1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Portions Copyright (c) 2011 Martin Matuska <mm@FreeBSD.org>
24 * Copyright (c) 2011, 2014 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 */
28
29 #include <sys/dmu_objset.h>
30 #include <sys/dsl_dataset.h>
31 #include <sys/dsl_dir.h>
32 #include <sys/dsl_prop.h>
33 #include <sys/dsl_synctask.h>
34 #include <sys/dmu_traverse.h>
35 #include <sys/dmu_impl.h>
36 #include <sys/dmu_tx.h>
37 #include <sys/arc.h>
38 #include <sys/zio.h>
39 #include <sys/zap.h>
40 #include <sys/zfeature.h>
41 #include <sys/unique.h>
42 #include <sys/zfs_context.h>
43 #include <sys/zfs_ioctl.h>
44 #include <sys/spa.h>
45 #include <sys/zfs_znode.h>
46 #include <sys/zfs_onexit.h>
47 #include <sys/zvol.h>
48 #include <sys/dsl_scan.h>
49 #include <sys/dsl_deadlist.h>
50 #include <sys/dsl_destroy.h>
51 #include <sys/dsl_userhold.h>
52 #include <sys/dsl_bookmark.h>
53
54 SYSCTL_DECL(_vfs_zfs);
55
56 /*
57 * The SPA supports block sizes up to 16MB. However, very large blocks
58 * can have an impact on i/o latency (e.g. tying up a spinning disk for
59 * ~300ms), and also potentially on the memory allocator. Therefore,
60 * we do not allow the recordsize to be set larger than zfs_max_recordsize
61 * (default 1MB). Larger blocks can be created by changing this tunable,
62 * and pools with larger blocks can always be imported and used, regardless
63 * of this setting.
64 */
65 int zfs_max_recordsize = 1 * 1024 * 1024;
66 SYSCTL_INT(_vfs_zfs, OID_AUTO, max_recordsize, CTLFLAG_RWTUN,
67 &zfs_max_recordsize, 0,
68 "Maximum block size. Expect dragons when tuning this.");
69
70 #define SWITCH64(x, y) \
71 { \
72 uint64_t __tmp = (x); \
73 (x) = (y); \
74 (y) = __tmp; \
75 }
76
77 #define DS_REF_MAX (1ULL << 62)
78
79 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
80 extern inline boolean_t dsl_dataset_is_snapshot(dsl_dataset_t *ds);
81
82 /*
83 * Figure out how much of this delta should be propogated to the dsl_dir
84 * layer. If there's a refreservation, that space has already been
85 * partially accounted for in our ancestors.
86 */
87 static int64_t
parent_delta(dsl_dataset_t * ds,int64_t delta)88 parent_delta(dsl_dataset_t *ds, int64_t delta)
89 {
90 dsl_dataset_phys_t *ds_phys;
91 uint64_t old_bytes, new_bytes;
92
93 if (ds->ds_reserved == 0)
94 return (delta);
95
96 ds_phys = dsl_dataset_phys(ds);
97 old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
98 new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
99
100 ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
101 return (new_bytes - old_bytes);
102 }
103
104 void
dsl_dataset_block_born(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx)105 dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
106 {
107 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
108 int compressed = BP_GET_PSIZE(bp);
109 int uncompressed = BP_GET_UCSIZE(bp);
110 int64_t delta;
111
112 dprintf_bp(bp, "ds=%p", ds);
113
114 ASSERT(dmu_tx_is_syncing(tx));
115 /* It could have been compressed away to nothing */
116 if (BP_IS_HOLE(bp))
117 return;
118 ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
119 ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
120 if (ds == NULL) {
121 dsl_pool_mos_diduse_space(tx->tx_pool,
122 used, compressed, uncompressed);
123 return;
124 }
125
126 dmu_buf_will_dirty(ds->ds_dbuf, tx);
127 mutex_enter(&ds->ds_lock);
128 delta = parent_delta(ds, used);
129 dsl_dataset_phys(ds)->ds_referenced_bytes += used;
130 dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
131 dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
132 dsl_dataset_phys(ds)->ds_unique_bytes += used;
133 if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE)
134 ds->ds_need_large_blocks = B_TRUE;
135 mutex_exit(&ds->ds_lock);
136 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
