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) 2012 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/zfeature.h>
39 #include <sys/unique.h>
40 #include <sys/zfs_context.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/spa.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_onexit.h>
45 #include <sys/zvol.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/dsl_deadlist.h>
48
49 static char *dsl_reaper = "the grim reaper";
50
51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
54
55 #define SWITCH64(x, y) \
56 { \
57 uint64_t __tmp = (x); \
58 (x) = (y); \
59 (y) = __tmp; \
60 }
61
62 #define DS_REF_MAX (1ULL << 62)
63
64 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
65
66 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67
68
69 /*
70 * Figure out how much of this delta should be propogated to the dsl_dir
71 * layer. If there's a refreservation, that space has already been
72 * partially accounted for in our ancestors.
73 */
74 static int64_t
75 parent_delta(dsl_dataset_t *ds, int64_t delta)
76 {
77 uint64_t old_bytes, new_bytes;
78
79 if (ds->ds_reserved == 0)
80 return (delta);
81
82 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
83 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
84
85 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
86 return (new_bytes - old_bytes);
87 }
88
89 void
90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
91 {
92 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
93 int compressed = BP_GET_PSIZE(bp);
94 int uncompressed = BP_GET_UCSIZE(bp);
95 int64_t delta;
96
97 dprintf_bp(bp, "ds=%p", ds);
98
99 ASSERT(dmu_tx_is_syncing(tx));
100 /* It could have been compressed away to nothing */
101 if (BP_IS_HOLE(bp))
102 return;
103 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
104 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
105 if (ds == NULL) {
106 /*
107 * Account for the meta-objset space in its placeholder
108 * dsl_dir.
109 */
110 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
111 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
112 used, compressed, uncompressed, tx);
113 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
114 return;
115 }
116 dmu_buf_will_dirty(ds->ds_dbuf, tx);
117
118 mutex_enter(&ds->ds_dir->dd_lock);
119 mutex_enter(&ds->ds_lock);
120 delta = parent_delta(ds, used);
121 ds->ds_phys->ds_referenced_bytes += used;
122 ds->ds_phys->ds_compressed_bytes += compressed;
123 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 ds->ds_phys->ds_unique_bytes += used;
125 mutex_exit(&ds->ds_lock);
126 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
127 compressed, uncompressed, tx);
128 dsl_dir_transfer_space(ds->ds_dir, used - delta,
129 DD_USED_REFRSRV, DD_USED_HEAD, tx);
130 mutex_exit(&ds->ds_dir->dd_lock);
131 }
132
133 int
134 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
135 boolean_t async)
136 {
137 if (BP_IS_HOLE(bp))
138 return (0);
139
140 ASSERT(dmu_tx_is_syncing(tx));
141 ASSERT(bp->blk_birth <= tx->tx_txg);
142
143 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
144 int compressed = BP_GET_PSIZE(bp);
145 int uncompressed = BP_GET_UCSIZE(bp);
146
147 ASSERT(used > 0);
148 if (ds == NULL) {
149 /*
150 * Account for the meta-objset space in its placeholder
151 * dataset.
152 */
153 dsl_free(tx->tx_pool, tx->tx_txg, bp);
154
155 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
156 -used, -compressed, -uncompressed, tx);
157 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
158 return (used);
159 }
160 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
161
162 ASSERT(!dsl_dataset_is_snapshot(ds));
163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
164
165 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
166 int64_t delta;
167
168 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
169 dsl_free(tx->tx_pool, tx->tx_txg, bp);
170
171 mutex_enter(&ds->ds_dir->dd_lock);
172 mutex_enter(&ds->ds_lock);
173 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
174 !DS_UNIQUE_IS_ACCURATE(ds));
175 delta = parent_delta(ds, -used);
176 ds->ds_phys->ds_unique_bytes -= used;
177 mutex_exit(&ds->ds_lock);
178 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
179 delta, -compressed, -uncompressed, tx);
180 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
181 DD_USED_REFRSRV, DD_USED_HEAD, tx);
182 mutex_exit(&ds->ds_dir->dd_lock);
183 } else {
184 dprintf_bp(bp, "putting on dead list: %s", "");
185 if (async) {
186 /*
187 * We are here as part of zio's write done callback,
188 * which means we're a zio interrupt thread. We can't
189 * call dsl_deadlist_insert() now because it may block
190 * waiting for I/O. Instead, put bp on the deferred
191 * queue and let dsl_pool_sync() finish the job.
192 */
193 bplist_append(&ds->ds_pending_deadlist, bp);
194 } else {
195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 }
197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 ds->ds_phys->ds_prev_snap_obj);
199 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
202 ds->ds_object && bp->blk_birth >
203 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 mutex_enter(&ds->ds_prev->ds_lock);
206 ds->ds_prev->ds_phys->ds_unique_bytes += used;
207 mutex_exit(&ds->ds_prev->ds_lock);
208 }
209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 dsl_dir_transfer_space(ds->ds_dir, used,
211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 }
213 }
214 mutex_enter(&ds->ds_lock);
215 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
216 ds->ds_phys->ds_referenced_bytes -= used;
217 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
218 ds->ds_phys->ds_compressed_bytes -= compressed;
219 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
220 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
221 mutex_exit(&ds->ds_lock);
222
223 return (used);
224 }
225
226 uint64_t
227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 {
229 uint64_t trysnap = 0;
230
231 if (ds == NULL)
232 return (0);
233 /*
234 * The snapshot creation could fail, but that would cause an
235 * incorrect FALSE return, which would only result in an
236 * overestimation of the amount of space that an operation would
237 * consume, which is OK.
238 *
239 * There's also a small window where we could miss a pending
240 * snapshot, because we could set the sync task in the quiescing
241 * phase. So this should only be used as a guess.
242 */
243 if (ds->ds_trysnap_txg >
244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 trysnap = ds->ds_trysnap_txg;
246 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
247 }
248
249 boolean_t
250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 uint64_t blk_birth)
252 {
253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
254 return (B_FALSE);
255
256 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
257
258 return (B_TRUE);
259 }
260
261 /* ARGSUSED */
262 static void
263 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
264 {
265 dsl_dataset_t *ds = dsv;
266
267 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
268
269 unique_remove(ds->ds_fsid_guid);
270
271 if (ds->ds_objset != NULL)
272 dmu_objset_evict(ds->ds_objset);
273
274 if (ds->ds_prev) {
275 dsl_dataset_drop_ref(ds->ds_prev, ds);
276 ds->ds_prev = NULL;
277 }
278
279 bplist_destroy(&ds->ds_pending_deadlist);
280 if (db != NULL) {
281 dsl_deadlist_close(&ds->ds_deadlist);
282 } else {
283 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
284 ASSERT(!ds->ds_deadlist.dl_oldfmt);
285 }
286 if (ds->ds_dir)
287 dsl_dir_close(ds->ds_dir, ds);
288
289 ASSERT(!list_link_active(&ds->ds_synced_link));
290
291 mutex_destroy(&ds->ds_lock);
292 mutex_destroy(&ds->ds_recvlock);
293 mutex_destroy(&ds->ds_opening_lock);
294 rw_destroy(&ds->ds_rwlock);
295 cv_destroy(&ds->ds_exclusive_cv);
296
297 kmem_free(ds, sizeof (dsl_dataset_t));
298 }
299
300 static int
301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
302 {
303 dsl_dataset_phys_t *headphys;
304 int err;
305 dmu_buf_t *headdbuf;
306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 objset_t *mos = dp->dp_meta_objset;
308
309 if (ds->ds_snapname[0])
310 return (0);
311 if (ds->ds_phys->ds_next_snap_obj == 0)
312 return (0);
313
314 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
315 FTAG, &headdbuf);
316 if (err)
317 return (err);
318 headphys = headdbuf->db_data;
319 err = zap_value_search(dp->dp_meta_objset,
320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 dmu_buf_rele(headdbuf, FTAG);
322 return (err);
323 }
324
325 static int
326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
327 {
328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
330 matchtype_t mt;
331 int err;
332
333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
334 mt = MT_FIRST;
335 else
336 mt = MT_EXACT;
337
338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
342 return (err);
343 }
344
345 static int
346 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
347 {
348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
350 matchtype_t mt;
351 int err;
352
353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354
355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
356 mt = MT_FIRST;
357 else
358 mt = MT_EXACT;
359
360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
361 if (err == ENOTSUP && mt == MT_FIRST)
362 err = zap_remove(mos, snapobj, name, tx);
363 return (err);
364 }
365
366 static int
367 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
368 dsl_dataset_t **dsp)
369 {
370 objset_t *mos = dp->dp_meta_objset;
371 dmu_buf_t *dbuf;
372 dsl_dataset_t *ds;
373 int err;
374 dmu_object_info_t doi;
375
376 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
377 dsl_pool_sync_context(dp));
378
379 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
380 if (err)
381 return (err);
382
383 /* Make sure dsobj has the correct object type. */
384 dmu_object_info_from_db(dbuf, &doi);
385 if (doi.doi_type != DMU_OT_DSL_DATASET)
386 return (EINVAL);
387
388 ds = dmu_buf_get_user(dbuf);
389 if (ds == NULL) {
390 dsl_dataset_t *winner;
391
392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
393 ds->ds_dbuf = dbuf;
394 ds->ds_object = dsobj;
395 ds->ds_phys = dbuf->db_data;
396
397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
400 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
401
402 rw_init(&ds->ds_rwlock, 0, 0, 0);
403 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
404
405 bplist_create(&ds->ds_pending_deadlist);
406 dsl_deadlist_open(&ds->ds_deadlist,
407 mos, ds->ds_phys->ds_deadlist_obj);
408
409 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
410 offsetof(dmu_sendarg_t, dsa_link));
411
412 if (err == 0) {
413 err = dsl_dir_open_obj(dp,
414 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
415 }
416 if (err) {
417 mutex_destroy(&ds->ds_lock);
418 mutex_destroy(&ds->ds_recvlock);
419 mutex_destroy(&ds->ds_opening_lock);
420 rw_destroy(&ds->ds_rwlock);
421 cv_destroy(&ds->ds_exclusive_cv);
422 bplist_destroy(&ds->ds_pending_deadlist);
423 dsl_deadlist_close(&ds->ds_deadlist);
424 kmem_free(ds, sizeof (dsl_dataset_t));
425 dmu_buf_rele(dbuf, tag);
426 return (err);
427 }
428
429 if (!dsl_dataset_is_snapshot(ds)) {
430 ds->ds_snapname[0] = '\0';
431 if (ds->ds_phys->ds_prev_snap_obj) {
432 err = dsl_dataset_get_ref(dp,
433 ds->ds_phys->ds_prev_snap_obj,
434 ds, &ds->ds_prev);
435 }
436 } else {
437 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
438 err = dsl_dataset_get_snapname(ds);
439 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
440 err = zap_count(
441 ds->ds_dir->dd_pool->dp_meta_objset,
442 ds->ds_phys->ds_userrefs_obj,
443 &ds->ds_userrefs);
444 }
445 }
446
447 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
448 /*
449 * In sync context, we're called with either no lock
450 * or with the write lock. If we're not syncing,
451 * we're always called with the read lock held.
452 */
453 boolean_t need_lock =
454 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
455 dsl_pool_sync_context(dp);
456
457 if (need_lock)
458 rw_enter(&dp->dp_config_rwlock, RW_READER);
459
460 err = dsl_prop_get_ds(ds,
461 "refreservation", sizeof (uint64_t), 1,
462 &ds->ds_reserved, NULL);
463 if (err == 0) {
464 err = dsl_prop_get_ds(ds,
465 "refquota", sizeof (uint64_t), 1,
466 &ds->ds_quota, NULL);
467 }
468
469 if (need_lock)
470 rw_exit(&dp->dp_config_rwlock);
471 } else {
472 ds->ds_reserved = ds->ds_quota = 0;
473 }
474
475 if (err == 0) {
476 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
477 dsl_dataset_evict);
478 }
479 if (err || winner) {
480 bplist_destroy(&ds->ds_pending_deadlist);
481 dsl_deadlist_close(&ds->ds_deadlist);
482 if (ds->ds_prev)
483 dsl_dataset_drop_ref(ds->ds_prev, ds);
484 dsl_dir_close(ds->ds_dir, ds);
485 mutex_destroy(&ds->ds_lock);
486 mutex_destroy(&ds->ds_recvlock);
487 mutex_destroy(&ds->ds_opening_lock);
488 rw_destroy(&ds->ds_rwlock);
489 cv_destroy(&ds->ds_exclusive_cv);
490 kmem_free(ds, sizeof (dsl_dataset_t));
491 if (err) {
492 dmu_buf_rele(dbuf, tag);
493 return (err);
494 }
495 ds = winner;
496 } else {
497 ds->ds_fsid_guid =
498 unique_insert(ds->ds_phys->ds_fsid_guid);
499 }
500 }
501 ASSERT3P(ds->ds_dbuf, ==, dbuf);
502 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
503 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
504 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
505 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
506 mutex_enter(&ds->ds_lock);
507 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
508 mutex_exit(&ds->ds_lock);
509 dmu_buf_rele(ds->ds_dbuf, tag);
510 return (ENOENT);
511 }
512 mutex_exit(&ds->ds_lock);
513 *dsp = ds;
514 return (0);
515 }
516
517 static int
518 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
519 {
520 dsl_pool_t *dp = ds->ds_dir->dd_pool;
521
522 /*
523 * In syncing context we don't want the rwlock lock: there
524 * may be an existing writer waiting for sync phase to
525 * finish. We don't need to worry about such writers, since
526 * sync phase is single-threaded, so the writer can't be
527 * doing anything while we are active.
528 */
529 if (dsl_pool_sync_context(dp)) {
530 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
531 return (0);
532 }
533
534 /*
535 * Normal users will hold the ds_rwlock as a READER until they
536 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
537 * drop their READER lock after they set the ds_owner field.
538 *
539 * If the dataset is being destroyed, the destroy thread will
540 * obtain a WRITER lock for exclusive access after it's done its
541 * open-context work and then change the ds_owner to
542 * dsl_reaper once destruction is assured. So threads
543 * may block here temporarily, until the "destructability" of
544 * the dataset is determined.
545 */
546 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
547 mutex_enter(&ds->ds_lock);
548 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
549 rw_exit(&dp->dp_config_rwlock);
550 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
551 if (DSL_DATASET_IS_DESTROYED(ds)) {
552 mutex_exit(&ds->ds_lock);
553 dsl_dataset_drop_ref(ds, tag);
554 rw_enter(&dp->dp_config_rwlock, RW_READER);
555 return (ENOENT);
556 }
557 /*
558 * The dp_config_rwlock lives above the ds_lock. And
559 * we need to check DSL_DATASET_IS_DESTROYED() while
560 * holding the ds_lock, so we have to drop and reacquire
561 * the ds_lock here.
