1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
24 * Copyright (c) 2012, Joyent, Inc. All rights reserved.
25 */
26
27 #include <sys/dmu_objset.h>
28 #include <sys/dsl_dataset.h>
29 #include <sys/dsl_dir.h>
30 #include <sys/dsl_prop.h>
31 #include <sys/dsl_synctask.h>
32 #include <sys/dmu_traverse.h>
33 #include <sys/dmu_impl.h>
34 #include <sys/dmu_tx.h>
35 #include <sys/arc.h>
36 #include <sys/zio.h>
37 #include <sys/zap.h>
38 #include <sys/zfeature.h>
39 #include <sys/unique.h>
40 #include <sys/zfs_context.h>
41 #include <sys/zfs_ioctl.h>
42 #include <sys/spa.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_onexit.h>
45 #include <sys/zvol.h>
46 #include <sys/dsl_scan.h>
47 #include <sys/dsl_deadlist.h>
48
49 static char *dsl_reaper = "the grim reaper";
50
51 static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
52 static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
53 static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
54
55 #define SWITCH64(x, y) \
56 { \
57 uint64_t __tmp = (x); \
58 (x) = (y); \
59 (y) = __tmp; \
60 }
61
62 #define DS_REF_MAX (1ULL << 62)
63
64 #define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
65
66 #define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
67
68
69 /*
70 * Figure out how much of this delta should be propogated to the dsl_dir
71 * layer. If there's a refreservation, that space has already been
72 * partially accounted for in our ancestors.
73 */
74 static int64_t
75 parent_delta(dsl_dataset_t *ds, int64_t delta)
76 {
77 uint64_t old_bytes, new_bytes;
78
79 if (ds->ds_reserved == 0)
80 return (delta);
81
82 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
83 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
84
85 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
86 return (new_bytes - old_bytes);
87 }
88
89 void
90 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
91 {
92 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
93 int compressed = BP_GET_PSIZE(bp);
94 int uncompressed = BP_GET_UCSIZE(bp);
95 int64_t delta;
96
97 dprintf_bp(bp, "ds=%p", ds);
98
99 ASSERT(dmu_tx_is_syncing(tx));
100 /* It could have been compressed away to nothing */
101 if (BP_IS_HOLE(bp))
102 return;
103 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
104 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
105 if (ds == NULL) {
106 /*
107 * Account for the meta-objset space in its placeholder
108 * dsl_dir.
109 */
110 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
111 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
112 used, compressed, uncompressed, tx);
113 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
114 return;
115 }
116 dmu_buf_will_dirty(ds->ds_dbuf, tx);
117
118 mutex_enter(&ds->ds_dir->dd_lock);
119 mutex_enter(&ds->ds_lock);
120 delta = parent_delta(ds, used);
121 ds->ds_phys->ds_referenced_bytes += used;
122 ds->ds_phys->ds_compressed_bytes += compressed;
123 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
124 ds->ds_phys->ds_unique_bytes += used;
125 mutex_exit(&ds->ds_lock);
126 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
127 compressed, uncompressed, tx);
128 dsl_dir_transfer_space(ds->ds_dir, used - delta,
129 DD_USED_REFRSRV, DD_USED_HEAD, tx);
130 mutex_exit(&ds->ds_dir->dd_lock);
131 }
132
133 int
134 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
135 boolean_t async)
136 {
137 if (BP_IS_HOLE(bp))
138 return (0);
139
140 ASSERT(dmu_tx_is_syncing(tx));
141 ASSERT(bp->blk_birth <= tx->tx_txg);
142
143 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
144 int compressed = BP_GET_PSIZE(bp);
145 int uncompressed = BP_GET_UCSIZE(bp);
146
147 ASSERT(used > 0);
148 if (ds == NULL) {
149 /*
150 * Account for the meta-objset space in its placeholder
151 * dataset.
152 */
153 dsl_free(tx->tx_pool, tx->tx_txg, bp);
154
155 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
156 -used, -compressed, -uncompressed, tx);
157 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
158 return (used);
159 }
160 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
161
162 ASSERT(!dsl_dataset_is_snapshot(ds));
163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
164
165 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
166 int64_t delta;
167
168 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
169 dsl_free(tx->tx_pool, tx->tx_txg, bp);
170
171 mutex_enter(&ds->ds_dir->dd_lock);
172 mutex_enter(&ds->ds_lock);
173 ASSERT(ds->ds_phys->ds_unique_bytes >= used ||
174 !DS_UNIQUE_IS_ACCURATE(ds));
175 delta = parent_delta(ds, -used);
176 ds->ds_phys->ds_unique_bytes -= used;
177 mutex_exit(&ds->ds_lock);
178 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
179 delta, -compressed, -uncompressed, tx);
180 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
181 DD_USED_REFRSRV, DD_USED_HEAD, tx);
182 mutex_exit(&ds->ds_dir->dd_lock);
183 } else {
184 dprintf_bp(bp, "putting on dead list: %s", "");
185 if (async) {
186 /*
187 * We are here as part of zio's write done callback,
188 * which means we're a zio interrupt thread. We can't
189 * call dsl_deadlist_insert() now because it may block
190 * waiting for I/O. Instead, put bp on the deferred
191 * queue and let dsl_pool_sync() finish the job.
192 */
193 bplist_append(&ds->ds_pending_deadlist, bp);
194 } else {
195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 }
197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 ds->ds_phys->ds_prev_snap_obj);
199 ASSERT(ds->ds_prev->ds_phys->ds_num_children > 0);
200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 if (ds->ds_prev->ds_phys->ds_next_snap_obj ==
202 ds->ds_object && bp->blk_birth >
203 ds->ds_prev->ds_phys->ds_prev_snap_txg) {
204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 mutex_enter(&ds->ds_prev->ds_lock);
206 ds->ds_prev->ds_phys->ds_unique_bytes += used;
207 mutex_exit(&ds->ds_prev->ds_lock);
208 }
209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 dsl_dir_transfer_space(ds->ds_dir, used,
211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 }
213 }
214 mutex_enter(&ds->ds_lock);
215 ASSERT3U(ds->ds_phys->ds_referenced_bytes, >=, used);
216 ds->ds_phys->ds_referenced_bytes -= used;
217 ASSERT3U(ds->ds_phys->ds_compressed_bytes, >=, compressed);
218 ds->ds_phys->ds_compressed_bytes -= compressed;
219 ASSERT3U(ds->ds_phys->ds_uncompressed_bytes, >=, uncompressed);
220 ds->ds_phys->ds_uncompressed_bytes -= uncompressed;
221 mutex_exit(&ds->ds_lock);
222
223 return (used);
224 }
225
226 uint64_t
227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 {
229 uint64_t trysnap = 0;
230
231 if (ds == NULL)
232 return (0);
233 /*
234 * The snapshot creation could fail, but that would cause an
235 * incorrect FALSE return, which would only result in an
236 * overestimation of the amount of space that an operation would
237 * consume, which is OK.
238 *
239 * There's also a small window where we could miss a pending
240 * snapshot, because we could set the sync task in the quiescing
241 * phase. So this should only be used as a guess.
242 */
243 if (ds->ds_trysnap_txg >
244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 trysnap = ds->ds_trysnap_txg;
246 return (MAX(ds->ds_phys->ds_prev_snap_txg, trysnap));
247 }
248
249 boolean_t
250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 uint64_t blk_birth)
252 {
253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
254 return (B_FALSE);
255
256 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
257
258 return (B_TRUE);
259 }
260
261 /* ARGSUSED */
262 static void
263 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
264 {
265 dsl_dataset_t *ds = dsv;
266
267 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
268
269 unique_remove(ds->ds_fsid_guid);
270
271 if (ds->ds_objset != NULL)
272 dmu_objset_evict(ds->ds_objset);
273
274 if (ds->ds_prev) {
275 dsl_dataset_drop_ref(ds->ds_prev, ds);
276 ds->ds_prev = NULL;
277 }
278
279 bplist_destroy(&ds->ds_pending_deadlist);
280 if (db != NULL) {
281 dsl_deadlist_close(&ds->ds_deadlist);
282 } else {
283 ASSERT(ds->ds_deadlist.dl_dbuf == NULL);
284 ASSERT(!ds->ds_deadlist.dl_oldfmt);
285 }
286 if (ds->ds_dir)
287 dsl_dir_close(ds->ds_dir, ds);
288
289 ASSERT(!list_link_active(&ds->ds_synced_link));
290
291 mutex_destroy(&ds->ds_lock);
292 mutex_destroy(&ds->ds_recvlock);
293 mutex_destroy(&ds->ds_opening_lock);
294 rw_destroy(&ds->ds_rwlock);
295 cv_destroy(&ds->ds_exclusive_cv);
296
297 kmem_free(ds, sizeof (dsl_dataset_t));
298 }
299
300 static int
301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
302 {
303 dsl_dataset_phys_t *headphys;
304 int err;
305 dmu_buf_t *headdbuf;
306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 objset_t *mos = dp->dp_meta_objset;
308
309 if (ds->ds_snapname[0])
310 return (0);
311 if (ds->ds_phys->ds_next_snap_obj == 0)
312 return (0);
313
314 err = dmu_bonus_hold(mos, ds->ds_dir->dd_phys->dd_head_dataset_obj,
315 FTAG, &headdbuf);
316 if (err)
317 return (err);
318 headphys = headdbuf->db_data;
319 err = zap_value_search(dp->dp_meta_objset,
320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 dmu_buf_rele(headdbuf, FTAG);
322 return (err);
323 }
324
325 static int
326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
327 {
328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
330 matchtype_t mt;
331 int err;
332
333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
334 mt = MT_FIRST;
335 else
336 mt = MT_EXACT;
337
338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
342 return (err);
343 }
344
345 static int
346 dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
347 {
348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
350 matchtype_t mt;
351 int err;
352
353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354
355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
356 mt = MT_FIRST;
357 else
358 mt = MT_EXACT;
359
360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
361 if (err == ENOTSUP && mt == MT_FIRST)
362 err = zap_remove(mos, snapobj, name, tx);
363 return (err);
364 }
365
366 static int
367 dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
368 dsl_dataset_t **dsp)
369 {
370 objset_t *mos = dp->dp_meta_objset;
371 dmu_buf_t *dbuf;
372 dsl_dataset_t *ds;
373 int err;
374 dmu_object_info_t doi;
375
376 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
377 dsl_pool_sync_context(dp));
378
379 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
380 if (err)
381 return (err);
382
383 /* Make sure dsobj has the correct object type. */
384 dmu_object_info_from_db(dbuf, &doi);
385 if (doi.doi_type != DMU_OT_DSL_DATASET)
386 return (EINVAL);
387
388 ds = dmu_buf_get_user(dbuf);
389 if (ds == NULL) {
390 dsl_dataset_t *winner;
391
392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
393 ds->ds_dbuf = dbuf;
394 ds->ds_object = dsobj;
395 ds->ds_phys = dbuf->db_data;
396
397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
400 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
401
402 rw_init(&ds->ds_rwlock, 0, 0, 0);
403 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
404
405 bplist_create(&ds->ds_pending_deadlist);
406 dsl_deadlist_open(&ds->ds_deadlist,
407 mos, ds->ds_phys->ds_deadlist_obj);
408
409 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
410 offsetof(dmu_sendarg_t, dsa_link));
411
412 if (err == 0) {
413 err = dsl_dir_open_obj(dp,
414 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
415 }
416 if (err) {
417 mutex_destroy(&ds->ds_lock);
418 mutex_destroy(&ds->ds_recvlock);
419 mutex_destroy(&ds->ds_opening_lock);
420 rw_destroy(&ds->ds_rwlock);
421 cv_destroy(&ds->ds_exclusive_cv);
422 bplist_destroy(&ds->ds_pending_deadlist);
423 dsl_deadlist_close(&ds->ds_deadlist);
424 kmem_free(ds, sizeof (dsl_dataset_t));
425 dmu_buf_rele(dbuf, tag);
426 return (err);
427 }
428
429 if (!dsl_dataset_is_snapshot(ds)) {
430 ds->ds_snapname[0] = '\0';
431 if (ds->ds_phys->ds_prev_snap_obj) {
432 err = dsl_dataset_get_ref(dp,
433 ds->ds_phys->ds_prev_snap_obj,
434 ds, &ds->ds_prev);
435 }
436 } else {
437 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
438 err = dsl_dataset_get_snapname(ds);
439 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
440 err = zap_count(
441 ds->ds_dir->dd_pool->dp_meta_objset,
442 ds->ds_phys->ds_userrefs_obj,
443 &ds->ds_userrefs);
444 }
445 }
446
447 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
448 /*
449 * In sync context, we're called with either no lock
450 * or with the write lock. If we're not syncing,
451 * we're always called with the read lock held.
452 */
453 boolean_t need_lock =
454 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
455 dsl_pool_sync_context(dp);
456
457 if (need_lock)
458 rw_enter(&dp->dp_config_rwlock, RW_READER);
459
460 err = dsl_prop_get_ds(ds,
461 "refreservation", sizeof (uint64_t), 1,
462 &ds->ds_reserved, NULL);
463 if (err == 0) {
464 err = dsl_prop_get_ds(ds,
465 "refquota", sizeof (uint64_t), 1,
466 &ds->ds_quota, NULL);
467 }
468
469 if (need_lock)
470 rw_exit(&dp->dp_config_rwlock);
471 } else {
472 ds->ds_reserved = ds->ds_quota = 0;
473 }
474
475 if (err == 0) {
476 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
477 dsl_dataset_evict);
478 }
479 if (err || winner) {
480 bplist_destroy(&ds->ds_pending_deadlist);
481 dsl_deadlist_close(&ds->ds_deadlist);
482 if (ds->ds_prev)
483 dsl_dataset_drop_ref(ds->ds_prev, ds);
484 dsl_dir_close(ds->ds_dir, ds);
485 mutex_destroy(&ds->ds_lock);
486 mutex_destroy(&ds->ds_recvlock);
487 mutex_destroy(&ds->ds_opening_lock);
488 rw_destroy(&ds->ds_rwlock);
489 cv_destroy(&ds->ds_exclusive_cv);
490 kmem_free(ds, sizeof (dsl_dataset_t));
491 if (err) {
492 dmu_buf_rele(dbuf, tag);
493 return (err);
494 }
495 ds = winner;
496 } else {
497 ds->ds_fsid_guid =
498 unique_insert(ds->ds_phys->ds_fsid_guid);
499 }
500 }
501 ASSERT3P(ds->ds_dbuf, ==, dbuf);
502 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
503 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
504 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
505 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
506 mutex_enter(&ds->ds_lock);
507 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
508 mutex_exit(&ds->ds_lock);
509 dmu_buf_rele(ds->ds_dbuf, tag);
510 return (ENOENT);
511 }
512 mutex_exit(&ds->ds_lock);
513 *dsp = ds;
514 return (0);
515 }
516
517 static int
518 dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
519 {
520 dsl_pool_t *dp = ds->ds_dir->dd_pool;
521
522 /*
523 * In syncing context we don't want the rwlock lock: there
524 * may be an existing writer waiting for sync phase to
525 * finish. We don't need to worry about such writers, since
526 * sync phase is single-threaded, so the writer can't be
527 * doing anything while we are active.
528 */
529 if (dsl_pool_sync_context(dp)) {
530 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
531 return (0);
532 }
533
534 /*
535 * Normal users will hold the ds_rwlock as a READER until they
536 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
537 * drop their READER lock after they set the ds_owner field.
538 *
539 * If the dataset is being destroyed, the destroy thread will
540 * obtain a WRITER lock for exclusive access after it's done its
541 * open-context work and then change the ds_owner to
542 * dsl_reaper once destruction is assured. So threads
543 * may block here temporarily, until the "destructability" of
544 * the dataset is determined.
545 */
546 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
547 mutex_enter(&ds->ds_lock);
548 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
549 rw_exit(&dp->dp_config_rwlock);
550 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
551 if (DSL_DATASET_IS_DESTROYED(ds)) {
552 mutex_exit(&ds->ds_lock);
553 dsl_dataset_drop_ref(ds, tag);
554 rw_enter(&dp->dp_config_rwlock, RW_READER);
555 return (ENOENT);
556 }
557 /*
558 * The dp_config_rwlock lives above the ds_lock. And
559 * we need to check DSL_DATASET_IS_DESTROYED() while
560 * holding the ds_lock, so we have to drop and reacquire
561 * the ds_lock here.
