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) 2011, 2015 by Delphix. All rights reserved.
24 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
25 * Copyright (c) 2014 RackTop Systems.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright (c) 2014 Integros [integros.com]
28 */
29
30 #include <sys/dmu_objset.h>
31 #include <sys/dsl_dataset.h>
32 #include <sys/dsl_dir.h>
33 #include <sys/dsl_prop.h>
34 #include <sys/dsl_synctask.h>
35 #include <sys/dmu_traverse.h>
36 #include <sys/dmu_impl.h>
37 #include <sys/dmu_tx.h>
38 #include <sys/arc.h>
39 #include <sys/zio.h>
40 #include <sys/zap.h>
41 #include <sys/zfeature.h>
42 #include <sys/unique.h>
43 #include <sys/zfs_context.h>
44 #include <sys/zfs_ioctl.h>
45 #include <sys/spa.h>
46 #include <sys/zfs_znode.h>
47 #include <sys/zfs_onexit.h>
48 #include <sys/zvol.h>
49 #include <sys/dsl_scan.h>
50 #include <sys/dsl_deadlist.h>
51 #include <sys/dsl_destroy.h>
52 #include <sys/dsl_userhold.h>
53 #include <sys/dsl_bookmark.h>
54 #include <sys/dmu_send.h>
55 #include <sys/zio_checksum.h>
56 #include <sys/zio_compress.h>
57 #include <zfs_fletcher.h>
58
59 /*
60 * The SPA supports block sizes up to 16MB. However, very large blocks
61 * can have an impact on i/o latency (e.g. tying up a spinning disk for
62 * ~300ms), and also potentially on the memory allocator. Therefore,
63 * we do not allow the recordsize to be set larger than zfs_max_recordsize
64 * (default 1MB). Larger blocks can be created by changing this tunable,
65 * and pools with larger blocks can always be imported and used, regardless
66 * of this setting.
67 */
68 int zfs_max_recordsize = 1 * 1024 * 1024;
69
70 #define SWITCH64(x, y) \
71 { \
72 uint64_t __tmp = (x); \
73 (x) = (y); \
74 (y) = __tmp; \
75 }
76
77 #define DS_REF_MAX (1ULL << 62)
78
79 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
80
81 /*
82 * Figure out how much of this delta should be propogated to the dsl_dir
83 * layer. If there's a refreservation, that space has already been
84 * partially accounted for in our ancestors.
85 */
86 static int64_t
87 parent_delta(dsl_dataset_t *ds, int64_t delta)
88 {
89 dsl_dataset_phys_t *ds_phys;
90 uint64_t old_bytes, new_bytes;
91
92 if (ds->ds_reserved == 0)
93 return (delta);
94
95 ds_phys = dsl_dataset_phys(ds);
96 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
97 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
98
99 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
100 return (new_bytes - old_bytes);
101 }
102
103 void
104 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
105 {
106 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
107 int compressed = BP_GET_PSIZE(bp);
108 int uncompressed = BP_GET_UCSIZE(bp);
109 int64_t delta;
110
111 dprintf_bp(bp, "ds=%p", ds);
112
113 ASSERT(dmu_tx_is_syncing(tx));
114 /* It could have been compressed away to nothing */
115 if (BP_IS_HOLE(bp))
116 return;
117 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
118 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
119 if (ds == NULL) {
120 dsl_pool_mos_diduse_space(tx->tx_pool,
121 used, compressed, uncompressed);
122 return;
123 }
124
125 dmu_buf_will_dirty(ds->ds_dbuf, tx);
126 mutex_enter(&ds->ds_lock);
127 delta = parent_delta(ds, used);
128 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
129 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
130 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
131 dsl_dataset_phys(ds)->ds_unique_bytes += used;
132
133 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
134 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
135 B_TRUE;
136 }
137
138 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
139 if (f != SPA_FEATURE_NONE)
140 ds->ds_feature_activation_needed[f] = B_TRUE;
141
142 mutex_exit(&ds->ds_lock);
143 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
144 compressed, uncompressed, tx);
145 dsl_dir_transfer_space(ds->ds_dir, used - delta,
146 DD_USED_REFRSRV, DD_USED_HEAD, tx);
147 }
148
149 int
150 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
151 boolean_t async)
152 {
153 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
154 int compressed = BP_GET_PSIZE(bp);
155 int uncompressed = BP_GET_UCSIZE(bp);
156
157 if (BP_IS_HOLE(bp))
158 return (0);
159
160 ASSERT(dmu_tx_is_syncing(tx));
161 ASSERT(bp->blk_birth <= tx->tx_txg);
162
163 if (ds == NULL) {
164 dsl_free(tx->tx_pool, tx->tx_txg, bp);
165 dsl_pool_mos_diduse_space(tx->tx_pool,
166 -used, -compressed, -uncompressed);
167 return (used);
168 }
169 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
170
171 ASSERT(!ds->ds_is_snapshot);
172 dmu_buf_will_dirty(ds->ds_dbuf, tx);
173
174 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
175 int64_t delta;
176
177 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
178 dsl_free(tx->tx_pool, tx->tx_txg, bp);
179
180 mutex_enter(&ds->ds_lock);
181 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
182 !DS_UNIQUE_IS_ACCURATE(ds));
183 delta = parent_delta(ds, -used);
184 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
185 mutex_exit(&ds->ds_lock);
186 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
187 delta, -compressed, -uncompressed, tx);
188 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
189 DD_USED_REFRSRV, DD_USED_HEAD, tx);
190 } else {
191 dprintf_bp(bp, "putting on dead list: %s", "");
192 if (async) {
193 /*
194 * We are here as part of zio's write done callback,
195 * which means we're a zio interrupt thread. We can't
196 * call dsl_deadlist_insert() now because it may block
197 * waiting for I/O. Instead, put bp on the deferred
198 * queue and let dsl_pool_sync() finish the job.
199 */
200 bplist_append(&ds->ds_pending_deadlist, bp);
201 } else {
202 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
203 }
204 ASSERT3U(ds->ds_prev->ds_object, ==,
205 dsl_dataset_phys(ds)->ds_prev_snap_obj);
206 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
207 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
208 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
209 ds->ds_object && bp->blk_birth >
210 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
211 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
212 mutex_enter(&ds->ds_prev->ds_lock);
213 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
214 mutex_exit(&ds->ds_prev->ds_lock);
215 }
216 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
217 dsl_dir_transfer_space(ds->ds_dir, used,
218 DD_USED_HEAD, DD_USED_SNAP, tx);
219 }
220 }
221 mutex_enter(&ds->ds_lock);
222 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
223 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
224 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
225 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
226 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
227 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
228 mutex_exit(&ds->ds_lock);
229
230 return (used);
231 }
232
233 uint64_t
234 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
235 {
236 uint64_t trysnap = 0;
237
238 if (ds == NULL)
239 return (0);
240 /*
241 * The snapshot creation could fail, but that would cause an
242 * incorrect FALSE return, which would only result in an
243 * overestimation of the amount of space that an operation would
244 * consume, which is OK.
245 *
246 * There's also a small window where we could miss a pending
247 * snapshot, because we could set the sync task in the quiescing
248 * phase. So this should only be used as a guess.
249 */
250 if (ds->ds_trysnap_txg >
251 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
252 trysnap = ds->ds_trysnap_txg;
253 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
254 }
255
256 boolean_t
257 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
258 uint64_t blk_birth)
259 {
260 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
261 (bp != NULL && BP_IS_HOLE(bp)))
262 return (B_FALSE);
263
264 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
265
266 return (B_TRUE);
267 }
268
269 static void
270 dsl_dataset_evict(void *dbu)
271 {
272 dsl_dataset_t *ds = dbu;
273
274 ASSERT(ds->ds_owner == NULL);
275
276 ds->ds_dbuf = NULL;
277
278 unique_remove(ds->ds_fsid_guid);
279
280 if (ds->ds_objset != NULL)
281 dmu_objset_evict(ds->ds_objset);
282
283 if (ds->ds_prev) {
284 dsl_dataset_rele(ds->ds_prev, ds);
285 ds->ds_prev = NULL;
286 }
287
288 bplist_destroy(&ds->ds_pending_deadlist);
289 if (ds->ds_deadlist.dl_os != NULL)
290 dsl_deadlist_close(&ds->ds_deadlist);
291 if (ds->ds_dir)
292 dsl_dir_async_rele(ds->ds_dir, ds);
293
294 ASSERT(!list_link_active(&ds->ds_synced_link));
295
296 list_destroy(&ds->ds_prop_cbs);
297 mutex_destroy(&ds->ds_lock);
298 mutex_destroy(&ds->ds_opening_lock);
299 mutex_destroy(&ds->ds_sendstream_lock);
300 refcount_destroy(&ds->ds_longholds);
301
302 kmem_free(ds, sizeof (dsl_dataset_t));
303 }
304
305 int
306 dsl_dataset_get_snapname(dsl_dataset_t *ds)
307 {
308 dsl_dataset_phys_t *headphys;
309 int err;
310 dmu_buf_t *headdbuf;
311 dsl_pool_t *dp = ds->ds_dir->dd_pool;
312 objset_t *mos = dp->dp_meta_objset;
313
314 if (ds->ds_snapname[0])
315 return (0);
316 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
317 return (0);
318
319 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
320 FTAG, &headdbuf);
321 if (err != 0)
322 return (err);
323 headphys = headdbuf->db_data;
324 err = zap_value_search(dp->dp_meta_objset,
325 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
326 dmu_buf_rele(headdbuf, FTAG);
327 return (err);
328 }
329
330 int
331 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
332 {
333 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
334 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
335 matchtype_t mt;
336 int err;
337
338 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
339 mt = MT_FIRST;
340 else
341 mt = MT_EXACT;
342
343 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
344 value, mt, NULL, 0, NULL);
345 if (err == ENOTSUP && mt == MT_FIRST)
346 err = zap_lookup(mos, snapobj, name, 8, 1, value);
347 return (err);
348 }
349
350 int
351 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
352 boolean_t adj_cnt)
353 {
354 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
355 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
356 matchtype_t mt;
357 int err;
358
359 dsl_dir_snap_cmtime_update(ds->ds_dir);
360
361 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
362 mt = MT_FIRST;
363 else
364 mt = MT_EXACT;
365
366 err = zap_remove_norm(mos, snapobj, name, mt, tx);
367 if (err == ENOTSUP && mt == MT_FIRST)
368 err = zap_remove(mos, snapobj, name, tx);
369
370 if (err == 0 && adj_cnt)
371 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
372 DD_FIELD_SNAPSHOT_COUNT, tx);
373
374 return (err);
375 }
376
377 boolean_t
378 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
379 {
380 dmu_buf_t *dbuf = ds->ds_dbuf;
381 boolean_t result = B_FALSE;
382
383 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
384 ds->ds_object, DMU_BONUS_BLKID, tag)) {
385
386 if (ds == dmu_buf_get_user(dbuf))
387 result = B_TRUE;
388 else
389 dmu_buf_rele(dbuf, tag);
390 }
391
392 return (result);
393 }
394
395 int
396 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
397 dsl_dataset_t **dsp)
398 {
399 objset_t *mos = dp->dp_meta_objset;
400 dmu_buf_t *dbuf;
401 dsl_dataset_t *ds;
402 int err;
403 dmu_object_info_t doi;
404
405 ASSERT(dsl_pool_config_held(dp));
406
407 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
408 if (err != 0)
409 return (err);
410
411 /* Make sure dsobj has the correct object type. */
412 dmu_object_info_from_db(dbuf, &doi);
413 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
414 dmu_buf_rele(dbuf, tag);
415 return (SET_ERROR(EINVAL));
416 }
417
418 ds = dmu_buf_get_user(dbuf);
419 if (ds == NULL) {
420 dsl_dataset_t *winner = NULL;
421
422 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
423 ds->ds_dbuf = dbuf;
424 ds->ds_object = dsobj;
425 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
426
427 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
428 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
429 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
430 refcount_create(&ds->ds_longholds);
431
432 bplist_create(&ds->ds_pending_deadlist);
433 dsl_deadlist_open(&ds->ds_deadlist,
434 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
435
436 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
437 offsetof(dmu_sendarg_t, dsa_link));
438
439 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
440 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
441
442 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
443 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
444 if (!(spa_feature_table[f].fi_flags &
445 ZFEATURE_FLAG_PER_DATASET))
446 continue;
447 err = zap_contains(mos, dsobj,
448 spa_feature_table[f].fi_guid);
449 if (err == 0) {
450 ds->ds_feature_inuse[f] = B_TRUE;
