1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 */
26
27 #include <sys/dmu_objset.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/arc.h>
36 #include <sys/zio.h>
37 #include <sys/zap.h>
38 #include <sys/unique.h>
39 #include <sys/zfs_context.h>
40 #include <sys/zfs_ioctl.h>
41 #include <sys/spa.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_onexit.h>
44 #include <sys/zvol.h>
45 #include <sys/dsl_scan.h>
46 #include <sys/dsl_deadlist.h>
47
48 static char *dsl_reaper = "the grim reaper";
49
50 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
51 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
52 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
53
54 #define SWITCH64(x, y) \
55 { \
56 uint64_t __tmp = (x); \
57 (x) = (y); \
58 (y) = __tmp; \
59 }
60
61 #define DS_REF_MAX (1ULL << 62)
62
63 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
64
65 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
66
67
68 /*
69 * Figure out how much of this delta should be propogated to the dsl_dir
70 * layer. If there's a refreservation, that space has already been
71 * partially accounted for in our ancestors.
72 */
73 static int64_t
74 parent_delta(dsl_dataset_t *ds, int64_t delta)
75 {
76 uint64_t old_bytes, new_bytes;
77
78 if (ds->ds_reserved == 0)
79 return (delta);
80
81 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
82 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
83
84 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
85 return (new_bytes - old_bytes);
86 }
87
88 void
89 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
90 {
91 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
92 int compressed = BP_GET_PSIZE(bp);
93 int uncompressed = BP_GET_UCSIZE(bp);
94 int64_t delta;
95
96 dprintf_bp(bp, "ds=%p", ds);
97
98 ASSERT(dmu_tx_is_syncing(tx));
99 /* It could have been compressed away to nothing */
100 if (BP_IS_HOLE(bp))
101 return;
102 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
103 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
104 if (ds == NULL) {
105 /*
106 * Account for the meta-objset space in its placeholder
107 * dsl_dir.
108 */
109 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
110 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
111 used, compressed, uncompressed, tx);
112 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
113 return;
114 }
115 dmu_buf_will_dirty(ds->ds_dbuf, tx);
116
117 mutex_enter(&ds->ds_dir->dd_lock);
118 mutex_enter(&ds->ds_lock);
119 delta = parent_delta(ds, used);
120 ds->ds_phys->ds_used_bytes += used;
121 ds->ds_phys->ds_compressed_bytes += compressed;
122 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
123 ds->ds_phys->ds_unique_bytes += used;
124 mutex_exit(&ds->ds_lock);
125 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
126 compressed, uncompressed, tx);
127 dsl_dir_transfer_space(ds->ds_dir, used - delta,
128 DD_USED_REFRSRV, DD_USED_HEAD, tx);
129 mutex_exit(&ds->ds_dir->dd_lock);
130 }
131
132 int
133 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
134 boolean_t async)
135 {
136 if (BP_IS_HOLE(bp))
137 return (0);
138
139 ASSERT(dmu_tx_is_syncing(tx));
140 ASSERT(bp->blk_birth <= tx->tx_txg);
141
142 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
143 int compressed = BP_GET_PSIZE(bp);
144 int uncompressed = BP_GET_UCSIZE(bp);
145
146 ASSERT(used > 0);
147 if (ds == NULL) {
148 /*
149 * Account for the meta-objset space in its placeholder
150 * dataset.
151 */
152 dsl_free(tx->tx_pool, tx->tx_txg, bp);
153
154 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
155 -used, -compressed, -uncompressed, tx);
156 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
157 return (used);
158 }
159 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
160
161 ASSERT(!dsl_dataset_is_snapshot(ds));
162 dmu_buf_will_dirty(ds->ds_dbuf, tx);
163
164 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
165 int64_t delta;
166
167 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
168 dsl_free(tx->tx_pool, tx->tx_txg, bp);
169
170 mutex_enter(&ds->ds_dir->dd_lock);
171 mutex_enter(&ds->ds_lock);
172 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
173 !DS_UNIQUE_IS_ACCURATE(ds));
174 delta = parent_delta(ds, -used);
175 ds->ds_phys->ds_unique_bytes -= used;
176 mutex_exit(&ds->ds_lock);
177 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
178 delta, -compressed, -uncompressed, tx);
179 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
180 DD_USED_REFRSRV, DD_USED_HEAD, tx);
181 mutex_exit(&ds->ds_dir->dd_lock);
182 } else {
183 dprintf_bp(bp, "putting on dead list: %s", "");
184 if (async) {
185 /*
186 * We are here as part of zio's write done callback,
187 * which means we're a zio interrupt thread. We can't
188 * call dsl_deadlist_insert() now because it may block
189 * waiting for I/O. Instead, put bp on the deferred
190 * queue and let dsl_pool_sync() finish the job.
191 */
192 bplist_append(&ds->ds_pending_deadlist, bp);
193 } else {
194 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
195 }
196 ASSERT3U(ds->ds_prev->ds_object, ==,
197 ds->ds_phys->ds_prev_snap_obj);
198 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
199 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
200 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
201 ds->ds_object && bp->blk_birth >
202 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
203 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
204 mutex_enter(&ds->ds_prev->ds_lock);
205 ds->ds_prev->ds_phys->ds_unique_bytes += used;
206 mutex_exit(&ds->ds_prev->ds_lock);
207 }
208 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
209 dsl_dir_transfer_space(ds->ds_dir, used,
210 DD_USED_HEAD, DD_USED_SNAP, tx);
211 }
212 }
213 mutex_enter(&ds->ds_lock);
214 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
215 ds->ds_phys->ds_used_bytes -= used;
216 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
217 ds->ds_phys->ds_compressed_bytes -= compressed;
218 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
219 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
220 mutex_exit(&ds->ds_lock);
221
222 return (used);
223 }
224
225 uint64_t
226 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
227 {
228 uint64_t trysnap = 0;
229
230 if (ds == NULL)
231 return (0);
232 /*
233 * The snapshot creation could fail, but that would cause an
234 * incorrect FALSE return, which would only result in an
235 * overestimation of the amount of space that an operation would
236 * consume, which is OK.
237 *
238 * There's also a small window where we could miss a pending
239 * snapshot, because we could set the sync task in the quiescing
240 * phase. So this should only be used as a guess.
241 */
242 if (ds->ds_trysnap_txg >
243 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
244 trysnap = ds->ds_trysnap_txg;
245 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
246 }
247
248 boolean_t
249 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
250 uint64_t blk_birth)
251 {
252 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
253 return (B_FALSE);
254
255 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
256
257 return (B_TRUE);
258 }
259
260 /* ARGSUSED */
261 static void
262 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
263 {
264 dsl_dataset_t *ds = dsv;
265
266 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
267
268 unique_remove(ds->ds_fsid_guid);
269
270 if (ds->ds_objset != NULL)
271 dmu_objset_evict(ds->ds_objset);
272
273 if (ds->ds_prev) {
274 dsl_dataset_drop_ref(ds->ds_prev, ds);
275 ds->ds_prev = NULL;
276 }
277
278 bplist_destroy(&ds->ds_pending_deadlist);
279 if (db != NULL) {
280 dsl_deadlist_close(&ds->ds_deadlist);
281 } else {
282 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
283 ASSERT(!ds->ds_deadlist.dl_oldfmt);
284 }
285 if (ds->ds_dir)
286 dsl_dir_close(ds->ds_dir, ds);
287
288 ASSERT(!list_link_active(&ds->ds_synced_link));
289
290 mutex_destroy(&ds->ds_lock);
291 mutex_destroy(&ds->ds_recvlock);
292 mutex_destroy(&ds->ds_opening_lock);
293 rw_destroy(&ds->ds_rwlock);
294 cv_destroy(&ds->ds_exclusive_cv);
295
296 kmem_free(ds, sizeof (dsl_dataset_t));
297 }
298
299 static int
300 dsl_dataset_get_snapname(dsl_dataset_t *ds)
301 {
302 dsl_dataset_phys_t *headphys;
303 int err;
304 dmu_buf_t *headdbuf;
305 dsl_pool_t *dp = ds->ds_dir->dd_pool;
306 objset_t *mos = dp->dp_meta_objset;
307
308 if (ds->ds_snapname[0])
309 return (0);
310 if (ds->ds_phys->ds_next_snap_obj == 0)
311 return (0);
312
313 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
314 FTAG, &headdbuf);
315 if (err)
316 return (err);
317 headphys = headdbuf->db_data;
318 err = zap_value_search(dp->dp_meta_objset,
319 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
320 dmu_buf_rele(headdbuf, FTAG);
321 return (err);
322 }
323
324 static int
325 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
326 {
327 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
328 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
329 matchtype_t mt;
330 int err;
331
332 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
333 mt = MT_FIRST;
334 else
335 mt = MT_EXACT;
336
337 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
338 value, mt, NULL, 0, NULL);
339 if (err == ENOTSUP && mt == MT_FIRST)
340 err = zap_lookup(mos, snapobj, name, 8, 1, value);
341 return (err);
342 }
343
344 static int
345 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
346 {
347 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
348 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
349 matchtype_t mt;
350 int err;
351
352 dsl_dir_snap_cmtime_update(ds->ds_dir);
353
354 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
355 mt = MT_FIRST;
356 else
357 mt = MT_EXACT;
358
359 err = zap_remove_norm(mos, snapobj, name, mt, tx);
360 if (err == ENOTSUP && mt == MT_FIRST)
361 err = zap_remove(mos, snapobj, name, tx);
362 return (err);
363 }
364
365 static int
366 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
367 dsl_dataset_t **dsp)
368 {
369 objset_t *mos = dp->dp_meta_objset;
370 dmu_buf_t *dbuf;
371 dsl_dataset_t *ds;
372 int err;
373 dmu_object_info_t doi;
374
375 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
376 dsl_pool_sync_context(dp));
377
378 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
379 if (err)
380 return (err);
381
382 /* Make sure dsobj has the correct object type. */
383 dmu_object_info_from_db(dbuf, &doi);
384 if (doi.doi_type != DMU_OT_DSL_DATASET)
385 return (EINVAL);
386
387 ds = dmu_buf_get_user(dbuf);
388 if (ds == NULL) {
389 dsl_dataset_t *winner;
390
391 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
392 ds->ds_dbuf = dbuf;
393 ds->ds_object = dsobj;
394 ds->ds_phys = dbuf->db_data;
395
396 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
397 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
398 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
399 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
400
401 rw_init(&ds->ds_rwlock, 0, 0, 0);
402 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
403
404 bplist_create(&ds->ds_pending_deadlist);
405 dsl_deadlist_open(&ds->ds_deadlist,
406 mos, ds->ds_phys->ds_deadlist_obj);
407
408 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
409 offsetof(dmu_sendarg_t, dsa_link));
410
411 if (err == 0) {
412 err = dsl_dir_open_obj(dp,
413 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
414 }
415 if (err) {
416 mutex_destroy(&ds->ds_lock);
417 mutex_destroy(&ds->ds_recvlock);
418 mutex_destroy(&ds->ds_opening_lock);
419 rw_destroy(&ds->ds_rwlock);
420 cv_destroy(&ds->ds_exclusive_cv);
421 bplist_destroy(&ds->ds_pending_deadlist);
422 dsl_deadlist_close(&ds->ds_deadlist);
423 kmem_free(ds, sizeof (dsl_dataset_t));
424 dmu_buf_rele(dbuf, tag);
425 return (err);
426 }
427
428 if (!dsl_dataset_is_snapshot(ds)) {
429 ds->ds_snapname[0] = '\0';
430 if (ds->ds_phys->ds_prev_snap_obj) {
431 err = dsl_dataset_get_ref(dp,
432 ds->ds_phys->ds_prev_snap_obj,
433 ds, &ds->ds_prev);
434 }
435 } else {
436 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
437 err = dsl_dataset_get_snapname(ds);
438 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
439 err = zap_count(
440 ds->ds_dir->dd_pool->dp_meta_objset,
441 ds->ds_phys->ds_userrefs_obj,
442 &ds->ds_userrefs);
443 }
444 }
445
446 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
447 /*
448 * In sync context, we're called with either no lock
449 * or with the write lock. If we're not syncing,
450 * we're always called with the read lock held.
451 */
452 boolean_t need_lock =
453 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
454 dsl_pool_sync_context(dp);
455
456 if (need_lock)
457 rw_enter(&dp->dp_config_rwlock, RW_READER);
458
459 err = dsl_prop_get_ds(ds,
460 "refreservation", sizeof (uint64_t), 1,
461 &ds->ds_reserved, NULL);
462 if (err == 0) {
463 err = dsl_prop_get_ds(ds,
464 "refquota", sizeof (uint64_t), 1,
465 &ds->ds_quota, NULL);
466 }
467
468 if (need_lock)
469 rw_exit(&dp->dp_config_rwlock);
470 } else {
471 ds->ds_reserved = ds->ds_quota = 0;
472 }
473
474 if (err == 0) {
475 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
476 dsl_dataset_evict);
477 }
478 if (err || winner) {
479 bplist_destroy(&ds->ds_pending_deadlist);
480 dsl_deadlist_close(&ds->ds_deadlist);
481 if (ds->ds_prev)
482 dsl_dataset_drop_ref(ds->ds_prev, ds);
483 dsl_dir_close(ds->ds_dir, ds);
484 mutex_destroy(&ds->ds_lock);
485 mutex_destroy(&ds->ds_recvlock);
486 mutex_destroy(&ds->ds_opening_lock);
487 rw_destroy(&ds->ds_rwlock);
488 cv_destroy(&ds->ds_exclusive_cv);
489 kmem_free(ds, sizeof (dsl_dataset_t));
490 if (err) {
491 dmu_buf_rele(dbuf, tag);
492 return (err);
493 }
494 ds = winner;
495 } else {
496 ds->ds_fsid_guid =
497 unique_insert(ds->ds_phys->ds_fsid_guid);
498 }
499 }
500 ASSERT3P(ds->ds_dbuf, ==, dbuf);
501 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
502 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
503 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
504 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
505 mutex_enter(&ds->ds_lock);
506 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
507 mutex_exit(&ds->ds_lock);
508 dmu_buf_rele(ds->ds_dbuf, tag);
509 return (ENOENT);
510 }
511 mutex_exit(&ds->ds_lock);
512 *dsp = ds;
513 return (0);
514 }
515
516 static int
517 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
518 {
519 dsl_pool_t *dp = ds->ds_dir->dd_pool;
520
521 /*
522 * In syncing context we don't want the rwlock lock: there
523 * may be an existing writer waiting for sync phase to
524 * finish. We don't need to worry about such writers, since
525 * sync phase is single-threaded, so the writer can't be
526 * doing anything while we are active.
527 */
528 if (dsl_pool_sync_context(dp)) {
529 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
530 return (0);
531 }
532
533 /*
534 * Normal users will hold the ds_rwlock as a READER until they
535 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
536 * drop their READER lock after they set the ds_owner field.
537 *
538 * If the dataset is being destroyed, the destroy thread will
539 * obtain a WRITER lock for exclusive access after it's done its
540 * open-context work and then change the ds_owner to
541 * dsl_reaper once destruction is assured. So threads
542 * may block here temporarily, until the "destructability" of
543 * the dataset is determined.
