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 https://opensource.org/licenses/CDDL-1.0.
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 /*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2020 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2016 Actifio, Inc. All rights reserved.
29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
31 * Copyright (c) 2019, Klara Inc.
32 * Copyright (c) 2019, Allan Jude
33 * Copyright (c) 2020 The FreeBSD Foundation [1]
34 *
35 * [1] Portions of this software were developed by Allan Jude
36 * under sponsorship from the FreeBSD Foundation.
37 */
38
39 #include <sys/dmu_objset.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_dir.h>
42 #include <sys/dsl_prop.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/dmu_traverse.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/dmu_tx.h>
47 #include <sys/arc.h>
48 #include <sys/zio.h>
49 #include <sys/zap.h>
50 #include <sys/zfeature.h>
51 #include <sys/unique.h>
52 #include <sys/zfs_context.h>
53 #include <sys/zfs_ioctl.h>
54 #include <sys/spa.h>
55 #include <sys/spa_impl.h>
56 #include <sys/vdev.h>
57 #include <sys/zfs_znode.h>
58 #include <sys/zfs_onexit.h>
59 #include <sys/zvol.h>
60 #include <sys/dsl_scan.h>
61 #include <sys/dsl_deadlist.h>
62 #include <sys/dsl_destroy.h>
63 #include <sys/dsl_userhold.h>
64 #include <sys/dsl_bookmark.h>
65 #include <sys/policy.h>
66 #include <sys/dmu_send.h>
67 #include <sys/dmu_recv.h>
68 #include <sys/zio_compress.h>
69 #include <zfs_fletcher.h>
70 #include <sys/zio_checksum.h>
71
72 /*
73 * The SPA supports block sizes up to 16MB. However, very large blocks
74 * can have an impact on i/o latency (e.g. tying up a spinning disk for
75 * ~300ms), and also potentially on the memory allocator. Therefore,
76 * we did not allow the recordsize to be set larger than zfs_max_recordsize
77 * (former default: 1MB). Larger blocks could be created by changing this
78 * tunable, and pools with larger blocks could always be imported and used,
79 * regardless of this setting.
80 *
81 * We do, however, still limit it by default to 1M on x86_32, because Linux's
82 * 3/1 memory split doesn't leave much room for 16M chunks.
83 */
84 #ifdef _ILP32
85 uint_t zfs_max_recordsize = 1 * 1024 * 1024;
86 #else
87 uint_t zfs_max_recordsize = 16 * 1024 * 1024;
88 #endif
89 static int zfs_allow_redacted_dataset_mount = 0;
90
91 int zfs_snapshot_history_enabled = 1;
92
93 #define SWITCH64(x, y) \
94 { \
95 uint64_t __tmp = (x); \
96 (x) = (y); \
97 (y) = __tmp; \
98 }
99
100 #define DS_REF_MAX (1ULL << 62)
101
102 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
103 uint64_t obj, dmu_tx_t *tx);
104 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
105 dmu_tx_t *tx);
106
107 static void unload_zfeature(dsl_dataset_t *ds, spa_feature_t f);
108
109 extern uint_t spa_asize_inflation;
110
111 static zil_header_t zero_zil;
112
113 /*
114 * Figure out how much of this delta should be propagated to the dsl_dir
115 * layer. If there's a refreservation, that space has already been
116 * partially accounted for in our ancestors.
117 */
118 static int64_t
parent_delta(dsl_dataset_t * ds,int64_t delta)119 parent_delta(dsl_dataset_t *ds, int64_t delta)
120 {
121 dsl_dataset_phys_t *ds_phys;
122 uint64_t old_bytes, new_bytes;
123
124 if (ds->ds_reserved == 0)
125 return (delta);
126
127 ds_phys = dsl_dataset_phys(ds);
128 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
129 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
130
131 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
132 return (new_bytes - old_bytes);
133 }
134
135 void
dsl_dataset_block_born(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx)136 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
137 {
138 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
139 int used = bp_get_dsize_sync(spa, bp);
140 int compressed = BP_GET_PSIZE(bp);
141 int uncompressed = BP_GET_UCSIZE(bp);
142 int64_t delta;
143 spa_feature_t f;
144
145 dprintf_bp(bp, "ds=%p", ds);
146
147 ASSERT(dmu_tx_is_syncing(tx));
148 /* It could have been compressed away to nothing */
149 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
150 return;
151 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
152 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
153 if (ds == NULL) {
154 dsl_pool_mos_diduse_space(tx->tx_pool,
155 used, compressed, uncompressed);
156 return;
157 }
158
159 ASSERT3U(BP_GET_LOGICAL_BIRTH(bp), >,
160 dsl_dataset_phys(ds)->ds_prev_snap_txg);
161 dmu_buf_will_dirty(ds->ds_dbuf, tx);
162 mutex_enter(&ds->ds_lock);
163 delta = parent_delta(ds, used);
164 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
165 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
166 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
167 dsl_dataset_phys(ds)->ds_unique_bytes += used;
168
169 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
170 ds->ds_feature_activation[SPA_FEATURE_LARGE_BLOCKS] =
171 (void *)B_TRUE;
172 }
173
174
175 f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
176 if (f != SPA_FEATURE_NONE) {
177 ASSERT3S(spa_feature_table[f].fi_type, ==,
178 ZFEATURE_TYPE_BOOLEAN);
179 ds->ds_feature_activation[f] = (void *)B_TRUE;
180 }
181
182 f = zio_compress_to_feature(BP_GET_COMPRESS(bp));
183 if (f != SPA_FEATURE_NONE) {
184 ASSERT3S(spa_feature_table[f].fi_type, ==,
185 ZFEATURE_TYPE_BOOLEAN);
186 ds->ds_feature_activation[f] = (void *)B_TRUE;
187 }
188
189 /*
190 * Track block for livelist, but ignore embedded blocks because
191 * they do not need to be freed.
192 */
193 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
194 BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg &&
195 !(BP_IS_EMBEDDED(bp))) {
196 ASSERT(dsl_dir_is_clone(ds->ds_dir));
197 ASSERT(spa_feature_is_enabled(spa,
198 SPA_FEATURE_LIVELIST));
199 bplist_append(&ds->ds_dir->dd_pending_allocs, bp);
200 }
201
202 mutex_exit(&ds->ds_lock);
203 dsl_dir_diduse_transfer_space(ds->ds_dir, delta,
204 compressed, uncompressed, used,
205 DD_USED_REFRSRV, DD_USED_HEAD, tx);
206 }
207
208 /*
209 * Called when the specified segment has been remapped, and is thus no
210 * longer referenced in the head dataset. The vdev must be indirect.
211 *
212 * If the segment is referenced by a snapshot, put it on the remap deadlist.
213 * Otherwise, add this segment to the obsolete spacemap.
214 */
215 void
dsl_dataset_block_remapped(dsl_dataset_t * ds,uint64_t vdev,uint64_t offset,uint64_t size,uint64_t birth,dmu_tx_t * tx)216 dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
217 uint64_t size, uint64_t birth, dmu_tx_t *tx)
218 {
219 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
220
221 ASSERT(dmu_tx_is_syncing(tx));
222 ASSERT(birth <= tx->tx_txg);
223 ASSERT(!ds->ds_is_snapshot);
224
225 if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
226 spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
227 } else {
228 blkptr_t fakebp;
229 dva_t *dva = &fakebp.blk_dva[0];
230
231 ASSERT(ds != NULL);
232
233 mutex_enter(&ds->ds_remap_deadlist_lock);
234 if (!dsl_dataset_remap_deadlist_exists(ds)) {
235 dsl_dataset_create_remap_deadlist(ds, tx);
236 }
237 mutex_exit(&ds->ds_remap_deadlist_lock);
238
239 BP_ZERO(&fakebp);
240 BP_SET_LOGICAL_BIRTH(&fakebp, birth);
241 DVA_SET_VDEV(dva, vdev);
242 DVA_SET_OFFSET(dva, offset);
243 DVA_SET_ASIZE(dva, size);
244 dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, B_FALSE,
245 tx);
246 }
247 }
248
249 int
dsl_dataset_block_kill(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx,boolean_t async)250 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
251 boolean_t async)
252 {
253 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
254
255 int used = bp_get_dsize_sync(spa, bp);
256 int compressed = BP_GET_PSIZE(bp);
257 int uncompressed = BP_GET_UCSIZE(bp);
258
259 if (BP_IS_HOLE(bp) || BP_IS_REDACTED(bp))
260 return (0);
261
262 ASSERT(dmu_tx_is_syncing(tx));
263 ASSERT(BP_GET_LOGICAL_BIRTH(bp) <= tx->tx_txg);
264
265 if (ds == NULL) {
266 dsl_free(tx->tx_pool, tx->tx_txg, bp);
267 dsl_pool_mos_diduse_space(tx->tx_pool,
268 -used, -compressed, -uncompressed);
269 return (used);
270 }
271 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
272
273 ASSERT(!ds->ds_is_snapshot);
274 dmu_buf_will_dirty(ds->ds_dbuf, tx);
275
276 /*
277 * Track block for livelist, but ignore embedded blocks because
278 * they do not need to be freed.
279 */
280 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist) &&
281 BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg &&
282 !(BP_IS_EMBEDDED(bp))) {
283 ASSERT(dsl_dir_is_clone(ds->ds_dir));
284 ASSERT(spa_feature_is_enabled(spa,
285 SPA_FEATURE_LIVELIST));
286 bplist_append(&ds->ds_dir->dd_pending_frees, bp);
287 }
288
289 if (BP_GET_LOGICAL_BIRTH(bp) > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
290 int64_t delta;
291
292 dprintf_bp(bp, "freeing ds=%llu", (u_longlong_t)ds->ds_object);
293 dsl_free(tx->tx_pool, tx->tx_txg, bp);
294
295 mutex_enter(&ds->ds_lock);
296 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
297 !DS_UNIQUE_IS_ACCURATE(ds));
298 delta = parent_delta(ds, -used);
299 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
300 mutex_exit(&ds->ds_lock);
301 dsl_dir_diduse_transfer_space(ds->ds_dir,
302 delta, -compressed, -uncompressed, -used,
303 DD_USED_REFRSRV, DD_USED_HEAD, tx);
304 } else {
305 dprintf_bp(bp, "putting on dead list: %s", "");
306 if (async) {
307 /*
308 * We are here as part of zio's write done callback,
309 * which means we're a zio interrupt thread. We can't
310 * call dsl_deadlist_insert() now because it may block
311 * waiting for I/O. Instead, put bp on the deferred
312 * queue and let dsl_pool_sync() finish the job.
313 */
314 bplist_append(&ds->ds_pending_deadlist, bp);
315 } else {
316 dsl_deadlist_insert(&ds->ds_deadlist, bp, B_FALSE, tx);
317 }
318 ASSERT3U(ds->ds_prev->ds_object, ==,
319 dsl_dataset_phys(ds)->ds_prev_snap_obj);
320 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
321 /* if (logical birth > prev prev snap txg) prev unique += bs */
322 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
323 ds->ds_object && BP_GET_LOGICAL_BIRTH(bp) >
324 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
325 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
326 mutex_enter(&ds->ds_prev->ds_lock);
327 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
328 mutex_exit(&ds->ds_prev->ds_lock);
329 }
330 if (BP_GET_LOGICAL_BIRTH(bp) > ds->ds_dir->dd_origin_txg) {
331 dsl_dir_transfer_space(ds->ds_dir, used,
332 DD_USED_HEAD, DD_USED_SNAP, tx);
333 }
334 }
335
336 dsl_bookmark_block_killed(ds, bp, tx);
337
338 mutex_enter(&ds->ds_lock);
339 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
340 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
341 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
342 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
343 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
344 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
345 mutex_exit(&ds->ds_lock);
346
347 return (used);
348 }
349
350 struct feature_type_uint64_array_arg {
351 uint64_t length;
352 uint64_t *array;
353 };
354
355 static void
unload_zfeature(dsl_dataset_t * ds,spa_feature_t f)356 unload_zfeature(dsl_dataset_t *ds, spa_feature_t f)
357 {
358 switch (spa_feature_table[f].fi_type) {
359 case ZFEATURE_TYPE_BOOLEAN:
360 break;
361 case ZFEATURE_TYPE_UINT64_ARRAY:
362 {
363 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
364 kmem_free(ftuaa->array, ftuaa->length * sizeof (uint64_t));
365 kmem_free(ftuaa, sizeof (*ftuaa));
366 break;
367 }
368 default:
369 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
370 }
371 }
372
373 static int
load_zfeature(objset_t * mos,dsl_dataset_t * ds,spa_feature_t f)374 load_zfeature(objset_t *mos, dsl_dataset_t *ds, spa_feature_t f)
375 {
376 int err = 0;
377 switch (spa_feature_table[f].fi_type) {
378 case ZFEATURE_TYPE_BOOLEAN:
379 err = zap_contains(mos, ds->ds_object,
380 spa_feature_table[f].fi_guid);
381 if (err == 0) {
382 ds->ds_feature[f] = (void *)B_TRUE;
383 } else {
384 ASSERT3U(err, ==, ENOENT);
385 err = 0;
386 }
387 break;
388 case ZFEATURE_TYPE_UINT64_ARRAY:
389 {
390 uint64_t int_size, num_int;
391 uint64_t *data;
392 err = zap_length(mos, ds->ds_object,
393 spa_feature_table[f].fi_guid, &int_size, &num_int);
394 if (err != 0) {
395 ASSERT3U(err, ==, ENOENT);
396 err = 0;
397 break;
398 }
399 ASSERT3U(int_size, ==, sizeof (uint64_t));
400 data = kmem_alloc(int_size * num_int, KM_SLEEP);
401 VERIFY0(zap_lookup(mos, ds->ds_object,
402 spa_feature_table[f].fi_guid, int_size, num_int, data));
403 struct feature_type_uint64_array_arg *ftuaa =
404 kmem_alloc(sizeof (*ftuaa), KM_SLEEP);
405 ftuaa->length = num_int;
406 ftuaa->array = data;
407 ds->ds_feature[f] = ftuaa;
408 break;
409 }
410 default:
411 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
412 }
413 return (err);
414 }
415
416 /*
417 * We have to release the fsid synchronously or we risk that a subsequent
418 * mount of the same dataset will fail to unique_insert the fsid. This
419 * failure would manifest itself as the fsid of this dataset changing
420 * between mounts which makes NFS clients quite unhappy.
421 */
422 static void
dsl_dataset_evict_sync(void * dbu)423 dsl_dataset_evict_sync(void *dbu)
424 {
425 dsl_dataset_t *ds = dbu;
426
427 ASSERT(ds->ds_owner == NULL);
428
429 unique_remove(ds->ds_fsid_guid);
430 }
431
432 static void
dsl_dataset_evict_async(void * dbu)433 dsl_dataset_evict_async(void *dbu)
434 {
435 dsl_dataset_t *ds = dbu;
436
437 ASSERT(ds->ds_owner == NULL);
438
439 ds->ds_dbuf = NULL;
440
441 if (ds->ds_objset != NULL)
442 dmu_objset_evict(ds->ds_objset);
443
444 if (ds->ds_prev) {
445 dsl_dataset_rele(ds->ds_prev, ds);
446 ds->ds_prev = NULL;
447 }
448
449 dsl_bookmark_fini_ds(ds);
450
451 bplist_destroy(&ds->ds_pending_deadlist);
452 if (dsl_deadlist_is_open(&ds->ds_deadlist))
453 dsl_deadlist_close(&ds->ds_deadlist);
454 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
455 dsl_deadlist_close(&ds->ds_remap_deadlist);
456 if (ds->ds_dir)
457 dsl_dir_async_rele(ds->ds_dir, ds);
458
459 ASSERT(!list_link_active(&ds->ds_synced_link));
460
461 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
462 if (dsl_dataset_feature_is_active(ds, f))
463 unload_zfeature(ds, f);
464 }
465
466 list_destroy(&ds->ds_prop_cbs);
467 mutex_destroy(&ds->ds_lock);
468 mutex_destroy(&ds->ds_opening_lock);
469 mutex_destroy(&ds->ds_sendstream_lock);
470 mutex_destroy(&ds->ds_remap_deadlist_lock);
471 zfs_refcount_destroy(&ds->ds_longholds);
472 rrw_destroy(&ds->ds_bp_rwlock);
473
474 kmem_free(ds, sizeof (dsl_dataset_t));
475 }
476
477 int
dsl_dataset_get_snapname(dsl_dataset_t * ds)478 dsl_dataset_get_snapname(dsl_dataset_t *ds)
479 {
480 dsl_dataset_phys_t *headphys;
481 int err;
482 dmu_buf_t *headdbuf;
483 dsl_pool_t *dp = ds->ds_dir->dd_pool;
484 objset_t *mos = dp->dp_meta_objset;
485
486 if (ds->ds_snapname[0])
487 return (0);
488 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
489 return (0);
490
491 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
492 FTAG, &headdbuf);
493 if (err != 0)
494 return (err);
495 headphys = headdbuf->db_data;
496 err = zap_value_search(dp->dp_meta_objset,
497 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
498 if (err != 0 && zfs_recover == B_TRUE) {
499 err = 0;
500 (void) snprintf(ds->ds_snapname, sizeof (ds->ds_snapname),
501 "SNAPOBJ=%llu-ERR=%d",
502 (unsigned long long)ds->ds_object, err);
503 }
504 dmu_buf_rele(headdbuf, FTAG);
505 return (err);
506 }
507
508 int
dsl_dataset_snap_lookup(dsl_dataset_t * ds,const char * name,uint64_t * value)509 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
510 {
511 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
512 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
513 matchtype_t mt = 0;
514 int err;
515
516 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
517 mt = MT_NORMALIZE;
518
519 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
520 value, mt, NULL, 0, NULL);
521 if (err == ENOTSUP && (mt & MT_NORMALIZE))
522 err = zap_lookup(mos, snapobj, name, 8, 1, value);
523 return (err);
524 }
525
526 int
dsl_dataset_snap_remove(dsl_dataset_t * ds,const char * name,dmu_tx_t * tx,boolean_t adj_cnt)527 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
528 boolean_t adj_cnt)
529 {
530 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
531 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
532 matchtype_t mt = 0;
533 int err;
534
535 dsl_dir_snap_cmtime_update(ds->ds_dir, tx);
536
537 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
538 mt = MT_NORMALIZE;
539
540 err = zap_remove_norm(mos, snapobj, name, mt, tx);
541 if (err == ENOTSUP && (mt & MT_NORMALIZE))
542 err = zap_remove(mos, snapobj, name, tx);
543
544 if (err == 0 && adj_cnt)
545 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
546 DD_FIELD_SNAPSHOT_COUNT, tx);
547
548 return (err);
549 }
550
551 boolean_t
dsl_dataset_try_add_ref(dsl_pool_t * dp,dsl_dataset_t * ds,const void * tag)552 dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, const void *tag)
553 {
554 dmu_buf_t *dbuf = ds->ds_dbuf;
555 boolean_t result = B_FALSE;
556
557 if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
558 ds->ds_object, DMU_BONUS_BLKID, tag)) {
559
560 if (ds == dmu_buf_get_user(dbuf))
561 result = B_TRUE;
562 else
563 dmu_buf_rele(dbuf, tag);
564 }
565
566 return (result);
567 }
568
569 int
dsl_dataset_hold_obj(dsl_pool_t * dp,uint64_t dsobj,const void * tag,dsl_dataset_t ** dsp)570 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, const void *tag,
571 dsl_dataset_t **dsp)
572 {
573 objset_t *mos = dp->dp_meta_objset;
574 dmu_buf_t *dbuf;
575 dsl_dataset_t *ds;
576 int err;
577 dmu_object_info_t doi;
578
579 ASSERT(dsl_pool_config_held(dp));
580
581 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
582 if (err != 0)
583 return (err);
584
585 /* Make sure dsobj has the correct object type. */
586 dmu_object_info_from_db(dbuf, &doi);
587 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
588 dmu_buf_rele(dbuf, tag);
589 return (SET_ERROR(EINVAL));
590 }
591
592 ds = dmu_buf_get_user(dbuf);
593 if (ds == NULL) {
594 dsl_dataset_t *winner = NULL;
595
596 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
597 ds->ds_dbuf = dbuf;
598 ds->ds_object = dsobj;
599 ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
600 list_link_init(&ds->ds_synced_link);
601
602 err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
603 NULL, ds, &ds->ds_dir);
604 if (err != 0) {
605 kmem_free(ds, sizeof (dsl_dataset_t));
606 dmu_buf_rele(dbuf, tag);
607 return (err);
608 }
609
610 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
611 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
612 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
613 mutex_init(&ds->ds_remap_deadlist_lock,
614 NULL, MUTEX_DEFAULT, NULL);
615 rrw_init(&ds->ds_bp_rwlock, B_FALSE);
616 zfs_refcount_create(&ds->ds_longholds);
617
618 bplist_create(&ds->ds_pending_deadlist);
619
620 list_create(&ds->ds_sendstreams, sizeof (dmu_sendstatus_t),
621 offsetof(dmu_sendstatus_t, dss_link));
622
623 list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
624 offsetof(dsl_prop_cb_record_t, cbr_ds_node));
625
626 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
627 spa_feature_t f;
628
629 for (f = 0; f < SPA_FEATURES; f++) {
630 if (!(spa_feature_table[f].fi_flags &
631 ZFEATURE_FLAG_PER_DATASET))
632 continue;
633 err = load_zfeature(mos, ds, f);
634 }
635 }
636
637 if (!ds->ds_is_snapshot) {
638 ds->ds_snapname[0] = '\0';
639 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
640 err = dsl_dataset_hold_obj(dp,
641 dsl_dataset_phys(ds)->ds_prev_snap_obj,
642 ds, &ds->ds_prev);
643 }
644 if (err != 0)
645 goto after_dsl_bookmark_fini;
646 err = dsl_bookmark_init_ds(ds);
647 } else {
648 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
649 err = dsl_dataset_get_snapname(ds);
650 if (err == 0 &&
651 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
652 err = zap_count(
653 ds->ds_dir->dd_pool->dp_meta_objset,
654 dsl_dataset_phys(ds)->ds_userrefs_obj,
655 &ds->ds_userrefs);
656 }
657 }
658
659 if (err == 0 && !ds->ds_is_snapshot) {
660 err = dsl_prop_get_int_ds(ds,
661 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
662 &ds->ds_reserved);
663 if (err == 0) {
664 err = dsl_prop_get_int_ds(ds,
665 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
666 &ds->ds_quota);
667 }
668 } else {
669 ds->ds_reserved = ds->ds_quota = 0;
670 }
671
672 if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 &&
673 ds->ds_is_snapshot &&
674 zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) {
675 dp->dp_spa->spa_errata =
676 ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
677 }
678
679 dsl_deadlist_open(&ds->ds_deadlist,
