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 */
28
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/arc.h>
38 #include <sys/zio.h>
39 #include <sys/zap.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/spa.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
47 #include <sys/zvol.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
53 #include <sys/dmu_send.h>
54 #include <sys/zio_checksum.h>
55 #include <sys/zio_compress.h>
56 #include <zfs_fletcher.h>
57
58 /*
59 * The SPA supports block sizes up to 16MB. However, very large blocks
60 * can have an impact on i/o latency (e.g. tying up a spinning disk for
61 * ~300ms), and also potentially on the memory allocator. Therefore,
62 * we do not allow the recordsize to be set larger than zfs_max_recordsize
63 * (default 1MB). Larger blocks can be created by changing this tunable,
64 * and pools with larger blocks can always be imported and used, regardless
65 * of this setting.
66 */
67 int zfs_max_recordsize = 1 * 1024 * 1024;
68
69 #define SWITCH64(x, y) \
70 { \
71 uint64_t __tmp = (x); \
72 (x) = (y); \
73 (y) = __tmp; \
74 }
75
76 #define DS_REF_MAX (1ULL << 62)
77
78 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
79
80 /*
81 * Figure out how much of this delta should be propogated to the dsl_dir
82 * layer. If there's a refreservation, that space has already been
83 * partially accounted for in our ancestors.
84 */
85 static int64_t
86 parent_delta(dsl_dataset_t *ds, int64_t delta)
87 {
88 dsl_dataset_phys_t *ds_phys;
89 uint64_t old_bytes, new_bytes;
90
91 if (ds->ds_reserved == 0)
92 return (delta);
93
94 ds_phys = dsl_dataset_phys(ds);
95 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
96 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
97
98 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
99 return (new_bytes - old_bytes);
100 }
101
102 void
103 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
104 {
105 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
106 int compressed = BP_GET_PSIZE(bp);
107 int uncompressed = BP_GET_UCSIZE(bp);
108 int64_t delta;
109
110 dprintf_bp(bp, "ds=%p", ds);
111
112 ASSERT(dmu_tx_is_syncing(tx));
113 /* It could have been compressed away to nothing */
114 if (BP_IS_HOLE(bp))
115 return;
116 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
117 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
118 if (ds == NULL) {
119 dsl_pool_mos_diduse_space(tx->tx_pool,
120 used, compressed, uncompressed);
121 return;
122 }
123
124 dmu_buf_will_dirty(ds->ds_dbuf, tx);
125 mutex_enter(&ds->ds_lock);
126 delta = parent_delta(ds, used);
127 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
128 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
129 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
130 dsl_dataset_phys(ds)->ds_unique_bytes += used;
131
132 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
133 ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
134 B_TRUE;
135 }
136
137 spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
138 if (f != SPA_FEATURE_NONE)
139 ds->ds_feature_activation_needed[f] = B_TRUE;
140
141 mutex_exit(&ds->ds_lock);
142 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
143 compressed, uncompressed, tx);
144 dsl_dir_transfer_space(ds->ds_dir, used - delta,
145 DD_USED_REFRSRV, DD_USED_HEAD, tx);
146 }
147
148 int
149 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
150 boolean_t async)
151 {
152 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
153 int compressed = BP_GET_PSIZE(bp);
154 int uncompressed = BP_GET_UCSIZE(bp);
155
156 if (BP_IS_HOLE(bp))
157 return (0);
158
159 ASSERT(dmu_tx_is_syncing(tx));
160 ASSERT(bp->blk_birth <= tx->tx_txg);
161
162 if (ds == NULL) {
163 dsl_free(tx->tx_pool, tx->tx_txg, bp);
164 dsl_pool_mos_diduse_space(tx->tx_pool,
165 -used, -compressed, -uncompressed);
166 return (used);
167 }
168 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
169
170 ASSERT(!ds->ds_is_snapshot);
171 dmu_buf_will_dirty(ds->ds_dbuf, tx);
172
173 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
174 int64_t delta;
175
176 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
177 dsl_free(tx->tx_pool, tx->tx_txg, bp);
178
179 mutex_enter(&ds->ds_lock);
180 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
181 !DS_UNIQUE_IS_ACCURATE(ds));
182 delta = parent_delta(ds, -used);
183 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
184 mutex_exit(&ds->ds_lock);
185 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
186 delta, -compressed, -uncompressed, tx);
187 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
188 DD_USED_REFRSRV, DD_USED_HEAD, tx);
189 } else {
190 dprintf_bp(bp, "putting on dead list: %s", "");
191 if (async) {
192 /*
193 * We are here as part of zio's write done callback,
194 * which means we're a zio interrupt thread. We can't
195 * call dsl_deadlist_insert() now because it may block
196 * waiting for I/O. Instead, put bp on the deferred
197 * queue and let dsl_pool_sync() finish the job.
198 */
199 bplist_append(&ds->ds_pending_deadlist, bp);
200 } else {
201 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
202 }
203 ASSERT3U(ds->ds_prev->ds_object, ==,
204 dsl_dataset_phys(ds)->ds_prev_snap_obj);
205 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
206 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
207 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
208 ds->ds_object && bp->blk_birth >
209 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
210 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
211 mutex_enter(&ds->ds_prev->ds_lock);
212 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
213 mutex_exit(&ds->ds_prev->ds_lock);
214 }
215 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
216 dsl_dir_transfer_space(ds->ds_dir, used,
217 DD_USED_HEAD, DD_USED_SNAP, tx);
218 }
219 }
220 mutex_enter(&ds->ds_lock);
221 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
222 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
223 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
224 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
225 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
226 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
227 mutex_exit(&ds->ds_lock);
228
229 return (used);
230 }
231
232 uint64_t
233 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
234 {
235 uint64_t trysnap = 0;
236
237 if (ds == NULL)
238 return (0);
239 /*
240 * The snapshot creation could fail, but that would cause an
241 * incorrect FALSE return, which would only result in an
242 * overestimation of the amount of space that an operation would
243 * consume, which is OK.
244 *
245 * There's also a small window where we could miss a pending
246 * snapshot, because we could set the sync task in the quiescing
247 * phase. So this should only be used as a guess.
248 */
249 if (ds->ds_trysnap_txg >
250 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
251 trysnap = ds->ds_trysnap_txg;
252 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
253 }
254
255 boolean_t
256 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
257 uint64_t blk_birth)
258 {
259 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
260 (bp != NULL && BP_IS_HOLE(bp)))
261 return (B_FALSE);
262
263 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
264
265 return (B_TRUE);
266 }
267
268 static void
269 dsl_dataset_evict(void *dbu)
270 {
271 dsl_dataset_t *ds = dbu;
272
273 ASSERT(ds->ds_owner == NULL);
274
275 ds->ds_dbuf = NULL;
276
277 unique_remove(ds->ds_fsid_guid);
278
279 if (ds->ds_objset != NULL)
280 dmu_objset_evict(ds->ds_objset);
281
282 if (ds->ds_prev) {
283 dsl_dataset_rele(ds->ds_prev, ds);
284 ds->ds_prev = NULL;
285 }
286
287 bplist_destroy(&ds->ds_pending_deadlist);
288 if (ds->ds_deadlist.dl_os != NULL)
289 dsl_deadlist_close(&ds->ds_deadlist);
290 if (ds->ds_dir)
291 dsl_dir_async_rele(ds->ds_dir, ds);
292
293 ASSERT(!list_link_active(&ds->ds_synced_link));
294
295 list_destroy(&ds->ds_prop_cbs);
296 mutex_destroy(&ds->ds_lock);
297 mutex_destroy(&ds->ds_opening_lock);
298 mutex_destroy(&ds->ds_sendstream_lock);
299 refcount_destroy(&ds->ds_longholds);
300
301 kmem_free(ds, sizeof (dsl_dataset_t));
302 }
303
304 int
305 dsl_dataset_get_snapname(dsl_dataset_t *ds)
306 {
307 dsl_dataset_phys_t *headphys;
308 int err;
309 dmu_buf_t *headdbuf;
310 dsl_pool_t *dp = ds->ds_dir->dd_pool;
311 objset_t *mos = dp->dp_meta_objset;
312
313 if (ds->ds_snapname[0])
314 return (0);
315 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
316 return (0);
317
318 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
319 FTAG, &headdbuf);
320 if (err != 0)
321 return (err);
322 headphys = headdbuf->db_data;
323 err = zap_value_search(dp->dp_meta_objset,
324 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
325 dmu_buf_rele(headdbuf, FTAG);
326 return (err);
327 }
328
329 int
330 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
331 {
332 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
333 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
334 matchtype_t mt;
335 int err;
336
337 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
338 mt = MT_FIRST;
339 else
340 mt = MT_EXACT;
341
342 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
343 value, mt, NULL, 0, NULL);
344 if (err == ENOTSUP && mt == MT_FIRST)
345 err = zap_lookup(mos, snapobj, name, 8, 1, value);
346 return (err);
347 }
348
349 int
350 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
351 boolean_t adj_cnt)
352 {
353 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
354 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
355 matchtype_t mt;
356 int err;
357
358 dsl_dir_snap_cmtime_update(ds->ds_dir);
359
360 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
361 mt = MT_FIRST;
362 else
363 mt = MT_EXACT;
364
365 err = zap_remove_norm(mos, snapobj, name, mt, tx);
366 if (err == ENOTSUP && mt == MT_FIRST)
367 err = zap_remove(mos, snapobj, name, tx);
368
369 if (err == 0 && adj_cnt)
370 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
371 DD_FIELD_SNAPSHOT_COUNT, tx);
372
373 return (err);
374 }
375
376 boolean_t
377 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
378 {
379 dmu_buf_t *dbuf = ds->ds_dbuf;
380 boolean_t result = B_FALSE;
381
382 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
383 ds->ds_object, DMU_BONUS_BLKID, tag)) {
384
385 if (ds == dmu_buf_get_user(dbuf))
386 result = B_TRUE;
387 else
388 dmu_buf_rele(dbuf, tag);
389 }
390
391 return (result);
392 }
393
394 int
395 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
396 dsl_dataset_t **dsp)
397 {
398 objset_t *mos = dp->dp_meta_objset;
399 dmu_buf_t *dbuf;
400 dsl_dataset_t *ds;
401 int err;
402 dmu_object_info_t doi;
403
404 ASSERT(dsl_pool_config_held(dp));
405
406 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
407 if (err != 0)
408 return (err);
409
410 /* Make sure dsobj has the correct object type. */
411 dmu_object_info_from_db(dbuf, &doi);
412 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
413 dmu_buf_rele(dbuf, tag);
414 return (SET_ERROR(EINVAL));
415 }
416
417 ds = dmu_buf_get_user(dbuf);
418 if (ds == NULL) {
419 dsl_dataset_t *winner = NULL;
420
421 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
422 ds->ds_dbuf = dbuf;
423 ds->ds_object = dsobj;
424 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
425
426 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
427 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
428 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
429 refcount_create(&ds->ds_longholds);
430
431 bplist_create(&ds->ds_pending_deadlist);
432 dsl_deadlist_open(&ds->ds_deadlist,
433 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
434
435 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
436 offsetof(dmu_sendarg_t, dsa_link));
437
438 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
439 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
440
441 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
442 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
443 if (!(spa_feature_table[f].fi_flags &
444 ZFEATURE_FLAG_PER_DATASET))
445 continue;
446 err = zap_contains(mos, dsobj,
447 spa_feature_table[f].fi_guid);
448 if (err == 0) {
449 ds->ds_feature_inuse[f] = B_TRUE;
450 } else {
451 ASSERT3U(err, ==, ENOENT);
452 err = 0;
453 }
454 }
455 }
456
457 err = dsl_dir_hold_obj(dp,
458 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds, &ds->ds_dir);
459 if (err != 0) {
460 mutex_destroy(&ds->ds_lock);
461 mutex_destroy(&ds->ds_opening_lock);
462 mutex_destroy(&ds->ds_sendstream_lock);
463 refcount_destroy(&ds->ds_longholds);
464 bplist_destroy(&ds->ds_pending_deadlist);
