]> 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
Fix gcc c90 compliance warnings
[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) {
390 dsl_dataset_t *winner;
391
392 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
393 ds->ds_dbuf = dbuf;
394 ds->ds_object = dsobj;
395 ds->ds_phys = dbuf->db_data;
396
397 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
45d1cae3 398 mutex_init(&ds->ds_recvlock, NULL, MUTEX_DEFAULT, NULL);
34dc7c2f 399 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
b128c09f
BB		 400 rw_init(&ds->ds_rwlock, 0, 0, 0);
401 cv_init(&ds->ds_exclusive_cv, NULL, CV_DEFAULT, NULL);
34dc7c2f 402
428870ff
BB		 403 bplist_create(&ds->ds_pending_deadlist);
404 dsl_deadlist_open(&ds->ds_deadlist,
34dc7c2f 405 mos, ds->ds_phys->ds_deadlist_obj);
428870ff 406
34dc7c2f
BB		 407 if (err == 0) {
408 err = dsl_dir_open_obj(dp,
409 ds->ds_phys->ds_dir_obj, NULL, ds, &ds->ds_dir);
410 }
411 if (err) {
34dc7c2f 412 mutex_destroy(&ds->ds_lock);
45d1cae3 413 mutex_destroy(&ds->ds_recvlock);
34dc7c2f 414 mutex_destroy(&ds->ds_opening_lock);
b128c09f
BB		 415 rw_destroy(&ds->ds_rwlock);
416 cv_destroy(&ds->ds_exclusive_cv);
428870ff
BB		 417 bplist_destroy(&ds->ds_pending_deadlist);
418 dsl_deadlist_close(&ds->ds_deadlist);
34dc7c2f
BB		 419 kmem_free(ds, sizeof (dsl_dataset_t));
420 dmu_buf_rele(dbuf, tag);
421 return (err);
422 }
423
b128c09f 424 if (!dsl_dataset_is_snapshot(ds)) {
34dc7c2f
BB		 425 ds->ds_snapname[0] = '\0';
426 if (ds->ds_phys->ds_prev_snap_obj) {
b128c09f
BB		 427 err = dsl_dataset_get_ref(dp,
428 ds->ds_phys->ds_prev_snap_obj,
429 ds, &ds->ds_prev);
34dc7c2f 430 }
45d1cae3
BB		 431 } else {
432 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
433 err = dsl_dataset_get_snapname(ds);
434 if (err == 0 && ds->ds_phys->ds_userrefs_obj != 0) {
435 err = zap_count(
436 ds->ds_dir->dd_pool->dp_meta_objset,
437 ds->ds_phys->ds_userrefs_obj,
438 &ds->ds_userrefs);
439 }
34dc7c2f
BB		 440 }
441
b128c09f 442 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
34dc7c2f
BB		 443 /*
444 * In sync context, we're called with either no lock
445 * or with the write lock. If we're not syncing,
446 * we're always called with the read lock held.
447 */
448 boolean_t need_lock =
449 !RW_WRITE_HELD(&dp->dp_config_rwlock) &&
450 dsl_pool_sync_context(dp);
451
452 if (need_lock)
453 rw_enter(&dp->dp_config_rwlock, RW_READER);
454
b128c09f 455 err = dsl_prop_get_ds(ds,
34dc7c2f
BB		 456 "refreservation", sizeof (uint64_t), 1,
457 &ds->ds_reserved, NULL);
458 if (err == 0) {
b128c09f 459 err = dsl_prop_get_ds(ds,
34dc7c2f
BB		 460 "refquota", sizeof (uint64_t), 1,
461 &ds->ds_quota, NULL);
462 }
463
464 if (need_lock)
465 rw_exit(&dp->dp_config_rwlock);
466 } else {
467 ds->ds_reserved = ds->ds_quota = 0;
468 }
469
470 if (err == 0) {
471 winner = dmu_buf_set_user_ie(dbuf, ds, &ds->ds_phys,
472 dsl_dataset_evict);
473 }
474 if (err || winner) {
428870ff
BB		 475 bplist_destroy(&ds->ds_pending_deadlist);
476 dsl_deadlist_close(&ds->ds_deadlist);
b128c09f
BB		 477 if (ds->ds_prev)
478 dsl_dataset_drop_ref(ds->ds_prev, ds);
34dc7c2f
BB		 479 dsl_dir_close(ds->ds_dir, ds);
480 mutex_destroy(&ds->ds_lock);
45d1cae3 481 mutex_destroy(&ds->ds_recvlock);
34dc7c2f 482 mutex_destroy(&ds->ds_opening_lock);
b128c09f
BB		 483 rw_destroy(&ds->ds_rwlock);
484 cv_destroy(&ds->ds_exclusive_cv);
34dc7c2f
BB		 485 kmem_free(ds, sizeof (dsl_dataset_t));
486 if (err) {
487 dmu_buf_rele(dbuf, tag);
488 return (err);
489 }
490 ds = winner;
491 } else {
492 ds->ds_fsid_guid =
493 unique_insert(ds->ds_phys->ds_fsid_guid);
494 }
495 }
496 ASSERT3P(ds->ds_dbuf, ==, dbuf);
497 ASSERT3P(ds->ds_phys, ==, dbuf->db_data);
b128c09f
BB		 498 ASSERT(ds->ds_phys->ds_prev_snap_obj != 0 ||
499 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
500 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
34dc7c2f 501 mutex_enter(&ds->ds_lock);
b128c09f 502 if (!dsl_pool_sync_context(dp) && DSL_DATASET_IS_DESTROYED(ds)) {
34dc7c2f 503 mutex_exit(&ds->ds_lock);
b128c09f
BB		 504 dmu_buf_rele(ds->ds_dbuf, tag);
505 return (ENOENT);
34dc7c2f 506 }
34dc7c2f 507 mutex_exit(&ds->ds_lock);
34dc7c2f
BB		 508 *dsp = ds;
509 return (0);
510}
511
b128c09f
BB		 512static int
513dsl_dataset_hold_ref(dsl_dataset_t *ds, void *tag)
514{
515 dsl_pool_t *dp = ds->ds_dir->dd_pool;
516
517 /*
518 * In syncing context we don't want the rwlock lock: there
519 * may be an existing writer waiting for sync phase to
520 * finish. We don't need to worry about such writers, since
521 * sync phase is single-threaded, so the writer can't be
522 * doing anything while we are active.
523 */
524 if (dsl_pool_sync_context(dp)) {
525 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
526 return (0);
527 }
528
529 /*
530 * Normal users will hold the ds_rwlock as a READER until they
531 * are finished (i.e., call dsl_dataset_rele()). "Owners" will
532 * drop their READER lock after they set the ds_owner field.
533 *
534 * If the dataset is being destroyed, the destroy thread will
535 * obtain a WRITER lock for exclusive access after it's done its
536 * open-context work and then change the ds_owner to
537 * dsl_reaper once destruction is assured. So threads
538 * may block here temporarily, until the "destructability" of
539 * the dataset is determined.
540 */
541 ASSERT(!RW_WRITE_HELD(&dp->dp_config_rwlock));
542 mutex_enter(&ds->ds_lock);
543 while (!rw_tryenter(&ds->ds_rwlock, RW_READER)) {
544 rw_exit(&dp->dp_config_rwlock);
545 cv_wait(&ds->ds_exclusive_cv, &ds->ds_lock);
546 if (DSL_DATASET_IS_DESTROYED(ds)) {
547 mutex_exit(&ds->ds_lock);
548 dsl_dataset_drop_ref(ds, tag);
549 rw_enter(&dp->dp_config_rwlock, RW_READER);
550 return (ENOENT);
551 }
428870ff
BB		 552 /*
553 * The dp_config_rwlock lives above the ds_lock. And
554 * we need to check DSL_DATASET_IS_DESTROYED() while
555 * holding the ds_lock, so we have to drop and reacquire
556 * the ds_lock here.
557 */
558 mutex_exit(&ds->ds_lock);
b128c09f 559 rw_enter(&dp->dp_config_rwlock, RW_READER);
428870ff 560 mutex_enter(&ds->ds_lock);
b128c09f
BB		 561 }
562 mutex_exit(&ds->ds_lock);
563 return (0);
564}
565
34dc7c2f 566int
b128c09f
BB		 567dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
568 dsl_dataset_t **dsp)
569{
570 int err = dsl_dataset_get_ref(dp, dsobj, tag, dsp);
571
572 if (err)
573 return (err);
574 return (dsl_dataset_hold_ref(*dsp, tag));
575}
576
577int
428870ff
BB		 578dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, boolean_t inconsistentok,
579 void *tag, dsl_dataset_t **dsp)
b128c09f 580{
428870ff 581 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
b128c09f
BB		 582 if (err)
583 return (err);
428870ff
BB		 584 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
585 dsl_dataset_rele(*dsp, tag);
9babb374 586 *dsp = NULL;
b128c09f
BB		 587 return (EBUSY);
588 }
589 return (0);
590}
591
592int
593dsl_dataset_hold(const char *name, void *tag, dsl_dataset_t **dsp)
34dc7c2f
BB		 594{
595 dsl_dir_t *dd;
596 dsl_pool_t *dp;
b128c09f 597 const char *snapname;
34dc7c2f 598 uint64_t obj;
34dc7c2f
BB		 599 int err = 0;
600
b128c09f 601 err = dsl_dir_open_spa(NULL, name, FTAG, &dd, &snapname);
34dc7c2f
BB		 602 if (err)
603 return (err);
604
605 dp = dd->dd_pool;
606 obj = dd->dd_phys->dd_head_dataset_obj;
607 rw_enter(&dp->dp_config_rwlock, RW_READER);
b128c09f
BB		 608 if (obj)
609 err = dsl_dataset_get_ref(dp, obj, tag, dsp);
610 else
34dc7c2f 611 err = ENOENT;
b128c09f 612 if (err)
34dc7c2f 613 goto out;
34dc7c2f 614
b128c09f 615 err = dsl_dataset_hold_ref(*dsp, tag);
34dc7c2f 616
b128c09f
BB		 617 /* we may be looking for a snapshot */
618 if (err == 0 && snapname != NULL) {
619 dsl_dataset_t *ds = NULL;
34dc7c2f 620
b128c09f
BB		 621 if (*snapname++ != '@') {
622 dsl_dataset_rele(*dsp, tag);
34dc7c2f
BB		 623 err = ENOENT;
624 goto out;
625 }
34dc7c2f 626
b128c09f
BB		 627 dprintf("looking for snapshot '%s'\n", snapname);
628 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
629 if (err == 0)
630 err = dsl_dataset_get_ref(dp, obj, tag, &ds);
631 dsl_dataset_rele(*dsp, tag);
632
633 ASSERT3U((err == 0), ==, (ds != NULL));
634
635 if (ds) {
636 mutex_enter(&ds->ds_lock);
637 if (ds->ds_snapname[0] == 0)
638 (void) strlcpy(ds->ds_snapname, snapname,
639 sizeof (ds->ds_snapname));
640 mutex_exit(&ds->ds_lock);
641 err = dsl_dataset_hold_ref(ds, tag);
642 *dsp = err ? NULL : ds;
34dc7c2f 643 }
34dc7c2f 644 }
34dc7c2f
BB		 645out:
646 rw_exit(&dp->dp_config_rwlock);
647 dsl_dir_close(dd, FTAG);
34dc7c2f
BB		 648 return (err);
649}
650
651int
428870ff
BB		 652dsl_dataset_own(const char *name, boolean_t inconsistentok,
653 void *tag, dsl_dataset_t **dsp)
34dc7c2f 654{
428870ff 655 int err = dsl_dataset_hold(name, tag, dsp);
b128c09f
BB		 656 if (err)
657 return (err);
428870ff
BB		 658 if (!dsl_dataset_tryown(*dsp, inconsistentok, tag)) {
659 dsl_dataset_rele(*dsp, tag);
b128c09f
BB		 660 return (EBUSY);
661 }
662 return (0);
34dc7c2f
BB		 663}
664
665void
666dsl_dataset_name(dsl_dataset_t *ds, char *name)
667{
668 if (ds == NULL) {
669 (void) strcpy(name, "mos");
670 } else {
671 dsl_dir_name(ds->ds_dir, name);
672 VERIFY(0 == dsl_dataset_get_snapname(ds));
673 if (ds->ds_snapname[0]) {
674 (void) strcat(name, "@");
b128c09f
BB		 675 /*
676 * We use a "recursive" mutex so that we
677 * can call dprintf_ds() with ds_lock held.
678 */
34dc7c2f 679 if (!MUTEX_HELD(&ds->ds_lock)) {
34dc7c2f
BB		 680 mutex_enter(&ds->ds_lock);
681 (void) strcat(name, ds->ds_snapname);
682 mutex_exit(&ds->ds_lock);
683 } else {
684 (void) strcat(name, ds->ds_snapname);
685 }
686 }
687 }
688}
689
690static int
691dsl_dataset_namelen(dsl_dataset_t *ds)
692{
693 int result;
694
695 if (ds == NULL) {
696 result = 3; /* "mos" */
697 } else {
698 result = dsl_dir_namelen(ds->ds_dir);
699 VERIFY(0 == dsl_dataset_get_snapname(ds));
700 if (ds->ds_snapname[0]) {
701 ++result; /* adding one for the @-sign */
702 if (!MUTEX_HELD(&ds->ds_lock)) {
34dc7c2f
BB		 703 mutex_enter(&ds->ds_lock);
704 result += strlen(ds->ds_snapname);
705 mutex_exit(&ds->ds_lock);
706 } else {
707 result += strlen(ds->ds_snapname);
708 }
709 }
710 }
711
712 return (result);
713}
714
715void
b128c09f 716dsl_dataset_drop_ref(dsl_dataset_t *ds, void *tag)
34dc7c2f 717{
34dc7c2f
BB		 718 dmu_buf_rele(ds->ds_dbuf, tag);
719}
720
721void
b128c09f
BB		 722dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
723{
724 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool)) {
725 rw_exit(&ds->ds_rwlock);
726 }
727 dsl_dataset_drop_ref(ds, tag);
728}
729
730void
428870ff 731dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
34dc7c2f 732{
428870ff 733 ASSERT((ds->ds_owner == tag && ds->ds_dbuf) ||
b128c09f
BB		 734 (DSL_DATASET_IS_DESTROYED(ds) && ds->ds_dbuf == NULL));
735
34dc7c2f 736 mutex_enter(&ds->ds_lock);
b128c09f
BB		 737 ds->ds_owner = NULL;
738 if (RW_WRITE_HELD(&ds->ds_rwlock)) {
739 rw_exit(&ds->ds_rwlock);
740 cv_broadcast(&ds->ds_exclusive_cv);
741 }
34dc7c2f 742 mutex_exit(&ds->ds_lock);
b128c09f 743 if (ds->ds_dbuf)
428870ff 744 dsl_dataset_drop_ref(ds, tag);
b128c09f 745 else
428870ff 746 dsl_dataset_evict(NULL, ds);
34dc7c2f
BB		 747}
748
749boolean_t
428870ff 750dsl_dataset_tryown(dsl_dataset_t *ds, boolean_t inconsistentok, void *tag)
34dc7c2f 751{
b128c09f
BB		 752 boolean_t gotit = FALSE;
753
34dc7c2f 754 mutex_enter(&ds->ds_lock);
b128c09f
BB		 755 if (ds->ds_owner == NULL &&
756 (!DS_IS_INCONSISTENT(ds) || inconsistentok)) {
428870ff 757 ds->ds_owner = tag;
b128c09f
BB		 758 if (!dsl_pool_sync_context(ds->ds_dir->dd_pool))
759 rw_exit(&ds->ds_rwlock);
760 gotit = TRUE;
34dc7c2f
BB		 761 }
762 mutex_exit(&ds->ds_lock);
b128c09f 763 return (gotit);
34dc7c2f
BB		 764}
765
766void
b128c09f 767dsl_dataset_make_exclusive(dsl_dataset_t *ds, void *owner)
34dc7c2f 768{
b128c09f
BB		 769 ASSERT3P(owner, ==, ds->ds_owner);
770 if (!RW_WRITE_HELD(&ds->ds_rwlock))
771 rw_enter(&ds->ds_rwlock, RW_WRITER);
34dc7c2f
BB		 772}
773
774uint64_t
b128c09f 775dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
34dc7c2f
BB		 776 uint64_t flags, dmu_tx_t *tx)
777{
778 dsl_pool_t *dp = dd->dd_pool;
779 dmu_buf_t *dbuf;
780 dsl_dataset_phys_t *dsphys;
781 uint64_t dsobj;
782 objset_t *mos = dp->dp_meta_objset;
783
b128c09f
BB		 784 if (origin == NULL)
785 origin = dp->dp_origin_snap;
786
34dc7c2f
BB		 787 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
788 ASSERT(origin == NULL || origin->ds_phys->ds_num_children > 0);
789 ASSERT(dmu_tx_is_syncing(tx));
790 ASSERT(dd->dd_phys->dd_head_dataset_obj == 0);
791
792 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
793 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
794 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
795 dmu_buf_will_dirty(dbuf, tx);
796 dsphys = dbuf->db_data;
b128c09f 797 bzero(dsphys, sizeof (dsl_dataset_phys_t));
34dc7c2f
BB		 798 dsphys->ds_dir_obj = dd->dd_object;
799 dsphys->ds_flags = flags;
800 dsphys->ds_fsid_guid = unique_create();
801 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
802 sizeof (dsphys->ds_guid));
803 dsphys->ds_snapnames_zapobj =
804 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
805 DMU_OT_NONE, 0, tx);
806 dsphys->ds_creation_time = gethrestime_sec();
b128c09f 807 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
34dc7c2f 808
428870ff
BB		 809 if (origin == NULL) {
810 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
811 } else {
812 dsl_dataset_t *ohds;
813
34dc7c2f
BB		 814 dsphys->ds_prev_snap_obj = origin->ds_object;
815 dsphys->ds_prev_snap_txg =
816 origin->ds_phys->ds_creation_txg;
817 dsphys->ds_used_bytes =
818 origin->ds_phys->ds_used_bytes;
819 dsphys->ds_compressed_bytes =
820 origin->ds_phys->ds_compressed_bytes;
821 dsphys->ds_uncompressed_bytes =
822 origin->ds_phys->ds_uncompressed_bytes;
823 dsphys->ds_bp = origin->ds_phys->ds_bp;
824 dsphys->ds_flags |= origin->ds_phys->ds_flags;
825
826 dmu_buf_will_dirty(origin->ds_dbuf, tx);
827 origin->ds_phys->ds_num_children++;
828
428870ff
BB		 829 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
830 origin->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ohds));
831 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
832 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
833 dsl_dataset_rele(ohds, FTAG);
834
b128c09f
BB		 835 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
836 if (origin->ds_phys->ds_next_clones_obj == 0) {
837 origin->ds_phys->ds_next_clones_obj =
838 zap_create(mos,
839 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
840 }
841 VERIFY(0 == zap_add_int(mos,
842 origin->ds_phys->ds_next_clones_obj,
843 dsobj, tx));
844 }
845
34dc7c2f
BB		 846 dmu_buf_will_dirty(dd->dd_dbuf, tx);
847 dd->dd_phys->dd_origin_obj = origin->ds_object;
428870ff
BB		 848 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
849 if (origin->ds_dir->dd_phys->dd_clones == 0) {
850 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
851 origin->ds_dir->dd_phys->dd_clones =
852 zap_create(mos,
853 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
854 }
855 VERIFY3U(0, ==, zap_add_int(mos,
856 origin->ds_dir->dd_phys->dd_clones, dsobj, tx));
857 }
34dc7c2f
BB		 858 }
859
860 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
861 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
862
863 dmu_buf_rele(dbuf, FTAG);
864
865 dmu_buf_will_dirty(dd->dd_dbuf, tx);
866 dd->dd_phys->dd_head_dataset_obj = dsobj;
867
868 return (dsobj);
869}
870
871uint64_t
872dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
873 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
874{
875 dsl_pool_t *dp = pdd->dd_pool;
876 uint64_t dsobj, ddobj;
877 dsl_dir_t *dd;
878
879 ASSERT(lastname[0] != '@');
880
b128c09f 881 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
34dc7c2f
BB		 882 VERIFY(0 == dsl_dir_open_obj(dp, ddobj, lastname, FTAG, &dd));
883
b128c09f 884 dsobj = dsl_dataset_create_sync_dd(dd, origin, flags, tx);
34dc7c2f
BB		 885
886 dsl_deleg_set_create_perms(dd, tx, cr);
887
888 dsl_dir_close(dd, FTAG);
889
572e2857
BB		 890 /*
891 * If we are creating a clone, make sure we zero out any stale
892 * data from the origin snapshots zil header.