544 */
545 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
546 mutex_enter(&ds->ds_lock);
547 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
548 rw_exit(&dp->dp_config_rwlock);
549 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
550 if (DSL_DATASET_IS_DESTROYED(ds)) {
551 mutex_exit(&ds->ds_lock);
552 dsl_dataset_drop_ref(ds, tag);
553 rw_enter(&dp->dp_config_rwlock, RW_READER);
554 return (ENOENT);
555 }
556 /*
557 * The dp_config_rwlock lives above the ds_lock. And
558 * we need to check DSL_DATASET_IS_DESTROYED() while
559 * holding the ds_lock, so we have to drop and reacquire
560 * the ds_lock here.
561 */
562 mutex_exit(&ds->ds_lock);
563 rw_enter(&dp->dp_config_rwlock, RW_READER);
564 mutex_enter(&ds->ds_lock);
565 }
566 mutex_exit(&ds->ds_lock);
567 return (0);
568 }
569
570 int
571 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
572 dsl_dataset_t **dsp)
573 {
574 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
575
576 if (err)
577 return (err);
578 return (dsl_dataset_hold_ref(*dsp, tag));
579 }
580
581 int
582 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
583 void *tag, dsl_dataset_t **dsp)
584 {
585 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
586 if (err)
587 return (err);
588 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
589 dsl_dataset_rele(*dsp, tag);
590 *dsp = NULL;
591 return (EBUSY);
592 }
593 return (0);
594 }
595
596 int
597 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
598 {
599 dsl_dir_t *dd;
600 dsl_pool_t *dp;
601 const char *snapname;
602 uint64_t obj;
603 int err = 0;
604
605 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
606 if (err)
607 return (err);
608
609 dp = dd->dd_pool;
610 obj = dd->dd_phys->dd_head_dataset_obj;
611 rw_enter(&dp->dp_config_rwlock, RW_READER);
612 if (obj)
613 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
614 else
615 err = ENOENT;
616 if (err)
617 goto out;
618
619 err = dsl_dataset_hold_ref(*dsp, tag);
620
621 /* we may be looking for a snapshot */
622 if (err == 0 && snapname != NULL) {
623 dsl_dataset_t *ds = NULL;
624
625 if (*snapname++ != '@') {
626 dsl_dataset_rele(*dsp, tag);
627 err = ENOENT;
628 goto out;
629 }
630
631 dprintf("looking for snapshot '%s'\n", snapname);
632 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
633 if (err == 0)
634 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
635 dsl_dataset_rele(*dsp, tag);
636
637 ASSERT3U((err == 0), ==, (ds != NULL));
638
639 if (ds) {
640 mutex_enter(&ds->ds_lock);
641 if (ds->ds_snapname[0] == 0)
642 (void) strlcpy(ds->ds_snapname, snapname,
643 sizeof (ds->ds_snapname));
644 mutex_exit(&ds->ds_lock);
645 err = dsl_dataset_hold_ref(ds, tag);
646 *dsp = err ? NULL : ds;
647 }
648 }
649 out:
650 rw_exit(&dp->dp_config_rwlock);
651 dsl_dir_close(dd, FTAG);
652 return (err);
653 }
654
655 int
656 dsl_dataset_own(const char *name, boolean_t inconsistentok,
657 void *tag, dsl_dataset_t **dsp)
658 {
659 int err = dsl_dataset_hold(name, tag, dsp);
660 if (err)
661 return (err);
662 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
663 dsl_dataset_rele(*dsp, tag);
664 return (EBUSY);
665 }
666 return (0);
667 }
668
669 void
670 dsl_dataset_name(dsl_dataset_t *ds, char *name)
671 {
672 if (ds == NULL) {
673 (void) strcpy(name, "mos");
674 } else {
675 dsl_dir_name(ds->ds_dir, name);
676 VERIFY(0 == dsl_dataset_get_snapname(ds));
677 if (ds->ds_snapname[0]) {
678 (void) strcat(name, "@");
679 /*
680 * We use a "recursive" mutex so that we
681 * can call dprintf_ds() with ds_lock held.
682 */
683 if (!MUTEX_HELD(&ds->ds_lock)) {
684 mutex_enter(&ds->ds_lock);
685 (void) strcat(name, ds->ds_snapname);
686 mutex_exit(&ds->ds_lock);
687 } else {
688 (void) strcat(name, ds->ds_snapname);
689 }
690 }
691 }
692 }
693
694 static int
695 dsl_dataset_namelen(dsl_dataset_t *ds)
696 {
697 int result;
698
699 if (ds == NULL) {
700 result = 3; /* "mos" */
701 } else {
702 result = dsl_dir_namelen(ds->ds_dir);
703 VERIFY(0 == dsl_dataset_get_snapname(ds));
704 if (ds->ds_snapname[0]) {
705 ++result; /* adding one for the @-sign */
706 if (!MUTEX_HELD(&ds->ds_lock)) {
707 mutex_enter(&ds->ds_lock);
708 result += strlen(ds->ds_snapname);
709 mutex_exit(&ds->ds_lock);
710 } else {
711 result += strlen(ds->ds_snapname);
712 }
713 }
714 }
715
716 return (result);
717 }
718
719 void
720 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
721 {
722 dmu_buf_rele(ds->ds_dbuf, tag);
723 }
724
725 void
726 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
727 {
728 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
729 rw_exit(&ds->ds_rwlock);
730 }
731 dsl_dataset_drop_ref(ds, tag);
732 }
733
734 void
735 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
736 {
737 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
738 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
739
740 mutex_enter(&ds->ds_lock);
741 ds->ds_owner = NULL;
742 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
743 rw_exit(&ds->ds_rwlock);
744 cv_broadcast(&ds->ds_exclusive_cv);
745 }
746 mutex_exit(&ds->ds_lock);
747 if (ds->ds_dbuf)
748 dsl_dataset_drop_ref(ds, tag);
749 else
750 dsl_dataset_evict(NULL, ds);
751 }
752
753 boolean_t
754 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
755 {
756 boolean_t gotit = FALSE;
757
758 mutex_enter(&ds->ds_lock);
759 if (ds->ds_owner == NULL &&
760 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
761 ds->ds_owner = tag;
762 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
763 rw_exit(&ds->ds_rwlock);
764 gotit = TRUE;
765 }
766 mutex_exit(&ds->ds_lock);
767 return (gotit);
768 }
769
770 void
771 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
772 {
773 ASSERT3P(owner, ==, ds->ds_owner);
774 if (!RW_WRITE_HELD(&ds->ds_rwlock))
775 rw_enter(&ds->ds_rwlock, RW_WRITER);
776 }
777
778 uint64_t
779 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
780 uint64_t flags, dmu_tx_t *tx)
781 {
782 dsl_pool_t *dp = dd->dd_pool;
783 dmu_buf_t *dbuf;
784 dsl_dataset_phys_t *dsphys;
785 uint64_t dsobj;
786 objset_t *mos = dp->dp_meta_objset;
787
788 if (origin == NULL)
789 origin = dp->dp_origin_snap;
790
791 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
792 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
793 ASSERT(dmu_tx_is_syncing(tx));
794 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
795
796 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
797 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
798 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
799 dmu_buf_will_dirty(dbuf, tx);
800 dsphys = dbuf->db_data;
801 bzero(dsphys, sizeof (dsl_dataset_phys_t));
802 dsphys->ds_dir_obj = dd->dd_object;
803 dsphys->ds_flags = flags;
804 dsphys->ds_fsid_guid = unique_create();
805 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
806 sizeof (dsphys->ds_guid));
807 dsphys->ds_snapnames_zapobj =
808 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
809 DMU_OT_NONE, 0, tx);
810 dsphys->ds_creation_time = gethrestime_sec();
811 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
812
813 if (origin == NULL) {
814 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
815 } else {
816 dsl_dataset_t *ohds;
817
818 dsphys->ds_prev_snap_obj = origin->ds_object;
819 dsphys->ds_prev_snap_txg =
820 origin->ds_phys->ds_creation_txg;
821 dsphys->ds_used_bytes =
822 origin->ds_phys->ds_used_bytes;
823 dsphys->ds_compressed_bytes =
824 origin->ds_phys->ds_compressed_bytes;
825 dsphys->ds_uncompressed_bytes =
826 origin->ds_phys->ds_uncompressed_bytes;
827 dsphys->ds_bp = origin->ds_phys->ds_bp;
828 dsphys->ds_flags |= origin->ds_phys->ds_flags;
829
830 dmu_buf_will_dirty(origin->ds_dbuf, tx);
831 origin->ds_phys->ds_num_children++;
832
833 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
834 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
835 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
836 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
837 dsl_dataset_rele(ohds, FTAG);
838
839 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
840 if (origin->ds_phys->ds_next_clones_obj == 0) {
841 origin->ds_phys->ds_next_clones_obj =
842 zap_create(mos,
843 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
844 }
845 VERIFY(0 == zap_add_int(mos,
846 origin->ds_phys->ds_next_clones_obj,
847 dsobj, tx));
848 }
849
850 dmu_buf_will_dirty(dd->dd_dbuf, tx);
851 dd->dd_phys->dd_origin_obj = origin->ds_object;
852 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
853 if (origin->ds_dir->dd_phys->dd_clones == 0) {
854 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
855 origin->ds_dir->dd_phys->dd_clones =
856 zap_create(mos,
857 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
858 }
859 VERIFY3U(0, ==, zap_add_int(mos,
860 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
861 }
862 }
863
864 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
865 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
866
867 dmu_buf_rele(dbuf, FTAG);
868
869 dmu_buf_will_dirty(dd->dd_dbuf, tx);
870 dd->dd_phys->dd_head_dataset_obj = dsobj;
871
872 return (dsobj);
873 }
874
875 uint64_t
876 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
877 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
878 {
879 dsl_pool_t *dp = pdd->dd_pool;
880 uint64_t dsobj, ddobj;
881 dsl_dir_t *dd;
882
883 ASSERT(lastname[0] != '@');
884
885 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
886 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
887
888 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
889
890 dsl_deleg_set_create_perms(dd, tx, cr);
891
892 dsl_dir_close(dd, FTAG);
893
894 /*
895 * If we are creating a clone, make sure we zero out any stale
896 * data from the origin snapshots zil header.
897 */
898 if (origin != NULL) {
899 dsl_dataset_t *ds;
900 objset_t *os;
901
902 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
903 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
904 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
905 dsl_dataset_dirty(ds, tx);
906 dsl_dataset_rele(ds, FTAG);
907 }
908
909 return (dsobj);
910 }
911
912 /*
913 * The snapshots must all be in the same pool.
914 */
915 int
916 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
917 {
918 int err;
919 dsl_sync_task_t *dst;
920 spa_t *spa;
921 nvpair_t *pair;
922 dsl_sync_task_group_t *dstg;
923
924 pair = nvlist_next_nvpair(snaps, NULL);
925 if (pair == NULL)
926 return (0);
927
928 err = spa_open(nvpair_name(pair), &spa, FTAG);
929 if (err)
930 return (err);
931 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
932
933 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
934 pair = nvlist_next_nvpair(snaps, pair)) {
935 dsl_dataset_t *ds;
936 int err;
937
938 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
939 if (err == 0) {
940 struct dsl_ds_destroyarg *dsda;
941
942 dsl_dataset_make_exclusive(ds, dstg);
943 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
944 KM_SLEEP);
945 dsda->ds = ds;
946 dsda->defer = defer;
947 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
948 dsl_dataset_destroy_sync, dsda, dstg, 0);
949 } else if (err == ENOENT) {
950 err = 0;
951 } else {
952 (void) strcpy(failed, nvpair_name(pair));
953 break;
954 }
955 }
956
957 if (err == 0)
958 err = dsl_sync_task_group_wait(dstg);
959
960 for (dst = list_head(&dstg->dstg_tasks); dst;
961 dst = list_next(&dstg->dstg_tasks, dst)) {
962 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
963 dsl_dataset_t *ds = dsda->ds;
964
965 /*
966 * Return the file system name that triggered the error
967 */
968 if (dst->dst_err) {
969 dsl_dataset_name(ds, failed);
970 }
971 ASSERT3P(dsda->rm_origin, ==, NULL);
972 dsl_dataset_disown(ds, dstg);
973 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
974 }
975
976 dsl_sync_task_group_destroy(dstg);
977 spa_close(spa, FTAG);
978 return (err);
979
980 }
981
982 static boolean_t
983 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
984 {
985 boolean_t might_destroy = B_FALSE;
986
987 mutex_enter(&ds->ds_lock);
988 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
989 DS_IS_DEFER_DESTROY(ds))
990 might_destroy = B_TRUE;
991 mutex_exit(&ds->ds_lock);
992
993 return (might_destroy);
994 }
995
996 /*
997 * If we're removing a clone, and these three conditions are true:
998 * 1) the clone's origin has no other children
999 * 2) the clone's origin has no user references
1000 * 3) the clone's origin has been marked for deferred destruction
1001 * Then, prepare to remove the origin as part of this sync task group.
1002 */
1003 static int
1004 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1005 {
1006 dsl_dataset_t *ds = dsda->ds;
1007 dsl_dataset_t *origin = ds->ds_prev;
1008
1009 if (dsl_dataset_might_destroy_origin(origin)) {
1010 char *name;
1011 int namelen;
1012 int error;
1013
1014 namelen = dsl_dataset_namelen(origin) + 1;
1015 name = kmem_alloc(namelen, KM_SLEEP);
1016 dsl_dataset_name(origin, name);
1017 #ifdef _KERNEL
1018 error = zfs_unmount_snap(name, NULL);
1019 if (error) {
1020 kmem_free(name, namelen);
1021 return (error);
1022 }
1023 #endif
1024 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1025 kmem_free(name, namelen);
1026 if (error)
1027 return (error);
1028 dsda->rm_origin = origin;
1029 dsl_dataset_make_exclusive(origin, tag);
1030 }
1031
1032 return (0);
1033 }
1034
1035 /*
1036 * ds must be opened as OWNER. On return (whether successful or not),
1037 * ds will be closed and caller can no longer dereference it.
1038 */
1039 int
1040 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1041 {
1042 int err;
1043 dsl_sync_task_group_t *dstg;
1044 objset_t *os;
1045 dsl_dir_t *dd;
1046 uint64_t obj;
1047 struct dsl_ds_destroyarg dsda = { 0 };
1048 dsl_dataset_t dummy_ds = { 0 };
1049
1050 dsda.ds = ds;
1051
1052 if (dsl_dataset_is_snapshot(ds)) {
1053 /* Destroying a snapshot is simpler */
1054 dsl_dataset_make_exclusive(ds, tag);
1055
1056 dsda.defer = defer;
1057 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1058 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1059 &dsda, tag, 0);
1060 ASSERT3P(dsda.rm_origin, ==, NULL);
1061 goto out;
1062 } else if (defer) {
1063 err = EINVAL;
1064 goto out;
1065 }
1066
1067 dd = ds->ds_dir;
1068 dummy_ds.ds_dir = dd;
1069 dummy_ds.ds_object = ds->ds_object;
1070
1071 /*
1072 * Check for errors and mark this ds as inconsistent, in
1073 * case we crash while freeing the objects.
1074 */
1075 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1076 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1077 if (err)
1078 goto out;
1079
1080 err = dmu_objset_from_ds(ds, &os);
1081 if (err)
1082 goto out;
1083
1084 /*
1085 * remove the objects in open context, so that we won't
1086 * have too much to do in syncing context.
