1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 */
26
27 #include <sys/dmu_objset.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/arc.h>
36 #include <sys/zio.h>
37 #include <sys/zap.h>
38 #include <sys/zfeature.h>
39 #include <sys/unique.h>
40 #include <sys/zfs_context.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/spa.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_onexit.h>
45 #include <sys/zvol.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/dsl_deadlist.h>
48 #include <sys/dsl_destroy.h>
49 #include <sys/dsl_userhold.h>
50
51 #define SWITCH64(x, y) \
52 { \
53 uint64_t __tmp = (x); \
54 (x) = (y); \
55 (y) = __tmp; \
56 }
57
58 #define DS_REF_MAX (1ULL << 62)
59
60 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
61
62 /*
63 * Figure out how much of this delta should be propogated to the dsl_dir
64 * layer. If there's a refreservation, that space has already been
65 * partially accounted for in our ancestors.
66 */
67 static int64_t
68 parent_delta(dsl_dataset_t *ds, int64_t delta)
69 {
70 uint64_t old_bytes, new_bytes;
71
72 if (ds->ds_reserved == 0)
73 return (delta);
74
75 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
76 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
77
78 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
79 return (new_bytes - old_bytes);
80 }
81
82 void
83 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
84 {
85 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
86 int compressed = BP_GET_PSIZE(bp);
87 int uncompressed = BP_GET_UCSIZE(bp);
88 int64_t delta;
89
90 dprintf_bp(bp, "ds=%p", ds);
91
92 ASSERT(dmu_tx_is_syncing(tx));
93 /* It could have been compressed away to nothing */
94 if (BP_IS_HOLE(bp))
95 return;
96 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
97 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
98 if (ds == NULL) {
99 dsl_pool_mos_diduse_space(tx->tx_pool,
100 used, compressed, uncompressed);
101 return;
102 }
103 dmu_buf_will_dirty(ds->ds_dbuf, tx);
104
105 mutex_enter(&ds->ds_dir->dd_lock);
106 mutex_enter(&ds->ds_lock);
107 delta = parent_delta(ds, used);
108 ds->ds_phys->ds_referenced_bytes += used;
109 ds->ds_phys->ds_compressed_bytes += compressed;
110 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
111 ds->ds_phys->ds_unique_bytes += used;
112 mutex_exit(&ds->ds_lock);
113 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
114 compressed, uncompressed, tx);
115 dsl_dir_transfer_space(ds->ds_dir, used - delta,
116 DD_USED_REFRSRV, DD_USED_HEAD, tx);
117 mutex_exit(&ds->ds_dir->dd_lock);
118 }
119
120 int
121 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
122 boolean_t async)
123 {
124 if (BP_IS_HOLE(bp))
125 return (0);
126
127 ASSERT(dmu_tx_is_syncing(tx));
128 ASSERT(bp->blk_birth <= tx->tx_txg);
129
130 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
131 int compressed = BP_GET_PSIZE(bp);
132 int uncompressed = BP_GET_UCSIZE(bp);
133
134 ASSERT(used > 0);
135 if (ds == NULL) {
136 dsl_free(tx->tx_pool, tx->tx_txg, bp);
137 dsl_pool_mos_diduse_space(tx->tx_pool,
138 -used, -compressed, -uncompressed);
139 return (used);
140 }
141 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
142
143 ASSERT(!dsl_dataset_is_snapshot(ds));
144 dmu_buf_will_dirty(ds->ds_dbuf, tx);
145
146 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
147 int64_t delta;
148
149 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
150 dsl_free(tx->tx_pool, tx->tx_txg, bp);
151
152 mutex_enter(&ds->ds_dir->dd_lock);
153 mutex_enter(&ds->ds_lock);
154 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
155 !DS_UNIQUE_IS_ACCURATE(ds));
156 delta = parent_delta(ds, -used);
157 ds->ds_phys->ds_unique_bytes -= used;
158 mutex_exit(&ds->ds_lock);
159 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
160 delta, -compressed, -uncompressed, tx);
161 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
162 DD_USED_REFRSRV, DD_USED_HEAD, tx);
163 mutex_exit(&ds->ds_dir->dd_lock);
164 } else {
165 dprintf_bp(bp, "putting on dead list: %s", "");
166 if (async) {
167 /*
168 * We are here as part of zio's write done callback,
169 * which means we're a zio interrupt thread. We can't
170 * call dsl_deadlist_insert() now because it may block
171 * waiting for I/O. Instead, put bp on the deferred
172 * queue and let dsl_pool_sync() finish the job.
173 */
174 bplist_append(&ds->ds_pending_deadlist, bp);
175 } else {
176 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
177 }
178 ASSERT3U(ds->ds_prev->ds_object, ==,
179 ds->ds_phys->ds_prev_snap_obj);
180 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
181 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
182 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
183 ds->ds_object && bp->blk_birth >
184 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
185 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
186 mutex_enter(&ds->ds_prev->ds_lock);
187 ds->ds_prev->ds_phys->ds_unique_bytes += used;
188 mutex_exit(&ds->ds_prev->ds_lock);
189 }
190 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
191 dsl_dir_transfer_space(ds->ds_dir, used,
192 DD_USED_HEAD, DD_USED_SNAP, tx);
193 }
194 }
195 mutex_enter(&ds->ds_lock);
196 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
197 ds->ds_phys->ds_referenced_bytes -= used;
198 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
199 ds->ds_phys->ds_compressed_bytes -= compressed;
200 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
201 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
202 mutex_exit(&ds->ds_lock);
203
204 return (used);
205 }
206
207 uint64_t
208 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
209 {
210 uint64_t trysnap = 0;
211
212 if (ds == NULL)
213 return (0);
214 /*
215 * The snapshot creation could fail, but that would cause an
216 * incorrect FALSE return, which would only result in an
217 * overestimation of the amount of space that an operation would
218 * consume, which is OK.
219 *
220 * There's also a small window where we could miss a pending
221 * snapshot, because we could set the sync task in the quiescing
222 * phase. So this should only be used as a guess.
223 */
224 if (ds->ds_trysnap_txg >
225 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
226 trysnap = ds->ds_trysnap_txg;
227 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
228 }
229
230 boolean_t
231 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
232 uint64_t blk_birth)
233 {
234 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
235 return (B_FALSE);
236
237 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
238
239 return (B_TRUE);
240 }
241
242 /* ARGSUSED */
243 static void
244 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
245 {
246 dsl_dataset_t *ds = dsv;
247
248 ASSERT(ds->ds_owner == NULL);
249
250 unique_remove(ds->ds_fsid_guid);
251
252 if (ds->ds_objset != NULL)
253 dmu_objset_evict(ds->ds_objset);
254
255 if (ds->ds_prev) {
256 dsl_dataset_rele(ds->ds_prev, ds);
257 ds->ds_prev = NULL;
258 }
259
260 bplist_destroy(&ds->ds_pending_deadlist);
261 if (ds->ds_phys->ds_deadlist_obj != 0)
262 dsl_deadlist_close(&ds->ds_deadlist);
263 if (ds->ds_dir)
264 dsl_dir_rele(ds->ds_dir, ds);
265
266 ASSERT(!list_link_active(&ds->ds_synced_link));
267
268 mutex_destroy(&ds->ds_lock);
269 mutex_destroy(&ds->ds_opening_lock);
270 refcount_destroy(&ds->ds_longholds);
271
272 kmem_free(ds, sizeof (dsl_dataset_t));
273 }
274
275 int
276 dsl_dataset_get_snapname(dsl_dataset_t *ds)
277 {
278 dsl_dataset_phys_t *headphys;
279 int err;
280 dmu_buf_t *headdbuf;
281 dsl_pool_t *dp = ds->ds_dir->dd_pool;
282 objset_t *mos = dp->dp_meta_objset;
283
284 if (ds->ds_snapname[0])
285 return (0);
286 if (ds->ds_phys->ds_next_snap_obj == 0)
287 return (0);
288
289 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
290 FTAG, &headdbuf);
291 if (err != 0)
292 return (err);
293 headphys = headdbuf->db_data;
294 err = zap_value_search(dp->dp_meta_objset,
295 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
296 dmu_buf_rele(headdbuf, FTAG);
297 return (err);
298 }
299
300 int
301 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
302 {
303 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
304 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
305 matchtype_t mt;
306 int err;
307
308 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
309 mt = MT_FIRST;
310 else
311 mt = MT_EXACT;
312
313 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
314 value, mt, NULL, 0, NULL);
315 if (err == ENOTSUP && mt == MT_FIRST)
316 err = zap_lookup(mos, snapobj, name, 8, 1, value);
317 return (err);
318 }
319
320 int
321 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
322 {
323 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
324 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
325 matchtype_t mt;
326 int err;
327
328 dsl_dir_snap_cmtime_update(ds->ds_dir);
329
330 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
331 mt = MT_FIRST;
332 else
333 mt = MT_EXACT;
334
335 err = zap_remove_norm(mos, snapobj, name, mt, tx);
336 if (err == ENOTSUP && mt == MT_FIRST)
337 err = zap_remove(mos, snapobj, name, tx);
338 return (err);
339 }
340
341 int
342 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
343 dsl_dataset_t **dsp)
344 {
345 objset_t *mos = dp->dp_meta_objset;
346 dmu_buf_t *dbuf;
347 dsl_dataset_t *ds;
348 int err;
349 dmu_object_info_t doi;
350
351 ASSERT(dsl_pool_config_held(dp));
352
353 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
354 if (err != 0)
355 return (err);
356
357 /* Make sure dsobj has the correct object type. */
358 dmu_object_info_from_db(dbuf, &doi);
359 if (doi.doi_type != DMU_OT_DSL_DATASET) {
360 dmu_buf_rele(dbuf, tag);
361 return (SET_ERROR(EINVAL));
362 }
363
364 ds = dmu_buf_get_user(dbuf);
365 if (ds == NULL) {
366 dsl_dataset_t *winner = NULL;
367
368 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
369 ds->ds_dbuf = dbuf;
370 ds->ds_object = dsobj;
371 ds->ds_phys = dbuf->db_data;
372
373 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
374 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
375 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
376 refcount_create(&ds->ds_longholds);
377
378 bplist_create(&ds->ds_pending_deadlist);
379 dsl_deadlist_open(&ds->ds_deadlist,
380 mos, ds->ds_phys->ds_deadlist_obj);
381
382 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
383 offsetof(dmu_sendarg_t, dsa_link));
384
385 if (err == 0) {
386 err = dsl_dir_hold_obj(dp,
387 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
388 }
389 if (err != 0) {
390 mutex_destroy(&ds->ds_lock);
391 mutex_destroy(&ds->ds_opening_lock);
392 refcount_destroy(&ds->ds_longholds);
393 bplist_destroy(&ds->ds_pending_deadlist);
394 dsl_deadlist_close(&ds->ds_deadlist);
395 kmem_free(ds, sizeof (dsl_dataset_t));
396 dmu_buf_rele(dbuf, tag);
397 return (err);
398 }
399
400 if (!dsl_dataset_is_snapshot(ds)) {