562 */
563 mutex_exit(&ds->ds_lock);
564 rw_enter(&dp->dp_config_rwlock, RW_READER);
565 mutex_enter(&ds->ds_lock);
566 }
567 mutex_exit(&ds->ds_lock);
568 return (0);
569 }
570
571 int
572 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
573 dsl_dataset_t **dsp)
574 {
575 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
576
577 if (err)
578 return (err);
579 return (dsl_dataset_hold_ref(*dsp, tag));
580 }
581
582 int
583 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
584 void *tag, dsl_dataset_t **dsp)
585 {
586 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
587 if (err)
588 return (err);
589 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
590 dsl_dataset_rele(*dsp, tag);
591 *dsp = NULL;
592 return (EBUSY);
593 }
594 return (0);
595 }
596
597 int
598 dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
599 {
600 dsl_dir_t *dd;
601 dsl_pool_t *dp;
602 const char *snapname;
603 uint64_t obj;
604 int err = 0;
605
606 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
607 if (err)
608 return (err);
609
610 dp = dd->dd_pool;
611 obj = dd->dd_phys->dd_head_dataset_obj;
612 rw_enter(&dp->dp_config_rwlock, RW_READER);
613 if (obj)
614 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
615 else
616 err = ENOENT;
617 if (err)
618 goto out;
619
620 err = dsl_dataset_hold_ref(*dsp, tag);
621
622 /* we may be looking for a snapshot */
623 if (err == 0 && snapname != NULL) {
624 dsl_dataset_t *ds = NULL;
625
626 if (*snapname++ != '@') {
627 dsl_dataset_rele(*dsp, tag);
628 err = ENOENT;
629 goto out;
630 }
631
632 dprintf("looking for snapshot '%s'\n", snapname);
633 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
634 if (err == 0)
635 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
636 dsl_dataset_rele(*dsp, tag);
637
638 ASSERT3U((err == 0), ==, (ds != NULL));
639
640 if (ds) {
641 mutex_enter(&ds->ds_lock);
642 if (ds->ds_snapname[0] == 0)
643 (void) strlcpy(ds->ds_snapname, snapname,
644 sizeof (ds->ds_snapname));
645 mutex_exit(&ds->ds_lock);
646 err = dsl_dataset_hold_ref(ds, tag);
647 *dsp = err ? NULL : ds;
648 }
649 }
650 out:
651 rw_exit(&dp->dp_config_rwlock);
652 dsl_dir_close(dd, FTAG);
653 return (err);
654 }
655
656 int
657 dsl_dataset_own(const char *name, boolean_t inconsistentok,
658 void *tag, dsl_dataset_t **dsp)
659 {
660 int err = dsl_dataset_hold(name, tag, dsp);
661 if (err)
662 return (err);
663 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
664 dsl_dataset_rele(*dsp, tag);
665 return (EBUSY);
666 }
667 return (0);
668 }
669
670 void
671 dsl_dataset_name(dsl_dataset_t *ds, char *name)
672 {
673 if (ds == NULL) {
674 (void) strcpy(name, "mos");
675 } else {
676 dsl_dir_name(ds->ds_dir, name);
677 VERIFY(0 == dsl_dataset_get_snapname(ds));
678 if (ds->ds_snapname[0]) {
679 (void) strcat(name, "@");
680 /*
681 * We use a "recursive" mutex so that we
682 * can call dprintf_ds() with ds_lock held.
683 */
684 if (!MUTEX_HELD(&ds->ds_lock)) {
685 mutex_enter(&ds->ds_lock);
686 (void) strcat(name, ds->ds_snapname);
687 mutex_exit(&ds->ds_lock);
688 } else {
689 (void) strcat(name, ds->ds_snapname);
690 }
691 }
692 }
693 }
694
695 static int
696 dsl_dataset_namelen(dsl_dataset_t *ds)
697 {
698 int result;
699
700 if (ds == NULL) {
701 result = 3; /* "mos" */
702 } else {
703 result = dsl_dir_namelen(ds->ds_dir);
704 VERIFY(0 == dsl_dataset_get_snapname(ds));
705 if (ds->ds_snapname[0]) {
706 ++result; /* adding one for the @-sign */
707 if (!MUTEX_HELD(&ds->ds_lock)) {
708 mutex_enter(&ds->ds_lock);
709 result += strlen(ds->ds_snapname);
710 mutex_exit(&ds->ds_lock);
711 } else {
712 result += strlen(ds->ds_snapname);
713 }
714 }
715 }
716
717 return (result);
718 }
719
720 void
721 dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
722 {
723 dmu_buf_rele(ds->ds_dbuf, tag);
724 }
725
726 void
727 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
728 {
729 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
730 rw_exit(&ds->ds_rwlock);
731 }
732 dsl_dataset_drop_ref(ds, tag);
733 }
734
735 void
736 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
737 {
738 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
739 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
740
741 mutex_enter(&ds->ds_lock);
742 ds->ds_owner = NULL;
743 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
744 rw_exit(&ds->ds_rwlock);
745 cv_broadcast(&ds->ds_exclusive_cv);
746 }
747 mutex_exit(&ds->ds_lock);
748 if (ds->ds_dbuf)
749 dsl_dataset_drop_ref(ds, tag);
750 else
751 dsl_dataset_evict(NULL, ds);
752 }
753
754 boolean_t
755 dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
756 {
757 boolean_t gotit = FALSE;
758
759 mutex_enter(&ds->ds_lock);
760 if (ds->ds_owner == NULL &&
761 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
762 ds->ds_owner = tag;
763 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
764 rw_exit(&ds->ds_rwlock);
765 gotit = TRUE;
766 }
767 mutex_exit(&ds->ds_lock);
768 return (gotit);
769 }
770
771 void
772 dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
773 {
774 ASSERT3P(owner, ==, ds->ds_owner);
775 if (!RW_WRITE_HELD(&ds->ds_rwlock))
776 rw_enter(&ds->ds_rwlock, RW_WRITER);
777 }
778
779 uint64_t
780 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
781 uint64_t flags, dmu_tx_t *tx)
782 {
783 dsl_pool_t *dp = dd->dd_pool;
784 dmu_buf_t *dbuf;
785 dsl_dataset_phys_t *dsphys;
786 uint64_t dsobj;
787 objset_t *mos = dp->dp_meta_objset;
788
789 if (origin == NULL)
790 origin = dp->dp_origin_snap;
791
792 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
793 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
794 ASSERT(dmu_tx_is_syncing(tx));
795 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
796
797 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
798 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
799 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
800 dmu_buf_will_dirty(dbuf, tx);
801 dsphys = dbuf->db_data;
802 bzero(dsphys, sizeof (dsl_dataset_phys_t));
803 dsphys->ds_dir_obj = dd->dd_object;
804 dsphys->ds_flags = flags;
805 dsphys->ds_fsid_guid = unique_create();
806 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
807 sizeof (dsphys->ds_guid));
808 dsphys->ds_snapnames_zapobj =
809 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
810 DMU_OT_NONE, 0, tx);
811 dsphys->ds_creation_time = gethrestime_sec();
812 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
813
814 if (origin == NULL) {
815 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
816 } else {
817 dsl_dataset_t *ohds;
818
819 dsphys->ds_prev_snap_obj = origin->ds_object;
820 dsphys->ds_prev_snap_txg =
821 origin->ds_phys->ds_creation_txg;
822 dsphys->ds_referenced_bytes =
823 origin->ds_phys->ds_referenced_bytes;
824 dsphys->ds_compressed_bytes =
825 origin->ds_phys->ds_compressed_bytes;
826 dsphys->ds_uncompressed_bytes =
827 origin->ds_phys->ds_uncompressed_bytes;
828 dsphys->ds_bp = origin->ds_phys->ds_bp;
829 dsphys->ds_flags |= origin->ds_phys->ds_flags;
830
831 dmu_buf_will_dirty(origin->ds_dbuf, tx);
832 origin->ds_phys->ds_num_children++;
833
834 VERIFY0(dsl_dataset_hold_obj(dp,
835 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
836 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
837 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
838 dsl_dataset_rele(ohds, FTAG);
839
840 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
841 if (origin->ds_phys->ds_next_clones_obj == 0) {
842 origin->ds_phys->ds_next_clones_obj =
843 zap_create(mos,
844 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
845 }
846 VERIFY(0 == zap_add_int(mos,
847 origin->ds_phys->ds_next_clones_obj,
848 dsobj, tx));
849 }
850
851 dmu_buf_will_dirty(dd->dd_dbuf, tx);
852 dd->dd_phys->dd_origin_obj = origin->ds_object;
853 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
854 if (origin->ds_dir->dd_phys->dd_clones == 0) {
855 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
856 origin->ds_dir->dd_phys->dd_clones =
857 zap_create(mos,
858 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
859 }
860 VERIFY0(zap_add_int(mos,
861 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
862 }
863 }
864
865 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
866 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
867
868 dmu_buf_rele(dbuf, FTAG);
869
870 dmu_buf_will_dirty(dd->dd_dbuf, tx);
871 dd->dd_phys->dd_head_dataset_obj = dsobj;
872
873 return (dsobj);
874 }
875
876 uint64_t
877 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
878 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
879 {
880 dsl_pool_t *dp = pdd->dd_pool;
881 uint64_t dsobj, ddobj;
882 dsl_dir_t *dd;
883
884 ASSERT(lastname[0] != '@');
885
886 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
887 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
888
889 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
890
891 dsl_deleg_set_create_perms(dd, tx, cr);
892
893 dsl_dir_close(dd, FTAG);
894
895 /*
896 * If we are creating a clone, make sure we zero out any stale
897 * data from the origin snapshots zil header.
898 */
899 if (origin != NULL) {
900 dsl_dataset_t *ds;
901 objset_t *os;
902
903 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
904 VERIFY0(dmu_objset_from_ds(ds, &os));
905 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
906 dsl_dataset_dirty(ds, tx);
907 dsl_dataset_rele(ds, FTAG);
908 }
909
910 return (dsobj);
911 }
912
913 /*
914 * The snapshots must all be in the same pool.
915 */
916 int
917 dmu_snapshots_destroy_nvl(nvlist_t *snaps, boolean_t defer,
918 nvlist_t *errlist)
919 {
920 int err;
921 dsl_sync_task_t *dst;
922 spa_t *spa;
923 nvpair_t *pair;
924 dsl_sync_task_group_t *dstg;
925
926 pair = nvlist_next_nvpair(snaps, NULL);
927 if (pair == NULL)
928 return (0);
929
930 err = spa_open(nvpair_name(pair), &spa, FTAG);
931 if (err)
932 return (err);
933 dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
934
935 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
936 pair = nvlist_next_nvpair(snaps, pair)) {
937 dsl_dataset_t *ds;
938
939 err = dsl_dataset_own(nvpair_name(pair), B_TRUE, dstg, &ds);
940 if (err == 0) {
941 struct dsl_ds_destroyarg *dsda;
942
943 dsl_dataset_make_exclusive(ds, dstg);
944 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg),
945 KM_SLEEP);
946 dsda->ds = ds;
947 dsda->defer = defer;
948 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
949 dsl_dataset_destroy_sync, dsda, dstg, 0);
950 } else if (err == ENOENT) {
951 err = 0;
952 } else {
953 fnvlist_add_int32(errlist, nvpair_name(pair), err);
954 break;
955 }
956 }
957
958 if (err == 0)
959 err = dsl_sync_task_group_wait(dstg);
960
961 for (dst = list_head(&dstg->dstg_tasks); dst;
962 dst = list_next(&dstg->dstg_tasks, dst)) {
963 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
964 dsl_dataset_t *ds = dsda->ds;
965
966 /*
967 * Return the snapshots that triggered the error.
968 */
969 if (dst->dst_err != 0) {
970 char name[ZFS_MAXNAMELEN];
971 dsl_dataset_name(ds, name);
972 fnvlist_add_int32(errlist, name, dst->dst_err);
973 }
974 ASSERT3P(dsda->rm_origin, ==, NULL);
975 dsl_dataset_disown(ds, dstg);
976 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
977 }
978
979 dsl_sync_task_group_destroy(dstg);
980 spa_close(spa, FTAG);
981 return (err);
982
983 }
984
985 static boolean_t
986 dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
987 {
988 boolean_t might_destroy = B_FALSE;
989
990 mutex_enter(&ds->ds_lock);
991 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
992 DS_IS_DEFER_DESTROY(ds))
993 might_destroy = B_TRUE;
994 mutex_exit(&ds->ds_lock);
995
996 return (might_destroy);
997 }
998
999 /*
1000 * If we're removing a clone, and these three conditions are true:
1001 * 1) the clone's origin has no other children
1002 * 2) the clone's origin has no user references
1003 * 3) the clone's origin has been marked for deferred destruction
1004 * Then, prepare to remove the origin as part of this sync task group.
1005 */
1006 static int
1007 dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1008 {
1009 dsl_dataset_t *ds = dsda->ds;
1010 dsl_dataset_t *origin = ds->ds_prev;
1011
1012 if (dsl_dataset_might_destroy_origin(origin)) {
1013 char *name;
1014 int namelen;
1015 int error;
1016
1017 namelen = dsl_dataset_namelen(origin) + 1;
1018 name = kmem_alloc(namelen, KM_SLEEP);
1019 dsl_dataset_name(origin, name);
1020 #ifdef _KERNEL
1021 error = zfs_unmount_snap(name, NULL);
1022 if (error) {
1023 kmem_free(name, namelen);
1024 return (error);
1025 }
1026 #endif
1027 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
1028 kmem_free(name, namelen);
1029 if (error)
1030 return (error);
1031 dsda->rm_origin = origin;
1032 dsl_dataset_make_exclusive(origin, tag);
1033 }
1034
1035 return (0);
1036 }
1037
1038 /*
1039 * ds must be opened as OWNER. On return (whether successful or not),
1040 * ds will be closed and caller can no longer dereference it.
1041 */
1042 int
1043 dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
1044 {
1045 int err;
1046 dsl_sync_task_group_t *dstg;
1047 objset_t *os;
1048 dsl_dir_t *dd;
1049 uint64_t obj;
1050 struct dsl_ds_destroyarg dsda = { 0 };
1051
1052 dsda.ds = ds;
1053
1054 if (dsl_dataset_is_snapshot(ds)) {
1055 /* Destroying a snapshot is simpler */
1056 dsl_dataset_make_exclusive(ds, tag);
1057
1058 dsda.defer = defer;
1059 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1060 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
1061 &dsda, tag, 0);
1062 ASSERT3P(dsda.rm_origin, ==, NULL);
1063 goto out;
1064 } else if (defer) {
1065 err = EINVAL;
1066 goto out;
1067 }
1068
1069 dd = ds->ds_dir;
1070
1071 /*
1072 * Check for errors and mark this ds as inconsistent, in
1073 * case we crash while freeing the objects.
1074 */
1075 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1076 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1077 if (err)
1078 goto out;
1079
1080 err = dmu_objset_from_ds(ds, &os);
1081 if (err)
1082 goto out;
1083
1084 /*
1085 * If async destruction is not enabled try to remove all objects
1086 * while in the open context so that there is less work to do in
1087 * the syncing context.
1088 */
1089 if (!spa_feature_is_enabled(dsl_dataset_get_spa(ds),
1090 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
1091 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1092 ds->ds_phys->ds_prev_snap_txg)) {
1093 /*
1094 * Ignore errors, if there is not enough disk space
1095 * we will deal with it in dsl_dataset_destroy_sync().
1096 */
1097 (void) dmu_free_object(os, obj);
1098 }
1099 if (err != ESRCH)
1100 goto out;
1101 }
1102
1103 /*
1104 * Only the ZIL knows how to free log blocks.
1105 */
1106 zil_destroy(dmu_objset_zil(os), B_FALSE);
1107
1108 /*
1109 * Sync out all in-flight IO.
1110 */
1111 txg_wait_synced(dd->dd_pool, 0);
1112
1113 /*
1114 * If we managed to free all the objects in open
1115 * context, the user space accounting should be zero.
