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_impl(dsl_dataset_t *ds, boolean_t evict_deadlist)
245 {
246
247 ASSERT(ds->ds_owner == NULL);
248
249 ds->ds_dbuf = NULL;
250
251 unique_remove(ds->ds_fsid_guid);
252
253 if (ds->ds_objset != NULL)
254 dmu_objset_evict(ds->ds_objset);
255
256 if (ds->ds_prev) {
257 dsl_dataset_rele(ds->ds_prev, ds);
258 ds->ds_prev = NULL;
259 }
260
261 bplist_destroy(&ds->ds_pending_deadlist);
262 if (evict_deadlist)
263 dsl_deadlist_close(&ds->ds_deadlist);
264 if (ds->ds_dir)
265 dsl_dir_rele(ds->ds_dir, ds);
266
267 ASSERT(!list_link_active(&ds->ds_synced_link));
268
269 mutex_destroy(&ds->ds_lock);
270 mutex_destroy(&ds->ds_opening_lock);
271 refcount_destroy(&ds->ds_longholds);
272
273 kmem_free(ds, sizeof (dsl_dataset_t));
274 }
275
276 /* ARGSUSED */
277 static void
278 dsl_dataset_evict(dmu_buf_user_t *dbu)
279 {
280 dsl_dataset_evict_impl((dsl_dataset_t *)dbu, B_TRUE);
281 }
282
283 int
284 dsl_dataset_get_snapname(dsl_dataset_t *ds)
285 {
286 dsl_dataset_phys_t *headphys;
287 int err;
288 dmu_buf_t *headdbuf;
289 dsl_pool_t *dp = ds->ds_dir->dd_pool;
290 objset_t *mos = dp->dp_meta_objset;
291
292 if (ds->ds_snapname[0])
293 return (0);
294 if (ds->ds_phys->ds_next_snap_obj == 0)
295 return (0);
296
297 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
298 FTAG, &headdbuf);
299 if (err != 0)
300 return (err);
301 headphys = headdbuf->db_data;
302 err = zap_value_search(dp->dp_meta_objset,
303 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
304 dmu_buf_rele(headdbuf, FTAG);
305 return (err);
306 }
307
308 int
309 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
310 {
311 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
312 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
313 matchtype_t mt;
314 int err;
315
316 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
317 mt = MT_FIRST;
318 else
319 mt = MT_EXACT;
320
321 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
322 value, mt, NULL, 0, NULL);
323 if (err == ENOTSUP && mt == MT_FIRST)
324 err = zap_lookup(mos, snapobj, name, 8, 1, value);
325 return (err);
326 }
327
328 int
329 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx)
330 {
331 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
332 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
333 matchtype_t mt;
334 int err;
335
336 dsl_dir_snap_cmtime_update(ds->ds_dir);
337
338 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
339 mt = MT_FIRST;
340 else
341 mt = MT_EXACT;
342
343 err = zap_remove_norm(mos, snapobj, name, mt, tx);
344 if (err == ENOTSUP && mt == MT_FIRST)
345 err = zap_remove(mos, snapobj, name, tx);
346 return (err);
347 }
348
349 int
350 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
351 dsl_dataset_t **dsp)
352 {
353 objset_t *mos = dp->dp_meta_objset;
354 dmu_buf_t *dbuf;
355 dsl_dataset_t *ds;
356 int err;
357 dmu_object_info_t doi;
358
359 ASSERT(dsl_pool_config_held(dp));
360
361 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
362 if (err != 0)
363 return (err);
364
365 /* Make sure dsobj has the correct object type. */
366 dmu_object_info_from_db(dbuf, &doi);
367 if (doi.doi_type != DMU_OT_DSL_DATASET)
368 return (SET_ERROR(EINVAL));
369
370 ds = (dsl_dataset_t *)dmu_buf_get_user(dbuf);
371 if (ds == NULL) {
372 dsl_dataset_t *winner = NULL;
373
374 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
375 ds->ds_dbuf = dbuf;
376 ds->ds_object = dsobj;
377
378 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
379 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
380 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
381 refcount_create(&ds->ds_longholds);
382
383 bplist_create(&ds->ds_pending_deadlist);
384 dsl_deadlist_open(&ds->ds_deadlist,
385 mos, ds->ds_phys->ds_deadlist_obj);
386
387 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
388 offsetof(dmu_sendarg_t, dsa_link));
389
390 if (err == 0) {
391 err = dsl_dir_hold_obj(dp,
392 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
393 }
394 if (err != 0) {
395 mutex_destroy(&ds->ds_lock);
396 mutex_destroy(&ds->ds_opening_lock);
397 refcount_destroy(&ds->ds_longholds);
398 bplist_destroy(&ds->ds_pending_deadlist);
399 dsl_deadlist_close(&ds->ds_deadlist);
400 kmem_free(ds, sizeof (dsl_dataset_t));
401 dmu_buf_rele(dbuf, tag);
402 return (err);
403 }
404
405 if (!dsl_dataset_is_snapshot(ds)) {
406 ds->ds_snapname[0] = '\0';
407 if (ds->ds_phys->ds_prev_snap_obj != 0) {
408 err = dsl_dataset_hold_obj(dp,
409 ds->ds_phys->ds_prev_snap_obj,
410 ds, &ds->ds_prev);
411 }
412 } else {
413 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
414 err = dsl_dataset_get_snapname(ds);
415 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
416 err = zap_count(
417 ds->ds_dir->dd_pool->dp_meta_objset,
418 ds->ds_phys->ds_userrefs_obj,
419 &ds->ds_userrefs);
420 }
421 }
422
423 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
424 err = dsl_prop_get_int_ds(ds,
425 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
426 &ds->ds_reserved);
427 if (err == 0) {
428 err = dsl_prop_get_int_ds(ds,
429 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
430 &ds->ds_quota);
431 }
432 } else {
433 ds->ds_reserved = ds->ds_quota = 0;
434 }
435
436 dmu_buf_init_user(&ds->db_evict, dsl_dataset_evict);
437 if (err == 0)
438 winner = (dsl_dataset_t *)
439 dmu_buf_set_user_ie(dbuf, &ds->db_evict);
440
441 if (err || winner) {
442 bplist_destroy(&ds->ds_pending_deadlist);
443 dsl_deadlist_close(&ds->ds_deadlist);
444 if (ds->ds_prev)
445 dsl_dataset_rele(ds->ds_prev, ds);
446 dsl_dir_rele(ds->ds_dir, ds);
447 mutex_destroy(&ds->ds_lock);
448 mutex_destroy(&ds->ds_opening_lock);
449 refcount_destroy(&ds->ds_longholds);
450 kmem_free(ds, sizeof (dsl_dataset_t));
451 if (err != 0) {
452 dmu_buf_rele(dbuf, tag);
453 return (err);
454 }
455 ds = winner;
456 } else {
457 ds->ds_fsid_guid =
458 unique_insert(ds->ds_phys->ds_fsid_guid);
459 }
460 }
461 ASSERT3P(ds->ds_dbuf, ==, dbuf);
462 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
463 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
464 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
465 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
466 *dsp = ds;
467 return (0);
468 }
469
470 int
471 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
472 void *tag, dsl_dataset_t **dsp)
473 {
474 dsl_dir_t *dd;
475 const char *snapname;
476 uint64_t obj;
477 int err = 0;
478
479 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
480 if (err != 0)
481 return (err);
482
483 ASSERT(dsl_pool_config_held(dp));
484 obj = dd->dd_phys->dd_head_dataset_obj;
485 if (obj != 0)
486 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
487 else
488 err = SET_ERROR(ENOENT);
489
490 /* we may be looking for a snapshot */
491 if (err == 0 && snapname != NULL) {
492 dsl_dataset_t *ds;
493
494 if (*snapname++ != '@') {
495 dsl_dataset_rele(*dsp, tag);
496 dsl_dir_rele(dd, FTAG);
497 return (SET_ERROR(ENOENT));
498 }
499
500 dprintf("looking for snapshot '%s'\n", snapname);
501 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
502 if (err == 0)
503 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
504 dsl_dataset_rele(*dsp, tag);
505
506 if (err == 0) {
507 mutex_enter(&ds->ds_lock);
508 if (ds->ds_snapname[0] == 0)
509 (void) strlcpy(ds->ds_snapname, snapname,
510 sizeof (ds->ds_snapname));
511 mutex_exit(&ds->ds_lock);
512 *dsp = ds;
513 }
514 }
515
516 dsl_dir_rele(dd, FTAG);
517 return (err);
518 }
519
520 int
521 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
522 void *tag, dsl_dataset_t **dsp)
523 {
524 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
525 if (err != 0)
526 return (err);
527 if (!dsl_dataset_tryown(*dsp, tag)) {
528 dsl_dataset_rele(*dsp, tag);
529 *dsp = NULL;
530 return (SET_ERROR(EBUSY));
531 }
532 return (0);
533 }
534
535 int
536 dsl_dataset_own(dsl_pool_t *dp, const char *name,
537 void *tag, dsl_dataset_t **dsp)
538 {
539 int err = dsl_dataset_hold(dp, name, tag, dsp);
540 if (err != 0)
541 return (err);
542 if (!dsl_dataset_tryown(*dsp, tag)) {
543 dsl_dataset_rele(*dsp, tag);
544 return (SET_ERROR(EBUSY));
545 }
546 return (0);
547 }
548
549 /*
550 * See the comment above dsl_pool_hold() for details. In summary, a long
551 * hold is used to prevent destruction of a dataset while the pool hold
552 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
553 *
554 * The dataset and pool must be held when this function is called. After it
555 * is called, the pool hold may be released while the dataset is still held
556 * and accessed.
557 */
558 void
559 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
560 {
561 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
562 (void) refcount_add(&ds->ds_longholds, tag);
563 }
564
565 void
566 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
567 {
568 (void) refcount_remove(&ds->ds_longholds, tag);
569 }
570
571 /* Return B_TRUE if there are any long holds on this dataset. */
572 boolean_t
573 dsl_dataset_long_held(dsl_dataset_t *ds)
574 {
575 return (!refcount_is_zero(&ds->ds_longholds));
576 }
577
578 void
579 dsl_dataset_name(dsl_dataset_t *ds, char *name)
580 {
581 if (ds == NULL) {
582 (void) strcpy(name, "mos");
583 } else {
584 dsl_dir_name(ds->ds_dir, name);
585 VERIFY0(dsl_dataset_get_snapname(ds));
586 if (ds->ds_snapname[0]) {
587 (void) strcat(name, "@");
588 /*
589 * We use a "recursive" mutex so that we
590 * can call dprintf_ds() with ds_lock held.