401 ds->ds_snapname[0] = '\0';
402 if (ds->ds_phys->ds_prev_snap_obj != 0) {
403 err = dsl_dataset_hold_obj(dp,
404 ds->ds_phys->ds_prev_snap_obj,
405 ds, &ds->ds_prev);
406 }
407 } else {
408 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
409 err = dsl_dataset_get_snapname(ds);
410 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
411 err = zap_count(
412 ds->ds_dir->dd_pool->dp_meta_objset,
413 ds->ds_phys->ds_userrefs_obj,
414 &ds->ds_userrefs);
415 }
416 }
417
418 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
419 err = dsl_prop_get_int_ds(ds,
420 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
421 &ds->ds_reserved);
422 if (err == 0) {
423 err = dsl_prop_get_int_ds(ds,
424 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
425 &ds->ds_quota);
426 }
427 } else {
428 ds->ds_reserved = ds->ds_quota = 0;
429 }
430
431 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
432 &ds->ds_phys, dsl_dataset_evict)) != NULL) {
433 bplist_destroy(&ds->ds_pending_deadlist);
434 dsl_deadlist_close(&ds->ds_deadlist);
435 if (ds->ds_prev)
436 dsl_dataset_rele(ds->ds_prev, ds);
437 dsl_dir_rele(ds->ds_dir, ds);
438 mutex_destroy(&ds->ds_lock);
439 mutex_destroy(&ds->ds_opening_lock);
440 refcount_destroy(&ds->ds_longholds);
441 kmem_free(ds, sizeof (dsl_dataset_t));
442 if (err != 0) {
443 dmu_buf_rele(dbuf, tag);
444 return (err);
445 }
446 ds = winner;
447 } else {
448 ds->ds_fsid_guid =
449 unique_insert(ds->ds_phys->ds_fsid_guid);
450 }
451 }
452 ASSERT3P(ds->ds_dbuf, ==, dbuf);
453 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
454 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
455 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
456 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
457 *dsp = ds;
458 return (0);
459 }
460
461 int
462 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
463 void *tag, dsl_dataset_t **dsp)
464 {
465 dsl_dir_t *dd;
466 const char *snapname;
467 uint64_t obj;
468 int err = 0;
469
470 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
471 if (err != 0)
472 return (err);
473
474 ASSERT(dsl_pool_config_held(dp));
475 obj = dd->dd_phys->dd_head_dataset_obj;
476 if (obj != 0)
477 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
478 else
479 err = SET_ERROR(ENOENT);
480
481 /* we may be looking for a snapshot */
482 if (err == 0 && snapname != NULL) {
483 dsl_dataset_t *ds;
484
485 if (*snapname++ != '@') {
486 dsl_dataset_rele(*dsp, tag);
487 dsl_dir_rele(dd, FTAG);
488 return (SET_ERROR(ENOENT));
489 }
490
491 dprintf("looking for snapshot '%s'\n", snapname);
492 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
493 if (err == 0)
494 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
495 dsl_dataset_rele(*dsp, tag);
496
497 if (err == 0) {
498 mutex_enter(&ds->ds_lock);
499 if (ds->ds_snapname[0] == 0)
500 (void) strlcpy(ds->ds_snapname, snapname,
501 sizeof (ds->ds_snapname));
502 mutex_exit(&ds->ds_lock);
503 *dsp = ds;
504 }
505 }
506
507 dsl_dir_rele(dd, FTAG);
508 return (err);
509 }
510
511 int
512 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
513 void *tag, dsl_dataset_t **dsp)
514 {
515 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
516 if (err != 0)
517 return (err);
518 if (!dsl_dataset_tryown(*dsp, tag)) {
519 dsl_dataset_rele(*dsp, tag);
520 *dsp = NULL;
521 return (SET_ERROR(EBUSY));
522 }
523 return (0);
524 }
525
526 int
527 dsl_dataset_own(dsl_pool_t *dp, const char *name,
528 void *tag, dsl_dataset_t **dsp)
529 {
530 int err = dsl_dataset_hold(dp, name, tag, dsp);
531 if (err != 0)
532 return (err);
533 if (!dsl_dataset_tryown(*dsp, tag)) {
534 dsl_dataset_rele(*dsp, tag);
535 return (SET_ERROR(EBUSY));
536 }
537 return (0);
538 }
539
540 /*
541 * See the comment above dsl_pool_hold() for details. In summary, a long
542 * hold is used to prevent destruction of a dataset while the pool hold
543 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
544 *
545 * The dataset and pool must be held when this function is called. After it
546 * is called, the pool hold may be released while the dataset is still held
547 * and accessed.
548 */
549 void
550 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
551 {
552 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
553 (void) refcount_add(&ds->ds_longholds, tag);
554 }
555
556 void
557 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
558 {
559 (void) refcount_remove(&ds->ds_longholds, tag);
560 }
561
562 /* Return B_TRUE if there are any long holds on this dataset. */
563 boolean_t
564 dsl_dataset_long_held(dsl_dataset_t *ds)
565 {
566 return (!refcount_is_zero(&ds->ds_longholds));
567 }
568
569 void
570 dsl_dataset_name(dsl_dataset_t *ds, char *name)
571 {
572 if (ds == NULL) {
573 (void) strcpy(name, "mos");
574 } else {
575 dsl_dir_name(ds->ds_dir, name);
576 VERIFY0(dsl_dataset_get_snapname(ds));
577 if (ds->ds_snapname[0]) {
578 (void) strcat(name, "@");
579 /*
580 * We use a "recursive" mutex so that we
581 * can call dprintf_ds() with ds_lock held.
582 */
583 if (!MUTEX_HELD(&ds->ds_lock)) {
584 mutex_enter(&ds->ds_lock);
585 (void) strcat(name, ds->ds_snapname);
586 mutex_exit(&ds->ds_lock);
587 } else {
588 (void) strcat(name, ds->ds_snapname);
589 }
590 }
591 }
592 }
593
594 static int
595 dsl_dataset_namelen(dsl_dataset_t *ds)
596 {
597 int result;
598
599 if (ds == NULL) {
600 result = 3; /* "mos" */
601 } else {
602 result = dsl_dir_namelen(ds->ds_dir);
603 VERIFY0(dsl_dataset_get_snapname(ds));
604 if (ds->ds_snapname[0]) {
605 ++result; /* adding one for the @-sign */
606 if (!MUTEX_HELD(&ds->ds_lock)) {
607 mutex_enter(&ds->ds_lock);
608 result += strlen(ds->ds_snapname);
609 mutex_exit(&ds->ds_lock);
610 } else {
611 result += strlen(ds->ds_snapname);
612 }
613 }
614 }
615
616 return (result);
617 }
618
619 void
620 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
621 {
622 dmu_buf_rele(ds->ds_dbuf, tag);
623 }
624
625 void
626 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
627 {
628 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
629
630 mutex_enter(&ds->ds_lock);
631 ds->ds_owner = NULL;
632 mutex_exit(&ds->ds_lock);
633 dsl_dataset_long_rele(ds, tag);
634 if (ds->ds_dbuf != NULL)
635 dsl_dataset_rele(ds, tag);
636 else
637 dsl_dataset_evict(NULL, ds);
638 }
639
640 boolean_t
641 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
642 {
643 boolean_t gotit = FALSE;
644
645 mutex_enter(&ds->ds_lock);
646 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
647 ds->ds_owner = tag;
648 dsl_dataset_long_hold(ds, tag);
649 gotit = TRUE;
650 }
651 mutex_exit(&ds->ds_lock);
652 return (gotit);
653 }
654
655 uint64_t
656 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
657 uint64_t flags, dmu_tx_t *tx)
658 {
659 dsl_pool_t *dp = dd->dd_pool;
660 dmu_buf_t *dbuf;
661 dsl_dataset_phys_t *dsphys;
662 uint64_t dsobj;
663 objset_t *mos = dp->dp_meta_objset;
664
665 if (origin == NULL)
666 origin = dp->dp_origin_snap;
667
668 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
669 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
670 ASSERT(dmu_tx_is_syncing(tx));
671 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
672
673 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
674 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
675 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
676 dmu_buf_will_dirty(dbuf, tx);
677 dsphys = dbuf->db_data;
678 bzero(dsphys, sizeof (dsl_dataset_phys_t));
679 dsphys->ds_dir_obj = dd->dd_object;
680 dsphys->ds_flags = flags;
681 dsphys->ds_fsid_guid = unique_create();
682 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
683 sizeof (dsphys->ds_guid));
684 dsphys->ds_snapnames_zapobj =
685 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
686 DMU_OT_NONE, 0, tx);
687 dsphys->ds_creation_time = gethrestime_sec();
688 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
689
690 if (origin == NULL) {
691 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
692 } else {
693 dsl_dataset_t *ohds; /* head of the origin snapshot */
694
695 dsphys->ds_prev_snap_obj = origin->ds_object;
696 dsphys->ds_prev_snap_txg =
697 origin->ds_phys->ds_creation_txg;
698 dsphys->ds_referenced_bytes =
699 origin->ds_phys->ds_referenced_bytes;
700 dsphys->ds_compressed_bytes =
701 origin->ds_phys->ds_compressed_bytes;
702 dsphys->ds_uncompressed_bytes =
703 origin->ds_phys->ds_uncompressed_bytes;
704 dsphys->ds_bp = origin->ds_phys->ds_bp;
705 dsphys->ds_flags |= origin->ds_phys->ds_flags;
706
707 dmu_buf_will_dirty(origin->ds_dbuf, tx);
708 origin->ds_phys->ds_num_children++;
709
710 VERIFY0(dsl_dataset_hold_obj(dp,
711 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
712 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
713 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
714 dsl_dataset_rele(ohds, FTAG);
715
716 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
717 if (origin->ds_phys->ds_next_clones_obj == 0) {
718 origin->ds_phys->ds_next_clones_obj =
719 zap_create(mos,
720 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
721 }
722 VERIFY0(zap_add_int(mos,
723 origin->ds_phys->ds_next_clones_obj, dsobj, tx));
724 }
725
726 dmu_buf_will_dirty(dd->dd_dbuf, tx);
727 dd->dd_phys->dd_origin_obj = origin->ds_object;
728 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
729 if (origin->ds_dir->dd_phys->dd_clones == 0) {
730 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
731 origin->ds_dir->dd_phys->dd_clones =
732 zap_create(mos,
733 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
734 }
735 VERIFY0(zap_add_int(mos,
736 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
737 }
738 }
739
740 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
741 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
742
743 dmu_buf_rele(dbuf, FTAG);
744
745 dmu_buf_will_dirty(dd->dd_dbuf, tx);
746 dd->dd_phys->dd_head_dataset_obj = dsobj;
747
748 return (dsobj);
749 }
750
751 static void
752 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
753 {
754 objset_t *os;
755
756 VERIFY0(dmu_objset_from_ds(ds, &os));
757 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
758 dsl_dataset_dirty(ds, tx);
759 }
760
761 uint64_t
762 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
763 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
764 {
765 dsl_pool_t *dp = pdd->dd_pool;
766 uint64_t dsobj, ddobj;
767 dsl_dir_t *dd;
768
769 ASSERT(dmu_tx_is_syncing(tx));
770 ASSERT(lastname[0] != '@');
771
772 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
773 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
774
775 dsobj = dsl_dataset_create_sync_dd(dd, origin,
776 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
777
778 dsl_deleg_set_create_perms(dd, tx, cr);
779
780 dsl_dir_rele(dd, FTAG);
781
782 /*
783 * If we are creating a clone, make sure we zero out any stale
784 * data from the origin snapshots zil header.
