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 do {
807 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
808 sizeof (dsphys->ds_guid));
809 } while (dsphys->ds_guid == 0);
810 dsphys->ds_snapnames_zapobj =
811 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
812 DMU_OT_NONE, 0, tx);
813 dsphys->ds_creation_time = gethrestime_sec();
814 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
815
816 if (origin == NULL) {
817 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
818 } else {
819 dsl_dataset_t *ohds;
820
821 dsphys->ds_prev_snap_obj = origin->ds_object;
822 dsphys->ds_prev_snap_txg =
823 origin->ds_phys->ds_creation_txg;
824 dsphys->ds_referenced_bytes =
825 origin->ds_phys->ds_referenced_bytes;
826 dsphys->ds_compressed_bytes =
827 origin->ds_phys->ds_compressed_bytes;
828 dsphys->ds_uncompressed_bytes =
829 origin->ds_phys->ds_uncompressed_bytes;
830 dsphys->ds_bp = origin->ds_phys->ds_bp;
831 dsphys->ds_flags |= origin->ds_phys->ds_flags;
832
833 dmu_buf_will_dirty(origin->ds_dbuf, tx);
834 origin->ds_phys->ds_num_children++;
835
836 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
837 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
838 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
839 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
840 dsl_dataset_rele(ohds, FTAG);
841
842 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
843 if (origin->ds_phys->ds_next_clones_obj == 0) {
844 origin->ds_phys->ds_next_clones_obj =
845 zap_create(mos,
846 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
847 }
848 VERIFY(0 == zap_add_int(mos,
849 origin->ds_phys->ds_next_clones_obj,
850 dsobj, tx));
851 }
852
853 dmu_buf_will_dirty(dd->dd_dbuf, tx);
854 dd->dd_phys->dd_origin_obj = origin->ds_object;
855 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
856 if (origin->ds_dir->dd_phys->dd_clones == 0) {
857 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
858 origin->ds_dir->dd_phys->dd_clones =
859 zap_create(mos,
860 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
861 }
862 VERIFY3U(0, ==, zap_add_int(mos,
863 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
864 }
865 }
866
867 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
868 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
869
870 dmu_buf_rele(dbuf, FTAG);
871
872 dmu_buf_will_dirty(dd->dd_dbuf, tx);
873 dd->dd_phys->dd_head_dataset_obj = dsobj;
874
875 return (dsobj);
876 }
877
878 uint64_t
879 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
880 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
881 {
882 dsl_pool_t *dp = pdd->dd_pool;
883 uint64_t dsobj, ddobj;
884 dsl_dir_t *dd;
885
886 ASSERT(lastname[0] != '@');
887
888 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
889 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
890
891 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
892
893 dsl_deleg_set_create_perms(dd, tx, cr);
894
895 dsl_dir_close(dd, FTAG);
896
897 /*
898 * If we are creating a clone, make sure we zero out any stale
899 * data from the origin snapshots zil header.
900 */
901 if (origin != NULL) {
902 dsl_dataset_t *ds;
903 objset_t *os;
904
905 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
906 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
907 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
908 dsl_dataset_dirty(ds, tx);
909 dsl_dataset_rele(ds, FTAG);
910 }
911
912 return (dsobj);
913 }
914
915 /*
916 * The snapshots must all be in the same pool.
917 */
918 int
919 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer, char *failed)
920 {
921 int err;
922 dsl_sync_task_t *dst;
923 spa_t *spa;
924 nvpair_t *pair;
925 dsl_sync_task_group_t *dstg;
926
927 pair = nvlist_next_nvpair(snaps, NULL);
928 if (pair == NULL)
929 return (0);
930
931 err = spa_open(nvpair_name(pair), &spa, FTAG);
932 if (err)
933 return (err);
934 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
935
936 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
937 pair = nvlist_next_nvpair(snaps, pair)) {
938 dsl_dataset_t *ds;
939
940 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
941 if (err == 0) {
942 struct dsl_ds_destroyarg *dsda;
943
944 dsl_dataset_make_exclusive(ds, dstg);
945 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
946 KM_SLEEP);
947 dsda->ds = ds;
948 dsda->defer = defer;
949 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
950 dsl_dataset_destroy_sync, dsda, dstg, 0);
951 } else if (err == ENOENT) {
952 err = 0;
953 } else {
954 (void) strcpy(failed, nvpair_name(pair));
955 break;
956 }
957 }
958
959 if (err == 0)
960 err = dsl_sync_task_group_wait(dstg);
961
962 for (dst = list_head(&dstg->dstg_tasks); dst;
963 dst = list_next(&dstg->dstg_tasks, dst)) {
964 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
965 dsl_dataset_t *ds = dsda->ds;
966
967 /*
968 * Return the file system name that triggered the error
969 */
970 if (dst->dst_err) {
971 dsl_dataset_name(ds, failed);
972 }
973 ASSERT3P(dsda->rm_origin, ==, NULL);
974 dsl_dataset_disown(ds, dstg);
975 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
976 }
977
978 dsl_sync_task_group_destroy(dstg);
979 spa_close(spa, FTAG);
980 return (err);
981
982 }
983
984 static boolean_t
985 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
986 {
987 boolean_t might_destroy = B_FALSE;
988
989 mutex_enter(&ds->ds_lock);
990 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
991 DS_IS_DEFER_DESTROY(ds))
992 might_destroy = B_TRUE;
993 mutex_exit(&ds->ds_lock);
994
995 return (might_destroy);
996 }
997
998 /*
999 * If we're removing a clone, and these three conditions are true:
1000 * 1) the clone's origin has no other children
1001 * 2) the clone's origin has no user references
1002 * 3) the clone's origin has been marked for deferred destruction
1003 * Then, prepare to remove the origin as part of this sync task group.
1004 */
1005 static int
1006 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1007 {
1008 dsl_dataset_t *ds = dsda->ds;
1009 dsl_dataset_t *origin = ds->ds_prev;
1010
1011 if (dsl_dataset_might_destroy_origin(origin)) {
1012 char *name;
1013 int namelen;
1014 int error;
1015
1016 namelen = dsl_dataset_namelen(origin) + 1;
1017 name = kmem_alloc(namelen, KM_SLEEP);
1018 dsl_dataset_name(origin, name);
1019 #ifdef _KERNEL
1020 error = zfs_unmount_snap(name, NULL);
1021 if (error) {
1022 kmem_free(name, namelen);
1023 return (error);
1024 }
1025 #endif
1026 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1027 kmem_free(name, namelen);
1028 if (error)
1029 return (error);
1030 dsda->rm_origin = origin;
1031 dsl_dataset_make_exclusive(origin, tag);
1032 }
1033
1034 return (0);
1035 }
1036
1037 /*
1038 * ds must be opened as OWNER. On return (whether successful or not),
1039 * ds will be closed and caller can no longer dereference it.
1040 */
1041 int
1042 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1043 {
1044 int err;
1045 dsl_sync_task_group_t *dstg;
1046 objset_t *os;
1047 dsl_dir_t *dd;
1048 uint64_t obj;
1049 struct dsl_ds_destroyarg dsda = { 0 };
1050 dsl_dataset_t dummy_ds = { 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 dummy_ds.ds_dir = dd;
1071 dummy_ds.ds_object = ds->ds_object;
1072
1073 /*
1074 * Check for errors and mark this ds as inconsistent, in
1075 * case we crash while freeing the objects.
1076 */
1077 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1078 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1079 if (err)
1080 goto out;
1081
1082 err = dmu_objset_from_ds(ds, &os);
1083 if (err)
1084 goto out;
1085
1086 /*
1087 * If async destruction is not enabled try to remove all objects
1088 * while in the open context so that there is less work to do in
1089 * the syncing context.
1090 */
1091 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1092 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1093 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1094 ds->ds_phys->ds_prev_snap_txg)) {
1095 /*
1096 * Ignore errors, if there is not enough disk space
1097 * we will deal with it in dsl_dataset_destroy_sync().
1098 */
1099 (void) dmu_free_object(os, obj);
1100 }
1101 if (err != ESRCH)
1102 goto out;
1103 }
1104
1105 /*
1106 * Only the ZIL knows how to free log blocks.
1107 */
1108 zil_destroy(dmu_objset_zil(os), B_FALSE);
1109
1110 /*
1111 * Sync out all in-flight IO.
1112 */
1113 txg_wait_synced(dd->dd_pool, 0);
1114
1115 /*
1116 * If we managed to free all the objects in open
1117 * context, the user space accounting should be zero.
1118 */
1119 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1120 dmu_objset_userused_enabled(os)) {
1121 uint64_t count;
1122
1123 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1124 count == 0);
1125 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1126 count == 0);
1127 }
1128
1129 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1130 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1131 rw_exit(&dd->dd_pool->dp_config_rwlock);
1132
1133 if (err)
1134 goto out;
1135
1136 /*
1137 * Blow away the dsl_dir + head dataset.
1138 */
1139 dsl_dataset_make_exclusive(ds, tag);
1140 /*
1141 * If we're removing a clone, we might also need to remove its
1142 * origin.
1143 */
1144 do {
1145 dsda.need_prep = B_FALSE;
1146 if (dsl_dir_is_clone(dd)) {
1147 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1148 if (err) {
1149 dsl_dir_close(dd, FTAG);
1150 goto out;
1151 }
1152 }
1153
1154 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1155 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1156 dsl_dataset_destroy_sync, &dsda, tag, 0);
1157 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1158 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
1159 err = dsl_sync_task_group_wait(dstg);
1160 dsl_sync_task_group_destroy(dstg);
1161
1162 /*
1163 * We could be racing against 'zfs release' or 'zfs destroy -d'
1164 * on the origin snap, in which case we can get EBUSY if we
1165 * needed to destroy the origin snap but were not ready to
1166 * do so.
1167 */
1168 if (dsda.need_prep) {
1169 ASSERT(err == EBUSY);
1170 ASSERT(dsl_dir_is_clone(dd));
1171 ASSERT(dsda.rm_origin == NULL);
1172 }
1173 } while (dsda.need_prep);
1174
1175 if (dsda.rm_origin != NULL)
1176 dsl_dataset_disown(dsda.rm_origin, tag);
1177
1178 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1179 if (err)
1180 dsl_dir_close(dd, FTAG);
1181 out:
1182 dsl_dataset_disown(ds, tag);
1183 return (err);
1184 }
1185
1186 blkptr_t *
1187 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1188 {
1189 return (&ds->ds_phys->ds_bp);
1190 }
1191
1192 void
1193 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1194 {
1195 ASSERT(dmu_tx_is_syncing(tx));
1196 /* If it's the meta-objset, set dp_meta_rootbp */
1197 if (ds == NULL) {
1198 tx->tx_pool->dp_meta_rootbp = *bp;
1199 } else {
1200 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1201 ds->ds_phys->ds_bp = *bp;
1202 }
1203 }
1204
1205 spa_t *
1206 dsl_dataset_get_spa(dsl_dataset_t *ds)
1207 {
1208 return (ds->ds_dir->dd_pool->dp_spa);
1209 }
1210
1211 void
1212 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1213 {
1214 dsl_pool_t *dp;
1215
1216 if (ds == NULL) /* this is the meta-objset */
1217 return;
1218
1219 ASSERT(ds->ds_objset != NULL);
1220
1221 if (ds->ds_phys->ds_next_snap_obj != 0)
1222 panic("dirtying snapshot!");
1223
1224 dp = ds->ds_dir->dd_pool;
1225
1226 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1227 /* up the hold count until we can be written out */
1228 dmu_buf_add_ref(ds->ds_dbuf, ds);
1229 }
1230 }
1231
1232 /*
1233 * The unique space in the head dataset can be calculated by subtracting
1234 * the space used in the most recent snapshot, that is still being used
1235 * in this file system, from the space currently in use. To figure out
1236 * the space in the most recent snapshot still in use, we need to take
1237 * the total space used in the snapshot and subtract out the space that
1238 * has been freed up since the snapshot was taken.