562 */
563 mutex_exit(&ds->ds_lock);
564 rw_enter(&dp->dp_config_rwlock, RW_READER);
565 mutex_enter(&ds->ds_lock);
566 }
567 mutex_exit(&ds->ds_lock);
568 return (0);
569 }
570
571 int
572 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
573 dsl_dataset_t **dsp)
574 {
575 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
576
577 if (err)
578 return (err);
579 return (dsl_dataset_hold_ref(*dsp, tag));
580 }
581
582 int
583 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
584 void *tag, dsl_dataset_t **dsp)
585 {
586 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
587 if (err)
588 return (err);
589 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
590 dsl_dataset_rele(*dsp, tag);
591 *dsp = NULL;
592 return (EBUSY);
593 }
594 return (0);
595 }
596
597 int
598 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
599 {
600 dsl_dir_t *dd;
601 dsl_pool_t *dp;
602 const char *snapname;
603 uint64_t obj;
604 int err = 0;
605
606 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
607 if (err)
608 return (err);
609
610 dp = dd->dd_pool;
611 obj = dd->dd_phys->dd_head_dataset_obj;
612 rw_enter(&dp->dp_config_rwlock, RW_READER);
613 if (obj)
614 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
615 else
616 err = ENOENT;
617 if (err)
618 goto out;
619
620 err = dsl_dataset_hold_ref(*dsp, tag);
621
622 /* we may be looking for a snapshot */
623 if (err == 0 && snapname != NULL) {
624 dsl_dataset_t *ds = NULL;
625
626 if (*snapname++ != '@') {
627 dsl_dataset_rele(*dsp, tag);
628 err = ENOENT;
629 goto out;
630 }
631
632 dprintf("looking for snapshot '%s'\n", snapname);
633 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
634 if (err == 0)
635 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
636 dsl_dataset_rele(*dsp, tag);
637
638 ASSERT3U((err == 0), ==, (ds != NULL));
639
640 if (ds) {
641 mutex_enter(&ds->ds_lock);
642 if (ds->ds_snapname[0] == 0)
643 (void) strlcpy(ds->ds_snapname, snapname,
644 sizeof (ds->ds_snapname));
645 mutex_exit(&ds->ds_lock);
646 err = dsl_dataset_hold_ref(ds, tag);
647 *dsp = err ? NULL : ds;
648 }
649 }
650 out:
651 rw_exit(&dp->dp_config_rwlock);
652 dsl_dir_close(dd, FTAG);
653 return (err);
654 }
655
656 int
657 dsl_dataset_own(const char *name, boolean_t inconsistentok,
658 void *tag, dsl_dataset_t **dsp)
659 {
660 int err = dsl_dataset_hold(name, tag, dsp);
661 if (err)
662 return (err);
663 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
664 dsl_dataset_rele(*dsp, tag);
665 return (EBUSY);
666 }
667 return (0);
668 }
669
670 void
671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
672 {
673 if (ds == NULL) {
674 (void) strcpy(name, "mos");
675 } else {
676 dsl_dir_name(ds->ds_dir, name);
677 VERIFY(0 == dsl_dataset_get_snapname(ds));
678 if (ds->ds_snapname[0]) {
679 (void) strcat(name, "@");
680 /*
681 * We use a "recursive" mutex so that we
682 * can call dprintf_ds() with ds_lock held.
683 */
684 if (!MUTEX_HELD(&ds->ds_lock)) {
685 mutex_enter(&ds->ds_lock);
686 (void) strcat(name, ds->ds_snapname);
687 mutex_exit(&ds->ds_lock);
688 } else {
689 (void) strcat(name, ds->ds_snapname);
690 }
691 }
692 }
693 }
694
695 static int
696 dsl_dataset_namelen(dsl_dataset_t *ds)
697 {
698 int result;
699
700 if (ds == NULL) {
701 result = 3; /* "mos" */
702 } else {
703 result = dsl_dir_namelen(ds->ds_dir);
704 VERIFY(0 == dsl_dataset_get_snapname(ds));
705 if (ds->ds_snapname[0]) {
706 ++result; /* adding one for the @-sign */
707 if (!MUTEX_HELD(&ds->ds_lock)) {
708 mutex_enter(&ds->ds_lock);
709 result += strlen(ds->ds_snapname);
710 mutex_exit(&ds->ds_lock);
711 } else {
712 result += strlen(ds->ds_snapname);
713 }
714 }
715 }
716
717 return (result);
718 }
719
720 void
721 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
722 {
723 dmu_buf_rele(ds->ds_dbuf, tag);
724 }
725
726 void
727 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
728 {
729 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
730 rw_exit(&ds->ds_rwlock);
731 }
732 dsl_dataset_drop_ref(ds, tag);
733 }
734
735 void
736 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
737 {
738 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
739 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
740
741 mutex_enter(&ds->ds_lock);
742 ds->ds_owner = NULL;
743 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
744 rw_exit(&ds->ds_rwlock);
745 cv_broadcast(&ds->ds_exclusive_cv);
746 }
747 mutex_exit(&ds->ds_lock);
748 if (ds->ds_dbuf)
749 dsl_dataset_drop_ref(ds, tag);
750 else
751 dsl_dataset_evict(NULL, ds);
752 }
753
754 boolean_t
755 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
756 {
757 boolean_t gotit = FALSE;
758
759 mutex_enter(&ds->ds_lock);
760 if (ds->ds_owner == NULL &&
761 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
762 ds->ds_owner = tag;
763 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
764 rw_exit(&ds->ds_rwlock);
765 gotit = TRUE;
766 }
767 mutex_exit(&ds->ds_lock);
768 return (gotit);
769 }
770
771 void
772 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
773 {
774 ASSERT3P(owner, ==, ds->ds_owner);
775 if (!RW_WRITE_HELD(&ds->ds_rwlock))
776 rw_enter(&ds->ds_rwlock, RW_WRITER);
777 }
778
779 uint64_t
780 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
781 uint64_t flags, dmu_tx_t *tx)
782 {
783 dsl_pool_t *dp = dd->dd_pool;
784 dmu_buf_t *dbuf;
785 dsl_dataset_phys_t *dsphys;
786 uint64_t dsobj;
787 objset_t *mos = dp->dp_meta_objset;
788
789 if (origin == NULL)
790 origin = dp->dp_origin_snap;
791
792 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
793 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
794 ASSERT(dmu_tx_is_syncing(tx));
795 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
796
797 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
798 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
799 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
800 dmu_buf_will_dirty(dbuf, tx);
801 dsphys = dbuf->db_data;
802 bzero(dsphys, sizeof (dsl_dataset_phys_t));
803 dsphys->ds_dir_obj = dd->dd_object;
804 dsphys->ds_flags = flags;
805 dsphys->ds_fsid_guid = unique_create();
806 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
807 sizeof (dsphys->ds_guid));
808 dsphys->ds_snapnames_zapobj =
809 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
810 DMU_OT_NONE, 0, tx);
811 dsphys->ds_creation_time = gethrestime_sec();
812 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
813
814 if (origin == NULL) {
815 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
816 } else {
817 dsl_dataset_t *ohds;
818
819 dsphys->ds_prev_snap_obj = origin->ds_object;
820 dsphys->ds_prev_snap_txg =
821 origin->ds_phys->ds_creation_txg;
822 dsphys->ds_referenced_bytes =
823 origin->ds_phys->ds_referenced_bytes;
824 dsphys->ds_compressed_bytes =
825 origin->ds_phys->ds_compressed_bytes;
826 dsphys->ds_uncompressed_bytes =
827 origin->ds_phys->ds_uncompressed_bytes;
828 dsphys->ds_bp = origin->ds_phys->ds_bp;
829 dsphys->ds_flags |= origin->ds_phys->ds_flags;
830
831 dmu_buf_will_dirty(origin->ds_dbuf, tx);
832 origin->ds_phys->ds_num_children++;
833
834 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
835 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
836 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
837 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
838 dsl_dataset_rele(ohds, FTAG);
839
840 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
841 if (origin->ds_phys->ds_next_clones_obj == 0) {
842 origin->ds_phys->ds_next_clones_obj =
843 zap_create(mos,
844 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
845 }
846 VERIFY(0 == zap_add_int(mos,
847 origin->ds_phys->ds_next_clones_obj,
848 dsobj, tx));
849 }
850
851 dmu_buf_will_dirty(dd->dd_dbuf, tx);
852 dd->dd_phys->dd_origin_obj = origin->ds_object;
853 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
854 if (origin->ds_dir->dd_phys->dd_clones == 0) {
855 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
856 origin->ds_dir->dd_phys->dd_clones =
857 zap_create(mos,
858 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
859 }
860 VERIFY3U(0, ==, zap_add_int(mos,
861 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
862 }
863 }
864
865 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
866 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867
868 dmu_buf_rele(dbuf, FTAG);
869
870 dmu_buf_will_dirty(dd->dd_dbuf, tx);
871 dd->dd_phys->dd_head_dataset_obj = dsobj;
872
873 return (dsobj);
874 }
875
876 uint64_t
877 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
878 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
879 {
880 dsl_pool_t *dp = pdd->dd_pool;
881 uint64_t dsobj, ddobj;
882 dsl_dir_t *dd;
883
884 ASSERT(lastname[0] != '@');
885
886 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
887 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
888
889 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
890
891 dsl_deleg_set_create_perms(dd, tx, cr);
892
893 dsl_dir_close(dd, FTAG);
894
895 /*
896 * If we are creating a clone, make sure we zero out any stale
897 * data from the origin snapshots zil header.
898 */
899 if (origin != NULL) {
900 dsl_dataset_t *ds;
901 objset_t *os;
902
903 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
904 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
905 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
906 dsl_dataset_dirty(ds, tx);
907 dsl_dataset_rele(ds, FTAG);
908 }
909
910 return (dsobj);
911 }
912
913 /*
914 * The snapshots must all be in the same pool.
915 */
916 int
917 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer,
918 nvlist_t *errlist)
919 {
920 int err;
921 dsl_sync_task_t *dst;
922 spa_t *spa;
923 nvpair_t *pair;
924 dsl_sync_task_group_t *dstg;
925
926 pair = nvlist_next_nvpair(snaps, NULL);
927 if (pair == NULL)
928 return (0);
929
930 err = spa_open(nvpair_name(pair), &spa, FTAG);
931 if (err)
932 return (err);
933 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
934
935 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
936 pair = nvlist_next_nvpair(snaps, pair)) {
937 dsl_dataset_t *ds;
938
939 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
940 if (err == 0) {
941 struct dsl_ds_destroyarg *dsda;
942
943 dsl_dataset_make_exclusive(ds, dstg);
944 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
945 KM_SLEEP);
946 dsda->ds = ds;
947 dsda->defer = defer;
948 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
949 dsl_dataset_destroy_sync, dsda, dstg, 0);
950 } else if (err == ENOENT) {
951 err = 0;
952 } else {
953 fnvlist_add_int32(errlist, nvpair_name(pair), err);
954 break;
955 }
956 }
957
958 if (err == 0)
959 err = dsl_sync_task_group_wait(dstg);
960
961 for (dst = list_head(&dstg->dstg_tasks); dst;
962 dst = list_next(&dstg->dstg_tasks, dst)) {
963 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
964 dsl_dataset_t *ds = dsda->ds;
965
966 /*
967 * Return the snapshots that triggered the error.
968 */
969 if (dst->dst_err != 0) {
970 char name[ZFS_MAXNAMELEN];
971 dsl_dataset_name(ds, name);
972 fnvlist_add_int32(errlist, name, dst->dst_err);
973 }
974 ASSERT3P(dsda->rm_origin, ==, NULL);
975 dsl_dataset_disown(ds, dstg);
976 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
977 }
978
979 dsl_sync_task_group_destroy(dstg);
980 spa_close(spa, FTAG);
981 return (err);
982
983 }
984
985 static boolean_t
986 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
987 {
988 boolean_t might_destroy = B_FALSE;
989
990 mutex_enter(&ds->ds_lock);
991 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
992 DS_IS_DEFER_DESTROY(ds))
993 might_destroy = B_TRUE;
994 mutex_exit(&ds->ds_lock);
995
996 return (might_destroy);
997 }
998
999 /*
1000 * If we're removing a clone, and these three conditions are true:
1001 * 1) the clone's origin has no other children
1002 * 2) the clone's origin has no user references
1003 * 3) the clone's origin has been marked for deferred destruction
1004 * Then, prepare to remove the origin as part of this sync task group.
1005 */
1006 static int
1007 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1008 {
1009 dsl_dataset_t *ds = dsda->ds;
1010 dsl_dataset_t *origin = ds->ds_prev;
1011
1012 if (dsl_dataset_might_destroy_origin(origin)) {
1013 char *name;
1014 int namelen;
1015 int error;
1016
1017 namelen = dsl_dataset_namelen(origin) + 1;
1018 name = kmem_alloc(namelen, KM_SLEEP);
1019 dsl_dataset_name(origin, name);
1020 #ifdef _KERNEL
1021 error = zfs_unmount_snap(name, NULL);
1022 if (error) {
1023 kmem_free(name, namelen);
1024 return (error);
1025 }
1026 #endif
1027 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1028 kmem_free(name, namelen);
1029 if (error)
1030 return (error);
1031 dsda->rm_origin = origin;
1032 dsl_dataset_make_exclusive(origin, tag);
1033 }
1034
1035 return (0);
1036 }
1037
1038 /*
1039 * ds must be opened as OWNER. On return (whether successful or not),
1040 * ds will be closed and caller can no longer dereference it.
1041 */
1042 int
1043 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1044 {
1045 int err;
1046 dsl_sync_task_group_t *dstg;
1047 objset_t *os;
1048 dsl_dir_t *dd;
1049 uint64_t obj;
1050 struct dsl_ds_destroyarg dsda = { 0 };
1051
1052 dsda.ds = ds;
1053
1054 if (dsl_dataset_is_snapshot(ds)) {
1055 /* Destroying a snapshot is simpler */
1056 dsl_dataset_make_exclusive(ds, tag);
1057
1058 dsda.defer = defer;
1059 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1060 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1061 &dsda, tag, 0);
1062 ASSERT3P(dsda.rm_origin, ==, NULL);
1063 goto out;
1064 } else if (defer) {
1065 err = EINVAL;
1066 goto out;
1067 }
1068
1069 dd = ds->ds_dir;
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 * If async destruction is not enabled try to remove all objects
1086 * while in the open context so that there is less work to do in
1087 * the syncing context.
1088 */
1089 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1090 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1091 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1092 ds->ds_phys->ds_prev_snap_txg)) {
1093 /*
1094 * Ignore errors, if there is not enough disk space
1095 * we will deal with it in dsl_dataset_destroy_sync().
1096 */
1097 (void) dmu_free_object(os, obj);
1098 }
1099 if (err != ESRCH)
1100 goto out;
1101 }
1102
1103 /*
1104 * Only the ZIL knows how to free log blocks.
1105 */
1106 zil_destroy(dmu_objset_zil(os), B_FALSE);
1107
1108 /*
1109 * Sync out all in-flight IO.
1110 */
1111 txg_wait_synced(dd->dd_pool, 0);
1112
1113 /*
1114 * If we managed to free all the objects in open
1115 * context, the user space accounting should be zero.
1116 */
1117 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1118 dmu_objset_userused_enabled(os)) {
1119 uint64_t count;
1120
1121 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1122 count == 0);
1123 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1124 count == 0);
1125 }
1126
1127 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1128 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1129 rw_exit(&dd->dd_pool->dp_config_rwlock);
1130
1131 if (err)
1132 goto out;
1133
1134 /*
1135 * Blow away the dsl_dir + head dataset.
1136 */
1137 dsl_dataset_make_exclusive(ds, tag);
1138 /*
1139 * If we're removing a clone, we might also need to remove its
1140 * origin.
