]> git.proxmox.com Git - mirror_zfs.git/blame - module/zfs/dsl_dataset.c
projects / mirror_zfs.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Add linux kernel module support
[mirror_zfs.git] / module / zfs / dsl_dataset.c
CommitLineData
34dc7c2f
BB		 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/*
428870ff 22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
34dc7c2f
BB		 23 */
24
34dc7c2f
BB		 25#include <sys/dmu_objset.h>
26#include <sys/dsl_dataset.h>
27#include <sys/dsl_dir.h>
28#include <sys/dsl_prop.h>
29#include <sys/dsl_synctask.h>
30#include <sys/dmu_traverse.h>
31#include <sys/dmu_tx.h>
32#include <sys/arc.h>
33#include <sys/zio.h>
34#include <sys/zap.h>
35#include <sys/unique.h>
36#include <sys/zfs_context.h>
37#include <sys/zfs_ioctl.h>
38#include <sys/spa.h>
b128c09f 39#include <sys/zfs_znode.h>
572e2857 40#include <sys/zfs_onexit.h>
45d1cae3 41#include <sys/zvol.h>
428870ff
BB		 42#include <sys/dsl_scan.h>
43#include <sys/dsl_deadlist.h>
44
b128c09f
BB		 45static char *dsl_reaper = "the grim reaper";
46
34dc7c2f
BB		 47static dsl_checkfunc_t dsl_dataset_destroy_begin_check;
48static dsl_syncfunc_t dsl_dataset_destroy_begin_sync;
34dc7c2f
BB		 49static dsl_syncfunc_t dsl_dataset_set_reservation_sync;
50
428870ff
BB		 51#define SWITCH64(x, y) \
52 { \
53 uint64_t __tmp = (x); \
54 (x) = (y); \
55 (y) = __tmp; \
56 }
57
34dc7c2f
BB		 58#define DS_REF_MAX (1ULL << 62)
59
60#define DSL_DEADLIST_BLOCKSIZE SPA_MAXBLOCKSIZE
61
b128c09f
BB		 62#define DSL_DATASET_IS_DESTROYED(ds) ((ds)->ds_owner == dsl_reaper)
63
34dc7c2f
BB		 64
65/*
66 * Figure out how much of this delta should be propogated to the dsl_dir
67 * layer. If there's a refreservation, that space has already been
68 * partially accounted for in our ancestors.
69 */
70static int64_t
71parent_delta(dsl_dataset_t *ds, int64_t delta)
72{
73 uint64_t old_bytes, new_bytes;
74
75 if (ds->ds_reserved == 0)
76 return (delta);
77
78 old_bytes = MAX(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
79 new_bytes = MAX(ds->ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
80
81 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
82 return (new_bytes - old_bytes);
83}
84
85void
428870ff 86dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
34dc7c2f 87{
d6320ddb 88 int used, compressed, uncompressed;
34dc7c2f
BB		 89 int64_t delta;
90
d6320ddb
BB		 91 used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
92 compressed = BP_GET_PSIZE(bp);
93 uncompressed = BP_GET_UCSIZE(bp);
94
428870ff 95 dprintf_bp(bp, "ds=%p", ds);
34dc7c2f
BB		 96
97 ASSERT(dmu_tx_is_syncing(tx));
98 /* It could have been compressed away to nothing */
99 if (BP_IS_HOLE(bp))
100 return;
101 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
102 ASSERT3U(BP_GET_TYPE(bp), <, DMU_OT_NUMTYPES);
103 if (ds == NULL) {
104 /*
105 * Account for the meta-objset space in its placeholder
106 * dsl_dir.
107 */
108 ASSERT3U(compressed, ==, uncompressed); /* it's all metadata */
b128c09f 109 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
34dc7c2f
BB		 110 used, compressed, uncompressed, tx);
111 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
112 return;
113 }
114 dmu_buf_will_dirty(ds->ds_dbuf, tx);
428870ff 115
b128c09f 116 mutex_enter(&ds->ds_dir->dd_lock);
34dc7c2f
BB		 117 mutex_enter(&ds->ds_lock);
118 delta = parent_delta(ds, used);
119 ds->ds_phys->ds_used_bytes += used;
120 ds->ds_phys->ds_compressed_bytes += compressed;
121 ds->ds_phys->ds_uncompressed_bytes += uncompressed;
122 ds->ds_phys->ds_unique_bytes += used;
123 mutex_exit(&ds->ds_lock);
b128c09f
BB		 124 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
125 compressed, uncompressed, tx);
126 dsl_dir_transfer_space(ds->ds_dir, used - delta,
127 DD_USED_REFRSRV, DD_USED_HEAD, tx);
128 mutex_exit(&ds->ds_dir->dd_lock);
34dc7c2f
BB		 129}
130
b128c09f 131int
428870ff
BB		 132dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
133 boolean_t async)
34dc7c2f 134{
d6320ddb
BB		 135 int used, compressed, uncompressed;
136
34dc7c2f 137 if (BP_IS_HOLE(bp))
b128c09f 138 return (0);
34dc7c2f 139
428870ff
BB		 140 ASSERT(dmu_tx_is_syncing(tx));
141 ASSERT(bp->blk_birth <= tx->tx_txg);
142
d6320ddb
BB		 143 used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
144 compressed = BP_GET_PSIZE(bp);
145 uncompressed = BP_GET_UCSIZE(bp);
428870ff 146
34dc7c2f
BB		 147 ASSERT(used > 0);
148 if (ds == NULL) {
34dc7c2f
BB		 149 /*
150 * Account for the meta-objset space in its placeholder
151 * dataset.
152 */
428870ff 153 dsl_free(tx->tx_pool, tx->tx_txg, bp);
34dc7c2f 154
b128c09f 155 dsl_dir_diduse_space(tx->tx_pool->dp_mos_dir, DD_USED_HEAD,
34dc7c2f
BB		 156 -used, -compressed, -uncompressed, tx);
157 dsl_dir_dirty(tx->tx_pool->dp_mos_dir, tx);
b128c09f 158 return (used);
34dc7c2f
BB		 159 }
160 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
161
b128c09f 162 ASSERT(!dsl_dataset_is_snapshot(ds));
34dc7c2f
BB		 163 dmu_buf_will_dirty(ds->ds_dbuf, tx);
164
165 if (bp->blk_birth > ds->ds_phys->ds_prev_snap_txg) {
34dc7c2f
BB		 166 int64_t delta;
167
428870ff
BB		 168 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
169 dsl_free(tx->tx_pool, tx->tx_txg, bp);
34dc7c2f 170
b128c09f 171 mutex_enter(&ds->ds_dir->dd_lock);
34dc7c2f
BB		 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);
b128c09f 178 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
34dc7c2f 179 delta, -compressed, -uncompressed, tx);
b128c09f
BB		 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);
34dc7c2f
BB		 183 } else {
184 dprintf_bp(bp, "putting on dead list: %s", "");
428870ff
BB		 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 }
34dc7c2f
BB		 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 }
428870ff 209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
b128c09f
BB		 210 dsl_dir_transfer_space(ds->ds_dir, used,
211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 }
34dc7c2f
BB		 213 }
214 mutex_enter(&ds->ds_lock);
215 ASSERT3U(ds->ds_phys->ds_used_bytes, >=, used);
216 ds->ds_phys->ds_used_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);
b128c09f
BB		 222
223 return (used);
34dc7c2f
BB		 224}
225
226uint64_t
227dsl_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
9babb374 249boolean_t
428870ff
BB		 250dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 uint64_t blk_birth)
34dc7c2f 252{
428870ff
BB		 253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds))
254 return (B_FALSE);
255
572e2857 256 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
428870ff
BB		 257
258 return (B_TRUE);
34dc7c2f
BB		 259}
260
261/* ARGSUSED */
262static void
263dsl_dataset_evict(dmu_buf_t *db, void *dsv)
264{
265 dsl_dataset_t *ds = dsv;
266
b128c09f 267 ASSERT(ds->ds_owner == NULL || DSL_DATASET_IS_DESTROYED(ds));
34dc7c2f 268
34dc7c2f
BB		 269 unique_remove(ds->ds_fsid_guid);
270
428870ff
BB		 271 if (ds->ds_objset != NULL)
272 dmu_objset_evict(ds->ds_objset);
34dc7c2f
BB		 273
274 if (ds->ds_prev) {
b128c09f 275 dsl_dataset_drop_ref(ds->ds_prev, ds);
34dc7c2f
BB		 276 ds->ds_prev = NULL;
277 }
278
428870ff
BB		 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 }
b128c09f
BB		 286 if (ds->ds_dir)
287 dsl_dir_close(ds->ds_dir, ds);
34dc7c2f
BB		 288
289 ASSERT(!list_link_active(&ds->ds_synced_link));
290
291 mutex_destroy(&ds->ds_lock);
45d1cae3 292 mutex_destroy(&ds->ds_recvlock);
34dc7c2f 293 mutex_destroy(&ds->ds_opening_lock);
b128c09f
BB		 294 rw_destroy(&ds->ds_rwlock);
295 cv_destroy(&ds->ds_exclusive_cv);
34dc7c2f
BB		 296
297 kmem_free(ds, sizeof (dsl_dataset_t));
298}
299
300static int
301dsl_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
325static int
b128c09f 326dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
34dc7c2f 327{
b128c09f
BB		 328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
34dc7c2f
BB		 330 matchtype_t mt;
331 int err;
332
b128c09f 333 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
34dc7c2f
BB		 334 mt = MT_FIRST;
335 else
336 mt = MT_EXACT;
337
b128c09f 338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
34dc7c2f
BB		 339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
b128c09f 341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
34dc7c2f
BB		 342 return (err);
343}
344
345static int
b128c09f 346dsl_dataset_snap_remove(dsl_dataset_t *ds, char *name, dmu_tx_t *tx)
34dc7c2f 347{
b128c09f
BB		 348 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
349 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
34dc7c2f
BB		 350 matchtype_t mt;
351 int err;
352
428870ff
BB		 353 dsl_dir_snap_cmtime_update(ds->ds_dir);
354
b128c09f 355 if (ds->ds_phys->ds_flags & DS_FLAG_CI_DATASET)
34dc7c2f
BB		 356 mt = MT_FIRST;
357 else
358 mt = MT_EXACT;
359
b128c09f 360 err = zap_remove_norm(mos, snapobj, name, mt, tx);
34dc7c2f 361 if (err == ENOTSUP && mt == MT_FIRST)
b128c09f 362 err = zap_remove(mos, snapobj, name, tx);
34dc7c2f
BB		 363 return (err);
364}
365
b128c09f
BB		 366static int
367dsl_dataset_get_ref(dsl_pool_t *dp, uint64_t dsobj, void *tag,
368 dsl_dataset_t **dsp)
34dc7c2f 369{
34dc7c2f
BB		 370 objset_t *mos = dp->dp_meta_objset;
371 dmu_buf_t *dbuf;
372 dsl_dataset_t *ds;
373 int err;
572e2857 374 dmu_object_info_t doi;
34dc7c2f
BB		 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);
572e2857
BB		 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
34dc7c2f
BB		 388 ds = dmu_buf_get_user(dbuf);
389 if (ds == NULL) {
d4ed6673 390 dsl_dataset_t *winner = NULL;
34dc7c2f
BB		 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;
98f72a53 396 list_link_init(&ds->ds_synced_link);
34dc7c2f
BB		 397
398 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
45d1cae3 399 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
34dc7c2f 400 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
ef5319df 401 rw_init(&ds->ds_rwlock, NULL, RW_DEFAULT, NULL);
b128c09f 402 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
34dc7c2f 403
428870ff
BB		 404 bplist_create(&ds->ds_pending_deadlist);
405 dsl_deadlist_open(&ds->ds_deadlist,
34dc7c2f 406 mos, ds->ds_phys->ds_deadlist_obj);
428870ff 407
34dc7c2f
BB		 408 if (err == 0) {
409 err = dsl_dir_open_obj(dp,
410 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
411 }
412 if (err) {
34dc7c2f 413 mutex_destroy(&ds->ds_lock);
45d1cae3 414 mutex_destroy(&ds->ds_recvlock);
34dc7c2f 415 mutex_destroy(&ds->ds_opening_lock);
b128c09f
BB		 416 rw_destroy(&ds->ds_rwlock);
417 cv_destroy(&ds->ds_exclusive_cv);
428870ff
BB		 418 bplist_destroy(&ds->ds_pending_deadlist);
419 dsl_deadlist_close(&ds->ds_deadlist);
34dc7c2f
BB		 420 kmem_free(ds, sizeof (dsl_dataset_t));
421 dmu_buf_rele(dbuf, tag);
422 return (err);
423 }
424
b128c09f 425 if (!dsl_dataset_is_snapshot(ds)) {
34dc7c2f
BB		 426 ds->ds_snapname[0] = '\0';
427 if (ds->ds_phys->ds_prev_snap_obj) {
b128c09f
BB		 428 err = dsl_dataset_get_ref(dp,
429 ds->ds_phys->ds_prev_snap_obj,
430 ds, &ds->ds_prev);
34dc7c2f 431 }
45d1cae3
BB		 432 } else {
433 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
434 err = dsl_dataset_get_snapname(ds);
435 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
436 err = zap_count(
437 ds->ds_dir->dd_pool->dp_meta_objset,
438 ds->ds_phys->ds_userrefs_obj,
439 &ds->ds_userrefs);
440 }
34dc7c2f
BB		 441 }
442
b128c09f 443 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
34dc7c2f
BB		 444 /*
445 * In sync context, we're called with either no lock
446 * or with the write lock. If we're not syncing,
447 * we're always called with the read lock held.
448 */
449 boolean_t need_lock =
450 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
451 dsl_pool_sync_context(dp);
452
453 if (need_lock)
454 rw_enter(&dp->dp_config_rwlock, RW_READER);
455
b128c09f 456 err = dsl_prop_get_ds(ds,
34dc7c2f
BB		 457 "refreservation", sizeof (uint64_t), 1,
458 &ds->ds_reserved, NULL);
459 if (err == 0) {
b128c09f 460 err = dsl_prop_get_ds(ds,
34dc7c2f
BB		 461 "refquota", sizeof (uint64_t), 1,
462 &ds->ds_quota, NULL);
463 }
464
465 if (need_lock)
466 rw_exit(&dp->dp_config_rwlock);
467 } else {
468 ds->ds_reserved = ds->ds_quota = 0;
469 }
470
471 if (err == 0) {
472 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
473 dsl_dataset_evict);
474 }
475 if (err || winner) {
428870ff
BB		 476 bplist_destroy(&ds->ds_pending_deadlist);
477 dsl_deadlist_close(&ds->ds_deadlist);
b128c09f
BB		 478 if (ds->ds_prev)
479 dsl_dataset_drop_ref(ds->ds_prev, ds);
34dc7c2f
BB		 480 dsl_dir_close(ds->ds_dir, ds);
481 mutex_destroy(&ds->ds_lock);
45d1cae3 482 mutex_destroy(&ds->ds_recvlock);
34dc7c2f 483 mutex_destroy(&ds->ds_opening_lock);
b128c09f
BB		 484 rw_destroy(&ds->ds_rwlock);
485 cv_destroy(&ds->ds_exclusive_cv);
34dc7c2f
BB		 486 kmem_free(ds, sizeof (dsl_dataset_t));
487 if (err) {
488 dmu_buf_rele(dbuf, tag);
489 return (err);
490 }
491 ds = winner;
492 } else {
493 ds->ds_fsid_guid =
494 unique_insert(ds->ds_phys->ds_fsid_guid);
495 }
496 }
497 ASSERT3P(ds->ds_dbuf, ==, dbuf);
498 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
b128c09f
BB		 499 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
500 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
501 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
34dc7c2f 502 mutex_enter(&ds->ds_lock);
b128c09f 503 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
34dc7c2f 504 mutex_exit(&ds->ds_lock);
b128c09f
BB		 505 dmu_buf_rele(ds->ds_dbuf, tag);
506 return (ENOENT);
34dc7c2f 507 }
34dc7c2f 508 mutex_exit(&ds->ds_lock);
34dc7c2f
BB		 509 *dsp = ds;
510 return (0);
511}
512
b128c09f
BB		 513static int
514dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
515{
516 dsl_pool_t *dp = ds->ds_dir->dd_pool;
517
518 /*
519 * In syncing context we don't want the rwlock lock: there
520 * may be an existing writer waiting for sync phase to
521 * finish. We don't need to worry about such writers, since
522 * sync phase is single-threaded, so the writer can't be
523 * doing anything while we are active.
524 */
525 if (dsl_pool_sync_context(dp)) {
526 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
527 return (0);
528 }
529
530 /*
531 * Normal users will hold the ds_rwlock as a READER until they
532 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
533 * drop their READER lock after they set the ds_owner field.
534 *
535 * If the dataset is being destroyed, the destroy thread will
536 * obtain a WRITER lock for exclusive access after it's done its
537 * open-context work and then change the ds_owner to
538 * dsl_reaper once destruction is assured. So threads
539 * may block here temporarily, until the "destructability" of
540 * the dataset is determined.
