]> 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 9337 - zfs get all is slow due to uncached metadata
[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/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 "zfs_comutil.h"
60
61 enum {
62 TOKEN_RO,
63 TOKEN_RW,
64 TOKEN_SETUID,
65 TOKEN_NOSETUID,
66 TOKEN_EXEC,
67 TOKEN_NOEXEC,
68 TOKEN_DEVICES,
69 TOKEN_NODEVICES,
70 TOKEN_DIRXATTR,
71 TOKEN_SAXATTR,
72 TOKEN_XATTR,
73 TOKEN_NOXATTR,
74 TOKEN_ATIME,
75 TOKEN_NOATIME,
76 TOKEN_RELATIME,
77 TOKEN_NORELATIME,
78 TOKEN_NBMAND,
79 TOKEN_NONBMAND,
80 TOKEN_MNTPOINT,
81 TOKEN_LAST,
82 };
83
84 static const match_table_t zpl_tokens = {
85 { TOKEN_RO, MNTOPT_RO },
86 { TOKEN_RW, MNTOPT_RW },
87 { TOKEN_SETUID, MNTOPT_SETUID },
88 { TOKEN_NOSETUID, MNTOPT_NOSETUID },
89 { TOKEN_EXEC, MNTOPT_EXEC },
90 { TOKEN_NOEXEC, MNTOPT_NOEXEC },
91 { TOKEN_DEVICES, MNTOPT_DEVICES },
92 { TOKEN_NODEVICES, MNTOPT_NODEVICES },
93 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR },
94 { TOKEN_SAXATTR, MNTOPT_SAXATTR },
95 { TOKEN_XATTR, MNTOPT_XATTR },
96 { TOKEN_NOXATTR, MNTOPT_NOXATTR },
97 { TOKEN_ATIME, MNTOPT_ATIME },
98 { TOKEN_NOATIME, MNTOPT_NOATIME },
99 { TOKEN_RELATIME, MNTOPT_RELATIME },
100 { TOKEN_NORELATIME, MNTOPT_NORELATIME },
101 { TOKEN_NBMAND, MNTOPT_NBMAND },
102 { TOKEN_NONBMAND, MNTOPT_NONBMAND },
103 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" },
104 { TOKEN_LAST, NULL },
105 };
106
107 static void
108 zfsvfs_vfs_free(vfs_t *vfsp)
109 {
110 if (vfsp != NULL) {
111 if (vfsp->vfs_mntpoint != NULL)
112 strfree(vfsp->vfs_mntpoint);
113
114 kmem_free(vfsp, sizeof (vfs_t));
115 }
116 }
117
118 static int
119 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
120 {
121 switch (token) {
122 case TOKEN_RO:
123 vfsp->vfs_readonly = B_TRUE;
124 vfsp->vfs_do_readonly = B_TRUE;
125 break;
126 case TOKEN_RW:
127 vfsp->vfs_readonly = B_FALSE;
128 vfsp->vfs_do_readonly = B_TRUE;
129 break;
130 case TOKEN_SETUID:
131 vfsp->vfs_setuid = B_TRUE;
132 vfsp->vfs_do_setuid = B_TRUE;
133 break;
134 case TOKEN_NOSETUID:
135 vfsp->vfs_setuid = B_FALSE;
136 vfsp->vfs_do_setuid = B_TRUE;
137 break;
138 case TOKEN_EXEC:
139 vfsp->vfs_exec = B_TRUE;
140 vfsp->vfs_do_exec = B_TRUE;
141 break;
142 case TOKEN_NOEXEC:
143 vfsp->vfs_exec = B_FALSE;
144 vfsp->vfs_do_exec = B_TRUE;
145 break;
146 case TOKEN_DEVICES:
147 vfsp->vfs_devices = B_TRUE;
148 vfsp->vfs_do_devices = B_TRUE;
149 break;
150 case TOKEN_NODEVICES:
151 vfsp->vfs_devices = B_FALSE;
152 vfsp->vfs_do_devices = B_TRUE;
153 break;
154 case TOKEN_DIRXATTR:
155 vfsp->vfs_xattr = ZFS_XATTR_DIR;
156 vfsp->vfs_do_xattr = B_TRUE;
157 break;
158 case TOKEN_SAXATTR:
159 vfsp->vfs_xattr = ZFS_XATTR_SA;
160 vfsp->vfs_do_xattr = B_TRUE;
161 break;
162 case TOKEN_XATTR:
163 vfsp->vfs_xattr = ZFS_XATTR_DIR;
164 vfsp->vfs_do_xattr = B_TRUE;
165 break;
166 case TOKEN_NOXATTR:
167 vfsp->vfs_xattr = ZFS_XATTR_OFF;
168 vfsp->vfs_do_xattr = B_TRUE;
169 break;
170 case TOKEN_ATIME:
171 vfsp->vfs_atime = B_TRUE;
172 vfsp->vfs_do_atime = B_TRUE;
173 break;
174 case TOKEN_NOATIME:
175 vfsp->vfs_atime = B_FALSE;
176 vfsp->vfs_do_atime = B_TRUE;
177 break;
178 case TOKEN_RELATIME:
179 vfsp->vfs_relatime = B_TRUE;
180 vfsp->vfs_do_relatime = B_TRUE;
181 break;
182 case TOKEN_NORELATIME:
183 vfsp->vfs_relatime = B_FALSE;
184 vfsp->vfs_do_relatime = B_TRUE;
185 break;
186 case TOKEN_NBMAND:
187 vfsp->vfs_nbmand = B_TRUE;
188 vfsp->vfs_do_nbmand = B_TRUE;
189 break;
190 case TOKEN_NONBMAND:
191 vfsp->vfs_nbmand = B_FALSE;
192 vfsp->vfs_do_nbmand = B_TRUE;
193 break;
194 case TOKEN_MNTPOINT:
195 vfsp->vfs_mntpoint = match_strdup(&args[0]);
196 if (vfsp->vfs_mntpoint == NULL)
197 return (SET_ERROR(ENOMEM));
198
199 break;
200 default:
201 break;
202 }
203
204 return (0);
205 }
206
207 /*
208 * Parse the raw mntopts and return a vfs_t describing the options.
209 */
210 static int
211 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
212 {
213 vfs_t *tmp_vfsp;
214 int error;
215
216 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
217
218 if (mntopts != NULL) {
219 substring_t args[MAX_OPT_ARGS];
220 char *tmp_mntopts, *p, *t;
221 int token;
222
223 tmp_mntopts = t = strdup(mntopts);
224 if (tmp_mntopts == NULL)
225 return (SET_ERROR(ENOMEM));
226
227 while ((p = strsep(&t, ",")) != NULL) {
228 if (!*p)
229 continue;
230
231 args[0].to = args[0].from = NULL;
232 token = match_token(p, zpl_tokens, args);
233 error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
234 if (error) {
235 strfree(tmp_mntopts);
236 zfsvfs_vfs_free(tmp_vfsp);
237 return (error);
238 }
239 }
240
241 strfree(tmp_mntopts);
242 }
243
244 *vfsp = tmp_vfsp;
245
246 return (0);
247 }
248
249 boolean_t
250 zfs_is_readonly(zfsvfs_t *zfsvfs)
251 {
252 return (!!(zfsvfs->z_sb->s_flags & MS_RDONLY));
253 }
254
255 /*ARGSUSED*/
256 int
257 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
258 {
259 zfsvfs_t *zfsvfs = sb->s_fs_info;
260
261 /*
262 * Data integrity is job one. We don't want a compromised kernel
263 * writing to the storage pool, so we never sync during panic.
264 */
265 if (unlikely(oops_in_progress))
266 return (0);
267
268 /*
269 * Semantically, the only requirement is that the sync be initiated.
270 * The DMU syncs out txgs frequently, so there's nothing to do.
271 */
272 if (!wait)
273 return (0);
274
275 if (zfsvfs != NULL) {
276 /*
277 * Sync a specific filesystem.
278 */
279 dsl_pool_t *dp;
280
281 ZFS_ENTER(zfsvfs);
282 dp = dmu_objset_pool(zfsvfs->z_os);
283
284 /*
285 * If the system is shutting down, then skip any
286 * filesystems which may exist on a suspended pool.
287 */
288 if (spa_suspended(dp->dp_spa)) {
289 ZFS_EXIT(zfsvfs);
290 return (0);
291 }
292
293 if (zfsvfs->z_log != NULL)
294 zil_commit(zfsvfs->z_log, 0);
295
296 ZFS_EXIT(zfsvfs);
297 } else {
298 /*
299 * Sync all ZFS filesystems. This is what happens when you
300 * run sync(1M). Unlike other filesystems, ZFS honors the
301 * request by waiting for all pools to commit all dirty data.