1141 */
1142 do {
1143 dsda.need_prep = B_FALSE;
1144 if (dsl_dir_is_clone(dd)) {
1145 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1146 if (err) {
1147 dsl_dir_close(dd, FTAG);
1148 goto out;
1149 }
1150 }
1151
1152 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1153 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1154 dsl_dataset_destroy_sync, &dsda, tag, 0);
1155 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1156 dsl_dir_destroy_sync, dd, FTAG, 0);
1157 err = dsl_sync_task_group_wait(dstg);
1158 dsl_sync_task_group_destroy(dstg);
1159
1160 /*
1161 * We could be racing against 'zfs release' or 'zfs destroy -d'
1162 * on the origin snap, in which case we can get EBUSY if we
1163 * needed to destroy the origin snap but were not ready to
1164 * do so.
1165 */
1166 if (dsda.need_prep) {
1167 ASSERT(err == EBUSY);
1168 ASSERT(dsl_dir_is_clone(dd));
1169 ASSERT(dsda.rm_origin == NULL);
1170 }
1171 } while (dsda.need_prep);
1172
1173 if (dsda.rm_origin != NULL)
1174 dsl_dataset_disown(dsda.rm_origin, tag);
1175
1176 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1177 if (err)
1178 dsl_dir_close(dd, FTAG);
1179 out:
1180 dsl_dataset_disown(ds, tag);
1181 return (err);
1182 }
1183
1184 blkptr_t *
1185 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1186 {
1187 return (&ds->ds_phys->ds_bp);
1188 }
1189
1190 void
1191 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1192 {
1193 ASSERT(dmu_tx_is_syncing(tx));
1194 /* If it's the meta-objset, set dp_meta_rootbp */
1195 if (ds == NULL) {
1196 tx->tx_pool->dp_meta_rootbp = *bp;
1197 } else {
1198 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1199 ds->ds_phys->ds_bp = *bp;
1200 }
1201 }
1202
1203 spa_t *
1204 dsl_dataset_get_spa(dsl_dataset_t *ds)
1205 {
1206 return (ds->ds_dir->dd_pool->dp_spa);
1207 }
1208
1209 void
1210 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1211 {
1212 dsl_pool_t *dp;
1213
1214 if (ds == NULL) /* this is the meta-objset */
1215 return;
1216
1217 ASSERT(ds->ds_objset != NULL);
1218
1219 if (ds->ds_phys->ds_next_snap_obj != 0)
1220 panic("dirtying snapshot!");
1221
1222 dp = ds->ds_dir->dd_pool;
1223
1224 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1225 /* up the hold count until we can be written out */
1226 dmu_buf_add_ref(ds->ds_dbuf, ds);
1227 }
1228 }
1229
1230 /*
1231 * The unique space in the head dataset can be calculated by subtracting
1232 * the space used in the most recent snapshot, that is still being used
1233 * in this file system, from the space currently in use. To figure out
1234 * the space in the most recent snapshot still in use, we need to take
1235 * the total space used in the snapshot and subtract out the space that
1236 * has been freed up since the snapshot was taken.
1237 */
1238 static void
1239 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1240 {
1241 uint64_t mrs_used;
1242 uint64_t dlused, dlcomp, dluncomp;
1243
1244 ASSERT(!dsl_dataset_is_snapshot(ds));
1245
1246 if (ds->ds_phys->ds_prev_snap_obj != 0)
1247 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1248 else
1249 mrs_used = 0;
1250
1251 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1252
1253 ASSERT3U(dlused, <=, mrs_used);
1254 ds->ds_phys->ds_unique_bytes =
1255 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1256
1257 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1258 SPA_VERSION_UNIQUE_ACCURATE)
1259 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1260 }
1261
1262 struct killarg {
1263 dsl_dataset_t *ds;
1264 dmu_tx_t *tx;
1265 };
1266
1267 /* ARGSUSED */
1268 static int
1269 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1270 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1271 {
1272 struct killarg *ka = arg;
1273 dmu_tx_t *tx = ka->tx;
1274
1275 if (bp == NULL)
1276 return (0);
1277
1278 if (zb->zb_level == ZB_ZIL_LEVEL) {
1279 ASSERT(zilog != NULL);
1280 /*
1281 * It's a block in the intent log. It has no
1282 * accounting, so just free it.
1283 */
1284 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1285 } else {
1286 ASSERT(zilog == NULL);
1287 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1288 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1289 }
1290
1291 return (0);
1292 }
1293
1294 /* ARGSUSED */
1295 static int
1296 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1297 {
1298 dsl_dataset_t *ds = arg1;
1299 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1300 uint64_t count;
1301 int err;
1302
1303 /*
1304 * Can't delete a head dataset if there are snapshots of it.
1305 * (Except if the only snapshots are from the branch we cloned
1306 * from.)
1307 */
1308 if (ds->ds_prev != NULL &&
1309 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1310 return (EBUSY);
1311
1312 /*
1313 * This is really a dsl_dir thing, but check it here so that
1314 * we'll be less likely to leave this dataset inconsistent &
1315 * nearly destroyed.
1316 */
1317 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1318 if (err)
1319 return (err);
1320 if (count != 0)
1321 return (EEXIST);
1322
1323 return (0);
1324 }
1325
1326 /* ARGSUSED */
1327 static void
1328 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1329 {
1330 dsl_dataset_t *ds = arg1;
1331
1332 /* Mark it as inconsistent on-disk, in case we crash */
1333 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1334 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1335
1336 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1337 }
1338
1339 static int
1340 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1341 dmu_tx_t *tx)
1342 {
1343 dsl_dataset_t *ds = dsda->ds;
1344 dsl_dataset_t *ds_prev = ds->ds_prev;
1345
1346 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1347 struct dsl_ds_destroyarg ndsda = {0};
1348
1349 /*
1350 * If we're not prepared to remove the origin, don't remove
1351 * the clone either.
1352 */
1353 if (dsda->rm_origin == NULL) {
1354 dsda->need_prep = B_TRUE;
1355 return (EBUSY);
1356 }
1357
1358 ndsda.ds = ds_prev;
1359 ndsda.is_origin_rm = B_TRUE;
1360 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1361 }
1362
1363 /*
1364 * If we're not going to remove the origin after all,
1365 * undo the open context setup.
1366 */
1367 if (dsda->rm_origin != NULL) {
1368 dsl_dataset_disown(dsda->rm_origin, tag);
1369 dsda->rm_origin = NULL;
1370 }
1371
1372 return (0);
1373 }
1374
1375 /*
1376 * If you add new checks here, you may need to add
1377 * additional checks to the "temporary" case in
1378 * snapshot_check() in dmu_objset.c.
1379 */
1380 /* ARGSUSED */
1381 int
1382 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1383 {
1384 struct dsl_ds_destroyarg *dsda = arg1;
1385 dsl_dataset_t *ds = dsda->ds;
1386
1387 /* we have an owner hold, so noone else can destroy us */
1388 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1389
1390 /*
1391 * Only allow deferred destroy on pools that support it.
1392 * NOTE: deferred destroy is only supported on snapshots.
1393 */
1394 if (dsda->defer) {
1395 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1396 SPA_VERSION_USERREFS)
1397 return (ENOTSUP);
1398 ASSERT(dsl_dataset_is_snapshot(ds));
1399 return (0);
1400 }
1401
1402 /*
1403 * Can't delete a head dataset if there are snapshots of it.
1404 * (Except if the only snapshots are from the branch we cloned
1405 * from.)
1406 */
1407 if (ds->ds_prev != NULL &&
1408 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1409 return (EBUSY);
1410
1411 /*
1412 * If we made changes this txg, traverse_dsl_dataset won't find
1413 * them. Try again.
1414 */
1415 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1416 return (EAGAIN);
1417
1418 if (dsl_dataset_is_snapshot(ds)) {
1419 /*
1420 * If this snapshot has an elevated user reference count,
1421 * we can't destroy it yet.
1422 */
1423 if (ds->ds_userrefs > 0 && !dsda->releasing)
1424 return (EBUSY);
1425
1426 mutex_enter(&ds->ds_lock);
1427 /*
1428 * Can't delete a branch point. However, if we're destroying
1429 * a clone and removing its origin due to it having a user
1430 * hold count of 0 and having been marked for deferred destroy,
1431 * it's OK for the origin to have a single clone.
1432 */
1433 if (ds->ds_phys->ds_num_children >
1434 (dsda->is_origin_rm ? 2 : 1)) {
1435 mutex_exit(&ds->ds_lock);
1436 return (EEXIST);
1437 }
1438 mutex_exit(&ds->ds_lock);
1439 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1440 return (dsl_dataset_origin_check(dsda, arg2, tx));
1441 }
1442
1443 /* XXX we should do some i/o error checking... */
1444 return (0);
1445 }
1446
1447 struct refsarg {
1448 kmutex_t lock;
1449 boolean_t gone;
1450 kcondvar_t cv;
1451 };
1452
1453 /* ARGSUSED */
1454 static void
1455 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1456 {
1457 struct refsarg *arg = argv;
1458
1459 mutex_enter(&arg->lock);
1460 arg->gone = TRUE;
1461 cv_signal(&arg->cv);
1462 mutex_exit(&arg->lock);
1463 }
1464
1465 static void
1466 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1467 {
1468 struct refsarg arg;
1469
1470 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1471 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1472 arg.gone = FALSE;
1473 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1474 dsl_dataset_refs_gone);
1475 dmu_buf_rele(ds->ds_dbuf, tag);
1476 mutex_enter(&arg.lock);
1477 while (!arg.gone)
1478 cv_wait(&arg.cv, &arg.lock);
1479 ASSERT(arg.gone);
1480 mutex_exit(&arg.lock);
1481 ds->ds_dbuf = NULL;
1482 ds->ds_phys = NULL;
1483 mutex_destroy(&arg.lock);
1484 cv_destroy(&arg.cv);
1485 }
1486
1487 static void
1488 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1489 {
1490 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1491 uint64_t count;
1492 int err;
1493
1494 ASSERT(ds->ds_phys->ds_num_children >= 2);
1495 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1496 /*
1497 * The err should not be ENOENT, but a bug in a previous version
1498 * of the code could cause upgrade_clones_cb() to not set
1499 * ds_next_snap_obj when it should, leading to a missing entry.
1500 * If we knew that the pool was created after
1501 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1502 * ENOENT. However, at least we can check that we don't have
1503 * too many entries in the next_clones_obj even after failing to
1504 * remove this one.
1505 */
1506 if (err != ENOENT) {
1507 VERIFY3U(err, ==, 0);
1508 }
1509 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1510 &count));
1511 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1512 }
1513
1514 static void
1515 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1516 {
1517 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1518 zap_cursor_t zc;
1519 zap_attribute_t za;
1520
1521 /*
1522 * If it is the old version, dd_clones doesn't exist so we can't
1523 * find the clones, but deadlist_remove_key() is a no-op so it
1524 * doesn't matter.
1525 */
1526 if (ds->ds_dir->dd_phys->dd_clones == 0)
1527 return;
1528
1529 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1530 zap_cursor_retrieve(&zc, &za) == 0;
1531 zap_cursor_advance(&zc)) {
1532 dsl_dataset_t *clone;
1533
1534 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1535 za.za_first_integer, FTAG, &clone));
1536 if (clone->ds_dir->dd_origin_txg > mintxg) {
1537 dsl_deadlist_remove_key(&clone->ds_deadlist,
1538 mintxg, tx);
1539 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1540 }
1541 dsl_dataset_rele(clone, FTAG);
1542 }
1543 zap_cursor_fini(&zc);
1544 }
1545
1546 struct process_old_arg {
1547 dsl_dataset_t *ds;
1548 dsl_dataset_t *ds_prev;
1549 boolean_t after_branch_point;
1550 zio_t *pio;
1551 uint64_t used, comp, uncomp;
1552 };
1553
1554 static int
1555 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1556 {
1557 struct process_old_arg *poa = arg;
1558 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1559
1560 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1561 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1562 if (poa->ds_prev && !poa->after_branch_point &&
1563 bp->blk_birth >
1564 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1565 poa->ds_prev->ds_phys->ds_unique_bytes +=
1566 bp_get_dsize_sync(dp->dp_spa, bp);
1567 }
1568 } else {
1569 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1570 poa->comp += BP_GET_PSIZE(bp);
1571 poa->uncomp += BP_GET_UCSIZE(bp);
1572 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1573 }
1574 return (0);
1575 }
1576
1577 static void
1578 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1579 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1580 {
1581 struct process_old_arg poa = { 0 };
1582 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1583 objset_t *mos = dp->dp_meta_objset;
1584
1585 ASSERT(ds->ds_deadlist.dl_oldfmt);
1586 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1587
1588 poa.ds = ds;
1589 poa.ds_prev = ds_prev;
1590 poa.after_branch_point = after_branch_point;
1591 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1592 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1593 process_old_cb, &poa, tx));
1594 VERIFY3U(zio_wait(poa.pio), ==, 0);
1595 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1596
1597 /* change snapused */
1598 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1599 -poa.used, -poa.comp, -poa.uncomp, tx);
1600
1601 /* swap next's deadlist to our deadlist */
1602 dsl_deadlist_close(&ds->ds_deadlist);
1603 dsl_deadlist_close(&ds_next->ds_deadlist);
1604 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1605 ds->ds_phys->ds_deadlist_obj);
1606 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1607 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1608 ds_next->ds_phys->ds_deadlist_obj);
1609 }
1610
1611 static int
1612 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1613 {
1614 int err;
1615 struct killarg ka;
1616
1617 /*
1618 * Free everything that we point to (that's born after
1619 * the previous snapshot, if we are a clone)
1620 *
1621 * NB: this should be very quick, because we already
1622 * freed all the objects in open context.
1623 */
1624 ka.ds = ds;
1625 ka.tx = tx;
1626 err = traverse_dataset(ds,
1627 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1628 kill_blkptr, &ka);
1629 ASSERT3U(err, ==, 0);
1630 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1631
1632 return (err);
1633 }
1634
1635 void
1636 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1637 {
1638 struct dsl_ds_destroyarg *dsda = arg1;
1639 dsl_dataset_t *ds = dsda->ds;
1640 int err;
1641 int after_branch_point = FALSE;
1642 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1643 objset_t *mos = dp->dp_meta_objset;
1644 dsl_dataset_t *ds_prev = NULL;
1645 boolean_t wont_destroy;
1646 uint64_t obj;
1647
1648 wont_destroy = (dsda->defer &&
1649 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1650
1651 ASSERT(ds->ds_owner || wont_destroy);
1652 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1653 ASSERT(ds->ds_prev == NULL ||
1654 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1655 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1656
1657 if (wont_destroy) {
1658 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1659 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1660 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1661 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
1662 return;
1663 }
1664
1665 /* We need to log before removing it from the namespace. */
1666 spa_history_log_internal_ds(ds, "destroy", tx, "");
1667
1668 /* signal any waiters that this dataset is going away */
1669 mutex_enter(&ds->ds_lock);
1670 ds->ds_owner = dsl_reaper;
1671 cv_broadcast(&ds->ds_exclusive_cv);
1672 mutex_exit(&ds->ds_lock);
1673
1674 /* Remove our reservation */
1675 if (ds->ds_reserved != 0) {
1676 dsl_prop_setarg_t psa;
1677 uint64_t value = 0;
1678
1679 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1680 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1681 &value);
1682 psa.psa_effective_value = 0; /* predict default value */
1683
1684 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1685 ASSERT3U(ds->ds_reserved, ==, 0);
1686 }
1687
1688 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1689
1690 dsl_scan_ds_destroyed(ds, tx);
1691
1692 obj = ds->ds_object;
1693
1694 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1695 if (ds->ds_prev) {
1696 ds_prev = ds->ds_prev;
1697 } else {
1698 VERIFY(0 == dsl_dataset_hold_obj(dp,
1699 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1700 }
1701 after_branch_point =
1702 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1703
1704 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1705 if (after_branch_point &&
1706 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1707 remove_from_next_clones(ds_prev, obj, tx);
1708 if (ds->ds_phys->ds_next_snap_obj != 0) {
1709 VERIFY(0 == zap_add_int(mos,
1710 ds_prev->ds_phys->ds_next_clones_obj,
1711 ds->ds_phys->ds_next_snap_obj, tx));
1712 }
1713 }
1714 if (after_branch_point &&
1715 ds->ds_phys->ds_next_snap_obj == 0) {
1716 /* This clone is toast. */
1717 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1718 ds_prev->ds_phys->ds_num_children--;
1719
1720 /*
1721 * If the clone's origin has no other clones, no
1722 * user holds, and has been marked for deferred
1723 * deletion, then we should have done the necessary
1724 * destroy setup for it.