541 */
542 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
543 mutex_enter(&ds->ds_lock);
544 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
545 rw_exit(&dp->dp_config_rwlock);
546 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
547 if (DSL_DATASET_IS_DESTROYED(ds)) {
548 mutex_exit(&ds->ds_lock);
549 dsl_dataset_drop_ref(ds, tag);
550 rw_enter(&dp->dp_config_rwlock, RW_READER);
551 return (ENOENT);
552 }
428870ff
BB		 553 /*
554 * The dp_config_rwlock lives above the ds_lock. And
555 * we need to check DSL_DATASET_IS_DESTROYED() while
556 * holding the ds_lock, so we have to drop and reacquire
557 * the ds_lock here.
558 */
559 mutex_exit(&ds->ds_lock);
b128c09f 560 rw_enter(&dp->dp_config_rwlock, RW_READER);
428870ff 561 mutex_enter(&ds->ds_lock);
b128c09f
BB		 562 }
563 mutex_exit(&ds->ds_lock);
564 return (0);
565}
566
34dc7c2f 567int
b128c09f
BB		 568dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
569 dsl_dataset_t **dsp)
570{
571 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
572
573 if (err)
574 return (err);
575 return (dsl_dataset_hold_ref(*dsp, tag));
576}
577
578int
428870ff
BB		 579dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
580 void *tag, dsl_dataset_t **dsp)
b128c09f 581{
428870ff 582 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
b128c09f
BB		 583 if (err)
584 return (err);
428870ff
BB		 585 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
586 dsl_dataset_rele(*dsp, tag);
9babb374 587 *dsp = NULL;
b128c09f
BB		 588 return (EBUSY);
589 }
590 return (0);
591}
592
593int
594dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
34dc7c2f
BB		 595{
596 dsl_dir_t *dd;
597 dsl_pool_t *dp;
b128c09f 598 const char *snapname;
34dc7c2f 599 uint64_t obj;
34dc7c2f
BB		 600 int err = 0;
601
b128c09f 602 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
34dc7c2f
BB		 603 if (err)
604 return (err);
605
606 dp = dd->dd_pool;
607 obj = dd->dd_phys->dd_head_dataset_obj;
608 rw_enter(&dp->dp_config_rwlock, RW_READER);
b128c09f
BB		 609 if (obj)
610 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
611 else
34dc7c2f 612 err = ENOENT;
b128c09f 613 if (err)
34dc7c2f 614 goto out;
34dc7c2f 615
b128c09f 616 err = dsl_dataset_hold_ref(*dsp, tag);
34dc7c2f 617
b128c09f
BB		 618 /* we may be looking for a snapshot */
619 if (err == 0 && snapname != NULL) {
620 dsl_dataset_t *ds = NULL;
34dc7c2f 621
b128c09f
BB		 622 if (*snapname++ != '@') {
623 dsl_dataset_rele(*dsp, tag);
34dc7c2f
BB		 624 err = ENOENT;
625 goto out;
626 }
34dc7c2f 627
b128c09f
BB		 628 dprintf("looking for snapshot '%s'\n", snapname);
629 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
630 if (err == 0)
631 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
632 dsl_dataset_rele(*dsp, tag);
633
634 ASSERT3U((err == 0), ==, (ds != NULL));
635
636 if (ds) {
637 mutex_enter(&ds->ds_lock);
638 if (ds->ds_snapname[0] == 0)
639 (void) strlcpy(ds->ds_snapname, snapname,
640 sizeof (ds->ds_snapname));
641 mutex_exit(&ds->ds_lock);
642 err = dsl_dataset_hold_ref(ds, tag);
643 *dsp = err ? NULL : ds;
34dc7c2f 644 }
34dc7c2f 645 }
34dc7c2f
BB		 646out:
647 rw_exit(&dp->dp_config_rwlock);
648 dsl_dir_close(dd, FTAG);
34dc7c2f
BB		 649 return (err);
650}
651
652int
428870ff
BB		 653dsl_dataset_own(const char *name, boolean_t inconsistentok,
654 void *tag, dsl_dataset_t **dsp)
34dc7c2f 655{
428870ff 656 int err = dsl_dataset_hold(name, tag, dsp);
b128c09f
BB		 657 if (err)
658 return (err);
428870ff
BB		 659 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
660 dsl_dataset_rele(*dsp, tag);
b128c09f
BB		 661 return (EBUSY);
662 }
663 return (0);
34dc7c2f
BB		 664}
665
666void
667dsl_dataset_name(dsl_dataset_t *ds, char *name)
668{
669 if (ds == NULL) {
670 (void) strcpy(name, "mos");
671 } else {
672 dsl_dir_name(ds->ds_dir, name);
673 VERIFY(0 == dsl_dataset_get_snapname(ds));
674 if (ds->ds_snapname[0]) {
675 (void) strcat(name, "@");
b128c09f
BB		 676 /*
677 * We use a "recursive" mutex so that we
678 * can call dprintf_ds() with ds_lock held.
679 */
34dc7c2f 680 if (!MUTEX_HELD(&ds->ds_lock)) {
34dc7c2f
BB		 681 mutex_enter(&ds->ds_lock);
682 (void) strcat(name, ds->ds_snapname);
683 mutex_exit(&ds->ds_lock);
684 } else {
685 (void) strcat(name, ds->ds_snapname);
686 }
687 }
688 }
689}
690
691static int
692dsl_dataset_namelen(dsl_dataset_t *ds)
693{
694 int result;
695
696 if (ds == NULL) {
697 result = 3; /* "mos" */
698 } else {
699 result = dsl_dir_namelen(ds->ds_dir);
700 VERIFY(0 == dsl_dataset_get_snapname(ds));
701 if (ds->ds_snapname[0]) {
702 ++result; /* adding one for the @-sign */
703 if (!MUTEX_HELD(&ds->ds_lock)) {
34dc7c2f
BB		 704 mutex_enter(&ds->ds_lock);
705 result += strlen(ds->ds_snapname);
706 mutex_exit(&ds->ds_lock);
707 } else {
708 result += strlen(ds->ds_snapname);
709 }
710 }
711 }
712
713 return (result);
714}
715
716void
b128c09f 717dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
34dc7c2f 718{
34dc7c2f
BB		 719 dmu_buf_rele(ds->ds_dbuf, tag);
720}
721
722void
b128c09f
BB		 723dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
724{
725 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
726 rw_exit(&ds->ds_rwlock);
727 }
728 dsl_dataset_drop_ref(ds, tag);
729}
730
731void
428870ff 732dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
34dc7c2f 733{
428870ff 734 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
b128c09f
BB		 735 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
736
34dc7c2f 737 mutex_enter(&ds->ds_lock);
b128c09f
BB		 738 ds->ds_owner = NULL;
739 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
740 rw_exit(&ds->ds_rwlock);
741 cv_broadcast(&ds->ds_exclusive_cv);
742 }
34dc7c2f 743 mutex_exit(&ds->ds_lock);
b128c09f 744 if (ds->ds_dbuf)
428870ff 745 dsl_dataset_drop_ref(ds, tag);
b128c09f 746 else
428870ff 747 dsl_dataset_evict(NULL, ds);
34dc7c2f
BB		 748}
749
750boolean_t
428870ff 751dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
34dc7c2f 752{
b128c09f
BB		 753 boolean_t gotit = FALSE;
754
34dc7c2f 755 mutex_enter(&ds->ds_lock);
b128c09f
BB		 756 if (ds->ds_owner == NULL &&
757 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
428870ff 758 ds->ds_owner = tag;
b128c09f
BB		 759 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
760 rw_exit(&ds->ds_rwlock);
761 gotit = TRUE;
34dc7c2f
BB		 762 }
763 mutex_exit(&ds->ds_lock);
b128c09f 764 return (gotit);
34dc7c2f
BB		 765}
766
767void
b128c09f 768dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
34dc7c2f 769{
b128c09f
BB		 770 ASSERT3P(owner, ==, ds->ds_owner);
771 if (!RW_WRITE_HELD(&ds->ds_rwlock))
772 rw_enter(&ds->ds_rwlock, RW_WRITER);
34dc7c2f
BB		 773}
774
775uint64_t
b128c09f 776dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
34dc7c2f
BB		 777 uint64_t flags, dmu_tx_t *tx)
778{
779 dsl_pool_t *dp = dd->dd_pool;
780 dmu_buf_t *dbuf;
781 dsl_dataset_phys_t *dsphys;
782 uint64_t dsobj;
783 objset_t *mos = dp->dp_meta_objset;
784
b128c09f
BB		 785 if (origin == NULL)
786 origin = dp->dp_origin_snap;
787
34dc7c2f
BB		 788 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
789 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
790 ASSERT(dmu_tx_is_syncing(tx));
791 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
792
793 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
794 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
795 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
796 dmu_buf_will_dirty(dbuf, tx);
797 dsphys = dbuf->db_data;
b128c09f 798 bzero(dsphys, sizeof (dsl_dataset_phys_t));
34dc7c2f
BB		 799 dsphys->ds_dir_obj = dd->dd_object;
800 dsphys->ds_flags = flags;
801 dsphys->ds_fsid_guid = unique_create();
802 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
803 sizeof (dsphys->ds_guid));
804 dsphys->ds_snapnames_zapobj =
805 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
806 DMU_OT_NONE, 0, tx);
807 dsphys->ds_creation_time = gethrestime_sec();
b128c09f 808 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
34dc7c2f 809
428870ff
BB		 810 if (origin == NULL) {
811 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
812 } else {
813 dsl_dataset_t *ohds;
814
34dc7c2f
BB		 815 dsphys->ds_prev_snap_obj = origin->ds_object;
816 dsphys->ds_prev_snap_txg =
817 origin->ds_phys->ds_creation_txg;
818 dsphys->ds_used_bytes =
819 origin->ds_phys->ds_used_bytes;
820 dsphys->ds_compressed_bytes =
821 origin->ds_phys->ds_compressed_bytes;
822 dsphys->ds_uncompressed_bytes =
823 origin->ds_phys->ds_uncompressed_bytes;
824 dsphys->ds_bp = origin->ds_phys->ds_bp;
825 dsphys->ds_flags |= origin->ds_phys->ds_flags;
826
827 dmu_buf_will_dirty(origin->ds_dbuf, tx);
828 origin->ds_phys->ds_num_children++;
829
428870ff
BB		 830 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
831 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
832 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
833 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
834 dsl_dataset_rele(ohds, FTAG);
835
b128c09f
BB		 836 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
837 if (origin->ds_phys->ds_next_clones_obj == 0) {
838 origin->ds_phys->ds_next_clones_obj =
839 zap_create(mos,
840 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
841 }
842 VERIFY(0 == zap_add_int(mos,
843 origin->ds_phys->ds_next_clones_obj,
844 dsobj, tx));
845 }
846
34dc7c2f
BB		 847 dmu_buf_will_dirty(dd->dd_dbuf, tx);
848 dd->dd_phys->dd_origin_obj = origin->ds_object;
428870ff
BB		 849 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
850 if (origin->ds_dir->dd_phys->dd_clones == 0) {
851 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
852 origin->ds_dir->dd_phys->dd_clones =
853 zap_create(mos,
854 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
855 }
856 VERIFY3U(0, ==, zap_add_int(mos,
857 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
858 }
34dc7c2f
BB		 859 }
860
861 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
862 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
863
864 dmu_buf_rele(dbuf, FTAG);
865
866 dmu_buf_will_dirty(dd->dd_dbuf, tx);
867 dd->dd_phys->dd_head_dataset_obj = dsobj;
868
869 return (dsobj);
870}
871
872uint64_t
873dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
874 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
875{
876 dsl_pool_t *dp = pdd->dd_pool;
877 uint64_t dsobj, ddobj;
878 dsl_dir_t *dd;
879
880 ASSERT(lastname[0] != '@');
881
b128c09f 882 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
34dc7c2f
BB		 883 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
884
b128c09f 885 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
34dc7c2f
BB		 886
887 dsl_deleg_set_create_perms(dd, tx, cr);
888
889 dsl_dir_close(dd, FTAG);
890
572e2857
BB		 891 /*
892 * If we are creating a clone, make sure we zero out any stale
893 * data from the origin snapshots zil header.
894 */
895 if (origin != NULL) {
896 dsl_dataset_t *ds;
897 objset_t *os;
898
899 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
900 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
901 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
902 dsl_dataset_dirty(ds, tx);
903 dsl_dataset_rele(ds, FTAG);
904 }
905
34dc7c2f
BB		 906 return (dsobj);
907}
908
909struct destroyarg {
910 dsl_sync_task_group_t *dstg;
911 char *snapname;
912 char *failed;
45d1cae3 913 boolean_t defer;
34dc7c2f
BB		 914};
915
916static int
428870ff 917dsl_snapshot_destroy_one(const char *name, void *arg)
34dc7c2f
BB		 918{
919 struct destroyarg *da = arg;
920 dsl_dataset_t *ds;
34dc7c2f 921 int err;
45d1cae3 922 char *dsname;
428870ff
BB		 923
924 dsname = kmem_asprintf("%s@%s", name, da->snapname);
925 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
926 strfree(dsname);
b128c09f 927 if (err == 0) {
45d1cae3
BB		 928 struct dsl_ds_destroyarg *dsda;
929
b128c09f 930 dsl_dataset_make_exclusive(ds, da->dstg);
45d1cae3
BB		 931 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
932 dsda->ds = ds;
933 dsda->defer = da->defer;
b128c09f 934 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
45d1cae3 935 dsl_dataset_destroy_sync, dsda, da->dstg, 0);
b128c09f
BB		 936 } else if (err == ENOENT) {
937 err = 0;
938 } else {
34dc7c2f 939 (void) strcpy(da->failed, name);
34dc7c2f 940 }
b128c09f 941 return (err);
34dc7c2f
BB		 942}
943
944/*
945 * Destroy 'snapname' in all descendants of 'fsname'.
946 */
947#pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
948int
45d1cae3 949dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
34dc7c2f
BB		 950{
951 int err;
952 struct destroyarg da;
953 dsl_sync_task_t *dst;
954 spa_t *spa;
955
956 err = spa_open(fsname, &spa, FTAG);
957 if (err)
958 return (err);
959 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
960 da.snapname = snapname;
961 da.failed = fsname;
45d1cae3 962 da.defer = defer;
34dc7c2f
BB		 963
964 err = dmu_objset_find(fsname,
965 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
966
967 if (err == 0)
968 err = dsl_sync_task_group_wait(da.dstg);
969
970 for (dst = list_head(&da.dstg->dstg_tasks); dst;
971 dst = list_next(&da.dstg->dstg_tasks, dst)) {
45d1cae3
BB		 972 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
973 dsl_dataset_t *ds = dsda->ds;
974
b128c09f
BB		 975 /*
976 * Return the file system name that triggered the error
977 */
34dc7c2f
BB		 978 if (dst->dst_err) {
979 dsl_dataset_name(ds, fsname);
980 *strchr(fsname, '@') = '\0';
981 }
45d1cae3 982 ASSERT3P(dsda->rm_origin, ==, NULL);
b128c09f 983 dsl_dataset_disown(ds, da.dstg);
45d1cae3 984 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
34dc7c2f
BB		 985 }
986
987 dsl_sync_task_group_destroy(da.dstg);
988 spa_close(spa, FTAG);
989 return (err);
990}
991
45d1cae3
BB		 992static boolean_t
993dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
994{
995 boolean_t might_destroy = B_FALSE;
996
997 mutex_enter(&ds->ds_lock);
998 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
999 DS_IS_DEFER_DESTROY(ds))
1000 might_destroy = B_TRUE;
1001 mutex_exit(&ds->ds_lock);
1002
1003 return (might_destroy);
1004}
1005
45d1cae3
BB		 1006/*
1007 * If we're removing a clone, and these three conditions are true:
1008 * 1) the clone's origin has no other children
1009 * 2) the clone's origin has no user references
1010 * 3) the clone's origin has been marked for deferred destruction
1011 * Then, prepare to remove the origin as part of this sync task group.
1012 */
1013static int
1014dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1015{
1016 dsl_dataset_t *ds = dsda->ds;
1017 dsl_dataset_t *origin = ds->ds_prev;
1018
1019 if (dsl_dataset_might_destroy_origin(origin)) {
1020 char *name;
1021 int namelen;
1022 int error;
1023
1024 namelen = dsl_dataset_namelen(origin) + 1;
1025 name = kmem_alloc(namelen, KM_SLEEP);
1026 dsl_dataset_name(origin, name);
1027#ifdef _KERNEL
1028 error = zfs_unmount_snap(name, NULL);
1029 if (error) {
1030 kmem_free(name, namelen);
1031 return (error);
1032 }
45d1cae3 1033#endif
428870ff 1034 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
45d1cae3
BB		 1035 kmem_free(name, namelen);
1036 if (error)
1037 return (error);
1038 dsda->rm_origin = origin;
1039 dsl_dataset_make_exclusive(origin, tag);
1040 }
1041
1042 return (0);
1043}
1044
34dc7c2f 1045/*
b128c09f
BB		 1046 * ds must be opened as OWNER. On return (whether successful or not),
1047 * ds will be closed and caller can no longer dereference it.