302 */
303 spa_sync_allpools();
304 }
305
306 return (0);
307 }
308
309 static void
310 atime_changed_cb(void *arg, uint64_t newval)
311 {
312 ((zfsvfs_t *)arg)->z_atime = newval;
313 }
314
315 static void
316 relatime_changed_cb(void *arg, uint64_t newval)
317 {
318 ((zfsvfs_t *)arg)->z_relatime = newval;
319 }
320
321 static void
322 xattr_changed_cb(void *arg, uint64_t newval)
323 {
324 zfsvfs_t *zfsvfs = arg;
325
326 if (newval == ZFS_XATTR_OFF) {
327 zfsvfs->z_flags &= ~ZSB_XATTR;
328 } else {
329 zfsvfs->z_flags |= ZSB_XATTR;
330
331 if (newval == ZFS_XATTR_SA)
332 zfsvfs->z_xattr_sa = B_TRUE;
333 else
334 zfsvfs->z_xattr_sa = B_FALSE;
335 }
336 }
337
338 static void
339 acltype_changed_cb(void *arg, uint64_t newval)
340 {
341 zfsvfs_t *zfsvfs = arg;
342
343 switch (newval) {
344 case ZFS_ACLTYPE_OFF:
345 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
346 zfsvfs->z_sb->s_flags &= ~MS_POSIXACL;
347 break;
348 case ZFS_ACLTYPE_POSIXACL:
349 #ifdef CONFIG_FS_POSIX_ACL
350 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIXACL;
351 zfsvfs->z_sb->s_flags |= MS_POSIXACL;
352 #else
353 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
354 zfsvfs->z_sb->s_flags &= ~MS_POSIXACL;
355 #endif /* CONFIG_FS_POSIX_ACL */
356 break;
357 default:
358 break;
359 }
360 }
361
362 static void
363 blksz_changed_cb(void *arg, uint64_t newval)
364 {
365 zfsvfs_t *zfsvfs = arg;
366 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
367 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
368 ASSERT(ISP2(newval));
369
370 zfsvfs->z_max_blksz = newval;
371 }
372
373 static void
374 readonly_changed_cb(void *arg, uint64_t newval)
375 {
376 zfsvfs_t *zfsvfs = arg;
377 struct super_block *sb = zfsvfs->z_sb;
378
379 if (sb == NULL)
380 return;
381
382 if (newval)
383 sb->s_flags |= MS_RDONLY;
384 else
385 sb->s_flags &= ~MS_RDONLY;
386 }
387
388 static void
389 devices_changed_cb(void *arg, uint64_t newval)
390 {
391 }
392
393 static void
394 setuid_changed_cb(void *arg, uint64_t newval)
395 {
396 }
397
398 static void
399 exec_changed_cb(void *arg, uint64_t newval)
400 {
401 }
402
403 static void
404 nbmand_changed_cb(void *arg, uint64_t newval)
405 {
406 zfsvfs_t *zfsvfs = arg;
407 struct super_block *sb = zfsvfs->z_sb;
408
409 if (sb == NULL)
410 return;
411
412 if (newval == TRUE)
413 sb->s_flags |= MS_MANDLOCK;
414 else
415 sb->s_flags &= ~MS_MANDLOCK;
416 }
417
418 static void
419 snapdir_changed_cb(void *arg, uint64_t newval)
420 {
421 ((zfsvfs_t *)arg)->z_show_ctldir = newval;
422 }
423
424 static void
425 vscan_changed_cb(void *arg, uint64_t newval)
426 {
427 ((zfsvfs_t *)arg)->z_vscan = newval;
428 }
429
430 static void
431 acl_inherit_changed_cb(void *arg, uint64_t newval)
432 {
433 ((zfsvfs_t *)arg)->z_acl_inherit = newval;
434 }
435
436 static int
437 zfs_register_callbacks(vfs_t *vfsp)
438 {
439 struct dsl_dataset *ds = NULL;
440 objset_t *os = NULL;
441 zfsvfs_t *zfsvfs = NULL;
442 int error = 0;
443
444 ASSERT(vfsp);
445 zfsvfs = vfsp->vfs_data;
446 ASSERT(zfsvfs);
447 os = zfsvfs->z_os;
448
449 /*
450 * The act of registering our callbacks will destroy any mount
451 * options we may have. In order to enable temporary overrides
452 * of mount options, we stash away the current values and
453 * restore them after we register the callbacks.
454 */
455 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
456 vfsp->vfs_do_readonly = B_TRUE;
457 vfsp->vfs_readonly = B_TRUE;
458 }
459
460 /*
461 * Register property callbacks.
462 *
463 * It would probably be fine to just check for i/o error from
464 * the first prop_register(), but I guess I like to go
465 * overboard...
466 */
467 ds = dmu_objset_ds(os);
468 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
469 error = dsl_prop_register(ds,
470 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
471 error = error ? error : dsl_prop_register(ds,
472 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
473 error = error ? error : dsl_prop_register(ds,
474 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
475 error = error ? error : dsl_prop_register(ds,
476 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
477 error = error ? error : dsl_prop_register(ds,
478 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
479 error = error ? error : dsl_prop_register(ds,
480 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
481 error = error ? error : dsl_prop_register(ds,
482 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
483 error = error ? error : dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
485 error = error ? error : dsl_prop_register(ds,
486 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
487 error = error ? error : dsl_prop_register(ds,
488 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
489 error = error ? error : dsl_prop_register(ds,
490 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
491 zfsvfs);
492 error = error ? error : dsl_prop_register(ds,
493 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
494 error = error ? error : dsl_prop_register(ds,
495 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
496 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
497 if (error)
498 goto unregister;
499
500 /*
501 * Invoke our callbacks to restore temporary mount options.
502 */
503 if (vfsp->vfs_do_readonly)
504 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
505 if (vfsp->vfs_do_setuid)
506 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
507 if (vfsp->vfs_do_exec)
508 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
509 if (vfsp->vfs_do_devices)
510 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
511 if (vfsp->vfs_do_xattr)
512 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
513 if (vfsp->vfs_do_atime)
514 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
515 if (vfsp->vfs_do_relatime)
516 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
517 if (vfsp->vfs_do_nbmand)
518 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
519
520 return (0);
521
522 unregister:
523 dsl_prop_unregister_all(ds, zfsvfs);
524 return (error);
525 }
526
527 static int
528 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
529 uint64_t *userp, uint64_t *groupp, uint64_t *projectp)
530 {
531 sa_hdr_phys_t sa;
532 sa_hdr_phys_t *sap = data;
533 uint64_t flags;
534 int hdrsize;
535 boolean_t swap = B_FALSE;
536
537 /*
538 * Is it a valid type of object to track?
539 */
540 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
541 return (SET_ERROR(ENOENT));
542
543 /*
544 * If we have a NULL data pointer
545 * then assume the id's aren't changing and
546 * return EEXIST to the dmu to let it know to
547 * use the same ids
548 */
549 if (data == NULL)
550 return (SET_ERROR(EEXIST));
551
552 if (bonustype == DMU_OT_ZNODE) {
553 znode_phys_t *znp = data;
554 *userp = znp->zp_uid;
555 *groupp = znp->zp_gid;
556 *projectp = ZFS_DEFAULT_PROJID;
557 return (0);
558 }
559
560 if (sap->sa_magic == 0) {
561 /*
562 * This should only happen for newly created files
563 * that haven't had the znode data filled in yet.