1116 */
1117 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
1118 dmu_objset_userused_enabled(os)) {
1119 uint64_t count;
1120
1121 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1122 count == 0);
1123 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1124 count == 0);
1125 }
1126
1127 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1128 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1129 rw_exit(&dd->dd_pool->dp_config_rwlock);
1130
1131 if (err)
1132 goto out;
1133
1134 /*
1135 * Blow away the dsl_dir + head dataset.
1136 */
1137 dsl_dataset_make_exclusive(ds, tag);
1138 /*
1139 * If we're removing a clone, we might also need to remove its
1140 * origin.
1141 */
1142 do {
1143 dsda.need_prep = B_FALSE;
1144 if (dsl_dir_is_clone(dd)) {
1145 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1146 if (err) {
1147 dsl_dir_close(dd, FTAG);
1148 goto out;
1149 }
1150 }
1151
1152 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1153 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1154 dsl_dataset_destroy_sync, &dsda, tag, 0);
1155 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
1156 dsl_dir_destroy_sync, dd, FTAG, 0);
1157 err = dsl_sync_task_group_wait(dstg);
1158 dsl_sync_task_group_destroy(dstg);
1159
1160 /*
1161 * We could be racing against 'zfs release' or 'zfs destroy -d'
1162 * on the origin snap, in which case we can get EBUSY if we
1163 * needed to destroy the origin snap but were not ready to
1164 * do so.
1165 */
1166 if (dsda.need_prep) {
1167 ASSERT(err == EBUSY);
1168 ASSERT(dsl_dir_is_clone(dd));
1169 ASSERT(dsda.rm_origin == NULL);
1170 }
1171 } while (dsda.need_prep);
1172
1173 if (dsda.rm_origin != NULL)
1174 dsl_dataset_disown(dsda.rm_origin, tag);
1175
1176 /* if it is successful, dsl_dir_destroy_sync will close the dd */
1177 if (err)
1178 dsl_dir_close(dd, FTAG);
1179 out:
1180 dsl_dataset_disown(ds, tag);
1181 return (err);
1182 }
1183
1184 blkptr_t *
1185 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1186 {
1187 return (&ds->ds_phys->ds_bp);
1188 }
1189
1190 void
1191 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1192 {
1193 ASSERT(dmu_tx_is_syncing(tx));
1194 /* If it's the meta-objset, set dp_meta_rootbp */
1195 if (ds == NULL) {
1196 tx->tx_pool->dp_meta_rootbp = *bp;
1197 } else {
1198 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1199 ds->ds_phys->ds_bp = *bp;
1200 }
1201 }
1202
1203 spa_t *
1204 dsl_dataset_get_spa(dsl_dataset_t *ds)
1205 {
1206 return (ds->ds_dir->dd_pool->dp_spa);
1207 }
1208
1209 void
1210 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1211 {
1212 dsl_pool_t *dp;
1213
1214 if (ds == NULL) /* this is the meta-objset */
1215 return;
1216
1217 ASSERT(ds->ds_objset != NULL);
1218
1219 if (ds->ds_phys->ds_next_snap_obj != 0)
1220 panic("dirtying snapshot!");
1221
1222 dp = ds->ds_dir->dd_pool;
1223
1224 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1225 /* up the hold count until we can be written out */
1226 dmu_buf_add_ref(ds->ds_dbuf, ds);
1227 }
1228 }
1229
1230 /*
1231 * The unique space in the head dataset can be calculated by subtracting
1232 * the space used in the most recent snapshot, that is still being used
1233 * in this file system, from the space currently in use. To figure out
1234 * the space in the most recent snapshot still in use, we need to take
1235 * the total space used in the snapshot and subtract out the space that
1236 * has been freed up since the snapshot was taken.
1237 */
1238 static void
1239 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1240 {
1241 uint64_t mrs_used;
1242 uint64_t dlused, dlcomp, dluncomp;
1243
1244 ASSERT(!dsl_dataset_is_snapshot(ds));
1245
1246 if (ds->ds_phys->ds_prev_snap_obj != 0)
1247 mrs_used = ds->ds_prev->ds_phys->ds_referenced_bytes;
1248 else
1249 mrs_used = 0;
1250
1251 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1252
1253 ASSERT3U(dlused, <=, mrs_used);
1254 ds->ds_phys->ds_unique_bytes =
1255 ds->ds_phys->ds_referenced_bytes - (mrs_used - dlused);
1256
1257 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1258 SPA_VERSION_UNIQUE_ACCURATE)
1259 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1260 }
1261
1262 struct killarg {
1263 dsl_dataset_t *ds;
1264 dmu_tx_t *tx;
1265 };
1266
1267 /* ARGSUSED */
1268 static int
1269 kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1270 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
1271 {
1272 struct killarg *ka = arg;
1273 dmu_tx_t *tx = ka->tx;
1274
1275 if (bp == NULL)
1276 return (0);
1277
1278 if (zb->zb_level == ZB_ZIL_LEVEL) {
1279 ASSERT(zilog != NULL);
1280 /*
1281 * It's a block in the intent log. It has no
1282 * accounting, so just free it.
1283 */
1284 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
1285 } else {
1286 ASSERT(zilog == NULL);
1287 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
1288 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
1289 }
1290
1291 return (0);
1292 }
1293
1294 /* ARGSUSED */
1295 static int
1296 dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1297 {
1298 dsl_dataset_t *ds = arg1;
1299 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1300 uint64_t count;
1301 int err;
1302
1303 /*
1304 * Can't delete a head dataset if there are snapshots of it.
1305 * (Except if the only snapshots are from the branch we cloned
1306 * from.)
1307 */
1308 if (ds->ds_prev != NULL &&
1309 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1310 return (EBUSY);
1311
1312 /*
1313 * This is really a dsl_dir thing, but check it here so that
1314 * we'll be less likely to leave this dataset inconsistent &
1315 * nearly destroyed.
1316 */
1317 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1318 if (err)
1319 return (err);
1320 if (count != 0)
1321 return (EEXIST);
1322
1323 return (0);
1324 }
1325
1326 /* ARGSUSED */
1327 static void
1328 dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
1329 {
1330 dsl_dataset_t *ds = arg1;
1331
1332 /* Mark it as inconsistent on-disk, in case we crash */
1333 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1334 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1335
1336 spa_history_log_internal_ds(ds, "destroy begin", tx, "");
1337 }
1338
1339 static int
1340 dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1341 dmu_tx_t *tx)
1342 {
1343 dsl_dataset_t *ds = dsda->ds;
1344 dsl_dataset_t *ds_prev = ds->ds_prev;
1345
1346 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1347 struct dsl_ds_destroyarg ndsda = {0};
1348
1349 /*
1350 * If we're not prepared to remove the origin, don't remove
1351 * the clone either.
1352 */
1353 if (dsda->rm_origin == NULL) {
1354 dsda->need_prep = B_TRUE;
1355 return (EBUSY);
1356 }
1357
1358 ndsda.ds = ds_prev;
1359 ndsda.is_origin_rm = B_TRUE;
1360 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1361 }
1362
1363 /*
1364 * If we're not going to remove the origin after all,
1365 * undo the open context setup.
1366 */
1367 if (dsda->rm_origin != NULL) {
1368 dsl_dataset_disown(dsda->rm_origin, tag);
1369 dsda->rm_origin = NULL;
1370 }
1371
1372 return (0);
1373 }
1374
1375 /*
1376 * If you add new checks here, you may need to add
1377 * additional checks to the "temporary" case in
1378 * snapshot_check() in dmu_objset.c.
1379 */
1380 /* ARGSUSED */
1381 int
1382 dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1383 {
1384 struct dsl_ds_destroyarg *dsda = arg1;
1385 dsl_dataset_t *ds = dsda->ds;
1386
1387 /* we have an owner hold, so noone else can destroy us */
1388 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1389
1390 /*
1391 * Only allow deferred destroy on pools that support it.
1392 * NOTE: deferred destroy is only supported on snapshots.
1393 */
1394 if (dsda->defer) {
1395 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1396 SPA_VERSION_USERREFS)
1397 return (ENOTSUP);
1398 ASSERT(dsl_dataset_is_snapshot(ds));
1399 return (0);
1400 }
1401
1402 /*
1403 * Can't delete a head dataset if there are snapshots of it.
1404 * (Except if the only snapshots are from the branch we cloned
1405 * from.)
1406 */
1407 if (ds->ds_prev != NULL &&
1408 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
1409 return (EBUSY);
1410
1411 /*
1412 * If we made changes this txg, traverse_dsl_dataset won't find
1413 * them. Try again.
1414 */
1415 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1416 return (EAGAIN);
1417
1418 if (dsl_dataset_is_snapshot(ds)) {
1419 /*
1420 * If this snapshot has an elevated user reference count,
1421 * we can't destroy it yet.
1422 */
1423 if (ds->ds_userrefs > 0 && !dsda->releasing)
1424 return (EBUSY);
1425
1426 mutex_enter(&ds->ds_lock);
1427 /*
1428 * Can't delete a branch point. However, if we're destroying
1429 * a clone and removing its origin due to it having a user
1430 * hold count of 0 and having been marked for deferred destroy,
1431 * it's OK for the origin to have a single clone.
1432 */
1433 if (ds->ds_phys->ds_num_children >
1434 (dsda->is_origin_rm ? 2 : 1)) {
1435 mutex_exit(&ds->ds_lock);
1436 return (EEXIST);
1437 }
1438 mutex_exit(&ds->ds_lock);
1439 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1440 return (dsl_dataset_origin_check(dsda, arg2, tx));
1441 }
1442
1443 /* XXX we should do some i/o error checking... */
1444 return (0);
1445 }
1446
1447 struct refsarg {
1448 kmutex_t lock;
1449 boolean_t gone;
1450 kcondvar_t cv;
1451 };
1452
1453 /* ARGSUSED */
1454 static void
1455 dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1456 {
1457 struct refsarg *arg = argv;
1458
1459 mutex_enter(&arg->lock);
1460 arg->gone = TRUE;
1461 cv_signal(&arg->cv);
1462 mutex_exit(&arg->lock);
1463 }
1464
1465 static void
1466 dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1467 {
1468 struct refsarg arg;
1469
1470 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1471 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1472 arg.gone = FALSE;
1473 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1474 dsl_dataset_refs_gone);
1475 dmu_buf_rele(ds->ds_dbuf, tag);
1476 mutex_enter(&arg.lock);
1477 while (!arg.gone)
1478 cv_wait(&arg.cv, &arg.lock);
1479 ASSERT(arg.gone);
1480 mutex_exit(&arg.lock);
1481 ds->ds_dbuf = NULL;
1482 ds->ds_phys = NULL;
1483 mutex_destroy(&arg.lock);
1484 cv_destroy(&arg.cv);
1485 }
1486
1487 static void
1488 remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1489 {
1490 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1491 uint64_t count;
1492 int err;
1493
1494 ASSERT(ds->ds_phys->ds_num_children >= 2);
1495 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1496 /*
1497 * The err should not be ENOENT, but a bug in a previous version
1498 * of the code could cause upgrade_clones_cb() to not set
1499 * ds_next_snap_obj when it should, leading to a missing entry.
1500 * If we knew that the pool was created after
1501 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1502 * ENOENT. However, at least we can check that we don't have
1503 * too many entries in the next_clones_obj even after failing to
1504 * remove this one.
1505 */
1506 if (err != ENOENT) {
1507 VERIFY0(err);
1508 }
1509 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj, &count));
1510 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1511 }
1512
1513 static void
1514 dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1515 {
1516 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1517 zap_cursor_t zc;
1518 zap_attribute_t za;
1519
1520 /*
1521 * If it is the old version, dd_clones doesn't exist so we can't
1522 * find the clones, but deadlist_remove_key() is a no-op so it
1523 * doesn't matter.
1524 */
1525 if (ds->ds_dir->dd_phys->dd_clones == 0)
1526 return;
1527
1528 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1529 zap_cursor_retrieve(&zc, &za) == 0;
1530 zap_cursor_advance(&zc)) {
1531 dsl_dataset_t *clone;
1532
1533 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1534 za.za_first_integer, FTAG, &clone));
1535 if (clone->ds_dir->dd_origin_txg > mintxg) {
1536 dsl_deadlist_remove_key(&clone->ds_deadlist,
1537 mintxg, tx);
1538 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1539 }
1540 dsl_dataset_rele(clone, FTAG);
1541 }
1542 zap_cursor_fini(&zc);
1543 }
1544
1545 struct process_old_arg {
1546 dsl_dataset_t *ds;
1547 dsl_dataset_t *ds_prev;
1548 boolean_t after_branch_point;
1549 zio_t *pio;
1550 uint64_t used, comp, uncomp;
1551 };
1552
1553 static int
1554 process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1555 {
1556 struct process_old_arg *poa = arg;
1557 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1558
1559 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1560 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1561 if (poa->ds_prev && !poa->after_branch_point &&
1562 bp->blk_birth >
1563 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1564 poa->ds_prev->ds_phys->ds_unique_bytes +=
1565 bp_get_dsize_sync(dp->dp_spa, bp);
1566 }
1567 } else {
1568 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1569 poa->comp += BP_GET_PSIZE(bp);
1570 poa->uncomp += BP_GET_UCSIZE(bp);
1571 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1572 }
1573 return (0);
1574 }
1575
1576 static void
1577 process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1578 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1579 {
1580 struct process_old_arg poa = { 0 };
1581 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1582 objset_t *mos = dp->dp_meta_objset;
1583
1584 ASSERT(ds->ds_deadlist.dl_oldfmt);
1585 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1586
1587 poa.ds = ds;
1588 poa.ds_prev = ds_prev;
1589 poa.after_branch_point = after_branch_point;
1590 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1591 VERIFY0(bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1592 process_old_cb, &poa, tx));
1593 VERIFY0(zio_wait(poa.pio));
1594 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1595
1596 /* change snapused */
1597 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1598 -poa.used, -poa.comp, -poa.uncomp, tx);
1599
1600 /* swap next's deadlist to our deadlist */
1601 dsl_deadlist_close(&ds->ds_deadlist);
1602 dsl_deadlist_close(&ds_next->ds_deadlist);
1603 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1604 ds->ds_phys->ds_deadlist_obj);
1605 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1606 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1607 ds_next->ds_phys->ds_deadlist_obj);
1608 }
1609
1610 static int
1611 old_synchronous_dataset_destroy(dsl_dataset_t *ds, dmu_tx_t *tx)
1612 {
1613 int err;
1614 struct killarg ka;
1615
1616 /*
1617 * Free everything that we point to (that's born after
1618 * the previous snapshot, if we are a clone)
1619 *
1620 * NB: this should be very quick, because we already
1621 * freed all the objects in open context.