680 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
681 uint64_t remap_deadlist_obj =
682 dsl_dataset_get_remap_deadlist_object(ds);
683 if (remap_deadlist_obj != 0) {
684 dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
685 remap_deadlist_obj);
686 }
687
688 dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
689 dsl_dataset_evict_async, &ds->ds_dbuf);
690 if (err == 0)
691 winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
692
693 if (err != 0 || winner != NULL) {
694 dsl_deadlist_close(&ds->ds_deadlist);
695 if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
696 dsl_deadlist_close(&ds->ds_remap_deadlist);
697 dsl_bookmark_fini_ds(ds);
698 after_dsl_bookmark_fini:
699 if (ds->ds_prev)
700 dsl_dataset_rele(ds->ds_prev, ds);
701 dsl_dir_rele(ds->ds_dir, ds);
702 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
703 if (dsl_dataset_feature_is_active(ds, f))
704 unload_zfeature(ds, f);
705 }
706
707 list_destroy(&ds->ds_prop_cbs);
708 list_destroy(&ds->ds_sendstreams);
709 bplist_destroy(&ds->ds_pending_deadlist);
710 mutex_destroy(&ds->ds_lock);
711 mutex_destroy(&ds->ds_opening_lock);
712 mutex_destroy(&ds->ds_sendstream_lock);
713 mutex_destroy(&ds->ds_remap_deadlist_lock);
714 zfs_refcount_destroy(&ds->ds_longholds);
715 rrw_destroy(&ds->ds_bp_rwlock);
716 kmem_free(ds, sizeof (dsl_dataset_t));
717 if (err != 0) {
718 dmu_buf_rele(dbuf, tag);
719 return (err);
720 }
721 ds = winner;
722 } else {
723 ds->ds_fsid_guid =
724 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
725 if (ds->ds_fsid_guid !=
726 dsl_dataset_phys(ds)->ds_fsid_guid) {
727 zfs_dbgmsg("ds_fsid_guid changed from "
728 "%llx to %llx for pool %s dataset id %llu",
729 (long long)
730 dsl_dataset_phys(ds)->ds_fsid_guid,
731 (long long)ds->ds_fsid_guid,
732 spa_name(dp->dp_spa),
733 (u_longlong_t)dsobj);
734 }
735 }
736 }
737
738 ASSERT3P(ds->ds_dbuf, ==, dbuf);
739 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
740 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
741 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
742 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
743 *dsp = ds;
744
745 return (0);
746 }
747
748 int
dsl_dataset_create_key_mapping(dsl_dataset_t * ds)749 dsl_dataset_create_key_mapping(dsl_dataset_t *ds)
750 {
751 dsl_dir_t *dd = ds->ds_dir;
752
753 if (dd->dd_crypto_obj == 0)
754 return (0);
755
756 return (spa_keystore_create_mapping(dd->dd_pool->dp_spa,
757 ds, ds, &ds->ds_key_mapping));
758 }
759
760 int
dsl_dataset_hold_obj_flags(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)761 dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj,
762 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp)
763 {
764 int err;
765
766 err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
767 if (err != 0)
768 return (err);
769
770 ASSERT3P(*dsp, !=, NULL);
771
772 if (flags & DS_HOLD_FLAG_DECRYPT) {
773 err = dsl_dataset_create_key_mapping(*dsp);
774 if (err != 0)
775 dsl_dataset_rele(*dsp, tag);
776 }
777
778 return (err);
779 }
780
781 int
dsl_dataset_hold_flags(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)782 dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
783 const void *tag, dsl_dataset_t **dsp)
784 {
785 dsl_dir_t *dd;
786 const char *snapname;
787 uint64_t obj;
788 int err = 0;
789 dsl_dataset_t *ds;
790
791 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
792 if (err != 0)
793 return (err);
794
795 ASSERT(dsl_pool_config_held(dp));
796 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
797 if (obj != 0)
798 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds);
799 else
800 err = SET_ERROR(ENOENT);
801
802 /* we may be looking for a snapshot */
803 if (err == 0 && snapname != NULL) {
804 dsl_dataset_t *snap_ds;
805
806 if (*snapname++ != '@') {
807 dsl_dataset_rele_flags(ds, flags, tag);
808 dsl_dir_rele(dd, FTAG);
809 return (SET_ERROR(ENOENT));
810 }
811
812 dprintf("looking for snapshot '%s'\n", snapname);
813 err = dsl_dataset_snap_lookup(ds, snapname, &obj);
814 if (err == 0) {
815 err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag,
816 &snap_ds);
817 }
818 dsl_dataset_rele_flags(ds, flags, tag);
819
820 if (err == 0) {
821 mutex_enter(&snap_ds->ds_lock);
822 if (snap_ds->ds_snapname[0] == 0)
823 (void) strlcpy(snap_ds->ds_snapname, snapname,
824 sizeof (snap_ds->ds_snapname));
825 mutex_exit(&snap_ds->ds_lock);
826 ds = snap_ds;
827 }
828 }
829 if (err == 0)
830 *dsp = ds;
831 dsl_dir_rele(dd, FTAG);
832 return (err);
833 }
834
835 int
dsl_dataset_hold(dsl_pool_t * dp,const char * name,const void * tag,dsl_dataset_t ** dsp)836 dsl_dataset_hold(dsl_pool_t *dp, const char *name, const void *tag,
837 dsl_dataset_t **dsp)
838 {
839 return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp));
840 }
841
842 static int
dsl_dataset_own_obj_impl(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,const void * tag,boolean_t override,dsl_dataset_t ** dsp)843 dsl_dataset_own_obj_impl(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
844 const void *tag, boolean_t override, dsl_dataset_t **dsp)
845 {
846 int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp);
847 if (err != 0)
848 return (err);
849 if (!dsl_dataset_tryown(*dsp, tag, override)) {
850 dsl_dataset_rele_flags(*dsp, flags, tag);
851 *dsp = NULL;
852 return (SET_ERROR(EBUSY));
853 }
854 return (0);
855 }
856
857
858 int
dsl_dataset_own_obj(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)859 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
860 const void *tag, dsl_dataset_t **dsp)
861 {
862 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_FALSE, dsp));
863 }
864
865 int
dsl_dataset_own_obj_force(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)866 dsl_dataset_own_obj_force(dsl_pool_t *dp, uint64_t dsobj,
867 ds_hold_flags_t flags, const void *tag, dsl_dataset_t **dsp)
868 {
869 return (dsl_dataset_own_obj_impl(dp, dsobj, flags, tag, B_TRUE, dsp));
870 }
871
872 static int
dsl_dataset_own_impl(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,const void * tag,boolean_t override,dsl_dataset_t ** dsp)873 dsl_dataset_own_impl(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
874 const void *tag, boolean_t override, dsl_dataset_t **dsp)
875 {
876 int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp);
877 if (err != 0)
878 return (err);
879 if (!dsl_dataset_tryown(*dsp, tag, override)) {
880 dsl_dataset_rele_flags(*dsp, flags, tag);
881 return (SET_ERROR(EBUSY));
882 }
883 return (0);
884 }
885
886 int
dsl_dataset_own_force(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)887 dsl_dataset_own_force(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
888 const void *tag, dsl_dataset_t **dsp)
889 {
890 return (dsl_dataset_own_impl(dp, name, flags, tag, B_TRUE, dsp));
891 }
892
893 int
dsl_dataset_own(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,const void * tag,dsl_dataset_t ** dsp)894 dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
895 const void *tag, dsl_dataset_t **dsp)
896 {
897 return (dsl_dataset_own_impl(dp, name, flags, tag, B_FALSE, dsp));
898 }
899
900 /*
901 * See the comment above dsl_pool_hold() for details. In summary, a long
902 * hold is used to prevent destruction of a dataset while the pool hold
903 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
904 *
905 * The dataset and pool must be held when this function is called. After it
906 * is called, the pool hold may be released while the dataset is still held
907 * and accessed.
908 */
909 void
dsl_dataset_long_hold(dsl_dataset_t * ds,const void * tag)910 dsl_dataset_long_hold(dsl_dataset_t *ds, const void *tag)
911 {
912 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
913 (void) zfs_refcount_add(&ds->ds_longholds, tag);
914 }
915
916 void
dsl_dataset_long_rele(dsl_dataset_t * ds,const void * tag)917 dsl_dataset_long_rele(dsl_dataset_t *ds, const void *tag)
918 {
919 (void) zfs_refcount_remove(&ds->ds_longholds, tag);
920 }
921
922 /* Return B_TRUE if there are any long holds on this dataset. */
923 boolean_t
dsl_dataset_long_held(dsl_dataset_t * ds)924 dsl_dataset_long_held(dsl_dataset_t *ds)
925 {
926 return (!zfs_refcount_is_zero(&ds->ds_longholds));
927 }
928
929 void
dsl_dataset_name(dsl_dataset_t * ds,char * name)930 dsl_dataset_name(dsl_dataset_t *ds, char *name)
931 {
932 if (ds == NULL) {
933 (void) strlcpy(name, "mos", ZFS_MAX_DATASET_NAME_LEN);
934 } else {
935 dsl_dir_name(ds->ds_dir, name);
936 VERIFY0(dsl_dataset_get_snapname(ds));
937 if (ds->ds_snapname[0]) {
938 VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
939 <, ZFS_MAX_DATASET_NAME_LEN);
940 /*
941 * We use a "recursive" mutex so that we
942 * can call dprintf_ds() with ds_lock held.
943 */
944 if (!MUTEX_HELD(&ds->ds_lock)) {
945 mutex_enter(&ds->ds_lock);
946 VERIFY3U(strlcat(name, ds->ds_snapname,
947 ZFS_MAX_DATASET_NAME_LEN), <,
948 ZFS_MAX_DATASET_NAME_LEN);
949 mutex_exit(&ds->ds_lock);
950 } else {
951 VERIFY3U(strlcat(name, ds->ds_snapname,
952 ZFS_MAX_DATASET_NAME_LEN), <,
953 ZFS_MAX_DATASET_NAME_LEN);
954 }
955 }
956 }
957 }
958
959 int
dsl_dataset_namelen(dsl_dataset_t * ds)960 dsl_dataset_namelen(dsl_dataset_t *ds)
961 {
962 VERIFY0(dsl_dataset_get_snapname(ds));
963 mutex_enter(&ds->ds_lock);
964 int len = strlen(ds->ds_snapname);
965 mutex_exit(&ds->ds_lock);
966 /* add '@' if ds is a snap */
967 if (len > 0)
968 len++;
969 len += dsl_dir_namelen(ds->ds_dir);
970 return (len);
971 }
972
973 void
dsl_dataset_rele(dsl_dataset_t * ds,const void * tag)974 dsl_dataset_rele(dsl_dataset_t *ds, const void *tag)
975 {
976 dmu_buf_rele(ds->ds_dbuf, tag);
977 }
978
979 void
dsl_dataset_remove_key_mapping(dsl_dataset_t * ds)980 dsl_dataset_remove_key_mapping(dsl_dataset_t *ds)
981 {
982 dsl_dir_t *dd = ds->ds_dir;
983
984 if (dd == NULL || dd->dd_crypto_obj == 0)
985 return;
986
987 (void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa,
988 ds->ds_object, ds);
989 }
990
991 void
dsl_dataset_rele_flags(dsl_dataset_t * ds,ds_hold_flags_t flags,const void * tag)992 dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags,
993 const void *tag)
994 {
995 if (flags & DS_HOLD_FLAG_DECRYPT)
996 dsl_dataset_remove_key_mapping(ds);
997
998 dsl_dataset_rele(ds, tag);
999 }
1000
1001 void
dsl_dataset_disown(dsl_dataset_t * ds,ds_hold_flags_t flags,const void * tag)1002 dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, const void *tag)
1003 {
1004 ASSERT3P(ds->ds_owner, ==, tag);
1005 ASSERT(ds->ds_dbuf != NULL);
1006
1007 mutex_enter(&ds->ds_lock);
1008 ds->ds_owner = NULL;
1009 mutex_exit(&ds->ds_lock);
1010 dsl_dataset_long_rele(ds, tag);
1011 dsl_dataset_rele_flags(ds, flags, tag);
1012 }
1013
1014 boolean_t
dsl_dataset_tryown(dsl_dataset_t * ds,const void * tag,boolean_t override)1015 dsl_dataset_tryown(dsl_dataset_t *ds, const void *tag, boolean_t override)
1016 {
1017 boolean_t gotit = FALSE;
1018
1019 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1020 mutex_enter(&ds->ds_lock);
1021 if (ds->ds_owner == NULL && (override || !(DS_IS_INCONSISTENT(ds) ||
1022 (dsl_dataset_feature_is_active(ds,
1023 SPA_FEATURE_REDACTED_DATASETS) &&
1024 !zfs_allow_redacted_dataset_mount)))) {
1025 ds->ds_owner = tag;
1026 dsl_dataset_long_hold(ds, tag);
1027 gotit = TRUE;
1028 }
1029 mutex_exit(&ds->ds_lock);
1030 return (gotit);
1031 }
1032
1033 boolean_t
dsl_dataset_has_owner(dsl_dataset_t * ds)1034 dsl_dataset_has_owner(dsl_dataset_t *ds)
1035 {
1036 boolean_t rv;
1037 mutex_enter(&ds->ds_lock);
1038 rv = (ds->ds_owner != NULL);
1039 mutex_exit(&ds->ds_lock);
1040 return (rv);
1041 }
1042
1043 static boolean_t
zfeature_active(spa_feature_t f,void * arg)1044 zfeature_active(spa_feature_t f, void *arg)
1045 {
1046 switch (spa_feature_table[f].fi_type) {
1047 case ZFEATURE_TYPE_BOOLEAN: {
1048 boolean_t val = (boolean_t)(uintptr_t)arg;
1049 ASSERT(val == B_FALSE || val == B_TRUE);
1050 return (val);
1051 }
1052 case ZFEATURE_TYPE_UINT64_ARRAY:
1053 /*
1054 * In this case, arg is a uint64_t array. The feature is active
1055 * if the array is non-null.
1056 */
1057 return (arg != NULL);
1058 default:
1059 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1060 return (B_FALSE);
1061 }
1062 }
1063
1064 boolean_t
dsl_dataset_feature_is_active(dsl_dataset_t * ds,spa_feature_t f)1065 dsl_dataset_feature_is_active(dsl_dataset_t *ds, spa_feature_t f)
1066 {
1067 return (zfeature_active(f, ds->ds_feature[f]));
1068 }
1069
1070 /*
1071 * The buffers passed out by this function are references to internal buffers;
1072 * they should not be freed by callers of this function, and they should not be
1073 * used after the dataset has been released.
1074 */
1075 boolean_t
dsl_dataset_get_uint64_array_feature(dsl_dataset_t * ds,spa_feature_t f,uint64_t * outlength,uint64_t ** outp)1076 dsl_dataset_get_uint64_array_feature(dsl_dataset_t *ds, spa_feature_t f,
1077 uint64_t *outlength, uint64_t **outp)
1078 {
1079 VERIFY(spa_feature_table[f].fi_type & ZFEATURE_TYPE_UINT64_ARRAY);
1080 if (!dsl_dataset_feature_is_active(ds, f)) {
1081 return (B_FALSE);
1082 }
1083 struct feature_type_uint64_array_arg *ftuaa = ds->ds_feature[f];
1084 *outp = ftuaa->array;
1085 *outlength = ftuaa->length;
1086 return (B_TRUE);
1087 }
1088
1089 void
dsl_dataset_activate_feature(uint64_t dsobj,spa_feature_t f,void * arg,dmu_tx_t * tx)1090 dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, void *arg,
1091 dmu_tx_t *tx)
1092 {
1093 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1094 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1095 uint64_t zero = 0;
1096
1097 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1098
1099 spa_feature_incr(spa, f, tx);
1100 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1101
1102 switch (spa_feature_table[f].fi_type) {
1103 case ZFEATURE_TYPE_BOOLEAN:
1104 ASSERT3S((boolean_t)(uintptr_t)arg, ==, B_TRUE);
1105 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1106 sizeof (zero), 1, &zero, tx));
1107 break;
1108 case ZFEATURE_TYPE_UINT64_ARRAY:
1109 {
1110 struct feature_type_uint64_array_arg *ftuaa = arg;
1111 VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
1112 sizeof (uint64_t), ftuaa->length, ftuaa->array, tx));
1113 break;
1114 }
1115 default:
1116 panic("Invalid zfeature type %d", spa_feature_table[f].fi_type);
1117 }
1118 }
1119
1120 static void
dsl_dataset_deactivate_feature_impl(dsl_dataset_t * ds,spa_feature_t f,dmu_tx_t * tx)1121 dsl_dataset_deactivate_feature_impl(dsl_dataset_t *ds, spa_feature_t f,
1122 dmu_tx_t *tx)
1123 {
1124 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
1125 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
1126 uint64_t dsobj = ds->ds_object;
1127
1128 VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
1129
1130 VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
1131 spa_feature_decr(spa, f, tx);
1132 ds->ds_feature[f] = NULL;
1133 }
1134
1135 void
dsl_dataset_deactivate_feature(dsl_dataset_t * ds,spa_feature_t f,dmu_tx_t * tx)1136 dsl_dataset_deactivate_feature(dsl_dataset_t *ds, spa_feature_t f, dmu_tx_t *tx)
1137 {
1138 unload_zfeature(ds, f);
1139 dsl_dataset_deactivate_feature_impl(ds, f, tx);
1140 }
1141
1142 uint64_t
dsl_dataset_create_sync_dd(dsl_dir_t * dd,dsl_dataset_t * origin,dsl_crypto_params_t * dcp,uint64_t flags,dmu_tx_t * tx)1143 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
1144 dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx)
1145 {
1146 dsl_pool_t *dp = dd->dd_pool;
1147 dmu_buf_t *dbuf;
1148 dsl_dataset_phys_t *dsphys;
1149 uint64_t dsobj;
1150 objset_t *mos = dp->dp_meta_objset;
1151
1152 if (origin == NULL)
1153 origin = dp->dp_origin_snap;
1154
1155 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
1156 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
1157 ASSERT(dmu_tx_is_syncing(tx));
1158 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
1159
1160 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1161 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1162 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1163 dmu_buf_will_dirty(dbuf, tx);
1164 dsphys = dbuf->db_data;
1165 memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1166 dsphys->ds_dir_obj = dd->dd_object;
1167 dsphys->ds_flags = flags;
1168 dsphys->ds_fsid_guid = unique_create();
1169 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1170 sizeof (dsphys->ds_guid));
1171 dsphys->ds_snapnames_zapobj =
1172 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
1173 DMU_OT_NONE, 0, tx);
1174 dsphys->ds_creation_time = gethrestime_sec();
1175 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
1176
1177 if (origin == NULL) {
1178 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
1179 } else {
1180 dsl_dataset_t *ohds; /* head of the origin snapshot */
1181
1182 dsphys->ds_prev_snap_obj = origin->ds_object;
1183 dsphys->ds_prev_snap_txg =
1184 dsl_dataset_phys(origin)->ds_creation_txg;
1185 dsphys->ds_referenced_bytes =
1186 dsl_dataset_phys(origin)->ds_referenced_bytes;
1187 dsphys->ds_compressed_bytes =
1188 dsl_dataset_phys(origin)->ds_compressed_bytes;
1189 dsphys->ds_uncompressed_bytes =
1190 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
1191 rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
1192 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
1193 rrw_exit(&origin->ds_bp_rwlock, FTAG);
1194
1195 /*
1196 * Inherit flags that describe the dataset's contents
1197 * (INCONSISTENT) or properties (Case Insensitive).
1198 */
1199 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
1200 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
1201
1202 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1203 if (zfeature_active(f, origin->ds_feature[f])) {
1204 dsl_dataset_activate_feature(dsobj, f,
1205 origin->ds_feature[f], tx);
1206 }
1207 }
1208
1209 dmu_buf_will_dirty(origin->ds_dbuf, tx);
1210 dsl_dataset_phys(origin)->ds_num_children++;
1211
1212 VERIFY0(dsl_dataset_hold_obj(dp,
1213 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
1214 FTAG, &ohds));
1215 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
1216 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
1217 dsl_dataset_rele(ohds, FTAG);
1218
1219 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
1220 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
1221 dsl_dataset_phys(origin)->ds_next_clones_obj =
1222 zap_create(mos,
1223 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
1224 }
1225 VERIFY0(zap_add_int(mos,
1226 dsl_dataset_phys(origin)->ds_next_clones_obj,
1227 dsobj, tx));
1228 }
1229
1230 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1231 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
1232 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1233 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
1234 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
1235 dsl_dir_phys(origin->ds_dir)->dd_clones =
1236 zap_create(mos,
1237 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
1238 }
1239 VERIFY0(zap_add_int(mos,
1240 dsl_dir_phys(origin->ds_dir)->dd_clones,
1241 dsobj, tx));
1242 }
1243 }
1244
1245 /* handle encryption */
1246 dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx);
1247
1248 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1249 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1250
1251 dmu_buf_rele(dbuf, FTAG);
1252
1253 dmu_buf_will_dirty(dd->dd_dbuf, tx);
1254 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
1255
1256 return (dsobj);
1257 }
1258
1259 static void
dsl_dataset_zero_zil(dsl_dataset_t * ds,dmu_tx_t * tx)1260 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
1261 {
1262 objset_t *os;
1263
1264 VERIFY0(dmu_objset_from_ds(ds, &os));
1265 if (memcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
1266 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1267 zio_t *zio;
1268
1269 memset(&os->os_zil_header, 0, sizeof (os->os_zil_header));
1270 if (os->os_encrypted)
1271 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
1272
1273 zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1274 dsl_dataset_sync(ds, zio, tx);
1275 VERIFY0(zio_wait(zio));
1276 dsl_dataset_sync_done(ds, tx);
1277 }
1278 }
1279
1280 uint64_t
dsl_dataset_create_sync(dsl_dir_t * pdd,const char * lastname,dsl_dataset_t * origin,uint64_t flags,cred_t * cr,dsl_crypto_params_t * dcp,dmu_tx_t * tx)1281 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
1282 dsl_dataset_t *origin, uint64_t flags, cred_t *cr,
1283 dsl_crypto_params_t *dcp, dmu_tx_t *tx)
1284 {
1285 dsl_pool_t *dp = pdd->dd_pool;
1286 uint64_t dsobj, ddobj;
1287 dsl_dir_t *dd;
1288
1289 ASSERT(dmu_tx_is_syncing(tx));
1290 ASSERT(lastname[0] != '@');
1291 /*
1292 * Filesystems will eventually have their origin set to dp_origin_snap,
1293 * but that's taken care of in dsl_dataset_create_sync_dd. When
1294 * creating a filesystem, this function is called with origin equal to
1295 * NULL.
