]> git.proxmox.com Git - mirror_zfs.git/blob - module/zfs/dsl_dataset.c
projects / mirror_zfs.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Undo c89 workarounds to match with upstream
[mirror_zfs.git] / module / zfs / dsl_dataset.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 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2016 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
31 */
32
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_dir.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dmu_traverse.h>
39 #include <sys/dmu_impl.h>
40 #include <sys/dmu_tx.h>
41 #include <sys/arc.h>
42 #include <sys/zio.h>
43 #include <sys/zap.h>
44 #include <sys/zfeature.h>
45 #include <sys/unique.h>
46 #include <sys/zfs_context.h>
47 #include <sys/zfs_ioctl.h>
48 #include <sys/spa.h>
49 #include <sys/zfs_znode.h>
50 #include <sys/zfs_onexit.h>
51 #include <sys/zvol.h>
52 #include <sys/dsl_scan.h>
53 #include <sys/dsl_deadlist.h>
54 #include <sys/dsl_destroy.h>
55 #include <sys/dsl_userhold.h>
56 #include <sys/dsl_bookmark.h>
57 #include <sys/policy.h>
58 #include <sys/dmu_send.h>
59 #include <sys/zio_compress.h>
60 #include <zfs_fletcher.h>
61 #include <sys/zio_checksum.h>
62
63 /*
64 * The SPA supports block sizes up to 16MB. However, very large blocks
65 * can have an impact on i/o latency (e.g. tying up a spinning disk for
66 * ~300ms), and also potentially on the memory allocator. Therefore,
67 * we do not allow the recordsize to be set larger than zfs_max_recordsize
68 * (default 1MB). Larger blocks can be created by changing this tunable,
69 * and pools with larger blocks can always be imported and used, regardless
70 * of this setting.
71 */
72 int zfs_max_recordsize = 1 * 1024 * 1024;
73
74 #define SWITCH64(x, y) \
75 { \
76 uint64_t __tmp = (x); \
77 (x) = (y); \
78 (y) = __tmp; \
79 }
80
81 #define DS_REF_MAX (1ULL << 62)
82
83 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
84
85 extern int spa_asize_inflation;
86
87 static zil_header_t zero_zil;
88
89 /*
90 * Figure out how much of this delta should be propagated to the dsl_dir
91 * layer. If there's a refreservation, that space has already been
92 * partially accounted for in our ancestors.
93 */
94 static int64_t
95 parent_delta(dsl_dataset_t *ds, int64_t delta)
96 {
97 dsl_dataset_phys_t *ds_phys;
98 uint64_t old_bytes, new_bytes;
99
100 if (ds->ds_reserved == 0)
101 return (delta);
102
103 ds_phys = dsl_dataset_phys(ds);
104 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
105 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
106
107 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
108 return (new_bytes - old_bytes);
109 }
110
111 void
112 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
113 {
114 int used, compressed, uncompressed;
115 int64_t delta;
116
117 used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
118 compressed = BP_GET_PSIZE(bp);
119 uncompressed = BP_GET_UCSIZE(bp);
120
121 dprintf_bp(bp, "ds=%p", ds);
122
123 ASSERT(dmu_tx_is_syncing(tx));
124 /* It could have been compressed away to nothing */
125 if (BP_IS_HOLE(bp))
126 return;
127 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
128 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
129 if (ds == NULL) {
130 dsl_pool_mos_diduse_space(tx->tx_pool,
131 used, compressed, uncompressed);
132 return;
133 }
134
135 ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
136 dmu_buf_will_dirty(ds->ds_dbuf, tx);
137 mutex_enter(&ds->ds_lock);
138 delta = parent_delta(ds, used);
139 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
140 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
141 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
142 dsl_dataset_phys(ds)->ds_unique_bytes += used;
143
144 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
145 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
146 B_TRUE;
147 }
148
149 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
150 if (f != SPA_FEATURE_NONE)
151 ds->ds_feature_activation_needed[f] = B_TRUE;
152
153 mutex_exit(&ds->ds_lock);
154 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
155 compressed, uncompressed, tx);
156 dsl_dir_transfer_space(ds->ds_dir, used - delta,
157 DD_USED_REFRSRV, DD_USED_HEAD, tx);
158 }
159
160 int
161 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
162 boolean_t async)
163 {
164 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
165 int compressed = BP_GET_PSIZE(bp);
166 int uncompressed = BP_GET_UCSIZE(bp);
167
168 if (BP_IS_HOLE(bp))
169 return (0);
170
171 ASSERT(dmu_tx_is_syncing(tx));
172 ASSERT(bp->blk_birth <= tx->tx_txg);
173
174 if (ds == NULL) {
175 dsl_free(tx->tx_pool, tx->tx_txg, bp);
176 dsl_pool_mos_diduse_space(tx->tx_pool,
177 -used, -compressed, -uncompressed);
178 return (used);
179 }
180 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
181
182 ASSERT(!ds->ds_is_snapshot);
183 dmu_buf_will_dirty(ds->ds_dbuf, tx);
184
185 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
186 int64_t delta;
187
188 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
189 dsl_free(tx->tx_pool, tx->tx_txg, bp);
190
191 mutex_enter(&ds->ds_lock);
192 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
193 !DS_UNIQUE_IS_ACCURATE(ds));
194 delta = parent_delta(ds, -used);
195 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
196 mutex_exit(&ds->ds_lock);
197 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
198 delta, -compressed, -uncompressed, tx);
199 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
200 DD_USED_REFRSRV, DD_USED_HEAD, tx);
201 } else {
202 dprintf_bp(bp, "putting on dead list: %s", "");
203 if (async) {
204 /*
205 * We are here as part of zio's write done callback,
206 * which means we're a zio interrupt thread. We can't
207 * call dsl_deadlist_insert() now because it may block
208 * waiting for I/O. Instead, put bp on the deferred
209 * queue and let dsl_pool_sync() finish the job.
210 */
211 bplist_append(&ds->ds_pending_deadlist, bp);
212 } else {
213 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
214 }
215 ASSERT3U(ds->ds_prev->ds_object, ==,
216 dsl_dataset_phys(ds)->ds_prev_snap_obj);
217 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
218 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
219 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
220 ds->ds_object && bp->blk_birth >
221 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
222 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
223 mutex_enter(&ds->ds_prev->ds_lock);
224 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
225 mutex_exit(&ds->ds_prev->ds_lock);
226 }
227 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
228 dsl_dir_transfer_space(ds->ds_dir, used,
229 DD_USED_HEAD, DD_USED_SNAP, tx);
230 }
231 }
232 mutex_enter(&ds->ds_lock);
233 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
234 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
235 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
236 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
237 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
238 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
239 mutex_exit(&ds->ds_lock);
240
241 return (used);
242 }
243
244 /*
245 * We have to release the fsid syncronously or we risk that a subsequent
246 * mount of the same dataset will fail to unique_insert the fsid. This
247 * failure would manifest itself as the fsid of this dataset changing
248 * between mounts which makes NFS clients quite unhappy.
249 */
250 static void
251 dsl_dataset_evict_sync(void *dbu)
252 {
253 dsl_dataset_t *ds = dbu;
254
255 ASSERT(ds->ds_owner == NULL);
256
257 unique_remove(ds->ds_fsid_guid);
258 }
259
260 static void
261 dsl_dataset_evict_async(void *dbu)
262 {
263 dsl_dataset_t *ds = dbu;
264
265 ASSERT(ds->ds_owner == NULL);
266
267 ds->ds_dbuf = NULL;
268
269 if (ds->ds_objset != NULL)
270 dmu_objset_evict(ds->ds_objset);
271
272 if (ds->ds_prev) {
273 dsl_dataset_rele(ds->ds_prev, ds);
274 ds->ds_prev = NULL;
275 }
276
277 bplist_destroy(&ds->ds_pending_deadlist);
278 if (ds->ds_deadlist.dl_os != NULL)
279 dsl_deadlist_close(&ds->ds_deadlist);
280 if (ds->ds_dir)
281 dsl_dir_async_rele(ds->ds_dir, ds);
282
283 ASSERT(!list_link_active(&ds->ds_synced_link));
284
285 list_destroy(&ds->ds_prop_cbs);
286 mutex_destroy(&ds->ds_lock);
287 mutex_destroy(&ds->ds_opening_lock);
288 mutex_destroy(&ds->ds_sendstream_lock);
289 refcount_destroy(&ds->ds_longholds);
290 rrw_destroy(&ds->ds_bp_rwlock);
291
292 kmem_free(ds, sizeof (dsl_dataset_t));
293 }
294
295 int
296 dsl_dataset_get_snapname(dsl_dataset_t *ds)
297 {
298 dsl_dataset_phys_t *headphys;
299 int err;
300 dmu_buf_t *headdbuf;
301 dsl_pool_t *dp = ds->ds_dir->dd_pool;
302 objset_t *mos = dp->dp_meta_objset;
303
304 if (ds->ds_snapname[0])
305 return (0);
306 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
307 return (0);
308
309 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
310 FTAG, &headdbuf);
311 if (err != 0)
312 return (err);
313 headphys = headdbuf->db_data;
314 err = zap_value_search(dp->dp_meta_objset,
315 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
316 if (err != 0 && zfs_recover == B_TRUE) {
317 err = 0;
318 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),
319 "SNAPOBJ=%llu-ERR=%d",
320 (unsigned long long)ds->ds_object, err);
321 }
322 dmu_buf_rele(headdbuf, FTAG);
323 return (err);
324 }
325
326 int
327 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
328 {
329 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
330 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
331 matchtype_t mt = 0;
332 int err;
333
334 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
335 mt = MT_NORMALIZE;
336
337 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
338 value, mt, NULL, 0, NULL);
339 if (err == ENOTSUP && (mt & MT_NORMALIZE))
340 err = zap_lookup(mos, snapobj, name, 8, 1, value);
341 return (err);
342 }
343
344 int
345 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
346 boolean_t adj_cnt)
347 {
348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
350 matchtype_t mt = 0;
351 int err;
352
353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354
355 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
356 mt = MT_NORMALIZE;
357
358 err = zap_remove_norm(mos, snapobj, name, mt, tx);
359 if (err == ENOTSUP && (mt & MT_NORMALIZE))
360 err = zap_remove(mos, snapobj, name, tx);
361
362 if (err == 0 && adj_cnt)
363 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
364 DD_FIELD_SNAPSHOT_COUNT, tx);
365
366 return (err);
367 }
368
369 boolean_t
370 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
371 {
372 dmu_buf_t *dbuf = ds->ds_dbuf;
373 boolean_t result = B_FALSE;
374
375 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
376 ds->ds_object, DMU_BONUS_BLKID, tag)) {
377
378 if (ds == dmu_buf_get_user(dbuf))
379 result = B_TRUE;
380 else
381 dmu_buf_rele(dbuf, tag);
382 }
383
384 return (result);
385 }
386
387 int
388 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj,
389 ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp)
390 {
391 objset_t *mos = dp->dp_meta_objset;
392 dmu_buf_t *dbuf;
393 dsl_dataset_t *ds;
394 int err;
395 dmu_object_info_t doi;
396
397 ASSERT(dsl_pool_config_held(dp));
398
399 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
400 if (err != 0)
401 return (err);
402
403 /* Make sure dsobj has the correct object type. */
404 dmu_object_info_from_db(dbuf, &doi);
405 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
406 dmu_buf_rele(dbuf, tag);
407 return (SET_ERROR(EINVAL));
408 }
409
410 ds = dmu_buf_get_user(dbuf);
411 if (ds == NULL) {
412 dsl_dataset_t *winner = NULL;
413
414 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
415 ds->ds_dbuf = dbuf;
416 ds->ds_object = dsobj;
417 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
418 list_link_init(&ds->ds_synced_link);
419
420 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
421 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
422 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
423 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
424 refcount_create(&ds->ds_longholds);
425
426 bplist_create(&ds->ds_pending_deadlist);
427 dsl_deadlist_open(&ds->ds_deadlist,
428 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
429
430 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
431 offsetof(dmu_sendarg_t, dsa_link));
432
433 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
434 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
435
436 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
437 spa_feature_t f;
438
439 for (f = 0; f < SPA_FEATURES; f++) {
440 if (!(spa_feature_table[f].fi_flags &
441 ZFEATURE_FLAG_PER_DATASET))
442 continue;
443 err = zap_contains(mos, dsobj,
444 spa_feature_table[f].fi_guid);
445 if (err == 0) {
446 ds->ds_feature_inuse[f] = B_TRUE;
447 } else {
448 ASSERT3U(err, ==, ENOENT);
449 err = 0;
450 }
451 }
452 }
453
454 err = dsl_dir_hold_obj(dp,
455 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
456 if (err != 0) {
457 mutex_destroy(&ds->ds_lock);
458 mutex_destroy(&ds->ds_opening_lock);
459 mutex_destroy(&ds->ds_sendstream_lock);
460 refcount_destroy(&ds->ds_longholds);
461 bplist_destroy(&ds->ds_pending_deadlist);
462 dsl_deadlist_close(&ds->ds_deadlist);
463 kmem_free(ds, sizeof (dsl_dataset_t));
464 dmu_buf_rele(dbuf, tag);
465 return (err);
466 }
467
468 if (!ds->ds_is_snapshot) {
469 ds->ds_snapname[0] = '\0';
470 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
471 err = dsl_dataset_hold_obj(dp,
472 dsl_dataset_phys(ds)->ds_prev_snap_obj,
473 ds, &ds->ds_prev);
474 }
475 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
476 int zaperr = zap_lookup(mos, ds->ds_object,
477 DS_FIELD_BOOKMARK_NAMES,
478 sizeof (ds->ds_bookmarks), 1,
479 &ds->ds_bookmarks);
480 if (zaperr != ENOENT)
481 VERIFY0(zaperr);
482 }
483 } else {
484 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
485 err = dsl_dataset_get_snapname(ds);
486 if (err == 0 &&
487 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
488 err = zap_count(
489 ds->ds_dir->dd_pool->dp_meta_objset,
490 dsl_dataset_phys(ds)->ds_userrefs_obj,