1622 */
1623 ka.ds = ds;
1624 ka.tx = tx;
1625 err = traverse_dataset(ds,
1626 ds->ds_phys->ds_prev_snap_txg, TRAVERSE_POST,
1627 kill_blkptr, &ka);
1628 ASSERT0(err);
1629 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) || ds->ds_phys->ds_unique_bytes == 0);
1630
1631 return (err);
1632 }
1633
1634 void
1635 dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
1636 {
1637 struct dsl_ds_destroyarg *dsda = arg1;
1638 dsl_dataset_t *ds = dsda->ds;
1639 int err;
1640 int after_branch_point = FALSE;
1641 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1642 objset_t *mos = dp->dp_meta_objset;
1643 dsl_dataset_t *ds_prev = NULL;
1644 boolean_t wont_destroy;
1645 uint64_t obj;
1646
1647 wont_destroy = (dsda->defer &&
1648 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1649
1650 ASSERT(ds->ds_owner || wont_destroy);
1651 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
1652 ASSERT(ds->ds_prev == NULL ||
1653 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1654 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1655
1656 if (wont_destroy) {
1657 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
1658 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1659 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1660 spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
1661 return;
1662 }
1663
1664 /* We need to log before removing it from the namespace. */
1665 spa_history_log_internal_ds(ds, "destroy", tx, "");
1666
1667 /* signal any waiters that this dataset is going away */
1668 mutex_enter(&ds->ds_lock);
1669 ds->ds_owner = dsl_reaper;
1670 cv_broadcast(&ds->ds_exclusive_cv);
1671 mutex_exit(&ds->ds_lock);
1672
1673 /* Remove our reservation */
1674 if (ds->ds_reserved != 0) {
1675 dsl_prop_setarg_t psa;
1676 uint64_t value = 0;
1677
1678 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1679 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1680 &value);
1681 psa.psa_effective_value = 0; /* predict default value */
1682
1683 dsl_dataset_set_reservation_sync(ds, &psa, tx);
1684 ASSERT0(ds->ds_reserved);
1685 }
1686
1687 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1688
1689 dsl_scan_ds_destroyed(ds, tx);
1690
1691 obj = ds->ds_object;
1692
1693 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1694 if (ds->ds_prev) {
1695 ds_prev = ds->ds_prev;
1696 } else {
1697 VERIFY(0 == dsl_dataset_hold_obj(dp,
1698 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
1699 }
1700 after_branch_point =
1701 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1702
1703 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
1704 if (after_branch_point &&
1705 ds_prev->ds_phys->ds_next_clones_obj != 0) {
1706 remove_from_next_clones(ds_prev, obj, tx);
1707 if (ds->ds_phys->ds_next_snap_obj != 0) {
1708 VERIFY(0 == zap_add_int(mos,
1709 ds_prev->ds_phys->ds_next_clones_obj,
1710 ds->ds_phys->ds_next_snap_obj, tx));
1711 }
1712 }
1713 if (after_branch_point &&
1714 ds->ds_phys->ds_next_snap_obj == 0) {
1715 /* This clone is toast. */
1716 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1717 ds_prev->ds_phys->ds_num_children--;
1718
1719 /*
1720 * If the clone's origin has no other clones, no
1721 * user holds, and has been marked for deferred
1722 * deletion, then we should have done the necessary
1723 * destroy setup for it.
1724 */
1725 if (ds_prev->ds_phys->ds_num_children == 1 &&
1726 ds_prev->ds_userrefs == 0 &&
1727 DS_IS_DEFER_DESTROY(ds_prev)) {
1728 ASSERT3P(dsda->rm_origin, !=, NULL);
1729 } else {
1730 ASSERT3P(dsda->rm_origin, ==, NULL);
1731 }
1732 } else if (!after_branch_point) {
1733 ds_prev->ds_phys->ds_next_snap_obj =
1734 ds->ds_phys->ds_next_snap_obj;
1735 }
1736 }
1737
1738 if (dsl_dataset_is_snapshot(ds)) {
1739 dsl_dataset_t *ds_next;
1740 uint64_t old_unique;
1741 uint64_t used = 0, comp = 0, uncomp = 0;
1742
1743 VERIFY(0 == dsl_dataset_hold_obj(dp,
1744 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
1745 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1746
1747 old_unique = ds_next->ds_phys->ds_unique_bytes;
1748
1749 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1750 ds_next->ds_phys->ds_prev_snap_obj =
1751 ds->ds_phys->ds_prev_snap_obj;
1752 ds_next->ds_phys->ds_prev_snap_txg =
1753 ds->ds_phys->ds_prev_snap_txg;
1754 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1755 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1756
1757
1758 if (ds_next->ds_deadlist.dl_oldfmt) {
1759 process_old_deadlist(ds, ds_prev, ds_next,
1760 after_branch_point, tx);
1761 } else {
1762 /* Adjust prev's unique space. */
1763 if (ds_prev && !after_branch_point) {
1764 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1765 ds_prev->ds_phys->ds_prev_snap_txg,
1766 ds->ds_phys->ds_prev_snap_txg,
1767 &used, &comp, &uncomp);
1768 ds_prev->ds_phys->ds_unique_bytes += used;
1769 }
1770
1771 /* Adjust snapused. */
1772 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1773 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1774 &used, &comp, &uncomp);
1775 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1776 -used, -comp, -uncomp, tx);
1777
1778 /* Move blocks to be freed to pool's free list. */
1779 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1780 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1781 tx);
1782 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1783 DD_USED_HEAD, used, comp, uncomp, tx);
1784
1785 /* Merge our deadlist into next's and free it. */
1786 dsl_deadlist_merge(&ds_next->ds_deadlist,
1787 ds->ds_phys->ds_deadlist_obj, tx);
1788 }
1789 dsl_deadlist_close(&ds->ds_deadlist);
1790 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1791
1792 /* Collapse range in clone heads */
1793 dsl_dataset_remove_clones_key(ds,
1794 ds->ds_phys->ds_creation_txg, tx);
1795
1796 if (dsl_dataset_is_snapshot(ds_next)) {
1797 dsl_dataset_t *ds_nextnext;
1798
1799 /*
1800 * Update next's unique to include blocks which
1801 * were previously shared by only this snapshot
1802 * and it. Those blocks will be born after the
1803 * prev snap and before this snap, and will have
1804 * died after the next snap and before the one
1805 * after that (ie. be on the snap after next's
1806 * deadlist).
1807 */
1808 VERIFY(0 == dsl_dataset_hold_obj(dp,
1809 ds_next->ds_phys->ds_next_snap_obj,
1810 FTAG, &ds_nextnext));
1811 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
1812 ds->ds_phys->ds_prev_snap_txg,
1813 ds->ds_phys->ds_creation_txg,
1814 &used, &comp, &uncomp);
1815 ds_next->ds_phys->ds_unique_bytes += used;
1816 dsl_dataset_rele(ds_nextnext, FTAG);
1817 ASSERT3P(ds_next->ds_prev, ==, NULL);
1818
1819 /* Collapse range in this head. */
1820 dsl_dataset_t *hds;
1821 VERIFY0(dsl_dataset_hold_obj(dp,
1822 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1823 FTAG, &hds));
1824 dsl_deadlist_remove_key(&hds->ds_deadlist,
1825 ds->ds_phys->ds_creation_txg, tx);
1826 dsl_dataset_rele(hds, FTAG);
1827
1828 } else {
1829 ASSERT3P(ds_next->ds_prev, ==, ds);
1830 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1831 ds_next->ds_prev = NULL;
1832 if (ds_prev) {
1833 VERIFY(0 == dsl_dataset_get_ref(dp,
1834 ds->ds_phys->ds_prev_snap_obj,
1835 ds_next, &ds_next->ds_prev));
1836 }
1837
1838 dsl_dataset_recalc_head_uniq(ds_next);
1839
1840 /*
1841 * Reduce the amount of our unconsmed refreservation
1842 * being charged to our parent by the amount of
1843 * new unique data we have gained.
1844 */
1845 if (old_unique < ds_next->ds_reserved) {
1846 int64_t mrsdelta;
1847 uint64_t new_unique =
1848 ds_next->ds_phys->ds_unique_bytes;
1849
1850 ASSERT(old_unique <= new_unique);
1851 mrsdelta = MIN(new_unique - old_unique,
1852 ds_next->ds_reserved - old_unique);
1853 dsl_dir_diduse_space(ds->ds_dir,
1854 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
1855 }
1856 }
1857 dsl_dataset_rele(ds_next, FTAG);
1858 } else {
1859 zfeature_info_t *async_destroy =
1860 &spa_feature_table[SPA_FEATURE_ASYNC_DESTROY];
1861
1862 /*
1863 * There's no next snapshot, so this is a head dataset.
1864 * Destroy the deadlist. Unless it's a clone, the
1865 * deadlist should be empty. (If it's a clone, it's
1866 * safe to ignore the deadlist contents.)
1867 */
1868 dsl_deadlist_close(&ds->ds_deadlist);
1869 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
1870 ds->ds_phys->ds_deadlist_obj = 0;
1871
1872 if (!spa_feature_is_enabled(dp->dp_spa, async_destroy)) {
1873 err = old_synchronous_dataset_destroy(ds, tx);
1874 } else {
1875 /*
1876 * Move the bptree into the pool's list of trees to
1877 * clean up and update space accounting information.
1878 */
1879 uint64_t used, comp, uncomp;
1880
1881 ASSERT(err == 0 || err == EBUSY);
1882 if (!spa_feature_is_active(dp->dp_spa, async_destroy)) {
1883 spa_feature_incr(dp->dp_spa, async_destroy, tx);
1884 dp->dp_bptree_obj = bptree_alloc(
1885 dp->dp_meta_objset, tx);
1886 VERIFY(zap_add(dp->dp_meta_objset,
1887 DMU_POOL_DIRECTORY_OBJECT,
1888 DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
1889 &dp->dp_bptree_obj, tx) == 0);
1890 }
1891
1892 used = ds->ds_dir->dd_phys->dd_used_bytes;
1893 comp = ds->ds_dir->dd_phys->dd_compressed_bytes;
1894 uncomp = ds->ds_dir->dd_phys->dd_uncompressed_bytes;
1895
1896 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
1897 ds->ds_phys->ds_unique_bytes == used);
1898
1899 bptree_add(dp->dp_meta_objset, dp->dp_bptree_obj,
1900 &ds->ds_phys->ds_bp, ds->ds_phys->ds_prev_snap_txg,
1901 used, comp, uncomp, tx);
1902 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
1903 -used, -comp, -uncomp, tx);
1904 dsl_dir_diduse_space(dp->dp_free_dir, DD_USED_HEAD,
1905 used, comp, uncomp, tx);
1906 }
1907
1908 if (ds->ds_prev != NULL) {
1909 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1910 VERIFY0(zap_remove_int(mos,
1911 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1912 ds->ds_object, tx));
1913 }
1914 dsl_dataset_rele(ds->ds_prev, ds);
1915 ds->ds_prev = ds_prev = NULL;
1916 }
1917 }
1918
1919 /*
1920 * This must be done after the dsl_traverse(), because it will
1921 * re-open the objset.
1922 */
1923 if (ds->ds_objset) {
1924 dmu_objset_evict(ds->ds_objset);
1925 ds->ds_objset = NULL;
1926 }
1927
1928 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
1929 /* Erase the link in the dir */
1930 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1931 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
1932 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1933 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1934 ASSERT(err == 0);
1935 } else {
1936 /* remove from snapshot namespace */
1937 dsl_dataset_t *ds_head;
1938 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1939 VERIFY(0 == dsl_dataset_hold_obj(dp,
1940 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
1941 VERIFY(0 == dsl_dataset_get_snapname(ds));
1942 #ifdef ZFS_DEBUG
1943 {
1944 uint64_t val;
1945
1946 err = dsl_dataset_snap_lookup(ds_head,
1947 ds->ds_snapname, &val);
1948 ASSERT0(err);
1949 ASSERT3U(val, ==, obj);
1950 }
1951 #endif
1952 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
1953 ASSERT(err == 0);
1954 dsl_dataset_rele(ds_head, FTAG);
1955 }
1956
1957 if (ds_prev && ds->ds_prev != ds_prev)
1958 dsl_dataset_rele(ds_prev, FTAG);
1959
1960 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
1961
1962 if (ds->ds_phys->ds_next_clones_obj != 0) {
1963 uint64_t count;
1964 ASSERT(0 == zap_count(mos,
1965 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1966 VERIFY(0 == dmu_object_free(mos,
1967 ds->ds_phys->ds_next_clones_obj, tx));
1968 }
1969 if (ds->ds_phys->ds_props_obj != 0)
1970 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
1971 if (ds->ds_phys->ds_userrefs_obj != 0)
1972 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
1973 dsl_dir_close(ds->ds_dir, ds);
1974 ds->ds_dir = NULL;
1975 dsl_dataset_drain_refs(ds, tag);
1976 VERIFY(0 == dmu_object_free(mos, obj, tx));
1977
1978 if (dsda->rm_origin) {
1979 /*
1980 * Remove the origin of the clone we just destroyed.
1981 */
1982 struct dsl_ds_destroyarg ndsda = {0};
1983
1984 ndsda.ds = dsda->rm_origin;
1985 dsl_dataset_destroy_sync(&ndsda, tag, tx);
1986 }
1987 }
1988
1989 static int
1990 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1991 {
1992 uint64_t asize;
1993
1994 if (!dmu_tx_is_syncing(tx))
1995 return (0);
1996
1997 /*
1998 * If there's an fs-only reservation, any blocks that might become
1999 * owned by the snapshot dataset must be accommodated by space
2000 * outside of the reservation.
2001 */
2002 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
2003 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2004 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
2005 return (ENOSPC);
2006
2007 /*
2008 * Propagate any reserved space for this snapshot to other
2009 * snapshot checks in this sync group.
2010 */
2011 if (asize > 0)
2012 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
2013
2014 return (0);
2015 }
2016
2017 int
2018 dsl_dataset_snapshot_check(dsl_dataset_t *ds, const char *snapname,
2019 dmu_tx_t *tx)
2020 {
2021 int err;
2022 uint64_t value;
2023
2024 /*
2025 * We don't allow multiple snapshots of the same txg. If there
2026 * is already one, try again.
2027 */
2028 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
2029 return (EAGAIN);
2030
2031 /*
2032 * Check for conflicting snapshot name.
2033 */
2034 err = dsl_dataset_snap_lookup(ds, snapname, &value);
2035 if (err == 0)
2036 return (EEXIST);
2037 if (err != ENOENT)
2038 return (err);
2039
2040 /*
2041 * Check that the dataset's name is not too long. Name consists
2042 * of the dataset's length + 1 for the @-sign + snapshot name's length
2043 */
2044 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2045 return (ENAMETOOLONG);
2046
2047 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2048 if (err)
2049 return (err);
2050
2051 ds->ds_trysnap_txg = tx->tx_txg;
2052 return (0);
2053 }
2054
2055 void
2056 dsl_dataset_snapshot_sync(dsl_dataset_t *ds, const char *snapname,
2057 dmu_tx_t *tx)
2058 {
2059 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2060 dmu_buf_t *dbuf;
2061 dsl_dataset_phys_t *dsphys;
2062 uint64_t dsobj, crtxg;
2063 objset_t *mos = dp->dp_meta_objset;
2064 int err;
2065
2066 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2067
2068 /*
2069 * The origin's ds_creation_txg has to be < TXG_INITIAL
2070 */
2071 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2072 crtxg = 1;
2073 else
2074 crtxg = tx->tx_txg;
2075
2076 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2077 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2078 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2079 dmu_buf_will_dirty(dbuf, tx);
2080 dsphys = dbuf->db_data;
2081 bzero(dsphys, sizeof (dsl_dataset_phys_t));
2082 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2083 dsphys->ds_fsid_guid = unique_create();
2084 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2085 sizeof (dsphys->ds_guid));
2086 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2087 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2088 dsphys->ds_next_snap_obj = ds->ds_object;
2089 dsphys->ds_num_children = 1;
2090 dsphys->ds_creation_time = gethrestime_sec();
2091 dsphys->ds_creation_txg = crtxg;
2092 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2093 dsphys->ds_referenced_bytes = ds->ds_phys->ds_referenced_bytes;
2094 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2095 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2096 dsphys->ds_flags = ds->ds_phys->ds_flags;
2097 dsphys->ds_bp = ds->ds_phys->ds_bp;
2098 dmu_buf_rele(dbuf, FTAG);
2099
2100 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2101 if (ds->ds_prev) {
2102 uint64_t next_clones_obj =
2103 ds->ds_prev->ds_phys->ds_next_clones_obj;
2104 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2105 ds->ds_object ||
2106 ds->ds_prev->ds_phys->ds_num_children > 1);
2107 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2108 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2109 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2110 ds->ds_prev->ds_phys->ds_creation_txg);
2111 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
2112 } else if (next_clones_obj != 0) {
2113 remove_from_next_clones(ds->ds_prev,
2114 dsphys->ds_next_snap_obj, tx);
2115 VERIFY0(zap_add_int(mos,
2116 next_clones_obj, dsobj, tx));
2117 }
2118 }
2119
2120 /*
2121 * If we have a reference-reservation on this dataset, we will
2122 * need to increase the amount of refreservation being charged
2123 * since our unique space is going to zero.