591 */
592 if (!MUTEX_HELD(&ds->ds_lock)) {
593 mutex_enter(&ds->ds_lock);
594 (void) strcat(name, ds->ds_snapname);
595 mutex_exit(&ds->ds_lock);
596 } else {
597 (void) strcat(name, ds->ds_snapname);
598 }
599 }
600 }
601 }
602
603 static int
604 dsl_dataset_namelen(dsl_dataset_t *ds)
605 {
606 int result;
607
608 if (ds == NULL) {
609 result = 3; /* "mos" */
610 } else {
611 result = dsl_dir_namelen(ds->ds_dir);
612 VERIFY0(dsl_dataset_get_snapname(ds));
613 if (ds->ds_snapname[0]) {
614 ++result; /* adding one for the @-sign */
615 if (!MUTEX_HELD(&ds->ds_lock)) {
616 mutex_enter(&ds->ds_lock);
617 result += strlen(ds->ds_snapname);
618 mutex_exit(&ds->ds_lock);
619 } else {
620 result += strlen(ds->ds_snapname);
621 }
622 }
623 }
624
625 return (result);
626 }
627
628 void
629 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
630 {
631 dmu_buf_rele(ds->ds_dbuf, tag);
632 }
633
634 void
635 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
636 {
637 ASSERT(ds->ds_owner == tag && ds->ds_dbuf != NULL);
638
639 mutex_enter(&ds->ds_lock);
640 ds->ds_owner = NULL;
641 mutex_exit(&ds->ds_lock);
642 dsl_dataset_long_rele(ds, tag);
643 if (ds->ds_dbuf != NULL)
644 dsl_dataset_rele(ds, tag);
645 else
646 dsl_dataset_evict_impl(ds, B_FALSE);
647 }
648
649 boolean_t
650 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
651 {
652 boolean_t gotit = FALSE;
653
654 mutex_enter(&ds->ds_lock);
655 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
656 ds->ds_owner = tag;
657 dsl_dataset_long_hold(ds, tag);
658 gotit = TRUE;
659 }
660 mutex_exit(&ds->ds_lock);
661 return (gotit);
662 }
663
664 uint64_t
665 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
666 uint64_t flags, dmu_tx_t *tx)
667 {
668 dsl_pool_t *dp = dd->dd_pool;
669 dmu_buf_t *dbuf;
670 dsl_dataset_phys_t *dsphys;
671 uint64_t dsobj;
672 objset_t *mos = dp->dp_meta_objset;
673
674 if (origin == NULL)
675 origin = dp->dp_origin_snap;
676
677 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
678 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
679 ASSERT(dmu_tx_is_syncing(tx));
680 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
681
682 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
683 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
684 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
685 dmu_buf_will_dirty(dbuf, tx);
686 dsphys = dbuf->db_data;
687 bzero(dsphys, sizeof (dsl_dataset_phys_t));
688 dsphys->ds_dir_obj = dd->dd_object;
689 dsphys->ds_flags = flags;
690 dsphys->ds_fsid_guid = unique_create();
691 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
692 sizeof (dsphys->ds_guid));
693 dsphys->ds_snapnames_zapobj =
694 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
695 DMU_OT_NONE, 0, tx);
696 dsphys->ds_creation_time = gethrestime_sec();
697 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
698
699 if (origin == NULL) {
700 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
701 } else {
702 dsl_dataset_t *ohds; /* head of the origin snapshot */
703
704 dsphys->ds_prev_snap_obj = origin->ds_object;
705 dsphys->ds_prev_snap_txg =
706 origin->ds_phys->ds_creation_txg;
707 dsphys->ds_referenced_bytes =
708 origin->ds_phys->ds_referenced_bytes;
709 dsphys->ds_compressed_bytes =
710 origin->ds_phys->ds_compressed_bytes;
711 dsphys->ds_uncompressed_bytes =
712 origin->ds_phys->ds_uncompressed_bytes;
713 dsphys->ds_bp = origin->ds_phys->ds_bp;
714 dsphys->ds_flags |= origin->ds_phys->ds_flags;
715
716 dmu_buf_will_dirty(origin->ds_dbuf, tx);
717 origin->ds_phys->ds_num_children++;
718
719 VERIFY0(dsl_dataset_hold_obj(dp,
720 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
721 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
722 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
723 dsl_dataset_rele(ohds, FTAG);
724
725 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
726 if (origin->ds_phys->ds_next_clones_obj == 0) {
727 origin->ds_phys->ds_next_clones_obj =
728 zap_create(mos,
729 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
730 }
731 VERIFY0(zap_add_int(mos,
732 origin->ds_phys->ds_next_clones_obj, dsobj, tx));
733 }
734
735 dmu_buf_will_dirty(dd->dd_dbuf, tx);
736 dd->dd_phys->dd_origin_obj = origin->ds_object;
737 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
738 if (origin->ds_dir->dd_phys->dd_clones == 0) {
739 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
740 origin->ds_dir->dd_phys->dd_clones =
741 zap_create(mos,
742 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
743 }
744 VERIFY0(zap_add_int(mos,
745 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
746 }
747 }
748
749 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
750 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
751
752 dmu_buf_rele(dbuf, FTAG);
753
754 dmu_buf_will_dirty(dd->dd_dbuf, tx);
755 dd->dd_phys->dd_head_dataset_obj = dsobj;
756
757 return (dsobj);
758 }
759
760 static void
761 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
762 {
763 objset_t *os;
764
765 VERIFY0(dmu_objset_from_ds(ds, &os));
766 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
767 dsl_dataset_dirty(ds, tx);
768 }
769
770 uint64_t
771 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
772 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
773 {
774 dsl_pool_t *dp = pdd->dd_pool;
775 uint64_t dsobj, ddobj;
776 dsl_dir_t *dd;
777
778 ASSERT(dmu_tx_is_syncing(tx));
779 ASSERT(lastname[0] != '@');
780
781 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
782 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
783
784 dsobj = dsl_dataset_create_sync_dd(dd, origin,
785 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
786
787 dsl_deleg_set_create_perms(dd, tx, cr);
788
789 dsl_dir_rele(dd, FTAG);
790
791 /*
792 * If we are creating a clone, make sure we zero out any stale
793 * data from the origin snapshots zil header.
794 */
795 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
796 dsl_dataset_t *ds;
797
798 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
799 dsl_dataset_zero_zil(ds, tx);
800 dsl_dataset_rele(ds, FTAG);
801 }
802
803 return (dsobj);
804 }
805
806 /*
807 * The unique space in the head dataset can be calculated by subtracting
808 * the space used in the most recent snapshot, that is still being used
809 * in this file system, from the space currently in use. To figure out
810 * the space in the most recent snapshot still in use, we need to take
811 * the total space used in the snapshot and subtract out the space that
812 * has been freed up since the snapshot was taken.
813 */
814 void
815 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
816 {
817 uint64_t mrs_used;
818 uint64_t dlused, dlcomp, dluncomp;
819
820 ASSERT(!dsl_dataset_is_snapshot(ds));
821
822 if (ds->ds_phys->ds_prev_snap_obj != 0)
823 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
824 else
825 mrs_used = 0;
826
827 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
828
829 ASSERT3U(dlused, <=, mrs_used);
830 ds->ds_phys->ds_unique_bytes =
831 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
832
833 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
834 SPA_VERSION_UNIQUE_ACCURATE)
835 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
836 }
837
838 void
839 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
840 dmu_tx_t *tx)
841 {
842 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
843 uint64_t count;
844 int err;
845
846 ASSERT(ds->ds_phys->ds_num_children >= 2);
847 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
848 /*
849 * The err should not be ENOENT, but a bug in a previous version
850 * of the code could cause upgrade_clones_cb() to not set
851 * ds_next_snap_obj when it should, leading to a missing entry.
852 * If we knew that the pool was created after
853 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
854 * ENOENT. However, at least we can check that we don't have
855 * too many entries in the next_clones_obj even after failing to
856 * remove this one.
857 */
858 if (err != ENOENT)
859 VERIFY0(err);
860 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
861 &count));
862 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
863 }
864
865
866 blkptr_t *
867 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
868 {
869 return (&ds->ds_phys->ds_bp);
870 }
871
872 void
873 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
874 {
875 ASSERT(dmu_tx_is_syncing(tx));
876 /* If it's the meta-objset, set dp_meta_rootbp */
877 if (ds == NULL) {
878 tx->tx_pool->dp_meta_rootbp = *bp;
879 } else {
880 dmu_buf_will_dirty(ds->ds_dbuf, tx);
881 ds->ds_phys->ds_bp = *bp;
882 }
883 }
884
885 spa_t *
886 dsl_dataset_get_spa(dsl_dataset_t *ds)
887 {
888 return (ds->ds_dir->dd_pool->dp_spa);
889 }
890
891 void
892 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
893 {
894 dsl_pool_t *dp;
895
896 if (ds == NULL) /* this is the meta-objset */
897 return;
898
899 ASSERT(ds->ds_objset != NULL);
900
901 if (ds->ds_phys->ds_next_snap_obj != 0)
902 panic("dirtying snapshot!");
903
904 dp = ds->ds_dir->dd_pool;
905
906 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
907 /* up the hold count until we can be written out */
908 dmu_buf_add_ref(ds->ds_dbuf, ds);
909 }
910 }
911
912 boolean_t
913 dsl_dataset_is_dirty(dsl_dataset_t *ds)
914 {
915 for (int t = 0; t < TXG_SIZE; t++) {
916 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
917 ds, t))
918 return (B_TRUE);
919 }
920 return (B_FALSE);
921 }
922
923 static int
924 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
925 {
926 uint64_t asize;
927
928 if (!dmu_tx_is_syncing(tx))
929 return (0);
930
931 /*
932 * If there's an fs-only reservation, any blocks that might become
933 * owned by the snapshot dataset must be accommodated by space
934 * outside of the reservation.
935 */
936 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
937 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
938 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
939 return (SET_ERROR(ENOSPC));
940
941 /*
942 * Propagate any reserved space for this snapshot to other
943 * snapshot checks in this sync group.
