]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - zfs/module/zfs/dmu_objset.c
projects / mirror_ubuntu-bionic-kernel.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
UBUNTU: SAUCE: (noup) Update zfs to 0.7.5-1ubuntu16.6
[mirror_ubuntu-bionic-kernel.git] / zfs / module / zfs / dmu_objset.c
CommitLineData
70e083d2
TG		 1/*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
86e3c28a 21
70e083d2
TG		 22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
86e3c28a 24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
70e083d2
TG		 25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
70e083d2
TG		 28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
86e3c28a 30 * Copyright 2017 Nexenta Systems, Inc.
70e083d2
TG		 31 */
32
33/* Portions Copyright 2010 Robert Milkowski */
34
86e3c28a 35#include <sys/zfeature.h>
70e083d2
TG		 36#include <sys/cred.h>
37#include <sys/zfs_context.h>
38#include <sys/dmu_objset.h>
39#include <sys/dsl_dir.h>
40#include <sys/dsl_dataset.h>
41#include <sys/dsl_prop.h>
42#include <sys/dsl_pool.h>
43#include <sys/dsl_synctask.h>
44#include <sys/dsl_deleg.h>
45#include <sys/dnode.h>
46#include <sys/dbuf.h>
47#include <sys/zvol.h>
48#include <sys/dmu_tx.h>
49#include <sys/zap.h>
50#include <sys/zil.h>
51#include <sys/dmu_impl.h>
52#include <sys/zfs_ioctl.h>
53#include <sys/sa.h>
54#include <sys/zfs_onexit.h>
55#include <sys/dsl_destroy.h>
56#include <sys/vdev.h>
86e3c28a
CIK		 57#include <sys/policy.h>
58#include <sys/spa_impl.h>
70e083d2
TG		 59
60/*
61 * Needed to close a window in dnode_move() that allows the objset to be freed
62 * before it can be safely accessed.
63 */
64krwlock_t os_lock;
65
66/*
86e3c28a 67 * Tunable to overwrite the maximum number of threads for the parallelization
70e083d2
TG		 68 * of dmu_objset_find_dp, needed to speed up the import of pools with many
69 * datasets.
70 * Default is 4 times the number of leaf vdevs.
71 */
72int dmu_find_threads = 0;
73
86e3c28a
CIK		 74/*
75 * Backfill lower metadnode objects after this many have been freed.
76 * Backfilling negatively impacts object creation rates, so only do it
77 * if there are enough holes to fill.
78 */
79int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT;
80
70e083d2
TG		 81static void dmu_objset_find_dp_cb(void *arg);
82
86e3c28a
CIK		 83static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb);
84static void dmu_objset_upgrade_stop(objset_t *os);
85
70e083d2
TG		 86void
87dmu_objset_init(void)
88{
89 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
90}
91
92void
93dmu_objset_fini(void)
94{
95 rw_destroy(&os_lock);
96}
97
98spa_t *
99dmu_objset_spa(objset_t *os)
100{
101 return (os->os_spa);
102}
103
104zilog_t *
105dmu_objset_zil(objset_t *os)
106{
107 return (os->os_zil);
108}
109
110dsl_pool_t *
111dmu_objset_pool(objset_t *os)
112{
113 dsl_dataset_t *ds;
114
115 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
116 return (ds->ds_dir->dd_pool);
117 else
118 return (spa_get_dsl(os->os_spa));
119}
120
121dsl_dataset_t *
122dmu_objset_ds(objset_t *os)
123{
124 return (os->os_dsl_dataset);
125}
126
127dmu_objset_type_t
128dmu_objset_type(objset_t *os)
129{
130 return (os->os_phys->os_type);
131}
132
133void
134dmu_objset_name(objset_t *os, char *buf)
135{
136 dsl_dataset_name(os->os_dsl_dataset, buf);
137}
138
139uint64_t
140dmu_objset_id(objset_t *os)
141{
142 dsl_dataset_t *ds = os->os_dsl_dataset;
143
144 return (ds ? ds->ds_object : 0);
145}
146
86e3c28a
CIK		 147uint64_t
148dmu_objset_dnodesize(objset_t *os)
149{
150 return (os->os_dnodesize);
151}
152
70e083d2
TG		 153zfs_sync_type_t
154dmu_objset_syncprop(objset_t *os)
155{
156 return (os->os_sync);
157}
158
159zfs_logbias_op_t
160dmu_objset_logbias(objset_t *os)
161{
162 return (os->os_logbias);
163}
164
165static void
166checksum_changed_cb(void *arg, uint64_t newval)
167{
168 objset_t *os = arg;
169
170 /*
171 * Inheritance should have been done by now.
172 */
173 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
174
175 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
176}
177
178static void
179compression_changed_cb(void *arg, uint64_t newval)
180{
181 objset_t *os = arg;
182
183 /*
184 * Inheritance and range checking should have been done by now.
185 */
186 ASSERT(newval != ZIO_COMPRESS_INHERIT);
187
188 os->os_compress = zio_compress_select(os->os_spa, newval,
189 ZIO_COMPRESS_ON);
190}
191
192static void
193copies_changed_cb(void *arg, uint64_t newval)
194{
195 objset_t *os = arg;
196
197 /*
198 * Inheritance and range checking should have been done by now.
199 */
200 ASSERT(newval > 0);
201 ASSERT(newval <= spa_max_replication(os->os_spa));
202
203 os->os_copies = newval;
204}
205
206static void
207dedup_changed_cb(void *arg, uint64_t newval)
208{
209 objset_t *os = arg;
210 spa_t *spa = os->os_spa;
211 enum zio_checksum checksum;
212
213 /*
214 * Inheritance should have been done by now.
215 */
216 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
217
218 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
219
220 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
221 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
222}
223
224static void
225primary_cache_changed_cb(void *arg, uint64_t newval)
226{
227 objset_t *os = arg;
228
229 /*
230 * Inheritance and range checking should have been done by now.
231 */
232 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
233 newval == ZFS_CACHE_METADATA);
234
235 os->os_primary_cache = newval;
236}
237
238static void
239secondary_cache_changed_cb(void *arg, uint64_t newval)
240{
241 objset_t *os = arg;
242
243 /*
244 * Inheritance and range checking should have been done by now.
245 */
246 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
247 newval == ZFS_CACHE_METADATA);
248
249 os->os_secondary_cache = newval;
250}
251
252static void
253sync_changed_cb(void *arg, uint64_t newval)
254{
255 objset_t *os = arg;
256
257 /*
258 * Inheritance and range checking should have been done by now.
259 */
260 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
261 newval == ZFS_SYNC_DISABLED);
262
263 os->os_sync = newval;
264 if (os->os_zil)
265 zil_set_sync(os->os_zil, newval);
266}
267
268static void
269redundant_metadata_changed_cb(void *arg, uint64_t newval)
270{
271 objset_t *os = arg;
272
273 /*
274 * Inheritance and range checking should have been done by now.
275 */
276 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
277 newval == ZFS_REDUNDANT_METADATA_MOST);
278
279 os->os_redundant_metadata = newval;
280}
281
86e3c28a
CIK		 282static void
283dnodesize_changed_cb(void *arg, uint64_t newval)
284{
285 objset_t *os = arg;
286
287 switch (newval) {
288 case ZFS_DNSIZE_LEGACY:
289 os->os_dnodesize = DNODE_MIN_SIZE;
290 break;
291 case ZFS_DNSIZE_AUTO:
292 /*
293 * Choose a dnode size that will work well for most
294 * workloads if the user specified "auto". Future code
295 * improvements could dynamically select a dnode size
296 * based on observed workload patterns.
297 */
298 os->os_dnodesize = DNODE_MIN_SIZE * 2;
299 break;
300 case ZFS_DNSIZE_1K:
301 case ZFS_DNSIZE_2K:
302 case ZFS_DNSIZE_4K:
303 case ZFS_DNSIZE_8K:
304 case ZFS_DNSIZE_16K:
305 os->os_dnodesize = newval;
306 break;
307 }
308}
309
70e083d2
TG		 310static void
311logbias_changed_cb(void *arg, uint64_t newval)
312{
313 objset_t *os = arg;
314
315 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
316 newval == ZFS_LOGBIAS_THROUGHPUT);
317 os->os_logbias = newval;
318 if (os->os_zil)
319 zil_set_logbias(os->os_zil, newval);
320}
321
322static void
323recordsize_changed_cb(void *arg, uint64_t newval)
324{
325 objset_t *os = arg;
326
327 os->os_recordsize = newval;
328}
329
330void
331dmu_objset_byteswap(void *buf, size_t size)
332{
333 objset_phys_t *osp = buf;
334
335 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
336 dnode_byteswap(&osp->os_meta_dnode);
337 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
338 osp->os_type = BSWAP_64(osp->os_type);
339 osp->os_flags = BSWAP_64(osp->os_flags);
340 if (size == sizeof (objset_phys_t)) {
341 dnode_byteswap(&osp->os_userused_dnode);
342 dnode_byteswap(&osp->os_groupused_dnode);
343 }
344}
345
86e3c28a
CIK		 346/*
347 * The hash is a CRC-based hash of the objset_t pointer and the object number.
348 */
349static uint64_t
350dnode_hash(const objset_t *os, uint64_t obj)
351{
352 uintptr_t osv = (uintptr_t)os;
353 uint64_t crc = -1ULL;
354
355 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
356 /*
357 * The low 6 bits of the pointer don't have much entropy, because
358 * the objset_t is larger than 2^6 bytes long.