2124 */
2125 if (ds->ds_reserved) {
2126 int64_t delta;
2127 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2128 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
2129 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
2130 delta, 0, 0, tx);
2131 }
2132
2133 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2134 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2135 ds->ds_dir->dd_myname, snapname, dsobj,
2136 ds->ds_phys->ds_prev_snap_txg);
2137 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2138 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2139 dsl_deadlist_close(&ds->ds_deadlist);
2140 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2141 dsl_deadlist_add_key(&ds->ds_deadlist,
2142 ds->ds_phys->ds_prev_snap_txg, tx);
2143
2144 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2145 ds->ds_phys->ds_prev_snap_obj = dsobj;
2146 ds->ds_phys->ds_prev_snap_txg = crtxg;
2147 ds->ds_phys->ds_unique_bytes = 0;
2148 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2149 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
2150
2151 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2152 snapname, 8, 1, &dsobj, tx);
2153 ASSERT(err == 0);
2154
2155 if (ds->ds_prev)
2156 dsl_dataset_drop_ref(ds->ds_prev, ds);
2157 VERIFY(0 == dsl_dataset_get_ref(dp,
2158 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2159
2160 dsl_scan_ds_snapshotted(ds, tx);
2161
2162 dsl_dir_snap_cmtime_update(ds->ds_dir);
2163
2164 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
2165 }
2166
2167 void
2168 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2169 {
2170 ASSERT(dmu_tx_is_syncing(tx));
2171 ASSERT(ds->ds_objset != NULL);
2172 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2173
2174 /*
2175 * in case we had to change ds_fsid_guid when we opened it,
2176 * sync it out now.
2177 */
2178 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2179 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2180
2181 dsl_dir_dirty(ds->ds_dir, tx);
2182 dmu_objset_sync(ds->ds_objset, zio, tx);
2183 }
2184
2185 static void
2186 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
2187 {
2188 uint64_t count = 0;
2189 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
2190 zap_cursor_t zc;
2191 zap_attribute_t za;
2192 nvlist_t *propval;
2193 nvlist_t *val;
2194
2195 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2196 VERIFY(nvlist_alloc(&propval, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2197 VERIFY(nvlist_alloc(&val, NV_UNIQUE_NAME, KM_SLEEP) == 0);
2198
2199 /*
2200 * There may me missing entries in ds_next_clones_obj
2201 * due to a bug in a previous version of the code.
2202 * Only trust it if it has the right number of entries.
2203 */
2204 if (ds->ds_phys->ds_next_clones_obj != 0) {
2205 ASSERT0(zap_count(mos, ds->ds_phys->ds_next_clones_obj,
2206 &count));
2207 }
2208 if (count != ds->ds_phys->ds_num_children - 1) {
2209 goto fail;
2210 }
2211 for (zap_cursor_init(&zc, mos, ds->ds_phys->ds_next_clones_obj);
2212 zap_cursor_retrieve(&zc, &za) == 0;
2213 zap_cursor_advance(&zc)) {
2214 dsl_dataset_t *clone;
2215 char buf[ZFS_MAXNAMELEN];
2216 /*
2217 * Even though we hold the dp_config_rwlock, the dataset
2218 * may fail to open, returning ENOENT. If there is a
2219 * thread concurrently attempting to destroy this
2220 * dataset, it will have the ds_rwlock held for
2221 * RW_WRITER. Our call to dsl_dataset_hold_obj() ->
2222 * dsl_dataset_hold_ref() will fail its
2223 * rw_tryenter(&ds->ds_rwlock, RW_READER), drop the
2224 * dp_config_rwlock, and wait for the destroy progress
2225 * and signal ds_exclusive_cv. If the destroy was
2226 * successful, we will see that
2227 * DSL_DATASET_IS_DESTROYED(), and return ENOENT.
2228 */
2229 if (dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
2230 za.za_first_integer, FTAG, &clone) != 0)
2231 continue;
2232 dsl_dir_name(clone->ds_dir, buf);
2233 VERIFY(nvlist_add_boolean(val, buf) == 0);
2234 dsl_dataset_rele(clone, FTAG);
2235 }
2236 zap_cursor_fini(&zc);
2237 VERIFY(nvlist_add_nvlist(propval, ZPROP_VALUE, val) == 0);
2238 VERIFY(nvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
2239 propval) == 0);
2240 fail:
2241 nvlist_free(val);
2242 nvlist_free(propval);
2243 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2244 }
2245
2246 void
2247 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2248 {
2249 uint64_t refd, avail, uobjs, aobjs, ratio;
2250
2251 ratio = ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2252 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2253 ds->ds_phys->ds_compressed_bytes);
2254
2255 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
2256
2257 if (dsl_dataset_is_snapshot(ds)) {
2258 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
2259 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2260 ds->ds_phys->ds_unique_bytes);
2261 get_clones_stat(ds, nv);
2262 } else {
2263 dsl_dir_stats(ds->ds_dir, nv);
2264 }
2265
2266 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2267 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2268 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2269
2270 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2271 ds->ds_phys->ds_creation_time);
2272 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2273 ds->ds_phys->ds_creation_txg);
2274 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2275 ds->ds_quota);
2276 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2277 ds->ds_reserved);
2278 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2279 ds->ds_phys->ds_guid);
2280 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2281 ds->ds_phys->ds_unique_bytes);
2282 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2283 ds->ds_object);
2284 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2285 ds->ds_userrefs);
2286 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2287 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2288
2289 if (ds->ds_phys->ds_prev_snap_obj != 0) {
2290 uint64_t written, comp, uncomp;
2291 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2292 dsl_dataset_t *prev;
2293
2294 rw_enter(&dp->dp_config_rwlock, RW_READER);
2295 int err = dsl_dataset_hold_obj(dp,
2296 ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
2297 rw_exit(&dp->dp_config_rwlock);
2298 if (err == 0) {
2299 err = dsl_dataset_space_written(prev, ds, &written,
2300 &comp, &uncomp);
2301 dsl_dataset_rele(prev, FTAG);
2302 if (err == 0) {
2303 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2304 written);
2305 }
2306 }
2307 }
2308
2309 }
2310
2311 void
2312 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2313 {
2314 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2315 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2316 stat->dds_guid = ds->ds_phys->ds_guid;
2317 stat->dds_origin[0] = '\0';
2318 if (dsl_dataset_is_snapshot(ds)) {
2319 stat->dds_is_snapshot = B_TRUE;
2320 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
2321 } else {
2322 stat->dds_is_snapshot = B_FALSE;
2323 stat->dds_num_clones = 0;
2324
2325 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
2326 if (dsl_dir_is_clone(ds->ds_dir)) {
2327 dsl_dataset_t *ods;
2328
2329 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2330 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
2331 dsl_dataset_name(ods, stat->dds_origin);
2332 dsl_dataset_drop_ref(ods, FTAG);
2333 }
2334 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2335 }
2336 }
2337
2338 uint64_t
2339 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2340 {
2341 return (ds->ds_fsid_guid);
2342 }
2343
2344 void
2345 dsl_dataset_space(dsl_dataset_t *ds,
2346 uint64_t *refdbytesp, uint64_t *availbytesp,
2347 uint64_t *usedobjsp, uint64_t *availobjsp)
2348 {
2349 *refdbytesp = ds->ds_phys->ds_referenced_bytes;
2350 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2351 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2352 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2353 if (ds->ds_quota != 0) {
2354 /*
2355 * Adjust available bytes according to refquota
2356 */
2357 if (*refdbytesp < ds->ds_quota)
2358 *availbytesp = MIN(*availbytesp,
2359 ds->ds_quota - *refdbytesp);
2360 else
2361 *availbytesp = 0;
2362 }
2363 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2364 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2365 }
2366
2367 boolean_t
2368 dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2369 {
2370 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2371
2372 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2373 dsl_pool_sync_context(dp));
2374 if (ds->ds_prev == NULL)
2375 return (B_FALSE);
2376 if (ds->ds_phys->ds_bp.blk_birth >
2377 ds->ds_prev->ds_phys->ds_creation_txg) {
2378 objset_t *os, *os_prev;
2379 /*
2380 * It may be that only the ZIL differs, because it was
2381 * reset in the head. Don't count that as being
2382 * modified.
2383 */
2384 if (dmu_objset_from_ds(ds, &os) != 0)
2385 return (B_TRUE);
2386 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2387 return (B_TRUE);
2388 return (bcmp(&os->os_phys->os_meta_dnode,
2389 &os_prev->os_phys->os_meta_dnode,
2390 sizeof (os->os_phys->os_meta_dnode)) != 0);
2391 }
2392 return (B_FALSE);
2393 }
2394
2395 /* ARGSUSED */
2396 static int
2397 dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2398 {
2399 dsl_dataset_t *ds = arg1;
2400 char *newsnapname = arg2;
2401 dsl_dir_t *dd = ds->ds_dir;
2402 dsl_dataset_t *hds;
2403 uint64_t val;
2404 int err;
2405
2406 err = dsl_dataset_hold_obj(dd->dd_pool,
2407 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
2408 if (err)
2409 return (err);
2410
2411 /* new name better not be in use */
2412 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2413 dsl_dataset_rele(hds, FTAG);
2414
2415 if (err == 0)
2416 err = EEXIST;
2417 else if (err == ENOENT)
2418 err = 0;
2419
2420 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2421 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2422 err = ENAMETOOLONG;
2423
2424 return (err);
2425 }
2426
2427 static void
2428 dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2429 {
2430 dsl_dataset_t *ds = arg1;
2431 const char *newsnapname = arg2;
2432 dsl_dir_t *dd = ds->ds_dir;
2433 objset_t *mos = dd->dd_pool->dp_meta_objset;
2434 dsl_dataset_t *hds;
2435 int err;
2436
2437 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2438
2439 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2440 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
2441
2442 VERIFY(0 == dsl_dataset_get_snapname(ds));
2443 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
2444 ASSERT0(err);
2445 mutex_enter(&ds->ds_lock);
2446 (void) strcpy(ds->ds_snapname, newsnapname);
2447 mutex_exit(&ds->ds_lock);
2448 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2449 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2450 ASSERT0(err);
2451
2452 spa_history_log_internal_ds(ds, "rename", tx,
2453 "-> @%s", newsnapname);
2454 dsl_dataset_rele(hds, FTAG);
2455 }
2456
2457 struct renamesnaparg {
2458 dsl_sync_task_group_t *dstg;
2459 char failed[MAXPATHLEN];
2460 char *oldsnap;
2461 char *newsnap;
2462 };
2463
2464 static int
2465 dsl_snapshot_rename_one(const char *name, void *arg)
2466 {
2467 struct renamesnaparg *ra = arg;
2468 dsl_dataset_t *ds = NULL;
2469 char *snapname;
2470 int err;
2471
2472 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2473 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
2474
2475 /*
2476 * For recursive snapshot renames the parent won't be changing
2477 * so we just pass name for both the to/from argument.
2478 */
2479 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2480 if (err != 0) {
2481 strfree(snapname);
2482 return (err == ENOENT ? 0 : err);
2483 }
2484
2485 #ifdef _KERNEL
2486 /*
2487 * For all filesystems undergoing rename, we'll need to unmount it.
2488 */
2489 (void) zfs_unmount_snap(snapname, NULL);
2490 #endif
2491 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2492 strfree(snapname);
2493 if (err != 0)
2494 return (err == ENOENT ? 0 : err);
2495
2496 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2497 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2498
2499 return (0);
2500 }
2501
2502 static int
2503 dsl_recursive_rename(char *oldname, const char *newname)
2504 {
2505 int err;
2506 struct renamesnaparg *ra;
2507 dsl_sync_task_t *dst;
2508 spa_t *spa;
2509 char *cp, *fsname = spa_strdup(oldname);
2510 int len = strlen(oldname) + 1;
2511
2512 /* truncate the snapshot name to get the fsname */
2513 cp = strchr(fsname, '@');
2514 *cp = '\0';
2515
2516 err = spa_open(fsname, &spa, FTAG);
2517 if (err) {
2518 kmem_free(fsname, len);
2519 return (err);
2520 }
2521 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2522 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2523
2524 ra->oldsnap = strchr(oldname, '@') + 1;
2525 ra->newsnap = strchr(newname, '@') + 1;
2526 *ra->failed = '\0';
2527
2528 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2529 DS_FIND_CHILDREN);
2530 kmem_free(fsname, len);
2531
2532 if (err == 0) {
2533 err = dsl_sync_task_group_wait(ra->dstg);
2534 }
2535
2536 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2537 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2538 dsl_dataset_t *ds = dst->dst_arg1;
2539 if (dst->dst_err) {
2540 dsl_dir_name(ds->ds_dir, ra->failed);
2541 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2542 (void) strlcat(ra->failed, ra->newsnap,
2543 sizeof (ra->failed));
2544 }
2545 dsl_dataset_rele(ds, ra->dstg);
2546 }
2547
2548 if (err)
2549 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
2550
2551 dsl_sync_task_group_destroy(ra->dstg);
2552 kmem_free(ra, sizeof (struct renamesnaparg));
2553 spa_close(spa, FTAG);
2554 return (err);
2555 }
2556
2557 static int
2558 dsl_valid_rename(const char *oldname, void *arg)
2559 {
2560 int delta = *(int *)arg;
2561
2562 if (strlen(oldname) + delta >= MAXNAMELEN)
2563 return (ENAMETOOLONG);
2564
2565 return (0);
2566 }
2567
2568 #pragma weak dmu_objset_rename = dsl_dataset_rename
2569 int
2570 dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
2571 {
2572 dsl_dir_t *dd;
2573 dsl_dataset_t *ds;
2574 const char *tail;
2575 int err;
2576
2577 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2578 if (err)
2579 return (err);
2580
2581 if (tail == NULL) {
2582 int delta = strlen(newname) - strlen(oldname);
2583
2584 /* if we're growing, validate child name lengths */
2585 if (delta > 0)
2586 err = dmu_objset_find(oldname, dsl_valid_rename,
2587 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2588
2589 if (err == 0)
2590 err = dsl_dir_rename(dd, newname);
2591 dsl_dir_close(dd, FTAG);
2592 return (err);
2593 }
2594
2595 if (tail[0] != '@') {
2596 /* the name ended in a nonexistent component */
2597 dsl_dir_close(dd, FTAG);
2598 return (ENOENT);
2599 }
2600
2601 dsl_dir_close(dd, FTAG);
2602
2603 /* new name must be snapshot in same filesystem */
2604 tail = strchr(newname, '@');
2605 if (tail == NULL)
2606 return (EINVAL);
2607 tail++;
2608 if (strncmp(oldname, newname, tail - newname) != 0)
2609 return (EXDEV);
2610
2611 if (recursive) {
2612 err = dsl_recursive_rename(oldname, newname);
2613 } else {
2614 err = dsl_dataset_hold(oldname, FTAG, &ds);
2615 if (err)
2616 return (err);
2617
2618 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2619 dsl_dataset_snapshot_rename_check,
2620 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2621
2622 dsl_dataset_rele(ds, FTAG);
2623 }
2624
2625 return (err);
2626 }
2627
2628 struct promotenode {
2629 list_node_t link;
2630 dsl_dataset_t *ds;
2631 };
2632
2633 struct promotearg {
2634 list_t shared_snaps, origin_snaps, clone_snaps;
2635 dsl_dataset_t *origin_origin;
2636 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2637 char *err_ds;
2638 };
2639
2640 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2641 static boolean_t snaplist_unstable(list_t *l);
2642
2643 static int
2644 dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2645 {
2646 dsl_dataset_t *hds = arg1;
2647 struct promotearg *pa = arg2;
2648 struct promotenode *snap = list_head(&pa->shared_snaps);
2649 dsl_dataset_t *origin_ds = snap->ds;
2650 int err;
2651 uint64_t unused;
2652
2653 /* Check that it is a real clone */
2654 if (!dsl_dir_is_clone(hds->ds_dir))
2655 return (EINVAL);
2656
2657 /* Since this is so expensive, don't do the preliminary check */
2658 if (!dmu_tx_is_syncing(tx))
2659 return (0);
2660
2661 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2662 return (EXDEV);
2663
2664 /* compute origin's new unique space */
2665 snap = list_tail(&pa->clone_snaps);
2666 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2667 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2668 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2669 &pa->unique, &unused, &unused);
2670
2671 /*
2672 * Walk the snapshots that we are moving
2673 *
2674 * Compute space to transfer. Consider the incremental changes
2675 * to used for each snapshot:
2676 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2677 * So each snapshot gave birth to:
2678 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2679 * So a sequence would look like:
2680 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2681 * Which simplifies to:
2682 * uN + kN + kN-1 + ... + k1 + k0
2683 * Note however, if we stop before we reach the ORIGIN we get:
2684 * uN + kN + kN-1 + ... + kM - uM-1
2685 */
2686 pa->used = origin_ds->ds_phys->ds_referenced_bytes;
2687 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2688 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2689 for (snap = list_head(&pa->shared_snaps); snap;
2690 snap = list_next(&pa->shared_snaps, snap)) {
2691 uint64_t val, dlused, dlcomp, dluncomp;
2692 dsl_dataset_t *ds = snap->ds;
2693
2694 /* Check that the snapshot name does not conflict */
2695 VERIFY(0 == dsl_dataset_get_snapname(ds));
2696 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2697 if (err == 0) {
2698 err = EEXIST;
2699 goto out;
2700 }
2701 if (err != ENOENT)
2702 goto out;
2703
2704 /* The very first snapshot does not have a deadlist */
2705 if (ds->ds_phys->ds_prev_snap_obj == 0)
2706 continue;
2707
2708 dsl_deadlist_space(&ds->ds_deadlist,
2709 &dlused, &dlcomp, &dluncomp);
2710 pa->used += dlused;
2711 pa->comp += dlcomp;
2712 pa->uncomp += dluncomp;
2713 }
2714
2715 /*
2716 * If we are a clone of a clone then we never reached ORIGIN,
2717 * so we need to subtract out the clone origin's used space.