1239 */
1240 static void
1241 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1242 {
1243 uint64_t mrs_used;
1244 uint64_t dlused, dlcomp, dluncomp;
1245
1246 ASSERT(!dsl_dataset_is_snapshot(ds));
1247
1248 if (ds->ds_phys->ds_prev_snap_obj != 0)
1249 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1250 else
1251 mrs_used = 0;
1252
1253 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1254
1255 ASSERT3U(dlused, <=, mrs_used);
1256 ds->ds_phys->ds_unique_bytes =
1257 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1258
1259 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1260 SPA_VERSION_UNIQUE_ACCURATE)
1261 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1262 }
1263
1264 struct killarg {
1265 dsl_dataset_t *ds;
1266 dmu_tx_t *tx;
1267 };
1268
1269 /* ARGSUSED */
1270 static int
1271 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1272 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1273 {
1274 struct killarg *ka = arg;
1275 dmu_tx_t *tx = ka->tx;
1276
1277 if (bp == NULL)
1278 return (0);
1279
1280 if (zb->zb_level == ZB_ZIL_LEVEL) {
1281 ASSERT(zilog != NULL);
1282 /*
1283 * It's a block in the intent log. It has no
1284 * accounting, so just free it.
1285 */
1286 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1287 } else {
1288 ASSERT(zilog == NULL);
1289 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1290 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1291 }
1292
1293 return (0);
1294 }
1295
1296 /* ARGSUSED */
1297 static int
1298 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1299 {
1300 dsl_dataset_t *ds = arg1;
1301 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1302 uint64_t count;
1303 int err;
1304
1305 /*
1306 * Can't delete a head dataset if there are snapshots of it.
1307 * (Except if the only snapshots are from the branch we cloned
1308 * from.)
1309 */
1310 if (ds->ds_prev != NULL &&
1311 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1312 return (EBUSY);
1313
1314 /*
1315 * This is really a dsl_dir thing, but check it here so that
1316 * we'll be less likely to leave this dataset inconsistent &
1317 * nearly destroyed.
1318 */
1319 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1320 if (err)
1321 return (err);
1322 if (count != 0)
1323 return (EEXIST);
1324
1325 return (0);
1326 }
1327
1328 /* ARGSUSED */
1329 static void
1330 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1331 {
1332 dsl_dataset_t *ds = arg1;
1333 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1334
1335 /* Mark it as inconsistent on-disk, in case we crash */
1336 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1337 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1338
1339 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1340 "dataset = %llu", ds->ds_object);
1341 }
1342
1343 static int
1344 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1345 dmu_tx_t *tx)
1346 {
1347 dsl_dataset_t *ds = dsda->ds;
1348 dsl_dataset_t *ds_prev = ds->ds_prev;
1349
1350 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1351 struct dsl_ds_destroyarg ndsda = {0};
1352
1353 /*
1354 * If we're not prepared to remove the origin, don't remove
1355 * the clone either.
1356 */
1357 if (dsda->rm_origin == NULL) {
1358 dsda->need_prep = B_TRUE;
1359 return (EBUSY);
1360 }
1361
1362 ndsda.ds = ds_prev;
1363 ndsda.is_origin_rm = B_TRUE;
1364 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1365 }
1366
1367 /*
1368 * If we're not going to remove the origin after all,
1369 * undo the open context setup.
1370 */
1371 if (dsda->rm_origin != NULL) {
1372 dsl_dataset_disown(dsda->rm_origin, tag);
1373 dsda->rm_origin = NULL;
1374 }
1375
1376 return (0);
1377 }
1378
1379 /*
1380 * If you add new checks here, you may need to add
1381 * additional checks to the "temporary" case in
1382 * snapshot_check() in dmu_objset.c.
1383 */
1384 /* ARGSUSED */
1385 int
1386 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1387 {
1388 struct dsl_ds_destroyarg *dsda = arg1;
1389 dsl_dataset_t *ds = dsda->ds;
1390
1391 /* we have an owner hold, so noone else can destroy us */
1392 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1393
1394 /*
1395 * Only allow deferred destroy on pools that support it.
1396 * NOTE: deferred destroy is only supported on snapshots.
1397 */
1398 if (dsda->defer) {
1399 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1400 SPA_VERSION_USERREFS)
1401 return (ENOTSUP);
1402 ASSERT(dsl_dataset_is_snapshot(ds));
1403 return (0);
1404 }
1405
1406 /*
1407 * Can't delete a head dataset if there are snapshots of it.
1408 * (Except if the only snapshots are from the branch we cloned
1409 * from.)
1410 */
1411 if (ds->ds_prev != NULL &&
1412 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1413 return (EBUSY);
1414
1415 /*
1416 * If we made changes this txg, traverse_dsl_dataset won't find
1417 * them. Try again.
1418 */
1419 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1420 return (EAGAIN);
1421
1422 if (dsl_dataset_is_snapshot(ds)) {
1423 /*
1424 * If this snapshot has an elevated user reference count,
1425 * we can't destroy it yet.
1426 */
1427 if (ds->ds_userrefs > 0 && !dsda->releasing)
1428 return (EBUSY);
1429
1430 mutex_enter(&ds->ds_lock);
1431 /*
1432 * Can't delete a branch point. However, if we're destroying
1433 * a clone and removing its origin due to it having a user
1434 * hold count of 0 and having been marked for deferred destroy,
1435 * it's OK for the origin to have a single clone.
1436 */
1437 if (ds->ds_phys->ds_num_children >
1438 (dsda->is_origin_rm ? 2 : 1)) {
1439 mutex_exit(&ds->ds_lock);
1440 return (EEXIST);
1441 }
1442 mutex_exit(&ds->ds_lock);
1443 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1444 return (dsl_dataset_origin_check(dsda, arg2, tx));
1445 }
1446
1447 /* XXX we should do some i/o error checking... */
1448 return (0);
1449 }
1450
1451 struct refsarg {
1452 kmutex_t lock;
1453 boolean_t gone;
1454 kcondvar_t cv;
1455 };
1456
1457 /* ARGSUSED */
1458 static void
1459 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1460 {
1461 struct refsarg *arg = argv;
1462
1463 mutex_enter(&arg->lock);
1464 arg->gone = TRUE;
1465 cv_signal(&arg->cv);
1466 mutex_exit(&arg->lock);
1467 }
1468
1469 static void
1470 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1471 {
1472 struct refsarg arg;
1473
1474 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1475 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1476 arg.gone = FALSE;
1477 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1478 dsl_dataset_refs_gone);
1479 dmu_buf_rele(ds->ds_dbuf, tag);
1480 mutex_enter(&arg.lock);
1481 while (!arg.gone)
1482 cv_wait(&arg.cv, &arg.lock);
1483 ASSERT(arg.gone);
1484 mutex_exit(&arg.lock);
1485 ds->ds_dbuf = NULL;
1486 ds->ds_phys = NULL;
1487 mutex_destroy(&arg.lock);
1488 cv_destroy(&arg.cv);
1489 }
1490
1491 static void
1492 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1493 {
1494 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1495 uint64_t count;
1496 int err;
1497
1498 ASSERT(ds->ds_phys->ds_num_children >= 2);
1499 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1500 /*
1501 * The err should not be ENOENT, but a bug in a previous version
1502 * of the code could cause upgrade_clones_cb() to not set
1503 * ds_next_snap_obj when it should, leading to a missing entry.
1504 * If we knew that the pool was created after
1505 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1506 * ENOENT. However, at least we can check that we don't have
1507 * too many entries in the next_clones_obj even after failing to
1508 * remove this one.
1509 */
1510 if (err != ENOENT) {
1511 VERIFY3U(err, ==, 0);
1512 }
1513 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1514 &count));
1515 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1516 }
1517
1518 static void
1519 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1520 {
1521 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1522 zap_cursor_t zc;
1523 zap_attribute_t za;
1524
1525 /*
1526 * If it is the old version, dd_clones doesn't exist so we can't
1527 * find the clones, but deadlist_remove_key() is a no-op so it
1528 * doesn't matter.
1529 */
1530 if (ds->ds_dir->dd_phys->dd_clones == 0)
1531 return;
1532
1533 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1534 zap_cursor_retrieve(&zc, &za) == 0;
1535 zap_cursor_advance(&zc)) {
1536 dsl_dataset_t *clone;
1537
1538 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1539 za.za_first_integer, FTAG, &clone));
1540 if (clone->ds_dir->dd_origin_txg > mintxg) {
1541 dsl_deadlist_remove_key(&clone->ds_deadlist,
1542 mintxg, tx);
1543 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1544 }
1545 dsl_dataset_rele(clone, FTAG);
1546 }
1547 zap_cursor_fini(&zc);
1548 }
1549
1550 struct process_old_arg {
1551 dsl_dataset_t *ds;
1552 dsl_dataset_t *ds_prev;
1553 boolean_t after_branch_point;
1554 zio_t *pio;
1555 uint64_t used, comp, uncomp;
1556 };
1557
1558 static int
1559 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1560 {
1561 struct process_old_arg *poa = arg;
1562 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1563
1564 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1565 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1566 if (poa->ds_prev && !poa->after_branch_point &&
1567 bp->blk_birth >
1568 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1569 poa->ds_prev->ds_phys->ds_unique_bytes +=
1570 bp_get_dsize_sync(dp->dp_spa, bp);
1571 }
1572 } else {
1573 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1574 poa->comp += BP_GET_PSIZE(bp);
1575 poa->uncomp += BP_GET_UCSIZE(bp);
1576 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1577 }
1578 return (0);
1579 }
1580
1581 static void
1582 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1583 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1584 {
1585 struct process_old_arg poa = { 0 };
1586 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1587 objset_t *mos = dp->dp_meta_objset;
1588
1589 ASSERT(ds->ds_deadlist.dl_oldfmt);
1590 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1591
1592 poa.ds = ds;
1593 poa.ds_prev = ds_prev;
1594 poa.after_branch_point = after_branch_point;
1595 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1596 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1597 process_old_cb, &poa, tx));
1598 VERIFY3U(zio_wait(poa.pio), ==, 0);
1599 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1600
1601 /* change snapused */
1602 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1603 -poa.used, -poa.comp, -poa.uncomp, tx);
1604
1605 /* swap next's deadlist to our deadlist */
1606 dsl_deadlist_close(&ds->ds_deadlist);
1607 dsl_deadlist_close(&ds_next->ds_deadlist);
1608 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1609 ds->ds_phys->ds_deadlist_obj);
1610 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1611 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1612 ds_next->ds_phys->ds_deadlist_obj);
1613 }
1614
1615 static int
1616 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1617 {
1618 int err;
1619 struct killarg ka;
1620
1621 /*
1622 * Free everything that we point to (that's born after
1623 * the previous snapshot, if we are a clone)
1624 *
1625 * NB: this should be very quick, because we already
1626 * freed all the objects in open context.