359 */
360 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
361 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
362 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
363 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
364
365 crc ^= (osv>>14) ^ (obj>>24);
366
367 return (crc);
368}
369
370unsigned int
371dnode_multilist_index_func(multilist_t *ml, void *obj)
372{
373 dnode_t *dn = obj;
374 return (dnode_hash(dn->dn_objset, dn->dn_object) %
375 multilist_get_num_sublists(ml));
376}
377
70e083d2
TG		 378int
379dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
380 objset_t **osp)
381{
382 objset_t *os;
383 int i, err;
384
385 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
386
387 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
388 os->os_dsl_dataset = ds;
389 os->os_spa = spa;
390 os->os_rootbp = bp;
391 if (!BP_IS_HOLE(os->os_rootbp)) {
392 arc_flags_t aflags = ARC_FLAG_WAIT;
393 zbookmark_phys_t zb;
394 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
395 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
396
397 if (DMU_OS_IS_L2CACHEABLE(os))
398 aflags |= ARC_FLAG_L2CACHE;
70e083d2
TG		 399
400 dprintf_bp(os->os_rootbp, "reading %s", "");
401 err = arc_read(NULL, spa, os->os_rootbp,
402 arc_getbuf_func, &os->os_phys_buf,
403 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
404 if (err != 0) {
405 kmem_free(os, sizeof (objset_t));
406 /* convert checksum errors into IO errors */
407 if (err == ECKSUM)
408 err = SET_ERROR(EIO);
409 return (err);
410 }
411
412 /* Increase the blocksize if we are permitted. */
413 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
414 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
86e3c28a
CIK		 415 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
416 ARC_BUFC_METADATA, sizeof (objset_phys_t));
70e083d2
TG		 417 bzero(buf->b_data, sizeof (objset_phys_t));
418 bcopy(os->os_phys_buf->b_data, buf->b_data,
419 arc_buf_size(os->os_phys_buf));
86e3c28a 420 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
70e083d2
TG		 421 os->os_phys_buf = buf;
422 }
423
424 os->os_phys = os->os_phys_buf->b_data;
425 os->os_flags = os->os_phys->os_flags;
426 } else {
427 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
428 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
86e3c28a
CIK		 429 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
430 ARC_BUFC_METADATA, size);
70e083d2
TG		 431 os->os_phys = os->os_phys_buf->b_data;
432 bzero(os->os_phys, size);
433 }
434
435 /*
436 * Note: the changed_cb will be called once before the register
437 * func returns, thus changing the checksum/compression from the
438 * default (fletcher2/off). Snapshots don't need to know about
439 * checksum/compression/copies.
440 */
441 if (ds != NULL) {
86e3c28a
CIK		 442 boolean_t needlock = B_FALSE;
443
444 /*
445 * Note: it's valid to open the objset if the dataset is
446 * long-held, in which case the pool_config lock will not
447 * be held.
448 */
449 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
450 needlock = B_TRUE;
451 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
452 }
70e083d2
TG		 453 err = dsl_prop_register(ds,
454 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
455 primary_cache_changed_cb, os);
456 if (err == 0) {
457 err = dsl_prop_register(ds,
458 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
459 secondary_cache_changed_cb, os);
460 }
461 if (!ds->ds_is_snapshot) {
462 if (err == 0) {
463 err = dsl_prop_register(ds,
464 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
465 checksum_changed_cb, os);
466 }
467 if (err == 0) {
468 err = dsl_prop_register(ds,
469 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
470 compression_changed_cb, os);
471 }
472 if (err == 0) {
473 err = dsl_prop_register(ds,
474 zfs_prop_to_name(ZFS_PROP_COPIES),
475 copies_changed_cb, os);
476 }
477 if (err == 0) {
478 err = dsl_prop_register(ds,
479 zfs_prop_to_name(ZFS_PROP_DEDUP),
480 dedup_changed_cb, os);
481 }
482 if (err == 0) {
483 err = dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
485 logbias_changed_cb, os);
486 }
487 if (err == 0) {
488 err = dsl_prop_register(ds,
489 zfs_prop_to_name(ZFS_PROP_SYNC),
490 sync_changed_cb, os);
491 }
492 if (err == 0) {
493 err = dsl_prop_register(ds,
494 zfs_prop_to_name(
495 ZFS_PROP_REDUNDANT_METADATA),
496 redundant_metadata_changed_cb, os);
497 }
498 if (err == 0) {
499 err = dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
501 recordsize_changed_cb, os);
502 }
86e3c28a
CIK		 503 if (err == 0) {
504 err = dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_DNODESIZE),
506 dnodesize_changed_cb, os);
507 }
70e083d2 508 }
86e3c28a
CIK		 509 if (needlock)
510 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
70e083d2 511 if (err != 0) {
86e3c28a 512 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
70e083d2
TG		 513 kmem_free(os, sizeof (objset_t));
514 return (err);
515 }
516 } else {
517 /* It's the meta-objset. */
518 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
519 os->os_compress = ZIO_COMPRESS_ON;
520 os->os_copies = spa_max_replication(spa);
521 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
522 os->os_dedup_verify = B_FALSE;
523 os->os_logbias = ZFS_LOGBIAS_LATENCY;
524 os->os_sync = ZFS_SYNC_STANDARD;
525 os->os_primary_cache = ZFS_CACHE_ALL;
526 os->os_secondary_cache = ZFS_CACHE_ALL;
86e3c28a 527 os->os_dnodesize = DNODE_MIN_SIZE;
70e083d2
TG		 528 }
529
530 if (ds == NULL || !ds->ds_is_snapshot)
531 os->os_zil_header = os->os_phys->os_zil_header;
532 os->os_zil = zil_alloc(os, &os->os_zil_header);
533
534 for (i = 0; i < TXG_SIZE; i++) {
86e3c28a
CIK		 535 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
536 offsetof(dnode_t, dn_dirty_link[i]),
537 dnode_multilist_index_func);
70e083d2
TG		 538 }
539 list_create(&os->os_dnodes, sizeof (dnode_t),
540 offsetof(dnode_t, dn_link));
541 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
542 offsetof(dmu_buf_impl_t, db_link));
543
544 list_link_init(&os->os_evicting_node);
545
546 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
86e3c28a 547 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
70e083d2
TG		 548 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
549 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
86e3c28a
CIK		 550 os->os_obj_next_percpu_len = boot_ncpus;
551 os->os_obj_next_percpu = kmem_zalloc(os->os_obj_next_percpu_len *
552 sizeof (os->os_obj_next_percpu[0]), KM_SLEEP);
70e083d2
TG		 553
554 dnode_special_open(os, &os->os_phys->os_meta_dnode,
555 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
556 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
557 dnode_special_open(os, &os->os_phys->os_userused_dnode,
558 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
559 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
560 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
561 }
562
86e3c28a
CIK		 563 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL);
564
70e083d2
TG		 565 *osp = os;
566 return (0);
567}
568
569int
570dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
571{
572 int err = 0;
573
86e3c28a
CIK		 574 /*
575 * We shouldn't be doing anything with dsl_dataset_t's unless the
576 * pool_config lock is held, or the dataset is long-held.
577 */
578 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
579 dsl_dataset_long_held(ds));
580
70e083d2
TG		 581 mutex_enter(&ds->ds_opening_lock);
582 if (ds->ds_objset == NULL) {
583 objset_t *os;
86e3c28a 584 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
70e083d2
TG		 585 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
586 ds, dsl_dataset_get_blkptr(ds), &os);
86e3c28a 587 rrw_exit(&ds->ds_bp_rwlock, FTAG);
70e083d2
TG		 588
589 if (err == 0) {
590 mutex_enter(&ds->ds_lock);
591 ASSERT(ds->ds_objset == NULL);
592 ds->ds_objset = os;
593 mutex_exit(&ds->ds_lock);
594 }
595 }
596 *osp = ds->ds_objset;
597 mutex_exit(&ds->ds_opening_lock);
598 return (err);
599}
600
601/*
602 * Holds the pool while the objset is held. Therefore only one objset
603 * can be held at a time.
604 */
605int
606dmu_objset_hold(const char *name, void *tag, objset_t **osp)
607{
608 dsl_pool_t *dp;
609 dsl_dataset_t *ds;
610 int err;
611
612 err = dsl_pool_hold(name, tag, &dp);
613 if (err != 0)
614 return (err);
615 err = dsl_dataset_hold(dp, name, tag, &ds);
616 if (err != 0) {
617 dsl_pool_rele(dp, tag);
618 return (err);
619 }
620
621 err = dmu_objset_from_ds(ds, osp);
622 if (err != 0) {
623 dsl_dataset_rele(ds, tag);
624 dsl_pool_rele(dp, tag);
625 }
626
627 return (err);
628}
629
630static int
631dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
632 boolean_t readonly, void *tag, objset_t **osp)
633{
634 int err;
635
636 err = dmu_objset_from_ds(ds, osp);
637 if (err != 0) {
638 dsl_dataset_disown(ds, tag);
639 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
640 dsl_dataset_disown(ds, tag);
641 return (SET_ERROR(EINVAL));
642 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
643 dsl_dataset_disown(ds, tag);
644 return (SET_ERROR(EROFS));
645 }
646 return (err);
647}
648
649/*
650 * dsl_pool must not be held when this is called.
651 * Upon successful return, there will be a longhold on the dataset,
652 * and the dsl_pool will not be held.
653 */
654int
655dmu_objset_own(const char *name, dmu_objset_type_t type,
656 boolean_t readonly, void *tag, objset_t **osp)
657{
658 dsl_pool_t *dp;
659 dsl_dataset_t *ds;
660 int err;
661
662 err = dsl_pool_hold(name, FTAG, &dp);
663 if (err != 0)
664 return (err);
665 err = dsl_dataset_own(dp, name, tag, &ds);
666 if (err != 0) {
667 dsl_pool_rele(dp, FTAG);
668 return (err);
669 }
670 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
671 dsl_pool_rele(dp, FTAG);
672
86e3c28a
CIK		 673 if (err == 0 && dmu_objset_userobjspace_upgradable(*osp))
674 dmu_objset_userobjspace_upgrade(*osp);
675
70e083d2
TG		 676 return (err);
677}
678
679int
680dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
681 boolean_t readonly, void *tag, objset_t **osp)
682{
683 dsl_dataset_t *ds;
684 int err;
685
686 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
687 if (err != 0)
688 return (err);
689
690 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
691}
692
693void
694dmu_objset_rele(objset_t *os, void *tag)
695{
696 dsl_pool_t *dp = dmu_objset_pool(os);
697 dsl_dataset_rele(os->os_dsl_dataset, tag);
698 dsl_pool_rele(dp, tag);
699}
700
701/*
702 * When we are called, os MUST refer to an objset associated with a dataset
703 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
704 * == tag. We will then release and reacquire ownership of the dataset while
705 * holding the pool config_rwlock to avoid intervening namespace or ownership
706 * changes may occur.