2718 */
2719 if (pa->origin_origin) {
2720 pa->used -= pa->origin_origin->ds_phys->ds_referenced_bytes;
2721 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2722 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
2723 }
2724
2725 /* Check that there is enough space here */
2726 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2727 pa->used);
2728 if (err)
2729 return (err);
2730
2731 /*
2732 * Compute the amounts of space that will be used by snapshots
2733 * after the promotion (for both origin and clone). For each,
2734 * it is the amount of space that will be on all of their
2735 * deadlists (that was not born before their new origin).
2736 */
2737 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2738 uint64_t space;
2739
2740 /*
2741 * Note, typically this will not be a clone of a clone,
2742 * so dd_origin_txg will be < TXG_INITIAL, so
2743 * these snaplist_space() -> dsl_deadlist_space_range()
2744 * calls will be fast because they do not have to
2745 * iterate over all bps.
2746 */
2747 snap = list_head(&pa->origin_snaps);
2748 err = snaplist_space(&pa->shared_snaps,
2749 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
2750 if (err)
2751 return (err);
2752
2753 err = snaplist_space(&pa->clone_snaps,
2754 snap->ds->ds_dir->dd_origin_txg, &space);
2755 if (err)
2756 return (err);
2757 pa->cloneusedsnap += space;
2758 }
2759 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2760 err = snaplist_space(&pa->origin_snaps,
2761 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2762 if (err)
2763 return (err);
2764 }
2765
2766 return (0);
2767 out:
2768 pa->err_ds = snap->ds->ds_snapname;
2769 return (err);
2770 }
2771
2772 static void
2773 dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
2774 {
2775 dsl_dataset_t *hds = arg1;
2776 struct promotearg *pa = arg2;
2777 struct promotenode *snap = list_head(&pa->shared_snaps);
2778 dsl_dataset_t *origin_ds = snap->ds;
2779 dsl_dataset_t *origin_head;
2780 dsl_dir_t *dd = hds->ds_dir;
2781 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2782 dsl_dir_t *odd = NULL;
2783 uint64_t oldnext_obj;
2784 int64_t delta;
2785
2786 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2787
2788 snap = list_head(&pa->origin_snaps);
2789 origin_head = snap->ds;
2790
2791 /*
2792 * We need to explicitly open odd, since origin_ds's dd will be
2793 * changing.
2794 */
2795 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2796 NULL, FTAG, &odd));
2797
2798 /* change origin's next snap */
2799 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2800 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2801 snap = list_tail(&pa->clone_snaps);
2802 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2803 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2804
2805 /* change the origin's next clone */
2806 if (origin_ds->ds_phys->ds_next_clones_obj) {
2807 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
2808 VERIFY0(zap_add_int(dp->dp_meta_objset,
2809 origin_ds->ds_phys->ds_next_clones_obj,
2810 oldnext_obj, tx));
2811 }
2812
2813 /* change origin */
2814 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2815 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2816 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
2817 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2818 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2819 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
2820 origin_head->ds_dir->dd_origin_txg =
2821 origin_ds->ds_phys->ds_creation_txg;
2822
2823 /* change dd_clone entries */
2824 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2825 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2826 odd->dd_phys->dd_clones, hds->ds_object, tx));
2827 VERIFY0(zap_add_int(dp->dp_meta_objset,
2828 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2829 hds->ds_object, tx));
2830
2831 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2832 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2833 origin_head->ds_object, tx));
2834 if (dd->dd_phys->dd_clones == 0) {
2835 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2836 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2837 }
2838 VERIFY0(zap_add_int(dp->dp_meta_objset,
2839 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2840
2841 }
2842
2843 /* move snapshots to this dir */
2844 for (snap = list_head(&pa->shared_snaps); snap;
2845 snap = list_next(&pa->shared_snaps, snap)) {
2846 dsl_dataset_t *ds = snap->ds;
2847
2848 /* unregister props as dsl_dir is changing */
2849 if (ds->ds_objset) {
2850 dmu_objset_evict(ds->ds_objset);
2851 ds->ds_objset = NULL;
2852 }
2853 /* move snap name entry */
2854 VERIFY(0 == dsl_dataset_get_snapname(ds));
2855 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2856 ds->ds_snapname, tx));
2857 VERIFY(0 == zap_add(dp->dp_meta_objset,
2858 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2859 8, 1, &ds->ds_object, tx));
2860
2861 /* change containing dsl_dir */
2862 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2863 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2864 ds->ds_phys->ds_dir_obj = dd->dd_object;
2865 ASSERT3P(ds->ds_dir, ==, odd);
2866 dsl_dir_close(ds->ds_dir, ds);
2867 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2868 NULL, ds, &ds->ds_dir));
2869
2870 /* move any clone references */
2871 if (ds->ds_phys->ds_next_clones_obj &&
2872 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2873 zap_cursor_t zc;
2874 zap_attribute_t za;
2875
2876 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2877 ds->ds_phys->ds_next_clones_obj);
2878 zap_cursor_retrieve(&zc, &za) == 0;
2879 zap_cursor_advance(&zc)) {
2880 dsl_dataset_t *cnds;
2881 uint64_t o;
2882
2883 if (za.za_first_integer == oldnext_obj) {
2884 /*
2885 * We've already moved the
2886 * origin's reference.
2887 */
2888 continue;
2889 }
2890
2891 VERIFY0(dsl_dataset_hold_obj(dp,
2892 za.za_first_integer, FTAG, &cnds));
2893 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2894
2895 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2896 odd->dd_phys->dd_clones, o, tx), ==, 0);
2897 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2898 dd->dd_phys->dd_clones, o, tx), ==, 0);
2899 dsl_dataset_rele(cnds, FTAG);
2900 }
2901 zap_cursor_fini(&zc);
2902 }
2903
2904 ASSERT0(dsl_prop_numcb(ds));
2905 }
2906
2907 /*
2908 * Change space accounting.
2909 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2910 * both be valid, or both be 0 (resulting in delta == 0). This
2911 * is true for each of {clone,origin} independently.
2912 */
2913
2914 delta = pa->cloneusedsnap -
2915 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2916 ASSERT3S(delta, >=, 0);
2917 ASSERT3U(pa->used, >=, delta);
2918 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2919 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2920 pa->used - delta, pa->comp, pa->uncomp, tx);
2921
2922 delta = pa->originusedsnap -
2923 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2924 ASSERT3S(delta, <=, 0);
2925 ASSERT3U(pa->used, >=, -delta);
2926 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2927 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2928 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
2929
2930 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2931
2932 /* log history record */
2933 spa_history_log_internal_ds(hds, "promote", tx, "");
2934
2935 dsl_dir_close(odd, FTAG);
2936 }
2937
2938 static char *snaplist_tag = "snaplist";
2939 /*
2940 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2941 * (exclusive) and last_obj (inclusive). The list will be in reverse
2942 * order (last_obj will be the list_head()). If first_obj == 0, do all
2943 * snapshots back to this dataset's origin.
2944 */
2945 static int
2946 snaplist_make(dsl_pool_t *dp, boolean_t own,
2947 uint64_t first_obj, uint64_t last_obj, list_t *l)
2948 {
2949 uint64_t obj = last_obj;
2950
2951 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2952
2953 list_create(l, sizeof (struct promotenode),
2954 offsetof(struct promotenode, link));
2955
2956 while (obj != first_obj) {
2957 dsl_dataset_t *ds;
2958 struct promotenode *snap;
2959 int err;
2960
2961 if (own) {
2962 err = dsl_dataset_own_obj(dp, obj,
2963 0, snaplist_tag, &ds);
2964 if (err == 0)
2965 dsl_dataset_make_exclusive(ds, snaplist_tag);
2966 } else {
2967 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2968 }
2969 if (err == ENOENT) {
2970 /* lost race with snapshot destroy */
2971 struct promotenode *last = list_tail(l);
2972 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2973 obj = last->ds->ds_phys->ds_prev_snap_obj;
2974 continue;
2975 } else if (err) {
2976 return (err);
2977 }
2978
2979 if (first_obj == 0)
2980 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2981
2982 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2983 snap->ds = ds;
2984 list_insert_tail(l, snap);
2985 obj = ds->ds_phys->ds_prev_snap_obj;
2986 }
2987
2988 return (0);
2989 }
2990
2991 static int
2992 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2993 {
2994 struct promotenode *snap;
2995
2996 *spacep = 0;
2997 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2998 uint64_t used, comp, uncomp;
2999 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3000 mintxg, UINT64_MAX, &used, &comp, &uncomp);
3001 *spacep += used;
3002 }
3003 return (0);
3004 }
3005
3006 static void
3007 snaplist_destroy(list_t *l, boolean_t own)
3008 {
3009 struct promotenode *snap;
3010
3011 if (!l || !list_link_active(&l->list_head))
3012 return;
3013
3014 while ((snap = list_tail(l)) != NULL) {
3015 list_remove(l, snap);
3016 if (own)
3017 dsl_dataset_disown(snap->ds, snaplist_tag);
3018 else
3019 dsl_dataset_rele(snap->ds, snaplist_tag);
3020 kmem_free(snap, sizeof (struct promotenode));
3021 }
3022 list_destroy(l);
3023 }
3024
3025 /*
3026 * Promote a clone. Nomenclature note:
3027 * "clone" or "cds": the original clone which is being promoted
3028 * "origin" or "ods": the snapshot which is originally clone's origin
3029 * "origin head" or "ohds": the dataset which is the head
3030 * (filesystem/volume) for the origin
3031 * "origin origin": the origin of the origin's filesystem (typically
3032 * NULL, indicating that the clone is not a clone of a clone).
3033 */
3034 int
3035 dsl_dataset_promote(const char *name, char *conflsnap)
3036 {
3037 dsl_dataset_t *ds;
3038 dsl_dir_t *dd;
3039 dsl_pool_t *dp;
3040 dmu_object_info_t doi;
3041 struct promotearg pa = { 0 };
3042 struct promotenode *snap;
3043 int err;
3044
3045 err = dsl_dataset_hold(name, FTAG, &ds);
3046 if (err)
3047 return (err);
3048 dd = ds->ds_dir;
3049 dp = dd->dd_pool;
3050
3051 err = dmu_object_info(dp->dp_meta_objset,
3052 ds->ds_phys->ds_snapnames_zapobj, &doi);
3053 if (err) {
3054 dsl_dataset_rele(ds, FTAG);
3055 return (err);
3056 }
3057
3058 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
3059 dsl_dataset_rele(ds, FTAG);
3060 return (EINVAL);
3061 }
3062
3063 /*
3064 * We are going to inherit all the snapshots taken before our
3065 * origin (i.e., our new origin will be our parent's origin).
3066 * Take ownership of them so that we can rename them into our
3067 * namespace.
3068 */
3069 rw_enter(&dp->dp_config_rwlock, RW_READER);
3070
3071 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
3072 &pa.shared_snaps);
3073 if (err != 0)
3074 goto out;
3075
3076 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
3077 if (err != 0)
3078 goto out;
3079
3080 snap = list_head(&pa.shared_snaps);
3081 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
3082 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
3083 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
3084 if (err != 0)
3085 goto out;
3086
3087 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
3088 err = dsl_dataset_hold_obj(dp,
3089 snap->ds->ds_dir->dd_phys->dd_origin_obj,
3090 FTAG, &pa.origin_origin);
3091 if (err != 0)
3092 goto out;
3093 }
3094
3095 out:
3096 rw_exit(&dp->dp_config_rwlock);
3097
3098 /*
3099 * Add in 128x the snapnames zapobj size, since we will be moving
3100 * a bunch of snapnames to the promoted ds, and dirtying their
3101 * bonus buffers.
3102 */
3103 if (err == 0) {
3104 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
3105 dsl_dataset_promote_sync, ds, &pa,
3106 2 + 2 * doi.doi_physical_blocks_512);
3107 if (err && pa.err_ds && conflsnap)
3108 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
3109 }
3110
3111 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3112 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3113 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3114 if (pa.origin_origin)
3115 dsl_dataset_rele(pa.origin_origin, FTAG);
3116 dsl_dataset_rele(ds, FTAG);
3117 return (err);
3118 }
3119
3120 struct cloneswaparg {
3121 dsl_dataset_t *cds; /* clone dataset */
3122 dsl_dataset_t *ohds; /* origin's head dataset */
3123 boolean_t force;
3124 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3125 };
3126
3127 /* ARGSUSED */
3128 static int
3129 dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3130 {
3131 struct cloneswaparg *csa = arg1;
3132
3133 /* they should both be heads */
3134 if (dsl_dataset_is_snapshot(csa->cds) ||
3135 dsl_dataset_is_snapshot(csa->ohds))
3136 return (EINVAL);
3137
3138 /* the branch point should be just before them */
3139 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3140 return (EINVAL);
3141
3142 /* cds should be the clone (unless they are unrelated) */
3143 if (csa->cds->ds_prev != NULL &&
3144 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3145 csa->ohds->ds_object !=
3146 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
3147 return (EINVAL);
3148
3149 /* the clone should be a child of the origin */
3150 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3151 return (EINVAL);
3152
3153 /* ohds shouldn't be modified unless 'force' */
3154 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3155 return (ETXTBSY);
3156
3157 /* adjust amount of any unconsumed refreservation */
3158 csa->unused_refres_delta =
3159 (int64_t)MIN(csa->ohds->ds_reserved,
3160 csa->ohds->ds_phys->ds_unique_bytes) -
3161 (int64_t)MIN(csa->ohds->ds_reserved,
3162 csa->cds->ds_phys->ds_unique_bytes);
3163
3164 if (csa->unused_refres_delta > 0 &&
3165 csa->unused_refres_delta >
3166 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3167 return (ENOSPC);
3168
3169 if (csa->ohds->ds_quota != 0 &&
3170 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3171 return (EDQUOT);
3172
3173 return (0);
3174 }
3175
3176 /* ARGSUSED */
3177 static void
3178 dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3179 {
3180 struct cloneswaparg *csa = arg1;
3181 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
3182
3183 ASSERT(csa->cds->ds_reserved == 0);
3184 ASSERT(csa->ohds->ds_quota == 0 ||
3185 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
3186
3187 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3188 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
3189
3190 if (csa->cds->ds_objset != NULL) {
3191 dmu_objset_evict(csa->cds->ds_objset);
3192 csa->cds->ds_objset = NULL;
3193 }
3194
3195 if (csa->ohds->ds_objset != NULL) {
3196 dmu_objset_evict(csa->ohds->ds_objset);
3197 csa->ohds->ds_objset = NULL;
3198 }
3199
3200 /*
3201 * Reset origin's unique bytes, if it exists.