491 &ds->ds_userrefs);
492 }
493 }
494
495 if (err == 0 && !ds->ds_is_snapshot) {
496 err = dsl_prop_get_int_ds(ds,
497 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
498 &ds->ds_reserved);
499 if (err == 0) {
500 err = dsl_prop_get_int_ds(ds,
501 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
502 &ds->ds_quota);
503 }
504 } else {
505 ds->ds_reserved = ds->ds_quota = 0;
506 }
507
508 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
509 dsl_dataset_evict_async, &ds->ds_dbuf);
510 if (err == 0)
511 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
512
513 if (err != 0 || winner != NULL) {
514 bplist_destroy(&ds->ds_pending_deadlist);
515 dsl_deadlist_close(&ds->ds_deadlist);
516 if (ds->ds_prev)
517 dsl_dataset_rele(ds->ds_prev, ds);
518 dsl_dir_rele(ds->ds_dir, ds);
519 mutex_destroy(&ds->ds_lock);
520 mutex_destroy(&ds->ds_opening_lock);
521 mutex_destroy(&ds->ds_sendstream_lock);
522 refcount_destroy(&ds->ds_longholds);
523 kmem_free(ds, sizeof (dsl_dataset_t));
524 if (err != 0) {
525 dmu_buf_rele(dbuf, tag);
526 return (err);
527 }
528 ds = winner;
529 } else {
530 ds->ds_fsid_guid =
531 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
532 if (ds->ds_fsid_guid !=
533 dsl_dataset_phys(ds)->ds_fsid_guid) {
534 zfs_dbgmsg("ds_fsid_guid changed from "
535 "%llx to %llx for pool %s dataset id %llu",
536 (long long)
537 dsl_dataset_phys(ds)->ds_fsid_guid,
538 (long long)ds->ds_fsid_guid,
539 spa_name(dp->dp_spa),
540 dsobj);
541 }
542 }
543 }
544 ASSERT3P(ds->ds_dbuf, ==, dbuf);
545 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
546 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
547 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
548 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
549 *dsp = ds;
550
551 if ((flags & DS_HOLD_FLAG_DECRYPT) && ds->ds_dir->dd_crypto_obj != 0) {
552 err = spa_keystore_create_mapping(dp->dp_spa, ds, ds);
553 if (err != 0) {
554 dsl_dataset_rele(ds, tag);
555 return (SET_ERROR(EACCES));
556 }
557 }
558
559 return (0);
560 }
561
562 int
563 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
564 dsl_dataset_t **dsp)
565 {
566 return (dsl_dataset_hold_obj_flags(dp, dsobj, 0, tag, dsp));
567 }
568
569 int
570 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
571 void *tag, dsl_dataset_t **dsp)
572 {
573 dsl_dir_t *dd;
574 const char *snapname;
575 uint64_t obj;
576 int err = 0;
577 dsl_dataset_t *ds;
578
579 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
580 if (err != 0)
581 return (err);
582
583 ASSERT(dsl_pool_config_held(dp));
584 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
585 if (obj != 0)
586 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds);
587 else
588 err = SET_ERROR(ENOENT);
589
590 /* we may be looking for a snapshot */
591 if (err == 0 && snapname != NULL) {
592 dsl_dataset_t *snap_ds;
593
594 if (*snapname++ != '@') {
595 dsl_dataset_rele_flags(ds, flags, tag);
596 dsl_dir_rele(dd, FTAG);
597 return (SET_ERROR(ENOENT));
598 }
599
600 dprintf("looking for snapshot '%s'\n", snapname);
601 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
602 if (err == 0) {
603 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag,
604 &snap_ds);
605 }
606 dsl_dataset_rele_flags(ds, flags, tag);
607
608 if (err == 0) {
609 mutex_enter(&snap_ds->ds_lock);
610 if (snap_ds->ds_snapname[0] == 0)
611 (void) strlcpy(snap_ds->ds_snapname, snapname,
612 sizeof (snap_ds->ds_snapname));
613 mutex_exit(&snap_ds->ds_lock);
614 ds = snap_ds;
615 }
616 }
617 if (err == 0)
618 *dsp = ds;
619 dsl_dir_rele(dd, FTAG);
620 return (err);
621 }
622
623 int
624 dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag,
625 dsl_dataset_t **dsp)
626 {
627 return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp));
628 }
629
630 int
631 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
632 void *tag, dsl_dataset_t **dsp)
633 {
634 int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp);
635 if (err != 0)
636 return (err);
637 if (!dsl_dataset_tryown(*dsp, tag)) {
638 dsl_dataset_rele_flags(*dsp, flags, tag);
639 *dsp = NULL;
640 return (SET_ERROR(EBUSY));
641 }
642 return (0);
643 }
644
645 int
646 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
647 void *tag, dsl_dataset_t **dsp)
648 {
649 int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp);
650 if (err != 0)
651 return (err);
652 if (!dsl_dataset_tryown(*dsp, tag)) {
653 dsl_dataset_rele_flags(*dsp, flags, tag);
654 return (SET_ERROR(EBUSY));
655 }
656 return (0);
657 }
658
659 /*
660 * See the comment above dsl_pool_hold() for details. In summary, a long
661 * hold is used to prevent destruction of a dataset while the pool hold
662 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
663 *
664 * The dataset and pool must be held when this function is called. After it
665 * is called, the pool hold may be released while the dataset is still held
666 * and accessed.
667 */
668 void
669 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
670 {
671 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
672 (void) refcount_add(&ds->ds_longholds, tag);
673 }
674
675 void
676 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
677 {
678 (void) refcount_remove(&ds->ds_longholds, tag);
679 }
680
681 /* Return B_TRUE if there are any long holds on this dataset. */
682 boolean_t
683 dsl_dataset_long_held(dsl_dataset_t *ds)
684 {
685 return (!refcount_is_zero(&ds->ds_longholds));
686 }
687
688 void
689 dsl_dataset_name(dsl_dataset_t *ds, char *name)
690 {
691 if (ds == NULL) {
692 (void) strcpy(name, "mos");
693 } else {
694 dsl_dir_name(ds->ds_dir, name);
695 VERIFY0(dsl_dataset_get_snapname(ds));
696 if (ds->ds_snapname[0]) {
697 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
698 <, ZFS_MAX_DATASET_NAME_LEN);
699 /*
700 * We use a "recursive" mutex so that we
701 * can call dprintf_ds() with ds_lock held.
702 */
703 if (!MUTEX_HELD(&ds->ds_lock)) {
704 mutex_enter(&ds->ds_lock);
705 VERIFY3U(strlcat(name, ds->ds_snapname,
706 ZFS_MAX_DATASET_NAME_LEN), <,
707 ZFS_MAX_DATASET_NAME_LEN);
708 mutex_exit(&ds->ds_lock);
709 } else {
710 VERIFY3U(strlcat(name, ds->ds_snapname,
711 ZFS_MAX_DATASET_NAME_LEN), <,
712 ZFS_MAX_DATASET_NAME_LEN);
713 }
714 }
715 }
716 }
717
718 int
719 dsl_dataset_namelen(dsl_dataset_t *ds)
720 {
721 VERIFY0(dsl_dataset_get_snapname(ds));
722 mutex_enter(&ds->ds_lock);
723 int len = strlen(ds->ds_snapname);
724 /* add '@' if ds is a snap */
725 if (len > 0)
726 len++;
727 len += dsl_dir_namelen(ds->ds_dir);
728 mutex_exit(&ds->ds_lock);
729 return (len);
730 }
731
732 void
733 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
734 {
735 if (ds->ds_dir != NULL && ds->ds_dir->dd_crypto_obj != 0 &&
736 (flags & DS_HOLD_FLAG_DECRYPT)) {
737 (void) spa_keystore_remove_mapping(ds->ds_dir->dd_pool->dp_spa,
738 ds->ds_object, ds);
739 }
740
741 dmu_buf_rele(ds->ds_dbuf, tag);
742 }
743
744 void
745 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
746 {
747 dsl_dataset_rele_flags(ds, 0, tag);
748 }
749
750 void
751 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
752 {
753 ASSERT3P(ds->ds_owner, ==, tag);
754 ASSERT(ds->ds_dbuf != NULL);
755
756 mutex_enter(&ds->ds_lock);
757 ds->ds_owner = NULL;
758 mutex_exit(&ds->ds_lock);
759 dsl_dataset_long_rele(ds, tag);
760 dsl_dataset_rele_flags(ds, flags, tag);
761 }
762
763 boolean_t
764 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
765 {
766 boolean_t gotit = FALSE;
767
768 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
769 mutex_enter(&ds->ds_lock);
770 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
771 ds->ds_owner = tag;
772 dsl_dataset_long_hold(ds, tag);
773 gotit = TRUE;
774 }
775 mutex_exit(&ds->ds_lock);
776 return (gotit);
777 }
778
779 boolean_t
780 dsl_dataset_has_owner(dsl_dataset_t *ds)
781 {
782 boolean_t rv;
783 mutex_enter(&ds->ds_lock);
784 rv = (ds->ds_owner != NULL);
785 mutex_exit(&ds->ds_lock);
786 return (rv);
787 }
788
789 void
790 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
791 {
792 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
793 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
794 uint64_t zero = 0;
795
796 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
797
798 spa_feature_incr(spa, f, tx);
799 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
800
801 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
802 sizeof (zero), 1, &zero, tx));
803 }
804
805 void
806 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
807 {
808 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
809 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
810
811 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
812
813 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
814 spa_feature_decr(spa, f, tx);
815 }
816
817 uint64_t
818 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
819 dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx)
820 {
821 dsl_pool_t *dp = dd->dd_pool;
822 dmu_buf_t *dbuf;
823 dsl_dataset_phys_t *dsphys;
824 uint64_t dsobj;
825 objset_t *mos = dp->dp_meta_objset;
826
827 if (origin == NULL)
828 origin = dp->dp_origin_snap;
829
830 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
831 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
832 ASSERT(dmu_tx_is_syncing(tx));
833 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
834
835 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
836 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
837 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
838 dmu_buf_will_dirty(dbuf, tx);
839 dsphys = dbuf->db_data;
840 bzero(dsphys, sizeof (dsl_dataset_phys_t));
841 dsphys->ds_dir_obj = dd->dd_object;
842 dsphys->ds_flags = flags;
843 dsphys->ds_fsid_guid = unique_create();
844 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
845 sizeof (dsphys->ds_guid));
846 dsphys->ds_snapnames_zapobj =
847 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
848 DMU_OT_NONE, 0, tx);
849 dsphys->ds_creation_time = gethrestime_sec();
850 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
851
852 if (origin == NULL) {
853 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
854 } else {
855 dsl_dataset_t *ohds; /* head of the origin snapshot */
856
857 dsphys->ds_prev_snap_obj = origin->ds_object;
858 dsphys->ds_prev_snap_txg =
859 dsl_dataset_phys(origin)->ds_creation_txg;
860 dsphys->ds_referenced_bytes =
861 dsl_dataset_phys(origin)->ds_referenced_bytes;
862 dsphys->ds_compressed_bytes =
863 dsl_dataset_phys(origin)->ds_compressed_bytes;
864 dsphys->ds_uncompressed_bytes =
865 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
866 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
867 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
868 rrw_exit(&origin->ds_bp_rwlock, FTAG);
869
870 /*
871 * Inherit flags that describe the dataset's contents
872 * (INCONSISTENT) or properties (Case Insensitive).
873 */
874 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
875 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
876
877 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
878 if (origin->ds_feature_inuse[f])
879 dsl_dataset_activate_feature(dsobj, f, tx);
880 }
881
882 dmu_buf_will_dirty(origin->ds_dbuf, tx);
883 dsl_dataset_phys(origin)->ds_num_children++;
884
885 VERIFY0(dsl_dataset_hold_obj(dp,
886 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
887 FTAG, &ohds));
888 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
889 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
890 dsl_dataset_rele(ohds, FTAG);
891
892 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
893 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
894 dsl_dataset_phys(origin)->ds_next_clones_obj =
895 zap_create(mos,
896 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
897 }
898 VERIFY0(zap_add_int(mos,
899 dsl_dataset_phys(origin)->ds_next_clones_obj,
900 dsobj, tx));
901 }
902
903 dmu_buf_will_dirty(dd->dd_dbuf, tx);
904 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
905 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
906 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
907 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
908 dsl_dir_phys(origin->ds_dir)->dd_clones =
909 zap_create(mos,
910 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
911 }
912 VERIFY0(zap_add_int(mos,
913 dsl_dir_phys(origin->ds_dir)->dd_clones,
914 dsobj, tx));
915 }
916 }
917
918 /* handle encryption */
919 dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx);
920
921 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
922 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
923
924 dmu_buf_rele(dbuf, FTAG);
925
926 dmu_buf_will_dirty(dd->dd_dbuf, tx);
927 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
928
929 return (dsobj);
930 }
931
932 static void
933 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
934 {
935 objset_t *os;
936
937 VERIFY0(dmu_objset_from_ds(ds, &os));
938 if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
939 dsl_pool_t *dp = ds->ds_dir->dd_pool;
940 zio_t *zio;
941
942 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
943 if (os->os_encrypted)
944 os->os_next_write_raw = B_TRUE;
945
946 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
947 dsl_dataset_sync(ds, zio, tx);
948 VERIFY0(zio_wait(zio));
949
950 /* dsl_dataset_sync_done will drop this reference. */
951 dmu_buf_add_ref(ds->ds_dbuf, ds);
952 dsl_dataset_sync_done(ds, tx);
953 }
954 }
955
956 uint64_t
957 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
958 dsl_dataset_t *origin, uint64_t flags, cred_t *cr,
959 dsl_crypto_params_t *dcp, dmu_tx_t *tx)
960 {
961 dsl_pool_t *dp = pdd->dd_pool;
962 uint64_t dsobj, ddobj;
963 dsl_dir_t *dd;
964
965 ASSERT(dmu_tx_is_syncing(tx));
966 ASSERT(lastname[0] != '@');
967
968 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
969 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
970
971 dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp,
972 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
973
974 dsl_deleg_set_create_perms(dd, tx, cr);
975
976 /*
977 * Since we're creating a new node we know it's a leaf, so we can
978 * initialize the counts if the limit feature is active.