465 dsl_deadlist_close(&ds->ds_deadlist);
466 kmem_free(ds, sizeof (dsl_dataset_t));
467 dmu_buf_rele(dbuf, tag);
468 return (err);
469 }
470
471 if (!ds->ds_is_snapshot) {
472 ds->ds_snapname[0] = '\0';
473 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
474 err = dsl_dataset_hold_obj(dp,
475 dsl_dataset_phys(ds)->ds_prev_snap_obj,
476 ds, &ds->ds_prev);
477 }
478 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
479 int zaperr = zap_lookup(mos, ds->ds_object,
480 DS_FIELD_BOOKMARK_NAMES,
481 sizeof (ds->ds_bookmarks), 1,
482 &ds->ds_bookmarks);
483 if (zaperr != ENOENT)
484 VERIFY0(zaperr);
485 }
486 } else {
487 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
488 err = dsl_dataset_get_snapname(ds);
489 if (err == 0 &&
490 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
491 err = zap_count(
492 ds->ds_dir->dd_pool->dp_meta_objset,
493 dsl_dataset_phys(ds)->ds_userrefs_obj,
494 &ds->ds_userrefs);
495 }
496 }
497
498 if (err == 0 && !ds->ds_is_snapshot) {
499 err = dsl_prop_get_int_ds(ds,
500 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
501 &ds->ds_reserved);
502 if (err == 0) {
503 err = dsl_prop_get_int_ds(ds,
504 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
505 &ds->ds_quota);
506 }
507 } else {
508 ds->ds_reserved = ds->ds_quota = 0;
509 }
510
511 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict, &ds->ds_dbuf);
512 if (err == 0)
513 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
514
515 if (err != 0 || winner != NULL) {
516 bplist_destroy(&ds->ds_pending_deadlist);
517 dsl_deadlist_close(&ds->ds_deadlist);
518 if (ds->ds_prev)
519 dsl_dataset_rele(ds->ds_prev, ds);
520 dsl_dir_rele(ds->ds_dir, ds);
521 mutex_destroy(&ds->ds_lock);
522 mutex_destroy(&ds->ds_opening_lock);
523 mutex_destroy(&ds->ds_sendstream_lock);
524 refcount_destroy(&ds->ds_longholds);
525 kmem_free(ds, sizeof (dsl_dataset_t));
526 if (err != 0) {
527 dmu_buf_rele(dbuf, tag);
528 return (err);
529 }
530 ds = winner;
531 } else {
532 ds->ds_fsid_guid =
533 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
534 }
535 }
536 ASSERT3P(ds->ds_dbuf, ==, dbuf);
537 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
538 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
539 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
540 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
541 *dsp = ds;
542 return (0);
543 }
544
545 int
546 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
547 void *tag, dsl_dataset_t **dsp)
548 {
549 dsl_dir_t *dd;
550 const char *snapname;
551 uint64_t obj;
552 int err = 0;
553 dsl_dataset_t *ds;
554
555 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
556 if (err != 0)
557 return (err);
558
559 ASSERT(dsl_pool_config_held(dp));
560 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
561 if (obj != 0)
562 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
563 else
564 err = SET_ERROR(ENOENT);
565
566 /* we may be looking for a snapshot */
567 if (err == 0 && snapname != NULL) {
568 dsl_dataset_t *snap_ds;
569
570 if (*snapname++ != '@') {
571 dsl_dataset_rele(ds, tag);
572 dsl_dir_rele(dd, FTAG);
573 return (SET_ERROR(ENOENT));
574 }
575
576 dprintf("looking for snapshot '%s'\n", snapname);
577 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
578 if (err == 0)
579 err = dsl_dataset_hold_obj(dp, obj, tag, &snap_ds);
580 dsl_dataset_rele(ds, tag);
581
582 if (err == 0) {
583 mutex_enter(&snap_ds->ds_lock);
584 if (snap_ds->ds_snapname[0] == 0)
585 (void) strlcpy(snap_ds->ds_snapname, snapname,
586 sizeof (snap_ds->ds_snapname));
587 mutex_exit(&snap_ds->ds_lock);
588 ds = snap_ds;
589 }
590 }
591 if (err == 0)
592 *dsp = ds;
593 dsl_dir_rele(dd, FTAG);
594 return (err);
595 }
596
597 int
598 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
599 void *tag, dsl_dataset_t **dsp)
600 {
601 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
602 if (err != 0)
603 return (err);
604 if (!dsl_dataset_tryown(*dsp, tag)) {
605 dsl_dataset_rele(*dsp, tag);
606 *dsp = NULL;
607 return (SET_ERROR(EBUSY));
608 }
609 return (0);
610 }
611
612 int
613 dsl_dataset_own(dsl_pool_t *dp, const char *name,
614 void *tag, dsl_dataset_t **dsp)
615 {
616 int err = dsl_dataset_hold(dp, name, tag, dsp);
617 if (err != 0)
618 return (err);
619 if (!dsl_dataset_tryown(*dsp, tag)) {
620 dsl_dataset_rele(*dsp, tag);
621 return (SET_ERROR(EBUSY));
622 }
623 return (0);
624 }
625
626 /*
627 * See the comment above dsl_pool_hold() for details. In summary, a long
628 * hold is used to prevent destruction of a dataset while the pool hold
629 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
630 *
631 * The dataset and pool must be held when this function is called. After it
632 * is called, the pool hold may be released while the dataset is still held
633 * and accessed.
634 */
635 void
636 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
637 {
638 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
639 (void) refcount_add(&ds->ds_longholds, tag);
640 }
641
642 void
643 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
644 {
645 (void) refcount_remove(&ds->ds_longholds, tag);
646 }
647
648 /* Return B_TRUE if there are any long holds on this dataset. */
649 boolean_t
650 dsl_dataset_long_held(dsl_dataset_t *ds)
651 {
652 return (!refcount_is_zero(&ds->ds_longholds));
653 }
654
655 void
656 dsl_dataset_name(dsl_dataset_t *ds, char *name)
657 {
658 if (ds == NULL) {
659 (void) strcpy(name, "mos");
660 } else {
661 dsl_dir_name(ds->ds_dir, name);
662 VERIFY0(dsl_dataset_get_snapname(ds));
663 if (ds->ds_snapname[0]) {
664 (void) strcat(name, "@");
665 /*
666 * We use a "recursive" mutex so that we
667 * can call dprintf_ds() with ds_lock held.
668 */
669 if (!MUTEX_HELD(&ds->ds_lock)) {
670 mutex_enter(&ds->ds_lock);
671 VERIFY3U(strlen(name) +
672 strlen(ds->ds_snapname) + 1, <=,
673 ZFS_MAXNAMELEN);
674 (void) strcat(name, ds->ds_snapname);
675 mutex_exit(&ds->ds_lock);
676 } else {
677 VERIFY3U(strlen(name) +
678 strlen(ds->ds_snapname) + 1, <=,
679 ZFS_MAXNAMELEN);
680 (void) strcat(name, ds->ds_snapname);
681 }
682 }
683 }
684 }
685
686 void
687 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
688 {
689 dmu_buf_rele(ds->ds_dbuf, tag);
690 }
691
692 void
693 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
694 {
695 ASSERT3P(ds->ds_owner, ==, tag);
696 ASSERT(ds->ds_dbuf != NULL);
697
698 mutex_enter(&ds->ds_lock);
699 ds->ds_owner = NULL;
700 mutex_exit(&ds->ds_lock);
701 dsl_dataset_long_rele(ds, tag);
702 dsl_dataset_rele(ds, tag);
703 }
704
705 boolean_t
706 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
707 {
708 boolean_t gotit = FALSE;
709
710 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
711 mutex_enter(&ds->ds_lock);
712 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
713 ds->ds_owner = tag;
714 dsl_dataset_long_hold(ds, tag);
715 gotit = TRUE;
716 }
717 mutex_exit(&ds->ds_lock);
718 return (gotit);
719 }
720
721 boolean_t
722 dsl_dataset_has_owner(dsl_dataset_t *ds)
723 {
724 boolean_t rv;
725 mutex_enter(&ds->ds_lock);
726 rv = (ds->ds_owner != NULL);
727 mutex_exit(&ds->ds_lock);
728 return (rv);
729 }
730
731 static void
732 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
733 {
734 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
735 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
736 uint64_t zero = 0;
737
738 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
739
740 spa_feature_incr(spa, f, tx);
741 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
742
743 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
744 sizeof (zero), 1, &zero, tx));
745 }
746
747 void
748 dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
749 {
750 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
751 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
752
753 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
754
755 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
756 spa_feature_decr(spa, f, tx);
757 }
758
759 uint64_t
760 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
761 uint64_t flags, dmu_tx_t *tx)
762 {
763 dsl_pool_t *dp = dd->dd_pool;
764 dmu_buf_t *dbuf;
765 dsl_dataset_phys_t *dsphys;
766 uint64_t dsobj;
767 objset_t *mos = dp->dp_meta_objset;
768
769 if (origin == NULL)
770 origin = dp->dp_origin_snap;
771
772 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
773 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
774 ASSERT(dmu_tx_is_syncing(tx));
775 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
776
777 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
778 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
779 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
780 dmu_buf_will_dirty(dbuf, tx);
781 dsphys = dbuf->db_data;
782 bzero(dsphys, sizeof (dsl_dataset_phys_t));
783 dsphys->ds_dir_obj = dd->dd_object;
784 dsphys->ds_flags = flags;
785 dsphys->ds_fsid_guid = unique_create();
786 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
787 sizeof (dsphys->ds_guid));
788 dsphys->ds_snapnames_zapobj =
789 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
790 DMU_OT_NONE, 0, tx);
791 dsphys->ds_creation_time = gethrestime_sec();
792 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
793
794 if (origin == NULL) {
795 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
796 } else {
797 dsl_dataset_t *ohds; /* head of the origin snapshot */
798
799 dsphys->ds_prev_snap_obj = origin->ds_object;
800 dsphys->ds_prev_snap_txg =
801 dsl_dataset_phys(origin)->ds_creation_txg;
802 dsphys->ds_referenced_bytes =
803 dsl_dataset_phys(origin)->ds_referenced_bytes;
804 dsphys->ds_compressed_bytes =
805 dsl_dataset_phys(origin)->ds_compressed_bytes;
806 dsphys->ds_uncompressed_bytes =
807 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
808 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
809
810 /*
811 * Inherit flags that describe the dataset's contents
812 * (INCONSISTENT) or properties (Case Insensitive).
813 */
814 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
815 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
816
817 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
818 if (origin->ds_feature_inuse[f])
819 dsl_dataset_activate_feature(dsobj, f, tx);
820 }
821
822 dmu_buf_will_dirty(origin->ds_dbuf, tx);
823 dsl_dataset_phys(origin)->ds_num_children++;
824
825 VERIFY0(dsl_dataset_hold_obj(dp,
826 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
827 FTAG, &ohds));
828 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
829 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
830 dsl_dataset_rele(ohds, FTAG);
831
832 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
833 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
834 dsl_dataset_phys(origin)->ds_next_clones_obj =
835 zap_create(mos,
836 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
837 }
838 VERIFY0(zap_add_int(mos,
839 dsl_dataset_phys(origin)->ds_next_clones_obj,
840 dsobj, tx));
841 }
842
843 dmu_buf_will_dirty(dd->dd_dbuf, tx);
844 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
845 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
846 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
847 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
848 dsl_dir_phys(origin->ds_dir)->dd_clones =
849 zap_create(mos,
850 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
851 }
852 VERIFY0(zap_add_int(mos,
853 dsl_dir_phys(origin->ds_dir)->dd_clones,
854 dsobj, tx));
855 }
856 }
857
858 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
859 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
860
861 dmu_buf_rele(dbuf, FTAG);
862
863 dmu_buf_will_dirty(dd->dd_dbuf, tx);
864 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
865
866 return (dsobj);
867 }
868
869 static void
870 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
871 {
872 objset_t *os;
873
874 VERIFY0(dmu_objset_from_ds(ds, &os));
875 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
876 dsl_dataset_dirty(ds, tx);
877 }
878
879 uint64_t
880 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
881 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
882 {
883 dsl_pool_t *dp = pdd->dd_pool;
884 uint64_t dsobj, ddobj;
885 dsl_dir_t *dd;
886
887 ASSERT(dmu_tx_is_syncing(tx));
888 ASSERT(lastname[0] != '@');
889
890 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
891 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
892
893 dsobj = dsl_dataset_create_sync_dd(dd, origin,
894 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
895
896 dsl_deleg_set_create_perms(dd, tx, cr);
897
898 /*
899 * Since we're creating a new node we know it's a leaf, so we can
900 * initialize the counts if the limit feature is active.