451 } else {
452 ASSERT3U(err, ==, ENOENT);
453 err = 0;
454 }
455 }
456 }
457
458 err = dsl_dir_hold_obj(dp,
459 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
460 if (err != 0) {
461 mutex_destroy(&ds->ds_lock);
462 mutex_destroy(&ds->ds_opening_lock);
463 mutex_destroy(&ds->ds_sendstream_lock);
464 refcount_destroy(&ds->ds_longholds);
465 bplist_destroy(&ds->ds_pending_deadlist);
466 dsl_deadlist_close(&ds->ds_deadlist);
467 kmem_free(ds, sizeof (dsl_dataset_t));
468 dmu_buf_rele(dbuf, tag);
469 return (err);
470 }
471
472 if (!ds->ds_is_snapshot) {
473 ds->ds_snapname[0] = '\0';
474 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
475 err = dsl_dataset_hold_obj(dp,
476 dsl_dataset_phys(ds)->ds_prev_snap_obj,
477 ds, &ds->ds_prev);
478 }
479 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
480 int zaperr = zap_lookup(mos, ds->ds_object,
481 DS_FIELD_BOOKMARK_NAMES,
482 sizeof (ds->ds_bookmarks), 1,
483 &ds->ds_bookmarks);
484 if (zaperr != ENOENT)
485 VERIFY0(zaperr);
486 }
487 } else {
488 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
489 err = dsl_dataset_get_snapname(ds);
490 if (err == 0 &&
491 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
492 err = zap_count(
493 ds->ds_dir->dd_pool->dp_meta_objset,
494 dsl_dataset_phys(ds)->ds_userrefs_obj,
495 &ds->ds_userrefs);
496 }
497 }
498
499 if (err == 0 && !ds->ds_is_snapshot) {
500 err = dsl_prop_get_int_ds(ds,
501 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
502 &ds->ds_reserved);
503 if (err == 0) {
504 err = dsl_prop_get_int_ds(ds,
505 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
506 &ds->ds_quota);
507 }
508 } else {
509 ds->ds_reserved = ds->ds_quota = 0;
510 }
511
512 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
513 if (err == 0)
514 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
515
516 if (err != 0 || winner != NULL) {
517 bplist_destroy(&ds->ds_pending_deadlist);
518 dsl_deadlist_close(&ds->ds_deadlist);
519 if (ds->ds_prev)
520 dsl_dataset_rele(ds->ds_prev, ds);
521 dsl_dir_rele(ds->ds_dir, ds);
522 mutex_destroy(&ds->ds_lock);
523 mutex_destroy(&ds->ds_opening_lock);
524 mutex_destroy(&ds->ds_sendstream_lock);
525 refcount_destroy(&ds->ds_longholds);
526 kmem_free(ds, sizeof (dsl_dataset_t));
527 if (err != 0) {
528 dmu_buf_rele(dbuf, tag);
529 return (err);
530 }
531 ds = winner;
532 } else {
533 ds->ds_fsid_guid =
534 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
535 }
536 }
537 ASSERT3P(ds->ds_dbuf, ==, dbuf);
538 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
539 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
540 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
541 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
542 *dsp = ds;
543 return (0);
544 }
545
546 int
547 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
548 void *tag, dsl_dataset_t **dsp)
549 {
550 dsl_dir_t *dd;
551 const char *snapname;
552 uint64_t obj;
553 int err = 0;
554 dsl_dataset_t *ds;
555
556 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
557 if (err != 0)
558 return (err);
559
560 ASSERT(dsl_pool_config_held(dp));
561 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
562 if (obj != 0)
563 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
564 else
565 err = SET_ERROR(ENOENT);
566
567 /* we may be looking for a snapshot */
568 if (err == 0 && snapname != NULL) {
569 dsl_dataset_t *snap_ds;
570
571 if (*snapname++ != '@') {
572 dsl_dataset_rele(ds, tag);
573 dsl_dir_rele(dd, FTAG);
574 return (SET_ERROR(ENOENT));
575 }
576
577 dprintf("looking for snapshot '%s'\n", snapname);
578 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
579 if (err == 0)
580 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
581 dsl_dataset_rele(ds, tag);
582
583 if (err == 0) {
584 mutex_enter(&snap_ds->ds_lock);
585 if (snap_ds->ds_snapname[0] == 0)
586 (void) strlcpy(snap_ds->ds_snapname, snapname,
587 sizeof (snap_ds->ds_snapname));
588 mutex_exit(&snap_ds->ds_lock);
589 ds = snap_ds;
590 }
591 }
592 if (err == 0)
593 *dsp = ds;
594 dsl_dir_rele(dd, FTAG);
595 return (err);
596 }
597
598 int
599 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
600 void *tag, dsl_dataset_t **dsp)
601 {
602 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
603 if (err != 0)
604 return (err);
605 if (!dsl_dataset_tryown(*dsp, tag)) {
606 dsl_dataset_rele(*dsp, tag);
607 *dsp = NULL;
608 return (SET_ERROR(EBUSY));
609 }
610 return (0);
611 }
612
613 int
614 dsl_dataset_own(dsl_pool_t *dp, const char *name,
615 void *tag, dsl_dataset_t **dsp)
616 {
617 int err = dsl_dataset_hold(dp, name, tag, dsp);
618 if (err != 0)
619 return (err);
620 if (!dsl_dataset_tryown(*dsp, tag)) {
621 dsl_dataset_rele(*dsp, tag);
622 return (SET_ERROR(EBUSY));
623 }
624 return (0);
625 }
626
627 /*
628 * See the comment above dsl_pool_hold() for details. In summary, a long
629 * hold is used to prevent destruction of a dataset while the pool hold
630 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
631 *
632 * The dataset and pool must be held when this function is called. After it
633 * is called, the pool hold may be released while the dataset is still held
634 * and accessed.
635 */
636 void
637 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
638 {
639 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
640 (void) refcount_add(&ds->ds_longholds, tag);
641 }
642
643 void
644 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
645 {
646 (void) refcount_remove(&ds->ds_longholds, tag);
647 }
648
649 /* Return B_TRUE if there are any long holds on this dataset. */
650 boolean_t
651 dsl_dataset_long_held(dsl_dataset_t *ds)
652 {
653 return (!refcount_is_zero(&ds->ds_longholds));
654 }
655
656 void
657 dsl_dataset_name(dsl_dataset_t *ds, char *name)
658 {
659 if (ds == NULL) {
660 (void) strcpy(name, "mos");
661 } else {
662 dsl_dir_name(ds->ds_dir, name);
663 VERIFY0(dsl_dataset_get_snapname(ds));
664 if (ds->ds_snapname[0]) {
665 (void) strcat(name, "@");
666 /*
667 * We use a "recursive" mutex so that we
668 * can call dprintf_ds() with ds_lock held.
669 */
670 if (!MUTEX_HELD(&ds->ds_lock)) {
671 mutex_enter(&ds->ds_lock);
672 (void) strcat(name, ds->ds_snapname);
673 mutex_exit(&ds->ds_lock);
674 } else {
675 (void) strcat(name, ds->ds_snapname);
676 }
677 }
678 }
679 }
680
681 void
682 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
683 {
684 dmu_buf_rele(ds->ds_dbuf, tag);
685 }
686
687 void
688 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
689 {
690 ASSERT3P(ds->ds_owner, ==, tag);
691 ASSERT(ds->ds_dbuf != NULL);
692
693 mutex_enter(&ds->ds_lock);
694 ds->ds_owner = NULL;
695 mutex_exit(&ds->ds_lock);
696 dsl_dataset_long_rele(ds, tag);
697 dsl_dataset_rele(ds, tag);
698 }
699
700 boolean_t
701 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
702 {
703 boolean_t gotit = FALSE;
704
705 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
706 mutex_enter(&ds->ds_lock);
707 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
708 ds->ds_owner = tag;
709 dsl_dataset_long_hold(ds, tag);
710 gotit = TRUE;
711 }
712 mutex_exit(&ds->ds_lock);
713 return (gotit);
714 }
715
716 boolean_t
717 dsl_dataset_has_owner(dsl_dataset_t *ds)
718 {
719 boolean_t rv;
720 mutex_enter(&ds->ds_lock);
721 rv = (ds->ds_owner != NULL);
722 mutex_exit(&ds->ds_lock);
723 return (rv);
724 }
725
726 static void
727 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
728 {
729 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
730 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
731 uint64_t zero = 0;
732
733 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
734
735 spa_feature_incr(spa, f, tx);
736 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
737
738 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
739 sizeof (zero), 1, &zero, tx));
740 }
741
742 void
743 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
744 {
745 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
746 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
747
748 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
749
750 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
751 spa_feature_decr(spa, f, tx);
752 }
753
754 uint64_t
755 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
756 uint64_t flags, dmu_tx_t *tx)
757 {
758 dsl_pool_t *dp = dd->dd_pool;
759 dmu_buf_t *dbuf;
760 dsl_dataset_phys_t *dsphys;
761 uint64_t dsobj;
762 objset_t *mos = dp->dp_meta_objset;
763
764 if (origin == NULL)
765 origin = dp->dp_origin_snap;
766
767 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
768 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
769 ASSERT(dmu_tx_is_syncing(tx));
770 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
771
772 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
773 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
774 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
775 dmu_buf_will_dirty(dbuf, tx);
776 dsphys = dbuf->db_data;
777 bzero(dsphys, sizeof (dsl_dataset_phys_t));
778 dsphys->ds_dir_obj = dd->dd_object;
779 dsphys->ds_flags = flags;
780 dsphys->ds_fsid_guid = unique_create();
781 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
782 sizeof (dsphys->ds_guid));
783 dsphys->ds_snapnames_zapobj =
784 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
785 DMU_OT_NONE, 0, tx);
786 dsphys->ds_creation_time = gethrestime_sec();
787 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
788
789 if (origin == NULL) {
790 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
791 } else {
792 dsl_dataset_t *ohds; /* head of the origin snapshot */
793
794 dsphys->ds_prev_snap_obj = origin->ds_object;
795 dsphys->ds_prev_snap_txg =
796 dsl_dataset_phys(origin)->ds_creation_txg;
797 dsphys->ds_referenced_bytes =
798 dsl_dataset_phys(origin)->ds_referenced_bytes;
799 dsphys->ds_compressed_bytes =
800 dsl_dataset_phys(origin)->ds_compressed_bytes;
801 dsphys->ds_uncompressed_bytes =
802 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
803 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
804
805 /*
806 * Inherit flags that describe the dataset's contents
807 * (INCONSISTENT) or properties (Case Insensitive).
808 */
809 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
810 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
811
812 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
813 if (origin->ds_feature_inuse[f])
814 dsl_dataset_activate_feature(dsobj, f, tx);
815 }
816
817 dmu_buf_will_dirty(origin->ds_dbuf, tx);
818 dsl_dataset_phys(origin)->ds_num_children++;
819
820 VERIFY0(dsl_dataset_hold_obj(dp,
821 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
822 FTAG, &ohds));
823 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
824 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
825 dsl_dataset_rele(ohds, FTAG);
826
827 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
828 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
829 dsl_dataset_phys(origin)->ds_next_clones_obj =
830 zap_create(mos,
831 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
832 }
833 VERIFY0(zap_add_int(mos,
834 dsl_dataset_phys(origin)->ds_next_clones_obj,
835 dsobj, tx));
836 }
837
838 dmu_buf_will_dirty(dd->dd_dbuf, tx);
839 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
840 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
841 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
842 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
843 dsl_dir_phys(origin->ds_dir)->dd_clones =
844 zap_create(mos,
845 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
846 }
847 VERIFY0(zap_add_int(mos,
848 dsl_dir_phys(origin->ds_dir)->dd_clones,
849 dsobj, tx));
850 }
851 }
852
853 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
854 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
855
856 dmu_buf_rele(dbuf, FTAG);
857
858 dmu_buf_will_dirty(dd->dd_dbuf, tx);
859 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
860
861 return (dsobj);
862 }
863
864 static void
865 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
866 {
867 objset_t *os;
868
869 VERIFY0(dmu_objset_from_ds(ds, &os));
870 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
871 dsl_dataset_dirty(ds, tx);
872 }
873
874 uint64_t
875 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
876 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
877 {
878 dsl_pool_t *dp = pdd->dd_pool;
879 uint64_t dsobj, ddobj;
880 dsl_dir_t *dd;
881
882 ASSERT(dmu_tx_is_syncing(tx));
883 ASSERT(lastname[0] != '@');
884
885 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
886 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
887
888 dsobj = dsl_dataset_create_sync_dd(dd, origin,
889 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
890
891 dsl_deleg_set_create_perms(dd, tx, cr);
892
893 /*
894 * Since we're creating a new node we know it's a leaf, so we can
895 * initialize the counts if the limit feature is active.
