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