1087 */
1088 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1089 ds->ds_phys->ds_prev_snap_txg)) {
1090 /*
1091 * Ignore errors, if there is not enough disk space
1092 * we will deal with it in dsl_dataset_destroy_sync().
1093 */
1094 (void) dmu_free_object(os, obj);
1095 }
1096 if (err != ESRCH)
1097 goto out;
1098
1099 /*
1100 * Only the ZIL knows how to free log blocks.
1101 */
1102 zil_destroy(dmu_objset_zil(os), B_FALSE);
1103
1104 /*
1105 * Sync out all in-flight IO.
1106 */
1107 txg_wait_synced(dd->dd_pool, 0);
1108
1109 /*
1110 * If we managed to free all the objects in open
1111 * context, the user space accounting should be zero.
1112 */
1113 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1114 dmu_objset_userused_enabled(os)) {
1115 uint64_t count;
1116
1117 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1118 count == 0);
1119 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1120 count == 0);
1121 }
1122
1123 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1124 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1125 rw_exit(&dd->dd_pool->dp_config_rwlock);
1126
1127 if (err)
1128 goto out;
1129
1130 /*
1131 * Blow away the dsl_dir + head dataset.
1132 */
1133 dsl_dataset_make_exclusive(ds, tag);
1134 /*
1135 * If we're removing a clone, we might also need to remove its
1136 * origin.
1137 */
1138 do {
1139 dsda.need_prep = B_FALSE;
1140 if (dsl_dir_is_clone(dd)) {
1141 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1142 if (err) {
1143 dsl_dir_close(dd, FTAG);
1144 goto out;
1145 }
1146 }
1147
1148 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1149 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1150 dsl_dataset_destroy_sync, &dsda, tag, 0);
1151 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1152 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1153 err = dsl_sync_task_group_wait(dstg);
1154 dsl_sync_task_group_destroy(dstg);
1155
1156 /*
1157 * We could be racing against 'zfs release' or 'zfs destroy -d'
1158 * on the origin snap, in which case we can get EBUSY if we
1159 * needed to destroy the origin snap but were not ready to
1160 * do so.
1161 */
1162 if (dsda.need_prep) {
1163 ASSERT(err == EBUSY);
1164 ASSERT(dsl_dir_is_clone(dd));
1165 ASSERT(dsda.rm_origin == NULL);
1166 }
1167 } while (dsda.need_prep);
1168
1169 if (dsda.rm_origin != NULL)
1170 dsl_dataset_disown(dsda.rm_origin, tag);
1171
1172 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1173 if (err)
1174 dsl_dir_close(dd, FTAG);
1175 out:
1176 dsl_dataset_disown(ds, tag);
1177 return (err);
1178 }
1179
1180 blkptr_t *
1181 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1182 {
1183 return (&ds->ds_phys->ds_bp);
1184 }
1185
1186 void
1187 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1188 {
1189 ASSERT(dmu_tx_is_syncing(tx));
1190 /* If it's the meta-objset, set dp_meta_rootbp */
1191 if (ds == NULL) {
1192 tx->tx_pool->dp_meta_rootbp = *bp;
1193 } else {
1194 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1195 ds->ds_phys->ds_bp = *bp;
1196 }
1197 }
1198
1199 spa_t *
1200 dsl_dataset_get_spa(dsl_dataset_t *ds)
1201 {
1202 return (ds->ds_dir->dd_pool->dp_spa);
1203 }
1204
1205 void
1206 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1207 {
1208 dsl_pool_t *dp;
1209
1210 if (ds == NULL) /* this is the meta-objset */
1211 return;
1212
1213 ASSERT(ds->ds_objset != NULL);
1214
1215 if (ds->ds_phys->ds_next_snap_obj != 0)
1216 panic("dirtying snapshot!");
1217
1218 dp = ds->ds_dir->dd_pool;
1219
1220 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1221 /* up the hold count until we can be written out */
1222 dmu_buf_add_ref(ds->ds_dbuf, ds);
1223 }
1224 }
1225
1226 /*
1227 * The unique space in the head dataset can be calculated by subtracting
1228 * the space used in the most recent snapshot, that is still being used
1229 * in this file system, from the space currently in use. To figure out
1230 * the space in the most recent snapshot still in use, we need to take
1231 * the total space used in the snapshot and subtract out the space that
1232 * has been freed up since the snapshot was taken.
1233 */
1234 static void
1235 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1236 {
1237 uint64_t mrs_used;
1238 uint64_t dlused, dlcomp, dluncomp;
1239
1240 ASSERT(!dsl_dataset_is_snapshot(ds));
1241
1242 if (ds->ds_phys->ds_prev_snap_obj != 0)
1243 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1244 else
1245 mrs_used = 0;
1246
1247 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1248
1249 ASSERT3U(dlused, <=, mrs_used);
1250 ds->ds_phys->ds_unique_bytes =
1251 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1252
1253 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1254 SPA_VERSION_UNIQUE_ACCURATE)
1255 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1256 }
1257
1258 struct killarg {
1259 dsl_dataset_t *ds;
1260 dmu_tx_t *tx;
1261 };
1262
1263 /* ARGSUSED */
1264 static int
1265 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1266 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1267 {
1268 struct killarg *ka = arg;
1269 dmu_tx_t *tx = ka->tx;
1270
1271 if (bp == NULL)
1272 return (0);
1273
1274 if (zb->zb_level == ZB_ZIL_LEVEL) {
1275 ASSERT(zilog != NULL);
1276 /*
1277 * It's a block in the intent log. It has no
1278 * accounting, so just free it.
1279 */
1280 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1281 } else {
1282 ASSERT(zilog == NULL);
1283 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1284 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1285 }
1286
1287 return (0);
1288 }
1289
1290 /* ARGSUSED */
1291 static int
1292 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1293 {
1294 dsl_dataset_t *ds = arg1;
1295 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1296 uint64_t count;
1297 int err;
1298
1299 /*
1300 * Can't delete a head dataset if there are snapshots of it.
1301 * (Except if the only snapshots are from the branch we cloned
1302 * from.)
1303 */
1304 if (ds->ds_prev != NULL &&
1305 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1306 return (EBUSY);
1307
1308 /*
1309 * This is really a dsl_dir thing, but check it here so that
1310 * we'll be less likely to leave this dataset inconsistent &
1311 * nearly destroyed.
1312 */
1313 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1314 if (err)
1315 return (err);
1316 if (count != 0)
1317 return (EEXIST);
1318
1319 return (0);
1320 }
1321
1322 /* ARGSUSED */
1323 static void
1324 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1325 {
1326 dsl_dataset_t *ds = arg1;
1327 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1328
1329 /* Mark it as inconsistent on-disk, in case we crash */
1330 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1331 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1332
1333 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1334 "dataset = %llu", ds->ds_object);
1335 }
1336
1337 static int
1338 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1339 dmu_tx_t *tx)
1340 {
1341 dsl_dataset_t *ds = dsda->ds;
1342 dsl_dataset_t *ds_prev = ds->ds_prev;
1343
1344 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1345 struct dsl_ds_destroyarg ndsda = {0};
1346
1347 /*
1348 * If we're not prepared to remove the origin, don't remove
1349 * the clone either.
1350 */
1351 if (dsda->rm_origin == NULL) {
1352 dsda->need_prep = B_TRUE;
1353 return (EBUSY);
1354 }
1355
1356 ndsda.ds = ds_prev;
1357 ndsda.is_origin_rm = B_TRUE;
1358 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1359 }
1360
1361 /*
1362 * If we're not going to remove the origin after all,
1363 * undo the open context setup.
1364 */
1365 if (dsda->rm_origin != NULL) {
1366 dsl_dataset_disown(dsda->rm_origin, tag);
1367 dsda->rm_origin = NULL;
1368 }
1369
1370 return (0);
1371 }
1372
1373 /*
1374 * If you add new checks here, you may need to add
1375 * additional checks to the "temporary" case in
1376 * snapshot_check() in dmu_objset.c.
1377 */
1378 /* ARGSUSED */
1379 int
1380 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1381 {
1382 struct dsl_ds_destroyarg *dsda = arg1;
1383 dsl_dataset_t *ds = dsda->ds;
1384
1385 /* we have an owner hold, so noone else can destroy us */
1386 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1387
1388 /*
1389 * Only allow deferred destroy on pools that support it.
1390 * NOTE: deferred destroy is only supported on snapshots.
1391 */
1392 if (dsda->defer) {
1393 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1394 SPA_VERSION_USERREFS)
1395 return (ENOTSUP);
1396 ASSERT(dsl_dataset_is_snapshot(ds));
1397 return (0);
1398 }
1399
1400 /*
1401 * Can't delete a head dataset if there are snapshots of it.
1402 * (Except if the only snapshots are from the branch we cloned
1403 * from.)
1404 */
1405 if (ds->ds_prev != NULL &&
1406 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1407 return (EBUSY);
1408
1409 /*
1410 * If we made changes this txg, traverse_dsl_dataset won't find
1411 * them. Try again.
1412 */
1413 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1414 return (EAGAIN);
1415
1416 if (dsl_dataset_is_snapshot(ds)) {
1417 /*
1418 * If this snapshot has an elevated user reference count,
1419 * we can't destroy it yet.
1420 */
1421 if (ds->ds_userrefs > 0 && !dsda->releasing)
1422 return (EBUSY);
1423
1424 mutex_enter(&ds->ds_lock);
1425 /*
1426 * Can't delete a branch point. However, if we're destroying
1427 * a clone and removing its origin due to it having a user
1428 * hold count of 0 and having been marked for deferred destroy,
1429 * it's OK for the origin to have a single clone.
1430 */
1431 if (ds->ds_phys->ds_num_children >
1432 (dsda->is_origin_rm ? 2 : 1)) {
1433 mutex_exit(&ds->ds_lock);
1434 return (EEXIST);
1435 }
1436 mutex_exit(&ds->ds_lock);
1437 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1438 return (dsl_dataset_origin_check(dsda, arg2, tx));
1439 }
1440
1441 /* XXX we should do some i/o error checking... */
1442 return (0);
1443 }
1444
1445 struct refsarg {
1446 kmutex_t lock;
1447 boolean_t gone;
1448 kcondvar_t cv;
1449 };
1450
1451 /* ARGSUSED */
1452 static void
1453 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1454 {
1455 struct refsarg *arg = argv;
1456
1457 mutex_enter(&arg->lock);
1458 arg->gone = TRUE;
1459 cv_signal(&arg->cv);
1460 mutex_exit(&arg->lock);
1461 }
1462
1463 static void
1464 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1465 {
1466 struct refsarg arg;
1467
1468 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1469 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1470 arg.gone = FALSE;
1471 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1472 dsl_dataset_refs_gone);
1473 dmu_buf_rele(ds->ds_dbuf, tag);
1474 mutex_enter(&arg.lock);
1475 while (!arg.gone)
1476 cv_wait(&arg.cv, &arg.lock);
1477 ASSERT(arg.gone);
1478 mutex_exit(&arg.lock);
1479 ds->ds_dbuf = NULL;
1480 ds->ds_phys = NULL;
1481 mutex_destroy(&arg.lock);
1482 cv_destroy(&arg.cv);
1483 }
1484
1485 static void
1486 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1487 {
1488 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1489 uint64_t count;
1490 int err;
1491
1492 ASSERT(ds->ds_phys->ds_num_children >= 2);
1493 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1494 /*
1495 * The err should not be ENOENT, but a bug in a previous version
1496 * of the code could cause upgrade_clones_cb() to not set
1497 * ds_next_snap_obj when it should, leading to a missing entry.
1498 * If we knew that the pool was created after
1499 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1500 * ENOENT. However, at least we can check that we don't have
1501 * too many entries in the next_clones_obj even after failing to
1502 * remove this one.
1503 */
1504 if (err != ENOENT) {
1505 VERIFY3U(err, ==, 0);
1506 }
1507 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1508 &count));
1509 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1510 }
1511
1512 static void
1513 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1514 {
1515 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1516 zap_cursor_t zc;
1517 zap_attribute_t za;
1518
1519 /*
1520 * If it is the old version, dd_clones doesn't exist so we can't
1521 * find the clones, but deadlist_remove_key() is a no-op so it
1522 * doesn't matter.
1523 */
1524 if (ds->ds_dir->dd_phys->dd_clones == 0)
1525 return;
1526
1527 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1528 zap_cursor_retrieve(&zc, &za) == 0;
1529 zap_cursor_advance(&zc)) {
1530 dsl_dataset_t *clone;
1531
1532 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1533 za.za_first_integer, FTAG, &clone));
1534 if (clone->ds_dir->dd_origin_txg > mintxg) {
1535 dsl_deadlist_remove_key(&clone->ds_deadlist,
1536 mintxg, tx);
1537 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1538 }
1539 dsl_dataset_rele(clone, FTAG);
1540 }
1541 zap_cursor_fini(&zc);
1542 }
1543
1544 struct process_old_arg {
1545 dsl_dataset_t *ds;
1546 dsl_dataset_t *ds_prev;
1547 boolean_t after_branch_point;
1548 zio_t *pio;
1549 uint64_t used, comp, uncomp;
1550 };
1551
1552 static int
1553 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1554 {
1555 struct process_old_arg *poa = arg;
1556 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1557
1558 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1559 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1560 if (poa->ds_prev && !poa->after_branch_point &&
1561 bp->blk_birth >
1562 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1563 poa->ds_prev->ds_phys->ds_unique_bytes +=
1564 bp_get_dsize_sync(dp->dp_spa, bp);
1565 }
1566 } else {
1567 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1568 poa->comp += BP_GET_PSIZE(bp);
1569 poa->uncomp += BP_GET_UCSIZE(bp);
1570 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1571 }
1572 return (0);
1573 }
1574
1575 static void
1576 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1577 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1578 {
1579 struct process_old_arg poa = { 0 };
1580 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1581 objset_t *mos = dp->dp_meta_objset;
1582
1583 ASSERT(ds->ds_deadlist.dl_oldfmt);
1584 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1585
1586 poa.ds = ds;
1587 poa.ds_prev = ds_prev;
1588 poa.after_branch_point = after_branch_point;
1589 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1590 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1591 process_old_cb, &poa, tx));
1592 VERIFY3U(zio_wait(poa.pio), ==, 0);
1593 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1594
1595 /* change snapused */
1596 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1597 -poa.used, -poa.comp, -poa.uncomp, tx);
1598
1599 /* swap next's deadlist to our deadlist */
1600 dsl_deadlist_close(&ds->ds_deadlist);
1601 dsl_deadlist_close(&ds_next->ds_deadlist);
1602 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1603 ds->ds_phys->ds_deadlist_obj);
1604 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1605 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1606 ds_next->ds_phys->ds_deadlist_obj);
1607 }
1608
1609 void
1610 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1611 {
1612 struct dsl_ds_destroyarg *dsda = arg1;
1613 dsl_dataset_t *ds = dsda->ds;
1614 int err;
1615 int after_branch_point = FALSE;
1616 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1617 objset_t *mos = dp->dp_meta_objset;
1618 dsl_dataset_t *ds_prev = NULL;
1619 boolean_t wont_destroy;
1620 uint64_t obj;
1621
1622 wont_destroy = (dsda->defer &&
1623 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1624
1625 ASSERT(ds->ds_owner || wont_destroy);
1626 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1627 ASSERT(ds->ds_prev == NULL ||
1628 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1629 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1630
1631 if (wont_destroy) {
1632 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1633 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1634 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1635 return;
1636 }
1637
1638 /* signal any waiters that this dataset is going away */
1639 mutex_enter(&ds->ds_lock);
1640 ds->ds_owner = dsl_reaper;
1641 cv_broadcast(&ds->ds_exclusive_cv);
1642 mutex_exit(&ds->ds_lock);
1643
1644 /* Remove our reservation */
1645 if (ds->ds_reserved != 0) {
1646 dsl_prop_setarg_t psa;
1647 uint64_t value = 0;
1648
1649 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1650 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1651 &value);
1652 psa.psa_effective_value = 0; /* predict default value */
1653
1654 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1655 ASSERT3U(ds->ds_reserved, ==, 0);
1656 }
1657
1658 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1659
1660 dsl_scan_ds_destroyed(ds, tx);
1661
1662 obj = ds->ds_object;
1663
1664 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1665 if (ds->ds_prev) {
1666 ds_prev = ds->ds_prev;
1667 } else {
1668 VERIFY(0 == dsl_dataset_hold_obj(dp,
1669 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1670 }
1671 after_branch_point =
1672 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1673
1674 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1675 if (after_branch_point &&
1676 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1677 remove_from_next_clones(ds_prev, obj, tx);
1678 if (ds->ds_phys->ds_next_snap_obj != 0) {
1679 VERIFY(0 == zap_add_int(mos,
1680 ds_prev->ds_phys->ds_next_clones_obj,
1681 ds->ds_phys->ds_next_snap_obj, tx));
1682 }
1683 }
1684 if (after_branch_point &&
1685 ds->ds_phys->ds_next_snap_obj == 0) {
1686 /* This clone is toast. */
1687 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1688 ds_prev->ds_phys->ds_num_children--;
1689
1690 /*
1691 * If the clone's origin has no other clones, no
1692 * user holds, and has been marked for deferred
1693 * deletion, then we should have done the necessary
1694 * destroy setup for it.