944 */
945 if (asize > 0)
946 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
947
948 return (0);
949 }
950
951 typedef struct dsl_dataset_snapshot_arg {
952 nvlist_t *ddsa_snaps;
953 nvlist_t *ddsa_props;
954 nvlist_t *ddsa_errors;
955 } dsl_dataset_snapshot_arg_t;
956
957 int
958 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
959 dmu_tx_t *tx)
960 {
961 int error;
962 uint64_t value;
963
964 ds->ds_trysnap_txg = tx->tx_txg;
965
966 if (!dmu_tx_is_syncing(tx))
967 return (0);
968
969 /*
970 * We don't allow multiple snapshots of the same txg. If there
971 * is already one, try again.
972 */
973 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
974 return (SET_ERROR(EAGAIN));
975
976 /*
977 * Check for conflicting snapshot name.
978 */
979 error = dsl_dataset_snap_lookup(ds, snapname, &value);
980 if (error == 0)
981 return (SET_ERROR(EEXIST));
982 if (error != ENOENT)
983 return (error);
984
985 error = dsl_dataset_snapshot_reserve_space(ds, tx);
986 if (error != 0)
987 return (error);
988
989 return (0);
990 }
991
992 static int
993 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
994 {
995 dsl_dataset_snapshot_arg_t *ddsa = arg;
996 dsl_pool_t *dp = dmu_tx_pool(tx);
997 nvpair_t *pair;
998 int rv = 0;
999
1000 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1001 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1002 int error = 0;
1003 dsl_dataset_t *ds;
1004 char *name, *atp;
1005 char dsname[MAXNAMELEN];
1006
1007 name = nvpair_name(pair);
1008 if (strlen(name) >= MAXNAMELEN)
1009 error = SET_ERROR(ENAMETOOLONG);
1010 if (error == 0) {
1011 atp = strchr(name, '@');
1012 if (atp == NULL)
1013 error = SET_ERROR(EINVAL);
1014 if (error == 0)
1015 (void) strlcpy(dsname, name, atp - name + 1);
1016 }
1017 if (error == 0)
1018 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1019 if (error == 0) {
1020 error = dsl_dataset_snapshot_check_impl(ds,
1021 atp + 1, tx);
1022 dsl_dataset_rele(ds, FTAG);
1023 }
1024
1025 if (error != 0) {
1026 if (ddsa->ddsa_errors != NULL) {
1027 fnvlist_add_int32(ddsa->ddsa_errors,
1028 name, error);
1029 }
1030 rv = error;
1031 }
1032 }
1033 return (rv);
1034 }
1035
1036 void
1037 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1038 dmu_tx_t *tx)
1039 {
1040 static zil_header_t zero_zil;
1041
1042 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1043 dmu_buf_t *dbuf;
1044 dsl_dataset_phys_t *dsphys;
1045 uint64_t dsobj, crtxg;
1046 objset_t *mos = dp->dp_meta_objset;
1047 objset_t *os;
1048
1049 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1050
1051 /*
1052 * If we are on an old pool, the zil must not be active, in which
1053 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1054 */
1055 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1056 dmu_objset_from_ds(ds, &os) != 0 ||
1057 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1058 sizeof (zero_zil)) == 0);
1059
1060
1061 /*
1062 * The origin's ds_creation_txg has to be < TXG_INITIAL
1063 */
1064 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1065 crtxg = 1;
1066 else
1067 crtxg = tx->tx_txg;
1068
1069 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1070 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1071 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1072 dmu_buf_will_dirty(dbuf, tx);
1073 dsphys = dbuf->db_data;
1074 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1075 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1076 dsphys->ds_fsid_guid = unique_create();
1077 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1078 sizeof (dsphys->ds_guid));
1079 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
1080 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
1081 dsphys->ds_next_snap_obj = ds->ds_object;
1082 dsphys->ds_num_children = 1;
1083 dsphys->ds_creation_time = gethrestime_sec();
1084 dsphys->ds_creation_txg = crtxg;
1085 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
1086 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
1087 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
1088 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
1089 dsphys->ds_flags = ds->ds_phys->ds_flags;
1090 dsphys->ds_bp = ds->ds_phys->ds_bp;
1091 dmu_buf_rele(dbuf, FTAG);
1092
1093 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
1094 if (ds->ds_prev) {
1095 uint64_t next_clones_obj =
1096 ds->ds_prev->ds_phys->ds_next_clones_obj;
1097 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
1098 ds->ds_object ||
1099 ds->ds_prev->ds_phys->ds_num_children > 1);
1100 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
1101 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1102 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1103 ds->ds_prev->ds_phys->ds_creation_txg);
1104 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
1105 } else if (next_clones_obj != 0) {
1106 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1107 dsphys->ds_next_snap_obj, tx);
1108 VERIFY0(zap_add_int(mos,
1109 next_clones_obj, dsobj, tx));
1110 }
1111 }
1112
1113 /*
1114 * If we have a reference-reservation on this dataset, we will
1115 * need to increase the amount of refreservation being charged
1116 * since our unique space is going to zero.
1117 */
1118 if (ds->ds_reserved) {
1119 int64_t delta;
1120 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1121 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1122 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1123 delta, 0, 0, tx);
1124 }
1125
1126 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1127 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
1128 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
1129 dsl_deadlist_close(&ds->ds_deadlist);
1130 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1131 dsl_deadlist_add_key(&ds->ds_deadlist,
1132 ds->ds_phys->ds_prev_snap_txg, tx);
1133
1134 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
1135 ds->ds_phys->ds_prev_snap_obj = dsobj;
1136 ds->ds_phys->ds_prev_snap_txg = crtxg;
1137 ds->ds_phys->ds_unique_bytes = 0;
1138 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1139 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1140
1141 VERIFY0(zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
1142 snapname, 8, 1, &dsobj, tx));
1143
1144 if (ds->ds_prev)
1145 dsl_dataset_rele(ds->ds_prev, ds);
1146 VERIFY0(dsl_dataset_hold_obj(dp,
1147 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
1148
1149 dsl_scan_ds_snapshotted(ds, tx);
1150
1151 dsl_dir_snap_cmtime_update(ds->ds_dir);
1152
1153 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1154 }
1155
1156 static void
1157 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1158 {
1159 dsl_dataset_snapshot_arg_t *ddsa = arg;
1160 dsl_pool_t *dp = dmu_tx_pool(tx);
1161 nvpair_t *pair;
1162
1163 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1164 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1165 dsl_dataset_t *ds;
1166 char *name, *atp;
1167 char dsname[MAXNAMELEN];
1168
1169 name = nvpair_name(pair);
1170 atp = strchr(name, '@');
1171 (void) strlcpy(dsname, name, atp - name + 1);
1172 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1173
1174 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1175 if (ddsa->ddsa_props != NULL) {
1176 dsl_props_set_sync_impl(ds->ds_prev,
1177 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1178 }
1179 dsl_dataset_rele(ds, FTAG);
1180 }
1181 }
1182
1183 /*
1184 * The snapshots must all be in the same pool.
1185 * All-or-nothing: if there are any failures, nothing will be modified.
1186 */
1187 int
1188 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1189 {
1190 dsl_dataset_snapshot_arg_t ddsa;
1191 nvpair_t *pair;
1192 boolean_t needsuspend;
1193 int error;
1194 spa_t *spa;
1195 char *firstname;
1196 nvlist_t *suspended = NULL;
1197
1198 pair = nvlist_next_nvpair(snaps, NULL);
1199 if (pair == NULL)
1200 return (0);
1201 firstname = nvpair_name(pair);
1202
1203 error = spa_open(firstname, &spa, FTAG);
1204 if (error != 0)
1205 return (error);
1206 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1207 spa_close(spa, FTAG);
1208
1209 if (needsuspend) {
1210 suspended = fnvlist_alloc();
1211 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1212 pair = nvlist_next_nvpair(snaps, pair)) {
1213 char fsname[MAXNAMELEN];
1214 char *snapname = nvpair_name(pair);
1215 char *atp;
1216 void *cookie;
1217
1218 atp = strchr(snapname, '@');
1219 if (atp == NULL) {
1220 error = SET_ERROR(EINVAL);
1221 break;
1222 }
1223 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1224
1225 error = zil_suspend(fsname, &cookie);
1226 if (error != 0)
1227 break;
1228 fnvlist_add_uint64(suspended, fsname,
1229 (uintptr_t)cookie);
1230 }
1231 }
1232
1233 ddsa.ddsa_snaps = snaps;
1234 ddsa.ddsa_props = props;
1235 ddsa.ddsa_errors = errors;
1236
1237 if (error == 0) {
1238 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1239 dsl_dataset_snapshot_sync, &ddsa,
1240 fnvlist_num_pairs(snaps) * 3);
1241 }
1242
1243 if (suspended != NULL) {
1244 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1245 pair = nvlist_next_nvpair(suspended, pair)) {
1246 zil_resume((void *)(uintptr_t)
1247 fnvpair_value_uint64(pair));
1248 }
1249 fnvlist_free(suspended);
1250 }
1251
1252 return (error);
1253 }
1254
1255 typedef struct dsl_dataset_snapshot_tmp_arg {
1256 const char *ddsta_fsname;
1257 const char *ddsta_snapname;
1258 minor_t ddsta_cleanup_minor;
1259 const char *ddsta_htag;
1260 } dsl_dataset_snapshot_tmp_arg_t;
1261
1262 static int
1263 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1264 {
1265 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1266 dsl_pool_t *dp = dmu_tx_pool(tx);
1267 dsl_dataset_t *ds;
1268 int error;
1269
1270 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1271 if (error != 0)
1272 return (error);
1273
1274 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname, tx);
1275 if (error != 0) {
1276 dsl_dataset_rele(ds, FTAG);
1277 return (error);
1278 }
1279
1280 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1281 dsl_dataset_rele(ds, FTAG);
1282 return (SET_ERROR(ENOTSUP));
1283 }
1284 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1285 B_TRUE, tx);
1286 if (error != 0) {
1287 dsl_dataset_rele(ds, FTAG);
1288 return (error);
1289 }
1290
1291 dsl_dataset_rele(ds, FTAG);
1292 return (0);
1293 }
1294
1295 static void
1296 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1297 {
1298 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1299 dsl_pool_t *dp = dmu_tx_pool(tx);
1300 dsl_dataset_t *ds;
1301
1302 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1303
1304 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1305 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1306 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1307 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1308
1309 dsl_dataset_rele(ds, FTAG);
1310 }
1311
1312 int
1313 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1314 minor_t cleanup_minor, const char *htag)
1315 {
1316 dsl_dataset_snapshot_tmp_arg_t ddsta;
1317 int error;
1318 spa_t *spa;
1319 boolean_t needsuspend;
1320 void *cookie;
1321
1322 ddsta.ddsta_fsname = fsname;
1323 ddsta.ddsta_snapname = snapname;
1324 ddsta.ddsta_cleanup_minor = cleanup_minor;
1325 ddsta.ddsta_htag = htag;
1326
1327 error = spa_open(fsname, &spa, FTAG);
1328 if (error != 0)
1329 return (error);
1330 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1331 spa_close(spa, FTAG);
1332
1333 if (needsuspend) {
1334 error = zil_suspend(fsname, &cookie);
1335 if (error != 0)
1336 return (error);
1337 }
1338
1339 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1340 dsl_dataset_snapshot_tmp_sync, &ddsta, 3);
1341
1342 if (needsuspend)
1343 zil_resume(cookie);
1344 return (error);
1345 }
1346
1347
1348 void
1349 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1350 {
1351 ASSERT(dmu_tx_is_syncing(tx));
1352 ASSERT(ds->ds_objset != NULL);
1353 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
1354
1355 /*
1356 * in case we had to change ds_fsid_guid when we opened it,
1357 * sync it out now.