785 */
786 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
787 dsl_dataset_t *ds;
788
789 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
790 dsl_dataset_zero_zil(ds, tx);
791 dsl_dataset_rele(ds, FTAG);
792 }
793
794 return (dsobj);
795 }
796
797 /*
798 * The unique space in the head dataset can be calculated by subtracting
799 * the space used in the most recent snapshot, that is still being used
800 * in this file system, from the space currently in use. To figure out
801 * the space in the most recent snapshot still in use, we need to take
802 * the total space used in the snapshot and subtract out the space that
803 * has been freed up since the snapshot was taken.
804 */
805 void
806 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
807 {
808 uint64_t mrs_used;
809 uint64_t dlused, dlcomp, dluncomp;
810
811 ASSERT(!dsl_dataset_is_snapshot(ds));
812
813 if (ds->ds_phys->ds_prev_snap_obj != 0)
814 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
815 else
816 mrs_used = 0;
817
818 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
819
820 ASSERT3U(dlused, <=, mrs_used);
821 ds->ds_phys->ds_unique_bytes =
822 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
823
824 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
825 SPA_VERSION_UNIQUE_ACCURATE)
826 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
827 }
828
829 void
830 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
831 dmu_tx_t *tx)
832 {
833 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
834 uint64_t count;
835 int err;
836
837 ASSERT(ds->ds_phys->ds_num_children >= 2);
838 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
839 /*
840 * The err should not be ENOENT, but a bug in a previous version
841 * of the code could cause upgrade_clones_cb() to not set
842 * ds_next_snap_obj when it should, leading to a missing entry.
843 * If we knew that the pool was created after
844 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
845 * ENOENT. However, at least we can check that we don't have
846 * too many entries in the next_clones_obj even after failing to
847 * remove this one.
848 */
849 if (err != ENOENT)
850 VERIFY0(err);
851 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
852 &count));
853 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
854 }
855
856
857 blkptr_t *
858 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
859 {
860 return (&ds->ds_phys->ds_bp);
861 }
862
863 void
864 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
865 {
866 ASSERT(dmu_tx_is_syncing(tx));
867 /* If it's the meta-objset, set dp_meta_rootbp */
868 if (ds == NULL) {
869 tx->tx_pool->dp_meta_rootbp = *bp;
870 } else {
871 dmu_buf_will_dirty(ds->ds_dbuf, tx);
872 ds->ds_phys->ds_bp = *bp;
873 }
874 }
875
876 spa_t *
877 dsl_dataset_get_spa(dsl_dataset_t *ds)
878 {
879 return (ds->ds_dir->dd_pool->dp_spa);
880 }
881
882 void
883 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
884 {
885 dsl_pool_t *dp;
886
887 if (ds == NULL) /* this is the meta-objset */
888 return;
889
890 ASSERT(ds->ds_objset != NULL);
891
892 if (ds->ds_phys->ds_next_snap_obj != 0)
893 panic("dirtying snapshot!");
894
895 dp = ds->ds_dir->dd_pool;
896
897 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
898 /* up the hold count until we can be written out */
899 dmu_buf_add_ref(ds->ds_dbuf, ds);
900 }
901 }
902
903 boolean_t
904 dsl_dataset_is_dirty(dsl_dataset_t *ds)
905 {
906 for (int t = 0; t < TXG_SIZE; t++) {
907 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
908 ds, t))
909 return (B_TRUE);
910 }
911 return (B_FALSE);
912 }
913
914 static int
915 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
916 {
917 uint64_t asize;
918
919 if (!dmu_tx_is_syncing(tx))
920 return (0);
921
922 /*
923 * If there's an fs-only reservation, any blocks that might become
924 * owned by the snapshot dataset must be accommodated by space
925 * outside of the reservation.
926 */
927 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
928 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
929 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
930 return (SET_ERROR(ENOSPC));
931
932 /*
933 * Propagate any reserved space for this snapshot to other
934 * snapshot checks in this sync group.
935 */
936 if (asize > 0)
937 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
938
939 return (0);
940 }
941
942 typedef struct dsl_dataset_snapshot_arg {
943 nvlist_t *ddsa_snaps;
944 nvlist_t *ddsa_props;
945 nvlist_t *ddsa_errors;
946 } dsl_dataset_snapshot_arg_t;
947
948 int
949 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
950 dmu_tx_t *tx)
951 {
952 int error;
953 uint64_t value;
954
955 ds->ds_trysnap_txg = tx->tx_txg;
956
957 if (!dmu_tx_is_syncing(tx))
958 return (0);
959
960 /*
961 * We don't allow multiple snapshots of the same txg. If there
962 * is already one, try again.
963 */
964 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
965 return (SET_ERROR(EAGAIN));
966
967 /*
968 * Check for conflicting snapshot name.
969 */
970 error = dsl_dataset_snap_lookup(ds, snapname, &value);
971 if (error == 0)
972 return (SET_ERROR(EEXIST));
973 if (error != ENOENT)
974 return (error);
975
976 error = dsl_dataset_snapshot_reserve_space(ds, tx);
977 if (error != 0)
978 return (error);
979
980 return (0);
981 }
982
983 static int
984 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
985 {
986 dsl_dataset_snapshot_arg_t *ddsa = arg;
987 dsl_pool_t *dp = dmu_tx_pool(tx);
988 nvpair_t *pair;
989 int rv = 0;
990
991 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
992 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
993 int error = 0;
994 dsl_dataset_t *ds;
995 char *name, *atp;
996 char dsname[MAXNAMELEN];
997
998 name = nvpair_name(pair);
999 if (strlen(name) >= MAXNAMELEN)
1000 error = SET_ERROR(ENAMETOOLONG);
1001 if (error == 0) {
1002 atp = strchr(name, '@');
1003 if (atp == NULL)
1004 error = SET_ERROR(EINVAL);
1005 if (error == 0)
1006 (void) strlcpy(dsname, name, atp - name + 1);
1007 }
1008 if (error == 0)
1009 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1010 if (error == 0) {
1011 error = dsl_dataset_snapshot_check_impl(ds,
1012 atp + 1, tx);
1013 dsl_dataset_rele(ds, FTAG);
1014 }
1015
1016 if (error != 0) {
1017 if (ddsa->ddsa_errors != NULL) {
1018 fnvlist_add_int32(ddsa->ddsa_errors,
1019 name, error);
1020 }
1021 rv = error;
1022 }
1023 }
1024 return (rv);
1025 }
1026
1027 void
1028 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1029 dmu_tx_t *tx)
1030 {
1031 static zil_header_t zero_zil;
1032
1033 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1034 dmu_buf_t *dbuf;
1035 dsl_dataset_phys_t *dsphys;
1036 uint64_t dsobj, crtxg;
1037 objset_t *mos = dp->dp_meta_objset;
1038 objset_t *os;
1039
1040 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1041
1042 /*
1043 * If we are on an old pool, the zil must not be active, in which
1044 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1045 */
1046 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1047 dmu_objset_from_ds(ds, &os) != 0 ||
1048 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1049 sizeof (zero_zil)) == 0);
1050
1051
1052 /*
1053 * The origin's ds_creation_txg has to be < TXG_INITIAL
1054 */
1055 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1056 crtxg = 1;
1057 else
1058 crtxg = tx->tx_txg;
1059
1060 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1061 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1062 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1063 dmu_buf_will_dirty(dbuf, tx);
1064 dsphys = dbuf->db_data;
1065 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1066 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1067 dsphys->ds_fsid_guid = unique_create();
1068 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1069 sizeof (dsphys->ds_guid));
1070 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1071 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1072 dsphys->ds_next_snap_obj = ds->ds_object;
1073 dsphys->ds_num_children = 1;
1074 dsphys->ds_creation_time = gethrestime_sec();
1075 dsphys->ds_creation_txg = crtxg;
1076 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1077 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1078 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1079 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1080 dsphys->ds_flags = ds->ds_phys->ds_flags;
1081 dsphys->ds_bp = ds->ds_phys->ds_bp;
1082 dmu_buf_rele(dbuf, FTAG);
1083
1084 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1085 if (ds->ds_prev) {
1086 uint64_t next_clones_obj =
1087 ds->ds_prev->ds_phys->ds_next_clones_obj;
1088 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1089 ds->ds_object ||
1090 ds->ds_prev->ds_phys->ds_num_children > 1);
1091 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1092 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1093 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1094 ds->ds_prev->ds_phys->ds_creation_txg);
1095 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1096 } else if (next_clones_obj != 0) {
1097 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1098 dsphys->ds_next_snap_obj, tx);
1099 VERIFY0(zap_add_int(mos,
1100 next_clones_obj, dsobj, tx));
1101 }
1102 }
1103
1104 /*
1105 * If we have a reference-reservation on this dataset, we will
1106 * need to increase the amount of refreservation being charged
1107 * since our unique space is going to zero.