34dc7c2f
BB		 1048 */
1049int
45d1cae3 1050dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
34dc7c2f
BB		 1051{
1052 int err;
1053 dsl_sync_task_group_t *dstg;
1054 objset_t *os;
1055 dsl_dir_t *dd;
1056 uint64_t obj;
428870ff 1057 struct dsl_ds_destroyarg dsda = { 0 };
81a49663 1058 dsl_dataset_t *dummy_ds;
45d1cae3
BB		 1059
1060 dsda.ds = ds;
34dc7c2f 1061
34dc7c2f
BB		 1062 if (dsl_dataset_is_snapshot(ds)) {
1063 /* Destroying a snapshot is simpler */
b128c09f
BB		 1064 dsl_dataset_make_exclusive(ds, tag);
1065
45d1cae3 1066 dsda.defer = defer;
34dc7c2f
BB		 1067 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1068 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
45d1cae3
BB		 1069 &dsda, tag, 0);
1070 ASSERT3P(dsda.rm_origin, ==, NULL);
34dc7c2f 1071 goto out;
428870ff
BB		 1072 } else if (defer) {
1073 err = EINVAL;
1074 goto out;
34dc7c2f
BB		 1075 }
1076
1077 dd = ds->ds_dir;
81a49663
BB		 1078 dummy_ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
1079 dummy_ds->ds_dir = dd;
1080 dummy_ds->ds_object = ds->ds_object;
34dc7c2f
BB		 1081
1082 /*
1083 * Check for errors and mark this ds as inconsistent, in
1084 * case we crash while freeing the objects.
1085 */
1086 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1087 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1088 if (err)
81a49663 1089 goto out_free;
34dc7c2f 1090
428870ff 1091 err = dmu_objset_from_ds(ds, &os);
34dc7c2f 1092 if (err)
81a49663 1093 goto out_free;
34dc7c2f
BB		 1094
1095 /*
1096 * remove the objects in open context, so that we won't
1097 * have too much to do in syncing context.
1098 */
1099 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1100 ds->ds_phys->ds_prev_snap_txg)) {
b128c09f
BB		 1101 /*
1102 * Ignore errors, if there is not enough disk space
1103 * we will deal with it in dsl_dataset_destroy_sync().
1104 */
1105 (void) dmu_free_object(os, obj);
34dc7c2f 1106 }
572e2857 1107 if (err != ESRCH)
81a49663 1108 goto out_free;
34dc7c2f 1109
9babb374 1110 /*
572e2857
BB		 1111 * Only the ZIL knows how to free log blocks.
1112 */
1113 zil_destroy(dmu_objset_zil(os), B_FALSE);
1114
1115 /*
1116 * Sync out all in-flight IO.
9babb374
BB		 1117 */
1118 txg_wait_synced(dd->dd_pool, 0);
1119
1120 /*
1121 * If we managed to free all the objects in open
1122 * context, the user space accounting should be zero.
1123 */
1124 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
428870ff 1125 dmu_objset_userused_enabled(os)) {
1fde1e37 1126 ASSERTV(uint64_t count);
9babb374
BB		 1127 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1128 count == 0);
1129 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1130 count == 0);
1131 }
1132
34dc7c2f
BB		 1133 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1134 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1135 rw_exit(&dd->dd_pool->dp_config_rwlock);
1136
1137 if (err)
81a49663 1138 goto out_free;
34dc7c2f
BB		 1139
1140 /*
1141 * Blow away the dsl_dir + head dataset.
1142 */
b128c09f 1143 dsl_dataset_make_exclusive(ds, tag);
45d1cae3
BB		 1144 /*
1145 * If we're removing a clone, we might also need to remove its
1146 * origin.
1147 */
1148 do {
1149 dsda.need_prep = B_FALSE;
1150 if (dsl_dir_is_clone(dd)) {
1151 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1152 if (err) {
1153 dsl_dir_close(dd, FTAG);
81a49663 1154 goto out_free;
45d1cae3
BB		 1155 }
1156 }
1157
1158 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1159 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1160 dsl_dataset_destroy_sync, &dsda, tag, 0);
1161 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
81a49663 1162 dsl_dir_destroy_sync, dummy_ds, FTAG, 0);
45d1cae3
BB		 1163 err = dsl_sync_task_group_wait(dstg);
1164 dsl_sync_task_group_destroy(dstg);
1165
1166 /*
1167 * We could be racing against 'zfs release' or 'zfs destroy -d'
1168 * on the origin snap, in which case we can get EBUSY if we
1169 * needed to destroy the origin snap but were not ready to
1170 * do so.
1171 */
1172 if (dsda.need_prep) {
1173 ASSERT(err == EBUSY);
1174 ASSERT(dsl_dir_is_clone(dd));
1175 ASSERT(dsda.rm_origin == NULL);
1176 }
1177 } while (dsda.need_prep);
1178
1179 if (dsda.rm_origin != NULL)
1180 dsl_dataset_disown(dsda.rm_origin, tag);
1181
b128c09f 1182 /* if it is successful, dsl_dir_destroy_sync will close the dd */
34dc7c2f
BB		 1183 if (err)
1184 dsl_dir_close(dd, FTAG);
81a49663
BB		 1185
1186out_free:
1187 kmem_free(dummy_ds, sizeof (dsl_dataset_t));
34dc7c2f 1188out:
b128c09f 1189 dsl_dataset_disown(ds, tag);
34dc7c2f
BB		 1190 return (err);
1191}
1192
34dc7c2f
BB		 1193blkptr_t *
1194dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1195{
1196 return (&ds->ds_phys->ds_bp);
1197}
1198
1199void
1200dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1201{
1202 ASSERT(dmu_tx_is_syncing(tx));
1203 /* If it's the meta-objset, set dp_meta_rootbp */
1204 if (ds == NULL) {
1205 tx->tx_pool->dp_meta_rootbp = *bp;
1206 } else {
1207 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1208 ds->ds_phys->ds_bp = *bp;
1209 }
1210}
1211
1212spa_t *
1213dsl_dataset_get_spa(dsl_dataset_t *ds)
1214{
1215 return (ds->ds_dir->dd_pool->dp_spa);
1216}
1217
1218void
1219dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1220{
1221 dsl_pool_t *dp;
1222
1223 if (ds == NULL) /* this is the meta-objset */
1224 return;
1225
428870ff 1226 ASSERT(ds->ds_objset != NULL);
34dc7c2f
BB		 1227
1228 if (ds->ds_phys->ds_next_snap_obj != 0)
1229 panic("dirtying snapshot!");
1230
1231 dp = ds->ds_dir->dd_pool;
1232
1233 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1234 /* up the hold count until we can be written out */
1235 dmu_buf_add_ref(ds->ds_dbuf, ds);
1236 }
1237}
1238
1239/*
1240 * The unique space in the head dataset can be calculated by subtracting
1241 * the space used in the most recent snapshot, that is still being used
1242 * in this file system, from the space currently in use. To figure out
1243 * the space in the most recent snapshot still in use, we need to take
1244 * the total space used in the snapshot and subtract out the space that
1245 * has been freed up since the snapshot was taken.
1246 */
1247static void
1248dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1249{
1250 uint64_t mrs_used;
1251 uint64_t dlused, dlcomp, dluncomp;
1252
428870ff 1253 ASSERT(!dsl_dataset_is_snapshot(ds));
34dc7c2f
BB		 1254
1255 if (ds->ds_phys->ds_prev_snap_obj != 0)
1256 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1257 else
1258 mrs_used = 0;
1259
428870ff 1260 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
34dc7c2f
BB		 1261
1262 ASSERT3U(dlused, <=, mrs_used);
1263 ds->ds_phys->ds_unique_bytes =
1264 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1265
428870ff 1266 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
34dc7c2f
BB		 1267 SPA_VERSION_UNIQUE_ACCURATE)
1268 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1269}
1270
34dc7c2f 1271struct killarg {
b128c09f 1272 dsl_dataset_t *ds;
34dc7c2f
BB		 1273 dmu_tx_t *tx;
1274};
1275
b128c09f 1276/* ARGSUSED */
34dc7c2f 1277static int
428870ff
BB		 1278kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1279 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
34dc7c2f
BB		 1280{
1281 struct killarg *ka = arg;
428870ff 1282 dmu_tx_t *tx = ka->tx;
34dc7c2f 1283
b128c09f
BB		 1284 if (bp == NULL)
1285 return (0);
1286
428870ff
BB		 1287 if (zb->zb_level == ZB_ZIL_LEVEL) {
1288 ASSERT(zilog != NULL);
9babb374
BB		 1289 /*
1290 * It's a block in the intent log. It has no
1291 * accounting, so just free it.
1292 */
428870ff 1293 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
9babb374 1294 } else {
428870ff 1295 ASSERT(zilog == NULL);
9babb374 1296 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
428870ff 1297 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
9babb374 1298 }
34dc7c2f 1299
34dc7c2f
BB		 1300 return (0);
1301}
1302
34dc7c2f
BB		 1303/* ARGSUSED */
1304static int
1305dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1306{
1307 dsl_dataset_t *ds = arg1;
1308 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1309 uint64_t count;
1310 int err;
1311
1312 /*
1313 * Can't delete a head dataset if there are snapshots of it.
1314 * (Except if the only snapshots are from the branch we cloned
1315 * from.)
1316 */
1317 if (ds->ds_prev != NULL &&
1318 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
428870ff 1319 return (EBUSY);
34dc7c2f
BB		 1320
1321 /*
1322 * This is really a dsl_dir thing, but check it here so that
1323 * we'll be less likely to leave this dataset inconsistent &
1324 * nearly destroyed.
1325 */
1326 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1327 if (err)
1328 return (err);
1329 if (count != 0)
1330 return (EEXIST);
1331
1332 return (0);
1333}
1334
1335/* ARGSUSED */
1336static void
428870ff 1337dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 1338{
1339 dsl_dataset_t *ds = arg1;
1340 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1341
1342 /* Mark it as inconsistent on-disk, in case we crash */
1343 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1344 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1345
428870ff
BB		 1346 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1347 "dataset = %llu", ds->ds_object);
34dc7c2f
BB		 1348}
1349
45d1cae3
BB		 1350static int
1351dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1352 dmu_tx_t *tx)
1353{
1354 dsl_dataset_t *ds = dsda->ds;
1355 dsl_dataset_t *ds_prev = ds->ds_prev;
1356
1357 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1358 struct dsl_ds_destroyarg ndsda = {0};
1359
1360 /*
1361 * If we're not prepared to remove the origin, don't remove
1362 * the clone either.
1363 */
1364 if (dsda->rm_origin == NULL) {
1365 dsda->need_prep = B_TRUE;
1366 return (EBUSY);
1367 }
1368
1369 ndsda.ds = ds_prev;
1370 ndsda.is_origin_rm = B_TRUE;
1371 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1372 }
1373
1374 /*
1375 * If we're not going to remove the origin after all,
1376 * undo the open context setup.
1377 */
1378 if (dsda->rm_origin != NULL) {
1379 dsl_dataset_disown(dsda->rm_origin, tag);
1380 dsda->rm_origin = NULL;
1381 }
1382
1383 return (0);
1384}
1385
572e2857
BB		 1386/*
1387 * If you add new checks here, you may need to add
1388 * additional checks to the "temporary" case in
1389 * snapshot_check() in dmu_objset.c.
1390 */
34dc7c2f
BB		 1391/* ARGSUSED */
1392int
1393dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1394{
45d1cae3
BB		 1395 struct dsl_ds_destroyarg *dsda = arg1;
1396 dsl_dataset_t *ds = dsda->ds;
34dc7c2f 1397
b128c09f
BB		 1398 /* we have an owner hold, so noone else can destroy us */
1399 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1400
45d1cae3
BB		 1401 /*
1402 * Only allow deferred destroy on pools that support it.
1403 * NOTE: deferred destroy is only supported on snapshots.
1404 */
1405 if (dsda->defer) {
1406 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1407 SPA_VERSION_USERREFS)
1408 return (ENOTSUP);
1409 ASSERT(dsl_dataset_is_snapshot(ds));
1410 return (0);
1411 }
34dc7c2f
BB		 1412
1413 /*
1414 * Can't delete a head dataset if there are snapshots of it.
1415 * (Except if the only snapshots are from the branch we cloned
1416 * from.)
1417 */
1418 if (ds->ds_prev != NULL &&
1419 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
428870ff 1420 return (EBUSY);
34dc7c2f
BB		 1421
1422 /*
1423 * If we made changes this txg, traverse_dsl_dataset won't find
1424 * them. Try again.
1425 */
1426 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1427 return (EAGAIN);
1428
45d1cae3
BB		 1429 if (dsl_dataset_is_snapshot(ds)) {
1430 /*
1431 * If this snapshot has an elevated user reference count,
1432 * we can't destroy it yet.
1433 */
1434 if (ds->ds_userrefs > 0 && !dsda->releasing)
1435 return (EBUSY);
1436
1437 mutex_enter(&ds->ds_lock);
1438 /*
1439 * Can't delete a branch point. However, if we're destroying
1440 * a clone and removing its origin due to it having a user
1441 * hold count of 0 and having been marked for deferred destroy,
1442 * it's OK for the origin to have a single clone.
1443 */
1444 if (ds->ds_phys->ds_num_children >
1445 (dsda->is_origin_rm ? 2 : 1)) {
1446 mutex_exit(&ds->ds_lock);
1447 return (EEXIST);
1448 }
1449 mutex_exit(&ds->ds_lock);
1450 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1451 return (dsl_dataset_origin_check(dsda, arg2, tx));
1452 }
1453
34dc7c2f
BB		 1454 /* XXX we should do some i/o error checking... */
1455 return (0);
1456}
1457
b128c09f
BB		 1458struct refsarg {
1459 kmutex_t lock;
1460 boolean_t gone;
1461 kcondvar_t cv;
1462};
1463
1464/* ARGSUSED */
1465static void
1466dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1467{
1468 struct refsarg *arg = argv;
1469
1470 mutex_enter(&arg->lock);
1471 arg->gone = TRUE;
1472 cv_signal(&arg->cv);
1473 mutex_exit(&arg->lock);
1474}
1475
1476static void
1477dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1478{
1479 struct refsarg arg;
1480
1481 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1482 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1483 arg.gone = FALSE;
1484 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1485 dsl_dataset_refs_gone);
1486 dmu_buf_rele(ds->ds_dbuf, tag);
1487 mutex_enter(&arg.lock);
1488 while (!arg.gone)
1489 cv_wait(&arg.cv, &arg.lock);
1490 ASSERT(arg.gone);
1491 mutex_exit(&arg.lock);
1492 ds->ds_dbuf = NULL;
1493 ds->ds_phys = NULL;
1494 mutex_destroy(&arg.lock);
1495 cv_destroy(&arg.cv);
1496}
1497
428870ff
BB		 1498static void
1499remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1500{
1501 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
428870ff 1502 int err;
1fde1e37 1503 ASSERTV(uint64_t count);
428870ff
BB		 1504
1505 ASSERT(ds->ds_phys->ds_num_children >= 2);
1506 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1507 /*
1508 * The err should not be ENOENT, but a bug in a previous version
1509 * of the code could cause upgrade_clones_cb() to not set
1510 * ds_next_snap_obj when it should, leading to a missing entry.
1511 * If we knew that the pool was created after
1512 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1513 * ENOENT. However, at least we can check that we don't have
1514 * too many entries in the next_clones_obj even after failing to
1515 * remove this one.
1516 */
1517 if (err != ENOENT) {
1518 VERIFY3U(err, ==, 0);
1519 }
1520 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1521 &count));
1522 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1523}
1524
1525static void
1526dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1527{
1528 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1529 zap_cursor_t zc;
1530 zap_attribute_t za;
1531
1532 /*
1533 * If it is the old version, dd_clones doesn't exist so we can't
1534 * find the clones, but deadlist_remove_key() is a no-op so it
1535 * doesn't matter.