707 *
708 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
709 * release the hold on its dataset and acquire a new one on the dataset of the
710 * same name so that it can be partially torn down and reconstructed.
711 */
712void
713dmu_objset_refresh_ownership(objset_t *os, void *tag)
714{
715 dsl_pool_t *dp;
716 dsl_dataset_t *ds, *newds;
86e3c28a 717 char name[ZFS_MAX_DATASET_NAME_LEN];
70e083d2
TG		 718
719 ds = os->os_dsl_dataset;
720 VERIFY3P(ds, !=, NULL);
721 VERIFY3P(ds->ds_owner, ==, tag);
722 VERIFY(dsl_dataset_long_held(ds));
723
724 dsl_dataset_name(ds, name);
725 dp = dmu_objset_pool(os);
726 dsl_pool_config_enter(dp, FTAG);
727 dmu_objset_disown(os, tag);
728 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
729 VERIFY3P(newds, ==, os->os_dsl_dataset);
730 dsl_pool_config_exit(dp, FTAG);
731}
732
733void
734dmu_objset_disown(objset_t *os, void *tag)
735{
86e3c28a
CIK		 736 /*
737 * Stop upgrading thread
738 */
739 dmu_objset_upgrade_stop(os);
70e083d2
TG		 740 dsl_dataset_disown(os->os_dsl_dataset, tag);
741}
742
743void
744dmu_objset_evict_dbufs(objset_t *os)
745{
746 dnode_t *dn_marker;
747 dnode_t *dn;
748
749 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
750
751 mutex_enter(&os->os_lock);
752 dn = list_head(&os->os_dnodes);
753 while (dn != NULL) {
754 /*
755 * Skip dnodes without holds. We have to do this dance
756 * because dnode_add_ref() only works if there is already a
757 * hold. If the dnode has no holds, then it has no dbufs.
758 */
759 if (dnode_add_ref(dn, FTAG)) {
760 list_insert_after(&os->os_dnodes, dn, dn_marker);
761 mutex_exit(&os->os_lock);
762
763 dnode_evict_dbufs(dn);
764 dnode_rele(dn, FTAG);
765
766 mutex_enter(&os->os_lock);
767 dn = list_next(&os->os_dnodes, dn_marker);
768 list_remove(&os->os_dnodes, dn_marker);
769 } else {
770 dn = list_next(&os->os_dnodes, dn);
771 }
772 }
773 mutex_exit(&os->os_lock);
774
775 kmem_free(dn_marker, sizeof (dnode_t));
776
777 if (DMU_USERUSED_DNODE(os) != NULL) {
778 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
779 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
780 }
781 dnode_evict_dbufs(DMU_META_DNODE(os));
782}
783
784/*
785 * Objset eviction processing is split into into two pieces.
786 * The first marks the objset as evicting, evicts any dbufs that
787 * have a refcount of zero, and then queues up the objset for the
788 * second phase of eviction. Once os->os_dnodes has been cleared by
789 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
790 * The second phase closes the special dnodes, dequeues the objset from
791 * the list of those undergoing eviction, and finally frees the objset.
792 *
793 * NOTE: Due to asynchronous eviction processing (invocation of
794 * dnode_buf_pageout()), it is possible for the meta dnode for the
795 * objset to have no holds even though os->os_dnodes is not empty.
796 */
797void
798dmu_objset_evict(objset_t *os)
799{
800 int t;
801
802 dsl_dataset_t *ds = os->os_dsl_dataset;
803
804 for (t = 0; t < TXG_SIZE; t++)
805 ASSERT(!dmu_objset_is_dirty(os, t));
806
86e3c28a
CIK		 807 if (ds)
808 dsl_prop_unregister_all(ds, os);
70e083d2
TG		 809
810 if (os->os_sa)
811 sa_tear_down(os);
812
813 dmu_objset_evict_dbufs(os);
814
815 mutex_enter(&os->os_lock);
816 spa_evicting_os_register(os->os_spa, os);
817 if (list_is_empty(&os->os_dnodes)) {
818 mutex_exit(&os->os_lock);
819 dmu_objset_evict_done(os);
820 } else {
821 mutex_exit(&os->os_lock);
822 }
823}
824
825void
826dmu_objset_evict_done(objset_t *os)
827{
828 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
829
830 dnode_special_close(&os->os_meta_dnode);
831 if (DMU_USERUSED_DNODE(os)) {
832 dnode_special_close(&os->os_userused_dnode);
833 dnode_special_close(&os->os_groupused_dnode);
834 }
835 zil_free(os->os_zil);
836
86e3c28a 837 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
70e083d2
TG		 838
839 /*
840 * This is a barrier to prevent the objset from going away in
841 * dnode_move() until we can safely ensure that the objset is still in
842 * use. We consider the objset valid before the barrier and invalid
843 * after the barrier.
844 */
845 rw_enter(&os_lock, RW_READER);
846 rw_exit(&os_lock);
847
86e3c28a
CIK		 848 kmem_free(os->os_obj_next_percpu,
849 os->os_obj_next_percpu_len * sizeof (os->os_obj_next_percpu[0]));
850
70e083d2 851 mutex_destroy(&os->os_lock);
86e3c28a 852 mutex_destroy(&os->os_userused_lock);
70e083d2
TG		 853 mutex_destroy(&os->os_obj_lock);
854 mutex_destroy(&os->os_user_ptr_lock);
86e3c28a
CIK		 855 mutex_destroy(&os->os_upgrade_lock);
856 for (int i = 0; i < TXG_SIZE; i++) {
857 multilist_destroy(os->os_dirty_dnodes[i]);
858 }
70e083d2
TG		 859 spa_evicting_os_deregister(os->os_spa, os);
860 kmem_free(os, sizeof (objset_t));
861}
862
863timestruc_t
864dmu_objset_snap_cmtime(objset_t *os)
865{
866 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
867}
868
869/* called from dsl for meta-objset */
870objset_t *
871dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
872 dmu_objset_type_t type, dmu_tx_t *tx)
873{
874 objset_t *os;
875 dnode_t *mdn;
876
877 ASSERT(dmu_tx_is_syncing(tx));
878
879 if (ds != NULL)
880 VERIFY0(dmu_objset_from_ds(ds, &os));
881 else
882 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
883
884 mdn = DMU_META_DNODE(os);
885
86e3c28a
CIK		 886 dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT,
887 DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx);
70e083d2
TG		 888
889 /*
890 * We don't want to have to increase the meta-dnode's nlevels
891 * later, because then we could do it in quescing context while
892 * we are also accessing it in open context.
893 *
894 * This precaution is not necessary for the MOS (ds == NULL),
895 * because the MOS is only updated in syncing context.
896 * This is most fortunate: the MOS is the only objset that
897 * needs to be synced multiple times as spa_sync() iterates
898 * to convergence, so minimizing its dn_nlevels matters.
899 */
900 if (ds != NULL) {
901 int levels = 1;
902
903 /*
904 * Determine the number of levels necessary for the meta-dnode
86e3c28a
CIK		 905 * to contain DN_MAX_OBJECT dnodes. Note that in order to
906 * ensure that we do not overflow 64 bits, there has to be
907 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
908 * but < 2^64. Therefore,
909 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
910 * less than (64 - log2(DN_MAX_OBJECT)) (16).