979 */
980 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
981 uint64_t cnt = 0;
982 objset_t *os = dd->dd_pool->dp_meta_objset;
983
984 dsl_dir_zapify(dd, tx);
985 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
986 sizeof (cnt), 1, &cnt, tx));
987 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
988 sizeof (cnt), 1, &cnt, tx));
989 }
990
991 dsl_dir_rele(dd, FTAG);
992
993 /*
994 * If we are creating a clone, make sure we zero out any stale
995 * data from the origin snapshots zil header.
996 */
997 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
998 dsl_dataset_t *ds;
999
1000 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1001 dsl_dataset_zero_zil(ds, tx);
1002 dsl_dataset_rele(ds, FTAG);
1003 }
1004
1005 return (dsobj);
1006 }
1007
1008 /*
1009 * The unique space in the head dataset can be calculated by subtracting
1010 * the space used in the most recent snapshot, that is still being used
1011 * in this file system, from the space currently in use. To figure out
1012 * the space in the most recent snapshot still in use, we need to take
1013 * the total space used in the snapshot and subtract out the space that
1014 * has been freed up since the snapshot was taken.
1015 */
1016 void
1017 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1018 {
1019 uint64_t mrs_used;
1020 uint64_t dlused, dlcomp, dluncomp;
1021
1022 ASSERT(!ds->ds_is_snapshot);
1023
1024 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1025 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1026 else
1027 mrs_used = 0;
1028
1029 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1030
1031 ASSERT3U(dlused, <=, mrs_used);
1032 dsl_dataset_phys(ds)->ds_unique_bytes =
1033 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1034
1035 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1036 SPA_VERSION_UNIQUE_ACCURATE)
1037 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1038 }
1039
1040 void
1041 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1042 dmu_tx_t *tx)
1043 {
1044 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1045 ASSERTV(uint64_t count);
1046 int err;
1047
1048 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1049 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1050 obj, tx);
1051 /*
1052 * The err should not be ENOENT, but a bug in a previous version
1053 * of the code could cause upgrade_clones_cb() to not set
1054 * ds_next_snap_obj when it should, leading to a missing entry.
1055 * If we knew that the pool was created after
1056 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1057 * ENOENT. However, at least we can check that we don't have
1058 * too many entries in the next_clones_obj even after failing to
1059 * remove this one.
1060 */
1061 if (err != ENOENT)
1062 VERIFY0(err);
1063 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1064 &count));
1065 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1066 }
1067
1068
1069 blkptr_t *
1070 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1071 {
1072 return (&dsl_dataset_phys(ds)->ds_bp);
1073 }
1074
1075 spa_t *
1076 dsl_dataset_get_spa(dsl_dataset_t *ds)
1077 {
1078 return (ds->ds_dir->dd_pool->dp_spa);
1079 }
1080
1081 void
1082 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1083 {
1084 dsl_pool_t *dp;
1085
1086 if (ds == NULL) /* this is the meta-objset */
1087 return;
1088
1089 ASSERT(ds->ds_objset != NULL);
1090
1091 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1092 panic("dirtying snapshot!");
1093
1094 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1095 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1096
1097 dp = ds->ds_dir->dd_pool;
1098 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1099 /* up the hold count until we can be written out */
1100 dmu_buf_add_ref(ds->ds_dbuf, ds);
1101 }
1102 }
1103
1104 boolean_t
1105 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1106 {
1107 for (int t = 0; t < TXG_SIZE; t++) {
1108 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1109 ds, t))
1110 return (B_TRUE);
1111 }
1112 return (B_FALSE);
1113 }
1114
1115 static int
1116 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1117 {
1118 uint64_t asize;
1119
1120 if (!dmu_tx_is_syncing(tx))
1121 return (0);
1122
1123 /*
1124 * If there's an fs-only reservation, any blocks that might become
1125 * owned by the snapshot dataset must be accommodated by space
1126 * outside of the reservation.
1127 */
1128 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1129 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1130 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1131 return (SET_ERROR(ENOSPC));
1132
1133 /*
1134 * Propagate any reserved space for this snapshot to other
1135 * snapshot checks in this sync group.
1136 */
1137 if (asize > 0)
1138 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1139
1140 return (0);
1141 }
1142
1143 typedef struct dsl_dataset_snapshot_arg {
1144 nvlist_t *ddsa_snaps;
1145 nvlist_t *ddsa_props;
1146 nvlist_t *ddsa_errors;
1147 cred_t *ddsa_cr;
1148 } dsl_dataset_snapshot_arg_t;
1149
1150 int
1151 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1152 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1153 {
1154 int error;
1155 uint64_t value;
1156
1157 ds->ds_trysnap_txg = tx->tx_txg;
1158
1159 if (!dmu_tx_is_syncing(tx))
1160 return (0);
1161
1162 /*
1163 * We don't allow multiple snapshots of the same txg. If there
1164 * is already one, try again.
1165 */
1166 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1167 return (SET_ERROR(EAGAIN));
1168
1169 /*
1170 * Check for conflicting snapshot name.
1171 */
1172 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1173 if (error == 0)
1174 return (SET_ERROR(EEXIST));
1175 if (error != ENOENT)
1176 return (error);
1177
1178 /*
1179 * We don't allow taking snapshots of inconsistent datasets, such as
1180 * those into which we are currently receiving. However, if we are
1181 * creating this snapshot as part of a receive, this check will be
1182 * executed atomically with respect to the completion of the receive
1183 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1184 * case we ignore this, knowing it will be fixed up for us shortly in
1185 * dmu_recv_end_sync().
1186 */
1187 if (!recv && DS_IS_INCONSISTENT(ds))
1188 return (SET_ERROR(EBUSY));
1189
1190 /*
1191 * Skip the check for temporary snapshots or if we have already checked
1192 * the counts in dsl_dataset_snapshot_check. This means we really only
1193 * check the count here when we're receiving a stream.
1194 */
1195 if (cnt != 0 && cr != NULL) {
1196 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1197 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1198 if (error != 0)
1199 return (error);
1200 }
1201
1202 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1203 if (error != 0)
1204 return (error);
1205
1206 return (0);
1207 }
1208
1209 static int
1210 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1211 {
1212 dsl_dataset_snapshot_arg_t *ddsa = arg;
1213 dsl_pool_t *dp = dmu_tx_pool(tx);
1214 nvpair_t *pair;
1215 int rv = 0;
1216
1217 /*
1218 * Pre-compute how many total new snapshots will be created for each
1219 * level in the tree and below. This is needed for validating the
1220 * snapshot limit when either taking a recursive snapshot or when
1221 * taking multiple snapshots.
1222 *
1223 * The problem is that the counts are not actually adjusted when
1224 * we are checking, only when we finally sync. For a single snapshot,
1225 * this is easy, the count will increase by 1 at each node up the tree,
1226 * but its more complicated for the recursive/multiple snapshot case.
1227 *
1228 * The dsl_fs_ss_limit_check function does recursively check the count
1229 * at each level up the tree but since it is validating each snapshot
1230 * independently we need to be sure that we are validating the complete
1231 * count for the entire set of snapshots. We do this by rolling up the
1232 * counts for each component of the name into an nvlist and then
1233 * checking each of those cases with the aggregated count.
1234 *
1235 * This approach properly handles not only the recursive snapshot
1236 * case (where we get all of those on the ddsa_snaps list) but also
1237 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1238 * validate the limit on 'a' using a count of 2).
1239 *
1240 * We validate the snapshot names in the third loop and only report
1241 * name errors once.
1242 */
1243 if (dmu_tx_is_syncing(tx)) {
1244 char *nm;
1245 nvlist_t *cnt_track = NULL;
1246 cnt_track = fnvlist_alloc();
1247
1248 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1249
1250 /* Rollup aggregated counts into the cnt_track list */
1251 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1252 pair != NULL;
1253 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1254 char *pdelim;
1255 uint64_t val;
1256
1257 (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1258 pdelim = strchr(nm, '@');
1259 if (pdelim == NULL)
1260 continue;
1261 *pdelim = '\0';
1262
1263 do {
1264 if (nvlist_lookup_uint64(cnt_track, nm,
1265 &val) == 0) {
1266 /* update existing entry */
1267 fnvlist_add_uint64(cnt_track, nm,
1268 val + 1);
1269 } else {
1270 /* add to list */
1271 fnvlist_add_uint64(cnt_track, nm, 1);
1272 }
1273
1274 pdelim = strrchr(nm, '/');
1275 if (pdelim != NULL)
1276 *pdelim = '\0';
1277 } while (pdelim != NULL);
1278 }
1279
1280 kmem_free(nm, MAXPATHLEN);
1281
1282 /* Check aggregated counts at each level */
1283 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1284 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1285 int error = 0;
1286 char *name;
1287 uint64_t cnt = 0;
1288 dsl_dataset_t *ds;
1289
1290 name = nvpair_name(pair);
1291 cnt = fnvpair_value_uint64(pair);
1292 ASSERT(cnt > 0);
1293
1294 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1295 if (error == 0) {
1296 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1297 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1298 ddsa->ddsa_cr);
1299 dsl_dataset_rele(ds, FTAG);
1300 }
1301
1302 if (error != 0) {
1303 if (ddsa->ddsa_errors != NULL)
1304 fnvlist_add_int32(ddsa->ddsa_errors,
1305 name, error);
1306 rv = error;
1307 /* only report one error for this check */
1308 break;
1309 }
1310 }
1311 nvlist_free(cnt_track);
1312 }
1313
1314 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1315 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1316 int error = 0;
1317 dsl_dataset_t *ds;
1318 char *name, *atp = NULL;
1319 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1320
1321 name = nvpair_name(pair);
1322 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1323 error = SET_ERROR(ENAMETOOLONG);
1324 if (error == 0) {
1325 atp = strchr(name, '@');
1326 if (atp == NULL)
1327 error = SET_ERROR(EINVAL);
1328 if (error == 0)
1329 (void) strlcpy(dsname, name, atp - name + 1);
1330 }
1331 if (error == 0)
1332 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1333 if (error == 0) {
1334 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1335 error = dsl_dataset_snapshot_check_impl(ds,
1336 atp + 1, tx, B_FALSE, 0, NULL);
1337 dsl_dataset_rele(ds, FTAG);
1338 }
1339
1340 if (error != 0) {
1341 if (ddsa->ddsa_errors != NULL) {
1342 fnvlist_add_int32(ddsa->ddsa_errors,
1343 name, error);
1344 }
1345 rv = error;
1346 }
1347 }
1348
1349 return (rv);
1350 }
1351
1352 void
1353 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1354 dmu_tx_t *tx)
1355 {
1356 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1357 dmu_buf_t *dbuf;
1358 dsl_dataset_phys_t *dsphys;
1359 uint64_t dsobj, crtxg;
1360 objset_t *mos = dp->dp_meta_objset;
1361 ASSERTV(static zil_header_t zero_zil);
1362 ASSERTV(objset_t *os);
1363
1364 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1365
1366 /*
1367 * If we are on an old pool, the zil must not be active, in which
1368 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1369 */
1370 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1371 dmu_objset_from_ds(ds, &os) != 0 ||
1372 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1373 sizeof (zero_zil)) == 0);
1374
1375 /* Should not snapshot a dirty dataset. */
1376 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1377 ds, tx->tx_txg));
1378
1379 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1380
1381 /*
1382 * The origin's ds_creation_txg has to be < TXG_INITIAL
1383 */
1384 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1385 crtxg = 1;
1386 else
1387 crtxg = tx->tx_txg;
1388
1389 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1390 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1391 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1392 dmu_buf_will_dirty(dbuf, tx);
1393 dsphys = dbuf->db_data;
1394 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1395 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1396 dsphys->ds_fsid_guid = unique_create();
1397 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1398 sizeof (dsphys->ds_guid));
1399 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1400 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1401 dsphys->ds_next_snap_obj = ds->ds_object;
1402 dsphys->ds_num_children = 1;
1403 dsphys->ds_creation_time = gethrestime_sec();
1404 dsphys->ds_creation_txg = crtxg;
1405 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1406 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1407 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1408 dsphys->ds_uncompressed_bytes =
1409 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1410 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1411 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1412 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1413 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1414 dmu_buf_rele(dbuf, FTAG);
1415
1416 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1417 if (ds->ds_feature_inuse[f])
1418 dsl_dataset_activate_feature(dsobj, f, tx);
1419 }
1420
1421 ASSERT3U(ds->ds_prev != 0, ==,
1422 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1423 if (ds->ds_prev) {
1424 uint64_t next_clones_obj =
1425 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1426 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1427 ds->ds_object ||
1428 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1429 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1430 ds->ds_object) {
1431 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1432 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1433 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1434 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1435 } else if (next_clones_obj != 0) {
1436 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1437 dsphys->ds_next_snap_obj, tx);
1438 VERIFY0(zap_add_int(mos,
1439 next_clones_obj, dsobj, tx));
1440 }
1441 }
1442
1443 /*
1444 * If we have a reference-reservation on this dataset, we will
1445 * need to increase the amount of refreservation being charged
1446 * since our unique space is going to zero.