1627 */
1628 ka.ds = ds;
1629 ka.tx = tx;
1630 err = traverse_dataset(ds,
1631 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1632 kill_blkptr, &ka);
1633 ASSERT3U(err, ==, 0);
1634 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1635
1636 return (err);
1637 }
1638
1639 void
1640 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1641 {
1642 struct dsl_ds_destroyarg *dsda = arg1;
1643 dsl_dataset_t *ds = dsda->ds;
1644 int err;
1645 int after_branch_point = FALSE;
1646 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1647 objset_t *mos = dp->dp_meta_objset;
1648 dsl_dataset_t *ds_prev = NULL;
1649 boolean_t wont_destroy;
1650 uint64_t obj;
1651
1652 wont_destroy = (dsda->defer &&
1653 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1654
1655 ASSERT(ds->ds_owner || wont_destroy);
1656 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1657 ASSERT(ds->ds_prev == NULL ||
1658 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1659 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1660
1661 if (wont_destroy) {
1662 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1663 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1664 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1665 return;
1666 }
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 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1963 "dataset = %llu", ds->ds_object);
1964
1965 if (ds->ds_phys->ds_next_clones_obj != 0) {
1966 uint64_t count;
1967 ASSERT(0 == zap_count(mos,
1968 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1969 VERIFY(0 == dmu_object_free(mos,
1970 ds->ds_phys->ds_next_clones_obj, tx));
1971 }
1972 if (ds->ds_phys->ds_props_obj != 0)
1973 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1974 if (ds->ds_phys->ds_userrefs_obj != 0)
1975 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1976 dsl_dir_close(ds->ds_dir, ds);
1977 ds->ds_dir = NULL;
1978 dsl_dataset_drain_refs(ds, tag);
1979 VERIFY(0 == dmu_object_free(mos, obj, tx));
1980
1981 if (dsda->rm_origin) {
1982 /*
1983 * Remove the origin of the clone we just destroyed.
1984 */
1985 struct dsl_ds_destroyarg ndsda = {0};
1986
1987 ndsda.ds = dsda->rm_origin;
1988 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1989 }
1990 }
1991
1992 static int
1993 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1994 {
1995 uint64_t asize;
1996
1997 if (!dmu_tx_is_syncing(tx))
1998 return (0);
1999
2000 /*
2001 * If there's an fs-only reservation, any blocks that might become
2002 * owned by the snapshot dataset must be accommodated by space
2003 * outside of the reservation.
2004 */
2005 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2006 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2007 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2008 return (ENOSPC);
2009
2010 /*
2011 * Propogate any reserved space for this snapshot to other
2012 * snapshot checks in this sync group.
2013 */
2014 if (asize > 0)
2015 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2016
2017 return (0);
2018 }
2019
2020 int
2021 dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
2022 {
2023 dsl_dataset_t *ds = arg1;
2024 const char *snapname = arg2;
2025 int err;
2026 uint64_t value;
2027
2028 /*
2029 * We don't allow multiple snapshots of the same txg. If there
2030 * is already one, try again.
2031 */
2032 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2033 return (EAGAIN);
2034
2035 /*
2036 * Check for conflicting name snapshot name.
2037 */
2038 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2039 if (err == 0)
2040 return (EEXIST);
2041 if (err != ENOENT)
2042 return (err);
2043
2044 /*
2045 * Check that the dataset's name is not too long. Name consists
2046 * of the dataset's length + 1 for the @-sign + snapshot name's length
2047 */
2048 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2049 return (ENAMETOOLONG);
2050
2051 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2052 if (err)
2053 return (err);
2054
2055 ds->ds_trysnap_txg = tx->tx_txg;
2056 return (0);
2057 }
2058
2059 void
2060 dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2061 {
2062 dsl_dataset_t *ds = arg1;
2063 const char *snapname = arg2;
2064 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2065 dmu_buf_t *dbuf;
2066 dsl_dataset_phys_t *dsphys;
2067 uint64_t dsobj, crtxg;
2068 objset_t *mos = dp->dp_meta_objset;
2069 int err;
2070
2071 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2072
2073 /*
2074 * The origin's ds_creation_txg has to be < TXG_INITIAL
2075 */
2076 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2077 crtxg = 1;
2078 else
2079 crtxg = tx->tx_txg;
2080
2081 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2082 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2083 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2084 dmu_buf_will_dirty(dbuf, tx);
2085 dsphys = dbuf->db_data;
2086 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2087 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2088 dsphys->ds_fsid_guid = unique_create();
2089 do {
2090 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2091 sizeof (dsphys->ds_guid));
2092 } while (dsphys->ds_guid == 0);
2093 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2094 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2095 dsphys->ds_next_snap_obj = ds->ds_object;
2096 dsphys->ds_num_children = 1;
2097 dsphys->ds_creation_time = gethrestime_sec();
2098 dsphys->ds_creation_txg = crtxg;
2099 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2100 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2101 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2102 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2103 dsphys->ds_flags = ds->ds_phys->ds_flags;
2104 dsphys->ds_bp = ds->ds_phys->ds_bp;
2105 dmu_buf_rele(dbuf, FTAG);
2106
2107 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2108 if (ds->ds_prev) {
2109 uint64_t next_clones_obj =
2110 ds->ds_prev->ds_phys->ds_next_clones_obj;
2111 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2112 ds->ds_object ||
2113 ds->ds_prev->ds_phys->ds_num_children > 1);
2114 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2115 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2116 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2117 ds->ds_prev->ds_phys->ds_creation_txg);
2118 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2119 } else if (next_clones_obj != 0) {
2120 remove_from_next_clones(ds->ds_prev,
2121 dsphys->ds_next_snap_obj, tx);
2122 VERIFY3U(0, ==, zap_add_int(mos,
2123 next_clones_obj, dsobj, tx));
2124 }
2125 }
2126
2127 /*
2128 * If we have a reference-reservation on this dataset, we will
2129 * need to increase the amount of refreservation being charged
2130 * since our unique space is going to zero.
2131 */
2132 if (ds->ds_reserved) {
2133 int64_t delta;
2134 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2135 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2136 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2137 delta, 0, 0, tx);
2138 }
2139
2140 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2141 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2142 ds->ds_dir->dd_myname, snapname, dsobj,
2143 ds->ds_phys->ds_prev_snap_txg);
2144 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2145 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2146 dsl_deadlist_close(&ds->ds_deadlist);
2147 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2148 dsl_deadlist_add_key(&ds->ds_deadlist,
2149 ds->ds_phys->ds_prev_snap_txg, tx);
2150
2151 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2152 ds->ds_phys->ds_prev_snap_obj = dsobj;
2153 ds->ds_phys->ds_prev_snap_txg = crtxg;
2154 ds->ds_phys->ds_unique_bytes = 0;
2155 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2156 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2157
2158 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2159 snapname, 8, 1, &dsobj, tx);
2160 ASSERT(err == 0);
2161
2162 if (ds->ds_prev)
2163 dsl_dataset_drop_ref(ds->ds_prev, ds);
2164 VERIFY(0 == dsl_dataset_get_ref(dp,
2165 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2166
2167 dsl_scan_ds_snapshotted(ds, tx);
2168
2169 dsl_dir_snap_cmtime_update(ds->ds_dir);
2170
2171 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
2172 "dataset = %llu", dsobj);
2173 }
2174
2175 void
2176 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2177 {
2178 ASSERT(dmu_tx_is_syncing(tx));
2179 ASSERT(ds->ds_objset != NULL);
2180 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2181
2182 /*
2183 * in case we had to change ds_fsid_guid when we opened it,
2184 * sync it out now.
2185 */
2186 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2187 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2188
2189 dsl_dir_dirty(ds->ds_dir, tx);
2190 dmu_objset_sync(ds->ds_objset, zio, tx);
2191 }
2192
2193 static void
2194 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2195 {
2196 uint64_t count = 0;
2197 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2198 zap_cursor_t zc;
2199 zap_attribute_t za;
2200 nvlist_t *propval;
2201 nvlist_t *val;
2202
2203 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2204 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2205 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2206
2207 /*
2208 * There may me missing entries in ds_next_clones_obj
2209 * due to a bug in a previous version of the code.
2210 * Only trust it if it has the right number of entries.
2211 */
2212 if (ds->ds_phys->ds_next_clones_obj != 0) {
2213 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2214 &count));
2215 }
2216 if (count != ds->ds_phys->ds_num_children - 1) {
2217 goto fail;
2218 }
2219 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2220 zap_cursor_retrieve(&zc, &za) == 0;
2221 zap_cursor_advance(&zc)) {
2222 dsl_dataset_t *clone;
2223 char buf[ZFS_MAXNAMELEN];
2224 /*
2225 * Even though we hold the dp_config_rwlock, the dataset
2226 * may fail to open, returning ENOENT. If there is a
2227 * thread concurrently attempting to destroy this
2228 * dataset, it will have the ds_rwlock held for
2229 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2230 * dsl_dataset_hold_ref() will fail its
2231 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2232 * dp_config_rwlock, and wait for the destroy progress
2233 * and signal ds_exclusive_cv. If the destroy was
2234 * successful, we will see that
2235 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2236 */
2237 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2238 za.za_first_integer, FTAG, &clone) != 0)
2239 continue;
2240 dsl_dir_name(clone->ds_dir, buf);
2241 VERIFY(nvlist_add_boolean(val, buf) == 0);
2242 dsl_dataset_rele(clone, FTAG);
2243 }
2244 zap_cursor_fini(&zc);
2245 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2246 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2247 propval) == 0);
2248 fail:
2249 nvlist_free(val);
2250 nvlist_free(propval);
2251 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2252 }
2253
2254 void
2255 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2256 {
2257 uint64_t refd, avail, uobjs, aobjs, ratio;
2258
2259 dsl_dir_stats(ds->ds_dir, nv);
2260
2261 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2262 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2263 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2264
2265 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2266 ds->ds_phys->ds_creation_time);
2267 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2268 ds->ds_phys->ds_creation_txg);
2269 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2270 ds->ds_quota);
2271 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2272 ds->ds_reserved);
2273 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2274 ds->ds_phys->ds_guid);
2275 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2276 ds->ds_phys->ds_unique_bytes);
2277 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2278 ds->ds_object);
2279 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2280 ds->ds_userrefs);
2281 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2282 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2283
2284 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2285 uint64_t written, comp, uncomp;
2286 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2287 dsl_dataset_t *prev;
2288
2289 rw_enter(&dp->dp_config_rwlock, RW_READER);
2290 int err = dsl_dataset_hold_obj(dp,
2291 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2292 rw_exit(&dp->dp_config_rwlock);
2293 if (err == 0) {
2294 err = dsl_dataset_space_written(prev, ds, &written,
2295 &comp, &uncomp);
2296 dsl_dataset_rele(prev, FTAG);
2297 if (err == 0) {
2298 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2299 written);
2300 }
2301 }
2302 }
2303
2304 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2305 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2306 ds->ds_phys->ds_compressed_bytes);
2307 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2308
2309 if (ds->ds_phys->ds_next_snap_obj) {
2310 /*
2311 * This is a snapshot; override the dd's space used with
2312 * our unique space and compression ratio.