893 */
894 if (origin != NULL) {
895 dsl_dataset_t *ds;
896 objset_t *os;
897
898 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
899 VERIFY3U(0, ==, dmu_objset_from_ds(ds, &os));
900 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
901 dsl_dataset_dirty(ds, tx);
902 dsl_dataset_rele(ds, FTAG);
903 }
904
34dc7c2f
BB		 905 return (dsobj);
906}
907
908struct destroyarg {
909 dsl_sync_task_group_t *dstg;
910 char *snapname;
911 char *failed;
45d1cae3 912 boolean_t defer;
34dc7c2f
BB		 913};
914
915static int
428870ff 916dsl_snapshot_destroy_one(const char *name, void *arg)
34dc7c2f
BB		 917{
918 struct destroyarg *da = arg;
919 dsl_dataset_t *ds;
34dc7c2f 920 int err;
45d1cae3 921 char *dsname;
428870ff
BB		 922
923 dsname = kmem_asprintf("%s@%s", name, da->snapname);
924 err = dsl_dataset_own(dsname, B_TRUE, da->dstg, &ds);
925 strfree(dsname);
b128c09f 926 if (err == 0) {
45d1cae3
BB		 927 struct dsl_ds_destroyarg *dsda;
928
b128c09f 929 dsl_dataset_make_exclusive(ds, da->dstg);
45d1cae3
BB		 930 dsda = kmem_zalloc(sizeof (struct dsl_ds_destroyarg), KM_SLEEP);
931 dsda->ds = ds;
932 dsda->defer = da->defer;
b128c09f 933 dsl_sync_task_create(da->dstg, dsl_dataset_destroy_check,
45d1cae3 934 dsl_dataset_destroy_sync, dsda, da->dstg, 0);
b128c09f
BB		 935 } else if (err == ENOENT) {
936 err = 0;
937 } else {
34dc7c2f 938 (void) strcpy(da->failed, name);
34dc7c2f 939 }
b128c09f 940 return (err);
34dc7c2f
BB		 941}
942
943/*
944 * Destroy 'snapname' in all descendants of 'fsname'.
945 */
946#pragma weak dmu_snapshots_destroy = dsl_snapshots_destroy
947int
45d1cae3 948dsl_snapshots_destroy(char *fsname, char *snapname, boolean_t defer)
34dc7c2f
BB		 949{
950 int err;
951 struct destroyarg da;
952 dsl_sync_task_t *dst;
953 spa_t *spa;
954
955 err = spa_open(fsname, &spa, FTAG);
956 if (err)
957 return (err);
958 da.dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
959 da.snapname = snapname;
960 da.failed = fsname;
45d1cae3 961 da.defer = defer;
34dc7c2f
BB		 962
963 err = dmu_objset_find(fsname,
964 dsl_snapshot_destroy_one, &da, DS_FIND_CHILDREN);
965
966 if (err == 0)
967 err = dsl_sync_task_group_wait(da.dstg);
968
969 for (dst = list_head(&da.dstg->dstg_tasks); dst;
970 dst = list_next(&da.dstg->dstg_tasks, dst)) {
45d1cae3
BB		 971 struct dsl_ds_destroyarg *dsda = dst->dst_arg1;
972 dsl_dataset_t *ds = dsda->ds;
973
b128c09f
BB		 974 /*
975 * Return the file system name that triggered the error
976 */
34dc7c2f
BB		 977 if (dst->dst_err) {
978 dsl_dataset_name(ds, fsname);
979 *strchr(fsname, '@') = '\0';
980 }
45d1cae3 981 ASSERT3P(dsda->rm_origin, ==, NULL);
b128c09f 982 dsl_dataset_disown(ds, da.dstg);
45d1cae3 983 kmem_free(dsda, sizeof (struct dsl_ds_destroyarg));
34dc7c2f
BB		 984 }
985
986 dsl_sync_task_group_destroy(da.dstg);
987 spa_close(spa, FTAG);
988 return (err);
989}
990
45d1cae3
BB		 991static boolean_t
992dsl_dataset_might_destroy_origin(dsl_dataset_t *ds)
993{
994 boolean_t might_destroy = B_FALSE;
995
996 mutex_enter(&ds->ds_lock);
997 if (ds->ds_phys->ds_num_children == 2 && ds->ds_userrefs == 0 &&
998 DS_IS_DEFER_DESTROY(ds))
999 might_destroy = B_TRUE;
1000 mutex_exit(&ds->ds_lock);
1001
1002 return (might_destroy);
1003}
1004
45d1cae3
BB		 1005/*
1006 * If we're removing a clone, and these three conditions are true:
1007 * 1) the clone's origin has no other children
1008 * 2) the clone's origin has no user references
1009 * 3) the clone's origin has been marked for deferred destruction
1010 * Then, prepare to remove the origin as part of this sync task group.
1011 */
1012static int
1013dsl_dataset_origin_rm_prep(struct dsl_ds_destroyarg *dsda, void *tag)
1014{
1015 dsl_dataset_t *ds = dsda->ds;
1016 dsl_dataset_t *origin = ds->ds_prev;
1017
1018 if (dsl_dataset_might_destroy_origin(origin)) {
1019 char *name;
1020 int namelen;
1021 int error;
1022
1023 namelen = dsl_dataset_namelen(origin) + 1;
1024 name = kmem_alloc(namelen, KM_SLEEP);
1025 dsl_dataset_name(origin, name);
1026#ifdef _KERNEL
1027 error = zfs_unmount_snap(name, NULL);
1028 if (error) {
1029 kmem_free(name, namelen);
1030 return (error);
1031 }
45d1cae3 1032#endif
428870ff 1033 error = dsl_dataset_own(name, B_TRUE, tag, &origin);
45d1cae3
BB		 1034 kmem_free(name, namelen);
1035 if (error)
1036 return (error);
1037 dsda->rm_origin = origin;
1038 dsl_dataset_make_exclusive(origin, tag);
1039 }
1040
1041 return (0);
1042}
1043
34dc7c2f 1044/*
b128c09f
BB		 1045 * ds must be opened as OWNER. On return (whether successful or not),
1046 * ds will be closed and caller can no longer dereference it.
34dc7c2f
BB		 1047 */
1048int
45d1cae3 1049dsl_dataset_destroy(dsl_dataset_t *ds, void *tag, boolean_t defer)
34dc7c2f
BB		 1050{
1051 int err;
1052 dsl_sync_task_group_t *dstg;
1053 objset_t *os;
1054 dsl_dir_t *dd;
1055 uint64_t obj;
428870ff
BB		 1056 struct dsl_ds_destroyarg dsda = { 0 };
1057 dsl_dataset_t dummy_ds = { 0 };
45d1cae3
BB		 1058
1059 dsda.ds = ds;
34dc7c2f 1060
34dc7c2f
BB		 1061 if (dsl_dataset_is_snapshot(ds)) {
1062 /* Destroying a snapshot is simpler */
b128c09f
BB		 1063 dsl_dataset_make_exclusive(ds, tag);
1064
45d1cae3 1065 dsda.defer = defer;
34dc7c2f
BB		 1066 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
1067 dsl_dataset_destroy_check, dsl_dataset_destroy_sync,
45d1cae3
BB		 1068 &dsda, tag, 0);
1069 ASSERT3P(dsda.rm_origin, ==, NULL);
34dc7c2f 1070 goto out;
428870ff
BB		 1071 } else if (defer) {
1072 err = EINVAL;
1073 goto out;
34dc7c2f
BB		 1074 }
1075
1076 dd = ds->ds_dir;
428870ff
BB		 1077 dummy_ds.ds_dir = dd;
1078 dummy_ds.ds_object = ds->ds_object;
34dc7c2f
BB		 1079
1080 /*
1081 * Check for errors and mark this ds as inconsistent, in
1082 * case we crash while freeing the objects.
1083 */
1084 err = dsl_sync_task_do(dd->dd_pool, dsl_dataset_destroy_begin_check,
1085 dsl_dataset_destroy_begin_sync, ds, NULL, 0);
1086 if (err)
1087 goto out;
1088
428870ff 1089 err = dmu_objset_from_ds(ds, &os);
34dc7c2f
BB		 1090 if (err)
1091 goto out;
1092
1093 /*
1094 * remove the objects in open context, so that we won't
1095 * have too much to do in syncing context.
1096 */
1097 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE,
1098 ds->ds_phys->ds_prev_snap_txg)) {
b128c09f
BB		 1099 /*
1100 * Ignore errors, if there is not enough disk space
1101 * we will deal with it in dsl_dataset_destroy_sync().
1102 */
1103 (void) dmu_free_object(os, obj);
34dc7c2f 1104 }
572e2857
BB		 1105 if (err != ESRCH)
1106 goto out;
34dc7c2f 1107
9babb374 1108 /*
572e2857
BB		 1109 * Only the ZIL knows how to free log blocks.
1110 */
1111 zil_destroy(dmu_objset_zil(os), B_FALSE);
1112
1113 /*
1114 * Sync out all in-flight IO.
9babb374
BB		 1115 */
1116 txg_wait_synced(dd->dd_pool, 0);
1117
1118 /*
1119 * If we managed to free all the objects in open
1120 * context, the user space accounting should be zero.
1121 */
1122 if (ds->ds_phys->ds_bp.blk_fill == 0 &&
428870ff 1123 dmu_objset_userused_enabled(os)) {
9babb374
BB		 1124 uint64_t count;
1125
1126 ASSERT(zap_count(os, DMU_USERUSED_OBJECT, &count) != 0 ||
1127 count == 0);
1128 ASSERT(zap_count(os, DMU_GROUPUSED_OBJECT, &count) != 0 ||
1129 count == 0);
1130 }
1131
34dc7c2f
BB		 1132 rw_enter(&dd->dd_pool->dp_config_rwlock, RW_READER);
1133 err = dsl_dir_open_obj(dd->dd_pool, dd->dd_object, NULL, FTAG, &dd);
1134 rw_exit(&dd->dd_pool->dp_config_rwlock);
1135
1136 if (err)
1137 goto out;
1138
1139 /*
1140 * Blow away the dsl_dir + head dataset.
1141 */
b128c09f 1142 dsl_dataset_make_exclusive(ds, tag);
45d1cae3
BB		 1143 /*
1144 * If we're removing a clone, we might also need to remove its
1145 * origin.
1146 */
1147 do {
1148 dsda.need_prep = B_FALSE;
1149 if (dsl_dir_is_clone(dd)) {
1150 err = dsl_dataset_origin_rm_prep(&dsda, tag);
1151 if (err) {
1152 dsl_dir_close(dd, FTAG);
1153 goto out;
1154 }
1155 }
1156
1157 dstg = dsl_sync_task_group_create(ds->ds_dir->dd_pool);
1158 dsl_sync_task_create(dstg, dsl_dataset_destroy_check,
1159 dsl_dataset_destroy_sync, &dsda, tag, 0);
1160 dsl_sync_task_create(dstg, dsl_dir_destroy_check,
428870ff 1161 dsl_dir_destroy_sync, &dummy_ds, FTAG, 0);
45d1cae3
BB		 1162 err = dsl_sync_task_group_wait(dstg);
1163 dsl_sync_task_group_destroy(dstg);
1164
1165 /*
1166 * We could be racing against 'zfs release' or 'zfs destroy -d'
1167 * on the origin snap, in which case we can get EBUSY if we
1168 * needed to destroy the origin snap but were not ready to
1169 * do so.
1170 */
1171 if (dsda.need_prep) {
1172 ASSERT(err == EBUSY);
1173 ASSERT(dsl_dir_is_clone(dd));
1174 ASSERT(dsda.rm_origin == NULL);
1175 }
1176 } while (dsda.need_prep);
1177
1178 if (dsda.rm_origin != NULL)
1179 dsl_dataset_disown(dsda.rm_origin, tag);
1180
b128c09f 1181 /* if it is successful, dsl_dir_destroy_sync will close the dd */
34dc7c2f
BB		 1182 if (err)
1183 dsl_dir_close(dd, FTAG);
1184out:
b128c09f 1185 dsl_dataset_disown(ds, tag);
34dc7c2f
BB		 1186 return (err);
1187}
1188
34dc7c2f
BB		 1189blkptr_t *
1190dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1191{
1192 return (&ds->ds_phys->ds_bp);
1193}
1194
1195void
1196dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
1197{
1198 ASSERT(dmu_tx_is_syncing(tx));
1199 /* If it's the meta-objset, set dp_meta_rootbp */
1200 if (ds == NULL) {
1201 tx->tx_pool->dp_meta_rootbp = *bp;
1202 } else {
1203 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1204 ds->ds_phys->ds_bp = *bp;
1205 }
1206}
1207
1208spa_t *
1209dsl_dataset_get_spa(dsl_dataset_t *ds)
1210{
1211 return (ds->ds_dir->dd_pool->dp_spa);
1212}
1213
1214void
1215dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1216{
1217 dsl_pool_t *dp;
1218
1219 if (ds == NULL) /* this is the meta-objset */
1220 return;
1221
428870ff 1222 ASSERT(ds->ds_objset != NULL);
34dc7c2f
BB		 1223
1224 if (ds->ds_phys->ds_next_snap_obj != 0)
1225 panic("dirtying snapshot!");
1226
1227 dp = ds->ds_dir->dd_pool;
1228
1229 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg) == 0) {
1230 /* up the hold count until we can be written out */
1231 dmu_buf_add_ref(ds->ds_dbuf, ds);
1232 }
1233}
1234
1235/*
1236 * The unique space in the head dataset can be calculated by subtracting
1237 * the space used in the most recent snapshot, that is still being used
1238 * in this file system, from the space currently in use. To figure out
1239 * the space in the most recent snapshot still in use, we need to take
1240 * the total space used in the snapshot and subtract out the space that
1241 * has been freed up since the snapshot was taken.
1242 */
1243static void
1244dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1245{
1246 uint64_t mrs_used;
1247 uint64_t dlused, dlcomp, dluncomp;
1248
428870ff 1249 ASSERT(!dsl_dataset_is_snapshot(ds));
34dc7c2f
BB		 1250
1251 if (ds->ds_phys->ds_prev_snap_obj != 0)
1252 mrs_used = ds->ds_prev->ds_phys->ds_used_bytes;
1253 else
1254 mrs_used = 0;
1255
428870ff 1256 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
34dc7c2f
BB		 1257
1258 ASSERT3U(dlused, <=, mrs_used);
1259 ds->ds_phys->ds_unique_bytes =
1260 ds->ds_phys->ds_used_bytes - (mrs_used - dlused);
1261
428870ff 1262 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
34dc7c2f
BB		 1263 SPA_VERSION_UNIQUE_ACCURATE)
1264 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1265}
1266
34dc7c2f 1267struct killarg {
b128c09f 1268 dsl_dataset_t *ds;
34dc7c2f
BB		 1269 dmu_tx_t *tx;
1270};
1271
b128c09f 1272/* ARGSUSED */
34dc7c2f 1273static int
428870ff
BB		 1274kill_blkptr(spa_t *spa, zilog_t *zilog, const blkptr_t *bp, arc_buf_t *pbuf,
1275 const zbookmark_t *zb, const dnode_phys_t *dnp, void *arg)
34dc7c2f
BB		 1276{
1277 struct killarg *ka = arg;
428870ff 1278 dmu_tx_t *tx = ka->tx;
34dc7c2f 1279
b128c09f
BB		 1280 if (bp == NULL)
1281 return (0);
1282
428870ff
BB		 1283 if (zb->zb_level == ZB_ZIL_LEVEL) {
1284 ASSERT(zilog != NULL);
9babb374
BB		 1285 /*
1286 * It's a block in the intent log. It has no
1287 * accounting, so just free it.
1288 */
428870ff 1289 dsl_free(ka->tx->tx_pool, ka->tx->tx_txg, bp);
9babb374 1290 } else {
428870ff 1291 ASSERT(zilog == NULL);
9babb374 1292 ASSERT3U(bp->blk_birth, >, ka->ds->ds_phys->ds_prev_snap_txg);
428870ff 1293 (void) dsl_dataset_block_kill(ka->ds, bp, tx, B_FALSE);
9babb374 1294 }
34dc7c2f 1295
34dc7c2f
BB		 1296 return (0);
1297}
1298
34dc7c2f
BB		 1299/* ARGSUSED */
1300static int
1301dsl_dataset_destroy_begin_check(void *arg1, void *arg2, dmu_tx_t *tx)
1302{
1303 dsl_dataset_t *ds = arg1;
1304 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1305 uint64_t count;
1306 int err;
1307
1308 /*
1309 * Can't delete a head dataset if there are snapshots of it.
1310 * (Except if the only snapshots are from the branch we cloned
1311 * from.)
1312 */
1313 if (ds->ds_prev != NULL &&
1314 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
428870ff 1315 return (EBUSY);
34dc7c2f
BB		 1316
1317 /*
1318 * This is really a dsl_dir thing, but check it here so that
1319 * we'll be less likely to leave this dataset inconsistent &
1320 * nearly destroyed.
1321 */
1322 err = zap_count(mos, ds->ds_dir->dd_phys->dd_child_dir_zapobj, &count);
1323 if (err)
1324 return (err);
1325 if (count != 0)
1326 return (EEXIST);
1327
1328 return (0);
1329}
1330
1331/* ARGSUSED */
1332static void
428870ff 1333dsl_dataset_destroy_begin_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 1334{
1335 dsl_dataset_t *ds = arg1;
1336 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1337
1338 /* Mark it as inconsistent on-disk, in case we crash */
1339 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1340 ds->ds_phys->ds_flags |= DS_FLAG_INCONSISTENT;
1341
428870ff
BB		 1342 spa_history_log_internal(LOG_DS_DESTROY_BEGIN, dp->dp_spa, tx,
1343 "dataset = %llu", ds->ds_object);
34dc7c2f
BB		 1344}
1345
45d1cae3
BB		 1346static int
1347dsl_dataset_origin_check(struct dsl_ds_destroyarg *dsda, void *tag,
1348 dmu_tx_t *tx)
1349{
1350 dsl_dataset_t *ds = dsda->ds;
1351 dsl_dataset_t *ds_prev = ds->ds_prev;
1352
1353 if (dsl_dataset_might_destroy_origin(ds_prev)) {
1354 struct dsl_ds_destroyarg ndsda = {0};
1355
1356 /*
1357 * If we're not prepared to remove the origin, don't remove
1358 * the clone either.