1108 */
1109 if (ds->ds_reserved) {
1110 int64_t delta;
1111 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1112 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1113 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1114 delta, 0, 0, tx);
1115 }
1116
1117 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1118 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1119 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1120 dsl_deadlist_close(&ds->ds_deadlist);
1121 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1122 dsl_deadlist_add_key(&ds->ds_deadlist,
1123 ds->ds_phys->ds_prev_snap_txg, tx);
1124
1125 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1126 ds->ds_phys->ds_prev_snap_obj = dsobj;
1127 ds->ds_phys->ds_prev_snap_txg = crtxg;
1128 ds->ds_phys->ds_unique_bytes = 0;
1129 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1130 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1131
1132 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1133 snapname, 8, 1, &dsobj, tx));
1134
1135 if (ds->ds_prev)
1136 dsl_dataset_rele(ds->ds_prev, ds);
1137 VERIFY0(dsl_dataset_hold_obj(dp,
1138 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1139
1140 dsl_scan_ds_snapshotted(ds, tx);
1141
1142 dsl_dir_snap_cmtime_update(ds->ds_dir);
1143
1144 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1145 }
1146
1147 static void
1148 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1149 {
1150 dsl_dataset_snapshot_arg_t *ddsa = arg;
1151 dsl_pool_t *dp = dmu_tx_pool(tx);
1152 nvpair_t *pair;
1153
1154 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1155 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1156 dsl_dataset_t *ds;
1157 char *name, *atp;
1158 char dsname[MAXNAMELEN];
1159
1160 name = nvpair_name(pair);
1161 atp = strchr(name, '@');
1162 (void) strlcpy(dsname, name, atp - name + 1);
1163 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1164
1165 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1166 if (ddsa->ddsa_props != NULL) {
1167 dsl_props_set_sync_impl(ds->ds_prev,
1168 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1169 }
1170 dsl_dataset_rele(ds, FTAG);
1171 }
1172 }
1173
1174 /*
1175 * The snapshots must all be in the same pool.
1176 * All-or-nothing: if there are any failures, nothing will be modified.
1177 */
1178 int
1179 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1180 {
1181 dsl_dataset_snapshot_arg_t ddsa;
1182 nvpair_t *pair;
1183 boolean_t needsuspend;
1184 int error;
1185 spa_t *spa;
1186 char *firstname;
1187 nvlist_t *suspended = NULL;
1188
1189 pair = nvlist_next_nvpair(snaps, NULL);
1190 if (pair == NULL)
1191 return (0);
1192 firstname = nvpair_name(pair);
1193
1194 error = spa_open(firstname, &spa, FTAG);
1195 if (error != 0)
1196 return (error);
1197 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1198 spa_close(spa, FTAG);
1199
1200 if (needsuspend) {
1201 suspended = fnvlist_alloc();
1202 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1203 pair = nvlist_next_nvpair(snaps, pair)) {
1204 char fsname[MAXNAMELEN];
1205 char *snapname = nvpair_name(pair);
1206 char *atp;
1207 void *cookie;
1208
1209 atp = strchr(snapname, '@');
1210 if (atp == NULL) {
1211 error = SET_ERROR(EINVAL);
1212 break;
1213 }
1214 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1215
1216 error = zil_suspend(fsname, &cookie);
1217 if (error != 0)
1218 break;
1219 fnvlist_add_uint64(suspended, fsname,
1220 (uintptr_t)cookie);
1221 }
1222 }
1223
1224 ddsa.ddsa_snaps = snaps;
1225 ddsa.ddsa_props = props;
1226 ddsa.ddsa_errors = errors;
1227
1228 if (error == 0) {
1229 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1230 dsl_dataset_snapshot_sync, &ddsa,
1231 fnvlist_num_pairs(snaps) * 3);
1232 }
1233
1234 if (suspended != NULL) {
1235 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1236 pair = nvlist_next_nvpair(suspended, pair)) {
1237 zil_resume((void *)(uintptr_t)
1238 fnvpair_value_uint64(pair));
1239 }
1240 fnvlist_free(suspended);
1241 }
1242
1243 return (error);
1244 }
1245
1246 typedef struct dsl_dataset_snapshot_tmp_arg {
1247 const char *ddsta_fsname;
1248 const char *ddsta_snapname;
1249 minor_t ddsta_cleanup_minor;
1250 const char *ddsta_htag;
1251 } dsl_dataset_snapshot_tmp_arg_t;
1252
1253 static int
1254 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1255 {
1256 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1257 dsl_pool_t *dp = dmu_tx_pool(tx);
1258 dsl_dataset_t *ds;
1259 int error;
1260
1261 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1262 if (error != 0)
1263 return (error);
1264
1265 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx);
1266 if (error != 0) {
1267 dsl_dataset_rele(ds, FTAG);
1268 return (error);
1269 }
1270
1271 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1272 dsl_dataset_rele(ds, FTAG);
1273 return (SET_ERROR(ENOTSUP));
1274 }
1275 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1276 B_TRUE, tx);
1277 if (error != 0) {
1278 dsl_dataset_rele(ds, FTAG);
1279 return (error);
1280 }
1281
1282 dsl_dataset_rele(ds, FTAG);
1283 return (0);
1284 }
1285
1286 static void
1287 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1288 {
1289 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1290 dsl_pool_t *dp = dmu_tx_pool(tx);
1291 dsl_dataset_t *ds;
1292
1293 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1294
1295 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1296 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1297 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1298 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1299
1300 dsl_dataset_rele(ds, FTAG);
1301 }
1302
1303 int
1304 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1305 minor_t cleanup_minor, const char *htag)
1306 {
1307 dsl_dataset_snapshot_tmp_arg_t ddsta;
1308 int error;
1309 spa_t *spa;
1310 boolean_t needsuspend;
1311 void *cookie;
1312
1313 ddsta.ddsta_fsname = fsname;
1314 ddsta.ddsta_snapname = snapname;
1315 ddsta.ddsta_cleanup_minor = cleanup_minor;
1316 ddsta.ddsta_htag = htag;
1317
1318 error = spa_open(fsname, &spa, FTAG);
1319 if (error != 0)
1320 return (error);
1321 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1322 spa_close(spa, FTAG);
1323
1324 if (needsuspend) {
1325 error = zil_suspend(fsname, &cookie);
1326 if (error != 0)
1327 return (error);
1328 }
1329
1330 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1331 dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1332
1333 if (needsuspend)
1334 zil_resume(cookie);
1335 return (error);
1336 }
1337
1338
1339 void
1340 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1341 {
1342 ASSERT(dmu_tx_is_syncing(tx));
1343 ASSERT(ds->ds_objset != NULL);
1344 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1345
1346 /*
1347 * in case we had to change ds_fsid_guid when we opened it,
1348 * sync it out now.
1349 */
1350 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1351 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1352
1353 dmu_objset_sync(ds->ds_objset, zio, tx);
1354 }
1355
1356 static void
1357 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1358 {
1359 uint64_t count = 0;
1360 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1361 zap_cursor_t zc;
1362 zap_attribute_t za;
1363 nvlist_t *propval = fnvlist_alloc();
1364 nvlist_t *val = fnvlist_alloc();
1365
1366 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1367
1368 /*
1369 * There may be missing entries in ds_next_clones_obj
1370 * due to a bug in a previous version of the code.
1371 * Only trust it if it has the right number of entries.
1372 */
1373 if (ds->ds_phys->ds_next_clones_obj != 0) {
1374 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1375 &count));
1376 }
1377 if (count != ds->ds_phys->ds_num_children - 1)
1378 goto fail;
1379 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1380 zap_cursor_retrieve(&zc, &za) == 0;
1381 zap_cursor_advance(&zc)) {
1382 dsl_dataset_t *clone;
1383 char buf[ZFS_MAXNAMELEN];
1384 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1385 za.za_first_integer, FTAG, &clone));
1386 dsl_dir_name(clone->ds_dir, buf);
1387 fnvlist_add_boolean(val, buf);
1388 dsl_dataset_rele(clone, FTAG);
1389 }
1390 zap_cursor_fini(&zc);
1391 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1392 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1393 fail:
1394 nvlist_free(val);
1395 nvlist_free(propval);
1396 }
1397
1398 void
1399 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1400 {
1401 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1402 uint64_t refd, avail, uobjs, aobjs, ratio;
1403
1404 ASSERT(dsl_pool_config_held(dp));
1405
1406 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1407 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1408 ds->ds_phys->ds_compressed_bytes);
1409
1410 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1411 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1412 ds->ds_phys->ds_uncompressed_bytes);
1413
1414 if (dsl_dataset_is_snapshot(ds)) {
1415 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1416 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1417 ds->ds_phys->ds_unique_bytes);
1418 get_clones_stat(ds, nv);
1419 } else {
1420 dsl_dir_stats(ds->ds_dir, nv);
1421 }
1422
1423 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1424 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1425 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1426
1427 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1428 ds->ds_phys->ds_creation_time);
1429 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1430 ds->ds_phys->ds_creation_txg);
1431 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1432 ds->ds_quota);
1433 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1434 ds->ds_reserved);
1435 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1436 ds->ds_phys->ds_guid);
1437 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1438 ds->ds_phys->ds_unique_bytes);
1439 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1440 ds->ds_object);
1441 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1442 ds->ds_userrefs);
1443 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1444 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1445
1446 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1447 uint64_t written, comp, uncomp;
1448 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1449 dsl_dataset_t *prev;
1450
1451 int err = dsl_dataset_hold_obj(dp,
1452 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1453 if (err == 0) {
1454 err = dsl_dataset_space_written(prev, ds, &written,
1455 &comp, &uncomp);
1456 dsl_dataset_rele(prev, FTAG);
1457 if (err == 0) {
1458 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1459 written);
1460 }
1461 }
1462 }
1463 }
1464
1465 void
1466 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1467 {
1468 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1469 ASSERT(dsl_pool_config_held(dp));
1470
1471 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1472 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1473 stat->dds_guid = ds->ds_phys->ds_guid;
1474 stat->dds_origin[0] = '\0';
1475 if (dsl_dataset_is_snapshot(ds)) {
1476 stat->dds_is_snapshot = B_TRUE;
1477 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1478 } else {
1479 stat->dds_is_snapshot = B_FALSE;
1480 stat->dds_num_clones = 0;
1481
1482 if (dsl_dir_is_clone(ds->ds_dir)) {
1483 dsl_dataset_t *ods;
1484
1485 VERIFY0(dsl_dataset_hold_obj(dp,
1486 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1487 dsl_dataset_name(ods, stat->dds_origin);
1488 dsl_dataset_rele(ods, FTAG);
1489 }
1490 }
1491 }
1492
1493 uint64_t
1494 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1495 {
1496 return (ds->ds_fsid_guid);
1497 }
1498
1499 void
1500 dsl_dataset_space(dsl_dataset_t *ds,
1501 uint64_t *refdbytesp, uint64_t *availbytesp,
1502 uint64_t *usedobjsp, uint64_t *availobjsp)
1503 {
1504 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
1505 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1506 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1507 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1508 if (ds->ds_quota != 0) {
1509 /*
1510 * Adjust available bytes according to refquota
1511 */
1512 if (*refdbytesp < ds->ds_quota)
1513 *availbytesp = MIN(*availbytesp,
1514 ds->ds_quota - *refdbytesp);
1515 else
1516 *availbytesp = 0;
1517 }
1518 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1519 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1520 }
1521
1522 boolean_t
1523 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1524 {
1525 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1526
1527 ASSERT(dsl_pool_config_held(dp));
1528 if (ds->ds_prev == NULL)
1529 return (B_FALSE);
1530 if (ds->ds_phys->ds_bp.blk_birth >
1531 ds->ds_prev->ds_phys->ds_creation_txg) {
1532 objset_t *os, *os_prev;
1533 /*
1534 * It may be that only the ZIL differs, because it was
1535 * reset in the head. Don't count that as being
1536 * modified.