896 */
897 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
898 uint64_t cnt = 0;
899 objset_t *os = dd->dd_pool->dp_meta_objset;
900
901 dsl_dir_zapify(dd, tx);
902 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
903 sizeof (cnt), 1, &cnt, tx));
904 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
905 sizeof (cnt), 1, &cnt, tx));
906 }
907
908 dsl_dir_rele(dd, FTAG);
909
910 /*
911 * If we are creating a clone, make sure we zero out any stale
912 * data from the origin snapshots zil header.
913 */
914 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
915 dsl_dataset_t *ds;
916
917 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
918 dsl_dataset_zero_zil(ds, tx);
919 dsl_dataset_rele(ds, FTAG);
920 }
921
922 return (dsobj);
923 }
924
925 /*
926 * The unique space in the head dataset can be calculated by subtracting
927 * the space used in the most recent snapshot, that is still being used
928 * in this file system, from the space currently in use. To figure out
929 * the space in the most recent snapshot still in use, we need to take
930 * the total space used in the snapshot and subtract out the space that
931 * has been freed up since the snapshot was taken.
932 */
933 void
934 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
935 {
936 uint64_t mrs_used;
937 uint64_t dlused, dlcomp, dluncomp;
938
939 ASSERT(!ds->ds_is_snapshot);
940
941 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
942 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
943 else
944 mrs_used = 0;
945
946 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
947
948 ASSERT3U(dlused, <=, mrs_used);
949 dsl_dataset_phys(ds)->ds_unique_bytes =
950 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
951
952 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
953 SPA_VERSION_UNIQUE_ACCURATE)
954 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
955 }
956
957 void
958 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
959 dmu_tx_t *tx)
960 {
961 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
962 uint64_t count;
963 int err;
964
965 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
966 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
967 obj, tx);
968 /*
969 * The err should not be ENOENT, but a bug in a previous version
970 * of the code could cause upgrade_clones_cb() to not set
971 * ds_next_snap_obj when it should, leading to a missing entry.
972 * If we knew that the pool was created after
973 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
974 * ENOENT. However, at least we can check that we don't have
975 * too many entries in the next_clones_obj even after failing to
976 * remove this one.
977 */
978 if (err != ENOENT)
979 VERIFY0(err);
980 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
981 &count));
982 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
983 }
984
985
986 blkptr_t *
987 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
988 {
989 return (&dsl_dataset_phys(ds)->ds_bp);
990 }
991
992 void
993 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
994 {
995 ASSERT(dmu_tx_is_syncing(tx));
996 /* If it's the meta-objset, set dp_meta_rootbp */
997 if (ds == NULL) {
998 tx->tx_pool->dp_meta_rootbp = *bp;
999 } else {
1000 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1001 dsl_dataset_phys(ds)->ds_bp = *bp;
1002 }
1003 }
1004
1005 spa_t *
1006 dsl_dataset_get_spa(dsl_dataset_t *ds)
1007 {
1008 return (ds->ds_dir->dd_pool->dp_spa);
1009 }
1010
1011 void
1012 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1013 {
1014 dsl_pool_t *dp;
1015
1016 if (ds == NULL) /* this is the meta-objset */
1017 return;
1018
1019 ASSERT(ds->ds_objset != NULL);
1020
1021 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1022 panic("dirtying snapshot!");
1023
1024 dp = ds->ds_dir->dd_pool;
1025
1026 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1027 /* up the hold count until we can be written out */
1028 dmu_buf_add_ref(ds->ds_dbuf, ds);
1029 }
1030 }
1031
1032 boolean_t
1033 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1034 {
1035 for (int t = 0; t < TXG_SIZE; t++) {
1036 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1037 ds, t))
1038 return (B_TRUE);
1039 }
1040 return (B_FALSE);
1041 }
1042
1043 static int
1044 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1045 {
1046 uint64_t asize;
1047
1048 if (!dmu_tx_is_syncing(tx))
1049 return (0);
1050
1051 /*
1052 * If there's an fs-only reservation, any blocks that might become
1053 * owned by the snapshot dataset must be accommodated by space
1054 * outside of the reservation.
1055 */
1056 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1057 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1058 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1059 return (SET_ERROR(ENOSPC));
1060
1061 /*
1062 * Propagate any reserved space for this snapshot to other
1063 * snapshot checks in this sync group.
1064 */
1065 if (asize > 0)
1066 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1067
1068 return (0);
1069 }
1070
1071 typedef struct dsl_dataset_snapshot_arg {
1072 nvlist_t *ddsa_snaps;
1073 nvlist_t *ddsa_props;
1074 nvlist_t *ddsa_errors;
1075 cred_t *ddsa_cr;
1076 } dsl_dataset_snapshot_arg_t;
1077
1078 int
1079 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1080 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1081 {
1082 int error;
1083 uint64_t value;
1084
1085 ds->ds_trysnap_txg = tx->tx_txg;
1086
1087 if (!dmu_tx_is_syncing(tx))
1088 return (0);
1089
1090 /*
1091 * We don't allow multiple snapshots of the same txg. If there
1092 * is already one, try again.
1093 */
1094 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1095 return (SET_ERROR(EAGAIN));
1096
1097 /*
1098 * Check for conflicting snapshot name.
1099 */
1100 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1101 if (error == 0)
1102 return (SET_ERROR(EEXIST));
1103 if (error != ENOENT)
1104 return (error);
1105
1106 /*
1107 * We don't allow taking snapshots of inconsistent datasets, such as
1108 * those into which we are currently receiving. However, if we are
1109 * creating this snapshot as part of a receive, this check will be
1110 * executed atomically with respect to the completion of the receive
1111 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1112 * case we ignore this, knowing it will be fixed up for us shortly in
1113 * dmu_recv_end_sync().
1114 */
1115 if (!recv && DS_IS_INCONSISTENT(ds))
1116 return (SET_ERROR(EBUSY));
1117
1118 /*
1119 * Skip the check for temporary snapshots or if we have already checked
1120 * the counts in dsl_dataset_snapshot_check. This means we really only
1121 * check the count here when we're receiving a stream.
1122 */
1123 if (cnt != 0 && cr != NULL) {
1124 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1125 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1126 if (error != 0)
1127 return (error);
1128 }
1129
1130 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1131 if (error != 0)
1132 return (error);
1133
1134 return (0);
1135 }
1136
1137 static int
1138 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1139 {
1140 dsl_dataset_snapshot_arg_t *ddsa = arg;
1141 dsl_pool_t *dp = dmu_tx_pool(tx);
1142 nvpair_t *pair;
1143 int rv = 0;
1144
1145 /*
1146 * Pre-compute how many total new snapshots will be created for each
1147 * level in the tree and below. This is needed for validating the
1148 * snapshot limit when either taking a recursive snapshot or when
1149 * taking multiple snapshots.
1150 *
1151 * The problem is that the counts are not actually adjusted when
1152 * we are checking, only when we finally sync. For a single snapshot,
1153 * this is easy, the count will increase by 1 at each node up the tree,
1154 * but its more complicated for the recursive/multiple snapshot case.
1155 *
1156 * The dsl_fs_ss_limit_check function does recursively check the count
1157 * at each level up the tree but since it is validating each snapshot
1158 * independently we need to be sure that we are validating the complete
1159 * count for the entire set of snapshots. We do this by rolling up the
1160 * counts for each component of the name into an nvlist and then
1161 * checking each of those cases with the aggregated count.
1162 *
1163 * This approach properly handles not only the recursive snapshot
1164 * case (where we get all of those on the ddsa_snaps list) but also
1165 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1166 * validate the limit on 'a' using a count of 2).
1167 *
1168 * We validate the snapshot names in the third loop and only report
1169 * name errors once.
1170 */
1171 if (dmu_tx_is_syncing(tx)) {
1172 nvlist_t *cnt_track = NULL;
1173 cnt_track = fnvlist_alloc();
1174
1175 /* Rollup aggregated counts into the cnt_track list */
1176 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1177 pair != NULL;
1178 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1179 char *pdelim;
1180 uint64_t val;
1181 char nm[MAXPATHLEN];
1182
1183 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1184 pdelim = strchr(nm, '@');
1185 if (pdelim == NULL)
1186 continue;
1187 *pdelim = '\0';
1188
1189 do {
1190 if (nvlist_lookup_uint64(cnt_track, nm,
1191 &val) == 0) {
1192 /* update existing entry */
1193 fnvlist_add_uint64(cnt_track, nm,
1194 val + 1);
1195 } else {
1196 /* add to list */
1197 fnvlist_add_uint64(cnt_track, nm, 1);
1198 }
1199
1200 pdelim = strrchr(nm, '/');
1201 if (pdelim != NULL)
1202 *pdelim = '\0';
1203 } while (pdelim != NULL);
1204 }
1205
1206 /* Check aggregated counts at each level */
1207 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1208 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1209 int error = 0;
1210 char *name;
1211 uint64_t cnt = 0;
1212 dsl_dataset_t *ds;
1213
1214 name = nvpair_name(pair);
1215 cnt = fnvpair_value_uint64(pair);
1216 ASSERT(cnt > 0);
1217
1218 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1219 if (error == 0) {
1220 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1221 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1222 ddsa->ddsa_cr);
1223 dsl_dataset_rele(ds, FTAG);
1224 }
1225
1226 if (error != 0) {
1227 if (ddsa->ddsa_errors != NULL)
1228 fnvlist_add_int32(ddsa->ddsa_errors,
1229 name, error);
1230 rv = error;
1231 /* only report one error for this check */
1232 break;
1233 }
1234 }
1235 nvlist_free(cnt_track);
1236 }
1237
1238 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1239 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1240 int error = 0;
1241 dsl_dataset_t *ds;
1242 char *name, *atp;
1243 char dsname[MAXNAMELEN];
1244
1245 name = nvpair_name(pair);
1246 if (strlen(name) >= MAXNAMELEN)
1247 error = SET_ERROR(ENAMETOOLONG);
1248 if (error == 0) {
1249 atp = strchr(name, '@');
1250 if (atp == NULL)
1251 error = SET_ERROR(EINVAL);
1252 if (error == 0)
1253 (void) strlcpy(dsname, name, atp - name + 1);
1254 }
1255 if (error == 0)
1256 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1257 if (error == 0) {
1258 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1259 error = dsl_dataset_snapshot_check_impl(ds,
1260 atp + 1, tx, B_FALSE, 0, NULL);
1261 dsl_dataset_rele(ds, FTAG);
1262 }
1263
1264 if (error != 0) {
1265 if (ddsa->ddsa_errors != NULL) {
1266 fnvlist_add_int32(ddsa->ddsa_errors,
1267 name, error);
1268 }
1269 rv = error;
1270 }
1271 }
1272
1273 return (rv);
1274 }
1275
1276 void
1277 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1278 dmu_tx_t *tx)
1279 {
1280 static zil_header_t zero_zil;
1281
1282 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1283 dmu_buf_t *dbuf;
1284 dsl_dataset_phys_t *dsphys;
1285 uint64_t dsobj, crtxg;
1286 objset_t *mos = dp->dp_meta_objset;
1287 objset_t *os;
1288
1289 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1290
1291 /*
1292 * If we are on an old pool, the zil must not be active, in which
1293 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1294 */
1295 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1296 dmu_objset_from_ds(ds, &os) != 0 ||
1297 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1298 sizeof (zero_zil)) == 0);
1299
1300 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1301
1302 /*
1303 * The origin's ds_creation_txg has to be < TXG_INITIAL
1304 */
1305 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1306 crtxg = 1;
1307 else
1308 crtxg = tx->tx_txg;
1309
1310 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1311 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1312 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1313 dmu_buf_will_dirty(dbuf, tx);
1314 dsphys = dbuf->db_data;
1315 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1316 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1317 dsphys->ds_fsid_guid = unique_create();
1318 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1319 sizeof (dsphys->ds_guid));
1320 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1321 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1322 dsphys->ds_next_snap_obj = ds->ds_object;
1323 dsphys->ds_num_children = 1;
1324 dsphys->ds_creation_time = gethrestime_sec();
1325 dsphys->ds_creation_txg = crtxg;
1326 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1327 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1328 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1329 dsphys->ds_uncompressed_bytes =
1330 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1331 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1332 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1333 dmu_buf_rele(dbuf, FTAG);
1334
1335 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1336 if (ds->ds_feature_inuse[f])
1337 dsl_dataset_activate_feature(dsobj, f, tx);
1338 }
1339
1340 ASSERT3U(ds->ds_prev != 0, ==,
1341 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1342 if (ds->ds_prev) {
1343 uint64_t next_clones_obj =
1344 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1345 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1346 ds->ds_object ||
1347 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1348 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1349 ds->ds_object) {
1350 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1351 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1352 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1353 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1354 } else if (next_clones_obj != 0) {
1355 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1356 dsphys->ds_next_snap_obj, tx);
1357 VERIFY0(zap_add_int(mos,
1358 next_clones_obj, dsobj, tx));
1359 }
1360 }
1361
1362 /*
1363 * If we have a reference-reservation on this dataset, we will
1364 * need to increase the amount of refreservation being charged
1365 * since our unique space is going to zero.