1359 */
1360 if (dsda->rm_origin == NULL) {
1361 dsda->need_prep = B_TRUE;
1362 return (EBUSY);
1363 }
1364
1365 ndsda.ds = ds_prev;
1366 ndsda.is_origin_rm = B_TRUE;
1367 return (dsl_dataset_destroy_check(&ndsda, tag, tx));
1368 }
1369
1370 /*
1371 * If we're not going to remove the origin after all,
1372 * undo the open context setup.
1373 */
1374 if (dsda->rm_origin != NULL) {
1375 dsl_dataset_disown(dsda->rm_origin, tag);
1376 dsda->rm_origin = NULL;
1377 }
1378
1379 return (0);
1380}
1381
572e2857
BB		 1382/*
1383 * If you add new checks here, you may need to add
1384 * additional checks to the "temporary" case in
1385 * snapshot_check() in dmu_objset.c.
1386 */
34dc7c2f
BB		 1387/* ARGSUSED */
1388int
1389dsl_dataset_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
1390{
45d1cae3
BB		 1391 struct dsl_ds_destroyarg *dsda = arg1;
1392 dsl_dataset_t *ds = dsda->ds;
34dc7c2f 1393
b128c09f
BB		 1394 /* we have an owner hold, so noone else can destroy us */
1395 ASSERT(!DSL_DATASET_IS_DESTROYED(ds));
1396
45d1cae3
BB		 1397 /*
1398 * Only allow deferred destroy on pools that support it.
1399 * NOTE: deferred destroy is only supported on snapshots.
1400 */
1401 if (dsda->defer) {
1402 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
1403 SPA_VERSION_USERREFS)
1404 return (ENOTSUP);
1405 ASSERT(dsl_dataset_is_snapshot(ds));
1406 return (0);
1407 }
34dc7c2f
BB		 1408
1409 /*
1410 * Can't delete a head dataset if there are snapshots of it.
1411 * (Except if the only snapshots are from the branch we cloned
1412 * from.)
1413 */
1414 if (ds->ds_prev != NULL &&
1415 ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object)
428870ff 1416 return (EBUSY);
34dc7c2f
BB		 1417
1418 /*
1419 * If we made changes this txg, traverse_dsl_dataset won't find
1420 * them. Try again.
1421 */
1422 if (ds->ds_phys->ds_bp.blk_birth >= tx->tx_txg)
1423 return (EAGAIN);
1424
45d1cae3
BB		 1425 if (dsl_dataset_is_snapshot(ds)) {
1426 /*
1427 * If this snapshot has an elevated user reference count,
1428 * we can't destroy it yet.
1429 */
1430 if (ds->ds_userrefs > 0 && !dsda->releasing)
1431 return (EBUSY);
1432
1433 mutex_enter(&ds->ds_lock);
1434 /*
1435 * Can't delete a branch point. However, if we're destroying
1436 * a clone and removing its origin due to it having a user
1437 * hold count of 0 and having been marked for deferred destroy,
1438 * it's OK for the origin to have a single clone.
1439 */
1440 if (ds->ds_phys->ds_num_children >
1441 (dsda->is_origin_rm ? 2 : 1)) {
1442 mutex_exit(&ds->ds_lock);
1443 return (EEXIST);
1444 }
1445 mutex_exit(&ds->ds_lock);
1446 } else if (dsl_dir_is_clone(ds->ds_dir)) {
1447 return (dsl_dataset_origin_check(dsda, arg2, tx));
1448 }
1449
34dc7c2f
BB		 1450 /* XXX we should do some i/o error checking... */
1451 return (0);
1452}
1453
b128c09f
BB		 1454struct refsarg {
1455 kmutex_t lock;
1456 boolean_t gone;
1457 kcondvar_t cv;
1458};
1459
1460/* ARGSUSED */
1461static void
1462dsl_dataset_refs_gone(dmu_buf_t *db, void *argv)
1463{
1464 struct refsarg *arg = argv;
1465
1466 mutex_enter(&arg->lock);
1467 arg->gone = TRUE;
1468 cv_signal(&arg->cv);
1469 mutex_exit(&arg->lock);
1470}
1471
1472static void
1473dsl_dataset_drain_refs(dsl_dataset_t *ds, void *tag)
1474{
1475 struct refsarg arg;
1476
1477 mutex_init(&arg.lock, NULL, MUTEX_DEFAULT, NULL);
1478 cv_init(&arg.cv, NULL, CV_DEFAULT, NULL);
1479 arg.gone = FALSE;
1480 (void) dmu_buf_update_user(ds->ds_dbuf, ds, &arg, &ds->ds_phys,
1481 dsl_dataset_refs_gone);
1482 dmu_buf_rele(ds->ds_dbuf, tag);
1483 mutex_enter(&arg.lock);
1484 while (!arg.gone)
1485 cv_wait(&arg.cv, &arg.lock);
1486 ASSERT(arg.gone);
1487 mutex_exit(&arg.lock);
1488 ds->ds_dbuf = NULL;
1489 ds->ds_phys = NULL;
1490 mutex_destroy(&arg.lock);
1491 cv_destroy(&arg.cv);
1492}
1493
428870ff
BB		 1494static void
1495remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj, dmu_tx_t *tx)
1496{
1497 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1498 uint64_t count;
1499 int err;
1500
1501 ASSERT(ds->ds_phys->ds_num_children >= 2);
1502 err = zap_remove_int(mos, ds->ds_phys->ds_next_clones_obj, obj, tx);
1503 /*
1504 * The err should not be ENOENT, but a bug in a previous version
1505 * of the code could cause upgrade_clones_cb() to not set
1506 * ds_next_snap_obj when it should, leading to a missing entry.
1507 * If we knew that the pool was created after
1508 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1509 * ENOENT. However, at least we can check that we don't have
1510 * too many entries in the next_clones_obj even after failing to
1511 * remove this one.
1512 */
1513 if (err != ENOENT) {
1514 VERIFY3U(err, ==, 0);
1515 }
1516 ASSERT3U(0, ==, zap_count(mos, ds->ds_phys->ds_next_clones_obj,
1517 &count));
1518 ASSERT3U(count, <=, ds->ds_phys->ds_num_children - 2);
1519}
1520
1521static void
1522dsl_dataset_remove_clones_key(dsl_dataset_t *ds, uint64_t mintxg, dmu_tx_t *tx)
1523{
1524 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1525 zap_cursor_t zc;
1526 zap_attribute_t za;
1527
1528 /*
1529 * If it is the old version, dd_clones doesn't exist so we can't
1530 * find the clones, but deadlist_remove_key() is a no-op so it
1531 * doesn't matter.
1532 */
1533 if (ds->ds_dir->dd_phys->dd_clones == 0)
1534 return;
1535
1536 for (zap_cursor_init(&zc, mos, ds->ds_dir->dd_phys->dd_clones);
1537 zap_cursor_retrieve(&zc, &za) == 0;
1538 zap_cursor_advance(&zc)) {
1539 dsl_dataset_t *clone;
1540
1541 VERIFY3U(0, ==, dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1542 za.za_first_integer, FTAG, &clone));
1543 if (clone->ds_dir->dd_origin_txg > mintxg) {
1544 dsl_deadlist_remove_key(&clone->ds_deadlist,
1545 mintxg, tx);
1546 dsl_dataset_remove_clones_key(clone, mintxg, tx);
1547 }
1548 dsl_dataset_rele(clone, FTAG);
1549 }
1550 zap_cursor_fini(&zc);
1551}
1552
1553struct process_old_arg {
1554 dsl_dataset_t *ds;
1555 dsl_dataset_t *ds_prev;
1556 boolean_t after_branch_point;
1557 zio_t *pio;
1558 uint64_t used, comp, uncomp;
1559};
1560
1561static int
1562process_old_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1563{
1564 struct process_old_arg *poa = arg;
1565 dsl_pool_t *dp = poa->ds->ds_dir->dd_pool;
1566
1567 if (bp->blk_birth <= poa->ds->ds_phys->ds_prev_snap_txg) {
1568 dsl_deadlist_insert(&poa->ds->ds_deadlist, bp, tx);
1569 if (poa->ds_prev && !poa->after_branch_point &&
1570 bp->blk_birth >
1571 poa->ds_prev->ds_phys->ds_prev_snap_txg) {
1572 poa->ds_prev->ds_phys->ds_unique_bytes +=
1573 bp_get_dsize_sync(dp->dp_spa, bp);
1574 }
1575 } else {
1576 poa->used += bp_get_dsize_sync(dp->dp_spa, bp);
1577 poa->comp += BP_GET_PSIZE(bp);
1578 poa->uncomp += BP_GET_UCSIZE(bp);
1579 dsl_free_sync(poa->pio, dp, tx->tx_txg, bp);
1580 }
1581 return (0);
1582}
1583
1584static void
1585process_old_deadlist(dsl_dataset_t *ds, dsl_dataset_t *ds_prev,
1586 dsl_dataset_t *ds_next, boolean_t after_branch_point, dmu_tx_t *tx)
1587{
1588 struct process_old_arg poa = { 0 };
1589 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1590 objset_t *mos = dp->dp_meta_objset;
1591
1592 ASSERT(ds->ds_deadlist.dl_oldfmt);
1593 ASSERT(ds_next->ds_deadlist.dl_oldfmt);
1594
1595 poa.ds = ds;
1596 poa.ds_prev = ds_prev;
1597 poa.after_branch_point = after_branch_point;
1598 poa.pio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1599 VERIFY3U(0, ==, bpobj_iterate(&ds_next->ds_deadlist.dl_bpobj,
1600 process_old_cb, &poa, tx));
1601 VERIFY3U(zio_wait(poa.pio), ==, 0);
1602 ASSERT3U(poa.used, ==, ds->ds_phys->ds_unique_bytes);
1603
1604 /* change snapused */
1605 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1606 -poa.used, -poa.comp, -poa.uncomp, tx);
1607
1608 /* swap next's deadlist to our deadlist */
1609 dsl_deadlist_close(&ds->ds_deadlist);
1610 dsl_deadlist_close(&ds_next->ds_deadlist);
1611 SWITCH64(ds_next->ds_phys->ds_deadlist_obj,
1612 ds->ds_phys->ds_deadlist_obj);
1613 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
1614 dsl_deadlist_open(&ds_next->ds_deadlist, mos,
1615 ds_next->ds_phys->ds_deadlist_obj);
1616}
1617
34dc7c2f 1618void
428870ff 1619dsl_dataset_destroy_sync(void *arg1, void *tag, dmu_tx_t *tx)
34dc7c2f 1620{
45d1cae3
BB		 1621 struct dsl_ds_destroyarg *dsda = arg1;
1622 dsl_dataset_t *ds = dsda->ds;
34dc7c2f
BB		 1623 int err;
1624 int after_branch_point = FALSE;
1625 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1626 objset_t *mos = dp->dp_meta_objset;
1627 dsl_dataset_t *ds_prev = NULL;
572e2857 1628 boolean_t wont_destroy;
34dc7c2f
BB		 1629 uint64_t obj;
1630
572e2857
BB		 1631 wont_destroy = (dsda->defer &&
1632 (ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1));
1633
1634 ASSERT(ds->ds_owner || wont_destroy);
45d1cae3 1635 ASSERT(dsda->defer || ds->ds_phys->ds_num_children <= 1);
34dc7c2f
BB		 1636 ASSERT(ds->ds_prev == NULL ||
1637 ds->ds_prev->ds_phys->ds_next_snap_obj != ds->ds_object);
1638 ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
1639
572e2857 1640 if (wont_destroy) {
45d1cae3 1641 ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
572e2857
BB		 1642 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1643 ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
1644 return;
45d1cae3
BB		 1645 }
1646
b128c09f
BB		 1647 /* signal any waiters that this dataset is going away */
1648 mutex_enter(&ds->ds_lock);
1649 ds->ds_owner = dsl_reaper;
1650 cv_broadcast(&ds->ds_exclusive_cv);
1651 mutex_exit(&ds->ds_lock);
1652
34dc7c2f
BB		 1653 /* Remove our reservation */
1654 if (ds->ds_reserved != 0) {
428870ff
BB		 1655 dsl_prop_setarg_t psa;
1656 uint64_t value = 0;
1657
1658 dsl_prop_setarg_init_uint64(&psa, "refreservation",
1659 (ZPROP_SRC_NONE | ZPROP_SRC_LOCAL | ZPROP_SRC_RECEIVED),
1660 &value);
1661 psa.psa_effective_value = 0; /* predict default value */
1662
1663 dsl_dataset_set_reservation_sync(ds, &psa, tx);
34dc7c2f
BB		 1664 ASSERT3U(ds->ds_reserved, ==, 0);
1665 }
1666
1667 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
1668
428870ff 1669 dsl_scan_ds_destroyed(ds, tx);
b128c09f 1670
34dc7c2f
BB		 1671 obj = ds->ds_object;
1672
1673 if (ds->ds_phys->ds_prev_snap_obj != 0) {
1674 if (ds->ds_prev) {
1675 ds_prev = ds->ds_prev;
1676 } else {
b128c09f
BB		 1677 VERIFY(0 == dsl_dataset_hold_obj(dp,
1678 ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
34dc7c2f
BB		 1679 }
1680 after_branch_point =
1681 (ds_prev->ds_phys->ds_next_snap_obj != obj);
1682
1683 dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
b128c09f
BB		 1684 if (after_branch_point &&
1685 ds_prev->ds_phys->ds_next_clones_obj != 0) {
428870ff 1686 remove_from_next_clones(ds_prev, obj, tx);
b128c09f
BB		 1687 if (ds->ds_phys->ds_next_snap_obj != 0) {
1688 VERIFY(0 == zap_add_int(mos,
1689 ds_prev->ds_phys->ds_next_clones_obj,
1690 ds->ds_phys->ds_next_snap_obj, tx));
1691 }
1692 }
34dc7c2f
BB		 1693 if (after_branch_point &&
1694 ds->ds_phys->ds_next_snap_obj == 0) {
1695 /* This clone is toast. */
1696 ASSERT(ds_prev->ds_phys->ds_num_children > 1);
1697 ds_prev->ds_phys->ds_num_children--;
45d1cae3
BB		 1698
1699 /*
1700 * If the clone's origin has no other clones, no
1701 * user holds, and has been marked for deferred
1702 * deletion, then we should have done the necessary
1703 * destroy setup for it.
1704 */
1705 if (ds_prev->ds_phys->ds_num_children == 1 &&
1706 ds_prev->ds_userrefs == 0 &&
1707 DS_IS_DEFER_DESTROY(ds_prev)) {
1708 ASSERT3P(dsda->rm_origin, !=, NULL);
1709 } else {
1710 ASSERT3P(dsda->rm_origin, ==, NULL);
1711 }
34dc7c2f
BB		 1712 } else if (!after_branch_point) {
1713 ds_prev->ds_phys->ds_next_snap_obj =
1714 ds->ds_phys->ds_next_snap_obj;
1715 }
1716 }
1717
428870ff 1718 if (dsl_dataset_is_snapshot(ds)) {
34dc7c2f 1719 dsl_dataset_t *ds_next;
34dc7c2f 1720 uint64_t old_unique;
428870ff 1721 uint64_t used = 0, comp = 0, uncomp = 0;
34dc7c2f 1722
b128c09f
BB		 1723 VERIFY(0 == dsl_dataset_hold_obj(dp,
1724 ds->ds_phys->ds_next_snap_obj, FTAG, &ds_next));
34dc7c2f
BB		 1725 ASSERT3U(ds_next->ds_phys->ds_prev_snap_obj, ==, obj);
1726
428870ff 1727 old_unique = ds_next->ds_phys->ds_unique_bytes;
34dc7c2f
BB		 1728
1729 dmu_buf_will_dirty(ds_next->ds_dbuf, tx);
1730 ds_next->ds_phys->ds_prev_snap_obj =
1731 ds->ds_phys->ds_prev_snap_obj;
1732 ds_next->ds_phys->ds_prev_snap_txg =
1733 ds->ds_phys->ds_prev_snap_txg;
1734 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
1735 ds_prev ? ds_prev->ds_phys->ds_creation_txg : 0);
1736
34dc7c2f 1737
428870ff
BB		 1738 if (ds_next->ds_deadlist.dl_oldfmt) {
1739 process_old_deadlist(ds, ds_prev, ds_next,
1740 after_branch_point, tx);
1741 } else {
1742 /* Adjust prev's unique space. */
1743 if (ds_prev && !after_branch_point) {
1744 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1745 ds_prev->ds_phys->ds_prev_snap_txg,
1746 ds->ds_phys->ds_prev_snap_txg,
1747 &used, &comp, &uncomp);
1748 ds_prev->ds_phys->ds_unique_bytes += used;
1749 }
b128c09f 1750
428870ff
BB		 1751 /* Adjust snapused. */
1752 dsl_deadlist_space_range(&ds_next->ds_deadlist,
1753 ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
1754 &used, &comp, &uncomp);
1755 dsl_dir_diduse_space(ds->ds_dir, DD_USED_SNAP,
1756 -used, -comp, -uncomp, tx);
1757
1758 /* Move blocks to be freed to pool's free list. */
1759 dsl_deadlist_move_bpobj(&ds_next->ds_deadlist,
1760 &dp->dp_free_bpobj, ds->ds_phys->ds_prev_snap_txg,
1761 tx);
1762 dsl_dir_diduse_space(tx->tx_pool->dp_free_dir,
1763 DD_USED_HEAD, used, comp, uncomp, tx);
1764 dsl_dir_dirty(tx->tx_pool->dp_free_dir, tx);
1765
1766 /* Merge our deadlist into next's and free it. */
1767 dsl_deadlist_merge(&ds_next->ds_deadlist,
1768 ds->ds_phys->ds_deadlist_obj, tx);
1769 }
1770 dsl_deadlist_close(&ds->ds_deadlist);
1771 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
b128c09f 1772
428870ff
BB		 1773 /* Collapse range in clone heads */
1774 dsl_dataset_remove_clones_key(ds,
1775 ds->ds_phys->ds_creation_txg, tx);
34dc7c2f 1776
428870ff
BB		 1777 if (dsl_dataset_is_snapshot(ds_next)) {
1778 dsl_dataset_t *ds_nextnext;
d6320ddb 1779 dsl_dataset_t *hds;
34dc7c2f 1780
34dc7c2f
BB		 1781 /*
1782 * Update next's unique to include blocks which
1783 * were previously shared by only this snapshot
1784 * and it. Those blocks will be born after the
1785 * prev snap and before this snap, and will have
1786 * died after the next snap and before the one
1787 * after that (ie. be on the snap after next's
1788 * deadlist).