1358 */
1359 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1360 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
1361
1362 dmu_objset_sync(ds->ds_objset, zio, tx);
1363 }
1364
1365 static void
1366 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1367 {
1368 uint64_t count = 0;
1369 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1370 zap_cursor_t zc;
1371 zap_attribute_t za;
1372 nvlist_t *propval = fnvlist_alloc();
1373 nvlist_t *val = fnvlist_alloc();
1374
1375 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1376
1377 /*
1378 * There may be missing entries in ds_next_clones_obj
1379 * due to a bug in a previous version of the code.
1380 * Only trust it if it has the right number of entries.
1381 */
1382 if (ds->ds_phys->ds_next_clones_obj != 0) {
1383 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1384 &count));
1385 }
1386 if (count != ds->ds_phys->ds_num_children - 1)
1387 goto fail;
1388 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
1389 zap_cursor_retrieve(&zc, &za) == 0;
1390 zap_cursor_advance(&zc)) {
1391 dsl_dataset_t *clone;
1392 char buf[ZFS_MAXNAMELEN];
1393 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1394 za.za_first_integer, FTAG, &clone));
1395 dsl_dir_name(clone->ds_dir, buf);
1396 fnvlist_add_boolean(val, buf);
1397 dsl_dataset_rele(clone, FTAG);
1398 }
1399 zap_cursor_fini(&zc);
1400 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1401 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1402 fail:
1403 nvlist_free(val);
1404 nvlist_free(propval);
1405 }
1406
1407 void
1408 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1409 {
1410 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1411 uint64_t refd, avail, uobjs, aobjs, ratio;
1412
1413 ASSERT(dsl_pool_config_held(dp));
1414
1415 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
1416 (ds->ds_phys->ds_uncompressed_bytes * 100 /
1417 ds->ds_phys->ds_compressed_bytes);
1418
1419 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1420 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1421 ds->ds_phys->ds_uncompressed_bytes);
1422
1423 if (dsl_dataset_is_snapshot(ds)) {
1424 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1425 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1426 ds->ds_phys->ds_unique_bytes);
1427 get_clones_stat(ds, nv);
1428 } else {
1429 dsl_dir_stats(ds->ds_dir, nv);
1430 }
1431
1432 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1433 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1434 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1435
1436 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1437 ds->ds_phys->ds_creation_time);
1438 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1439 ds->ds_phys->ds_creation_txg);
1440 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1441 ds->ds_quota);
1442 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1443 ds->ds_reserved);
1444 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1445 ds->ds_phys->ds_guid);
1446 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1447 ds->ds_phys->ds_unique_bytes);
1448 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1449 ds->ds_object);
1450 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1451 ds->ds_userrefs);
1452 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1453 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1454
1455 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1456 uint64_t written, comp, uncomp;
1457 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1458 dsl_dataset_t *prev;
1459
1460 int err = dsl_dataset_hold_obj(dp,
1461 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
1462 if (err == 0) {
1463 err = dsl_dataset_space_written(prev, ds, &written,
1464 &comp, &uncomp);
1465 dsl_dataset_rele(prev, FTAG);
1466 if (err == 0) {
1467 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1468 written);
1469 }
1470 }
1471 }
1472 }
1473
1474 void
1475 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1476 {
1477 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1478 ASSERT(dsl_pool_config_held(dp));
1479
1480 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
1481 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
1482 stat->dds_guid = ds->ds_phys->ds_guid;
1483 stat->dds_origin[0] = '\0';
1484 if (dsl_dataset_is_snapshot(ds)) {
1485 stat->dds_is_snapshot = B_TRUE;
1486 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
1487 } else {
1488 stat->dds_is_snapshot = B_FALSE;
1489 stat->dds_num_clones = 0;
1490
1491 if (dsl_dir_is_clone(ds->ds_dir)) {
1492 dsl_dataset_t *ods;
1493
1494 VERIFY0(dsl_dataset_hold_obj(dp,
1495 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
1496 dsl_dataset_name(ods, stat->dds_origin);
1497 dsl_dataset_rele(ods, FTAG);
1498 }
1499 }
1500 }
1501
1502 uint64_t
1503 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1504 {
1505 return (ds->ds_fsid_guid);
1506 }
1507
1508 void
1509 dsl_dataset_space(dsl_dataset_t *ds,
1510 uint64_t *refdbytesp, uint64_t *availbytesp,
1511 uint64_t *usedobjsp, uint64_t *availobjsp)
1512 {
1513 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
1514 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1515 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
1516 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
1517 if (ds->ds_quota != 0) {
1518 /*
1519 * Adjust available bytes according to refquota
1520 */
1521 if (*refdbytesp < ds->ds_quota)
1522 *availbytesp = MIN(*availbytesp,
1523 ds->ds_quota - *refdbytesp);
1524 else
1525 *availbytesp = 0;
1526 }
1527 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
1528 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1529 }
1530
1531 boolean_t
1532 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
1533 {
1534 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1535
1536 ASSERT(dsl_pool_config_held(dp));
1537 if (ds->ds_prev == NULL)
1538 return (B_FALSE);
1539 if (ds->ds_phys->ds_bp.blk_birth >
1540 ds->ds_prev->ds_phys->ds_creation_txg) {
1541 objset_t *os, *os_prev;
1542 /*
1543 * It may be that only the ZIL differs, because it was
1544 * reset in the head. Don't count that as being
1545 * modified.
1546 */
1547 if (dmu_objset_from_ds(ds, &os) != 0)
1548 return (B_TRUE);
1549 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
1550 return (B_TRUE);
1551 return (bcmp(&os->os_phys->os_meta_dnode,
1552 &os_prev->os_phys->os_meta_dnode,
1553 sizeof (os->os_phys->os_meta_dnode)) != 0);
1554 }
1555 return (B_FALSE);
1556 }
1557
1558 typedef struct dsl_dataset_rename_snapshot_arg {
1559 const char *ddrsa_fsname;
1560 const char *ddrsa_oldsnapname;
1561 const char *ddrsa_newsnapname;
1562 boolean_t ddrsa_recursive;
1563 dmu_tx_t *ddrsa_tx;
1564 } dsl_dataset_rename_snapshot_arg_t;
1565
1566 /* ARGSUSED */
1567 static int
1568 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1569 dsl_dataset_t *hds, void *arg)
1570 {
1571 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1572 int error;
1573 uint64_t val;
1574
1575 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1576 if (error != 0) {
1577 /* ignore nonexistent snapshots */
1578 return (error == ENOENT ? 0 : error);
1579 }
1580
1581 /* new name should not exist */
1582 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1583 if (error == 0)
1584 error = SET_ERROR(EEXIST);
1585 else if (error == ENOENT)
1586 error = 0;
1587
1588 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1589 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1590 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1591 error = SET_ERROR(ENAMETOOLONG);
1592
1593 return (error);
1594 }
1595
1596 static int
1597 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1598 {
1599 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1600 dsl_pool_t *dp = dmu_tx_pool(tx);
1601 dsl_dataset_t *hds;
1602 int error;
1603
1604 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1605 if (error != 0)
1606 return (error);
1607
1608 if (ddrsa->ddrsa_recursive) {
1609 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1610 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1611 DS_FIND_CHILDREN);
1612 } else {
1613 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1614 }
1615 dsl_dataset_rele(hds, FTAG);
1616 return (error);
1617 }
1618
1619 static int
1620 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1621 dsl_dataset_t *hds, void *arg)
1622 {
1623 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1624 dsl_dataset_t *ds;
1625 uint64_t val;
1626 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1627 int error;
1628
1629 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1630 ASSERT(error == 0 || error == ENOENT);
1631 if (error == ENOENT) {
1632 /* ignore nonexistent snapshots */
1633 return (0);
1634 }
1635
1636 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1637
1638 /* log before we change the name */
1639 spa_history_log_internal_ds(ds, "rename", tx,
1640 "-> @%s", ddrsa->ddrsa_newsnapname);
1641
1642 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx));
1643 mutex_enter(&ds->ds_lock);
1644 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1645 mutex_exit(&ds->ds_lock);
1646 VERIFY0(zap_add(dp->dp_meta_objset, hds->ds_phys->ds_snapnames_zapobj,
1647 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1648
1649 dsl_dataset_rele(ds, FTAG);
1650 return (0);
1651 }
1652
1653 static void
1654 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1655 {
1656 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1657 dsl_pool_t *dp = dmu_tx_pool(tx);
1658 dsl_dataset_t *hds;
1659
1660 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1661 ddrsa->ddrsa_tx = tx;
1662 if (ddrsa->ddrsa_recursive) {
1663 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1664 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1665 DS_FIND_CHILDREN));
1666 } else {
1667 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1668 }
1669 dsl_dataset_rele(hds, FTAG);
1670 }
1671
1672 int
1673 dsl_dataset_rename_snapshot(const char *fsname,
1674 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1675 {
1676 dsl_dataset_rename_snapshot_arg_t ddrsa;
1677
1678 ddrsa.ddrsa_fsname = fsname;
1679 ddrsa.ddrsa_oldsnapname = oldsnapname;
1680 ddrsa.ddrsa_newsnapname = newsnapname;
1681 ddrsa.ddrsa_recursive = recursive;
1682
1683 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1684 dsl_dataset_rename_snapshot_sync, &ddrsa, 1));
1685 }
1686
1687 static int
1688 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1689 {
1690 const char *fsname = arg;
1691 dsl_pool_t *dp = dmu_tx_pool(tx);
1692 dsl_dataset_t *ds;
1693 int64_t unused_refres_delta;
1694 int error;
1695
1696 error = dsl_dataset_hold(dp, fsname, FTAG, &ds);
1697 if (error != 0)
1698 return (error);
1699
1700 /* must not be a snapshot */
1701 if (dsl_dataset_is_snapshot(ds)) {
1702 dsl_dataset_rele(ds, FTAG);
1703 return (SET_ERROR(EINVAL));
1704 }
1705
1706 /* must have a most recent snapshot */
1707 if (ds->ds_phys->ds_prev_snap_txg < TXG_INITIAL) {
1708 dsl_dataset_rele(ds, FTAG);
1709 return (SET_ERROR(EINVAL));
1710 }
1711
1712 if (dsl_dataset_long_held(ds)) {
1713 dsl_dataset_rele(ds, FTAG);
1714 return (SET_ERROR(EBUSY));
1715 }
1716
1717 /*
1718 * Check if the snap we are rolling back to uses more than
1719 * the refquota.