1296 */
1297 if (origin != NULL)
1298 ASSERT3P(origin, !=, dp->dp_origin_snap);
1299
1300 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
1301 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
1302
1303 dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp,
1304 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
1305
1306 dsl_deleg_set_create_perms(dd, tx, cr);
1307
1308 /*
1309 * If we are creating a clone and the livelist feature is enabled,
1310 * add the entry DD_FIELD_LIVELIST to ZAP.
1311 */
1312 if (origin != NULL &&
1313 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LIVELIST)) {
1314 objset_t *mos = dd->dd_pool->dp_meta_objset;
1315 dsl_dir_zapify(dd, tx);
1316 uint64_t obj = dsl_deadlist_alloc(mos, tx);
1317 VERIFY0(zap_add(mos, dd->dd_object, DD_FIELD_LIVELIST,
1318 sizeof (uint64_t), 1, &obj, tx));
1319 spa_feature_incr(dp->dp_spa, SPA_FEATURE_LIVELIST, tx);
1320 }
1321
1322 /*
1323 * Since we're creating a new node we know it's a leaf, so we can
1324 * initialize the counts if the limit feature is active.
1325 */
1326 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
1327 uint64_t cnt = 0;
1328 objset_t *os = dd->dd_pool->dp_meta_objset;
1329
1330 dsl_dir_zapify(dd, tx);
1331 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1332 sizeof (cnt), 1, &cnt, tx));
1333 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1334 sizeof (cnt), 1, &cnt, tx));
1335 }
1336
1337 dsl_dir_rele(dd, FTAG);
1338
1339 /*
1340 * If we are creating a clone, make sure we zero out any stale
1341 * data from the origin snapshots zil header.
1342 */
1343 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
1344 dsl_dataset_t *ds;
1345
1346 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1347 dsl_dataset_zero_zil(ds, tx);
1348 dsl_dataset_rele(ds, FTAG);
1349 }
1350
1351 return (dsobj);
1352 }
1353
1354 /*
1355 * The unique space in the head dataset can be calculated by subtracting
1356 * the space used in the most recent snapshot, that is still being used
1357 * in this file system, from the space currently in use. To figure out
1358 * the space in the most recent snapshot still in use, we need to take
1359 * the total space used in the snapshot and subtract out the space that
1360 * has been freed up since the snapshot was taken.
1361 */
1362 void
dsl_dataset_recalc_head_uniq(dsl_dataset_t * ds)1363 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1364 {
1365 uint64_t mrs_used;
1366 uint64_t dlused, dlcomp, dluncomp;
1367
1368 ASSERT(!ds->ds_is_snapshot);
1369
1370 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1371 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1372 else
1373 mrs_used = 0;
1374
1375 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1376
1377 ASSERT3U(dlused, <=, mrs_used);
1378 dsl_dataset_phys(ds)->ds_unique_bytes =
1379 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1380
1381 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1382 SPA_VERSION_UNIQUE_ACCURATE)
1383 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1384 }
1385
1386 void
dsl_dataset_remove_from_next_clones(dsl_dataset_t * ds,uint64_t obj,dmu_tx_t * tx)1387 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1388 dmu_tx_t *tx)
1389 {
1390 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1391 uint64_t count __maybe_unused;
1392 int err;
1393
1394 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1395 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1396 obj, tx);
1397 /*
1398 * The err should not be ENOENT, but a bug in a previous version
1399 * of the code could cause upgrade_clones_cb() to not set
1400 * ds_next_snap_obj when it should, leading to a missing entry.
1401 * If we knew that the pool was created after
1402 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1403 * ENOENT. However, at least we can check that we don't have
1404 * too many entries in the next_clones_obj even after failing to
1405 * remove this one.
1406 */
1407 if (err != ENOENT)
1408 VERIFY0(err);
1409 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1410 &count));
1411 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1412 }
1413
1414
1415 blkptr_t *
dsl_dataset_get_blkptr(dsl_dataset_t * ds)1416 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1417 {
1418 return (&dsl_dataset_phys(ds)->ds_bp);
1419 }
1420
1421 spa_t *
dsl_dataset_get_spa(dsl_dataset_t * ds)1422 dsl_dataset_get_spa(dsl_dataset_t *ds)
1423 {
1424 return (ds->ds_dir->dd_pool->dp_spa);
1425 }
1426
1427 void
dsl_dataset_dirty(dsl_dataset_t * ds,dmu_tx_t * tx)1428 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1429 {
1430 dsl_pool_t *dp;
1431
1432 if (ds == NULL) /* this is the meta-objset */
1433 return;
1434
1435 ASSERT(ds->ds_objset != NULL);
1436
1437 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1438 panic("dirtying snapshot!");
1439
1440 /* Must not dirty a dataset in the same txg where it got snapshotted. */
1441 ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1442
1443 dp = ds->ds_dir->dd_pool;
1444 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1445 objset_t *os = ds->ds_objset;
1446
1447 /* up the hold count until we can be written out */
1448 dmu_buf_add_ref(ds->ds_dbuf, ds);
1449
1450 /* if this dataset is encrypted, grab a reference to the DCK */
1451 if (ds->ds_dir->dd_crypto_obj != 0 &&
1452 !os->os_raw_receive &&
1453 !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1454 ASSERT3P(ds->ds_key_mapping, !=, NULL);
1455 key_mapping_add_ref(ds->ds_key_mapping, ds);
1456 }
1457 }
1458 }
1459
1460 static int
dsl_dataset_snapshot_reserve_space(dsl_dataset_t * ds,dmu_tx_t * tx)1461 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1462 {
1463 uint64_t asize;
1464
1465 if (!dmu_tx_is_syncing(tx))
1466 return (0);
1467
1468 /*
1469 * If there's an fs-only reservation, any blocks that might become
1470 * owned by the snapshot dataset must be accommodated by space
1471 * outside of the reservation.
1472 */
1473 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1474 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1475 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1476 return (SET_ERROR(ENOSPC));
1477
1478 /*
1479 * Propagate any reserved space for this snapshot to other
1480 * snapshot checks in this sync group.
1481 */
1482 if (asize > 0)
1483 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1484
1485 return (0);
1486 }
1487
1488 int
dsl_dataset_snapshot_check_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx,boolean_t recv,uint64_t cnt,cred_t * cr,proc_t * proc)1489 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1490 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr, proc_t *proc)
1491 {
1492 int error;
1493 uint64_t value;
1494
1495 ds->ds_trysnap_txg = tx->tx_txg;
1496
1497 if (!dmu_tx_is_syncing(tx))
1498 return (0);
1499
1500 /*
1501 * We don't allow multiple snapshots of the same txg. If there
1502 * is already one, try again.
1503 */
1504 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1505 return (SET_ERROR(EAGAIN));
1506
1507 /*
1508 * Check for conflicting snapshot name.
1509 */
1510 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1511 if (error == 0)
1512 return (SET_ERROR(EEXIST));
1513 if (error != ENOENT)
1514 return (error);
1515
1516 /*
1517 * We don't allow taking snapshots of inconsistent datasets, such as
1518 * those into which we are currently receiving. However, if we are
1519 * creating this snapshot as part of a receive, this check will be
1520 * executed atomically with respect to the completion of the receive
1521 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1522 * case we ignore this, knowing it will be fixed up for us shortly in
1523 * dmu_recv_end_sync().
1524 */
1525 if (!recv && DS_IS_INCONSISTENT(ds))
1526 return (SET_ERROR(EBUSY));
1527
1528 /*
1529 * Skip the check for temporary snapshots or if we have already checked
1530 * the counts in dsl_dataset_snapshot_check. This means we really only
1531 * check the count here when we're receiving a stream.
1532 */
1533 if (cnt != 0 && cr != NULL) {
1534 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1535 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr, proc);
1536 if (error != 0)
1537 return (error);
1538 }
1539
1540 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1541 if (error != 0)
1542 return (error);
1543
1544 return (0);
1545 }
1546
1547 int
dsl_dataset_snapshot_check(void * arg,dmu_tx_t * tx)1548 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1549 {
1550 dsl_dataset_snapshot_arg_t *ddsa = arg;
1551 dsl_pool_t *dp = dmu_tx_pool(tx);
1552 nvpair_t *pair;
1553 int rv = 0;
1554
1555 /*
1556 * Pre-compute how many total new snapshots will be created for each
1557 * level in the tree and below. This is needed for validating the
1558 * snapshot limit when either taking a recursive snapshot or when
1559 * taking multiple snapshots.
1560 *
1561 * The problem is that the counts are not actually adjusted when
1562 * we are checking, only when we finally sync. For a single snapshot,
1563 * this is easy, the count will increase by 1 at each node up the tree,
1564 * but its more complicated for the recursive/multiple snapshot case.
1565 *
1566 * The dsl_fs_ss_limit_check function does recursively check the count
1567 * at each level up the tree but since it is validating each snapshot
1568 * independently we need to be sure that we are validating the complete
1569 * count for the entire set of snapshots. We do this by rolling up the
1570 * counts for each component of the name into an nvlist and then
1571 * checking each of those cases with the aggregated count.
1572 *
1573 * This approach properly handles not only the recursive snapshot
1574 * case (where we get all of those on the ddsa_snaps list) but also
1575 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1576 * validate the limit on 'a' using a count of 2).
1577 *
1578 * We validate the snapshot names in the third loop and only report
1579 * name errors once.
1580 */
1581 if (dmu_tx_is_syncing(tx)) {
1582 char *nm;
1583 nvlist_t *cnt_track = NULL;
1584 cnt_track = fnvlist_alloc();
1585
1586 nm = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1587
1588 /* Rollup aggregated counts into the cnt_track list */
1589 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1590 pair != NULL;
1591 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1592 char *pdelim;
1593 uint64_t val;
1594
1595 (void) strlcpy(nm, nvpair_name(pair), MAXPATHLEN);
1596 pdelim = strchr(nm, '@');
1597 if (pdelim == NULL)
1598 continue;
1599 *pdelim = '\0';
1600
1601 do {
1602 if (nvlist_lookup_uint64(cnt_track, nm,
1603 &val) == 0) {
1604 /* update existing entry */
1605 fnvlist_add_uint64(cnt_track, nm,
1606 val + 1);
1607 } else {
1608 /* add to list */
1609 fnvlist_add_uint64(cnt_track, nm, 1);
1610 }
1611
1612 pdelim = strrchr(nm, '/');
1613 if (pdelim != NULL)
1614 *pdelim = '\0';
1615 } while (pdelim != NULL);
1616 }
1617
1618 kmem_free(nm, MAXPATHLEN);
1619
1620 /* Check aggregated counts at each level */
1621 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1622 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1623 int error = 0;
1624 const char *name;
1625 uint64_t cnt = 0;
1626 dsl_dataset_t *ds;
1627
1628 name = nvpair_name(pair);
1629 cnt = fnvpair_value_uint64(pair);
1630 ASSERT(cnt > 0);
1631
1632 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1633 if (error == 0) {
1634 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1635 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1636 ddsa->ddsa_cr, ddsa->ddsa_proc);
1637 dsl_dataset_rele(ds, FTAG);
1638 }
1639
1640 if (error != 0) {
1641 if (ddsa->ddsa_errors != NULL)
1642 fnvlist_add_int32(ddsa->ddsa_errors,
1643 name, error);
1644 rv = error;
1645 /* only report one error for this check */
1646 break;
1647 }
1648 }
1649 nvlist_free(cnt_track);
1650 }
1651
1652 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1653 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1654 int error = 0;
1655 dsl_dataset_t *ds;
1656 const char *name, *atp = NULL;
1657 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1658
1659 name = nvpair_name(pair);
1660 if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1661 error = SET_ERROR(ENAMETOOLONG);
1662 if (error == 0) {
1663 atp = strchr(name, '@');
1664 if (atp == NULL)
1665 error = SET_ERROR(EINVAL);
1666 if (error == 0)
1667 (void) strlcpy(dsname, name, atp - name + 1);
1668 }
1669 if (error == 0)
1670 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1671 if (error == 0) {
1672 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1673 error = dsl_dataset_snapshot_check_impl(ds,
1674 atp + 1, tx, B_FALSE, 0, NULL, NULL);
1675 dsl_dataset_rele(ds, FTAG);
1676 }
1677
1678 if (error != 0) {
1679 if (ddsa->ddsa_errors != NULL) {
1680 fnvlist_add_int32(ddsa->ddsa_errors,
1681 name, error);
1682 }
1683 rv = error;
1684 }
1685 }
1686
1687 return (rv);
1688 }
1689
1690 void
dsl_dataset_snapshot_sync_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx)1691 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1692 dmu_tx_t *tx)
1693 {
1694 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1695 dmu_buf_t *dbuf;
1696 dsl_dataset_phys_t *dsphys;
1697 uint64_t dsobj, crtxg;
1698 objset_t *mos = dp->dp_meta_objset;
1699 objset_t *os __maybe_unused;
1700
1701 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1702
1703 /*
1704 * If we are on an old pool, the zil must not be active, in which
1705 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1706 */
1707 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1708 dmu_objset_from_ds(ds, &os) != 0 ||
1709 memcmp(&os->os_phys->os_zil_header, &zero_zil,
1710 sizeof (zero_zil)) == 0);
1711
1712 /* Should not snapshot a dirty dataset. */
1713 ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1714 ds, tx->tx_txg));
1715
1716 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1717
1718 /*
1719 * The origin's ds_creation_txg has to be < TXG_INITIAL
1720 */
1721 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1722 crtxg = 1;
1723 else
1724 crtxg = tx->tx_txg;
1725
1726 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1727 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1728 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1729 dmu_buf_will_dirty(dbuf, tx);
1730 dsphys = dbuf->db_data;
1731 memset(dsphys, 0, sizeof (dsl_dataset_phys_t));
1732 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1733 dsphys->ds_fsid_guid = unique_create();
1734 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1735 sizeof (dsphys->ds_guid));
1736 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1737 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1738 dsphys->ds_next_snap_obj = ds->ds_object;
1739 dsphys->ds_num_children = 1;
1740 dsphys->ds_creation_time = gethrestime_sec();
1741 dsphys->ds_creation_txg = crtxg;
1742 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1743 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1744 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1745 dsphys->ds_uncompressed_bytes =
1746 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1747 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1748 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1749 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1750 rrw_exit(&ds->ds_bp_rwlock, FTAG);
1751 dmu_buf_rele(dbuf, FTAG);
1752
1753 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1754 if (zfeature_active(f, ds->ds_feature[f])) {
1755 dsl_dataset_activate_feature(dsobj, f,
1756 ds->ds_feature[f], tx);
1757 }
1758 }
1759
1760 ASSERT3U(ds->ds_prev != 0, ==,
1761 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1762 if (ds->ds_prev) {
1763 uint64_t next_clones_obj =
1764 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1765 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1766 ds->ds_object ||
1767 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1768 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1769 ds->ds_object) {
1770 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1771 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1772 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1773 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1774 } else if (next_clones_obj != 0) {
1775 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1776 dsphys->ds_next_snap_obj, tx);
1777 VERIFY0(zap_add_int(mos,
1778 next_clones_obj, dsobj, tx));
1779 }
1780 }
1781
1782 /*
1783 * If we have a reference-reservation on this dataset, we will
1784 * need to increase the amount of refreservation being charged
1785 * since our unique space is going to zero.
1786 */
1787 if (ds->ds_reserved) {
1788 int64_t delta;
1789 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1790 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1791 ds->ds_reserved);
1792 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1793 delta, 0, 0, tx);
1794 }
1795
1796 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1797 dsl_dataset_phys(ds)->ds_deadlist_obj =
1798 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1799 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1800 dsl_deadlist_close(&ds->ds_deadlist);
1801 dsl_deadlist_open(&ds->ds_deadlist, mos,
1802 dsl_dataset_phys(ds)->ds_deadlist_obj);
1803 dsl_deadlist_add_key(&ds->ds_deadlist,
1804 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1805 dsl_bookmark_snapshotted(ds, tx);
1806
1807 if (dsl_dataset_remap_deadlist_exists(ds)) {
1808 uint64_t remap_deadlist_obj =
1809 dsl_dataset_get_remap_deadlist_object(ds);
1810 /*
1811 * Move the remap_deadlist to the snapshot. The head
1812 * will create a new remap deadlist on demand, from
1813 * dsl_dataset_block_remapped().
1814 */
1815 dsl_dataset_unset_remap_deadlist_object(ds, tx);
1816 dsl_deadlist_close(&ds->ds_remap_deadlist);
1817
1818 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1819 VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1820 sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1821 }
1822
1823 /*
1824 * Create a ivset guid for this snapshot if the dataset is
1825 * encrypted. This may be overridden by a raw receive. A
1826 * previous implementation of this code did not have this
1827 * field as part of the on-disk format for ZFS encryption
1828 * (see errata #4). As part of the remediation for this
1829 * issue, we ask the user to enable the bookmark_v2 feature
1830 * which is now a dependency of the encryption feature. We
1831 * use this as a heuristic to determine when the user has
1832 * elected to correct any datasets created with the old code.
1833 * As a result, we only do this step if the bookmark_v2
1834 * feature is enabled, which limits the number of states a
1835 * given pool / dataset can be in with regards to terms of
1836 * correcting the issue.
1837 */
1838 if (ds->ds_dir->dd_crypto_obj != 0 &&
1839 spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) {
1840 uint64_t ivset_guid = unique_create();
1841
1842 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1843 VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID,
1844 sizeof (ivset_guid), 1, &ivset_guid, tx));
1845 }
1846
1847 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1848 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1849 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1850 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1851
1852 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1853 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1854
1855 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1856 snapname, 8, 1, &dsobj, tx));
1857
1858 if (ds->ds_prev)
1859 dsl_dataset_rele(ds->ds_prev, ds);
1860 VERIFY0(dsl_dataset_hold_obj(dp,
1861 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1862
1863 dsl_scan_ds_snapshotted(ds, tx);
1864
1865 dsl_dir_snap_cmtime_update(ds->ds_dir, tx);
1866
1867 if (zfs_snapshot_history_enabled)
1868 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, " ");
1869 }
1870
1871 void
dsl_dataset_snapshot_sync(void * arg,dmu_tx_t * tx)1872 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1873 {
1874 dsl_dataset_snapshot_arg_t *ddsa = arg;
1875 dsl_pool_t *dp = dmu_tx_pool(tx);
1876 nvpair_t *pair;
1877
1878 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1879 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1880 dsl_dataset_t *ds;
1881 const char *name, *atp;
1882 char dsname[ZFS_MAX_DATASET_NAME_LEN];
1883
1884 name = nvpair_name(pair);
1885 atp = strchr(name, '@');
1886 (void) strlcpy(dsname, name, atp - name + 1);
1887 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1888
1889 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1890 if (ddsa->ddsa_props != NULL) {
1891 dsl_props_set_sync_impl(ds->ds_prev,
1892 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1893 }
1894 dsl_dataset_rele(ds, FTAG);
1895 }
1896 }
1897
1898 /*
1899 * The snapshots must all be in the same pool.
1900 * All-or-nothing: if there are any failures, nothing will be modified.