1536 */
1537 if (ds->ds_dir->dd_phys->dd_clones == 0)
1538 return;
1539
1540 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1541 zap_cursor_retrieve(&zc, &za) == 0;
1542 zap_cursor_advance(&zc)) {
1543 dsl_dataset_t *clone;
1544
1545 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1546 za.za_first_integer, FTAG, &clone));
1547 if (clone->ds_dir->dd_origin_txg > mintxg) {
1548 dsl_deadlist_remove_key(&clone->ds_deadlist,
1549 mintxg, tx);
1550 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1551 }
1552 dsl_dataset_rele(clone, FTAG);
1553 }
1554 zap_cursor_fini(&zc);
1555}
1556
1557struct process_old_arg {
1558 dsl_dataset_t *ds;
1559 dsl_dataset_t *ds_prev;
1560 boolean_t after_branch_point;
1561 zio_t *pio;
1562 uint64_t used, comp, uncomp;
1563};
1564
1565static int
1566process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1567{
1568 struct process_old_arg *poa = arg;
1569 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1570
1571 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1572 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1573 if (poa->ds_prev && !poa->after_branch_point &&
1574 bp->blk_birth >
1575 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1576 poa->ds_prev->ds_phys->ds_unique_bytes +=
1577 bp_get_dsize_sync(dp->dp_spa, bp);
1578 }
1579 } else {
1580 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1581 poa->comp += BP_GET_PSIZE(bp);
1582 poa->uncomp += BP_GET_UCSIZE(bp);
1583 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1584 }
1585 return (0);
1586}
1587
1588static void
1589process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1590 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1591{
1592 struct process_old_arg poa = { 0 };
1593 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1594 objset_t *mos = dp->dp_meta_objset;
1595
1596 ASSERT(ds->ds_deadlist.dl_oldfmt);
1597 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1598
1599 poa.ds = ds;
1600 poa.ds_prev = ds_prev;
1601 poa.after_branch_point = after_branch_point;
1602 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1603 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1604 process_old_cb, &poa, tx));
1605 VERIFY3U(zio_wait(poa.pio), ==, 0);
1606 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1607
1608 /* change snapused */
1609 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1610 -poa.used, -poa.comp, -poa.uncomp, tx);
1611
1612 /* swap next's deadlist to our deadlist */
1613 dsl_deadlist_close(&ds->ds_deadlist);
1614 dsl_deadlist_close(&ds_next->ds_deadlist);
1615 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1616 ds->ds_phys->ds_deadlist_obj);
1617 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1618 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1619 ds_next->ds_phys->ds_deadlist_obj);
1620}
1621
34dc7c2f 1622void
428870ff 1623dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
34dc7c2f 1624{
45d1cae3
BB		 1625 struct dsl_ds_destroyarg *dsda = arg1;
1626 dsl_dataset_t *ds = dsda->ds;
34dc7c2f
BB		 1627 int err;
1628 int after_branch_point = FALSE;
1629 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1630 objset_t *mos = dp->dp_meta_objset;
1631 dsl_dataset_t *ds_prev = NULL;
572e2857 1632 boolean_t wont_destroy;
34dc7c2f
BB		 1633 uint64_t obj;
1634
572e2857
BB		 1635 wont_destroy = (dsda->defer &&
1636 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1637
1638 ASSERT(ds->ds_owner || wont_destroy);
45d1cae3 1639 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
34dc7c2f
BB		 1640 ASSERT(ds->ds_prev == NULL ||
1641 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1642 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1643
572e2857 1644 if (wont_destroy) {
45d1cae3 1645 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
572e2857
BB		 1646 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1647 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1648 return;
45d1cae3
BB		 1649 }
1650
b128c09f
BB		 1651 /* signal any waiters that this dataset is going away */
1652 mutex_enter(&ds->ds_lock);
1653 ds->ds_owner = dsl_reaper;
1654 cv_broadcast(&ds->ds_exclusive_cv);
1655 mutex_exit(&ds->ds_lock);
1656
34dc7c2f
BB		 1657 /* Remove our reservation */
1658 if (ds->ds_reserved != 0) {
428870ff
BB		 1659 dsl_prop_setarg_t psa;
1660 uint64_t value = 0;
1661
1662 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1663 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1664 &value);
1665 psa.psa_effective_value = 0; /* predict default value */
1666
1667 dsl_dataset_set_reservation_sync(ds, &psa, tx);
34dc7c2f
BB		 1668 ASSERT3U(ds->ds_reserved, ==, 0);
1669 }
1670
1671 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1672
428870ff 1673 dsl_scan_ds_destroyed(ds, tx);
b128c09f 1674
34dc7c2f
BB		 1675 obj = ds->ds_object;
1676
1677 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1678 if (ds->ds_prev) {
1679 ds_prev = ds->ds_prev;
1680 } else {
b128c09f
BB		 1681 VERIFY(0 == dsl_dataset_hold_obj(dp,
1682 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
34dc7c2f
BB		 1683 }
1684 after_branch_point =
1685 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1686
1687 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
b128c09f
BB		 1688 if (after_branch_point &&
1689 ds_prev->ds_phys->ds_next_clones_obj != 0) {
428870ff 1690 remove_from_next_clones(ds_prev, obj, tx);
b128c09f
BB		 1691 if (ds->ds_phys->ds_next_snap_obj != 0) {
1692 VERIFY(0 == zap_add_int(mos,
1693 ds_prev->ds_phys->ds_next_clones_obj,
1694 ds->ds_phys->ds_next_snap_obj, tx));
1695 }
1696 }
34dc7c2f
BB		 1697 if (after_branch_point &&
1698 ds->ds_phys->ds_next_snap_obj == 0) {
1699 /* This clone is toast. */
1700 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1701 ds_prev->ds_phys->ds_num_children--;
45d1cae3
BB		 1702
1703 /*
1704 * If the clone's origin has no other clones, no
1705 * user holds, and has been marked for deferred
1706 * deletion, then we should have done the necessary
1707 * destroy setup for it.
1708 */
1709 if (ds_prev->ds_phys->ds_num_children == 1 &&
1710 ds_prev->ds_userrefs == 0 &&
1711 DS_IS_DEFER_DESTROY(ds_prev)) {
1712 ASSERT3P(dsda->rm_origin, !=, NULL);
1713 } else {
1714 ASSERT3P(dsda->rm_origin, ==, NULL);
1715 }
34dc7c2f
BB		 1716 } else if (!after_branch_point) {
1717 ds_prev->ds_phys->ds_next_snap_obj =
1718 ds->ds_phys->ds_next_snap_obj;
1719 }
1720 }
1721
428870ff 1722 if (dsl_dataset_is_snapshot(ds)) {
34dc7c2f 1723 dsl_dataset_t *ds_next;
34dc7c2f 1724 uint64_t old_unique;
428870ff 1725 uint64_t used = 0, comp = 0, uncomp = 0;
34dc7c2f 1726
b128c09f
BB		 1727 VERIFY(0 == dsl_dataset_hold_obj(dp,
1728 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
34dc7c2f
BB		 1729 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1730
428870ff 1731 old_unique = ds_next->ds_phys->ds_unique_bytes;
34dc7c2f
BB		 1732
1733 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1734 ds_next->ds_phys->ds_prev_snap_obj =
1735 ds->ds_phys->ds_prev_snap_obj;
1736 ds_next->ds_phys->ds_prev_snap_txg =
1737 ds->ds_phys->ds_prev_snap_txg;
1738 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1739 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1740
34dc7c2f 1741
428870ff
BB		 1742 if (ds_next->ds_deadlist.dl_oldfmt) {
1743 process_old_deadlist(ds, ds_prev, ds_next,
1744 after_branch_point, tx);
1745 } else {
1746 /* Adjust prev's unique space. */
1747 if (ds_prev && !after_branch_point) {
1748 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1749 ds_prev->ds_phys->ds_prev_snap_txg,
1750 ds->ds_phys->ds_prev_snap_txg,
1751 &used, &comp, &uncomp);
1752 ds_prev->ds_phys->ds_unique_bytes += used;
1753 }
b128c09f 1754
428870ff
BB		 1755 /* Adjust snapused. */
1756 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1757 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1758 &used, &comp, &uncomp);
1759 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1760 -used, -comp, -uncomp, tx);
1761
1762 /* Move blocks to be freed to pool's free list. */
1763 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1764 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1765 tx);
1766 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1767 DD_USED_HEAD, used, comp, uncomp, tx);
1768 dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1769
1770 /* Merge our deadlist into next's and free it. */
1771 dsl_deadlist_merge(&ds_next->ds_deadlist,
1772 ds->ds_phys->ds_deadlist_obj, tx);
1773 }
1774 dsl_deadlist_close(&ds->ds_deadlist);
1775 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
b128c09f 1776
428870ff
BB		 1777 /* Collapse range in clone heads */
1778 dsl_dataset_remove_clones_key(ds,
1779 ds->ds_phys->ds_creation_txg, tx);
34dc7c2f 1780
428870ff
BB		 1781 if (dsl_dataset_is_snapshot(ds_next)) {
1782 dsl_dataset_t *ds_nextnext;
d6320ddb 1783 dsl_dataset_t *hds;
34dc7c2f 1784
34dc7c2f
BB		 1785 /*
1786 * Update next's unique to include blocks which
1787 * were previously shared by only this snapshot
1788 * and it. Those blocks will be born after the
1789 * prev snap and before this snap, and will have
1790 * died after the next snap and before the one
1791 * after that (ie. be on the snap after next's
1792 * deadlist).
34dc7c2f 1793 */
b128c09f
BB		 1794 VERIFY(0 == dsl_dataset_hold_obj(dp,
1795 ds_next->ds_phys->ds_next_snap_obj,
428870ff
BB		 1796 FTAG, &ds_nextnext));
1797 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
b128c09f 1798 ds->ds_phys->ds_prev_snap_txg,
428870ff
BB		 1799 ds->ds_phys->ds_creation_txg,
1800 &used, &comp, &uncomp);
1801 ds_next->ds_phys->ds_unique_bytes += used;
1802 dsl_dataset_rele(ds_nextnext, FTAG);
34dc7c2f 1803 ASSERT3P(ds_next->ds_prev, ==, NULL);
428870ff
BB		 1804
1805 /* Collapse range in this head. */
428870ff
BB		 1806 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1807 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1808 FTAG, &hds));
1809 dsl_deadlist_remove_key(&hds->ds_deadlist,
1810 ds->ds_phys->ds_creation_txg, tx);
1811 dsl_dataset_rele(hds, FTAG);
1812
34dc7c2f
BB		 1813 } else {
1814 ASSERT3P(ds_next->ds_prev, ==, ds);
b128c09f
BB		 1815 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1816 ds_next->ds_prev = NULL;
34dc7c2f 1817 if (ds_prev) {
b128c09f
BB		 1818 VERIFY(0 == dsl_dataset_get_ref(dp,
1819 ds->ds_phys->ds_prev_snap_obj,
1820 ds_next, &ds_next->ds_prev));
34dc7c2f
BB		 1821 }
1822
1823 dsl_dataset_recalc_head_uniq(ds_next);
1824
1825 /*
1826 * Reduce the amount of our unconsmed refreservation
1827 * being charged to our parent by the amount of
1828 * new unique data we have gained.
1829 */
1830 if (old_unique < ds_next->ds_reserved) {
1831 int64_t mrsdelta;
1832 uint64_t new_unique =
1833 ds_next->ds_phys->ds_unique_bytes;
1834
1835 ASSERT(old_unique <= new_unique);
1836 mrsdelta = MIN(new_unique - old_unique,
1837 ds_next->ds_reserved - old_unique);
b128c09f
BB		 1838 dsl_dir_diduse_space(ds->ds_dir,
1839 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
34dc7c2f
BB		 1840 }
1841 }
b128c09f 1842 dsl_dataset_rele(ds_next, FTAG);
34dc7c2f
BB		 1843 } else {
1844 /*
1845 * There's no next snapshot, so this is a head dataset.
1846 * Destroy the deadlist. Unless it's a clone, the
1847 * deadlist should be empty. (If it's a clone, it's
1848 * safe to ignore the deadlist contents.)
1849 */
1850 struct killarg ka;
1851
428870ff
BB		 1852 dsl_deadlist_close(&ds->ds_deadlist);
1853 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
34dc7c2f
BB		 1854 ds->ds_phys->ds_deadlist_obj = 0;
1855
1856 /*
1857 * Free everything that we point to (that's born after
1858 * the previous snapshot, if we are a clone)
1859 *
b128c09f
BB		 1860 * NB: this should be very quick, because we already
1861 * freed all the objects in open context.
34dc7c2f 1862 */
b128c09f 1863 ka.ds = ds;
34dc7c2f 1864 ka.tx = tx;
b128c09f
BB		 1865 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1866 TRAVERSE_POST, kill_blkptr, &ka);
34dc7c2f 1867 ASSERT3U(err, ==, 0);
9babb374 1868 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
b128c09f 1869 ds->ds_phys->ds_unique_bytes == 0);
34dc7c2f 1870
428870ff
BB		 1871 if (ds->ds_prev != NULL) {
1872 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1873 VERIFY3U(0, ==, zap_remove_int(mos,
1874 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1875 ds->ds_object, tx));
1876 }
1877 dsl_dataset_rele(ds->ds_prev, ds);
1878 ds->ds_prev = ds_prev = NULL;
1879 }
1880 }
34dc7c2f 1881
572e2857
BB		 1882 /*
1883 * This must be done after the dsl_traverse(), because it will
1884 * re-open the objset.
1885 */
1886 if (ds->ds_objset) {
1887 dmu_objset_evict(ds->ds_objset);
1888 ds->ds_objset = NULL;
1889 }
1890
34dc7c2f 1891 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
b128c09f 1892 /* Erase the link in the dir */
34dc7c2f
BB		 1893 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1894 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
b128c09f
BB		 1895 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1896 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1897 ASSERT(err == 0);
34dc7c2f
BB		 1898 } else {
1899 /* remove from snapshot namespace */
1900 dsl_dataset_t *ds_head;
b128c09f
BB		 1901 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1902 VERIFY(0 == dsl_dataset_hold_obj(dp,
1903 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
34dc7c2f
BB		 1904 VERIFY(0 == dsl_dataset_get_snapname(ds));
1905#ifdef ZFS_DEBUG
1906 {
1907 uint64_t val;
1908
b128c09f 1909 err = dsl_dataset_snap_lookup(ds_head,
34dc7c2f
BB		 1910 ds->ds_snapname, &val);
1911 ASSERT3U(err, ==, 0);
1912 ASSERT3U(val, ==, obj);
1913 }
1914#endif
b128c09f 1915 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
34dc7c2f 1916 ASSERT(err == 0);
b128c09f 1917 dsl_dataset_rele(ds_head, FTAG);
34dc7c2f
BB		 1918 }
1919
1920 if (ds_prev && ds->ds_prev != ds_prev)
b128c09f 1921 dsl_dataset_rele(ds_prev, FTAG);
34dc7c2f
BB		 1922
1923 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
428870ff
BB		 1924 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1925 "dataset = %llu", ds->ds_object);
34dc7c2f 1926
b128c09f 1927 if (ds->ds_phys->ds_next_clones_obj != 0) {
1fde1e37 1928 ASSERTV(uint64_t count);
b128c09f
BB		 1929 ASSERT(0 == zap_count(mos,
1930 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1931 VERIFY(0 == dmu_object_free(mos,
1932 ds->ds_phys->ds_next_clones_obj, tx));
1933 }
1934 if (ds->ds_phys->ds_props_obj != 0)
1935 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
45d1cae3
BB		 1936 if (ds->ds_phys->ds_userrefs_obj != 0)
1937 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
b128c09f
BB		 1938 dsl_dir_close(ds->ds_dir, ds);
1939 ds->ds_dir = NULL;
1940 dsl_dataset_drain_refs(ds, tag);
34dc7c2f 1941 VERIFY(0 == dmu_object_free(mos, obj, tx));
45d1cae3
BB		 1942
1943 if (dsda->rm_origin) {
1944 /*
1945 * Remove the origin of the clone we just destroyed.
1946 */
45d1cae3
BB		 1947 struct dsl_ds_destroyarg ndsda = {0};
1948
428870ff
BB		 1949 ndsda.ds = dsda->rm_origin;
1950 dsl_dataset_destroy_sync(&ndsda, tag, tx);
45d1cae3 1951 }
34dc7c2f
BB		 1952}
1953
1954static int
1955dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1956{
1957 uint64_t asize;
1958
1959 if (!dmu_tx_is_syncing(tx))
1960 return (0);
1961
1962 /*
1963 * If there's an fs-only reservation, any blocks that might become
1964 * owned by the snapshot dataset must be accommodated by space
1965 * outside of the reservation.
1966 */
428870ff
BB		 1967 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1968 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
572e2857 1969 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
34dc7c2f
BB		 1970 return (ENOSPC);
1971
1972 /*
1973 * Propogate any reserved space for this snapshot to other
1974 * snapshot checks in this sync group.
1975 */
1976 if (asize > 0)
1977 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1978
1979 return (0);
1980}
1981
34dc7c2f
BB		 1982int
1983dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1984{
1985 dsl_dataset_t *ds = arg1;
1986 const char *snapname = arg2;
34dc7c2f
BB		 1987 int err;
1988 uint64_t value;
1989
1990 /*
1991 * We don't allow multiple snapshots of the same txg. If there
1992 * is already one, try again.
1993 */
1994 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1995 return (EAGAIN);
1996
1997 /*
1998 * Check for conflicting name snapshot name.
1999 */
b128c09f 2000 err = dsl_dataset_snap_lookup(ds, snapname, &value);
34dc7c2f
BB		 2001 if (err == 0)
2002 return (EEXIST);
2003 if (err != ENOENT)
2004 return (err);
2005
2006 /*
2007 * Check that the dataset's name is not too long. Name consists
2008 * of the dataset's length + 1 for the @-sign + snapshot name's length
2009 */
2010 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2011 return (ENAMETOOLONG);
2012
2013 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2014 if (err)
2015 return (err);
2016
2017 ds->ds_trysnap_txg = tx->tx_txg;
2018 return (0);
2019}
2020
2021void
428870ff 2022dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2023{
2024 dsl_dataset_t *ds = arg1;
2025 const char *snapname = arg2;
2026 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2027 dmu_buf_t *dbuf;
2028 dsl_dataset_phys_t *dsphys;
b128c09f 2029 uint64_t dsobj, crtxg;
34dc7c2f
BB		 2030 objset_t *mos = dp->dp_meta_objset;
2031 int err;
2032
34dc7c2f
BB		 2033 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2034
b128c09f
BB		 2035 /*
2036 * The origin's ds_creation_txg has to be < TXG_INITIAL
2037 */
2038 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2039 crtxg = 1;
2040 else
2041 crtxg = tx->tx_txg;
2042
34dc7c2f
BB		 2043 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2044 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2045 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2046 dmu_buf_will_dirty(dbuf, tx);
2047 dsphys = dbuf->db_data;
b128c09f 2048 bzero(dsphys, sizeof (dsl_dataset_phys_t));
34dc7c2f
BB		 2049 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2050 dsphys->ds_fsid_guid = unique_create();
2051 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2052 sizeof (dsphys->ds_guid));
2053 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2054 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2055 dsphys->ds_next_snap_obj = ds->ds_object;
2056 dsphys->ds_num_children = 1;
2057 dsphys->ds_creation_time = gethrestime_sec();
b128c09f 2058 dsphys->ds_creation_txg = crtxg;
34dc7c2f
BB		 2059 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2060 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2061 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2062 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2063 dsphys->ds_flags = ds->ds_phys->ds_flags;
2064 dsphys->ds_bp = ds->ds_phys->ds_bp;
2065 dmu_buf_rele(dbuf, FTAG);
2066
2067 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2068 if (ds->ds_prev) {
b128c09f
BB		 2069 uint64_t next_clones_obj =
2070 ds->ds_prev->ds_phys->ds_next_clones_obj;
34dc7c2f
BB		 2071 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2072 ds->ds_object ||
2073 ds->ds_prev->ds_phys->ds_num_children > 1);
2074 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2075 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2076 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2077 ds->ds_prev->ds_phys->ds_creation_txg);
2078 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
b128c09f 2079 } else if (next_clones_obj != 0) {
428870ff
BB		 2080 remove_from_next_clones(ds->ds_prev,
2081 dsphys->ds_next_snap_obj, tx);
b128c09f
BB		 2082 VERIFY3U(0, ==, zap_add_int(mos,
2083 next_clones_obj, dsobj, tx));
34dc7c2f
BB		 2084 }
2085 }
2086
2087 /*
2088 * If we have a reference-reservation on this dataset, we will
2089 * need to increase the amount of refreservation being charged
2090 * since our unique space is going to zero.