1720 */
1721 if (ds->ds_quota != 0 &&
1722 ds->ds_prev->ds_phys->ds_referenced_bytes > ds->ds_quota) {
1723 dsl_dataset_rele(ds, FTAG);
1724 return (SET_ERROR(EDQUOT));
1725 }
1726
1727 /*
1728 * When we do the clone swap, we will temporarily use more space
1729 * due to the refreservation (the head will no longer have any
1730 * unique space, so the entire amount of the refreservation will need
1731 * to be free). We will immediately destroy the clone, freeing
1732 * this space, but the freeing happens over many txg's.
1733 */
1734 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
1735 ds->ds_phys->ds_unique_bytes);
1736
1737 if (unused_refres_delta > 0 &&
1738 unused_refres_delta >
1739 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
1740 dsl_dataset_rele(ds, FTAG);
1741 return (SET_ERROR(ENOSPC));
1742 }
1743
1744 dsl_dataset_rele(ds, FTAG);
1745 return (0);
1746 }
1747
1748 static void
1749 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
1750 {
1751 const char *fsname = arg;
1752 dsl_pool_t *dp = dmu_tx_pool(tx);
1753 dsl_dataset_t *ds, *clone;
1754 uint64_t cloneobj;
1755
1756 VERIFY0(dsl_dataset_hold(dp, fsname, FTAG, &ds));
1757
1758 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
1759 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
1760
1761 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
1762
1763 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
1764 dsl_dataset_zero_zil(ds, tx);
1765
1766 dsl_destroy_head_sync_impl(clone, tx);
1767
1768 dsl_dataset_rele(clone, FTAG);
1769 dsl_dataset_rele(ds, FTAG);
1770 }
1771
1772 int
1773 dsl_dataset_rollback(const char *fsname)
1774 {
1775 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
1776 dsl_dataset_rollback_sync, (void *)fsname, 1));
1777 }
1778
1779 struct promotenode {
1780 list_node_t link;
1781 dsl_dataset_t *ds;
1782 };
1783
1784 typedef struct dsl_dataset_promote_arg {
1785 const char *ddpa_clonename;
1786 dsl_dataset_t *ddpa_clone;
1787 list_t shared_snaps, origin_snaps, clone_snaps;
1788 dsl_dataset_t *origin_origin; /* origin of the origin */
1789 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
1790 char *err_ds;
1791 } dsl_dataset_promote_arg_t;
1792
1793 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
1794 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
1795 void *tag);
1796 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
1797
1798 static int
1799 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
1800 {
1801 dsl_dataset_promote_arg_t *ddpa = arg;
1802 dsl_pool_t *dp = dmu_tx_pool(tx);
1803 dsl_dataset_t *hds;
1804 struct promotenode *snap;
1805 dsl_dataset_t *origin_ds;
1806 int err;
1807 uint64_t unused;
1808
1809 err = promote_hold(ddpa, dp, FTAG);
1810 if (err != 0)
1811 return (err);
1812
1813 hds = ddpa->ddpa_clone;
1814
1815 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE) {
1816 promote_rele(ddpa, FTAG);
1817 return (SET_ERROR(EXDEV));
1818 }
1819
1820 /*
1821 * Compute and check the amount of space to transfer. Since this is
1822 * so expensive, don't do the preliminary check.
1823 */
1824 if (!dmu_tx_is_syncing(tx)) {
1825 promote_rele(ddpa, FTAG);
1826 return (0);
1827 }
1828
1829 snap = list_head(&ddpa->shared_snaps);
1830 origin_ds = snap->ds;
1831
1832 /* compute origin's new unique space */
1833 snap = list_tail(&ddpa->clone_snaps);
1834 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1835 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
1836 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1837 &ddpa->unique, &unused, &unused);
1838
1839 /*
1840 * Walk the snapshots that we are moving
1841 *
1842 * Compute space to transfer. Consider the incremental changes
1843 * to used by each snapshot:
1844 * (my used) = (prev's used) + (blocks born) - (blocks killed)
1845 * So each snapshot gave birth to:
1846 * (blocks born) = (my used) - (prev's used) + (blocks killed)
1847 * So a sequence would look like:
1848 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
1849 * Which simplifies to:
1850 * uN + kN + kN-1 + ... + k1 + k0
1851 * Note however, if we stop before we reach the ORIGIN we get:
1852 * uN + kN + kN-1 + ... + kM - uM-1
1853 */
1854 ddpa->used = origin_ds->ds_phys->ds_referenced_bytes;
1855 ddpa->comp = origin_ds->ds_phys->ds_compressed_bytes;
1856 ddpa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
1857 for (snap = list_head(&ddpa->shared_snaps); snap;
1858 snap = list_next(&ddpa->shared_snaps, snap)) {
1859 uint64_t val, dlused, dlcomp, dluncomp;
1860 dsl_dataset_t *ds = snap->ds;
1861
1862 /*
1863 * If there are long holds, we won't be able to evict
1864 * the objset.
1865 */
1866 if (dsl_dataset_long_held(ds)) {
1867 err = SET_ERROR(EBUSY);
1868 goto out;
1869 }
1870
1871 /* Check that the snapshot name does not conflict */
1872 VERIFY0(dsl_dataset_get_snapname(ds));
1873 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
1874 if (err == 0) {
1875 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
1876 err = SET_ERROR(EEXIST);
1877 goto out;
1878 }
1879 if (err != ENOENT)
1880 goto out;
1881
1882 /* The very first snapshot does not have a deadlist */
1883 if (ds->ds_phys->ds_prev_snap_obj == 0)
1884 continue;
1885
1886 dsl_deadlist_space(&ds->ds_deadlist,
1887 &dlused, &dlcomp, &dluncomp);
1888 ddpa->used += dlused;
1889 ddpa->comp += dlcomp;
1890 ddpa->uncomp += dluncomp;
1891 }
1892
1893 /*
1894 * If we are a clone of a clone then we never reached ORIGIN,
1895 * so we need to subtract out the clone origin's used space.
1896 */
1897 if (ddpa->origin_origin) {
1898 ddpa->used -= ddpa->origin_origin->ds_phys->ds_referenced_bytes;
1899 ddpa->comp -= ddpa->origin_origin->ds_phys->ds_compressed_bytes;
1900 ddpa->uncomp -=
1901 ddpa->origin_origin->ds_phys->ds_uncompressed_bytes;
1902 }
1903
1904 /* Check that there is enough space here */
1905 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
1906 ddpa->used);
1907 if (err != 0)
1908 goto out;
1909
1910 /*
1911 * Compute the amounts of space that will be used by snapshots
1912 * after the promotion (for both origin and clone). For each,
1913 * it is the amount of space that will be on all of their
1914 * deadlists (that was not born before their new origin).
1915 */
1916 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1917 uint64_t space;
1918
1919 /*
1920 * Note, typically this will not be a clone of a clone,
1921 * so dd_origin_txg will be < TXG_INITIAL, so
1922 * these snaplist_space() -> dsl_deadlist_space_range()
1923 * calls will be fast because they do not have to
1924 * iterate over all bps.
1925 */
1926 snap = list_head(&ddpa->origin_snaps);
1927 err = snaplist_space(&ddpa->shared_snaps,
1928 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
1929 if (err != 0)
1930 goto out;
1931
1932 err = snaplist_space(&ddpa->clone_snaps,
1933 snap->ds->ds_dir->dd_origin_txg, &space);
1934 if (err != 0)
1935 goto out;
1936 ddpa->cloneusedsnap += space;
1937 }
1938 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
1939 err = snaplist_space(&ddpa->origin_snaps,
1940 origin_ds->ds_phys->ds_creation_txg, &ddpa->originusedsnap);
1941 if (err != 0)
1942 goto out;
1943 }
1944
1945 out:
1946 promote_rele(ddpa, FTAG);
1947 return (err);
1948 }
1949
1950 static void
1951 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
1952 {
1953 dsl_dataset_promote_arg_t *ddpa = arg;
1954 dsl_pool_t *dp = dmu_tx_pool(tx);
1955 dsl_dataset_t *hds;
1956 struct promotenode *snap;
1957 dsl_dataset_t *origin_ds;
1958 dsl_dataset_t *origin_head;
1959 dsl_dir_t *dd;
1960 dsl_dir_t *odd = NULL;
1961 uint64_t oldnext_obj;
1962 int64_t delta;
1963
1964 VERIFY0(promote_hold(ddpa, dp, FTAG));
1965 hds = ddpa->ddpa_clone;
1966
1967 ASSERT0(hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE);
1968
1969 snap = list_head(&ddpa->shared_snaps);
1970 origin_ds = snap->ds;
1971 dd = hds->ds_dir;
1972
1973 snap = list_head(&ddpa->origin_snaps);
1974 origin_head = snap->ds;
1975
1976 /*
1977 * We need to explicitly open odd, since origin_ds's dd will be
1978 * changing.