1537 */
1538 if (dmu_objset_from_ds(ds, &os) != 0)
1539 return (B_TRUE);
1540 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
1541 return (B_TRUE);
1542 return (bcmp(&os->os_phys->os_meta_dnode,
1543 &os_prev->os_phys->os_meta_dnode,
1544 sizeof (os->os_phys->os_meta_dnode)) != 0);
1545 }
1546 return (B_FALSE);
1547 }
1548
1549 typedef struct dsl_dataset_rename_snapshot_arg {
1550 const char *ddrsa_fsname;
1551 const char *ddrsa_oldsnapname;
1552 const char *ddrsa_newsnapname;
1553 boolean_t ddrsa_recursive;
1554 dmu_tx_t *ddrsa_tx;
1555 } dsl_dataset_rename_snapshot_arg_t;
1556
1557 /* ARGSUSED */
1558 static int
1559 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1560 dsl_dataset_t *hds, void *arg)
1561 {
1562 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1563 int error;
1564 uint64_t val;
1565
1566 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1567 if (error != 0) {
1568 /* ignore nonexistent snapshots */
1569 return (error == ENOENT ? 0 : error);
1570 }
1571
1572 /* new name should not exist */
1573 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1574 if (error == 0)
1575 error = SET_ERROR(EEXIST);
1576 else if (error == ENOENT)
1577 error = 0;
1578
1579 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1580 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1581 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1582 error = SET_ERROR(ENAMETOOLONG);
1583
1584 return (error);
1585 }
1586
1587 static int
1588 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1589 {
1590 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1591 dsl_pool_t *dp = dmu_tx_pool(tx);
1592 dsl_dataset_t *hds;
1593 int error;
1594
1595 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1596 if (error != 0)
1597 return (error);
1598
1599 if (ddrsa->ddrsa_recursive) {
1600 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1601 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1602 DS_FIND_CHILDREN);
1603 } else {
1604 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1605 }
1606 dsl_dataset_rele(hds, FTAG);
1607 return (error);
1608 }
1609
1610 static int
1611 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1612 dsl_dataset_t *hds, void *arg)
1613 {
1614 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1615 dsl_dataset_t *ds;
1616 uint64_t val;
1617 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1618 int error;
1619
1620 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1621 ASSERT(error == 0 || error == ENOENT);
1622 if (error == ENOENT) {
1623 /* ignore nonexistent snapshots */
1624 return (0);
1625 }
1626
1627 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1628
1629 /* log before we change the name */
1630 spa_history_log_internal_ds(ds, "rename", tx,
1631 "-> @%s", ddrsa->ddrsa_newsnapname);
1632
1633 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1634 mutex_enter(&ds->ds_lock);
1635 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1636 mutex_exit(&ds->ds_lock);
1637 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1638 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1639
1640 dsl_dataset_rele(ds, FTAG);
1641 return (0);
1642 }
1643
1644 static void
1645 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1646 {
1647 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1648 dsl_pool_t *dp = dmu_tx_pool(tx);
1649 dsl_dataset_t *hds;
1650
1651 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1652 ddrsa->ddrsa_tx = tx;
1653 if (ddrsa->ddrsa_recursive) {
1654 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1655 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1656 DS_FIND_CHILDREN));
1657 } else {
1658 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1659 }
1660 dsl_dataset_rele(hds, FTAG);
1661 }
1662
1663 int
1664 dsl_dataset_rename_snapshot(const char *fsname,
1665 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1666 {
1667 dsl_dataset_rename_snapshot_arg_t ddrsa;
1668
1669 ddrsa.ddrsa_fsname = fsname;
1670 ddrsa.ddrsa_oldsnapname = oldsnapname;
1671 ddrsa.ddrsa_newsnapname = newsnapname;
1672 ddrsa.ddrsa_recursive = recursive;
1673
1674 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1675 dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1676 }
1677
1678 static int
1679 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1680 {
1681 const char *fsname = arg;
1682 dsl_pool_t *dp = dmu_tx_pool(tx);
1683 dsl_dataset_t *ds;
1684 int64_t unused_refres_delta;
1685 int error;
1686
1687 error = dsl_dataset_hold(dp, fsname, FTAG, &ds);
1688 if (error != 0)
1689 return (error);
1690
1691 /* must not be a snapshot */
1692 if (dsl_dataset_is_snapshot(ds)) {
1693 dsl_dataset_rele(ds, FTAG);
1694 return (SET_ERROR(EINVAL));
1695 }
1696
1697 /* must have a most recent snapshot */
1698 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1699 dsl_dataset_rele(ds, FTAG);
1700 return (SET_ERROR(EINVAL));
1701 }
1702
1703 if (dsl_dataset_long_held(ds)) {
1704 dsl_dataset_rele(ds, FTAG);
1705 return (SET_ERROR(EBUSY));
1706 }
1707
1708 /*
1709 * Check if the snap we are rolling back to uses more than
1710 * the refquota.
1711 */
1712 if (ds->ds_quota != 0 &&
1713 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1714 dsl_dataset_rele(ds, FTAG);
1715 return (SET_ERROR(EDQUOT));
1716 }
1717
1718 /*
1719 * When we do the clone swap, we will temporarily use more space
1720 * due to the refreservation (the head will no longer have any
1721 * unique space, so the entire amount of the refreservation will need
1722 * to be free). We will immediately destroy the clone, freeing
1723 * this space, but the freeing happens over many txg's.
1724 */
1725 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1726 ds->ds_phys->ds_unique_bytes);
1727
1728 if (unused_refres_delta > 0 &&
1729 unused_refres_delta >
1730 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1731 dsl_dataset_rele(ds, FTAG);
1732 return (SET_ERROR(ENOSPC));
1733 }
1734
1735 dsl_dataset_rele(ds, FTAG);
1736 return (0);
1737 }
1738
1739 static void
1740 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1741 {
1742 const char *fsname = arg;
1743 dsl_pool_t *dp = dmu_tx_pool(tx);
1744 dsl_dataset_t *ds, *clone;
1745 uint64_t cloneobj;
1746
1747 VERIFY0(dsl_dataset_hold(dp, fsname, FTAG, &ds));
1748
1749 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1750 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1751
1752 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1753
1754 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1755 dsl_dataset_zero_zil(ds, tx);
1756
1757 dsl_destroy_head_sync_impl(clone, tx);
1758
1759 dsl_dataset_rele(clone, FTAG);
1760 dsl_dataset_rele(ds, FTAG);
1761 }
1762
1763 int
1764 dsl_dataset_rollback(const char *fsname)
1765 {
1766 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1767 dsl_dataset_rollback_sync, (void *)fsname, 1));
1768 }
1769
1770 struct promotenode {
1771 list_node_t link;
1772 dsl_dataset_t *ds;
1773 };
1774
1775 typedef struct dsl_dataset_promote_arg {
1776 const char *ddpa_clonename;
1777 dsl_dataset_t *ddpa_clone;
1778 list_t shared_snaps, origin_snaps, clone_snaps;
1779 dsl_dataset_t *origin_origin; /* origin of the origin */
1780 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1781 char *err_ds;
1782 } dsl_dataset_promote_arg_t;
1783
1784 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1785 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1786 void *tag);
1787 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1788
1789 static int
1790 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1791 {
1792 dsl_dataset_promote_arg_t *ddpa = arg;
1793 dsl_pool_t *dp = dmu_tx_pool(tx);
1794 dsl_dataset_t *hds;
1795 struct promotenode *snap;
1796 dsl_dataset_t *origin_ds;
1797 int err;
1798 uint64_t unused;
1799
1800 err = promote_hold(ddpa, dp, FTAG);
1801 if (err != 0)
1802 return (err);
1803
1804 hds = ddpa->ddpa_clone;
1805
1806 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1807 promote_rele(ddpa, FTAG);
1808 return (SET_ERROR(EXDEV));
1809 }
1810
1811 /*
1812 * Compute and check the amount of space to transfer. Since this is
1813 * so expensive, don't do the preliminary check.
1814 */
1815 if (!dmu_tx_is_syncing(tx)) {
1816 promote_rele(ddpa, FTAG);
1817 return (0);
1818 }
1819
1820 snap = list_head(&ddpa->shared_snaps);
1821 origin_ds = snap->ds;
1822
1823 /* compute origin's new unique space */
1824 snap = list_tail(&ddpa->clone_snaps);
1825 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1826 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1827 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1828 &ddpa->unique, &unused, &unused);
1829
1830 /*
1831 * Walk the snapshots that we are moving
1832 *
1833 * Compute space to transfer. Consider the incremental changes
1834 * to used by each snapshot:
1835 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1836 * So each snapshot gave birth to:
1837 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1838 * So a sequence would look like:
1839 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1840 * Which simplifies to:
1841 * uN + kN + kN-1 + ... + k1 + k0
1842 * Note however, if we stop before we reach the ORIGIN we get:
1843 * uN + kN + kN-1 + ... + kM - uM-1
1844 */
1845 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1846 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1847 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1848 for (snap = list_head(&ddpa->shared_snaps); snap;
1849 snap = list_next(&ddpa->shared_snaps, snap)) {
1850 uint64_t val, dlused, dlcomp, dluncomp;
1851 dsl_dataset_t *ds = snap->ds;
1852
1853 /*
1854 * If there are long holds, we won't be able to evict
1855 * the objset.
1856 */
1857 if (dsl_dataset_long_held(ds)) {
1858 err = SET_ERROR(EBUSY);
1859 goto out;
1860 }
1861
1862 /* Check that the snapshot name does not conflict */
1863 VERIFY0(dsl_dataset_get_snapname(ds));
1864 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1865 if (err == 0) {
1866 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1867 err = SET_ERROR(EEXIST);
1868 goto out;
1869 }
1870 if (err != ENOENT)
1871 goto out;
1872
1873 /* The very first snapshot does not have a deadlist */
1874 if (ds->ds_phys->ds_prev_snap_obj == 0)
1875 continue;
1876
1877 dsl_deadlist_space(&ds->ds_deadlist,
1878 &dlused, &dlcomp, &dluncomp);
1879 ddpa->used += dlused;
1880 ddpa->comp += dlcomp;
1881 ddpa->uncomp += dluncomp;
1882 }
1883
1884 /*
1885 * If we are a clone of a clone then we never reached ORIGIN,
1886 * so we need to subtract out the clone origin's used space.