1366 */
1367 if (ds->ds_reserved) {
1368 int64_t delta;
1369 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1370 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1371 ds->ds_reserved);
1372 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1373 delta, 0, 0, tx);
1374 }
1375
1376 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1377 dsl_dataset_phys(ds)->ds_deadlist_obj =
1378 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1379 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1380 dsl_deadlist_close(&ds->ds_deadlist);
1381 dsl_deadlist_open(&ds->ds_deadlist, mos,
1382 dsl_dataset_phys(ds)->ds_deadlist_obj);
1383 dsl_deadlist_add_key(&ds->ds_deadlist,
1384 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1385
1386 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1387 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1388 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1389 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1390 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1391 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1392
1393 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1394 snapname, 8, 1, &dsobj, tx));
1395
1396 if (ds->ds_prev)
1397 dsl_dataset_rele(ds->ds_prev, ds);
1398 VERIFY0(dsl_dataset_hold_obj(dp,
1399 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1400
1401 dsl_scan_ds_snapshotted(ds, tx);
1402
1403 dsl_dir_snap_cmtime_update(ds->ds_dir);
1404
1405 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1406 }
1407
1408 static void
1409 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1410 {
1411 dsl_dataset_snapshot_arg_t *ddsa = arg;
1412 dsl_pool_t *dp = dmu_tx_pool(tx);
1413 nvpair_t *pair;
1414
1415 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1416 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1417 dsl_dataset_t *ds;
1418 char *name, *atp;
1419 char dsname[MAXNAMELEN];
1420
1421 name = nvpair_name(pair);
1422 atp = strchr(name, '@');
1423 (void) strlcpy(dsname, name, atp - name + 1);
1424 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1425
1426 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1427 if (ddsa->ddsa_props != NULL) {
1428 dsl_props_set_sync_impl(ds->ds_prev,
1429 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1430 }
1431 dsl_dataset_rele(ds, FTAG);
1432 }
1433 }
1434
1435 /*
1436 * The snapshots must all be in the same pool.
1437 * All-or-nothing: if there are any failures, nothing will be modified.
1438 */
1439 int
1440 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1441 {
1442 dsl_dataset_snapshot_arg_t ddsa;
1443 nvpair_t *pair;
1444 boolean_t needsuspend;
1445 int error;
1446 spa_t *spa;
1447 char *firstname;
1448 nvlist_t *suspended = NULL;
1449
1450 pair = nvlist_next_nvpair(snaps, NULL);
1451 if (pair == NULL)
1452 return (0);
1453 firstname = nvpair_name(pair);
1454
1455 error = spa_open(firstname, &spa, FTAG);
1456 if (error != 0)
1457 return (error);
1458 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1459 spa_close(spa, FTAG);
1460
1461 if (needsuspend) {
1462 suspended = fnvlist_alloc();
1463 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1464 pair = nvlist_next_nvpair(snaps, pair)) {
1465 char fsname[MAXNAMELEN];
1466 char *snapname = nvpair_name(pair);
1467 char *atp;
1468 void *cookie;
1469
1470 atp = strchr(snapname, '@');
1471 if (atp == NULL) {
1472 error = SET_ERROR(EINVAL);
1473 break;
1474 }
1475 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1476
1477 error = zil_suspend(fsname, &cookie);
1478 if (error != 0)
1479 break;
1480 fnvlist_add_uint64(suspended, fsname,
1481 (uintptr_t)cookie);
1482 }
1483 }
1484
1485 ddsa.ddsa_snaps = snaps;
1486 ddsa.ddsa_props = props;
1487 ddsa.ddsa_errors = errors;
1488 ddsa.ddsa_cr = CRED();
1489
1490 if (error == 0) {
1491 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1492 dsl_dataset_snapshot_sync, &ddsa,
1493 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1494 }
1495
1496 if (suspended != NULL) {
1497 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1498 pair = nvlist_next_nvpair(suspended, pair)) {
1499 zil_resume((void *)(uintptr_t)
1500 fnvpair_value_uint64(pair));
1501 }
1502 fnvlist_free(suspended);
1503 }
1504
1505 return (error);
1506 }
1507
1508 typedef struct dsl_dataset_snapshot_tmp_arg {
1509 const char *ddsta_fsname;
1510 const char *ddsta_snapname;
1511 minor_t ddsta_cleanup_minor;
1512 const char *ddsta_htag;
1513 } dsl_dataset_snapshot_tmp_arg_t;
1514
1515 static int
1516 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1517 {
1518 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1519 dsl_pool_t *dp = dmu_tx_pool(tx);
1520 dsl_dataset_t *ds;
1521 int error;
1522
1523 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1524 if (error != 0)
1525 return (error);
1526
1527 /* NULL cred means no limit check for tmp snapshot */
1528 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1529 tx, B_FALSE, 0, NULL);
1530 if (error != 0) {
1531 dsl_dataset_rele(ds, FTAG);
1532 return (error);
1533 }
1534
1535 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1536 dsl_dataset_rele(ds, FTAG);
1537 return (SET_ERROR(ENOTSUP));
1538 }
1539 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1540 B_TRUE, tx);
1541 if (error != 0) {
1542 dsl_dataset_rele(ds, FTAG);
1543 return (error);
1544 }
1545
1546 dsl_dataset_rele(ds, FTAG);
1547 return (0);
1548 }
1549
1550 static void
1551 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1552 {
1553 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1554 dsl_pool_t *dp = dmu_tx_pool(tx);
1555 dsl_dataset_t *ds;
1556
1557 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1558
1559 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1560 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1561 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1562 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1563
1564 dsl_dataset_rele(ds, FTAG);
1565 }
1566
1567 int
1568 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1569 minor_t cleanup_minor, const char *htag)
1570 {
1571 dsl_dataset_snapshot_tmp_arg_t ddsta;
1572 int error;
1573 spa_t *spa;
1574 boolean_t needsuspend;
1575 void *cookie;
1576
1577 ddsta.ddsta_fsname = fsname;
1578 ddsta.ddsta_snapname = snapname;
1579 ddsta.ddsta_cleanup_minor = cleanup_minor;
1580 ddsta.ddsta_htag = htag;
1581
1582 error = spa_open(fsname, &spa, FTAG);
1583 if (error != 0)
1584 return (error);
1585 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1586 spa_close(spa, FTAG);
1587
1588 if (needsuspend) {
1589 error = zil_suspend(fsname, &cookie);
1590 if (error != 0)
1591 return (error);
1592 }
1593
1594 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1595 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1596
1597 if (needsuspend)
1598 zil_resume(cookie);
1599 return (error);
1600 }
1601
1602
1603 void
1604 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1605 {
1606 ASSERT(dmu_tx_is_syncing(tx));
1607 ASSERT(ds->ds_objset != NULL);
1608 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1609
1610 /*
1611 * in case we had to change ds_fsid_guid when we opened it,
1612 * sync it out now.
1613 */
1614 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1615 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1616
1617 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1618 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1619 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1620 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1621 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1622 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1623 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1624 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1625 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1626 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1627 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1628 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1629 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1630 }
1631
1632 dmu_objset_sync(ds->ds_objset, zio, tx);
1633
1634 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1635 if (ds->ds_feature_activation_needed[f]) {
1636 if (ds->ds_feature_inuse[f])
1637 continue;
1638 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1639 ds->ds_feature_inuse[f] = B_TRUE;
1640 }
1641 }
1642 }
1643
1644 static void
1645 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1646 {
1647 uint64_t count = 0;
1648 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1649 zap_cursor_t zc;
1650 zap_attribute_t za;
1651 nvlist_t *propval = fnvlist_alloc();
1652 nvlist_t *val = fnvlist_alloc();
1653
1654 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1655
1656 /*
1657 * There may be missing entries in ds_next_clones_obj
1658 * due to a bug in a previous version of the code.
1659 * Only trust it if it has the right number of entries.
1660 */
1661 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1662 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1663 &count));
1664 }
1665 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1666 goto fail;
1667 for (zap_cursor_init(&zc, mos,
1668 dsl_dataset_phys(ds)->ds_next_clones_obj);
1669 zap_cursor_retrieve(&zc, &za) == 0;
1670 zap_cursor_advance(&zc)) {
1671 dsl_dataset_t *clone;
1672 char buf[ZFS_MAXNAMELEN];
1673 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1674 za.za_first_integer, FTAG, &clone));
1675 dsl_dir_name(clone->ds_dir, buf);
1676 fnvlist_add_boolean(val, buf);
1677 dsl_dataset_rele(clone, FTAG);
1678 }
1679 zap_cursor_fini(&zc);
1680 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1681 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1682 fail:
1683 nvlist_free(val);
1684 nvlist_free(propval);
1685 }
1686
1687 static void
1688 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1689 {
1690 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1691
1692 if (dsl_dataset_has_resume_receive_state(ds)) {
1693 char *str;
1694 void *packed;
1695 uint8_t *compressed;
1696 uint64_t val;
1697 nvlist_t *token_nv = fnvlist_alloc();
1698 size_t packed_size, compressed_size;
1699
1700 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1701 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1702 fnvlist_add_uint64(token_nv, "fromguid", val);
1703 }
1704 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1705 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1706 fnvlist_add_uint64(token_nv, "object", val);
1707 }
1708 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1709 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1710 fnvlist_add_uint64(token_nv, "offset", val);
1711 }
1712 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1713 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1714 fnvlist_add_uint64(token_nv, "bytes", val);
1715 }
1716 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1717 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1718 fnvlist_add_uint64(token_nv, "toguid", val);
1719 }
1720 char buf[256];
1721 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1722 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1723 fnvlist_add_string(token_nv, "toname", buf);
1724 }
1725 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1726 DS_FIELD_RESUME_EMBEDOK) == 0) {
1727 fnvlist_add_boolean(token_nv, "embedok");
1728 }
1729 packed = fnvlist_pack(token_nv, &packed_size);
1730 fnvlist_free(token_nv);
1731 compressed = kmem_alloc(packed_size, KM_SLEEP);
1732
1733 compressed_size = gzip_compress(packed, compressed,
1734 packed_size, packed_size, 6);
1735
1736 zio_cksum_t cksum;
1737 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1738
1739 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1740 for (int i = 0; i < compressed_size; i++) {
1741 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1742 }
1743 str[compressed_size * 2] = '\0';
1744 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1745 ZFS_SEND_RESUME_TOKEN_VERSION,
1746 (longlong_t)cksum.zc_word[0],
1747 (longlong_t)packed_size, str);
1748 dsl_prop_nvlist_add_string(nv,
1749 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1750 kmem_free(packed, packed_size);
1751 kmem_free(str, compressed_size * 2 + 1);
1752 kmem_free(compressed, packed_size);
1753 strfree(propval);
1754 }
1755 }
1756
1757 void
1758 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1759 {
1760 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1761 uint64_t refd, avail, uobjs, aobjs, ratio;
1762
1763 ASSERT(dsl_pool_config_held(dp));
1764
1765 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1766 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1767 dsl_dataset_phys(ds)->ds_compressed_bytes);
1768
1769 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1770 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1771 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1772
1773 if (ds->ds_is_snapshot) {
1774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1775 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1776 dsl_dataset_phys(ds)->ds_unique_bytes);
1777 get_clones_stat(ds, nv);
1778 } else {
1779 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1780 char buf[MAXNAMELEN];
1781 dsl_dataset_name(ds->ds_prev, buf);
1782 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1783 }
1784
1785 dsl_dir_stats(ds->ds_dir, nv);
1786 }
1787
1788 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1789 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1790 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1791
1792 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1793 dsl_dataset_phys(ds)->ds_creation_time);
1794 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1795 dsl_dataset_phys(ds)->ds_creation_txg);
1796 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1797 ds->ds_quota);
1798 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1799 ds->ds_reserved);
1800 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1801 dsl_dataset_phys(ds)->ds_guid);
1802 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1803 dsl_dataset_phys(ds)->ds_unique_bytes);
1804 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1805 ds->ds_object);
1806 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1807 ds->ds_userrefs);
1808 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1809 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1810
1811 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1812 uint64_t written, comp, uncomp;
1813 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1814 dsl_dataset_t *prev;
1815
1816 int err = dsl_dataset_hold_obj(dp,
1817 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1818 if (err == 0) {
1819 err = dsl_dataset_space_written(prev, ds, &written,
1820 &comp, &uncomp);
1821 dsl_dataset_rele(prev, FTAG);
1822 if (err == 0) {
1823 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1824 written);
1825 }
1826 }
1827 }
1828
1829 if (!dsl_dataset_is_snapshot(ds)) {
1830 /*
1831 * A failed "newfs" (e.g. full) resumable receive leaves
1832 * the stats set on this dataset. Check here for the prop.