564 */
565 *userp = 0;
566 *groupp = 0;
567 *projectp = ZFS_DEFAULT_PROJID;
568 return (0);
569 }
570
571 sa = *sap;
572 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
573 sa.sa_magic = SA_MAGIC;
574 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
575 swap = B_TRUE;
576 } else {
577 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
578 }
579
580 hdrsize = sa_hdrsize(&sa);
581 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
582
583 *userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET));
584 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET));
585 flags = *((uint64_t *)((uintptr_t)data + hdrsize + SA_FLAGS_OFFSET));
586 if (swap)
587 flags = BSWAP_64(flags);
588
589 if (flags & ZFS_PROJID)
590 *projectp = *((uint64_t *)((uintptr_t)data + hdrsize +
591 SA_PROJID_OFFSET));
592 else
593 *projectp = ZFS_DEFAULT_PROJID;
594
595 if (swap) {
596 *userp = BSWAP_64(*userp);
597 *groupp = BSWAP_64(*groupp);
598 *projectp = BSWAP_64(*projectp);
599 }
600 return (0);
601 }
602
603 static void
604 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
605 char *domainbuf, int buflen, uid_t *ridp)
606 {
607 uint64_t fuid;
608 const char *domain;
609
610 fuid = zfs_strtonum(fuidstr, NULL);
611
612 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
613 if (domain)
614 (void) strlcpy(domainbuf, domain, buflen);
615 else
616 domainbuf[0] = '\0';
617 *ridp = FUID_RID(fuid);
618 }
619
620 static uint64_t
621 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
622 {
623 switch (type) {
624 case ZFS_PROP_USERUSED:
625 case ZFS_PROP_USEROBJUSED:
626 return (DMU_USERUSED_OBJECT);
627 case ZFS_PROP_GROUPUSED:
628 case ZFS_PROP_GROUPOBJUSED:
629 return (DMU_GROUPUSED_OBJECT);
630 case ZFS_PROP_PROJECTUSED:
631 case ZFS_PROP_PROJECTOBJUSED:
632 return (DMU_PROJECTUSED_OBJECT);
633 case ZFS_PROP_USERQUOTA:
634 return (zfsvfs->z_userquota_obj);
635 case ZFS_PROP_GROUPQUOTA:
636 return (zfsvfs->z_groupquota_obj);
637 case ZFS_PROP_USEROBJQUOTA:
638 return (zfsvfs->z_userobjquota_obj);
639 case ZFS_PROP_GROUPOBJQUOTA:
640 return (zfsvfs->z_groupobjquota_obj);
641 case ZFS_PROP_PROJECTQUOTA:
642 return (zfsvfs->z_projectquota_obj);
643 case ZFS_PROP_PROJECTOBJQUOTA:
644 return (zfsvfs->z_projectobjquota_obj);
645 default:
646 return (ZFS_NO_OBJECT);
647 }
648 }
649
650 int
651 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
652 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
653 {
654 int error;
655 zap_cursor_t zc;
656 zap_attribute_t za;
657 zfs_useracct_t *buf = vbuf;
658 uint64_t obj;
659 int offset = 0;
660
661 if (!dmu_objset_userspace_present(zfsvfs->z_os))
662 return (SET_ERROR(ENOTSUP));
663
664 if ((type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
665 type == ZFS_PROP_PROJECTOBJQUOTA ||
666 type == ZFS_PROP_PROJECTOBJUSED) &&
667 !dmu_objset_projectquota_present(zfsvfs->z_os))
668 return (SET_ERROR(ENOTSUP));
669
670 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
671 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
672 type == ZFS_PROP_PROJECTOBJUSED ||
673 type == ZFS_PROP_PROJECTOBJQUOTA) &&
674 !dmu_objset_userobjspace_present(zfsvfs->z_os))
675 return (SET_ERROR(ENOTSUP));
676
677 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
678 if (obj == ZFS_NO_OBJECT) {
679 *bufsizep = 0;
680 return (0);
681 }
682
683 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
684 type == ZFS_PROP_PROJECTOBJUSED)
685 offset = DMU_OBJACCT_PREFIX_LEN;
686
687 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
688 (error = zap_cursor_retrieve(&zc, &za)) == 0;
689 zap_cursor_advance(&zc)) {
690 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
691 *bufsizep)
692 break;
693
694 /*
695 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
696 * when dealing with block quota and vice versa.
697 */
698 if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
699 DMU_OBJACCT_PREFIX_LEN) == 0))
700 continue;
701
702 fuidstr_to_sid(zfsvfs, za.za_name + offset,
703 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
704
705 buf->zu_space = za.za_first_integer;
706 buf++;
707 }
708 if (error == ENOENT)
709 error = 0;
710
711 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
712 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
713 *cookiep = zap_cursor_serialize(&zc);
714 zap_cursor_fini(&zc);
715 return (error);
716 }
717
718 /*
719 * buf must be big enough (eg, 32 bytes)
720 */
721 static int
722 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
723 char *buf, boolean_t addok)
724 {
725 uint64_t fuid;
726 int domainid = 0;
727
728 if (domain && domain[0]) {
729 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
730 if (domainid == -1)
731 return (SET_ERROR(ENOENT));
732 }
733 fuid = FUID_ENCODE(domainid, rid);
734 (void) sprintf(buf, "%llx", (longlong_t)fuid);
735 return (0);
736 }
737
738 int
739 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
740 const char *domain, uint64_t rid, uint64_t *valp)
741 {
742 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
743 int offset = 0;
744 int err;
745 uint64_t obj;
746
747 *valp = 0;
748
749 if (!dmu_objset_userspace_present(zfsvfs->z_os))
750 return (SET_ERROR(ENOTSUP));
751
752 if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
753 type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
754 type == ZFS_PROP_PROJECTOBJUSED ||
755 type == ZFS_PROP_PROJECTOBJQUOTA) &&
756 !dmu_objset_userobjspace_present(zfsvfs->z_os))
757 return (SET_ERROR(ENOTSUP));
758
759 if (type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
760 type == ZFS_PROP_PROJECTOBJQUOTA ||
761 type == ZFS_PROP_PROJECTOBJUSED) {
762 if (!dmu_objset_projectquota_present(zfsvfs->z_os))
763 return (SET_ERROR(ENOTSUP));
764 if (!zpl_is_valid_projid(rid))
765 return (SET_ERROR(EINVAL));
766 }
767
768 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
769 if (obj == ZFS_NO_OBJECT)
770 return (0);
771
772 if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
773 type == ZFS_PROP_PROJECTOBJUSED) {
774 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
775 offset = DMU_OBJACCT_PREFIX_LEN;
776 }
777
778 err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
779 if (err)
780 return (err);
781
782 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
783 if (err == ENOENT)
784 err = 0;
785 return (err);
786 }
787
788 int
789 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
790 const char *domain, uint64_t rid, uint64_t quota)
791 {
792 char buf[32];
793 int err;
794 dmu_tx_t *tx;
795 uint64_t *objp;
796 boolean_t fuid_dirtied;
797
798 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
799 return (SET_ERROR(ENOTSUP));
800
801 switch (type) {
802 case ZFS_PROP_USERQUOTA:
803 objp = &zfsvfs->z_userquota_obj;
804 break;
805 case ZFS_PROP_GROUPQUOTA:
806 objp = &zfsvfs->z_groupquota_obj;
807 break;
808 case ZFS_PROP_USEROBJQUOTA:
809 objp = &zfsvfs->z_userobjquota_obj;
810 break;
811 case ZFS_PROP_GROUPOBJQUOTA:
812 objp = &zfsvfs->z_groupobjquota_obj;
813 break;
814 case ZFS_PROP_PROJECTQUOTA:
815 if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
816 return (SET_ERROR(ENOTSUP));
817 if (!zpl_is_valid_projid(rid))
818 return (SET_ERROR(EINVAL));
819
820 objp = &zfsvfs->z_projectquota_obj;
821 break;
822 case ZFS_PROP_PROJECTOBJQUOTA:
823 if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
824 return (SET_ERROR(ENOTSUP));
825 if (!zpl_is_valid_projid(rid))
826 return (SET_ERROR(EINVAL));
827
828 objp = &zfsvfs->z_projectobjquota_obj;
829 break;
830 default:
831 return (SET_ERROR(EINVAL));
832 }
833
834 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
835 if (err)
836 return (err);
837 fuid_dirtied = zfsvfs->z_fuid_dirty;
838
839 tx = dmu_tx_create(zfsvfs->z_os);
840 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
841 if (*objp == 0) {
842 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
843 zfs_userquota_prop_prefixes[type]);
844 }
845 if (fuid_dirtied)
846 zfs_fuid_txhold(zfsvfs, tx);
847 err = dmu_tx_assign(tx, TXG_WAIT);
848 if (err) {
849 dmu_tx_abort(tx);
850 return (err);
851 }
852
853 mutex_enter(&zfsvfs->z_lock);
854 if (*objp == 0) {
855 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
856 DMU_OT_NONE, 0, tx);
857 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
858 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
859 }
860 mutex_exit(&zfsvfs->z_lock);
861
862 if (quota == 0) {
863 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
864 if (err == ENOENT)
865 err = 0;
866 } else {
867 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
868 }
869 ASSERT(err == 0);
870 if (fuid_dirtied)
871 zfs_fuid_sync(zfsvfs, tx);
872 dmu_tx_commit(tx);
873 return (err);
874 }
875
876 boolean_t
877 zfs_id_overobjquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
878 {
879 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
880 uint64_t used, quota, quotaobj;
881 int err;
882
883 if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
884 if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os)) {
885 dsl_pool_config_enter(
886 dmu_objset_pool(zfsvfs->z_os), FTAG);
887 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
888 dsl_pool_config_exit(
889 dmu_objset_pool(zfsvfs->z_os), FTAG);
890 }
891 return (B_FALSE);
892 }
893
894 if (usedobj == DMU_PROJECTUSED_OBJECT) {
895 if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
896 if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
897 dsl_pool_config_enter(
898 dmu_objset_pool(zfsvfs->z_os), FTAG);
899 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
900 dsl_pool_config_exit(
901 dmu_objset_pool(zfsvfs->z_os), FTAG);
902 }
903 return (B_FALSE);
904 }
905 quotaobj = zfsvfs->z_projectobjquota_obj;
906 } else if (usedobj == DMU_USERUSED_OBJECT) {
907 quotaobj = zfsvfs->z_userobjquota_obj;
908 } else if (usedobj == DMU_GROUPUSED_OBJECT) {
909 quotaobj = zfsvfs->z_groupobjquota_obj;
910 } else {
911 return (B_FALSE);
912 }
913 if (quotaobj == 0 || zfsvfs->z_replay)
914 return (B_FALSE);
915
916 (void) sprintf(buf, "%llx", (longlong_t)id);
917 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
918 if (err != 0)
919 return (B_FALSE);
920
921 (void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)id);
922 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
923 if (err != 0)
924 return (B_FALSE);
925 return (used >= quota);
926 }
927
928 boolean_t
929 zfs_id_overblockquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
930 {
931 char buf[20];
932 uint64_t used, quota, quotaobj;
933 int err;
934
935 if (usedobj == DMU_PROJECTUSED_OBJECT) {
936 if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
937 if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
938 dsl_pool_config_enter(
939 dmu_objset_pool(zfsvfs->z_os), FTAG);
940 dmu_objset_id_quota_upgrade(zfsvfs->z_os);
941 dsl_pool_config_exit(
942 dmu_objset_pool(zfsvfs->z_os), FTAG);
943 }
944 return (B_FALSE);
945 }
946 quotaobj = zfsvfs->z_projectquota_obj;
947 } else if (usedobj == DMU_USERUSED_OBJECT) {
948 quotaobj = zfsvfs->z_userquota_obj;
949 } else if (usedobj == DMU_GROUPUSED_OBJECT) {
950 quotaobj = zfsvfs->z_groupquota_obj;
951 } else {
952 return (B_FALSE);
953 }
954 if (quotaobj == 0 || zfsvfs->z_replay)
955 return (B_FALSE);
956
957 (void) sprintf(buf, "%llx", (longlong_t)id);
958 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
959 if (err != 0)
960 return (B_FALSE);
961
962 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
963 if (err != 0)
964 return (B_FALSE);
965 return (used >= quota);
966 }
967
968 boolean_t
969 zfs_id_overquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
970 {
971 return (zfs_id_overblockquota(zfsvfs, usedobj, id) ||
972 zfs_id_overobjquota(zfsvfs, usedobj, id));
973 }
974
975 /*
976 * Associate this zfsvfs with the given objset, which must be owned.