2313 */
2314 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2315 ds->ds_phys->ds_unique_bytes);
2316 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2317
2318 get_clones_stat(ds, nv);
2319 }
2320 }
2321
2322 void
2323 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2324 {
2325 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2326 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2327 stat->dds_guid = ds->ds_phys->ds_guid;
2328 if (ds->ds_phys->ds_next_snap_obj) {
2329 stat->dds_is_snapshot = B_TRUE;
2330 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2331 } else {
2332 stat->dds_is_snapshot = B_FALSE;
2333 stat->dds_num_clones = 0;
2334 }
2335
2336 /* clone origin is really a dsl_dir thing... */
2337 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2338 if (dsl_dir_is_clone(ds->ds_dir)) {
2339 dsl_dataset_t *ods;
2340
2341 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2342 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2343 dsl_dataset_name(ods, stat->dds_origin);
2344 dsl_dataset_drop_ref(ods, FTAG);
2345 } else {
2346 stat->dds_origin[0] = '\0';
2347 }
2348 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2349 }
2350
2351 uint64_t
2352 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2353 {
2354 return (ds->ds_fsid_guid);
2355 }
2356
2357 void
2358 dsl_dataset_space(dsl_dataset_t *ds,
2359 uint64_t *refdbytesp, uint64_t *availbytesp,
2360 uint64_t *usedobjsp, uint64_t *availobjsp)
2361 {
2362 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2363 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2364 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2365 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2366 if (ds->ds_quota != 0) {
2367 /*
2368 * Adjust available bytes according to refquota
2369 */
2370 if (*refdbytesp < ds->ds_quota)
2371 *availbytesp = MIN(*availbytesp,
2372 ds->ds_quota - *refdbytesp);
2373 else
2374 *availbytesp = 0;
2375 }
2376 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2377 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2378 }
2379
2380 boolean_t
2381 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2382 {
2383 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2384
2385 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2386 dsl_pool_sync_context(dp));
2387 if (ds->ds_prev == NULL)
2388 return (B_FALSE);
2389 if (ds->ds_phys->ds_bp.blk_birth >
2390 ds->ds_prev->ds_phys->ds_creation_txg) {
2391 objset_t *os, *os_prev;
2392 /*
2393 * It may be that only the ZIL differs, because it was
2394 * reset in the head. Don't count that as being
2395 * modified.
2396 */
2397 if (dmu_objset_from_ds(ds, &os) != 0)
2398 return (B_TRUE);
2399 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2400 return (B_TRUE);
2401 return (bcmp(&os->os_phys->os_meta_dnode,
2402 &os_prev->os_phys->os_meta_dnode,
2403 sizeof (os->os_phys->os_meta_dnode)) != 0);
2404 }
2405 return (B_FALSE);
2406 }
2407
2408 /* ARGSUSED */
2409 static int
2410 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2411 {
2412 dsl_dataset_t *ds = arg1;
2413 char *newsnapname = arg2;
2414 dsl_dir_t *dd = ds->ds_dir;
2415 dsl_dataset_t *hds;
2416 uint64_t val;
2417 int err;
2418
2419 err = dsl_dataset_hold_obj(dd->dd_pool,
2420 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2421 if (err)
2422 return (err);
2423
2424 /* new name better not be in use */
2425 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2426 dsl_dataset_rele(hds, FTAG);
2427
2428 if (err == 0)
2429 err = EEXIST;
2430 else if (err == ENOENT)
2431 err = 0;
2432
2433 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2434 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2435 err = ENAMETOOLONG;
2436
2437 return (err);
2438 }
2439
2440 static void
2441 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2442 {
2443 dsl_dataset_t *ds = arg1;
2444 const char *newsnapname = arg2;
2445 dsl_dir_t *dd = ds->ds_dir;
2446 objset_t *mos = dd->dd_pool->dp_meta_objset;
2447 dsl_dataset_t *hds;
2448 int err;
2449
2450 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2451
2452 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2453 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2454
2455 VERIFY(0 == dsl_dataset_get_snapname(ds));
2456 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2457 ASSERT3U(err, ==, 0);
2458 mutex_enter(&ds->ds_lock);
2459 (void) strcpy(ds->ds_snapname, newsnapname);
2460 mutex_exit(&ds->ds_lock);
2461 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2462 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2463 ASSERT3U(err, ==, 0);
2464
2465 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2466 "dataset = %llu", ds->ds_object);
2467 dsl_dataset_rele(hds, FTAG);
2468 }
2469
2470 struct renamesnaparg {
2471 dsl_sync_task_group_t *dstg;
2472 char failed[MAXPATHLEN];
2473 char *oldsnap;
2474 char *newsnap;
2475 };
2476
2477 static int
2478 dsl_snapshot_rename_one(const char *name, void *arg)
2479 {
2480 struct renamesnaparg *ra = arg;
2481 dsl_dataset_t *ds = NULL;
2482 char *snapname;
2483 int err;
2484
2485 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2486 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2487
2488 /*
2489 * For recursive snapshot renames the parent won't be changing
2490 * so we just pass name for both the to/from argument.
2491 */
2492 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2493 if (err != 0) {
2494 strfree(snapname);
2495 return (err == ENOENT ? 0 : err);
2496 }
2497
2498 #ifdef _KERNEL
2499 /*
2500 * For all filesystems undergoing rename, we'll need to unmount it.
2501 */
2502 (void) zfs_unmount_snap(snapname, NULL);
2503 #endif
2504 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2505 strfree(snapname);
2506 if (err != 0)
2507 return (err == ENOENT ? 0 : err);
2508
2509 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2510 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2511
2512 return (0);
2513 }
2514
2515 static int
2516 dsl_recursive_rename(char *oldname, const char *newname)
2517 {
2518 int err;
2519 struct renamesnaparg *ra;
2520 dsl_sync_task_t *dst;
2521 spa_t *spa;
2522 char *cp, *fsname = spa_strdup(oldname);
2523 int len = strlen(oldname) + 1;
2524
2525 /* truncate the snapshot name to get the fsname */
2526 cp = strchr(fsname, '@');
2527 *cp = '\0';
2528
2529 err = spa_open(fsname, &spa, FTAG);
2530 if (err) {
2531 kmem_free(fsname, len);
2532 return (err);
2533 }
2534 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2535 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2536
2537 ra->oldsnap = strchr(oldname, '@') + 1;
2538 ra->newsnap = strchr(newname, '@') + 1;
2539 *ra->failed = '\0';
2540
2541 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2542 DS_FIND_CHILDREN);
2543 kmem_free(fsname, len);
2544
2545 if (err == 0) {
2546 err = dsl_sync_task_group_wait(ra->dstg);
2547 }
2548
2549 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2550 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2551 dsl_dataset_t *ds = dst->dst_arg1;
2552 if (dst->dst_err) {
2553 dsl_dir_name(ds->ds_dir, ra->failed);
2554 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2555 (void) strlcat(ra->failed, ra->newsnap,
2556 sizeof (ra->failed));
2557 }
2558 dsl_dataset_rele(ds, ra->dstg);
2559 }
2560
2561 if (err)
2562 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2563
2564 dsl_sync_task_group_destroy(ra->dstg);
2565 kmem_free(ra, sizeof (struct renamesnaparg));
2566 spa_close(spa, FTAG);
2567 return (err);
2568 }
2569
2570 static int
2571 dsl_valid_rename(const char *oldname, void *arg)
2572 {
2573 int delta = *(int *)arg;
2574
2575 if (strlen(oldname) + delta >= MAXNAMELEN)
2576 return (ENAMETOOLONG);
2577
2578 return (0);
2579 }
2580
2581 #pragma weak dmu_objset_rename = dsl_dataset_rename
2582 int
2583 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2584 {
2585 dsl_dir_t *dd;
2586 dsl_dataset_t *ds;
2587 const char *tail;
2588 int err;
2589
2590 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2591 if (err)
2592 return (err);
2593
2594 if (tail == NULL) {
2595 int delta = strlen(newname) - strlen(oldname);
2596
2597 /* if we're growing, validate child name lengths */
2598 if (delta > 0)
2599 err = dmu_objset_find(oldname, dsl_valid_rename,
2600 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2601
2602 if (err == 0)
2603 err = dsl_dir_rename(dd, newname);
2604 dsl_dir_close(dd, FTAG);
2605 return (err);
2606 }
2607
2608 if (tail[0] != '@') {
2609 /* the name ended in a nonexistent component */
2610 dsl_dir_close(dd, FTAG);
2611 return (ENOENT);
2612 }
2613
2614 dsl_dir_close(dd, FTAG);
2615
2616 /* new name must be snapshot in same filesystem */
2617 tail = strchr(newname, '@');
2618 if (tail == NULL)
2619 return (EINVAL);
2620 tail++;
2621 if (strncmp(oldname, newname, tail - newname) != 0)
2622 return (EXDEV);
2623
2624 if (recursive) {
2625 err = dsl_recursive_rename(oldname, newname);
2626 } else {
2627 err = dsl_dataset_hold(oldname, FTAG, &ds);
2628 if (err)
2629 return (err);
2630
2631 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2632 dsl_dataset_snapshot_rename_check,
2633 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2634
2635 dsl_dataset_rele(ds, FTAG);
2636 }
2637
2638 return (err);
2639 }
2640
2641 struct promotenode {
2642 list_node_t link;
2643 dsl_dataset_t *ds;
2644 };
2645
2646 struct promotearg {
2647 list_t shared_snaps, origin_snaps, clone_snaps;
2648 dsl_dataset_t *origin_origin;
2649 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2650 char *err_ds;
2651 };
2652
2653 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2654 static boolean_t snaplist_unstable(list_t *l);
2655
2656 static int
2657 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2658 {
2659 dsl_dataset_t *hds = arg1;
2660 struct promotearg *pa = arg2;
2661 struct promotenode *snap = list_head(&pa->shared_snaps);
2662 dsl_dataset_t *origin_ds = snap->ds;
2663 int err;
2664 uint64_t unused;
2665
2666 /* Check that it is a real clone */
2667 if (!dsl_dir_is_clone(hds->ds_dir))
2668 return (EINVAL);
2669
2670 /* Since this is so expensive, don't do the preliminary check */
2671 if (!dmu_tx_is_syncing(tx))
2672 return (0);
2673
2674 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2675 return (EXDEV);
2676
2677 /* compute origin's new unique space */
2678 snap = list_tail(&pa->clone_snaps);
2679 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2680 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2681 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2682 &pa->unique, &unused, &unused);
2683
2684 /*
2685 * Walk the snapshots that we are moving
2686 *
2687 * Compute space to transfer. Consider the incremental changes
2688 * to used for each snapshot:
2689 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2690 * So each snapshot gave birth to:
2691 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2692 * So a sequence would look like:
2693 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2694 * Which simplifies to:
2695 * uN + kN + kN-1 + ... + k1 + k0
2696 * Note however, if we stop before we reach the ORIGIN we get:
2697 * uN + kN + kN-1 + ... + kM - uM-1
2698 */
2699 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2700 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2701 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2702 for (snap = list_head(&pa->shared_snaps); snap;
2703 snap = list_next(&pa->shared_snaps, snap)) {
2704 uint64_t val, dlused, dlcomp, dluncomp;
2705 dsl_dataset_t *ds = snap->ds;
2706
2707 /* Check that the snapshot name does not conflict */
2708 VERIFY(0 == dsl_dataset_get_snapname(ds));
2709 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2710 if (err == 0) {
2711 err = EEXIST;
2712 goto out;
2713 }
2714 if (err != ENOENT)
2715 goto out;
2716
2717 /* The very first snapshot does not have a deadlist */
2718 if (ds->ds_phys->ds_prev_snap_obj == 0)
2719 continue;
2720
2721 dsl_deadlist_space(&ds->ds_deadlist,
2722 &dlused, &dlcomp, &dluncomp);
2723 pa->used += dlused;
2724 pa->comp += dlcomp;
2725 pa->uncomp += dluncomp;
2726 }
2727
2728 /*
2729 * If we are a clone of a clone then we never reached ORIGIN,
2730 * so we need to subtract out the clone origin's used space.