1695 */
1696 if (ds_prev->ds_phys->ds_num_children == 1 &&
1697 ds_prev->ds_userrefs == 0 &&
1698 DS_IS_DEFER_DESTROY(ds_prev)) {
1699 ASSERT3P(dsda->rm_origin, !=, NULL);
1700 } else {
1701 ASSERT3P(dsda->rm_origin, ==, NULL);
1702 }
1703 } else if (!after_branch_point) {
1704 ds_prev->ds_phys->ds_next_snap_obj =
1705 ds->ds_phys->ds_next_snap_obj;
1706 }
1707 }
1708
1709 if (dsl_dataset_is_snapshot(ds)) {
1710 dsl_dataset_t *ds_next;
1711 uint64_t old_unique;
1712 uint64_t used = 0, comp = 0, uncomp = 0;
1713
1714 VERIFY(0 == dsl_dataset_hold_obj(dp,
1715 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1716 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1717
1718 old_unique = ds_next->ds_phys->ds_unique_bytes;
1719
1720 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1721 ds_next->ds_phys->ds_prev_snap_obj =
1722 ds->ds_phys->ds_prev_snap_obj;
1723 ds_next->ds_phys->ds_prev_snap_txg =
1724 ds->ds_phys->ds_prev_snap_txg;
1725 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1726 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1727
1728
1729 if (ds_next->ds_deadlist.dl_oldfmt) {
1730 process_old_deadlist(ds, ds_prev, ds_next,
1731 after_branch_point, tx);
1732 } else {
1733 /* Adjust prev's unique space. */
1734 if (ds_prev && !after_branch_point) {
1735 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1736 ds_prev->ds_phys->ds_prev_snap_txg,
1737 ds->ds_phys->ds_prev_snap_txg,
1738 &used, &comp, &uncomp);
1739 ds_prev->ds_phys->ds_unique_bytes += used;
1740 }
1741
1742 /* Adjust snapused. */
1743 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1744 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1745 &used, &comp, &uncomp);
1746 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1747 -used, -comp, -uncomp, tx);
1748
1749 /* Move blocks to be freed to pool's free list. */
1750 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1751 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1752 tx);
1753 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1754 DD_USED_HEAD, used, comp, uncomp, tx);
1755 dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1756
1757 /* Merge our deadlist into next's and free it. */
1758 dsl_deadlist_merge(&ds_next->ds_deadlist,
1759 ds->ds_phys->ds_deadlist_obj, tx);
1760 }
1761 dsl_deadlist_close(&ds->ds_deadlist);
1762 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1763
1764 /* Collapse range in clone heads */
1765 dsl_dataset_remove_clones_key(ds,
1766 ds->ds_phys->ds_creation_txg, tx);
1767
1768 if (dsl_dataset_is_snapshot(ds_next)) {
1769 dsl_dataset_t *ds_nextnext;
1770
1771 /*
1772 * Update next's unique to include blocks which
1773 * were previously shared by only this snapshot
1774 * and it. Those blocks will be born after the
1775 * prev snap and before this snap, and will have
1776 * died after the next snap and before the one
1777 * after that (ie. be on the snap after next's
1778 * deadlist).
1779 */
1780 VERIFY(0 == dsl_dataset_hold_obj(dp,
1781 ds_next->ds_phys->ds_next_snap_obj,
1782 FTAG, &ds_nextnext));
1783 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1784 ds->ds_phys->ds_prev_snap_txg,
1785 ds->ds_phys->ds_creation_txg,
1786 &used, &comp, &uncomp);
1787 ds_next->ds_phys->ds_unique_bytes += used;
1788 dsl_dataset_rele(ds_nextnext, FTAG);
1789 ASSERT3P(ds_next->ds_prev, ==, NULL);
1790
1791 /* Collapse range in this head. */
1792 dsl_dataset_t *hds;
1793 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1794 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1795 FTAG, &hds));
1796 dsl_deadlist_remove_key(&hds->ds_deadlist,
1797 ds->ds_phys->ds_creation_txg, tx);
1798 dsl_dataset_rele(hds, FTAG);
1799
1800 } else {
1801 ASSERT3P(ds_next->ds_prev, ==, ds);
1802 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1803 ds_next->ds_prev = NULL;
1804 if (ds_prev) {
1805 VERIFY(0 == dsl_dataset_get_ref(dp,
1806 ds->ds_phys->ds_prev_snap_obj,
1807 ds_next, &ds_next->ds_prev));
1808 }
1809
1810 dsl_dataset_recalc_head_uniq(ds_next);
1811
1812 /*
1813 * Reduce the amount of our unconsmed refreservation
1814 * being charged to our parent by the amount of
1815 * new unique data we have gained.
1816 */
1817 if (old_unique < ds_next->ds_reserved) {
1818 int64_t mrsdelta;
1819 uint64_t new_unique =
1820 ds_next->ds_phys->ds_unique_bytes;
1821
1822 ASSERT(old_unique <= new_unique);
1823 mrsdelta = MIN(new_unique - old_unique,
1824 ds_next->ds_reserved - old_unique);
1825 dsl_dir_diduse_space(ds->ds_dir,
1826 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1827 }
1828 }
1829 dsl_dataset_rele(ds_next, FTAG);
1830 } else {
1831 /*
1832 * There's no next snapshot, so this is a head dataset.
1833 * Destroy the deadlist. Unless it's a clone, the
1834 * deadlist should be empty. (If it's a clone, it's
1835 * safe to ignore the deadlist contents.)
1836 */
1837 struct killarg ka;
1838
1839 dsl_deadlist_close(&ds->ds_deadlist);
1840 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1841 ds->ds_phys->ds_deadlist_obj = 0;
1842
1843 /*
1844 * Free everything that we point to (that's born after
1845 * the previous snapshot, if we are a clone)
1846 *
1847 * NB: this should be very quick, because we already
1848 * freed all the objects in open context.
1849 */
1850 ka.ds = ds;
1851 ka.tx = tx;
1852 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1853 TRAVERSE_POST, kill_blkptr, &ka);
1854 ASSERT3U(err, ==, 0);
1855 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1856 ds->ds_phys->ds_unique_bytes == 0);
1857
1858 if (ds->ds_prev != NULL) {
1859 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1860 VERIFY3U(0, ==, zap_remove_int(mos,
1861 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1862 ds->ds_object, tx));
1863 }
1864 dsl_dataset_rele(ds->ds_prev, ds);
1865 ds->ds_prev = ds_prev = NULL;
1866 }
1867 }
1868
1869 /*
1870 * This must be done after the dsl_traverse(), because it will
1871 * re-open the objset.
1872 */
1873 if (ds->ds_objset) {
1874 dmu_objset_evict(ds->ds_objset);
1875 ds->ds_objset = NULL;
1876 }
1877
1878 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1879 /* Erase the link in the dir */
1880 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1881 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1882 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1883 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1884 ASSERT(err == 0);
1885 } else {
1886 /* remove from snapshot namespace */
1887 dsl_dataset_t *ds_head;
1888 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1889 VERIFY(0 == dsl_dataset_hold_obj(dp,
1890 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1891 VERIFY(0 == dsl_dataset_get_snapname(ds));
1892 #ifdef ZFS_DEBUG
1893 {
1894 uint64_t val;
1895
1896 err = dsl_dataset_snap_lookup(ds_head,
1897 ds->ds_snapname, &val);
1898 ASSERT3U(err, ==, 0);
1899 ASSERT3U(val, ==, obj);
1900 }
1901 #endif
1902 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1903 ASSERT(err == 0);
1904 dsl_dataset_rele(ds_head, FTAG);
1905 }
1906
1907 if (ds_prev && ds->ds_prev != ds_prev)
1908 dsl_dataset_rele(ds_prev, FTAG);
1909
1910 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1911 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1912 "dataset = %llu", ds->ds_object);
1913
1914 if (ds->ds_phys->ds_next_clones_obj != 0) {
1915 uint64_t count;
1916 ASSERT(0 == zap_count(mos,
1917 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1918 VERIFY(0 == dmu_object_free(mos,
1919 ds->ds_phys->ds_next_clones_obj, tx));
1920 }
1921 if (ds->ds_phys->ds_props_obj != 0)
1922 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1923 if (ds->ds_phys->ds_userrefs_obj != 0)
1924 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1925 dsl_dir_close(ds->ds_dir, ds);
1926 ds->ds_dir = NULL;
1927 dsl_dataset_drain_refs(ds, tag);
1928 VERIFY(0 == dmu_object_free(mos, obj, tx));
1929
1930 if (dsda->rm_origin) {
1931 /*
1932 * Remove the origin of the clone we just destroyed.
1933 */
1934 struct dsl_ds_destroyarg ndsda = {0};
1935
1936 ndsda.ds = dsda->rm_origin;
1937 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1938 }
1939 }
1940
1941 static int
1942 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1943 {
1944 uint64_t asize;
1945
1946 if (!dmu_tx_is_syncing(tx))
1947 return (0);
1948
1949 /*
1950 * If there's an fs-only reservation, any blocks that might become
1951 * owned by the snapshot dataset must be accommodated by space
1952 * outside of the reservation.
1953 */
1954 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1955 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
1956 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1957 return (ENOSPC);
1958
1959 /*
1960 * Propogate any reserved space for this snapshot to other
1961 * snapshot checks in this sync group.
1962 */
1963 if (asize > 0)
1964 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1965
1966 return (0);
1967 }
1968
1969 int
1970 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1971 {
1972 dsl_dataset_t *ds = arg1;
1973 const char *snapname = arg2;
1974 int err;
1975 uint64_t value;
1976
1977 /*
1978 * We don't allow multiple snapshots of the same txg. If there
1979 * is already one, try again.
1980 */
1981 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1982 return (EAGAIN);
1983
1984 /*
1985 * Check for conflicting name snapshot name.
1986 */
1987 err = dsl_dataset_snap_lookup(ds, snapname, &value);
1988 if (err == 0)
1989 return (EEXIST);
1990 if (err != ENOENT)
1991 return (err);
1992
1993 /*
1994 * Check that the dataset's name is not too long. Name consists
1995 * of the dataset's length + 1 for the @-sign + snapshot name's length
1996 */
1997 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
1998 return (ENAMETOOLONG);
1999
2000 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2001 if (err)
2002 return (err);
2003
2004 ds->ds_trysnap_txg = tx->tx_txg;
2005 return (0);
2006 }
2007
2008 void
2009 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2010 {
2011 dsl_dataset_t *ds = arg1;
2012 const char *snapname = arg2;
2013 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2014 dmu_buf_t *dbuf;
2015 dsl_dataset_phys_t *dsphys;
2016 uint64_t dsobj, crtxg;
2017 objset_t *mos = dp->dp_meta_objset;
2018 int err;
2019
2020 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2021
2022 /*
2023 * The origin's ds_creation_txg has to be < TXG_INITIAL
2024 */
2025 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2026 crtxg = 1;
2027 else
2028 crtxg = tx->tx_txg;
2029
2030 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2031 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2032 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2033 dmu_buf_will_dirty(dbuf, tx);
2034 dsphys = dbuf->db_data;
2035 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2036 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2037 dsphys->ds_fsid_guid = unique_create();
2038 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2039 sizeof (dsphys->ds_guid));
2040 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2041 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2042 dsphys->ds_next_snap_obj = ds->ds_object;
2043 dsphys->ds_num_children = 1;
2044 dsphys->ds_creation_time = gethrestime_sec();
2045 dsphys->ds_creation_txg = crtxg;
2046 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2047 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2048 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2049 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2050 dsphys->ds_flags = ds->ds_phys->ds_flags;
2051 dsphys->ds_bp = ds->ds_phys->ds_bp;
2052 dmu_buf_rele(dbuf, FTAG);
2053
2054 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2055 if (ds->ds_prev) {
2056 uint64_t next_clones_obj =
2057 ds->ds_prev->ds_phys->ds_next_clones_obj;
2058 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2059 ds->ds_object ||
2060 ds->ds_prev->ds_phys->ds_num_children > 1);
2061 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2062 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2063 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2064 ds->ds_prev->ds_phys->ds_creation_txg);
2065 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2066 } else if (next_clones_obj != 0) {
2067 remove_from_next_clones(ds->ds_prev,
2068 dsphys->ds_next_snap_obj, tx);
2069 VERIFY3U(0, ==, zap_add_int(mos,
2070 next_clones_obj, dsobj, tx));
2071 }
2072 }
2073
2074 /*
2075 * If we have a reference-reservation on this dataset, we will
2076 * need to increase the amount of refreservation being charged
2077 * since our unique space is going to zero.