2091 */
2092 if (ds->ds_reserved) {
428870ff
BB		 2093 int64_t delta;
2094 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2095 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
b128c09f 2096 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
428870ff 2097 delta, 0, 0, tx);
34dc7c2f
BB		 2098 }
2099
34dc7c2f 2100 dmu_buf_will_dirty(ds->ds_dbuf, tx);
428870ff
BB		 2101 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2102 ds->ds_dir->dd_myname, snapname, dsobj,
2103 ds->ds_phys->ds_prev_snap_txg);
2104 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2105 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2106 dsl_deadlist_close(&ds->ds_deadlist);
2107 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2108 dsl_deadlist_add_key(&ds->ds_deadlist,
2109 ds->ds_phys->ds_prev_snap_txg, tx);
2110
34dc7c2f
BB		 2111 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2112 ds->ds_phys->ds_prev_snap_obj = dsobj;
b128c09f 2113 ds->ds_phys->ds_prev_snap_txg = crtxg;
34dc7c2f
BB		 2114 ds->ds_phys->ds_unique_bytes = 0;
2115 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2116 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
34dc7c2f 2117
34dc7c2f
BB		 2118 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2119 snapname, 8, 1, &dsobj, tx);
2120 ASSERT(err == 0);
2121
2122 if (ds->ds_prev)
b128c09f
BB		 2123 dsl_dataset_drop_ref(ds->ds_prev, ds);
2124 VERIFY(0 == dsl_dataset_get_ref(dp,
2125 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2126
428870ff
BB		 2127 dsl_scan_ds_snapshotted(ds, tx);
2128
2129 dsl_dir_snap_cmtime_update(ds->ds_dir);
34dc7c2f 2130
428870ff 2131 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
34dc7c2f
BB		 2132 "dataset = %llu", dsobj);
2133}
2134
2135void
2136dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2137{
2138 ASSERT(dmu_tx_is_syncing(tx));
428870ff 2139 ASSERT(ds->ds_objset != NULL);
34dc7c2f
BB		 2140 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2141
2142 /*
2143 * in case we had to change ds_fsid_guid when we opened it,
2144 * sync it out now.
2145 */
2146 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2147 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2148
2149 dsl_dir_dirty(ds->ds_dir, tx);
428870ff 2150 dmu_objset_sync(ds->ds_objset, zio, tx);
34dc7c2f
BB		 2151}
2152
2153void
2154dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2155{
2156 uint64_t refd, avail, uobjs, aobjs;
2157
2158 dsl_dir_stats(ds->ds_dir, nv);
2159
2160 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2161 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2162 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2163
2164 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2165 ds->ds_phys->ds_creation_time);
2166 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2167 ds->ds_phys->ds_creation_txg);
2168 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2169 ds->ds_quota);
2170 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2171 ds->ds_reserved);
b128c09f
BB		 2172 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2173 ds->ds_phys->ds_guid);
428870ff
BB		 2174 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2175 ds->ds_phys->ds_unique_bytes);
2176 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2177 ds->ds_object);
2178 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2179 ds->ds_userrefs);
45d1cae3
BB		 2180 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2181 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
34dc7c2f
BB		 2182
2183 if (ds->ds_phys->ds_next_snap_obj) {
2184 /*
2185 * This is a snapshot; override the dd's space used with
2186 * our unique space and compression ratio.
2187 */
2188 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2189 ds->ds_phys->ds_unique_bytes);
2190 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2191 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2192 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2193 ds->ds_phys->ds_compressed_bytes));
2194 }
2195}
2196
2197void
2198dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2199{
2200 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2201 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2202 stat->dds_guid = ds->ds_phys->ds_guid;
2203 if (ds->ds_phys->ds_next_snap_obj) {
2204 stat->dds_is_snapshot = B_TRUE;
2205 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
fb5f0bc8
BB		 2206 } else {
2207 stat->dds_is_snapshot = B_FALSE;
2208 stat->dds_num_clones = 0;
34dc7c2f
BB		 2209 }
2210
2211 /* clone origin is really a dsl_dir thing... */
2212 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
b128c09f 2213 if (dsl_dir_is_clone(ds->ds_dir)) {
34dc7c2f
BB		 2214 dsl_dataset_t *ods;
2215
b128c09f
BB		 2216 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2217 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
34dc7c2f 2218 dsl_dataset_name(ods, stat->dds_origin);
b128c09f 2219 dsl_dataset_drop_ref(ods, FTAG);
fb5f0bc8
BB		 2220 } else {
2221 stat->dds_origin[0] = '\0';
34dc7c2f
BB		 2222 }
2223 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2224}
2225
2226uint64_t
2227dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2228{
2229 return (ds->ds_fsid_guid);
2230}
2231
2232void
2233dsl_dataset_space(dsl_dataset_t *ds,
2234 uint64_t *refdbytesp, uint64_t *availbytesp,
2235 uint64_t *usedobjsp, uint64_t *availobjsp)
2236{
2237 *refdbytesp = ds->ds_phys->ds_used_bytes;
2238 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2239 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2240 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2241 if (ds->ds_quota != 0) {
2242 /*
2243 * Adjust available bytes according to refquota
2244 */
2245 if (*refdbytesp < ds->ds_quota)
2246 *availbytesp = MIN(*availbytesp,
2247 ds->ds_quota - *refdbytesp);
2248 else
2249 *availbytesp = 0;
2250 }
2251 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2252 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2253}
2254
2255boolean_t
2256dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2257{
1fde1e37 2258 ASSERTV(dsl_pool_t *dp = ds->ds_dir->dd_pool);
34dc7c2f
BB		 2259
2260 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2261 dsl_pool_sync_context(dp));
2262 if (ds->ds_prev == NULL)
2263 return (B_FALSE);
2264 if (ds->ds_phys->ds_bp.blk_birth >
572e2857
BB		 2265 ds->ds_prev->ds_phys->ds_creation_txg) {
2266 objset_t *os, *os_prev;
2267 /*
2268 * It may be that only the ZIL differs, because it was
2269 * reset in the head. Don't count that as being
2270 * modified.
2271 */
2272 if (dmu_objset_from_ds(ds, &os) != 0)
2273 return (B_TRUE);
2274 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2275 return (B_TRUE);
2276 return (bcmp(&os->os_phys->os_meta_dnode,
2277 &os_prev->os_phys->os_meta_dnode,
2278 sizeof (os->os_phys->os_meta_dnode)) != 0);
2279 }
34dc7c2f
BB		 2280 return (B_FALSE);
2281}
2282
2283/* ARGSUSED */
2284static int
2285dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2286{
2287 dsl_dataset_t *ds = arg1;
2288 char *newsnapname = arg2;
2289 dsl_dir_t *dd = ds->ds_dir;
34dc7c2f
BB		 2290 dsl_dataset_t *hds;
2291 uint64_t val;
2292 int err;
2293
b128c09f
BB		 2294 err = dsl_dataset_hold_obj(dd->dd_pool,
2295 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
34dc7c2f
BB		 2296 if (err)
2297 return (err);
2298
2299 /* new name better not be in use */
b128c09f
BB		 2300 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2301 dsl_dataset_rele(hds, FTAG);
34dc7c2f
BB		 2302
2303 if (err == 0)
2304 err = EEXIST;
2305 else if (err == ENOENT)
2306 err = 0;
2307
2308 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2309 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2310 err = ENAMETOOLONG;
2311
2312 return (err);
2313}
2314
2315static void
428870ff 2316dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2317{
2318 dsl_dataset_t *ds = arg1;
2319 const char *newsnapname = arg2;
2320 dsl_dir_t *dd = ds->ds_dir;
2321 objset_t *mos = dd->dd_pool->dp_meta_objset;
2322 dsl_dataset_t *hds;
2323 int err;
2324
2325 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2326
b128c09f
BB		 2327 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2328 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
34dc7c2f
BB		 2329
2330 VERIFY(0 == dsl_dataset_get_snapname(ds));
b128c09f 2331 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
34dc7c2f
BB		 2332 ASSERT3U(err, ==, 0);
2333 mutex_enter(&ds->ds_lock);
2334 (void) strcpy(ds->ds_snapname, newsnapname);
2335 mutex_exit(&ds->ds_lock);
2336 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2337 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2338 ASSERT3U(err, ==, 0);
2339
428870ff
BB		 2340 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2341 "dataset = %llu", ds->ds_object);
b128c09f 2342 dsl_dataset_rele(hds, FTAG);
34dc7c2f
BB		 2343}
2344
2345struct renamesnaparg {
2346 dsl_sync_task_group_t *dstg;
2347 char failed[MAXPATHLEN];
2348 char *oldsnap;
2349 char *newsnap;
2350};
2351
2352static int
428870ff 2353dsl_snapshot_rename_one(const char *name, void *arg)
34dc7c2f
BB		 2354{
2355 struct renamesnaparg *ra = arg;
2356 dsl_dataset_t *ds = NULL;
428870ff 2357 char *snapname;
34dc7c2f
BB		 2358 int err;
2359
428870ff
BB		 2360 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2361 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
34dc7c2f
BB		 2362
2363 /*
2364 * For recursive snapshot renames the parent won't be changing
2365 * so we just pass name for both the to/from argument.
2366 */
c28b2279 2367#ifdef HAVE_ZPL
428870ff
BB		 2368 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2369 if (err != 0) {
2370 strfree(snapname);
2371 return (err == ENOENT ? 0 : err);
34dc7c2f 2372 }
c28b2279 2373#endif
34dc7c2f 2374
c28b2279
BB		 2375/* XXX: Ignore for SPL version until mounting the FS is supported */
2376#if defined(_KERNEL) && !defined(HAVE_SPL)
b128c09f
BB		 2377 /*
2378 * For all filesystems undergoing rename, we'll need to unmount it.
2379 */
428870ff 2380 (void) zfs_unmount_snap(snapname, NULL);
34dc7c2f 2381#endif
428870ff
BB		 2382 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2383 strfree(snapname);
2384 if (err != 0)
2385 return (err == ENOENT ? 0 : err);
34dc7c2f
BB		 2386
2387 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2388 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2389
2390 return (0);
2391}
2392
2393static int
2394dsl_recursive_rename(char *oldname, const char *newname)
2395{
2396 int err;
2397 struct renamesnaparg *ra;
2398 dsl_sync_task_t *dst;
2399 spa_t *spa;
2400 char *cp, *fsname = spa_strdup(oldname);
428870ff 2401 int len = strlen(oldname) + 1;
34dc7c2f
BB		 2402
2403 /* truncate the snapshot name to get the fsname */
2404 cp = strchr(fsname, '@');
2405 *cp = '\0';
2406
2407 err = spa_open(fsname, &spa, FTAG);
2408 if (err) {
428870ff 2409 kmem_free(fsname, len);
34dc7c2f
BB		 2410 return (err);
2411 }
2412 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2413 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2414
2415 ra->oldsnap = strchr(oldname, '@') + 1;
2416 ra->newsnap = strchr(newname, '@') + 1;
2417 *ra->failed = '\0';
2418
2419 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2420 DS_FIND_CHILDREN);
428870ff 2421 kmem_free(fsname, len);
34dc7c2f
BB		 2422
2423 if (err == 0) {
2424 err = dsl_sync_task_group_wait(ra->dstg);
2425 }
2426
2427 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2428 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2429 dsl_dataset_t *ds = dst->dst_arg1;
2430 if (dst->dst_err) {
2431 dsl_dir_name(ds->ds_dir, ra->failed);
428870ff
BB		 2432 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2433 (void) strlcat(ra->failed, ra->newsnap,
2434 sizeof (ra->failed));
34dc7c2f 2435 }
b128c09f 2436 dsl_dataset_rele(ds, ra->dstg);
34dc7c2f
BB		 2437 }
2438
2439 if (err)
428870ff 2440 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
34dc7c2f
BB		 2441
2442 dsl_sync_task_group_destroy(ra->dstg);
2443 kmem_free(ra, sizeof (struct renamesnaparg));
2444 spa_close(spa, FTAG);
2445 return (err);
2446}
2447
2448static int
428870ff 2449dsl_valid_rename(const char *oldname, void *arg)
34dc7c2f
BB		 2450{
2451 int delta = *(int *)arg;
2452
2453 if (strlen(oldname) + delta >= MAXNAMELEN)
2454 return (ENAMETOOLONG);
2455
2456 return (0);
2457}
2458
2459#pragma weak dmu_objset_rename = dsl_dataset_rename
2460int
b128c09f 2461dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
34dc7c2f
BB		 2462{
2463 dsl_dir_t *dd;
2464 dsl_dataset_t *ds;
2465 const char *tail;
2466 int err;
2467
2468 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2469 if (err)
2470 return (err);
428870ff 2471
34dc7c2f
BB		 2472 if (tail == NULL) {
2473 int delta = strlen(newname) - strlen(oldname);
2474
b128c09f 2475 /* if we're growing, validate child name lengths */
34dc7c2f
BB		 2476 if (delta > 0)
2477 err = dmu_objset_find(oldname, dsl_valid_rename,
2478 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2479
428870ff 2480 if (err == 0)
34dc7c2f
BB		 2481 err = dsl_dir_rename(dd, newname);
2482 dsl_dir_close(dd, FTAG);
2483 return (err);
2484 }
428870ff 2485
34dc7c2f 2486 if (tail[0] != '@') {
428870ff 2487 /* the name ended in a nonexistent component */
34dc7c2f
BB		 2488 dsl_dir_close(dd, FTAG);
2489 return (ENOENT);
2490 }
2491
2492 dsl_dir_close(dd, FTAG);
2493
2494 /* new name must be snapshot in same filesystem */
2495 tail = strchr(newname, '@');
2496 if (tail == NULL)
2497 return (EINVAL);
2498 tail++;
2499 if (strncmp(oldname, newname, tail - newname) != 0)
2500 return (EXDEV);
2501
2502 if (recursive) {
2503 err = dsl_recursive_rename(oldname, newname);
2504 } else {
b128c09f 2505 err = dsl_dataset_hold(oldname, FTAG, &ds);
34dc7c2f
BB		 2506 if (err)
2507 return (err);
2508
2509 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2510 dsl_dataset_snapshot_rename_check,
2511 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2512
b128c09f 2513 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 2514 }
2515
2516 return (err);
2517}
2518
b128c09f
BB		 2519struct promotenode {
2520 list_node_t link;
2521 dsl_dataset_t *ds;
2522};
2523
34dc7c2f 2524struct promotearg {
b128c09f 2525 list_t shared_snaps, origin_snaps, clone_snaps;
428870ff 2526 dsl_dataset_t *origin_origin;
b128c09f 2527 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
428870ff 2528 char *err_ds;
34dc7c2f
BB		 2529};
2530
b128c09f
BB		 2531static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2532
34dc7c2f
BB		 2533static int
2534dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2535{
2536 dsl_dataset_t *hds = arg1;
2537 struct promotearg *pa = arg2;
b128c09f
BB		 2538 struct promotenode *snap = list_head(&pa->shared_snaps);
2539 dsl_dataset_t *origin_ds = snap->ds;
34dc7c2f 2540 int err;
428870ff 2541 uint64_t unused;
34dc7c2f 2542
b128c09f
BB		 2543 /* Check that it is a real clone */
2544 if (!dsl_dir_is_clone(hds->ds_dir))
34dc7c2f
BB		 2545 return (EINVAL);
2546
2547 /* Since this is so expensive, don't do the preliminary check */
2548 if (!dmu_tx_is_syncing(tx))
2549 return (0);
2550
b128c09f
BB		 2551 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2552 return (EXDEV);
34dc7c2f
BB		 2553
2554 /* compute origin's new unique space */
b128c09f
BB		 2555 snap = list_tail(&pa->clone_snaps);
2556 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
428870ff
BB		 2557 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2558 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2559 &pa->unique, &unused, &unused);
34dc7c2f 2560
b128c09f
BB		 2561 /*
2562 * Walk the snapshots that we are moving
2563 *
2564 * Compute space to transfer. Consider the incremental changes
2565 * to used for each snapshot:
2566 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2567 * So each snapshot gave birth to:
2568 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2569 * So a sequence would look like:
2570 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2571 * Which simplifies to:
2572 * uN + kN + kN-1 + ... + k1 + k0
2573 * Note however, if we stop before we reach the ORIGIN we get:
2574 * uN + kN + kN-1 + ... + kM - uM-1
2575 */
2576 pa->used = origin_ds->ds_phys->ds_used_bytes;
2577 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2578 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2579 for (snap = list_head(&pa->shared_snaps); snap;
2580 snap = list_next(&pa->shared_snaps, snap)) {
34dc7c2f 2581 uint64_t val, dlused, dlcomp, dluncomp;
b128c09f 2582 dsl_dataset_t *ds = snap->ds;
34dc7c2f
BB		 2583
2584 /* Check that the snapshot name does not conflict */
b128c09f
BB		 2585 VERIFY(0 == dsl_dataset_get_snapname(ds));
2586 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
428870ff
BB		 2587 if (err == 0) {
2588 err = EEXIST;
2589 goto out;
2590 }
b128c09f 2591 if (err != ENOENT)
428870ff 2592 goto out;
34dc7c2f 2593
b128c09f 2594 /* The very first snapshot does not have a deadlist */
34dc7c2f 2595 if (ds->ds_phys->ds_prev_snap_obj == 0)
b128c09f 2596 continue;
34dc7c2f 2597
428870ff
BB		 2598 dsl_deadlist_space(&ds->ds_deadlist,
2599 &dlused, &dlcomp, &dluncomp);
b128c09f
BB		 2600 pa->used += dlused;
2601 pa->comp += dlcomp;
2602 pa->uncomp += dluncomp;
2603 }
34dc7c2f 2604
b128c09f
BB		 2605 /*
2606 * If we are a clone of a clone then we never reached ORIGIN,
2607 * so we need to subtract out the clone origin's used space.