1447 */
1448 if (ds->ds_reserved) {
1449 int64_t delta;
1450 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1451 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1452 ds->ds_reserved);
1453 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1454 delta, 0, 0, tx);
1455 }
1456
1457 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1458 dsl_dataset_phys(ds)->ds_deadlist_obj =
1459 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1460 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1461 dsl_deadlist_close(&ds->ds_deadlist);
1462 dsl_deadlist_open(&ds->ds_deadlist, mos,
1463 dsl_dataset_phys(ds)->ds_deadlist_obj);
1464 dsl_deadlist_add_key(&ds->ds_deadlist,
1465 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1466
1467 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1468 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1469 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1470 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1471 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1472 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1473
1474 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1475 snapname, 8, 1, &dsobj, tx));
1476
1477 if (ds->ds_prev)
1478 dsl_dataset_rele(ds->ds_prev, ds);
1479 VERIFY0(dsl_dataset_hold_obj(dp,
1480 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1481
1482 dsl_scan_ds_snapshotted(ds, tx);
1483
1484 dsl_dir_snap_cmtime_update(ds->ds_dir);
1485
1486 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1487 }
1488
1489 static void
1490 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1491 {
1492 dsl_dataset_snapshot_arg_t *ddsa = arg;
1493 dsl_pool_t *dp = dmu_tx_pool(tx);
1494 nvpair_t *pair;
1495
1496 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1497 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1498 dsl_dataset_t *ds;
1499 char *name, *atp;
1500 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1501
1502 name = nvpair_name(pair);
1503 atp = strchr(name, '@');
1504 (void) strlcpy(dsname, name, atp - name + 1);
1505 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1506
1507 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1508 if (ddsa->ddsa_props != NULL) {
1509 dsl_props_set_sync_impl(ds->ds_prev,
1510 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1511 }
1512 zvol_create_minors(dp->dp_spa, nvpair_name(pair), B_TRUE);
1513 dsl_dataset_rele(ds, FTAG);
1514 }
1515 }
1516
1517 /*
1518 * The snapshots must all be in the same pool.
1519 * All-or-nothing: if there are any failures, nothing will be modified.
1520 */
1521 int
1522 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1523 {
1524 dsl_dataset_snapshot_arg_t ddsa;
1525 nvpair_t *pair;
1526 boolean_t needsuspend;
1527 int error;
1528 spa_t *spa;
1529 char *firstname;
1530 nvlist_t *suspended = NULL;
1531
1532 pair = nvlist_next_nvpair(snaps, NULL);
1533 if (pair == NULL)
1534 return (0);
1535 firstname = nvpair_name(pair);
1536
1537 error = spa_open(firstname, &spa, FTAG);
1538 if (error != 0)
1539 return (error);
1540 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1541 spa_close(spa, FTAG);
1542
1543 if (needsuspend) {
1544 suspended = fnvlist_alloc();
1545 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1546 pair = nvlist_next_nvpair(snaps, pair)) {
1547 char fsname[ZFS_MAX_DATASET_NAME_LEN];
1548 char *snapname = nvpair_name(pair);
1549 char *atp;
1550 void *cookie;
1551
1552 atp = strchr(snapname, '@');
1553 if (atp == NULL) {
1554 error = SET_ERROR(EINVAL);
1555 break;
1556 }
1557 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1558
1559 error = zil_suspend(fsname, &cookie);
1560 if (error != 0)
1561 break;
1562 fnvlist_add_uint64(suspended, fsname,
1563 (uintptr_t)cookie);
1564 }
1565 }
1566
1567 ddsa.ddsa_snaps = snaps;
1568 ddsa.ddsa_props = props;
1569 ddsa.ddsa_errors = errors;
1570 ddsa.ddsa_cr = CRED();
1571
1572 if (error == 0) {
1573 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1574 dsl_dataset_snapshot_sync, &ddsa,
1575 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1576 }
1577
1578 if (suspended != NULL) {
1579 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1580 pair = nvlist_next_nvpair(suspended, pair)) {
1581 zil_resume((void *)(uintptr_t)
1582 fnvpair_value_uint64(pair));
1583 }
1584 fnvlist_free(suspended);
1585 }
1586
1587 return (error);
1588 }
1589
1590 typedef struct dsl_dataset_snapshot_tmp_arg {
1591 const char *ddsta_fsname;
1592 const char *ddsta_snapname;
1593 minor_t ddsta_cleanup_minor;
1594 const char *ddsta_htag;
1595 } dsl_dataset_snapshot_tmp_arg_t;
1596
1597 static int
1598 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1599 {
1600 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1601 dsl_pool_t *dp = dmu_tx_pool(tx);
1602 dsl_dataset_t *ds;
1603 int error;
1604
1605 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1606 if (error != 0)
1607 return (error);
1608
1609 /* NULL cred means no limit check for tmp snapshot */
1610 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1611 tx, B_FALSE, 0, NULL);
1612 if (error != 0) {
1613 dsl_dataset_rele(ds, FTAG);
1614 return (error);
1615 }
1616
1617 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1618 dsl_dataset_rele(ds, FTAG);
1619 return (SET_ERROR(ENOTSUP));
1620 }
1621 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1622 B_TRUE, tx);
1623 if (error != 0) {
1624 dsl_dataset_rele(ds, FTAG);
1625 return (error);
1626 }
1627
1628 dsl_dataset_rele(ds, FTAG);
1629 return (0);
1630 }
1631
1632 static void
1633 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1634 {
1635 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1636 dsl_pool_t *dp = dmu_tx_pool(tx);
1637 dsl_dataset_t *ds = NULL;
1638
1639 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1640
1641 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1642 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1643 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1644 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1645
1646 dsl_dataset_rele(ds, FTAG);
1647 }
1648
1649 int
1650 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1651 minor_t cleanup_minor, const char *htag)
1652 {
1653 dsl_dataset_snapshot_tmp_arg_t ddsta;
1654 int error;
1655 spa_t *spa;
1656 boolean_t needsuspend;
1657 void *cookie;
1658
1659 ddsta.ddsta_fsname = fsname;
1660 ddsta.ddsta_snapname = snapname;
1661 ddsta.ddsta_cleanup_minor = cleanup_minor;
1662 ddsta.ddsta_htag = htag;
1663
1664 error = spa_open(fsname, &spa, FTAG);
1665 if (error != 0)
1666 return (error);
1667 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1668 spa_close(spa, FTAG);
1669
1670 if (needsuspend) {
1671 error = zil_suspend(fsname, &cookie);
1672 if (error != 0)
1673 return (error);
1674 }
1675
1676 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1677 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1678
1679 if (needsuspend)
1680 zil_resume(cookie);
1681 return (error);
1682 }
1683
1684
1685 void
1686 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1687 {
1688 ASSERT(dmu_tx_is_syncing(tx));
1689 ASSERT(ds->ds_objset != NULL);
1690 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1691
1692 /*
1693 * in case we had to change ds_fsid_guid when we opened it,
1694 * sync it out now.
1695 */
1696 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1697 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1698
1699 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1700 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1701 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1702 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1703 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1704 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1705 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1706 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1707 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1708 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1709 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1710 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1711 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1712 }
1713
1714 dmu_objset_sync(ds->ds_objset, zio, tx);
1715
1716 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1717 if (ds->ds_feature_activation_needed[f]) {
1718 if (ds->ds_feature_inuse[f])
1719 continue;
1720 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1721 ds->ds_feature_inuse[f] = B_TRUE;
1722 }
1723 }
1724 }
1725
1726 static int
1727 deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1728 {
1729 dsl_deadlist_t *dl = arg;
1730 dsl_deadlist_insert(dl, bp, tx);
1731 return (0);
1732 }
1733
1734 void
1735 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1736 {
1737 objset_t *os = ds->ds_objset;
1738
1739 bplist_iterate(&ds->ds_pending_deadlist,
1740 deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1741
1742 if (os->os_synced_dnodes != NULL) {
1743 multilist_destroy(os->os_synced_dnodes);
1744 os->os_synced_dnodes = NULL;
1745 }
1746
1747 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1748
1749 dmu_buf_rele(ds->ds_dbuf, ds);
1750 }
1751
1752 static void
1753 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1754 {
1755 uint64_t count = 0;
1756 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1757 zap_cursor_t zc;
1758 zap_attribute_t za;
1759 nvlist_t *propval = fnvlist_alloc();
1760 nvlist_t *val;
1761
1762 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1763
1764 /*
1765 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1766 * latter would allocate the list with NV_UNIQUE_NAME flag.
1767 * As a result, every time a clone name is appended to the list
1768 * it would be (linearly) searched for for a duplicate name.
1769 * We already know that all clone names must be unique and we
1770 * want avoid the quadratic complexity of double-checking that
1771 * because we can have a large number of clones.
1772 */
1773 VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1774
1775 /*
1776 * There may be missing entries in ds_next_clones_obj
1777 * due to a bug in a previous version of the code.
1778 * Only trust it if it has the right number of entries.
1779 */
1780 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1781 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1782 &count));
1783 }
1784 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1785 goto fail;
1786 for (zap_cursor_init(&zc, mos,
1787 dsl_dataset_phys(ds)->ds_next_clones_obj);
1788 zap_cursor_retrieve(&zc, &za) == 0;
1789 zap_cursor_advance(&zc)) {
1790 dsl_dataset_t *clone;
1791 char buf[ZFS_MAX_DATASET_NAME_LEN];
1792 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1793 za.za_first_integer, FTAG, &clone));
1794 dsl_dir_name(clone->ds_dir, buf);
1795 fnvlist_add_boolean(val, buf);
1796 dsl_dataset_rele(clone, FTAG);
1797 }
1798 zap_cursor_fini(&zc);
1799 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1800 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1801 fail:
1802 nvlist_free(val);
1803 nvlist_free(propval);
1804 }
1805
1806 static void
1807 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1808 {
1809 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1810
1811 if (dsl_dataset_has_resume_receive_state(ds)) {
1812 char *str;
1813 void *packed;
1814 uint8_t *compressed;
1815 uint64_t val;
1816 nvlist_t *token_nv = fnvlist_alloc();
1817 size_t packed_size, compressed_size;
1818
1819 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1820 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1821 fnvlist_add_uint64(token_nv, "fromguid", val);
1822 }
1823 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1824 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1825 fnvlist_add_uint64(token_nv, "object", val);
1826 }
1827 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1828 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1829 fnvlist_add_uint64(token_nv, "offset", val);
1830 }
1831 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1832 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1833 fnvlist_add_uint64(token_nv, "bytes", val);
1834 }
1835 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1836 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1837 fnvlist_add_uint64(token_nv, "toguid", val);
1838 }
1839 char buf[MAXNAMELEN];
1840 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1841 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1842 fnvlist_add_string(token_nv, "toname", buf);
1843 }
1844 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1845 DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1846 fnvlist_add_boolean(token_nv, "largeblockok");
1847 }
1848 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1849 DS_FIELD_RESUME_EMBEDOK) == 0) {
1850 fnvlist_add_boolean(token_nv, "embedok");
1851 }
1852 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1853 DS_FIELD_RESUME_COMPRESSOK) == 0) {
1854 fnvlist_add_boolean(token_nv, "compressok");
1855 }
1856 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1857 DS_FIELD_RESUME_RAWOK) == 0) {
1858 fnvlist_add_boolean(token_nv, "rawok");
1859 }
1860 packed = fnvlist_pack(token_nv, &packed_size);
1861 fnvlist_free(token_nv);
1862 compressed = kmem_alloc(packed_size, KM_SLEEP);
1863
1864 compressed_size = gzip_compress(packed, compressed,
1865 packed_size, packed_size, 6);
1866
1867 zio_cksum_t cksum;
1868 fletcher_4_native_varsize(compressed, compressed_size, &cksum);
1869
1870 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1871 for (int i = 0; i < compressed_size; i++) {
1872 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1873 }
1874 str[compressed_size * 2] = '\0';
1875 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1876 ZFS_SEND_RESUME_TOKEN_VERSION,
1877 (longlong_t)cksum.zc_word[0],
1878 (longlong_t)packed_size, str);
1879 dsl_prop_nvlist_add_string(nv,
1880 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1881 kmem_free(packed, packed_size);
1882 kmem_free(str, compressed_size * 2 + 1);
1883 kmem_free(compressed, packed_size);
1884 strfree(propval);
1885 }
1886 }
1887
1888 void
1889 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1890 {
1891 int err;
1892 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1893 uint64_t refd, avail, uobjs, aobjs, ratio;
1894
1895 ASSERT(dsl_pool_config_held(dp));
1896
1897 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1898 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1899 dsl_dataset_phys(ds)->ds_compressed_bytes);
1900
1901 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1902 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1903 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1904
1905 if (ds->ds_is_snapshot) {
1906 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1907 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1908 dsl_dataset_phys(ds)->ds_unique_bytes);
1909 get_clones_stat(ds, nv);
1910 } else {
1911 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1912 char buf[ZFS_MAX_DATASET_NAME_LEN];
1913 dsl_dataset_name(ds->ds_prev, buf);
1914 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1915 }
1916
1917 dsl_dir_stats(ds->ds_dir, nv);
1918 }
1919
1920 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1921 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1922 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1923
1924 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1925 dsl_dataset_phys(ds)->ds_creation_time);
1926 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1927 dsl_dataset_phys(ds)->ds_creation_txg);
1928 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1929 ds->ds_quota);
1930 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1931 ds->ds_reserved);
1932 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1933 dsl_dataset_phys(ds)->ds_guid);
1934 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1935 dsl_dataset_phys(ds)->ds_unique_bytes);
1936 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1937 ds->ds_object);
1938 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1939 ds->ds_userrefs);
1940 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1941 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1942 dsl_dataset_crypt_stats(ds, nv);
1943
1944 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1945 uint64_t written, comp, uncomp;
1946 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1947 dsl_dataset_t *prev;
1948
1949 err = dsl_dataset_hold_obj(dp,
1950 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1951 if (err == 0) {
1952 err = dsl_dataset_space_written(prev, ds, &written,
1953 &comp, &uncomp);
1954 dsl_dataset_rele(prev, FTAG);
1955 if (err == 0) {
1956 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1957 written);
1958 }
1959 }
1960 }
1961
1962 if (!dsl_dataset_is_snapshot(ds)) {
1963 /*
1964 * A failed "newfs" (e.g. full) resumable receive leaves
1965 * the stats set on this dataset. Check here for the prop.