1901 */
1902 int
dsl_dataset_snapshot(nvlist_t * snaps,nvlist_t * props,nvlist_t * errors)1903 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1904 {
1905 dsl_dataset_snapshot_arg_t ddsa;
1906 nvpair_t *pair;
1907 boolean_t needsuspend;
1908 int error;
1909 spa_t *spa;
1910 const char *firstname;
1911 nvlist_t *suspended = NULL;
1912
1913 pair = nvlist_next_nvpair(snaps, NULL);
1914 if (pair == NULL)
1915 return (0);
1916 firstname = nvpair_name(pair);
1917
1918 error = spa_open(firstname, &spa, FTAG);
1919 if (error != 0)
1920 return (error);
1921 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1922 spa_close(spa, FTAG);
1923
1924 if (needsuspend) {
1925 suspended = fnvlist_alloc();
1926 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1927 pair = nvlist_next_nvpair(snaps, pair)) {
1928 char fsname[ZFS_MAX_DATASET_NAME_LEN];
1929 const char *snapname = nvpair_name(pair);
1930 const char *atp;
1931 void *cookie;
1932
1933 atp = strchr(snapname, '@');
1934 if (atp == NULL) {
1935 error = SET_ERROR(EINVAL);
1936 break;
1937 }
1938 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1939
1940 error = zil_suspend(fsname, &cookie);
1941 if (error != 0)
1942 break;
1943 fnvlist_add_uint64(suspended, fsname,
1944 (uintptr_t)cookie);
1945 }
1946 }
1947
1948 ddsa.ddsa_snaps = snaps;
1949 ddsa.ddsa_props = props;
1950 ddsa.ddsa_errors = errors;
1951 ddsa.ddsa_cr = CRED();
1952 ddsa.ddsa_proc = curproc;
1953
1954 if (error == 0) {
1955 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1956 dsl_dataset_snapshot_sync, &ddsa,
1957 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1958 }
1959
1960 if (suspended != NULL) {
1961 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1962 pair = nvlist_next_nvpair(suspended, pair)) {
1963 zil_resume((void *)(uintptr_t)
1964 fnvpair_value_uint64(pair));
1965 }
1966 fnvlist_free(suspended);
1967 }
1968
1969 if (error == 0) {
1970 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1971 pair = nvlist_next_nvpair(snaps, pair)) {
1972 zvol_create_minor(nvpair_name(pair));
1973 }
1974 }
1975
1976 return (error);
1977 }
1978
1979 typedef struct dsl_dataset_snapshot_tmp_arg {
1980 const char *ddsta_fsname;
1981 const char *ddsta_snapname;
1982 minor_t ddsta_cleanup_minor;
1983 const char *ddsta_htag;
1984 } dsl_dataset_snapshot_tmp_arg_t;
1985
1986 static int
dsl_dataset_snapshot_tmp_check(void * arg,dmu_tx_t * tx)1987 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1988 {
1989 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1990 dsl_pool_t *dp = dmu_tx_pool(tx);
1991 dsl_dataset_t *ds;
1992 int error;
1993
1994 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1995 if (error != 0)
1996 return (error);
1997
1998 /* NULL cred means no limit check for tmp snapshot */
1999 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
2000 tx, B_FALSE, 0, NULL, NULL);
2001 if (error != 0) {
2002 dsl_dataset_rele(ds, FTAG);
2003 return (error);
2004 }
2005
2006 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
2007 dsl_dataset_rele(ds, FTAG);
2008 return (SET_ERROR(ENOTSUP));
2009 }
2010 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
2011 B_TRUE, tx);
2012 if (error != 0) {
2013 dsl_dataset_rele(ds, FTAG);
2014 return (error);
2015 }
2016
2017 dsl_dataset_rele(ds, FTAG);
2018 return (0);
2019 }
2020
2021 static void
dsl_dataset_snapshot_tmp_sync(void * arg,dmu_tx_t * tx)2022 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
2023 {
2024 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
2025 dsl_pool_t *dp = dmu_tx_pool(tx);
2026 dsl_dataset_t *ds = NULL;
2027
2028 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
2029
2030 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
2031 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
2032 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
2033 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
2034
2035 dsl_dataset_rele(ds, FTAG);
2036 }
2037
2038 int
dsl_dataset_snapshot_tmp(const char * fsname,const char * snapname,minor_t cleanup_minor,const char * htag)2039 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
2040 minor_t cleanup_minor, const char *htag)
2041 {
2042 dsl_dataset_snapshot_tmp_arg_t ddsta;
2043 int error;
2044 spa_t *spa;
2045 boolean_t needsuspend;
2046 void *cookie;
2047
2048 ddsta.ddsta_fsname = fsname;
2049 ddsta.ddsta_snapname = snapname;
2050 ddsta.ddsta_cleanup_minor = cleanup_minor;
2051 ddsta.ddsta_htag = htag;
2052
2053 error = spa_open(fsname, &spa, FTAG);
2054 if (error != 0)
2055 return (error);
2056 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
2057 spa_close(spa, FTAG);
2058
2059 if (needsuspend) {
2060 error = zil_suspend(fsname, &cookie);
2061 if (error != 0)
2062 return (error);
2063 }
2064
2065 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
2066 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
2067
2068 if (needsuspend)
2069 zil_resume(cookie);
2070 return (error);
2071 }
2072
2073 /* Nonblocking dataset sync. Assumes dataset:objset is always 1:1 */
2074 void
dsl_dataset_sync(dsl_dataset_t * ds,zio_t * rio,dmu_tx_t * tx)2075 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *rio, dmu_tx_t *tx)
2076 {
2077 ASSERT(dmu_tx_is_syncing(tx));
2078 ASSERT(ds->ds_objset != NULL);
2079 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
2080
2081 /*
2082 * in case we had to change ds_fsid_guid when we opened it,
2083 * sync it out now.
2084 */
2085 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2086 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
2087
2088 if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
2089 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2090 ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
2091 &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
2092 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2093 ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
2094 &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
2095 VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
2096 ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
2097 &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
2098 ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
2099 ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
2100 ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
2101 }
2102
2103 dmu_objset_sync(ds->ds_objset, rio, tx);
2104 }
2105
2106 /*
2107 * Check if the percentage of blocks shared between the clone and the
2108 * snapshot (as opposed to those that are clone only) is below a certain
2109 * threshold
2110 */
2111 static boolean_t
dsl_livelist_should_disable(dsl_dataset_t * ds)2112 dsl_livelist_should_disable(dsl_dataset_t *ds)
2113 {
2114 uint64_t used, referenced;
2115 int percent_shared;
2116
2117 used = dsl_dir_get_usedds(ds->ds_dir);
2118 referenced = dsl_get_referenced(ds);
2119 if (referenced == 0)
2120 return (B_FALSE);
2121 percent_shared = (100 * (referenced - used)) / referenced;
2122 if (percent_shared <= zfs_livelist_min_percent_shared)
2123 return (B_TRUE);
2124 return (B_FALSE);
2125 }
2126
2127 /*
2128 * Check if it is possible to combine two livelist entries into one.
2129 * This is the case if the combined number of 'live' blkptrs (ALLOCs that
2130 * don't have a matching FREE) is under the maximum sublist size.
2131 * We check this by subtracting twice the total number of frees from the total
2132 * number of blkptrs. FREEs are counted twice because each FREE blkptr
2133 * will cancel out an ALLOC blkptr when the livelist is processed.
2134 */
2135 static boolean_t
dsl_livelist_should_condense(dsl_deadlist_entry_t * first,dsl_deadlist_entry_t * next)2136 dsl_livelist_should_condense(dsl_deadlist_entry_t *first,
2137 dsl_deadlist_entry_t *next)
2138 {
2139 uint64_t total_free = first->dle_bpobj.bpo_phys->bpo_num_freed +
2140 next->dle_bpobj.bpo_phys->bpo_num_freed;
2141 uint64_t total_entries = first->dle_bpobj.bpo_phys->bpo_num_blkptrs +
2142 next->dle_bpobj.bpo_phys->bpo_num_blkptrs;
2143 if ((total_entries - (2 * total_free)) < zfs_livelist_max_entries)
2144 return (B_TRUE);
2145 return (B_FALSE);
2146 }
2147
2148 typedef struct try_condense_arg {
2149 spa_t *spa;
2150 dsl_dataset_t *ds;
2151 } try_condense_arg_t;
2152
2153 /*
2154 * Iterate over the livelist entries, searching for a pair to condense.
2155 * A nonzero return value means stop, 0 means keep looking.
2156 */
2157 static int
dsl_livelist_try_condense(void * arg,dsl_deadlist_entry_t * first)2158 dsl_livelist_try_condense(void *arg, dsl_deadlist_entry_t *first)
2159 {
2160 try_condense_arg_t *tca = arg;
2161 spa_t *spa = tca->spa;
2162 dsl_dataset_t *ds = tca->ds;
2163 dsl_deadlist_t *ll = &ds->ds_dir->dd_livelist;
2164 dsl_deadlist_entry_t *next;
2165
2166 /* The condense thread has not yet been created at import */
2167 if (spa->spa_livelist_condense_zthr == NULL)
2168 return (1);
2169
2170 /* A condense is already in progress */
2171 if (spa->spa_to_condense.ds != NULL)
2172 return (1);
2173
2174 next = AVL_NEXT(&ll->dl_tree, &first->dle_node);
2175 /* The livelist has only one entry - don't condense it */
2176 if (next == NULL)
2177 return (1);
2178
2179 /* Next is the newest entry - don't condense it */
2180 if (AVL_NEXT(&ll->dl_tree, &next->dle_node) == NULL)
2181 return (1);
2182
2183 /* This pair is not ready to condense but keep looking */
2184 if (!dsl_livelist_should_condense(first, next))
2185 return (0);
2186
2187 /*
2188 * Add a ref to prevent the dataset from being evicted while
2189 * the condense zthr or synctask are running. Ref will be
2190 * released at the end of the condense synctask
2191 */
2192 dmu_buf_add_ref(ds->ds_dbuf, spa);
2193
2194 spa->spa_to_condense.ds = ds;
2195 spa->spa_to_condense.first = first;
2196 spa->spa_to_condense.next = next;
2197 spa->spa_to_condense.syncing = B_FALSE;
2198 spa->spa_to_condense.cancelled = B_FALSE;
2199
2200 zthr_wakeup(spa->spa_livelist_condense_zthr);
2201 return (1);
2202 }
2203
2204 static void
dsl_flush_pending_livelist(dsl_dataset_t * ds,dmu_tx_t * tx)2205 dsl_flush_pending_livelist(dsl_dataset_t *ds, dmu_tx_t *tx)
2206 {
2207 dsl_dir_t *dd = ds->ds_dir;
2208 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
2209 dsl_deadlist_entry_t *last = dsl_deadlist_last(&dd->dd_livelist);
2210
2211 /* Check if we need to add a new sub-livelist */
2212 if (last == NULL) {
2213 /* The livelist is empty */
2214 dsl_deadlist_add_key(&dd->dd_livelist,
2215 tx->tx_txg - 1, tx);
2216 } else if (spa_sync_pass(spa) == 1) {
2217 /*
2218 * Check if the newest entry is full. If it is, make a new one.
2219 * We only do this once per sync because we could overfill a
2220 * sublist in one sync pass and don't want to add another entry
2221 * for a txg that is already represented. This ensures that
2222 * blkptrs born in the same txg are stored in the same sublist.
2223 */
2224 bpobj_t bpobj = last->dle_bpobj;
2225 uint64_t all = bpobj.bpo_phys->bpo_num_blkptrs;
2226 uint64_t free = bpobj.bpo_phys->bpo_num_freed;
2227 uint64_t alloc = all - free;
2228 if (alloc > zfs_livelist_max_entries) {
2229 dsl_deadlist_add_key(&dd->dd_livelist,
2230 tx->tx_txg - 1, tx);
2231 }
2232 }
2233
2234 /* Insert each entry into the on-disk livelist */
2235 bplist_iterate(&dd->dd_pending_allocs,
2236 dsl_deadlist_insert_alloc_cb, &dd->dd_livelist, tx);
2237 bplist_iterate(&dd->dd_pending_frees,
2238 dsl_deadlist_insert_free_cb, &dd->dd_livelist, tx);
2239
2240 /* Attempt to condense every pair of adjacent entries */
2241 try_condense_arg_t arg = {
2242 .spa = spa,
2243 .ds = ds
2244 };
2245 dsl_deadlist_iterate(&dd->dd_livelist, dsl_livelist_try_condense,
2246 &arg);
2247 }
2248
2249 void
dsl_dataset_sync_done(dsl_dataset_t * ds,dmu_tx_t * tx)2250 dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
2251 {
2252 objset_t *os = ds->ds_objset;
2253
2254 bplist_iterate(&ds->ds_pending_deadlist,
2255 dsl_deadlist_insert_alloc_cb, &ds->ds_deadlist, tx);
2256
2257 if (dsl_deadlist_is_open(&ds->ds_dir->dd_livelist)) {
2258 dsl_flush_pending_livelist(ds, tx);
2259 if (dsl_livelist_should_disable(ds)) {
2260 dsl_dir_remove_livelist(ds->ds_dir, tx, B_TRUE);
2261 }
2262 }
2263
2264 dsl_bookmark_sync_done(ds, tx);
2265
2266 multilist_destroy(&os->os_synced_dnodes);
2267
2268 if (os->os_encrypted)
2269 os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE;
2270 else
2271 ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]);
2272
2273 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
2274 if (zfeature_active(f,
2275 ds->ds_feature_activation[f])) {
2276 if (zfeature_active(f, ds->ds_feature[f]))
2277 continue;
2278 dsl_dataset_activate_feature(ds->ds_object, f,
2279 ds->ds_feature_activation[f], tx);
2280 ds->ds_feature[f] = ds->ds_feature_activation[f];
2281 }
2282 }
2283
2284 ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
2285 }
2286
2287 int
get_clones_stat_impl(dsl_dataset_t * ds,nvlist_t * val)2288 get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
2289 {
2290 uint64_t count = 0;
2291 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2292 zap_cursor_t zc;
2293 zap_attribute_t za;
2294
2295 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
2296
2297 /*
2298 * There may be missing entries in ds_next_clones_obj
2299 * due to a bug in a previous version of the code.
2300 * Only trust it if it has the right number of entries.
2301 */
2302 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
2303 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
2304 &count));
2305 }
2306 if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
2307 return (SET_ERROR(ENOENT));
2308 }
2309 for (zap_cursor_init(&zc, mos,
2310 dsl_dataset_phys(ds)->ds_next_clones_obj);
2311 zap_cursor_retrieve(&zc, &za) == 0;
2312 zap_cursor_advance(&zc)) {
2313 dsl_dataset_t *clone;
2314 char buf[ZFS_MAX_DATASET_NAME_LEN];
2315 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2316 za.za_first_integer, FTAG, &clone));
2317 dsl_dir_name(clone->ds_dir, buf);
2318 fnvlist_add_boolean(val, buf);
2319 dsl_dataset_rele(clone, FTAG);
2320 }
2321 zap_cursor_fini(&zc);
2322 return (0);
2323 }
2324
2325 void
get_clones_stat(dsl_dataset_t * ds,nvlist_t * nv)2326 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2327 {
2328 nvlist_t *propval = fnvlist_alloc();
2329 nvlist_t *val = fnvlist_alloc();
2330
2331 if (get_clones_stat_impl(ds, val) == 0) {
2332 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
2333 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2334 propval);
2335 }
2336
2337 nvlist_free(val);
2338 nvlist_free(propval);
2339 }
2340
2341 static char *
get_receive_resume_token_impl(dsl_dataset_t * ds)2342 get_receive_resume_token_impl(dsl_dataset_t *ds)
2343 {
2344 if (!dsl_dataset_has_resume_receive_state(ds))
2345 return (NULL);
2346
2347 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2348 char *str;
2349 void *packed;
2350 uint8_t *compressed;
2351 uint64_t val;
2352 nvlist_t *token_nv = fnvlist_alloc();
2353 size_t packed_size, compressed_size;
2354
2355 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2356 DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
2357 fnvlist_add_uint64(token_nv, "fromguid", val);
2358 }
2359 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2360 DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
2361 fnvlist_add_uint64(token_nv, "object", val);
2362 }
2363 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2364 DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
2365 fnvlist_add_uint64(token_nv, "offset", val);
2366 }
2367 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2368 DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
2369 fnvlist_add_uint64(token_nv, "bytes", val);
2370 }
2371 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2372 DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
2373 fnvlist_add_uint64(token_nv, "toguid", val);
2374 }
2375 char buf[MAXNAMELEN];
2376 if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
2377 DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
2378 fnvlist_add_string(token_nv, "toname", buf);
2379 }
2380 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2381 DS_FIELD_RESUME_LARGEBLOCK) == 0) {
2382 fnvlist_add_boolean(token_nv, "largeblockok");
2383 }
2384 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2385 DS_FIELD_RESUME_EMBEDOK) == 0) {
2386 fnvlist_add_boolean(token_nv, "embedok");
2387 }
2388 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2389 DS_FIELD_RESUME_COMPRESSOK) == 0) {
2390 fnvlist_add_boolean(token_nv, "compressok");
2391 }
2392 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2393 DS_FIELD_RESUME_RAWOK) == 0) {
2394 fnvlist_add_boolean(token_nv, "rawok");
2395 }
2396 if (dsl_dataset_feature_is_active(ds,
2397 SPA_FEATURE_REDACTED_DATASETS)) {
2398 uint64_t num_redact_snaps = 0;
2399 uint64_t *redact_snaps = NULL;
2400 VERIFY3B(dsl_dataset_get_uint64_array_feature(ds,
2401 SPA_FEATURE_REDACTED_DATASETS, &num_redact_snaps,
2402 &redact_snaps), ==, B_TRUE);
2403 fnvlist_add_uint64_array(token_nv, "redact_snaps",
2404 redact_snaps, num_redact_snaps);
2405 }
2406 if (zap_contains(dp->dp_meta_objset, ds->ds_object,
2407 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS) == 0) {
2408 uint64_t num_redact_snaps = 0, int_size = 0;
2409 uint64_t *redact_snaps = NULL;
2410 VERIFY0(zap_length(dp->dp_meta_objset, ds->ds_object,
2411 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, &int_size,
2412 &num_redact_snaps));
2413 ASSERT3U(int_size, ==, sizeof (uint64_t));
2414
2415 redact_snaps = kmem_alloc(int_size * num_redact_snaps,
2416 KM_SLEEP);
2417 VERIFY0(zap_lookup(dp->dp_meta_objset, ds->ds_object,
2418 DS_FIELD_RESUME_REDACT_BOOKMARK_SNAPS, int_size,
2419 num_redact_snaps, redact_snaps));
2420 fnvlist_add_uint64_array(token_nv, "book_redact_snaps",
2421 redact_snaps, num_redact_snaps);
2422 kmem_free(redact_snaps, int_size * num_redact_snaps);
2423 }
2424 packed = fnvlist_pack(token_nv, &packed_size);
2425 fnvlist_free(token_nv);
2426 compressed = kmem_alloc(packed_size, KM_SLEEP);
2427
2428 compressed_size = gzip_compress(packed, compressed,
2429 packed_size, packed_size, 6);
2430
2431 zio_cksum_t cksum;
2432 fletcher_4_native_varsize(compressed, compressed_size, &cksum);
2433
2434 size_t alloc_size = compressed_size * 2 + 1;
2435 str = kmem_alloc(alloc_size, KM_SLEEP);
2436 for (int i = 0; i < compressed_size; i++) {
2437 size_t offset = i * 2;
2438 (void) snprintf(str + offset, alloc_size - offset,
2439 "%02x", compressed[i]);
2440 }
2441 str[compressed_size * 2] = '\0';
2442 char *propval = kmem_asprintf("%u-%llx-%llx-%s",
2443 ZFS_SEND_RESUME_TOKEN_VERSION,
2444 (longlong_t)cksum.zc_word[0],
2445 (longlong_t)packed_size, str);
2446 kmem_free(packed, packed_size);
2447 kmem_free(str, alloc_size);
2448 kmem_free(compressed, packed_size);
2449 return (propval);
2450 }
2451
2452 /*
2453 * Returns a string that represents the receive resume state token. It should
2454 * be freed with strfree(). NULL is returned if no resume state is present.
2455 */
2456 char *
get_receive_resume_token(dsl_dataset_t * ds)2457 get_receive_resume_token(dsl_dataset_t *ds)
2458 {
2459 /*
2460 * A failed "newfs" (e.g. full) resumable receive leaves
2461 * the stats set on this dataset. Check here for the prop.
2462 */
2463 char *token = get_receive_resume_token_impl(ds);
2464 if (token != NULL)
2465 return (token);
2466 /*
2467 * A failed incremental resumable receive leaves the
2468 * stats set on our child named "%recv". Check the child
2469 * for the prop.