901 */
902 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
903 uint64_t cnt = 0;
904 objset_t *os = dd->dd_pool->dp_meta_objset;
905
906 dsl_dir_zapify(dd, tx);
907 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
908 sizeof (cnt), 1, &cnt, tx));
909 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
910 sizeof (cnt), 1, &cnt, tx));
911 }
912
913 dsl_dir_rele(dd, FTAG);
914
915 /*
916 * If we are creating a clone, make sure we zero out any stale
917 * data from the origin snapshots zil header.
918 */
919 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
920 dsl_dataset_t *ds;
921
922 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
923 dsl_dataset_zero_zil(ds, tx);
924 dsl_dataset_rele(ds, FTAG);
925 }
926
927 return (dsobj);
928 }
929
930 /*
931 * The unique space in the head dataset can be calculated by subtracting
932 * the space used in the most recent snapshot, that is still being used
933 * in this file system, from the space currently in use. To figure out
934 * the space in the most recent snapshot still in use, we need to take
935 * the total space used in the snapshot and subtract out the space that
936 * has been freed up since the snapshot was taken.
937 */
938 void
939 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
940 {
941 uint64_t mrs_used;
942 uint64_t dlused, dlcomp, dluncomp;
943
944 ASSERT(!ds->ds_is_snapshot);
945
946 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
947 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
948 else
949 mrs_used = 0;
950
951 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
952
953 ASSERT3U(dlused, <=, mrs_used);
954 dsl_dataset_phys(ds)->ds_unique_bytes =
955 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
956
957 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
958 SPA_VERSION_UNIQUE_ACCURATE)
959 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
960 }
961
962 void
963 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
964 dmu_tx_t *tx)
965 {
966 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
967 uint64_t count;
968 int err;
969
970 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
971 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
972 obj, tx);
973 /*
974 * The err should not be ENOENT, but a bug in a previous version
975 * of the code could cause upgrade_clones_cb() to not set
976 * ds_next_snap_obj when it should, leading to a missing entry.
977 * If we knew that the pool was created after
978 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
979 * ENOENT. However, at least we can check that we don't have
980 * too many entries in the next_clones_obj even after failing to
981 * remove this one.
982 */
983 if (err != ENOENT)
984 VERIFY0(err);
985 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
986 &count));
987 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
988 }
989
990
991 blkptr_t *
992 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
993 {
994 return (&dsl_dataset_phys(ds)->ds_bp);
995 }
996
997 void
998 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
999 {
1000 ASSERT(dmu_tx_is_syncing(tx));
1001 /* If it's the meta-objset, set dp_meta_rootbp */
1002 if (ds == NULL) {
1003 tx->tx_pool->dp_meta_rootbp = *bp;
1004 } else {
1005 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1006 dsl_dataset_phys(ds)->ds_bp = *bp;
1007 }
1008 }
1009
1010 spa_t *
1011 dsl_dataset_get_spa(dsl_dataset_t *ds)
1012 {
1013 return (ds->ds_dir->dd_pool->dp_spa);
1014 }
1015
1016 void
1017 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1018 {
1019 dsl_pool_t *dp;
1020
1021 if (ds == NULL) /* this is the meta-objset */
1022 return;
1023
1024 ASSERT(ds->ds_objset != NULL);
1025
1026 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1027 panic("dirtying snapshot!");
1028
1029 dp = ds->ds_dir->dd_pool;
1030
1031 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1032 /* up the hold count until we can be written out */
1033 dmu_buf_add_ref(ds->ds_dbuf, ds);
1034 }
1035 }
1036
1037 boolean_t
1038 dsl_dataset_is_dirty(dsl_dataset_t *ds)
1039 {
1040 for (int t = 0; t < TXG_SIZE; t++) {
1041 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1042 ds, t))
1043 return (B_TRUE);
1044 }
1045 return (B_FALSE);
1046 }
1047
1048 static int
1049 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1050 {
1051 uint64_t asize;
1052
1053 if (!dmu_tx_is_syncing(tx))
1054 return (0);
1055
1056 /*
1057 * If there's an fs-only reservation, any blocks that might become
1058 * owned by the snapshot dataset must be accommodated by space
1059 * outside of the reservation.
1060 */
1061 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1062 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1063 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1064 return (SET_ERROR(ENOSPC));
1065
1066 /*
1067 * Propagate any reserved space for this snapshot to other
1068 * snapshot checks in this sync group.
1069 */
1070 if (asize > 0)
1071 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1072
1073 return (0);
1074 }
1075
1076 typedef struct dsl_dataset_snapshot_arg {
1077 nvlist_t *ddsa_snaps;
1078 nvlist_t *ddsa_props;
1079 nvlist_t *ddsa_errors;
1080 cred_t *ddsa_cr;
1081 } dsl_dataset_snapshot_arg_t;
1082
1083 int
1084 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1085 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1086 {
1087 int error;
1088 uint64_t value;
1089
1090 ds->ds_trysnap_txg = tx->tx_txg;
1091
1092 if (!dmu_tx_is_syncing(tx))
1093 return (0);
1094
1095 /*
1096 * We don't allow multiple snapshots of the same txg. If there
1097 * is already one, try again.
1098 */
1099 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1100 return (SET_ERROR(EAGAIN));
1101
1102 /*
1103 * Check for conflicting snapshot name.
1104 */
1105 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1106 if (error == 0)
1107 return (SET_ERROR(EEXIST));
1108 if (error != ENOENT)
1109 return (error);
1110
1111 /*
1112 * We don't allow taking snapshots of inconsistent datasets, such as
1113 * those into which we are currently receiving. However, if we are
1114 * creating this snapshot as part of a receive, this check will be
1115 * executed atomically with respect to the completion of the receive
1116 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1117 * case we ignore this, knowing it will be fixed up for us shortly in
1118 * dmu_recv_end_sync().
1119 */
1120 if (!recv && DS_IS_INCONSISTENT(ds))
1121 return (SET_ERROR(EBUSY));
1122
1123 /*
1124 * Skip the check for temporary snapshots or if we have already checked
1125 * the counts in dsl_dataset_snapshot_check. This means we really only
1126 * check the count here when we're receiving a stream.
1127 */
1128 if (cnt != 0 && cr != NULL) {
1129 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1130 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1131 if (error != 0)
1132 return (error);
1133 }
1134
1135 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1136 if (error != 0)
1137 return (error);
1138
1139 return (0);
1140 }
1141
1142 static int
1143 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1144 {
1145 dsl_dataset_snapshot_arg_t *ddsa = arg;
1146 dsl_pool_t *dp = dmu_tx_pool(tx);
1147 nvpair_t *pair;
1148 int rv = 0;
1149
1150 /*
1151 * Pre-compute how many total new snapshots will be created for each
1152 * level in the tree and below. This is needed for validating the
1153 * snapshot limit when either taking a recursive snapshot or when
1154 * taking multiple snapshots.
1155 *
1156 * The problem is that the counts are not actually adjusted when
1157 * we are checking, only when we finally sync. For a single snapshot,
1158 * this is easy, the count will increase by 1 at each node up the tree,
1159 * but its more complicated for the recursive/multiple snapshot case.
1160 *
1161 * The dsl_fs_ss_limit_check function does recursively check the count
1162 * at each level up the tree but since it is validating each snapshot
1163 * independently we need to be sure that we are validating the complete
1164 * count for the entire set of snapshots. We do this by rolling up the
1165 * counts for each component of the name into an nvlist and then
1166 * checking each of those cases with the aggregated count.
1167 *
1168 * This approach properly handles not only the recursive snapshot
1169 * case (where we get all of those on the ddsa_snaps list) but also
1170 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1171 * validate the limit on 'a' using a count of 2).
1172 *
1173 * We validate the snapshot names in the third loop and only report
1174 * name errors once.
1175 */
1176 if (dmu_tx_is_syncing(tx)) {
1177 nvlist_t *cnt_track = NULL;
1178 cnt_track = fnvlist_alloc();
1179
1180 /* Rollup aggregated counts into the cnt_track list */
1181 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1182 pair != NULL;
1183 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1184 char *pdelim;
1185 uint64_t val;
1186 char nm[MAXPATHLEN];
1187
1188 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1189 pdelim = strchr(nm, '@');
1190 if (pdelim == NULL)
1191 continue;
1192 *pdelim = '\0';
1193
1194 do {
1195 if (nvlist_lookup_uint64(cnt_track, nm,
1196 &val) == 0) {
1197 /* update existing entry */
1198 fnvlist_add_uint64(cnt_track, nm,
1199 val + 1);
1200 } else {
1201 /* add to list */
1202 fnvlist_add_uint64(cnt_track, nm, 1);
1203 }
1204
1205 pdelim = strrchr(nm, '/');
1206 if (pdelim != NULL)
1207 *pdelim = '\0';
1208 } while (pdelim != NULL);
1209 }
1210
1211 /* Check aggregated counts at each level */
1212 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1213 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1214 int error = 0;
1215 char *name;
1216 uint64_t cnt = 0;
1217 dsl_dataset_t *ds;
1218
1219 name = nvpair_name(pair);
1220 cnt = fnvpair_value_uint64(pair);
1221 ASSERT(cnt > 0);
1222
1223 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1224 if (error == 0) {
1225 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1226 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1227 ddsa->ddsa_cr);
1228 dsl_dataset_rele(ds, FTAG);
1229 }
1230
1231 if (error != 0) {
1232 if (ddsa->ddsa_errors != NULL)
1233 fnvlist_add_int32(ddsa->ddsa_errors,
1234 name, error);
1235 rv = error;
1236 /* only report one error for this check */
1237 break;
1238 }
1239 }
1240 nvlist_free(cnt_track);
1241 }
1242
1243 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1244 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1245 int error = 0;
1246 dsl_dataset_t *ds;
1247 char *name, *atp;
1248 char dsname[MAXNAMELEN];
1249
1250 name = nvpair_name(pair);
1251 if (strlen(name) >= MAXNAMELEN)
1252 error = SET_ERROR(ENAMETOOLONG);
1253 if (error == 0) {
1254 atp = strchr(name, '@');
1255 if (atp == NULL)
1256 error = SET_ERROR(EINVAL);
1257 if (error == 0)
1258 (void) strlcpy(dsname, name, atp - name + 1);
1259 }
1260 if (error == 0)
1261 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1262 if (error == 0) {
1263 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1264 error = dsl_dataset_snapshot_check_impl(ds,
1265 atp + 1, tx, B_FALSE, 0, NULL);
1266 dsl_dataset_rele(ds, FTAG);
1267 }
1268
1269 if (error != 0) {
1270 if (ddsa->ddsa_errors != NULL) {
1271 fnvlist_add_int32(ddsa->ddsa_errors,
1272 name, error);
1273 }
1274 rv = error;
1275 }
1276 }
1277
1278 return (rv);
1279 }
1280
1281 void
1282 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1283 dmu_tx_t *tx)
1284 {
1285 static zil_header_t zero_zil;
1286
1287 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1288 dmu_buf_t *dbuf;
1289 dsl_dataset_phys_t *dsphys;
1290 uint64_t dsobj, crtxg;
1291 objset_t *mos = dp->dp_meta_objset;
1292 objset_t *os;
1293
1294 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1295
1296 /*
1297 * If we are on an old pool, the zil must not be active, in which
1298 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1299 */
1300 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1301 dmu_objset_from_ds(ds, &os) != 0 ||
1302 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1303 sizeof (zero_zil)) == 0);
1304
1305 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1306
1307 /*
1308 * The origin's ds_creation_txg has to be < TXG_INITIAL
1309 */
1310 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1311 crtxg = 1;
1312 else
1313 crtxg = tx->tx_txg;
1314
1315 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1316 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1317 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1318 dmu_buf_will_dirty(dbuf, tx);
1319 dsphys = dbuf->db_data;
1320 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1321 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1322 dsphys->ds_fsid_guid = unique_create();
1323 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1324 sizeof (dsphys->ds_guid));
1325 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1326 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1327 dsphys->ds_next_snap_obj = ds->ds_object;
1328 dsphys->ds_num_children = 1;
1329 dsphys->ds_creation_time = gethrestime_sec();
1330 dsphys->ds_creation_txg = crtxg;
1331 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1332 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1333 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1334 dsphys->ds_uncompressed_bytes =
1335 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1336 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1337 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1338 dmu_buf_rele(dbuf, FTAG);
1339
1340 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1341 if (ds->ds_feature_inuse[f])
1342 dsl_dataset_activate_feature(dsobj, f, tx);
1343 }
1344
1345 ASSERT3U(ds->ds_prev != 0, ==,
1346 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1347 if (ds->ds_prev) {
1348 uint64_t next_clones_obj =
1349 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1350 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1351 ds->ds_object ||
1352 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1353 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1354 ds->ds_object) {
1355 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1356 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1357 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1358 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1359 } else if (next_clones_obj != 0) {
1360 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1361 dsphys->ds_next_snap_obj, tx);
1362 VERIFY0(zap_add_int(mos,
1363 next_clones_obj, dsobj, tx));
1364 }
1365 }
1366
1367 /*
1368 * If we have a reference-reservation on this dataset, we will
1369 * need to increase the amount of refreservation being charged
1370 * since our unique space is going to zero.