2731 */
2732 if (pa->origin_origin) {
2733 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2734 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2735 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2736 }
2737
2738 /* Check that there is enough space here */
2739 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2740 pa->used);
2741 if (err)
2742 return (err);
2743
2744 /*
2745 * Compute the amounts of space that will be used by snapshots
2746 * after the promotion (for both origin and clone). For each,
2747 * it is the amount of space that will be on all of their
2748 * deadlists (that was not born before their new origin).
2749 */
2750 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2751 uint64_t space;
2752
2753 /*
2754 * Note, typically this will not be a clone of a clone,
2755 * so dd_origin_txg will be < TXG_INITIAL, so
2756 * these snaplist_space() -> dsl_deadlist_space_range()
2757 * calls will be fast because they do not have to
2758 * iterate over all bps.
2759 */
2760 snap = list_head(&pa->origin_snaps);
2761 err = snaplist_space(&pa->shared_snaps,
2762 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2763 if (err)
2764 return (err);
2765
2766 err = snaplist_space(&pa->clone_snaps,
2767 snap->ds->ds_dir->dd_origin_txg, &space);
2768 if (err)
2769 return (err);
2770 pa->cloneusedsnap += space;
2771 }
2772 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2773 err = snaplist_space(&pa->origin_snaps,
2774 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2775 if (err)
2776 return (err);
2777 }
2778
2779 return (0);
2780 out:
2781 pa->err_ds = snap->ds->ds_snapname;
2782 return (err);
2783 }
2784
2785 static void
2786 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2787 {
2788 dsl_dataset_t *hds = arg1;
2789 struct promotearg *pa = arg2;
2790 struct promotenode *snap = list_head(&pa->shared_snaps);
2791 dsl_dataset_t *origin_ds = snap->ds;
2792 dsl_dataset_t *origin_head;
2793 dsl_dir_t *dd = hds->ds_dir;
2794 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2795 dsl_dir_t *odd = NULL;
2796 uint64_t oldnext_obj;
2797 int64_t delta;
2798
2799 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2800
2801 snap = list_head(&pa->origin_snaps);
2802 origin_head = snap->ds;
2803
2804 /*
2805 * We need to explicitly open odd, since origin_ds's dd will be
2806 * changing.
2807 */
2808 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2809 NULL, FTAG, &odd));
2810
2811 /* change origin's next snap */
2812 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2813 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2814 snap = list_tail(&pa->clone_snaps);
2815 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2816 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2817
2818 /* change the origin's next clone */
2819 if (origin_ds->ds_phys->ds_next_clones_obj) {
2820 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2821 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2822 origin_ds->ds_phys->ds_next_clones_obj,
2823 oldnext_obj, tx));
2824 }
2825
2826 /* change origin */
2827 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2828 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2829 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2830 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2831 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2832 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2833 origin_head->ds_dir->dd_origin_txg =
2834 origin_ds->ds_phys->ds_creation_txg;
2835
2836 /* change dd_clone entries */
2837 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2838 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2839 odd->dd_phys->dd_clones, hds->ds_object, tx));
2840 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2841 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2842 hds->ds_object, tx));
2843
2844 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2845 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2846 origin_head->ds_object, tx));
2847 if (dd->dd_phys->dd_clones == 0) {
2848 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2849 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2850 }
2851 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2852 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2853
2854 }
2855
2856 /* move snapshots to this dir */
2857 for (snap = list_head(&pa->shared_snaps); snap;
2858 snap = list_next(&pa->shared_snaps, snap)) {
2859 dsl_dataset_t *ds = snap->ds;
2860
2861 /* unregister props as dsl_dir is changing */
2862 if (ds->ds_objset) {
2863 dmu_objset_evict(ds->ds_objset);
2864 ds->ds_objset = NULL;
2865 }
2866 /* move snap name entry */
2867 VERIFY(0 == dsl_dataset_get_snapname(ds));
2868 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2869 ds->ds_snapname, tx));
2870 VERIFY(0 == zap_add(dp->dp_meta_objset,
2871 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2872 8, 1, &ds->ds_object, tx));
2873
2874 /* change containing dsl_dir */
2875 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2876 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2877 ds->ds_phys->ds_dir_obj = dd->dd_object;
2878 ASSERT3P(ds->ds_dir, ==, odd);
2879 dsl_dir_close(ds->ds_dir, ds);
2880 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2881 NULL, ds, &ds->ds_dir));
2882
2883 /* move any clone references */
2884 if (ds->ds_phys->ds_next_clones_obj &&
2885 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2886 zap_cursor_t zc;
2887 zap_attribute_t za;
2888
2889 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2890 ds->ds_phys->ds_next_clones_obj);
2891 zap_cursor_retrieve(&zc, &za) == 0;
2892 zap_cursor_advance(&zc)) {
2893 dsl_dataset_t *cnds;
2894 uint64_t o;
2895
2896 if (za.za_first_integer == oldnext_obj) {
2897 /*
2898 * We've already moved the
2899 * origin's reference.
2900 */
2901 continue;
2902 }
2903
2904 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2905 za.za_first_integer, FTAG, &cnds));
2906 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2907
2908 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2909 odd->dd_phys->dd_clones, o, tx), ==, 0);
2910 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2911 dd->dd_phys->dd_clones, o, tx), ==, 0);
2912 dsl_dataset_rele(cnds, FTAG);
2913 }
2914 zap_cursor_fini(&zc);
2915 }
2916
2917 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
2918 }
2919
2920 /*
2921 * Change space accounting.
2922 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2923 * both be valid, or both be 0 (resulting in delta == 0). This
2924 * is true for each of {clone,origin} independently.
2925 */
2926
2927 delta = pa->cloneusedsnap -
2928 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2929 ASSERT3S(delta, >=, 0);
2930 ASSERT3U(pa->used, >=, delta);
2931 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2932 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2933 pa->used - delta, pa->comp, pa->uncomp, tx);
2934
2935 delta = pa->originusedsnap -
2936 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2937 ASSERT3S(delta, <=, 0);
2938 ASSERT3U(pa->used, >=, -delta);
2939 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2940 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2941 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2942
2943 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2944
2945 /* log history record */
2946 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2947 "dataset = %llu", hds->ds_object);
2948
2949 dsl_dir_close(odd, FTAG);
2950 }
2951
2952 static char *snaplist_tag = "snaplist";
2953 /*
2954 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2955 * (exclusive) and last_obj (inclusive). The list will be in reverse
2956 * order (last_obj will be the list_head()). If first_obj == 0, do all
2957 * snapshots back to this dataset's origin.
2958 */
2959 static int
2960 snaplist_make(dsl_pool_t *dp, boolean_t own,
2961 uint64_t first_obj, uint64_t last_obj, list_t *l)
2962 {
2963 uint64_t obj = last_obj;
2964
2965 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2966
2967 list_create(l, sizeof (struct promotenode),
2968 offsetof(struct promotenode, link));
2969
2970 while (obj != first_obj) {
2971 dsl_dataset_t *ds;
2972 struct promotenode *snap;
2973 int err;
2974
2975 if (own) {
2976 err = dsl_dataset_own_obj(dp, obj,
2977 0, snaplist_tag, &ds);
2978 if (err == 0)
2979 dsl_dataset_make_exclusive(ds, snaplist_tag);
2980 } else {
2981 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2982 }
2983 if (err == ENOENT) {
2984 /* lost race with snapshot destroy */
2985 struct promotenode *last = list_tail(l);
2986 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2987 obj = last->ds->ds_phys->ds_prev_snap_obj;
2988 continue;
2989 } else if (err) {
2990 return (err);
2991 }
2992
2993 if (first_obj == 0)
2994 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2995
2996 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2997 snap->ds = ds;
2998 list_insert_tail(l, snap);
2999 obj = ds->ds_phys->ds_prev_snap_obj;
3000 }
3001
3002 return (0);
3003 }
3004
3005 static int
3006 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3007 {
3008 struct promotenode *snap;
3009
3010 *spacep = 0;
3011 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3012 uint64_t used, comp, uncomp;
3013 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3014 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3015 *spacep += used;
3016 }
3017 return (0);
3018 }
3019
3020 static void
3021 snaplist_destroy(list_t *l, boolean_t own)
3022 {
3023 struct promotenode *snap;
3024
3025 if (!l || !list_link_active(&l->list_head))
3026 return;
3027
3028 while ((snap = list_tail(l)) != NULL) {
3029 list_remove(l, snap);
3030 if (own)
3031 dsl_dataset_disown(snap->ds, snaplist_tag);
3032 else
3033 dsl_dataset_rele(snap->ds, snaplist_tag);
3034 kmem_free(snap, sizeof (struct promotenode));
3035 }
3036 list_destroy(l);
3037 }
3038
3039 /*
3040 * Promote a clone. Nomenclature note:
3041 * "clone" or "cds": the original clone which is being promoted
3042 * "origin" or "ods": the snapshot which is originally clone's origin
3043 * "origin head" or "ohds": the dataset which is the head
3044 * (filesystem/volume) for the origin
3045 * "origin origin": the origin of the origin's filesystem (typically
3046 * NULL, indicating that the clone is not a clone of a clone).
3047 */
3048 int
3049 dsl_dataset_promote(const char *name, char *conflsnap)
3050 {
3051 dsl_dataset_t *ds;
3052 dsl_dir_t *dd;
3053 dsl_pool_t *dp;
3054 dmu_object_info_t doi;
3055 struct promotearg pa = { 0 };
3056 struct promotenode *snap;
3057 int err;
3058
3059 err = dsl_dataset_hold(name, FTAG, &ds);
3060 if (err)
3061 return (err);
3062 dd = ds->ds_dir;
3063 dp = dd->dd_pool;
3064
3065 err = dmu_object_info(dp->dp_meta_objset,
3066 ds->ds_phys->ds_snapnames_zapobj, &doi);
3067 if (err) {
3068 dsl_dataset_rele(ds, FTAG);
3069 return (err);
3070 }
3071
3072 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3073 dsl_dataset_rele(ds, FTAG);
3074 return (EINVAL);
3075 }
3076
3077 /*
3078 * We are going to inherit all the snapshots taken before our
3079 * origin (i.e., our new origin will be our parent's origin).
3080 * Take ownership of them so that we can rename them into our
3081 * namespace.
3082 */
3083 rw_enter(&dp->dp_config_rwlock, RW_READER);
3084
3085 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3086 &pa.shared_snaps);
3087 if (err != 0)
3088 goto out;
3089
3090 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3091 if (err != 0)
3092 goto out;
3093
3094 snap = list_head(&pa.shared_snaps);
3095 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3096 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3097 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3098 if (err != 0)
3099 goto out;
3100
3101 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3102 err = dsl_dataset_hold_obj(dp,
3103 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3104 FTAG, &pa.origin_origin);
3105 if (err != 0)
3106 goto out;
3107 }
3108
3109 out:
3110 rw_exit(&dp->dp_config_rwlock);
3111
3112 /*
3113 * Add in 128x the snapnames zapobj size, since we will be moving
3114 * a bunch of snapnames to the promoted ds, and dirtying their
3115 * bonus buffers.