1833 */
1834 get_receive_resume_stats(ds, nv);
1835
1836 /*
1837 * A failed incremental resumable receive leaves the
1838 * stats set on our child named "%recv". Check the child
1839 * for the prop.
1840 */
1841 char recvname[ZFS_MAXNAMELEN];
1842 dsl_dataset_t *recv_ds;
1843 dsl_dataset_name(ds, recvname);
1844 (void) strcat(recvname, "/");
1845 (void) strcat(recvname, recv_clone_name);
1846 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1847 get_receive_resume_stats(recv_ds, nv);
1848 dsl_dataset_rele(recv_ds, FTAG);
1849 }
1850 }
1851 }
1852
1853 void
1854 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1855 {
1856 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1857 ASSERT(dsl_pool_config_held(dp));
1858
1859 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1860 stat->dds_inconsistent =
1861 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1862 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1863 stat->dds_origin[0] = '\0';
1864 if (ds->ds_is_snapshot) {
1865 stat->dds_is_snapshot = B_TRUE;
1866 stat->dds_num_clones =
1867 dsl_dataset_phys(ds)->ds_num_children - 1;
1868 } else {
1869 stat->dds_is_snapshot = B_FALSE;
1870 stat->dds_num_clones = 0;
1871
1872 if (dsl_dir_is_clone(ds->ds_dir)) {
1873 dsl_dataset_t *ods;
1874
1875 VERIFY0(dsl_dataset_hold_obj(dp,
1876 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1877 FTAG, &ods));
1878 dsl_dataset_name(ods, stat->dds_origin);
1879 dsl_dataset_rele(ods, FTAG);
1880 }
1881 }
1882 }
1883
1884 uint64_t
1885 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1886 {
1887 return (ds->ds_fsid_guid);
1888 }
1889
1890 void
1891 dsl_dataset_space(dsl_dataset_t *ds,
1892 uint64_t *refdbytesp, uint64_t *availbytesp,
1893 uint64_t *usedobjsp, uint64_t *availobjsp)
1894 {
1895 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1896 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1897 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1898 *availbytesp +=
1899 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1900 if (ds->ds_quota != 0) {
1901 /*
1902 * Adjust available bytes according to refquota
1903 */
1904 if (*refdbytesp < ds->ds_quota)
1905 *availbytesp = MIN(*availbytesp,
1906 ds->ds_quota - *refdbytesp);
1907 else
1908 *availbytesp = 0;
1909 }
1910 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1911 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1912 }
1913
1914 boolean_t
1915 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1916 {
1917 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1918
1919 ASSERT(dsl_pool_config_held(dp));
1920 if (snap == NULL)
1921 return (B_FALSE);
1922 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1923 dsl_dataset_phys(snap)->ds_creation_txg) {
1924 objset_t *os, *os_snap;
1925 /*
1926 * It may be that only the ZIL differs, because it was
1927 * reset in the head. Don't count that as being
1928 * modified.
1929 */
1930 if (dmu_objset_from_ds(ds, &os) != 0)
1931 return (B_TRUE);
1932 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1933 return (B_TRUE);
1934 return (bcmp(&os->os_phys->os_meta_dnode,
1935 &os_snap->os_phys->os_meta_dnode,
1936 sizeof (os->os_phys->os_meta_dnode)) != 0);
1937 }
1938 return (B_FALSE);
1939 }
1940
1941 typedef struct dsl_dataset_rename_snapshot_arg {
1942 const char *ddrsa_fsname;
1943 const char *ddrsa_oldsnapname;
1944 const char *ddrsa_newsnapname;
1945 boolean_t ddrsa_recursive;
1946 dmu_tx_t *ddrsa_tx;
1947 } dsl_dataset_rename_snapshot_arg_t;
1948
1949 /* ARGSUSED */
1950 static int
1951 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1952 dsl_dataset_t *hds, void *arg)
1953 {
1954 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1955 int error;
1956 uint64_t val;
1957
1958 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1959 if (error != 0) {
1960 /* ignore nonexistent snapshots */
1961 return (error == ENOENT ? 0 : error);
1962 }
1963
1964 /* new name should not exist */
1965 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1966 if (error == 0)
1967 error = SET_ERROR(EEXIST);
1968 else if (error == ENOENT)
1969 error = 0;
1970
1971 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1972 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1973 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1974 error = SET_ERROR(ENAMETOOLONG);
1975
1976 return (error);
1977 }
1978
1979 static int
1980 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1981 {
1982 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1983 dsl_pool_t *dp = dmu_tx_pool(tx);
1984 dsl_dataset_t *hds;
1985 int error;
1986
1987 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1988 if (error != 0)
1989 return (error);
1990
1991 if (ddrsa->ddrsa_recursive) {
1992 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1993 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1994 DS_FIND_CHILDREN);
1995 } else {
1996 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1997 }
1998 dsl_dataset_rele(hds, FTAG);
1999 return (error);
2000 }
2001
2002 static int
2003 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2004 dsl_dataset_t *hds, void *arg)
2005 {
2006 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2007 dsl_dataset_t *ds;
2008 uint64_t val;
2009 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2010 int error;
2011
2012 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2013 ASSERT(error == 0 || error == ENOENT);
2014 if (error == ENOENT) {
2015 /* ignore nonexistent snapshots */
2016 return (0);
2017 }
2018
2019 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2020
2021 /* log before we change the name */
2022 spa_history_log_internal_ds(ds, "rename", tx,
2023 "-> @%s", ddrsa->ddrsa_newsnapname);
2024
2025 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2026 B_FALSE));
2027 mutex_enter(&ds->ds_lock);
2028 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2029 mutex_exit(&ds->ds_lock);
2030 VERIFY0(zap_add(dp->dp_meta_objset,
2031 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2032 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2033
2034 dsl_dataset_rele(ds, FTAG);
2035 return (0);
2036 }
2037
2038 static void
2039 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2040 {
2041 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2042 dsl_pool_t *dp = dmu_tx_pool(tx);
2043 dsl_dataset_t *hds;
2044
2045 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2046 ddrsa->ddrsa_tx = tx;
2047 if (ddrsa->ddrsa_recursive) {
2048 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2049 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2050 DS_FIND_CHILDREN));
2051 } else {
2052 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2053 }
2054 dsl_dataset_rele(hds, FTAG);
2055 }
2056
2057 int
2058 dsl_dataset_rename_snapshot(const char *fsname,
2059 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2060 {
2061 dsl_dataset_rename_snapshot_arg_t ddrsa;
2062
2063 ddrsa.ddrsa_fsname = fsname;
2064 ddrsa.ddrsa_oldsnapname = oldsnapname;
2065 ddrsa.ddrsa_newsnapname = newsnapname;
2066 ddrsa.ddrsa_recursive = recursive;
2067
2068 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2069 dsl_dataset_rename_snapshot_sync, &ddrsa,
2070 1, ZFS_SPACE_CHECK_RESERVED));
2071 }
2072
2073 /*
2074 * If we're doing an ownership handoff, we need to make sure that there is
2075 * only one long hold on the dataset. We're not allowed to change anything here
2076 * so we don't permanently release the long hold or regular hold here. We want
2077 * to do this only when syncing to avoid the dataset unexpectedly going away
2078 * when we release the long hold.
2079 */
2080 static int
2081 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2082 {
2083 boolean_t held;
2084
2085 if (!dmu_tx_is_syncing(tx))
2086 return (0);
2087
2088 if (owner != NULL) {
2089 VERIFY3P(ds->ds_owner, ==, owner);
2090 dsl_dataset_long_rele(ds, owner);
2091 }
2092
2093 held = dsl_dataset_long_held(ds);
2094
2095 if (owner != NULL)
2096 dsl_dataset_long_hold(ds, owner);
2097
2098 if (held)
2099 return (SET_ERROR(EBUSY));
2100
2101 return (0);
2102 }
2103
2104 typedef struct dsl_dataset_rollback_arg {
2105 const char *ddra_fsname;
2106 void *ddra_owner;
2107 nvlist_t *ddra_result;
2108 } dsl_dataset_rollback_arg_t;
2109
2110 static int
2111 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2112 {
2113 dsl_dataset_rollback_arg_t *ddra = arg;
2114 dsl_pool_t *dp = dmu_tx_pool(tx);
2115 dsl_dataset_t *ds;
2116 int64_t unused_refres_delta;
2117 int error;
2118
2119 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2120 if (error != 0)
2121 return (error);
2122
2123 /* must not be a snapshot */
2124 if (ds->ds_is_snapshot) {
2125 dsl_dataset_rele(ds, FTAG);
2126 return (SET_ERROR(EINVAL));
2127 }
2128
2129 /* must have a most recent snapshot */
2130 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2131 dsl_dataset_rele(ds, FTAG);
2132 return (SET_ERROR(EINVAL));
2133 }
2134
2135 /* must not have any bookmarks after the most recent snapshot */
2136 nvlist_t *proprequest = fnvlist_alloc();
2137 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2138 nvlist_t *bookmarks = fnvlist_alloc();
2139 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2140 fnvlist_free(proprequest);
2141 if (error != 0)
2142 return (error);
2143 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2144 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2145 nvlist_t *valuenv =
2146 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2147 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2148 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2149 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2150 fnvlist_free(bookmarks);
2151 dsl_dataset_rele(ds, FTAG);
2152 return (SET_ERROR(EEXIST));
2153 }
2154 }
2155 fnvlist_free(bookmarks);
2156
2157 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2158 if (error != 0) {
2159 dsl_dataset_rele(ds, FTAG);
2160 return (error);
2161 }
2162
2163 /*
2164 * Check if the snap we are rolling back to uses more than
2165 * the refquota.
2166 */
2167 if (ds->ds_quota != 0 &&
2168 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2169 dsl_dataset_rele(ds, FTAG);
2170 return (SET_ERROR(EDQUOT));
2171 }
2172
2173 /*
2174 * When we do the clone swap, we will temporarily use more space
2175 * due to the refreservation (the head will no longer have any
2176 * unique space, so the entire amount of the refreservation will need
2177 * to be free). We will immediately destroy the clone, freeing
2178 * this space, but the freeing happens over many txg's.
2179 */
2180 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2181 dsl_dataset_phys(ds)->ds_unique_bytes);
2182
2183 if (unused_refres_delta > 0 &&
2184 unused_refres_delta >
2185 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2186 dsl_dataset_rele(ds, FTAG);
2187 return (SET_ERROR(ENOSPC));
2188 }
2189
2190 dsl_dataset_rele(ds, FTAG);
2191 return (0);
2192 }
2193
2194 static void
2195 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2196 {
2197 dsl_dataset_rollback_arg_t *ddra = arg;
2198 dsl_pool_t *dp = dmu_tx_pool(tx);
2199 dsl_dataset_t *ds, *clone;
2200 uint64_t cloneobj;
2201 char namebuf[ZFS_MAXNAMELEN];
2202
2203 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2204
2205 dsl_dataset_name(ds->ds_prev, namebuf);
2206 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2207
2208 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2209 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2210
2211 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2212
2213 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2214 dsl_dataset_zero_zil(ds, tx);
2215
2216 dsl_destroy_head_sync_impl(clone, tx);
2217
2218 dsl_dataset_rele(clone, FTAG);
2219 dsl_dataset_rele(ds, FTAG);
2220 }
2221
2222 /*
2223 * Rolls back the given filesystem or volume to the most recent snapshot.