1979 */
1980 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
1981 NULL, FTAG, &odd));
1982
1983 /* change origin's next snap */
1984 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
1985 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
1986 snap = list_tail(&ddpa->clone_snaps);
1987 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
1988 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
1989
1990 /* change the origin's next clone */
1991 if (origin_ds->ds_phys->ds_next_clones_obj) {
1992 dsl_dataset_remove_from_next_clones(origin_ds,
1993 snap->ds->ds_object, tx);
1994 VERIFY0(zap_add_int(dp->dp_meta_objset,
1995 origin_ds->ds_phys->ds_next_clones_obj,
1996 oldnext_obj, tx));
1997 }
1998
1999 /* change origin */
2000 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2001 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2002 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2003 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2004 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2005 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2006 origin_head->ds_dir->dd_origin_txg =
2007 origin_ds->ds_phys->ds_creation_txg;
2008
2009 /* change dd_clone entries */
2010 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2011 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2012 odd->dd_phys->dd_clones, hds->ds_object, tx));
2013 VERIFY0(zap_add_int(dp->dp_meta_objset,
2014 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2015 hds->ds_object, tx));
2016
2017 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2018 ddpa->origin_origin->ds_dir->dd_phys->dd_clones,
2019 origin_head->ds_object, tx));
2020 if (dd->dd_phys->dd_clones == 0) {
2021 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2022 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2023 }
2024 VERIFY0(zap_add_int(dp->dp_meta_objset,
2025 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2026 }
2027
2028 /* move snapshots to this dir */
2029 for (snap = list_head(&ddpa->shared_snaps); snap;
2030 snap = list_next(&ddpa->shared_snaps, snap)) {
2031 dsl_dataset_t *ds = snap->ds;
2032
2033 /*
2034 * Property callbacks are registered to a particular
2035 * dsl_dir. Since ours is changing, evict the objset
2036 * so that they will be unregistered from the old dsl_dir.
2037 */
2038 if (ds->ds_objset) {
2039 dmu_objset_evict(ds->ds_objset);
2040 ds->ds_objset = NULL;
2041 }
2042
2043 /* move snap name entry */
2044 VERIFY0(dsl_dataset_get_snapname(ds));
2045 VERIFY0(dsl_dataset_snap_remove(origin_head,
2046 ds->ds_snapname, tx));
2047 VERIFY0(zap_add(dp->dp_meta_objset,
2048 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2049 8, 1, &ds->ds_object, tx));
2050
2051 /* change containing dsl_dir */
2052 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2053 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2054 ds->ds_phys->ds_dir_obj = dd->dd_object;
2055 ASSERT3P(ds->ds_dir, ==, odd);
2056 dsl_dir_rele(ds->ds_dir, ds);
2057 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2058 NULL, ds, &ds->ds_dir));
2059
2060 /* move any clone references */
2061 if (ds->ds_phys->ds_next_clones_obj &&
2062 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2063 zap_cursor_t zc;
2064 zap_attribute_t za;
2065
2066 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2067 ds->ds_phys->ds_next_clones_obj);
2068 zap_cursor_retrieve(&zc, &za) == 0;
2069 zap_cursor_advance(&zc)) {
2070 dsl_dataset_t *cnds;
2071 uint64_t o;
2072
2073 if (za.za_first_integer == oldnext_obj) {
2074 /*
2075 * We've already moved the
2076 * origin's reference.
2077 */
2078 continue;
2079 }
2080
2081 VERIFY0(dsl_dataset_hold_obj(dp,
2082 za.za_first_integer, FTAG, &cnds));
2083 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2084
2085 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2086 odd->dd_phys->dd_clones, o, tx));
2087 VERIFY0(zap_add_int(dp->dp_meta_objset,
2088 dd->dd_phys->dd_clones, o, tx));
2089 dsl_dataset_rele(cnds, FTAG);
2090 }
2091 zap_cursor_fini(&zc);
2092 }
2093
2094 ASSERT(!dsl_prop_hascb(ds));
2095 }
2096
2097 /*
2098 * Change space accounting.
2099 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2100 * both be valid, or both be 0 (resulting in delta == 0). This
2101 * is true for each of {clone,origin} independently.
2102 */
2103
2104 delta = ddpa->cloneusedsnap -
2105 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2106 ASSERT3S(delta, >=, 0);
2107 ASSERT3U(ddpa->used, >=, delta);
2108 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2109 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2110 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2111
2112 delta = ddpa->originusedsnap -
2113 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2114 ASSERT3S(delta, <=, 0);
2115 ASSERT3U(ddpa->used, >=, -delta);
2116 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2117 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2118 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2119
2120 origin_ds->ds_phys->ds_unique_bytes = ddpa->unique;
2121
2122 /* log history record */
2123 spa_history_log_internal_ds(hds, "promote", tx, "");
2124
2125 dsl_dir_rele(odd, FTAG);
2126 promote_rele(ddpa, FTAG);
2127 }
2128
2129 /*
2130 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2131 * (exclusive) and last_obj (inclusive). The list will be in reverse
2132 * order (last_obj will be the list_head()). If first_obj == 0, do all
2133 * snapshots back to this dataset's origin.
2134 */
2135 static int
2136 snaplist_make(dsl_pool_t *dp,
2137 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2138 {
2139 uint64_t obj = last_obj;
2140
2141 list_create(l, sizeof (struct promotenode),
2142 offsetof(struct promotenode, link));
2143
2144 while (obj != first_obj) {
2145 dsl_dataset_t *ds;
2146 struct promotenode *snap;
2147 int err;
2148
2149 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2150 ASSERT(err != ENOENT);
2151 if (err != 0)
2152 return (err);
2153
2154 if (first_obj == 0)
2155 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2156
2157 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2158 snap->ds = ds;
2159 list_insert_tail(l, snap);
2160 obj = ds->ds_phys->ds_prev_snap_obj;
2161 }
2162
2163 return (0);
2164 }
2165
2166 static int
2167 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2168 {
2169 struct promotenode *snap;
2170
2171 *spacep = 0;
2172 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2173 uint64_t used, comp, uncomp;
2174 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2175 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2176 *spacep += used;
2177 }
2178 return (0);
2179 }
2180
2181 static void
2182 snaplist_destroy(list_t *l, void *tag)
2183 {
2184 struct promotenode *snap;
2185
2186 if (l == NULL || !list_link_active(&l->list_head))
2187 return;
2188
2189 while ((snap = list_tail(l)) != NULL) {
2190 list_remove(l, snap);
2191 dsl_dataset_rele(snap->ds, tag);
2192 kmem_free(snap, sizeof (*snap));
2193 }
2194 list_destroy(l);
2195 }
2196
2197 static int
2198 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2199 {
2200 int error;
2201 dsl_dir_t *dd;
2202 struct promotenode *snap;
2203
2204 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2205 &ddpa->ddpa_clone);
2206 if (error != 0)
2207 return (error);
2208 dd = ddpa->ddpa_clone->ds_dir;
2209
2210 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2211 !dsl_dir_is_clone(dd)) {
2212 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2213 return (SET_ERROR(EINVAL));
2214 }
2215
2216 error = snaplist_make(dp, 0, dd->dd_phys->dd_origin_obj,
2217 &ddpa->shared_snaps, tag);
2218 if (error != 0)
2219 goto out;
2220
2221 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2222 &ddpa->clone_snaps, tag);
2223 if (error != 0)
2224 goto out;
2225
2226 snap = list_head(&ddpa->shared_snaps);
2227 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2228 error = snaplist_make(dp, dd->dd_phys->dd_origin_obj,
2229 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj,
2230 &ddpa->origin_snaps, tag);
2231 if (error != 0)
2232 goto out;
2233
2234 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2235 error = dsl_dataset_hold_obj(dp,
2236 snap->ds->ds_dir->dd_phys->dd_origin_obj,
2237 tag, &ddpa->origin_origin);
2238 if (error != 0)
2239 goto out;
2240 }
2241 out:
2242 if (error != 0)
2243 promote_rele(ddpa, tag);
2244 return (error);
2245 }
2246
2247 static void
2248 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2249 {
2250 snaplist_destroy(&ddpa->shared_snaps, tag);
2251 snaplist_destroy(&ddpa->clone_snaps, tag);
2252 snaplist_destroy(&ddpa->origin_snaps, tag);
2253 if (ddpa->origin_origin != NULL)
2254 dsl_dataset_rele(ddpa->origin_origin, tag);
2255 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2256 }
2257
2258 /*
2259 * Promote a clone.
2260 *
2261 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2262 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2263 */
2264 int
2265 dsl_dataset_promote(const char *name, char *conflsnap)
2266 {
2267 dsl_dataset_promote_arg_t ddpa = { 0 };
2268 uint64_t numsnaps;
2269 int error;
2270 objset_t *os;
2271
2272 /*
2273 * We will modify space proportional to the number of
2274 * snapshots. Compute numsnaps.