34dc7c2f 1789 */
b128c09f
BB		 1790 VERIFY(0 == dsl_dataset_hold_obj(dp,
1791 ds_next->ds_phys->ds_next_snap_obj,
428870ff
BB		 1792 FTAG, &ds_nextnext));
1793 dsl_deadlist_space_range(&ds_nextnext->ds_deadlist,
b128c09f 1794 ds->ds_phys->ds_prev_snap_txg,
428870ff
BB		 1795 ds->ds_phys->ds_creation_txg,
1796 &used, &comp, &uncomp);
1797 ds_next->ds_phys->ds_unique_bytes += used;
1798 dsl_dataset_rele(ds_nextnext, FTAG);
34dc7c2f 1799 ASSERT3P(ds_next->ds_prev, ==, NULL);
428870ff
BB		 1800
1801 /* Collapse range in this head. */
428870ff
BB		 1802 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
1803 ds->ds_dir->dd_phys->dd_head_dataset_obj,
1804 FTAG, &hds));
1805 dsl_deadlist_remove_key(&hds->ds_deadlist,
1806 ds->ds_phys->ds_creation_txg, tx);
1807 dsl_dataset_rele(hds, FTAG);
1808
34dc7c2f
BB		 1809 } else {
1810 ASSERT3P(ds_next->ds_prev, ==, ds);
b128c09f
BB		 1811 dsl_dataset_drop_ref(ds_next->ds_prev, ds_next);
1812 ds_next->ds_prev = NULL;
34dc7c2f 1813 if (ds_prev) {
b128c09f
BB		 1814 VERIFY(0 == dsl_dataset_get_ref(dp,
1815 ds->ds_phys->ds_prev_snap_obj,
1816 ds_next, &ds_next->ds_prev));
34dc7c2f
BB		 1817 }
1818
1819 dsl_dataset_recalc_head_uniq(ds_next);
1820
1821 /*
1822 * Reduce the amount of our unconsmed refreservation
1823 * being charged to our parent by the amount of
1824 * new unique data we have gained.
1825 */
1826 if (old_unique < ds_next->ds_reserved) {
1827 int64_t mrsdelta;
1828 uint64_t new_unique =
1829 ds_next->ds_phys->ds_unique_bytes;
1830
1831 ASSERT(old_unique <= new_unique);
1832 mrsdelta = MIN(new_unique - old_unique,
1833 ds_next->ds_reserved - old_unique);
b128c09f
BB		 1834 dsl_dir_diduse_space(ds->ds_dir,
1835 DD_USED_REFRSRV, -mrsdelta, 0, 0, tx);
34dc7c2f
BB		 1836 }
1837 }
b128c09f 1838 dsl_dataset_rele(ds_next, FTAG);
34dc7c2f
BB		 1839 } else {
1840 /*
1841 * There's no next snapshot, so this is a head dataset.
1842 * Destroy the deadlist. Unless it's a clone, the
1843 * deadlist should be empty. (If it's a clone, it's
1844 * safe to ignore the deadlist contents.)
1845 */
1846 struct killarg ka;
1847
428870ff
BB		 1848 dsl_deadlist_close(&ds->ds_deadlist);
1849 dsl_deadlist_free(mos, ds->ds_phys->ds_deadlist_obj, tx);
34dc7c2f
BB		 1850 ds->ds_phys->ds_deadlist_obj = 0;
1851
1852 /*
1853 * Free everything that we point to (that's born after
1854 * the previous snapshot, if we are a clone)
1855 *
b128c09f
BB		 1856 * NB: this should be very quick, because we already
1857 * freed all the objects in open context.
34dc7c2f 1858 */
b128c09f 1859 ka.ds = ds;
34dc7c2f 1860 ka.tx = tx;
b128c09f
BB		 1861 err = traverse_dataset(ds, ds->ds_phys->ds_prev_snap_txg,
1862 TRAVERSE_POST, kill_blkptr, &ka);
34dc7c2f 1863 ASSERT3U(err, ==, 0);
9babb374 1864 ASSERT(!DS_UNIQUE_IS_ACCURATE(ds) ||
b128c09f 1865 ds->ds_phys->ds_unique_bytes == 0);
34dc7c2f 1866
428870ff
BB		 1867 if (ds->ds_prev != NULL) {
1868 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
1869 VERIFY3U(0, ==, zap_remove_int(mos,
1870 ds->ds_prev->ds_dir->dd_phys->dd_clones,
1871 ds->ds_object, tx));
1872 }
1873 dsl_dataset_rele(ds->ds_prev, ds);
1874 ds->ds_prev = ds_prev = NULL;
1875 }
1876 }
34dc7c2f 1877
572e2857
BB		 1878 /*
1879 * This must be done after the dsl_traverse(), because it will
1880 * re-open the objset.
1881 */
1882 if (ds->ds_objset) {
1883 dmu_objset_evict(ds->ds_objset);
1884 ds->ds_objset = NULL;
1885 }
1886
34dc7c2f 1887 if (ds->ds_dir->dd_phys->dd_head_dataset_obj == ds->ds_object) {
b128c09f 1888 /* Erase the link in the dir */
34dc7c2f
BB		 1889 dmu_buf_will_dirty(ds->ds_dir->dd_dbuf, tx);
1890 ds->ds_dir->dd_phys->dd_head_dataset_obj = 0;
b128c09f
BB		 1891 ASSERT(ds->ds_phys->ds_snapnames_zapobj != 0);
1892 err = zap_destroy(mos, ds->ds_phys->ds_snapnames_zapobj, tx);
1893 ASSERT(err == 0);
34dc7c2f
BB		 1894 } else {
1895 /* remove from snapshot namespace */
1896 dsl_dataset_t *ds_head;
b128c09f
BB		 1897 ASSERT(ds->ds_phys->ds_snapnames_zapobj == 0);
1898 VERIFY(0 == dsl_dataset_hold_obj(dp,
1899 ds->ds_dir->dd_phys->dd_head_dataset_obj, FTAG, &ds_head));
34dc7c2f
BB		 1900 VERIFY(0 == dsl_dataset_get_snapname(ds));
1901#ifdef ZFS_DEBUG
1902 {
1903 uint64_t val;
1904
b128c09f 1905 err = dsl_dataset_snap_lookup(ds_head,
34dc7c2f
BB		 1906 ds->ds_snapname, &val);
1907 ASSERT3U(err, ==, 0);
1908 ASSERT3U(val, ==, obj);
1909 }
1910#endif
b128c09f 1911 err = dsl_dataset_snap_remove(ds_head, ds->ds_snapname, tx);
34dc7c2f 1912 ASSERT(err == 0);
b128c09f 1913 dsl_dataset_rele(ds_head, FTAG);
34dc7c2f
BB		 1914 }
1915
1916 if (ds_prev && ds->ds_prev != ds_prev)
b128c09f 1917 dsl_dataset_rele(ds_prev, FTAG);
34dc7c2f
BB		 1918
1919 spa_prop_clear_bootfs(dp->dp_spa, ds->ds_object, tx);
428870ff
BB		 1920 spa_history_log_internal(LOG_DS_DESTROY, dp->dp_spa, tx,
1921 "dataset = %llu", ds->ds_object);
34dc7c2f 1922
b128c09f
BB		 1923 if (ds->ds_phys->ds_next_clones_obj != 0) {
1924 uint64_t count;
1925 ASSERT(0 == zap_count(mos,
1926 ds->ds_phys->ds_next_clones_obj, &count) && count == 0);
1927 VERIFY(0 == dmu_object_free(mos,
1928 ds->ds_phys->ds_next_clones_obj, tx));
1929 }
1930 if (ds->ds_phys->ds_props_obj != 0)
1931 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_props_obj, tx));
45d1cae3
BB		 1932 if (ds->ds_phys->ds_userrefs_obj != 0)
1933 VERIFY(0 == zap_destroy(mos, ds->ds_phys->ds_userrefs_obj, tx));
b128c09f
BB		 1934 dsl_dir_close(ds->ds_dir, ds);
1935 ds->ds_dir = NULL;
1936 dsl_dataset_drain_refs(ds, tag);
34dc7c2f 1937 VERIFY(0 == dmu_object_free(mos, obj, tx));
45d1cae3
BB		 1938
1939 if (dsda->rm_origin) {
1940 /*
1941 * Remove the origin of the clone we just destroyed.
1942 */
45d1cae3
BB		 1943 struct dsl_ds_destroyarg ndsda = {0};
1944
428870ff
BB		 1945 ndsda.ds = dsda->rm_origin;
1946 dsl_dataset_destroy_sync(&ndsda, tag, tx);
45d1cae3 1947 }
34dc7c2f
BB		 1948}
1949
1950static int
1951dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1952{
1953 uint64_t asize;
1954
1955 if (!dmu_tx_is_syncing(tx))
1956 return (0);
1957
1958 /*
1959 * If there's an fs-only reservation, any blocks that might become
1960 * owned by the snapshot dataset must be accommodated by space
1961 * outside of the reservation.
1962 */
428870ff
BB		 1963 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1964 asize = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
572e2857 1965 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
34dc7c2f
BB		 1966 return (ENOSPC);
1967
1968 /*
1969 * Propogate any reserved space for this snapshot to other
1970 * snapshot checks in this sync group.
1971 */
1972 if (asize > 0)
1973 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1974
1975 return (0);
1976}
1977
34dc7c2f
BB		 1978int
1979dsl_dataset_snapshot_check(void *arg1, void *arg2, dmu_tx_t *tx)
1980{
1981 dsl_dataset_t *ds = arg1;
1982 const char *snapname = arg2;
34dc7c2f
BB		 1983 int err;
1984 uint64_t value;
1985
1986 /*
1987 * We don't allow multiple snapshots of the same txg. If there
1988 * is already one, try again.
1989 */
1990 if (ds->ds_phys->ds_prev_snap_txg >= tx->tx_txg)
1991 return (EAGAIN);
1992
1993 /*
1994 * Check for conflicting name snapshot name.
1995 */
b128c09f 1996 err = dsl_dataset_snap_lookup(ds, snapname, &value);
34dc7c2f
BB		 1997 if (err == 0)
1998 return (EEXIST);
1999 if (err != ENOENT)
2000 return (err);
2001
2002 /*
2003 * Check that the dataset's name is not too long. Name consists
2004 * of the dataset's length + 1 for the @-sign + snapshot name's length
2005 */
2006 if (dsl_dataset_namelen(ds) + 1 + strlen(snapname) >= MAXNAMELEN)
2007 return (ENAMETOOLONG);
2008
2009 err = dsl_dataset_snapshot_reserve_space(ds, tx);
2010 if (err)
2011 return (err);
2012
2013 ds->ds_trysnap_txg = tx->tx_txg;
2014 return (0);
2015}
2016
2017void
428870ff 2018dsl_dataset_snapshot_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2019{
2020 dsl_dataset_t *ds = arg1;
2021 const char *snapname = arg2;
2022 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2023 dmu_buf_t *dbuf;
2024 dsl_dataset_phys_t *dsphys;
b128c09f 2025 uint64_t dsobj, crtxg;
34dc7c2f
BB		 2026 objset_t *mos = dp->dp_meta_objset;
2027 int err;
2028
34dc7c2f
BB		 2029 ASSERT(RW_WRITE_HELD(&dp->dp_config_rwlock));
2030
b128c09f
BB		 2031 /*
2032 * The origin's ds_creation_txg has to be < TXG_INITIAL
2033 */
2034 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
2035 crtxg = 1;
2036 else
2037 crtxg = tx->tx_txg;
2038
34dc7c2f
BB		 2039 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
2040 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
2041 VERIFY(0 == dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
2042 dmu_buf_will_dirty(dbuf, tx);
2043 dsphys = dbuf->db_data;
b128c09f 2044 bzero(dsphys, sizeof (dsl_dataset_phys_t));
34dc7c2f
BB		 2045 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
2046 dsphys->ds_fsid_guid = unique_create();
2047 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
2048 sizeof (dsphys->ds_guid));
2049 dsphys->ds_prev_snap_obj = ds->ds_phys->ds_prev_snap_obj;
2050 dsphys->ds_prev_snap_txg = ds->ds_phys->ds_prev_snap_txg;
2051 dsphys->ds_next_snap_obj = ds->ds_object;
2052 dsphys->ds_num_children = 1;
2053 dsphys->ds_creation_time = gethrestime_sec();
b128c09f 2054 dsphys->ds_creation_txg = crtxg;
34dc7c2f
BB		 2055 dsphys->ds_deadlist_obj = ds->ds_phys->ds_deadlist_obj;
2056 dsphys->ds_used_bytes = ds->ds_phys->ds_used_bytes;
2057 dsphys->ds_compressed_bytes = ds->ds_phys->ds_compressed_bytes;
2058 dsphys->ds_uncompressed_bytes = ds->ds_phys->ds_uncompressed_bytes;
2059 dsphys->ds_flags = ds->ds_phys->ds_flags;
2060 dsphys->ds_bp = ds->ds_phys->ds_bp;
2061 dmu_buf_rele(dbuf, FTAG);
2062
2063 ASSERT3U(ds->ds_prev != 0, ==, ds->ds_phys->ds_prev_snap_obj != 0);
2064 if (ds->ds_prev) {
b128c09f
BB		 2065 uint64_t next_clones_obj =
2066 ds->ds_prev->ds_phys->ds_next_clones_obj;
34dc7c2f
BB		 2067 ASSERT(ds->ds_prev->ds_phys->ds_next_snap_obj ==
2068 ds->ds_object ||
2069 ds->ds_prev->ds_phys->ds_num_children > 1);
2070 if (ds->ds_prev->ds_phys->ds_next_snap_obj == ds->ds_object) {
2071 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
2072 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, ==,
2073 ds->ds_prev->ds_phys->ds_creation_txg);
2074 ds->ds_prev->ds_phys->ds_next_snap_obj = dsobj;
b128c09f 2075 } else if (next_clones_obj != 0) {
428870ff
BB		 2076 remove_from_next_clones(ds->ds_prev,
2077 dsphys->ds_next_snap_obj, tx);
b128c09f
BB		 2078 VERIFY3U(0, ==, zap_add_int(mos,
2079 next_clones_obj, dsobj, tx));
34dc7c2f
BB		 2080 }
2081 }
2082
2083 /*
2084 * If we have a reference-reservation on this dataset, we will
2085 * need to increase the amount of refreservation being charged
2086 * since our unique space is going to zero.
2087 */
2088 if (ds->ds_reserved) {
428870ff
BB		 2089 int64_t delta;
2090 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
2091 delta = MIN(ds->ds_phys->ds_unique_bytes, ds->ds_reserved);
b128c09f 2092 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
428870ff 2093 delta, 0, 0, tx);
34dc7c2f
BB		 2094 }
2095
34dc7c2f 2096 dmu_buf_will_dirty(ds->ds_dbuf, tx);
428870ff
BB		 2097 zfs_dbgmsg("taking snapshot %s@%s/%llu; newkey=%llu",
2098 ds->ds_dir->dd_myname, snapname, dsobj,
2099 ds->ds_phys->ds_prev_snap_txg);
2100 ds->ds_phys->ds_deadlist_obj = dsl_deadlist_clone(&ds->ds_deadlist,
2101 UINT64_MAX, ds->ds_phys->ds_prev_snap_obj, tx);
2102 dsl_deadlist_close(&ds->ds_deadlist);
2103 dsl_deadlist_open(&ds->ds_deadlist, mos, ds->ds_phys->ds_deadlist_obj);
2104 dsl_deadlist_add_key(&ds->ds_deadlist,
2105 ds->ds_phys->ds_prev_snap_txg, tx);
2106
34dc7c2f
BB		 2107 ASSERT3U(ds->ds_phys->ds_prev_snap_txg, <, tx->tx_txg);
2108 ds->ds_phys->ds_prev_snap_obj = dsobj;
b128c09f 2109 ds->ds_phys->ds_prev_snap_txg = crtxg;
34dc7c2f
BB		 2110 ds->ds_phys->ds_unique_bytes = 0;
2111 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
2112 ds->ds_phys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
34dc7c2f 2113
34dc7c2f
BB		 2114 err = zap_add(mos, ds->ds_phys->ds_snapnames_zapobj,
2115 snapname, 8, 1, &dsobj, tx);
2116 ASSERT(err == 0);
2117
2118 if (ds->ds_prev)
b128c09f
BB		 2119 dsl_dataset_drop_ref(ds->ds_prev, ds);
2120 VERIFY(0 == dsl_dataset_get_ref(dp,
2121 ds->ds_phys->ds_prev_snap_obj, ds, &ds->ds_prev));
2122
428870ff
BB		 2123 dsl_scan_ds_snapshotted(ds, tx);
2124
2125 dsl_dir_snap_cmtime_update(ds->ds_dir);
34dc7c2f 2126
428870ff 2127 spa_history_log_internal(LOG_DS_SNAPSHOT, dp->dp_spa, tx,
34dc7c2f
BB		 2128 "dataset = %llu", dsobj);
2129}
2130
2131void
2132dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
2133{
2134 ASSERT(dmu_tx_is_syncing(tx));
428870ff 2135 ASSERT(ds->ds_objset != NULL);
34dc7c2f
BB		 2136 ASSERT(ds->ds_phys->ds_next_snap_obj == 0);
2137
2138 /*
2139 * in case we had to change ds_fsid_guid when we opened it,
2140 * sync it out now.
2141 */
2142 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2143 ds->ds_phys->ds_fsid_guid = ds->ds_fsid_guid;
2144
2145 dsl_dir_dirty(ds->ds_dir, tx);
428870ff 2146 dmu_objset_sync(ds->ds_objset, zio, tx);
34dc7c2f
BB		 2147}
2148
2149void
2150dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2151{
2152 uint64_t refd, avail, uobjs, aobjs;
2153
2154 dsl_dir_stats(ds->ds_dir, nv);
2155
2156 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
2157 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
2158 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
2159
2160 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2161 ds->ds_phys->ds_creation_time);
2162 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2163 ds->ds_phys->ds_creation_txg);
2164 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2165 ds->ds_quota);
2166 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2167 ds->ds_reserved);
b128c09f
BB		 2168 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2169 ds->ds_phys->ds_guid);
428870ff
BB		 2170 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2171 ds->ds_phys->ds_unique_bytes);
2172 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2173 ds->ds_object);
2174 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2175 ds->ds_userrefs);
45d1cae3
BB		 2176 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2177 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
34dc7c2f
BB		 2178
2179 if (ds->ds_phys->ds_next_snap_obj) {
2180 /*
2181 * This is a snapshot; override the dd's space used with
2182 * our unique space and compression ratio.