1725 */
1726 if (ds_prev->ds_phys->ds_num_children == 1 &&
1727 ds_prev->ds_userrefs == 0 &&
1728 DS_IS_DEFER_DESTROY(ds_prev)) {
1729 ASSERT3P(dsda->rm_origin, !=, NULL);
1730 } else {
1731 ASSERT3P(dsda->rm_origin, ==, NULL);
1732 }
1733 } else if (!after_branch_point) {
1734 ds_prev->ds_phys->ds_next_snap_obj =
1735 ds->ds_phys->ds_next_snap_obj;
1736 }
1737 }
1738
1739 if (dsl_dataset_is_snapshot(ds)) {
1740 dsl_dataset_t *ds_next;
1741 uint64_t old_unique;
1742 uint64_t used = 0, comp = 0, uncomp = 0;
1743
1744 VERIFY(0 == dsl_dataset_hold_obj(dp,
1745 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1746 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1747
1748 old_unique = ds_next->ds_phys->ds_unique_bytes;
1749
1750 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1751 ds_next->ds_phys->ds_prev_snap_obj =
1752 ds->ds_phys->ds_prev_snap_obj;
1753 ds_next->ds_phys->ds_prev_snap_txg =
1754 ds->ds_phys->ds_prev_snap_txg;
1755 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1756 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1757
1758
1759 if (ds_next->ds_deadlist.dl_oldfmt) {
1760 process_old_deadlist(ds, ds_prev, ds_next,
1761 after_branch_point, tx);
1762 } else {
1763 /* Adjust prev's unique space. */
1764 if (ds_prev && !after_branch_point) {
1765 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1766 ds_prev->ds_phys->ds_prev_snap_txg,
1767 ds->ds_phys->ds_prev_snap_txg,
1768 &used, &comp, &uncomp);
1769 ds_prev->ds_phys->ds_unique_bytes += used;
1770 }
1771
1772 /* Adjust snapused. */
1773 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1774 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1775 &used, &comp, &uncomp);
1776 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1777 -used, -comp, -uncomp, tx);
1778
1779 /* Move blocks to be freed to pool's free list. */
1780 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1781 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1782 tx);
1783 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1784 DD_USED_HEAD, used, comp, uncomp, tx);
1785
1786 /* Merge our deadlist into next's and free it. */
1787 dsl_deadlist_merge(&ds_next->ds_deadlist,
1788 ds->ds_phys->ds_deadlist_obj, tx);
1789 }
1790 dsl_deadlist_close(&ds->ds_deadlist);
1791 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1792
1793 /* Collapse range in clone heads */
1794 dsl_dataset_remove_clones_key(ds,
1795 ds->ds_phys->ds_creation_txg, tx);
1796
1797 if (dsl_dataset_is_snapshot(ds_next)) {
1798 dsl_dataset_t *ds_nextnext;
1799
1800 /*
1801 * Update next's unique to include blocks which
1802 * were previously shared by only this snapshot
1803 * and it. Those blocks will be born after the
1804 * prev snap and before this snap, and will have
1805 * died after the next snap and before the one
1806 * after that (ie. be on the snap after next's
1807 * deadlist).
1808 */
1809 VERIFY(0 == dsl_dataset_hold_obj(dp,
1810 ds_next->ds_phys->ds_next_snap_obj,
1811 FTAG, &ds_nextnext));
1812 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1813 ds->ds_phys->ds_prev_snap_txg,
1814 ds->ds_phys->ds_creation_txg,
1815 &used, &comp, &uncomp);
1816 ds_next->ds_phys->ds_unique_bytes += used;
1817 dsl_dataset_rele(ds_nextnext, FTAG);
1818 ASSERT3P(ds_next->ds_prev, ==, NULL);
1819
1820 /* Collapse range in this head. */
1821 dsl_dataset_t *hds;
1822 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1823 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1824 FTAG, &hds));
1825 dsl_deadlist_remove_key(&hds->ds_deadlist,
1826 ds->ds_phys->ds_creation_txg, tx);
1827 dsl_dataset_rele(hds, FTAG);
1828
1829 } else {
1830 ASSERT3P(ds_next->ds_prev, ==, ds);
1831 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1832 ds_next->ds_prev = NULL;
1833 if (ds_prev) {
1834 VERIFY(0 == dsl_dataset_get_ref(dp,
1835 ds->ds_phys->ds_prev_snap_obj,
1836 ds_next, &ds_next->ds_prev));
1837 }
1838
1839 dsl_dataset_recalc_head_uniq(ds_next);
1840
1841 /*
1842 * Reduce the amount of our unconsmed refreservation
1843 * being charged to our parent by the amount of
1844 * new unique data we have gained.
1845 */
1846 if (old_unique < ds_next->ds_reserved) {
1847 int64_t mrsdelta;
1848 uint64_t new_unique =
1849 ds_next->ds_phys->ds_unique_bytes;
1850
1851 ASSERT(old_unique <= new_unique);
1852 mrsdelta = MIN(new_unique - old_unique,
1853 ds_next->ds_reserved - old_unique);
1854 dsl_dir_diduse_space(ds->ds_dir,
1855 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1856 }
1857 }
1858 dsl_dataset_rele(ds_next, FTAG);
1859 } else {
1860 zfeature_info_t *async_destroy =
1861 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1862
1863 /*
1864 * There's no next snapshot, so this is a head dataset.
1865 * Destroy the deadlist. Unless it's a clone, the
1866 * deadlist should be empty. (If it's a clone, it's
1867 * safe to ignore the deadlist contents.)
1868 */
1869 dsl_deadlist_close(&ds->ds_deadlist);
1870 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1871 ds->ds_phys->ds_deadlist_obj = 0;
1872
1873 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1874 err = old_synchronous_dataset_destroy(ds, tx);
1875 } else {
1876 /*
1877 * Move the bptree into the pool's list of trees to
1878 * clean up and update space accounting information.
1879 */
1880 uint64_t used, comp, uncomp;
1881
1882 ASSERT(err == 0 || err == EBUSY);
1883 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1884 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1885 dp->dp_bptree_obj = bptree_alloc(
1886 dp->dp_meta_objset, tx);
1887 VERIFY(zap_add(dp->dp_meta_objset,
1888 DMU_POOL_DIRECTORY_OBJECT,
1889 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1890 &dp->dp_bptree_obj, tx) == 0);
1891 }
1892
1893 used = ds->ds_dir->dd_phys->dd_used_bytes;
1894 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1895 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1896
1897 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1898 ds->ds_phys->ds_unique_bytes == used);
1899
1900 bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
1901 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1902 used, comp, uncomp, tx);
1903 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1904 -used, -comp, -uncomp, tx);
1905 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1906 used, comp, uncomp, tx);
1907 }
1908
1909 if (ds->ds_prev != NULL) {
1910 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1911 VERIFY3U(0, ==, zap_remove_int(mos,
1912 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1913 ds->ds_object, tx));
1914 }
1915 dsl_dataset_rele(ds->ds_prev, ds);
1916 ds->ds_prev = ds_prev = NULL;
1917 }
1918 }
1919
1920 /*
1921 * This must be done after the dsl_traverse(), because it will
1922 * re-open the objset.
1923 */
1924 if (ds->ds_objset) {
1925 dmu_objset_evict(ds->ds_objset);
1926 ds->ds_objset = NULL;
1927 }
1928
1929 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1930 /* Erase the link in the dir */
1931 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1932 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1933 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1934 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1935 ASSERT(err == 0);
1936 } else {
1937 /* remove from snapshot namespace */
1938 dsl_dataset_t *ds_head;
1939 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1940 VERIFY(0 == dsl_dataset_hold_obj(dp,
1941 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1942 VERIFY(0 == dsl_dataset_get_snapname(ds));
1943 #ifdef ZFS_DEBUG
1944 {
1945 uint64_t val;
1946
1947 err = dsl_dataset_snap_lookup(ds_head,
1948 ds->ds_snapname, &val);
1949 ASSERT3U(err, ==, 0);
1950 ASSERT3U(val, ==, obj);
1951 }
1952 #endif
1953 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1954 ASSERT(err == 0);
1955 dsl_dataset_rele(ds_head, FTAG);
1956 }
1957
1958 if (ds_prev && ds->ds_prev != ds_prev)
1959 dsl_dataset_rele(ds_prev, FTAG);
1960
1961 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1962
1963 if (ds->ds_phys->ds_next_clones_obj != 0) {
1964 uint64_t count;
1965 ASSERT(0 == zap_count(mos,
1966 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1967 VERIFY(0 == dmu_object_free(mos,
1968 ds->ds_phys->ds_next_clones_obj, tx));
1969 }
1970 if (ds->ds_phys->ds_props_obj != 0)
1971 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1972 if (ds->ds_phys->ds_userrefs_obj != 0)
1973 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1974 dsl_dir_close(ds->ds_dir, ds);
1975 ds->ds_dir = NULL;
1976 dsl_dataset_drain_refs(ds, tag);
1977 VERIFY(0 == dmu_object_free(mos, obj, tx));
1978
1979 if (dsda->rm_origin) {
1980 /*
1981 * Remove the origin of the clone we just destroyed.
1982 */
1983 struct dsl_ds_destroyarg ndsda = {0};
1984
1985 ndsda.ds = dsda->rm_origin;
1986 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1987 }
1988 }
1989
1990 static int
1991 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1992 {
1993 uint64_t asize;
1994
1995 if (!dmu_tx_is_syncing(tx))
1996 return (0);
1997
1998 /*
1999 * If there's an fs-only reservation, any blocks that might become
2000 * owned by the snapshot dataset must be accommodated by space
2001 * outside of the reservation.
2002 */
2003 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2004 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2005 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2006 return (ENOSPC);
2007
2008 /*
2009 * Propagate any reserved space for this snapshot to other
2010 * snapshot checks in this sync group.
2011 */
2012 if (asize > 0)
2013 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2014
2015 return (0);
2016 }
2017
2018 int
2019 dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname,
2020 dmu_tx_t *tx)
2021 {
2022 int err;
2023 uint64_t value;
2024
2025 /*
2026 * We don't allow multiple snapshots of the same txg. If there
2027 * is already one, try again.
2028 */
2029 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2030 return (EAGAIN);
2031
2032 /*
2033 * Check for conflicting snapshot name.
2034 */
2035 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2036 if (err == 0)
2037 return (EEXIST);
2038 if (err != ENOENT)
2039 return (err);
2040
2041 /*
2042 * Check that the dataset's name is not too long. Name consists
2043 * of the dataset's length + 1 for the @-sign + snapshot name's length
2044 */
2045 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2046 return (ENAMETOOLONG);
2047
2048 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2049 if (err)
2050 return (err);
2051
2052 ds->ds_trysnap_txg = tx->tx_txg;
2053 return (0);
2054 }
2055
2056 void
2057 dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname,
2058 dmu_tx_t *tx)
2059 {
2060 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2061 dmu_buf_t *dbuf;
2062 dsl_dataset_phys_t *dsphys;
2063 uint64_t dsobj, crtxg;
2064 objset_t *mos = dp->dp_meta_objset;
2065 int err;
2066
2067 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2068
2069 /*
2070 * The origin's ds_creation_txg has to be < TXG_INITIAL
2071 */
2072 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2073 crtxg = 1;
2074 else
2075 crtxg = tx->tx_txg;
2076
2077 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2078 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2079 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2080 dmu_buf_will_dirty(dbuf, tx);
2081 dsphys = dbuf->db_data;
2082 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2083 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2084 dsphys->ds_fsid_guid = unique_create();
2085 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2086 sizeof (dsphys->ds_guid));
2087 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2088 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2089 dsphys->ds_next_snap_obj = ds->ds_object;
2090 dsphys->ds_num_children = 1;
2091 dsphys->ds_creation_time = gethrestime_sec();
2092 dsphys->ds_creation_txg = crtxg;
2093 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2094 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2095 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2096 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2097 dsphys->ds_flags = ds->ds_phys->ds_flags;
2098 dsphys->ds_bp = ds->ds_phys->ds_bp;
2099 dmu_buf_rele(dbuf, FTAG);
2100
2101 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2102 if (ds->ds_prev) {
2103 uint64_t next_clones_obj =
2104 ds->ds_prev->ds_phys->ds_next_clones_obj;
2105 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2106 ds->ds_object ||
2107 ds->ds_prev->ds_phys->ds_num_children > 1);
2108 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2109 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2110 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2111 ds->ds_prev->ds_phys->ds_creation_txg);
2112 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2113 } else if (next_clones_obj != 0) {
2114 remove_from_next_clones(ds->ds_prev,
2115 dsphys->ds_next_snap_obj, tx);
2116 VERIFY3U(0, ==, zap_add_int(mos,
2117 next_clones_obj, dsobj, tx));
2118 }
2119 }
2120
2121 /*
2122 * If we have a reference-reservation on this dataset, we will
2123 * need to increase the amount of refreservation being charged
2124 * since our unique space is going to zero.
2125 */
2126 if (ds->ds_reserved) {
2127 int64_t delta;
2128 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2129 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2130 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2131 delta, 0, 0, tx);
2132 }
2133
2134 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2135 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2136 ds->ds_dir->dd_myname, snapname, dsobj,
2137 ds->ds_phys->ds_prev_snap_txg);
2138 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2139 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2140 dsl_deadlist_close(&ds->ds_deadlist);
2141 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2142 dsl_deadlist_add_key(&ds->ds_deadlist,
2143 ds->ds_phys->ds_prev_snap_txg, tx);
2144
2145 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2146 ds->ds_phys->ds_prev_snap_obj = dsobj;
2147 ds->ds_phys->ds_prev_snap_txg = crtxg;
2148 ds->ds_phys->ds_unique_bytes = 0;
2149 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2150 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2151
2152 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2153 snapname, 8, 1, &dsobj, tx);
2154 ASSERT(err == 0);
2155
2156 if (ds->ds_prev)
2157 dsl_dataset_drop_ref(ds->ds_prev, ds);
2158 VERIFY(0 == dsl_dataset_get_ref(dp,
2159 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2160
2161 dsl_scan_ds_snapshotted(ds, tx);
2162
2163 dsl_dir_snap_cmtime_update(ds->ds_dir);
2164
2165 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
2166 }
2167
2168 void
2169 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2170 {
2171 ASSERT(dmu_tx_is_syncing(tx));
2172 ASSERT(ds->ds_objset != NULL);
2173 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2174
2175 /*
2176 * in case we had to change ds_fsid_guid when we opened it,
2177 * sync it out now.