137 compressed, uncompressed, tx);
138 dsl_dir_transfer_space(ds->ds_dir, used - delta,
139 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
140 }
141
142 int
dsl_dataset_block_kill(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx,boolean_t async)143 dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
144 boolean_t async)
145 {
146 int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
147 int compressed = BP_GET_PSIZE(bp);
148 int uncompressed = BP_GET_UCSIZE(bp);
149
150 if (BP_IS_HOLE(bp))
151 return (0);
152
153 ASSERT(dmu_tx_is_syncing(tx));
154 ASSERT(bp->blk_birth <= tx->tx_txg);
155
156 if (ds == NULL) {
157 dsl_free(tx->tx_pool, tx->tx_txg, bp);
158 dsl_pool_mos_diduse_space(tx->tx_pool,
159 -used, -compressed, -uncompressed);
160 return (used);
161 }
162 ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
163
164 ASSERT(!dsl_dataset_is_snapshot(ds));
165 dmu_buf_will_dirty(ds->ds_dbuf, tx);
166
167 if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
168 int64_t delta;
169
170 dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
171 dsl_free(tx->tx_pool, tx->tx_txg, bp);
172
173 mutex_enter(&ds->ds_lock);
174 ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
175 !DS_UNIQUE_IS_ACCURATE(ds));
176 delta = parent_delta(ds, -used);
177 dsl_dataset_phys(ds)->ds_unique_bytes -= used;
178 mutex_exit(&ds->ds_lock);
179 dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
180 delta, -compressed, -uncompressed, tx);
181 dsl_dir_transfer_space(ds->ds_dir, -used - delta,
182 DD_USED_REFRSRV, DD_USED_HEAD, NULL);
183 } else {
184 dprintf_bp(bp, "putting on dead list: %s", "");
185 if (async) {
186 /*
187 * We are here as part of zio's write done callback,
188 * which means we're a zio interrupt thread. We can't
189 * call dsl_deadlist_insert() now because it may block
190 * waiting for I/O. Instead, put bp on the deferred
191 * queue and let dsl_pool_sync() finish the job.
192 */
193 bplist_append(&ds->ds_pending_deadlist, bp);
194 } else {
195 dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
196 }
197 ASSERT3U(ds->ds_prev->ds_object, ==,
198 dsl_dataset_phys(ds)->ds_prev_snap_obj);
199 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
200 /* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
201 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
202 ds->ds_object && bp->blk_birth >
203 dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
204 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
205 mutex_enter(&ds->ds_prev->ds_lock);
206 dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
207 mutex_exit(&ds->ds_prev->ds_lock);
208 }
209 if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
210 dsl_dir_transfer_space(ds->ds_dir, used,
211 DD_USED_HEAD, DD_USED_SNAP, tx);
212 }
213 }
214 mutex_enter(&ds->ds_lock);
215 ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
216 dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
217 ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
218 dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
219 ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
220 dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
221 mutex_exit(&ds->ds_lock);
222
223 return (used);
224 }
225
226 uint64_t
dsl_dataset_prev_snap_txg(dsl_dataset_t * ds)227 dsl_dataset_prev_snap_txg(dsl_dataset_t *ds)
228 {
229 uint64_t trysnap = 0;
230
231 if (ds == NULL)
232 return (0);
233 /*
234 * The snapshot creation could fail, but that would cause an
235 * incorrect FALSE return, which would only result in an
236 * overestimation of the amount of space that an operation would
237 * consume, which is OK.
238 *
239 * There's also a small window where we could miss a pending
240 * snapshot, because we could set the sync task in the quiescing
241 * phase. So this should only be used as a guess.