2183 */
2184 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2185 ds->ds_phys->ds_unique_bytes);
2186 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2187 ds->ds_phys->ds_compressed_bytes == 0 ? 100 :
2188 (ds->ds_phys->ds_uncompressed_bytes * 100 /
2189 ds->ds_phys->ds_compressed_bytes));
2190 }
2191}
2192
2193void
2194dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2195{
2196 stat->dds_creation_txg = ds->ds_phys->ds_creation_txg;
2197 stat->dds_inconsistent = ds->ds_phys->ds_flags & DS_FLAG_INCONSISTENT;
2198 stat->dds_guid = ds->ds_phys->ds_guid;
2199 if (ds->ds_phys->ds_next_snap_obj) {
2200 stat->dds_is_snapshot = B_TRUE;
2201 stat->dds_num_clones = ds->ds_phys->ds_num_children - 1;
fb5f0bc8
BB		 2202 } else {
2203 stat->dds_is_snapshot = B_FALSE;
2204 stat->dds_num_clones = 0;
34dc7c2f
BB		 2205 }
2206
2207 /* clone origin is really a dsl_dir thing... */
2208 rw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER);
b128c09f 2209 if (dsl_dir_is_clone(ds->ds_dir)) {
34dc7c2f
BB		 2210 dsl_dataset_t *ods;
2211
b128c09f
BB		 2212 VERIFY(0 == dsl_dataset_get_ref(ds->ds_dir->dd_pool,
2213 ds->ds_dir->dd_phys->dd_origin_obj, FTAG, &ods));
34dc7c2f 2214 dsl_dataset_name(ods, stat->dds_origin);
b128c09f 2215 dsl_dataset_drop_ref(ods, FTAG);
fb5f0bc8
BB		 2216 } else {
2217 stat->dds_origin[0] = '\0';
34dc7c2f
BB		 2218 }
2219 rw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock);
2220}
2221
2222uint64_t
2223dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2224{
2225 return (ds->ds_fsid_guid);
2226}
2227
2228void
2229dsl_dataset_space(dsl_dataset_t *ds,
2230 uint64_t *refdbytesp, uint64_t *availbytesp,
2231 uint64_t *usedobjsp, uint64_t *availobjsp)
2232{
2233 *refdbytesp = ds->ds_phys->ds_used_bytes;
2234 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2235 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes)
2236 *availbytesp += ds->ds_reserved - ds->ds_phys->ds_unique_bytes;
2237 if (ds->ds_quota != 0) {
2238 /*
2239 * Adjust available bytes according to refquota
2240 */
2241 if (*refdbytesp < ds->ds_quota)
2242 *availbytesp = MIN(*availbytesp,
2243 ds->ds_quota - *refdbytesp);
2244 else
2245 *availbytesp = 0;
2246 }
2247 *usedobjsp = ds->ds_phys->ds_bp.blk_fill;
2248 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
2249}
2250
2251boolean_t
2252dsl_dataset_modified_since_lastsnap(dsl_dataset_t *ds)
2253{
2254 dsl_pool_t *dp = ds->ds_dir->dd_pool;
2255
2256 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
2257 dsl_pool_sync_context(dp));
2258 if (ds->ds_prev == NULL)
2259 return (B_FALSE);
2260 if (ds->ds_phys->ds_bp.blk_birth >
572e2857
BB		 2261 ds->ds_prev->ds_phys->ds_creation_txg) {
2262 objset_t *os, *os_prev;
2263 /*
2264 * It may be that only the ZIL differs, because it was
2265 * reset in the head. Don't count that as being
2266 * modified.
2267 */
2268 if (dmu_objset_from_ds(ds, &os) != 0)
2269 return (B_TRUE);
2270 if (dmu_objset_from_ds(ds->ds_prev, &os_prev) != 0)
2271 return (B_TRUE);
2272 return (bcmp(&os->os_phys->os_meta_dnode,
2273 &os_prev->os_phys->os_meta_dnode,
2274 sizeof (os->os_phys->os_meta_dnode)) != 0);
2275 }
34dc7c2f
BB		 2276 return (B_FALSE);
2277}
2278
2279/* ARGSUSED */
2280static int
2281dsl_dataset_snapshot_rename_check(void *arg1, void *arg2, dmu_tx_t *tx)
2282{
2283 dsl_dataset_t *ds = arg1;
2284 char *newsnapname = arg2;
2285 dsl_dir_t *dd = ds->ds_dir;
34dc7c2f
BB		 2286 dsl_dataset_t *hds;
2287 uint64_t val;
2288 int err;
2289
b128c09f
BB		 2290 err = dsl_dataset_hold_obj(dd->dd_pool,
2291 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds);
34dc7c2f
BB		 2292 if (err)
2293 return (err);
2294
2295 /* new name better not be in use */
b128c09f
BB		 2296 err = dsl_dataset_snap_lookup(hds, newsnapname, &val);
2297 dsl_dataset_rele(hds, FTAG);
34dc7c2f
BB		 2298
2299 if (err == 0)
2300 err = EEXIST;
2301 else if (err == ENOENT)
2302 err = 0;
2303
2304 /* dataset name + 1 for the "@" + the new snapshot name must fit */
2305 if (dsl_dir_namelen(ds->ds_dir) + 1 + strlen(newsnapname) >= MAXNAMELEN)
2306 err = ENAMETOOLONG;
2307
2308 return (err);
2309}
2310
2311static void
428870ff 2312dsl_dataset_snapshot_rename_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2313{
2314 dsl_dataset_t *ds = arg1;
2315 const char *newsnapname = arg2;
2316 dsl_dir_t *dd = ds->ds_dir;
2317 objset_t *mos = dd->dd_pool->dp_meta_objset;
2318 dsl_dataset_t *hds;
2319 int err;
2320
2321 ASSERT(ds->ds_phys->ds_next_snap_obj != 0);
2322
b128c09f
BB		 2323 VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool,
2324 dd->dd_phys->dd_head_dataset_obj, FTAG, &hds));
34dc7c2f
BB		 2325
2326 VERIFY(0 == dsl_dataset_get_snapname(ds));
b128c09f 2327 err = dsl_dataset_snap_remove(hds, ds->ds_snapname, tx);
34dc7c2f
BB		 2328 ASSERT3U(err, ==, 0);
2329 mutex_enter(&ds->ds_lock);
2330 (void) strcpy(ds->ds_snapname, newsnapname);
2331 mutex_exit(&ds->ds_lock);
2332 err = zap_add(mos, hds->ds_phys->ds_snapnames_zapobj,
2333 ds->ds_snapname, 8, 1, &ds->ds_object, tx);
2334 ASSERT3U(err, ==, 0);
2335
428870ff
BB		 2336 spa_history_log_internal(LOG_DS_RENAME, dd->dd_pool->dp_spa, tx,
2337 "dataset = %llu", ds->ds_object);
b128c09f 2338 dsl_dataset_rele(hds, FTAG);
34dc7c2f
BB		 2339}
2340
2341struct renamesnaparg {
2342 dsl_sync_task_group_t *dstg;
2343 char failed[MAXPATHLEN];
2344 char *oldsnap;
2345 char *newsnap;
2346};
2347
2348static int
428870ff 2349dsl_snapshot_rename_one(const char *name, void *arg)
34dc7c2f
BB		 2350{
2351 struct renamesnaparg *ra = arg;
2352 dsl_dataset_t *ds = NULL;
428870ff 2353 char *snapname;
34dc7c2f
BB		 2354 int err;
2355
428870ff
BB		 2356 snapname = kmem_asprintf("%s@%s", name, ra->oldsnap);
2357 (void) strlcpy(ra->failed, snapname, sizeof (ra->failed));
34dc7c2f
BB		 2358
2359 /*
2360 * For recursive snapshot renames the parent won't be changing
2361 * so we just pass name for both the to/from argument.
2362 */
428870ff
BB		 2363 err = zfs_secpolicy_rename_perms(snapname, snapname, CRED());
2364 if (err != 0) {
2365 strfree(snapname);
2366 return (err == ENOENT ? 0 : err);
34dc7c2f
BB		 2367 }
2368
2369#ifdef _KERNEL
b128c09f
BB		 2370 /*
2371 * For all filesystems undergoing rename, we'll need to unmount it.
2372 */
428870ff 2373 (void) zfs_unmount_snap(snapname, NULL);
34dc7c2f 2374#endif
428870ff
BB		 2375 err = dsl_dataset_hold(snapname, ra->dstg, &ds);
2376 strfree(snapname);
2377 if (err != 0)
2378 return (err == ENOENT ? 0 : err);
34dc7c2f
BB		 2379
2380 dsl_sync_task_create(ra->dstg, dsl_dataset_snapshot_rename_check,
2381 dsl_dataset_snapshot_rename_sync, ds, ra->newsnap, 0);
2382
2383 return (0);
2384}
2385
2386static int
2387dsl_recursive_rename(char *oldname, const char *newname)
2388{
2389 int err;
2390 struct renamesnaparg *ra;
2391 dsl_sync_task_t *dst;
2392 spa_t *spa;
2393 char *cp, *fsname = spa_strdup(oldname);
428870ff 2394 int len = strlen(oldname) + 1;
34dc7c2f
BB		 2395
2396 /* truncate the snapshot name to get the fsname */
2397 cp = strchr(fsname, '@');
2398 *cp = '\0';
2399
2400 err = spa_open(fsname, &spa, FTAG);
2401 if (err) {
428870ff 2402 kmem_free(fsname, len);
34dc7c2f
BB		 2403 return (err);
2404 }
2405 ra = kmem_alloc(sizeof (struct renamesnaparg), KM_SLEEP);
2406 ra->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
2407
2408 ra->oldsnap = strchr(oldname, '@') + 1;
2409 ra->newsnap = strchr(newname, '@') + 1;
2410 *ra->failed = '\0';
2411
2412 err = dmu_objset_find(fsname, dsl_snapshot_rename_one, ra,
2413 DS_FIND_CHILDREN);
428870ff 2414 kmem_free(fsname, len);
34dc7c2f
BB		 2415
2416 if (err == 0) {
2417 err = dsl_sync_task_group_wait(ra->dstg);
2418 }
2419
2420 for (dst = list_head(&ra->dstg->dstg_tasks); dst;
2421 dst = list_next(&ra->dstg->dstg_tasks, dst)) {
2422 dsl_dataset_t *ds = dst->dst_arg1;
2423 if (dst->dst_err) {
2424 dsl_dir_name(ds->ds_dir, ra->failed);
428870ff
BB		 2425 (void) strlcat(ra->failed, "@", sizeof (ra->failed));
2426 (void) strlcat(ra->failed, ra->newsnap,
2427 sizeof (ra->failed));
34dc7c2f 2428 }
b128c09f 2429 dsl_dataset_rele(ds, ra->dstg);
34dc7c2f
BB		 2430 }
2431
2432 if (err)
428870ff 2433 (void) strlcpy(oldname, ra->failed, sizeof (ra->failed));
34dc7c2f
BB		 2434
2435 dsl_sync_task_group_destroy(ra->dstg);
2436 kmem_free(ra, sizeof (struct renamesnaparg));
2437 spa_close(spa, FTAG);
2438 return (err);
2439}
2440
2441static int
428870ff 2442dsl_valid_rename(const char *oldname, void *arg)
34dc7c2f
BB		 2443{
2444 int delta = *(int *)arg;
2445
2446 if (strlen(oldname) + delta >= MAXNAMELEN)
2447 return (ENAMETOOLONG);
2448
2449 return (0);
2450}
2451
2452#pragma weak dmu_objset_rename = dsl_dataset_rename
2453int
b128c09f 2454dsl_dataset_rename(char *oldname, const char *newname, boolean_t recursive)
34dc7c2f
BB		 2455{
2456 dsl_dir_t *dd;
2457 dsl_dataset_t *ds;
2458 const char *tail;
2459 int err;
2460
2461 err = dsl_dir_open(oldname, FTAG, &dd, &tail);
2462 if (err)
2463 return (err);
428870ff 2464
34dc7c2f
BB		 2465 if (tail == NULL) {
2466 int delta = strlen(newname) - strlen(oldname);
2467
b128c09f 2468 /* if we're growing, validate child name lengths */
34dc7c2f
BB		 2469 if (delta > 0)
2470 err = dmu_objset_find(oldname, dsl_valid_rename,
2471 &delta, DS_FIND_CHILDREN | DS_FIND_SNAPSHOTS);
2472
428870ff 2473 if (err == 0)
34dc7c2f
BB		 2474 err = dsl_dir_rename(dd, newname);
2475 dsl_dir_close(dd, FTAG);
2476 return (err);
2477 }
428870ff 2478
34dc7c2f 2479 if (tail[0] != '@') {
428870ff 2480 /* the name ended in a nonexistent component */
34dc7c2f
BB		 2481 dsl_dir_close(dd, FTAG);
2482 return (ENOENT);
2483 }
2484
2485 dsl_dir_close(dd, FTAG);
2486
2487 /* new name must be snapshot in same filesystem */
2488 tail = strchr(newname, '@');
2489 if (tail == NULL)
2490 return (EINVAL);
2491 tail++;
2492 if (strncmp(oldname, newname, tail - newname) != 0)
2493 return (EXDEV);
2494
2495 if (recursive) {
2496 err = dsl_recursive_rename(oldname, newname);
2497 } else {
b128c09f 2498 err = dsl_dataset_hold(oldname, FTAG, &ds);
34dc7c2f
BB		 2499 if (err)
2500 return (err);
2501
2502 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
2503 dsl_dataset_snapshot_rename_check,
2504 dsl_dataset_snapshot_rename_sync, ds, (char *)tail, 1);
2505
b128c09f 2506 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 2507 }
2508
2509 return (err);
2510}
2511
b128c09f
BB		 2512struct promotenode {
2513 list_node_t link;
2514 dsl_dataset_t *ds;
2515};
2516
34dc7c2f 2517struct promotearg {
b128c09f 2518 list_t shared_snaps, origin_snaps, clone_snaps;
428870ff 2519 dsl_dataset_t *origin_origin;
b128c09f 2520 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
428870ff 2521 char *err_ds;
34dc7c2f
BB		 2522};
2523
b128c09f 2524static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
428870ff 2525static boolean_t snaplist_unstable(list_t *l);
b128c09f 2526
34dc7c2f
BB		 2527static int
2528dsl_dataset_promote_check(void *arg1, void *arg2, dmu_tx_t *tx)
2529{
2530 dsl_dataset_t *hds = arg1;
2531 struct promotearg *pa = arg2;
b128c09f
BB		 2532 struct promotenode *snap = list_head(&pa->shared_snaps);
2533 dsl_dataset_t *origin_ds = snap->ds;
34dc7c2f 2534 int err;
428870ff 2535 uint64_t unused;
34dc7c2f 2536
b128c09f
BB		 2537 /* Check that it is a real clone */
2538 if (!dsl_dir_is_clone(hds->ds_dir))
34dc7c2f
BB		 2539 return (EINVAL);
2540
2541 /* Since this is so expensive, don't do the preliminary check */
2542 if (!dmu_tx_is_syncing(tx))
2543 return (0);
2544
b128c09f
BB		 2545 if (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE)
2546 return (EXDEV);
34dc7c2f
BB		 2547
2548 /* compute origin's new unique space */
b128c09f
BB		 2549 snap = list_tail(&pa->clone_snaps);
2550 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
428870ff
BB		 2551 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2552 origin_ds->ds_phys->ds_prev_snap_txg, UINT64_MAX,
2553 &pa->unique, &unused, &unused);
34dc7c2f 2554
b128c09f
BB		 2555 /*
2556 * Walk the snapshots that we are moving
2557 *
2558 * Compute space to transfer. Consider the incremental changes
2559 * to used for each snapshot:
2560 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2561 * So each snapshot gave birth to:
2562 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2563 * So a sequence would look like:
2564 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2565 * Which simplifies to:
2566 * uN + kN + kN-1 + ... + k1 + k0
2567 * Note however, if we stop before we reach the ORIGIN we get:
2568 * uN + kN + kN-1 + ... + kM - uM-1
2569 */
2570 pa->used = origin_ds->ds_phys->ds_used_bytes;
2571 pa->comp = origin_ds->ds_phys->ds_compressed_bytes;
2572 pa->uncomp = origin_ds->ds_phys->ds_uncompressed_bytes;
2573 for (snap = list_head(&pa->shared_snaps); snap;
2574 snap = list_next(&pa->shared_snaps, snap)) {
34dc7c2f 2575 uint64_t val, dlused, dlcomp, dluncomp;
b128c09f 2576 dsl_dataset_t *ds = snap->ds;
34dc7c2f
BB		 2577
2578 /* Check that the snapshot name does not conflict */
b128c09f
BB		 2579 VERIFY(0 == dsl_dataset_get_snapname(ds));
2580 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
428870ff
BB		 2581 if (err == 0) {
2582 err = EEXIST;
2583 goto out;
2584 }
b128c09f 2585 if (err != ENOENT)
428870ff 2586 goto out;
34dc7c2f 2587
b128c09f 2588 /* The very first snapshot does not have a deadlist */
34dc7c2f 2589 if (ds->ds_phys->ds_prev_snap_obj == 0)
b128c09f 2590 continue;
34dc7c2f 2591
428870ff
BB		 2592 dsl_deadlist_space(&ds->ds_deadlist,
2593 &dlused, &dlcomp, &dluncomp);
b128c09f
BB		 2594 pa->used += dlused;
2595 pa->comp += dlcomp;
2596 pa->uncomp += dluncomp;
2597 }
34dc7c2f 2598
b128c09f
BB		 2599 /*
2600 * If we are a clone of a clone then we never reached ORIGIN,
2601 * so we need to subtract out the clone origin's used space.
2602 */
2603 if (pa->origin_origin) {
2604 pa->used -= pa->origin_origin->ds_phys->ds_used_bytes;
2605 pa->comp -= pa->origin_origin->ds_phys->ds_compressed_bytes;
2606 pa->uncomp -= pa->origin_origin->ds_phys->ds_uncompressed_bytes;
34dc7c2f
BB		 2607 }
2608
2609 /* Check that there is enough space here */
b128c09f
BB		 2610 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2611 pa->used);
2612 if (err)
2613 return (err);
34dc7c2f 2614
b128c09f
BB		 2615 /*
2616 * Compute the amounts of space that will be used by snapshots
2617 * after the promotion (for both origin and clone). For each,
2618 * it is the amount of space that will be on all of their
2619 * deadlists (that was not born before their new origin).
2620 */
2621 if (hds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2622 uint64_t space;
2623
2624 /*
2625 * Note, typically this will not be a clone of a clone,
428870ff
BB		 2626 * so dd_origin_txg will be < TXG_INITIAL, so
2627 * these snaplist_space() -> dsl_deadlist_space_range()
b128c09f
BB		 2628 * calls will be fast because they do not have to
2629 * iterate over all bps.
2630 */
2631 snap = list_head(&pa->origin_snaps);
2632 err = snaplist_space(&pa->shared_snaps,
428870ff 2633 snap->ds->ds_dir->dd_origin_txg, &pa->cloneusedsnap);
b128c09f
BB		 2634 if (err)
2635 return (err);
2636
2637 err = snaplist_space(&pa->clone_snaps,
428870ff 2638 snap->ds->ds_dir->dd_origin_txg, &space);
b128c09f
BB		 2639 if (err)
2640 return (err);
2641 pa->cloneusedsnap += space;
2642 }
2643 if (origin_ds->ds_dir->dd_phys->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2644 err = snaplist_space(&pa->origin_snaps,
2645 origin_ds->ds_phys->ds_creation_txg, &pa->originusedsnap);
2646 if (err)
2647 return (err);
2648 }
2649
2650 return (0);
428870ff
BB		 2651out:
2652 pa->err_ds = snap->ds->ds_snapname;
2653 return (err);
34dc7c2f
BB		 2654}
2655
2656static void
428870ff 2657dsl_dataset_promote_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 2658{
2659 dsl_dataset_t *hds = arg1;
2660 struct promotearg *pa = arg2;
b128c09f
BB		 2661 struct promotenode *snap = list_head(&pa->shared_snaps);
2662 dsl_dataset_t *origin_ds = snap->ds;
2663 dsl_dataset_t *origin_head;
34dc7c2f
BB		 2664 dsl_dir_t *dd = hds->ds_dir;
2665 dsl_pool_t *dp = hds->ds_dir->dd_pool;
2666 dsl_dir_t *odd = NULL;
b128c09f
BB		 2667 uint64_t oldnext_obj;
2668 int64_t delta;
34dc7c2f 2669
34dc7c2f
BB		 2670 ASSERT(0 == (hds->ds_phys->ds_flags & DS_FLAG_NOPROMOTE));
2671
b128c09f
BB		 2672 snap = list_head(&pa->origin_snaps);
2673 origin_head = snap->ds;
2674
34dc7c2f
BB		 2675 /*
2676 * We need to explicitly open odd, since origin_ds's dd will be
2677 * changing.