2178 */
2179 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2180 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2181
2182 dsl_dir_dirty(ds->ds_dir, tx);
2183 dmu_objset_sync(ds->ds_objset, zio, tx);
2184 }
2185
2186 static void
2187 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2188 {
2189 uint64_t count = 0;
2190 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2191 zap_cursor_t zc;
2192 zap_attribute_t za;
2193 nvlist_t *propval;
2194 nvlist_t *val;
2195
2196 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2197 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2198 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2199
2200 /*
2201 * There may me missing entries in ds_next_clones_obj
2202 * due to a bug in a previous version of the code.
2203 * Only trust it if it has the right number of entries.
2204 */
2205 if (ds->ds_phys->ds_next_clones_obj != 0) {
2206 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2207 &count));
2208 }
2209 if (count != ds->ds_phys->ds_num_children - 1) {
2210 goto fail;
2211 }
2212 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2213 zap_cursor_retrieve(&zc, &za) == 0;
2214 zap_cursor_advance(&zc)) {
2215 dsl_dataset_t *clone;
2216 char buf[ZFS_MAXNAMELEN];
2217 /*
2218 * Even though we hold the dp_config_rwlock, the dataset
2219 * may fail to open, returning ENOENT. If there is a
2220 * thread concurrently attempting to destroy this
2221 * dataset, it will have the ds_rwlock held for
2222 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2223 * dsl_dataset_hold_ref() will fail its
2224 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2225 * dp_config_rwlock, and wait for the destroy progress
2226 * and signal ds_exclusive_cv. If the destroy was
2227 * successful, we will see that
2228 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2229 */
2230 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2231 za.za_first_integer, FTAG, &clone) != 0)
2232 continue;
2233 dsl_dir_name(clone->ds_dir, buf);
2234 VERIFY(nvlist_add_boolean(val, buf) == 0);
2235 dsl_dataset_rele(clone, FTAG);
2236 }
2237 zap_cursor_fini(&zc);
2238 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2239 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2240 propval) == 0);
2241 fail:
2242 nvlist_free(val);
2243 nvlist_free(propval);
2244 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2245 }
2246
2247 void
2248 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2249 {
2250 uint64_t refd, avail, uobjs, aobjs, ratio;
2251
2252 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2253 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2254 ds->ds_phys->ds_compressed_bytes);
2255
2256 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2257
2258 if (dsl_dataset_is_snapshot(ds)) {
2259 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2260 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2261 ds->ds_phys->ds_unique_bytes);
2262 get_clones_stat(ds, nv);
2263 } else {
2264 dsl_dir_stats(ds->ds_dir, nv);
2265 }
2266
2267 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2268 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2269 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2270
2271 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2272 ds->ds_phys->ds_creation_time);
2273 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2274 ds->ds_phys->ds_creation_txg);
2275 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2276 ds->ds_quota);
2277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2278 ds->ds_reserved);
2279 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2280 ds->ds_phys->ds_guid);
2281 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2282 ds->ds_phys->ds_unique_bytes);
2283 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2284 ds->ds_object);
2285 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2286 ds->ds_userrefs);
2287 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2288 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2289
2290 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2291 uint64_t written, comp, uncomp;
2292 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2293 dsl_dataset_t *prev;
2294
2295 rw_enter(&dp->dp_config_rwlock, RW_READER);
2296 int err = dsl_dataset_hold_obj(dp,
2297 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2298 rw_exit(&dp->dp_config_rwlock);
2299 if (err == 0) {
2300 err = dsl_dataset_space_written(prev, ds, &written,
2301 &comp, &uncomp);
2302 dsl_dataset_rele(prev, FTAG);
2303 if (err == 0) {
2304 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2305 written);
2306 }
2307 }
2308 }
2309
2310 }
2311
2312 void
2313 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2314 {
2315 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2316 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2317 stat->dds_guid = ds->ds_phys->ds_guid;
2318 stat->dds_origin[0] = '\0';
2319 if (dsl_dataset_is_snapshot(ds)) {
2320 stat->dds_is_snapshot = B_TRUE;
2321 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2322 } else {
2323 stat->dds_is_snapshot = B_FALSE;
2324 stat->dds_num_clones = 0;
2325
2326 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2327 if (dsl_dir_is_clone(ds->ds_dir)) {
2328 dsl_dataset_t *ods;
2329
2330 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2331 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2332 dsl_dataset_name(ods, stat->dds_origin);
2333 dsl_dataset_drop_ref(ods, FTAG);
2334 }
2335 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2336 }
2337 }
2338
2339 uint64_t
2340 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2341 {
2342 return (ds->ds_fsid_guid);
2343 }
2344
2345 void
2346 dsl_dataset_space(dsl_dataset_t *ds,
2347 uint64_t *refdbytesp, uint64_t *availbytesp,
2348 uint64_t *usedobjsp, uint64_t *availobjsp)
2349 {
2350 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2351 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2352 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2353 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2354 if (ds->ds_quota != 0) {
2355 /*
2356 * Adjust available bytes according to refquota
2357 */
2358 if (*refdbytesp < ds->ds_quota)
2359 *availbytesp = MIN(*availbytesp,
2360 ds->ds_quota - *refdbytesp);
2361 else
2362 *availbytesp = 0;
2363 }
2364 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2365 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2366 }
2367
2368 boolean_t
2369 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2370 {
2371 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2372
2373 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2374 dsl_pool_sync_context(dp));
2375 if (ds->ds_prev == NULL)
2376 return (B_FALSE);
2377 if (ds->ds_phys->ds_bp.blk_birth >
2378 ds->ds_prev->ds_phys->ds_creation_txg) {
2379 objset_t *os, *os_prev;
2380 /*
2381 * It may be that only the ZIL differs, because it was
2382 * reset in the head. Don't count that as being
2383 * modified.
2384 */
2385 if (dmu_objset_from_ds(ds, &os) != 0)
2386 return (B_TRUE);
2387 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2388 return (B_TRUE);
2389 return (bcmp(&os->os_phys->os_meta_dnode,
2390 &os_prev->os_phys->os_meta_dnode,
2391 sizeof (os->os_phys->os_meta_dnode)) != 0);
2392 }
2393 return (B_FALSE);
2394 }
2395
2396 /* ARGSUSED */
2397 static int
2398 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2399 {
2400 dsl_dataset_t *ds = arg1;
2401 char *newsnapname = arg2;
2402 dsl_dir_t *dd = ds->ds_dir;
2403 dsl_dataset_t *hds;
2404 uint64_t val;
2405 int err;
2406
2407 err = dsl_dataset_hold_obj(dd->dd_pool,
2408 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2409 if (err)
2410 return (err);
2411
2412 /* new name better not be in use */
2413 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2414 dsl_dataset_rele(hds, FTAG);
2415
2416 if (err == 0)
2417 err = EEXIST;
2418 else if (err == ENOENT)
2419 err = 0;
2420
2421 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2422 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2423 err = ENAMETOOLONG;
2424
2425 return (err);
2426 }
2427
2428 static void
2429 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2430 {
2431 dsl_dataset_t *ds = arg1;
2432 const char *newsnapname = arg2;
2433 dsl_dir_t *dd = ds->ds_dir;
2434 objset_t *mos = dd->dd_pool->dp_meta_objset;
2435 dsl_dataset_t *hds;
2436 int err;
2437
2438 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2439
2440 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2441 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2442
2443 VERIFY(0 == dsl_dataset_get_snapname(ds));
2444 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2445 ASSERT3U(err, ==, 0);
2446 mutex_enter(&ds->ds_lock);
2447 (void) strcpy(ds->ds_snapname, newsnapname);
2448 mutex_exit(&ds->ds_lock);
2449 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2450 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2451 ASSERT3U(err, ==, 0);
2452
2453 spa_history_log_internal_ds(ds, "rename", tx,
2454 "-> @%s", newsnapname);
2455 dsl_dataset_rele(hds, FTAG);
2456 }
2457
2458 struct renamesnaparg {
2459 dsl_sync_task_group_t *dstg;
2460 char failed[MAXPATHLEN];
2461 char *oldsnap;
2462 char *newsnap;
2463 };
2464
2465 static int
2466 dsl_snapshot_rename_one(const char *name, void *arg)
2467 {
2468 struct renamesnaparg *ra = arg;
2469 dsl_dataset_t *ds = NULL;
2470 char *snapname;
2471 int err;
2472
2473 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2474 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2475
2476 /*
2477 * For recursive snapshot renames the parent won't be changing
2478 * so we just pass name for both the to/from argument.
2479 */
2480 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2481 if (err != 0) {
2482 strfree(snapname);
2483 return (err == ENOENT ? 0 : err);
2484 }
2485
2486 #ifdef _KERNEL
2487 /*
2488 * For all filesystems undergoing rename, we'll need to unmount it.
2489 */
2490 (void) zfs_unmount_snap(snapname, NULL);
2491 #endif
2492 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2493 strfree(snapname);
2494 if (err != 0)
2495 return (err == ENOENT ? 0 : err);
2496
2497 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2498 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2499
2500 return (0);
2501 }
2502
2503 static int
2504 dsl_recursive_rename(char *oldname, const char *newname)
2505 {
2506 int err;
2507 struct renamesnaparg *ra;
2508 dsl_sync_task_t *dst;
2509 spa_t *spa;
2510 char *cp, *fsname = spa_strdup(oldname);
2511 int len = strlen(oldname) + 1;
2512
2513 /* truncate the snapshot name to get the fsname */
2514 cp = strchr(fsname, '@');
2515 *cp = '\0';
2516
2517 err = spa_open(fsname, &spa, FTAG);
2518 if (err) {
2519 kmem_free(fsname, len);
2520 return (err);
2521 }
2522 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2523 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2524
2525 ra->oldsnap = strchr(oldname, '@') + 1;
2526 ra->newsnap = strchr(newname, '@') + 1;
2527 *ra->failed = '\0';
2528
2529 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2530 DS_FIND_CHILDREN);
2531 kmem_free(fsname, len);
2532
2533 if (err == 0) {
2534 err = dsl_sync_task_group_wait(ra->dstg);
2535 }
2536
2537 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2538 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2539 dsl_dataset_t *ds = dst->dst_arg1;
2540 if (dst->dst_err) {
2541 dsl_dir_name(ds->ds_dir, ra->failed);
2542 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2543 (void) strlcat(ra->failed, ra->newsnap,
2544 sizeof (ra->failed));
2545 }
2546 dsl_dataset_rele(ds, ra->dstg);
2547 }
2548
2549 if (err)
2550 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2551
2552 dsl_sync_task_group_destroy(ra->dstg);
2553 kmem_free(ra, sizeof (struct renamesnaparg));
2554 spa_close(spa, FTAG);
2555 return (err);
2556 }
2557
2558 static int
2559 dsl_valid_rename(const char *oldname, void *arg)
2560 {
2561 int delta = *(int *)arg;
2562
2563 if (strlen(oldname) + delta >= MAXNAMELEN)
2564 return (ENAMETOOLONG);
2565
2566 return (0);
2567 }
2568
2569 #pragma weak dmu_objset_rename = dsl_dataset_rename
2570 int
2571 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2572 {
2573 dsl_dir_t *dd;
2574 dsl_dataset_t *ds;
2575 const char *tail;
2576 int err;
2577
2578 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2579 if (err)
2580 return (err);
2581
2582 if (tail == NULL) {
2583 int delta = strlen(newname) - strlen(oldname);
2584
2585 /* if we're growing, validate child name lengths */
2586 if (delta > 0)
2587 err = dmu_objset_find(oldname, dsl_valid_rename,
2588 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2589
2590 if (err == 0)
2591 err = dsl_dir_rename(dd, newname);
2592 dsl_dir_close(dd, FTAG);
2593 return (err);
2594 }
2595
2596 if (tail[0] != '@') {
2597 /* the name ended in a nonexistent component */
2598 dsl_dir_close(dd, FTAG);
2599 return (ENOENT);
2600 }
2601
2602 dsl_dir_close(dd, FTAG);
2603
2604 /* new name must be snapshot in same filesystem */
2605 tail = strchr(newname, '@');
2606 if (tail == NULL)
2607 return (EINVAL);
2608 tail++;
2609 if (strncmp(oldname, newname, tail - newname) != 0)
2610 return (EXDEV);
2611
2612 if (recursive) {
2613 err = dsl_recursive_rename(oldname, newname);
2614 } else {
2615 err = dsl_dataset_hold(oldname, FTAG, &ds);
2616 if (err)
2617 return (err);
2618
2619 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2620 dsl_dataset_snapshot_rename_check,
2621 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2622
2623 dsl_dataset_rele(ds, FTAG);
2624 }
2625
2626 return (err);
2627 }
2628
2629 struct promotenode {
2630 list_node_t link;
2631 dsl_dataset_t *ds;
2632 };
2633
2634 struct promotearg {
2635 list_t shared_snaps, origin_snaps, clone_snaps;
2636 dsl_dataset_t *origin_origin;
2637 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2638 char *err_ds;
2639 };
2640
2641 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2642 static boolean_t snaplist_unstable(list_t *l);
2643
2644 static int
2645 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2646 {
2647 dsl_dataset_t *hds = arg1;
2648 struct promotearg *pa = arg2;
2649 struct promotenode *snap = list_head(&pa->shared_snaps);
2650 dsl_dataset_t *origin_ds = snap->ds;
2651 int err;
2652 uint64_t unused;
2653
2654 /* Check that it is a real clone */
2655 if (!dsl_dir_is_clone(hds->ds_dir))
2656 return (EINVAL);
2657
2658 /* Since this is so expensive, don't do the preliminary check */
2659 if (!dmu_tx_is_syncing(tx))
2660 return (0);
2661
2662 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2663 return (EXDEV);
2664
2665 /* compute origin's new unique space */
2666 snap = list_tail(&pa->clone_snaps);
2667 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2668 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2669 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2670 &pa->unique, &unused, &unused);
2671
2672 /*
2673 * Walk the snapshots that we are moving
2674 *
2675 * Compute space to transfer. Consider the incremental changes
2676 * to used for each snapshot:
2677 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2678 * So each snapshot gave birth to:
2679 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2680 * So a sequence would look like:
2681 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2682 * Which simplifies to:
2683 * uN + kN + kN-1 + ... + k1 + k0
2684 * Note however, if we stop before we reach the ORIGIN we get:
2685 * uN + kN + kN-1 + ... + kM - uM-1
2686 */
2687 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2688 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2689 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2690 for (snap = list_head(&pa->shared_snaps); snap;
2691 snap = list_next(&pa->shared_snaps, snap)) {
2692 uint64_t val, dlused, dlcomp, dluncomp;
2693 dsl_dataset_t *ds = snap->ds;
2694
2695 /* Check that the snapshot name does not conflict */
2696 VERIFY(0 == dsl_dataset_get_snapname(ds));
2697 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2698 if (err == 0) {
2699 err = EEXIST;
2700 goto out;
2701 }
2702 if (err != ENOENT)
2703 goto out;
2704
2705 /* The very first snapshot does not have a deadlist */
2706 if (ds->ds_phys->ds_prev_snap_obj == 0)
2707 continue;
2708
2709 dsl_deadlist_space(&ds->ds_deadlist,
2710 &dlused, &dlcomp, &dluncomp);
2711 pa->used += dlused;
2712 pa->comp += dlcomp;
2713 pa->uncomp += dluncomp;
2714 }
2715
2716 /*
2717 * If we are a clone of a clone then we never reached ORIGIN,
2718 * so we need to subtract out the clone origin's used space.