2078 */
2079 if (ds->ds_reserved) {
2080 int64_t delta;
2081 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2082 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2083 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2084 delta, 0, 0, tx);
2085 }
2086
2087 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2088 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2089 ds->ds_dir->dd_myname, snapname, dsobj,
2090 ds->ds_phys->ds_prev_snap_txg);
2091 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2092 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2093 dsl_deadlist_close(&ds->ds_deadlist);
2094 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2095 dsl_deadlist_add_key(&ds->ds_deadlist,
2096 ds->ds_phys->ds_prev_snap_txg, tx);
2097
2098 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2099 ds->ds_phys->ds_prev_snap_obj = dsobj;
2100 ds->ds_phys->ds_prev_snap_txg = crtxg;
2101 ds->ds_phys->ds_unique_bytes = 0;
2102 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2103 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2104
2105 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2106 snapname, 8, 1, &dsobj, tx);
2107 ASSERT(err == 0);
2108
2109 if (ds->ds_prev)
2110 dsl_dataset_drop_ref(ds->ds_prev, ds);
2111 VERIFY(0 == dsl_dataset_get_ref(dp,
2112 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2113
2114 dsl_scan_ds_snapshotted(ds, tx);
2115
2116 dsl_dir_snap_cmtime_update(ds->ds_dir);
2117
2118 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2119 "dataset = %llu", dsobj);
2120 }
2121
2122 void
2123 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2124 {
2125 ASSERT(dmu_tx_is_syncing(tx));
2126 ASSERT(ds->ds_objset != NULL);
2127 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2128
2129 /*
2130 * in case we had to change ds_fsid_guid when we opened it,
2131 * sync it out now.
2132 */
2133 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2134 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2135
2136 dsl_dir_dirty(ds->ds_dir, tx);
2137 dmu_objset_sync(ds->ds_objset, zio, tx);
2138 }
2139
2140 static void
2141 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2142 {
2143 uint64_t count = 0;
2144 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2145 zap_cursor_t zc;
2146 zap_attribute_t za;
2147 nvlist_t *propval;
2148 nvlist_t *val;
2149
2150 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2151 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2152 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2153
2154 /*
2155 * There may me missing entries in ds_next_clones_obj
2156 * due to a bug in a previous version of the code.
2157 * Only trust it if it has the right number of entries.
2158 */
2159 if (ds->ds_phys->ds_next_clones_obj != 0) {
2160 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2161 &count));
2162 }
2163 if (count != ds->ds_phys->ds_num_children - 1) {
2164 goto fail;
2165 }
2166 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2167 zap_cursor_retrieve(&zc, &za) == 0;
2168 zap_cursor_advance(&zc)) {
2169 dsl_dataset_t *clone;
2170 char buf[ZFS_MAXNAMELEN];
2171 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2172 za.za_first_integer, FTAG, &clone) != 0) {
2173 goto fail;
2174 }
2175 dsl_dir_name(clone->ds_dir, buf);
2176 VERIFY(nvlist_add_boolean(val, buf) == 0);
2177 dsl_dataset_rele(clone, FTAG);
2178 }
2179 zap_cursor_fini(&zc);
2180 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2181 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2182 propval) == 0);
2183 fail:
2184 nvlist_free(val);
2185 nvlist_free(propval);
2186 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2187 }
2188
2189 void
2190 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2191 {
2192 uint64_t refd, avail, uobjs, aobjs, ratio;
2193
2194 dsl_dir_stats(ds->ds_dir, nv);
2195
2196 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2197 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2198 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2199
2200 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2201 ds->ds_phys->ds_creation_time);
2202 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2203 ds->ds_phys->ds_creation_txg);
2204 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2205 ds->ds_quota);
2206 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2207 ds->ds_reserved);
2208 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2209 ds->ds_phys->ds_guid);
2210 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2211 ds->ds_phys->ds_unique_bytes);
2212 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2213 ds->ds_object);
2214 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2215 ds->ds_userrefs);
2216 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2217 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2218
2219 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2220 uint64_t written, comp, uncomp;
2221 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2222 dsl_dataset_t *prev;
2223
2224 rw_enter(&dp->dp_config_rwlock, RW_READER);
2225 int err = dsl_dataset_hold_obj(dp,
2226 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2227 rw_exit(&dp->dp_config_rwlock);
2228 if (err == 0) {
2229 err = dsl_dataset_space_written(prev, ds, &written,
2230 &comp, &uncomp);
2231 dsl_dataset_rele(prev, FTAG);
2232 if (err == 0) {
2233 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2234 written);
2235 }
2236 }
2237 }
2238
2239 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2240 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2241 ds->ds_phys->ds_compressed_bytes);
2242 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2243
2244 if (ds->ds_phys->ds_next_snap_obj) {
2245 /*
2246 * This is a snapshot; override the dd's space used with
2247 * our unique space and compression ratio.
2248 */
2249 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2250 ds->ds_phys->ds_unique_bytes);
2251 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2252
2253 get_clones_stat(ds, nv);
2254 }
2255 }
2256
2257 void
2258 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2259 {
2260 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2261 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2262 stat->dds_guid = ds->ds_phys->ds_guid;
2263 if (ds->ds_phys->ds_next_snap_obj) {
2264 stat->dds_is_snapshot = B_TRUE;
2265 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2266 } else {
2267 stat->dds_is_snapshot = B_FALSE;
2268 stat->dds_num_clones = 0;
2269 }
2270
2271 /* clone origin is really a dsl_dir thing... */
2272 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2273 if (dsl_dir_is_clone(ds->ds_dir)) {
2274 dsl_dataset_t *ods;
2275
2276 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2277 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2278 dsl_dataset_name(ods, stat->dds_origin);
2279 dsl_dataset_drop_ref(ods, FTAG);
2280 } else {
2281 stat->dds_origin[0] = '\0';
2282 }
2283 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2284 }
2285
2286 uint64_t
2287 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2288 {
2289 return (ds->ds_fsid_guid);
2290 }
2291
2292 void
2293 dsl_dataset_space(dsl_dataset_t *ds,
2294 uint64_t *refdbytesp, uint64_t *availbytesp,
2295 uint64_t *usedobjsp, uint64_t *availobjsp)
2296 {
2297 *refdbytesp = ds->ds_phys->ds_used_bytes;
2298 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2299 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2300 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2301 if (ds->ds_quota != 0) {
2302 /*
2303 * Adjust available bytes according to refquota
2304 */
2305 if (*refdbytesp < ds->ds_quota)
2306 *availbytesp = MIN(*availbytesp,
2307 ds->ds_quota - *refdbytesp);
2308 else
2309 *availbytesp = 0;
2310 }
2311 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2312 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2313 }
2314
2315 boolean_t
2316 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2317 {
2318 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2319
2320 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2321 dsl_pool_sync_context(dp));
2322 if (ds->ds_prev == NULL)
2323 return (B_FALSE);
2324 if (ds->ds_phys->ds_bp.blk_birth >
2325 ds->ds_prev->ds_phys->ds_creation_txg) {
2326 objset_t *os, *os_prev;
2327 /*
2328 * It may be that only the ZIL differs, because it was
2329 * reset in the head. Don't count that as being
2330 * modified.
2331 */
2332 if (dmu_objset_from_ds(ds, &os) != 0)
2333 return (B_TRUE);
2334 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2335 return (B_TRUE);
2336 return (bcmp(&os->os_phys->os_meta_dnode,
2337 &os_prev->os_phys->os_meta_dnode,
2338 sizeof (os->os_phys->os_meta_dnode)) != 0);
2339 }
2340 return (B_FALSE);
2341 }
2342
2343 /* ARGSUSED */
2344 static int
2345 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2346 {
2347 dsl_dataset_t *ds = arg1;
2348 char *newsnapname = arg2;
2349 dsl_dir_t *dd = ds->ds_dir;
2350 dsl_dataset_t *hds;
2351 uint64_t val;
2352 int err;
2353
2354 err = dsl_dataset_hold_obj(dd->dd_pool,
2355 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2356 if (err)
2357 return (err);
2358
2359 /* new name better not be in use */
2360 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2361 dsl_dataset_rele(hds, FTAG);
2362
2363 if (err == 0)
2364 err = EEXIST;
2365 else if (err == ENOENT)
2366 err = 0;
2367
2368 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2369 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2370 err = ENAMETOOLONG;
2371
2372 return (err);
2373 }
2374
2375 static void
2376 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2377 {
2378 dsl_dataset_t *ds = arg1;
2379 const char *newsnapname = arg2;
2380 dsl_dir_t *dd = ds->ds_dir;
2381 objset_t *mos = dd->dd_pool->dp_meta_objset;
2382 dsl_dataset_t *hds;
2383 int err;
2384
2385 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2386
2387 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2388 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2389
2390 VERIFY(0 == dsl_dataset_get_snapname(ds));
2391 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2392 ASSERT3U(err, ==, 0);
2393 mutex_enter(&ds->ds_lock);
2394 (void) strcpy(ds->ds_snapname, newsnapname);
2395 mutex_exit(&ds->ds_lock);
2396 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2397 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2398 ASSERT3U(err, ==, 0);
2399
2400 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2401 "dataset = %llu", ds->ds_object);
2402 dsl_dataset_rele(hds, FTAG);
2403 }
2404
2405 struct renamesnaparg {
2406 dsl_sync_task_group_t *dstg;
2407 char failed[MAXPATHLEN];
2408 char *oldsnap;
2409 char *newsnap;
2410 };
2411
2412 static int
2413 dsl_snapshot_rename_one(const char *name, void *arg)
2414 {
2415 struct renamesnaparg *ra = arg;
2416 dsl_dataset_t *ds = NULL;
2417 char *snapname;
2418 int err;
2419
2420 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2421 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2422
2423 /*
2424 * For recursive snapshot renames the parent won't be changing
2425 * so we just pass name for both the to/from argument.
2426 */
2427 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2428 if (err != 0) {
2429 strfree(snapname);
2430 return (err == ENOENT ? 0 : err);
2431 }
2432
2433 #ifdef _KERNEL
2434 /*
2435 * For all filesystems undergoing rename, we'll need to unmount it.
2436 */
2437 (void) zfs_unmount_snap(snapname, NULL);
2438 #endif
2439 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2440 strfree(snapname);
2441 if (err != 0)
2442 return (err == ENOENT ? 0 : err);
2443
2444 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2445 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2446
2447 return (0);
2448 }
2449
2450 static int
2451 dsl_recursive_rename(char *oldname, const char *newname)
2452 {
2453 int err;
2454 struct renamesnaparg *ra;
2455 dsl_sync_task_t *dst;
2456 spa_t *spa;
2457 char *cp, *fsname = spa_strdup(oldname);
2458 int len = strlen(oldname) + 1;
2459
2460 /* truncate the snapshot name to get the fsname */
2461 cp = strchr(fsname, '@');
2462 *cp = '\0';
2463
2464 err = spa_open(fsname, &spa, FTAG);
2465 if (err) {
2466 kmem_free(fsname, len);
2467 return (err);
2468 }
2469 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2470 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2471
2472 ra->oldsnap = strchr(oldname, '@') + 1;
2473 ra->newsnap = strchr(newname, '@') + 1;
2474 *ra->failed = '\0';
2475
2476 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2477 DS_FIND_CHILDREN);
2478 kmem_free(fsname, len);
2479
2480 if (err == 0) {
2481 err = dsl_sync_task_group_wait(ra->dstg);
2482 }
2483
2484 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2485 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2486 dsl_dataset_t *ds = dst->dst_arg1;
2487 if (dst->dst_err) {
2488 dsl_dir_name(ds->ds_dir, ra->failed);
2489 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2490 (void) strlcat(ra->failed, ra->newsnap,
2491 sizeof (ra->failed));
2492 }
2493 dsl_dataset_rele(ds, ra->dstg);
2494 }
2495
2496 if (err)
2497 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2498
2499 dsl_sync_task_group_destroy(ra->dstg);
2500 kmem_free(ra, sizeof (struct renamesnaparg));
2501 spa_close(spa, FTAG);
2502 return (err);
2503 }
2504
2505 static int
2506 dsl_valid_rename(const char *oldname, void *arg)
2507 {
2508 int delta = *(int *)arg;
2509
2510 if (strlen(oldname) + delta >= MAXNAMELEN)
2511 return (ENAMETOOLONG);
2512
2513 return (0);
2514 }
2515
2516 #pragma weak dmu_objset_rename = dsl_dataset_rename
2517 int
2518 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2519 {
2520 dsl_dir_t *dd;
2521 dsl_dataset_t *ds;
2522 const char *tail;
2523 int err;
2524
2525 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2526 if (err)
2527 return (err);
2528
2529 if (tail == NULL) {
2530 int delta = strlen(newname) - strlen(oldname);
2531
2532 /* if we're growing, validate child name lengths */
2533 if (delta > 0)
2534 err = dmu_objset_find(oldname, dsl_valid_rename,
2535 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2536
2537 if (err == 0)
2538 err = dsl_dir_rename(dd, newname);
2539 dsl_dir_close(dd, FTAG);
2540 return (err);
2541 }
2542
2543 if (tail[0] != '@') {
2544 /* the name ended in a nonexistent component */
2545 dsl_dir_close(dd, FTAG);
2546 return (ENOENT);
2547 }
2548
2549 dsl_dir_close(dd, FTAG);
2550
2551 /* new name must be snapshot in same filesystem */
2552 tail = strchr(newname, '@');
2553 if (tail == NULL)
2554 return (EINVAL);
2555 tail++;
2556 if (strncmp(oldname, newname, tail - newname) != 0)
2557 return (EXDEV);
2558
2559 if (recursive) {
2560 err = dsl_recursive_rename(oldname, newname);
2561 } else {
2562 err = dsl_dataset_hold(oldname, FTAG, &ds);
2563 if (err)
2564 return (err);
2565
2566 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2567 dsl_dataset_snapshot_rename_check,
2568 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2569
2570 dsl_dataset_rele(ds, FTAG);
2571 }
2572
2573 return (err);
2574 }
2575
2576 struct promotenode {
2577 list_node_t link;
2578 dsl_dataset_t *ds;
2579 };
2580
2581 struct promotearg {
2582 list_t shared_snaps, origin_snaps, clone_snaps;
2583 dsl_dataset_t *origin_origin;
2584 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2585 char *err_ds;
2586 };
2587
2588 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2589 static boolean_t snaplist_unstable(list_t *l);
2590
2591 static int
2592 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2593 {
2594 dsl_dataset_t *hds = arg1;
2595 struct promotearg *pa = arg2;
2596 struct promotenode *snap = list_head(&pa->shared_snaps);
2597 dsl_dataset_t *origin_ds = snap->ds;
2598 int err;
2599 uint64_t unused;
2600
2601 /* Check that it is a real clone */
2602 if (!dsl_dir_is_clone(hds->ds_dir))
2603 return (EINVAL);
2604
2605 /* Since this is so expensive, don't do the preliminary check */
2606 if (!dmu_tx_is_syncing(tx))
2607 return (0);
2608
2609 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2610 return (EXDEV);
2611
2612 /* compute origin's new unique space */
2613 snap = list_tail(&pa->clone_snaps);
2614 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2615 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2616 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2617 &pa->unique, &unused, &unused);
2618
2619 /*
2620 * Walk the snapshots that we are moving
2621 *
2622 * Compute space to transfer. Consider the incremental changes
2623 * to used for each snapshot:
2624 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2625 * So each snapshot gave birth to:
2626 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2627 * So a sequence would look like:
2628 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2629 * Which simplifies to:
2630 * uN + kN + kN-1 + ... + k1 + k0
2631 * Note however, if we stop before we reach the ORIGIN we get:
2632 * uN + kN + kN-1 + ... + kM - uM-1
2633 */
2634 pa->used = origin_ds->ds_phys->ds_used_bytes;
2635 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2636 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2637 for (snap = list_head(&pa->shared_snaps); snap;
2638 snap = list_next(&pa->shared_snaps, snap)) {
2639 uint64_t val, dlused, dlcomp, dluncomp;
2640 dsl_dataset_t *ds = snap->ds;
2641
2642 /* Check that the snapshot name does not conflict */
2643 VERIFY(0 == dsl_dataset_get_snapname(ds));
2644 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2645 if (err == 0) {
2646 err = EEXIST;
2647 goto out;
2648 }
2649 if (err != ENOENT)
2650 goto out;
2651
2652 /* The very first snapshot does not have a deadlist */
2653 if (ds->ds_phys->ds_prev_snap_obj == 0)
2654 continue;
2655
2656 dsl_deadlist_space(&ds->ds_deadlist,
2657 &dlused, &dlcomp, &dluncomp);
2658 pa->used += dlused;
2659 pa->comp += dlcomp;
2660 pa->uncomp += dluncomp;
2661 }
2662
2663 /*
2664 * If we are a clone of a clone then we never reached ORIGIN,
2665 * so we need to subtract out the clone origin's used space.