2470 */
2471 /* 6 extra bytes for /%recv */
2472 char name[ZFS_MAX_DATASET_NAME_LEN + 6];
2473 dsl_dataset_t *recv_ds;
2474 dsl_dataset_name(ds, name);
2475 if (strlcat(name, "/", sizeof (name)) < sizeof (name) &&
2476 strlcat(name, recv_clone_name, sizeof (name)) < sizeof (name) &&
2477 dsl_dataset_hold(ds->ds_dir->dd_pool, name, FTAG, &recv_ds) == 0) {
2478 token = get_receive_resume_token_impl(recv_ds);
2479 dsl_dataset_rele(recv_ds, FTAG);
2480 }
2481 return (token);
2482 }
2483
2484 uint64_t
dsl_get_refratio(dsl_dataset_t * ds)2485 dsl_get_refratio(dsl_dataset_t *ds)
2486 {
2487 uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
2488 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
2489 dsl_dataset_phys(ds)->ds_compressed_bytes);
2490 return (ratio);
2491 }
2492
2493 uint64_t
dsl_get_logicalreferenced(dsl_dataset_t * ds)2494 dsl_get_logicalreferenced(dsl_dataset_t *ds)
2495 {
2496 return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2497 }
2498
2499 uint64_t
dsl_get_compressratio(dsl_dataset_t * ds)2500 dsl_get_compressratio(dsl_dataset_t *ds)
2501 {
2502 if (ds->ds_is_snapshot) {
2503 return (dsl_get_refratio(ds));
2504 } else {
2505 dsl_dir_t *dd = ds->ds_dir;
2506 mutex_enter(&dd->dd_lock);
2507 uint64_t val = dsl_dir_get_compressratio(dd);
2508 mutex_exit(&dd->dd_lock);
2509 return (val);
2510 }
2511 }
2512
2513 uint64_t
dsl_get_used(dsl_dataset_t * ds)2514 dsl_get_used(dsl_dataset_t *ds)
2515 {
2516 if (ds->ds_is_snapshot) {
2517 return (dsl_dataset_phys(ds)->ds_unique_bytes);
2518 } else {
2519 dsl_dir_t *dd = ds->ds_dir;
2520 mutex_enter(&dd->dd_lock);
2521 uint64_t val = dsl_dir_get_used(dd);
2522 mutex_exit(&dd->dd_lock);
2523 return (val);
2524 }
2525 }
2526
2527 uint64_t
dsl_get_creation(dsl_dataset_t * ds)2528 dsl_get_creation(dsl_dataset_t *ds)
2529 {
2530 return (dsl_dataset_phys(ds)->ds_creation_time);
2531 }
2532
2533 uint64_t
dsl_get_creationtxg(dsl_dataset_t * ds)2534 dsl_get_creationtxg(dsl_dataset_t *ds)
2535 {
2536 return (dsl_dataset_phys(ds)->ds_creation_txg);
2537 }
2538
2539 uint64_t
dsl_get_refquota(dsl_dataset_t * ds)2540 dsl_get_refquota(dsl_dataset_t *ds)
2541 {
2542 return (ds->ds_quota);
2543 }
2544
2545 uint64_t
dsl_get_refreservation(dsl_dataset_t * ds)2546 dsl_get_refreservation(dsl_dataset_t *ds)
2547 {
2548 return (ds->ds_reserved);
2549 }
2550
2551 uint64_t
dsl_get_guid(dsl_dataset_t * ds)2552 dsl_get_guid(dsl_dataset_t *ds)
2553 {
2554 return (dsl_dataset_phys(ds)->ds_guid);
2555 }
2556
2557 uint64_t
dsl_get_unique(dsl_dataset_t * ds)2558 dsl_get_unique(dsl_dataset_t *ds)
2559 {
2560 return (dsl_dataset_phys(ds)->ds_unique_bytes);
2561 }
2562
2563 uint64_t
dsl_get_objsetid(dsl_dataset_t * ds)2564 dsl_get_objsetid(dsl_dataset_t *ds)
2565 {
2566 return (ds->ds_object);
2567 }
2568
2569 uint64_t
dsl_get_userrefs(dsl_dataset_t * ds)2570 dsl_get_userrefs(dsl_dataset_t *ds)
2571 {
2572 return (ds->ds_userrefs);
2573 }
2574
2575 uint64_t
dsl_get_defer_destroy(dsl_dataset_t * ds)2576 dsl_get_defer_destroy(dsl_dataset_t *ds)
2577 {
2578 return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2579 }
2580
2581 uint64_t
dsl_get_referenced(dsl_dataset_t * ds)2582 dsl_get_referenced(dsl_dataset_t *ds)
2583 {
2584 return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2585 }
2586
2587 uint64_t
dsl_get_numclones(dsl_dataset_t * ds)2588 dsl_get_numclones(dsl_dataset_t *ds)
2589 {
2590 ASSERT(ds->ds_is_snapshot);
2591 return (dsl_dataset_phys(ds)->ds_num_children - 1);
2592 }
2593
2594 uint64_t
dsl_get_inconsistent(dsl_dataset_t * ds)2595 dsl_get_inconsistent(dsl_dataset_t *ds)
2596 {
2597 return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2598 1 : 0);
2599 }
2600
2601 uint64_t
dsl_get_redacted(dsl_dataset_t * ds)2602 dsl_get_redacted(dsl_dataset_t *ds)
2603 {
2604 return (dsl_dataset_feature_is_active(ds,
2605 SPA_FEATURE_REDACTED_DATASETS));
2606 }
2607
2608 uint64_t
dsl_get_available(dsl_dataset_t * ds)2609 dsl_get_available(dsl_dataset_t *ds)
2610 {
2611 uint64_t refdbytes = dsl_get_referenced(ds);
2612 uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2613 NULL, 0, TRUE);
2614 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2615 availbytes +=
2616 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2617 }
2618 if (ds->ds_quota != 0) {
2619 /*
2620 * Adjust available bytes according to refquota
2621 */
2622 if (refdbytes < ds->ds_quota) {
2623 availbytes = MIN(availbytes,
2624 ds->ds_quota - refdbytes);
2625 } else {
2626 availbytes = 0;
2627 }
2628 }
2629 return (availbytes);
2630 }
2631
2632 int
dsl_get_written(dsl_dataset_t * ds,uint64_t * written)2633 dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2634 {
2635 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2636 dsl_dataset_t *prev;
2637 int err = dsl_dataset_hold_obj(dp,
2638 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2639 if (err == 0) {
2640 uint64_t comp, uncomp;
2641 err = dsl_dataset_space_written(prev, ds, written,
2642 &comp, &uncomp);
2643 dsl_dataset_rele(prev, FTAG);
2644 }
2645 return (err);
2646 }
2647
2648 /*
2649 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2650 */
2651 int
dsl_get_prev_snap(dsl_dataset_t * ds,char * snap)2652 dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2653 {
2654 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2655 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2656 dsl_dataset_name(ds->ds_prev, snap);
2657 return (0);
2658 } else {
2659 return (SET_ERROR(ENOENT));
2660 }
2661 }
2662
2663 void
dsl_get_redact_snaps(dsl_dataset_t * ds,nvlist_t * propval)2664 dsl_get_redact_snaps(dsl_dataset_t *ds, nvlist_t *propval)
2665 {
2666 uint64_t nsnaps;
2667 uint64_t *snaps;
2668 if (dsl_dataset_get_uint64_array_feature(ds,
2669 SPA_FEATURE_REDACTED_DATASETS, &nsnaps, &snaps)) {
2670 fnvlist_add_uint64_array(propval, ZPROP_VALUE, snaps,
2671 nsnaps);
2672 }
2673 }
2674
2675 /*
2676 * Returns the mountpoint property and source for the given dataset in the value
2677 * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2678 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2679 * Returns 0 on success and an error on failure.
2680 */
2681 int
dsl_get_mountpoint(dsl_dataset_t * ds,const char * dsname,char * value,char * source)2682 dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2683 char *source)
2684 {
2685 int error;
2686 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2687
2688 /* Retrieve the mountpoint value stored in the zap object */
2689 error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2690 ZAP_MAXVALUELEN, value, source);
2691 if (error != 0) {
2692 return (error);
2693 }
2694
2695 /*
2696 * Process the dsname and source to find the full mountpoint string.
2697 * Can be skipped for 'legacy' or 'none'.
2698 */
2699 if (value[0] == '/') {
2700 char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2701 char *root = buf;
2702 const char *relpath;
2703
2704 /*
2705 * If we inherit the mountpoint, even from a dataset
2706 * with a received value, the source will be the path of
2707 * the dataset we inherit from. If source is
2708 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2709 * inherited.
2710 */
2711 if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2712 relpath = "";
2713 } else {
2714 ASSERT0(strncmp(dsname, source, strlen(source)));
2715 relpath = dsname + strlen(source);
2716 if (relpath[0] == '/')
2717 relpath++;
2718 }
2719
2720 spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2721
2722 /*
2723 * Special case an alternate root of '/'. This will
2724 * avoid having multiple leading slashes in the
2725 * mountpoint path.
2726 */
2727 if (strcmp(root, "/") == 0)
2728 root++;
2729
2730 /*
2731 * If the mountpoint is '/' then skip over this
2732 * if we are obtaining either an alternate root or
2733 * an inherited mountpoint.
2734 */
2735 char *mnt = value;
2736 if (value[1] == '\0' && (root[0] != '\0' ||
2737 relpath[0] != '\0'))
2738 mnt = value + 1;
2739
2740 mnt = kmem_strdup(mnt);
2741
2742 if (relpath[0] == '\0') {
2743 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2744 root, mnt);
2745 } else {
2746 (void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2747 root, mnt, relpath[0] == '@' ? "" : "/",
2748 relpath);
2749 }
2750 kmem_free(buf, ZAP_MAXVALUELEN);
2751 kmem_strfree(mnt);
2752 }
2753
2754 return (0);
2755 }
2756
2757 void
dsl_dataset_stats(dsl_dataset_t * ds,nvlist_t * nv)2758 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2759 {
2760 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2761
2762 ASSERT(dsl_pool_config_held(dp));
2763
2764 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2765 dsl_get_refratio(ds));
2766 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2767 dsl_get_logicalreferenced(ds));
2768 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2769 dsl_get_compressratio(ds));
2770 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2771 dsl_get_used(ds));
2772
2773 if (ds->ds_is_snapshot) {
2774 get_clones_stat(ds, nv);
2775 } else {
2776 char buf[ZFS_MAX_DATASET_NAME_LEN];
2777 if (dsl_get_prev_snap(ds, buf) == 0)
2778 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2779 buf);
2780 dsl_dir_stats(ds->ds_dir, nv);
2781 }
2782
2783 nvlist_t *propval = fnvlist_alloc();
2784 dsl_get_redact_snaps(ds, propval);
2785 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_REDACT_SNAPS),
2786 propval);
2787 nvlist_free(propval);
2788
2789 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2790 dsl_get_available(ds));
2791 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2792 dsl_get_referenced(ds));
2793 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2794 dsl_get_creation(ds));
2795 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2796 dsl_get_creationtxg(ds));
2797 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2798 dsl_get_refquota(ds));
2799 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2800 dsl_get_refreservation(ds));
2801 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2802 dsl_get_guid(ds));
2803 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2804 dsl_get_unique(ds));
2805 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2806 dsl_get_objsetid(ds));
2807 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2808 dsl_get_userrefs(ds));
2809 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2810 dsl_get_defer_destroy(ds));
2811 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_SNAPSHOTS_CHANGED,
2812 dsl_dir_snap_cmtime(ds->ds_dir).tv_sec);
2813 dsl_dataset_crypt_stats(ds, nv);
2814
2815 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2816 uint64_t written;
2817 if (dsl_get_written(ds, &written) == 0) {
2818 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2819 written);
2820 }
2821 }
2822
2823 if (!dsl_dataset_is_snapshot(ds)) {
2824 char *token = get_receive_resume_token(ds);
2825 if (token != NULL) {
2826 dsl_prop_nvlist_add_string(nv,
2827 ZFS_PROP_RECEIVE_RESUME_TOKEN, token);
2828 kmem_strfree(token);
2829 }
2830 }
2831 }
2832
2833 void
dsl_dataset_fast_stat(dsl_dataset_t * ds,dmu_objset_stats_t * stat)2834 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2835 {
2836 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2837 ASSERT(dsl_pool_config_held(dp));
2838
2839 stat->dds_creation_txg = dsl_get_creationtxg(ds);
2840 stat->dds_inconsistent = dsl_get_inconsistent(ds);
2841 stat->dds_guid = dsl_get_guid(ds);
2842 stat->dds_redacted = dsl_get_redacted(ds);
2843 stat->dds_origin[0] = '\0';
2844 if (ds->ds_is_snapshot) {
2845 stat->dds_is_snapshot = B_TRUE;
2846 stat->dds_num_clones = dsl_get_numclones(ds);
2847 } else {
2848 stat->dds_is_snapshot = B_FALSE;
2849 stat->dds_num_clones = 0;
2850
2851 if (dsl_dir_is_clone(ds->ds_dir)) {
2852 dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2853 }
2854 }
2855 }
2856
2857 uint64_t
dsl_dataset_fsid_guid(dsl_dataset_t * ds)2858 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2859 {
2860 return (ds->ds_fsid_guid);
2861 }
2862
2863 void
dsl_dataset_space(dsl_dataset_t * ds,uint64_t * refdbytesp,uint64_t * availbytesp,uint64_t * usedobjsp,uint64_t * availobjsp)2864 dsl_dataset_space(dsl_dataset_t *ds,
2865 uint64_t *refdbytesp, uint64_t *availbytesp,
2866 uint64_t *usedobjsp, uint64_t *availobjsp)
2867 {
2868 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2869 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2870 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2871 *availbytesp +=
2872 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2873 if (ds->ds_quota != 0) {
2874 /*
2875 * Adjust available bytes according to refquota
2876 */
2877 if (*refdbytesp < ds->ds_quota)
2878 *availbytesp = MIN(*availbytesp,
2879 ds->ds_quota - *refdbytesp);
2880 else
2881 *availbytesp = 0;
2882 }
2883 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2884 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2885 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2886 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2887 }
2888
2889 boolean_t
dsl_dataset_modified_since_snap(dsl_dataset_t * ds,dsl_dataset_t * snap)2890 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2891 {
2892 dsl_pool_t *dp __maybe_unused = ds->ds_dir->dd_pool;
2893 uint64_t birth;
2894
2895 ASSERT(dsl_pool_config_held(dp));
2896 if (snap == NULL)
2897 return (B_FALSE);
2898 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2899 birth = BP_GET_LOGICAL_BIRTH(dsl_dataset_get_blkptr(ds));
2900 rrw_exit(&ds->ds_bp_rwlock, FTAG);
2901 if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2902 objset_t *os, *os_snap;
2903 /*
2904 * It may be that only the ZIL differs, because it was
2905 * reset in the head. Don't count that as being
2906 * modified.
2907 */
2908 if (dmu_objset_from_ds(ds, &os) != 0)
2909 return (B_TRUE);
2910 if (dmu_objset_from_ds(snap, &os_snap) != 0)
2911 return (B_TRUE);
2912 return (memcmp(&os->os_phys->os_meta_dnode,
2913 &os_snap->os_phys->os_meta_dnode,
2914 sizeof (os->os_phys->os_meta_dnode)) != 0);
2915 }
2916 return (B_FALSE);
2917 }
2918
2919 static int
dsl_dataset_rename_snapshot_check_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)2920 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2921 dsl_dataset_t *hds, void *arg)
2922 {
2923 (void) dp;
2924 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2925 int error;
2926 uint64_t val;
2927
2928 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2929 if (error != 0) {
2930 /* ignore nonexistent snapshots */
2931 return (error == ENOENT ? 0 : error);
2932 }
2933
2934 /* new name should not exist */
2935 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2936 if (error == 0)
2937 error = SET_ERROR(EEXIST);
2938 else if (error == ENOENT)
2939 error = 0;
2940
2941 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2942 if (dsl_dir_namelen(hds->ds_dir) + 1 +
2943 strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2944 error = SET_ERROR(ENAMETOOLONG);
2945
2946 return (error);
2947 }
2948
2949 int
dsl_dataset_rename_snapshot_check(void * arg,dmu_tx_t * tx)2950 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2951 {
2952 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2953 dsl_pool_t *dp = dmu_tx_pool(tx);
2954 dsl_dataset_t *hds;
2955 int error;
2956
2957 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2958 if (error != 0)
2959 return (error);
2960
2961 if (ddrsa->ddrsa_recursive) {
2962 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2963 dsl_dataset_rename_snapshot_check_impl, ddrsa,
2964 DS_FIND_CHILDREN);
2965 } else {
2966 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2967 }
2968 dsl_dataset_rele(hds, FTAG);
2969 return (error);
2970 }
2971
2972 static int
dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)2973 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2974 dsl_dataset_t *hds, void *arg)
2975 {
2976 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2977 dsl_dataset_t *ds;
2978 uint64_t val;
2979 dmu_tx_t *tx = ddrsa->ddrsa_tx;
2980 int error;
2981
2982 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2983 ASSERT(error == 0 || error == ENOENT);
2984 if (error == ENOENT) {
2985 /* ignore nonexistent snapshots */
2986 return (0);
2987 }
2988
2989 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2990
2991 /* log before we change the name */
2992 spa_history_log_internal_ds(ds, "rename", tx,
2993 "-> @%s", ddrsa->ddrsa_newsnapname);
2994
2995 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2996 B_FALSE));
2997 mutex_enter(&ds->ds_lock);
2998 (void) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname,
2999 sizeof (ds->ds_snapname));
3000 mutex_exit(&ds->ds_lock);
3001 VERIFY0(zap_add(dp->dp_meta_objset,
3002 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
3003 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
3004 zvol_rename_minors(dp->dp_spa, ddrsa->ddrsa_oldsnapname,
3005 ddrsa->ddrsa_newsnapname, B_TRUE);
3006
3007 dsl_dataset_rele(ds, FTAG);
3008 return (0);
3009 }
3010
3011 void
dsl_dataset_rename_snapshot_sync(void * arg,dmu_tx_t * tx)3012 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
3013 {
3014 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
3015 dsl_pool_t *dp = dmu_tx_pool(tx);
3016 dsl_dataset_t *hds = NULL;
3017
3018 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
3019 ddrsa->ddrsa_tx = tx;
3020 if (ddrsa->ddrsa_recursive) {
3021 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
3022 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
3023 DS_FIND_CHILDREN));
3024 } else {
3025 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
3026 }
3027 dsl_dataset_rele(hds, FTAG);
3028 }
3029
3030 int
dsl_dataset_rename_snapshot(const char * fsname,const char * oldsnapname,const char * newsnapname,boolean_t recursive)3031 dsl_dataset_rename_snapshot(const char *fsname,
3032 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
3033 {
3034 dsl_dataset_rename_snapshot_arg_t ddrsa;
3035
3036 ddrsa.ddrsa_fsname = fsname;
3037 ddrsa.ddrsa_oldsnapname = oldsnapname;
3038 ddrsa.ddrsa_newsnapname = newsnapname;
3039 ddrsa.ddrsa_recursive = recursive;
3040
3041 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
3042 dsl_dataset_rename_snapshot_sync, &ddrsa,
3043 1, ZFS_SPACE_CHECK_RESERVED));
3044 }
3045
3046 /*
3047 * If we're doing an ownership handoff, we need to make sure that there is
3048 * only one long hold on the dataset. We're not allowed to change anything here
3049 * so we don't permanently release the long hold or regular hold here. We want
3050 * to do this only when syncing to avoid the dataset unexpectedly going away
3051 * when we release the long hold.
3052 */
3053 static int
dsl_dataset_handoff_check(dsl_dataset_t * ds,void * owner,dmu_tx_t * tx)3054 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
3055 {
3056 boolean_t held = B_FALSE;
3057
3058 if (!dmu_tx_is_syncing(tx))
3059 return (0);
3060
3061 dsl_dir_t *dd = ds->ds_dir;
3062 mutex_enter(&dd->dd_activity_lock);
3063 uint64_t holds = zfs_refcount_count(&ds->ds_longholds) -
3064 (owner != NULL ? 1 : 0);
3065 /*
3066 * The value of dd_activity_waiters can chance as soon as we drop the
3067 * lock, but we're fine with that; new waiters coming in or old
3068 * waiters leaving doesn't cause problems, since we're going to cancel
3069 * waiters later anyway. The goal of this check is to verify that no
3070 * non-waiters have long-holds, and all new long-holds will be
3071 * prevented because we're holding the pool config as writer.
3072 */
3073 if (holds != dd->dd_activity_waiters)
3074 held = B_TRUE;
3075 mutex_exit(&dd->dd_activity_lock);
3076
3077 if (held)
3078 return (SET_ERROR(EBUSY));
3079
3080 return (0);
3081 }
3082
3083 int
dsl_dataset_rollback_check(void * arg,dmu_tx_t * tx)3084 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
3085 {
3086 dsl_dataset_rollback_arg_t *ddra = arg;
3087 dsl_pool_t *dp = dmu_tx_pool(tx);
3088 dsl_dataset_t *ds;
3089 int64_t unused_refres_delta;
3090 int error;
3091
3092 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
3093 if (error != 0)
3094 return (error);
3095
3096 /* must not be a snapshot */
3097 if (ds->ds_is_snapshot) {
3098 dsl_dataset_rele(ds, FTAG);
3099 return (SET_ERROR(EINVAL));
3100 }
3101
3102 /* must have a most recent snapshot */
3103 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
3104 dsl_dataset_rele(ds, FTAG);
3105 return (SET_ERROR(ESRCH));
3106 }
3107
3108 /*
3109 * No rollback to a snapshot created in the current txg, because
3110 * the rollback may dirty the dataset and create blocks that are
3111 * not reachable from the rootbp while having a birth txg that
3112 * falls into the snapshot's range.
3113 */
3114 if (dmu_tx_is_syncing(tx) &&
3115 dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
3116 dsl_dataset_rele(ds, FTAG);
3117 return (SET_ERROR(EAGAIN));
3118 }
3119
3120 /*
3121 * If the expected target snapshot is specified, then check that
3122 * the latest snapshot is it.
3123 */
3124 if (ddra->ddra_tosnap != NULL) {
3125 dsl_dataset_t *snapds;
3126
3127 /* Check if the target snapshot exists at all. */
3128 error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
3129 if (error != 0) {
3130 /*
3131 * ESRCH is used to signal that the target snapshot does
3132 * not exist, while ENOENT is used to report that
3133 * the rolled back dataset does not exist.
3134 * ESRCH is also used to cover other cases where the
3135 * target snapshot is not related to the dataset being
3136 * rolled back such as being in a different pool.
3137 */
3138 if (error == ENOENT || error == EXDEV)
3139 error = SET_ERROR(ESRCH);
3140 dsl_dataset_rele(ds, FTAG);
3141 return (error);
3142 }
3143 ASSERT(snapds->ds_is_snapshot);
3144
3145 /* Check if the snapshot is the latest snapshot indeed. */
3146 if (snapds != ds->ds_prev) {
3147 /*
3148 * Distinguish between the case where the only problem
3149 * is intervening snapshots (EEXIST) vs the snapshot
3150 * not being a valid target for rollback (ESRCH).
3151 */
3152 if (snapds->ds_dir == ds->ds_dir ||
3153 (dsl_dir_is_clone(ds->ds_dir) &&
3154 dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
3155 snapds->ds_object)) {
3156 error = SET_ERROR(EEXIST);
3157 } else {
3158 error = SET_ERROR(ESRCH);
3159 }
3160 dsl_dataset_rele(snapds, FTAG);
3161 dsl_dataset_rele(ds, FTAG);
3162 return (error);
3163 }
3164 dsl_dataset_rele(snapds, FTAG);
3165 }
3166
3167 /* must not have any bookmarks after the most recent snapshot */
3168 if (dsl_bookmark_latest_txg(ds) >
3169 dsl_dataset_phys(ds)->ds_prev_snap_txg) {
3170 dsl_dataset_rele(ds, FTAG);
3171 return (SET_ERROR(EEXIST));
3172 }
3173
3174 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
3175 if (error != 0) {
3176 dsl_dataset_rele(ds, FTAG);
3177 return (error);
3178 }
3179
3180 /*
3181 * Check if the snap we are rolling back to uses more than
3182 * the refquota.
3183 */
3184 if (ds->ds_quota != 0 &&
3185 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
3186 dsl_dataset_rele(ds, FTAG);
3187 return (SET_ERROR(EDQUOT));
3188 }
3189
3190 /*
3191 * When we do the clone swap, we will temporarily use more space
3192 * due to the refreservation (the head will no longer have any
3193 * unique space, so the entire amount of the refreservation will need
3194 * to be free). We will immediately destroy the clone, freeing
3195 * this space, but the freeing happens over many txg's.