977 * This will cache a bunch of on-disk state from the objset in the
978 * zfsvfs.
979 */
980 static int
981 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
982 {
983 int error;
984 uint64_t val;
985
986 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
987 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
988 zfsvfs->z_os = os;
989
990 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
991 if (error != 0)
992 return (error);
993 if (zfsvfs->z_version >
994 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
995 (void) printk("Can't mount a version %lld file system "
996 "on a version %lld pool\n. Pool must be upgraded to mount "
997 "this file system.", (u_longlong_t)zfsvfs->z_version,
998 (u_longlong_t)spa_version(dmu_objset_spa(os)));
999 return (SET_ERROR(ENOTSUP));
1000 }
1001 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1002 if (error != 0)
1003 return (error);
1004 zfsvfs->z_norm = (int)val;
1005
1006 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1007 if (error != 0)
1008 return (error);
1009 zfsvfs->z_utf8 = (val != 0);
1010
1011 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1012 if (error != 0)
1013 return (error);
1014 zfsvfs->z_case = (uint_t)val;
1015
1016 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
1017 return (error);
1018 zfsvfs->z_acl_type = (uint_t)val;
1019
1020 /*
1021 * Fold case on file systems that are always or sometimes case
1022 * insensitive.
1023 */
1024 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1025 zfsvfs->z_case == ZFS_CASE_MIXED)
1026 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1027
1028 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1029 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1030
1031 uint64_t sa_obj = 0;
1032 if (zfsvfs->z_use_sa) {
1033 /* should either have both of these objects or none */
1034 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1035 &sa_obj);
1036 if (error != 0)
1037 return (error);
1038
1039 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
1040 if ((error == 0) && (val == ZFS_XATTR_SA))
1041 zfsvfs->z_xattr_sa = B_TRUE;
1042 }
1043
1044 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1045 &zfsvfs->z_attr_table);
1046 if (error != 0)
1047 return (error);
1048
1049 if (zfsvfs->z_version >= ZPL_VERSION_SA)
1050 sa_register_update_callback(os, zfs_sa_upgrade);
1051
1052 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1053 &zfsvfs->z_root);
1054 if (error != 0)
1055 return (error);
1056 ASSERT(zfsvfs->z_root != 0);
1057
1058 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1059 &zfsvfs->z_unlinkedobj);
1060 if (error != 0)
1061 return (error);
1062
1063 error = zap_lookup(os, MASTER_NODE_OBJ,
1064 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1065 8, 1, &zfsvfs->z_userquota_obj);
1066 if (error == ENOENT)
1067 zfsvfs->z_userquota_obj = 0;
1068 else if (error != 0)
1069 return (error);
1070
1071 error = zap_lookup(os, MASTER_NODE_OBJ,
1072 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1073 8, 1, &zfsvfs->z_groupquota_obj);
1074 if (error == ENOENT)
1075 zfsvfs->z_groupquota_obj = 0;
1076 else if (error != 0)
1077 return (error);
1078
1079 error = zap_lookup(os, MASTER_NODE_OBJ,
1080 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
1081 8, 1, &zfsvfs->z_projectquota_obj);
1082 if (error == ENOENT)
1083 zfsvfs->z_projectquota_obj = 0;
1084 else if (error != 0)
1085 return (error);
1086
1087 error = zap_lookup(os, MASTER_NODE_OBJ,
1088 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1089 8, 1, &zfsvfs->z_userobjquota_obj);
1090 if (error == ENOENT)
1091 zfsvfs->z_userobjquota_obj = 0;
1092 else if (error != 0)
1093 return (error);
1094
1095 error = zap_lookup(os, MASTER_NODE_OBJ,
1096 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1097 8, 1, &zfsvfs->z_groupobjquota_obj);
1098 if (error == ENOENT)
1099 zfsvfs->z_groupobjquota_obj = 0;
1100 else if (error != 0)
1101 return (error);
1102
1103 error = zap_lookup(os, MASTER_NODE_OBJ,
1104 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
1105 8, 1, &zfsvfs->z_projectobjquota_obj);
1106 if (error == ENOENT)
1107 zfsvfs->z_projectobjquota_obj = 0;
1108 else if (error != 0)
1109 return (error);
1110
1111 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1112 &zfsvfs->z_fuid_obj);
1113 if (error == ENOENT)
1114 zfsvfs->z_fuid_obj = 0;
1115 else if (error != 0)
1116 return (error);
1117
1118 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1119 &zfsvfs->z_shares_dir);
1120 if (error == ENOENT)
1121 zfsvfs->z_shares_dir = 0;
1122 else if (error != 0)
1123 return (error);
1124
1125 return (0);
1126 }
1127
1128 int
1129 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
1130 {
1131 objset_t *os;
1132 zfsvfs_t *zfsvfs;
1133 int error;
1134 boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
1135
1136 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1137
1138 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
1139 if (error != 0) {
1140 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1141 return (error);
1142 }
1143
1144 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1145 if (error != 0) {
1146 dmu_objset_disown(os, B_TRUE, zfsvfs);
1147 }
1148 return (error);
1149 }
1150
1151 int
1152 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1153 {
1154 int error;
1155
1156 zfsvfs->z_vfs = NULL;
1157 zfsvfs->z_sb = NULL;
1158 zfsvfs->z_parent = zfsvfs;
1159
1160 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1161 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1162 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1163 offsetof(znode_t, z_link_node));
1164 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1165 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1166 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1167
1168 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
1169 ZFS_OBJ_MTX_MAX);
1170 zfsvfs->z_hold_size = size;
1171 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
1172 KM_SLEEP);
1173 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
1174 for (int i = 0; i != size; i++) {
1175 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
1176 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
1177 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
1178 }
1179
1180 error = zfsvfs_init(zfsvfs, os);
1181 if (error != 0) {
1182 *zfvp = NULL;
1183 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1184 return (error);
1185 }
1186
1187 *zfvp = zfsvfs;
1188 return (0);
1189 }
1190
1191 static int
1192 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1193 {
1194 int error;
1195 boolean_t readonly = zfs_is_readonly(zfsvfs);
1196
1197 error = zfs_register_callbacks(zfsvfs->z_vfs);
1198 if (error)
1199 return (error);
1200
1201 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1202
1203 /*
1204 * If we are not mounting (ie: online recv), then we don't
1205 * have to worry about replaying the log as we blocked all
1206 * operations out since we closed the ZIL.
1207 */
1208 if (mounting) {
1209 /*
1210 * During replay we remove the read only flag to
1211 * allow replays to succeed.
1212 */
1213 if (readonly != 0)
1214 readonly_changed_cb(zfsvfs, B_FALSE);
1215 else
1216 zfs_unlinked_drain(zfsvfs);
1217
1218 /*
1219 * Parse and replay the intent log.
1220 *
1221 * Because of ziltest, this must be done after
1222 * zfs_unlinked_drain(). (Further note: ziltest
1223 * doesn't use readonly mounts, where
1224 * zfs_unlinked_drain() isn't called.) This is because
1225 * ziltest causes spa_sync() to think it's committed,
1226 * but actually it is not, so the intent log contains
1227 * many txg's worth of changes.