70e083d2 911 */
86e3c28a
CIK		 912 while ((uint64_t)mdn->dn_nblkptr <<
913 (mdn->dn_datablkshift - DNODE_SHIFT +
70e083d2 914 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
86e3c28a 915 DN_MAX_OBJECT)
70e083d2
TG		 916 levels++;
917
918 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
919 mdn->dn_nlevels = levels;
920 }
921
922 ASSERT(type != DMU_OST_NONE);
923 ASSERT(type != DMU_OST_ANY);
924 ASSERT(type < DMU_OST_NUMTYPES);
925 os->os_phys->os_type = type;
926 if (dmu_objset_userused_enabled(os)) {
927 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
86e3c28a
CIK		 928 if (dmu_objset_userobjused_enabled(os)) {
929 ds->ds_feature_activation_needed[
930 SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE;
931 os->os_phys->os_flags |=
932 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
933 }
70e083d2
TG		 934 os->os_flags = os->os_phys->os_flags;
935 }
936
937 dsl_dataset_dirty(ds, tx);
938
939 return (os);
940}
941
942typedef struct dmu_objset_create_arg {
943 const char *doca_name;
944 cred_t *doca_cred;
945 void (*doca_userfunc)(objset_t *os, void *arg,
946 cred_t *cr, dmu_tx_t *tx);
947 void *doca_userarg;
948 dmu_objset_type_t doca_type;
949 uint64_t doca_flags;
950} dmu_objset_create_arg_t;
951
952/*ARGSUSED*/
953static int
954dmu_objset_create_check(void *arg, dmu_tx_t *tx)
955{
956 dmu_objset_create_arg_t *doca = arg;
957 dsl_pool_t *dp = dmu_tx_pool(tx);
958 dsl_dir_t *pdd;
959 const char *tail;
960 int error;
961
962 if (strchr(doca->doca_name, '@') != NULL)
963 return (SET_ERROR(EINVAL));
964
86e3c28a
CIK		 965 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
966 return (SET_ERROR(ENAMETOOLONG));
967
70e083d2
TG		 968 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
969 if (error != 0)
970 return (error);
971 if (tail == NULL) {
972 dsl_dir_rele(pdd, FTAG);
973 return (SET_ERROR(EEXIST));
974 }
975 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
976 doca->doca_cred);
977 dsl_dir_rele(pdd, FTAG);
978
979 return (error);
980}
981
982static void
983dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
984{
985 dmu_objset_create_arg_t *doca = arg;
986 dsl_pool_t *dp = dmu_tx_pool(tx);
987 dsl_dir_t *pdd;
988 const char *tail;
989 dsl_dataset_t *ds;
990 uint64_t obj;
991 blkptr_t *bp;
992 objset_t *os;
993
994 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
995
996 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
997 doca->doca_cred, tx);
998
999 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
86e3c28a 1000 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
70e083d2
TG		 1001 bp = dsl_dataset_get_blkptr(ds);
1002 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
1003 ds, bp, doca->doca_type, tx);
86e3c28a 1004 rrw_exit(&ds->ds_bp_rwlock, FTAG);
70e083d2
TG		 1005
1006 if (doca->doca_userfunc != NULL) {
1007 doca->doca_userfunc(os, doca->doca_userarg,
1008 doca->doca_cred, tx);
1009 }
1010
1011 spa_history_log_internal_ds(ds, "create", tx, "");
1012 zvol_create_minors(dp->dp_spa, doca->doca_name, B_TRUE);
1013
1014 dsl_dataset_rele(ds, FTAG);
1015 dsl_dir_rele(pdd, FTAG);
1016}
1017
1018int
1019dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1020 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
1021{
1022 dmu_objset_create_arg_t doca;
1023
1024 doca.doca_name = name;
1025 doca.doca_cred = CRED();
1026 doca.doca_flags = flags;
1027 doca.doca_userfunc = func;
1028 doca.doca_userarg = arg;
1029 doca.doca_type = type;
1030
1031 return (dsl_sync_task(name,
1032 dmu_objset_create_check, dmu_objset_create_sync, &doca,
1033 5, ZFS_SPACE_CHECK_NORMAL));
1034}
1035
1036typedef struct dmu_objset_clone_arg {
1037 const char *doca_clone;
1038 const char *doca_origin;
1039 cred_t *doca_cred;
1040} dmu_objset_clone_arg_t;
1041
1042/*ARGSUSED*/
1043static int
1044dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1045{
1046 dmu_objset_clone_arg_t *doca = arg;
1047 dsl_dir_t *pdd;
1048 const char *tail;
1049 int error;
1050 dsl_dataset_t *origin;
1051 dsl_pool_t *dp = dmu_tx_pool(tx);
1052
1053 if (strchr(doca->doca_clone, '@') != NULL)
1054 return (SET_ERROR(EINVAL));
1055
86e3c28a
CIK		 1056 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1057 return (SET_ERROR(ENAMETOOLONG));
1058
70e083d2
TG		 1059 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1060 if (error != 0)
1061 return (error);
1062 if (tail == NULL) {
1063 dsl_dir_rele(pdd, FTAG);
1064 return (SET_ERROR(EEXIST));
1065 }
1066
1067 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1068 doca->doca_cred);
1069 if (error != 0) {
1070 dsl_dir_rele(pdd, FTAG);
1071 return (SET_ERROR(EDQUOT));
1072 }
1073 dsl_dir_rele(pdd, FTAG);
1074
1075 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1076 if (error != 0)
1077 return (error);
1078
1079 /* You can only clone snapshots, not the head datasets. */
1080 if (!origin->ds_is_snapshot) {
1081 dsl_dataset_rele(origin, FTAG);
1082 return (SET_ERROR(EINVAL));
1083 }
1084 dsl_dataset_rele(origin, FTAG);
1085
1086 return (0);
1087}
1088
1089static void
1090dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1091{
1092 dmu_objset_clone_arg_t *doca = arg;
1093 dsl_pool_t *dp = dmu_tx_pool(tx);
1094 dsl_dir_t *pdd;
1095 const char *tail;
1096 dsl_dataset_t *origin, *ds;
1097 uint64_t obj;
86e3c28a 1098 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
70e083d2
TG		 1099
1100 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1101 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1102
1103 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1104 doca->doca_cred, tx);
1105
1106 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1107 dsl_dataset_name(origin, namebuf);
1108 spa_history_log_internal_ds(ds, "clone", tx,
1109 "origin=%s (%llu)", namebuf, origin->ds_object);
1110 zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE);
1111 dsl_dataset_rele(ds, FTAG);
1112 dsl_dataset_rele(origin, FTAG);
1113 dsl_dir_rele(pdd, FTAG);
1114}
1115
1116int
1117dmu_objset_clone(const char *clone, const char *origin)
1118{
1119 dmu_objset_clone_arg_t doca;
1120
1121 doca.doca_clone = clone;
1122 doca.doca_origin = origin;
1123 doca.doca_cred = CRED();
1124
1125 return (dsl_sync_task(clone,
1126 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1127 5, ZFS_SPACE_CHECK_NORMAL));
1128}
1129
1130int
1131dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1132{
1133 int err;
1134 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1135 nvlist_t *snaps = fnvlist_alloc();
1136
1137 fnvlist_add_boolean(snaps, longsnap);
1138 strfree(longsnap);
1139 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1140 fnvlist_free(snaps);
1141 return (err);
1142}
1143
1144static void
86e3c28a
CIK		 1145dmu_objset_upgrade_task_cb(void *data)
1146{
1147 objset_t *os = data;
1148
1149 mutex_enter(&os->os_upgrade_lock);
1150 os->os_upgrade_status = EINTR;
1151 if (!os->os_upgrade_exit) {
1152 mutex_exit(&os->os_upgrade_lock);
1153
1154 os->os_upgrade_status = os->os_upgrade_cb(os);
1155 mutex_enter(&os->os_upgrade_lock);
1156 }
1157 os->os_upgrade_exit = B_TRUE;
1158 os->os_upgrade_id = 0;
1159 mutex_exit(&os->os_upgrade_lock);
1160}
1161
1162static void
1163dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb)
1164{
1165 if (os->os_upgrade_id != 0)
1166 return;
1167
1168 mutex_enter(&os->os_upgrade_lock);
1169 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) {
1170 os->os_upgrade_exit = B_FALSE;
1171 os->os_upgrade_cb = cb;
1172 os->os_upgrade_id = taskq_dispatch(
1173 os->os_spa->spa_upgrade_taskq,
1174 dmu_objset_upgrade_task_cb, os, TQ_SLEEP);
1175 if (os->os_upgrade_id == TASKQID_INVALID)
1176 os->os_upgrade_status = ENOMEM;
1177 }
1178 mutex_exit(&os->os_upgrade_lock);
1179}
1180
1181static void
1182dmu_objset_upgrade_stop(objset_t *os)
1183{
1184 mutex_enter(&os->os_upgrade_lock);
1185 os->os_upgrade_exit = B_TRUE;
1186 if (os->os_upgrade_id != 0) {
1187 taskqid_t id = os->os_upgrade_id;
1188
1189 os->os_upgrade_id = 0;
1190 mutex_exit(&os->os_upgrade_lock);
1191
1192 taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id);
1193 } else {
1194 mutex_exit(&os->os_upgrade_lock);
1195 }
1196}
1197
1198static void
1199dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
70e083d2
TG		 1200{
1201 dnode_t *dn;
1202
86e3c28a 1203 while ((dn = multilist_sublist_head(list)) != NULL) {
70e083d2
TG		 1204 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1205 ASSERT(dn->dn_dbuf->db_data_pending);
1206 /*
1207 * Initialize dn_zio outside dnode_sync() because the
1208 * meta-dnode needs to set it ouside dnode_sync().
1209 */
1210 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1211 ASSERT(dn->dn_zio);
1212
1213 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
86e3c28a 1214 multilist_sublist_remove(list, dn);
70e083d2 1215
86e3c28a
CIK		 1216 multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1217 if (newlist != NULL) {
70e083d2 1218 (void) dnode_add_ref(dn, newlist);
86e3c28a 1219 multilist_insert(newlist, dn);
70e083d2
TG		 1220 }
1221
1222 dnode_sync(dn, tx);
1223 }
1224}
1225
1226/* ARGSUSED */
1227static void
1228dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1229{
1230 int i;
1231
1232 blkptr_t *bp = zio->io_bp;
1233 objset_t *os = arg;
1234 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1235
1236 ASSERT(!BP_IS_EMBEDDED(bp));
70e083d2
TG		 1237 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1238 ASSERT0(BP_GET_LEVEL(bp));
1239
1240 /*
1241 * Update rootbp fill count: it should be the number of objects
1242 * allocated in the object set (not counting the "special"
1243 * objects that are stored in the objset_phys_t -- the meta
1244 * dnode and user/group accounting objects).
1245 */
1246 bp->blk_fill = 0;
1247 for (i = 0; i < dnp->dn_nblkptr; i++)
1248 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
86e3c28a
CIK		 1249 if (os->os_dsl_dataset != NULL)
1250 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1251 *os->os_rootbp = *bp;
1252 if (os->os_dsl_dataset != NULL)
1253 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
70e083d2
TG		 1254}
1255
1256/* ARGSUSED */
1257static void
1258dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1259{
1260 blkptr_t *bp = zio->io_bp;
1261 blkptr_t *bp_orig = &zio->io_bp_orig;
1262 objset_t *os = arg;
1263
1264 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1265 ASSERT(BP_EQUAL(bp, bp_orig));
1266 } else {
1267 dsl_dataset_t *ds = os->os_dsl_dataset;
1268 dmu_tx_t *tx = os->os_synctx;
1269
1270 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1271 dsl_dataset_block_born(ds, bp, tx);
1272 }
86e3c28a 1273 kmem_free(bp, sizeof (*bp));
70e083d2
TG		 1274}
1275
86e3c28a
CIK		 1276typedef struct sync_dnodes_arg {
1277 multilist_t *sda_list;
1278 int sda_sublist_idx;
1279 multilist_t *sda_newlist;
1280 dmu_tx_t *sda_tx;
1281} sync_dnodes_arg_t;
1282
1283static void
1284sync_dnodes_task(void *arg)
1285{
1286 sync_dnodes_arg_t *sda = arg;
1287
1288 multilist_sublist_t *ms =
1289 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1290
1291 dmu_objset_sync_dnodes(ms, sda->sda_tx);
1292
1293 multilist_sublist_unlock(ms);
1294
1295 kmem_free(sda, sizeof (*sda));
1296}
1297
1298
70e083d2
TG		 1299/* called from dsl */
1300void
1301dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1302{
1303 int txgoff;
1304 zbookmark_phys_t zb;
1305 zio_prop_t zp;
1306 zio_t *zio;
1307 list_t *list;
70e083d2 1308 dbuf_dirty_record_t *dr;
86e3c28a
CIK		 1309 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1310 *blkptr_copy = *os->os_rootbp;
70e083d2
TG		 1311
1312 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1313
1314 ASSERT(dmu_tx_is_syncing(tx));
1315 /* XXX the write_done callback should really give us the tx... */
1316 os->os_synctx = tx;
1317
1318 if (os->os_dsl_dataset == NULL) {
1319 /*
1320 * This is the MOS. If we have upgraded,
1321 * spa_max_replication() could change, so reset
1322 * os_copies here.