3196 */
3197 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
3198 dsl_dataset_phys(ds)->ds_unique_bytes);
3199
3200 if (unused_refres_delta > 0 &&
3201 unused_refres_delta >
3202 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
3203 dsl_dataset_rele(ds, FTAG);
3204 return (SET_ERROR(ENOSPC));
3205 }
3206
3207 dsl_dataset_rele(ds, FTAG);
3208 return (0);
3209 }
3210
3211 void
dsl_dataset_rollback_sync(void * arg,dmu_tx_t * tx)3212 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
3213 {
3214 dsl_dataset_rollback_arg_t *ddra = arg;
3215 dsl_pool_t *dp = dmu_tx_pool(tx);
3216 dsl_dataset_t *ds, *clone;
3217 uint64_t cloneobj;
3218 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
3219
3220 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
3221
3222 dsl_dataset_name(ds->ds_prev, namebuf);
3223 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
3224
3225 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
3226 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
3227
3228 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
3229
3230 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
3231 dsl_dataset_zero_zil(ds, tx);
3232
3233 dsl_destroy_head_sync_impl(clone, tx);
3234
3235 dsl_dataset_rele(clone, FTAG);
3236 dsl_dataset_rele(ds, FTAG);
3237 }
3238
3239 /*
3240 * Rolls back the given filesystem or volume to the most recent snapshot.
3241 * The name of the most recent snapshot will be returned under key "target"
3242 * in the result nvlist.
3243 *
3244 * If owner != NULL:
3245 * - The existing dataset MUST be owned by the specified owner at entry
3246 * - Upon return, dataset will still be held by the same owner, whether we
3247 * succeed or not.
3248 *
3249 * This mode is required any time the existing filesystem is mounted. See
3250 * notes above zfs_suspend_fs() for further details.
3251 */
3252 int
dsl_dataset_rollback(const char * fsname,const char * tosnap,void * owner,nvlist_t * result)3253 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
3254 nvlist_t *result)
3255 {
3256 dsl_dataset_rollback_arg_t ddra;
3257
3258 ddra.ddra_fsname = fsname;
3259 ddra.ddra_tosnap = tosnap;
3260 ddra.ddra_owner = owner;
3261 ddra.ddra_result = result;
3262
3263 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
3264 dsl_dataset_rollback_sync, &ddra,
3265 1, ZFS_SPACE_CHECK_RESERVED));
3266 }
3267
3268 struct promotenode {
3269 list_node_t link;
3270 dsl_dataset_t *ds;
3271 };
3272
3273 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
3274 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
3275 const void *tag);
3276 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag);
3277
3278 int
dsl_dataset_promote_check(void * arg,dmu_tx_t * tx)3279 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
3280 {
3281 dsl_dataset_promote_arg_t *ddpa = arg;
3282 dsl_pool_t *dp = dmu_tx_pool(tx);
3283 dsl_dataset_t *hds;
3284 struct promotenode *snap;
3285 int err;
3286 uint64_t unused;
3287 uint64_t ss_mv_cnt;
3288 size_t max_snap_len;
3289 boolean_t conflicting_snaps;
3290
3291 err = promote_hold(ddpa, dp, FTAG);
3292 if (err != 0)
3293 return (err);
3294
3295 hds = ddpa->ddpa_clone;
3296 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
3297
3298 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
3299 promote_rele(ddpa, FTAG);
3300 return (SET_ERROR(EXDEV));
3301 }
3302
3303 snap = list_head(&ddpa->shared_snaps);
3304 if (snap == NULL) {
3305 err = SET_ERROR(ENOENT);
3306 goto out;
3307 }
3308 dsl_dataset_t *const origin_ds = snap->ds;
3309
3310 /*
3311 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
3312 * When doing a promote we must make sure the encryption root for
3313 * both the target and the target's origin does not change to avoid
3314 * needing to rewrap encryption keys
3315 */
3316 err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
3317 if (err != 0)
3318 goto out;
3319
3320 /*
3321 * Compute and check the amount of space to transfer. Since this is
3322 * so expensive, don't do the preliminary check.
3323 */
3324 if (!dmu_tx_is_syncing(tx)) {
3325 promote_rele(ddpa, FTAG);
3326 return (0);
3327 }
3328
3329 /* compute origin's new unique space */
3330 snap = list_tail(&ddpa->clone_snaps);
3331 ASSERT(snap != NULL);
3332 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3333 origin_ds->ds_object);
3334 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3335 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
3336 &ddpa->unique, &unused, &unused);
3337
3338 /*
3339 * Walk the snapshots that we are moving
3340 *
3341 * Compute space to transfer. Consider the incremental changes
3342 * to used by each snapshot:
3343 * (my used) = (prev's used) + (blocks born) - (blocks killed)
3344 * So each snapshot gave birth to:
3345 * (blocks born) = (my used) - (prev's used) + (blocks killed)
3346 * So a sequence would look like:
3347 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
3348 * Which simplifies to:
3349 * uN + kN + kN-1 + ... + k1 + k0
3350 * Note however, if we stop before we reach the ORIGIN we get:
3351 * uN + kN + kN-1 + ... + kM - uM-1
3352 */
3353 conflicting_snaps = B_FALSE;
3354 ss_mv_cnt = 0;
3355 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
3356 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
3357 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
3358 for (snap = list_head(&ddpa->shared_snaps); snap;
3359 snap = list_next(&ddpa->shared_snaps, snap)) {
3360 uint64_t val, dlused, dlcomp, dluncomp;
3361 dsl_dataset_t *ds = snap->ds;
3362
3363 ss_mv_cnt++;
3364
3365 /*
3366 * If there are long holds, we won't be able to evict
3367 * the objset.
3368 */
3369 if (dsl_dataset_long_held(ds)) {
3370 err = SET_ERROR(EBUSY);
3371 goto out;
3372 }
3373
3374 /* Check that the snapshot name does not conflict */
3375 VERIFY0(dsl_dataset_get_snapname(ds));
3376 if (strlen(ds->ds_snapname) >= max_snap_len) {
3377 err = SET_ERROR(ENAMETOOLONG);
3378 goto out;
3379 }
3380 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
3381 if (err == 0) {
3382 fnvlist_add_boolean(ddpa->err_ds,
3383 snap->ds->ds_snapname);
3384 conflicting_snaps = B_TRUE;
3385 } else if (err != ENOENT) {
3386 goto out;
3387 }
3388
3389 /* The very first snapshot does not have a deadlist */
3390 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
3391 continue;
3392
3393 dsl_deadlist_space(&ds->ds_deadlist,
3394 &dlused, &dlcomp, &dluncomp);
3395 ddpa->used += dlused;
3396 ddpa->comp += dlcomp;
3397 ddpa->uncomp += dluncomp;
3398 }
3399
3400 /*
3401 * Check that bookmarks that are being transferred don't have
3402 * name conflicts.
3403 */
3404 for (dsl_bookmark_node_t *dbn = avl_first(&origin_ds->ds_bookmarks);
3405 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3406 dsl_dataset_phys(origin_ds)->ds_creation_txg;
3407 dbn = AVL_NEXT(&origin_ds->ds_bookmarks, dbn)) {
3408 if (strlen(dbn->dbn_name) >= max_snap_len) {
3409 err = SET_ERROR(ENAMETOOLONG);
3410 goto out;
3411 }
3412 zfs_bookmark_phys_t bm;
3413 err = dsl_bookmark_lookup_impl(ddpa->ddpa_clone,
3414 dbn->dbn_name, &bm);
3415
3416 if (err == 0) {
3417 fnvlist_add_boolean(ddpa->err_ds, dbn->dbn_name);
3418 conflicting_snaps = B_TRUE;
3419 } else if (err == ESRCH) {
3420 err = 0;
3421 }
3422 if (err != 0) {
3423 goto out;
3424 }
3425 }
3426
3427 /*
3428 * In order to return the full list of conflicting snapshots, we check
3429 * whether there was a conflict after traversing all of them.
3430 */
3431 if (conflicting_snaps) {
3432 err = SET_ERROR(EEXIST);
3433 goto out;
3434 }
3435
3436 /*
3437 * If we are a clone of a clone then we never reached ORIGIN,
3438 * so we need to subtract out the clone origin's used space.
3439 */
3440 if (ddpa->origin_origin) {
3441 ddpa->used -=
3442 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
3443 ddpa->comp -=
3444 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
3445 ddpa->uncomp -=
3446 dsl_dataset_phys(ddpa->origin_origin)->
3447 ds_uncompressed_bytes;
3448 }
3449
3450 /* Check that there is enough space and limit headroom here */
3451 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
3452 0, ss_mv_cnt, ddpa->used, ddpa->cr, ddpa->proc);
3453 if (err != 0)
3454 goto out;
3455
3456 /*
3457 * Compute the amounts of space that will be used by snapshots
3458 * after the promotion (for both origin and clone). For each,
3459 * it is the amount of space that will be on all of their
3460 * deadlists (that was not born before their new origin).
3461 */
3462 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
3463 uint64_t space;
3464
3465 /*
3466 * Note, typically this will not be a clone of a clone,
3467 * so dd_origin_txg will be < TXG_INITIAL, so
3468 * these snaplist_space() -> dsl_deadlist_space_range()
3469 * calls will be fast because they do not have to
3470 * iterate over all bps.
3471 */
3472 snap = list_head(&ddpa->origin_snaps);
3473 if (snap == NULL) {
3474 err = SET_ERROR(ENOENT);
3475 goto out;
3476 }
3477 err = snaplist_space(&ddpa->shared_snaps,
3478 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
3479 if (err != 0)
3480 goto out;
3481
3482 err = snaplist_space(&ddpa->clone_snaps,
3483 snap->ds->ds_dir->dd_origin_txg, &space);
3484 if (err != 0)
3485 goto out;
3486 ddpa->cloneusedsnap += space;
3487 }
3488 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
3489 DD_FLAG_USED_BREAKDOWN) {
3490 err = snaplist_space(&ddpa->origin_snaps,
3491 dsl_dataset_phys(origin_ds)->ds_creation_txg,
3492 &ddpa->originusedsnap);
3493 if (err != 0)
3494 goto out;
3495 }
3496
3497 out:
3498 promote_rele(ddpa, FTAG);
3499 return (err);
3500 }
3501
3502 void
dsl_dataset_promote_sync(void * arg,dmu_tx_t * tx)3503 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3504 {
3505 dsl_dataset_promote_arg_t *ddpa = arg;
3506 dsl_pool_t *dp = dmu_tx_pool(tx);
3507 dsl_dataset_t *hds;
3508 struct promotenode *snap;
3509 dsl_dataset_t *origin_ds;
3510 dsl_dataset_t *origin_head;
3511 dsl_dir_t *dd;
3512 dsl_dir_t *odd = NULL;
3513 uint64_t oldnext_obj;
3514 int64_t delta;
3515
3516 ASSERT(nvlist_empty(ddpa->err_ds));
3517
3518 VERIFY0(promote_hold(ddpa, dp, FTAG));
3519 hds = ddpa->ddpa_clone;
3520
3521 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3522
3523 snap = list_head(&ddpa->shared_snaps);
3524 origin_ds = snap->ds;
3525 dd = hds->ds_dir;
3526
3527 snap = list_head(&ddpa->origin_snaps);
3528 origin_head = snap->ds;
3529
3530 /*
3531 * We need to explicitly open odd, since origin_ds's dd will be
3532 * changing.
3533 */
3534 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3535 NULL, FTAG, &odd));
3536
3537 dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
3538
3539 /* change origin's next snap */
3540 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3541 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3542 snap = list_tail(&ddpa->clone_snaps);
3543 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3544 origin_ds->ds_object);
3545 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3546
3547 /* change the origin's next clone */
3548 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3549 dsl_dataset_remove_from_next_clones(origin_ds,
3550 snap->ds->ds_object, tx);
3551 VERIFY0(zap_add_int(dp->dp_meta_objset,
3552 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3553 oldnext_obj, tx));
3554 }
3555
3556 /* change origin */
3557 dmu_buf_will_dirty(dd->dd_dbuf, tx);
3558 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3559 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3560 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3561 dmu_buf_will_dirty(odd->dd_dbuf, tx);
3562 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3563 origin_head->ds_dir->dd_origin_txg =
3564 dsl_dataset_phys(origin_ds)->ds_creation_txg;
3565
3566 /* change dd_clone entries */
3567 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3568 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3569 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3570 VERIFY0(zap_add_int(dp->dp_meta_objset,
3571 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3572 hds->ds_object, tx));
3573
3574 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3575 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3576 origin_head->ds_object, tx));
3577 if (dsl_dir_phys(dd)->dd_clones == 0) {
3578 dsl_dir_phys(dd)->dd_clones =
3579 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3580 DMU_OT_NONE, 0, tx);
3581 }
3582 VERIFY0(zap_add_int(dp->dp_meta_objset,
3583 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3584 }
3585
3586 /*
3587 * Move bookmarks to this dir.
3588 */
3589 dsl_bookmark_node_t *dbn_next;
3590 for (dsl_bookmark_node_t *dbn = avl_first(&origin_head->ds_bookmarks);
3591 dbn != NULL && dbn->dbn_phys.zbm_creation_txg <=
3592 dsl_dataset_phys(origin_ds)->ds_creation_txg;
3593 dbn = dbn_next) {
3594 dbn_next = AVL_NEXT(&origin_head->ds_bookmarks, dbn);
3595
3596 avl_remove(&origin_head->ds_bookmarks, dbn);
3597 VERIFY0(zap_remove(dp->dp_meta_objset,
3598 origin_head->ds_bookmarks_obj, dbn->dbn_name, tx));
3599
3600 dsl_bookmark_node_add(hds, dbn, tx);
3601 }
3602
3603 dsl_bookmark_next_changed(hds, origin_ds, tx);
3604
3605 /* move snapshots to this dir */
3606 for (snap = list_head(&ddpa->shared_snaps); snap;
3607 snap = list_next(&ddpa->shared_snaps, snap)) {
3608 dsl_dataset_t *ds = snap->ds;
3609
3610 /*
3611 * Property callbacks are registered to a particular
3612 * dsl_dir. Since ours is changing, evict the objset
3613 * so that they will be unregistered from the old dsl_dir.
3614 */
3615 if (ds->ds_objset) {
3616 dmu_objset_evict(ds->ds_objset);
3617 ds->ds_objset = NULL;
3618 }
3619
3620 /* move snap name entry */
3621 VERIFY0(dsl_dataset_get_snapname(ds));
3622 VERIFY0(dsl_dataset_snap_remove(origin_head,
3623 ds->ds_snapname, tx, B_TRUE));
3624 VERIFY0(zap_add(dp->dp_meta_objset,
3625 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3626 8, 1, &ds->ds_object, tx));
3627 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3628 DD_FIELD_SNAPSHOT_COUNT, tx);
3629
3630 /* change containing dsl_dir */
3631 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3632 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3633 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3634 ASSERT3P(ds->ds_dir, ==, odd);
3635 dsl_dir_rele(ds->ds_dir, ds);
3636 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3637 NULL, ds, &ds->ds_dir));
3638
3639 /* move any clone references */
3640 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3641 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3642 zap_cursor_t zc;
3643 zap_attribute_t za;
3644
3645 for (zap_cursor_init(&zc, dp->dp_meta_objset,
3646 dsl_dataset_phys(ds)->ds_next_clones_obj);
3647 zap_cursor_retrieve(&zc, &za) == 0;
3648 zap_cursor_advance(&zc)) {
3649 dsl_dataset_t *cnds;
3650 uint64_t o;
3651
3652 if (za.za_first_integer == oldnext_obj) {
3653 /*
3654 * We've already moved the
3655 * origin's reference.
3656 */
3657 continue;
3658 }
3659
3660 VERIFY0(dsl_dataset_hold_obj(dp,
3661 za.za_first_integer, FTAG, &cnds));
3662 o = dsl_dir_phys(cnds->ds_dir)->
3663 dd_head_dataset_obj;
3664
3665 VERIFY0(zap_remove_int(dp->dp_meta_objset,
3666 dsl_dir_phys(odd)->dd_clones, o, tx));
3667 VERIFY0(zap_add_int(dp->dp_meta_objset,
3668 dsl_dir_phys(dd)->dd_clones, o, tx));
3669 dsl_dataset_rele(cnds, FTAG);
3670 }
3671 zap_cursor_fini(&zc);
3672 }
3673
3674 ASSERT(!dsl_prop_hascb(ds));
3675 }
3676
3677 /*
3678 * Change space accounting.
3679 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3680 * both be valid, or both be 0 (resulting in delta == 0). This
3681 * is true for each of {clone,origin} independently.
3682 */
3683
3684 delta = ddpa->cloneusedsnap -
3685 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3686 ASSERT3S(delta, >=, 0);
3687 ASSERT3U(ddpa->used, >=, delta);
3688 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3689 dsl_dir_diduse_space(dd, DD_USED_HEAD,
3690 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3691
3692 delta = ddpa->originusedsnap -
3693 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3694 ASSERT3S(delta, <=, 0);
3695 ASSERT3U(ddpa->used, >=, -delta);
3696 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3697 dsl_dir_diduse_space(odd, DD_USED_HEAD,
3698 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3699
3700 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3701
3702 /*
3703 * Since livelists are specific to a clone's origin txg, they
3704 * are no longer accurate. Destroy the livelist from the clone being
3705 * promoted. If the origin dataset is a clone, destroy its livelist
3706 * as well.
3707 */
3708 dsl_dir_remove_livelist(dd, tx, B_TRUE);
3709 dsl_dir_remove_livelist(odd, tx, B_TRUE);
3710
3711 /* log history record */
3712 spa_history_log_internal_ds(hds, "promote", tx, " ");
3713
3714 dsl_dir_rele(odd, FTAG);
3715
3716 /*
3717 * Transfer common error blocks from old head to new head, before
3718 * calling promote_rele() on ddpa since we need to dereference
3719 * origin_head and hds.
3720 */
3721 if (spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_HEAD_ERRLOG)) {
3722 uint64_t old_head = origin_head->ds_object;
3723 uint64_t new_head = hds->ds_object;
3724 spa_swap_errlog(dp->dp_spa, new_head, old_head, tx);
3725 }
3726
3727 promote_rele(ddpa, FTAG);
3728 }
3729
3730 /*
3731 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3732 * (exclusive) and last_obj (inclusive). The list will be in reverse
3733 * order (last_obj will be the list_head()). If first_obj == 0, do all
3734 * snapshots back to this dataset's origin.
3735 */
3736 static int
snaplist_make(dsl_pool_t * dp,uint64_t first_obj,uint64_t last_obj,list_t * l,const void * tag)3737 snaplist_make(dsl_pool_t *dp,
3738 uint64_t first_obj, uint64_t last_obj, list_t *l, const void *tag)
3739 {
3740 uint64_t obj = last_obj;
3741
3742 list_create(l, sizeof (struct promotenode),
3743 offsetof(struct promotenode, link));
3744
3745 while (obj != first_obj) {
3746 dsl_dataset_t *ds;
3747 struct promotenode *snap;
3748 int err;
3749
3750 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3751 ASSERT(err != ENOENT);
3752 if (err != 0)
3753 return (err);
3754
3755 if (first_obj == 0)
3756 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3757
3758 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3759 snap->ds = ds;
3760 list_insert_tail(l, snap);
3761 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3762 }
3763
3764 return (0);
3765 }
3766
3767 static int
snaplist_space(list_t * l,uint64_t mintxg,uint64_t * spacep)3768 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3769 {
3770 struct promotenode *snap;
3771
3772 *spacep = 0;
3773 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3774 uint64_t used, comp, uncomp;
3775 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3776 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3777 *spacep += used;
3778 }
3779 return (0);
3780 }
3781
3782 static void
snaplist_destroy(list_t * l,const void * tag)3783 snaplist_destroy(list_t *l, const void *tag)
3784 {
3785 struct promotenode *snap;
3786
3787 if (l == NULL || !list_link_active(&l->list_head))
3788 return;
3789
3790 while ((snap = list_remove_tail(l)) != NULL) {
3791 dsl_dataset_rele(snap->ds, tag);
3792 kmem_free(snap, sizeof (*snap));
3793 }
3794 list_destroy(l);
3795 }
3796
3797 static int
promote_hold(dsl_dataset_promote_arg_t * ddpa,dsl_pool_t * dp,const void * tag)3798 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, const void *tag)
3799 {
3800 int error;
3801 dsl_dir_t *dd;
3802 struct promotenode *snap;
3803
3804 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3805 &ddpa->ddpa_clone);
3806 if (error != 0)
3807 return (error);
3808 dd = ddpa->ddpa_clone->ds_dir;
3809
3810 if (ddpa->ddpa_clone->ds_is_snapshot ||
3811 !dsl_dir_is_clone(dd)) {
3812 dsl_dataset_rele(ddpa->ddpa_clone, tag);
3813 return (SET_ERROR(EINVAL));
3814 }
3815
3816 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3817 &ddpa->shared_snaps, tag);
3818 if (error != 0)
3819 goto out;
3820
3821 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3822 &ddpa->clone_snaps, tag);
3823 if (error != 0)
3824 goto out;
3825
3826 snap = list_head(&ddpa->shared_snaps);
3827 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3828 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3829 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3830 &ddpa->origin_snaps, tag);
3831 if (error != 0)
3832 goto out;
3833
3834 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3835 error = dsl_dataset_hold_obj(dp,
3836 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3837 tag, &ddpa->origin_origin);
3838 if (error != 0)
3839 goto out;
3840 }
3841 out:
3842 if (error != 0)
3843 promote_rele(ddpa, tag);
3844 return (error);
3845 }
3846
3847 static void
promote_rele(dsl_dataset_promote_arg_t * ddpa,const void * tag)3848 promote_rele(dsl_dataset_promote_arg_t *ddpa, const void *tag)
3849 {
3850 snaplist_destroy(&ddpa->shared_snaps, tag);
3851 snaplist_destroy(&ddpa->clone_snaps, tag);
3852 snaplist_destroy(&ddpa->origin_snaps, tag);
3853 if (ddpa->origin_origin != NULL)
3854 dsl_dataset_rele(ddpa->origin_origin, tag);
3855 dsl_dataset_rele(ddpa->ddpa_clone, tag);
3856 }
3857
3858 /*
3859 * Promote a clone.