1371 */
1372 if (ds->ds_reserved) {
1373 int64_t delta;
1374 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1375 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1376 ds->ds_reserved);
1377 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1378 delta, 0, 0, tx);
1379 }
1380
1381 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1382 dsl_dataset_phys(ds)->ds_deadlist_obj =
1383 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1384 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1385 dsl_deadlist_close(&ds->ds_deadlist);
1386 dsl_deadlist_open(&ds->ds_deadlist, mos,
1387 dsl_dataset_phys(ds)->ds_deadlist_obj);
1388 dsl_deadlist_add_key(&ds->ds_deadlist,
1389 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1390
1391 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1392 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1393 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1394 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1395 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1396 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1397
1398 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1399 snapname, 8, 1, &dsobj, tx));
1400
1401 if (ds->ds_prev)
1402 dsl_dataset_rele(ds->ds_prev, ds);
1403 VERIFY0(dsl_dataset_hold_obj(dp,
1404 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1405
1406 dsl_scan_ds_snapshotted(ds, tx);
1407
1408 dsl_dir_snap_cmtime_update(ds->ds_dir);
1409
1410 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1411 }
1412
1413 static void
1414 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1415 {
1416 dsl_dataset_snapshot_arg_t *ddsa = arg;
1417 dsl_pool_t *dp = dmu_tx_pool(tx);
1418 nvpair_t *pair;
1419
1420 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1421 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1422 dsl_dataset_t *ds;
1423 char *name, *atp;
1424 char dsname[MAXNAMELEN];
1425
1426 name = nvpair_name(pair);
1427 atp = strchr(name, '@');
1428 (void) strlcpy(dsname, name, atp - name + 1);
1429 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1430
1431 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1432 if (ddsa->ddsa_props != NULL) {
1433 dsl_props_set_sync_impl(ds->ds_prev,
1434 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1435 }
1436 dsl_dataset_rele(ds, FTAG);
1437 }
1438 }
1439
1440 /*
1441 * The snapshots must all be in the same pool.
1442 * All-or-nothing: if there are any failures, nothing will be modified.
1443 */
1444 int
1445 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1446 {
1447 dsl_dataset_snapshot_arg_t ddsa;
1448 nvpair_t *pair;
1449 boolean_t needsuspend;
1450 int error;
1451 spa_t *spa;
1452 char *firstname;
1453 nvlist_t *suspended = NULL;
1454
1455 pair = nvlist_next_nvpair(snaps, NULL);
1456 if (pair == NULL)
1457 return (0);
1458 firstname = nvpair_name(pair);
1459
1460 error = spa_open(firstname, &spa, FTAG);
1461 if (error != 0)
1462 return (error);
1463 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1464 spa_close(spa, FTAG);
1465
1466 if (needsuspend) {
1467 suspended = fnvlist_alloc();
1468 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1469 pair = nvlist_next_nvpair(snaps, pair)) {
1470 char fsname[MAXNAMELEN];
1471 char *snapname = nvpair_name(pair);
1472 char *atp;
1473 void *cookie;
1474
1475 atp = strchr(snapname, '@');
1476 if (atp == NULL) {
1477 error = SET_ERROR(EINVAL);
1478 break;
1479 }
1480 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1481
1482 error = zil_suspend(fsname, &cookie);
1483 if (error != 0)
1484 break;
1485 fnvlist_add_uint64(suspended, fsname,
1486 (uintptr_t)cookie);
1487 }
1488 }
1489
1490 ddsa.ddsa_snaps = snaps;
1491 ddsa.ddsa_props = props;
1492 ddsa.ddsa_errors = errors;
1493 ddsa.ddsa_cr = CRED();
1494
1495 if (error == 0) {
1496 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1497 dsl_dataset_snapshot_sync, &ddsa,
1498 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1499 }
1500
1501 if (suspended != NULL) {
1502 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1503 pair = nvlist_next_nvpair(suspended, pair)) {
1504 zil_resume((void *)(uintptr_t)
1505 fnvpair_value_uint64(pair));
1506 }
1507 fnvlist_free(suspended);
1508 }
1509
1510 return (error);
1511 }
1512
1513 typedef struct dsl_dataset_snapshot_tmp_arg {
1514 const char *ddsta_fsname;
1515 const char *ddsta_snapname;
1516 minor_t ddsta_cleanup_minor;
1517 const char *ddsta_htag;
1518 } dsl_dataset_snapshot_tmp_arg_t;
1519
1520 static int
1521 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1522 {
1523 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1524 dsl_pool_t *dp = dmu_tx_pool(tx);
1525 dsl_dataset_t *ds;
1526 int error;
1527
1528 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1529 if (error != 0)
1530 return (error);
1531
1532 /* NULL cred means no limit check for tmp snapshot */
1533 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1534 tx, B_FALSE, 0, NULL);
1535 if (error != 0) {
1536 dsl_dataset_rele(ds, FTAG);
1537 return (error);
1538 }
1539
1540 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1541 dsl_dataset_rele(ds, FTAG);
1542 return (SET_ERROR(ENOTSUP));
1543 }
1544 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1545 B_TRUE, tx);
1546 if (error != 0) {
1547 dsl_dataset_rele(ds, FTAG);
1548 return (error);
1549 }
1550
1551 dsl_dataset_rele(ds, FTAG);
1552 return (0);
1553 }
1554
1555 static void
1556 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1557 {
1558 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1559 dsl_pool_t *dp = dmu_tx_pool(tx);
1560 dsl_dataset_t *ds;
1561
1562 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1563
1564 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1565 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1566 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1567 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1568
1569 dsl_dataset_rele(ds, FTAG);
1570 }
1571
1572 int
1573 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1574 minor_t cleanup_minor, const char *htag)
1575 {
1576 dsl_dataset_snapshot_tmp_arg_t ddsta;
1577 int error;
1578 spa_t *spa;
1579 boolean_t needsuspend;
1580 void *cookie;
1581
1582 ddsta.ddsta_fsname = fsname;
1583 ddsta.ddsta_snapname = snapname;
1584 ddsta.ddsta_cleanup_minor = cleanup_minor;
1585 ddsta.ddsta_htag = htag;
1586
1587 error = spa_open(fsname, &spa, FTAG);
1588 if (error != 0)
1589 return (error);
1590 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1591 spa_close(spa, FTAG);
1592
1593 if (needsuspend) {
1594 error = zil_suspend(fsname, &cookie);
1595 if (error != 0)
1596 return (error);
1597 }
1598
1599 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1600 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1601
1602 if (needsuspend)
1603 zil_resume(cookie);
1604 return (error);
1605 }
1606
1607
1608 void
1609 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1610 {
1611 ASSERT(dmu_tx_is_syncing(tx));
1612 ASSERT(ds->ds_objset != NULL);
1613 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1614
1615 /*
1616 * in case we had to change ds_fsid_guid when we opened it,
1617 * sync it out now.
1618 */
1619 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1620 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1621
1622 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1623 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1624 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1625 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1626 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1627 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1628 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1629 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1630 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1631 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1632 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1633 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1634 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1635 }
1636
1637 dmu_objset_sync(ds->ds_objset, zio, tx);
1638
1639 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1640 if (ds->ds_feature_activation_needed[f]) {
1641 if (ds->ds_feature_inuse[f])
1642 continue;
1643 dsl_dataset_activate_feature(ds->ds_object, f, tx);
1644 ds->ds_feature_inuse[f] = B_TRUE;
1645 }
1646 }
1647 }
1648
1649 static void
1650 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1651 {
1652 uint64_t count = 0;
1653 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1654 zap_cursor_t zc;
1655 zap_attribute_t za;
1656 nvlist_t *propval = fnvlist_alloc();
1657 nvlist_t *val = fnvlist_alloc();
1658
1659 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1660
1661 /*
1662 * There may be missing entries in ds_next_clones_obj
1663 * due to a bug in a previous version of the code.
1664 * Only trust it if it has the right number of entries.
1665 */
1666 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1667 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1668 &count));
1669 }
1670 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1671 goto fail;
1672 for (zap_cursor_init(&zc, mos,
1673 dsl_dataset_phys(ds)->ds_next_clones_obj);
1674 zap_cursor_retrieve(&zc, &za) == 0;
1675 zap_cursor_advance(&zc)) {
1676 dsl_dataset_t *clone;
1677 char buf[ZFS_MAXNAMELEN];
1678 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1679 za.za_first_integer, FTAG, &clone));
1680 dsl_dir_name(clone->ds_dir, buf);
1681 fnvlist_add_boolean(val, buf);
1682 dsl_dataset_rele(clone, FTAG);
1683 }
1684 zap_cursor_fini(&zc);
1685 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1686 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1687 fail:
1688 nvlist_free(val);
1689 nvlist_free(propval);
1690 }
1691
1692 static void
1693 get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
1694 {
1695 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1696
1697 if (dsl_dataset_has_resume_receive_state(ds)) {
1698 char *str;
1699 void *packed;
1700 uint8_t *compressed;
1701 uint64_t val;
1702 nvlist_t *token_nv = fnvlist_alloc();
1703 size_t packed_size, compressed_size;
1704
1705 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1706 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1707 fnvlist_add_uint64(token_nv, "fromguid", val);
1708 }
1709 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1710 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1711 fnvlist_add_uint64(token_nv, "object", val);
1712 }
1713 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1714 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1715 fnvlist_add_uint64(token_nv, "offset", val);
1716 }
1717 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1718 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1719 fnvlist_add_uint64(token_nv, "bytes", val);
1720 }
1721 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1722 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1723 fnvlist_add_uint64(token_nv, "toguid", val);
1724 }
1725 char buf[256];
1726 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1727 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1728 fnvlist_add_string(token_nv, "toname", buf);
1729 }
1730 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1731 DS_FIELD_RESUME_EMBEDOK) == 0) {
1732 fnvlist_add_boolean(token_nv, "embedok");
1733 }
1734 packed = fnvlist_pack(token_nv, &packed_size);
1735 fnvlist_free(token_nv);
1736 compressed = kmem_alloc(packed_size, KM_SLEEP);
1737
1738 compressed_size = gzip_compress(packed, compressed,
1739 packed_size, packed_size, 6);
1740
1741 zio_cksum_t cksum;
1742 fletcher_4_native(compressed, compressed_size, NULL, &cksum);
1743
1744 str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
1745 for (int i = 0; i < compressed_size; i++) {
1746 (void) sprintf(str + i * 2, "%02x", compressed[i]);
1747 }
1748 str[compressed_size * 2] = '\0';
1749 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
1750 ZFS_SEND_RESUME_TOKEN_VERSION,
1751 (longlong_t)cksum.zc_word[0],
1752 (longlong_t)packed_size, str);
1753 dsl_prop_nvlist_add_string(nv,
1754 ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
1755 kmem_free(packed, packed_size);
1756 kmem_free(str, compressed_size * 2 + 1);
1757 kmem_free(compressed, packed_size);
1758 strfree(propval);
1759 }
1760 }
1761
1762 void
1763 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1764 {
1765 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1766 uint64_t refd, avail, uobjs, aobjs, ratio;
1767
1768 ASSERT(dsl_pool_config_held(dp));
1769
1770 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1771 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1772 dsl_dataset_phys(ds)->ds_compressed_bytes);
1773
1774 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1775 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1776 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1777
1778 if (ds->ds_is_snapshot) {
1779 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1780 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1781 dsl_dataset_phys(ds)->ds_unique_bytes);
1782 get_clones_stat(ds, nv);
1783 } else {
1784 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1785 char buf[MAXNAMELEN];
1786 dsl_dataset_name(ds->ds_prev, buf);
1787 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1788 }
1789
1790 dsl_dir_stats(ds->ds_dir, nv);
1791 }
1792
1793 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1794 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1796
1797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1798 dsl_dataset_phys(ds)->ds_creation_time);
1799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1800 dsl_dataset_phys(ds)->ds_creation_txg);
1801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1802 ds->ds_quota);
1803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1804 ds->ds_reserved);
1805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1806 dsl_dataset_phys(ds)->ds_guid);
1807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1808 dsl_dataset_phys(ds)->ds_unique_bytes);
1809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1810 ds->ds_object);
1811 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1812 ds->ds_userrefs);
1813 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1814 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1815
1816 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1817 uint64_t written, comp, uncomp;
1818 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1819 dsl_dataset_t *prev;
1820
1821 int err = dsl_dataset_hold_obj(dp,
1822 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1823 if (err == 0) {
1824 err = dsl_dataset_space_written(prev, ds, &written,
1825 &comp, &uncomp);
1826 dsl_dataset_rele(prev, FTAG);
1827 if (err == 0) {
1828 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1829 written);
1830 }
1831 }
1832 }
1833
1834 if (!dsl_dataset_is_snapshot(ds)) {
1835 /*
1836 * A failed "newfs" (e.g. full) resumable receive leaves
1837 * the stats set on this dataset. Check here for the prop.