1887 */
1888 if (ddpa->origin_origin) {
1889 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1890 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1891 ddpa->uncomp -=
1892 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1893 }
1894
1895 /* Check that there is enough space here */
1896 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1897 ddpa->used);
1898 if (err != 0)
1899 goto out;
1900
1901 /*
1902 * Compute the amounts of space that will be used by snapshots
1903 * after the promotion (for both origin and clone). For each,
1904 * it is the amount of space that will be on all of their
1905 * deadlists (that was not born before their new origin).
1906 */
1907 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1908 uint64_t space;
1909
1910 /*
1911 * Note, typically this will not be a clone of a clone,
1912 * so dd_origin_txg will be < TXG_INITIAL, so
1913 * these snaplist_space() -> dsl_deadlist_space_range()
1914 * calls will be fast because they do not have to
1915 * iterate over all bps.
1916 */
1917 snap = list_head(&ddpa->origin_snaps);
1918 err = snaplist_space(&ddpa->shared_snaps,
1919 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1920 if (err != 0)
1921 goto out;
1922
1923 err = snaplist_space(&ddpa->clone_snaps,
1924 snap->ds->ds_dir->dd_origin_txg, &space);
1925 if (err != 0)
1926 goto out;
1927 ddpa->cloneusedsnap += space;
1928 }
1929 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1930 err = snaplist_space(&ddpa->origin_snaps,
1931 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
1932 if (err != 0)
1933 goto out;
1934 }
1935
1936 out:
1937 promote_rele(ddpa, FTAG);
1938 return (err);
1939 }
1940
1941 static void
1942 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
1943 {
1944 dsl_dataset_promote_arg_t *ddpa = arg;
1945 dsl_pool_t *dp = dmu_tx_pool(tx);
1946 dsl_dataset_t *hds;
1947 struct promotenode *snap;
1948 dsl_dataset_t *origin_ds;
1949 dsl_dataset_t *origin_head;
1950 dsl_dir_t *dd;
1951 dsl_dir_t *odd = NULL;
1952 uint64_t oldnext_obj;
1953 int64_t delta;
1954
1955 VERIFY0(promote_hold(ddpa, dp, FTAG));
1956 hds = ddpa->ddpa_clone;
1957
1958 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
1959
1960 snap = list_head(&ddpa->shared_snaps);
1961 origin_ds = snap->ds;
1962 dd = hds->ds_dir;
1963
1964 snap = list_head(&ddpa->origin_snaps);
1965 origin_head = snap->ds;
1966
1967 /*
1968 * We need to explicitly open odd, since origin_ds's dd will be
1969 * changing.
1970 */
1971 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
1972 NULL, FTAG, &odd));
1973
1974 /* change origin's next snap */
1975 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
1976 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
1977 snap = list_tail(&ddpa->clone_snaps);
1978 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1979 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
1980
1981 /* change the origin's next clone */
1982 if (origin_ds->ds_phys->ds_next_clones_obj) {
1983 dsl_dataset_remove_from_next_clones(origin_ds,
1984 snap->ds->ds_object, tx);
1985 VERIFY0(zap_add_int(dp->dp_meta_objset,
1986 origin_ds->ds_phys->ds_next_clones_obj,
1987 oldnext_obj, tx));
1988 }
1989
1990 /* change origin */
1991 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1992 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
1993 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
1994 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
1995 dmu_buf_will_dirty(odd->dd_dbuf, tx);
1996 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
1997 origin_head->ds_dir->dd_origin_txg =
1998 origin_ds->ds_phys->ds_creation_txg;
1999
2000 /* change dd_clone entries */
2001 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2002 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2003 odd->dd_phys->dd_clones, hds->ds_object, tx));
2004 VERIFY0(zap_add_int(dp->dp_meta_objset,
2005 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2006 hds->ds_object, tx));
2007
2008 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2009 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2010 origin_head->ds_object, tx));
2011 if (dd->dd_phys->dd_clones == 0) {
2012 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2013 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2014 }
2015 VERIFY0(zap_add_int(dp->dp_meta_objset,
2016 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2017 }
2018
2019 /* move snapshots to this dir */
2020 for (snap = list_head(&ddpa->shared_snaps); snap;
2021 snap = list_next(&ddpa->shared_snaps, snap)) {
2022 dsl_dataset_t *ds = snap->ds;
2023
2024 /*
2025 * Property callbacks are registered to a particular
2026 * dsl_dir. Since ours is changing, evict the objset
2027 * so that they will be unregistered from the old dsl_dir.
2028 */
2029 if (ds->ds_objset) {
2030 dmu_objset_evict(ds->ds_objset);
2031 ds->ds_objset = NULL;
2032 }
2033
2034 /* move snap name entry */
2035 VERIFY0(dsl_dataset_get_snapname(ds));
2036 VERIFY0(dsl_dataset_snap_remove(origin_head,
2037 ds->ds_snapname, tx));
2038 VERIFY0(zap_add(dp->dp_meta_objset,
2039 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2040 8, 1, &ds->ds_object, tx));
2041
2042 /* change containing dsl_dir */
2043 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2044 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2045 ds->ds_phys->ds_dir_obj = dd->dd_object;
2046 ASSERT3P(ds->ds_dir, ==, odd);
2047 dsl_dir_rele(ds->ds_dir, ds);
2048 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2049 NULL, ds, &ds->ds_dir));
2050
2051 /* move any clone references */
2052 if (ds->ds_phys->ds_next_clones_obj &&
2053 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2054 zap_cursor_t zc;
2055 zap_attribute_t za;
2056
2057 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2058 ds->ds_phys->ds_next_clones_obj);
2059 zap_cursor_retrieve(&zc, &za) == 0;
2060 zap_cursor_advance(&zc)) {
2061 dsl_dataset_t *cnds;
2062 uint64_t o;
2063
2064 if (za.za_first_integer == oldnext_obj) {
2065 /*
2066 * We've already moved the
2067 * origin's reference.
2068 */
2069 continue;
2070 }
2071
2072 VERIFY0(dsl_dataset_hold_obj(dp,
2073 za.za_first_integer, FTAG, &cnds));
2074 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2075
2076 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2077 odd->dd_phys->dd_clones, o, tx));
2078 VERIFY0(zap_add_int(dp->dp_meta_objset,
2079 dd->dd_phys->dd_clones, o, tx));
2080 dsl_dataset_rele(cnds, FTAG);
2081 }
2082 zap_cursor_fini(&zc);
2083 }
2084
2085 ASSERT(!dsl_prop_hascb(ds));
2086 }
2087
2088 /*
2089 * Change space accounting.
2090 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2091 * both be valid, or both be 0 (resulting in delta == 0). This
2092 * is true for each of {clone,origin} independently.
2093 */
2094
2095 delta = ddpa->cloneusedsnap -
2096 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2097 ASSERT3S(delta, >=, 0);
2098 ASSERT3U(ddpa->used, >=, delta);
2099 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2100 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2101 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2102
2103 delta = ddpa->originusedsnap -
2104 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2105 ASSERT3S(delta, <=, 0);
2106 ASSERT3U(ddpa->used, >=, -delta);
2107 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2108 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2109 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2110
2111 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2112
2113 /* log history record */
2114 spa_history_log_internal_ds(hds, "promote", tx, "");
2115
2116 dsl_dir_rele(odd, FTAG);
2117 promote_rele(ddpa, FTAG);
2118 }
2119
2120 /*
2121 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2122 * (exclusive) and last_obj (inclusive). The list will be in reverse
2123 * order (last_obj will be the list_head()). If first_obj == 0, do all
2124 * snapshots back to this dataset's origin.
2125 */
2126 static int
2127 snaplist_make(dsl_pool_t *dp,
2128 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2129 {
2130 uint64_t obj = last_obj;
2131
2132 list_create(l, sizeof (struct promotenode),
2133 offsetof(struct promotenode, link));
2134
2135 while (obj != first_obj) {
2136 dsl_dataset_t *ds;
2137 struct promotenode *snap;
2138 int err;
2139
2140 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2141 ASSERT(err != ENOENT);
2142 if (err != 0)
2143 return (err);
2144
2145 if (first_obj == 0)
2146 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2147
2148 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2149 snap->ds = ds;
2150 list_insert_tail(l, snap);
2151 obj = ds->ds_phys->ds_prev_snap_obj;
2152 }
2153
2154 return (0);
2155 }
2156
2157 static int
2158 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2159 {
2160 struct promotenode *snap;
2161
2162 *spacep = 0;
2163 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2164 uint64_t used, comp, uncomp;
2165 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2166 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2167 *spacep += used;
2168 }
2169 return (0);
2170 }
2171
2172 static void
2173 snaplist_destroy(list_t *l, void *tag)
2174 {
2175 struct promotenode *snap;
2176
2177 if (l == NULL || !list_link_active(&l->list_head))
2178 return;
2179
2180 while ((snap = list_tail(l)) != NULL) {
2181 list_remove(l, snap);
2182 dsl_dataset_rele(snap->ds, tag);
2183 kmem_free(snap, sizeof (*snap));
2184 }
2185 list_destroy(l);
2186 }
2187
2188 static int
2189 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2190 {
2191 int error;
2192 dsl_dir_t *dd;
2193 struct promotenode *snap;
2194
2195 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2196 &ddpa->ddpa_clone);
2197 if (error != 0)
2198 return (error);
2199 dd = ddpa->ddpa_clone->ds_dir;
2200
2201 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2202 !dsl_dir_is_clone(dd)) {
2203 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2204 return (SET_ERROR(EINVAL));
2205 }
2206
2207 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2208 &ddpa->shared_snaps, tag);
2209 if (error != 0)
2210 goto out;
2211
2212 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2213 &ddpa->clone_snaps, tag);
2214 if (error != 0)
2215 goto out;
2216
2217 snap = list_head(&ddpa->shared_snaps);
2218 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2219 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2220 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2221 &ddpa->origin_snaps, tag);
2222 if (error != 0)
2223 goto out;
2224
2225 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2226 error = dsl_dataset_hold_obj(dp,
2227 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2228 tag, &ddpa->origin_origin);
2229 if (error != 0)
2230 goto out;
2231 }
2232 out:
2233 if (error != 0)
2234 promote_rele(ddpa, tag);
2235 return (error);
2236 }
2237
2238 static void
2239 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2240 {
2241 snaplist_destroy(&ddpa->shared_snaps, tag);
2242 snaplist_destroy(&ddpa->clone_snaps, tag);
2243 snaplist_destroy(&ddpa->origin_snaps, tag);
2244 if (ddpa->origin_origin != NULL)
2245 dsl_dataset_rele(ddpa->origin_origin, tag);
2246 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2247 }
2248
2249 /*
2250 * Promote a clone.