1323 */
1324 os->os_copies = spa_max_replication(os->os_spa);
1325 }
1326
1327 /*
1328 * Create the root block IO
1329 */
1330 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1331 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1332 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1333 arc_release(os->os_phys_buf, &os->os_phys_buf);
1334
1335 dmu_write_policy(os, NULL, 0, 0, &zp);
1336
1337 zio = arc_write(pio, os->os_spa, tx->tx_txg,
86e3c28a
CIK		 1338 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1339 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1340 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
70e083d2
TG		 1341
1342 /*
1343 * Sync special dnodes - the parent IO for the sync is the root block
1344 */
1345 DMU_META_DNODE(os)->dn_zio = zio;
1346 dnode_sync(DMU_META_DNODE(os), tx);
1347
1348 os->os_phys->os_flags = os->os_flags;
1349
1350 if (DMU_USERUSED_DNODE(os) &&
1351 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1352 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1353 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1354 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1355 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1356 }
1357
1358 txgoff = tx->tx_txg & TXG_MASK;
1359
1360 if (dmu_objset_userused_enabled(os)) {
70e083d2
TG		 1361 /*
1362 * We must create the list here because it uses the
86e3c28a
CIK		 1363 * dn_dirty_link[] of this txg. But it may already
1364 * exist because we call dsl_dataset_sync() twice per txg.
70e083d2 1365 */
86e3c28a
CIK		 1366 if (os->os_synced_dnodes == NULL) {
1367 os->os_synced_dnodes =
1368 multilist_create(sizeof (dnode_t),
1369 offsetof(dnode_t, dn_dirty_link[txgoff]),
1370 dnode_multilist_index_func);
1371 } else {
1372 ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1373 offsetof(dnode_t, dn_dirty_link[txgoff]));
1374 }
70e083d2
TG		 1375 }
1376
86e3c28a
CIK		 1377 for (int i = 0;
1378 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) {
1379 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1380 sda->sda_list = os->os_dirty_dnodes[txgoff];
1381 sda->sda_sublist_idx = i;
1382 sda->sda_tx = tx;
1383 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1384 sync_dnodes_task, sda, 0);
1385 /* callback frees sda */
1386 }
1387 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
70e083d2
TG		 1388
1389 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
86e3c28a 1390 while ((dr = list_head(list)) != NULL) {
70e083d2
TG		 1391 ASSERT0(dr->dr_dbuf->db_level);
1392 list_remove(list, dr);
1393 if (dr->dr_zio)
1394 zio_nowait(dr->dr_zio);
1395 }
86e3c28a
CIK		 1396
1397 /* Enable dnode backfill if enough objects have been freed. */
1398 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1399 os->os_rescan_dnodes = B_TRUE;
1400 os->os_freed_dnodes = 0;
1401 }
1402
70e083d2
TG		 1403 /*
1404 * Free intent log blocks up to this tx.
1405 */
1406 zil_sync(os->os_zil, tx);
1407 os->os_phys->os_zil_header = os->os_zil_header;
1408 zio_nowait(zio);
1409}
1410
1411boolean_t
1412dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1413{
86e3c28a 1414 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
70e083d2
TG		 1415}
1416
1417static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1418
1419void
1420dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1421{
1422 used_cbs[ost] = cb;
1423}
1424
1425boolean_t
1426dmu_objset_userused_enabled(objset_t *os)
1427{
1428 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1429 used_cbs[os->os_phys->os_type] != NULL &&
1430 DMU_USERUSED_DNODE(os) != NULL);
1431}
1432
86e3c28a
CIK		 1433boolean_t
1434dmu_objset_userobjused_enabled(objset_t *os)
1435{
1436 return (dmu_objset_userused_enabled(os) &&
1437 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING));
1438}
1439
1440typedef struct userquota_node {
1441 /* must be in the first filed, see userquota_update_cache() */
1442 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN];
1443 int64_t uqn_delta;
1444 avl_node_t uqn_node;
1445} userquota_node_t;
1446
1447typedef struct userquota_cache {
1448 avl_tree_t uqc_user_deltas;
1449 avl_tree_t uqc_group_deltas;
1450} userquota_cache_t;
1451
1452static int
1453userquota_compare(const void *l, const void *r)
1454{
1455 const userquota_node_t *luqn = l;
1456 const userquota_node_t *ruqn = r;
1457 int rv;
1458
1459 /*
1460 * NB: can only access uqn_id because userquota_update_cache() doesn't
1461 * pass in an entire userquota_node_t.
1462 */
1463 rv = strcmp(luqn->uqn_id, ruqn->uqn_id);
1464
1465 return (AVL_ISIGN(rv));
1466}
1467
70e083d2 1468static void
86e3c28a
CIK		 1469do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1470{
1471 void *cookie;
1472 userquota_node_t *uqn;
1473
1474 ASSERT(dmu_tx_is_syncing(tx));
1475
1476 cookie = NULL;
1477 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1478 &cookie)) != NULL) {
1479 /*
1480 * os_userused_lock protects against concurrent calls to
1481 * zap_increment_int(). It's needed because zap_increment_int()
1482 * is not thread-safe (i.e. not atomic).
1483 */
1484 mutex_enter(&os->os_userused_lock);
1485 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT,
1486 uqn->uqn_id, uqn->uqn_delta, tx));
1487 mutex_exit(&os->os_userused_lock);
1488 kmem_free(uqn, sizeof (*uqn));
1489 }
1490 avl_destroy(&cache->uqc_user_deltas);
1491
1492 cookie = NULL;
1493 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1494 &cookie)) != NULL) {
1495 mutex_enter(&os->os_userused_lock);
1496 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT,
1497 uqn->uqn_id, uqn->uqn_delta, tx));
1498 mutex_exit(&os->os_userused_lock);
1499 kmem_free(uqn, sizeof (*uqn));
1500 }
1501 avl_destroy(&cache->uqc_group_deltas);
1502}
1503
1504static void
1505userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta)
1506{
1507 userquota_node_t *uqn;
1508 avl_index_t idx;
1509
1510 ASSERT(strlen(id) < sizeof (uqn->uqn_id));
1511 /*
1512 * Use id directly for searching because uqn_id is the first field of
1513 * userquota_node_t and fields after uqn_id won't be accessed in
1514 * avl_find().
1515 */
1516 uqn = avl_find(avl, (const void *)id, &idx);
1517 if (uqn == NULL) {
1518 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1519 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id));
1520 avl_insert(avl, uqn, idx);
1521 }
1522 uqn->uqn_delta += delta;
1523}
1524
1525static void
1526do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1527 uint64_t user, uint64_t group, boolean_t subtract)
70e083d2
TG		 1528{
1529 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
86e3c28a
CIK		 1530 int64_t delta = DNODE_MIN_SIZE + used;
1531 char name[20];
1532
70e083d2
TG		 1533 if (subtract)
1534 delta = -delta;
86e3c28a
CIK		 1535
1536 (void) sprintf(name, "%llx", (longlong_t)user);
1537 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1538
1539 (void) sprintf(name, "%llx", (longlong_t)group);
1540 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
70e083d2
TG		 1541 }
1542}
1543
86e3c28a
CIK		 1544static void
1545do_userobjquota_update(userquota_cache_t *cache, uint64_t flags,
1546 uint64_t user, uint64_t group, boolean_t subtract)
70e083d2 1547{
86e3c28a
CIK		 1548 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) {
1549 char name[20 + DMU_OBJACCT_PREFIX_LEN];
1550 int delta = subtract ? -1 : 1;
1551
1552 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1553 (longlong_t)user);
1554 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1555
1556 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1557 (longlong_t)group);
1558 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1559 }
1560}
1561
1562typedef struct userquota_updates_arg {
1563 objset_t *uua_os;
1564 int uua_sublist_idx;
1565 dmu_tx_t *uua_tx;
1566} userquota_updates_arg_t;
1567
1568static void
1569userquota_updates_task(void *arg)
1570{
1571 userquota_updates_arg_t *uua = arg;
1572 objset_t *os = uua->uua_os;
1573 dmu_tx_t *tx = uua->uua_tx;
70e083d2 1574 dnode_t *dn;
86e3c28a 1575 userquota_cache_t cache = { { 0 } };
70e083d2 1576
86e3c28a
CIK		 1577 multilist_sublist_t *list =
1578 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
70e083d2 1579
86e3c28a
CIK		 1580 ASSERT(multilist_sublist_head(list) == NULL ||
1581 dmu_objset_userused_enabled(os));
1582 avl_create(&cache.uqc_user_deltas, userquota_compare,
1583 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1584 avl_create(&cache.uqc_group_deltas, userquota_compare,
1585 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1586
1587 while ((dn = multilist_sublist_head(list)) != NULL) {
70e083d2
TG		 1588 int flags;
1589 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1590 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1591 dn->dn_phys->dn_flags &
1592 DNODE_FLAG_USERUSED_ACCOUNTED);
1593
70e083d2
TG		 1594 flags = dn->dn_id_flags;
1595 ASSERT(flags);
1596 if (flags & DN_ID_OLD_EXIST) {
86e3c28a
CIK		 1597 do_userquota_update(&cache,
1598 dn->dn_oldused, dn->dn_oldflags,
1599 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1600 do_userobjquota_update(&cache, dn->dn_oldflags,
1601 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
70e083d2
TG		 1602 }
1603 if (flags & DN_ID_NEW_EXIST) {
86e3c28a
CIK		 1604 do_userquota_update(&cache,
1605 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags,
1606 dn->dn_newuid, dn->dn_newgid, B_FALSE);
1607 do_userobjquota_update(&cache, dn->dn_phys->dn_flags,
1608 dn->dn_newuid, dn->dn_newgid, B_FALSE);
70e083d2
TG		 1609 }
1610
1611 mutex_enter(&dn->dn_mtx);
1612 dn->dn_oldused = 0;
1613 dn->dn_oldflags = 0;
1614 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1615 dn->dn_olduid = dn->dn_newuid;
1616 dn->dn_oldgid = dn->dn_newgid;
1617 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1618 if (dn->dn_bonuslen == 0)
1619 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1620 else
1621 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1622 }
1623 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1624 mutex_exit(&dn->dn_mtx);
1625
86e3c28a
CIK		 1626 multilist_sublist_remove(list, dn);
1627 dnode_rele(dn, os->os_synced_dnodes);
1628 }
1629 do_userquota_cacheflush(os, &cache, tx);
1630 multilist_sublist_unlock(list);
1631 kmem_free(uua, sizeof (*uua));
1632}
1633
1634void
1635dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1636{
1637 if (!dmu_objset_userused_enabled(os))
1638 return;
1639
1640 /* Allocate the user/groupused objects if necessary. */
1641 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1642 VERIFY0(zap_create_claim(os,
1643 DMU_USERUSED_OBJECT,
1644 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1645 VERIFY0(zap_create_claim(os,
1646 DMU_GROUPUSED_OBJECT,
1647 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1648 }
1649
1650 for (int i = 0;
1651 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) {
1652 userquota_updates_arg_t *uua =
1653 kmem_alloc(sizeof (*uua), KM_SLEEP);
1654 uua->uua_os = os;
1655 uua->uua_sublist_idx = i;
1656 uua->uua_tx = tx;
1657 /* note: caller does taskq_wait() */
1658 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1659 userquota_updates_task, uua, 0);
1660 /* callback frees uua */
70e083d2
TG		 1661 }
1662}
1663
1664/*
1665 * Returns a pointer to data to find uid/gid from
1666 *
1667 * If a dirty record for transaction group that is syncing can't
1668 * be found then NULL is returned. In the NULL case it is assumed
1669 * the uid/gid aren't changing.