1966 */
1967 get_receive_resume_stats(ds, nv);
1968
1969 /*
1970 * A failed incremental resumable receive leaves the
1971 * stats set on our child named "%recv". Check the child
1972 * for the prop.
1973 */
1974 /* 6 extra bytes for /%recv */
1975 char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
1976 dsl_dataset_t *recv_ds;
1977 dsl_dataset_name(ds, recvname);
1978 if (strlcat(recvname, "/", sizeof (recvname)) <
1979 sizeof (recvname) &&
1980 strlcat(recvname, recv_clone_name, sizeof (recvname)) <
1981 sizeof (recvname) &&
1982 dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1983 get_receive_resume_stats(recv_ds, nv);
1984 dsl_dataset_rele(recv_ds, FTAG);
1985 }
1986 }
1987 }
1988
1989 void
1990 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1991 {
1992 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1993 ASSERT(dsl_pool_config_held(dp));
1994
1995 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1996 stat->dds_inconsistent =
1997 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1998 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1999 stat->dds_origin[0] = '\0';
2000 if (ds->ds_is_snapshot) {
2001 stat->dds_is_snapshot = B_TRUE;
2002 stat->dds_num_clones =
2003 dsl_dataset_phys(ds)->ds_num_children - 1;
2004 } else {
2005 stat->dds_is_snapshot = B_FALSE;
2006 stat->dds_num_clones = 0;
2007
2008 if (dsl_dir_is_clone(ds->ds_dir)) {
2009 dsl_dataset_t *ods;
2010
2011 VERIFY0(dsl_dataset_hold_obj(dp,
2012 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
2013 FTAG, &ods));
2014 dsl_dataset_name(ods, stat->dds_origin);
2015 dsl_dataset_rele(ods, FTAG);
2016 }
2017 }
2018 }
2019
2020 uint64_t
2021 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2022 {
2023 return (ds->ds_fsid_guid);
2024 }
2025
2026 void
2027 dsl_dataset_space(dsl_dataset_t *ds,
2028 uint64_t *refdbytesp, uint64_t *availbytesp,
2029 uint64_t *usedobjsp, uint64_t *availobjsp)
2030 {
2031 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2032 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2033 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2034 *availbytesp +=
2035 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2036 if (ds->ds_quota != 0) {
2037 /*
2038 * Adjust available bytes according to refquota
2039 */
2040 if (*refdbytesp < ds->ds_quota)
2041 *availbytesp = MIN(*availbytesp,
2042 ds->ds_quota - *refdbytesp);
2043 else
2044 *availbytesp = 0;
2045 }
2046 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2047 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2048 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2049 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2050 }
2051
2052 boolean_t
2053 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2054 {
2055 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
2056 uint64_t birth;
2057
2058 ASSERT(dsl_pool_config_held(dp));
2059 if (snap == NULL)
2060 return (B_FALSE);
2061 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2062 birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2063 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2064 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2065 objset_t *os, *os_snap;
2066 /*
2067 * It may be that only the ZIL differs, because it was
2068 * reset in the head. Don't count that as being
2069 * modified.
2070 */
2071 if (dmu_objset_from_ds(ds, &os) != 0)
2072 return (B_TRUE);
2073 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2074 return (B_TRUE);
2075 return (bcmp(&os->os_phys->os_meta_dnode,
2076 &os_snap->os_phys->os_meta_dnode,
2077 sizeof (os->os_phys->os_meta_dnode)) != 0);
2078 }
2079 return (B_FALSE);
2080 }
2081
2082 typedef struct dsl_dataset_rename_snapshot_arg {
2083 const char *ddrsa_fsname;
2084 const char *ddrsa_oldsnapname;
2085 const char *ddrsa_newsnapname;
2086 boolean_t ddrsa_recursive;
2087 dmu_tx_t *ddrsa_tx;
2088 } dsl_dataset_rename_snapshot_arg_t;
2089
2090 /* ARGSUSED */
2091 static int
2092 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2093 dsl_dataset_t *hds, void *arg)
2094 {
2095 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2096 int error;
2097 uint64_t val;
2098
2099 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2100 if (error != 0) {
2101 /* ignore nonexistent snapshots */
2102 return (error == ENOENT ? 0 : error);
2103 }
2104
2105 /* new name should not exist */
2106 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2107 if (error == 0)
2108 error = SET_ERROR(EEXIST);
2109 else if (error == ENOENT)
2110 error = 0;
2111
2112 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2113 if (dsl_dir_namelen(hds->ds_dir) + 1 +
2114 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2115 error = SET_ERROR(ENAMETOOLONG);
2116
2117 return (error);
2118 }
2119
2120 static int
2121 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2122 {
2123 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2124 dsl_pool_t *dp = dmu_tx_pool(tx);
2125 dsl_dataset_t *hds;
2126 int error;
2127
2128 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2129 if (error != 0)
2130 return (error);
2131
2132 if (ddrsa->ddrsa_recursive) {
2133 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2134 dsl_dataset_rename_snapshot_check_impl, ddrsa,
2135 DS_FIND_CHILDREN);
2136 } else {
2137 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2138 }
2139 dsl_dataset_rele(hds, FTAG);
2140 return (error);
2141 }
2142
2143 static int
2144 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2145 dsl_dataset_t *hds, void *arg)
2146 {
2147 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2148 dsl_dataset_t *ds;
2149 uint64_t val;
2150 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2151 int error;
2152
2153 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2154 ASSERT(error == 0 || error == ENOENT);
2155 if (error == ENOENT) {
2156 /* ignore nonexistent snapshots */
2157 return (0);
2158 }
2159
2160 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2161
2162 /* log before we change the name */
2163 spa_history_log_internal_ds(ds, "rename", tx,
2164 "-> @%s", ddrsa->ddrsa_newsnapname);
2165
2166 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2167 B_FALSE));
2168 mutex_enter(&ds->ds_lock);
2169 (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname,
2170 sizeof (ds->ds_snapname));
2171 mutex_exit(&ds->ds_lock);
2172 VERIFY0(zap_add(dp->dp_meta_objset,
2173 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2174 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2175 zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname,
2176 ddrsa->ddrsa_newsnapname, B_TRUE);
2177
2178 dsl_dataset_rele(ds, FTAG);
2179 return (0);
2180 }
2181
2182 static void
2183 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2184 {
2185 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2186 dsl_pool_t *dp = dmu_tx_pool(tx);
2187 dsl_dataset_t *hds = NULL;
2188
2189 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2190 ddrsa->ddrsa_tx = tx;
2191 if (ddrsa->ddrsa_recursive) {
2192 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2193 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2194 DS_FIND_CHILDREN));
2195 } else {
2196 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2197 }
2198 dsl_dataset_rele(hds, FTAG);
2199 }
2200
2201 int
2202 dsl_dataset_rename_snapshot(const char *fsname,
2203 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2204 {
2205 dsl_dataset_rename_snapshot_arg_t ddrsa;
2206
2207 ddrsa.ddrsa_fsname = fsname;
2208 ddrsa.ddrsa_oldsnapname = oldsnapname;
2209 ddrsa.ddrsa_newsnapname = newsnapname;
2210 ddrsa.ddrsa_recursive = recursive;
2211
2212 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2213 dsl_dataset_rename_snapshot_sync, &ddrsa,
2214 1, ZFS_SPACE_CHECK_RESERVED));
2215 }
2216
2217 /*
2218 * If we're doing an ownership handoff, we need to make sure that there is
2219 * only one long hold on the dataset. We're not allowed to change anything here
2220 * so we don't permanently release the long hold or regular hold here. We want
2221 * to do this only when syncing to avoid the dataset unexpectedly going away
2222 * when we release the long hold.
2223 */
2224 static int
2225 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2226 {
2227 boolean_t held;
2228
2229 if (!dmu_tx_is_syncing(tx))
2230 return (0);
2231
2232 if (owner != NULL) {
2233 VERIFY3P(ds->ds_owner, ==, owner);
2234 dsl_dataset_long_rele(ds, owner);
2235 }
2236
2237 held = dsl_dataset_long_held(ds);
2238
2239 if (owner != NULL)
2240 dsl_dataset_long_hold(ds, owner);
2241
2242 if (held)
2243 return (SET_ERROR(EBUSY));
2244
2245 return (0);
2246 }
2247
2248 typedef struct dsl_dataset_rollback_arg {
2249 const char *ddra_fsname;
2250 const char *ddra_tosnap;
2251 void *ddra_owner;
2252 nvlist_t *ddra_result;
2253 } dsl_dataset_rollback_arg_t;
2254
2255 static int
2256 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2257 {
2258 dsl_dataset_rollback_arg_t *ddra = arg;
2259 dsl_pool_t *dp = dmu_tx_pool(tx);
2260 dsl_dataset_t *ds;
2261 int64_t unused_refres_delta;
2262 int error;
2263
2264 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2265 if (error != 0)
2266 return (error);
2267
2268 /* must not be a snapshot */
2269 if (ds->ds_is_snapshot) {
2270 dsl_dataset_rele(ds, FTAG);
2271 return (SET_ERROR(EINVAL));
2272 }
2273
2274 /* must have a most recent snapshot */
2275 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2276 dsl_dataset_rele(ds, FTAG);
2277 return (SET_ERROR(EINVAL));
2278 }
2279
2280 /*
2281 * No rollback to a snapshot created in the current txg, because
2282 * the rollback may dirty the dataset and create blocks that are
2283 * not reachable from the rootbp while having a birth txg that
2284 * falls into the snapshot's range.
2285 */
2286 if (dmu_tx_is_syncing(tx) &&
2287 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2288 dsl_dataset_rele(ds, FTAG);
2289 return (SET_ERROR(EAGAIN));
2290 }
2291
2292 /*
2293 * If the expected target snapshot is specified, then check that
2294 * the latest snapshot is it.
2295 */
2296 if (ddra->ddra_tosnap != NULL) {
2297 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2298
2299 dsl_dataset_name(ds->ds_prev, namebuf);
2300 if (strcmp(namebuf, ddra->ddra_tosnap) != 0)
2301 return (SET_ERROR(EXDEV));
2302 }
2303
2304 /* must not have any bookmarks after the most recent snapshot */
2305 nvlist_t *proprequest = fnvlist_alloc();
2306 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2307 nvlist_t *bookmarks = fnvlist_alloc();
2308 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2309 fnvlist_free(proprequest);
2310 if (error != 0)
2311 return (error);
2312 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2313 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2314 nvlist_t *valuenv =
2315 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2316 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2317 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2318 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2319 fnvlist_free(bookmarks);
2320 dsl_dataset_rele(ds, FTAG);
2321 return (SET_ERROR(EEXIST));
2322 }
2323 }
2324 fnvlist_free(bookmarks);
2325
2326 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2327 if (error != 0) {
2328 dsl_dataset_rele(ds, FTAG);
2329 return (error);
2330 }
2331
2332 /*
2333 * Check if the snap we are rolling back to uses more than
2334 * the refquota.
2335 */
2336 if (ds->ds_quota != 0 &&
2337 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2338 dsl_dataset_rele(ds, FTAG);
2339 return (SET_ERROR(EDQUOT));
2340 }
2341
2342 /*
2343 * When we do the clone swap, we will temporarily use more space
2344 * due to the refreservation (the head will no longer have any
2345 * unique space, so the entire amount of the refreservation will need
2346 * to be free). We will immediately destroy the clone, freeing
2347 * this space, but the freeing happens over many txg's.
2348 */
2349 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2350 dsl_dataset_phys(ds)->ds_unique_bytes);
2351
2352 if (unused_refres_delta > 0 &&
2353 unused_refres_delta >
2354 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2355 dsl_dataset_rele(ds, FTAG);
2356 return (SET_ERROR(ENOSPC));
2357 }
2358
2359 dsl_dataset_rele(ds, FTAG);
2360 return (0);
2361 }
2362
2363 static void
2364 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2365 {
2366 dsl_dataset_rollback_arg_t *ddra = arg;
2367 dsl_pool_t *dp = dmu_tx_pool(tx);
2368 dsl_dataset_t *ds, *clone;
2369 uint64_t cloneobj;
2370 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2371
2372 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2373
2374 dsl_dataset_name(ds->ds_prev, namebuf);
2375 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2376
2377 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2378 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
2379
2380 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2381
2382 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2383 dsl_dataset_zero_zil(ds, tx);
2384
2385 dsl_destroy_head_sync_impl(clone, tx);
2386
2387 dsl_dataset_rele(clone, FTAG);
2388 dsl_dataset_rele(ds, FTAG);
2389 }
2390
2391 /*
2392 * Rolls back the given filesystem or volume to the most recent snapshot.
2393 * The name of the most recent snapshot will be returned under key "target"
2394 * in the result nvlist.
2395 *
2396 * If owner != NULL:
2397 * - The existing dataset MUST be owned by the specified owner at entry
2398 * - Upon return, dataset will still be held by the same owner, whether we
2399 * succeed or not.
2400 *
2401 * This mode is required any time the existing filesystem is mounted. See
2402 * notes above zfs_suspend_fs() for further details.