2678 */
2679 VERIFY(0 == dsl_dir_open_obj(dp, origin_ds->ds_dir->dd_object,
2680 NULL, FTAG, &odd));
2681
b128c09f
BB		 2682 /* change origin's next snap */
2683 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2684 oldnext_obj = origin_ds->ds_phys->ds_next_snap_obj;
2685 snap = list_tail(&pa->clone_snaps);
2686 ASSERT3U(snap->ds->ds_phys->ds_prev_snap_obj, ==, origin_ds->ds_object);
2687 origin_ds->ds_phys->ds_next_snap_obj = snap->ds->ds_object;
2688
2689 /* change the origin's next clone */
2690 if (origin_ds->ds_phys->ds_next_clones_obj) {
428870ff 2691 remove_from_next_clones(origin_ds, snap->ds->ds_object, tx);
b128c09f
BB		 2692 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2693 origin_ds->ds_phys->ds_next_clones_obj,
2694 oldnext_obj, tx));
2695 }
2696
2697 /* change origin */
2698 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2699 ASSERT3U(dd->dd_phys->dd_origin_obj, ==, origin_ds->ds_object);
2700 dd->dd_phys->dd_origin_obj = odd->dd_phys->dd_origin_obj;
428870ff 2701 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
b128c09f
BB		 2702 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2703 odd->dd_phys->dd_origin_obj = origin_ds->ds_object;
428870ff
BB		 2704 origin_head->ds_dir->dd_origin_txg =
2705 origin_ds->ds_phys->ds_creation_txg;
2706
2707 /* change dd_clone entries */
2708 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2709 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2710 odd->dd_phys->dd_clones, hds->ds_object, tx));
2711 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2712 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2713 hds->ds_object, tx));
2714
2715 VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
2716 pa->origin_origin->ds_dir->dd_phys->dd_clones,
2717 origin_head->ds_object, tx));
2718 if (dd->dd_phys->dd_clones == 0) {
2719 dd->dd_phys->dd_clones = zap_create(dp->dp_meta_objset,
2720 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
2721 }
2722 VERIFY3U(0, ==, zap_add_int(dp->dp_meta_objset,
2723 dd->dd_phys->dd_clones, origin_head->ds_object, tx));
2724
2725 }
34dc7c2f 2726
b128c09f
BB		 2727 /* move snapshots to this dir */
2728 for (snap = list_head(&pa->shared_snaps); snap;
2729 snap = list_next(&pa->shared_snaps, snap)) {
2730 dsl_dataset_t *ds = snap->ds;
2731
2732 /* unregister props as dsl_dir is changing */
428870ff
BB		 2733 if (ds->ds_objset) {
2734 dmu_objset_evict(ds->ds_objset);
2735 ds->ds_objset = NULL;
b128c09f 2736 }
34dc7c2f 2737 /* move snap name entry */
b128c09f
BB		 2738 VERIFY(0 == dsl_dataset_get_snapname(ds));
2739 VERIFY(0 == dsl_dataset_snap_remove(origin_head,
2740 ds->ds_snapname, tx));
34dc7c2f
BB		 2741 VERIFY(0 == zap_add(dp->dp_meta_objset,
2742 hds->ds_phys->ds_snapnames_zapobj, ds->ds_snapname,
2743 8, 1, &ds->ds_object, tx));
428870ff 2744
34dc7c2f
BB		 2745 /* change containing dsl_dir */
2746 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2747 ASSERT3U(ds->ds_phys->ds_dir_obj, ==, odd->dd_object);
2748 ds->ds_phys->ds_dir_obj = dd->dd_object;
2749 ASSERT3P(ds->ds_dir, ==, odd);
2750 dsl_dir_close(ds->ds_dir, ds);
2751 VERIFY(0 == dsl_dir_open_obj(dp, dd->dd_object,
2752 NULL, ds, &ds->ds_dir));
2753
428870ff
BB		 2754 /* move any clone references */
2755 if (ds->ds_phys->ds_next_clones_obj &&
2756 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2757 zap_cursor_t zc;
2758 zap_attribute_t za;
2759
2760 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2761 ds->ds_phys->ds_next_clones_obj);
2762 zap_cursor_retrieve(&zc, &za) == 0;
2763 zap_cursor_advance(&zc)) {
2764 dsl_dataset_t *cnds;
2765 uint64_t o;
2766
2767 if (za.za_first_integer == oldnext_obj) {
2768 /*
2769 * We've already moved the
2770 * origin's reference.
2771 */
2772 continue;
2773 }
2774
2775 VERIFY3U(0, ==, dsl_dataset_hold_obj(dp,
2776 za.za_first_integer, FTAG, &cnds));
2777 o = cnds->ds_dir->dd_phys->dd_head_dataset_obj;
2778
2779 VERIFY3U(zap_remove_int(dp->dp_meta_objset,
2780 odd->dd_phys->dd_clones, o, tx), ==, 0);
2781 VERIFY3U(zap_add_int(dp->dp_meta_objset,
2782 dd->dd_phys->dd_clones, o, tx), ==, 0);
2783 dsl_dataset_rele(cnds, FTAG);
2784 }
2785 zap_cursor_fini(&zc);
2786 }
2787
34dc7c2f 2788 ASSERT3U(dsl_prop_numcb(ds), ==, 0);
34dc7c2f 2789 }
34dc7c2f 2790
b128c09f
BB		 2791 /*
2792 * Change space accounting.
2793 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2794 * both be valid, or both be 0 (resulting in delta == 0). This
2795 * is true for each of {clone,origin} independently.
2796 */
34dc7c2f 2797
b128c09f
BB		 2798 delta = pa->cloneusedsnap -
2799 dd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2800 ASSERT3S(delta, >=, 0);
2801 ASSERT3U(pa->used, >=, delta);
2802 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2803 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2804 pa->used - delta, pa->comp, pa->uncomp, tx);
2805
2806 delta = pa->originusedsnap -
2807 odd->dd_phys->dd_used_breakdown[DD_USED_SNAP];
2808 ASSERT3S(delta, <=, 0);
2809 ASSERT3U(pa->used, >=, -delta);
2810 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2811 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2812 -pa->used - delta, -pa->comp, -pa->uncomp, tx);
34dc7c2f 2813
34dc7c2f
BB		 2814 origin_ds->ds_phys->ds_unique_bytes = pa->unique;
2815
2816 /* log history record */
428870ff
BB		 2817 spa_history_log_internal(LOG_DS_PROMOTE, dd->dd_pool->dp_spa, tx,
2818 "dataset = %llu", hds->ds_object);
34dc7c2f
BB		 2819
2820 dsl_dir_close(odd, FTAG);
34dc7c2f
BB		 2821}
2822
b128c09f
BB		 2823static char *snaplist_tag = "snaplist";
2824/*
2825 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2826 * (exclusive) and last_obj (inclusive). The list will be in reverse
2827 * order (last_obj will be the list_head()). If first_obj == 0, do all
2828 * snapshots back to this dataset's origin.
2829 */
2830static int
2831snaplist_make(dsl_pool_t *dp, boolean_t own,
2832 uint64_t first_obj, uint64_t last_obj, list_t *l)
2833{
2834 uint64_t obj = last_obj;
2835
2836 ASSERT(RW_LOCK_HELD(&dp->dp_config_rwlock));
2837
2838 list_create(l, sizeof (struct promotenode),
2839 offsetof(struct promotenode, link));
2840
2841 while (obj != first_obj) {
2842 dsl_dataset_t *ds;
2843 struct promotenode *snap;
2844 int err;
2845
2846 if (own) {
2847 err = dsl_dataset_own_obj(dp, obj,
2848 0, snaplist_tag, &ds);
2849 if (err == 0)
2850 dsl_dataset_make_exclusive(ds, snaplist_tag);
2851 } else {
2852 err = dsl_dataset_hold_obj(dp, obj, snaplist_tag, &ds);
2853 }
2854 if (err == ENOENT) {
2855 /* lost race with snapshot destroy */
2856 struct promotenode *last = list_tail(l);
2857 ASSERT(obj != last->ds->ds_phys->ds_prev_snap_obj);
2858 obj = last->ds->ds_phys->ds_prev_snap_obj;
2859 continue;
2860 } else if (err) {
2861 return (err);
2862 }
2863
2864 if (first_obj == 0)
2865 first_obj = ds->ds_dir->dd_phys->dd_origin_obj;
2866
2867 snap = kmem_alloc(sizeof (struct promotenode), KM_SLEEP);
2868 snap->ds = ds;
2869 list_insert_tail(l, snap);
2870 obj = ds->ds_phys->ds_prev_snap_obj;
2871 }
2872
2873 return (0);
2874}
2875
2876static int
2877snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2878{
2879 struct promotenode *snap;
2880
2881 *spacep = 0;
2882 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
428870ff
BB		 2883 uint64_t used, comp, uncomp;
2884 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2885 mintxg, UINT64_MAX, &used, &comp, &uncomp);
b128c09f
BB		 2886 *spacep += used;
2887 }
2888 return (0);
2889}
2890
2891static void
2892snaplist_destroy(list_t *l, boolean_t own)
2893{
2894 struct promotenode *snap;
2895
9babb374 2896 if (!l || !list_link_active(&l->list_head))
b128c09f
BB		 2897 return;
2898
2899 while ((snap = list_tail(l)) != NULL) {
2900 list_remove(l, snap);
2901 if (own)
2902 dsl_dataset_disown(snap->ds, snaplist_tag);
2903 else
2904 dsl_dataset_rele(snap->ds, snaplist_tag);
2905 kmem_free(snap, sizeof (struct promotenode));
2906 }
2907 list_destroy(l);
2908}
2909
2910/*
2911 * Promote a clone. Nomenclature note:
2912 * "clone" or "cds": the original clone which is being promoted
2913 * "origin" or "ods": the snapshot which is originally clone's origin
2914 * "origin head" or "ohds": the dataset which is the head
2915 * (filesystem/volume) for the origin
2916 * "origin origin": the origin of the origin's filesystem (typically
2917 * NULL, indicating that the clone is not a clone of a clone).
2918 */
34dc7c2f 2919int
428870ff 2920dsl_dataset_promote(const char *name, char *conflsnap)
34dc7c2f
BB		 2921{
2922 dsl_dataset_t *ds;
b128c09f
BB		 2923 dsl_dir_t *dd;
2924 dsl_pool_t *dp;
34dc7c2f 2925 dmu_object_info_t doi;
b128c09f
BB		 2926 struct promotearg pa = { 0 };
2927 struct promotenode *snap;
2928 int err;
34dc7c2f 2929
b128c09f 2930 err = dsl_dataset_hold(name, FTAG, &ds);
34dc7c2f
BB		 2931 if (err)
2932 return (err);
b128c09f
BB		 2933 dd = ds->ds_dir;
2934 dp = dd->dd_pool;
34dc7c2f 2935
b128c09f 2936 err = dmu_object_info(dp->dp_meta_objset,
34dc7c2f
BB		 2937 ds->ds_phys->ds_snapnames_zapobj, &doi);
2938 if (err) {
b128c09f 2939 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 2940 return (err);
2941 }
2942
b128c09f
BB		 2943 if (dsl_dataset_is_snapshot(ds) || dd->dd_phys->dd_origin_obj == 0) {
2944 dsl_dataset_rele(ds, FTAG);
2945 return (EINVAL);
2946 }
2947
2948 /*
2949 * We are going to inherit all the snapshots taken before our
2950 * origin (i.e., our new origin will be our parent's origin).
2951 * Take ownership of them so that we can rename them into our
2952 * namespace.
2953 */
2954 rw_enter(&dp->dp_config_rwlock, RW_READER);
2955
2956 err = snaplist_make(dp, B_TRUE, 0, dd->dd_phys->dd_origin_obj,
2957 &pa.shared_snaps);
2958 if (err != 0)
2959 goto out;
2960
2961 err = snaplist_make(dp, B_FALSE, 0, ds->ds_object, &pa.clone_snaps);
2962 if (err != 0)
2963 goto out;
2964
2965 snap = list_head(&pa.shared_snaps);
2966 ASSERT3U(snap->ds->ds_object, ==, dd->dd_phys->dd_origin_obj);
2967 err = snaplist_make(dp, B_FALSE, dd->dd_phys->dd_origin_obj,
2968 snap->ds->ds_dir->dd_phys->dd_head_dataset_obj, &pa.origin_snaps);
2969 if (err != 0)
2970 goto out;
2971
428870ff
BB		 2972 if (snap->ds->ds_dir->dd_phys->dd_origin_obj != 0) {
2973 err = dsl_dataset_hold_obj(dp,
b128c09f 2974 snap->ds->ds_dir->dd_phys->dd_origin_obj,
428870ff 2975 FTAG, &pa.origin_origin);
b128c09f
BB		 2976 if (err != 0)
2977 goto out;
2978 }
2979
2980out:
2981 rw_exit(&dp->dp_config_rwlock);
2982
34dc7c2f
BB		 2983 /*
2984 * Add in 128x the snapnames zapobj size, since we will be moving
2985 * a bunch of snapnames to the promoted ds, and dirtying their
2986 * bonus buffers.
2987 */
b128c09f
BB		 2988 if (err == 0) {
2989 err = dsl_sync_task_do(dp, dsl_dataset_promote_check,
2990 dsl_dataset_promote_sync, ds, &pa,
428870ff
BB		 2991 2 + 2 * doi.doi_physical_blocks_512);
2992 if (err && pa.err_ds && conflsnap)
2993 (void) strncpy(conflsnap, pa.err_ds, MAXNAMELEN);
b128c09f
BB		 2994 }
2995
2996 snaplist_destroy(&pa.shared_snaps, B_TRUE);
2997 snaplist_destroy(&pa.clone_snaps, B_FALSE);
2998 snaplist_destroy(&pa.origin_snaps, B_FALSE);
2999 if (pa.origin_origin)
428870ff 3000 dsl_dataset_rele(pa.origin_origin, FTAG);
b128c09f 3001 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3002 return (err);
3003}
3004
3005struct cloneswaparg {
3006 dsl_dataset_t *cds; /* clone dataset */
3007 dsl_dataset_t *ohds; /* origin's head dataset */
3008 boolean_t force;
3009 int64_t unused_refres_delta; /* change in unconsumed refreservation */
3010};
3011
3012/* ARGSUSED */
3013static int
3014dsl_dataset_clone_swap_check(void *arg1, void *arg2, dmu_tx_t *tx)
3015{
3016 struct cloneswaparg *csa = arg1;
3017
3018 /* they should both be heads */
3019 if (dsl_dataset_is_snapshot(csa->cds) ||
3020 dsl_dataset_is_snapshot(csa->ohds))
3021 return (EINVAL);
3022
3023 /* the branch point should be just before them */
3024 if (csa->cds->ds_prev != csa->ohds->ds_prev)
3025 return (EINVAL);
3026
428870ff
BB		 3027 /* cds should be the clone (unless they are unrelated) */
3028 if (csa->cds->ds_prev != NULL &&
3029 csa->cds->ds_prev != csa->cds->ds_dir->dd_pool->dp_origin_snap &&
3030 csa->ohds->ds_object !=
3031 csa->cds->ds_prev->ds_phys->ds_next_snap_obj)
34dc7c2f
BB		 3032 return (EINVAL);
3033
3034 /* the clone should be a child of the origin */
3035 if (csa->cds->ds_dir->dd_parent != csa->ohds->ds_dir)
3036 return (EINVAL);
3037
3038 /* ohds shouldn't be modified unless 'force' */
3039 if (!csa->force && dsl_dataset_modified_since_lastsnap(csa->ohds))
3040 return (ETXTBSY);
3041
3042 /* adjust amount of any unconsumed refreservation */
3043 csa->unused_refres_delta =
3044 (int64_t)MIN(csa->ohds->ds_reserved,
3045 csa->ohds->ds_phys->ds_unique_bytes) -
3046 (int64_t)MIN(csa->ohds->ds_reserved,
3047 csa->cds->ds_phys->ds_unique_bytes);
3048
3049 if (csa->unused_refres_delta > 0 &&
3050 csa->unused_refres_delta >
3051 dsl_dir_space_available(csa->ohds->ds_dir, NULL, 0, TRUE))
3052 return (ENOSPC);
3053
428870ff
BB		 3054 if (csa->ohds->ds_quota != 0 &&
3055 csa->cds->ds_phys->ds_unique_bytes > csa->ohds->ds_quota)
3056 return (EDQUOT);
3057
34dc7c2f
BB		 3058 return (0);
3059}
3060
3061/* ARGSUSED */
3062static void
428870ff 3063dsl_dataset_clone_swap_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3064{
3065 struct cloneswaparg *csa = arg1;
3066 dsl_pool_t *dp = csa->cds->ds_dir->dd_pool;
34dc7c2f
BB		 3067
3068 ASSERT(csa->cds->ds_reserved == 0);
428870ff
BB		 3069 ASSERT(csa->ohds->ds_quota == 0 ||
3070 csa->cds->ds_phys->ds_unique_bytes <= csa->ohds->ds_quota);
34dc7c2f
BB		 3071
3072 dmu_buf_will_dirty(csa->cds->ds_dbuf, tx);
3073 dmu_buf_will_dirty(csa->ohds->ds_dbuf, tx);
34dc7c2f 3074
428870ff
BB		 3075 if (csa->cds->ds_objset != NULL) {
3076 dmu_objset_evict(csa->cds->ds_objset);
3077 csa->cds->ds_objset = NULL;
34dc7c2f
BB		 3078 }
3079
428870ff
BB		 3080 if (csa->ohds->ds_objset != NULL) {
3081 dmu_objset_evict(csa->ohds->ds_objset);
3082 csa->ohds->ds_objset = NULL;
34dc7c2f
BB		 3083 }
3084
428870ff
BB		 3085 /*
3086 * Reset origin's unique bytes, if it exists.
