfsvfs->z_vfs = vfsp;
1997 if (!issnap)
1998 (void) zfs_register_callbacks(vfsp);
1999
2000 return (error);
2001 }
2002
2003 int
2004 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
2005 {
2006 zfsvfs_t *zfsvfs = sb->s_fs_info;
2007 znode_t *zp;
2008 uint64_t object = 0;
2009 uint64_t fid_gen = 0;
2010 uint64_t gen_mask;
2011 uint64_t zp_gen;
2012 int i, err;
2013
2014 *ipp = NULL;
2015
2016 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2017 zfid_short_t *zfid = (zfid_short_t *)fidp;
2018
2019 for (i = 0; i < sizeof (zfid->zf_object); i++)
2020 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2021
2022 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2023 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2024 } else {
2025 return (SET_ERROR(EINVAL));
2026 }
2027
2028 /* LONG_FID_LEN means snapdirs */
2029 if (fidp->fid_len == LONG_FID_LEN) {
2030 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2031 uint64_t objsetid = 0;
2032 uint64_t setgen = 0;
2033
2034 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2035 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2036
2037 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2038 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2039
2040 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
2041 dprintf("snapdir fid: objsetid (%llu) != "
2042 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
2043 objsetid, ZFSCTL_INO_SNAPDIRS, object);
2044
2045 return (SET_ERROR(EINVAL));
2046 }
2047
2048 if (fid_gen > 1 || setgen != 0) {
2049 dprintf("snapdir fid: fid_gen (%llu) and setgen "
2050 "(%llu)\n", fid_gen, setgen);
2051 return (SET_ERROR(EINVAL));
2052 }
2053
2054 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
2055 }
2056
2057 ZFS_ENTER(zfsvfs);
2058 /* A zero fid_gen means we are in the .zfs control directories */
2059 if (fid_gen == 0 &&
2060 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2061 *ipp = zfsvfs->z_ctldir;
2062 ASSERT(*ipp != NULL);
2063 if (object == ZFSCTL_INO_SNAPDIR) {
2064 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
2065 0, kcred, NULL, NULL) == 0);
2066 } else {
2067 igrab(*ipp);
2068 }
2069 ZFS_EXIT(zfsvfs);
2070 return (0);
2071 }
2072
2073 gen_mask = -1ULL >> (64 - 8 * i);
2074
2075 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
2076 if ((err = zfs_zget(zfsvfs, object, &zp))) {
2077 ZFS_EXIT(zfsvfs);
2078 return (err);
2079 }
2080
2081 /* Don't export xattr stuff */
2082 if (zp->z_pflags & ZFS_XATTR) {
2083 iput(ZTOI(zp));
2084 ZFS_EXIT(zfsvfs);
2085 return (SET_ERROR(ENOENT));
2086 }
2087
2088 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2089 sizeof (uint64_t));
2090 zp_gen = zp_gen & gen_mask;
2091 if (zp_gen == 0)
2092 zp_gen = 1;
2093 if ((fid_gen == 0) && (zfsvfs->z_root == object))
2094 fid_gen = zp_gen;
2095 if (zp->z_unlinked || zp_gen != fid_gen) {
2096 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
2097 fid_gen);
2098 iput(ZTOI(zp));
2099 ZFS_EXIT(zfsvfs);
2100 return (SET_ERROR(ENOENT));
2101 }
2102
2103 *ipp = ZTOI(zp);
2104 if (*ipp)
2105 zfs_inode_update(ITOZ(*ipp));
2106
2107 ZFS_EXIT(zfsvfs);
2108 return (0);
2109 }
2110
2111 /*
2112 * Block out VFS ops and close zfsvfs_t
2113 *
2114 * Note, if successful, then we return with the 'z_teardown_lock' and
2115 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2116 * dataset and objset intact so that they can be atomically handed off during
2117 * a subsequent rollback or recv operation and the resume thereafter.
2118 */
2119 int
2120 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2121 {
2122 int error;
2123
2124 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2125 return (error);
2126
2127 return (0);
2128 }
2129
2130 /*
2131 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2132 * is an invariant across any of the operations that can be performed while the
2133 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2134 * are the same: the relevant objset and associated dataset are owned by
2135 * zfsvfs, held, and long held on entry.
2136 */
2137 int
2138 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2139 {
2140 int err, err2;
2141 znode_t *zp;
2142
2143 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2144 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2145
2146 /*
2147 * We already own this, so just update the objset_t, as the one we
2148 * had before may have been evicted.