2224 * The name of the most recent snapshot will be returned under key "target"
2225 * in the result nvlist.
2226 *
2227 * If owner != NULL:
2228 * - The existing dataset MUST be owned by the specified owner at entry
2229 * - Upon return, dataset will still be held by the same owner, whether we
2230 * succeed or not.
2231 *
2232 * This mode is required any time the existing filesystem is mounted. See
2233 * notes above zfs_suspend_fs() for further details.
2234 */
2235 int
2236 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2237 {
2238 dsl_dataset_rollback_arg_t ddra;
2239
2240 ddra.ddra_fsname = fsname;
2241 ddra.ddra_owner = owner;
2242 ddra.ddra_result = result;
2243
2244 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2245 dsl_dataset_rollback_sync, &ddra,
2246 1, ZFS_SPACE_CHECK_RESERVED));
2247 }
2248
2249 struct promotenode {
2250 list_node_t link;
2251 dsl_dataset_t *ds;
2252 };
2253
2254 typedef struct dsl_dataset_promote_arg {
2255 const char *ddpa_clonename;
2256 dsl_dataset_t *ddpa_clone;
2257 list_t shared_snaps, origin_snaps, clone_snaps;
2258 dsl_dataset_t *origin_origin; /* origin of the origin */
2259 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2260 char *err_ds;
2261 cred_t *cr;
2262 } dsl_dataset_promote_arg_t;
2263
2264 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2265 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2266 void *tag);
2267 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2268
2269 static int
2270 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2271 {
2272 dsl_dataset_promote_arg_t *ddpa = arg;
2273 dsl_pool_t *dp = dmu_tx_pool(tx);
2274 dsl_dataset_t *hds;
2275 struct promotenode *snap;
2276 dsl_dataset_t *origin_ds;
2277 int err;
2278 uint64_t unused;
2279 uint64_t ss_mv_cnt;
2280 size_t max_snap_len;
2281
2282 err = promote_hold(ddpa, dp, FTAG);
2283 if (err != 0)
2284 return (err);
2285
2286 hds = ddpa->ddpa_clone;
2287 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2288
2289 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2290 promote_rele(ddpa, FTAG);
2291 return (SET_ERROR(EXDEV));
2292 }
2293
2294 /*
2295 * Compute and check the amount of space to transfer. Since this is
2296 * so expensive, don't do the preliminary check.
2297 */
2298 if (!dmu_tx_is_syncing(tx)) {
2299 promote_rele(ddpa, FTAG);
2300 return (0);
2301 }
2302
2303 snap = list_head(&ddpa->shared_snaps);
2304 origin_ds = snap->ds;
2305
2306 /* compute origin's new unique space */
2307 snap = list_tail(&ddpa->clone_snaps);
2308 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2309 origin_ds->ds_object);
2310 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2311 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2312 &ddpa->unique, &unused, &unused);
2313
2314 /*
2315 * Walk the snapshots that we are moving
2316 *
2317 * Compute space to transfer. Consider the incremental changes
2318 * to used by each snapshot:
2319 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2320 * So each snapshot gave birth to:
2321 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2322 * So a sequence would look like:
2323 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2324 * Which simplifies to:
2325 * uN + kN + kN-1 + ... + k1 + k0
2326 * Note however, if we stop before we reach the ORIGIN we get:
2327 * uN + kN + kN-1 + ... + kM - uM-1
2328 */
2329 ss_mv_cnt = 0;
2330 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2331 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2332 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2333 for (snap = list_head(&ddpa->shared_snaps); snap;
2334 snap = list_next(&ddpa->shared_snaps, snap)) {
2335 uint64_t val, dlused, dlcomp, dluncomp;
2336 dsl_dataset_t *ds = snap->ds;
2337
2338 ss_mv_cnt++;
2339
2340 /*
2341 * If there are long holds, we won't be able to evict
2342 * the objset.
2343 */
2344 if (dsl_dataset_long_held(ds)) {
2345 err = SET_ERROR(EBUSY);
2346 goto out;
2347 }
2348
2349 /* Check that the snapshot name does not conflict */
2350 VERIFY0(dsl_dataset_get_snapname(ds));
2351 if (strlen(ds->ds_snapname) >= max_snap_len) {
2352 err = SET_ERROR(ENAMETOOLONG);
2353 goto out;
2354 }
2355 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2356 if (err == 0) {
2357 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2358 err = SET_ERROR(EEXIST);
2359 goto out;
2360 }
2361 if (err != ENOENT)
2362 goto out;
2363
2364 /* The very first snapshot does not have a deadlist */
2365 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2366 continue;
2367
2368 dsl_deadlist_space(&ds->ds_deadlist,
2369 &dlused, &dlcomp, &dluncomp);
2370 ddpa->used += dlused;
2371 ddpa->comp += dlcomp;
2372 ddpa->uncomp += dluncomp;
2373 }
2374
2375 /*
2376 * If we are a clone of a clone then we never reached ORIGIN,
2377 * so we need to subtract out the clone origin's used space.
2378 */
2379 if (ddpa->origin_origin) {
2380 ddpa->used -=
2381 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2382 ddpa->comp -=
2383 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2384 ddpa->uncomp -=
2385 dsl_dataset_phys(ddpa->origin_origin)->
2386 ds_uncompressed_bytes;
2387 }
2388
2389 /* Check that there is enough space and limit headroom here */
2390 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2391 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2392 if (err != 0)
2393 goto out;
2394
2395 /*
2396 * Compute the amounts of space that will be used by snapshots
2397 * after the promotion (for both origin and clone). For each,
2398 * it is the amount of space that will be on all of their
2399 * deadlists (that was not born before their new origin).
2400 */
2401 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2402 uint64_t space;
2403
2404 /*
2405 * Note, typically this will not be a clone of a clone,
2406 * so dd_origin_txg will be < TXG_INITIAL, so
2407 * these snaplist_space() -> dsl_deadlist_space_range()
2408 * calls will be fast because they do not have to
2409 * iterate over all bps.
2410 */
2411 snap = list_head(&ddpa->origin_snaps);
2412 err = snaplist_space(&ddpa->shared_snaps,
2413 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2414 if (err != 0)
2415 goto out;
2416
2417 err = snaplist_space(&ddpa->clone_snaps,
2418 snap->ds->ds_dir->dd_origin_txg, &space);
2419 if (err != 0)
2420 goto out;
2421 ddpa->cloneusedsnap += space;
2422 }
2423 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2424 DD_FLAG_USED_BREAKDOWN) {
2425 err = snaplist_space(&ddpa->origin_snaps,
2426 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2427 &ddpa->originusedsnap);
2428 if (err != 0)
2429 goto out;
2430 }
2431
2432 out:
2433 promote_rele(ddpa, FTAG);
2434 return (err);
2435 }
2436
2437 static void
2438 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2439 {
2440 dsl_dataset_promote_arg_t *ddpa = arg;
2441 dsl_pool_t *dp = dmu_tx_pool(tx);
2442 dsl_dataset_t *hds;
2443 struct promotenode *snap;
2444 dsl_dataset_t *origin_ds;
2445 dsl_dataset_t *origin_head;
2446 dsl_dir_t *dd;
2447 dsl_dir_t *odd = NULL;
2448 uint64_t oldnext_obj;
2449 int64_t delta;
2450
2451 VERIFY0(promote_hold(ddpa, dp, FTAG));
2452 hds = ddpa->ddpa_clone;
2453
2454 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2455
2456 snap = list_head(&ddpa->shared_snaps);
2457 origin_ds = snap->ds;
2458 dd = hds->ds_dir;
2459
2460 snap = list_head(&ddpa->origin_snaps);
2461 origin_head = snap->ds;
2462
2463 /*
2464 * We need to explicitly open odd, since origin_ds's dd will be
2465 * changing.
2466 */
2467 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2468 NULL, FTAG, &odd));
2469
2470 /* change origin's next snap */
2471 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2472 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2473 snap = list_tail(&ddpa->clone_snaps);
2474 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2475 origin_ds->ds_object);
2476 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2477
2478 /* change the origin's next clone */
2479 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2480 dsl_dataset_remove_from_next_clones(origin_ds,
2481 snap->ds->ds_object, tx);
2482 VERIFY0(zap_add_int(dp->dp_meta_objset,
2483 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2484 oldnext_obj, tx));
2485 }
2486
2487 /* change origin */
2488 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2489 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2490 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2491 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2492 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2493 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2494 origin_head->ds_dir->dd_origin_txg =
2495 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2496
2497 /* change dd_clone entries */
2498 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2499 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2500 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2501 VERIFY0(zap_add_int(dp->dp_meta_objset,
2502 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2503 hds->ds_object, tx));
2504
2505 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2506 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2507 origin_head->ds_object, tx));
2508 if (dsl_dir_phys(dd)->dd_clones == 0) {
2509 dsl_dir_phys(dd)->dd_clones =
2510 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2511 DMU_OT_NONE, 0, tx);
2512 }
2513 VERIFY0(zap_add_int(dp->dp_meta_objset,
2514 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2515 }
2516
2517 /* move snapshots to this dir */
2518 for (snap = list_head(&ddpa->shared_snaps); snap;
2519 snap = list_next(&ddpa->shared_snaps, snap)) {
2520 dsl_dataset_t *ds = snap->ds;
2521
2522 /*
2523 * Property callbacks are registered to a particular
2524 * dsl_dir. Since ours is changing, evict the objset
2525 * so that they will be unregistered from the old dsl_dir.
2526 */
2527 if (ds->ds_objset) {
2528 dmu_objset_evict(ds->ds_objset);
2529 ds->ds_objset = NULL;
2530 }
2531
2532 /* move snap name entry */
2533 VERIFY0(dsl_dataset_get_snapname(ds));
2534 VERIFY0(dsl_dataset_snap_remove(origin_head,
2535 ds->ds_snapname, tx, B_TRUE));
2536 VERIFY0(zap_add(dp->dp_meta_objset,
2537 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2538 8, 1, &ds->ds_object, tx));
2539 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2540 DD_FIELD_SNAPSHOT_COUNT, tx);
2541
2542 /* change containing dsl_dir */
2543 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2544 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2545 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2546 ASSERT3P(ds->ds_dir, ==, odd);
2547 dsl_dir_rele(ds->ds_dir, ds);
2548 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2549 NULL, ds, &ds->ds_dir));
2550
2551 /* move any clone references */
2552 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2553 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2554 zap_cursor_t zc;
2555 zap_attribute_t za;
2556
2557 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2558 dsl_dataset_phys(ds)->ds_next_clones_obj);
2559 zap_cursor_retrieve(&zc, &za) == 0;
2560 zap_cursor_advance(&zc)) {
2561 dsl_dataset_t *cnds;
2562 uint64_t o;
2563
2564 if (za.za_first_integer == oldnext_obj) {
2565 /*
2566 * We've already moved the
2567 * origin's reference.
2568 */
2569 continue;
2570 }
2571
2572 VERIFY0(dsl_dataset_hold_obj(dp,
2573 za.za_first_integer, FTAG, &cnds));
2574 o = dsl_dir_phys(cnds->ds_dir)->
2575 dd_head_dataset_obj;
2576
2577 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2578 dsl_dir_phys(odd)->dd_clones, o, tx));
2579 VERIFY0(zap_add_int(dp->dp_meta_objset,
2580 dsl_dir_phys(dd)->dd_clones, o, tx));
2581 dsl_dataset_rele(cnds, FTAG);
2582 }
2583 zap_cursor_fini(&zc);
2584 }
2585
2586 ASSERT(!dsl_prop_hascb(ds));
2587 }
2588
2589 /*
2590 * Change space accounting.
2591 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2592 * both be valid, or both be 0 (resulting in delta == 0). This
2593 * is true for each of {clone,origin} independently.
2594 */
2595
2596 delta = ddpa->cloneusedsnap -
2597 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2598 ASSERT3S(delta, >=, 0);
2599 ASSERT3U(ddpa->used, >=, delta);
2600 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2601 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2602 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2603
2604 delta = ddpa->originusedsnap -
2605 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2606 ASSERT3S(delta, <=, 0);
2607 ASSERT3U(ddpa->used, >=, -delta);
2608 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2609 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2610 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2611
2612 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2613
2614 /* log history record */
2615 spa_history_log_internal_ds(hds, "promote", tx, "");
2616
2617 dsl_dir_rele(odd, FTAG);
2618 promote_rele(ddpa, FTAG);
2619 }
2620
2621 /*
2622 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2623 * (exclusive) and last_obj (inclusive). The list will be in reverse
2624 * order (last_obj will be the list_head()). If first_obj == 0, do all
2625 * snapshots back to this dataset's origin.