3860 *
3861 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3862 * in with the name. (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3863 */
3864 int
dsl_dataset_promote(const char * name,char * conflsnap)3865 dsl_dataset_promote(const char *name, char *conflsnap)
3866 {
3867 dsl_dataset_promote_arg_t ddpa = { 0 };
3868 uint64_t numsnaps;
3869 int error;
3870 nvpair_t *snap_pair;
3871 objset_t *os;
3872
3873 /*
3874 * We will modify space proportional to the number of
3875 * snapshots. Compute numsnaps.
3876 */
3877 error = dmu_objset_hold(name, FTAG, &os);
3878 if (error != 0)
3879 return (error);
3880 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3881 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3882 &numsnaps);
3883 dmu_objset_rele(os, FTAG);
3884 if (error != 0)
3885 return (error);
3886
3887 ddpa.ddpa_clonename = name;
3888 ddpa.err_ds = fnvlist_alloc();
3889 ddpa.cr = CRED();
3890 ddpa.proc = curproc;
3891
3892 error = dsl_sync_task(name, dsl_dataset_promote_check,
3893 dsl_dataset_promote_sync, &ddpa,
3894 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3895
3896 /*
3897 * Return the first conflicting snapshot found.
3898 */
3899 snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3900 if (snap_pair != NULL && conflsnap != NULL)
3901 (void) strlcpy(conflsnap, nvpair_name(snap_pair),
3902 ZFS_MAX_DATASET_NAME_LEN);
3903
3904 fnvlist_free(ddpa.err_ds);
3905 return (error);
3906 }
3907
3908 int
dsl_dataset_clone_swap_check_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,boolean_t force,void * owner,dmu_tx_t * tx)3909 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3910 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3911 {
3912 /*
3913 * "slack" factor for received datasets with refquota set on them.
3914 * See the bottom of this function for details on its use.
3915 */
3916 uint64_t refquota_slack = (uint64_t)DMU_MAX_ACCESS *
3917 spa_asize_inflation;
3918 int64_t unused_refres_delta;
3919
3920 /* they should both be heads */
3921 if (clone->ds_is_snapshot ||
3922 origin_head->ds_is_snapshot)
3923 return (SET_ERROR(EINVAL));
3924
3925 /* if we are not forcing, the branch point should be just before them */
3926 if (!force && clone->ds_prev != origin_head->ds_prev)
3927 return (SET_ERROR(EINVAL));
3928
3929 /* clone should be the clone (unless they are unrelated) */
3930 if (clone->ds_prev != NULL &&
3931 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3932 origin_head->ds_dir != clone->ds_prev->ds_dir)
3933 return (SET_ERROR(EINVAL));
3934
3935 /* the clone should be a child of the origin */
3936 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3937 return (SET_ERROR(EINVAL));
3938
3939 /* origin_head shouldn't be modified unless 'force' */
3940 if (!force &&
3941 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3942 return (SET_ERROR(ETXTBSY));
3943
3944 /* origin_head should have no long holds (e.g. is not mounted) */
3945 if (dsl_dataset_handoff_check(origin_head, owner, tx))
3946 return (SET_ERROR(EBUSY));
3947
3948 /* check amount of any unconsumed refreservation */
3949 unused_refres_delta =
3950 (int64_t)MIN(origin_head->ds_reserved,
3951 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3952 (int64_t)MIN(origin_head->ds_reserved,
3953 dsl_dataset_phys(clone)->ds_unique_bytes);
3954
3955 if (unused_refres_delta > 0 &&
3956 unused_refres_delta >
3957 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3958 return (SET_ERROR(ENOSPC));
3959
3960 /*
3961 * The clone can't be too much over the head's refquota.
3962 *
3963 * To ensure that the entire refquota can be used, we allow one
3964 * transaction to exceed the refquota. Therefore, this check
3965 * needs to also allow for the space referenced to be more than the
3966 * refquota. The maximum amount of space that one transaction can use
3967 * on disk is DMU_MAX_ACCESS * spa_asize_inflation. Allowing this
3968 * overage ensures that we are able to receive a filesystem that
3969 * exceeds the refquota on the source system.
3970 *
3971 * So that overage is the refquota_slack we use below.
3972 */
3973 if (origin_head->ds_quota != 0 &&
3974 dsl_dataset_phys(clone)->ds_referenced_bytes >
3975 origin_head->ds_quota + refquota_slack)
3976 return (SET_ERROR(EDQUOT));
3977
3978 return (0);
3979 }
3980
3981 static void
dsl_dataset_swap_remap_deadlists(dsl_dataset_t * clone,dsl_dataset_t * origin,dmu_tx_t * tx)3982 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3983 dsl_dataset_t *origin, dmu_tx_t *tx)
3984 {
3985 uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3986 dsl_pool_t *dp = dmu_tx_pool(tx);
3987
3988 ASSERT(dsl_pool_sync_context(dp));
3989
3990 clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3991 origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3992
3993 if (clone_remap_dl_obj != 0) {
3994 dsl_deadlist_close(&clone->ds_remap_deadlist);
3995 dsl_dataset_unset_remap_deadlist_object(clone, tx);
3996 }
3997 if (origin_remap_dl_obj != 0) {
3998 dsl_deadlist_close(&origin->ds_remap_deadlist);
3999 dsl_dataset_unset_remap_deadlist_object(origin, tx);
4000 }
4001
4002 if (clone_remap_dl_obj != 0) {
4003 dsl_dataset_set_remap_deadlist_object(origin,
4004 clone_remap_dl_obj, tx);
4005 dsl_deadlist_open(&origin->ds_remap_deadlist,
4006 dp->dp_meta_objset, clone_remap_dl_obj);
4007 }
4008 if (origin_remap_dl_obj != 0) {
4009 dsl_dataset_set_remap_deadlist_object(clone,
4010 origin_remap_dl_obj, tx);
4011 dsl_deadlist_open(&clone->ds_remap_deadlist,
4012 dp->dp_meta_objset, origin_remap_dl_obj);
4013 }
4014 }
4015
4016 void
dsl_dataset_clone_swap_sync_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,dmu_tx_t * tx)4017 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
4018 dsl_dataset_t *origin_head, dmu_tx_t *tx)
4019 {
4020 dsl_pool_t *dp = dmu_tx_pool(tx);
4021 int64_t unused_refres_delta;
4022
4023 ASSERT(clone->ds_reserved == 0);
4024 /*
4025 * NOTE: On DEBUG kernels there could be a race between this and
4026 * the check function if spa_asize_inflation is adjusted...
4027 */
4028 ASSERT(origin_head->ds_quota == 0 ||
4029 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
4030 DMU_MAX_ACCESS * spa_asize_inflation);
4031 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
4032
4033 dsl_dir_cancel_waiters(origin_head->ds_dir);
4034
4035 /*
4036 * Swap per-dataset feature flags.
4037 */
4038 for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
4039 if (!(spa_feature_table[f].fi_flags &
4040 ZFEATURE_FLAG_PER_DATASET)) {
4041 ASSERT(!dsl_dataset_feature_is_active(clone, f));
4042 ASSERT(!dsl_dataset_feature_is_active(origin_head, f));
4043 continue;
4044 }
4045
4046 boolean_t clone_inuse = dsl_dataset_feature_is_active(clone, f);
4047 void *clone_feature = clone->ds_feature[f];
4048 boolean_t origin_head_inuse =
4049 dsl_dataset_feature_is_active(origin_head, f);
4050 void *origin_head_feature = origin_head->ds_feature[f];
4051
4052 if (clone_inuse)
4053 dsl_dataset_deactivate_feature_impl(clone, f, tx);
4054 if (origin_head_inuse)
4055 dsl_dataset_deactivate_feature_impl(origin_head, f, tx);
4056
4057 if (clone_inuse) {
4058 dsl_dataset_activate_feature(origin_head->ds_object, f,
4059 clone_feature, tx);
4060 origin_head->ds_feature[f] = clone_feature;
4061 }
4062 if (origin_head_inuse) {
4063 dsl_dataset_activate_feature(clone->ds_object, f,
4064 origin_head_feature, tx);
4065 clone->ds_feature[f] = origin_head_feature;
4066 }
4067 }
4068
4069 dmu_buf_will_dirty(clone->ds_dbuf, tx);
4070 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
4071
4072 if (clone->ds_objset != NULL) {
4073 dmu_objset_evict(clone->ds_objset);
4074 clone->ds_objset = NULL;
4075 }
4076
4077 if (origin_head->ds_objset != NULL) {
4078 dmu_objset_evict(origin_head->ds_objset);
4079 origin_head->ds_objset = NULL;
4080 }
4081
4082 unused_refres_delta =
4083 (int64_t)MIN(origin_head->ds_reserved,
4084 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
4085 (int64_t)MIN(origin_head->ds_reserved,
4086 dsl_dataset_phys(clone)->ds_unique_bytes);
4087
4088 /*
4089 * Reset origin's unique bytes.
4090 */
4091 {
4092 dsl_dataset_t *origin = clone->ds_prev;
4093 uint64_t comp, uncomp;
4094
4095 dmu_buf_will_dirty(origin->ds_dbuf, tx);
4096 dsl_deadlist_space_range(&clone->ds_deadlist,
4097 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
4098 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
4099 }
4100
4101 /* swap blkptrs */
4102 {
4103 rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
4104 rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
4105 blkptr_t tmp;
4106 tmp = dsl_dataset_phys(origin_head)->ds_bp;
4107 dsl_dataset_phys(origin_head)->ds_bp =
4108 dsl_dataset_phys(clone)->ds_bp;
4109 dsl_dataset_phys(clone)->ds_bp = tmp;
4110 rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
4111 rrw_exit(&clone->ds_bp_rwlock, FTAG);
4112 }
4113
4114 /* set dd_*_bytes */
4115 {
4116 int64_t dused, dcomp, duncomp;
4117 uint64_t cdl_used, cdl_comp, cdl_uncomp;
4118 uint64_t odl_used, odl_comp, odl_uncomp;
4119
4120 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
4121 dd_used_breakdown[DD_USED_SNAP], ==, 0);
4122
4123 dsl_deadlist_space(&clone->ds_deadlist,
4124 &cdl_used, &cdl_comp, &cdl_uncomp);
4125 dsl_deadlist_space(&origin_head->ds_deadlist,
4126 &odl_used, &odl_comp, &odl_uncomp);
4127
4128 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
4129 cdl_used -
4130 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
4131 odl_used);
4132 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
4133 cdl_comp -
4134 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
4135 odl_comp);
4136 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
4137 cdl_uncomp -
4138 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
4139 odl_uncomp);
4140
4141 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
4142 dused, dcomp, duncomp, tx);
4143 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
4144 -dused, -dcomp, -duncomp, tx);
4145
4146 /*
4147 * The difference in the space used by snapshots is the
4148 * difference in snapshot space due to the head's
4149 * deadlist (since that's the only thing that's
4150 * changing that affects the snapused).
4151 */
4152 dsl_deadlist_space_range(&clone->ds_deadlist,
4153 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4154 &cdl_used, &cdl_comp, &cdl_uncomp);
4155 dsl_deadlist_space_range(&origin_head->ds_deadlist,
4156 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
4157 &odl_used, &odl_comp, &odl_uncomp);
4158 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
4159 DD_USED_HEAD, DD_USED_SNAP, tx);
4160 }
4161
4162 /* swap ds_*_bytes */
4163 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
4164 dsl_dataset_phys(clone)->ds_referenced_bytes);
4165 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
4166 dsl_dataset_phys(clone)->ds_compressed_bytes);
4167 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
4168 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
4169 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
4170 dsl_dataset_phys(clone)->ds_unique_bytes);
4171
4172 /* apply any parent delta for change in unconsumed refreservation */
4173 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
4174 unused_refres_delta, 0, 0, tx);
4175
4176 /*
4177 * Swap deadlists.
4178 */
4179 dsl_deadlist_close(&clone->ds_deadlist);
4180 dsl_deadlist_close(&origin_head->ds_deadlist);
4181 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
4182 dsl_dataset_phys(clone)->ds_deadlist_obj);
4183 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
4184 dsl_dataset_phys(clone)->ds_deadlist_obj);
4185 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
4186 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
4187 dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
4188
4189 /*
4190 * If there is a bookmark at the origin, its "next dataset" is
4191 * changing, so we need to reset its FBN.
4192 */
4193 dsl_bookmark_next_changed(origin_head, origin_head->ds_prev, tx);
4194
4195 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
4196
4197 /*
4198 * Destroy any livelists associated with the clone or the origin,
4199 * since after the swap the corresponding livelists are no longer
4200 * valid.
4201 */
4202 dsl_dir_remove_livelist(clone->ds_dir, tx, B_TRUE);
4203 dsl_dir_remove_livelist(origin_head->ds_dir, tx, B_TRUE);
4204
4205 spa_history_log_internal_ds(clone, "clone swap", tx,
4206 "parent=%s", origin_head->ds_dir->dd_myname);
4207 }
4208
4209 /*
4210 * Given a pool name and a dataset object number in that pool,
4211 * return the name of that dataset.
4212 */
4213 int
dsl_dsobj_to_dsname(char * pname,uint64_t obj,char * buf)4214 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
4215 {
4216 dsl_pool_t *dp;
4217 dsl_dataset_t *ds;
4218 int error;
4219
4220 error = dsl_pool_hold(pname, FTAG, &dp);
4221 if (error != 0)
4222 return (error);
4223
4224 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
4225 if (error == 0) {
4226 dsl_dataset_name(ds, buf);
4227 dsl_dataset_rele(ds, FTAG);
4228 }
4229 dsl_pool_rele(dp, FTAG);
4230
4231 return (error);
4232 }
4233
4234 int
dsl_dataset_check_quota(dsl_dataset_t * ds,boolean_t check_quota,uint64_t asize,uint64_t inflight,uint64_t * used,uint64_t * ref_rsrv)4235 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
4236 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
4237 {
4238 int error = 0;
4239
4240 ASSERT3S(asize, >, 0);
4241
4242 /*
4243 * *ref_rsrv is the portion of asize that will come from any
4244 * unconsumed refreservation space.
4245 */
4246 *ref_rsrv = 0;
4247
4248 mutex_enter(&ds->ds_lock);
4249 /*
4250 * Make a space adjustment for reserved bytes.
4251 */
4252 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
4253 ASSERT3U(*used, >=,
4254 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4255 *used -=
4256 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
4257 *ref_rsrv =
4258 asize - MIN(asize, parent_delta(ds, asize + inflight));
4259 }
4260
4261 if (!check_quota || ds->ds_quota == 0) {
4262 mutex_exit(&ds->ds_lock);
4263 return (0);
4264 }
4265 /*
4266 * If they are requesting more space, and our current estimate
4267 * is over quota, they get to try again unless the actual
4268 * on-disk is over quota and there are no pending changes (which
4269 * may free up space for us).
4270 */
4271 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
4272 ds->ds_quota) {
4273 if (inflight > 0 ||
4274 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
4275 error = SET_ERROR(ERESTART);
4276 else
4277 error = SET_ERROR(EDQUOT);
4278 }
4279 mutex_exit(&ds->ds_lock);
4280
4281 return (error);
4282 }
4283
4284 typedef struct dsl_dataset_set_qr_arg {
4285 const char *ddsqra_name;
4286 zprop_source_t ddsqra_source;
4287 uint64_t ddsqra_value;
4288 } dsl_dataset_set_qr_arg_t;
4289
4290
4291 static int
dsl_dataset_set_refquota_check(void * arg,dmu_tx_t * tx)4292 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
4293 {
4294 dsl_dataset_set_qr_arg_t *ddsqra = arg;
4295 dsl_pool_t *dp = dmu_tx_pool(tx);
4296 dsl_dataset_t *ds;
4297 int error;
4298 uint64_t newval;
4299
4300 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
4301 return (SET_ERROR(ENOTSUP));
4302
4303 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4304 if (error != 0)
4305 return (error);
4306
4307 if (ds->ds_is_snapshot) {
4308 dsl_dataset_rele(ds, FTAG);
4309 return (SET_ERROR(EINVAL));
4310 }
4311
4312 error = dsl_prop_predict(ds->ds_dir,
4313 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4314 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4315 if (error != 0) {
4316 dsl_dataset_rele(ds, FTAG);
4317 return (error);
4318 }
4319
4320 if (newval == 0) {
4321 dsl_dataset_rele(ds, FTAG);
4322 return (0);
4323 }
4324
4325 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
4326 newval < ds->ds_reserved) {
4327 dsl_dataset_rele(ds, FTAG);
4328 return (SET_ERROR(ENOSPC));
4329 }
4330
4331 dsl_dataset_rele(ds, FTAG);
4332 return (0);
4333 }
4334
4335 static void
dsl_dataset_set_refquota_sync(void * arg,dmu_tx_t * tx)4336 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
4337 {
4338 dsl_dataset_set_qr_arg_t *ddsqra = arg;
4339 dsl_pool_t *dp = dmu_tx_pool(tx);
4340 dsl_dataset_t *ds = NULL;
4341 uint64_t newval;
4342
4343 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4344
4345 dsl_prop_set_sync_impl(ds,
4346 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
4347 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
4348 &ddsqra->ddsqra_value, tx);
4349
4350 VERIFY0(dsl_prop_get_int_ds(ds,
4351 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
4352
4353 if (ds->ds_quota != newval) {
4354 dmu_buf_will_dirty(ds->ds_dbuf, tx);
4355 ds->ds_quota = newval;
4356 }
4357 dsl_dataset_rele(ds, FTAG);
4358 }
4359
4360 int
dsl_dataset_set_refquota(const char * dsname,zprop_source_t source,uint64_t refquota)4361 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
4362 uint64_t refquota)
4363 {
4364 dsl_dataset_set_qr_arg_t ddsqra;
4365
4366 ddsqra.ddsqra_name = dsname;
4367 ddsqra.ddsqra_source = source;
4368 ddsqra.ddsqra_value = refquota;
4369
4370 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
4371 dsl_dataset_set_refquota_sync, &ddsqra, 0,
4372 ZFS_SPACE_CHECK_EXTRA_RESERVED));
4373 }
4374
4375 static int
dsl_dataset_set_refreservation_check(void * arg,dmu_tx_t * tx)4376 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
4377 {
4378 dsl_dataset_set_qr_arg_t *ddsqra = arg;
4379 dsl_pool_t *dp = dmu_tx_pool(tx);
4380 dsl_dataset_t *ds;
4381 int error;
4382 uint64_t newval, unique;
4383
4384 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
4385 return (SET_ERROR(ENOTSUP));
4386
4387 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
4388 if (error != 0)
4389 return (error);
4390
4391 if (ds->ds_is_snapshot) {
4392 dsl_dataset_rele(ds, FTAG);
4393 return (SET_ERROR(EINVAL));
4394 }
4395
4396 error = dsl_prop_predict(ds->ds_dir,
4397 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4398 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
4399 if (error != 0) {
4400 dsl_dataset_rele(ds, FTAG);
4401 return (error);
4402 }
4403
4404 /*
4405 * If we are doing the preliminary check in open context, the
4406 * space estimates may be inaccurate.