3116 */
3117 if (err == 0) {
3118 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3119 dsl_dataset_promote_sync, ds, &pa,
3120 2 + 2 * doi.doi_physical_blocks_512);
3121 if (err && pa.err_ds && conflsnap)
3122 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3123 }
3124
3125 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3126 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3127 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3128 if (pa.origin_origin)
3129 dsl_dataset_rele(pa.origin_origin, FTAG);
3130 dsl_dataset_rele(ds, FTAG);
3131 return (err);
3132 }
3133
3134 struct cloneswaparg {
3135 dsl_dataset_t *cds; /* clone dataset */
3136 dsl_dataset_t *ohds; /* origin's head dataset */
3137 boolean_t force;
3138 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3139 };
3140
3141 /* ARGSUSED */
3142 static int
3143 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3144 {
3145 struct cloneswaparg *csa = arg1;
3146
3147 /* they should both be heads */
3148 if (dsl_dataset_is_snapshot(csa->cds) ||
3149 dsl_dataset_is_snapshot(csa->ohds))
3150 return (EINVAL);
3151
3152 /* the branch point should be just before them */
3153 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3154 return (EINVAL);
3155
3156 /* cds should be the clone (unless they are unrelated) */
3157 if (csa->cds->ds_prev != NULL &&
3158 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3159 csa->ohds->ds_object !=
3160 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3161 return (EINVAL);
3162
3163 /* the clone should be a child of the origin */
3164 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3165 return (EINVAL);
3166
3167 /* ohds shouldn't be modified unless 'force' */
3168 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3169 return (ETXTBSY);
3170
3171 /* adjust amount of any unconsumed refreservation */
3172 csa->unused_refres_delta =
3173 (int64_t)MIN(csa->ohds->ds_reserved,
3174 csa->ohds->ds_phys->ds_unique_bytes) -
3175 (int64_t)MIN(csa->ohds->ds_reserved,
3176 csa->cds->ds_phys->ds_unique_bytes);
3177
3178 if (csa->unused_refres_delta > 0 &&
3179 csa->unused_refres_delta >
3180 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3181 return (ENOSPC);
3182
3183 if (csa->ohds->ds_quota != 0 &&
3184 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3185 return (EDQUOT);
3186
3187 return (0);
3188 }
3189
3190 /* ARGSUSED */
3191 static void
3192 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3193 {
3194 struct cloneswaparg *csa = arg1;
3195 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3196
3197 ASSERT(csa->cds->ds_reserved == 0);
3198 ASSERT(csa->ohds->ds_quota == 0 ||
3199 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3200
3201 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3202 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3203
3204 if (csa->cds->ds_objset != NULL) {
3205 dmu_objset_evict(csa->cds->ds_objset);
3206 csa->cds->ds_objset = NULL;
3207 }
3208
3209 if (csa->ohds->ds_objset != NULL) {
3210 dmu_objset_evict(csa->ohds->ds_objset);
3211 csa->ohds->ds_objset = NULL;
3212 }
3213
3214 /*
3215 * Reset origin's unique bytes, if it exists.
3216 */
3217 if (csa->cds->ds_prev) {
3218 dsl_dataset_t *origin = csa->cds->ds_prev;
3219 uint64_t comp, uncomp;
3220
3221 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3222 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3223 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3224 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3225 }
3226
3227 /* swap blkptrs */
3228 {
3229 blkptr_t tmp;
3230 tmp = csa->ohds->ds_phys->ds_bp;
3231 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3232 csa->cds->ds_phys->ds_bp = tmp;
3233 }
3234
3235 /* set dd_*_bytes */
3236 {
3237 int64_t dused, dcomp, duncomp;
3238 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3239 uint64_t odl_used, odl_comp, odl_uncomp;
3240
3241 ASSERT3U(csa->cds->ds_dir->dd_phys->
3242 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3243
3244 dsl_deadlist_space(&csa->cds->ds_deadlist,
3245 &cdl_used, &cdl_comp, &cdl_uncomp);
3246 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3247 &odl_used, &odl_comp, &odl_uncomp);
3248
3249 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3250 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3251 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3252 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3253 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3254 cdl_uncomp -
3255 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3256
3257 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3258 dused, dcomp, duncomp, tx);
3259 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3260 -dused, -dcomp, -duncomp, tx);
3261
3262 /*
3263 * The difference in the space used by snapshots is the
3264 * difference in snapshot space due to the head's
3265 * deadlist (since that's the only thing that's
3266 * changing that affects the snapused).
3267 */
3268 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3269 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3270 &cdl_used, &cdl_comp, &cdl_uncomp);
3271 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3272 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3273 &odl_used, &odl_comp, &odl_uncomp);
3274 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3275 DD_USED_HEAD, DD_USED_SNAP, tx);
3276 }
3277
3278 /* swap ds_*_bytes */
3279 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3280 csa->cds->ds_phys->ds_referenced_bytes);
3281 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3282 csa->cds->ds_phys->ds_compressed_bytes);
3283 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3284 csa->cds->ds_phys->ds_uncompressed_bytes);
3285 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3286 csa->cds->ds_phys->ds_unique_bytes);
3287
3288 /* apply any parent delta for change in unconsumed refreservation */
3289 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3290 csa->unused_refres_delta, 0, 0, tx);
3291
3292 /*
3293 * Swap deadlists.
3294 */
3295 dsl_deadlist_close(&csa->cds->ds_deadlist);
3296 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3297 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3298 csa->cds->ds_phys->ds_deadlist_obj);
3299 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3300 csa->cds->ds_phys->ds_deadlist_obj);
3301 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3302 csa->ohds->ds_phys->ds_deadlist_obj);
3303
3304 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3305 }
3306
3307 /*
3308 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3309 * recv" into an existing fs to swizzle the file system to the new
3310 * version, and by "zfs rollback". Can also be used to swap two
3311 * independent head datasets if neither has any snapshots.
3312 */
3313 int
3314 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3315 boolean_t force)
3316 {
3317 struct cloneswaparg csa;
3318 int error;
3319
3320 ASSERT(clone->ds_owner);
3321 ASSERT(origin_head->ds_owner);
3322 retry:
3323 /*
3324 * Need exclusive access for the swap. If we're swapping these
3325 * datasets back after an error, we already hold the locks.
3326 */
3327 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3328 rw_enter(&clone->ds_rwlock, RW_WRITER);
3329 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3330 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3331 rw_exit(&clone->ds_rwlock);
3332 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3333 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3334 rw_exit(&origin_head->ds_rwlock);
3335 goto retry;
3336 }
3337 }
3338 csa.cds = clone;
3339 csa.ohds = origin_head;
3340 csa.force = force;
3341 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3342 dsl_dataset_clone_swap_check,
3343 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3344 return (error);
3345 }
3346
3347 /*
3348 * Given a pool name and a dataset object number in that pool,
3349 * return the name of that dataset.
3350 */
3351 int
3352 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3353 {
3354 spa_t *spa;
3355 dsl_pool_t *dp;
3356 dsl_dataset_t *ds;
3357 int error;
3358
3359 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3360 return (error);
3361 dp = spa_get_dsl(spa);
3362 rw_enter(&dp->dp_config_rwlock, RW_READER);
3363 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3364 dsl_dataset_name(ds, buf);
3365 dsl_dataset_rele(ds, FTAG);
3366 }
3367 rw_exit(&dp->dp_config_rwlock);
3368 spa_close(spa, FTAG);
3369
3370 return (error);
3371 }
3372
3373 int
3374 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3375 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3376 {
3377 int error = 0;
3378
3379 ASSERT3S(asize, >, 0);
3380
3381 /*
3382 * *ref_rsrv is the portion of asize that will come from any
3383 * unconsumed refreservation space.
3384 */
3385 *ref_rsrv = 0;
3386
3387 mutex_enter(&ds->ds_lock);
3388 /*
3389 * Make a space adjustment for reserved bytes.
3390 */
3391 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3392 ASSERT3U(*used, >=,
3393 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3394 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3395 *ref_rsrv =
3396 asize - MIN(asize, parent_delta(ds, asize + inflight));
3397 }
3398
3399 if (!check_quota || ds->ds_quota == 0) {
3400 mutex_exit(&ds->ds_lock);
3401 return (0);
3402 }
3403 /*
3404 * If they are requesting more space, and our current estimate
3405 * is over quota, they get to try again unless the actual
3406 * on-disk is over quota and there are no pending changes (which
3407 * may free up space for us).
3408 */
3409 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3410 if (inflight > 0 ||
3411 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3412 error = ERESTART;
3413 else
3414 error = EDQUOT;
3415 }
3416 mutex_exit(&ds->ds_lock);
3417
3418 return (error);
3419 }
3420
3421 /* ARGSUSED */
3422 static int
3423 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3424 {
3425 dsl_dataset_t *ds = arg1;
3426 dsl_prop_setarg_t *psa = arg2;
3427 int err;
3428
3429 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3430 return (ENOTSUP);
3431
3432 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3433 return (err);
3434
3435 if (psa->psa_effective_value == 0)
3436 return (0);
3437
3438 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3439 psa->psa_effective_value < ds->ds_reserved)
3440 return (ENOSPC);
3441
3442 return (0);
3443 }
3444
3445 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3446
3447 void
3448 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3449 {
3450 dsl_dataset_t *ds = arg1;
3451 dsl_prop_setarg_t *psa = arg2;
3452 uint64_t effective_value = psa->psa_effective_value;
3453
3454 dsl_prop_set_sync(ds, psa, tx);
3455 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3456
3457 if (ds->ds_quota != effective_value) {
3458 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3459 ds->ds_quota = effective_value;
3460
3461 spa_history_log_internal(LOG_DS_REFQUOTA,
3462 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3463 (longlong_t)ds->ds_quota, ds->ds_object);
3464 }
3465 }
3466
3467 int
3468 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3469 {
3470 dsl_dataset_t *ds;
3471 dsl_prop_setarg_t psa;
3472 int err;
3473
3474 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3475
3476 err = dsl_dataset_hold(dsname, FTAG, &ds);
3477 if (err)
3478 return (err);
3479
3480 /*
3481 * If someone removes a file, then tries to set the quota, we
3482 * want to make sure the file freeing takes effect.
3483 */
3484 txg_wait_open(ds->ds_dir->dd_pool, 0);
3485
3486 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3487 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3488 ds, &psa, 0);
3489
3490 dsl_dataset_rele(ds, FTAG);
3491 return (err);
3492 }
3493
3494 static int
3495 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3496 {
3497 dsl_dataset_t *ds = arg1;
3498 dsl_prop_setarg_t *psa = arg2;
3499 uint64_t effective_value;
3500 uint64_t unique;
3501 int err;
3502
3503 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3504 SPA_VERSION_REFRESERVATION)
3505 return (ENOTSUP);
3506
3507 if (dsl_dataset_is_snapshot(ds))
3508 return (EINVAL);
3509
3510 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3511 return (err);
3512
3513 effective_value = psa->psa_effective_value;
3514
3515 /*
3516 * If we are doing the preliminary check in open context, the
3517 * space estimates may be inaccurate.