1838 */
1839 get_receive_resume_stats(ds, nv);
1840
1841 /*
1842 * A failed incremental resumable receive leaves the
1843 * stats set on our child named "%recv". Check the child
1844 * for the prop.
1845 */
1846 char recvname[ZFS_MAXNAMELEN];
1847 dsl_dataset_t *recv_ds;
1848 dsl_dataset_name(ds, recvname);
1849 (void) strcat(recvname, "/");
1850 (void) strcat(recvname, recv_clone_name);
1851 if (dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
1852 get_receive_resume_stats(recv_ds, nv);
1853 dsl_dataset_rele(recv_ds, FTAG);
1854 }
1855 }
1856 }
1857
1858 void
1859 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1860 {
1861 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1862 ASSERT(dsl_pool_config_held(dp));
1863
1864 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1865 stat->dds_inconsistent =
1866 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1867 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1868 stat->dds_origin[0] = '\0';
1869 if (ds->ds_is_snapshot) {
1870 stat->dds_is_snapshot = B_TRUE;
1871 stat->dds_num_clones =
1872 dsl_dataset_phys(ds)->ds_num_children - 1;
1873 } else {
1874 stat->dds_is_snapshot = B_FALSE;
1875 stat->dds_num_clones = 0;
1876
1877 if (dsl_dir_is_clone(ds->ds_dir)) {
1878 dsl_dataset_t *ods;
1879
1880 VERIFY0(dsl_dataset_hold_obj(dp,
1881 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1882 FTAG, &ods));
1883 dsl_dataset_name(ods, stat->dds_origin);
1884 dsl_dataset_rele(ods, FTAG);
1885 }
1886 }
1887 }
1888
1889 uint64_t
1890 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1891 {
1892 return (ds->ds_fsid_guid);
1893 }
1894
1895 void
1896 dsl_dataset_space(dsl_dataset_t *ds,
1897 uint64_t *refdbytesp, uint64_t *availbytesp,
1898 uint64_t *usedobjsp, uint64_t *availobjsp)
1899 {
1900 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1901 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1902 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1903 *availbytesp +=
1904 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1905 if (ds->ds_quota != 0) {
1906 /*
1907 * Adjust available bytes according to refquota
1908 */
1909 if (*refdbytesp < ds->ds_quota)
1910 *availbytesp = MIN(*availbytesp,
1911 ds->ds_quota - *refdbytesp);
1912 else
1913 *availbytesp = 0;
1914 }
1915 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1916 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1917 }
1918
1919 boolean_t
1920 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1921 {
1922 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1923
1924 ASSERT(dsl_pool_config_held(dp));
1925 if (snap == NULL)
1926 return (B_FALSE);
1927 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1928 dsl_dataset_phys(snap)->ds_creation_txg) {
1929 objset_t *os, *os_snap;
1930 /*
1931 * It may be that only the ZIL differs, because it was
1932 * reset in the head. Don't count that as being
1933 * modified.
1934 */
1935 if (dmu_objset_from_ds(ds, &os) != 0)
1936 return (B_TRUE);
1937 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1938 return (B_TRUE);
1939 return (bcmp(&os->os_phys->os_meta_dnode,
1940 &os_snap->os_phys->os_meta_dnode,
1941 sizeof (os->os_phys->os_meta_dnode)) != 0);
1942 }
1943 return (B_FALSE);
1944 }
1945
1946 typedef struct dsl_dataset_rename_snapshot_arg {
1947 const char *ddrsa_fsname;
1948 const char *ddrsa_oldsnapname;
1949 const char *ddrsa_newsnapname;
1950 boolean_t ddrsa_recursive;
1951 dmu_tx_t *ddrsa_tx;
1952 } dsl_dataset_rename_snapshot_arg_t;
1953
1954 /* ARGSUSED */
1955 static int
1956 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1957 dsl_dataset_t *hds, void *arg)
1958 {
1959 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1960 int error;
1961 uint64_t val;
1962
1963 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1964 if (error != 0) {
1965 /* ignore nonexistent snapshots */
1966 return (error == ENOENT ? 0 : error);
1967 }
1968
1969 /* new name should not exist */
1970 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1971 if (error == 0)
1972 error = SET_ERROR(EEXIST);
1973 else if (error == ENOENT)
1974 error = 0;
1975
1976 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1977 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1978 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1979 error = SET_ERROR(ENAMETOOLONG);
1980
1981 return (error);
1982 }
1983
1984 static int
1985 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1986 {
1987 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1988 dsl_pool_t *dp = dmu_tx_pool(tx);
1989 dsl_dataset_t *hds;
1990 int error;
1991
1992 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1993 if (error != 0)
1994 return (error);
1995
1996 if (ddrsa->ddrsa_recursive) {
1997 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1998 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1999 DS_FIND_CHILDREN);
2000 } else {
2001 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2002 }
2003 dsl_dataset_rele(hds, FTAG);
2004 return (error);
2005 }
2006
2007 static int
2008 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2009 dsl_dataset_t *hds, void *arg)
2010 {
2011 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2012 dsl_dataset_t *ds;
2013 uint64_t val;
2014 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2015 int error;
2016
2017 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2018 ASSERT(error == 0 || error == ENOENT);
2019 if (error == ENOENT) {
2020 /* ignore nonexistent snapshots */
2021 return (0);
2022 }
2023
2024 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2025
2026 /* log before we change the name */
2027 spa_history_log_internal_ds(ds, "rename", tx,
2028 "-> @%s", ddrsa->ddrsa_newsnapname);
2029
2030 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2031 B_FALSE));
2032 mutex_enter(&ds->ds_lock);
2033 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2034 mutex_exit(&ds->ds_lock);
2035 VERIFY0(zap_add(dp->dp_meta_objset,
2036 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2037 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2038
2039 dsl_dataset_rele(ds, FTAG);
2040 return (0);
2041 }
2042
2043 static void
2044 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2045 {
2046 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2047 dsl_pool_t *dp = dmu_tx_pool(tx);
2048 dsl_dataset_t *hds;
2049
2050 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2051 ddrsa->ddrsa_tx = tx;
2052 if (ddrsa->ddrsa_recursive) {
2053 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2054 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2055 DS_FIND_CHILDREN));
2056 } else {
2057 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2058 }
2059 dsl_dataset_rele(hds, FTAG);
2060 }
2061
2062 int
2063 dsl_dataset_rename_snapshot(const char *fsname,
2064 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2065 {
2066 dsl_dataset_rename_snapshot_arg_t ddrsa;
2067
2068 ddrsa.ddrsa_fsname = fsname;
2069 ddrsa.ddrsa_oldsnapname = oldsnapname;
2070 ddrsa.ddrsa_newsnapname = newsnapname;
2071 ddrsa.ddrsa_recursive = recursive;
2072
2073 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2074 dsl_dataset_rename_snapshot_sync, &ddrsa,
2075 1, ZFS_SPACE_CHECK_RESERVED));
2076 }
2077
2078 /*
2079 * If we're doing an ownership handoff, we need to make sure that there is
2080 * only one long hold on the dataset. We're not allowed to change anything here
2081 * so we don't permanently release the long hold or regular hold here. We want
2082 * to do this only when syncing to avoid the dataset unexpectedly going away
2083 * when we release the long hold.
2084 */
2085 static int
2086 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2087 {
2088 boolean_t held;
2089
2090 if (!dmu_tx_is_syncing(tx))
2091 return (0);
2092
2093 if (owner != NULL) {
2094 VERIFY3P(ds->ds_owner, ==, owner);
2095 dsl_dataset_long_rele(ds, owner);
2096 }
2097
2098 held = dsl_dataset_long_held(ds);
2099
2100 if (owner != NULL)
2101 dsl_dataset_long_hold(ds, owner);
2102
2103 if (held)
2104 return (SET_ERROR(EBUSY));
2105
2106 return (0);
2107 }
2108
2109 typedef struct dsl_dataset_rollback_arg {
2110 const char *ddra_fsname;
2111 void *ddra_owner;
2112 nvlist_t *ddra_result;
2113 } dsl_dataset_rollback_arg_t;
2114
2115 static int
2116 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2117 {
2118 dsl_dataset_rollback_arg_t *ddra = arg;
2119 dsl_pool_t *dp = dmu_tx_pool(tx);
2120 dsl_dataset_t *ds;
2121 int64_t unused_refres_delta;
2122 int error;
2123
2124 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2125 if (error != 0)
2126 return (error);
2127
2128 /* must not be a snapshot */
2129 if (ds->ds_is_snapshot) {
2130 dsl_dataset_rele(ds, FTAG);
2131 return (SET_ERROR(EINVAL));
2132 }
2133
2134 /* must have a most recent snapshot */
2135 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2136 dsl_dataset_rele(ds, FTAG);
2137 return (SET_ERROR(EINVAL));
2138 }
2139
2140 /* must not have any bookmarks after the most recent snapshot */
2141 nvlist_t *proprequest = fnvlist_alloc();
2142 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2143 nvlist_t *bookmarks = fnvlist_alloc();
2144 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2145 fnvlist_free(proprequest);
2146 if (error != 0)
2147 return (error);
2148 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2149 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2150 nvlist_t *valuenv =
2151 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2152 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2153 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2154 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2155 fnvlist_free(bookmarks);
2156 dsl_dataset_rele(ds, FTAG);
2157 return (SET_ERROR(EEXIST));
2158 }
2159 }
2160 fnvlist_free(bookmarks);
2161
2162 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2163 if (error != 0) {
2164 dsl_dataset_rele(ds, FTAG);
2165 return (error);
2166 }
2167
2168 /*
2169 * Check if the snap we are rolling back to uses more than
2170 * the refquota.
2171 */
2172 if (ds->ds_quota != 0 &&
2173 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2174 dsl_dataset_rele(ds, FTAG);
2175 return (SET_ERROR(EDQUOT));
2176 }
2177
2178 /*
2179 * When we do the clone swap, we will temporarily use more space
2180 * due to the refreservation (the head will no longer have any
2181 * unique space, so the entire amount of the refreservation will need
2182 * to be free). We will immediately destroy the clone, freeing
2183 * this space, but the freeing happens over many txg's.
2184 */
2185 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2186 dsl_dataset_phys(ds)->ds_unique_bytes);
2187
2188 if (unused_refres_delta > 0 &&
2189 unused_refres_delta >
2190 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2191 dsl_dataset_rele(ds, FTAG);
2192 return (SET_ERROR(ENOSPC));
2193 }
2194
2195 dsl_dataset_rele(ds, FTAG);
2196 return (0);
2197 }
2198
2199 static void
2200 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2201 {
2202 dsl_dataset_rollback_arg_t *ddra = arg;
2203 dsl_pool_t *dp = dmu_tx_pool(tx);
2204 dsl_dataset_t *ds, *clone;
2205 uint64_t cloneobj;
2206 char namebuf[ZFS_MAXNAMELEN];
2207
2208 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2209
2210 dsl_dataset_name(ds->ds_prev, namebuf);
2211 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2212
2213 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2214 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2215
2216 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2217
2218 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2219 dsl_dataset_zero_zil(ds, tx);
2220
2221 dsl_destroy_head_sync_impl(clone, tx);
2222
2223 dsl_dataset_rele(clone, FTAG);
2224 dsl_dataset_rele(ds, FTAG);
2225 }
2226
2227 /*
2228 * Rolls back the given filesystem or volume to the most recent snapshot.
2229 * The name of the most recent snapshot will be returned under key "target"
2230 * in the result nvlist.