1670 */
1671static void *
1672dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1673{
1674 dbuf_dirty_record_t *dr, **drp;
1675 void *data;
1676
1677 if (db->db_dirtycnt == 0)
1678 return (db->db.db_data); /* Nothing is changing */
1679
1680 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1681 if (dr->dr_txg == tx->tx_txg)
1682 break;
1683
1684 if (dr == NULL) {
1685 data = NULL;
1686 } else {
1687 dnode_t *dn;
1688
1689 DB_DNODE_ENTER(dr->dr_dbuf);
1690 dn = DB_DNODE(dr->dr_dbuf);
1691
1692 if (dn->dn_bonuslen == 0 &&
1693 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1694 data = dr->dt.dl.dr_data->b_data;
1695 else
1696 data = dr->dt.dl.dr_data;
1697
1698 DB_DNODE_EXIT(dr->dr_dbuf);
1699 }
1700
1701 return (data);
1702}
1703
1704void
1705dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1706{
1707 objset_t *os = dn->dn_objset;
1708 void *data = NULL;
1709 dmu_buf_impl_t *db = NULL;
1710 uint64_t *user = NULL;
1711 uint64_t *group = NULL;
1712 int flags = dn->dn_id_flags;
1713 int error;
1714 boolean_t have_spill = B_FALSE;
1715
1716 if (!dmu_objset_userused_enabled(dn->dn_objset))
1717 return;
1718
1719 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1720 DN_ID_CHKED_SPILL)))
1721 return;
1722
1723 if (before && dn->dn_bonuslen != 0)
1724 data = DN_BONUS(dn->dn_phys);
1725 else if (!before && dn->dn_bonuslen != 0) {
1726 if (dn->dn_bonus) {
1727 db = dn->dn_bonus;
1728 mutex_enter(&db->db_mtx);
1729 data = dmu_objset_userquota_find_data(db, tx);
1730 } else {
1731 data = DN_BONUS(dn->dn_phys);
1732 }
1733 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1734 int rf = 0;
1735
1736 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1737 rf |= DB_RF_HAVESTRUCT;
1738 error = dmu_spill_hold_by_dnode(dn,
1739 rf | DB_RF_MUST_SUCCEED,
1740 FTAG, (dmu_buf_t **)&db);
1741 ASSERT(error == 0);
1742 mutex_enter(&db->db_mtx);
1743 data = (before) ? db->db.db_data :
1744 dmu_objset_userquota_find_data(db, tx);
1745 have_spill = B_TRUE;
1746 } else {
1747 mutex_enter(&dn->dn_mtx);
1748 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1749 mutex_exit(&dn->dn_mtx);
1750 return;
1751 }
1752
1753 if (before) {
1754 ASSERT(data);
1755 user = &dn->dn_olduid;
1756 group = &dn->dn_oldgid;
1757 } else if (data) {
1758 user = &dn->dn_newuid;
1759 group = &dn->dn_newgid;
1760 }
1761
1762 /*
1763 * Must always call the callback in case the object
1764 * type has changed and that type isn't an object type to track
1765 */
1766 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1767 user, group);
1768
1769 /*
1770 * Preserve existing uid/gid when the callback can't determine
1771 * what the new uid/gid are and the callback returned EEXIST.
1772 * The EEXIST error tells us to just use the existing uid/gid.
1773 * If we don't know what the old values are then just assign
1774 * them to 0, since that is a new file being created.
1775 */
1776 if (!before && data == NULL && error == EEXIST) {
1777 if (flags & DN_ID_OLD_EXIST) {
1778 dn->dn_newuid = dn->dn_olduid;
1779 dn->dn_newgid = dn->dn_oldgid;
1780 } else {
1781 dn->dn_newuid = 0;
1782 dn->dn_newgid = 0;
1783 }
1784 error = 0;
1785 }
1786
1787 if (db)
1788 mutex_exit(&db->db_mtx);
1789
1790 mutex_enter(&dn->dn_mtx);
1791 if (error == 0 && before)
1792 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1793 if (error == 0 && !before)
1794 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1795
1796 if (have_spill) {
1797 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1798 } else {
1799 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1800 }
1801 mutex_exit(&dn->dn_mtx);
1802 if (have_spill)
1803 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1804}
1805
1806boolean_t
1807dmu_objset_userspace_present(objset_t *os)
1808{
1809 return (os->os_phys->os_flags &
1810 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1811}
1812
86e3c28a
CIK		 1813boolean_t
1814dmu_objset_userobjspace_present(objset_t *os)
1815{
1816 return (os->os_phys->os_flags &
1817 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE);
1818}
1819
1820static int
1821dmu_objset_space_upgrade(objset_t *os)
70e083d2
TG		 1822{
1823 uint64_t obj;
1824 int err = 0;
1825
70e083d2
TG		 1826 /*
1827 * We simply need to mark every object dirty, so that it will be
1828 * synced out and now accounted. If this is called
1829 * concurrently, or if we already did some work before crashing,
1830 * that's fine, since we track each object's accounted state
1831 * independently.
1832 */
1833
1834 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1835 dmu_tx_t *tx;
1836 dmu_buf_t *db;
1837 int objerr;
1838
86e3c28a
CIK		 1839 mutex_enter(&os->os_upgrade_lock);
1840 if (os->os_upgrade_exit)
1841 err = SET_ERROR(EINTR);
1842 mutex_exit(&os->os_upgrade_lock);
1843 if (err != 0)
1844 return (err);
1845
70e083d2
TG		 1846 if (issig(JUSTLOOKING) && issig(FORREAL))
1847 return (SET_ERROR(EINTR));
1848
1849 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1850 if (objerr != 0)
1851 continue;
1852 tx = dmu_tx_create(os);
1853 dmu_tx_hold_bonus(tx, obj);
1854 objerr = dmu_tx_assign(tx, TXG_WAIT);
1855 if (objerr != 0) {
b49151d6 1856 dmu_buf_rele(db, FTAG);
70e083d2
TG		 1857 dmu_tx_abort(tx);
1858 continue;
1859 }
1860 dmu_buf_will_dirty(db, tx);
1861 dmu_buf_rele(db, FTAG);
1862 dmu_tx_commit(tx);
1863 }
86e3c28a
CIK		 1864 return (0);
1865}
1866
1867int
1868dmu_objset_userspace_upgrade(objset_t *os)
1869{
1870 int err = 0;
1871
1872 if (dmu_objset_userspace_present(os))
1873 return (0);
1874 if (dmu_objset_is_snapshot(os))
1875 return (SET_ERROR(EINVAL));
1876 if (!dmu_objset_userused_enabled(os))
1877 return (SET_ERROR(ENOTSUP));
1878
1879 err = dmu_objset_space_upgrade(os);
1880 if (err)
1881 return (err);
70e083d2
TG		 1882
1883 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1884 txg_wait_synced(dmu_objset_pool(os), 0);
1885 return (0);
1886}
1887
86e3c28a
CIK		 1888static int
1889dmu_objset_userobjspace_upgrade_cb(objset_t *os)
1890{
1891 int err = 0;
1892
1893 if (dmu_objset_userobjspace_present(os))
1894 return (0);
1895 if (dmu_objset_is_snapshot(os))
1896 return (SET_ERROR(EINVAL));
1897 if (!dmu_objset_userobjused_enabled(os))
1898 return (SET_ERROR(ENOTSUP));
1899
1900 dmu_objset_ds(os)->ds_feature_activation_needed[
1901 SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE;
1902
1903 err = dmu_objset_space_upgrade(os);
1904 if (err)
1905 return (err);
1906
1907 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
1908 txg_wait_synced(dmu_objset_pool(os), 0);
1909 return (0);
1910}
1911
1912void
1913dmu_objset_userobjspace_upgrade(objset_t *os)
1914{
1915 dmu_objset_upgrade(os, dmu_objset_userobjspace_upgrade_cb);
1916}
1917
1918boolean_t
1919dmu_objset_userobjspace_upgradable(objset_t *os)
1920{
1921 return (dmu_objset_type(os) == DMU_OST_ZFS &&
1922 !dmu_objset_is_snapshot(os) &&
1923 dmu_objset_userobjused_enabled(os) &&
1924 !dmu_objset_userobjspace_present(os));
1925}
1926
70e083d2
TG		 1927void
1928dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1929 uint64_t *usedobjsp, uint64_t *availobjsp)
1930{
1931 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1932 usedobjsp, availobjsp);
1933}
1934
1935uint64_t
1936dmu_objset_fsid_guid(objset_t *os)
1937{
1938 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1939}
1940
1941void
1942dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1943{
1944 stat->dds_type = os->os_phys->os_type;
1945 if (os->os_dsl_dataset)
1946 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1947}
1948
1949void
1950dmu_objset_stats(objset_t *os, nvlist_t *nv)
1951{
1952 ASSERT(os->os_dsl_dataset ||
1953 os->os_phys->os_type == DMU_OST_META);
1954
1955 if (os->os_dsl_dataset != NULL)
1956 dsl_dataset_stats(os->os_dsl_dataset, nv);
1957
1958 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1959 os->os_phys->os_type);
1960 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1961 dmu_objset_userspace_present(os));
1962}
1963
1964int
1965dmu_objset_is_snapshot(objset_t *os)