4407 */
4408 if (!dmu_tx_is_syncing(tx)) {
4409 dsl_dataset_rele(ds, FTAG);
4410 return (0);
4411 }
4412
4413 mutex_enter(&ds->ds_lock);
4414 if (!DS_UNIQUE_IS_ACCURATE(ds))
4415 dsl_dataset_recalc_head_uniq(ds);
4416 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4417 mutex_exit(&ds->ds_lock);
4418
4419 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
4420 uint64_t delta = MAX(unique, newval) -
4421 MAX(unique, ds->ds_reserved);
4422
4423 if (delta >
4424 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
4425 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
4426 dsl_dataset_rele(ds, FTAG);
4427 return (SET_ERROR(ENOSPC));
4428 }
4429 }
4430
4431 dsl_dataset_rele(ds, FTAG);
4432 return (0);
4433 }
4434
4435 void
dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t * ds,zprop_source_t source,uint64_t value,dmu_tx_t * tx)4436 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
4437 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
4438 {
4439 uint64_t newval;
4440 uint64_t unique;
4441 int64_t delta;
4442
4443 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
4444 source, sizeof (value), 1, &value, tx);
4445
4446 VERIFY0(dsl_prop_get_int_ds(ds,
4447 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
4448
4449 dmu_buf_will_dirty(ds->ds_dbuf, tx);
4450 mutex_enter(&ds->ds_dir->dd_lock);
4451 mutex_enter(&ds->ds_lock);
4452 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
4453 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
4454 delta = MAX(0, (int64_t)(newval - unique)) -
4455 MAX(0, (int64_t)(ds->ds_reserved - unique));
4456 ds->ds_reserved = newval;
4457 mutex_exit(&ds->ds_lock);
4458
4459 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
4460 mutex_exit(&ds->ds_dir->dd_lock);
4461 }
4462
4463 static void
dsl_dataset_set_refreservation_sync(void * arg,dmu_tx_t * tx)4464 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
4465 {
4466 dsl_dataset_set_qr_arg_t *ddsqra = arg;
4467 dsl_pool_t *dp = dmu_tx_pool(tx);
4468 dsl_dataset_t *ds = NULL;
4469
4470 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
4471 dsl_dataset_set_refreservation_sync_impl(ds,
4472 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
4473 dsl_dataset_rele(ds, FTAG);
4474 }
4475
4476 int
dsl_dataset_set_refreservation(const char * dsname,zprop_source_t source,uint64_t refreservation)4477 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
4478 uint64_t refreservation)
4479 {
4480 dsl_dataset_set_qr_arg_t ddsqra;
4481
4482 ddsqra.ddsqra_name = dsname;
4483 ddsqra.ddsqra_source = source;
4484 ddsqra.ddsqra_value = refreservation;
4485
4486 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
4487 dsl_dataset_set_refreservation_sync, &ddsqra, 0,
4488 ZFS_SPACE_CHECK_EXTRA_RESERVED));
4489 }
4490
4491 typedef struct dsl_dataset_set_compression_arg {
4492 const char *ddsca_name;
4493 zprop_source_t ddsca_source;
4494 uint64_t ddsca_value;
4495 } dsl_dataset_set_compression_arg_t;
4496
4497 static int
dsl_dataset_set_compression_check(void * arg,dmu_tx_t * tx)4498 dsl_dataset_set_compression_check(void *arg, dmu_tx_t *tx)
4499 {
4500 dsl_dataset_set_compression_arg_t *ddsca = arg;
4501 dsl_pool_t *dp = dmu_tx_pool(tx);
4502
4503 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4504 spa_feature_t f = zio_compress_to_feature(compval);
4505
4506 if (f == SPA_FEATURE_NONE)
4507 return (SET_ERROR(EINVAL));
4508
4509 if (!spa_feature_is_enabled(dp->dp_spa, f))
4510 return (SET_ERROR(ENOTSUP));
4511
4512 return (0);
4513 }
4514
4515 static void
dsl_dataset_set_compression_sync(void * arg,dmu_tx_t * tx)4516 dsl_dataset_set_compression_sync(void *arg, dmu_tx_t *tx)
4517 {
4518 dsl_dataset_set_compression_arg_t *ddsca = arg;
4519 dsl_pool_t *dp = dmu_tx_pool(tx);
4520 dsl_dataset_t *ds = NULL;
4521
4522 uint64_t compval = ZIO_COMPRESS_ALGO(ddsca->ddsca_value);
4523 spa_feature_t f = zio_compress_to_feature(compval);
4524 ASSERT3S(f, !=, SPA_FEATURE_NONE);
4525 ASSERT3S(spa_feature_table[f].fi_type, ==, ZFEATURE_TYPE_BOOLEAN);
4526
4527 VERIFY0(dsl_dataset_hold(dp, ddsca->ddsca_name, FTAG, &ds));
4528 if (zfeature_active(f, ds->ds_feature[f]) != B_TRUE) {
4529 ds->ds_feature_activation[f] = (void *)B_TRUE;
4530 dsl_dataset_activate_feature(ds->ds_object, f,
4531 ds->ds_feature_activation[f], tx);
4532 ds->ds_feature[f] = ds->ds_feature_activation[f];
4533 }
4534 dsl_dataset_rele(ds, FTAG);
4535 }
4536
4537 int
dsl_dataset_set_compression(const char * dsname,zprop_source_t source,uint64_t compression)4538 dsl_dataset_set_compression(const char *dsname, zprop_source_t source,
4539 uint64_t compression)
4540 {
4541 dsl_dataset_set_compression_arg_t ddsca;
4542
4543 /*
4544 * The sync task is only required for zstd in order to activate
4545 * the feature flag when the property is first set.
4546 */
4547 if (ZIO_COMPRESS_ALGO(compression) != ZIO_COMPRESS_ZSTD)
4548 return (0);
4549
4550 ddsca.ddsca_name = dsname;
4551 ddsca.ddsca_source = source;
4552 ddsca.ddsca_value = compression;
4553
4554 return (dsl_sync_task(dsname, dsl_dataset_set_compression_check,
4555 dsl_dataset_set_compression_sync, &ddsca, 0,
4556 ZFS_SPACE_CHECK_EXTRA_RESERVED));
4557 }
4558
4559 /*
4560 * Return (in *usedp) the amount of space referenced by "new" that was not
4561 * referenced at the time the bookmark corresponds to. "New" may be a
4562 * snapshot or a head. The bookmark must be before new, in
4563 * new's filesystem (or its origin) -- caller verifies this.
4564 *
4565 * The written space is calculated by considering two components: First, we
4566 * ignore any freed space, and calculate the written as new's used space
4567 * minus old's used space. Next, we add in the amount of space that was freed
4568 * between the two time points, thus reducing new's used space relative to
4569 * old's. Specifically, this is the space that was born before
4570 * zbm_creation_txg, and freed before new (ie. on new's deadlist or a
4571 * previous deadlist).
4572 *
4573 * space freed [---------------------]
4574 * snapshots ---O-------O--------O-------O------
4575 * bookmark new
4576 *
4577 * Note, the bookmark's zbm_*_bytes_refd must be valid, but if the HAS_FBN
4578 * flag is not set, we will calculate the freed_before_next based on the
4579 * next snapshot's deadlist, rather than using zbm_*_freed_before_next_snap.
4580 */
4581 static int
dsl_dataset_space_written_impl(zfs_bookmark_phys_t * bmp,dsl_dataset_t * new,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4582 dsl_dataset_space_written_impl(zfs_bookmark_phys_t *bmp,
4583 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4584 {
4585 int err = 0;
4586 dsl_pool_t *dp = new->ds_dir->dd_pool;
4587
4588 ASSERT(dsl_pool_config_held(dp));
4589 if (dsl_dataset_is_snapshot(new)) {
4590 ASSERT3U(bmp->zbm_creation_txg, <,
4591 dsl_dataset_phys(new)->ds_creation_txg);
4592 }
4593
4594 *usedp = 0;
4595 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
4596 *usedp -= bmp->zbm_referenced_bytes_refd;
4597
4598 *compp = 0;
4599 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
4600 *compp -= bmp->zbm_compressed_bytes_refd;
4601
4602 *uncompp = 0;
4603 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
4604 *uncompp -= bmp->zbm_uncompressed_bytes_refd;
4605
4606 dsl_dataset_t *snap = new;
4607
4608 while (dsl_dataset_phys(snap)->ds_prev_snap_txg >
4609 bmp->zbm_creation_txg) {
4610 uint64_t used, comp, uncomp;
4611
4612 dsl_deadlist_space_range(&snap->ds_deadlist,
4613 0, bmp->zbm_creation_txg,
4614 &used, &comp, &uncomp);
4615 *usedp += used;
4616 *compp += comp;
4617 *uncompp += uncomp;
4618
4619 uint64_t snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4620 if (snap != new)
4621 dsl_dataset_rele(snap, FTAG);
4622 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4623 if (err != 0)
4624 break;
4625 }
4626
4627 /*
4628 * We might not have the FBN if we are calculating written from
4629 * a snapshot (because we didn't know the correct "next" snapshot
4630 * until now).
4631 */
4632 if (bmp->zbm_flags & ZBM_FLAG_HAS_FBN) {
4633 *usedp += bmp->zbm_referenced_freed_before_next_snap;
4634 *compp += bmp->zbm_compressed_freed_before_next_snap;
4635 *uncompp += bmp->zbm_uncompressed_freed_before_next_snap;
4636 } else {
4637 ASSERT3U(dsl_dataset_phys(snap)->ds_prev_snap_txg, ==,
4638 bmp->zbm_creation_txg);
4639 uint64_t used, comp, uncomp;
4640 dsl_deadlist_space(&snap->ds_deadlist, &used, &comp, &uncomp);
4641 *usedp += used;
4642 *compp += comp;
4643 *uncompp += uncomp;
4644 }
4645 if (snap != new)
4646 dsl_dataset_rele(snap, FTAG);
4647 return (err);
4648 }
4649
4650 /*
4651 * Return (in *usedp) the amount of space written in new that was not
4652 * present at the time the bookmark corresponds to. New may be a
4653 * snapshot or the head. Old must be a bookmark before new, in
4654 * new's filesystem (or its origin) -- caller verifies this.
4655 */
4656 int
dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t * bmp,dsl_dataset_t * new,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4657 dsl_dataset_space_written_bookmark(zfs_bookmark_phys_t *bmp,
4658 dsl_dataset_t *new, uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4659 {
4660 if (!(bmp->zbm_flags & ZBM_FLAG_HAS_FBN))
4661 return (SET_ERROR(ENOTSUP));
4662 return (dsl_dataset_space_written_impl(bmp, new,
4663 usedp, compp, uncompp));
4664 }
4665
4666 /*
4667 * Return (in *usedp) the amount of space written in new that is not
4668 * present in oldsnap. New may be a snapshot or the head. Old must be
4669 * a snapshot before new, in new's filesystem (or its origin). If not then
4670 * fail and return EINVAL.
4671 */
4672 int
dsl_dataset_space_written(dsl_dataset_t * oldsnap,dsl_dataset_t * new,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4673 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4674 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4675 {
4676 if (!dsl_dataset_is_before(new, oldsnap, 0))
4677 return (SET_ERROR(EINVAL));
4678
4679 zfs_bookmark_phys_t zbm = { 0 };
4680 dsl_dataset_phys_t *dsp = dsl_dataset_phys(oldsnap);
4681 zbm.zbm_guid = dsp->ds_guid;
4682 zbm.zbm_creation_txg = dsp->ds_creation_txg;
4683 zbm.zbm_creation_time = dsp->ds_creation_time;
4684 zbm.zbm_referenced_bytes_refd = dsp->ds_referenced_bytes;
4685 zbm.zbm_compressed_bytes_refd = dsp->ds_compressed_bytes;
4686 zbm.zbm_uncompressed_bytes_refd = dsp->ds_uncompressed_bytes;
4687
4688 /*
4689 * If oldsnap is the origin (or origin's origin, ...) of new,
4690 * we can't easily calculate the effective FBN. Therefore,
4691 * we do not set ZBM_FLAG_HAS_FBN, so that the _impl will calculate
4692 * it relative to the correct "next": the next snapshot towards "new",
4693 * rather than the next snapshot in oldsnap's dsl_dir.
4694 */
4695 return (dsl_dataset_space_written_impl(&zbm, new,
4696 usedp, compp, uncompp));
4697 }
4698
4699 /*
4700 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4701 * lastsnap, and all snapshots in between are deleted.
4702 *
4703 * blocks that would be freed [---------------------------]
4704 * snapshots ---O-------O--------O-------O--------O
4705 * firstsnap lastsnap
4706 *
4707 * This is the set of blocks that were born after the snap before firstsnap,
4708 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4709 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4710 * We calculate this by iterating over the relevant deadlists (from the snap
4711 * after lastsnap, backward to the snap after firstsnap), summing up the
4712 * space on the deadlist that was born after the snap before firstsnap.
4713 */
4714 int
dsl_dataset_space_wouldfree(dsl_dataset_t * firstsnap,dsl_dataset_t * lastsnap,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4715 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4716 dsl_dataset_t *lastsnap,
4717 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4718 {
4719 int err = 0;
4720 uint64_t snapobj;
4721 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4722
4723 ASSERT(firstsnap->ds_is_snapshot);
4724 ASSERT(lastsnap->ds_is_snapshot);
4725
4726 /*
4727 * Check that the snapshots are in the same dsl_dir, and firstsnap
4728 * is before lastsnap.
4729 */
4730 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4731 dsl_dataset_phys(firstsnap)->ds_creation_txg >
4732 dsl_dataset_phys(lastsnap)->ds_creation_txg)
4733 return (SET_ERROR(EINVAL));
4734
4735 *usedp = *compp = *uncompp = 0;
4736
4737 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4738 while (snapobj != firstsnap->ds_object) {
4739 dsl_dataset_t *ds;
4740 uint64_t used, comp, uncomp;
4741
4742 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4743 if (err != 0)
4744 break;
4745
4746 dsl_deadlist_space_range(&ds->ds_deadlist,
4747 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4748 &used, &comp, &uncomp);
4749 *usedp += used;
4750 *compp += comp;
4751 *uncompp += uncomp;
4752
4753 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4754 ASSERT3U(snapobj, !=, 0);
4755 dsl_dataset_rele(ds, FTAG);
4756 }
4757 return (err);
4758 }
4759
4760 /*
4761 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4762 * For example, they could both be snapshots of the same filesystem, and
4763 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
4764 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
4765 * filesystem. Or 'earlier' could be the origin's origin.
4766 *
4767 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4768 */
4769 boolean_t
dsl_dataset_is_before(dsl_dataset_t * later,dsl_dataset_t * earlier,uint64_t earlier_txg)4770 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4771 uint64_t earlier_txg)
4772 {
4773 dsl_pool_t *dp = later->ds_dir->dd_pool;
4774 int error;
4775 boolean_t ret;
4776
4777 ASSERT(dsl_pool_config_held(dp));
4778 ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4779
4780 if (earlier_txg == 0)
4781 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4782
4783 if (later->ds_is_snapshot &&
4784 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4785 return (B_FALSE);
4786
4787 if (later->ds_dir == earlier->ds_dir)
4788 return (B_TRUE);
4789
4790 /*
4791 * We check dd_origin_obj explicitly here rather than using
4792 * dsl_dir_is_clone() so that we will return TRUE if "earlier"
4793 * is $ORIGIN@$ORIGIN. dsl_dataset_space_written() depends on
4794 * this behavior.
4795 */
4796 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == 0)
4797 return (B_FALSE);
4798
4799 dsl_dataset_t *origin;
4800 error = dsl_dataset_hold_obj(dp,
4801 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4802 if (error != 0)
4803 return (B_FALSE);
4804 if (dsl_dataset_phys(origin)->ds_creation_txg == earlier_txg &&
4805 origin->ds_dir == earlier->ds_dir) {
4806 dsl_dataset_rele(origin, FTAG);
4807 return (B_TRUE);
4808 }
4809 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4810 dsl_dataset_rele(origin, FTAG);
4811 return (ret);
4812 }
4813
4814 void
dsl_dataset_zapify(dsl_dataset_t * ds,dmu_tx_t * tx)4815 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4816 {
4817 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4818 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4819 }
4820
4821 boolean_t
dsl_dataset_is_zapified(dsl_dataset_t * ds)4822 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4823 {
4824 dmu_object_info_t doi;
4825
4826 dmu_object_info_from_db(ds->ds_dbuf, &doi);
4827 return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4828 }
4829
4830 boolean_t
dsl_dataset_has_resume_receive_state(dsl_dataset_t * ds)4831 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4832 {
4833 return (dsl_dataset_is_zapified(ds) &&
4834 zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4835 ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4836 }
4837
4838 uint64_t
dsl_dataset_get_remap_deadlist_object(dsl_dataset_t * ds)4839 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4840 {
4841 uint64_t remap_deadlist_obj;
4842 int err;
4843
4844 if (!dsl_dataset_is_zapified(ds))
4845 return (0);
4846
4847 err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4848 DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4849 &remap_deadlist_obj);
4850
4851 if (err != 0) {
4852 VERIFY3S(err, ==, ENOENT);
4853 return (0);
4854 }
4855
4856 ASSERT(remap_deadlist_obj != 0);
4857 return (remap_deadlist_obj);
4858 }
4859
4860 boolean_t
dsl_dataset_remap_deadlist_exists(dsl_dataset_t * ds)4861 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4862 {
4863 EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4864 dsl_dataset_get_remap_deadlist_object(ds) != 0);
4865 return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4866 }
4867
4868 static void
dsl_dataset_set_remap_deadlist_object(dsl_dataset_t * ds,uint64_t obj,dmu_tx_t * tx)4869 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4870 dmu_tx_t *tx)
4871 {
4872 ASSERT(obj != 0);
4873 dsl_dataset_zapify(ds, tx);
4874 VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4875 DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4876 }
4877
4878 static void
dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t * ds,dmu_tx_t * tx)4879 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4880 {
4881 VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4882 ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4883 }
4884
4885 void
dsl_dataset_destroy_remap_deadlist(dsl_dataset_t * ds,dmu_tx_t * tx)4886 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4887 {
4888 uint64_t remap_deadlist_object;
4889 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4890
4891 ASSERT(dmu_tx_is_syncing(tx));
4892 ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4893
4894 remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4895 dsl_deadlist_close(&ds->ds_remap_deadlist);
4896 dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4897 dsl_dataset_unset_remap_deadlist_object(ds, tx);
4898 spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4899 }
4900
4901 void
dsl_dataset_create_remap_deadlist(dsl_dataset_t * ds,dmu_tx_t * tx)4902 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4903 {
4904 uint64_t remap_deadlist_obj;
4905 spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4906
4907 ASSERT(dmu_tx_is_syncing(tx));
4908 ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4909 /*
4910 * Currently we only create remap deadlists when there are indirect
4911 * vdevs with referenced mappings.
4912 */
4913 ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4914
4915 remap_deadlist_obj = dsl_deadlist_clone(
4916 &ds->ds_deadlist, UINT64_MAX,
4917 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4918 dsl_dataset_set_remap_deadlist_object(ds,
4919 remap_deadlist_obj, tx);
4920 dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4921 remap_deadlist_obj);
4922 spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4923 }
4924
4925 void
dsl_dataset_activate_redaction(dsl_dataset_t * ds,uint64_t * redact_snaps,uint64_t num_redact_snaps,dmu_tx_t * tx)4926 dsl_dataset_activate_redaction(dsl_dataset_t *ds, uint64_t *redact_snaps,
4927 uint64_t num_redact_snaps, dmu_tx_t *tx)
4928 {
4929 uint64_t dsobj = ds->ds_object;
4930 struct feature_type_uint64_array_arg *ftuaa =
4931 kmem_zalloc(sizeof (*ftuaa), KM_SLEEP);
4932 ftuaa->length = (int64_t)num_redact_snaps;
4933 if (num_redact_snaps > 0) {
4934 ftuaa->array = kmem_alloc(num_redact_snaps * sizeof (uint64_t),
4935 KM_SLEEP);
4936 memcpy(ftuaa->array, redact_snaps, num_redact_snaps *
4937 sizeof (uint64_t));
4938 }
4939 dsl_dataset_activate_feature(dsobj, SPA_FEATURE_REDACTED_DATASETS,
4940 ftuaa, tx);
4941 ds->ds_feature[SPA_FEATURE_REDACTED_DATASETS] = ftuaa;
4942 }
4943
4944 /*
4945 * Find and return (in *oldest_dsobj) the oldest snapshot of the dsobj
4946 * dataset whose birth time is >= min_txg.
4947 */
4948 int
dsl_dataset_oldest_snapshot(spa_t * spa,uint64_t head_ds,uint64_t min_txg,uint64_t * oldest_dsobj)4949 dsl_dataset_oldest_snapshot(spa_t *spa, uint64_t head_ds, uint64_t min_txg,
4950 uint64_t *oldest_dsobj)
4951 {
4952 dsl_dataset_t *ds;
4953 dsl_pool_t *dp = spa->spa_dsl_pool;
4954
4955 int error = dsl_dataset_hold_obj(dp, head_ds, FTAG, &ds);
4956 if (error != 0)
4957 return (error);
4958
4959 uint64_t prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4960 uint64_t prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
4961
4962 while (prev_obj != 0 && min_txg < prev_obj_txg) {
4963 dsl_dataset_rele(ds, FTAG);
4964 if ((error = dsl_dataset_hold_obj(dp, prev_obj,
4965 FTAG, &ds)) != 0)
4966 return (error);
4967 prev_obj_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
4968 prev_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4969 }
4970 *oldest_dsobj = ds->ds_object;
4971 dsl_dataset_rele(ds, FTAG);
4972 return (0);
4973 }
4974
4975 ZFS_MODULE_PARAM(zfs, zfs_, max_recordsize, UINT, ZMOD_RW,
4976 "Max allowed record size");
4977
4978 ZFS_MODULE_PARAM(zfs, zfs_, allow_redacted_dataset_mount, INT, ZMOD_RW,
4979 "Allow mounting of redacted datasets");
4980
4981 ZFS_MODULE_PARAM(zfs, zfs_, snapshot_history_enabled, INT, ZMOD_RW,
4982 "Include snapshot events in pool history/events");
4983
4984 EXPORT_SYMBOL(dsl_dataset_hold);
4985 EXPORT_SYMBOL(dsl_dataset_hold_flags);
4986 EXPORT_SYMBOL(dsl_dataset_hold_obj);
4987 EXPORT_SYMBOL(dsl_dataset_hold_obj_flags);
4988 EXPORT_SYMBOL(dsl_dataset_own);
4989 EXPORT_SYMBOL(dsl_dataset_own_obj);
4990 EXPORT_SYMBOL(dsl_dataset_name);
4991 EXPORT_SYMBOL(dsl_dataset_rele);
4992 EXPORT_SYMBOL(dsl_dataset_rele_flags);
4993 EXPORT_SYMBOL(dsl_dataset_disown);
4994 EXPORT_SYMBOL(dsl_dataset_tryown);
4995 EXPORT_SYMBOL(dsl_dataset_create_sync);
4996 EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
4997 EXPORT_SYMBOL(dsl_dataset_snapshot_check);
4998 EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
4999 EXPORT_SYMBOL(dsl_dataset_promote);
5000 EXPORT_SYMBOL(dsl_dataset_user_hold);
5001 EXPORT_SYMBOL(dsl_dataset_user_release);
5002 EXPORT_SYMBOL(dsl_dataset_get_holds);
5003 EXPORT_SYMBOL(dsl_dataset_get_blkptr);
5004 EXPORT_SYMBOL(dsl_dataset_get_spa);
5005 EXPORT_SYMBOL(dsl_dataset_modified_since_snap);
5006 EXPORT_SYMBOL(dsl_dataset_space_written);
5007 EXPORT_SYMBOL(dsl_dataset_space_wouldfree);
5008 EXPORT_SYMBOL(dsl_dataset_sync);
5009 EXPORT_SYMBOL(dsl_dataset_block_born);
5010 EXPORT_SYMBOL(dsl_dataset_block_kill);
5011 EXPORT_SYMBOL(dsl_dataset_dirty);
5012 EXPORT_SYMBOL(dsl_dataset_stats);
5013 EXPORT_SYMBOL(dsl_dataset_fast_stat);
5014 EXPORT_SYMBOL(dsl_dataset_space);
5015 EXPORT_SYMBOL(dsl_dataset_fsid_guid);
5016 EXPORT_SYMBOL(dsl_dsobj_to_dsname);
5017 EXPORT_SYMBOL(dsl_dataset_check_quota);
5018 EXPORT_SYMBOL(dsl_dataset_clone_swap_check_impl);
5019 EXPORT_SYMBOL(dsl_dataset_clone_swap_sync_impl);
5020