2231 *
2232 * If owner != NULL:
2233 * - The existing dataset MUST be owned by the specified owner at entry
2234 * - Upon return, dataset will still be held by the same owner, whether we
2235 * succeed or not.
2236 *
2237 * This mode is required any time the existing filesystem is mounted. See
2238 * notes above zfs_suspend_fs() for further details.
2239 */
2240 int
2241 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2242 {
2243 dsl_dataset_rollback_arg_t ddra;
2244
2245 ddra.ddra_fsname = fsname;
2246 ddra.ddra_owner = owner;
2247 ddra.ddra_result = result;
2248
2249 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2250 dsl_dataset_rollback_sync, &ddra,
2251 1, ZFS_SPACE_CHECK_RESERVED));
2252 }
2253
2254 struct promotenode {
2255 list_node_t link;
2256 dsl_dataset_t *ds;
2257 };
2258
2259 typedef struct dsl_dataset_promote_arg {
2260 const char *ddpa_clonename;
2261 dsl_dataset_t *ddpa_clone;
2262 list_t shared_snaps, origin_snaps, clone_snaps;
2263 dsl_dataset_t *origin_origin; /* origin of the origin */
2264 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2265 char *err_ds;
2266 cred_t *cr;
2267 } dsl_dataset_promote_arg_t;
2268
2269 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2270 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2271 void *tag);
2272 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2273
2274 static int
2275 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2276 {
2277 dsl_dataset_promote_arg_t *ddpa = arg;
2278 dsl_pool_t *dp = dmu_tx_pool(tx);
2279 dsl_dataset_t *hds;
2280 struct promotenode *snap;
2281 dsl_dataset_t *origin_ds;
2282 int err;
2283 uint64_t unused;
2284 uint64_t ss_mv_cnt;
2285 size_t max_snap_len;
2286
2287 err = promote_hold(ddpa, dp, FTAG);
2288 if (err != 0)
2289 return (err);
2290
2291 hds = ddpa->ddpa_clone;
2292 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2293
2294 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2295 promote_rele(ddpa, FTAG);
2296 return (SET_ERROR(EXDEV));
2297 }
2298
2299 /*
2300 * Compute and check the amount of space to transfer. Since this is
2301 * so expensive, don't do the preliminary check.
2302 */
2303 if (!dmu_tx_is_syncing(tx)) {
2304 promote_rele(ddpa, FTAG);
2305 return (0);
2306 }
2307
2308 snap = list_head(&ddpa->shared_snaps);
2309 origin_ds = snap->ds;
2310
2311 /* compute origin's new unique space */
2312 snap = list_tail(&ddpa->clone_snaps);
2313 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2314 origin_ds->ds_object);
2315 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2316 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2317 &ddpa->unique, &unused, &unused);
2318
2319 /*
2320 * Walk the snapshots that we are moving
2321 *
2322 * Compute space to transfer. Consider the incremental changes
2323 * to used by each snapshot:
2324 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2325 * So each snapshot gave birth to:
2326 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2327 * So a sequence would look like:
2328 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2329 * Which simplifies to:
2330 * uN + kN + kN-1 + ... + k1 + k0
2331 * Note however, if we stop before we reach the ORIGIN we get:
2332 * uN + kN + kN-1 + ... + kM - uM-1
2333 */
2334 ss_mv_cnt = 0;
2335 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2336 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2337 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2338 for (snap = list_head(&ddpa->shared_snaps); snap;
2339 snap = list_next(&ddpa->shared_snaps, snap)) {
2340 uint64_t val, dlused, dlcomp, dluncomp;
2341 dsl_dataset_t *ds = snap->ds;
2342
2343 ss_mv_cnt++;
2344
2345 /*
2346 * If there are long holds, we won't be able to evict
2347 * the objset.
2348 */
2349 if (dsl_dataset_long_held(ds)) {
2350 err = SET_ERROR(EBUSY);
2351 goto out;
2352 }
2353
2354 /* Check that the snapshot name does not conflict */
2355 VERIFY0(dsl_dataset_get_snapname(ds));
2356 if (strlen(ds->ds_snapname) >= max_snap_len) {
2357 err = SET_ERROR(ENAMETOOLONG);
2358 goto out;
2359 }
2360 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2361 if (err == 0) {
2362 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2363 err = SET_ERROR(EEXIST);
2364 goto out;
2365 }
2366 if (err != ENOENT)
2367 goto out;
2368
2369 /* The very first snapshot does not have a deadlist */
2370 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2371 continue;
2372
2373 dsl_deadlist_space(&ds->ds_deadlist,
2374 &dlused, &dlcomp, &dluncomp);
2375 ddpa->used += dlused;
2376 ddpa->comp += dlcomp;
2377 ddpa->uncomp += dluncomp;
2378 }
2379
2380 /*
2381 * If we are a clone of a clone then we never reached ORIGIN,
2382 * so we need to subtract out the clone origin's used space.
2383 */
2384 if (ddpa->origin_origin) {
2385 ddpa->used -=
2386 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2387 ddpa->comp -=
2388 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2389 ddpa->uncomp -=
2390 dsl_dataset_phys(ddpa->origin_origin)->
2391 ds_uncompressed_bytes;
2392 }
2393
2394 /* Check that there is enough space and limit headroom here */
2395 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2396 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2397 if (err != 0)
2398 goto out;
2399
2400 /*
2401 * Compute the amounts of space that will be used by snapshots
2402 * after the promotion (for both origin and clone). For each,
2403 * it is the amount of space that will be on all of their
2404 * deadlists (that was not born before their new origin).
2405 */
2406 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2407 uint64_t space;
2408
2409 /*
2410 * Note, typically this will not be a clone of a clone,
2411 * so dd_origin_txg will be < TXG_INITIAL, so
2412 * these snaplist_space() -> dsl_deadlist_space_range()
2413 * calls will be fast because they do not have to
2414 * iterate over all bps.
2415 */
2416 snap = list_head(&ddpa->origin_snaps);
2417 err = snaplist_space(&ddpa->shared_snaps,
2418 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2419 if (err != 0)
2420 goto out;
2421
2422 err = snaplist_space(&ddpa->clone_snaps,
2423 snap->ds->ds_dir->dd_origin_txg, &space);
2424 if (err != 0)
2425 goto out;
2426 ddpa->cloneusedsnap += space;
2427 }
2428 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2429 DD_FLAG_USED_BREAKDOWN) {
2430 err = snaplist_space(&ddpa->origin_snaps,
2431 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2432 &ddpa->originusedsnap);
2433 if (err != 0)
2434 goto out;
2435 }
2436
2437 out:
2438 promote_rele(ddpa, FTAG);
2439 return (err);
2440 }
2441
2442 static void
2443 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2444 {
2445 dsl_dataset_promote_arg_t *ddpa = arg;
2446 dsl_pool_t *dp = dmu_tx_pool(tx);
2447 dsl_dataset_t *hds;
2448 struct promotenode *snap;
2449 dsl_dataset_t *origin_ds;
2450 dsl_dataset_t *origin_head;
2451 dsl_dir_t *dd;
2452 dsl_dir_t *odd = NULL;
2453 uint64_t oldnext_obj;
2454 int64_t delta;
2455
2456 VERIFY0(promote_hold(ddpa, dp, FTAG));
2457 hds = ddpa->ddpa_clone;
2458
2459 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2460
2461 snap = list_head(&ddpa->shared_snaps);
2462 origin_ds = snap->ds;
2463 dd = hds->ds_dir;
2464
2465 snap = list_head(&ddpa->origin_snaps);
2466 origin_head = snap->ds;
2467
2468 /*
2469 * We need to explicitly open odd, since origin_ds's dd will be
2470 * changing.
2471 */
2472 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2473 NULL, FTAG, &odd));
2474
2475 /* change origin's next snap */
2476 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2477 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2478 snap = list_tail(&ddpa->clone_snaps);
2479 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2480 origin_ds->ds_object);
2481 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2482
2483 /* change the origin's next clone */
2484 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2485 dsl_dataset_remove_from_next_clones(origin_ds,
2486 snap->ds->ds_object, tx);
2487 VERIFY0(zap_add_int(dp->dp_meta_objset,
2488 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2489 oldnext_obj, tx));
2490 }
2491
2492 /* change origin */
2493 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2494 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2495 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2496 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2497 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2498 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2499 origin_head->ds_dir->dd_origin_txg =
2500 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2501
2502 /* change dd_clone entries */
2503 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2504 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2505 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2506 VERIFY0(zap_add_int(dp->dp_meta_objset,
2507 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2508 hds->ds_object, tx));
2509
2510 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2511 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2512 origin_head->ds_object, tx));
2513 if (dsl_dir_phys(dd)->dd_clones == 0) {
2514 dsl_dir_phys(dd)->dd_clones =
2515 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2516 DMU_OT_NONE, 0, tx);
2517 }
2518 VERIFY0(zap_add_int(dp->dp_meta_objset,
2519 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2520 }
2521
2522 /* move snapshots to this dir */
2523 for (snap = list_head(&ddpa->shared_snaps); snap;
2524 snap = list_next(&ddpa->shared_snaps, snap)) {
2525 dsl_dataset_t *ds = snap->ds;
2526
2527 /*
2528 * Property callbacks are registered to a particular
2529 * dsl_dir. Since ours is changing, evict the objset
2530 * so that they will be unregistered from the old dsl_dir.
2531 */
2532 if (ds->ds_objset) {
2533 dmu_objset_evict(ds->ds_objset);
2534 ds->ds_objset = NULL;
2535 }
2536
2537 /* move snap name entry */
2538 VERIFY0(dsl_dataset_get_snapname(ds));
2539 VERIFY0(dsl_dataset_snap_remove(origin_head,
2540 ds->ds_snapname, tx, B_TRUE));
2541 VERIFY0(zap_add(dp->dp_meta_objset,
2542 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2543 8, 1, &ds->ds_object, tx));
2544 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2545 DD_FIELD_SNAPSHOT_COUNT, tx);
2546
2547 /* change containing dsl_dir */
2548 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2549 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2550 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2551 ASSERT3P(ds->ds_dir, ==, odd);
2552 dsl_dir_rele(ds->ds_dir, ds);
2553 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2554 NULL, ds, &ds->ds_dir));
2555
2556 /* move any clone references */
2557 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2558 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2559 zap_cursor_t zc;
2560 zap_attribute_t za;
2561
2562 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2563 dsl_dataset_phys(ds)->ds_next_clones_obj);
2564 zap_cursor_retrieve(&zc, &za) == 0;
2565 zap_cursor_advance(&zc)) {
2566 dsl_dataset_t *cnds;
2567 uint64_t o;
2568
2569 if (za.za_first_integer == oldnext_obj) {
2570 /*
2571 * We've already moved the
2572 * origin's reference.
2573 */
2574 continue;
2575 }
2576
2577 VERIFY0(dsl_dataset_hold_obj(dp,
2578 za.za_first_integer, FTAG, &cnds));
2579 o = dsl_dir_phys(cnds->ds_dir)->
2580 dd_head_dataset_obj;
2581
2582 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2583 dsl_dir_phys(odd)->dd_clones, o, tx));
2584 VERIFY0(zap_add_int(dp->dp_meta_objset,
2585 dsl_dir_phys(dd)->dd_clones, o, tx));
2586 dsl_dataset_rele(cnds, FTAG);
2587 }
2588 zap_cursor_fini(&zc);
2589 }
2590
2591 ASSERT(!dsl_prop_hascb(ds));
2592 }
2593
2594 /*
2595 * Change space accounting.
2596 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2597 * both be valid, or both be 0 (resulting in delta == 0). This
2598 * is true for each of {clone,origin} independently.
2599 */
2600
2601 delta = ddpa->cloneusedsnap -
2602 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2603 ASSERT3S(delta, >=, 0);
2604 ASSERT3U(ddpa->used, >=, delta);
2605 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2606 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2607 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2608
2609 delta = ddpa->originusedsnap -
2610 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2611 ASSERT3S(delta, <=, 0);
2612 ASSERT3U(ddpa->used, >=, -delta);
2613 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2614 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2615 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2616
2617 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2618
2619 /* log history record */
2620 spa_history_log_internal_ds(hds, "promote", tx, "");
2621
2622 dsl_dir_rele(odd, FTAG);
2623 promote_rele(ddpa, FTAG);
2624 }
2625
2626 /*
2627 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2628 * (exclusive) and last_obj (inclusive). The list will be in reverse
2629 * order (last_obj will be the list_head()). If first_obj == 0, do all
2630 * snapshots back to this dataset's origin.