3202 */
3203 if (csa->cds->ds_prev) {
3204 dsl_dataset_t *origin = csa->cds->ds_prev;
3205 uint64_t comp, uncomp;
3206
3207 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3208 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3209 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3210 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3211 }
3212
3213 /* swap blkptrs */
3214 {
3215 blkptr_t tmp;
3216 tmp = csa->ohds->ds_phys->ds_bp;
3217 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3218 csa->cds->ds_phys->ds_bp = tmp;
3219 }
3220
3221 /* set dd_*_bytes */
3222 {
3223 int64_t dused, dcomp, duncomp;
3224 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3225 uint64_t odl_used, odl_comp, odl_uncomp;
3226
3227 ASSERT3U(csa->cds->ds_dir->dd_phys->
3228 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3229
3230 dsl_deadlist_space(&csa->cds->ds_deadlist,
3231 &cdl_used, &cdl_comp, &cdl_uncomp);
3232 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3233 &odl_used, &odl_comp, &odl_uncomp);
3234
3235 dused = csa->cds->ds_phys->ds_referenced_bytes + cdl_used -
3236 (csa->ohds->ds_phys->ds_referenced_bytes + odl_used);
3237 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3238 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3239 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3240 cdl_uncomp -
3241 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3242
3243 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
3244 dused, dcomp, duncomp, tx);
3245 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
3246 -dused, -dcomp, -duncomp, tx);
3247
3248 /*
3249 * The difference in the space used by snapshots is the
3250 * difference in snapshot space due to the head's
3251 * deadlist (since that's the only thing that's
3252 * changing that affects the snapused).
3253 */
3254 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3255 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3256 &cdl_used, &cdl_comp, &cdl_uncomp);
3257 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3258 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3259 &odl_used, &odl_comp, &odl_uncomp);
3260 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3261 DD_USED_HEAD, DD_USED_SNAP, tx);
3262 }
3263
3264 /* swap ds_*_bytes */
3265 SWITCH64(csa->ohds->ds_phys->ds_referenced_bytes,
3266 csa->cds->ds_phys->ds_referenced_bytes);
3267 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3268 csa->cds->ds_phys->ds_compressed_bytes);
3269 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3270 csa->cds->ds_phys->ds_uncompressed_bytes);
3271 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3272 csa->cds->ds_phys->ds_unique_bytes);
3273
3274 /* apply any parent delta for change in unconsumed refreservation */
3275 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3276 csa->unused_refres_delta, 0, 0, tx);
3277
3278 /*
3279 * Swap deadlists.
3280 */
3281 dsl_deadlist_close(&csa->cds->ds_deadlist);
3282 dsl_deadlist_close(&csa->ohds->ds_deadlist);
3283 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3284 csa->cds->ds_phys->ds_deadlist_obj);
3285 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3286 csa->cds->ds_phys->ds_deadlist_obj);
3287 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3288 csa->ohds->ds_phys->ds_deadlist_obj);
3289
3290 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
3291
3292 spa_history_log_internal_ds(csa->cds, "clone swap", tx,
3293 "parent=%s", csa->ohds->ds_dir->dd_myname);
3294 }
3295
3296 /*
3297 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3298 * recv" into an existing fs to swizzle the file system to the new
3299 * version, and by "zfs rollback". Can also be used to swap two
3300 * independent head datasets if neither has any snapshots.
3301 */
3302 int
3303 dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3304 boolean_t force)
3305 {
3306 struct cloneswaparg csa;
3307 int error;
3308
3309 ASSERT(clone->ds_owner);
3310 ASSERT(origin_head->ds_owner);
3311 retry:
3312 /*
3313 * Need exclusive access for the swap. If we're swapping these
3314 * datasets back after an error, we already hold the locks.
3315 */
3316 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3317 rw_enter(&clone->ds_rwlock, RW_WRITER);
3318 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3319 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
3320 rw_exit(&clone->ds_rwlock);
3321 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3322 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3323 rw_exit(&origin_head->ds_rwlock);
3324 goto retry;
3325 }
3326 }
3327 csa.cds = clone;
3328 csa.ohds = origin_head;
3329 csa.force = force;
3330 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
3331 dsl_dataset_clone_swap_check,
3332 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3333 return (error);
3334 }
3335
3336 /*
3337 * Given a pool name and a dataset object number in that pool,
3338 * return the name of that dataset.
3339 */
3340 int
3341 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3342 {
3343 spa_t *spa;
3344 dsl_pool_t *dp;
3345 dsl_dataset_t *ds;
3346 int error;
3347
3348 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3349 return (error);
3350 dp = spa_get_dsl(spa);
3351 rw_enter(&dp->dp_config_rwlock, RW_READER);
3352 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3353 dsl_dataset_name(ds, buf);
3354 dsl_dataset_rele(ds, FTAG);
3355 }
3356 rw_exit(&dp->dp_config_rwlock);
3357 spa_close(spa, FTAG);
3358
3359 return (error);
3360 }
3361
3362 int
3363 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3364 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3365 {
3366 int error = 0;
3367
3368 ASSERT3S(asize, >, 0);
3369
3370 /*
3371 * *ref_rsrv is the portion of asize that will come from any
3372 * unconsumed refreservation space.
3373 */
3374 *ref_rsrv = 0;
3375
3376 mutex_enter(&ds->ds_lock);
3377 /*
3378 * Make a space adjustment for reserved bytes.
3379 */
3380 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3381 ASSERT3U(*used, >=,
3382 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3383 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3384 *ref_rsrv =
3385 asize - MIN(asize, parent_delta(ds, asize + inflight));
3386 }
3387
3388 if (!check_quota || ds->ds_quota == 0) {
3389 mutex_exit(&ds->ds_lock);
3390 return (0);
3391 }
3392 /*
3393 * If they are requesting more space, and our current estimate
3394 * is over quota, they get to try again unless the actual
3395 * on-disk is over quota and there are no pending changes (which
3396 * may free up space for us).
3397 */
3398 if (ds->ds_phys->ds_referenced_bytes + inflight >= ds->ds_quota) {
3399 if (inflight > 0 ||
3400 ds->ds_phys->ds_referenced_bytes < ds->ds_quota)
3401 error = ERESTART;
3402 else
3403 error = EDQUOT;
3404 }
3405 mutex_exit(&ds->ds_lock);
3406
3407 return (error);
3408 }
3409
3410 /* ARGSUSED */
3411 static int
3412 dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3413 {
3414 dsl_dataset_t *ds = arg1;
3415 dsl_prop_setarg_t *psa = arg2;
3416 int err;
3417
3418 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3419 return (ENOTSUP);
3420
3421 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3422 return (err);
3423
3424 if (psa->psa_effective_value == 0)
3425 return (0);
3426
3427 if (psa->psa_effective_value < ds->ds_phys->ds_referenced_bytes ||
3428 psa->psa_effective_value < ds->ds_reserved)
3429 return (ENOSPC);
3430
3431 return (0);
3432 }
3433
3434 extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3435
3436 void
3437 dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3438 {
3439 dsl_dataset_t *ds = arg1;
3440 dsl_prop_setarg_t *psa = arg2;
3441 uint64_t effective_value = psa->psa_effective_value;
3442
3443 dsl_prop_set_sync(ds, psa, tx);
3444 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3445
3446 if (ds->ds_quota != effective_value) {
3447 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3448 ds->ds_quota = effective_value;
3449
3450 spa_history_log_internal_ds(ds, "set refquota", tx,
3451 "refquota=%lld", (longlong_t)ds->ds_quota);
3452 }
3453 }
3454
3455 int
3456 dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
3457 {
3458 dsl_dataset_t *ds;
3459 dsl_prop_setarg_t psa;
3460 int err;
3461
3462 dsl_prop_setarg_init_uint64(&psa, "refquota", source, "a);
3463
3464 err = dsl_dataset_hold(dsname, FTAG, &ds);
3465 if (err)
3466 return (err);
3467
3468 /*
3469 * If someone removes a file, then tries to set the quota, we
3470 * want to make sure the file freeing takes effect.
3471 */
3472 txg_wait_open(ds->ds_dir->dd_pool, 0);
3473
3474 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3475 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3476 ds, &psa, 0);
3477
3478 dsl_dataset_rele(ds, FTAG);
3479 return (err);
3480 }
3481
3482 static int
3483 dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3484 {
3485 dsl_dataset_t *ds = arg1;
3486 dsl_prop_setarg_t *psa = arg2;
3487 uint64_t effective_value;
3488 uint64_t unique;
3489 int err;
3490
3491 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3492 SPA_VERSION_REFRESERVATION)
3493 return (ENOTSUP);
3494
3495 if (dsl_dataset_is_snapshot(ds))
3496 return (EINVAL);
3497
3498 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3499 return (err);
3500
3501 effective_value = psa->psa_effective_value;
3502
3503 /*
3504 * If we are doing the preliminary check in open context, the
3505 * space estimates may be inaccurate.
3506 */
3507 if (!dmu_tx_is_syncing(tx))
3508 return (0);
3509
3510 mutex_enter(&ds->ds_lock);
3511 if (!DS_UNIQUE_IS_ACCURATE(ds))
3512 dsl_dataset_recalc_head_uniq(ds);
3513 unique = ds->ds_phys->ds_unique_bytes;
3514 mutex_exit(&ds->ds_lock);
3515
3516 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3517 uint64_t delta = MAX(unique, effective_value) -
3518 MAX(unique, ds->ds_reserved);
3519
3520 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3521 return (ENOSPC);
3522 if (ds->ds_quota > 0 &&
3523 effective_value > ds->ds_quota)
3524 return (ENOSPC);
3525 }
3526
3527 return (0);
3528 }
3529
3530 static void
3531 dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3532 {
3533 dsl_dataset_t *ds = arg1;
3534 dsl_prop_setarg_t *psa = arg2;
3535 uint64_t effective_value = psa->psa_effective_value;
3536 uint64_t unique;
3537 int64_t delta;
3538
3539 dsl_prop_set_sync(ds, psa, tx);
3540 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3541
3542 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3543
3544 mutex_enter(&ds->ds_dir->dd_lock);
3545 mutex_enter(&ds->ds_lock);
3546 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3547 unique = ds->ds_phys->ds_unique_bytes;
3548 delta = MAX(0, (int64_t)(effective_value - unique)) -
3549 MAX(0, (int64_t)(ds->ds_reserved - unique));
3550 ds->ds_reserved = effective_value;
3551 mutex_exit(&ds->ds_lock);
3552
3553 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3554 mutex_exit(&ds->ds_dir->dd_lock);
3555
3556 spa_history_log_internal_ds(ds, "set refreservation", tx,
3557 "refreservation=%lld", (longlong_t)effective_value);
3558 }
3559
3560 int
3561 dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3562 uint64_t reservation)
3563 {
3564 dsl_dataset_t *ds;
3565 dsl_prop_setarg_t psa;
3566 int err;
3567
3568 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3569 &reservation);
3570
3571 err = dsl_dataset_hold(dsname, FTAG, &ds);
3572 if (err)
3573 return (err);
3574
3575 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3576 dsl_dataset_set_reservation_check,
3577 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3578
3579 dsl_dataset_rele(ds, FTAG);
3580 return (err);
3581 }
3582
3583 typedef struct zfs_hold_cleanup_arg {
3584 dsl_pool_t *dp;
3585 uint64_t dsobj;
3586 char htag[MAXNAMELEN];
3587 } zfs_hold_cleanup_arg_t;
3588
3589 static void
3590 dsl_dataset_user_release_onexit(void *arg)
3591 {
3592 zfs_hold_cleanup_arg_t *ca = arg;
3593
3594 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3595 B_TRUE);
3596 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3597 }
3598
3599 void
3600 dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3601 minor_t minor)
3602 {
3603 zfs_hold_cleanup_arg_t *ca;
3604
3605 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3606 ca->dp = ds->ds_dir->dd_pool;
3607 ca->dsobj = ds->ds_object;
3608 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3609 VERIFY0(zfs_onexit_add_cb(minor,
3610 dsl_dataset_user_release_onexit, ca, NULL));
3611 }
3612
3613 /*
3614 * If you add new checks here, you may need to add
3615 * additional checks to the "temporary" case in
3616 * snapshot_check() in dmu_objset.c.
3617 */
3618 static int
3619 dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3620 {
3621 dsl_dataset_t *ds = arg1;
3622 struct dsl_ds_holdarg *ha = arg2;
3623 const char *htag = ha->htag;
3624 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3625 int error = 0;
3626
3627 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3628 return (ENOTSUP);
3629
3630 if (!dsl_dataset_is_snapshot(ds))
3631 return (EINVAL);
3632
3633 /* tags must be unique */
3634 mutex_enter(&ds->ds_lock);
3635 if (ds->ds_phys->ds_userrefs_obj) {
3636 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3637 8, 1, tx);
3638 if (error == 0)
3639 error = EEXIST;
3640 else if (error == ENOENT)
3641 error = 0;
3642 }
3643 mutex_exit(&ds->ds_lock);
3644
3645 if (error == 0 && ha->temphold &&
3646 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3647 error = E2BIG;
3648
3649 return (error);
3650 }
3651
3652 void
3653 dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
3654 {
3655 dsl_dataset_t *ds = arg1;
3656 struct dsl_ds_holdarg *ha = arg2;
3657 const char *htag = ha->htag;
3658 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3659 objset_t *mos = dp->dp_meta_objset;
3660 uint64_t now = gethrestime_sec();
3661 uint64_t zapobj;
3662
3663 mutex_enter(&ds->ds_lock);
3664 if (ds->ds_phys->ds_userrefs_obj == 0) {
3665 /*
3666 * This is the first user hold for this dataset. Create
3667 * the userrefs zap object.
3668 */
3669 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3670 zapobj = ds->ds_phys->ds_userrefs_obj =
3671 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3672 } else {
3673 zapobj = ds->ds_phys->ds_userrefs_obj;
3674 }
3675 ds->ds_userrefs++;
3676 mutex_exit(&ds->ds_lock);
3677
3678 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3679
3680 if (ha->temphold) {
3681 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3682 htag, &now, tx));
3683 }
3684
3685 spa_history_log_internal_ds(ds, "hold", tx,
3686 "tag = %s temp = %d holds now = %llu",
3687 htag, (int)ha->temphold, ds->ds_userrefs);
3688 }
3689
3690 static int
3691 dsl_dataset_user_hold_one(const char *dsname, void *arg)
3692 {
3693 struct dsl_ds_holdarg *ha = arg;
3694 dsl_dataset_t *ds;
3695 int error;
3696 char *name;
3697
3698 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
3699 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3700 error = dsl_dataset_hold(name, ha->dstg, &ds);
3701 strfree(name);
3702 if (error == 0) {
3703 ha->gotone = B_TRUE;
3704 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
3705 dsl_dataset_user_hold_sync, ds, ha, 0);
3706 } else if (error == ENOENT && ha->recursive) {
3707 error = 0;
3708 } else {
3709 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3710 }
3711 return (error);
3712 }
3713
3714 int
3715 dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3716 boolean_t temphold)
3717 {
3718 struct dsl_ds_holdarg *ha;
3719 int error;
3720
3721 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3722 ha->htag = htag;
3723 ha->temphold = temphold;
3724 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3725 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3726 ds, ha, 0);
3727 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3728
3729 return (error);
3730 }
3731
3732 int
3733 dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
3734 boolean_t recursive, boolean_t temphold, int cleanup_fd)
3735 {
3736 struct dsl_ds_holdarg *ha;
3737 dsl_sync_task_t *dst;
3738 spa_t *spa;
3739 int error;
3740 minor_t minor = 0;
3741
3742 if (cleanup_fd != -1) {
3743 /* Currently we only support cleanup-on-exit of tempholds. */
3744 if (!temphold)
3745 return (EINVAL);
3746 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3747 if (error)
3748 return (error);
3749 }
3750
3751 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3752
3753 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3754
3755 error = spa_open(dsname, &spa, FTAG);
3756 if (error) {
3757 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3758 if (cleanup_fd != -1)
3759 zfs_onexit_fd_rele(cleanup_fd);
3760 return (error);
3761 }
3762
3763 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3764 ha->htag = htag;
3765 ha->snapname = snapname;
3766 ha->recursive = recursive;
3767 ha->temphold = temphold;
3768
3769 if (recursive) {
3770 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3771 ha, DS_FIND_CHILDREN);
3772 } else {
3773 error = dsl_dataset_user_hold_one(dsname, ha);
3774 }
3775 if (error == 0)
3776 error = dsl_sync_task_group_wait(ha->dstg);
3777
3778 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3779 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3780 dsl_dataset_t *ds = dst->dst_arg1;
3781
3782 if (dst->dst_err) {
3783 dsl_dataset_name(ds, ha->failed);
3784 *strchr(ha->failed, '@') = '\0';
3785 } else if (error == 0 && minor != 0 && temphold) {
3786 /*
3787 * If this hold is to be released upon process exit,
3788 * register that action now.