2403 */
2404 int
2405 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2406 nvlist_t *result)
2407 {
2408 dsl_dataset_rollback_arg_t ddra;
2409
2410 ddra.ddra_fsname = fsname;
2411 ddra.ddra_tosnap = tosnap;
2412 ddra.ddra_owner = owner;
2413 ddra.ddra_result = result;
2414
2415 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2416 dsl_dataset_rollback_sync, &ddra,
2417 1, ZFS_SPACE_CHECK_RESERVED));
2418 }
2419
2420 struct promotenode {
2421 list_node_t link;
2422 dsl_dataset_t *ds;
2423 };
2424
2425 typedef struct dsl_dataset_promote_arg {
2426 const char *ddpa_clonename;
2427 dsl_dataset_t *ddpa_clone;
2428 list_t shared_snaps, origin_snaps, clone_snaps;
2429 dsl_dataset_t *origin_origin; /* origin of the origin */
2430 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2431 char *err_ds;
2432 cred_t *cr;
2433 } dsl_dataset_promote_arg_t;
2434
2435 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2436 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2437 void *tag);
2438 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2439
2440 static int
2441 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2442 {
2443 dsl_dataset_promote_arg_t *ddpa = arg;
2444 dsl_pool_t *dp = dmu_tx_pool(tx);
2445 dsl_dataset_t *hds;
2446 struct promotenode *snap;
2447 dsl_dataset_t *origin_ds;
2448 int err;
2449 uint64_t unused;
2450 uint64_t ss_mv_cnt;
2451 size_t max_snap_len;
2452
2453 err = promote_hold(ddpa, dp, FTAG);
2454 if (err != 0)
2455 return (err);
2456
2457 hds = ddpa->ddpa_clone;
2458 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2459
2460 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2461 promote_rele(ddpa, FTAG);
2462 return (SET_ERROR(EXDEV));
2463 }
2464
2465 snap = list_head(&ddpa->shared_snaps);
2466 if (snap == NULL) {
2467 err = SET_ERROR(ENOENT);
2468 goto out;
2469 }
2470 origin_ds = snap->ds;
2471
2472 /*
2473 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
2474 * When doing a promote we must make sure the encryption root for
2475 * both the target and the target's origin does not change to avoid
2476 * needing to rewrap encryption keys
2477 */
2478 err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
2479 if (err != 0)
2480 goto out;
2481
2482 /*
2483 * Compute and check the amount of space to transfer. Since this is
2484 * so expensive, don't do the preliminary check.
2485 */
2486 if (!dmu_tx_is_syncing(tx)) {
2487 promote_rele(ddpa, FTAG);
2488 return (0);
2489 }
2490
2491 /* compute origin's new unique space */
2492 snap = list_tail(&ddpa->clone_snaps);
2493 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2494 origin_ds->ds_object);
2495 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2496 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2497 &ddpa->unique, &unused, &unused);
2498
2499 /*
2500 * Walk the snapshots that we are moving
2501 *
2502 * Compute space to transfer. Consider the incremental changes
2503 * to used by each snapshot:
2504 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2505 * So each snapshot gave birth to:
2506 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2507 * So a sequence would look like:
2508 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2509 * Which simplifies to:
2510 * uN + kN + kN-1 + ... + k1 + k0
2511 * Note however, if we stop before we reach the ORIGIN we get:
2512 * uN + kN + kN-1 + ... + kM - uM-1
2513 */
2514 ss_mv_cnt = 0;
2515 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2516 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2517 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2518 for (snap = list_head(&ddpa->shared_snaps); snap;
2519 snap = list_next(&ddpa->shared_snaps, snap)) {
2520 uint64_t val, dlused, dlcomp, dluncomp;
2521 dsl_dataset_t *ds = snap->ds;
2522
2523 ss_mv_cnt++;
2524
2525 /*
2526 * If there are long holds, we won't be able to evict
2527 * the objset.
2528 */
2529 if (dsl_dataset_long_held(ds)) {
2530 err = SET_ERROR(EBUSY);
2531 goto out;
2532 }
2533
2534 /* Check that the snapshot name does not conflict */
2535 VERIFY0(dsl_dataset_get_snapname(ds));
2536 if (strlen(ds->ds_snapname) >= max_snap_len) {
2537 err = SET_ERROR(ENAMETOOLONG);
2538 goto out;
2539 }
2540 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2541 if (err == 0) {
2542 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2543 err = SET_ERROR(EEXIST);
2544 goto out;
2545 }
2546 if (err != ENOENT)
2547 goto out;
2548
2549 /* The very first snapshot does not have a deadlist */
2550 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2551 continue;
2552
2553 dsl_deadlist_space(&ds->ds_deadlist,
2554 &dlused, &dlcomp, &dluncomp);
2555 ddpa->used += dlused;
2556 ddpa->comp += dlcomp;
2557 ddpa->uncomp += dluncomp;
2558 }
2559
2560 /*
2561 * If we are a clone of a clone then we never reached ORIGIN,
2562 * so we need to subtract out the clone origin's used space.
2563 */
2564 if (ddpa->origin_origin) {
2565 ddpa->used -=
2566 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2567 ddpa->comp -=
2568 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2569 ddpa->uncomp -=
2570 dsl_dataset_phys(ddpa->origin_origin)->
2571 ds_uncompressed_bytes;
2572 }
2573
2574 /* Check that there is enough space and limit headroom here */
2575 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2576 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2577 if (err != 0)
2578 goto out;
2579
2580 /*
2581 * Compute the amounts of space that will be used by snapshots
2582 * after the promotion (for both origin and clone). For each,
2583 * it is the amount of space that will be on all of their
2584 * deadlists (that was not born before their new origin).
2585 */
2586 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2587 uint64_t space;
2588
2589 /*
2590 * Note, typically this will not be a clone of a clone,
2591 * so dd_origin_txg will be < TXG_INITIAL, so
2592 * these snaplist_space() -> dsl_deadlist_space_range()
2593 * calls will be fast because they do not have to
2594 * iterate over all bps.
2595 */
2596 snap = list_head(&ddpa->origin_snaps);
2597 if (snap == NULL) {
2598 err = SET_ERROR(ENOENT);
2599 goto out;
2600 }
2601 err = snaplist_space(&ddpa->shared_snaps,
2602 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2603 if (err != 0)
2604 goto out;
2605
2606 err = snaplist_space(&ddpa->clone_snaps,
2607 snap->ds->ds_dir->dd_origin_txg, &space);
2608 if (err != 0)
2609 goto out;
2610 ddpa->cloneusedsnap += space;
2611 }
2612 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2613 DD_FLAG_USED_BREAKDOWN) {
2614 err = snaplist_space(&ddpa->origin_snaps,
2615 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2616 &ddpa->originusedsnap);
2617 if (err != 0)
2618 goto out;
2619 }
2620
2621 out:
2622 promote_rele(ddpa, FTAG);
2623 return (err);
2624 }
2625
2626 static void
2627 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2628 {
2629 dsl_dataset_promote_arg_t *ddpa = arg;
2630 dsl_pool_t *dp = dmu_tx_pool(tx);
2631 dsl_dataset_t *hds;
2632 struct promotenode *snap;
2633 dsl_dataset_t *origin_ds;
2634 dsl_dataset_t *origin_head;
2635 dsl_dir_t *dd;
2636 dsl_dir_t *odd = NULL;
2637 uint64_t oldnext_obj;
2638 int64_t delta;
2639
2640 VERIFY0(promote_hold(ddpa, dp, FTAG));
2641 hds = ddpa->ddpa_clone;
2642
2643 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2644
2645 snap = list_head(&ddpa->shared_snaps);
2646 origin_ds = snap->ds;
2647 dd = hds->ds_dir;
2648
2649 snap = list_head(&ddpa->origin_snaps);
2650 origin_head = snap->ds;
2651
2652 /*
2653 * We need to explicitly open odd, since origin_ds's dd will be
2654 * changing.
2655 */
2656 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2657 NULL, FTAG, &odd));
2658
2659 dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
2660
2661 /* change origin's next snap */
2662 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2663 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2664 snap = list_tail(&ddpa->clone_snaps);
2665 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2666 origin_ds->ds_object);
2667 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2668
2669 /* change the origin's next clone */
2670 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2671 dsl_dataset_remove_from_next_clones(origin_ds,
2672 snap->ds->ds_object, tx);
2673 VERIFY0(zap_add_int(dp->dp_meta_objset,
2674 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2675 oldnext_obj, tx));
2676 }
2677
2678 /* change origin */
2679 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2680 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2681 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2682 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2683 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2684 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2685 origin_head->ds_dir->dd_origin_txg =
2686 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2687
2688 /* change dd_clone entries */
2689 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2690 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2691 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2692 VERIFY0(zap_add_int(dp->dp_meta_objset,
2693 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2694 hds->ds_object, tx));
2695
2696 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2697 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2698 origin_head->ds_object, tx));
2699 if (dsl_dir_phys(dd)->dd_clones == 0) {
2700 dsl_dir_phys(dd)->dd_clones =
2701 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2702 DMU_OT_NONE, 0, tx);
2703 }
2704 VERIFY0(zap_add_int(dp->dp_meta_objset,
2705 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2706 }
2707
2708 /* move snapshots to this dir */
2709 for (snap = list_head(&ddpa->shared_snaps); snap;
2710 snap = list_next(&ddpa->shared_snaps, snap)) {
2711 dsl_dataset_t *ds = snap->ds;
2712
2713 /*
2714 * Property callbacks are registered to a particular
2715 * dsl_dir. Since ours is changing, evict the objset
2716 * so that they will be unregistered from the old dsl_dir.
2717 */
2718 if (ds->ds_objset) {
2719 dmu_objset_evict(ds->ds_objset);
2720 ds->ds_objset = NULL;
2721 }
2722
2723 /* move snap name entry */
2724 VERIFY0(dsl_dataset_get_snapname(ds));
2725 VERIFY0(dsl_dataset_snap_remove(origin_head,
2726 ds->ds_snapname, tx, B_TRUE));
2727 VERIFY0(zap_add(dp->dp_meta_objset,
2728 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2729 8, 1, &ds->ds_object, tx));
2730 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2731 DD_FIELD_SNAPSHOT_COUNT, tx);
2732
2733 /* change containing dsl_dir */
2734 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2735 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2736 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2737 ASSERT3P(ds->ds_dir, ==, odd);
2738 dsl_dir_rele(ds->ds_dir, ds);
2739 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2740 NULL, ds, &ds->ds_dir));
2741
2742 /* move any clone references */
2743 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2744 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2745 zap_cursor_t zc;
2746 zap_attribute_t za;
2747
2748 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2749 dsl_dataset_phys(ds)->ds_next_clones_obj);
2750 zap_cursor_retrieve(&zc, &za) == 0;
2751 zap_cursor_advance(&zc)) {
2752 dsl_dataset_t *cnds;
2753 uint64_t o;
2754
2755 if (za.za_first_integer == oldnext_obj) {
2756 /*
2757 * We've already moved the
2758 * origin's reference.
2759 */
2760 continue;
2761 }
2762
2763 VERIFY0(dsl_dataset_hold_obj(dp,
2764 za.za_first_integer, FTAG, &cnds));
2765 o = dsl_dir_phys(cnds->ds_dir)->
2766 dd_head_dataset_obj;
2767
2768 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2769 dsl_dir_phys(odd)->dd_clones, o, tx));
2770 VERIFY0(zap_add_int(dp->dp_meta_objset,
2771 dsl_dir_phys(dd)->dd_clones, o, tx));
2772 dsl_dataset_rele(cnds, FTAG);
2773 }
2774 zap_cursor_fini(&zc);
2775 }
2776
2777 ASSERT(!dsl_prop_hascb(ds));
2778 }
2779
2780 /*
2781 * Change space accounting.
2782 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2783 * both be valid, or both be 0 (resulting in delta == 0). This
2784 * is true for each of {clone,origin} independently.
2785 */
2786
2787 delta = ddpa->cloneusedsnap -
2788 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2789 ASSERT3S(delta, >=, 0);
2790 ASSERT3U(ddpa->used, >=, delta);
2791 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2792 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2793 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2794
2795 delta = ddpa->originusedsnap -
2796 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2797 ASSERT3S(delta, <=, 0);
2798 ASSERT3U(ddpa->used, >=, -delta);
2799 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2800 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2801 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2802
2803 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2804
2805 /* log history record */
2806 spa_history_log_internal_ds(hds, "promote", tx, "");
2807
2808 dsl_dir_rele(odd, FTAG);
2809 promote_rele(ddpa, FTAG);
2810 }
2811
2812 /*
2813 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2814 * (exclusive) and last_obj (inclusive). The list will be in reverse
2815 * order (last_obj will be the list_head()). If first_obj == 0, do all
2816 * snapshots back to this dataset's origin.