2626 */
2627 static int
2628 snaplist_make(dsl_pool_t *dp,
2629 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2630 {
2631 uint64_t obj = last_obj;
2632
2633 list_create(l, sizeof (struct promotenode),
2634 offsetof(struct promotenode, link));
2635
2636 while (obj != first_obj) {
2637 dsl_dataset_t *ds;
2638 struct promotenode *snap;
2639 int err;
2640
2641 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2642 ASSERT(err != ENOENT);
2643 if (err != 0)
2644 return (err);
2645
2646 if (first_obj == 0)
2647 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2648
2649 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2650 snap->ds = ds;
2651 list_insert_tail(l, snap);
2652 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2653 }
2654
2655 return (0);
2656 }
2657
2658 static int
2659 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2660 {
2661 struct promotenode *snap;
2662
2663 *spacep = 0;
2664 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2665 uint64_t used, comp, uncomp;
2666 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2667 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2668 *spacep += used;
2669 }
2670 return (0);
2671 }
2672
2673 static void
2674 snaplist_destroy(list_t *l, void *tag)
2675 {
2676 struct promotenode *snap;
2677
2678 if (l == NULL || !list_link_active(&l->list_head))
2679 return;
2680
2681 while ((snap = list_tail(l)) != NULL) {
2682 list_remove(l, snap);
2683 dsl_dataset_rele(snap->ds, tag);
2684 kmem_free(snap, sizeof (*snap));
2685 }
2686 list_destroy(l);
2687 }
2688
2689 static int
2690 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2691 {
2692 int error;
2693 dsl_dir_t *dd;
2694 struct promotenode *snap;
2695
2696 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2697 &ddpa->ddpa_clone);
2698 if (error != 0)
2699 return (error);
2700 dd = ddpa->ddpa_clone->ds_dir;
2701
2702 if (ddpa->ddpa_clone->ds_is_snapshot ||
2703 !dsl_dir_is_clone(dd)) {
2704 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2705 return (SET_ERROR(EINVAL));
2706 }
2707
2708 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2709 &ddpa->shared_snaps, tag);
2710 if (error != 0)
2711 goto out;
2712
2713 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2714 &ddpa->clone_snaps, tag);
2715 if (error != 0)
2716 goto out;
2717
2718 snap = list_head(&ddpa->shared_snaps);
2719 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2720 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2721 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2722 &ddpa->origin_snaps, tag);
2723 if (error != 0)
2724 goto out;
2725
2726 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2727 error = dsl_dataset_hold_obj(dp,
2728 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2729 tag, &ddpa->origin_origin);
2730 if (error != 0)
2731 goto out;
2732 }
2733 out:
2734 if (error != 0)
2735 promote_rele(ddpa, tag);
2736 return (error);
2737 }
2738
2739 static void
2740 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2741 {
2742 snaplist_destroy(&ddpa->shared_snaps, tag);
2743 snaplist_destroy(&ddpa->clone_snaps, tag);
2744 snaplist_destroy(&ddpa->origin_snaps, tag);
2745 if (ddpa->origin_origin != NULL)
2746 dsl_dataset_rele(ddpa->origin_origin, tag);
2747 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2748 }
2749
2750 /*
2751 * Promote a clone.
2752 *
2753 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2754 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2755 */
2756 int
2757 dsl_dataset_promote(const char *name, char *conflsnap)
2758 {
2759 dsl_dataset_promote_arg_t ddpa = { 0 };
2760 uint64_t numsnaps;
2761 int error;
2762 objset_t *os;
2763
2764 /*
2765 * We will modify space proportional to the number of
2766 * snapshots. Compute numsnaps.
2767 */
2768 error = dmu_objset_hold(name, FTAG, &os);
2769 if (error != 0)
2770 return (error);
2771 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2772 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2773 &numsnaps);
2774 dmu_objset_rele(os, FTAG);
2775 if (error != 0)
2776 return (error);
2777
2778 ddpa.ddpa_clonename = name;
2779 ddpa.err_ds = conflsnap;
2780 ddpa.cr = CRED();
2781
2782 return (dsl_sync_task(name, dsl_dataset_promote_check,
2783 dsl_dataset_promote_sync, &ddpa,
2784 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2785 }
2786
2787 int
2788 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2789 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2790 {
2791 int64_t unused_refres_delta;
2792
2793 /* they should both be heads */
2794 if (clone->ds_is_snapshot ||
2795 origin_head->ds_is_snapshot)
2796 return (SET_ERROR(EINVAL));
2797
2798 /* if we are not forcing, the branch point should be just before them */
2799 if (!force && clone->ds_prev != origin_head->ds_prev)
2800 return (SET_ERROR(EINVAL));
2801
2802 /* clone should be the clone (unless they are unrelated) */
2803 if (clone->ds_prev != NULL &&
2804 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2805 origin_head->ds_dir != clone->ds_prev->ds_dir)
2806 return (SET_ERROR(EINVAL));
2807
2808 /* the clone should be a child of the origin */
2809 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2810 return (SET_ERROR(EINVAL));
2811
2812 /* origin_head shouldn't be modified unless 'force' */
2813 if (!force &&
2814 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2815 return (SET_ERROR(ETXTBSY));
2816
2817 /* origin_head should have no long holds (e.g. is not mounted) */
2818 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2819 return (SET_ERROR(EBUSY));
2820
2821 /* check amount of any unconsumed refreservation */
2822 unused_refres_delta =
2823 (int64_t)MIN(origin_head->ds_reserved,
2824 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2825 (int64_t)MIN(origin_head->ds_reserved,
2826 dsl_dataset_phys(clone)->ds_unique_bytes);
2827
2828 if (unused_refres_delta > 0 &&
2829 unused_refres_delta >
2830 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2831 return (SET_ERROR(ENOSPC));
2832
2833 /* clone can't be over the head's refquota */
2834 if (origin_head->ds_quota != 0 &&
2835 dsl_dataset_phys(clone)->ds_referenced_bytes >
2836 origin_head->ds_quota)
2837 return (SET_ERROR(EDQUOT));
2838
2839 return (0);
2840 }
2841
2842 void
2843 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2844 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2845 {
2846 dsl_pool_t *dp = dmu_tx_pool(tx);
2847 int64_t unused_refres_delta;
2848
2849 ASSERT(clone->ds_reserved == 0);
2850 ASSERT(origin_head->ds_quota == 0 ||
2851 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2852 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2853
2854 /*
2855 * Swap per-dataset feature flags.
2856 */
2857 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2858 if (!(spa_feature_table[f].fi_flags &
2859 ZFEATURE_FLAG_PER_DATASET)) {
2860 ASSERT(!clone->ds_feature_inuse[f]);
2861 ASSERT(!origin_head->ds_feature_inuse[f]);
2862 continue;
2863 }
2864
2865 boolean_t clone_inuse = clone->ds_feature_inuse[f];
2866 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2867
2868 if (clone_inuse) {
2869 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2870 clone->ds_feature_inuse[f] = B_FALSE;
2871 }
2872 if (origin_head_inuse) {
2873 dsl_dataset_deactivate_feature(origin_head->ds_object,
2874 f, tx);
2875 origin_head->ds_feature_inuse[f] = B_FALSE;
2876 }
2877 if (clone_inuse) {
2878 dsl_dataset_activate_feature(origin_head->ds_object,
2879 f, tx);
2880 origin_head->ds_feature_inuse[f] = B_TRUE;
2881 }
2882 if (origin_head_inuse) {
2883 dsl_dataset_activate_feature(clone->ds_object, f, tx);
2884 clone->ds_feature_inuse[f] = B_TRUE;
2885 }
2886 }
2887
2888 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2889 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2890
2891 if (clone->ds_objset != NULL) {
2892 dmu_objset_evict(clone->ds_objset);
2893 clone->ds_objset = NULL;
2894 }
2895
2896 if (origin_head->ds_objset != NULL) {
2897 dmu_objset_evict(origin_head->ds_objset);
2898 origin_head->ds_objset = NULL;
2899 }
2900
2901 unused_refres_delta =
2902 (int64_t)MIN(origin_head->ds_reserved,
2903 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2904 (int64_t)MIN(origin_head->ds_reserved,
2905 dsl_dataset_phys(clone)->ds_unique_bytes);
2906
2907 /*
2908 * Reset origin's unique bytes, if it exists.
2909 */
2910 if (clone->ds_prev) {
2911 dsl_dataset_t *origin = clone->ds_prev;
2912 uint64_t comp, uncomp;
2913
2914 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2915 dsl_deadlist_space_range(&clone->ds_deadlist,
2916 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2917 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2918 }
2919
2920 /* swap blkptrs */
2921 {
2922 blkptr_t tmp;
2923 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2924 dsl_dataset_phys(origin_head)->ds_bp =
2925 dsl_dataset_phys(clone)->ds_bp;
2926 dsl_dataset_phys(clone)->ds_bp = tmp;
2927 }
2928
2929 /* set dd_*_bytes */
2930 {
2931 int64_t dused, dcomp, duncomp;
2932 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2933 uint64_t odl_used, odl_comp, odl_uncomp;
2934
2935 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2936 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2937
2938 dsl_deadlist_space(&clone->ds_deadlist,
2939 &cdl_used, &cdl_comp, &cdl_uncomp);
2940 dsl_deadlist_space(&origin_head->ds_deadlist,
2941 &odl_used, &odl_comp, &odl_uncomp);
2942
2943 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2944 cdl_used -
2945 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2946 odl_used);
2947 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2948 cdl_comp -
2949 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2950 odl_comp);
2951 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2952 cdl_uncomp -
2953 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2954 odl_uncomp);
2955
2956 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2957 dused, dcomp, duncomp, tx);
2958 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2959 -dused, -dcomp, -duncomp, tx);
2960
2961 /*
2962 * The difference in the space used by snapshots is the
2963 * difference in snapshot space due to the head's
2964 * deadlist (since that's the only thing that's
2965 * changing that affects the snapused).
2966 */
2967 dsl_deadlist_space_range(&clone->ds_deadlist,
2968 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2969 &cdl_used, &cdl_comp, &cdl_uncomp);
2970 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2971 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2972 &odl_used, &odl_comp, &odl_uncomp);
2973 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2974 DD_USED_HEAD, DD_USED_SNAP, tx);
2975 }
2976
2977 /* swap ds_*_bytes */
2978 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2979 dsl_dataset_phys(clone)->ds_referenced_bytes);
2980 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2981 dsl_dataset_phys(clone)->ds_compressed_bytes);
2982 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2983 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2984 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2985 dsl_dataset_phys(clone)->ds_unique_bytes);
2986
2987 /* apply any parent delta for change in unconsumed refreservation */
2988 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2989 unused_refres_delta, 0, 0, tx);
2990
2991 /*
2992 * Swap deadlists.
2993 */
2994 dsl_deadlist_close(&clone->ds_deadlist);
2995 dsl_deadlist_close(&origin_head->ds_deadlist);
2996 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2997 dsl_dataset_phys(clone)->ds_deadlist_obj);
2998 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2999 dsl_dataset_phys(clone)->ds_deadlist_obj);
3000 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3001 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3002
3003 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3004
3005 spa_history_log_internal_ds(clone, "clone swap", tx,
3006 "parent=%s", origin_head->ds_dir->dd_myname);
3007 }
3008
3009 /*
3010 * Given a pool name and a dataset object number in that pool,
3011 * return the name of that dataset.
3012 */
3013 int
3014 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3015 {
3016 dsl_pool_t *dp;
3017 dsl_dataset_t *ds;
3018 int error;
3019
3020 error = dsl_pool_hold(pname, FTAG, &dp);
3021 if (error != 0)
3022 return (error);
3023
3024 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3025 if (error == 0) {
3026 dsl_dataset_name(ds, buf);
3027 dsl_dataset_rele(ds, FTAG);
3028 }
3029 dsl_pool_rele(dp, FTAG);
3030
3031 return (error);
3032 }
3033
3034 int
3035 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3036 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3037 {
3038 int error = 0;
3039
3040 ASSERT3S(asize, >, 0);
3041
3042 /*
3043 * *ref_rsrv is the portion of asize that will come from any
3044 * unconsumed refreservation space.
3045 */
3046 *ref_rsrv = 0;
3047
3048 mutex_enter(&ds->ds_lock);
3049 /*
3050 * Make a space adjustment for reserved bytes.