1966{
1967 if (os->os_dsl_dataset != NULL)
1968 return (os->os_dsl_dataset->ds_is_snapshot);
1969 else
1970 return (B_FALSE);
1971}
1972
1973int
1974dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1975 boolean_t *conflict)
1976{
1977 dsl_dataset_t *ds = os->os_dsl_dataset;
1978 uint64_t ignored;
1979
1980 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1981 return (SET_ERROR(ENOENT));
1982
1983 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1984 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
86e3c28a 1985 MT_NORMALIZE, real, maxlen, conflict));
70e083d2
TG		 1986}
1987
1988int
1989dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1990 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1991{
1992 dsl_dataset_t *ds = os->os_dsl_dataset;
1993 zap_cursor_t cursor;
1994 zap_attribute_t attr;
1995
1996 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1997
1998 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1999 return (SET_ERROR(ENOENT));
2000
2001 zap_cursor_init_serialized(&cursor,
2002 ds->ds_dir->dd_pool->dp_meta_objset,
2003 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
2004
2005 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2006 zap_cursor_fini(&cursor);
2007 return (SET_ERROR(ENOENT));
2008 }
2009
2010 if (strlen(attr.za_name) + 1 > namelen) {
2011 zap_cursor_fini(&cursor);
2012 return (SET_ERROR(ENAMETOOLONG));
2013 }
2014
2015 (void) strcpy(name, attr.za_name);
2016 if (idp)
2017 *idp = attr.za_first_integer;
2018 if (case_conflict)
2019 *case_conflict = attr.za_normalization_conflict;
2020 zap_cursor_advance(&cursor);
2021 *offp = zap_cursor_serialize(&cursor);
2022 zap_cursor_fini(&cursor);
2023
2024 return (0);
2025}
2026
2027int
2028dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
2029{
2030 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
2031}
2032
2033int
2034dmu_dir_list_next(objset_t *os, int namelen, char *name,
2035 uint64_t *idp, uint64_t *offp)
2036{
2037 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
2038 zap_cursor_t cursor;
2039 zap_attribute_t attr;
2040
2041 /* there is no next dir on a snapshot! */
2042 if (os->os_dsl_dataset->ds_object !=
2043 dsl_dir_phys(dd)->dd_head_dataset_obj)
2044 return (SET_ERROR(ENOENT));
2045
2046 zap_cursor_init_serialized(&cursor,
2047 dd->dd_pool->dp_meta_objset,
2048 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
2049
2050 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2051 zap_cursor_fini(&cursor);
2052 return (SET_ERROR(ENOENT));
2053 }
2054
2055 if (strlen(attr.za_name) + 1 > namelen) {
2056 zap_cursor_fini(&cursor);
2057 return (SET_ERROR(ENAMETOOLONG));
2058 }
2059
2060 (void) strcpy(name, attr.za_name);
2061 if (idp)
2062 *idp = attr.za_first_integer;
2063 zap_cursor_advance(&cursor);
2064 *offp = zap_cursor_serialize(&cursor);
2065 zap_cursor_fini(&cursor);
2066
2067 return (0);
2068}
2069
2070typedef struct dmu_objset_find_ctx {
2071 taskq_t *dc_tq;
2072 dsl_pool_t *dc_dp;
2073 uint64_t dc_ddobj;
86e3c28a 2074 char *dc_ddname; /* last component of ddobj's name */
70e083d2
TG		 2075 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
2076 void *dc_arg;
2077 int dc_flags;
2078 kmutex_t *dc_error_lock;
2079 int *dc_error;
2080} dmu_objset_find_ctx_t;
2081
2082static void
2083dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
2084{
2085 dsl_pool_t *dp = dcp->dc_dp;
2086 dmu_objset_find_ctx_t *child_dcp;
2087 dsl_dir_t *dd;
2088 dsl_dataset_t *ds;
2089 zap_cursor_t zc;
2090 zap_attribute_t *attr;
2091 uint64_t thisobj;
2092 int err = 0;
2093
2094 /* don't process if there already was an error */
2095 if (*dcp->dc_error != 0)
2096 goto out;
2097
86e3c28a
CIK		 2098 /*
2099 * Note: passing the name (dc_ddname) here is optional, but it
2100 * improves performance because we don't need to call
2101 * zap_value_search() to determine the name.
2102 */
2103 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
70e083d2
TG		 2104 if (err != 0)
2105 goto out;
2106
2107 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2108 if (dd->dd_myname[0] == '$') {
2109 dsl_dir_rele(dd, FTAG);
2110 goto out;
2111 }
2112
2113 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2114 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2115
2116 /*
2117 * Iterate over all children.
2118 */
2119 if (dcp->dc_flags & DS_FIND_CHILDREN) {
2120 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2121 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2122 zap_cursor_retrieve(&zc, attr) == 0;
2123 (void) zap_cursor_advance(&zc)) {
2124 ASSERT3U(attr->za_integer_length, ==,
2125 sizeof (uint64_t));
2126 ASSERT3U(attr->za_num_integers, ==, 1);
2127
86e3c28a
CIK		 2128 child_dcp =
2129 kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
70e083d2
TG		 2130 *child_dcp = *dcp;
2131 child_dcp->dc_ddobj = attr->za_first_integer;
86e3c28a 2132 child_dcp->dc_ddname = spa_strdup(attr->za_name);
70e083d2
TG		 2133 if (dcp->dc_tq != NULL)
2134 (void) taskq_dispatch(dcp->dc_tq,
2135 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2136 else
2137 dmu_objset_find_dp_impl(child_dcp);
2138 }
2139 zap_cursor_fini(&zc);
2140 }
2141
2142 /*
2143 * Iterate over all snapshots.
2144 */
2145 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2146 dsl_dataset_t *ds;
2147 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2148
2149 if (err == 0) {
2150 uint64_t snapobj;
2151
2152 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2153 dsl_dataset_rele(ds, FTAG);
2154
2155 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2156 zap_cursor_retrieve(&zc, attr) == 0;
2157 (void) zap_cursor_advance(&zc)) {
2158 ASSERT3U(attr->za_integer_length, ==,
2159 sizeof (uint64_t));
2160 ASSERT3U(attr->za_num_integers, ==, 1);
2161
2162 err = dsl_dataset_hold_obj(dp,
2163 attr->za_first_integer, FTAG, &ds);
2164 if (err != 0)
2165 break;
2166 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2167 dsl_dataset_rele(ds, FTAG);
2168 if (err != 0)
2169 break;
2170 }
2171 zap_cursor_fini(&zc);
2172 }
2173 }
2174
70e083d2
TG		 2175 kmem_free(attr, sizeof (zap_attribute_t));
2176
86e3c28a
CIK		 2177 if (err != 0) {
2178 dsl_dir_rele(dd, FTAG);
70e083d2 2179 goto out;
86e3c28a 2180 }
70e083d2
TG		 2181
2182 /*
2183 * Apply to self.
2184 */
2185 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
86e3c28a
CIK		 2186
2187 /*
2188 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2189 * that the dir will remain cached, and we won't have to re-instantiate
2190 * it (which could be expensive due to finding its name via
2191 * zap_value_search()).
2192 */
2193 dsl_dir_rele(dd, FTAG);
70e083d2
TG		 2194 if (err != 0)
2195 goto out;
2196 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2197 dsl_dataset_rele(ds, FTAG);
2198
2199out:
2200 if (err != 0) {
2201 mutex_enter(dcp->dc_error_lock);
2202 /* only keep first error */
2203 if (*dcp->dc_error == 0)
2204 *dcp->dc_error = err;
2205 mutex_exit(dcp->dc_error_lock);
2206 }
2207
86e3c28a
CIK		 2208 if (dcp->dc_ddname != NULL)
2209 spa_strfree(dcp->dc_ddname);
70e083d2
TG		 2210 kmem_free(dcp, sizeof (*dcp));
2211}
2212
2213static void
2214dmu_objset_find_dp_cb(void *arg)
2215{
2216 dmu_objset_find_ctx_t *dcp = arg;
2217 dsl_pool_t *dp = dcp->dc_dp;
2218
2219 /*
2220 * We need to get a pool_config_lock here, as there are several
2221 * asssert(pool_config_held) down the stack. Getting a lock via
2222 * dsl_pool_config_enter is risky, as it might be stalled by a
2223 * pending writer. This would deadlock, as the write lock can
2224 * only be granted when our parent thread gives up the lock.
2225 * The _prio interface gives us priority over a pending writer.
2226 */
2227 dsl_pool_config_enter_prio(dp, FTAG);
2228
2229 dmu_objset_find_dp_impl(dcp);
2230
2231 dsl_pool_config_exit(dp, FTAG);
2232}
2233
2234/*
2235 * Find objsets under and including ddobj, call func(ds) on each.
2236 * The order for the enumeration is completely undefined.
2237 * func is called with dsl_pool_config held.