2719 */
2720 if (pa->origin_origin) {
2721 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2722 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2723 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2724 }
2725
2726 /* Check that there is enough space here */
2727 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2728 pa->used);
2729 if (err)
2730 return (err);
2731
2732 /*
2733 * Compute the amounts of space that will be used by snapshots
2734 * after the promotion (for both origin and clone). For each,
2735 * it is the amount of space that will be on all of their
2736 * deadlists (that was not born before their new origin).
2737 */
2738 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2739 uint64_t space;
2740
2741 /*
2742 * Note, typically this will not be a clone of a clone,
2743 * so dd_origin_txg will be < TXG_INITIAL, so
2744 * these snaplist_space() -> dsl_deadlist_space_range()
2745 * calls will be fast because they do not have to
2746 * iterate over all bps.
2747 */
2748 snap = list_head(&pa->origin_snaps);
2749 err = snaplist_space(&pa->shared_snaps,
2750 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2751 if (err)
2752 return (err);
2753
2754 err = snaplist_space(&pa->clone_snaps,
2755 snap->ds->ds_dir->dd_origin_txg, &space);
2756 if (err)
2757 return (err);
2758 pa->cloneusedsnap += space;
2759 }
2760 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2761 err = snaplist_space(&pa->origin_snaps,
2762 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2763 if (err)
2764 return (err);
2765 }
2766
2767 return (0);
2768 out:
2769 pa->err_ds = snap->ds->ds_snapname;
2770 return (err);
2771 }
2772
2773 static void
2774 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2775 {
2776 dsl_dataset_t *hds = arg1;
2777 struct promotearg *pa = arg2;
2778 struct promotenode *snap = list_head(&pa->shared_snaps);
2779 dsl_dataset_t *origin_ds = snap->ds;
2780 dsl_dataset_t *origin_head;
2781 dsl_dir_t *dd = hds->ds_dir;
2782 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2783 dsl_dir_t *odd = NULL;
2784 uint64_t oldnext_obj;
2785 int64_t delta;
2786
2787 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2788
2789 snap = list_head(&pa->origin_snaps);
2790 origin_head = snap->ds;
2791
2792 /*
2793 * We need to explicitly open odd, since origin_ds's dd will be
2794 * changing.
2795 */
2796 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2797 NULL, FTAG, &odd));
2798
2799 /* change origin's next snap */
2800 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2801 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2802 snap = list_tail(&pa->clone_snaps);
2803 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2804 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2805
2806 /* change the origin's next clone */
2807 if (origin_ds->ds_phys->ds_next_clones_obj) {
2808 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2809 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2810 origin_ds->ds_phys->ds_next_clones_obj,
2811 oldnext_obj, tx));
2812 }
2813
2814 /* change origin */
2815 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2816 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2817 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2818 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2819 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2820 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2821 origin_head->ds_dir->dd_origin_txg =
2822 origin_ds->ds_phys->ds_creation_txg;
2823
2824 /* change dd_clone entries */
2825 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2826 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2827 odd->dd_phys->dd_clones, hds->ds_object, tx));
2828 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2829 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2830 hds->ds_object, tx));
2831
2832 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2833 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2834 origin_head->ds_object, tx));
2835 if (dd->dd_phys->dd_clones == 0) {
2836 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2837 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2838 }
2839 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2840 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2841
2842 }
2843
2844 /* move snapshots to this dir */
2845 for (snap = list_head(&pa->shared_snaps); snap;
2846 snap = list_next(&pa->shared_snaps, snap)) {
2847 dsl_dataset_t *ds = snap->ds;
2848
2849 /* unregister props as dsl_dir is changing */
2850 if (ds->ds_objset) {
2851 dmu_objset_evict(ds->ds_objset);
2852 ds->ds_objset = NULL;
2853 }
2854 /* move snap name entry */
2855 VERIFY(0 == dsl_dataset_get_snapname(ds));
2856 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2857 ds->ds_snapname, tx));
2858 VERIFY(0 == zap_add(dp->dp_meta_objset,
2859 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2860 8, 1, &ds->ds_object, tx));
2861
2862 /* change containing dsl_dir */
2863 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2864 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2865 ds->ds_phys->ds_dir_obj = dd->dd_object;
2866 ASSERT3P(ds->ds_dir, ==, odd);
2867 dsl_dir_close(ds->ds_dir, ds);
2868 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2869 NULL, ds, &ds->ds_dir));
2870
2871 /* move any clone references */
2872 if (ds->ds_phys->ds_next_clones_obj &&
2873 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2874 zap_cursor_t zc;
2875 zap_attribute_t za;
2876
2877 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2878 ds->ds_phys->ds_next_clones_obj);
2879 zap_cursor_retrieve(&zc, &za) == 0;
2880 zap_cursor_advance(&zc)) {
2881 dsl_dataset_t *cnds;
2882 uint64_t o;
2883
2884 if (za.za_first_integer == oldnext_obj) {
2885 /*
2886 * We've already moved the
2887 * origin's reference.
2888 */
2889 continue;
2890 }
2891
2892 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2893 za.za_first_integer, FTAG, &cnds));
2894 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2895
2896 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2897 odd->dd_phys->dd_clones, o, tx), ==, 0);
2898 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2899 dd->dd_phys->dd_clones, o, tx), ==, 0);
2900 dsl_dataset_rele(cnds, FTAG);
2901 }
2902 zap_cursor_fini(&zc);
2903 }
2904
2905 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2906 }
2907
2908 /*
2909 * Change space accounting.
2910 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2911 * both be valid, or both be 0 (resulting in delta == 0). This
2912 * is true for each of {clone,origin} independently.
2913 */
2914
2915 delta = pa->cloneusedsnap -
2916 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2917 ASSERT3S(delta, >=, 0);
2918 ASSERT3U(pa->used, >=, delta);
2919 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2920 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2921 pa->used - delta, pa->comp, pa->uncomp, tx);
2922
2923 delta = pa->originusedsnap -
2924 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2925 ASSERT3S(delta, <=, 0);
2926 ASSERT3U(pa->used, >=, -delta);
2927 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2928 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2929 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2930
2931 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2932
2933 /* log history record */
2934 spa_history_log_internal_ds(hds, "promote", tx, "");
2935
2936 dsl_dir_close(odd, FTAG);
2937 }
2938
2939 static char *snaplist_tag = "snaplist";
2940 /*
2941 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2942 * (exclusive) and last_obj (inclusive). The list will be in reverse
2943 * order (last_obj will be the list_head()). If first_obj == 0, do all
2944 * snapshots back to this dataset's origin.
2945 */
2946 static int
2947 snaplist_make(dsl_pool_t *dp, boolean_t own,
2948 uint64_t first_obj, uint64_t last_obj, list_t *l)
2949 {
2950 uint64_t obj = last_obj;
2951
2952 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2953
2954 list_create(l, sizeof (struct promotenode),
2955 offsetof(struct promotenode, link));
2956
2957 while (obj != first_obj) {
2958 dsl_dataset_t *ds;
2959 struct promotenode *snap;
2960 int err;
2961
2962 if (own) {
2963 err = dsl_dataset_own_obj(dp, obj,
2964 0, snaplist_tag, &ds);
2965 if (err == 0)
2966 dsl_dataset_make_exclusive(ds, snaplist_tag);
2967 } else {
2968 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2969 }
2970 if (err == ENOENT) {
2971 /* lost race with snapshot destroy */
2972 struct promotenode *last = list_tail(l);
2973 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2974 obj = last->ds->ds_phys->ds_prev_snap_obj;
2975 continue;
2976 } else if (err) {
2977 return (err);
2978 }
2979
2980 if (first_obj == 0)
2981 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2982
2983 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2984 snap->ds = ds;
2985 list_insert_tail(l, snap);
2986 obj = ds->ds_phys->ds_prev_snap_obj;
2987 }
2988
2989 return (0);
2990 }
2991
2992 static int
2993 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2994 {
2995 struct promotenode *snap;
2996
2997 *spacep = 0;
2998 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2999 uint64_t used, comp, uncomp;
3000 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3001 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3002 *spacep += used;
3003 }
3004 return (0);
3005 }
3006
3007 static void
3008 snaplist_destroy(list_t *l, boolean_t own)
3009 {
3010 struct promotenode *snap;
3011
3012 if (!l || !list_link_active(&l->list_head))
3013 return;
3014
3015 while ((snap = list_tail(l)) != NULL) {
3016 list_remove(l, snap);
3017 if (own)
3018 dsl_dataset_disown(snap->ds, snaplist_tag);
3019 else
3020 dsl_dataset_rele(snap->ds, snaplist_tag);
3021 kmem_free(snap, sizeof (struct promotenode));
3022 }
3023 list_destroy(l);
3024 }
3025
3026 /*
3027 * Promote a clone. Nomenclature note:
3028 * "clone" or "cds": the original clone which is being promoted
3029 * "origin" or "ods": the snapshot which is originally clone's origin
3030 * "origin head" or "ohds": the dataset which is the head
3031 * (filesystem/volume) for the origin
3032 * "origin origin": the origin of the origin's filesystem (typically
3033 * NULL, indicating that the clone is not a clone of a clone).
3034 */
3035 int
3036 dsl_dataset_promote(const char *name, char *conflsnap)
3037 {
3038 dsl_dataset_t *ds;
3039 dsl_dir_t *dd;
3040 dsl_pool_t *dp;
3041 dmu_object_info_t doi;
3042 struct promotearg pa = { 0 };
3043 struct promotenode *snap;
3044 int err;
3045
3046 err = dsl_dataset_hold(name, FTAG, &ds);
3047 if (err)
3048 return (err);
3049 dd = ds->ds_dir;
3050 dp = dd->dd_pool;
3051
3052 err = dmu_object_info(dp->dp_meta_objset,
3053 ds->ds_phys->ds_snapnames_zapobj, &doi);
3054 if (err) {
3055 dsl_dataset_rele(ds, FTAG);
3056 return (err);
3057 }
3058
3059 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3060 dsl_dataset_rele(ds, FTAG);
3061 return (EINVAL);
3062 }
3063
3064 /*
3065 * We are going to inherit all the snapshots taken before our
3066 * origin (i.e., our new origin will be our parent's origin).
3067 * Take ownership of them so that we can rename them into our
3068 * namespace.
3069 */
3070 rw_enter(&dp->dp_config_rwlock, RW_READER);
3071
3072 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3073 &pa.shared_snaps);
3074 if (err != 0)
3075 goto out;
3076
3077 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3078 if (err != 0)
3079 goto out;
3080
3081 snap = list_head(&pa.shared_snaps);
3082 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3083 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3084 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3085 if (err != 0)
3086 goto out;
3087
3088 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3089 err = dsl_dataset_hold_obj(dp,
3090 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3091 FTAG, &pa.origin_origin);
3092 if (err != 0)
3093 goto out;
3094 }
3095
3096 out:
3097 rw_exit(&dp->dp_config_rwlock);
3098
3099 /*
3100 * Add in 128x the snapnames zapobj size, since we will be moving
3101 * a bunch of snapnames to the promoted ds, and dirtying their
3102 * bonus buffers.
3103 */
3104 if (err == 0) {
3105 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3106 dsl_dataset_promote_sync, ds, &pa,
3107 2 + 2 * doi.doi_physical_blocks_512);
3108 if (err && pa.err_ds && conflsnap)
3109 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3110 }
3111
3112 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3113 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3114 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3115 if (pa.origin_origin)
3116 dsl_dataset_rele(pa.origin_origin, FTAG);
3117 dsl_dataset_rele(ds, FTAG);
3118 return (err);
3119 }
3120
3121 struct cloneswaparg {
3122 dsl_dataset_t *cds; /* clone dataset */
3123 dsl_dataset_t *ohds; /* origin's head dataset */
3124 boolean_t force;
3125 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3126 };
3127
3128 /* ARGSUSED */
3129 static int
3130 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3131 {
3132 struct cloneswaparg *csa = arg1;
3133
3134 /* they should both be heads */
3135 if (dsl_dataset_is_snapshot(csa->cds) ||
3136 dsl_dataset_is_snapshot(csa->ohds))
3137 return (EINVAL);
3138
3139 /* the branch point should be just before them */
3140 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3141 return (EINVAL);
3142
3143 /* cds should be the clone (unless they are unrelated) */
3144 if (csa->cds->ds_prev != NULL &&
3145 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3146 csa->ohds->ds_object !=
3147 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3148 return (EINVAL);
3149
3150 /* the clone should be a child of the origin */
3151 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3152 return (EINVAL);
3153
3154 /* ohds shouldn't be modified unless 'force' */
3155 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3156 return (ETXTBSY);
3157
3158 /* adjust amount of any unconsumed refreservation */
3159 csa->unused_refres_delta =
3160 (int64_t)MIN(csa->ohds->ds_reserved,
3161 csa->ohds->ds_phys->ds_unique_bytes) -
3162 (int64_t)MIN(csa->ohds->ds_reserved,
3163 csa->cds->ds_phys->ds_unique_bytes);
3164
3165 if (csa->unused_refres_delta > 0 &&
3166 csa->unused_refres_delta >
3167 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3168 return (ENOSPC);
3169
3170 if (csa->ohds->ds_quota != 0 &&
3171 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3172 return (EDQUOT);
3173
3174 return (0);
3175 }
3176
3177 /* ARGSUSED */
3178 static void
3179 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3180 {
3181 struct cloneswaparg *csa = arg1;
3182 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3183
3184 ASSERT(csa->cds->ds_reserved == 0);
3185 ASSERT(csa->ohds->ds_quota == 0 ||
3186 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3187
3188 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3189 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3190
3191 if (csa->cds->ds_objset != NULL) {
3192 dmu_objset_evict(csa->cds->ds_objset);
3193 csa->cds->ds_objset = NULL;
3194 }
3195
3196 if (csa->ohds->ds_objset != NULL) {
3197 dmu_objset_evict(csa->ohds->ds_objset);
3198 csa->ohds->ds_objset = NULL;
3199 }
3200
3201 /*
3202 * Reset origin's unique bytes, if it exists.