1228 *
1229 * In particular, if object N is in the unlinked set in
1230 * the last txg to actually sync, then it could be
1231 * actually freed in a later txg and then reallocated
1232 * in a yet later txg. This would write a "create
1233 * object N" record to the intent log. Normally, this
1234 * would be fine because the spa_sync() would have
1235 * written out the fact that object N is free, before
1236 * we could write the "create object N" intent log
1237 * record.
1238 *
1239 * But when we are in ziltest mode, we advance the "open
1240 * txg" without actually spa_sync()-ing the changes to
1241 * disk. So we would see that object N is still
1242 * allocated and in the unlinked set, and there is an
1243 * intent log record saying to allocate it.
1244 */
1245 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1246 if (zil_replay_disable) {
1247 zil_destroy(zfsvfs->z_log, B_FALSE);
1248 } else {
1249 zfsvfs->z_replay = B_TRUE;
1250 zil_replay(zfsvfs->z_os, zfsvfs,
1251 zfs_replay_vector);
1252 zfsvfs->z_replay = B_FALSE;
1253 }
1254 }
1255
1256 /* restore readonly bit */
1257 if (readonly != 0)
1258 readonly_changed_cb(zfsvfs, B_TRUE);
1259 }
1260
1261 /*
1262 * Set the objset user_ptr to track its zfsvfs.
1263 */
1264 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1265 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1266 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1267
1268 return (0);
1269 }
1270
1271 void
1272 zfsvfs_free(zfsvfs_t *zfsvfs)
1273 {
1274 int i, size = zfsvfs->z_hold_size;
1275
1276 zfs_fuid_destroy(zfsvfs);
1277
1278 mutex_destroy(&zfsvfs->z_znodes_lock);
1279 mutex_destroy(&zfsvfs->z_lock);
1280 list_destroy(&zfsvfs->z_all_znodes);
1281 rrm_destroy(&zfsvfs->z_teardown_lock);
1282 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1283 rw_destroy(&zfsvfs->z_fuid_lock);
1284 for (i = 0; i != size; i++) {
1285 avl_destroy(&zfsvfs->z_hold_trees[i]);
1286 mutex_destroy(&zfsvfs->z_hold_locks[i]);
1287 }
1288 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
1289 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
1290 zfsvfs_vfs_free(zfsvfs->z_vfs);
1291 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1292 }
1293
1294 static void
1295 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1296 {
1297 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1298 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1299 }
1300
1301 void
1302 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1303 {
1304 objset_t *os = zfsvfs->z_os;
1305
1306 if (!dmu_objset_is_snapshot(os))
1307 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1308 }
1309
1310 #ifdef HAVE_MLSLABEL
1311 /*
1312 * Check that the hex label string is appropriate for the dataset being
1313 * mounted into the global_zone proper.
1314 *
1315 * Return an error if the hex label string is not default or
1316 * admin_low/admin_high. For admin_low labels, the corresponding
1317 * dataset must be readonly.
1318 */
1319 int
1320 zfs_check_global_label(const char *dsname, const char *hexsl)
1321 {
1322 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1323 return (0);
1324 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1325 return (0);
1326 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1327 /* must be readonly */
1328 uint64_t rdonly;
1329
1330 if (dsl_prop_get_integer(dsname,
1331 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1332 return (SET_ERROR(EACCES));
1333 return (rdonly ? 0 : EACCES);
1334 }
1335 return (SET_ERROR(EACCES));
1336 }
1337 #endif /* HAVE_MLSLABEL */
1338
1339 static int
1340 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1341 uint32_t bshift)
1342 {
1343 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1344 uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1345 uint64_t quota;
1346 uint64_t used;
1347 int err;
1348
1349 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1350 err = id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, B_FALSE);
1351 if (err)
1352 return (err);
1353
1354 if (zfsvfs->z_projectquota_obj == 0)
1355 goto objs;
1356
1357 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1358 buf + offset, 8, 1, &quota);
1359 if (err == ENOENT)
1360 goto objs;
1361 else if (err)
1362 return (err);
1363
1364 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1365 buf + offset, 8, 1, &used);
1366 if (unlikely(err == ENOENT)) {
1367 uint32_t blksize;
1368 u_longlong_t nblocks;
1369
1370 /*
1371 * Quota accounting is async, so it is possible race case.
1372 * There is at least one object with the given project ID.
1373 */
1374 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1375 if (unlikely(zp->z_blksz == 0))
1376 blksize = zfsvfs->z_max_blksz;
1377
1378 used = blksize * nblocks;
1379 } else if (err) {
1380 return (err);
1381 }
1382
1383 statp->f_blocks = quota >> bshift;
1384 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1385 statp->f_bavail = statp->f_bfree;
1386
1387 objs:
1388 if (zfsvfs->z_projectobjquota_obj == 0)
1389 return (0);
1390
1391 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1392 buf + offset, 8, 1, &quota);
1393 if (err == ENOENT)
1394 return (0);
1395 else if (err)
1396 return (err);
1397
1398 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1399 buf, 8, 1, &used);
1400 if (unlikely(err == ENOENT)) {
1401 /*
1402 * Quota accounting is async, so it is possible race case.
1403 * There is at least one object with the given project ID.
1404 */
1405 used = 1;
1406 } else if (err) {
1407 return (err);
1408 }
1409
1410 statp->f_files = quota;
1411 statp->f_ffree = (quota > used) ? (quota - used) : 0;
1412
1413 return (0);
1414 }
1415
1416 int
1417 zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
1418 {
1419 zfsvfs_t *zfsvfs = dentry->d_sb->s_fs_info;
1420 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1421 uint64_t fsid;
1422 uint32_t bshift;
1423 int err = 0;
1424
1425 ZFS_ENTER(zfsvfs);
1426
1427 dmu_objset_space(zfsvfs->z_os,
1428 &refdbytes, &availbytes, &usedobjs, &availobjs);
1429
1430 fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1431 /*
1432 * The underlying storage pool actually uses multiple block
1433 * size. Under Solaris frsize (fragment size) is reported as
1434 * the smallest block size we support, and bsize (block size)
1435 * as the filesystem's maximum block size. Unfortunately,
1436 * under Linux the fragment size and block size are often used
1437 * interchangeably. Thus we are forced to report both of them
1438 * as the filesystem's maximum block size.
1439 */
1440 statp->f_frsize = zfsvfs->z_max_blksz;
1441 statp->f_bsize = zfsvfs->z_max_blksz;
1442 bshift = fls(statp->f_bsize) - 1;
1443
1444 /*
1445 * The following report "total" blocks of various kinds in
1446 * the file system, but reported in terms of f_bsize - the
1447 * "preferred" size.
1448 */
1449
1450 statp->f_blocks = (refdbytes + availbytes) >> bshift;
1451 statp->f_bfree = availbytes >> bshift;
1452 statp->f_bavail = statp->f_bfree; /* no root reservation */
1453
1454 /*
1455 * statvfs() should really be called statufs(), because it assumes
1456 * static metadata. ZFS doesn't preallocate files, so the best
1457 * we can do is report the max that could possibly fit in f_files,
1458 * and that minus the number actually used in f_ffree.
1459 * For f_ffree, report the smaller of the number of object available
1460 * and the number of blocks (each object will take at least a block).
1461 */
1462 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1463 statp->f_files = statp->f_ffree + usedobjs;
1464 statp->f_fsid.val[0] = (uint32_t)fsid;
1465 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1466 statp->f_type = ZFS_SUPER_MAGIC;
1467 statp->f_namelen = MAXNAMELEN - 1;
1468
1469 /*
1470 * We have all of 40 characters to stuff a string here.
1471 * Is there anything useful we could/should provide?
1472 */
1473 bzero(statp->f_spare, sizeof (statp->f_spare));
1474
1475 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1476 dmu_objset_projectquota_present(zfsvfs->z_os)) {
1477 znode_t *zp = ITOZ(dentry->d_inode);
1478
1479 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1480 zpl_is_valid_projid(zp->z_projid))
1481 err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1482 }
1483
1484 ZFS_EXIT(zfsvfs);
1485 return (err);
1486 }
1487
1488 int
1489 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1490 {
1491 znode_t *rootzp;
1492 int error;
1493
1494 ZFS_ENTER(zfsvfs);
1495
1496 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1497 if (error == 0)
1498 *ipp = ZTOI(rootzp);
1499
1500 ZFS_EXIT(zfsvfs);
1501 return (error);
1502 }
1503
1504 #ifdef HAVE_D_PRUNE_ALIASES
1505 /*
1506 * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1507 * To accommodate this we must improvise and manually walk the list of znodes
1508 * attempting to prune dentries in order to be able to drop the inodes.
1509 *
1510 * To avoid scanning the same znodes multiple times they are always rotated
1511 * to the end of the z_all_znodes list. New znodes are inserted at the
1512 * end of the list so we're always scanning the oldest znodes first.