2631 */
2632 static int
2633 snaplist_make(dsl_pool_t *dp,
2634 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2635 {
2636 uint64_t obj = last_obj;
2637
2638 list_create(l, sizeof (struct promotenode),
2639 offsetof(struct promotenode, link));
2640
2641 while (obj != first_obj) {
2642 dsl_dataset_t *ds;
2643 struct promotenode *snap;
2644 int err;
2645
2646 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2647 ASSERT(err != ENOENT);
2648 if (err != 0)
2649 return (err);
2650
2651 if (first_obj == 0)
2652 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2653
2654 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2655 snap->ds = ds;
2656 list_insert_tail(l, snap);
2657 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2658 }
2659
2660 return (0);
2661 }
2662
2663 static int
2664 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2665 {
2666 struct promotenode *snap;
2667
2668 *spacep = 0;
2669 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2670 uint64_t used, comp, uncomp;
2671 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2672 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2673 *spacep += used;
2674 }
2675 return (0);
2676 }
2677
2678 static void
2679 snaplist_destroy(list_t *l, void *tag)
2680 {
2681 struct promotenode *snap;
2682
2683 if (l == NULL || !list_link_active(&l->list_head))
2684 return;
2685
2686 while ((snap = list_tail(l)) != NULL) {
2687 list_remove(l, snap);
2688 dsl_dataset_rele(snap->ds, tag);
2689 kmem_free(snap, sizeof (*snap));
2690 }
2691 list_destroy(l);
2692 }
2693
2694 static int
2695 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2696 {
2697 int error;
2698 dsl_dir_t *dd;
2699 struct promotenode *snap;
2700
2701 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2702 &ddpa->ddpa_clone);
2703 if (error != 0)
2704 return (error);
2705 dd = ddpa->ddpa_clone->ds_dir;
2706
2707 if (ddpa->ddpa_clone->ds_is_snapshot ||
2708 !dsl_dir_is_clone(dd)) {
2709 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2710 return (SET_ERROR(EINVAL));
2711 }
2712
2713 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2714 &ddpa->shared_snaps, tag);
2715 if (error != 0)
2716 goto out;
2717
2718 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2719 &ddpa->clone_snaps, tag);
2720 if (error != 0)
2721 goto out;
2722
2723 snap = list_head(&ddpa->shared_snaps);
2724 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2725 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2726 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2727 &ddpa->origin_snaps, tag);
2728 if (error != 0)
2729 goto out;
2730
2731 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2732 error = dsl_dataset_hold_obj(dp,
2733 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2734 tag, &ddpa->origin_origin);
2735 if (error != 0)
2736 goto out;
2737 }
2738 out:
2739 if (error != 0)
2740 promote_rele(ddpa, tag);
2741 return (error);
2742 }
2743
2744 static void
2745 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2746 {
2747 snaplist_destroy(&ddpa->shared_snaps, tag);
2748 snaplist_destroy(&ddpa->clone_snaps, tag);
2749 snaplist_destroy(&ddpa->origin_snaps, tag);
2750 if (ddpa->origin_origin != NULL)
2751 dsl_dataset_rele(ddpa->origin_origin, tag);
2752 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2753 }
2754
2755 /*
2756 * Promote a clone.
2757 *
2758 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2759 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2760 */
2761 int
2762 dsl_dataset_promote(const char *name, char *conflsnap)
2763 {
2764 dsl_dataset_promote_arg_t ddpa = { 0 };
2765 uint64_t numsnaps;
2766 int error;
2767 objset_t *os;
2768
2769 /*
2770 * We will modify space proportional to the number of
2771 * snapshots. Compute numsnaps.
2772 */
2773 error = dmu_objset_hold(name, FTAG, &os);
2774 if (error != 0)
2775 return (error);
2776 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2777 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2778 &numsnaps);
2779 dmu_objset_rele(os, FTAG);
2780 if (error != 0)
2781 return (error);
2782
2783 ddpa.ddpa_clonename = name;
2784 ddpa.err_ds = conflsnap;
2785 ddpa.cr = CRED();
2786
2787 return (dsl_sync_task(name, dsl_dataset_promote_check,
2788 dsl_dataset_promote_sync, &ddpa,
2789 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2790 }
2791
2792 int
2793 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2794 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2795 {
2796 int64_t unused_refres_delta;
2797
2798 /* they should both be heads */
2799 if (clone->ds_is_snapshot ||
2800 origin_head->ds_is_snapshot)
2801 return (SET_ERROR(EINVAL));
2802
2803 /* if we are not forcing, the branch point should be just before them */
2804 if (!force && clone->ds_prev != origin_head->ds_prev)
2805 return (SET_ERROR(EINVAL));
2806
2807 /* clone should be the clone (unless they are unrelated) */
2808 if (clone->ds_prev != NULL &&
2809 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2810 origin_head->ds_dir != clone->ds_prev->ds_dir)
2811 return (SET_ERROR(EINVAL));
2812
2813 /* the clone should be a child of the origin */
2814 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2815 return (SET_ERROR(EINVAL));
2816
2817 /* origin_head shouldn't be modified unless 'force' */
2818 if (!force &&
2819 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2820 return (SET_ERROR(ETXTBSY));
2821
2822 /* origin_head should have no long holds (e.g. is not mounted) */
2823 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2824 return (SET_ERROR(EBUSY));
2825
2826 /* check amount of any unconsumed refreservation */
2827 unused_refres_delta =
2828 (int64_t)MIN(origin_head->ds_reserved,
2829 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2830 (int64_t)MIN(origin_head->ds_reserved,
2831 dsl_dataset_phys(clone)->ds_unique_bytes);
2832
2833 if (unused_refres_delta > 0 &&
2834 unused_refres_delta >
2835 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2836 return (SET_ERROR(ENOSPC));
2837
2838 /* clone can't be over the head's refquota */
2839 if (origin_head->ds_quota != 0 &&
2840 dsl_dataset_phys(clone)->ds_referenced_bytes >
2841 origin_head->ds_quota)
2842 return (SET_ERROR(EDQUOT));
2843
2844 return (0);
2845 }
2846
2847 void
2848 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2849 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2850 {
2851 dsl_pool_t *dp = dmu_tx_pool(tx);
2852 int64_t unused_refres_delta;
2853
2854 ASSERT(clone->ds_reserved == 0);
2855 ASSERT(origin_head->ds_quota == 0 ||
2856 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2857 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2858
2859 /*
2860 * Swap per-dataset feature flags.
2861 */
2862 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2863 if (!(spa_feature_table[f].fi_flags &
2864 ZFEATURE_FLAG_PER_DATASET)) {
2865 ASSERT(!clone->ds_feature_inuse[f]);
2866 ASSERT(!origin_head->ds_feature_inuse[f]);
2867 continue;
2868 }
2869
2870 boolean_t clone_inuse = clone->ds_feature_inuse[f];
2871 boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
2872
2873 if (clone_inuse) {
2874 dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
2875 clone->ds_feature_inuse[f] = B_FALSE;
2876 }
2877 if (origin_head_inuse) {
2878 dsl_dataset_deactivate_feature(origin_head->ds_object,
2879 f, tx);
2880 origin_head->ds_feature_inuse[f] = B_FALSE;
2881 }
2882 if (clone_inuse) {
2883 dsl_dataset_activate_feature(origin_head->ds_object,
2884 f, tx);
2885 origin_head->ds_feature_inuse[f] = B_TRUE;
2886 }
2887 if (origin_head_inuse) {
2888 dsl_dataset_activate_feature(clone->ds_object, f, tx);
2889 clone->ds_feature_inuse[f] = B_TRUE;
2890 }
2891 }
2892
2893 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2894 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2895
2896 if (clone->ds_objset != NULL) {
2897 dmu_objset_evict(clone->ds_objset);
2898 clone->ds_objset = NULL;
2899 }
2900
2901 if (origin_head->ds_objset != NULL) {
2902 dmu_objset_evict(origin_head->ds_objset);
2903 origin_head->ds_objset = NULL;
2904 }
2905
2906 unused_refres_delta =
2907 (int64_t)MIN(origin_head->ds_reserved,
2908 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2909 (int64_t)MIN(origin_head->ds_reserved,
2910 dsl_dataset_phys(clone)->ds_unique_bytes);
2911
2912 /*
2913 * Reset origin's unique bytes, if it exists.
2914 */
2915 if (clone->ds_prev) {
2916 dsl_dataset_t *origin = clone->ds_prev;
2917 uint64_t comp, uncomp;
2918
2919 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2920 dsl_deadlist_space_range(&clone->ds_deadlist,
2921 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2922 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2923 }
2924
2925 /* swap blkptrs */
2926 {
2927 blkptr_t tmp;
2928 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2929 dsl_dataset_phys(origin_head)->ds_bp =
2930 dsl_dataset_phys(clone)->ds_bp;
2931 dsl_dataset_phys(clone)->ds_bp = tmp;
2932 }
2933
2934 /* set dd_*_bytes */
2935 {
2936 int64_t dused, dcomp, duncomp;
2937 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2938 uint64_t odl_used, odl_comp, odl_uncomp;
2939
2940 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2941 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2942
2943 dsl_deadlist_space(&clone->ds_deadlist,
2944 &cdl_used, &cdl_comp, &cdl_uncomp);
2945 dsl_deadlist_space(&origin_head->ds_deadlist,
2946 &odl_used, &odl_comp, &odl_uncomp);
2947
2948 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2949 cdl_used -
2950 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2951 odl_used);
2952 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2953 cdl_comp -
2954 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2955 odl_comp);
2956 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2957 cdl_uncomp -
2958 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2959 odl_uncomp);
2960
2961 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2962 dused, dcomp, duncomp, tx);
2963 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2964 -dused, -dcomp, -duncomp, tx);
2965
2966 /*
2967 * The difference in the space used by snapshots is the
2968 * difference in snapshot space due to the head's
2969 * deadlist (since that's the only thing that's
2970 * changing that affects the snapused).
2971 */
2972 dsl_deadlist_space_range(&clone->ds_deadlist,
2973 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2974 &cdl_used, &cdl_comp, &cdl_uncomp);
2975 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2976 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2977 &odl_used, &odl_comp, &odl_uncomp);
2978 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2979 DD_USED_HEAD, DD_USED_SNAP, tx);
2980 }
2981
2982 /* swap ds_*_bytes */
2983 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2984 dsl_dataset_phys(clone)->ds_referenced_bytes);
2985 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2986 dsl_dataset_phys(clone)->ds_compressed_bytes);
2987 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2988 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2989 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2990 dsl_dataset_phys(clone)->ds_unique_bytes);
2991
2992 /* apply any parent delta for change in unconsumed refreservation */
2993 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2994 unused_refres_delta, 0, 0, tx);
2995
2996 /*
2997 * Swap deadlists.
2998 */
2999 dsl_deadlist_close(&clone->ds_deadlist);
3000 dsl_deadlist_close(&origin_head->ds_deadlist);
3001 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3002 dsl_dataset_phys(clone)->ds_deadlist_obj);
3003 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3004 dsl_dataset_phys(clone)->ds_deadlist_obj);
3005 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3006 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3007
3008 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3009
3010 spa_history_log_internal_ds(clone, "clone swap", tx,
3011 "parent=%s", origin_head->ds_dir->dd_myname);
3012 }
3013
3014 /*
3015 * Given a pool name and a dataset object number in that pool,
3016 * return the name of that dataset.
3017 */
3018 int
3019 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3020 {
3021 dsl_pool_t *dp;
3022 dsl_dataset_t *ds;
3023 int error;
3024
3025 error = dsl_pool_hold(pname, FTAG, &dp);
3026 if (error != 0)
3027 return (error);
3028
3029 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3030 if (error == 0) {
3031 dsl_dataset_name(ds, buf);
3032 dsl_dataset_rele(ds, FTAG);
3033 }
3034 dsl_pool_rele(dp, FTAG);
3035
3036 return (error);
3037 }
3038
3039 int
3040 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3041 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3042 {
3043 int error = 0;
3044
3045 ASSERT3S(asize, >, 0);
3046
3047 /*
3048 * *ref_rsrv is the portion of asize that will come from any
3049 * unconsumed refreservation space.
3050 */
3051 *ref_rsrv = 0;
3052
3053 mutex_enter(&ds->ds_lock);
3054 /*
3055 * Make a space adjustment for reserved bytes.