2666 */
2667 if (pa->origin_origin) {
2668 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2669 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2670 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2671 }
2672
2673 /* Check that there is enough space here */
2674 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2675 pa->used);
2676 if (err)
2677 return (err);
2678
2679 /*
2680 * Compute the amounts of space that will be used by snapshots
2681 * after the promotion (for both origin and clone). For each,
2682 * it is the amount of space that will be on all of their
2683 * deadlists (that was not born before their new origin).
2684 */
2685 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2686 uint64_t space;
2687
2688 /*
2689 * Note, typically this will not be a clone of a clone,
2690 * so dd_origin_txg will be < TXG_INITIAL, so
2691 * these snaplist_space() -> dsl_deadlist_space_range()
2692 * calls will be fast because they do not have to
2693 * iterate over all bps.
2694 */
2695 snap = list_head(&pa->origin_snaps);
2696 err = snaplist_space(&pa->shared_snaps,
2697 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2698 if (err)
2699 return (err);
2700
2701 err = snaplist_space(&pa->clone_snaps,
2702 snap->ds->ds_dir->dd_origin_txg, &space);
2703 if (err)
2704 return (err);
2705 pa->cloneusedsnap += space;
2706 }
2707 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2708 err = snaplist_space(&pa->origin_snaps,
2709 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2710 if (err)
2711 return (err);
2712 }
2713
2714 return (0);
2715 out:
2716 pa->err_ds = snap->ds->ds_snapname;
2717 return (err);
2718 }
2719
2720 static void
2721 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2722 {
2723 dsl_dataset_t *hds = arg1;
2724 struct promotearg *pa = arg2;
2725 struct promotenode *snap = list_head(&pa->shared_snaps);
2726 dsl_dataset_t *origin_ds = snap->ds;
2727 dsl_dataset_t *origin_head;
2728 dsl_dir_t *dd = hds->ds_dir;
2729 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2730 dsl_dir_t *odd = NULL;
2731 uint64_t oldnext_obj;
2732 int64_t delta;
2733
2734 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2735
2736 snap = list_head(&pa->origin_snaps);
2737 origin_head = snap->ds;
2738
2739 /*
2740 * We need to explicitly open odd, since origin_ds's dd will be
2741 * changing.
2742 */
2743 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2744 NULL, FTAG, &odd));
2745
2746 /* change origin's next snap */
2747 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2748 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2749 snap = list_tail(&pa->clone_snaps);
2750 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2751 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2752
2753 /* change the origin's next clone */
2754 if (origin_ds->ds_phys->ds_next_clones_obj) {
2755 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2756 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2757 origin_ds->ds_phys->ds_next_clones_obj,
2758 oldnext_obj, tx));
2759 }
2760
2761 /* change origin */
2762 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2763 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2764 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2765 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2766 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2767 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2768 origin_head->ds_dir->dd_origin_txg =
2769 origin_ds->ds_phys->ds_creation_txg;
2770
2771 /* change dd_clone entries */
2772 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2773 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2774 odd->dd_phys->dd_clones, hds->ds_object, tx));
2775 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2776 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2777 hds->ds_object, tx));
2778
2779 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2780 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2781 origin_head->ds_object, tx));
2782 if (dd->dd_phys->dd_clones == 0) {
2783 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2784 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2785 }
2786 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2787 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2788
2789 }
2790
2791 /* move snapshots to this dir */
2792 for (snap = list_head(&pa->shared_snaps); snap;
2793 snap = list_next(&pa->shared_snaps, snap)) {
2794 dsl_dataset_t *ds = snap->ds;
2795
2796 /* unregister props as dsl_dir is changing */
2797 if (ds->ds_objset) {
2798 dmu_objset_evict(ds->ds_objset);
2799 ds->ds_objset = NULL;
2800 }
2801 /* move snap name entry */
2802 VERIFY(0 == dsl_dataset_get_snapname(ds));
2803 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2804 ds->ds_snapname, tx));
2805 VERIFY(0 == zap_add(dp->dp_meta_objset,
2806 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2807 8, 1, &ds->ds_object, tx));
2808
2809 /* change containing dsl_dir */
2810 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2811 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2812 ds->ds_phys->ds_dir_obj = dd->dd_object;
2813 ASSERT3P(ds->ds_dir, ==, odd);
2814 dsl_dir_close(ds->ds_dir, ds);
2815 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2816 NULL, ds, &ds->ds_dir));
2817
2818 /* move any clone references */
2819 if (ds->ds_phys->ds_next_clones_obj &&
2820 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2821 zap_cursor_t zc;
2822 zap_attribute_t za;
2823
2824 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2825 ds->ds_phys->ds_next_clones_obj);
2826 zap_cursor_retrieve(&zc, &za) == 0;
2827 zap_cursor_advance(&zc)) {
2828 dsl_dataset_t *cnds;
2829 uint64_t o;
2830
2831 if (za.za_first_integer == oldnext_obj) {
2832 /*
2833 * We've already moved the
2834 * origin's reference.
2835 */
2836 continue;
2837 }
2838
2839 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2840 za.za_first_integer, FTAG, &cnds));
2841 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2842
2843 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2844 odd->dd_phys->dd_clones, o, tx), ==, 0);
2845 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2846 dd->dd_phys->dd_clones, o, tx), ==, 0);
2847 dsl_dataset_rele(cnds, FTAG);
2848 }
2849 zap_cursor_fini(&zc);
2850 }
2851
2852 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2853 }
2854
2855 /*
2856 * Change space accounting.
2857 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2858 * both be valid, or both be 0 (resulting in delta == 0). This
2859 * is true for each of {clone,origin} independently.
2860 */
2861
2862 delta = pa->cloneusedsnap -
2863 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2864 ASSERT3S(delta, >=, 0);
2865 ASSERT3U(pa->used, >=, delta);
2866 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2867 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2868 pa->used - delta, pa->comp, pa->uncomp, tx);
2869
2870 delta = pa->originusedsnap -
2871 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2872 ASSERT3S(delta, <=, 0);
2873 ASSERT3U(pa->used, >=, -delta);
2874 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2875 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2876 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2877
2878 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2879
2880 /* log history record */
2881 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2882 "dataset = %llu", hds->ds_object);
2883
2884 dsl_dir_close(odd, FTAG);
2885 }
2886
2887 static char *snaplist_tag = "snaplist";
2888 /*
2889 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2890 * (exclusive) and last_obj (inclusive). The list will be in reverse
2891 * order (last_obj will be the list_head()). If first_obj == 0, do all
2892 * snapshots back to this dataset's origin.
2893 */
2894 static int
2895 snaplist_make(dsl_pool_t *dp, boolean_t own,
2896 uint64_t first_obj, uint64_t last_obj, list_t *l)
2897 {
2898 uint64_t obj = last_obj;
2899
2900 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2901
2902 list_create(l, sizeof (struct promotenode),
2903 offsetof(struct promotenode, link));
2904
2905 while (obj != first_obj) {
2906 dsl_dataset_t *ds;
2907 struct promotenode *snap;
2908 int err;
2909
2910 if (own) {
2911 err = dsl_dataset_own_obj(dp, obj,
2912 0, snaplist_tag, &ds);
2913 if (err == 0)
2914 dsl_dataset_make_exclusive(ds, snaplist_tag);
2915 } else {
2916 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2917 }
2918 if (err == ENOENT) {
2919 /* lost race with snapshot destroy */
2920 struct promotenode *last = list_tail(l);
2921 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2922 obj = last->ds->ds_phys->ds_prev_snap_obj;
2923 continue;
2924 } else if (err) {
2925 return (err);
2926 }
2927
2928 if (first_obj == 0)
2929 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2930
2931 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2932 snap->ds = ds;
2933 list_insert_tail(l, snap);
2934 obj = ds->ds_phys->ds_prev_snap_obj;
2935 }
2936
2937 return (0);
2938 }
2939
2940 static int
2941 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2942 {
2943 struct promotenode *snap;
2944
2945 *spacep = 0;
2946 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2947 uint64_t used, comp, uncomp;
2948 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2949 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2950 *spacep += used;
2951 }
2952 return (0);
2953 }
2954
2955 static void
2956 snaplist_destroy(list_t *l, boolean_t own)
2957 {
2958 struct promotenode *snap;
2959
2960 if (!l || !list_link_active(&l->list_head))
2961 return;
2962
2963 while ((snap = list_tail(l)) != NULL) {
2964 list_remove(l, snap);
2965 if (own)
2966 dsl_dataset_disown(snap->ds, snaplist_tag);
2967 else
2968 dsl_dataset_rele(snap->ds, snaplist_tag);
2969 kmem_free(snap, sizeof (struct promotenode));
2970 }
2971 list_destroy(l);
2972 }
2973
2974 /*
2975 * Promote a clone. Nomenclature note:
2976 * "clone" or "cds": the original clone which is being promoted
2977 * "origin" or "ods": the snapshot which is originally clone's origin
2978 * "origin head" or "ohds": the dataset which is the head
2979 * (filesystem/volume) for the origin
2980 * "origin origin": the origin of the origin's filesystem (typically
2981 * NULL, indicating that the clone is not a clone of a clone).
2982 */
2983 int
2984 dsl_dataset_promote(const char *name, char *conflsnap)
2985 {
2986 dsl_dataset_t *ds;
2987 dsl_dir_t *dd;
2988 dsl_pool_t *dp;
2989 dmu_object_info_t doi;
2990 struct promotearg pa = { 0 };
2991 struct promotenode *snap;
2992 int err;
2993
2994 err = dsl_dataset_hold(name, FTAG, &ds);
2995 if (err)
2996 return (err);
2997 dd = ds->ds_dir;
2998 dp = dd->dd_pool;
2999
3000 err = dmu_object_info(dp->dp_meta_objset,
3001 ds->ds_phys->ds_snapnames_zapobj, &doi);
3002 if (err) {
3003 dsl_dataset_rele(ds, FTAG);
3004 return (err);
3005 }
3006
3007 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3008 dsl_dataset_rele(ds, FTAG);
3009 return (EINVAL);
3010 }
3011
3012 /*
3013 * We are going to inherit all the snapshots taken before our
3014 * origin (i.e., our new origin will be our parent's origin).
3015 * Take ownership of them so that we can rename them into our
3016 * namespace.
3017 */
3018 rw_enter(&dp->dp_config_rwlock, RW_READER);
3019
3020 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3021 &pa.shared_snaps);
3022 if (err != 0)
3023 goto out;
3024
3025 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3026 if (err != 0)
3027 goto out;
3028
3029 snap = list_head(&pa.shared_snaps);
3030 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3031 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3032 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3033 if (err != 0)
3034 goto out;
3035
3036 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3037 err = dsl_dataset_hold_obj(dp,
3038 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3039 FTAG, &pa.origin_origin);
3040 if (err != 0)
3041 goto out;
3042 }
3043
3044 out:
3045 rw_exit(&dp->dp_config_rwlock);
3046
3047 /*
3048 * Add in 128x the snapnames zapobj size, since we will be moving
3049 * a bunch of snapnames to the promoted ds, and dirtying their
3050 * bonus buffers.
3051 */
3052 if (err == 0) {
3053 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3054 dsl_dataset_promote_sync, ds, &pa,
3055 2 + 2 * doi.doi_physical_blocks_512);
3056 if (err && pa.err_ds && conflsnap)
3057 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3058 }
3059
3060 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3061 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3062 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3063 if (pa.origin_origin)
3064 dsl_dataset_rele(pa.origin_origin, FTAG);
3065 dsl_dataset_rele(ds, FTAG);
3066 return (err);
3067 }
3068
3069 struct cloneswaparg {
3070 dsl_dataset_t *cds; /* clone dataset */
3071 dsl_dataset_t *ohds; /* origin's head dataset */
3072 boolean_t force;
3073 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3074 };
3075
3076 /* ARGSUSED */
3077 static int
3078 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3079 {
3080 struct cloneswaparg *csa = arg1;
3081
3082 /* they should both be heads */
3083 if (dsl_dataset_is_snapshot(csa->cds) ||
3084 dsl_dataset_is_snapshot(csa->ohds))
3085 return (EINVAL);
3086
3087 /* the branch point should be just before them */
3088 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3089 return (EINVAL);
3090
3091 /* cds should be the clone (unless they are unrelated) */
3092 if (csa->cds->ds_prev != NULL &&
3093 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3094 csa->ohds->ds_object !=
3095 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3096 return (EINVAL);
3097
3098 /* the clone should be a child of the origin */
3099 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3100 return (EINVAL);
3101
3102 /* ohds shouldn't be modified unless 'force' */
3103 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3104 return (ETXTBSY);
3105
3106 /* adjust amount of any unconsumed refreservation */
3107 csa->unused_refres_delta =
3108 (int64_t)MIN(csa->ohds->ds_reserved,
3109 csa->ohds->ds_phys->ds_unique_bytes) -
3110 (int64_t)MIN(csa->ohds->ds_reserved,
3111 csa->cds->ds_phys->ds_unique_bytes);
3112
3113 if (csa->unused_refres_delta > 0 &&
3114 csa->unused_refres_delta >
3115 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3116 return (ENOSPC);
3117
3118 if (csa->ohds->ds_quota != 0 &&
3119 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3120 return (EDQUOT);
3121
3122 return (0);
3123 }
3124
3125 /* ARGSUSED */
3126 static void
3127 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3128 {
3129 struct cloneswaparg *csa = arg1;
3130 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3131
3132 ASSERT(csa->cds->ds_reserved == 0);
3133 ASSERT(csa->ohds->ds_quota == 0 ||
3134 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3135
3136 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3137 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3138
3139 if (csa->cds->ds_objset != NULL) {
3140 dmu_objset_evict(csa->cds->ds_objset);
3141 csa->cds->ds_objset = NULL;
3142 }
3143
3144 if (csa->ohds->ds_objset != NULL) {
3145 dmu_objset_evict(csa->ohds->ds_objset);
3146 csa->ohds->ds_objset = NULL;
3147 }
3148
3149 /*
3150 * Reset origin's unique bytes, if it exists.