3087 */
3088 if (csa->cds->ds_prev) {
3089 dsl_dataset_t *origin = csa->cds->ds_prev;
3090 uint64_t comp, uncomp;
3091
3092 dmu_buf_will_dirty(origin->ds_dbuf, tx);
3093 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3094 origin->ds_phys->ds_prev_snap_txg, UINT64_MAX,
3095 &origin->ds_phys->ds_unique_bytes, &comp, &uncomp);
3096 }
34dc7c2f
BB		 3097
3098 /* swap blkptrs */
3099 {
3100 blkptr_t tmp;
3101 tmp = csa->ohds->ds_phys->ds_bp;
3102 csa->ohds->ds_phys->ds_bp = csa->cds->ds_phys->ds_bp;
3103 csa->cds->ds_phys->ds_bp = tmp;
3104 }
3105
3106 /* set dd_*_bytes */
3107 {
3108 int64_t dused, dcomp, duncomp;
3109 uint64_t cdl_used, cdl_comp, cdl_uncomp;
3110 uint64_t odl_used, odl_comp, odl_uncomp;
3111
b128c09f
BB		 3112 ASSERT3U(csa->cds->ds_dir->dd_phys->
3113 dd_used_breakdown[DD_USED_SNAP], ==, 0);
3114
428870ff
BB		 3115 dsl_deadlist_space(&csa->cds->ds_deadlist,
3116 &cdl_used, &cdl_comp, &cdl_uncomp);
3117 dsl_deadlist_space(&csa->ohds->ds_deadlist,
3118 &odl_used, &odl_comp, &odl_uncomp);
b128c09f 3119
34dc7c2f
BB		 3120 dused = csa->cds->ds_phys->ds_used_bytes + cdl_used -
3121 (csa->ohds->ds_phys->ds_used_bytes + odl_used);
3122 dcomp = csa->cds->ds_phys->ds_compressed_bytes + cdl_comp -
3123 (csa->ohds->ds_phys->ds_compressed_bytes + odl_comp);
3124 duncomp = csa->cds->ds_phys->ds_uncompressed_bytes +
3125 cdl_uncomp -
3126 (csa->ohds->ds_phys->ds_uncompressed_bytes + odl_uncomp);
3127
b128c09f 3128 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_HEAD,
34dc7c2f 3129 dused, dcomp, duncomp, tx);
b128c09f 3130 dsl_dir_diduse_space(csa->cds->ds_dir, DD_USED_HEAD,
34dc7c2f 3131 -dused, -dcomp, -duncomp, tx);
b128c09f
BB		 3132
3133 /*
3134 * The difference in the space used by snapshots is the
3135 * difference in snapshot space due to the head's
3136 * deadlist (since that's the only thing that's
3137 * changing that affects the snapused).
3138 */
428870ff
BB		 3139 dsl_deadlist_space_range(&csa->cds->ds_deadlist,
3140 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3141 &cdl_used, &cdl_comp, &cdl_uncomp);
3142 dsl_deadlist_space_range(&csa->ohds->ds_deadlist,
3143 csa->ohds->ds_dir->dd_origin_txg, UINT64_MAX,
3144 &odl_used, &odl_comp, &odl_uncomp);
b128c09f
BB		 3145 dsl_dir_transfer_space(csa->ohds->ds_dir, cdl_used - odl_used,
3146 DD_USED_HEAD, DD_USED_SNAP, tx);
34dc7c2f
BB		 3147 }
3148
34dc7c2f
BB		 3149 /* swap ds_*_bytes */
3150 SWITCH64(csa->ohds->ds_phys->ds_used_bytes,
3151 csa->cds->ds_phys->ds_used_bytes);
3152 SWITCH64(csa->ohds->ds_phys->ds_compressed_bytes,
3153 csa->cds->ds_phys->ds_compressed_bytes);
3154 SWITCH64(csa->ohds->ds_phys->ds_uncompressed_bytes,
3155 csa->cds->ds_phys->ds_uncompressed_bytes);
3156 SWITCH64(csa->ohds->ds_phys->ds_unique_bytes,
3157 csa->cds->ds_phys->ds_unique_bytes);
3158
3159 /* apply any parent delta for change in unconsumed refreservation */
b128c09f
BB		 3160 dsl_dir_diduse_space(csa->ohds->ds_dir, DD_USED_REFRSRV,
3161 csa->unused_refres_delta, 0, 0, tx);
34dc7c2f 3162
428870ff
BB		 3163 /*
3164 * Swap deadlists.
3165 */
3166 dsl_deadlist_close(&csa->cds->ds_deadlist);
3167 dsl_deadlist_close(&csa->ohds->ds_deadlist);
34dc7c2f
BB		 3168 SWITCH64(csa->ohds->ds_phys->ds_deadlist_obj,
3169 csa->cds->ds_phys->ds_deadlist_obj);
428870ff
BB		 3170 dsl_deadlist_open(&csa->cds->ds_deadlist, dp->dp_meta_objset,
3171 csa->cds->ds_phys->ds_deadlist_obj);
3172 dsl_deadlist_open(&csa->ohds->ds_deadlist, dp->dp_meta_objset,
3173 csa->ohds->ds_phys->ds_deadlist_obj);
b128c09f 3174
428870ff 3175 dsl_scan_ds_clone_swapped(csa->ohds, csa->cds, tx);
34dc7c2f
BB		 3176}
3177
3178/*
428870ff
BB		 3179 * Swap 'clone' with its origin head datasets. Used at the end of "zfs
3180 * recv" into an existing fs to swizzle the file system to the new
3181 * version, and by "zfs rollback". Can also be used to swap two
3182 * independent head datasets if neither has any snapshots.
34dc7c2f
BB		 3183 */
3184int
3185dsl_dataset_clone_swap(dsl_dataset_t *clone, dsl_dataset_t *origin_head,
3186 boolean_t force)
3187{
3188 struct cloneswaparg csa;
b128c09f 3189 int error;
34dc7c2f 3190
b128c09f
BB		 3191 ASSERT(clone->ds_owner);
3192 ASSERT(origin_head->ds_owner);
3193retry:
572e2857
BB		 3194 /*
3195 * Need exclusive access for the swap. If we're swapping these
3196 * datasets back after an error, we already hold the locks.
3197 */
3198 if (!RW_WRITE_HELD(&clone->ds_rwlock))
3199 rw_enter(&clone->ds_rwlock, RW_WRITER);
3200 if (!RW_WRITE_HELD(&origin_head->ds_rwlock) &&
3201 !rw_tryenter(&origin_head->ds_rwlock, RW_WRITER)) {
b128c09f
BB		 3202 rw_exit(&clone->ds_rwlock);
3203 rw_enter(&origin_head->ds_rwlock, RW_WRITER);
3204 if (!rw_tryenter(&clone->ds_rwlock, RW_WRITER)) {
3205 rw_exit(&origin_head->ds_rwlock);
3206 goto retry;
3207 }
3208 }
34dc7c2f
BB		 3209 csa.cds = clone;
3210 csa.ohds = origin_head;
3211 csa.force = force;
b128c09f 3212 error = dsl_sync_task_do(clone->ds_dir->dd_pool,
34dc7c2f 3213 dsl_dataset_clone_swap_check,
b128c09f
BB		 3214 dsl_dataset_clone_swap_sync, &csa, NULL, 9);
3215 return (error);
34dc7c2f
BB		 3216}
3217
3218/*
3219 * Given a pool name and a dataset object number in that pool,
3220 * return the name of that dataset.
3221 */
3222int
3223dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3224{
3225 spa_t *spa;
3226 dsl_pool_t *dp;
b128c09f 3227 dsl_dataset_t *ds;
34dc7c2f
BB		 3228 int error;
3229
3230 if ((error = spa_open(pname, &spa, FTAG)) != 0)
3231 return (error);
3232 dp = spa_get_dsl(spa);
3233 rw_enter(&dp->dp_config_rwlock, RW_READER);
b128c09f
BB		 3234 if ((error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds)) == 0) {
3235 dsl_dataset_name(ds, buf);
3236 dsl_dataset_rele(ds, FTAG);
34dc7c2f 3237 }
34dc7c2f
BB		 3238 rw_exit(&dp->dp_config_rwlock);
3239 spa_close(spa, FTAG);
3240
b128c09f 3241 return (error);
34dc7c2f
BB		 3242}
3243
3244int
3245dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
b128c09f 3246 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
34dc7c2f
BB		 3247{
3248 int error = 0;
3249
3250 ASSERT3S(asize, >, 0);
3251
3252 /*
3253 * *ref_rsrv is the portion of asize that will come from any
3254 * unconsumed refreservation space.
3255 */
3256 *ref_rsrv = 0;
3257
3258 mutex_enter(&ds->ds_lock);
3259 /*
3260 * Make a space adjustment for reserved bytes.
3261 */
3262 if (ds->ds_reserved > ds->ds_phys->ds_unique_bytes) {
3263 ASSERT3U(*used, >=,
3264 ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3265 *used -= (ds->ds_reserved - ds->ds_phys->ds_unique_bytes);
3266 *ref_rsrv =
3267 asize - MIN(asize, parent_delta(ds, asize + inflight));
3268 }
3269
3270 if (!check_quota || ds->ds_quota == 0) {
3271 mutex_exit(&ds->ds_lock);
3272 return (0);
3273 }
3274 /*
3275 * If they are requesting more space, and our current estimate
3276 * is over quota, they get to try again unless the actual
3277 * on-disk is over quota and there are no pending changes (which
3278 * may free up space for us).
3279 */
3280 if (ds->ds_phys->ds_used_bytes + inflight >= ds->ds_quota) {
3281 if (inflight > 0 || ds->ds_phys->ds_used_bytes < ds->ds_quota)
3282 error = ERESTART;
3283 else
3284 error = EDQUOT;
3285 }
3286 mutex_exit(&ds->ds_lock);
3287
3288 return (error);
3289}
3290
3291/* ARGSUSED */
3292static int
3293dsl_dataset_set_quota_check(void *arg1, void *arg2, dmu_tx_t *tx)
3294{
3295 dsl_dataset_t *ds = arg1;
428870ff
BB		 3296 dsl_prop_setarg_t *psa = arg2;
3297 int err;
34dc7c2f
BB		 3298
3299 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_REFQUOTA)
3300 return (ENOTSUP);
3301
428870ff
BB		 3302 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3303 return (err);
3304
3305 if (psa->psa_effective_value == 0)
34dc7c2f
BB		 3306 return (0);
3307
428870ff
BB		 3308 if (psa->psa_effective_value < ds->ds_phys->ds_used_bytes ||
3309 psa->psa_effective_value < ds->ds_reserved)
34dc7c2f
BB		 3310 return (ENOSPC);
3311
3312 return (0);
3313}
3314
428870ff
BB		 3315extern void dsl_prop_set_sync(void *, void *, dmu_tx_t *);
3316
34dc7c2f 3317void
428870ff 3318dsl_dataset_set_quota_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3319{
3320 dsl_dataset_t *ds = arg1;
428870ff
BB		 3321 dsl_prop_setarg_t *psa = arg2;
3322 uint64_t effective_value = psa->psa_effective_value;
34dc7c2f 3323
428870ff
BB		 3324 dsl_prop_set_sync(ds, psa, tx);
3325 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
34dc7c2f 3326
428870ff
BB		 3327 if (ds->ds_quota != effective_value) {
3328 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3329 ds->ds_quota = effective_value;
34dc7c2f 3330
428870ff
BB		 3331 spa_history_log_internal(LOG_DS_REFQUOTA,
3332 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu ",
3333 (longlong_t)ds->ds_quota, ds->ds_object);
3334 }
34dc7c2f
BB		 3335}
3336
3337int
428870ff 3338dsl_dataset_set_quota(const char *dsname, zprop_source_t source, uint64_t quota)
34dc7c2f
BB		 3339{
3340 dsl_dataset_t *ds;
428870ff 3341 dsl_prop_setarg_t psa;
34dc7c2f
BB		 3342 int err;
3343
428870ff
BB		 3344 dsl_prop_setarg_init_uint64(&psa, "refquota", source, &quota);
3345
b128c09f 3346 err = dsl_dataset_hold(dsname, FTAG, &ds);
34dc7c2f
BB		 3347 if (err)
3348 return (err);
3349
428870ff
BB		 3350 /*
3351 * If someone removes a file, then tries to set the quota, we
3352 * want to make sure the file freeing takes effect.
3353 */
3354 txg_wait_open(ds->ds_dir->dd_pool, 0);
3355
3356 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3357 dsl_dataset_set_quota_check, dsl_dataset_set_quota_sync,
3358 ds, &psa, 0);
34dc7c2f 3359
b128c09f 3360 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3361 return (err);
3362}
3363
3364static int
3365dsl_dataset_set_reservation_check(void *arg1, void *arg2, dmu_tx_t *tx)
3366{
3367 dsl_dataset_t *ds = arg1;
428870ff
BB		 3368 dsl_prop_setarg_t *psa = arg2;
3369 uint64_t effective_value;
34dc7c2f 3370 uint64_t unique;
428870ff 3371 int err;
34dc7c2f 3372
34dc7c2f
BB		 3373 if (spa_version(ds->ds_dir->dd_pool->dp_spa) <
3374 SPA_VERSION_REFRESERVATION)
3375 return (ENOTSUP);
3376
3377 if (dsl_dataset_is_snapshot(ds))
3378 return (EINVAL);
3379
428870ff
BB		 3380 if ((err = dsl_prop_predict_sync(ds->ds_dir, psa)) != 0)
3381 return (err);
3382
3383 effective_value = psa->psa_effective_value;
3384
34dc7c2f
BB		 3385 /*
3386 * If we are doing the preliminary check in open context, the
3387 * space estimates may be inaccurate.
3388 */
3389 if (!dmu_tx_is_syncing(tx))
3390 return (0);
3391
3392 mutex_enter(&ds->ds_lock);
428870ff
BB		 3393 if (!DS_UNIQUE_IS_ACCURATE(ds))
3394 dsl_dataset_recalc_head_uniq(ds);
3395 unique = ds->ds_phys->ds_unique_bytes;
34dc7c2f
BB		 3396 mutex_exit(&ds->ds_lock);
3397
428870ff
BB		 3398 if (MAX(unique, effective_value) > MAX(unique, ds->ds_reserved)) {
3399 uint64_t delta = MAX(unique, effective_value) -
d164b209
BB		 3400 MAX(unique, ds->ds_reserved);
3401
3402 if (delta > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
3403 return (ENOSPC);
3404 if (ds->ds_quota > 0 &&
428870ff 3405 effective_value > ds->ds_quota)
d164b209
BB		 3406 return (ENOSPC);
3407 }
34dc7c2f
BB		 3408
3409 return (0);
3410}
3411
34dc7c2f 3412static void
428870ff 3413dsl_dataset_set_reservation_sync(void *arg1, void *arg2, dmu_tx_t *tx)
34dc7c2f
BB		 3414{
3415 dsl_dataset_t *ds = arg1;
428870ff
BB		 3416 dsl_prop_setarg_t *psa = arg2;
3417 uint64_t effective_value = psa->psa_effective_value;
34dc7c2f
BB		 3418 uint64_t unique;
3419 int64_t delta;
3420
428870ff
BB		 3421 dsl_prop_set_sync(ds, psa, tx);
3422 DSL_PROP_CHECK_PREDICTION(ds->ds_dir, psa);
3423
34dc7c2f
BB		 3424 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3425
b128c09f 3426 mutex_enter(&ds->ds_dir->dd_lock);
34dc7c2f 3427 mutex_enter(&ds->ds_lock);
428870ff
BB		 3428 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3429 unique = ds->ds_phys->ds_unique_bytes;
3430 delta = MAX(0, (int64_t)(effective_value - unique)) -
34dc7c2f 3431 MAX(0, (int64_t)(ds->ds_reserved - unique));
428870ff 3432 ds->ds_reserved = effective_value;
34dc7c2f
BB		 3433 mutex_exit(&ds->ds_lock);
3434
b128c09f
BB		 3435 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3436 mutex_exit(&ds->ds_dir->dd_lock);
34dc7c2f 3437
428870ff
BB		 3438 spa_history_log_internal(LOG_DS_REFRESERV,
3439 ds->ds_dir->dd_pool->dp_spa, tx, "%lld dataset = %llu",
3440 (longlong_t)effective_value, ds->ds_object);
34dc7c2f
BB		 3441}
3442
3443int
428870ff
BB		 3444dsl_dataset_set_reservation(const char *dsname, zprop_source_t source,
3445 uint64_t reservation)
34dc7c2f
BB		 3446{
3447 dsl_dataset_t *ds;
428870ff 3448 dsl_prop_setarg_t psa;
34dc7c2f
BB		 3449 int err;
3450
428870ff
BB		 3451 dsl_prop_setarg_init_uint64(&psa, "refreservation", source,
3452 &reservation);
3453
b128c09f 3454 err = dsl_dataset_hold(dsname, FTAG, &ds);
34dc7c2f
BB		 3455 if (err)
3456 return (err);
3457
3458 err = dsl_sync_task_do(ds->ds_dir->dd_pool,
3459 dsl_dataset_set_reservation_check,
428870ff
BB		 3460 dsl_dataset_set_reservation_sync, ds, &psa, 0);
3461
b128c09f 3462 dsl_dataset_rele(ds, FTAG);
34dc7c2f
BB		 3463 return (err);
3464}
45d1cae3 3465
572e2857
BB		 3466typedef struct zfs_hold_cleanup_arg {
3467 dsl_pool_t *dp;
3468 uint64_t dsobj;
3469 char htag[MAXNAMELEN];
3470} zfs_hold_cleanup_arg_t;
3471
3472static void
3473dsl_dataset_user_release_onexit(void *arg)
3474{
3475 zfs_hold_cleanup_arg_t *ca = arg;
3476
3477 (void) dsl_dataset_user_release_tmp(ca->dp, ca->dsobj, ca->htag,
3478 B_TRUE);
3479 kmem_free(ca, sizeof (zfs_hold_cleanup_arg_t));
3480}
3481
3482void
3483dsl_register_onexit_hold_cleanup(dsl_dataset_t *ds, const char *htag,
3484 minor_t minor)
3485{
3486 zfs_hold_cleanup_arg_t *ca;
3487
3488 ca = kmem_alloc(sizeof (zfs_hold_cleanup_arg_t), KM_SLEEP);
3489 ca->dp = ds->ds_dir->dd_pool;
3490 ca->dsobj = ds->ds_object;
3491 (void) strlcpy(ca->htag, htag, sizeof (ca->htag));
3492 VERIFY3U(0, ==, zfs_onexit_add_cb(minor,
3493 dsl_dataset_user_release_onexit, ca, NULL));
3494}
428870ff
BB		 3495
3496/*
572e2857
BB		 3497 * If you add new checks here, you may need to add
3498 * additional checks to the "temporary" case in
3499 * snapshot_check() in dmu_objset.c.
428870ff 3500 */
45d1cae3
BB		 3501static int
3502dsl_dataset_user_hold_check(void *arg1, void *arg2, dmu_tx_t *tx)
3503{
3504 dsl_dataset_t *ds = arg1;
428870ff
BB		 3505 struct dsl_ds_holdarg *ha = arg2;
3506 char *htag = ha->htag;
45d1cae3
BB		 3507 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3508 int error = 0;
3509
3510 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3511 return (ENOTSUP);
3512
3513 if (!dsl_dataset_is_snapshot(ds))
3514 return (EINVAL);
3515
45d1cae3
BB		 3516 /* tags must be unique */
3517 mutex_enter(&ds->ds_lock);
3518 if (ds->ds_phys->ds_userrefs_obj) {
3519 error = zap_lookup(mos, ds->ds_phys->ds_userrefs_obj, htag,
3520 8, 1, tx);
3521 if (error == 0)
3522 error = EEXIST;
3523 else if (error == ENOENT)
3524 error = 0;
3525 }
3526 mutex_exit(&ds->ds_lock);
3527
428870ff
BB		 3528 if (error == 0 && ha->temphold &&
3529 strlen(htag) + MAX_TAG_PREFIX_LEN >= MAXNAMELEN)
3530 error = E2BIG;
3531
45d1cae3
BB		 3532 return (error);
3533}
3534
572e2857 3535void
428870ff 3536dsl_dataset_user_hold_sync(void *arg1, void *arg2, dmu_tx_t *tx)
45d1cae3
BB		 3537{
3538 dsl_dataset_t *ds = arg1;
428870ff
BB		 3539 struct dsl_ds_holdarg *ha = arg2;
3540 char *htag = ha->htag;
3541 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3542 objset_t *mos = dp->dp_meta_objset;
3543 uint64_t now = gethrestime_sec();
45d1cae3
BB		 3544 uint64_t zapobj;
3545
3546 mutex_enter(&ds->ds_lock);
3547 if (ds->ds_phys->ds_userrefs_obj == 0) {
3548 /*
3549 * This is the first user hold for this dataset. Create
3550 * the userrefs zap object.