2608 */
2609 if (pa->origin_origin) {
2610 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2611 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2612 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
34dc7c2f
BB		 2613 }
2614
2615 /* Check that there is enough space here */
b128c09f
BB		 2616 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2617 pa->used);
2618 if (err)
2619 return (err);
34dc7c2f 2620
b128c09f
BB		 2621 /*
2622 * Compute the amounts of space that will be used by snapshots
2623 * after the promotion (for both origin and clone). For each,
2624 * it is the amount of space that will be on all of their
2625 * deadlists (that was not born before their new origin).
2626 */
2627 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2628 uint64_t space;
2629
2630 /*
2631 * Note, typically this will not be a clone of a clone,
428870ff
BB		 2632 * so dd_origin_txg will be < TXG_INITIAL, so
2633 * these snaplist_space() -> dsl_deadlist_space_range()
b128c09f
BB		 2634 * calls will be fast because they do not have to
2635 * iterate over all bps.
2636 */
2637 snap = list_head(&pa->origin_snaps);
2638 err = snaplist_space(&pa->shared_snaps,
428870ff 2639 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
b128c09f
BB		 2640 if (err)
2641 return (err);
2642
2643 err = snaplist_space(&pa->clone_snaps,
428870ff 2644 snap->ds->ds_dir->dd_origin_txg, &space);
b128c09f
BB		 2645 if (err)
2646 return (err);
2647 pa->cloneusedsnap += space;
2648 }
2649 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2650 err = snaplist_space(&pa->origin_snaps,
2651 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2652 if (err)
2653 return (err);
2654 }
2655
2656 return (0);
428870ff
BB		 2657out:
2658 pa->err_ds = snap->ds->ds_snapname;
2659 return (err);
34dc7c2f
BB		 2660}
2661
2662static void
428870ff 2663dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2664{
2665 dsl_dataset_t *hds = arg1;
2666 struct promotearg *pa = arg2;
b128c09f
BB		 2667 struct promotenode *snap = list_head(&pa->shared_snaps);
2668 dsl_dataset_t *origin_ds = snap->ds;
2669 dsl_dataset_t *origin_head;
34dc7c2f
BB		 2670 dsl_dir_t *dd = hds->ds_dir;
2671 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2672 dsl_dir_t *odd = NULL;
b128c09f
BB		 2673 uint64_t oldnext_obj;
2674 int64_t delta;
34dc7c2f 2675
34dc7c2f
BB		 2676 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2677
b128c09f
BB		 2678 snap = list_head(&pa->origin_snaps);
2679 origin_head = snap->ds;
2680
34dc7c2f
BB		 2681 /*
2682 * We need to explicitly open odd, since origin_ds's dd will be
2683 * changing.
2684 */
2685 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2686 NULL, FTAG, &odd));
2687
b128c09f
BB		 2688 /* change origin's next snap */
2689 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2690 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2691 snap = list_tail(&pa->clone_snaps);
2692 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2693 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2694
2695 /* change the origin's next clone */
2696 if (origin_ds->ds_phys->ds_next_clones_obj) {
428870ff 2697 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
b128c09f
BB		 2698 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2699 origin_ds->ds_phys->ds_next_clones_obj,
2700 oldnext_obj, tx));
2701 }
2702
2703 /* change origin */
2704 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2705 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2706 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
428870ff 2707 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
b128c09f
BB		 2708 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2709 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
428870ff
BB		 2710 origin_head->ds_dir->dd_origin_txg =
2711 origin_ds->ds_phys->ds_creation_txg;
2712
2713 /* change dd_clone entries */
2714 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2715 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2716 odd->dd_phys->dd_clones, hds->ds_object, tx));
2717 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2718 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2719 hds->ds_object, tx));
2720
2721 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2722 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2723 origin_head->ds_object, tx));
2724 if (dd->dd_phys->dd_clones == 0) {
2725 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2726 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2727 }
2728 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2729 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2730
2731 }
34dc7c2f 2732
b128c09f
BB		 2733 /* move snapshots to this dir */
2734 for (snap = list_head(&pa->shared_snaps); snap;
2735 snap = list_next(&pa->shared_snaps, snap)) {
2736 dsl_dataset_t *ds = snap->ds;
2737
2738 /* unregister props as dsl_dir is changing */
428870ff
BB		 2739 if (ds->ds_objset) {
2740 dmu_objset_evict(ds->ds_objset);
2741 ds->ds_objset = NULL;
b128c09f 2742 }
34dc7c2f 2743 /* move snap name entry */
b128c09f
BB		 2744 VERIFY(0 == dsl_dataset_get_snapname(ds));
2745 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2746 ds->ds_snapname, tx));
34dc7c2f
BB		 2747 VERIFY(0 == zap_add(dp->dp_meta_objset,
2748 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2749 8, 1, &ds->ds_object, tx));
428870ff 2750
34dc7c2f
BB		 2751 /* change containing dsl_dir */
2752 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2753 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2754 ds->ds_phys->ds_dir_obj = dd->dd_object;
2755 ASSERT3P(ds->ds_dir, ==, odd);
2756 dsl_dir_close(ds->ds_dir, ds);
2757 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2758 NULL, ds, &ds->ds_dir));
2759
428870ff
BB		 2760 /* move any clone references */
2761 if (ds->ds_phys->ds_next_clones_obj &&
2762 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2763 zap_cursor_t zc;
2764 zap_attribute_t za;
2765
2766 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2767 ds->ds_phys->ds_next_clones_obj);
2768 zap_cursor_retrieve(&zc, &za) == 0;
2769 zap_cursor_advance(&zc)) {
2770 dsl_dataset_t *cnds;
2771 uint64_t o;
2772
2773 if (za.za_first_integer == oldnext_obj) {
2774 /*
2775 * We've already moved the
2776 * origin's reference.
2777 */
2778 continue;
2779 }
2780
2781 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2782 za.za_first_integer, FTAG, &cnds));
2783 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2784
2785 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2786 odd->dd_phys->dd_clones, o, tx), ==, 0);
2787 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2788 dd->dd_phys->dd_clones, o, tx), ==, 0);
2789 dsl_dataset_rele(cnds, FTAG);
2790 }
2791 zap_cursor_fini(&zc);
2792 }
2793
34dc7c2f 2794 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
34dc7c2f 2795 }
34dc7c2f 2796
b128c09f
BB		 2797 /*
2798 * Change space accounting.
2799 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2800 * both be valid, or both be 0 (resulting in delta == 0). This
2801 * is true for each of {clone,origin} independently.
2802 */
34dc7c2f 2803
b128c09f
BB		 2804 delta = pa->cloneusedsnap -
2805 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2806 ASSERT3S(delta, >=, 0);
2807 ASSERT3U(pa->used, >=, delta);
2808 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2809 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2810 pa->used - delta, pa->comp, pa->uncomp, tx);
2811
2812 delta = pa->originusedsnap -
2813 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2814 ASSERT3S(delta, <=, 0);
2815 ASSERT3U(pa->used, >=, -delta);
2816 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2817 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2818 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
34dc7c2f 2819
34dc7c2f
BB		 2820 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2821
2822 /* log history record */
428870ff
BB		 2823 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2824 "dataset = %llu", hds->ds_object);
34dc7c2f
BB		 2825
2826 dsl_dir_close(odd, FTAG);
34dc7c2f
BB		 2827}
2828
b128c09f
BB		 2829static char *snaplist_tag = "snaplist";
2830/*
2831 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2832 * (exclusive) and last_obj (inclusive). The list will be in reverse
2833 * order (last_obj will be the list_head()). If first_obj == 0, do all
2834 * snapshots back to this dataset's origin.
2835 */
2836static int
2837snaplist_make(dsl_pool_t *dp, boolean_t own,
2838 uint64_t first_obj, uint64_t last_obj, list_t *l)
2839{
2840 uint64_t obj = last_obj;
2841
2842 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2843
2844 list_create(l, sizeof (struct promotenode),
2845 offsetof(struct promotenode, link));
2846
2847 while (obj != first_obj) {
2848 dsl_dataset_t *ds;
2849 struct promotenode *snap;
2850 int err;
2851
2852 if (own) {
2853 err = dsl_dataset_own_obj(dp, obj,
2854 0, snaplist_tag, &ds);
2855 if (err == 0)
2856 dsl_dataset_make_exclusive(ds, snaplist_tag);
2857 } else {
2858 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2859 }
2860 if (err == ENOENT) {
2861 /* lost race with snapshot destroy */
2862 struct promotenode *last = list_tail(l);
2863 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2864 obj = last->ds->ds_phys->ds_prev_snap_obj;
2865 continue;
2866 } else if (err) {
2867 return (err);
2868 }
2869
2870 if (first_obj == 0)
2871 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2872
2873 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2874 snap->ds = ds;
2875 list_insert_tail(l, snap);
2876 obj = ds->ds_phys->ds_prev_snap_obj;
2877 }
2878
2879 return (0);
2880}
2881
2882static int
2883snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2884{
2885 struct promotenode *snap;
2886
2887 *spacep = 0;
2888 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
428870ff
BB		 2889 uint64_t used, comp, uncomp;
2890 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2891 mintxg, UINT64_MAX, &used, &comp, &uncomp);
b128c09f
BB		 2892 *spacep += used;
2893 }
2894 return (0);
2895}
2896
2897static void
2898snaplist_destroy(list_t *l, boolean_t own)
2899{
2900 struct promotenode *snap;
2901
9babb374 2902 if (!l || !list_link_active(&l->list_head))
b128c09f
BB		 2903 return;
2904
2905 while ((snap = list_tail(l)) != NULL) {
2906 list_remove(l, snap);
2907 if (own)
2908 dsl_dataset_disown(snap->ds, snaplist_tag);
2909 else
2910 dsl_dataset_rele(snap->ds, snaplist_tag);
2911 kmem_free(snap, sizeof (struct promotenode));
2912 }
2913 list_destroy(l);
2914}
2915
2916/*
2917 * Promote a clone. Nomenclature note:
2918 * "clone" or "cds": the original clone which is being promoted
2919 * "origin" or "ods": the snapshot which is originally clone's origin
2920 * "origin head" or "ohds": the dataset which is the head
2921 * (filesystem/volume) for the origin
2922 * "origin origin": the origin of the origin's filesystem (typically
2923 * NULL, indicating that the clone is not a clone of a clone).
2924 */
34dc7c2f 2925int
428870ff 2926dsl_dataset_promote(const char *name, char *conflsnap)
34dc7c2f
BB		 2927{
2928 dsl_dataset_t *ds;
b128c09f
BB		 2929 dsl_dir_t *dd;
2930 dsl_pool_t *dp;
34dc7c2f 2931 dmu_object_info_t doi;
2598c001 2932 struct promotearg pa;
b128c09f
BB		 2933 struct promotenode *snap;
2934 int err;
34dc7c2f 2935
2598c001 2936 bzero(&pa, sizeof(struct promotearg));
b128c09f 2937 err = dsl_dataset_hold(name, FTAG, &ds);
34dc7c2f
BB		 2938 if (err)
2939 return (err);
b128c09f
BB		 2940 dd = ds->ds_dir;
2941 dp = dd->dd_pool;
34dc7c2f 2942
b128c09f 2943 err = dmu_object_info(dp->dp_meta_objset,
34dc7c2f
BB		 2944 ds->ds_phys->ds_snapnames_zapobj, &doi);
2945 if (err) {
b128c09f 2946 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 2947 return (err);
2948 }
2949
b128c09f
BB		 2950 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2951 dsl_dataset_rele(ds, FTAG);
2952 return (EINVAL);
2953 }
2954
2955 /*
2956 * We are going to inherit all the snapshots taken before our
2957 * origin (i.e., our new origin will be our parent's origin).
2958 * Take ownership of them so that we can rename them into our
2959 * namespace.
2960 */
2961 rw_enter(&dp->dp_config_rwlock, RW_READER);
2962
2963 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2964 &pa.shared_snaps);
2965 if (err != 0)
2966 goto out;
2967
2968 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2969 if (err != 0)
2970 goto out;
2971
2972 snap = list_head(&pa.shared_snaps);
2973 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2974 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2975 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2976 if (err != 0)
2977 goto out;
2978
428870ff
BB		 2979 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2980 err = dsl_dataset_hold_obj(dp,
b128c09f 2981 snap->ds->ds_dir->dd_phys->dd_origin_obj,
428870ff 2982 FTAG, &pa.origin_origin);
b128c09f
BB		 2983 if (err != 0)
2984 goto out;
2985 }
2986
2987out:
2988 rw_exit(&dp->dp_config_rwlock);
2989
34dc7c2f
BB		 2990 /*
2991 * Add in 128x the snapnames zapobj size, since we will be moving
2992 * a bunch of snapnames to the promoted ds, and dirtying their
2993 * bonus buffers.
2994 */
b128c09f
BB		 2995 if (err == 0) {
2996 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2997 dsl_dataset_promote_sync, ds, &pa,
428870ff
BB		 2998 2 + 2 * doi.doi_physical_blocks_512);
2999 if (err && pa.err_ds && conflsnap)
3000 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
b128c09f
BB		 3001 }
3002
3003 snaplist_destroy(&pa.shared_snaps, B_TRUE);
3004 snaplist_destroy(&pa.clone_snaps, B_FALSE);
3005 snaplist_destroy(&pa.origin_snaps, B_FALSE);
3006 if (pa.origin_origin)
428870ff 3007 dsl_dataset_rele(pa.origin_origin, FTAG);
b128c09f 3008 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3009 return (err);
3010}
3011
3012struct cloneswaparg {
3013 dsl_dataset_t *cds; /* clone dataset */
3014 dsl_dataset_t *ohds; /* origin's head dataset */
3015 boolean_t force;
3016 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3017};
3018
3019/* ARGSUSED */
3020static int
3021dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3022{
3023 struct cloneswaparg *csa = arg1;
3024
3025 /* they should both be heads */
3026 if (dsl_dataset_is_snapshot(csa->cds) ||
3027 dsl_dataset_is_snapshot(csa->ohds))
3028 return (EINVAL);
3029
3030 /* the branch point should be just before them */
3031 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3032 return (EINVAL);
3033
428870ff
BB		 3034 /* cds should be the clone (unless they are unrelated) */
3035 if (csa->cds->ds_prev != NULL &&
3036 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3037 csa->ohds->ds_object !=
3038 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
34dc7c2f
BB		 3039 return (EINVAL);
3040
3041 /* the clone should be a child of the origin */
3042 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3043 return (EINVAL);
3044
3045 /* ohds shouldn't be modified unless 'force' */
3046 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3047 return (ETXTBSY);
3048
3049 /* adjust amount of any unconsumed refreservation */
3050 csa->unused_refres_delta =
3051 (int64_t)MIN(csa->ohds->ds_reserved,
3052 csa->ohds->ds_phys->ds_unique_bytes) -
3053 (int64_t)MIN(csa->ohds->ds_reserved,
3054 csa->cds->ds_phys->ds_unique_bytes);
3055
3056 if (csa->unused_refres_delta > 0 &&
3057 csa->unused_refres_delta >
3058 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3059 return (ENOSPC);
3060
428870ff
BB		 3061 if (csa->ohds->ds_quota != 0 &&
3062 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3063 return (EDQUOT);
3064
34dc7c2f
BB		 3065 return (0);
3066}
3067
3068/* ARGSUSED */
3069static void
428870ff 3070dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3071{
3072 struct cloneswaparg *csa = arg1;
3073 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
34dc7c2f
BB		 3074
3075 ASSERT(csa->cds->ds_reserved == 0);
428870ff
BB		 3076 ASSERT(csa->ohds->ds_quota == 0 ||
3077 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
34dc7c2f
BB		 3078
3079 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3080 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
34dc7c2f 3081
428870ff
BB		 3082 if (csa->cds->ds_objset != NULL) {
3083 dmu_objset_evict(csa->cds->ds_objset);
3084 csa->cds->ds_objset = NULL;
34dc7c2f
BB		 3085 }
3086
428870ff
BB		 3087 if (csa->ohds->ds_objset != NULL) {
3088 dmu_objset_evict(csa->ohds->ds_objset);
3089 csa->ohds->ds_objset = NULL;
34dc7c2f
BB		 3090 }
3091
428870ff
BB		 3092 /*
3093 * Reset origin's unique bytes, if it exists.