2251 *
2252 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2253 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2254 */
2255 int
2256 dsl_dataset_promote(const char *name, char *conflsnap)
2257 {
2258 dsl_dataset_promote_arg_t ddpa = { 0 };
2259 uint64_t numsnaps;
2260 int error;
2261 objset_t *os;
2262
2263 /*
2264 * We will modify space proportional to the number of
2265 * snapshots. Compute numsnaps.
2266 */
2267 error = dmu_objset_hold(name, FTAG, &os);
2268 if (error != 0)
2269 return (error);
2270 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2271 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2272 dmu_objset_rele(os, FTAG);
2273 if (error != 0)
2274 return (error);
2275
2276 ddpa.ddpa_clonename = name;
2277 ddpa.err_ds = conflsnap;
2278
2279 return (dsl_sync_task(name, dsl_dataset_promote_check,
2280 dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2281 }
2282
2283 int
2284 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2285 dsl_dataset_t *origin_head, boolean_t force)
2286 {
2287 int64_t unused_refres_delta;
2288
2289 /* they should both be heads */
2290 if (dsl_dataset_is_snapshot(clone) ||
2291 dsl_dataset_is_snapshot(origin_head))
2292 return (SET_ERROR(EINVAL));
2293
2294 /* the branch point should be just before them */
2295 if (clone->ds_prev != origin_head->ds_prev)
2296 return (SET_ERROR(EINVAL));
2297
2298 /* clone should be the clone (unless they are unrelated) */
2299 if (clone->ds_prev != NULL &&
2300 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2301 origin_head->ds_object !=
2302 clone->ds_prev->ds_phys->ds_next_snap_obj)
2303 return (SET_ERROR(EINVAL));
2304
2305 /* the clone should be a child of the origin */
2306 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2307 return (SET_ERROR(EINVAL));
2308
2309 /* origin_head shouldn't be modified unless 'force' */
2310 if (!force && dsl_dataset_modified_since_lastsnap(origin_head))
2311 return (SET_ERROR(ETXTBSY));
2312
2313 /* origin_head should have no long holds (e.g. is not mounted) */
2314 if (dsl_dataset_long_held(origin_head))
2315 return (SET_ERROR(EBUSY));
2316
2317 /* check amount of any unconsumed refreservation */
2318 unused_refres_delta =
2319 (int64_t)MIN(origin_head->ds_reserved,
2320 origin_head->ds_phys->ds_unique_bytes) -
2321 (int64_t)MIN(origin_head->ds_reserved,
2322 clone->ds_phys->ds_unique_bytes);
2323
2324 if (unused_refres_delta > 0 &&
2325 unused_refres_delta >
2326 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2327 return (SET_ERROR(ENOSPC));
2328
2329 /* clone can't be over the head's refquota */
2330 if (origin_head->ds_quota != 0 &&
2331 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2332 return (SET_ERROR(EDQUOT));
2333
2334 return (0);
2335 }
2336
2337 void
2338 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2339 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2340 {
2341 dsl_pool_t *dp = dmu_tx_pool(tx);
2342 int64_t unused_refres_delta;
2343
2344 ASSERT(clone->ds_reserved == 0);
2345 ASSERT(origin_head->ds_quota == 0 ||
2346 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2347
2348 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2349 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2350
2351 if (clone->ds_objset != NULL) {
2352 dmu_objset_evict(clone->ds_objset);
2353 clone->ds_objset = NULL;
2354 }
2355
2356 if (origin_head->ds_objset != NULL) {
2357 dmu_objset_evict(origin_head->ds_objset);
2358 origin_head->ds_objset = NULL;
2359 }
2360
2361 unused_refres_delta =
2362 (int64_t)MIN(origin_head->ds_reserved,
2363 origin_head->ds_phys->ds_unique_bytes) -
2364 (int64_t)MIN(origin_head->ds_reserved,
2365 clone->ds_phys->ds_unique_bytes);
2366
2367 /*
2368 * Reset origin's unique bytes, if it exists.
2369 */
2370 if (clone->ds_prev) {
2371 dsl_dataset_t *origin = clone->ds_prev;
2372 uint64_t comp, uncomp;
2373
2374 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2375 dsl_deadlist_space_range(&clone->ds_deadlist,
2376 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2377 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2378 }
2379
2380 /* swap blkptrs */
2381 {
2382 blkptr_t tmp;
2383 tmp = origin_head->ds_phys->ds_bp;
2384 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2385 clone->ds_phys->ds_bp = tmp;
2386 }
2387
2388 /* set dd_*_bytes */
2389 {
2390 int64_t dused, dcomp, duncomp;
2391 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2392 uint64_t odl_used, odl_comp, odl_uncomp;
2393
2394 ASSERT3U(clone->ds_dir->dd_phys->
2395 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2396
2397 dsl_deadlist_space(&clone->ds_deadlist,
2398 &cdl_used, &cdl_comp, &cdl_uncomp);
2399 dsl_deadlist_space(&origin_head->ds_deadlist,
2400 &odl_used, &odl_comp, &odl_uncomp);
2401
2402 dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2403 (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2404 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2405 (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2406 duncomp = clone->ds_phys->ds_uncompressed_bytes +
2407 cdl_uncomp -
2408 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2409
2410 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2411 dused, dcomp, duncomp, tx);
2412 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2413 -dused, -dcomp, -duncomp, tx);
2414
2415 /*
2416 * The difference in the space used by snapshots is the
2417 * difference in snapshot space due to the head's
2418 * deadlist (since that's the only thing that's
2419 * changing that affects the snapused).
2420 */
2421 dsl_deadlist_space_range(&clone->ds_deadlist,
2422 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2423 &cdl_used, &cdl_comp, &cdl_uncomp);
2424 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2425 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2426 &odl_used, &odl_comp, &odl_uncomp);
2427 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2428 DD_USED_HEAD, DD_USED_SNAP, tx);
2429 }
2430
2431 /* swap ds_*_bytes */
2432 SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2433 clone->ds_phys->ds_referenced_bytes);
2434 SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2435 clone->ds_phys->ds_compressed_bytes);
2436 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2437 clone->ds_phys->ds_uncompressed_bytes);
2438 SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2439 clone->ds_phys->ds_unique_bytes);
2440
2441 /* apply any parent delta for change in unconsumed refreservation */
2442 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2443 unused_refres_delta, 0, 0, tx);
2444
2445 /*
2446 * Swap deadlists.
2447 */
2448 dsl_deadlist_close(&clone->ds_deadlist);
2449 dsl_deadlist_close(&origin_head->ds_deadlist);
2450 SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2451 clone->ds_phys->ds_deadlist_obj);
2452 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2453 clone->ds_phys->ds_deadlist_obj);
2454 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2455 origin_head->ds_phys->ds_deadlist_obj);
2456
2457 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2458
2459 spa_history_log_internal_ds(clone, "clone swap", tx,
2460 "parent=%s", origin_head->ds_dir->dd_myname);
2461 }
2462
2463 /*
2464 * Given a pool name and a dataset object number in that pool,
2465 * return the name of that dataset.
2466 */
2467 int
2468 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2469 {
2470 dsl_pool_t *dp;
2471 dsl_dataset_t *ds;
2472 int error;
2473
2474 error = dsl_pool_hold(pname, FTAG, &dp);
2475 if (error != 0)
2476 return (error);
2477
2478 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2479 if (error == 0) {
2480 dsl_dataset_name(ds, buf);
2481 dsl_dataset_rele(ds, FTAG);
2482 }
2483 dsl_pool_rele(dp, FTAG);
2484
2485 return (error);
2486 }
2487
2488 int
2489 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2490 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2491 {
2492 int error = 0;
2493
2494 ASSERT3S(asize, >, 0);
2495
2496 /*
2497 * *ref_rsrv is the portion of asize that will come from any
2498 * unconsumed refreservation space.
2499 */
2500 *ref_rsrv = 0;
2501
2502 mutex_enter(&ds->ds_lock);
2503 /*
2504 * Make a space adjustment for reserved bytes.
2505 */
2506 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2507 ASSERT3U(*used, >=,
2508 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2509 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2510 *ref_rsrv =
2511 asize - MIN(asize, parent_delta(ds, asize + inflight));
2512 }
2513
2514 if (!check_quota || ds->ds_quota == 0) {
2515 mutex_exit(&ds->ds_lock);
2516 return (0);
2517 }
2518 /*
2519 * If they are requesting more space, and our current estimate
2520 * is over quota, they get to try again unless the actual
2521 * on-disk is over quota and there are no pending changes (which
2522 * may free up space for us).