3151 */
3152 if (csa->cds->ds_prev) {
3153 dsl_dataset_t *origin = csa->cds->ds_prev;
3154 uint64_t comp, uncomp;
3155
3156 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3157 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3158 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3159 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3160 }
3161
3162 /* swap blkptrs */
3163 {
3164 blkptr_t tmp;
3165 tmp = csa->ohds->ds_phys->ds_bp;
3166 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3167 csa->cds->ds_phys->ds_bp = tmp;
3168 }
3169
3170 /* set dd_*_bytes */
3171 {
3172 int64_t dused, dcomp, duncomp;
3173 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3174 uint64_t odl_used, odl_comp, odl_uncomp;
3175
3176 ASSERT3U(csa->cds->ds_dir->dd_phys->
3177 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3178
3179 dsl_deadlist_space(&csa->cds->ds_deadlist,
3180 &cdl_used, &cdl_comp, &cdl_uncomp);
3181 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3182 &odl_used, &odl_comp, &odl_uncomp);
3183
3184 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3185 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3186 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3187 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3188 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3189 cdl_uncomp -
3190 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3191
3192 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3193 dused, dcomp, duncomp, tx);
3194 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3195 -dused, -dcomp, -duncomp, tx);
3196
3197 /*
3198 * The difference in the space used by snapshots is the
3199 * difference in snapshot space due to the head's
3200 * deadlist (since that's the only thing that's
3201 * changing that affects the snapused).
3202 */
3203 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3204 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3205 &cdl_used, &cdl_comp, &cdl_uncomp);
3206 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3207 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3208 &odl_used, &odl_comp, &odl_uncomp);
3209 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3210 DD_USED_HEAD, DD_USED_SNAP, tx);
3211 }
3212
3213 /* swap ds_*_bytes */
3214 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3215 csa->cds->ds_phys->ds_used_bytes);
3216 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3217 csa->cds->ds_phys->ds_compressed_bytes);
3218 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3219 csa->cds->ds_phys->ds_uncompressed_bytes);
3220 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3221 csa->cds->ds_phys->ds_unique_bytes);
3222
3223 /* apply any parent delta for change in unconsumed refreservation */
3224 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3225 csa->unused_refres_delta, 0, 0, tx);
3226
3227 /*
3228 * Swap deadlists.
3229 */
3230 dsl_deadlist_close(&csa->cds->ds_deadlist);
3231 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3232 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3233 csa->cds->ds_phys->ds_deadlist_obj);
3234 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3235 csa->cds->ds_phys->ds_deadlist_obj);
3236 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3237 csa->ohds->ds_phys->ds_deadlist_obj);
3238
3239 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3240 }
3241
3242 /*
3243 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3244 * recv" into an existing fs to swizzle the file system to the new
3245 * version, and by "zfs rollback". Can also be used to swap two
3246 * independent head datasets if neither has any snapshots.
3247 */
3248 int
3249 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3250 boolean_t force)
3251 {
3252 struct cloneswaparg csa;
3253 int error;
3254
3255 ASSERT(clone->ds_owner);
3256 ASSERT(origin_head->ds_owner);
3257 retry:
3258 /*
3259 * Need exclusive access for the swap. If we're swapping these
3260 * datasets back after an error, we already hold the locks.
3261 */
3262 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3263 rw_enter(&clone->ds_rwlock, RW_WRITER);
3264 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3265 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3266 rw_exit(&clone->ds_rwlock);
3267 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3268 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3269 rw_exit(&origin_head->ds_rwlock);
3270 goto retry;
3271 }
3272 }
3273 csa.cds = clone;
3274 csa.ohds = origin_head;
3275 csa.force = force;
3276 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3277 dsl_dataset_clone_swap_check,
3278 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3279 return (error);
3280 }
3281
3282 /*
3283 * Given a pool name and a dataset object number in that pool,
3284 * return the name of that dataset.
3285 */
3286 int
3287 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3288 {
3289 spa_t *spa;
3290 dsl_pool_t *dp;
3291 dsl_dataset_t *ds;
3292 int error;
3293
3294 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3295 return (error);
3296 dp = spa_get_dsl(spa);
3297 rw_enter(&dp->dp_config_rwlock, RW_READER);
3298 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3299 dsl_dataset_name(ds, buf);
3300 dsl_dataset_rele(ds, FTAG);
3301 }
3302 rw_exit(&dp->dp_config_rwlock);
3303 spa_close(spa, FTAG);
3304
3305 return (error);
3306 }
3307
3308 int
3309 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3310 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3311 {
3312 int error = 0;
3313
3314 ASSERT3S(asize, >, 0);
3315
3316 /*
3317 * *ref_rsrv is the portion of asize that will come from any
3318 * unconsumed refreservation space.
3319 */
3320 *ref_rsrv = 0;
3321
3322 mutex_enter(&ds->ds_lock);
3323 /*
3324 * Make a space adjustment for reserved bytes.
3325 */
3326 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3327 ASSERT3U(*used, >=,
3328 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3329 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3330 *ref_rsrv =
3331 asize - MIN(asize, parent_delta(ds, asize + inflight));
3332 }
3333
3334 if (!check_quota || ds->ds_quota == 0) {
3335 mutex_exit(&ds->ds_lock);
3336 return (0);
3337 }
3338 /*
3339 * If they are requesting more space, and our current estimate
3340 * is over quota, they get to try again unless the actual
3341 * on-disk is over quota and there are no pending changes (which
3342 * may free up space for us).
3343 */
3344 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3345 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3346 error = ERESTART;
3347 else
3348 error = EDQUOT;
3349 }
3350 mutex_exit(&ds->ds_lock);
3351
3352 return (error);
3353 }
3354
3355 /* ARGSUSED */
3356 static int
3357 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3358 {
3359 dsl_dataset_t *ds = arg1;
3360 dsl_prop_setarg_t *psa = arg2;
3361 int err;
3362
3363 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3364 return (ENOTSUP);
3365
3366 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3367 return (err);
3368
3369 if (psa->psa_effective_value == 0)
3370 return (0);
3371
3372 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3373 psa->psa_effective_value < ds->ds_reserved)
3374 return (ENOSPC);
3375
3376 return (0);
3377 }
3378
3379 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3380
3381 void
3382 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3383 {
3384 dsl_dataset_t *ds = arg1;
3385 dsl_prop_setarg_t *psa = arg2;
3386 uint64_t effective_value = psa->psa_effective_value;
3387
3388 dsl_prop_set_sync(ds, psa, tx);
3389 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3390
3391 if (ds->ds_quota != effective_value) {
3392 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3393 ds->ds_quota = effective_value;
3394
3395 spa_history_log_internal(LOG_DS_REFQUOTA,
3396 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3397 (longlong_t)ds->ds_quota, ds->ds_object);
3398 }
3399 }
3400
3401 int
3402 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3403 {
3404 dsl_dataset_t *ds;
3405 dsl_prop_setarg_t psa;
3406 int err;
3407
3408 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3409
3410 err = dsl_dataset_hold(dsname, FTAG, &ds);
3411 if (err)
3412 return (err);
3413
3414 /*
3415 * If someone removes a file, then tries to set the quota, we
3416 * want to make sure the file freeing takes effect.
3417 */
3418 txg_wait_open(ds->ds_dir->dd_pool, 0);
3419
3420 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3421 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3422 ds, &psa, 0);
3423
3424 dsl_dataset_rele(ds, FTAG);
3425 return (err);
3426 }
3427
3428 static int
3429 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3430 {
3431 dsl_dataset_t *ds = arg1;
3432 dsl_prop_setarg_t *psa = arg2;
3433 uint64_t effective_value;
3434 uint64_t unique;
3435 int err;
3436
3437 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3438 SPA_VERSION_REFRESERVATION)
3439 return (ENOTSUP);
3440
3441 if (dsl_dataset_is_snapshot(ds))
3442 return (EINVAL);
3443
3444 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3445 return (err);
3446
3447 effective_value = psa->psa_effective_value;
3448
3449 /*
3450 * If we are doing the preliminary check in open context, the
3451 * space estimates may be inaccurate.
3452 */
3453 if (!dmu_tx_is_syncing(tx))
3454 return (0);
3455
3456 mutex_enter(&ds->ds_lock);
3457 if (!DS_UNIQUE_IS_ACCURATE(ds))
3458 dsl_dataset_recalc_head_uniq(ds);
3459 unique = ds->ds_phys->ds_unique_bytes;
3460 mutex_exit(&ds->ds_lock);
3461
3462 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3463 uint64_t delta = MAX(unique, effective_value) -
3464 MAX(unique, ds->ds_reserved);
3465
3466 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3467 return (ENOSPC);
3468 if (ds->ds_quota > 0 &&
3469 effective_value > ds->ds_quota)
3470 return (ENOSPC);
3471 }
3472
3473 return (0);
3474 }
3475
3476 static void
3477 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3478 {
3479 dsl_dataset_t *ds = arg1;
3480 dsl_prop_setarg_t *psa = arg2;
3481 uint64_t effective_value = psa->psa_effective_value;
3482 uint64_t unique;
3483 int64_t delta;
3484
3485 dsl_prop_set_sync(ds, psa, tx);
3486 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3487
3488 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3489
3490 mutex_enter(&ds->ds_dir->dd_lock);
3491 mutex_enter(&ds->ds_lock);
3492 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3493 unique = ds->ds_phys->ds_unique_bytes;
3494 delta = MAX(0, (int64_t)(effective_value - unique)) -
3495 MAX(0, (int64_t)(ds->ds_reserved - unique));
3496 ds->ds_reserved = effective_value;
3497 mutex_exit(&ds->ds_lock);
3498
3499 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3500 mutex_exit(&ds->ds_dir->dd_lock);
3501
3502 spa_history_log_internal(LOG_DS_REFRESERV,
3503 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3504 (longlong_t)effective_value, ds->ds_object);
3505 }
3506
3507 int
3508 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3509 uint64_t reservation)
3510 {
3511 dsl_dataset_t *ds;
3512 dsl_prop_setarg_t psa;
3513 int err;
3514
3515 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3516 &reservation);
3517
3518 err = dsl_dataset_hold(dsname, FTAG, &ds);
3519 if (err)
3520 return (err);
3521
3522 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3523 dsl_dataset_set_reservation_check,
3524 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3525
3526 dsl_dataset_rele(ds, FTAG);
3527 return (err);
3528 }
3529
3530 typedef struct zfs_hold_cleanup_arg {
3531 dsl_pool_t *dp;
3532 uint64_t dsobj;
3533 char htag[MAXNAMELEN];
3534 } zfs_hold_cleanup_arg_t;
3535
3536 static void
3537 dsl_dataset_user_release_onexit(void *arg)
3538 {
3539 zfs_hold_cleanup_arg_t *ca = arg;
3540
3541 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3542 B_TRUE);
3543 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3544 }
3545
3546 void
3547 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3548 minor_t minor)
3549 {
3550 zfs_hold_cleanup_arg_t *ca;
3551
3552 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3553 ca->dp = ds->ds_dir->dd_pool;
3554 ca->dsobj = ds->ds_object;
3555 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3556 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3557 dsl_dataset_user_release_onexit, ca, NULL));
3558 }
3559
3560 /*
3561 * If you add new checks here, you may need to add
3562 * additional checks to the "temporary" case in
3563 * snapshot_check() in dmu_objset.c.
3564 */
3565 static int
3566 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3567 {
3568 dsl_dataset_t *ds = arg1;
3569 struct dsl_ds_holdarg *ha = arg2;
3570 char *htag = ha->htag;
3571 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3572 int error = 0;
3573
3574 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3575 return (ENOTSUP);
3576
3577 if (!dsl_dataset_is_snapshot(ds))
3578 return (EINVAL);
3579
3580 /* tags must be unique */
3581 mutex_enter(&ds->ds_lock);
3582 if (ds->ds_phys->ds_userrefs_obj) {
3583 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3584 8, 1, tx);
3585 if (error == 0)
3586 error = EEXIST;
3587 else if (error == ENOENT)
3588 error = 0;
3589 }
3590 mutex_exit(&ds->ds_lock);
3591
3592 if (error == 0 && ha->temphold &&
3593 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3594 error = E2BIG;
3595
3596 return (error);
3597 }
3598
3599 void
3600 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3601 {
3602 dsl_dataset_t *ds = arg1;
3603 struct dsl_ds_holdarg *ha = arg2;
3604 char *htag = ha->htag;
3605 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3606 objset_t *mos = dp->dp_meta_objset;
3607 uint64_t now = gethrestime_sec();
3608 uint64_t zapobj;
3609
3610 mutex_enter(&ds->ds_lock);
3611 if (ds->ds_phys->ds_userrefs_obj == 0) {
3612 /*
3613 * This is the first user hold for this dataset. Create
3614 * the userrefs zap object.
3615 */
3616 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3617 zapobj = ds->ds_phys->ds_userrefs_obj =
3618 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3619 } else {
3620 zapobj = ds->ds_phys->ds_userrefs_obj;
3621 }
3622 ds->ds_userrefs++;
3623 mutex_exit(&ds->ds_lock);
3624
3625 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3626
3627 if (ha->temphold) {
3628 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3629 htag, &now, tx));
3630 }
3631
3632 spa_history_log_internal(LOG_DS_USER_HOLD,
3633 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3634 (int)ha->temphold, ds->ds_object);
3635 }
3636
3637 static int
3638 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3639 {
3640 struct dsl_ds_holdarg *ha = arg;
3641 dsl_dataset_t *ds;
3642 int error;
3643 char *name;
3644
3645 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3646 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3647 error = dsl_dataset_hold(name, ha->dstg, &ds);
3648 strfree(name);
3649 if (error == 0) {
3650 ha->gotone = B_TRUE;
3651 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3652 dsl_dataset_user_hold_sync, ds, ha, 0);
3653 } else if (error == ENOENT && ha->recursive) {
3654 error = 0;
3655 } else {
3656 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3657 }
3658 return (error);
3659 }
3660
3661 int
3662 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3663 boolean_t temphold)
3664 {
3665 struct dsl_ds_holdarg *ha;
3666 int error;
3667
3668 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3669 ha->htag = htag;
3670 ha->temphold = temphold;
3671 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3672 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3673 ds, ha, 0);
3674 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3675
3676 return (error);
3677 }
3678
3679 int
3680 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3681 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3682 {
3683 struct dsl_ds_holdarg *ha;
3684 dsl_sync_task_t *dst;
3685 spa_t *spa;
3686 int error;
3687 minor_t minor = 0;
3688
3689 if (cleanup_fd != -1) {
3690 /* Currently we only support cleanup-on-exit of tempholds. */
3691 if (!temphold)
3692 return (EINVAL);
3693 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3694 if (error)
3695 return (error);
3696 }
3697
3698 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3699
3700 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3701
3702 error = spa_open(dsname, &spa, FTAG);
3703 if (error) {
3704 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3705 if (cleanup_fd != -1)
3706 zfs_onexit_fd_rele(cleanup_fd);
3707 return (error);
3708 }
3709
3710 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3711 ha->htag = htag;
3712 ha->snapname = snapname;
3713 ha->recursive = recursive;
3714 ha->temphold = temphold;
3715
3716 if (recursive) {
3717 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3718 ha, DS_FIND_CHILDREN);
3719 } else {
3720 error = dsl_dataset_user_hold_one(dsname, ha);
3721 }
3722 if (error == 0)
3723 error = dsl_sync_task_group_wait(ha->dstg);
3724
3725 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3726 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3727 dsl_dataset_t *ds = dst->dst_arg1;
3728
3729 if (dst->dst_err) {
3730 dsl_dataset_name(ds, ha->failed);
3731 *strchr(ha->failed, '@') = '\0';
3732 } else if (error == 0 && minor != 0 && temphold) {
3733 /*
3734 * If this hold is to be released upon process exit,
3735 * register that action now.