1513 */
1514 static int
1515 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1516 {
1517 znode_t **zp_array, *zp;
1518 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1519 int objects = 0;
1520 int i = 0, j = 0;
1521
1522 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1523
1524 mutex_enter(&zfsvfs->z_znodes_lock);
1525 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1526
1527 if ((i++ > nr_to_scan) || (j >= max_array))
1528 break;
1529
1530 ASSERT(list_link_active(&zp->z_link_node));
1531 list_remove(&zfsvfs->z_all_znodes, zp);
1532 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1533
1534 /* Skip active znodes and .zfs entries */
1535 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1536 continue;
1537
1538 if (igrab(ZTOI(zp)) == NULL)
1539 continue;
1540
1541 zp_array[j] = zp;
1542 j++;
1543 }
1544 mutex_exit(&zfsvfs->z_znodes_lock);
1545
1546 for (i = 0; i < j; i++) {
1547 zp = zp_array[i];
1548
1549 ASSERT3P(zp, !=, NULL);
1550 d_prune_aliases(ZTOI(zp));
1551
1552 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1553 objects++;
1554
1555 iput(ZTOI(zp));
1556 }
1557
1558 kmem_free(zp_array, max_array * sizeof (znode_t *));
1559
1560 return (objects);
1561 }
1562 #endif /* HAVE_D_PRUNE_ALIASES */
1563
1564 /*
1565 * The ARC has requested that the filesystem drop entries from the dentry
1566 * and inode caches. This can occur when the ARC needs to free meta data
1567 * blocks but can't because they are all pinned by entries in these caches.
1568 */
1569 int
1570 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1571 {
1572 zfsvfs_t *zfsvfs = sb->s_fs_info;
1573 int error = 0;
1574 #if defined(HAVE_SHRINK) || defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1575 struct shrinker *shrinker = &sb->s_shrink;
1576 struct shrink_control sc = {
1577 .nr_to_scan = nr_to_scan,
1578 .gfp_mask = GFP_KERNEL,
1579 };
1580 #endif
1581
1582 ZFS_ENTER(zfsvfs);
1583
1584 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1585 defined(SHRINK_CONTROL_HAS_NID) && \
1586 defined(SHRINKER_NUMA_AWARE)
1587 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1588 *objects = 0;
1589 for_each_online_node(sc.nid) {
1590 *objects += (*shrinker->scan_objects)(shrinker, &sc);
1591 }
1592 } else {
1593 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1594 }
1595
1596 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1597 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1598 #elif defined(HAVE_SHRINK)
1599 *objects = (*shrinker->shrink)(shrinker, &sc);
1600 #elif defined(HAVE_D_PRUNE_ALIASES)
1601 #define D_PRUNE_ALIASES_IS_DEFAULT
1602 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1603 #else
1604 #error "No available dentry and inode cache pruning mechanism."
1605 #endif
1606
1607 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1608 #undef D_PRUNE_ALIASES_IS_DEFAULT
1609 /*
1610 * Fall back to zfs_prune_aliases if the kernel's per-superblock
1611 * shrinker couldn't free anything, possibly due to the inodes being
1612 * allocated in a different memcg.
1613 */
1614 if (*objects == 0)
1615 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1616 #endif
1617
1618 ZFS_EXIT(zfsvfs);
1619
1620 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1621 "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1622 nr_to_scan, *objects, error);
1623
1624 return (error);
1625 }
1626
1627 /*
1628 * Teardown the zfsvfs_t.
1629 *
1630 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1631 * and 'z_teardown_inactive_lock' held.
1632 */
1633 static int
1634 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1635 {
1636 znode_t *zp;
1637
1638 /*
1639 * If someone has not already unmounted this file system,
1640 * drain the iput_taskq to ensure all active references to the
1641 * zfsvfs_t have been handled only then can it be safely destroyed.
1642 */
1643 if (zfsvfs->z_os) {
1644 /*
1645 * If we're unmounting we have to wait for the list to
1646 * drain completely.
1647 *
1648 * If we're not unmounting there's no guarantee the list
1649 * will drain completely, but iputs run from the taskq
1650 * may add the parents of dir-based xattrs to the taskq
1651 * so we want to wait for these.
1652 *
1653 * We can safely read z_nr_znodes without locking because the
1654 * VFS has already blocked operations which add to the
1655 * z_all_znodes list and thus increment z_nr_znodes.
1656 */
1657 int round = 0;
1658 while (zfsvfs->z_nr_znodes > 0) {
1659 taskq_wait_outstanding(dsl_pool_iput_taskq(
1660 dmu_objset_pool(zfsvfs->z_os)), 0);
1661 if (++round > 1 && !unmounting)
1662 break;
1663 }
1664 }
1665
1666 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1667
1668 if (!unmounting) {
1669 /*
1670 * We purge the parent filesystem's super block as the
1671 * parent filesystem and all of its snapshots have their
1672 * inode's super block set to the parent's filesystem's
1673 * super block. Note, 'z_parent' is self referential
1674 * for non-snapshots.
1675 */
1676 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1677 }
1678
1679 /*
1680 * Close the zil. NB: Can't close the zil while zfs_inactive
1681 * threads are blocked as zil_close can call zfs_inactive.
1682 */
1683 if (zfsvfs->z_log) {
1684 zil_close(zfsvfs->z_log);
1685 zfsvfs->z_log = NULL;
1686 }
1687
1688 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1689
1690 /*
1691 * If we are not unmounting (ie: online recv) and someone already
1692 * unmounted this file system while we were doing the switcheroo,
1693 * or a reopen of z_os failed then just bail out now.
1694 */
1695 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1696 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1697 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1698 return (SET_ERROR(EIO));
1699 }
1700
1701 /*
1702 * At this point there are no VFS ops active, and any new VFS ops
1703 * will fail with EIO since we have z_teardown_lock for writer (only
1704 * relevant for forced unmount).
1705 *
1706 * Release all holds on dbufs.
1707 */
1708 if (!unmounting) {
1709 mutex_enter(&zfsvfs->z_znodes_lock);
1710 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1711 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1712 if (zp->z_sa_hdl)
1713 zfs_znode_dmu_fini(zp);
1714 }
1715 mutex_exit(&zfsvfs->z_znodes_lock);
1716 }
1717
1718 /*
1719 * If we are unmounting, set the unmounted flag and let new VFS ops
1720 * unblock. zfs_inactive will have the unmounted behavior, and all
1721 * other VFS ops will fail with EIO.
1722 */
1723 if (unmounting) {
1724 zfsvfs->z_unmounted = B_TRUE;
1725 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1726 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1727 }
1728
1729 /*
1730 * z_os will be NULL if there was an error in attempting to reopen
1731 * zfsvfs, so just return as the properties had already been
1732 *
1733 * unregistered and cached data had been evicted before.
1734 */
1735 if (zfsvfs->z_os == NULL)
1736 return (0);
1737
1738 /*
1739 * Unregister properties.
1740 */
1741 zfs_unregister_callbacks(zfsvfs);
1742
1743 /*
1744 * Evict cached data
1745 */
1746 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1747 !zfs_is_readonly(zfsvfs))
1748 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1749 dmu_objset_evict_dbufs(zfsvfs->z_os);
1750
1751 return (0);
1752 }
1753
1754 #if !defined(HAVE_2ARGS_BDI_SETUP_AND_REGISTER) && \
1755 !defined(HAVE_3ARGS_BDI_SETUP_AND_REGISTER)
1756 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1757 #endif
1758
1759 int
1760 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1761 {
1762 const char *osname = zm->mnt_osname;
1763 struct inode *root_inode;
1764 uint64_t recordsize;
1765 int error = 0;
1766 zfsvfs_t *zfsvfs = NULL;
1767 vfs_t *vfs = NULL;
1768
1769 ASSERT(zm);
1770 ASSERT(osname);
1771
1772 error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1773 if (error)
1774 return (error);
1775
1776 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1777 if (error) {
1778 zfsvfs_vfs_free(vfs);
1779 goto out;
1780 }
1781
1782 if ((error = dsl_prop_get_integer(osname, "recordsize",
1783 &recordsize, NULL))) {
1784 zfsvfs_vfs_free(vfs);
1785 goto out;
1786 }
1787
1788 vfs->vfs_data = zfsvfs;
1789 zfsvfs->z_vfs = vfs;
1790 zfsvfs->z_sb = sb;
1791 sb->s_fs_info = zfsvfs;
1792 sb->s_magic = ZFS_SUPER_MAGIC;
1793 sb->s_maxbytes = MAX_LFS_FILESIZE;
1794 sb->s_time_gran = 1;
1795 sb->s_blocksize = recordsize;
1796 sb->s_blocksize_bits = ilog2(recordsize);
1797
1798 error = -zpl_bdi_setup(sb, "zfs");
1799 if (error)
1800 goto out;
1801
1802 sb->s_bdi->ra_pages = 0;
1803
1804 /* Set callback operations for the file system. */
1805 sb->s_op = &zpl_super_operations;
1806 sb->s_xattr = zpl_xattr_handlers;
1807 sb->s_export_op = &zpl_export_operations;
1808 #ifdef HAVE_S_D_OP
1809 sb->s_d_op = &zpl_dentry_operations;
1810 #endif /* HAVE_S_D_OP */
1811
1812 /* Set features for file system. */
1813 zfs_set_fuid_feature(zfsvfs);
1814
1815 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1816 uint64_t pval;
1817
1818 atime_changed_cb(zfsvfs, B_FALSE);
1819 readonly_changed_cb(zfsvfs, B_TRUE);
1820 if ((error = dsl_prop_get_integer(osname,
1821 "xattr", &pval, NULL)))
1822 goto out;
1823 xattr_changed_cb(zfsvfs, pval);
1824 if ((error = dsl_prop_get_integer(osname,
1825 "acltype", &pval, NULL)))
1826 goto out;
1827 acltype_changed_cb(zfsvfs, pval);
1828 zfsvfs->z_issnap = B_TRUE;
1829 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1830 zfsvfs->z_snap_defer_time = jiffies;
1831
1832 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1833 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1834 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1835 } else {
1836 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1837 goto out;
1838 }
1839
1840 /* Allocate a root inode for the filesystem. */
1841 error = zfs_root(zfsvfs, &root_inode);
1842 if (error) {
1843 (void) zfs_umount(sb);
1844 goto out;
1845 }
1846
1847 /* Allocate a root dentry for the filesystem */
1848 sb->s_root = d_make_root(root_inode);
1849 if (sb->s_root == NULL) {
1850 (void) zfs_umount(sb);
1851 error = SET_ERROR(ENOMEM);
1852 goto out;
1853 }
1854
1855 if (!zfsvfs->z_issnap)
1856 zfsctl_create(zfsvfs);
1857
1858 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1859 out:
1860 if (error) {
1861 if (zfsvfs != NULL) {
1862 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1863 zfsvfs_free(zfsvfs);
1864 }
1865 /*
1866 * make sure we don't have dangling sb->s_fs_info which
1867 * zfs_preumount will use.