2238 */
2239int
2240dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2241 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2242{
2243 int error = 0;
2244 taskq_t *tq = NULL;
2245 int ntasks;
2246 dmu_objset_find_ctx_t *dcp;
2247 kmutex_t err_lock;
2248
2249 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2250 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2251 dcp->dc_tq = NULL;
2252 dcp->dc_dp = dp;
2253 dcp->dc_ddobj = ddobj;
86e3c28a 2254 dcp->dc_ddname = NULL;
70e083d2
TG		 2255 dcp->dc_func = func;
2256 dcp->dc_arg = arg;
2257 dcp->dc_flags = flags;
2258 dcp->dc_error_lock = &err_lock;
2259 dcp->dc_error = &error;
2260
2261 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2262 /*
2263 * In case a write lock is held we can't make use of
2264 * parallelism, as down the stack of the worker threads
2265 * the lock is asserted via dsl_pool_config_held.
2266 * In case of a read lock this is solved by getting a read
2267 * lock in each worker thread, which isn't possible in case
2268 * of a writer lock. So we fall back to the synchronous path
2269 * here.
2270 * In the future it might be possible to get some magic into
2271 * dsl_pool_config_held in a way that it returns true for
2272 * the worker threads so that a single lock held from this
2273 * thread suffices. For now, stay single threaded.
2274 */
2275 dmu_objset_find_dp_impl(dcp);
86e3c28a 2276 mutex_destroy(&err_lock);
70e083d2
TG		 2277
2278 return (error);
2279 }
2280
2281 ntasks = dmu_find_threads;
2282 if (ntasks == 0)
2283 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2284 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks,
2285 INT_MAX, 0);
2286 if (tq == NULL) {
2287 kmem_free(dcp, sizeof (*dcp));
86e3c28a
CIK		 2288 mutex_destroy(&err_lock);
2289
70e083d2
TG		 2290 return (SET_ERROR(ENOMEM));
2291 }
2292 dcp->dc_tq = tq;
2293
2294 /* dcp will be freed by task */
2295 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2296
2297 /*
2298 * PORTING: this code relies on the property of taskq_wait to wait
2299 * until no more tasks are queued and no more tasks are active. As
2300 * we always queue new tasks from within other tasks, task_wait
2301 * reliably waits for the full recursion to finish, even though we
2302 * enqueue new tasks after taskq_wait has been called.
2303 * On platforms other than illumos, taskq_wait may not have this
2304 * property.
2305 */
2306 taskq_wait(tq);
2307 taskq_destroy(tq);
2308 mutex_destroy(&err_lock);
2309
2310 return (error);
2311}
2312
2313/*
2314 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2315 * The dp_config_rwlock must not be held when this is called, and it
2316 * will not be held when the callback is called.
2317 * Therefore this function should only be used when the pool is not changing
2318 * (e.g. in syncing context), or the callback can deal with the possible races.
2319 */
2320static int
2321dmu_objset_find_impl(spa_t *spa, const char *name,
2322 int func(const char *, void *), void *arg, int flags)
2323{
2324 dsl_dir_t *dd;
2325 dsl_pool_t *dp = spa_get_dsl(spa);
2326 dsl_dataset_t *ds;
2327 zap_cursor_t zc;
2328 zap_attribute_t *attr;
2329 char *child;
2330 uint64_t thisobj;
2331 int err;
2332
2333 dsl_pool_config_enter(dp, FTAG);
2334
2335 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2336 if (err != 0) {
2337 dsl_pool_config_exit(dp, FTAG);
2338 return (err);
2339 }
2340
2341 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2342 if (dd->dd_myname[0] == '$') {
2343 dsl_dir_rele(dd, FTAG);
2344 dsl_pool_config_exit(dp, FTAG);
2345 return (0);
2346 }
2347
2348 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2349 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2350
2351 /*
2352 * Iterate over all children.
2353 */
2354 if (flags & DS_FIND_CHILDREN) {
2355 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2356 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2357 zap_cursor_retrieve(&zc, attr) == 0;
2358 (void) zap_cursor_advance(&zc)) {
2359 ASSERT3U(attr->za_integer_length, ==,
2360 sizeof (uint64_t));
2361 ASSERT3U(attr->za_num_integers, ==, 1);
2362
2363 child = kmem_asprintf("%s/%s", name, attr->za_name);
2364 dsl_pool_config_exit(dp, FTAG);
2365 err = dmu_objset_find_impl(spa, child,
2366 func, arg, flags);
2367 dsl_pool_config_enter(dp, FTAG);
2368 strfree(child);
2369 if (err != 0)
2370 break;
2371 }
2372 zap_cursor_fini(&zc);
2373
2374 if (err != 0) {
2375 dsl_dir_rele(dd, FTAG);
2376 dsl_pool_config_exit(dp, FTAG);
2377 kmem_free(attr, sizeof (zap_attribute_t));
2378 return (err);
2379 }
2380 }
2381
2382 /*
2383 * Iterate over all snapshots.
2384 */
2385 if (flags & DS_FIND_SNAPSHOTS) {
2386 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2387
2388 if (err == 0) {
2389 uint64_t snapobj;
2390
2391 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2392 dsl_dataset_rele(ds, FTAG);
2393
2394 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2395 zap_cursor_retrieve(&zc, attr) == 0;
2396 (void) zap_cursor_advance(&zc)) {
2397 ASSERT3U(attr->za_integer_length, ==,
2398 sizeof (uint64_t));
2399 ASSERT3U(attr->za_num_integers, ==, 1);
2400
2401 child = kmem_asprintf("%s@%s",
2402 name, attr->za_name);
2403 dsl_pool_config_exit(dp, FTAG);
2404 err = func(child, arg);
2405 dsl_pool_config_enter(dp, FTAG);
2406 strfree(child);
2407 if (err != 0)
2408 break;
2409 }
2410 zap_cursor_fini(&zc);
2411 }
2412 }
2413
2414 dsl_dir_rele(dd, FTAG);
2415 kmem_free(attr, sizeof (zap_attribute_t));
2416 dsl_pool_config_exit(dp, FTAG);
2417
2418 if (err != 0)
2419 return (err);
2420
2421 /* Apply to self. */
2422 return (func(name, arg));
2423}
2424
2425/*
2426 * See comment above dmu_objset_find_impl().
2427 */
2428int
2429dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2430 int flags)
2431{
2432 spa_t *spa;
2433 int error;
2434
2435 error = spa_open(name, &spa, FTAG);
2436 if (error != 0)
2437 return (error);
2438 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2439 spa_close(spa, FTAG);
2440 return (error);
2441}
2442
2443void
2444dmu_objset_set_user(objset_t *os, void *user_ptr)
2445{
2446 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2447 os->os_user_ptr = user_ptr;
2448}
2449
2450void *
2451dmu_objset_get_user(objset_t *os)
2452{
2453 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2454 return (os->os_user_ptr);
2455}
2456
2457/*
2458 * Determine name of filesystem, given name of snapshot.
86e3c28a 2459 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
70e083d2
TG		 2460 */
2461int
2462dmu_fsname(const char *snapname, char *buf)
2463{
2464 char *atp = strchr(snapname, '@');
2465 if (atp == NULL)
2466 return (SET_ERROR(EINVAL));
86e3c28a 2467 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
70e083d2
TG		 2468 return (SET_ERROR(ENAMETOOLONG));
2469 (void) strlcpy(buf, snapname, atp - snapname + 1);
2470 return (0);
2471}
2472
86e3c28a
CIK		 2473/*
2474 * Call when we think we're going to write/free space in open context to track
2475 * the amount of dirty data in the open txg, which is also the amount
2476 * of memory that can not be evicted until this txg syncs.
2477 */
2478void
2479dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2480{
2481 dsl_dataset_t *ds = os->os_dsl_dataset;
2482 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2483
2484 if (ds != NULL) {
2485 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2486 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
2487 }
2488}
2489
70e083d2
TG		 2490#if defined(_KERNEL) && defined(HAVE_SPL)
2491EXPORT_SYMBOL(dmu_objset_zil);
2492EXPORT_SYMBOL(dmu_objset_pool);
2493EXPORT_SYMBOL(dmu_objset_ds);
2494EXPORT_SYMBOL(dmu_objset_type);
2495EXPORT_SYMBOL(dmu_objset_name);
2496EXPORT_SYMBOL(dmu_objset_hold);
2497EXPORT_SYMBOL(dmu_objset_own);
2498EXPORT_SYMBOL(dmu_objset_rele);
2499EXPORT_SYMBOL(dmu_objset_disown);
2500EXPORT_SYMBOL(dmu_objset_from_ds);
2501EXPORT_SYMBOL(dmu_objset_create);
2502EXPORT_SYMBOL(dmu_objset_clone);
2503EXPORT_SYMBOL(dmu_objset_stats);
2504EXPORT_SYMBOL(dmu_objset_fast_stat);
2505EXPORT_SYMBOL(dmu_objset_spa);
2506EXPORT_SYMBOL(dmu_objset_space);
2507EXPORT_SYMBOL(dmu_objset_fsid_guid);
2508EXPORT_SYMBOL(dmu_objset_find);
2509EXPORT_SYMBOL(dmu_objset_byteswap);
2510EXPORT_SYMBOL(dmu_objset_evict_dbufs);
2511EXPORT_SYMBOL(dmu_objset_snap_cmtime);
86e3c28a 2512EXPORT_SYMBOL(dmu_objset_dnodesize);
70e083d2
TG		 2513
2514EXPORT_SYMBOL(dmu_objset_sync);
2515EXPORT_SYMBOL(dmu_objset_is_dirty);
2516EXPORT_SYMBOL(dmu_objset_create_impl);
2517EXPORT_SYMBOL(dmu_objset_open_impl);
2518EXPORT_SYMBOL(dmu_objset_evict);
2519EXPORT_SYMBOL(dmu_objset_register_type);
2520EXPORT_SYMBOL(dmu_objset_do_userquota_updates);
2521EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
2522EXPORT_SYMBOL(dmu_objset_userused_enabled);
2523EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
2524EXPORT_SYMBOL(dmu_objset_userspace_present);
86e3c28a
CIK		 2525EXPORT_SYMBOL(dmu_objset_userobjused_enabled);
2526EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade);
2527EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable);
2528EXPORT_SYMBOL(dmu_objset_userobjspace_present);
70e083d2 2529#endif
ubuntu-bionic-kernel mirror