2275 */
2276 error = dmu_objset_hold(name, FTAG, &os);
2277 if (error != 0)
2278 return (error);
2279 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2280 dmu_objset_ds(os)->ds_phys->ds_snapnames_zapobj, &numsnaps);
2281 dmu_objset_rele(os, FTAG);
2282 if (error != 0)
2283 return (error);
2284
2285 ddpa.ddpa_clonename = name;
2286 ddpa.err_ds = conflsnap;
2287
2288 return (dsl_sync_task(name, dsl_dataset_promote_check,
2289 dsl_dataset_promote_sync, &ddpa, 2 + numsnaps));
2290 }
2291
2292 int
2293 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2294 dsl_dataset_t *origin_head, boolean_t force)
2295 {
2296 int64_t unused_refres_delta;
2297
2298 /* they should both be heads */
2299 if (dsl_dataset_is_snapshot(clone) ||
2300 dsl_dataset_is_snapshot(origin_head))
2301 return (SET_ERROR(EINVAL));
2302
2303 /* the branch point should be just before them */
2304 if (clone->ds_prev != origin_head->ds_prev)
2305 return (SET_ERROR(EINVAL));
2306
2307 /* clone should be the clone (unless they are unrelated) */
2308 if (clone->ds_prev != NULL &&
2309 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2310 origin_head->ds_object !=
2311 clone->ds_prev->ds_phys->ds_next_snap_obj)
2312 return (SET_ERROR(EINVAL));
2313
2314 /* the clone should be a child of the origin */
2315 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2316 return (SET_ERROR(EINVAL));
2317
2318 /* origin_head shouldn't be modified unless 'force' */
2319 if (!force && dsl_dataset_modified_since_lastsnap(origin_head))
2320 return (SET_ERROR(ETXTBSY));
2321
2322 /* origin_head should have no long holds (e.g. is not mounted) */
2323 if (dsl_dataset_long_held(origin_head))
2324 return (SET_ERROR(EBUSY));
2325
2326 /* check amount of any unconsumed refreservation */
2327 unused_refres_delta =
2328 (int64_t)MIN(origin_head->ds_reserved,
2329 origin_head->ds_phys->ds_unique_bytes) -
2330 (int64_t)MIN(origin_head->ds_reserved,
2331 clone->ds_phys->ds_unique_bytes);
2332
2333 if (unused_refres_delta > 0 &&
2334 unused_refres_delta >
2335 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2336 return (SET_ERROR(ENOSPC));
2337
2338 /* clone can't be over the head's refquota */
2339 if (origin_head->ds_quota != 0 &&
2340 clone->ds_phys->ds_referenced_bytes > origin_head->ds_quota)
2341 return (SET_ERROR(EDQUOT));
2342
2343 return (0);
2344 }
2345
2346 void
2347 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2348 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2349 {
2350 dsl_pool_t *dp = dmu_tx_pool(tx);
2351 int64_t unused_refres_delta;
2352
2353 ASSERT(clone->ds_reserved == 0);
2354 ASSERT(origin_head->ds_quota == 0 ||
2355 clone->ds_phys->ds_unique_bytes <= origin_head->ds_quota);
2356
2357 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2358 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2359
2360 if (clone->ds_objset != NULL) {
2361 dmu_objset_evict(clone->ds_objset);
2362 clone->ds_objset = NULL;
2363 }
2364
2365 if (origin_head->ds_objset != NULL) {
2366 dmu_objset_evict(origin_head->ds_objset);
2367 origin_head->ds_objset = NULL;
2368 }
2369
2370 unused_refres_delta =
2371 (int64_t)MIN(origin_head->ds_reserved,
2372 origin_head->ds_phys->ds_unique_bytes) -
2373 (int64_t)MIN(origin_head->ds_reserved,
2374 clone->ds_phys->ds_unique_bytes);
2375
2376 /*
2377 * Reset origin's unique bytes, if it exists.
2378 */
2379 if (clone->ds_prev) {
2380 dsl_dataset_t *origin = clone->ds_prev;
2381 uint64_t comp, uncomp;
2382
2383 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2384 dsl_deadlist_space_range(&clone->ds_deadlist,
2385 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2386 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
2387 }
2388
2389 /* swap blkptrs */
2390 {
2391 blkptr_t tmp;
2392 tmp = origin_head->ds_phys->ds_bp;
2393 origin_head->ds_phys->ds_bp = clone->ds_phys->ds_bp;
2394 clone->ds_phys->ds_bp = tmp;
2395 }
2396
2397 /* set dd_*_bytes */
2398 {
2399 int64_t dused, dcomp, duncomp;
2400 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2401 uint64_t odl_used, odl_comp, odl_uncomp;
2402
2403 ASSERT3U(clone->ds_dir->dd_phys->
2404 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2405
2406 dsl_deadlist_space(&clone->ds_deadlist,
2407 &cdl_used, &cdl_comp, &cdl_uncomp);
2408 dsl_deadlist_space(&origin_head->ds_deadlist,
2409 &odl_used, &odl_comp, &odl_uncomp);
2410
2411 dused = clone->ds_phys->ds_referenced_bytes + cdl_used -
2412 (origin_head->ds_phys->ds_referenced_bytes + odl_used);
2413 dcomp = clone->ds_phys->ds_compressed_bytes + cdl_comp -
2414 (origin_head->ds_phys->ds_compressed_bytes + odl_comp);
2415 duncomp = clone->ds_phys->ds_uncompressed_bytes +
2416 cdl_uncomp -
2417 (origin_head->ds_phys->ds_uncompressed_bytes + odl_uncomp);
2418
2419 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2420 dused, dcomp, duncomp, tx);
2421 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2422 -dused, -dcomp, -duncomp, tx);
2423
2424 /*
2425 * The difference in the space used by snapshots is the
2426 * difference in snapshot space due to the head's
2427 * deadlist (since that's the only thing that's
2428 * changing that affects the snapused).
2429 */
2430 dsl_deadlist_space_range(&clone->ds_deadlist,
2431 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2432 &cdl_used, &cdl_comp, &cdl_uncomp);
2433 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2434 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2435 &odl_used, &odl_comp, &odl_uncomp);
2436 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2437 DD_USED_HEAD, DD_USED_SNAP, tx);
2438 }
2439
2440 /* swap ds_*_bytes */
2441 SWITCH64(origin_head->ds_phys->ds_referenced_bytes,
2442 clone->ds_phys->ds_referenced_bytes);
2443 SWITCH64(origin_head->ds_phys->ds_compressed_bytes,
2444 clone->ds_phys->ds_compressed_bytes);
2445 SWITCH64(origin_head->ds_phys->ds_uncompressed_bytes,
2446 clone->ds_phys->ds_uncompressed_bytes);
2447 SWITCH64(origin_head->ds_phys->ds_unique_bytes,
2448 clone->ds_phys->ds_unique_bytes);
2449
2450 /* apply any parent delta for change in unconsumed refreservation */
2451 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2452 unused_refres_delta, 0, 0, tx);
2453
2454 /*
2455 * Swap deadlists.
2456 */
2457 dsl_deadlist_close(&clone->ds_deadlist);
2458 dsl_deadlist_close(&origin_head->ds_deadlist);
2459 SWITCH64(origin_head->ds_phys->ds_deadlist_obj,
2460 clone->ds_phys->ds_deadlist_obj);
2461 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2462 clone->ds_phys->ds_deadlist_obj);
2463 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2464 origin_head->ds_phys->ds_deadlist_obj);
2465
2466 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2467
2468 spa_history_log_internal_ds(clone, "clone swap", tx,
2469 "parent=%s", origin_head->ds_dir->dd_myname);
2470 }
2471
2472 /*
2473 * Given a pool name and a dataset object number in that pool,
2474 * return the name of that dataset.
2475 */
2476 int
2477 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2478 {
2479 dsl_pool_t *dp;
2480 dsl_dataset_t *ds;
2481 int error;
2482
2483 error = dsl_pool_hold(pname, FTAG, &dp);
2484 if (error != 0)
2485 return (error);
2486
2487 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2488 if (error == 0) {
2489 dsl_dataset_name(ds, buf);
2490 dsl_dataset_rele(ds, FTAG);
2491 }
2492 dsl_pool_rele(dp, FTAG);
2493
2494 return (error);
2495 }
2496
2497 int
2498 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2499 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2500 {
2501 int error = 0;
2502
2503 ASSERT3S(asize, >, 0);
2504
2505 /*
2506 * *ref_rsrv is the portion of asize that will come from any
2507 * unconsumed refreservation space.
2508 */
2509 *ref_rsrv = 0;
2510
2511 mutex_enter(&ds->ds_lock);
2512 /*
2513 * Make a space adjustment for reserved bytes.
2514 */
2515 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
2516 ASSERT3U(*used, >=,
2517 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2518 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
2519 *ref_rsrv =
2520 asize - MIN(asize, parent_delta(ds, asize + inflight));
2521 }
2522
2523 if (!check_quota || ds->ds_quota == 0) {
2524 mutex_exit(&ds->ds_lock);
2525 return (0);
2526 }
2527 /*
2528 * If they are requesting more space, and our current estimate
2529 * is over quota, they get to try again unless the actual
2530 * on-disk is over quota and there are no pending changes (which
2531 * may free up space for us).
2532 */
2533 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
2534 if (inflight > 0 ||
2535 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
2536 error = SET_ERROR(ERESTART);
2537 else
2538 error = SET_ERROR(EDQUOT);
2539 }
2540 mutex_exit(&ds->ds_lock);
2541
2542 return (error);
2543 }
2544
2545 typedef struct dsl_dataset_set_qr_arg {
2546 const char *ddsqra_name;
2547 zprop_source_t ddsqra_source;
2548 uint64_t ddsqra_value;
2549 } dsl_dataset_set_qr_arg_t;
2550
2551
2552 /* ARGSUSED */
2553 static int
2554 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2555 {
2556 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2557 dsl_pool_t *dp = dmu_tx_pool(tx);
2558 dsl_dataset_t *ds;
2559 int error;
2560 uint64_t newval;
2561
2562 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2563 return (SET_ERROR(ENOTSUP));
2564
2565 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2566 if (error != 0)
2567 return (error);
2568
2569 if (dsl_dataset_is_snapshot(ds)) {
2570 dsl_dataset_rele(ds, FTAG);
2571 return (SET_ERROR(EINVAL));
2572 }
2573
2574 error = dsl_prop_predict(ds->ds_dir,
2575 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2576 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2577 if (error != 0) {
2578 dsl_dataset_rele(ds, FTAG);
2579 return (error);
2580 }
2581
2582 if (newval == 0) {
2583 dsl_dataset_rele(ds, FTAG);
2584 return (0);
2585 }
2586
2587 if (newval < ds->ds_phys->ds_referenced_bytes ||
2588 newval < ds->ds_reserved) {
2589 dsl_dataset_rele(ds, FTAG);
2590 return (SET_ERROR(ENOSPC));
2591 }
2592
2593 dsl_dataset_rele(ds, FTAG);
2594 return (0);
2595 }
2596
2597 static void
2598 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
2599 {
2600 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2601 dsl_pool_t *dp = dmu_tx_pool(tx);
2602 dsl_dataset_t *ds;
2603 uint64_t newval;
2604
2605 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2606
2607 dsl_prop_set_sync_impl(ds,
2608 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2609 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
2610 &ddsqra->ddsqra_value, tx);
2611
2612 VERIFY0(dsl_prop_get_int_ds(ds,
2613 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
2614
2615 if (ds->ds_quota != newval) {
2616 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2617 ds->ds_quota = newval;
2618 }
2619 dsl_dataset_rele(ds, FTAG);
2620 }
2621
2622 int
2623 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
2624 uint64_t refquota)
2625 {
2626 dsl_dataset_set_qr_arg_t ddsqra;
2627
2628 ddsqra.ddsqra_name = dsname;
2629 ddsqra.ddsqra_source = source;
2630 ddsqra.ddsqra_value = refquota;
2631
2632 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
2633 dsl_dataset_set_refquota_sync, &ddsqra, 0));
2634 }
2635
2636 static int
2637 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
2638 {
2639 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2640 dsl_pool_t *dp = dmu_tx_pool(tx);
2641 dsl_dataset_t *ds;
2642 int error;
2643 uint64_t newval, unique;
2644
2645 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
2646 return (SET_ERROR(ENOTSUP));
2647
2648 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2649 if (error != 0)
2650 return (error);
2651
2652 if (dsl_dataset_is_snapshot(ds)) {
2653 dsl_dataset_rele(ds, FTAG);
2654 return (SET_ERROR(EINVAL));
2655 }
2656
2657 error = dsl_prop_predict(ds->ds_dir,
2658 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2659 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2660 if (error != 0) {
2661 dsl_dataset_rele(ds, FTAG);
2662 return (error);
2663 }
2664
2665 /*
2666 * If we are doing the preliminary check in open context, the
2667 * space estimates may be inaccurate.