1868 */
1869 sb->s_fs_info = NULL;
1870 }
1871
1872 return (error);
1873 }
1874
1875 /*
1876 * Called when an unmount is requested and certain sanity checks have
1877 * already passed. At this point no dentries or inodes have been reclaimed
1878 * from their respective caches. We drop the extra reference on the .zfs
1879 * control directory to allow everything to be reclaimed. All snapshots
1880 * must already have been unmounted to reach this point.
1881 */
1882 void
1883 zfs_preumount(struct super_block *sb)
1884 {
1885 zfsvfs_t *zfsvfs = sb->s_fs_info;
1886
1887 /* zfsvfs is NULL when zfs_domount fails during mount */
1888 if (zfsvfs) {
1889 zfsctl_destroy(sb->s_fs_info);
1890 /*
1891 * Wait for iput_async before entering evict_inodes in
1892 * generic_shutdown_super. The reason we must finish before
1893 * evict_inodes is when lazytime is on, or when zfs_purgedir
1894 * calls zfs_zget, iput would bump i_count from 0 to 1. This
1895 * would race with the i_count check in evict_inodes. This means
1896 * it could destroy the inode while we are still using it.
1897 *
1898 * We wait for two passes. xattr directories in the first pass
1899 * may add xattr entries in zfs_purgedir, so in the second pass
1900 * we wait for them. We don't use taskq_wait here because it is
1901 * a pool wide taskq. Other mounted filesystems can constantly
1902 * do iput_async and there's no guarantee when taskq will be
1903 * empty.
1904 */
1905 taskq_wait_outstanding(dsl_pool_iput_taskq(
1906 dmu_objset_pool(zfsvfs->z_os)), 0);
1907 taskq_wait_outstanding(dsl_pool_iput_taskq(
1908 dmu_objset_pool(zfsvfs->z_os)), 0);
1909 }
1910 }
1911
1912 /*
1913 * Called once all other unmount released tear down has occurred.
1914 * It is our responsibility to release any remaining infrastructure.
1915 */
1916 /*ARGSUSED*/
1917 int
1918 zfs_umount(struct super_block *sb)
1919 {
1920 zfsvfs_t *zfsvfs = sb->s_fs_info;
1921 objset_t *os;
1922
1923 if (zfsvfs->z_arc_prune != NULL)
1924 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1925 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1926 os = zfsvfs->z_os;
1927 zpl_bdi_destroy(sb);
1928
1929 /*
1930 * z_os will be NULL if there was an error in
1931 * attempting to reopen zfsvfs.
1932 */
1933 if (os != NULL) {
1934 /*
1935 * Unset the objset user_ptr.
1936 */
1937 mutex_enter(&os->os_user_ptr_lock);
1938 dmu_objset_set_user(os, NULL);
1939 mutex_exit(&os->os_user_ptr_lock);
1940
1941 /*
1942 * Finally release the objset
1943 */
1944 dmu_objset_disown(os, B_TRUE, zfsvfs);
1945 }
1946
1947 zfsvfs_free(zfsvfs);
1948 return (0);
1949 }
1950
1951 int
1952 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1953 {
1954 zfsvfs_t *zfsvfs = sb->s_fs_info;
1955 vfs_t *vfsp;
1956 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1957 int error;
1958
1959 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1960 !(*flags & MS_RDONLY)) {
1961 *flags |= MS_RDONLY;
1962 return (EROFS);
1963 }
1964
1965 error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1966 if (error)
1967 return (error);
1968
1969 zfs_unregister_callbacks(zfsvfs);
1970 zfsvfs_vfs_free(zfsvfs->z_vfs);
1971
1972 vfsp->vfs_data = zfsvfs;
1973 zfsvfs->z_vfs = vfsp;
1974 if (!issnap)
1975 (void) zfs_register_callbacks(vfsp);
1976
1977 return (error);
1978 }
1979
1980 int
1981 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1982 {
1983 zfsvfs_t *zfsvfs = sb->s_fs_info;
1984 znode_t *zp;
1985 uint64_t object = 0;
1986 uint64_t fid_gen = 0;
1987 uint64_t gen_mask;
1988 uint64_t zp_gen;
1989 int i, err;
1990
1991 *ipp = NULL;
1992
1993 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1994 zfid_short_t *zfid = (zfid_short_t *)fidp;
1995
1996 for (i = 0; i < sizeof (zfid->zf_object); i++)
1997 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1998
1999 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2000 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2001 } else {
2002 return (SET_ERROR(EINVAL));
2003 }
2004
2005 /* LONG_FID_LEN means snapdirs */
2006 if (fidp->fid_len == LONG_FID_LEN) {
2007 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2008 uint64_t objsetid = 0;
2009 uint64_t setgen = 0;
2010
2011 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2012 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2013
2014 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2015 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2016
2017 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
2018 dprintf("snapdir fid: objsetid (%llu) != "
2019 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
2020 objsetid, ZFSCTL_INO_SNAPDIRS, object);
2021
2022 return (SET_ERROR(EINVAL));
2023 }
2024
2025 if (fid_gen > 1 || setgen != 0) {
2026 dprintf("snapdir fid: fid_gen (%llu) and setgen "
2027 "(%llu)\n", fid_gen, setgen);
2028 return (SET_ERROR(EINVAL));
2029 }
2030
2031 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
2032 }
2033
2034 ZFS_ENTER(zfsvfs);
2035 /* A zero fid_gen means we are in the .zfs control directories */
2036 if (fid_gen == 0 &&
2037 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2038 *ipp = zfsvfs->z_ctldir;
2039 ASSERT(*ipp != NULL);
2040 if (object == ZFSCTL_INO_SNAPDIR) {
2041 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
2042 0, kcred, NULL, NULL) == 0);
2043 } else {
2044 igrab(*ipp);
2045 }
2046 ZFS_EXIT(zfsvfs);
2047 return (0);
2048 }
2049
2050 gen_mask = -1ULL >> (64 - 8 * i);
2051
2052 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
2053 if ((err = zfs_zget(zfsvfs, object, &zp))) {
2054 ZFS_EXIT(zfsvfs);
2055 return (err);
2056 }
2057
2058 /* Don't export xattr stuff */
2059 if (zp->z_pflags & ZFS_XATTR) {
2060 iput(ZTOI(zp));
2061 ZFS_EXIT(zfsvfs);
2062 return (SET_ERROR(ENOENT));
2063 }
2064
2065 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2066 sizeof (uint64_t));
2067 zp_gen = zp_gen & gen_mask;
2068 if (zp_gen == 0)
2069 zp_gen = 1;
2070 if ((fid_gen == 0) && (zfsvfs->z_root == object))
2071 fid_gen = zp_gen;
2072 if (zp->z_unlinked || zp_gen != fid_gen) {
2073 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
2074 fid_gen);
2075 iput(ZTOI(zp));
2076 ZFS_EXIT(zfsvfs);
2077 return (SET_ERROR(ENOENT));
2078 }
2079
2080 *ipp = ZTOI(zp);
2081 if (*ipp)
2082 zfs_inode_update(ITOZ(*ipp));
2083
2084 ZFS_EXIT(zfsvfs);
2085 return (0);
2086 }
2087
2088 /*
2089 * Block out VFS ops and close zfsvfs_t
2090 *
2091 * Note, if successful, then we return with the 'z_teardown_lock' and
2092 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2093 * dataset and objset intact so that they can be atomically handed off during
2094 * a subsequent rollback or recv operation and the resume thereafter.