2149 */
2150 objset_t *os;
2151 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2152 VERIFY(dsl_dataset_long_held(ds));
2153 VERIFY0(dmu_objset_from_ds(ds, &os));
2154
2155 err = zfsvfs_init(zfsvfs, os);
2156 if (err != 0)
2157 goto bail;
2158
2159 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2160
2161 zfs_set_fuid_feature(zfsvfs);
2162 zfsvfs->z_rollback_time = jiffies;
2163
2164 /*
2165 * Attempt to re-establish all the active inodes with their
2166 * dbufs. If a zfs_rezget() fails, then we unhash the inode
2167 * and mark it stale. This prevents a collision if a new
2168 * inode/object is created which must use the same inode
2169 * number. The stale inode will be be released when the
2170 * VFS prunes the dentry holding the remaining references
2171 * on the stale inode.
2172 */
2173 mutex_enter(&zfsvfs->z_znodes_lock);
2174 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2175 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2176 err2 = zfs_rezget(zp);
2177 if (err2) {
2178 remove_inode_hash(ZTOI(zp));
2179 zp->z_is_stale = B_TRUE;
2180 }
2181 }
2182 mutex_exit(&zfsvfs->z_znodes_lock);
2183
2184 if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
2185 /*
2186 * zfs_suspend_fs() could have interrupted freeing
2187 * of dnodes. We need to restart this freeing so
2188 * that we don't "leak" the space.
2189 */
2190 zfs_unlinked_drain(zfsvfs);
2191 }
2192
2193 bail:
2194 /* release the VFS ops */
2195 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2196 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2197
2198 if (err) {
2199 /*
2200 * Since we couldn't setup the sa framework, try to force
2201 * unmount this file system.
2202 */
2203 if (zfsvfs->z_os)
2204 (void) zfs_umount(zfsvfs->z_sb);
2205 }
2206 return (err);
2207 }
2208
2209 int
2210 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2211 {
2212 int error;
2213 objset_t *os = zfsvfs->z_os;
2214 dmu_tx_t *tx;
2215
2216 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2217 return (SET_ERROR(EINVAL));
2218
2219 if (newvers < zfsvfs->z_version)
2220 return (SET_ERROR(EINVAL));
2221
2222 if (zfs_spa_version_map(newvers) >
2223 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2224 return (SET_ERROR(ENOTSUP));
2225
2226 tx = dmu_tx_create(os);
2227 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2228 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2229 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2230 ZFS_SA_ATTRS);
2231 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2232 }
2233 error = dmu_tx_assign(tx, TXG_WAIT);
2234 if (error) {
2235 dmu_tx_abort(tx);
2236 return (error);
2237 }
2238
2239 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2240 8, 1, &newvers, tx);
2241
2242 if (error) {
2243 dmu_tx_commit(tx);
2244 return (error);
2245 }
2246
2247 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2248 uint64_t sa_obj;
2249
2250 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2251 SPA_VERSION_SA);
2252 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2253 DMU_OT_NONE, 0, tx);
2254
2255 error = zap_add(os, MASTER_NODE_OBJ,
2256 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2257 ASSERT0(error);
2258
2259 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2260 sa_register_update_callback(os, zfs_sa_upgrade);
2261 }
2262
2263 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2264 "from %llu to %llu", zfsvfs->z_version, newvers);
2265
2266 dmu_tx_commit(tx);
2267
2268 zfsvfs->z_version = newvers;
2269 os->os_version = newvers;
2270
2271 zfs_set_fuid_feature(zfsvfs);
2272
2273 return (0);
2274 }
2275
2276 /*
2277 * Read a property stored within the master node.
2278 */
2279 int
2280 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2281 {
2282 uint64_t *cached_copy = NULL;
2283
2284 /*
2285 * Figure out where in the objset_t the cached copy would live, if it
2286 * is available for the requested property.
2287 */
2288 if (os != NULL) {
2289 switch (prop) {
2290 case ZFS_PROP_VERSION:
2291 cached_copy = &os->os_version;
2292 break;
2293 case ZFS_PROP_NORMALIZE:
2294 cached_copy = &os->os_normalization;
2295 break;
2296 case ZFS_PROP_UTF8ONLY:
2297 cached_copy = &os->os_utf8only;
2298 break;
2299 case ZFS_PROP_CASE:
2300 cached_copy = &os->os_casesensitivity;
2301 break;
2302 default:
2303 break;
2304 }
2305 }
2306 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2307 *value = *cached_copy;
2308 return (0);
2309 }
2310
2311 /*
2312 * If the property wasn't cached, look up the file system's value for
2313 * the property. For the version property, we look up a slightly
2314 * different string.