3203 */
3204 if (csa->cds->ds_prev) {
3205 dsl_dataset_t *origin = csa->cds->ds_prev;
3206 uint64_t comp, uncomp;
3207
3208 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3209 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3210 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3211 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3212 }
3213
3214 /* swap blkptrs */
3215 {
3216 blkptr_t tmp;
3217 tmp = csa->ohds->ds_phys->ds_bp;
3218 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3219 csa->cds->ds_phys->ds_bp = tmp;
3220 }
3221
3222 /* set dd_*_bytes */
3223 {
3224 int64_t dused, dcomp, duncomp;
3225 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3226 uint64_t odl_used, odl_comp, odl_uncomp;
3227
3228 ASSERT3U(csa->cds->ds_dir->dd_phys->
3229 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3230
3231 dsl_deadlist_space(&csa->cds->ds_deadlist,
3232 &cdl_used, &cdl_comp, &cdl_uncomp);
3233 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3234 &odl_used, &odl_comp, &odl_uncomp);
3235
3236 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3237 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3238 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3239 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3240 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3241 cdl_uncomp -
3242 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3243
3244 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3245 dused, dcomp, duncomp, tx);
3246 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3247 -dused, -dcomp, -duncomp, tx);
3248
3249 /*
3250 * The difference in the space used by snapshots is the
3251 * difference in snapshot space due to the head's
3252 * deadlist (since that's the only thing that's
3253 * changing that affects the snapused).
3254 */
3255 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3256 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3257 &cdl_used, &cdl_comp, &cdl_uncomp);
3258 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3259 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3260 &odl_used, &odl_comp, &odl_uncomp);
3261 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3262 DD_USED_HEAD, DD_USED_SNAP, tx);
3263 }
3264
3265 /* swap ds_*_bytes */
3266 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3267 csa->cds->ds_phys->ds_referenced_bytes);
3268 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3269 csa->cds->ds_phys->ds_compressed_bytes);
3270 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3271 csa->cds->ds_phys->ds_uncompressed_bytes);
3272 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3273 csa->cds->ds_phys->ds_unique_bytes);
3274
3275 /* apply any parent delta for change in unconsumed refreservation */
3276 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3277 csa->unused_refres_delta, 0, 0, tx);
3278
3279 /*
3280 * Swap deadlists.
3281 */
3282 dsl_deadlist_close(&csa->cds->ds_deadlist);
3283 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3284 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3285 csa->cds->ds_phys->ds_deadlist_obj);
3286 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3287 csa->cds->ds_phys->ds_deadlist_obj);
3288 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3289 csa->ohds->ds_phys->ds_deadlist_obj);
3290
3291 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3292
3293 spa_history_log_internal_ds(csa->cds, "clone swap", tx,
3294 "parent=%s", csa->ohds->ds_dir->dd_myname);
3295 }
3296
3297 /*
3298 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3299 * recv" into an existing fs to swizzle the file system to the new
3300 * version, and by "zfs rollback". Can also be used to swap two
3301 * independent head datasets if neither has any snapshots.
3302 */
3303 int
3304 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3305 boolean_t force)
3306 {
3307 struct cloneswaparg csa;
3308 int error;
3309
3310 ASSERT(clone->ds_owner);
3311 ASSERT(origin_head->ds_owner);
3312 retry:
3313 /*
3314 * Need exclusive access for the swap. If we're swapping these
3315 * datasets back after an error, we already hold the locks.
3316 */
3317 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3318 rw_enter(&clone->ds_rwlock, RW_WRITER);
3319 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3320 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3321 rw_exit(&clone->ds_rwlock);
3322 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3323 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3324 rw_exit(&origin_head->ds_rwlock);
3325 goto retry;
3326 }
3327 }
3328 csa.cds = clone;
3329 csa.ohds = origin_head;
3330 csa.force = force;
3331 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3332 dsl_dataset_clone_swap_check,
3333 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3334 return (error);
3335 }
3336
3337 /*
3338 * Given a pool name and a dataset object number in that pool,
3339 * return the name of that dataset.
3340 */
3341 int
3342 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3343 {
3344 spa_t *spa;
3345 dsl_pool_t *dp;
3346 dsl_dataset_t *ds;
3347 int error;
3348
3349 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3350 return (error);
3351 dp = spa_get_dsl(spa);
3352 rw_enter(&dp->dp_config_rwlock, RW_READER);
3353 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3354 dsl_dataset_name(ds, buf);
3355 dsl_dataset_rele(ds, FTAG);
3356 }
3357 rw_exit(&dp->dp_config_rwlock);
3358 spa_close(spa, FTAG);
3359
3360 return (error);
3361 }
3362
3363 int
3364 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3365 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3366 {
3367 int error = 0;
3368
3369 ASSERT3S(asize, >, 0);
3370
3371 /*
3372 * *ref_rsrv is the portion of asize that will come from any
3373 * unconsumed refreservation space.
3374 */
3375 *ref_rsrv = 0;
3376
3377 mutex_enter(&ds->ds_lock);
3378 /*
3379 * Make a space adjustment for reserved bytes.
3380 */
3381 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3382 ASSERT3U(*used, >=,
3383 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3384 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3385 *ref_rsrv =
3386 asize - MIN(asize, parent_delta(ds, asize + inflight));
3387 }
3388
3389 if (!check_quota || ds->ds_quota == 0) {
3390 mutex_exit(&ds->ds_lock);
3391 return (0);
3392 }
3393 /*
3394 * If they are requesting more space, and our current estimate
3395 * is over quota, they get to try again unless the actual
3396 * on-disk is over quota and there are no pending changes (which
3397 * may free up space for us).
3398 */
3399 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3400 if (inflight > 0 ||
3401 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3402 error = ERESTART;
3403 else
3404 error = EDQUOT;
3405 }
3406 mutex_exit(&ds->ds_lock);
3407
3408 return (error);
3409 }
3410
3411 /* ARGSUSED */
3412 static int
3413 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3414 {
3415 dsl_dataset_t *ds = arg1;
3416 dsl_prop_setarg_t *psa = arg2;
3417 int err;
3418
3419 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3420 return (ENOTSUP);
3421
3422 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3423 return (err);
3424
3425 if (psa->psa_effective_value == 0)
3426 return (0);
3427
3428 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3429 psa->psa_effective_value < ds->ds_reserved)
3430 return (ENOSPC);
3431
3432 return (0);
3433 }
3434
3435 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3436
3437 void
3438 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3439 {
3440 dsl_dataset_t *ds = arg1;
3441 dsl_prop_setarg_t *psa = arg2;
3442 uint64_t effective_value = psa->psa_effective_value;
3443
3444 dsl_prop_set_sync(ds, psa, tx);
3445 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3446
3447 if (ds->ds_quota != effective_value) {
3448 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3449 ds->ds_quota = effective_value;
3450
3451 spa_history_log_internal_ds(ds, "set refquota", tx,
3452 "refquota=%lld", (longlong_t)ds->ds_quota);
3453 }
3454 }
3455
3456 int
3457 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3458 {
3459 dsl_dataset_t *ds;
3460 dsl_prop_setarg_t psa;
3461 int err;
3462
3463 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3464
3465 err = dsl_dataset_hold(dsname, FTAG, &ds);
3466 if (err)
3467 return (err);
3468
3469 /*
3470 * If someone removes a file, then tries to set the quota, we
3471 * want to make sure the file freeing takes effect.
3472 */
3473 txg_wait_open(ds->ds_dir->dd_pool, 0);
3474
3475 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3476 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3477 ds, &psa, 0);
3478
3479 dsl_dataset_rele(ds, FTAG);
3480 return (err);
3481 }
3482
3483 static int
3484 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3485 {
3486 dsl_dataset_t *ds = arg1;
3487 dsl_prop_setarg_t *psa = arg2;
3488 uint64_t effective_value;
3489 uint64_t unique;
3490 int err;
3491
3492 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3493 SPA_VERSION_REFRESERVATION)
3494 return (ENOTSUP);
3495
3496 if (dsl_dataset_is_snapshot(ds))
3497 return (EINVAL);
3498
3499 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3500 return (err);
3501
3502 effective_value = psa->psa_effective_value;
3503
3504 /*
3505 * If we are doing the preliminary check in open context, the
3506 * space estimates may be inaccurate.
3507 */
3508 if (!dmu_tx_is_syncing(tx))
3509 return (0);
3510
3511 mutex_enter(&ds->ds_lock);
3512 if (!DS_UNIQUE_IS_ACCURATE(ds))
3513 dsl_dataset_recalc_head_uniq(ds);
3514 unique = ds->ds_phys->ds_unique_bytes;
3515 mutex_exit(&ds->ds_lock);
3516
3517 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3518 uint64_t delta = MAX(unique, effective_value) -
3519 MAX(unique, ds->ds_reserved);
3520
3521 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3522 return (ENOSPC);
3523 if (ds->ds_quota > 0 &&
3524 effective_value > ds->ds_quota)
3525 return (ENOSPC);
3526 }
3527
3528 return (0);
3529 }
3530
3531 static void
3532 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3533 {
3534 dsl_dataset_t *ds = arg1;
3535 dsl_prop_setarg_t *psa = arg2;
3536 uint64_t effective_value = psa->psa_effective_value;
3537 uint64_t unique;
3538 int64_t delta;
3539
3540 dsl_prop_set_sync(ds, psa, tx);
3541 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3542
3543 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3544
3545 mutex_enter(&ds->ds_dir->dd_lock);
3546 mutex_enter(&ds->ds_lock);
3547 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3548 unique = ds->ds_phys->ds_unique_bytes;
3549 delta = MAX(0, (int64_t)(effective_value - unique)) -
3550 MAX(0, (int64_t)(ds->ds_reserved - unique));
3551 ds->ds_reserved = effective_value;
3552 mutex_exit(&ds->ds_lock);
3553
3554 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3555 mutex_exit(&ds->ds_dir->dd_lock);
3556
3557 spa_history_log_internal_ds(ds, "set refreservation", tx,
3558 "refreservation=%lld", (longlong_t)effective_value);
3559 }
3560
3561 int
3562 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3563 uint64_t reservation)
3564 {
3565 dsl_dataset_t *ds;
3566 dsl_prop_setarg_t psa;
3567 int err;
3568
3569 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3570 &reservation);
3571
3572 err = dsl_dataset_hold(dsname, FTAG, &ds);
3573 if (err)
3574 return (err);
3575
3576 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3577 dsl_dataset_set_reservation_check,
3578 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3579
3580 dsl_dataset_rele(ds, FTAG);
3581 return (err);
3582 }
3583
3584 typedef struct zfs_hold_cleanup_arg {
3585 dsl_pool_t *dp;
3586 uint64_t dsobj;
3587 char htag[MAXNAMELEN];
3588 } zfs_hold_cleanup_arg_t;
3589
3590 static void
3591 dsl_dataset_user_release_onexit(void *arg)
3592 {
3593 zfs_hold_cleanup_arg_t *ca = arg;
3594
3595 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3596 B_TRUE);
3597 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3598 }
3599
3600 void
3601 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3602 minor_t minor)
3603 {
3604 zfs_hold_cleanup_arg_t *ca;
3605
3606 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3607 ca->dp = ds->ds_dir->dd_pool;
3608 ca->dsobj = ds->ds_object;
3609 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3610 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3611 dsl_dataset_user_release_onexit, ca, NULL));
3612 }
3613
3614 /*
3615 * If you add new checks here, you may need to add
3616 * additional checks to the "temporary" case in
3617 * snapshot_check() in dmu_objset.c.
3618 */
3619 static int
3620 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3621 {
3622 dsl_dataset_t *ds = arg1;
3623 struct dsl_ds_holdarg *ha = arg2;
3624 const char *htag = ha->htag;
3625 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3626 int error = 0;
3627
3628 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3629 return (ENOTSUP);
3630
3631 if (!dsl_dataset_is_snapshot(ds))
3632 return (EINVAL);
3633
3634 /* tags must be unique */
3635 mutex_enter(&ds->ds_lock);
3636 if (ds->ds_phys->ds_userrefs_obj) {
3637 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3638 8, 1, tx);
3639 if (error == 0)
3640 error = EEXIST;
3641 else if (error == ENOENT)
3642 error = 0;
3643 }
3644 mutex_exit(&ds->ds_lock);
3645
3646 if (error == 0 && ha->temphold &&
3647 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3648 error = E2BIG;
3649
3650 return (error);
3651 }
3652
3653 void
3654 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3655 {
3656 dsl_dataset_t *ds = arg1;
3657 struct dsl_ds_holdarg *ha = arg2;
3658 const char *htag = ha->htag;
3659 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3660 objset_t *mos = dp->dp_meta_objset;
3661 uint64_t now = gethrestime_sec();
3662 uint64_t zapobj;
3663
3664 mutex_enter(&ds->ds_lock);
3665 if (ds->ds_phys->ds_userrefs_obj == 0) {
3666 /*
3667 * This is the first user hold for this dataset. Create
3668 * the userrefs zap object.
3669 */
3670 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3671 zapobj = ds->ds_phys->ds_userrefs_obj =
3672 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3673 } else {
3674 zapobj = ds->ds_phys->ds_userrefs_obj;
3675 }
3676 ds->ds_userrefs++;
3677 mutex_exit(&ds->ds_lock);
3678
3679 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3680
3681 if (ha->temphold) {
3682 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3683 htag, &now, tx));
3684 }
3685
3686 spa_history_log_internal_ds(ds, "hold", tx,
3687 "tag = %s temp = %d holds now = %llu",
3688 htag, (int)ha->temphold, ds->ds_userrefs);
3689 }
3690
3691 static int
3692 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3693 {
3694 struct dsl_ds_holdarg *ha = arg;
3695 dsl_dataset_t *ds;
3696 int error;
3697 char *name;
3698
3699 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3700 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3701 error = dsl_dataset_hold(name, ha->dstg, &ds);
3702 strfree(name);
3703 if (error == 0) {
3704 ha->gotone = B_TRUE;
3705 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3706 dsl_dataset_user_hold_sync, ds, ha, 0);
3707 } else if (error == ENOENT && ha->recursive) {
3708 error = 0;
3709 } else {
3710 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3711 }
3712 return (error);
3713 }
3714
3715 int
3716 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3717 boolean_t temphold)
3718 {
3719 struct dsl_ds_holdarg *ha;
3720 int error;
3721
3722 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3723 ha->htag = htag;
3724 ha->temphold = temphold;
3725 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3726 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3727 ds, ha, 0);
3728 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3729
3730 return (error);
3731 }
3732
3733 int
3734 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3735 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3736 {
3737 struct dsl_ds_holdarg *ha;
3738 dsl_sync_task_t *dst;
3739 spa_t *spa;
3740 int error;
3741 minor_t minor = 0;
3742
3743 if (cleanup_fd != -1) {
3744 /* Currently we only support cleanup-on-exit of tempholds. */
3745 if (!temphold)
3746 return (EINVAL);
3747 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3748 if (error)
3749 return (error);
3750 }
3751
3752 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3753
3754 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3755
3756 error = spa_open(dsname, &spa, FTAG);
3757 if (error) {
3758 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3759 if (cleanup_fd != -1)
3760 zfs_onexit_fd_rele(cleanup_fd);
3761 return (error);
3762 }
3763
3764 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3765 ha->htag = htag;
3766 ha->snapname = snapname;
3767 ha->recursive = recursive;
3768 ha->temphold = temphold;
3769
3770 if (recursive) {
3771 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3772 ha, DS_FIND_CHILDREN);
3773 } else {
3774 error = dsl_dataset_user_hold_one(dsname, ha);
3775 }
3776 if (error == 0)
3777 error = dsl_sync_task_group_wait(ha->dstg);
3778
3779 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3780 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3781 dsl_dataset_t *ds = dst->dst_arg1;
3782
3783 if (dst->dst_err) {
3784 dsl_dataset_name(ds, ha->failed);
3785 *strchr(ha->failed, '@') = '\0';
3786 } else if (error == 0 && minor != 0 && temphold) {
3787 /*
3788 * If this hold is to be released upon process exit,
3789 * register that action now.