2817 */
2818 static int
2819 snaplist_make(dsl_pool_t *dp,
2820 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2821 {
2822 uint64_t obj = last_obj;
2823
2824 list_create(l, sizeof (struct promotenode),
2825 offsetof(struct promotenode, link));
2826
2827 while (obj != first_obj) {
2828 dsl_dataset_t *ds;
2829 struct promotenode *snap;
2830 int err;
2831
2832 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2833 ASSERT(err != ENOENT);
2834 if (err != 0)
2835 return (err);
2836
2837 if (first_obj == 0)
2838 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2839
2840 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2841 snap->ds = ds;
2842 list_insert_tail(l, snap);
2843 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2844 }
2845
2846 return (0);
2847 }
2848
2849 static int
2850 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2851 {
2852 struct promotenode *snap;
2853
2854 *spacep = 0;
2855 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2856 uint64_t used, comp, uncomp;
2857 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2858 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2859 *spacep += used;
2860 }
2861 return (0);
2862 }
2863
2864 static void
2865 snaplist_destroy(list_t *l, void *tag)
2866 {
2867 struct promotenode *snap;
2868
2869 if (l == NULL || !list_link_active(&l->list_head))
2870 return;
2871
2872 while ((snap = list_tail(l)) != NULL) {
2873 list_remove(l, snap);
2874 dsl_dataset_rele(snap->ds, tag);
2875 kmem_free(snap, sizeof (*snap));
2876 }
2877 list_destroy(l);
2878 }
2879
2880 static int
2881 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2882 {
2883 int error;
2884 dsl_dir_t *dd;
2885 struct promotenode *snap;
2886
2887 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2888 &ddpa->ddpa_clone);
2889 if (error != 0)
2890 return (error);
2891 dd = ddpa->ddpa_clone->ds_dir;
2892
2893 if (ddpa->ddpa_clone->ds_is_snapshot ||
2894 !dsl_dir_is_clone(dd)) {
2895 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2896 return (SET_ERROR(EINVAL));
2897 }
2898
2899 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2900 &ddpa->shared_snaps, tag);
2901 if (error != 0)
2902 goto out;
2903
2904 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2905 &ddpa->clone_snaps, tag);
2906 if (error != 0)
2907 goto out;
2908
2909 snap = list_head(&ddpa->shared_snaps);
2910 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2911 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2912 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2913 &ddpa->origin_snaps, tag);
2914 if (error != 0)
2915 goto out;
2916
2917 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2918 error = dsl_dataset_hold_obj(dp,
2919 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2920 tag, &ddpa->origin_origin);
2921 if (error != 0)
2922 goto out;
2923 }
2924 out:
2925 if (error != 0)
2926 promote_rele(ddpa, tag);
2927 return (error);
2928 }
2929
2930 static void
2931 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2932 {
2933 snaplist_destroy(&ddpa->shared_snaps, tag);
2934 snaplist_destroy(&ddpa->clone_snaps, tag);
2935 snaplist_destroy(&ddpa->origin_snaps, tag);
2936 if (ddpa->origin_origin != NULL)
2937 dsl_dataset_rele(ddpa->origin_origin, tag);
2938 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2939 }
2940
2941 /*
2942 * Promote a clone.
2943 *
2944 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2945 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
2946 */
2947 int
2948 dsl_dataset_promote(const char *name, char *conflsnap)
2949 {
2950 dsl_dataset_promote_arg_t ddpa = { 0 };
2951 uint64_t numsnaps;
2952 int error;
2953 objset_t *os;
2954
2955 /*
2956 * We will modify space proportional to the number of
2957 * snapshots. Compute numsnaps.
2958 */
2959 error = dmu_objset_hold(name, FTAG, &os);
2960 if (error != 0)
2961 return (error);
2962 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2963 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2964 &numsnaps);
2965 dmu_objset_rele(os, FTAG);
2966 if (error != 0)
2967 return (error);
2968
2969 ddpa.ddpa_clonename = name;
2970 ddpa.err_ds = conflsnap;
2971 ddpa.cr = CRED();
2972
2973 return (dsl_sync_task(name, dsl_dataset_promote_check,
2974 dsl_dataset_promote_sync, &ddpa,
2975 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2976 }
2977
2978 int
2979 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2980 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2981 {
2982 /*
2983 * "slack" factor for received datasets with refquota set on them.
2984 * See the bottom of this function for details on its use.
2985 */
2986 uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS *
2987 spa_asize_inflation;
2988 int64_t unused_refres_delta;
2989
2990 /* they should both be heads */
2991 if (clone->ds_is_snapshot ||
2992 origin_head->ds_is_snapshot)
2993 return (SET_ERROR(EINVAL));
2994
2995 /* if we are not forcing, the branch point should be just before them */
2996 if (!force && clone->ds_prev != origin_head->ds_prev)
2997 return (SET_ERROR(EINVAL));
2998
2999 /* clone should be the clone (unless they are unrelated) */
3000 if (clone->ds_prev != NULL &&
3001 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3002 origin_head->ds_dir != clone->ds_prev->ds_dir)
3003 return (SET_ERROR(EINVAL));
3004
3005 /* the clone should be a child of the origin */
3006 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3007 return (SET_ERROR(EINVAL));
3008
3009 /* origin_head shouldn't be modified unless 'force' */
3010 if (!force &&
3011 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3012 return (SET_ERROR(ETXTBSY));
3013
3014 /* origin_head should have no long holds (e.g. is not mounted) */
3015 if (dsl_dataset_handoff_check(origin_head, owner, tx))
3016 return (SET_ERROR(EBUSY));
3017
3018 /* check amount of any unconsumed refreservation */
3019 unused_refres_delta =
3020 (int64_t)MIN(origin_head->ds_reserved,
3021 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3022 (int64_t)MIN(origin_head->ds_reserved,
3023 dsl_dataset_phys(clone)->ds_unique_bytes);
3024
3025 if (unused_refres_delta > 0 &&
3026 unused_refres_delta >
3027 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3028 return (SET_ERROR(ENOSPC));
3029
3030 /*
3031 * The clone can't be too much over the head's refquota.
3032 *
3033 * To ensure that the entire refquota can be used, we allow one
3034 * transaction to exceed the the refquota. Therefore, this check
3035 * needs to also allow for the space referenced to be more than the
3036 * refquota. The maximum amount of space that one transaction can use
3037 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3038 * overage ensures that we are able to receive a filesystem that
3039 * exceeds the refquota on the source system.
3040 *
3041 * So that overage is the refquota_slack we use below.
3042 */
3043 if (origin_head->ds_quota != 0 &&
3044 dsl_dataset_phys(clone)->ds_referenced_bytes >
3045 origin_head->ds_quota + refquota_slack)
3046 return (SET_ERROR(EDQUOT));
3047
3048 return (0);
3049 }
3050
3051 void
3052 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3053 dsl_dataset_t *origin_head, dmu_tx_t *tx)
3054 {
3055 dsl_pool_t *dp = dmu_tx_pool(tx);
3056 int64_t unused_refres_delta;
3057
3058 ASSERT(clone->ds_reserved == 0);
3059 /*
3060 * NOTE: On DEBUG kernels there could be a race between this and
3061 * the check function if spa_asize_inflation is adjusted...
3062 */
3063 ASSERT(origin_head->ds_quota == 0 ||
3064 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3065 DMU_MAX_ACCESS * spa_asize_inflation);
3066 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3067
3068 /*
3069 * Swap per-dataset feature flags.
3070 */
3071 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3072 if (!(spa_feature_table[f].fi_flags &
3073 ZFEATURE_FLAG_PER_DATASET)) {
3074 ASSERT(!clone->ds_feature_inuse[f]);
3075 ASSERT(!origin_head->ds_feature_inuse[f]);
3076 continue;
3077 }
3078
3079 boolean_t clone_inuse = clone->ds_feature_inuse[f];
3080 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3081
3082 if (clone_inuse) {
3083 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3084 clone->ds_feature_inuse[f] = B_FALSE;
3085 }
3086 if (origin_head_inuse) {
3087 dsl_dataset_deactivate_feature(origin_head->ds_object,
3088 f, tx);
3089 origin_head->ds_feature_inuse[f] = B_FALSE;
3090 }
3091 if (clone_inuse) {
3092 dsl_dataset_activate_feature(origin_head->ds_object,
3093 f, tx);
3094 origin_head->ds_feature_inuse[f] = B_TRUE;
3095 }
3096 if (origin_head_inuse) {
3097 dsl_dataset_activate_feature(clone->ds_object, f, tx);
3098 clone->ds_feature_inuse[f] = B_TRUE;
3099 }
3100 }
3101
3102 dmu_buf_will_dirty(clone->ds_dbuf, tx);
3103 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3104
3105 if (clone->ds_objset != NULL) {
3106 dmu_objset_evict(clone->ds_objset);
3107 clone->ds_objset = NULL;
3108 }
3109
3110 if (origin_head->ds_objset != NULL) {
3111 dmu_objset_evict(origin_head->ds_objset);
3112 origin_head->ds_objset = NULL;
3113 }
3114
3115 unused_refres_delta =
3116 (int64_t)MIN(origin_head->ds_reserved,
3117 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3118 (int64_t)MIN(origin_head->ds_reserved,
3119 dsl_dataset_phys(clone)->ds_unique_bytes);
3120
3121 /*
3122 * Reset origin's unique bytes, if it exists.
3123 */
3124 if (clone->ds_prev) {
3125 dsl_dataset_t *origin = clone->ds_prev;
3126 uint64_t comp, uncomp;
3127
3128 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3129 dsl_deadlist_space_range(&clone->ds_deadlist,
3130 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3131 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3132 }
3133
3134 /* swap blkptrs */
3135 {
3136 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3137 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3138 blkptr_t tmp;
3139 tmp = dsl_dataset_phys(origin_head)->ds_bp;
3140 dsl_dataset_phys(origin_head)->ds_bp =
3141 dsl_dataset_phys(clone)->ds_bp;
3142 dsl_dataset_phys(clone)->ds_bp = tmp;
3143 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3144 rrw_exit(&clone->ds_bp_rwlock, FTAG);
3145 }
3146
3147 /* set dd_*_bytes */
3148 {
3149 int64_t dused, dcomp, duncomp;
3150 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3151 uint64_t odl_used, odl_comp, odl_uncomp;
3152
3153 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3154 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3155
3156 dsl_deadlist_space(&clone->ds_deadlist,
3157 &cdl_used, &cdl_comp, &cdl_uncomp);
3158 dsl_deadlist_space(&origin_head->ds_deadlist,
3159 &odl_used, &odl_comp, &odl_uncomp);
3160
3161 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3162 cdl_used -
3163 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3164 odl_used);
3165 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3166 cdl_comp -
3167 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3168 odl_comp);
3169 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3170 cdl_uncomp -
3171 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3172 odl_uncomp);
3173
3174 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3175 dused, dcomp, duncomp, tx);
3176 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3177 -dused, -dcomp, -duncomp, tx);
3178
3179 /*
3180 * The difference in the space used by snapshots is the
3181 * difference in snapshot space due to the head's
3182 * deadlist (since that's the only thing that's
3183 * changing that affects the snapused).
3184 */
3185 dsl_deadlist_space_range(&clone->ds_deadlist,
3186 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3187 &cdl_used, &cdl_comp, &cdl_uncomp);
3188 dsl_deadlist_space_range(&origin_head->ds_deadlist,
3189 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3190 &odl_used, &odl_comp, &odl_uncomp);
3191 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3192 DD_USED_HEAD, DD_USED_SNAP, tx);
3193 }
3194
3195 /* swap ds_*_bytes */
3196 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3197 dsl_dataset_phys(clone)->ds_referenced_bytes);
3198 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3199 dsl_dataset_phys(clone)->ds_compressed_bytes);
3200 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3201 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3202 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3203 dsl_dataset_phys(clone)->ds_unique_bytes);
3204
3205 /* apply any parent delta for change in unconsumed refreservation */
3206 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3207 unused_refres_delta, 0, 0, tx);
3208
3209 /*
3210 * Swap deadlists.
3211 */
3212 dsl_deadlist_close(&clone->ds_deadlist);
3213 dsl_deadlist_close(&origin_head->ds_deadlist);
3214 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3215 dsl_dataset_phys(clone)->ds_deadlist_obj);
3216 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3217 dsl_dataset_phys(clone)->ds_deadlist_obj);
3218 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3219 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3220
3221 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3222
3223 spa_history_log_internal_ds(clone, "clone swap", tx,
3224 "parent=%s", origin_head->ds_dir->dd_myname);
3225 }
3226
3227 /*
3228 * Given a pool name and a dataset object number in that pool,
3229 * return the name of that dataset.
3230 */
3231 int
3232 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3233 {
3234 dsl_pool_t *dp;
3235 dsl_dataset_t *ds;
3236 int error;
3237
3238 error = dsl_pool_hold(pname, FTAG, &dp);
3239 if (error != 0)
3240 return (error);
3241
3242 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3243 if (error == 0) {
3244 dsl_dataset_name(ds, buf);
3245 dsl_dataset_rele(ds, FTAG);
3246 }
3247 dsl_pool_rele(dp, FTAG);
3248
3249 return (error);
3250 }
3251
3252 int
3253 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3254 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3255 {
3256 int error = 0;
3257
3258 ASSERT3S(asize, >, 0);
3259
3260 /*
3261 * *ref_rsrv is the portion of asize that will come from any
3262 * unconsumed refreservation space.
3263 */
3264 *ref_rsrv = 0;
3265
3266 mutex_enter(&ds->ds_lock);
3267 /*
3268 * Make a space adjustment for reserved bytes.
3269 */
3270 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3271 ASSERT3U(*used, >=,
3272 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3273 *used -=
3274 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3275 *ref_rsrv =
3276 asize - MIN(asize, parent_delta(ds, asize + inflight));
3277 }
3278
3279 if (!check_quota || ds->ds_quota == 0) {
3280 mutex_exit(&ds->ds_lock);
3281 return (0);
3282 }
3283 /*
3284 * If they are requesting more space, and our current estimate
3285 * is over quota, they get to try again unless the actual
3286 * on-disk is over quota and there are no pending changes (which
3287 * may free up space for us).