2315 */
2316 const char *pname;
2317 int error = ENOENT;
2318 if (prop == ZFS_PROP_VERSION)
2319 pname = ZPL_VERSION_STR;
2320 else
2321 pname = zfs_prop_to_name(prop);
2322
2323 if (os != NULL) {
2324 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2325 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2326 }
2327
2328 if (error == ENOENT) {
2329 /* No value set, use the default value */
2330 switch (prop) {
2331 case ZFS_PROP_VERSION:
2332 *value = ZPL_VERSION;
2333 break;
2334 case ZFS_PROP_NORMALIZE:
2335 case ZFS_PROP_UTF8ONLY:
2336 *value = 0;
2337 break;
2338 case ZFS_PROP_CASE:
2339 *value = ZFS_CASE_SENSITIVE;
2340 break;
2341 case ZFS_PROP_ACLTYPE:
2342 *value = ZFS_ACLTYPE_OFF;
2343 break;
2344 default:
2345 return (error);
2346 }
2347 error = 0;
2348 }
2349
2350 /*
2351 * If one of the methods for getting the property value above worked,
2352 * copy it into the objset_t's cache.
2353 */
2354 if (error == 0 && cached_copy != NULL) {
2355 *cached_copy = *value;
2356 }
2357
2358 return (error);
2359 }
2360
2361 /*
2362 * Return true if the coresponding vfs's unmounted flag is set.
2363 * Otherwise return false.
2364 * If this function returns true we know VFS unmount has been initiated.
2365 */
2366 boolean_t
2367 zfs_get_vfs_flag_unmounted(objset_t *os)
2368 {
2369 zfsvfs_t *zfvp;
2370 boolean_t unmounted = B_FALSE;
2371
2372 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2373
2374 mutex_enter(&os->os_user_ptr_lock);
2375 zfvp = dmu_objset_get_user(os);
2376 if (zfvp != NULL && zfvp->z_unmounted)
2377 unmounted = B_TRUE;
2378 mutex_exit(&os->os_user_ptr_lock);
2379
2380 return (unmounted);
2381 }
2382
2383 void
2384 zfs_init(void)
2385 {
2386 zfsctl_init();
2387 zfs_znode_init();
2388 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2389 register_filesystem(&zpl_fs_type);
2390 }
2391
2392 void
2393 zfs_fini(void)
2394 {
2395 /*
2396 * we don't use outstanding because zpl_posix_acl_free might add more.
2397 */
2398 taskq_wait(system_delay_taskq);
2399 taskq_wait(system_taskq);
2400 unregister_filesystem(&zpl_fs_type);
2401 zfs_znode_fini();
2402 zfsctl_fini();
2403 }
2404
2405 #if defined(_KERNEL)
2406 EXPORT_SYMBOL(zfs_suspend_fs);
2407 EXPORT_SYMBOL(zfs_resume_fs);
2408 EXPORT_SYMBOL(zfs_userspace_one);
2409 EXPORT_SYMBOL(zfs_userspace_many);
2410 EXPORT_SYMBOL(zfs_set_userquota);
2411 EXPORT_SYMBOL(zfs_id_overblockquota);
2412 EXPORT_SYMBOL(zfs_id_overobjquota);
2413 EXPORT_SYMBOL(zfs_id_overquota);
2414 EXPORT_SYMBOL(zfs_set_version);
2415 EXPORT_SYMBOL(zfsvfs_create);
2416 EXPORT_SYMBOL(zfsvfs_free);
2417 EXPORT_SYMBOL(zfs_is_readonly);
2418 EXPORT_SYMBOL(zfs_domount);
2419 EXPORT_SYMBOL(zfs_preumount);
2420 EXPORT_SYMBOL(zfs_umount);
2421 EXPORT_SYMBOL(zfs_remount);
2422 EXPORT_SYMBOL(zfs_statvfs);
2423 EXPORT_SYMBOL(zfs_vget);
2424 EXPORT_SYMBOL(zfs_prune);
2425 #endif
ZFS On Linux mirror for PVE