2668 */
2669 if (!dmu_tx_is_syncing(tx)) {
2670 dsl_dataset_rele(ds, FTAG);
2671 return (0);
2672 }
2673
2674 mutex_enter(&ds->ds_lock);
2675 if (!DS_UNIQUE_IS_ACCURATE(ds))
2676 dsl_dataset_recalc_head_uniq(ds);
2677 unique = ds->ds_phys->ds_unique_bytes;
2678 mutex_exit(&ds->ds_lock);
2679
2680 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
2681 uint64_t delta = MAX(unique, newval) -
2682 MAX(unique, ds->ds_reserved);
2683
2684 if (delta >
2685 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
2686 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
2687 dsl_dataset_rele(ds, FTAG);
2688 return (SET_ERROR(ENOSPC));
2689 }
2690 }
2691
2692 dsl_dataset_rele(ds, FTAG);
2693 return (0);
2694 }
2695
2696 void
2697 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
2698 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
2699 {
2700 uint64_t newval;
2701 uint64_t unique;
2702 int64_t delta;
2703
2704 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
2705 source, sizeof (value), 1, &value, tx);
2706
2707 VERIFY0(dsl_prop_get_int_ds(ds,
2708 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
2709
2710 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2711 mutex_enter(&ds->ds_dir->dd_lock);
2712 mutex_enter(&ds->ds_lock);
2713 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2714 unique = ds->ds_phys->ds_unique_bytes;
2715 delta = MAX(0, (int64_t)(newval - unique)) -
2716 MAX(0, (int64_t)(ds->ds_reserved - unique));
2717 ds->ds_reserved = newval;
2718 mutex_exit(&ds->ds_lock);
2719
2720 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
2721 mutex_exit(&ds->ds_dir->dd_lock);
2722 }
2723
2724 static void
2725 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
2726 {
2727 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2728 dsl_pool_t *dp = dmu_tx_pool(tx);
2729 dsl_dataset_t *ds;
2730
2731 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
2732 dsl_dataset_set_refreservation_sync_impl(ds,
2733 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
2734 dsl_dataset_rele(ds, FTAG);
2735 }
2736
2737 int
2738 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
2739 uint64_t refreservation)
2740 {
2741 dsl_dataset_set_qr_arg_t ddsqra;
2742
2743 ddsqra.ddsqra_name = dsname;
2744 ddsqra.ddsqra_source = source;
2745 ddsqra.ddsqra_value = refreservation;
2746
2747 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
2748 dsl_dataset_set_refreservation_sync, &ddsqra, 0));
2749 }
2750
2751 /*
2752 * Return (in *usedp) the amount of space written in new that is not
2753 * present in oldsnap. New may be a snapshot or the head. Old must be
2754 * a snapshot before new, in new's filesystem (or its origin). If not then
2755 * fail and return EINVAL.
2756 *
2757 * The written space is calculated by considering two components: First, we
2758 * ignore any freed space, and calculate the written as new's used space
2759 * minus old's used space. Next, we add in the amount of space that was freed
2760 * between the two snapshots, thus reducing new's used space relative to old's.
2761 * Specifically, this is the space that was born before old->ds_creation_txg,
2762 * and freed before new (ie. on new's deadlist or a previous deadlist).
2763 *
2764 * space freed [---------------------]
2765 * snapshots ---O-------O--------O-------O------
2766 * oldsnap new
2767 */
2768 int
2769 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
2770 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2771 {
2772 int err = 0;
2773 uint64_t snapobj;
2774 dsl_pool_t *dp = new->ds_dir->dd_pool;
2775
2776 ASSERT(dsl_pool_config_held(dp));
2777
2778 *usedp = 0;
2779 *usedp += new->ds_phys->ds_referenced_bytes;
2780 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
2781
2782 *compp = 0;
2783 *compp += new->ds_phys->ds_compressed_bytes;
2784 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
2785
2786 *uncompp = 0;
2787 *uncompp += new->ds_phys->ds_uncompressed_bytes;
2788 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
2789
2790 snapobj = new->ds_object;
2791 while (snapobj != oldsnap->ds_object) {
2792 dsl_dataset_t *snap;
2793 uint64_t used, comp, uncomp;
2794
2795 if (snapobj == new->ds_object) {
2796 snap = new;
2797 } else {
2798 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
2799 if (err != 0)
2800 break;
2801 }
2802
2803 if (snap->ds_phys->ds_prev_snap_txg ==
2804 oldsnap->ds_phys->ds_creation_txg) {
2805 /*
2806 * The blocks in the deadlist can not be born after
2807 * ds_prev_snap_txg, so get the whole deadlist space,
2808 * which is more efficient (especially for old-format
2809 * deadlists). Unfortunately the deadlist code
2810 * doesn't have enough information to make this
2811 * optimization itself.
2812 */
2813 dsl_deadlist_space(&snap->ds_deadlist,
2814 &used, &comp, &uncomp);
2815 } else {
2816 dsl_deadlist_space_range(&snap->ds_deadlist,
2817 0, oldsnap->ds_phys->ds_creation_txg,
2818 &used, &comp, &uncomp);
2819 }
2820 *usedp += used;
2821 *compp += comp;
2822 *uncompp += uncomp;
2823
2824 /*
2825 * If we get to the beginning of the chain of snapshots
2826 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
2827 * was not a snapshot of/before new.
2828 */
2829 snapobj = snap->ds_phys->ds_prev_snap_obj;
2830 if (snap != new)
2831 dsl_dataset_rele(snap, FTAG);
2832 if (snapobj == 0) {
2833 err = SET_ERROR(EINVAL);
2834 break;
2835 }
2836
2837 }
2838 return (err);
2839 }
2840
2841 /*
2842 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
2843 * lastsnap, and all snapshots in between are deleted.
2844 *
2845 * blocks that would be freed [---------------------------]
2846 * snapshots ---O-------O--------O-------O--------O
2847 * firstsnap lastsnap
2848 *
2849 * This is the set of blocks that were born after the snap before firstsnap,
2850 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
2851 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
2852 * We calculate this by iterating over the relevant deadlists (from the snap
2853 * after lastsnap, backward to the snap after firstsnap), summing up the
2854 * space on the deadlist that was born after the snap before firstsnap.
2855 */
2856 int
2857 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
2858 dsl_dataset_t *lastsnap,
2859 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
2860 {
2861 int err = 0;
2862 uint64_t snapobj;
2863 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
2864
2865 ASSERT(dsl_dataset_is_snapshot(firstsnap));
2866 ASSERT(dsl_dataset_is_snapshot(lastsnap));
2867
2868 /*
2869 * Check that the snapshots are in the same dsl_dir, and firstsnap
2870 * is before lastsnap.
2871 */
2872 if (firstsnap->ds_dir != lastsnap->ds_dir ||
2873 firstsnap->ds_phys->ds_creation_txg >
2874 lastsnap->ds_phys->ds_creation_txg)
2875 return (SET_ERROR(EINVAL));
2876
2877 *usedp = *compp = *uncompp = 0;
2878
2879 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
2880 while (snapobj != firstsnap->ds_object) {
2881 dsl_dataset_t *ds;
2882 uint64_t used, comp, uncomp;
2883
2884 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
2885 if (err != 0)
2886 break;
2887
2888 dsl_deadlist_space_range(&ds->ds_deadlist,
2889 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2890 &used, &comp, &uncomp);
2891 *usedp += used;
2892 *compp += comp;
2893 *uncompp += uncomp;
2894
2895 snapobj = ds->ds_phys->ds_prev_snap_obj;
2896 ASSERT3U(snapobj, !=, 0);
2897 dsl_dataset_rele(ds, FTAG);
2898 }
2899 return (err);
2900 }
2901
2902 /*
2903 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
2904 * For example, they could both be snapshots of the same filesystem, and
2905 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
2906 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
2907 * filesystem. Or 'earlier' could be the origin's origin.
2908 */
2909 boolean_t
2910 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier)
2911 {
2912 dsl_pool_t *dp = later->ds_dir->dd_pool;
2913 int error;
2914 boolean_t ret;
2915
2916 ASSERT(dsl_pool_config_held(dp));
2917
2918 if (earlier->ds_phys->ds_creation_txg >=
2919 later->ds_phys->ds_creation_txg)
2920 return (B_FALSE);
2921
2922 if (later->ds_dir == earlier->ds_dir)
2923 return (B_TRUE);
2924 if (!dsl_dir_is_clone(later->ds_dir))
2925 return (B_FALSE);
2926
2927 if (later->ds_dir->dd_phys->dd_origin_obj == earlier->ds_object)
2928 return (B_TRUE);
2929 dsl_dataset_t *origin;
2930 error = dsl_dataset_hold_obj(dp,
2931 later->ds_dir->dd_phys->dd_origin_obj, FTAG, &origin);
2932 if (error != 0)
2933 return (B_FALSE);
2934 ret = dsl_dataset_is_before(origin, earlier);
2935 dsl_dataset_rele(origin, FTAG);
2936 return (ret);
2937 }