2523 */
2524 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2525 if (inflight > 0 ||
2526 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2527 error = SET_ERROR(ERESTART);
2528 else
2529 error = SET_ERROR(EDQUOT);
2530 }
2531 mutex_exit(&ds->ds_lock);
2532
2533 return (error);
2534 }
2535
2536 typedef struct dsl_dataset_set_qr_arg {
2537 const char *ddsqra_name;
2538 zprop_source_t ddsqra_source;
2539 uint64_t ddsqra_value;
2540 } dsl_dataset_set_qr_arg_t;
2541
2542
2543 /* ARGSUSED */
2544 static int
2545 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2546 {
2547 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2548 dsl_pool_t *dp = dmu_tx_pool(tx);
2549 dsl_dataset_t *ds;
2550 int error;
2551 uint64_t newval;
2552
2553 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2554 return (SET_ERROR(ENOTSUP));
2555
2556 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2557 if (error != 0)
2558 return (error);
2559
2560 if (dsl_dataset_is_snapshot(ds)) {
2561 dsl_dataset_rele(ds, FTAG);
2562 return (SET_ERROR(EINVAL));
2563 }
2564
2565 error = dsl_prop_predict(ds->ds_dir,
2566 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2567 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2568 if (error != 0) {
2569 dsl_dataset_rele(ds, FTAG);
2570 return (error);
2571 }
2572
2573 if (newval == 0) {
2574 dsl_dataset_rele(ds, FTAG);
2575 return (0);
2576 }
2577
2578 if (newval < ds->ds_phys->ds_referenced_bytes ||
2579 newval < ds->ds_reserved) {
2580 dsl_dataset_rele(ds, FTAG);
2581 return (SET_ERROR(ENOSPC));
2582 }
2583
2584 dsl_dataset_rele(ds, FTAG);
2585 return (0);
2586 }
2587
2588 static void
2589 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2590 {
2591 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2592 dsl_pool_t *dp = dmu_tx_pool(tx);
2593 dsl_dataset_t *ds;
2594 uint64_t newval;
2595
2596 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2597
2598 dsl_prop_set_sync_impl(ds,
2599 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2600 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2601 &ddsqra->ddsqra_value, tx);
2602
2603 VERIFY0(dsl_prop_get_int_ds(ds,
2604 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2605
2606 if (ds->ds_quota != newval) {
2607 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2608 ds->ds_quota = newval;
2609 }
2610 dsl_dataset_rele(ds, FTAG);
2611 }
2612
2613 int
2614 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2615 uint64_t refquota)
2616 {
2617 dsl_dataset_set_qr_arg_t ddsqra;
2618
2619 ddsqra.ddsqra_name = dsname;
2620 ddsqra.ddsqra_source = source;
2621 ddsqra.ddsqra_value = refquota;
2622
2623 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2624 dsl_dataset_set_refquota_sync, &ddsqra, 0));
2625 }
2626
2627 static int
2628 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2629 {
2630 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2631 dsl_pool_t *dp = dmu_tx_pool(tx);
2632 dsl_dataset_t *ds;
2633 int error;
2634 uint64_t newval, unique;
2635
2636 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2637 return (SET_ERROR(ENOTSUP));
2638
2639 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2640 if (error != 0)
2641 return (error);
2642
2643 if (dsl_dataset_is_snapshot(ds)) {
2644 dsl_dataset_rele(ds, FTAG);
2645 return (SET_ERROR(EINVAL));
2646 }
2647
2648 error = dsl_prop_predict(ds->ds_dir,
2649 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2650 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2651 if (error != 0) {
2652 dsl_dataset_rele(ds, FTAG);
2653 return (error);
2654 }
2655
2656 /*
2657 * If we are doing the preliminary check in open context, the
2658 * space estimates may be inaccurate.
2659 */
2660 if (!dmu_tx_is_syncing(tx)) {
2661 dsl_dataset_rele(ds, FTAG);
2662 return (0);
2663 }
2664
2665 mutex_enter(&ds->ds_lock);
2666 if (!DS_UNIQUE_IS_ACCURATE(ds))
2667 dsl_dataset_recalc_head_uniq(ds);
2668 unique = ds->ds_phys->ds_unique_bytes;
2669 mutex_exit(&ds->ds_lock);
2670
2671 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2672 uint64_t delta = MAX(unique, newval) -
2673 MAX(unique, ds->ds_reserved);
2674
2675 if (delta >
2676 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2677 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2678 dsl_dataset_rele(ds, FTAG);
2679 return (SET_ERROR(ENOSPC));
2680 }
2681 }
2682
2683 dsl_dataset_rele(ds, FTAG);
2684 return (0);
2685 }
2686
2687 void
2688 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2689 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2690 {
2691 uint64_t newval;
2692 uint64_t unique;
2693 int64_t delta;
2694
2695 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2696 source, sizeof (value), 1, &value, tx);
2697
2698 VERIFY0(dsl_prop_get_int_ds(ds,
2699 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2700
2701 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2702 mutex_enter(&ds->ds_dir->dd_lock);
2703 mutex_enter(&ds->ds_lock);
2704 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2705 unique = ds->ds_phys->ds_unique_bytes;
2706 delta = MAX(0, (int64_t)(newval - unique)) -
2707 MAX(0, (int64_t)(ds->ds_reserved - unique));
2708 ds->ds_reserved = newval;
2709 mutex_exit(&ds->ds_lock);
2710
2711 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2712 mutex_exit(&ds->ds_dir->dd_lock);
2713 }
2714
2715 static void
2716 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2717 {
2718 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2719 dsl_pool_t *dp = dmu_tx_pool(tx);
2720 dsl_dataset_t *ds;
2721
2722 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2723 dsl_dataset_set_refreservation_sync_impl(ds,
2724 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2725 dsl_dataset_rele(ds, FTAG);
2726 }
2727
2728 int
2729 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2730 uint64_t refreservation)
2731 {
2732 dsl_dataset_set_qr_arg_t ddsqra;
2733
2734 ddsqra.ddsqra_name = dsname;
2735 ddsqra.ddsqra_source = source;
2736 ddsqra.ddsqra_value = refreservation;
2737
2738 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2739 dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2740 }
2741
2742 /*
2743 * Return (in *usedp) the amount of space written in new that is not
2744 * present in oldsnap. New may be a snapshot or the head. Old must be
2745 * a snapshot before new, in new's filesystem (or its origin). If not then
2746 * fail and return EINVAL.
2747 *
2748 * The written space is calculated by considering two components: First, we
2749 * ignore any freed space, and calculate the written as new's used space
2750 * minus old's used space. Next, we add in the amount of space that was freed
2751 * between the two snapshots, thus reducing new's used space relative to old's.
2752 * Specifically, this is the space that was born before old->ds_creation_txg,
2753 * and freed before new (ie. on new's deadlist or a previous deadlist).
2754 *
2755 * space freed [---------------------]
2756 * snapshots ---O-------O--------O-------O------
2757 * oldsnap new
2758 */
2759 int
2760 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2761 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2762 {
2763 int err = 0;
2764 uint64_t snapobj;
2765 dsl_pool_t *dp = new->ds_dir->dd_pool;
2766
2767 ASSERT(dsl_pool_config_held(dp));
2768
2769 *usedp = 0;
2770 *usedp += new->ds_phys->ds_referenced_bytes;
2771 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2772
2773 *compp = 0;
2774 *compp += new->ds_phys->ds_compressed_bytes;
2775 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
2776
2777 *uncompp = 0;
2778 *uncompp += new->ds_phys->ds_uncompressed_bytes;
2779 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2780
2781 snapobj = new->ds_object;
2782 while (snapobj != oldsnap->ds_object) {
2783 dsl_dataset_t *snap;
2784 uint64_t used, comp, uncomp;
2785
2786 if (snapobj == new->ds_object) {
2787 snap = new;
2788 } else {
2789 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2790 if (err != 0)
2791 break;
2792 }
2793
2794 if (snap->ds_phys->ds_prev_snap_txg ==
2795 oldsnap->ds_phys->ds_creation_txg) {
2796 /*
2797 * The blocks in the deadlist can not be born after
2798 * ds_prev_snap_txg, so get the whole deadlist space,
2799 * which is more efficient (especially for old-format
2800 * deadlists). Unfortunately the deadlist code
2801 * doesn't have enough information to make this
2802 * optimization itself.
2803 */
2804 dsl_deadlist_space(&snap->ds_deadlist,
2805 &used, &comp, &uncomp);
2806 } else {
2807 dsl_deadlist_space_range(&snap->ds_deadlist,
2808 0, oldsnap->ds_phys->ds_creation_txg,
2809 &used, &comp, &uncomp);
2810 }
2811 *usedp += used;
2812 *compp += comp;
2813 *uncompp += uncomp;
2814
2815 /*
2816 * If we get to the beginning of the chain of snapshots
2817 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2818 * was not a snapshot of/before new.
2819 */
2820 snapobj = snap->ds_phys->ds_prev_snap_obj;
2821 if (snap != new)
2822 dsl_dataset_rele(snap, FTAG);
2823 if (snapobj == 0) {
2824 err = SET_ERROR(EINVAL);
2825 break;
2826 }
2827
2828 }
2829 return (err);
2830 }
2831
2832 /*
2833 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2834 * lastsnap, and all snapshots in between are deleted.
2835 *
2836 * blocks that would be freed [---------------------------]
2837 * snapshots ---O-------O--------O-------O--------O
2838 * firstsnap lastsnap
2839 *
2840 * This is the set of blocks that were born after the snap before firstsnap,
2841 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2842 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2843 * We calculate this by iterating over the relevant deadlists (from the snap
2844 * after lastsnap, backward to the snap after firstsnap), summing up the
2845 * space on the deadlist that was born after the snap before firstsnap.
2846 */
2847 int
2848 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2849 dsl_dataset_t *lastsnap,
2850 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2851 {
2852 int err = 0;
2853 uint64_t snapobj;
2854 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2855
2856 ASSERT(dsl_dataset_is_snapshot(firstsnap));
2857 ASSERT(dsl_dataset_is_snapshot(lastsnap));
2858
2859 /*
2860 * Check that the snapshots are in the same dsl_dir, and firstsnap
2861 * is before lastsnap.
2862 */
2863 if (firstsnap->ds_dir != lastsnap->ds_dir ||
2864 firstsnap->ds_phys->ds_creation_txg >
2865 lastsnap->ds_phys->ds_creation_txg)
2866 return (SET_ERROR(EINVAL));
2867
2868 *usedp = *compp = *uncompp = 0;
2869
2870 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2871 while (snapobj != firstsnap->ds_object) {
2872 dsl_dataset_t *ds;
2873 uint64_t used, comp, uncomp;
2874
2875 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2876 if (err != 0)
2877 break;
2878
2879 dsl_deadlist_space_range(&ds->ds_deadlist,
2880 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2881 &used, &comp, &uncomp);
2882 *usedp += used;
2883 *compp += comp;
2884 *uncompp += uncomp;
2885
2886 snapobj = ds->ds_phys->ds_prev_snap_obj;
2887 ASSERT3U(snapobj, !=, 0);
2888 dsl_dataset_rele(ds, FTAG);
2889 }
2890 return (err);
2891 }
2892
2893 /*
2894 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2895 * For example, they could both be snapshots of the same filesystem, and
2896 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
2897 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
2898 * filesystem. Or 'earlier' could be the origin's origin.
2899 */
2900 boolean_t
2901 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2902 {
2903 dsl_pool_t *dp = later->ds_dir->dd_pool;
2904 int error;
2905 boolean_t ret;
2906
2907 ASSERT(dsl_pool_config_held(dp));
2908
2909 if (earlier->ds_phys->ds_creation_txg >=
2910 later->ds_phys->ds_creation_txg)
2911 return (B_FALSE);
2912
2913 if (later->ds_dir == earlier->ds_dir)
2914 return (B_TRUE);
2915 if (!dsl_dir_is_clone(later->ds_dir))
2916 return (B_FALSE);
2917
2918 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2919 return (B_TRUE);
2920 dsl_dataset_t *origin;
2921 error = dsl_dataset_hold_obj(dp,
2922 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2923 if (error != 0)
2924 return (B_FALSE);
2925 ret = dsl_dataset_is_before(origin, earlier);
2926 dsl_dataset_rele(origin, FTAG);
2927 return (ret);
2928 }