2095 */
2096 int
2097 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2098 {
2099 int error;
2100
2101 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2102 return (error);
2103
2104 return (0);
2105 }
2106
2107 /*
2108 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2109 * is an invariant across any of the operations that can be performed while the
2110 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2111 * are the same: the relevant objset and associated dataset are owned by
2112 * zfsvfs, held, and long held on entry.
2113 */
2114 int
2115 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2116 {
2117 int err, err2;
2118 znode_t *zp;
2119
2120 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2121 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2122
2123 /*
2124 * We already own this, so just update the objset_t, as the one we
2125 * had before may have been evicted.
2126 */
2127 objset_t *os;
2128 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2129 VERIFY(dsl_dataset_long_held(ds));
2130 VERIFY0(dmu_objset_from_ds(ds, &os));
2131
2132 err = zfsvfs_init(zfsvfs, os);
2133 if (err != 0)
2134 goto bail;
2135
2136 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2137
2138 zfs_set_fuid_feature(zfsvfs);
2139 zfsvfs->z_rollback_time = jiffies;
2140
2141 /*
2142 * Attempt to re-establish all the active inodes with their
2143 * dbufs. If a zfs_rezget() fails, then we unhash the inode
2144 * and mark it stale. This prevents a collision if a new
2145 * inode/object is created which must use the same inode
2146 * number. The stale inode will be be released when the
2147 * VFS prunes the dentry holding the remaining references
2148 * on the stale inode.
2149 */
2150 mutex_enter(&zfsvfs->z_znodes_lock);
2151 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2152 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2153 err2 = zfs_rezget(zp);
2154 if (err2) {
2155 remove_inode_hash(ZTOI(zp));
2156 zp->z_is_stale = B_TRUE;
2157 }
2158 }
2159 mutex_exit(&zfsvfs->z_znodes_lock);
2160
2161 bail:
2162 /* release the VFS ops */
2163 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2164 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2165
2166 if (err) {
2167 /*
2168 * Since we couldn't setup the sa framework, try to force
2169 * unmount this file system.
2170 */
2171 if (zfsvfs->z_os)
2172 (void) zfs_umount(zfsvfs->z_sb);
2173 }
2174 return (err);
2175 }
2176
2177 int
2178 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2179 {
2180 int error;
2181 objset_t *os = zfsvfs->z_os;
2182 dmu_tx_t *tx;
2183
2184 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2185 return (SET_ERROR(EINVAL));
2186
2187 if (newvers < zfsvfs->z_version)
2188 return (SET_ERROR(EINVAL));
2189
2190 if (zfs_spa_version_map(newvers) >
2191 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2192 return (SET_ERROR(ENOTSUP));
2193
2194 tx = dmu_tx_create(os);
2195 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2196 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2197 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2198 ZFS_SA_ATTRS);
2199 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2200 }
2201 error = dmu_tx_assign(tx, TXG_WAIT);
2202 if (error) {
2203 dmu_tx_abort(tx);
2204 return (error);
2205 }
2206
2207 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2208 8, 1, &newvers, tx);
2209
2210 if (error) {
2211 dmu_tx_commit(tx);
2212 return (error);
2213 }
2214
2215 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2216 uint64_t sa_obj;
2217
2218 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2219 SPA_VERSION_SA);
2220 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2221 DMU_OT_NONE, 0, tx);
2222
2223 error = zap_add(os, MASTER_NODE_OBJ,
2224 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2225 ASSERT0(error);
2226
2227 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2228 sa_register_update_callback(os, zfs_sa_upgrade);
2229 }
2230
2231 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2232 "from %llu to %llu", zfsvfs->z_version, newvers);
2233
2234 dmu_tx_commit(tx);
2235
2236 zfsvfs->z_version = newvers;
2237 os->os_version = newvers;
2238
2239 zfs_set_fuid_feature(zfsvfs);
2240
2241 return (0);
2242 }
2243
2244 /*
2245 * Read a property stored within the master node.
2246 */
2247 int
2248 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2249 {
2250 uint64_t *cached_copy = NULL;
2251
2252 /*
2253 * Figure out where in the objset_t the cached copy would live, if it
2254 * is available for the requested property.
2255 */
2256 if (os != NULL) {
2257 switch (prop) {
2258 case ZFS_PROP_VERSION:
2259 cached_copy = &os->os_version;
2260 break;
2261 case ZFS_PROP_NORMALIZE:
2262 cached_copy = &os->os_normalization;
2263 break;
2264 case ZFS_PROP_UTF8ONLY:
2265 cached_copy = &os->os_utf8only;
2266 break;
2267 case ZFS_PROP_CASE:
2268 cached_copy = &os->os_casesensitivity;
2269 break;
2270 default:
2271 break;
2272 }
2273 }
2274 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2275 *value = *cached_copy;
2276 return (0);
2277 }
2278
2279 /*
2280 * If the property wasn't cached, look up the file system's value for
2281 * the property. For the version property, we look up a slightly
2282 * different string.
2283 */
2284 const char *pname;
2285 int error = ENOENT;
2286 if (prop == ZFS_PROP_VERSION)
2287 pname = ZPL_VERSION_STR;
2288 else
2289 pname = zfs_prop_to_name(prop);
2290
2291 if (os != NULL) {
2292 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2293 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2294 }
2295
2296 if (error == ENOENT) {
2297 /* No value set, use the default value */
2298 switch (prop) {
2299 case ZFS_PROP_VERSION:
2300 *value = ZPL_VERSION;
2301 break;
2302 case ZFS_PROP_NORMALIZE:
2303 case ZFS_PROP_UTF8ONLY:
2304 *value = 0;
2305 break;
2306 case ZFS_PROP_CASE:
2307 *value = ZFS_CASE_SENSITIVE;
2308 break;
2309 case ZFS_PROP_ACLTYPE:
2310 *value = ZFS_ACLTYPE_OFF;
2311 break;
2312 default:
2313 return (error);
2314 }
2315 error = 0;
2316 }
2317
2318 /*
2319 * If one of the methods for getting the property value above worked,
2320 * copy it into the objset_t's cache.
2321 */
2322 if (error == 0 && cached_copy != NULL) {
2323 *cached_copy = *value;
2324 }
2325
2326 return (error);
2327 }
2328
2329 /*
2330 * Return true if the coresponding vfs's unmounted flag is set.
2331 * Otherwise return false.
2332 * If this function returns true we know VFS unmount has been initiated.
2333 */
2334 boolean_t
2335 zfs_get_vfs_flag_unmounted(objset_t *os)
2336 {
2337 zfsvfs_t *zfvp;
2338 boolean_t unmounted = B_FALSE;
2339
2340 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2341
2342 mutex_enter(&os->os_user_ptr_lock);
2343 zfvp = dmu_objset_get_user(os);
2344 if (zfvp != NULL && zfvp->z_unmounted)
2345 unmounted = B_TRUE;
2346 mutex_exit(&os->os_user_ptr_lock);
2347
2348 return (unmounted);
2349 }
2350
2351 void
2352 zfs_init(void)
2353 {
2354 zfsctl_init();
2355 zfs_znode_init();
2356 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2357 register_filesystem(&zpl_fs_type);
2358 }
2359
2360 void
2361 zfs_fini(void)
2362 {
2363 /*
2364 * we don't use outstanding because zpl_posix_acl_free might add more.
2365 */
2366 taskq_wait(system_delay_taskq);
2367 taskq_wait(system_taskq);
2368 unregister_filesystem(&zpl_fs_type);
2369 zfs_znode_fini();
2370 zfsctl_fini();
2371 }
2372
2373 #if defined(_KERNEL)
2374 EXPORT_SYMBOL(zfs_suspend_fs);
2375 EXPORT_SYMBOL(zfs_resume_fs);
2376 EXPORT_SYMBOL(zfs_userspace_one);
2377 EXPORT_SYMBOL(zfs_userspace_many);
2378 EXPORT_SYMBOL(zfs_set_userquota);
2379 EXPORT_SYMBOL(zfs_id_overblockquota);
2380 EXPORT_SYMBOL(zfs_id_overobjquota);
2381 EXPORT_SYMBOL(zfs_id_overquota);
2382 EXPORT_SYMBOL(zfs_set_version);
2383 EXPORT_SYMBOL(zfsvfs_create);
2384 EXPORT_SYMBOL(zfsvfs_free);
2385 EXPORT_SYMBOL(zfs_is_readonly);
2386 EXPORT_SYMBOL(zfs_domount);
2387 EXPORT_SYMBOL(zfs_preumount);
2388 EXPORT_SYMBOL(zfs_umount);
2389 EXPORT_SYMBOL(zfs_remount);
2390 EXPORT_SYMBOL(zfs_statvfs);
2391 EXPORT_SYMBOL(zfs_vget);
2392 EXPORT_SYMBOL(zfs_prune);
2393 #endif
ZFS On Linux mirror for PVE