3051 */
3052 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3053 ASSERT3U(*used, >=,
3054 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3055 *used -=
3056 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3057 *ref_rsrv =
3058 asize - MIN(asize, parent_delta(ds, asize + inflight));
3059 }
3060
3061 if (!check_quota || ds->ds_quota == 0) {
3062 mutex_exit(&ds->ds_lock);
3063 return (0);
3064 }
3065 /*
3066 * If they are requesting more space, and our current estimate
3067 * is over quota, they get to try again unless the actual
3068 * on-disk is over quota and there are no pending changes (which
3069 * may free up space for us).
3070 */
3071 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3072 ds->ds_quota) {
3073 if (inflight > 0 ||
3074 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3075 error = SET_ERROR(ERESTART);
3076 else
3077 error = SET_ERROR(EDQUOT);
3078 }
3079 mutex_exit(&ds->ds_lock);
3080
3081 return (error);
3082 }
3083
3084 typedef struct dsl_dataset_set_qr_arg {
3085 const char *ddsqra_name;
3086 zprop_source_t ddsqra_source;
3087 uint64_t ddsqra_value;
3088 } dsl_dataset_set_qr_arg_t;
3089
3090
3091 /* ARGSUSED */
3092 static int
3093 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3094 {
3095 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3096 dsl_pool_t *dp = dmu_tx_pool(tx);
3097 dsl_dataset_t *ds;
3098 int error;
3099 uint64_t newval;
3100
3101 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3102 return (SET_ERROR(ENOTSUP));
3103
3104 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3105 if (error != 0)
3106 return (error);
3107
3108 if (ds->ds_is_snapshot) {
3109 dsl_dataset_rele(ds, FTAG);
3110 return (SET_ERROR(EINVAL));
3111 }
3112
3113 error = dsl_prop_predict(ds->ds_dir,
3114 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3115 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3116 if (error != 0) {
3117 dsl_dataset_rele(ds, FTAG);
3118 return (error);
3119 }
3120
3121 if (newval == 0) {
3122 dsl_dataset_rele(ds, FTAG);
3123 return (0);
3124 }
3125
3126 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3127 newval < ds->ds_reserved) {
3128 dsl_dataset_rele(ds, FTAG);
3129 return (SET_ERROR(ENOSPC));
3130 }
3131
3132 dsl_dataset_rele(ds, FTAG);
3133 return (0);
3134 }
3135
3136 static void
3137 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3138 {
3139 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3140 dsl_pool_t *dp = dmu_tx_pool(tx);
3141 dsl_dataset_t *ds;
3142 uint64_t newval;
3143
3144 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3145
3146 dsl_prop_set_sync_impl(ds,
3147 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3148 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3149 &ddsqra->ddsqra_value, tx);
3150
3151 VERIFY0(dsl_prop_get_int_ds(ds,
3152 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3153
3154 if (ds->ds_quota != newval) {
3155 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3156 ds->ds_quota = newval;
3157 }
3158 dsl_dataset_rele(ds, FTAG);
3159 }
3160
3161 int
3162 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3163 uint64_t refquota)
3164 {
3165 dsl_dataset_set_qr_arg_t ddsqra;
3166
3167 ddsqra.ddsqra_name = dsname;
3168 ddsqra.ddsqra_source = source;
3169 ddsqra.ddsqra_value = refquota;
3170
3171 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3172 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3173 }
3174
3175 static int
3176 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3177 {
3178 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3179 dsl_pool_t *dp = dmu_tx_pool(tx);
3180 dsl_dataset_t *ds;
3181 int error;
3182 uint64_t newval, unique;
3183
3184 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3185 return (SET_ERROR(ENOTSUP));
3186
3187 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3188 if (error != 0)
3189 return (error);
3190
3191 if (ds->ds_is_snapshot) {
3192 dsl_dataset_rele(ds, FTAG);
3193 return (SET_ERROR(EINVAL));
3194 }
3195
3196 error = dsl_prop_predict(ds->ds_dir,
3197 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3198 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3199 if (error != 0) {
3200 dsl_dataset_rele(ds, FTAG);
3201 return (error);
3202 }
3203
3204 /*
3205 * If we are doing the preliminary check in open context, the
3206 * space estimates may be inaccurate.
3207 */
3208 if (!dmu_tx_is_syncing(tx)) {
3209 dsl_dataset_rele(ds, FTAG);
3210 return (0);
3211 }
3212
3213 mutex_enter(&ds->ds_lock);
3214 if (!DS_UNIQUE_IS_ACCURATE(ds))
3215 dsl_dataset_recalc_head_uniq(ds);
3216 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3217 mutex_exit(&ds->ds_lock);
3218
3219 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3220 uint64_t delta = MAX(unique, newval) -
3221 MAX(unique, ds->ds_reserved);
3222
3223 if (delta >
3224 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3225 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3226 dsl_dataset_rele(ds, FTAG);
3227 return (SET_ERROR(ENOSPC));
3228 }
3229 }
3230
3231 dsl_dataset_rele(ds, FTAG);
3232 return (0);
3233 }
3234
3235 void
3236 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3237 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3238 {
3239 uint64_t newval;
3240 uint64_t unique;
3241 int64_t delta;
3242
3243 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3244 source, sizeof (value), 1, &value, tx);
3245
3246 VERIFY0(dsl_prop_get_int_ds(ds,
3247 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3248
3249 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3250 mutex_enter(&ds->ds_dir->dd_lock);
3251 mutex_enter(&ds->ds_lock);
3252 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3253 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3254 delta = MAX(0, (int64_t)(newval - unique)) -
3255 MAX(0, (int64_t)(ds->ds_reserved - unique));
3256 ds->ds_reserved = newval;
3257 mutex_exit(&ds->ds_lock);
3258
3259 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3260 mutex_exit(&ds->ds_dir->dd_lock);
3261 }
3262
3263 static void
3264 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3265 {
3266 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3267 dsl_pool_t *dp = dmu_tx_pool(tx);
3268 dsl_dataset_t *ds;
3269
3270 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3271 dsl_dataset_set_refreservation_sync_impl(ds,
3272 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3273 dsl_dataset_rele(ds, FTAG);
3274 }
3275
3276 int
3277 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3278 uint64_t refreservation)
3279 {
3280 dsl_dataset_set_qr_arg_t ddsqra;
3281
3282 ddsqra.ddsqra_name = dsname;
3283 ddsqra.ddsqra_source = source;
3284 ddsqra.ddsqra_value = refreservation;
3285
3286 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3287 dsl_dataset_set_refreservation_sync, &ddsqra,
3288 0, ZFS_SPACE_CHECK_NONE));
3289 }
3290
3291 /*
3292 * Return (in *usedp) the amount of space written in new that is not
3293 * present in oldsnap. New may be a snapshot or the head. Old must be
3294 * a snapshot before new, in new's filesystem (or its origin). If not then
3295 * fail and return EINVAL.
3296 *
3297 * The written space is calculated by considering two components: First, we
3298 * ignore any freed space, and calculate the written as new's used space
3299 * minus old's used space. Next, we add in the amount of space that was freed
3300 * between the two snapshots, thus reducing new's used space relative to old's.
3301 * Specifically, this is the space that was born before old->ds_creation_txg,
3302 * and freed before new (ie. on new's deadlist or a previous deadlist).
3303 *
3304 * space freed [---------------------]
3305 * snapshots ---O-------O--------O-------O------
3306 * oldsnap new
3307 */
3308 int
3309 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3310 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3311 {
3312 int err = 0;
3313 uint64_t snapobj;
3314 dsl_pool_t *dp = new->ds_dir->dd_pool;
3315
3316 ASSERT(dsl_pool_config_held(dp));
3317
3318 *usedp = 0;
3319 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3320 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3321
3322 *compp = 0;
3323 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3324 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3325
3326 *uncompp = 0;
3327 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3328 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3329
3330 snapobj = new->ds_object;
3331 while (snapobj != oldsnap->ds_object) {
3332 dsl_dataset_t *snap;
3333 uint64_t used, comp, uncomp;
3334
3335 if (snapobj == new->ds_object) {
3336 snap = new;
3337 } else {
3338 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3339 if (err != 0)
3340 break;
3341 }
3342
3343 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3344 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3345 /*
3346 * The blocks in the deadlist can not be born after
3347 * ds_prev_snap_txg, so get the whole deadlist space,
3348 * which is more efficient (especially for old-format
3349 * deadlists). Unfortunately the deadlist code
3350 * doesn't have enough information to make this
3351 * optimization itself.
3352 */
3353 dsl_deadlist_space(&snap->ds_deadlist,
3354 &used, &comp, &uncomp);
3355 } else {
3356 dsl_deadlist_space_range(&snap->ds_deadlist,
3357 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3358 &used, &comp, &uncomp);
3359 }
3360 *usedp += used;
3361 *compp += comp;
3362 *uncompp += uncomp;
3363
3364 /*
3365 * If we get to the beginning of the chain of snapshots
3366 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3367 * was not a snapshot of/before new.
3368 */
3369 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3370 if (snap != new)
3371 dsl_dataset_rele(snap, FTAG);
3372 if (snapobj == 0) {
3373 err = SET_ERROR(EINVAL);
3374 break;
3375 }
3376
3377 }
3378 return (err);
3379 }
3380
3381 /*
3382 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3383 * lastsnap, and all snapshots in between are deleted.
3384 *
3385 * blocks that would be freed [---------------------------]
3386 * snapshots ---O-------O--------O-------O--------O
3387 * firstsnap lastsnap
3388 *
3389 * This is the set of blocks that were born after the snap before firstsnap,
3390 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3391 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3392 * We calculate this by iterating over the relevant deadlists (from the snap
3393 * after lastsnap, backward to the snap after firstsnap), summing up the
3394 * space on the deadlist that was born after the snap before firstsnap.
3395 */
3396 int
3397 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3398 dsl_dataset_t *lastsnap,
3399 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3400 {
3401 int err = 0;
3402 uint64_t snapobj;
3403 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3404
3405 ASSERT(firstsnap->ds_is_snapshot);
3406 ASSERT(lastsnap->ds_is_snapshot);
3407
3408 /*
3409 * Check that the snapshots are in the same dsl_dir, and firstsnap
3410 * is before lastsnap.
3411 */
3412 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3413 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3414 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3415 return (SET_ERROR(EINVAL));
3416
3417 *usedp = *compp = *uncompp = 0;
3418
3419 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3420 while (snapobj != firstsnap->ds_object) {
3421 dsl_dataset_t *ds;
3422 uint64_t used, comp, uncomp;
3423
3424 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3425 if (err != 0)
3426 break;
3427
3428 dsl_deadlist_space_range(&ds->ds_deadlist,
3429 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3430 &used, &comp, &uncomp);
3431 *usedp += used;
3432 *compp += comp;
3433 *uncompp += uncomp;
3434
3435 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3436 ASSERT3U(snapobj, !=, 0);
3437 dsl_dataset_rele(ds, FTAG);
3438 }
3439 return (err);
3440 }
3441
3442 /*
3443 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3444 * For example, they could both be snapshots of the same filesystem, and
3445 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3446 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3447 * filesystem. Or 'earlier' could be the origin's origin.
3448 *
3449 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3450 */
3451 boolean_t
3452 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3453 uint64_t earlier_txg)
3454 {
3455 dsl_pool_t *dp = later->ds_dir->dd_pool;
3456 int error;
3457 boolean_t ret;
3458
3459 ASSERT(dsl_pool_config_held(dp));
3460 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3461
3462 if (earlier_txg == 0)
3463 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3464
3465 if (later->ds_is_snapshot &&
3466 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3467 return (B_FALSE);
3468
3469 if (later->ds_dir == earlier->ds_dir)
3470 return (B_TRUE);
3471 if (!dsl_dir_is_clone(later->ds_dir))
3472 return (B_FALSE);
3473
3474 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3475 return (B_TRUE);
3476 dsl_dataset_t *origin;
3477 error = dsl_dataset_hold_obj(dp,
3478 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3479 if (error != 0)
3480 return (B_FALSE);
3481 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3482 dsl_dataset_rele(origin, FTAG);
3483 return (ret);
3484 }
3485
3486 void
3487 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3488 {
3489 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3490 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3491 }
3492
3493 boolean_t
3494 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3495 {
3496 dmu_object_info_t doi;
3497
3498 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3499 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3500 }
3501
3502 boolean_t
3503 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3504 {
3505 return (dsl_dataset_is_zapified(ds) &&
3506 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3507 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3508 }