3518 */
3519 if (!dmu_tx_is_syncing(tx))
3520 return (0);
3521
3522 mutex_enter(&ds->ds_lock);
3523 if (!DS_UNIQUE_IS_ACCURATE(ds))
3524 dsl_dataset_recalc_head_uniq(ds);
3525 unique = ds->ds_phys->ds_unique_bytes;
3526 mutex_exit(&ds->ds_lock);
3527
3528 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3529 uint64_t delta = MAX(unique, effective_value) -
3530 MAX(unique, ds->ds_reserved);
3531
3532 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3533 return (ENOSPC);
3534 if (ds->ds_quota > 0 &&
3535 effective_value > ds->ds_quota)
3536 return (ENOSPC);
3537 }
3538
3539 return (0);
3540 }
3541
3542 static void
3543 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3544 {
3545 dsl_dataset_t *ds = arg1;
3546 dsl_prop_setarg_t *psa = arg2;
3547 uint64_t effective_value = psa->psa_effective_value;
3548 uint64_t unique;
3549 int64_t delta;
3550
3551 dsl_prop_set_sync(ds, psa, tx);
3552 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3553
3554 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3555
3556 mutex_enter(&ds->ds_dir->dd_lock);
3557 mutex_enter(&ds->ds_lock);
3558 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3559 unique = ds->ds_phys->ds_unique_bytes;
3560 delta = MAX(0, (int64_t)(effective_value - unique)) -
3561 MAX(0, (int64_t)(ds->ds_reserved - unique));
3562 ds->ds_reserved = effective_value;
3563 mutex_exit(&ds->ds_lock);
3564
3565 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3566 mutex_exit(&ds->ds_dir->dd_lock);
3567
3568 spa_history_log_internal(LOG_DS_REFRESERV,
3569 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3570 (longlong_t)effective_value, ds->ds_object);
3571 }
3572
3573 int
3574 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3575 uint64_t reservation)
3576 {
3577 dsl_dataset_t *ds;
3578 dsl_prop_setarg_t psa;
3579 int err;
3580
3581 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3582 &reservation);
3583
3584 err = dsl_dataset_hold(dsname, FTAG, &ds);
3585 if (err)
3586 return (err);
3587
3588 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3589 dsl_dataset_set_reservation_check,
3590 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3591
3592 dsl_dataset_rele(ds, FTAG);
3593 return (err);
3594 }
3595
3596 typedef struct zfs_hold_cleanup_arg {
3597 dsl_pool_t *dp;
3598 uint64_t dsobj;
3599 char htag[MAXNAMELEN];
3600 } zfs_hold_cleanup_arg_t;
3601
3602 static void
3603 dsl_dataset_user_release_onexit(void *arg)
3604 {
3605 zfs_hold_cleanup_arg_t *ca = arg;
3606
3607 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3608 B_TRUE);
3609 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3610 }
3611
3612 void
3613 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3614 minor_t minor)
3615 {
3616 zfs_hold_cleanup_arg_t *ca;
3617
3618 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3619 ca->dp = ds->ds_dir->dd_pool;
3620 ca->dsobj = ds->ds_object;
3621 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3622 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3623 dsl_dataset_user_release_onexit, ca, NULL));
3624 }
3625
3626 /*
3627 * If you add new checks here, you may need to add
3628 * additional checks to the "temporary" case in
3629 * snapshot_check() in dmu_objset.c.
3630 */
3631 static int
3632 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3633 {
3634 dsl_dataset_t *ds = arg1;
3635 struct dsl_ds_holdarg *ha = arg2;
3636 char *htag = ha->htag;
3637 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3638 int error = 0;
3639
3640 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3641 return (ENOTSUP);
3642
3643 if (!dsl_dataset_is_snapshot(ds))
3644 return (EINVAL);
3645
3646 /* tags must be unique */
3647 mutex_enter(&ds->ds_lock);
3648 if (ds->ds_phys->ds_userrefs_obj) {
3649 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3650 8, 1, tx);
3651 if (error == 0)
3652 error = EEXIST;
3653 else if (error == ENOENT)
3654 error = 0;
3655 }
3656 mutex_exit(&ds->ds_lock);
3657
3658 if (error == 0 && ha->temphold &&
3659 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3660 error = E2BIG;
3661
3662 return (error);
3663 }
3664
3665 void
3666 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3667 {
3668 dsl_dataset_t *ds = arg1;
3669 struct dsl_ds_holdarg *ha = arg2;
3670 char *htag = ha->htag;
3671 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3672 objset_t *mos = dp->dp_meta_objset;
3673 uint64_t now = gethrestime_sec();
3674 uint64_t zapobj;
3675
3676 mutex_enter(&ds->ds_lock);
3677 if (ds->ds_phys->ds_userrefs_obj == 0) {
3678 /*
3679 * This is the first user hold for this dataset. Create
3680 * the userrefs zap object.
3681 */
3682 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3683 zapobj = ds->ds_phys->ds_userrefs_obj =
3684 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3685 } else {
3686 zapobj = ds->ds_phys->ds_userrefs_obj;
3687 }
3688 ds->ds_userrefs++;
3689 mutex_exit(&ds->ds_lock);
3690
3691 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3692
3693 if (ha->temphold) {
3694 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3695 htag, &now, tx));
3696 }
3697
3698 spa_history_log_internal(LOG_DS_USER_HOLD,
3699 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3700 (int)ha->temphold, ds->ds_object);
3701 }
3702
3703 static int
3704 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3705 {
3706 struct dsl_ds_holdarg *ha = arg;
3707 dsl_dataset_t *ds;
3708 int error;
3709 char *name;
3710
3711 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3712 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3713 error = dsl_dataset_hold(name, ha->dstg, &ds);
3714 strfree(name);
3715 if (error == 0) {
3716 ha->gotone = B_TRUE;
3717 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3718 dsl_dataset_user_hold_sync, ds, ha, 0);
3719 } else if (error == ENOENT && ha->recursive) {
3720 error = 0;
3721 } else {
3722 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3723 }
3724 return (error);
3725 }
3726
3727 int
3728 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3729 boolean_t temphold)
3730 {
3731 struct dsl_ds_holdarg *ha;
3732 int error;
3733
3734 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3735 ha->htag = htag;
3736 ha->temphold = temphold;
3737 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3738 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3739 ds, ha, 0);
3740 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3741
3742 return (error);
3743 }
3744
3745 int
3746 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3747 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3748 {
3749 struct dsl_ds_holdarg *ha;
3750 dsl_sync_task_t *dst;
3751 spa_t *spa;
3752 int error;
3753 minor_t minor = 0;
3754
3755 if (cleanup_fd != -1) {
3756 /* Currently we only support cleanup-on-exit of tempholds. */
3757 if (!temphold)
3758 return (EINVAL);
3759 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3760 if (error)
3761 return (error);
3762 }
3763
3764 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3765
3766 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3767
3768 error = spa_open(dsname, &spa, FTAG);
3769 if (error) {
3770 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3771 if (cleanup_fd != -1)
3772 zfs_onexit_fd_rele(cleanup_fd);
3773 return (error);
3774 }
3775
3776 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3777 ha->htag = htag;
3778 ha->snapname = snapname;
3779 ha->recursive = recursive;
3780 ha->temphold = temphold;
3781
3782 if (recursive) {
3783 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3784 ha, DS_FIND_CHILDREN);
3785 } else {
3786 error = dsl_dataset_user_hold_one(dsname, ha);
3787 }
3788 if (error == 0)
3789 error = dsl_sync_task_group_wait(ha->dstg);
3790
3791 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3792 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3793 dsl_dataset_t *ds = dst->dst_arg1;
3794
3795 if (dst->dst_err) {
3796 dsl_dataset_name(ds, ha->failed);
3797 *strchr(ha->failed, '@') = '\0';
3798 } else if (error == 0 && minor != 0 && temphold) {
3799 /*
3800 * If this hold is to be released upon process exit,
3801 * register that action now.
3802 */
3803 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3804 }
3805 dsl_dataset_rele(ds, ha->dstg);
3806 }
3807
3808 if (error == 0 && recursive && !ha->gotone)
3809 error = ENOENT;
3810
3811 if (error)
3812 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3813
3814 dsl_sync_task_group_destroy(ha->dstg);
3815
3816 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3817 spa_close(spa, FTAG);
3818 if (cleanup_fd != -1)
3819 zfs_onexit_fd_rele(cleanup_fd);
3820 return (error);
3821 }
3822
3823 struct dsl_ds_releasearg {
3824 dsl_dataset_t *ds;
3825 const char *htag;
3826 boolean_t own; /* do we own or just hold ds? */
3827 };
3828
3829 static int
3830 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3831 boolean_t *might_destroy)
3832 {
3833 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3834 uint64_t zapobj;
3835 uint64_t tmp;
3836 int error;
3837
3838 *might_destroy = B_FALSE;
3839
3840 mutex_enter(&ds->ds_lock);
3841 zapobj = ds->ds_phys->ds_userrefs_obj;
3842 if (zapobj == 0) {
3843 /* The tag can't possibly exist */
3844 mutex_exit(&ds->ds_lock);
3845 return (ESRCH);
3846 }
3847
3848 /* Make sure the tag exists */
3849 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3850 if (error) {
3851 mutex_exit(&ds->ds_lock);
3852 if (error == ENOENT)
3853 error = ESRCH;
3854 return (error);
3855 }
3856
3857 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3858 DS_IS_DEFER_DESTROY(ds))
3859 *might_destroy = B_TRUE;
3860
3861 mutex_exit(&ds->ds_lock);
3862 return (0);
3863 }
3864
3865 static int
3866 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3867 {
3868 struct dsl_ds_releasearg *ra = arg1;
3869 dsl_dataset_t *ds = ra->ds;
3870 boolean_t might_destroy;
3871 int error;
3872
3873 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3874 return (ENOTSUP);
3875
3876 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3877 if (error)
3878 return (error);
3879
3880 if (might_destroy) {
3881 struct dsl_ds_destroyarg dsda = {0};
3882
3883 if (dmu_tx_is_syncing(tx)) {
3884 /*
3885 * If we're not prepared to remove the snapshot,
3886 * we can't allow the release to happen right now.