3056 */
3057 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3058 ASSERT3U(*used, >=,
3059 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3060 *used -=
3061 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3062 *ref_rsrv =
3063 asize - MIN(asize, parent_delta(ds, asize + inflight));
3064 }
3065
3066 if (!check_quota || ds->ds_quota == 0) {
3067 mutex_exit(&ds->ds_lock);
3068 return (0);
3069 }
3070 /*
3071 * If they are requesting more space, and our current estimate
3072 * is over quota, they get to try again unless the actual
3073 * on-disk is over quota and there are no pending changes (which
3074 * may free up space for us).
3075 */
3076 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3077 ds->ds_quota) {
3078 if (inflight > 0 ||
3079 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3080 error = SET_ERROR(ERESTART);
3081 else
3082 error = SET_ERROR(EDQUOT);
3083 }
3084 mutex_exit(&ds->ds_lock);
3085
3086 return (error);
3087 }
3088
3089 typedef struct dsl_dataset_set_qr_arg {
3090 const char *ddsqra_name;
3091 zprop_source_t ddsqra_source;
3092 uint64_t ddsqra_value;
3093 } dsl_dataset_set_qr_arg_t;
3094
3095
3096 /* ARGSUSED */
3097 static int
3098 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3099 {
3100 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3101 dsl_pool_t *dp = dmu_tx_pool(tx);
3102 dsl_dataset_t *ds;
3103 int error;
3104 uint64_t newval;
3105
3106 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3107 return (SET_ERROR(ENOTSUP));
3108
3109 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3110 if (error != 0)
3111 return (error);
3112
3113 if (ds->ds_is_snapshot) {
3114 dsl_dataset_rele(ds, FTAG);
3115 return (SET_ERROR(EINVAL));
3116 }
3117
3118 error = dsl_prop_predict(ds->ds_dir,
3119 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3120 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3121 if (error != 0) {
3122 dsl_dataset_rele(ds, FTAG);
3123 return (error);
3124 }
3125
3126 if (newval == 0) {
3127 dsl_dataset_rele(ds, FTAG);
3128 return (0);
3129 }
3130
3131 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3132 newval < ds->ds_reserved) {
3133 dsl_dataset_rele(ds, FTAG);
3134 return (SET_ERROR(ENOSPC));
3135 }
3136
3137 dsl_dataset_rele(ds, FTAG);
3138 return (0);
3139 }
3140
3141 static void
3142 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3143 {
3144 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3145 dsl_pool_t *dp = dmu_tx_pool(tx);
3146 dsl_dataset_t *ds;
3147 uint64_t newval;
3148
3149 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3150
3151 dsl_prop_set_sync_impl(ds,
3152 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3153 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3154 &ddsqra->ddsqra_value, tx);
3155
3156 VERIFY0(dsl_prop_get_int_ds(ds,
3157 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3158
3159 if (ds->ds_quota != newval) {
3160 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3161 ds->ds_quota = newval;
3162 }
3163 dsl_dataset_rele(ds, FTAG);
3164 }
3165
3166 int
3167 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3168 uint64_t refquota)
3169 {
3170 dsl_dataset_set_qr_arg_t ddsqra;
3171
3172 ddsqra.ddsqra_name = dsname;
3173 ddsqra.ddsqra_source = source;
3174 ddsqra.ddsqra_value = refquota;
3175
3176 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3177 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3178 }
3179
3180 static int
3181 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3182 {
3183 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3184 dsl_pool_t *dp = dmu_tx_pool(tx);
3185 dsl_dataset_t *ds;
3186 int error;
3187 uint64_t newval, unique;
3188
3189 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3190 return (SET_ERROR(ENOTSUP));
3191
3192 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3193 if (error != 0)
3194 return (error);
3195
3196 if (ds->ds_is_snapshot) {
3197 dsl_dataset_rele(ds, FTAG);
3198 return (SET_ERROR(EINVAL));
3199 }
3200
3201 error = dsl_prop_predict(ds->ds_dir,
3202 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3203 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3204 if (error != 0) {
3205 dsl_dataset_rele(ds, FTAG);
3206 return (error);
3207 }
3208
3209 /*
3210 * If we are doing the preliminary check in open context, the
3211 * space estimates may be inaccurate.
3212 */
3213 if (!dmu_tx_is_syncing(tx)) {
3214 dsl_dataset_rele(ds, FTAG);
3215 return (0);
3216 }
3217
3218 mutex_enter(&ds->ds_lock);
3219 if (!DS_UNIQUE_IS_ACCURATE(ds))
3220 dsl_dataset_recalc_head_uniq(ds);
3221 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3222 mutex_exit(&ds->ds_lock);
3223
3224 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3225 uint64_t delta = MAX(unique, newval) -
3226 MAX(unique, ds->ds_reserved);
3227
3228 if (delta >
3229 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3230 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3231 dsl_dataset_rele(ds, FTAG);
3232 return (SET_ERROR(ENOSPC));
3233 }
3234 }
3235
3236 dsl_dataset_rele(ds, FTAG);
3237 return (0);
3238 }
3239
3240 void
3241 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3242 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3243 {
3244 uint64_t newval;
3245 uint64_t unique;
3246 int64_t delta;
3247
3248 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3249 source, sizeof (value), 1, &value, tx);
3250
3251 VERIFY0(dsl_prop_get_int_ds(ds,
3252 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3253
3254 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3255 mutex_enter(&ds->ds_dir->dd_lock);
3256 mutex_enter(&ds->ds_lock);
3257 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3258 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3259 delta = MAX(0, (int64_t)(newval - unique)) -
3260 MAX(0, (int64_t)(ds->ds_reserved - unique));
3261 ds->ds_reserved = newval;
3262 mutex_exit(&ds->ds_lock);
3263
3264 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3265 mutex_exit(&ds->ds_dir->dd_lock);
3266 }
3267
3268 static void
3269 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3270 {
3271 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3272 dsl_pool_t *dp = dmu_tx_pool(tx);
3273 dsl_dataset_t *ds;
3274
3275 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3276 dsl_dataset_set_refreservation_sync_impl(ds,
3277 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3278 dsl_dataset_rele(ds, FTAG);
3279 }
3280
3281 int
3282 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3283 uint64_t refreservation)
3284 {
3285 dsl_dataset_set_qr_arg_t ddsqra;
3286
3287 ddsqra.ddsqra_name = dsname;
3288 ddsqra.ddsqra_source = source;
3289 ddsqra.ddsqra_value = refreservation;
3290
3291 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3292 dsl_dataset_set_refreservation_sync, &ddsqra,
3293 0, ZFS_SPACE_CHECK_NONE));
3294 }
3295
3296 /*
3297 * Return (in *usedp) the amount of space written in new that is not
3298 * present in oldsnap. New may be a snapshot or the head. Old must be
3299 * a snapshot before new, in new's filesystem (or its origin). If not then
3300 * fail and return EINVAL.
3301 *
3302 * The written space is calculated by considering two components: First, we
3303 * ignore any freed space, and calculate the written as new's used space
3304 * minus old's used space. Next, we add in the amount of space that was freed
3305 * between the two snapshots, thus reducing new's used space relative to old's.
3306 * Specifically, this is the space that was born before old->ds_creation_txg,
3307 * and freed before new (ie. on new's deadlist or a previous deadlist).
3308 *
3309 * space freed [---------------------]
3310 * snapshots ---O-------O--------O-------O------
3311 * oldsnap new
3312 */
3313 int
3314 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3315 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3316 {
3317 int err = 0;
3318 uint64_t snapobj;
3319 dsl_pool_t *dp = new->ds_dir->dd_pool;
3320
3321 ASSERT(dsl_pool_config_held(dp));
3322
3323 *usedp = 0;
3324 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3325 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3326
3327 *compp = 0;
3328 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3329 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3330
3331 *uncompp = 0;
3332 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3333 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3334
3335 snapobj = new->ds_object;
3336 while (snapobj != oldsnap->ds_object) {
3337 dsl_dataset_t *snap;
3338 uint64_t used, comp, uncomp;
3339
3340 if (snapobj == new->ds_object) {
3341 snap = new;
3342 } else {
3343 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3344 if (err != 0)
3345 break;
3346 }
3347
3348 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3349 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3350 /*
3351 * The blocks in the deadlist can not be born after
3352 * ds_prev_snap_txg, so get the whole deadlist space,
3353 * which is more efficient (especially for old-format
3354 * deadlists). Unfortunately the deadlist code
3355 * doesn't have enough information to make this
3356 * optimization itself.
3357 */
3358 dsl_deadlist_space(&snap->ds_deadlist,
3359 &used, &comp, &uncomp);
3360 } else {
3361 dsl_deadlist_space_range(&snap->ds_deadlist,
3362 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3363 &used, &comp, &uncomp);
3364 }
3365 *usedp += used;
3366 *compp += comp;
3367 *uncompp += uncomp;
3368
3369 /*
3370 * If we get to the beginning of the chain of snapshots
3371 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3372 * was not a snapshot of/before new.
3373 */
3374 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3375 if (snap != new)
3376 dsl_dataset_rele(snap, FTAG);
3377 if (snapobj == 0) {
3378 err = SET_ERROR(EINVAL);
3379 break;
3380 }
3381
3382 }
3383 return (err);
3384 }
3385
3386 /*
3387 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3388 * lastsnap, and all snapshots in between are deleted.
3389 *
3390 * blocks that would be freed [---------------------------]
3391 * snapshots ---O-------O--------O-------O--------O
3392 * firstsnap lastsnap
3393 *
3394 * This is the set of blocks that were born after the snap before firstsnap,
3395 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3396 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3397 * We calculate this by iterating over the relevant deadlists (from the snap
3398 * after lastsnap, backward to the snap after firstsnap), summing up the
3399 * space on the deadlist that was born after the snap before firstsnap.
3400 */
3401 int
3402 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3403 dsl_dataset_t *lastsnap,
3404 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3405 {
3406 int err = 0;
3407 uint64_t snapobj;
3408 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3409
3410 ASSERT(firstsnap->ds_is_snapshot);
3411 ASSERT(lastsnap->ds_is_snapshot);
3412
3413 /*
3414 * Check that the snapshots are in the same dsl_dir, and firstsnap
3415 * is before lastsnap.
3416 */
3417 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3418 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3419 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3420 return (SET_ERROR(EINVAL));
3421
3422 *usedp = *compp = *uncompp = 0;
3423
3424 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3425 while (snapobj != firstsnap->ds_object) {
3426 dsl_dataset_t *ds;
3427 uint64_t used, comp, uncomp;
3428
3429 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3430 if (err != 0)
3431 break;
3432
3433 dsl_deadlist_space_range(&ds->ds_deadlist,
3434 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3435 &used, &comp, &uncomp);
3436 *usedp += used;
3437 *compp += comp;
3438 *uncompp += uncomp;
3439
3440 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3441 ASSERT3U(snapobj, !=, 0);
3442 dsl_dataset_rele(ds, FTAG);
3443 }
3444 return (err);
3445 }
3446
3447 /*
3448 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3449 * For example, they could both be snapshots of the same filesystem, and
3450 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3451 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3452 * filesystem. Or 'earlier' could be the origin's origin.
3453 *
3454 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3455 */
3456 boolean_t
3457 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3458 uint64_t earlier_txg)
3459 {
3460 dsl_pool_t *dp = later->ds_dir->dd_pool;
3461 int error;
3462 boolean_t ret;
3463
3464 ASSERT(dsl_pool_config_held(dp));
3465 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
3466
3467 if (earlier_txg == 0)
3468 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3469
3470 if (later->ds_is_snapshot &&
3471 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3472 return (B_FALSE);
3473
3474 if (later->ds_dir == earlier->ds_dir)
3475 return (B_TRUE);
3476 if (!dsl_dir_is_clone(later->ds_dir))
3477 return (B_FALSE);
3478
3479 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3480 return (B_TRUE);
3481 dsl_dataset_t *origin;
3482 error = dsl_dataset_hold_obj(dp,
3483 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3484 if (error != 0)
3485 return (B_FALSE);
3486 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3487 dsl_dataset_rele(origin, FTAG);
3488 return (ret);
3489 }
3490
3491 void
3492 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3493 {
3494 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3495 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3496 }
3497
3498 boolean_t
3499 dsl_dataset_is_zapified(dsl_dataset_t *ds)
3500 {
3501 dmu_object_info_t doi;
3502
3503 dmu_object_info_from_db(ds->ds_dbuf, &doi);
3504 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
3505 }
3506
3507 boolean_t
3508 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
3509 {
3510 return (dsl_dataset_is_zapified(ds) &&
3511 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
3512 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
3513 }