3094 */
3095 if (csa->cds->ds_prev) {
3096 dsl_dataset_t *origin = csa->cds->ds_prev;
3097 uint64_t comp, uncomp;
3098
3099 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3100 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3101 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3102 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3103 }
34dc7c2f
BB		 3104
3105 /* swap blkptrs */
3106 {
3107 blkptr_t tmp;
3108 tmp = csa->ohds->ds_phys->ds_bp;
3109 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3110 csa->cds->ds_phys->ds_bp = tmp;
3111 }
3112
3113 /* set dd_*_bytes */
3114 {
3115 int64_t dused, dcomp, duncomp;
3116 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3117 uint64_t odl_used, odl_comp, odl_uncomp;
3118
b128c09f
BB		 3119 ASSERT3U(csa->cds->ds_dir->dd_phys->
3120 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3121
428870ff
BB		 3122 dsl_deadlist_space(&csa->cds->ds_deadlist,
3123 &cdl_used, &cdl_comp, &cdl_uncomp);
3124 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3125 &odl_used, &odl_comp, &odl_uncomp);
b128c09f 3126
34dc7c2f
BB		 3127 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3128 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3129 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3130 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3131 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3132 cdl_uncomp -
3133 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3134
b128c09f 3135 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
34dc7c2f 3136 dused, dcomp, duncomp, tx);
b128c09f 3137 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
34dc7c2f 3138 -dused, -dcomp, -duncomp, tx);
b128c09f
BB		 3139
3140 /*
3141 * The difference in the space used by snapshots is the
3142 * difference in snapshot space due to the head's
3143 * deadlist (since that's the only thing that's
3144 * changing that affects the snapused).
3145 */
428870ff
BB		 3146 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3147 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3148 &cdl_used, &cdl_comp, &cdl_uncomp);
3149 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3150 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3151 &odl_used, &odl_comp, &odl_uncomp);
b128c09f
BB		 3152 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3153 DD_USED_HEAD, DD_USED_SNAP, tx);
34dc7c2f
BB		 3154 }
3155
34dc7c2f
BB		 3156 /* swap ds_*_bytes */
3157 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3158 csa->cds->ds_phys->ds_used_bytes);
3159 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3160 csa->cds->ds_phys->ds_compressed_bytes);
3161 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3162 csa->cds->ds_phys->ds_uncompressed_bytes);
3163 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3164 csa->cds->ds_phys->ds_unique_bytes);
3165
3166 /* apply any parent delta for change in unconsumed refreservation */
b128c09f
BB		 3167 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3168 csa->unused_refres_delta, 0, 0, tx);
34dc7c2f 3169
428870ff
BB		 3170 /*
3171 * Swap deadlists.
3172 */
3173 dsl_deadlist_close(&csa->cds->ds_deadlist);
3174 dsl_deadlist_close(&csa->ohds->ds_deadlist);
34dc7c2f
BB		 3175 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3176 csa->cds->ds_phys->ds_deadlist_obj);
428870ff
BB		 3177 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3178 csa->cds->ds_phys->ds_deadlist_obj);
3179 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3180 csa->ohds->ds_phys->ds_deadlist_obj);
b128c09f 3181
428870ff 3182 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
34dc7c2f
BB		 3183}
3184
3185/*
428870ff
BB		 3186 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3187 * recv" into an existing fs to swizzle the file system to the new
3188 * version, and by "zfs rollback". Can also be used to swap two
3189 * independent head datasets if neither has any snapshots.
34dc7c2f
BB		 3190 */
3191int
3192dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3193 boolean_t force)
3194{
3195 struct cloneswaparg csa;
b128c09f 3196 int error;
34dc7c2f 3197
b128c09f
BB		 3198 ASSERT(clone->ds_owner);
3199 ASSERT(origin_head->ds_owner);
3200retry:
572e2857
BB		 3201 /*
3202 * Need exclusive access for the swap. If we're swapping these
3203 * datasets back after an error, we already hold the locks.
3204 */
3205 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3206 rw_enter(&clone->ds_rwlock, RW_WRITER);
3207 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3208 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
b128c09f
BB		 3209 rw_exit(&clone->ds_rwlock);
3210 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3211 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3212 rw_exit(&origin_head->ds_rwlock);
3213 goto retry;
3214 }
3215 }
34dc7c2f
BB		 3216 csa.cds = clone;
3217 csa.ohds = origin_head;
3218 csa.force = force;
b128c09f 3219 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
34dc7c2f 3220 dsl_dataset_clone_swap_check,
b128c09f
BB		 3221 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3222 return (error);
34dc7c2f
BB		 3223}
3224
3225/*
3226 * Given a pool name and a dataset object number in that pool,
3227 * return the name of that dataset.
3228 */
3229int
3230dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3231{
3232 spa_t *spa;
3233 dsl_pool_t *dp;
b128c09f 3234 dsl_dataset_t *ds;
34dc7c2f
BB		 3235 int error;
3236
3237 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3238 return (error);
3239 dp = spa_get_dsl(spa);
3240 rw_enter(&dp->dp_config_rwlock, RW_READER);
b128c09f
BB		 3241 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3242 dsl_dataset_name(ds, buf);
3243 dsl_dataset_rele(ds, FTAG);
34dc7c2f 3244 }
34dc7c2f
BB		 3245 rw_exit(&dp->dp_config_rwlock);
3246 spa_close(spa, FTAG);
3247
b128c09f 3248 return (error);
34dc7c2f
BB		 3249}
3250
3251int
3252dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
b128c09f 3253 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
34dc7c2f
BB		 3254{
3255 int error = 0;
3256
3257 ASSERT3S(asize, >, 0);
3258
3259 /*
3260 * *ref_rsrv is the portion of asize that will come from any
3261 * unconsumed refreservation space.
3262 */
3263 *ref_rsrv = 0;
3264
3265 mutex_enter(&ds->ds_lock);
3266 /*
3267 * Make a space adjustment for reserved bytes.
3268 */
3269 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3270 ASSERT3U(*used, >=,
3271 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3272 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3273 *ref_rsrv =
3274 asize - MIN(asize, parent_delta(ds, asize + inflight));
3275 }
3276
3277 if (!check_quota || ds->ds_quota == 0) {
3278 mutex_exit(&ds->ds_lock);
3279 return (0);
3280 }
3281 /*
3282 * If they are requesting more space, and our current estimate
3283 * is over quota, they get to try again unless the actual
3284 * on-disk is over quota and there are no pending changes (which
3285 * may free up space for us).
3286 */
3287 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3288 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3289 error = ERESTART;
3290 else
3291 error = EDQUOT;
3292 }
3293 mutex_exit(&ds->ds_lock);
3294
3295 return (error);
3296}
3297
3298/* ARGSUSED */
3299static int
3300dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3301{
3302 dsl_dataset_t *ds = arg1;
428870ff
BB		 3303 dsl_prop_setarg_t *psa = arg2;
3304 int err;
34dc7c2f
BB		 3305
3306 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3307 return (ENOTSUP);
3308
428870ff
BB		 3309 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3310 return (err);
3311
3312 if (psa->psa_effective_value == 0)
34dc7c2f
BB		 3313 return (0);
3314
428870ff
BB		 3315 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3316 psa->psa_effective_value < ds->ds_reserved)
34dc7c2f
BB		 3317 return (ENOSPC);
3318
3319 return (0);
3320}
3321
428870ff
BB		 3322extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3323
34dc7c2f 3324void
428870ff 3325dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3326{
3327 dsl_dataset_t *ds = arg1;
428870ff
BB		 3328 dsl_prop_setarg_t *psa = arg2;
3329 uint64_t effective_value = psa->psa_effective_value;
34dc7c2f 3330
428870ff
BB		 3331 dsl_prop_set_sync(ds, psa, tx);
3332 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
34dc7c2f 3333
428870ff
BB		 3334 if (ds->ds_quota != effective_value) {
3335 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3336 ds->ds_quota = effective_value;
34dc7c2f 3337
428870ff
BB		 3338 spa_history_log_internal(LOG_DS_REFQUOTA,
3339 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3340 (longlong_t)ds->ds_quota, ds->ds_object);
3341 }
34dc7c2f
BB		 3342}
3343
3344int
428870ff 3345dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
34dc7c2f
BB		 3346{
3347 dsl_dataset_t *ds;
428870ff 3348 dsl_prop_setarg_t psa;
34dc7c2f
BB		 3349 int err;
3350
428870ff
BB		 3351 dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3352
b128c09f 3353 err = dsl_dataset_hold(dsname, FTAG, &ds);
34dc7c2f
BB		 3354 if (err)
3355 return (err);
3356
428870ff
BB		 3357 /*
3358 * If someone removes a file, then tries to set the quota, we
3359 * want to make sure the file freeing takes effect.
3360 */
3361 txg_wait_open(ds->ds_dir->dd_pool, 0);
3362
3363 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3364 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3365 ds, &psa, 0);
34dc7c2f 3366
b128c09f 3367 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3368 return (err);
3369}
3370
3371static int
3372dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3373{
3374 dsl_dataset_t *ds = arg1;
428870ff
BB		 3375 dsl_prop_setarg_t *psa = arg2;
3376 uint64_t effective_value;
34dc7c2f 3377 uint64_t unique;
428870ff 3378 int err;
34dc7c2f 3379
34dc7c2f
BB		 3380 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3381 SPA_VERSION_REFRESERVATION)
3382 return (ENOTSUP);
3383
3384 if (dsl_dataset_is_snapshot(ds))
3385 return (EINVAL);
3386
428870ff
BB		 3387 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3388 return (err);
3389
3390 effective_value = psa->psa_effective_value;
3391
34dc7c2f
BB		 3392 /*
3393 * If we are doing the preliminary check in open context, the
3394 * space estimates may be inaccurate.
3395 */
3396 if (!dmu_tx_is_syncing(tx))
3397 return (0);
3398
3399 mutex_enter(&ds->ds_lock);
428870ff
BB		 3400 if (!DS_UNIQUE_IS_ACCURATE(ds))
3401 dsl_dataset_recalc_head_uniq(ds);
3402 unique = ds->ds_phys->ds_unique_bytes;
34dc7c2f
BB		 3403 mutex_exit(&ds->ds_lock);
3404
428870ff
BB		 3405 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3406 uint64_t delta = MAX(unique, effective_value) -
d164b209
BB		 3407 MAX(unique, ds->ds_reserved);
3408
3409 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3410 return (ENOSPC);
3411 if (ds->ds_quota > 0 &&
428870ff 3412 effective_value > ds->ds_quota)
d164b209
BB		 3413 return (ENOSPC);
3414 }
34dc7c2f
BB		 3415
3416 return (0);
3417}
3418
34dc7c2f 3419static void
428870ff 3420dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3421{
3422 dsl_dataset_t *ds = arg1;
428870ff
BB		 3423 dsl_prop_setarg_t *psa = arg2;
3424 uint64_t effective_value = psa->psa_effective_value;
34dc7c2f
BB		 3425 uint64_t unique;
3426 int64_t delta;
3427
428870ff
BB		 3428 dsl_prop_set_sync(ds, psa, tx);
3429 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3430
34dc7c2f
BB		 3431 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3432
b128c09f 3433 mutex_enter(&ds->ds_dir->dd_lock);
34dc7c2f 3434 mutex_enter(&ds->ds_lock);
428870ff
BB		 3435 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3436 unique = ds->ds_phys->ds_unique_bytes;
3437 delta = MAX(0, (int64_t)(effective_value - unique)) -
34dc7c2f 3438 MAX(0, (int64_t)(ds->ds_reserved - unique));
428870ff 3439 ds->ds_reserved = effective_value;
34dc7c2f
BB		 3440 mutex_exit(&ds->ds_lock);
3441
b128c09f
BB		 3442 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3443 mutex_exit(&ds->ds_dir->dd_lock);
34dc7c2f 3444
428870ff
BB		 3445 spa_history_log_internal(LOG_DS_REFRESERV,
3446 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3447 (longlong_t)effective_value, ds->ds_object);
34dc7c2f
BB		 3448}
3449
3450int
428870ff
BB		 3451dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3452 uint64_t reservation)
34dc7c2f
BB		 3453{
3454 dsl_dataset_t *ds;
428870ff 3455 dsl_prop_setarg_t psa;
34dc7c2f
BB		 3456 int err;
3457
428870ff
BB		 3458 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3459 &reservation);
3460
b128c09f 3461 err = dsl_dataset_hold(dsname, FTAG, &ds);
34dc7c2f
BB		 3462 if (err)
3463 return (err);
3464
3465 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3466 dsl_dataset_set_reservation_check,
428870ff
BB		 3467 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3468
b128c09f 3469 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3470 return (err);
3471}
45d1cae3 3472
572e2857
BB		 3473typedef struct zfs_hold_cleanup_arg {
3474 dsl_pool_t *dp;
3475 uint64_t dsobj;
3476 char htag[MAXNAMELEN];
3477} zfs_hold_cleanup_arg_t;
3478
3479static void
3480dsl_dataset_user_release_onexit(void *arg)
3481{
3482 zfs_hold_cleanup_arg_t *ca = arg;
3483
3484 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3485 B_TRUE);
3486 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3487}
3488
3489void
3490dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3491 minor_t minor)
3492{
3493 zfs_hold_cleanup_arg_t *ca;
3494
3495 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3496 ca->dp = ds->ds_dir->dd_pool;
3497 ca->dsobj = ds->ds_object;
3498 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3499 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3500 dsl_dataset_user_release_onexit, ca, NULL));
3501}
428870ff
BB		 3502
3503/*
572e2857
BB		 3504 * If you add new checks here, you may need to add
3505 * additional checks to the "temporary" case in
3506 * snapshot_check() in dmu_objset.c.
428870ff 3507 */
45d1cae3
BB		 3508static int
3509dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3510{
3511 dsl_dataset_t *ds = arg1;
428870ff
BB		 3512 struct dsl_ds_holdarg *ha = arg2;
3513 char *htag = ha->htag;
45d1cae3
BB		 3514 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3515 int error = 0;
3516
3517 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3518 return (ENOTSUP);
3519
3520 if (!dsl_dataset_is_snapshot(ds))
3521 return (EINVAL);
3522
45d1cae3
BB		 3523 /* tags must be unique */
3524 mutex_enter(&ds->ds_lock);
3525 if (ds->ds_phys->ds_userrefs_obj) {
3526 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3527 8, 1, tx);
3528 if (error == 0)
3529 error = EEXIST;
3530 else if (error == ENOENT)
3531 error = 0;
3532 }
3533 mutex_exit(&ds->ds_lock);
3534
428870ff
BB		 3535 if (error == 0 && ha->temphold &&
3536 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3537 error = E2BIG;
3538
45d1cae3
BB		 3539 return (error);
3540}
3541
572e2857 3542void
428870ff 3543dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
45d1cae3
BB		 3544{
3545 dsl_dataset_t *ds = arg1;
428870ff
BB		 3546 struct dsl_ds_holdarg *ha = arg2;
3547 char *htag = ha->htag;
3548 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3549 objset_t *mos = dp->dp_meta_objset;
3550 uint64_t now = gethrestime_sec();
45d1cae3
BB		 3551 uint64_t zapobj;
3552
3553 mutex_enter(&ds->ds_lock);
3554 if (ds->ds_phys->ds_userrefs_obj == 0) {
3555 /*
3556 * This is the first user hold for this dataset. Create
3557 * the userrefs zap object.