3887 */
3888 if (!ra->own)
3889 return (EBUSY);
3890 }
3891 dsda.ds = ds;
3892 dsda.releasing = B_TRUE;
3893 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3894 }
3895
3896 return (0);
3897 }
3898
3899 static void
3900 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3901 {
3902 struct dsl_ds_releasearg *ra = arg1;
3903 dsl_dataset_t *ds = ra->ds;
3904 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3905 objset_t *mos = dp->dp_meta_objset;
3906 uint64_t zapobj;
3907 uint64_t dsobj = ds->ds_object;
3908 uint64_t refs;
3909 int error;
3910
3911 mutex_enter(&ds->ds_lock);
3912 ds->ds_userrefs--;
3913 refs = ds->ds_userrefs;
3914 mutex_exit(&ds->ds_lock);
3915 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3916 VERIFY(error == 0 || error == ENOENT);
3917 zapobj = ds->ds_phys->ds_userrefs_obj;
3918 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3919 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3920 DS_IS_DEFER_DESTROY(ds)) {
3921 struct dsl_ds_destroyarg dsda = {0};
3922
3923 ASSERT(ra->own);
3924 dsda.ds = ds;
3925 dsda.releasing = B_TRUE;
3926 /* We already did the destroy_check */
3927 dsl_dataset_destroy_sync(&dsda, tag, tx);
3928 }
3929
3930 spa_history_log_internal(LOG_DS_USER_RELEASE,
3931 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
3932 ra->htag, (longlong_t)refs, dsobj);
3933 }
3934
3935 static int
3936 dsl_dataset_user_release_one(const char *dsname, void *arg)
3937 {
3938 struct dsl_ds_holdarg *ha = arg;
3939 struct dsl_ds_releasearg *ra;
3940 dsl_dataset_t *ds;
3941 int error;
3942 void *dtag = ha->dstg;
3943 char *name;
3944 boolean_t own = B_FALSE;
3945 boolean_t might_destroy;
3946
3947 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3948 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3949 error = dsl_dataset_hold(name, dtag, &ds);
3950 strfree(name);
3951 if (error == ENOENT && ha->recursive)
3952 return (0);
3953 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3954 if (error)
3955 return (error);
3956
3957 ha->gotone = B_TRUE;
3958
3959 ASSERT(dsl_dataset_is_snapshot(ds));
3960
3961 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3962 if (error) {
3963 dsl_dataset_rele(ds, dtag);
3964 return (error);
3965 }
3966
3967 if (might_destroy) {
3968 #ifdef _KERNEL
3969 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3970 error = zfs_unmount_snap(name, NULL);
3971 strfree(name);
3972 if (error) {
3973 dsl_dataset_rele(ds, dtag);
3974 return (error);
3975 }
3976 #endif
3977 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3978 dsl_dataset_rele(ds, dtag);
3979 return (EBUSY);
3980 } else {
3981 own = B_TRUE;
3982 dsl_dataset_make_exclusive(ds, dtag);
3983 }
3984 }
3985
3986 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3987 ra->ds = ds;
3988 ra->htag = ha->htag;
3989 ra->own = own;
3990 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3991 dsl_dataset_user_release_sync, ra, dtag, 0);
3992
3993 return (0);
3994 }
3995
3996 int
3997 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3998 boolean_t recursive)
3999 {
4000 struct dsl_ds_holdarg *ha;
4001 dsl_sync_task_t *dst;
4002 spa_t *spa;
4003 int error;
4004
4005 top:
4006 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
4007
4008 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
4009
4010 error = spa_open(dsname, &spa, FTAG);
4011 if (error) {
4012 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4013 return (error);
4014 }
4015
4016 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4017 ha->htag = htag;
4018 ha->snapname = snapname;
4019 ha->recursive = recursive;
4020 if (recursive) {
4021 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4022 ha, DS_FIND_CHILDREN);
4023 } else {
4024 error = dsl_dataset_user_release_one(dsname, ha);
4025 }
4026 if (error == 0)
4027 error = dsl_sync_task_group_wait(ha->dstg);
4028
4029 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4030 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4031 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4032 dsl_dataset_t *ds = ra->ds;
4033
4034 if (dst->dst_err)
4035 dsl_dataset_name(ds, ha->failed);
4036
4037 if (ra->own)
4038 dsl_dataset_disown(ds, ha->dstg);
4039 else
4040 dsl_dataset_rele(ds, ha->dstg);
4041
4042 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4043 }
4044
4045 if (error == 0 && recursive && !ha->gotone)
4046 error = ENOENT;
4047
4048 if (error && error != EBUSY)
4049 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4050
4051 dsl_sync_task_group_destroy(ha->dstg);
4052 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4053 spa_close(spa, FTAG);
4054
4055 /*
4056 * We can get EBUSY if we were racing with deferred destroy and
4057 * dsl_dataset_user_release_check() hadn't done the necessary
4058 * open context setup. We can also get EBUSY if we're racing
4059 * with destroy and that thread is the ds_owner. Either way
4060 * the busy condition should be transient, and we should retry
4061 * the release operation.
4062 */
4063 if (error == EBUSY)
4064 goto top;
4065
4066 return (error);
4067 }
4068
4069 /*
4070 * Called at spa_load time (with retry == B_FALSE) to release a stale
4071 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4072 */
4073 int
4074 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4075 boolean_t retry)
4076 {
4077 dsl_dataset_t *ds;
4078 char *snap;
4079 char *name;
4080 int namelen;
4081 int error;
4082
4083 do {
4084 rw_enter(&dp->dp_config_rwlock, RW_READER);
4085 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4086 rw_exit(&dp->dp_config_rwlock);
4087 if (error)
4088 return (error);
4089 namelen = dsl_dataset_namelen(ds)+1;
4090 name = kmem_alloc(namelen, KM_SLEEP);
4091 dsl_dataset_name(ds, name);
4092 dsl_dataset_rele(ds, FTAG);
4093
4094 snap = strchr(name, '@');
4095 *snap = '\0';
4096 ++snap;
4097 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4098 kmem_free(name, namelen);
4099
4100 /*
4101 * The object can't have been destroyed because we have a hold,
4102 * but it might have been renamed, resulting in ENOENT. Retry
4103 * if we've been requested to do so.
4104 *
4105 * It would be nice if we could use the dsobj all the way
4106 * through and avoid ENOENT entirely. But we might need to
4107 * unmount the snapshot, and there's currently no way to lookup
4108 * a vfsp using a ZFS object id.
4109 */
4110 } while ((error == ENOENT) && retry);
4111
4112 return (error);
4113 }
4114
4115 int
4116 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4117 {
4118 dsl_dataset_t *ds;
4119 int err;
4120
4121 err = dsl_dataset_hold(dsname, FTAG, &ds);
4122 if (err)
4123 return (err);
4124
4125 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4126 if (ds->ds_phys->ds_userrefs_obj != 0) {
4127 zap_attribute_t *za;
4128 zap_cursor_t zc;
4129
4130 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4131 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4132 ds->ds_phys->ds_userrefs_obj);
4133 zap_cursor_retrieve(&zc, za) == 0;
4134 zap_cursor_advance(&zc)) {
4135 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4136 za->za_first_integer));
4137 }
4138 zap_cursor_fini(&zc);
4139 kmem_free(za, sizeof (zap_attribute_t));
4140 }
4141 dsl_dataset_rele(ds, FTAG);
4142 return (0);
4143 }
4144
4145 /*
4146 * Note, this function is used as the callback for dmu_objset_find(). We
4147 * always return 0 so that we will continue to find and process
4148 * inconsistent datasets, even if we encounter an error trying to
4149 * process one of them.
4150 */
4151 /* ARGSUSED */
4152 int
4153 dsl_destroy_inconsistent(const char *dsname, void *arg)
4154 {
4155 dsl_dataset_t *ds;
4156
4157 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4158 if (DS_IS_INCONSISTENT(ds))
4159 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4160 else
4161 dsl_dataset_disown(ds, FTAG);
4162 }
4163 return (0);
4164 }
4165
4166 /*
4167 * Return (in *usedp) the amount of space written in new that is not
4168 * present in oldsnap. New may be a snapshot or the head. Old must be
4169 * a snapshot before new, in new's filesystem (or its origin). If not then
4170 * fail and return EINVAL.
4171 *
4172 * The written space is calculated by considering two components: First, we
4173 * ignore any freed space, and calculate the written as new's used space
4174 * minus old's used space. Next, we add in the amount of space that was freed
4175 * between the two snapshots, thus reducing new's used space relative to old's.
4176 * Specifically, this is the space that was born before old->ds_creation_txg,
4177 * and freed before new (ie. on new's deadlist or a previous deadlist).
4178 *
4179 * space freed [---------------------]
4180 * snapshots ---O-------O--------O-------O------
4181 * oldsnap new
4182 */
4183 int
4184 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4185 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4186 {
4187 int err = 0;
4188 uint64_t snapobj;
4189 dsl_pool_t *dp = new->ds_dir->dd_pool;
4190
4191 *usedp = 0;
4192 *usedp += new->ds_phys->ds_referenced_bytes;
4193 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4194
4195 *compp = 0;
4196 *compp += new->ds_phys->ds_compressed_bytes;
4197 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4198
4199 *uncompp = 0;
4200 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4201 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4202
4203 rw_enter(&dp->dp_config_rwlock, RW_READER);
4204 snapobj = new->ds_object;
4205 while (snapobj != oldsnap->ds_object) {
4206 dsl_dataset_t *snap;
4207 uint64_t used, comp, uncomp;
4208
4209 if (snapobj == new->ds_object) {
4210 snap = new;
4211 } else {
4212 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4213 if (err != 0)
4214 break;
4215 }
4216
4217 if (snap->ds_phys->ds_prev_snap_txg ==
4218 oldsnap->ds_phys->ds_creation_txg) {
4219 /*
4220 * The blocks in the deadlist can not be born after
4221 * ds_prev_snap_txg, so get the whole deadlist space,
4222 * which is more efficient (especially for old-format
4223 * deadlists). Unfortunately the deadlist code
4224 * doesn't have enough information to make this
4225 * optimization itself.
4226 */
4227 dsl_deadlist_space(&snap->ds_deadlist,
4228 &used, &comp, &uncomp);
4229 } else {
4230 dsl_deadlist_space_range(&snap->ds_deadlist,
4231 0, oldsnap->ds_phys->ds_creation_txg,
4232 &used, &comp, &uncomp);
4233 }
4234 *usedp += used;
4235 *compp += comp;
4236 *uncompp += uncomp;
4237
4238 /*
4239 * If we get to the beginning of the chain of snapshots
4240 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4241 * was not a snapshot of/before new.
4242 */
4243 snapobj = snap->ds_phys->ds_prev_snap_obj;
4244 if (snap != new)
4245 dsl_dataset_rele(snap, FTAG);
4246 if (snapobj == 0) {
4247 err = EINVAL;
4248 break;
4249 }
4250
4251 }
4252 rw_exit(&dp->dp_config_rwlock);
4253 return (err);
4254 }
4255
4256 /*
4257 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4258 * lastsnap, and all snapshots in between are deleted.
4259 *
4260 * blocks that would be freed [---------------------------]
4261 * snapshots ---O-------O--------O-------O--------O
4262 * firstsnap lastsnap
4263 *
4264 * This is the set of blocks that were born after the snap before firstsnap,
4265 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4266 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4267 * We calculate this by iterating over the relevant deadlists (from the snap
4268 * after lastsnap, backward to the snap after firstsnap), summing up the
4269 * space on the deadlist that was born after the snap before firstsnap.
4270 */
4271 int
4272 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4273 dsl_dataset_t *lastsnap,
4274 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4275 {
4276 int err = 0;
4277 uint64_t snapobj;
4278 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4279
4280 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4281 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4282
4283 /*
4284 * Check that the snapshots are in the same dsl_dir, and firstsnap
4285 * is before lastsnap.
4286 */
4287 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4288 firstsnap->ds_phys->ds_creation_txg >
4289 lastsnap->ds_phys->ds_creation_txg)
4290 return (EINVAL);
4291
4292 *usedp = *compp = *uncompp = 0;
4293
4294 rw_enter(&dp->dp_config_rwlock, RW_READER);
4295 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4296 while (snapobj != firstsnap->ds_object) {
4297 dsl_dataset_t *ds;
4298 uint64_t used, comp, uncomp;
4299
4300 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4301 if (err != 0)
4302 break;
4303
4304 dsl_deadlist_space_range(&ds->ds_deadlist,
4305 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4306 &used, &comp, &uncomp);
4307 *usedp += used;
4308 *compp += comp;
4309 *uncompp += uncomp;
4310
4311 snapobj = ds->ds_phys->ds_prev_snap_obj;
4312 ASSERT3U(snapobj, !=, 0);
4313 dsl_dataset_rele(ds, FTAG);
4314 }
4315 rw_exit(&dp->dp_config_rwlock);
4316 return (err);
4317 }