3551 */
3552 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3553 zapobj = ds->ds_phys->ds_userrefs_obj =
3554 zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
3555 } else {
3556 zapobj = ds->ds_phys->ds_userrefs_obj;
3557 }
3558 ds->ds_userrefs++;
3559 mutex_exit(&ds->ds_lock);
3560
3561 VERIFY(0 == zap_add(mos, zapobj, htag, 8, 1, &now, tx));
3562
428870ff
BB		 3563 if (ha->temphold) {
3564 VERIFY(0 == dsl_pool_user_hold(dp, ds->ds_object,
3565 htag, &now, tx));
3566 }
45d1cae3 3567
428870ff
BB		 3568 spa_history_log_internal(LOG_DS_USER_HOLD,
3569 dp->dp_spa, tx, "<%s> temp = %d dataset = %llu", htag,
3570 (int)ha->temphold, ds->ds_object);
3571}
45d1cae3
BB		 3572
3573static int
428870ff 3574dsl_dataset_user_hold_one(const char *dsname, void *arg)
45d1cae3
BB		 3575{
3576 struct dsl_ds_holdarg *ha = arg;
3577 dsl_dataset_t *ds;
3578 int error;
3579 char *name;
45d1cae3
BB		 3580
3581 /* alloc a buffer to hold dsname@snapname plus terminating NULL */
428870ff 3582 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3583 error = dsl_dataset_hold(name, ha->dstg, &ds);
428870ff 3584 strfree(name);
45d1cae3 3585 if (error == 0) {
428870ff 3586 ha->gotone = B_TRUE;
45d1cae3 3587 dsl_sync_task_create(ha->dstg, dsl_dataset_user_hold_check,
428870ff 3588 dsl_dataset_user_hold_sync, ds, ha, 0);
45d1cae3
BB		 3589 } else if (error == ENOENT && ha->recursive) {
3590 error = 0;
3591 } else {
428870ff 3592 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
45d1cae3
BB		 3593 }
3594 return (error);
3595}
3596
572e2857
BB		 3597int
3598dsl_dataset_user_hold_for_send(dsl_dataset_t *ds, char *htag,
3599 boolean_t temphold)
3600{
3601 struct dsl_ds_holdarg *ha;
3602 int error;
3603
3604 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3605 ha->htag = htag;
3606 ha->temphold = temphold;
3607 error = dsl_sync_task_do(ds->ds_dir->dd_pool,
3608 dsl_dataset_user_hold_check, dsl_dataset_user_hold_sync,
3609 ds, ha, 0);
3610 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3611
3612 return (error);
3613}
3614
45d1cae3
BB		 3615int
3616dsl_dataset_user_hold(char *dsname, char *snapname, char *htag,
572e2857 3617 boolean_t recursive, boolean_t temphold, int cleanup_fd)
45d1cae3
BB		 3618{
3619 struct dsl_ds_holdarg *ha;
3620 dsl_sync_task_t *dst;
3621 spa_t *spa;
3622 int error;
572e2857
BB		 3623 minor_t minor = 0;
3624
3625 if (cleanup_fd != -1) {
3626 /* Currently we only support cleanup-on-exit of tempholds. */
3627 if (!temphold)
3628 return (EINVAL);
3629 error = zfs_onexit_fd_hold(cleanup_fd, &minor);
3630 if (error)
3631 return (error);
3632 }
45d1cae3
BB		 3633
3634 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3635
3636 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3637
3638 error = spa_open(dsname, &spa, FTAG);
3639 if (error) {
3640 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
572e2857
BB		 3641 if (cleanup_fd != -1)
3642 zfs_onexit_fd_rele(cleanup_fd);
45d1cae3
BB		 3643 return (error);
3644 }
3645
3646 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3647 ha->htag = htag;
3648 ha->snapname = snapname;
3649 ha->recursive = recursive;
428870ff 3650 ha->temphold = temphold;
572e2857 3651
45d1cae3
BB		 3652 if (recursive) {
3653 error = dmu_objset_find(dsname, dsl_dataset_user_hold_one,
3654 ha, DS_FIND_CHILDREN);
3655 } else {
3656 error = dsl_dataset_user_hold_one(dsname, ha);
3657 }
3658 if (error == 0)
3659 error = dsl_sync_task_group_wait(ha->dstg);
3660
3661 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3662 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3663 dsl_dataset_t *ds = dst->dst_arg1;
3664
3665 if (dst->dst_err) {
3666 dsl_dataset_name(ds, ha->failed);
3667 *strchr(ha->failed, '@') = '\0';
572e2857
BB		 3668 } else if (error == 0 && minor != 0 && temphold) {
3669 /*
3670 * If this hold is to be released upon process exit,
3671 * register that action now.
3672 */
3673 dsl_register_onexit_hold_cleanup(ds, htag, minor);
45d1cae3
BB		 3674 }
3675 dsl_dataset_rele(ds, ha->dstg);
3676 }
3677
428870ff
BB		 3678 if (error == 0 && recursive && !ha->gotone)
3679 error = ENOENT;
3680
45d1cae3 3681 if (error)
428870ff 3682 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
45d1cae3
BB		 3683
3684 dsl_sync_task_group_destroy(ha->dstg);
572e2857 3685
45d1cae3
BB		 3686 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3687 spa_close(spa, FTAG);
572e2857
BB		 3688 if (cleanup_fd != -1)
3689 zfs_onexit_fd_rele(cleanup_fd);
45d1cae3
BB		 3690 return (error);
3691}
3692
3693struct dsl_ds_releasearg {
3694 dsl_dataset_t *ds;
3695 const char *htag;
3696 boolean_t own; /* do we own or just hold ds? */
3697};
3698
3699static int
3700dsl_dataset_release_might_destroy(dsl_dataset_t *ds, const char *htag,
3701 boolean_t *might_destroy)
3702{
3703 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3704 uint64_t zapobj;
3705 uint64_t tmp;
3706 int error;
3707
3708 *might_destroy = B_FALSE;
3709
3710 mutex_enter(&ds->ds_lock);
3711 zapobj = ds->ds_phys->ds_userrefs_obj;
3712 if (zapobj == 0) {
3713 /* The tag can't possibly exist */
3714 mutex_exit(&ds->ds_lock);
3715 return (ESRCH);
3716 }
3717
3718 /* Make sure the tag exists */
3719 error = zap_lookup(mos, zapobj, htag, 8, 1, &tmp);
3720 if (error) {
3721 mutex_exit(&ds->ds_lock);
3722 if (error == ENOENT)
3723 error = ESRCH;
3724 return (error);
3725 }
3726
3727 if (ds->ds_userrefs == 1 && ds->ds_phys->ds_num_children == 1 &&
3728 DS_IS_DEFER_DESTROY(ds))
3729 *might_destroy = B_TRUE;
3730
3731 mutex_exit(&ds->ds_lock);
3732 return (0);
3733}
3734
3735static int
3736dsl_dataset_user_release_check(void *arg1, void *tag, dmu_tx_t *tx)
3737{
3738 struct dsl_ds_releasearg *ra = arg1;
3739 dsl_dataset_t *ds = ra->ds;
3740 boolean_t might_destroy;
3741 int error;
3742
3743 if (spa_version(ds->ds_dir->dd_pool->dp_spa) < SPA_VERSION_USERREFS)
3744 return (ENOTSUP);
3745
3746 error = dsl_dataset_release_might_destroy(ds, ra->htag, &might_destroy);
3747 if (error)
3748 return (error);
3749
3750 if (might_destroy) {
3751 struct dsl_ds_destroyarg dsda = {0};
3752
3753 if (dmu_tx_is_syncing(tx)) {
3754 /*
3755 * If we're not prepared to remove the snapshot,
3756 * we can't allow the release to happen right now.
3757 */
3758 if (!ra->own)
3759 return (EBUSY);
45d1cae3
BB		 3760 }
3761 dsda.ds = ds;
3762 dsda.releasing = B_TRUE;
3763 return (dsl_dataset_destroy_check(&dsda, tag, tx));
3764 }
3765
3766 return (0);
3767}
3768
3769static void
428870ff 3770dsl_dataset_user_release_sync(void *arg1, void *tag, dmu_tx_t *tx)
45d1cae3
BB		 3771{
3772 struct dsl_ds_releasearg *ra = arg1;
3773 dsl_dataset_t *ds = ra->ds;
428870ff
BB		 3774 dsl_pool_t *dp = ds->ds_dir->dd_pool;
3775 objset_t *mos = dp->dp_meta_objset;
45d1cae3
BB		 3776 uint64_t zapobj;
3777 uint64_t dsobj = ds->ds_object;
3778 uint64_t refs;
428870ff
BB		 3779 int error;
3780
45d1cae3
BB		 3781 mutex_enter(&ds->ds_lock);
3782 ds->ds_userrefs--;
3783 refs = ds->ds_userrefs;
3784 mutex_exit(&ds->ds_lock);
428870ff
BB		 3785 error = dsl_pool_user_release(dp, ds->ds_object, ra->htag, tx);
3786 VERIFY(error == 0 || error == ENOENT);
45d1cae3
BB		 3787 zapobj = ds->ds_phys->ds_userrefs_obj;
3788 VERIFY(0 == zap_remove(mos, zapobj, ra->htag, tx));
3789 if (ds->ds_userrefs == 0 && ds->ds_phys->ds_num_children == 1 &&
3790 DS_IS_DEFER_DESTROY(ds)) {
3791 struct dsl_ds_destroyarg dsda = {0};
3792
3793 ASSERT(ra->own);
3794 dsda.ds = ds;
3795 dsda.releasing = B_TRUE;
3796 /* We already did the destroy_check */
428870ff 3797 dsl_dataset_destroy_sync(&dsda, tag, tx);
45d1cae3
BB		 3798 }
3799
428870ff
BB		 3800 spa_history_log_internal(LOG_DS_USER_RELEASE,
3801 dp->dp_spa, tx, "<%s> %lld dataset = %llu",
45d1cae3
BB		 3802 ra->htag, (longlong_t)refs, dsobj);
3803}
3804
3805static int
428870ff 3806dsl_dataset_user_release_one(const char *dsname, void *arg)
45d1cae3
BB		 3807{
3808 struct dsl_ds_holdarg *ha = arg;
3809 struct dsl_ds_releasearg *ra;
3810 dsl_dataset_t *ds;
3811 int error;
3812 void *dtag = ha->dstg;
3813 char *name;
45d1cae3
BB		 3814 boolean_t own = B_FALSE;
3815 boolean_t might_destroy;
3816
45d1cae3 3817 /* alloc a buffer to hold dsname@snapname, plus the terminating NULL */
428870ff 3818 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3819 error = dsl_dataset_hold(name, dtag, &ds);
428870ff 3820 strfree(name);
45d1cae3
BB		 3821 if (error == ENOENT && ha->recursive)
3822 return (0);
428870ff 3823 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
45d1cae3
BB		 3824 if (error)
3825 return (error);
3826
428870ff
BB		 3827 ha->gotone = B_TRUE;
3828
45d1cae3
BB		 3829 ASSERT(dsl_dataset_is_snapshot(ds));
3830
3831 error = dsl_dataset_release_might_destroy(ds, ha->htag, &might_destroy);
3832 if (error) {
3833 dsl_dataset_rele(ds, dtag);
3834 return (error);
3835 }
3836
3837 if (might_destroy) {
3838#ifdef _KERNEL
428870ff 3839 name = kmem_asprintf("%s@%s", dsname, ha->snapname);
45d1cae3 3840 error = zfs_unmount_snap(name, NULL);
428870ff 3841 strfree(name);
45d1cae3
BB		 3842 if (error) {
3843 dsl_dataset_rele(ds, dtag);
3844 return (error);
3845 }
3846#endif
428870ff 3847 if (!dsl_dataset_tryown(ds, B_TRUE, dtag)) {
45d1cae3
BB		 3848 dsl_dataset_rele(ds, dtag);
3849 return (EBUSY);
3850 } else {
3851 own = B_TRUE;
3852 dsl_dataset_make_exclusive(ds, dtag);
3853 }
3854 }
3855
3856 ra = kmem_alloc(sizeof (struct dsl_ds_releasearg), KM_SLEEP);
3857 ra->ds = ds;
3858 ra->htag = ha->htag;
3859 ra->own = own;
3860 dsl_sync_task_create(ha->dstg, dsl_dataset_user_release_check,
3861 dsl_dataset_user_release_sync, ra, dtag, 0);
3862
3863 return (0);
3864}
3865
3866int
3867dsl_dataset_user_release(char *dsname, char *snapname, char *htag,
3868 boolean_t recursive)
3869{
3870 struct dsl_ds_holdarg *ha;
3871 dsl_sync_task_t *dst;
3872 spa_t *spa;
3873 int error;
3874
428870ff 3875top:
45d1cae3
BB		 3876 ha = kmem_zalloc(sizeof (struct dsl_ds_holdarg), KM_SLEEP);
3877
3878 (void) strlcpy(ha->failed, dsname, sizeof (ha->failed));
3879
3880 error = spa_open(dsname, &spa, FTAG);
3881 if (error) {
3882 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3883 return (error);
3884 }
3885
3886 ha->dstg = dsl_sync_task_group_create(spa_get_dsl(spa));
3887 ha->htag = htag;
3888 ha->snapname = snapname;
3889 ha->recursive = recursive;
3890 if (recursive) {
3891 error = dmu_objset_find(dsname, dsl_dataset_user_release_one,
3892 ha, DS_FIND_CHILDREN);
3893 } else {
3894 error = dsl_dataset_user_release_one(dsname, ha);
3895 }
3896 if (error == 0)
3897 error = dsl_sync_task_group_wait(ha->dstg);
3898
3899 for (dst = list_head(&ha->dstg->dstg_tasks); dst;
3900 dst = list_next(&ha->dstg->dstg_tasks, dst)) {
3901 struct dsl_ds_releasearg *ra = dst->dst_arg1;
3902 dsl_dataset_t *ds = ra->ds;
3903
3904 if (dst->dst_err)
3905 dsl_dataset_name(ds, ha->failed);
3906
3907 if (ra->own)
3908 dsl_dataset_disown(ds, ha->dstg);
3909 else
3910 dsl_dataset_rele(ds, ha->dstg);
3911
3912 kmem_free(ra, sizeof (struct dsl_ds_releasearg));
3913 }
3914
428870ff
BB		 3915 if (error == 0 && recursive && !ha->gotone)
3916 error = ENOENT;
3917
3918 if (error && error != EBUSY)
3919 (void) strlcpy(dsname, ha->failed, sizeof (ha->failed));
45d1cae3
BB		 3920
3921 dsl_sync_task_group_destroy(ha->dstg);
3922 kmem_free(ha, sizeof (struct dsl_ds_holdarg));
3923 spa_close(spa, FTAG);
428870ff
BB		 3924
3925 /*
3926 * We can get EBUSY if we were racing with deferred destroy and
3927 * dsl_dataset_user_release_check() hadn't done the necessary
3928 * open context setup. We can also get EBUSY if we're racing
3929 * with destroy and that thread is the ds_owner. Either way
3930 * the busy condition should be transient, and we should retry
3931 * the release operation.
3932 */
3933 if (error == EBUSY)
3934 goto top;
3935
45d1cae3
BB		 3936 return (error);
3937}
3938
428870ff 3939/*
572e2857
BB		 3940 * Called at spa_load time (with retry == B_FALSE) to release a stale
3941 * temporary user hold. Also called by the onexit code (with retry == B_TRUE).
428870ff
BB		 3942 */
3943int
572e2857
BB		 3944dsl_dataset_user_release_tmp(dsl_pool_t *dp, uint64_t dsobj, char *htag,
3945 boolean_t retry)
428870ff
BB		 3946{
3947 dsl_dataset_t *ds;
3948 char *snap;
3949 char *name;
3950 int namelen;
3951 int error;
3952
572e2857
BB		 3953 do {
3954 rw_enter(&dp->dp_config_rwlock, RW_READER);
3955 error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
3956 rw_exit(&dp->dp_config_rwlock);
3957 if (error)
3958 return (error);
3959 namelen = dsl_dataset_namelen(ds)+1;
3960 name = kmem_alloc(namelen, KM_SLEEP);
3961 dsl_dataset_name(ds, name);
3962 dsl_dataset_rele(ds, FTAG);
428870ff 3963
572e2857
BB		 3964 snap = strchr(name, '@');
3965 *snap = '\0';
3966 ++snap;
3967 error = dsl_dataset_user_release(name, snap, htag, B_FALSE);
3968 kmem_free(name, namelen);
3969
3970 /*
3971 * The object can't have been destroyed because we have a hold,
3972 * but it might have been renamed, resulting in ENOENT. Retry
3973 * if we've been requested to do so.
3974 *
3975 * It would be nice if we could use the dsobj all the way
3976 * through and avoid ENOENT entirely. But we might need to
3977 * unmount the snapshot, and there's currently no way to lookup
3978 * a vfsp using a ZFS object id.
3979 */
3980 } while ((error == ENOENT) && retry);
3981
3982 return (error);
428870ff
BB		 3983}
3984
45d1cae3
BB		 3985int
3986dsl_dataset_get_holds(const char *dsname, nvlist_t **nvp)
3987{
3988 dsl_dataset_t *ds;
3989 int err;
3990
3991 err = dsl_dataset_hold(dsname, FTAG, &ds);
3992 if (err)
3993 return (err);
3994
3995 VERIFY(0 == nvlist_alloc(nvp, NV_UNIQUE_NAME, KM_SLEEP));
3996 if (ds->ds_phys->ds_userrefs_obj != 0) {
3997 zap_attribute_t *za;
3998 zap_cursor_t zc;
3999
4000 za = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
4001 for (zap_cursor_init(&zc, ds->ds_dir->dd_pool->dp_meta_objset,
4002 ds->ds_phys->ds_userrefs_obj);
4003 zap_cursor_retrieve(&zc, za) == 0;
4004 zap_cursor_advance(&zc)) {
4005 VERIFY(0 == nvlist_add_uint64(*nvp, za->za_name,
4006 za->za_first_integer));
4007 }
4008 zap_cursor_fini(&zc);
4009 kmem_free(za, sizeof (zap_attribute_t));
4010 }
4011 dsl_dataset_rele(ds, FTAG);
4012 return (0);
4013}
428870ff
BB		 4014
4015/*
4016 * Note, this fuction is used as the callback for dmu_objset_find(). We
4017 * always return 0 so that we will continue to find and process
4018 * inconsistent datasets, even if we encounter an error trying to
4019 * process one of them.
4020 */
4021/* ARGSUSED */
4022int
4023dsl_destroy_inconsistent(const char *dsname, void *arg)
4024{
4025 dsl_dataset_t *ds;
4026
4027 if (dsl_dataset_own(dsname, B_TRUE, FTAG, &ds) == 0) {
4028 if (DS_IS_INCONSISTENT(ds))
4029 (void) dsl_dataset_destroy(ds, FTAG, B_FALSE);
4030 else
4031 dsl_dataset_disown(ds, FTAG);
4032 }
4033 return (0);
4034}
ZFS On Linux mirror for PVE