3790 */
3791 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3792 }
3793 dsl_dataset_rele(ds, ha->dstg);
3794 }
3795
3796 if (error == 0 && recursive && !ha->gotone)
3797 error = ENOENT;
3798
3799 if (error)
3800 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3801
3802 dsl_sync_task_group_destroy(ha->dstg);
3803
3804 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3805 spa_close(spa, FTAG);
3806 if (cleanup_fd != -1)
3807 zfs_onexit_fd_rele(cleanup_fd);
3808 return (error);
3809 }
3810
3811 struct dsl_ds_releasearg {
3812 dsl_dataset_t *ds;
3813 const char *htag;
3814 boolean_t own; /* do we own or just hold ds? */
3815 };
3816
3817 static int
3818 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3819 boolean_t *might_destroy)
3820 {
3821 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3822 uint64_t zapobj;
3823 uint64_t tmp;
3824 int error;
3825
3826 *might_destroy = B_FALSE;
3827
3828 mutex_enter(&ds->ds_lock);
3829 zapobj = ds->ds_phys->ds_userrefs_obj;
3830 if (zapobj == 0) {
3831 /* The tag can't possibly exist */
3832 mutex_exit(&ds->ds_lock);
3833 return (ESRCH);
3834 }
3835
3836 /* Make sure the tag exists */
3837 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3838 if (error) {
3839 mutex_exit(&ds->ds_lock);
3840 if (error == ENOENT)
3841 error = ESRCH;
3842 return (error);
3843 }
3844
3845 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3846 DS_IS_DEFER_DESTROY(ds))
3847 *might_destroy = B_TRUE;
3848
3849 mutex_exit(&ds->ds_lock);
3850 return (0);
3851 }
3852
3853 static int
3854 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3855 {
3856 struct dsl_ds_releasearg *ra = arg1;
3857 dsl_dataset_t *ds = ra->ds;
3858 boolean_t might_destroy;
3859 int error;
3860
3861 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3862 return (ENOTSUP);
3863
3864 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3865 if (error)
3866 return (error);
3867
3868 if (might_destroy) {
3869 struct dsl_ds_destroyarg dsda = {0};
3870
3871 if (dmu_tx_is_syncing(tx)) {
3872 /*
3873 * If we're not prepared to remove the snapshot,
3874 * we can't allow the release to happen right now.
3875 */
3876 if (!ra->own)
3877 return (EBUSY);
3878 }
3879 dsda.ds = ds;
3880 dsda.releasing = B_TRUE;
3881 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3882 }
3883
3884 return (0);
3885 }
3886
3887 static void
3888 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3889 {
3890 struct dsl_ds_releasearg *ra = arg1;
3891 dsl_dataset_t *ds = ra->ds;
3892 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3893 objset_t *mos = dp->dp_meta_objset;
3894 uint64_t zapobj;
3895 uint64_t refs;
3896 int error;
3897
3898 mutex_enter(&ds->ds_lock);
3899 ds->ds_userrefs--;
3900 refs = ds->ds_userrefs;
3901 mutex_exit(&ds->ds_lock);
3902 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3903 VERIFY(error == 0 || error == ENOENT);
3904 zapobj = ds->ds_phys->ds_userrefs_obj;
3905 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3906 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3907 DS_IS_DEFER_DESTROY(ds)) {
3908 struct dsl_ds_destroyarg dsda = {0};
3909
3910 ASSERT(ra->own);
3911 dsda.ds = ds;
3912 dsda.releasing = B_TRUE;
3913 /* We already did the destroy_check */
3914 dsl_dataset_destroy_sync(&dsda, tag, tx);
3915 }
3916
3917 spa_history_log_internal_ds(ds, "release", tx,
3918 "tag = %s refs now = %lld", ra->htag, (longlong_t)refs);
3919 }
3920
3921 static int
3922 dsl_dataset_user_release_one(const char *dsname, void *arg)
3923 {
3924 struct dsl_ds_holdarg *ha = arg;
3925 struct dsl_ds_releasearg *ra;
3926 dsl_dataset_t *ds;
3927 int error;
3928 void *dtag = ha->dstg;
3929 char *name;
3930 boolean_t own = B_FALSE;
3931 boolean_t might_destroy;
3932
3933 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3934 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3935 error = dsl_dataset_hold(name, dtag, &ds);
3936 strfree(name);
3937 if (error == ENOENT && ha->recursive)
3938 return (0);
3939 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3940 if (error)
3941 return (error);
3942
3943 ha->gotone = B_TRUE;
3944
3945 ASSERT(dsl_dataset_is_snapshot(ds));
3946
3947 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3948 if (error) {
3949 dsl_dataset_rele(ds, dtag);
3950 return (error);
3951 }
3952
3953 if (might_destroy) {
3954 #ifdef _KERNEL
3955 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3956 error = zfs_unmount_snap(name, NULL);
3957 strfree(name);
3958 if (error) {
3959 dsl_dataset_rele(ds, dtag);
3960 return (error);
3961 }
3962 #endif
3963 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3964 dsl_dataset_rele(ds, dtag);
3965 return (EBUSY);
3966 } else {
3967 own = B_TRUE;
3968 dsl_dataset_make_exclusive(ds, dtag);
3969 }
3970 }
3971
3972 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3973 ra->ds = ds;
3974 ra->htag = ha->htag;
3975 ra->own = own;
3976 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3977 dsl_dataset_user_release_sync, ra, dtag, 0);
3978
3979 return (0);
3980 }
3981
3982 int
3983 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3984 boolean_t recursive)
3985 {
3986 struct dsl_ds_holdarg *ha;
3987 dsl_sync_task_t *dst;
3988 spa_t *spa;
3989 int error;
3990
3991 top:
3992 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3993
3994 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3995
3996 error = spa_open(dsname, &spa, FTAG);
3997 if (error) {
3998 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3999 return (error);
4000 }
4001
4002 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4003 ha->htag = htag;
4004 ha->snapname = snapname;
4005 ha->recursive = recursive;
4006 if (recursive) {
4007 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4008 ha, DS_FIND_CHILDREN);
4009 } else {
4010 error = dsl_dataset_user_release_one(dsname, ha);
4011 }
4012 if (error == 0)
4013 error = dsl_sync_task_group_wait(ha->dstg);
4014
4015 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4016 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4017 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4018 dsl_dataset_t *ds = ra->ds;
4019
4020 if (dst->dst_err)
4021 dsl_dataset_name(ds, ha->failed);
4022
4023 if (ra->own)
4024 dsl_dataset_disown(ds, ha->dstg);
4025 else
4026 dsl_dataset_rele(ds, ha->dstg);
4027
4028 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4029 }
4030
4031 if (error == 0 && recursive && !ha->gotone)
4032 error = ENOENT;
4033
4034 if (error && error != EBUSY)
4035 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4036
4037 dsl_sync_task_group_destroy(ha->dstg);
4038 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4039 spa_close(spa, FTAG);
4040
4041 /*
4042 * We can get EBUSY if we were racing with deferred destroy and
4043 * dsl_dataset_user_release_check() hadn't done the necessary
4044 * open context setup. We can also get EBUSY if we're racing
4045 * with destroy and that thread is the ds_owner. Either way
4046 * the busy condition should be transient, and we should retry
4047 * the release operation.
4048 */
4049 if (error == EBUSY)
4050 goto top;
4051
4052 return (error);
4053 }
4054
4055 /*
4056 * Called at spa_load time (with retry == B_FALSE) to release a stale
4057 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4058 */
4059 int
4060 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4061 boolean_t retry)
4062 {
4063 dsl_dataset_t *ds;
4064 char *snap;
4065 char *name;
4066 int namelen;
4067 int error;
4068
4069 do {
4070 rw_enter(&dp->dp_config_rwlock, RW_READER);
4071 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4072 rw_exit(&dp->dp_config_rwlock);
4073 if (error)
4074 return (error);
4075 namelen = dsl_dataset_namelen(ds)+1;
4076 name = kmem_alloc(namelen, KM_SLEEP);
4077 dsl_dataset_name(ds, name);
4078 dsl_dataset_rele(ds, FTAG);
4079
4080 snap = strchr(name, '@');
4081 *snap = '\0';
4082 ++snap;
4083 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4084 kmem_free(name, namelen);
4085
4086 /*
4087 * The object can't have been destroyed because we have a hold,
4088 * but it might have been renamed, resulting in ENOENT. Retry
4089 * if we've been requested to do so.
4090 *
4091 * It would be nice if we could use the dsobj all the way
4092 * through and avoid ENOENT entirely. But we might need to
4093 * unmount the snapshot, and there's currently no way to lookup
4094 * a vfsp using a ZFS object id.
4095 */
4096 } while ((error == ENOENT) && retry);
4097
4098 return (error);
4099 }
4100
4101 int
4102 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4103 {
4104 dsl_dataset_t *ds;
4105 int err;
4106
4107 err = dsl_dataset_hold(dsname, FTAG, &ds);
4108 if (err)
4109 return (err);
4110
4111 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4112 if (ds->ds_phys->ds_userrefs_obj != 0) {
4113 zap_attribute_t *za;
4114 zap_cursor_t zc;
4115
4116 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4117 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4118 ds->ds_phys->ds_userrefs_obj);
4119 zap_cursor_retrieve(&zc, za) == 0;
4120 zap_cursor_advance(&zc)) {
4121 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4122 za->za_first_integer));
4123 }
4124 zap_cursor_fini(&zc);
4125 kmem_free(za, sizeof (zap_attribute_t));
4126 }
4127 dsl_dataset_rele(ds, FTAG);
4128 return (0);
4129 }
4130
4131 /*
4132 * Note, this function is used as the callback for dmu_objset_find(). We
4133 * always return 0 so that we will continue to find and process
4134 * inconsistent datasets, even if we encounter an error trying to
4135 * process one of them.
4136 */
4137 /* ARGSUSED */
4138 int
4139 dsl_destroy_inconsistent(const char *dsname, void *arg)
4140 {
4141 dsl_dataset_t *ds;
4142
4143 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4144 if (DS_IS_INCONSISTENT(ds))
4145 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4146 else
4147 dsl_dataset_disown(ds, FTAG);
4148 }
4149 return (0);
4150 }
4151
4152 /*
4153 * Return (in *usedp) the amount of space written in new that is not
4154 * present in oldsnap. New may be a snapshot or the head. Old must be
4155 * a snapshot before new, in new's filesystem (or its origin). If not then
4156 * fail and return EINVAL.
4157 *
4158 * The written space is calculated by considering two components: First, we
4159 * ignore any freed space, and calculate the written as new's used space
4160 * minus old's used space. Next, we add in the amount of space that was freed
4161 * between the two snapshots, thus reducing new's used space relative to old's.
4162 * Specifically, this is the space that was born before old->ds_creation_txg,
4163 * and freed before new (ie. on new's deadlist or a previous deadlist).
4164 *
4165 * space freed [---------------------]
4166 * snapshots ---O-------O--------O-------O------
4167 * oldsnap new
4168 */
4169 int
4170 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4171 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4172 {
4173 int err = 0;
4174 uint64_t snapobj;
4175 dsl_pool_t *dp = new->ds_dir->dd_pool;
4176
4177 *usedp = 0;
4178 *usedp += new->ds_phys->ds_referenced_bytes;
4179 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4180
4181 *compp = 0;
4182 *compp += new->ds_phys->ds_compressed_bytes;
4183 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4184
4185 *uncompp = 0;
4186 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4187 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4188
4189 rw_enter(&dp->dp_config_rwlock, RW_READER);
4190 snapobj = new->ds_object;
4191 while (snapobj != oldsnap->ds_object) {
4192 dsl_dataset_t *snap;
4193 uint64_t used, comp, uncomp;
4194
4195 if (snapobj == new->ds_object) {
4196 snap = new;
4197 } else {
4198 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4199 if (err != 0)
4200 break;
4201 }
4202
4203 if (snap->ds_phys->ds_prev_snap_txg ==
4204 oldsnap->ds_phys->ds_creation_txg) {
4205 /*
4206 * The blocks in the deadlist can not be born after
4207 * ds_prev_snap_txg, so get the whole deadlist space,
4208 * which is more efficient (especially for old-format
4209 * deadlists). Unfortunately the deadlist code
4210 * doesn't have enough information to make this
4211 * optimization itself.
4212 */
4213 dsl_deadlist_space(&snap->ds_deadlist,
4214 &used, &comp, &uncomp);
4215 } else {
4216 dsl_deadlist_space_range(&snap->ds_deadlist,
4217 0, oldsnap->ds_phys->ds_creation_txg,
4218 &used, &comp, &uncomp);
4219 }
4220 *usedp += used;
4221 *compp += comp;
4222 *uncompp += uncomp;
4223
4224 /*
4225 * If we get to the beginning of the chain of snapshots
4226 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4227 * was not a snapshot of/before new.
4228 */
4229 snapobj = snap->ds_phys->ds_prev_snap_obj;
4230 if (snap != new)
4231 dsl_dataset_rele(snap, FTAG);
4232 if (snapobj == 0) {
4233 err = EINVAL;
4234 break;
4235 }
4236
4237 }
4238 rw_exit(&dp->dp_config_rwlock);
4239 return (err);
4240 }
4241
4242 /*
4243 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4244 * lastsnap, and all snapshots in between are deleted.
4245 *
4246 * blocks that would be freed [---------------------------]
4247 * snapshots ---O-------O--------O-------O--------O
4248 * firstsnap lastsnap
4249 *
4250 * This is the set of blocks that were born after the snap before firstsnap,
4251 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4252 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4253 * We calculate this by iterating over the relevant deadlists (from the snap
4254 * after lastsnap, backward to the snap after firstsnap), summing up the
4255 * space on the deadlist that was born after the snap before firstsnap.
4256 */
4257 int
4258 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4259 dsl_dataset_t *lastsnap,
4260 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4261 {
4262 int err = 0;
4263 uint64_t snapobj;
4264 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4265
4266 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4267 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4268
4269 /*
4270 * Check that the snapshots are in the same dsl_dir, and firstsnap
4271 * is before lastsnap.
4272 */
4273 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4274 firstsnap->ds_phys->ds_creation_txg >
4275 lastsnap->ds_phys->ds_creation_txg)
4276 return (EINVAL);
4277
4278 *usedp = *compp = *uncompp = 0;
4279
4280 rw_enter(&dp->dp_config_rwlock, RW_READER);
4281 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4282 while (snapobj != firstsnap->ds_object) {
4283 dsl_dataset_t *ds;
4284 uint64_t used, comp, uncomp;
4285
4286 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4287 if (err != 0)
4288 break;
4289
4290 dsl_deadlist_space_range(&ds->ds_deadlist,
4291 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4292 &used, &comp, &uncomp);
4293 *usedp += used;
4294 *compp += comp;
4295 *uncompp += uncomp;
4296
4297 snapobj = ds->ds_phys->ds_prev_snap_obj;
4298 ASSERT3U(snapobj, !=, 0);
4299 dsl_dataset_rele(ds, FTAG);
4300 }
4301 rw_exit(&dp->dp_config_rwlock);
4302 return (err);
4303 }