3736 */
3737 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3738 }
3739 dsl_dataset_rele(ds, ha->dstg);
3740 }
3741
3742 if (error == 0 && recursive && !ha->gotone)
3743 error = ENOENT;
3744
3745 if (error)
3746 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3747
3748 dsl_sync_task_group_destroy(ha->dstg);
3749
3750 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3751 spa_close(spa, FTAG);
3752 if (cleanup_fd != -1)
3753 zfs_onexit_fd_rele(cleanup_fd);
3754 return (error);
3755 }
3756
3757 struct dsl_ds_releasearg {
3758 dsl_dataset_t *ds;
3759 const char *htag;
3760 boolean_t own; /* do we own or just hold ds? */
3761 };
3762
3763 static int
3764 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3765 boolean_t *might_destroy)
3766 {
3767 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3768 uint64_t zapobj;
3769 uint64_t tmp;
3770 int error;
3771
3772 *might_destroy = B_FALSE;
3773
3774 mutex_enter(&ds->ds_lock);
3775 zapobj = ds->ds_phys->ds_userrefs_obj;
3776 if (zapobj == 0) {
3777 /* The tag can't possibly exist */
3778 mutex_exit(&ds->ds_lock);
3779 return (ESRCH);
3780 }
3781
3782 /* Make sure the tag exists */
3783 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3784 if (error) {
3785 mutex_exit(&ds->ds_lock);
3786 if (error == ENOENT)
3787 error = ESRCH;
3788 return (error);
3789 }
3790
3791 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3792 DS_IS_DEFER_DESTROY(ds))
3793 *might_destroy = B_TRUE;
3794
3795 mutex_exit(&ds->ds_lock);
3796 return (0);
3797 }
3798
3799 static int
3800 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3801 {
3802 struct dsl_ds_releasearg *ra = arg1;
3803 dsl_dataset_t *ds = ra->ds;
3804 boolean_t might_destroy;
3805 int error;
3806
3807 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3808 return (ENOTSUP);
3809
3810 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3811 if (error)
3812 return (error);
3813
3814 if (might_destroy) {
3815 struct dsl_ds_destroyarg dsda = {0};
3816
3817 if (dmu_tx_is_syncing(tx)) {
3818 /*
3819 * If we're not prepared to remove the snapshot,
3820 * we can't allow the release to happen right now.
3821 */
3822 if (!ra->own)
3823 return (EBUSY);
3824 }
3825 dsda.ds = ds;
3826 dsda.releasing = B_TRUE;
3827 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3828 }
3829
3830 return (0);
3831 }
3832
3833 static void
3834 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3835 {
3836 struct dsl_ds_releasearg *ra = arg1;
3837 dsl_dataset_t *ds = ra->ds;
3838 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3839 objset_t *mos = dp->dp_meta_objset;
3840 uint64_t zapobj;
3841 uint64_t dsobj = ds->ds_object;
3842 uint64_t refs;
3843 int error;
3844
3845 mutex_enter(&ds->ds_lock);
3846 ds->ds_userrefs--;
3847 refs = ds->ds_userrefs;
3848 mutex_exit(&ds->ds_lock);
3849 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3850 VERIFY(error == 0 || error == ENOENT);
3851 zapobj = ds->ds_phys->ds_userrefs_obj;
3852 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3853 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3854 DS_IS_DEFER_DESTROY(ds)) {
3855 struct dsl_ds_destroyarg dsda = {0};
3856
3857 ASSERT(ra->own);
3858 dsda.ds = ds;
3859 dsda.releasing = B_TRUE;
3860 /* We already did the destroy_check */
3861 dsl_dataset_destroy_sync(&dsda, tag, tx);
3862 }
3863
3864 spa_history_log_internal(LOG_DS_USER_RELEASE,
3865 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3866 ra->htag, (longlong_t)refs, dsobj);
3867 }
3868
3869 static int
3870 dsl_dataset_user_release_one(const char *dsname, void *arg)
3871 {
3872 struct dsl_ds_holdarg *ha = arg;
3873 struct dsl_ds_releasearg *ra;
3874 dsl_dataset_t *ds;
3875 int error;
3876 void *dtag = ha->dstg;
3877 char *name;
3878 boolean_t own = B_FALSE;
3879 boolean_t might_destroy;
3880
3881 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3882 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3883 error = dsl_dataset_hold(name, dtag, &ds);
3884 strfree(name);
3885 if (error == ENOENT && ha->recursive)
3886 return (0);
3887 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3888 if (error)
3889 return (error);
3890
3891 ha->gotone = B_TRUE;
3892
3893 ASSERT(dsl_dataset_is_snapshot(ds));
3894
3895 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3896 if (error) {
3897 dsl_dataset_rele(ds, dtag);
3898 return (error);
3899 }
3900
3901 if (might_destroy) {
3902 #ifdef _KERNEL
3903 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3904 error = zfs_unmount_snap(name, NULL);
3905 strfree(name);
3906 if (error) {
3907 dsl_dataset_rele(ds, dtag);
3908 return (error);
3909 }
3910 #endif
3911 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3912 dsl_dataset_rele(ds, dtag);
3913 return (EBUSY);
3914 } else {
3915 own = B_TRUE;
3916 dsl_dataset_make_exclusive(ds, dtag);
3917 }
3918 }
3919
3920 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3921 ra->ds = ds;
3922 ra->htag = ha->htag;
3923 ra->own = own;
3924 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3925 dsl_dataset_user_release_sync, ra, dtag, 0);
3926
3927 return (0);
3928 }
3929
3930 int
3931 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3932 boolean_t recursive)
3933 {
3934 struct dsl_ds_holdarg *ha;
3935 dsl_sync_task_t *dst;
3936 spa_t *spa;
3937 int error;
3938
3939 top:
3940 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3941
3942 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3943
3944 error = spa_open(dsname, &spa, FTAG);
3945 if (error) {
3946 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3947 return (error);
3948 }
3949
3950 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3951 ha->htag = htag;
3952 ha->snapname = snapname;
3953 ha->recursive = recursive;
3954 if (recursive) {
3955 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3956 ha, DS_FIND_CHILDREN);
3957 } else {
3958 error = dsl_dataset_user_release_one(dsname, ha);
3959 }
3960 if (error == 0)
3961 error = dsl_sync_task_group_wait(ha->dstg);
3962
3963 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3964 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3965 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3966 dsl_dataset_t *ds = ra->ds;
3967
3968 if (dst->dst_err)
3969 dsl_dataset_name(ds, ha->failed);
3970
3971 if (ra->own)
3972 dsl_dataset_disown(ds, ha->dstg);
3973 else
3974 dsl_dataset_rele(ds, ha->dstg);
3975
3976 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3977 }
3978
3979 if (error == 0 && recursive && !ha->gotone)
3980 error = ENOENT;
3981
3982 if (error && error != EBUSY)
3983 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3984
3985 dsl_sync_task_group_destroy(ha->dstg);
3986 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3987 spa_close(spa, FTAG);
3988
3989 /*
3990 * We can get EBUSY if we were racing with deferred destroy and
3991 * dsl_dataset_user_release_check() hadn't done the necessary
3992 * open context setup. We can also get EBUSY if we're racing
3993 * with destroy and that thread is the ds_owner. Either way
3994 * the busy condition should be transient, and we should retry
3995 * the release operation.
3996 */
3997 if (error == EBUSY)
3998 goto top;
3999
4000 return (error);
4001 }
4002
4003 /*
4004 * Called at spa_load time (with retry == B_FALSE) to release a stale
4005 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4006 */
4007 int
4008 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4009 boolean_t retry)
4010 {
4011 dsl_dataset_t *ds;
4012 char *snap;
4013 char *name;
4014 int namelen;
4015 int error;
4016
4017 do {
4018 rw_enter(&dp->dp_config_rwlock, RW_READER);
4019 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4020 rw_exit(&dp->dp_config_rwlock);
4021 if (error)
4022 return (error);
4023 namelen = dsl_dataset_namelen(ds)+1;
4024 name = kmem_alloc(namelen, KM_SLEEP);
4025 dsl_dataset_name(ds, name);
4026 dsl_dataset_rele(ds, FTAG);
4027
4028 snap = strchr(name, '@');
4029 *snap = '\0';
4030 ++snap;
4031 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4032 kmem_free(name, namelen);
4033
4034 /*
4035 * The object can't have been destroyed because we have a hold,
4036 * but it might have been renamed, resulting in ENOENT. Retry
4037 * if we've been requested to do so.
4038 *
4039 * It would be nice if we could use the dsobj all the way
4040 * through and avoid ENOENT entirely. But we might need to
4041 * unmount the snapshot, and there's currently no way to lookup
4042 * a vfsp using a ZFS object id.
4043 */
4044 } while ((error == ENOENT) && retry);
4045
4046 return (error);
4047 }
4048
4049 int
4050 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4051 {
4052 dsl_dataset_t *ds;
4053 int err;
4054
4055 err = dsl_dataset_hold(dsname, FTAG, &ds);
4056 if (err)
4057 return (err);
4058
4059 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4060 if (ds->ds_phys->ds_userrefs_obj != 0) {
4061 zap_attribute_t *za;
4062 zap_cursor_t zc;
4063
4064 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4065 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4066 ds->ds_phys->ds_userrefs_obj);
4067 zap_cursor_retrieve(&zc, za) == 0;
4068 zap_cursor_advance(&zc)) {
4069 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4070 za->za_first_integer));
4071 }
4072 zap_cursor_fini(&zc);
4073 kmem_free(za, sizeof (zap_attribute_t));
4074 }
4075 dsl_dataset_rele(ds, FTAG);
4076 return (0);
4077 }
4078
4079 /*
4080 * Note, this function is used as the callback for dmu_objset_find(). We
4081 * always return 0 so that we will continue to find and process
4082 * inconsistent datasets, even if we encounter an error trying to
4083 * process one of them.
4084 */
4085 /* ARGSUSED */
4086 int
4087 dsl_destroy_inconsistent(const char *dsname, void *arg)
4088 {
4089 dsl_dataset_t *ds;
4090
4091 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4092 if (DS_IS_INCONSISTENT(ds))
4093 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4094 else
4095 dsl_dataset_disown(ds, FTAG);
4096 }
4097 return (0);
4098 }
4099
4100 /*
4101 * Return (in *usedp) the amount of space written in new that is not
4102 * present in oldsnap. New may be a snapshot or the head. Old must be
4103 * a snapshot before new, in new's filesystem (or its origin). If not then
4104 * fail and return EINVAL.
4105 *
4106 * The written space is calculated by considering two components: First, we
4107 * ignore any freed space, and calculate the written as new's used space
4108 * minus old's used space. Next, we add in the amount of space that was freed
4109 * between the two snapshots, thus reducing new's used space relative to old's.
4110 * Specifically, this is the space that was born before old->ds_creation_txg,
4111 * and freed before new (ie. on new's deadlist or a previous deadlist).
4112 *
4113 * space freed [---------------------]
4114 * snapshots ---O-------O--------O-------O------
4115 * oldsnap new
4116 */
4117 int
4118 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4119 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4120 {
4121 int err = 0;
4122 uint64_t snapobj;
4123 dsl_pool_t *dp = new->ds_dir->dd_pool;
4124
4125 *usedp = 0;
4126 *usedp += new->ds_phys->ds_used_bytes;
4127 *usedp -= oldsnap->ds_phys->ds_used_bytes;
4128
4129 *compp = 0;
4130 *compp += new->ds_phys->ds_compressed_bytes;
4131 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4132
4133 *uncompp = 0;
4134 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4135 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4136
4137 rw_enter(&dp->dp_config_rwlock, RW_READER);
4138 snapobj = new->ds_object;
4139 while (snapobj != oldsnap->ds_object) {
4140 dsl_dataset_t *snap;
4141 uint64_t used, comp, uncomp;
4142
4143 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4144 if (err != 0)
4145 break;
4146
4147 if (snap->ds_phys->ds_prev_snap_txg ==
4148 oldsnap->ds_phys->ds_creation_txg) {
4149 /*
4150 * The blocks in the deadlist can not be born after
4151 * ds_prev_snap_txg, so get the whole deadlist space,
4152 * which is more efficient (especially for old-format
4153 * deadlists). Unfortunately the deadlist code
4154 * doesn't have enough information to make this
4155 * optimization itself.
4156 */
4157 dsl_deadlist_space(&snap->ds_deadlist,
4158 &used, &comp, &uncomp);
4159 } else {
4160 dsl_deadlist_space_range(&snap->ds_deadlist,
4161 0, oldsnap->ds_phys->ds_creation_txg,
4162 &used, &comp, &uncomp);
4163 }
4164 *usedp += used;
4165 *compp += comp;
4166 *uncompp += uncomp;
4167
4168 /*
4169 * If we get to the beginning of the chain of snapshots
4170 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4171 * was not a snapshot of/before new.
4172 */
4173 snapobj = snap->ds_phys->ds_prev_snap_obj;
4174 dsl_dataset_rele(snap, FTAG);
4175 if (snapobj == 0) {
4176 err = EINVAL;
4177 break;
4178 }
4179
4180 }
4181 rw_exit(&dp->dp_config_rwlock);
4182 return (err);
4183 }
4184
4185 /*
4186 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4187 * lastsnap, and all snapshots in between are deleted.
4188 *
4189 * blocks that would be freed [---------------------------]
4190 * snapshots ---O-------O--------O-------O--------O
4191 * firstsnap lastsnap
4192 *
4193 * This is the set of blocks that were born after the snap before firstsnap,
4194 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4195 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4196 * We calculate this by iterating over the relevant deadlists (from the snap
4197 * after lastsnap, backward to the snap after firstsnap), summing up the
4198 * space on the deadlist that was born after the snap before firstsnap.
4199 */
4200 int
4201 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4202 dsl_dataset_t *lastsnap,
4203 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4204 {
4205 int err = 0;
4206 uint64_t snapobj;
4207 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4208
4209 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4210 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4211
4212 /*
4213 * Check that the snapshots are in the same dsl_dir, and firstsnap
4214 * is before lastsnap.
4215 */
4216 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4217 firstsnap->ds_phys->ds_creation_txg >
4218 lastsnap->ds_phys->ds_creation_txg)
4219 return (EINVAL);
4220
4221 *usedp = *compp = *uncompp = 0;
4222
4223 rw_enter(&dp->dp_config_rwlock, RW_READER);
4224 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4225 while (snapobj != firstsnap->ds_object) {
4226 dsl_dataset_t *ds;
4227 uint64_t used, comp, uncomp;
4228
4229 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4230 if (err != 0)
4231 break;
4232
4233 dsl_deadlist_space_range(&ds->ds_deadlist,
4234 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4235 &used, &comp, &uncomp);
4236 *usedp += used;
4237 *compp += comp;
4238 *uncompp += uncomp;
4239
4240 snapobj = ds->ds_phys->ds_prev_snap_obj;
4241 ASSERT3U(snapobj, !=, 0);
4242 dsl_dataset_rele(ds, FTAG);
4243 }
4244 rw_exit(&dp->dp_config_rwlock);
4245 return (err);
4246 }