3288 */
3289 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3290 ds->ds_quota) {
3291 if (inflight > 0 ||
3292 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3293 error = SET_ERROR(ERESTART);
3294 else
3295 error = SET_ERROR(EDQUOT);
3296 }
3297 mutex_exit(&ds->ds_lock);
3298
3299 return (error);
3300 }
3301
3302 typedef struct dsl_dataset_set_qr_arg {
3303 const char *ddsqra_name;
3304 zprop_source_t ddsqra_source;
3305 uint64_t ddsqra_value;
3306 } dsl_dataset_set_qr_arg_t;
3307
3308
3309 /* ARGSUSED */
3310 static int
3311 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3312 {
3313 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3314 dsl_pool_t *dp = dmu_tx_pool(tx);
3315 dsl_dataset_t *ds;
3316 int error;
3317 uint64_t newval;
3318
3319 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3320 return (SET_ERROR(ENOTSUP));
3321
3322 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3323 if (error != 0)
3324 return (error);
3325
3326 if (ds->ds_is_snapshot) {
3327 dsl_dataset_rele(ds, FTAG);
3328 return (SET_ERROR(EINVAL));
3329 }
3330
3331 error = dsl_prop_predict(ds->ds_dir,
3332 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3333 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3334 if (error != 0) {
3335 dsl_dataset_rele(ds, FTAG);
3336 return (error);
3337 }
3338
3339 if (newval == 0) {
3340 dsl_dataset_rele(ds, FTAG);
3341 return (0);
3342 }
3343
3344 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3345 newval < ds->ds_reserved) {
3346 dsl_dataset_rele(ds, FTAG);
3347 return (SET_ERROR(ENOSPC));
3348 }
3349
3350 dsl_dataset_rele(ds, FTAG);
3351 return (0);
3352 }
3353
3354 static void
3355 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3356 {
3357 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3358 dsl_pool_t *dp = dmu_tx_pool(tx);
3359 dsl_dataset_t *ds = NULL;
3360 uint64_t newval;
3361
3362 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3363
3364 dsl_prop_set_sync_impl(ds,
3365 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3366 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3367 &ddsqra->ddsqra_value, tx);
3368
3369 VERIFY0(dsl_prop_get_int_ds(ds,
3370 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3371
3372 if (ds->ds_quota != newval) {
3373 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3374 ds->ds_quota = newval;
3375 }
3376 dsl_dataset_rele(ds, FTAG);
3377 }
3378
3379 int
3380 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3381 uint64_t refquota)
3382 {
3383 dsl_dataset_set_qr_arg_t ddsqra;
3384
3385 ddsqra.ddsqra_name = dsname;
3386 ddsqra.ddsqra_source = source;
3387 ddsqra.ddsqra_value = refquota;
3388
3389 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3390 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3391 }
3392
3393 static int
3394 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3395 {
3396 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3397 dsl_pool_t *dp = dmu_tx_pool(tx);
3398 dsl_dataset_t *ds;
3399 int error;
3400 uint64_t newval, unique;
3401
3402 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3403 return (SET_ERROR(ENOTSUP));
3404
3405 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3406 if (error != 0)
3407 return (error);
3408
3409 if (ds->ds_is_snapshot) {
3410 dsl_dataset_rele(ds, FTAG);
3411 return (SET_ERROR(EINVAL));
3412 }
3413
3414 error = dsl_prop_predict(ds->ds_dir,
3415 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3416 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3417 if (error != 0) {
3418 dsl_dataset_rele(ds, FTAG);
3419 return (error);
3420 }
3421
3422 /*
3423 * If we are doing the preliminary check in open context, the
3424 * space estimates may be inaccurate.
3425 */
3426 if (!dmu_tx_is_syncing(tx)) {
3427 dsl_dataset_rele(ds, FTAG);
3428 return (0);
3429 }
3430
3431 mutex_enter(&ds->ds_lock);
3432 if (!DS_UNIQUE_IS_ACCURATE(ds))
3433 dsl_dataset_recalc_head_uniq(ds);
3434 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3435 mutex_exit(&ds->ds_lock);
3436
3437 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3438 uint64_t delta = MAX(unique, newval) -
3439 MAX(unique, ds->ds_reserved);
3440
3441 if (delta >
3442 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3443 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3444 dsl_dataset_rele(ds, FTAG);
3445 return (SET_ERROR(ENOSPC));
3446 }
3447 }
3448
3449 dsl_dataset_rele(ds, FTAG);
3450 return (0);
3451 }
3452
3453 void
3454 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3455 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3456 {
3457 uint64_t newval;
3458 uint64_t unique;
3459 int64_t delta;
3460
3461 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3462 source, sizeof (value), 1, &value, tx);
3463
3464 VERIFY0(dsl_prop_get_int_ds(ds,
3465 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3466
3467 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3468 mutex_enter(&ds->ds_dir->dd_lock);
3469 mutex_enter(&ds->ds_lock);
3470 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3471 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3472 delta = MAX(0, (int64_t)(newval - unique)) -
3473 MAX(0, (int64_t)(ds->ds_reserved - unique));
3474 ds->ds_reserved = newval;
3475 mutex_exit(&ds->ds_lock);
3476
3477 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3478 mutex_exit(&ds->ds_dir->dd_lock);
3479 }
3480
3481 static void
3482 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3483 {
3484 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3485 dsl_pool_t *dp = dmu_tx_pool(tx);
3486 dsl_dataset_t *ds = NULL;
3487
3488 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3489 dsl_dataset_set_refreservation_sync_impl(ds,
3490 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3491 dsl_dataset_rele(ds, FTAG);
3492 }
3493
3494 int
3495 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3496 uint64_t refreservation)
3497 {
3498 dsl_dataset_set_qr_arg_t ddsqra;
3499
3500 ddsqra.ddsqra_name = dsname;
3501 ddsqra.ddsqra_source = source;
3502 ddsqra.ddsqra_value = refreservation;
3503
3504 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3505 dsl_dataset_set_refreservation_sync, &ddsqra,
3506 0, ZFS_SPACE_CHECK_NONE));
3507 }
3508
3509 /*
3510 * Return (in *usedp) the amount of space written in new that is not
3511 * present in oldsnap. New may be a snapshot or the head. Old must be
3512 * a snapshot before new, in new's filesystem (or its origin). If not then
3513 * fail and return EINVAL.
3514 *
3515 * The written space is calculated by considering two components: First, we
3516 * ignore any freed space, and calculate the written as new's used space
3517 * minus old's used space. Next, we add in the amount of space that was freed
3518 * between the two snapshots, thus reducing new's used space relative to old's.
3519 * Specifically, this is the space that was born before old->ds_creation_txg,
3520 * and freed before new (ie. on new's deadlist or a previous deadlist).
3521 *
3522 * space freed [---------------------]
3523 * snapshots ---O-------O--------O-------O------
3524 * oldsnap new
3525 */
3526 int
3527 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3528 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3529 {
3530 int err = 0;
3531 uint64_t snapobj;
3532 dsl_pool_t *dp = new->ds_dir->dd_pool;
3533
3534 ASSERT(dsl_pool_config_held(dp));
3535
3536 *usedp = 0;
3537 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3538 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3539
3540 *compp = 0;
3541 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3542 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3543
3544 *uncompp = 0;
3545 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3546 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3547
3548 snapobj = new->ds_object;
3549 while (snapobj != oldsnap->ds_object) {
3550 dsl_dataset_t *snap;
3551 uint64_t used, comp, uncomp;
3552
3553 if (snapobj == new->ds_object) {
3554 snap = new;
3555 } else {
3556 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3557 if (err != 0)
3558 break;
3559 }
3560
3561 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3562 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3563 /*
3564 * The blocks in the deadlist can not be born after
3565 * ds_prev_snap_txg, so get the whole deadlist space,
3566 * which is more efficient (especially for old-format
3567 * deadlists). Unfortunately the deadlist code
3568 * doesn't have enough information to make this
3569 * optimization itself.
3570 */
3571 dsl_deadlist_space(&snap->ds_deadlist,
3572 &used, &comp, &uncomp);
3573 } else {
3574 dsl_deadlist_space_range(&snap->ds_deadlist,
3575 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3576 &used, &comp, &uncomp);
3577 }
3578 *usedp += used;
3579 *compp += comp;
3580 *uncompp += uncomp;
3581
3582 /*
3583 * If we get to the beginning of the chain of snapshots
3584 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3585 * was not a snapshot of/before new.
3586 */
3587 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3588 if (snap != new)
3589 dsl_dataset_rele(snap, FTAG);
3590 if (snapobj == 0) {
3591 err = SET_ERROR(EINVAL);
3592 break;
3593 }
3594
3595 }
3596 return (err);
3597 }
3598
3599 /*
3600 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3601 * lastsnap, and all snapshots in between are deleted.
3602 *
3603 * blocks that would be freed [---------------------------]
3604 * snapshots ---O-------O--------O-------O--------O
3605 * firstsnap lastsnap
3606 *
3607 * This is the set of blocks that were born after the snap before firstsnap,
3608 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3609 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3610 * We calculate this by iterating over the relevant deadlists (from the snap
3611 * after lastsnap, backward to the snap after firstsnap), summing up the
3612 * space on the deadlist that was born after the snap before firstsnap.
3613 */
3614 int
3615 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3616 dsl_dataset_t *lastsnap,
3617 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3618 {
3619 int err = 0;
3620 uint64_t snapobj;
3621 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3622
3623 ASSERT(firstsnap->ds_is_snapshot);
3624 ASSERT(lastsnap->ds_is_snapshot);
3625
3626 /*
3627 * Check that the snapshots are in the same dsl_dir, and firstsnap
3628 * is before lastsnap.
3629 */
3630 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3631 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3632 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3633 return (SET_ERROR(EINVAL));
3634
3635 *usedp = *compp = *uncompp = 0;
3636
3637 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3638 while (snapobj != firstsnap->ds_object) {
3639 dsl_dataset_t *ds;
3640 uint64_t used, comp, uncomp;
3641
3642 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3643 if (err != 0)
3644 break;
3645
3646 dsl_deadlist_space_range(&ds->ds_deadlist,
3647 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3648 &used, &comp, &uncomp);
3649 *usedp += used;
3650 *compp += comp;
3651 *uncompp += uncomp;
3652
3653 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3654 ASSERT3U(snapobj, !=, 0);
3655 dsl_dataset_rele(ds, FTAG);
3656 }
3657 return (err);
3658 }
3659
3660 /*
3661 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3662 * For example, they could both be snapshots of the same filesystem, and
3663 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3664 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3665 * filesystem. Or 'earlier' could be the origin's origin.
3666 *
3667 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3668 */
3669 boolean_t
3670 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3671 uint64_t earlier_txg)
3672 {
3673 dsl_pool_t *dp = later->ds_dir->dd_pool;
3674 int error;
3675 boolean_t ret;
3676
3677 ASSERT(dsl_pool_config_held(dp));
3678 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3679
3680 if (earlier_txg == 0)
3681 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3682
3683 if (later->ds_is_snapshot &&
3684 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3685 return (B_FALSE);
3686
3687 if (later->ds_dir == earlier->ds_dir)
3688 return (B_TRUE);
3689 if (!dsl_dir_is_clone(later->ds_dir))
3690 return (B_FALSE);
3691
3692 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3693 return (B_TRUE);
3694 dsl_dataset_t *origin;
3695 error = dsl_dataset_hold_obj(dp,
3696 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3697 if (error != 0)
3698 return (B_FALSE);
3699 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3700 dsl_dataset_rele(origin, FTAG);
3701 return (ret);
3702 }
3703
3704 void
3705 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3706 {
3707 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3708 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3709 }
3710
3711 boolean_t
3712 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3713 {
3714 dmu_object_info_t doi;
3715
3716 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3717 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3718 }
3719
3720 boolean_t
3721 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3722 {
3723 return (dsl_dataset_is_zapified(ds) &&
3724 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3725 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3726 }
3727
3728 #if defined(_KERNEL) && defined(HAVE_SPL)
3729 #if defined(_LP64)
3730 module_param(zfs_max_recordsize, int, 0644);
3731 MODULE_PARM_DESC(zfs_max_recordsize, "Max allowed record size");
3732 #else
3733 /* Limited to 1M on 32-bit platforms due to lack of virtual address space */
3734 module_param(zfs_max_recordsize, int, 0444);
3735 MODULE_PARM_DESC(zfs_max_recordsize, "Max allowed record size");
3736 #endif
3737
3738 EXPORT_SYMBOL(dsl_dataset_hold);
3739 EXPORT_SYMBOL(dsl_dataset_hold_flags);
3740 EXPORT_SYMBOL(dsl_dataset_hold_obj);
3741 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags);
3742 EXPORT_SYMBOL(dsl_dataset_own);
3743 EXPORT_SYMBOL(dsl_dataset_own_obj);
3744 EXPORT_SYMBOL(dsl_dataset_name);
3745 EXPORT_SYMBOL(dsl_dataset_rele);
3746 EXPORT_SYMBOL(dsl_dataset_rele_flags);
3747 EXPORT_SYMBOL(dsl_dataset_disown);
3748 EXPORT_SYMBOL(dsl_dataset_tryown);
3749 EXPORT_SYMBOL(dsl_dataset_create_sync);
3750 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
3751 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
3752 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
3753 EXPORT_SYMBOL(dsl_dataset_promote);
3754 EXPORT_SYMBOL(dsl_dataset_user_hold);
3755 EXPORT_SYMBOL(dsl_dataset_user_release);
3756 EXPORT_SYMBOL(dsl_dataset_get_holds);
3757 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
3758 EXPORT_SYMBOL(dsl_dataset_get_spa);
3759 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
3760 EXPORT_SYMBOL(dsl_dataset_space_written);
3761 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
3762 EXPORT_SYMBOL(dsl_dataset_sync);
3763 EXPORT_SYMBOL(dsl_dataset_block_born);
3764 EXPORT_SYMBOL(dsl_dataset_block_kill);
3765 EXPORT_SYMBOL(dsl_dataset_dirty);
3766 EXPORT_SYMBOL(dsl_dataset_stats);
3767 EXPORT_SYMBOL(dsl_dataset_fast_stat);
3768 EXPORT_SYMBOL(dsl_dataset_space);
3769 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
3770 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
3771 EXPORT_SYMBOL(dsl_dataset_check_quota);
3772 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
3773 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
3774 #endif
ZFS On Linux mirror for PVE