3789 */
3790 dsl_register_onexit_hold_cleanup(ds, htag, minor);
3791 }
3792 dsl_dataset_rele(ds, ha->dstg);
3793 }
3794
3795 if (error == 0 && recursive && !ha->gotone)
3796 error = ENOENT;
3797
3798 if (error)
3799 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
3800
3801 dsl_sync_task_group_destroy(ha->dstg);
3802
3803 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3804 spa_close(spa, FTAG);
3805 if (cleanup_fd != -1)
3806 zfs_onexit_fd_rele(cleanup_fd);
3807 return (error);
3808 }
3809
3810 struct dsl_ds_releasearg {
3811 dsl_dataset_t *ds;
3812 const char *htag;
3813 boolean_t own; /* do we own or just hold ds? */
3814 };
3815
3816 static int
3817 dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3818 boolean_t *might_destroy)
3819 {
3820 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3821 uint64_t zapobj;
3822 uint64_t tmp;
3823 int error;
3824
3825 *might_destroy = B_FALSE;
3826
3827 mutex_enter(&ds->ds_lock);
3828 zapobj = ds->ds_phys->ds_userrefs_obj;
3829 if (zapobj == 0) {
3830 /* The tag can't possibly exist */
3831 mutex_exit(&ds->ds_lock);
3832 return (ESRCH);
3833 }
3834
3835 /* Make sure the tag exists */
3836 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3837 if (error) {
3838 mutex_exit(&ds->ds_lock);
3839 if (error == ENOENT)
3840 error = ESRCH;
3841 return (error);
3842 }
3843
3844 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3845 DS_IS_DEFER_DESTROY(ds))
3846 *might_destroy = B_TRUE;
3847
3848 mutex_exit(&ds->ds_lock);
3849 return (0);
3850 }
3851
3852 static int
3853 dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3854 {
3855 struct dsl_ds_releasearg *ra = arg1;
3856 dsl_dataset_t *ds = ra->ds;
3857 boolean_t might_destroy;
3858 int error;
3859
3860 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3861 return (ENOTSUP);
3862
3863 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3864 if (error)
3865 return (error);
3866
3867 if (might_destroy) {
3868 struct dsl_ds_destroyarg dsda = {0};
3869
3870 if (dmu_tx_is_syncing(tx)) {
3871 /*
3872 * If we're not prepared to remove the snapshot,
3873 * we can't allow the release to happen right now.
3874 */
3875 if (!ra->own)
3876 return (EBUSY);
3877 }
3878 dsda.ds = ds;
3879 dsda.releasing = B_TRUE;
3880 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3881 }
3882
3883 return (0);
3884 }
3885
3886 static void
3887 dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
3888 {
3889 struct dsl_ds_releasearg *ra = arg1;
3890 dsl_dataset_t *ds = ra->ds;
3891 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3892 objset_t *mos = dp->dp_meta_objset;
3893 uint64_t zapobj;
3894 uint64_t refs;
3895 int error;
3896
3897 mutex_enter(&ds->ds_lock);
3898 ds->ds_userrefs--;
3899 refs = ds->ds_userrefs;
3900 mutex_exit(&ds->ds_lock);
3901 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3902 VERIFY(error == 0 || error == ENOENT);
3903 zapobj = ds->ds_phys->ds_userrefs_obj;
3904 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3905 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3906 DS_IS_DEFER_DESTROY(ds)) {
3907 struct dsl_ds_destroyarg dsda = {0};
3908
3909 ASSERT(ra->own);
3910 dsda.ds = ds;
3911 dsda.releasing = B_TRUE;
3912 /* We already did the destroy_check */
3913 dsl_dataset_destroy_sync(&dsda, tag, tx);
3914 }
3915
3916 spa_history_log_internal_ds(ds, "release", tx,
3917 "tag = %s refs now = %lld", ra->htag, (longlong_t)refs);
3918 }
3919
3920 static int
3921 dsl_dataset_user_release_one(const char *dsname, void *arg)
3922 {
3923 struct dsl_ds_holdarg *ha = arg;
3924 struct dsl_ds_releasearg *ra;
3925 dsl_dataset_t *ds;
3926 int error;
3927 void *dtag = ha->dstg;
3928 char *name;
3929 boolean_t own = B_FALSE;
3930 boolean_t might_destroy;
3931
3932 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
3933 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3934 error = dsl_dataset_hold(name, dtag, &ds);
3935 strfree(name);
3936 if (error == ENOENT && ha->recursive)
3937 return (0);
3938 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3939 if (error)
3940 return (error);
3941
3942 ha->gotone = B_TRUE;
3943
3944 ASSERT(dsl_dataset_is_snapshot(ds));
3945
3946 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3947 if (error) {
3948 dsl_dataset_rele(ds, dtag);
3949 return (error);
3950 }
3951
3952 if (might_destroy) {
3953 #ifdef _KERNEL
3954 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
3955 error = zfs_unmount_snap(name, NULL);
3956 strfree(name);
3957 if (error) {
3958 dsl_dataset_rele(ds, dtag);
3959 return (error);
3960 }
3961 #endif
3962 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
3963 dsl_dataset_rele(ds, dtag);
3964 return (EBUSY);
3965 } else {
3966 own = B_TRUE;
3967 dsl_dataset_make_exclusive(ds, dtag);
3968 }
3969 }
3970
3971 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3972 ra->ds = ds;
3973 ra->htag = ha->htag;
3974 ra->own = own;
3975 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3976 dsl_dataset_user_release_sync, ra, dtag, 0);
3977
3978 return (0);
3979 }
3980
3981 int
3982 dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3983 boolean_t recursive)
3984 {
3985 struct dsl_ds_holdarg *ha;
3986 dsl_sync_task_t *dst;
3987 spa_t *spa;
3988 int error;
3989
3990 top:
3991 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3992
3993 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3994
3995 error = spa_open(dsname, &spa, FTAG);
3996 if (error) {
3997 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3998 return (error);
3999 }
4000
4001 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
4002 ha->htag = htag;
4003 ha->snapname = snapname;
4004 ha->recursive = recursive;
4005 if (recursive) {
4006 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
4007 ha, DS_FIND_CHILDREN);
4008 } else {
4009 error = dsl_dataset_user_release_one(dsname, ha);
4010 }
4011 if (error == 0)
4012 error = dsl_sync_task_group_wait(ha->dstg);
4013
4014 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
4015 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
4016 struct dsl_ds_releasearg *ra = dst->dst_arg1;
4017 dsl_dataset_t *ds = ra->ds;
4018
4019 if (dst->dst_err)
4020 dsl_dataset_name(ds, ha->failed);
4021
4022 if (ra->own)
4023 dsl_dataset_disown(ds, ha->dstg);
4024 else
4025 dsl_dataset_rele(ds, ha->dstg);
4026
4027 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
4028 }
4029
4030 if (error == 0 && recursive && !ha->gotone)
4031 error = ENOENT;
4032
4033 if (error && error != EBUSY)
4034 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
4035
4036 dsl_sync_task_group_destroy(ha->dstg);
4037 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
4038 spa_close(spa, FTAG);
4039
4040 /*
4041 * We can get EBUSY if we were racing with deferred destroy and
4042 * dsl_dataset_user_release_check() hadn't done the necessary
4043 * open context setup. We can also get EBUSY if we're racing
4044 * with destroy and that thread is the ds_owner. Either way
4045 * the busy condition should be transient, and we should retry
4046 * the release operation.
4047 */
4048 if (error == EBUSY)
4049 goto top;
4050
4051 return (error);
4052 }
4053
4054 /*
4055 * Called at spa_load time (with retry == B_FALSE) to release a stale
4056 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
4057 */
4058 int
4059 dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
4060 boolean_t retry)
4061 {
4062 dsl_dataset_t *ds;
4063 char *snap;
4064 char *name;
4065 int namelen;
4066 int error;
4067
4068 do {
4069 rw_enter(&dp->dp_config_rwlock, RW_READER);
4070 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
4071 rw_exit(&dp->dp_config_rwlock);
4072 if (error)
4073 return (error);
4074 namelen = dsl_dataset_namelen(ds)+1;
4075 name = kmem_alloc(namelen, KM_SLEEP);
4076 dsl_dataset_name(ds, name);
4077 dsl_dataset_rele(ds, FTAG);
4078
4079 snap = strchr(name, '@');
4080 *snap = '\0';
4081 ++snap;
4082 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
4083 kmem_free(name, namelen);
4084
4085 /*
4086 * The object can't have been destroyed because we have a hold,
4087 * but it might have been renamed, resulting in ENOENT. Retry
4088 * if we've been requested to do so.
4089 *
4090 * It would be nice if we could use the dsobj all the way
4091 * through and avoid ENOENT entirely. But we might need to
4092 * unmount the snapshot, and there's currently no way to lookup
4093 * a vfsp using a ZFS object id.
4094 */
4095 } while ((error == ENOENT) && retry);
4096
4097 return (error);
4098 }
4099
4100 int
4101 dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
4102 {
4103 dsl_dataset_t *ds;
4104 int err;
4105
4106 err = dsl_dataset_hold(dsname, FTAG, &ds);
4107 if (err)
4108 return (err);
4109
4110 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4111 if (ds->ds_phys->ds_userrefs_obj != 0) {
4112 zap_attribute_t *za;
4113 zap_cursor_t zc;
4114
4115 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4116 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4117 ds->ds_phys->ds_userrefs_obj);
4118 zap_cursor_retrieve(&zc, za) == 0;
4119 zap_cursor_advance(&zc)) {
4120 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4121 za->za_first_integer));
4122 }
4123 zap_cursor_fini(&zc);
4124 kmem_free(za, sizeof (zap_attribute_t));
4125 }
4126 dsl_dataset_rele(ds, FTAG);
4127 return (0);
4128 }
4129
4130 /*
4131 * Note, this function is used as the callback for dmu_objset_find(). We
4132 * always return 0 so that we will continue to find and process
4133 * inconsistent datasets, even if we encounter an error trying to
4134 * process one of them.
4135 */
4136 /* ARGSUSED */
4137 int
4138 dsl_destroy_inconsistent(const char *dsname, void *arg)
4139 {
4140 dsl_dataset_t *ds;
4141
4142 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4143 if (DS_IS_INCONSISTENT(ds))
4144 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4145 else
4146 dsl_dataset_disown(ds, FTAG);
4147 }
4148 return (0);
4149 }
4150
4151 /*
4152 * Return (in *usedp) the amount of space written in new that is not
4153 * present in oldsnap. New may be a snapshot or the head. Old must be
4154 * a snapshot before new, in new's filesystem (or its origin). If not then
4155 * fail and return EINVAL.
4156 *
4157 * The written space is calculated by considering two components: First, we
4158 * ignore any freed space, and calculate the written as new's used space
4159 * minus old's used space. Next, we add in the amount of space that was freed
4160 * between the two snapshots, thus reducing new's used space relative to old's.
4161 * Specifically, this is the space that was born before old->ds_creation_txg,
4162 * and freed before new (ie. on new's deadlist or a previous deadlist).
4163 *
4164 * space freed [---------------------]
4165 * snapshots ---O-------O--------O-------O------
4166 * oldsnap new
4167 */
4168 int
4169 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4170 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4171 {
4172 int err = 0;
4173 uint64_t snapobj;
4174 dsl_pool_t *dp = new->ds_dir->dd_pool;
4175
4176 *usedp = 0;
4177 *usedp += new->ds_phys->ds_referenced_bytes;
4178 *usedp -= oldsnap->ds_phys->ds_referenced_bytes;
4179
4180 *compp = 0;
4181 *compp += new->ds_phys->ds_compressed_bytes;
4182 *compp -= oldsnap->ds_phys->ds_compressed_bytes;
4183
4184 *uncompp = 0;
4185 *uncompp += new->ds_phys->ds_uncompressed_bytes;
4186 *uncompp -= oldsnap->ds_phys->ds_uncompressed_bytes;
4187
4188 rw_enter(&dp->dp_config_rwlock, RW_READER);
4189 snapobj = new->ds_object;
4190 while (snapobj != oldsnap->ds_object) {
4191 dsl_dataset_t *snap;
4192 uint64_t used, comp, uncomp;
4193
4194 if (snapobj == new->ds_object) {
4195 snap = new;
4196 } else {
4197 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4198 if (err != 0)
4199 break;
4200 }
4201
4202 if (snap->ds_phys->ds_prev_snap_txg ==
4203 oldsnap->ds_phys->ds_creation_txg) {
4204 /*
4205 * The blocks in the deadlist can not be born after
4206 * ds_prev_snap_txg, so get the whole deadlist space,
4207 * which is more efficient (especially for old-format
4208 * deadlists). Unfortunately the deadlist code
4209 * doesn't have enough information to make this
4210 * optimization itself.
4211 */
4212 dsl_deadlist_space(&snap->ds_deadlist,
4213 &used, &comp, &uncomp);
4214 } else {
4215 dsl_deadlist_space_range(&snap->ds_deadlist,
4216 0, oldsnap->ds_phys->ds_creation_txg,
4217 &used, &comp, &uncomp);
4218 }
4219 *usedp += used;
4220 *compp += comp;
4221 *uncompp += uncomp;
4222
4223 /*
4224 * If we get to the beginning of the chain of snapshots
4225 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4226 * was not a snapshot of/before new.
4227 */
4228 snapobj = snap->ds_phys->ds_prev_snap_obj;
4229 if (snap != new)
4230 dsl_dataset_rele(snap, FTAG);
4231 if (snapobj == 0) {
4232 err = EINVAL;
4233 break;
4234 }
4235
4236 }
4237 rw_exit(&dp->dp_config_rwlock);
4238 return (err);
4239 }
4240
4241 /*
4242 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4243 * lastsnap, and all snapshots in between are deleted.
4244 *
4245 * blocks that would be freed [---------------------------]
4246 * snapshots ---O-------O--------O-------O--------O
4247 * firstsnap lastsnap
4248 *
4249 * This is the set of blocks that were born after the snap before firstsnap,
4250 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4251 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4252 * We calculate this by iterating over the relevant deadlists (from the snap
4253 * after lastsnap, backward to the snap after firstsnap), summing up the
4254 * space on the deadlist that was born after the snap before firstsnap.
4255 */
4256 int
4257 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4258 dsl_dataset_t *lastsnap,
4259 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4260 {
4261 int err = 0;
4262 uint64_t snapobj;
4263 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4264
4265 ASSERT(dsl_dataset_is_snapshot(firstsnap));
4266 ASSERT(dsl_dataset_is_snapshot(lastsnap));
4267
4268 /*
4269 * Check that the snapshots are in the same dsl_dir, and firstsnap
4270 * is before lastsnap.
4271 */
4272 if (firstsnap->ds_dir != lastsnap->ds_dir ||
4273 firstsnap->ds_phys->ds_creation_txg >
4274 lastsnap->ds_phys->ds_creation_txg)
4275 return (EINVAL);
4276
4277 *usedp = *compp = *uncompp = 0;
4278
4279 rw_enter(&dp->dp_config_rwlock, RW_READER);
4280 snapobj = lastsnap->ds_phys->ds_next_snap_obj;
4281 while (snapobj != firstsnap->ds_object) {
4282 dsl_dataset_t *ds;
4283 uint64_t used, comp, uncomp;
4284
4285 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4286 if (err != 0)
4287 break;
4288
4289 dsl_deadlist_space_range(&ds->ds_deadlist,
4290 firstsnap->ds_phys->ds_prev_snap_txg, UINT64_MAX,
4291 &used, &comp, &uncomp);
4292 *usedp += used;
4293 *compp += comp;
4294 *uncompp += uncomp;
4295
4296 snapobj = ds->ds_phys->ds_prev_snap_obj;
4297 ASSERT3U(snapobj, !=, 0);
4298 dsl_dataset_rele(ds, FTAG);
4299 }
4300 rw_exit(&dp->dp_config_rwlock);
4301 return (err);
4302 }