3558 */
3559 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3560 zapobj = ds->ds_phys->ds_userrefs_obj =
3561 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3562 } else {
3563 zapobj = ds->ds_phys->ds_userrefs_obj;
3564 }
3565 ds->ds_userrefs++;
3566 mutex_exit(&ds->ds_lock);
3567
3568 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3569
428870ff
BB		 3570 if (ha->temphold) {
3571 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3572 htag, &now, tx));
3573 }
45d1cae3 3574
428870ff
BB		 3575 spa_history_log_internal(LOG_DS_USER_HOLD,
3576 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3577 (int)ha->temphold, ds->ds_object);
3578}
45d1cae3
BB		 3579
3580static int
428870ff 3581dsl_dataset_user_hold_one(const char *dsname, void *arg)
45d1cae3
BB		 3582{
3583 struct dsl_ds_holdarg *ha = arg;
3584 dsl_dataset_t *ds;
3585 int error;
3586 char *name;
45d1cae3
BB		 3587
3588 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
428870ff 3589 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3590 error = dsl_dataset_hold(name, ha->dstg, &ds);
428870ff 3591 strfree(name);
45d1cae3 3592 if (error == 0) {
428870ff 3593 ha->gotone = B_TRUE;
45d1cae3 3594 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
428870ff 3595 dsl_dataset_user_hold_sync, ds, ha, 0);
45d1cae3
BB		 3596 } else if (error == ENOENT && ha->recursive) {
3597 error = 0;
3598 } else {
428870ff 3599 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
45d1cae3
BB		 3600 }
3601 return (error);
3602}
3603
572e2857
BB		 3604int
3605dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3606 boolean_t temphold)
3607{
3608 struct dsl_ds_holdarg *ha;
3609 int error;
3610
3611 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3612 ha->htag = htag;
3613 ha->temphold = temphold;
3614 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3615 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3616 ds, ha, 0);
3617 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3618
3619 return (error);
3620}
3621
45d1cae3
BB		 3622int
3623dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
572e2857 3624 boolean_t recursive, boolean_t temphold, int cleanup_fd)
45d1cae3
BB		 3625{
3626 struct dsl_ds_holdarg *ha;
3627 dsl_sync_task_t *dst;
3628 spa_t *spa;
3629 int error;
572e2857
BB		 3630 minor_t minor = 0;
3631
3632 if (cleanup_fd != -1) {
3633 /* Currently we only support cleanup-on-exit of tempholds. */
3634 if (!temphold)
3635 return (EINVAL);
3636 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3637 if (error)
3638 return (error);
3639 }
45d1cae3
BB		 3640
3641 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3642
3643 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3644
3645 error = spa_open(dsname, &spa, FTAG);
3646 if (error) {
3647 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
572e2857
BB		 3648 if (cleanup_fd != -1)
3649 zfs_onexit_fd_rele(cleanup_fd);
45d1cae3
BB		 3650 return (error);
3651 }
3652
3653 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3654 ha->htag = htag;
3655 ha->snapname = snapname;
3656 ha->recursive = recursive;
428870ff 3657 ha->temphold = temphold;
572e2857 3658
45d1cae3
BB		 3659 if (recursive) {
3660 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3661 ha, DS_FIND_CHILDREN);
3662 } else {
3663 error = dsl_dataset_user_hold_one(dsname, ha);
3664 }
3665 if (error == 0)
3666 error = dsl_sync_task_group_wait(ha->dstg);
3667
3668 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3669 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3670 dsl_dataset_t *ds = dst->dst_arg1;
3671
3672 if (dst->dst_err) {
3673 dsl_dataset_name(ds, ha->failed);
3674 *strchr(ha->failed, '@') = '\0';
572e2857
BB		 3675 } else if (error == 0 && minor != 0 && temphold) {
3676 /*
3677 * If this hold is to be released upon process exit,
3678 * register that action now.
3679 */
3680 dsl_register_onexit_hold_cleanup(ds, htag, minor);
45d1cae3
BB		 3681 }
3682 dsl_dataset_rele(ds, ha->dstg);
3683 }
3684
428870ff
BB		 3685 if (error == 0 && recursive && !ha->gotone)
3686 error = ENOENT;
3687
45d1cae3 3688 if (error)
428870ff 3689 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
45d1cae3
BB		 3690
3691 dsl_sync_task_group_destroy(ha->dstg);
572e2857 3692
45d1cae3
BB		 3693 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3694 spa_close(spa, FTAG);
572e2857
BB		 3695 if (cleanup_fd != -1)
3696 zfs_onexit_fd_rele(cleanup_fd);
45d1cae3
BB		 3697 return (error);
3698}
3699
3700struct dsl_ds_releasearg {
3701 dsl_dataset_t *ds;
3702 const char *htag;
3703 boolean_t own; /* do we own or just hold ds? */
3704};
3705
3706static int
3707dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3708 boolean_t *might_destroy)
3709{
3710 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3711 uint64_t zapobj;
3712 uint64_t tmp;
3713 int error;
3714
3715 *might_destroy = B_FALSE;
3716
3717 mutex_enter(&ds->ds_lock);
3718 zapobj = ds->ds_phys->ds_userrefs_obj;
3719 if (zapobj == 0) {
3720 /* The tag can't possibly exist */
3721 mutex_exit(&ds->ds_lock);
3722 return (ESRCH);
3723 }
3724
3725 /* Make sure the tag exists */
3726 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3727 if (error) {
3728 mutex_exit(&ds->ds_lock);
3729 if (error == ENOENT)
3730 error = ESRCH;
3731 return (error);
3732 }
3733
3734 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3735 DS_IS_DEFER_DESTROY(ds))
3736 *might_destroy = B_TRUE;
3737
3738 mutex_exit(&ds->ds_lock);
3739 return (0);
3740}
3741
3742static int
3743dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3744{
3745 struct dsl_ds_releasearg *ra = arg1;
3746 dsl_dataset_t *ds = ra->ds;
3747 boolean_t might_destroy;
3748 int error;
3749
3750 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3751 return (ENOTSUP);
3752
3753 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3754 if (error)
3755 return (error);
3756
3757 if (might_destroy) {
3758 struct dsl_ds_destroyarg dsda = {0};
3759
3760 if (dmu_tx_is_syncing(tx)) {
3761 /*
3762 * If we're not prepared to remove the snapshot,
3763 * we can't allow the release to happen right now.
3764 */
3765 if (!ra->own)
3766 return (EBUSY);
45d1cae3
BB		 3767 }
3768 dsda.ds = ds;
3769 dsda.releasing = B_TRUE;
3770 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3771 }
3772
3773 return (0);
3774}
3775
3776static void
428870ff 3777dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
45d1cae3
BB		 3778{
3779 struct dsl_ds_releasearg *ra = arg1;
3780 dsl_dataset_t *ds = ra->ds;
428870ff
BB		 3781 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3782 objset_t *mos = dp->dp_meta_objset;
45d1cae3
BB		 3783 uint64_t zapobj;
3784 uint64_t dsobj = ds->ds_object;
3785 uint64_t refs;
428870ff
BB		 3786 int error;
3787
45d1cae3
BB		 3788 mutex_enter(&ds->ds_lock);
3789 ds->ds_userrefs--;
3790 refs = ds->ds_userrefs;
3791 mutex_exit(&ds->ds_lock);
428870ff
BB		 3792 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3793 VERIFY(error == 0 || error == ENOENT);
45d1cae3
BB		 3794 zapobj = ds->ds_phys->ds_userrefs_obj;
3795 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3796 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3797 DS_IS_DEFER_DESTROY(ds)) {
3798 struct dsl_ds_destroyarg dsda = {0};
3799
3800 ASSERT(ra->own);
3801 dsda.ds = ds;
3802 dsda.releasing = B_TRUE;
3803 /* We already did the destroy_check */
428870ff 3804 dsl_dataset_destroy_sync(&dsda, tag, tx);
45d1cae3
BB		 3805 }
3806
428870ff
BB		 3807 spa_history_log_internal(LOG_DS_USER_RELEASE,
3808 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
45d1cae3
BB		 3809 ra->htag, (longlong_t)refs, dsobj);
3810}
3811
3812static int
428870ff 3813dsl_dataset_user_release_one(const char *dsname, void *arg)
45d1cae3
BB		 3814{
3815 struct dsl_ds_holdarg *ha = arg;
3816 struct dsl_ds_releasearg *ra;
3817 dsl_dataset_t *ds;
3818 int error;
3819 void *dtag = ha->dstg;
3820 char *name;
45d1cae3
BB		 3821 boolean_t own = B_FALSE;
3822 boolean_t might_destroy;
3823
45d1cae3 3824 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
428870ff 3825 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3826 error = dsl_dataset_hold(name, dtag, &ds);
428870ff 3827 strfree(name);
45d1cae3
BB		 3828 if (error == ENOENT && ha->recursive)
3829 return (0);
428870ff 3830 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
45d1cae3
BB		 3831 if (error)
3832 return (error);
3833
428870ff
BB		 3834 ha->gotone = B_TRUE;
3835
45d1cae3
BB		 3836 ASSERT(dsl_dataset_is_snapshot(ds));
3837
3838 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3839 if (error) {
3840 dsl_dataset_rele(ds, dtag);
3841 return (error);
3842 }
3843
3844 if (might_destroy) {
3845#ifdef _KERNEL
428870ff 3846 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3847 error = zfs_unmount_snap(name, NULL);
428870ff 3848 strfree(name);
45d1cae3
BB		 3849 if (error) {
3850 dsl_dataset_rele(ds, dtag);
3851 return (error);
3852 }
3853#endif
428870ff 3854 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
45d1cae3
BB		 3855 dsl_dataset_rele(ds, dtag);
3856 return (EBUSY);
3857 } else {
3858 own = B_TRUE;
3859 dsl_dataset_make_exclusive(ds, dtag);
3860 }
3861 }
3862
3863 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3864 ra->ds = ds;
3865 ra->htag = ha->htag;
3866 ra->own = own;
3867 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3868 dsl_dataset_user_release_sync, ra, dtag, 0);
3869
3870 return (0);
3871}
3872
3873int
3874dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3875 boolean_t recursive)
3876{
3877 struct dsl_ds_holdarg *ha;
3878 dsl_sync_task_t *dst;
3879 spa_t *spa;
3880 int error;
3881
428870ff 3882top:
45d1cae3
BB		 3883 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3884
3885 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3886
3887 error = spa_open(dsname, &spa, FTAG);
3888 if (error) {
3889 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3890 return (error);
3891 }
3892
3893 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3894 ha->htag = htag;
3895 ha->snapname = snapname;
3896 ha->recursive = recursive;
3897 if (recursive) {
3898 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3899 ha, DS_FIND_CHILDREN);
3900 } else {
3901 error = dsl_dataset_user_release_one(dsname, ha);
3902 }
3903 if (error == 0)
3904 error = dsl_sync_task_group_wait(ha->dstg);
3905
3906 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3907 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3908 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3909 dsl_dataset_t *ds = ra->ds;
3910
3911 if (dst->dst_err)
3912 dsl_dataset_name(ds, ha->failed);
3913
3914 if (ra->own)
3915 dsl_dataset_disown(ds, ha->dstg);
3916 else
3917 dsl_dataset_rele(ds, ha->dstg);
3918
3919 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3920 }
3921
428870ff
BB		 3922 if (error == 0 && recursive && !ha->gotone)
3923 error = ENOENT;
3924
3925 if (error && error != EBUSY)
3926 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
45d1cae3
BB		 3927
3928 dsl_sync_task_group_destroy(ha->dstg);
3929 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3930 spa_close(spa, FTAG);
428870ff
BB		 3931
3932 /*
3933 * We can get EBUSY if we were racing with deferred destroy and
3934 * dsl_dataset_user_release_check() hadn't done the necessary
3935 * open context setup. We can also get EBUSY if we're racing
3936 * with destroy and that thread is the ds_owner. Either way
3937 * the busy condition should be transient, and we should retry
3938 * the release operation.
3939 */
3940 if (error == EBUSY)
3941 goto top;
3942
45d1cae3
BB		 3943 return (error);
3944}
3945
428870ff 3946/*
572e2857
BB		 3947 * Called at spa_load time (with retry == B_FALSE) to release a stale
3948 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
428870ff
BB		 3949 */
3950int
572e2857
BB		 3951dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
3952 boolean_t retry)
428870ff
BB		 3953{
3954 dsl_dataset_t *ds;
3955 char *snap;
3956 char *name;
3957 int namelen;
3958 int error;
3959
572e2857
BB		 3960 do {
3961 rw_enter(&dp->dp_config_rwlock, RW_READER);
3962 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3963 rw_exit(&dp->dp_config_rwlock);
3964 if (error)
3965 return (error);
3966 namelen = dsl_dataset_namelen(ds)+1;
3967 name = kmem_alloc(namelen, KM_SLEEP);
3968 dsl_dataset_name(ds, name);
3969 dsl_dataset_rele(ds, FTAG);
428870ff 3970
572e2857
BB		 3971 snap = strchr(name, '@');
3972 *snap = '\0';
3973 ++snap;
3974 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
3975 kmem_free(name, namelen);
3976
3977 /*
3978 * The object can't have been destroyed because we have a hold,
3979 * but it might have been renamed, resulting in ENOENT. Retry
3980 * if we've been requested to do so.
3981 *
3982 * It would be nice if we could use the dsobj all the way
3983 * through and avoid ENOENT entirely. But we might need to
3984 * unmount the snapshot, and there's currently no way to lookup
3985 * a vfsp using a ZFS object id.
3986 */
3987 } while ((error == ENOENT) && retry);
3988
3989 return (error);
428870ff
BB		 3990}
3991
45d1cae3
BB		 3992int
3993dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3994{
3995 dsl_dataset_t *ds;
3996 int err;
3997
3998 err = dsl_dataset_hold(dsname, FTAG, &ds);
3999 if (err)
4000 return (err);
4001
4002 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
4003 if (ds->ds_phys->ds_userrefs_obj != 0) {
4004 zap_attribute_t *za;
4005 zap_cursor_t zc;
4006
4007 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4008 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4009 ds->ds_phys->ds_userrefs_obj);
4010 zap_cursor_retrieve(&zc, za) == 0;
4011 zap_cursor_advance(&zc)) {
4012 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4013 za->za_first_integer));
4014 }
4015 zap_cursor_fini(&zc);
4016 kmem_free(za, sizeof (zap_attribute_t));
4017 }
4018 dsl_dataset_rele(ds, FTAG);
4019 return (0);
4020}
428870ff
BB		 4021
4022/*
4023 * Note, this fuction is used as the callback for dmu_objset_find(). We
4024 * always return 0 so that we will continue to find and process
4025 * inconsistent datasets, even if we encounter an error trying to
4026 * process one of them.
4027 */
4028/* ARGSUSED */
4029int
4030dsl_destroy_inconsistent(const char *dsname, void *arg)
4031{
4032 dsl_dataset_t *ds;
4033
4034 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4035 if (DS_IS_INCONSISTENT(ds))
4036 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4037 else
4038 dsl_dataset_disown(ds, FTAG);
4039 }
4040 return (0);
4041}
c28b2279
BB		 4042
4043#if defined(_KERNEL) && defined(HAVE_SPL)
4044EXPORT_SYMBOL(dsl_dataset_hold);
4045EXPORT_SYMBOL(dsl_dataset_hold_obj);
4046EXPORT_SYMBOL(dsl_dataset_own);
4047EXPORT_SYMBOL(dsl_dataset_own_obj);
4048EXPORT_SYMBOL(dsl_dataset_name);
4049EXPORT_SYMBOL(dsl_dataset_rele);
4050EXPORT_SYMBOL(dsl_dataset_disown);
4051EXPORT_SYMBOL(dsl_dataset_drop_ref);
4052EXPORT_SYMBOL(dsl_dataset_tryown);
4053EXPORT_SYMBOL(dsl_dataset_make_exclusive);
4054EXPORT_SYMBOL(dsl_dataset_create_sync);
4055EXPORT_SYMBOL(dsl_dataset_create_sync_dd);
4056EXPORT_SYMBOL(dsl_dataset_destroy);
4057EXPORT_SYMBOL(dsl_snapshots_destroy);
4058EXPORT_SYMBOL(dsl_dataset_destroy_check);
4059EXPORT_SYMBOL(dsl_dataset_destroy_sync);
4060EXPORT_SYMBOL(dsl_dataset_snapshot_check);
4061EXPORT_SYMBOL(dsl_dataset_snapshot_sync);
4062EXPORT_SYMBOL(dsl_dataset_rename);
4063EXPORT_SYMBOL(dsl_dataset_promote);
4064EXPORT_SYMBOL(dsl_dataset_clone_swap);
4065EXPORT_SYMBOL(dsl_dataset_user_hold);
4066EXPORT_SYMBOL(dsl_dataset_user_release);
4067EXPORT_SYMBOL(dsl_dataset_user_release_tmp);
4068EXPORT_SYMBOL(dsl_dataset_get_holds);
4069EXPORT_SYMBOL(dsl_dataset_get_blkptr);
4070EXPORT_SYMBOL(dsl_dataset_set_blkptr);
4071EXPORT_SYMBOL(dsl_dataset_get_spa);
4072EXPORT_SYMBOL(dsl_dataset_modified_since_lastsnap);
4073EXPORT_SYMBOL(dsl_dataset_sync);
4074EXPORT_SYMBOL(dsl_dataset_block_born);
4075EXPORT_SYMBOL(dsl_dataset_block_kill);
4076EXPORT_SYMBOL(dsl_dataset_block_freeable);
4077EXPORT_SYMBOL(dsl_dataset_prev_snap_txg);
4078EXPORT_SYMBOL(dsl_dataset_dirty);
4079EXPORT_SYMBOL(dsl_dataset_stats);
4080EXPORT_SYMBOL(dsl_dataset_fast_stat);
4081EXPORT_SYMBOL(dsl_dataset_space);
4082EXPORT_SYMBOL(dsl_dataset_fsid_guid);
4083EXPORT_SYMBOL(dsl_dsobj_to_dsname);
4084EXPORT_SYMBOL(dsl_dataset_check_quota);
4085EXPORT_SYMBOL(dsl_dataset_set_quota);
4086EXPORT_SYMBOL(dsl_dataset_set_quota_sync);
4087EXPORT_SYMBOL(dsl_dataset_set_reservation);
4088EXPORT_SYMBOL(dsl_destroy_inconsistent);
4089#endif
ZFS On Linux mirror for PVE