242 */
243 if (ds->ds_trysnap_txg >
244 spa_last_synced_txg(ds->ds_dir->dd_pool->dp_spa))
245 trysnap = ds->ds_trysnap_txg;
246 return (MAX(dsl_dataset_phys(ds)->ds_prev_snap_txg, trysnap));
247 }
248
249 boolean_t
dsl_dataset_block_freeable(dsl_dataset_t * ds,const blkptr_t * bp,uint64_t blk_birth)250 dsl_dataset_block_freeable(dsl_dataset_t *ds, const blkptr_t *bp,
251 uint64_t blk_birth)
252 {
253 if (blk_birth <= dsl_dataset_prev_snap_txg(ds) ||
254 (bp != NULL && BP_IS_HOLE(bp)))
255 return (B_FALSE);
256
257 ddt_prefetch(dsl_dataset_get_spa(ds), bp);
258
259 return (B_TRUE);
260 }
261
262 /* ARGSUSED */
263 static void
dsl_dataset_evict(dmu_buf_t * db,void * dsv)264 dsl_dataset_evict(dmu_buf_t *db, void *dsv)
265 {
266 dsl_dataset_t *ds = dsv;
267
268 ASSERT(ds->ds_owner == NULL);
269
270 unique_remove(ds->ds_fsid_guid);
271
272 if (ds->ds_objset != NULL)
273 dmu_objset_evict(ds->ds_objset);
274
275 if (ds->ds_prev) {
276 dsl_dataset_rele(ds->ds_prev, ds);
277 ds->ds_prev = NULL;
278 }
279
280 bplist_destroy(&ds->ds_pending_deadlist);
281 if (dsl_dataset_phys(ds)->ds_deadlist_obj != 0)
282 dsl_deadlist_close(&ds->ds_deadlist);
283 if (ds->ds_dir)
284 dsl_dir_rele(ds->ds_dir, ds);
285
286 ASSERT(!list_link_active(&ds->ds_synced_link));
287
288 if (mutex_owned(&ds->ds_lock))
289 mutex_exit(&ds->ds_lock);
290 mutex_destroy(&ds->ds_lock);
291 if (mutex_owned(&ds->ds_opening_lock))
292 mutex_exit(&ds->ds_opening_lock);
293 mutex_destroy(&ds->ds_opening_lock);
294 mutex_destroy(&ds->ds_sendstream_lock);
295 refcount_destroy(&ds->ds_longholds);
296
297 kmem_free(ds, sizeof (dsl_dataset_t));
298 }
299
300 int
dsl_dataset_get_snapname(dsl_dataset_t * ds)301 dsl_dataset_get_snapname(dsl_dataset_t *ds)
302 {
303 dsl_dataset_phys_t *headphys;
304 int err;
305 dmu_buf_t *headdbuf;
306 dsl_pool_t *dp = ds->ds_dir->dd_pool;
307 objset_t *mos = dp->dp_meta_objset;
308
309 if (ds->ds_snapname[0])
310 return (0);
311 if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
312 return (0);
313
314 err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
315 FTAG, &headdbuf);
316 if (err != 0)
317 return (err);
318 headphys = headdbuf->db_data;
319 err = zap_value_search(dp->dp_meta_objset,
320 headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
321 dmu_buf_rele(headdbuf, FTAG);
322 return (err);
323 }
324
325 int
dsl_dataset_snap_lookup(dsl_dataset_t * ds,const char * name,uint64_t * value)326 dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
327 {
328 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
329 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
330 matchtype_t mt;
331 int err;
332
333 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
334 mt = MT_FIRST;
335 else
336 mt = MT_EXACT;
337
338 err = zap_lookup_norm(mos, snapobj, name, 8, 1,
339 value, mt, NULL, 0, NULL);
340 if (err == ENOTSUP && mt == MT_FIRST)
341 err = zap_lookup(mos, snapobj, name, 8, 1, value);
342 return (err);
343 }
344
345 int
dsl_dataset_snap_remove(dsl_dataset_t * ds,const char * name,dmu_tx_t * tx,boolean_t adj_cnt)346 dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
347 boolean_t adj_cnt)
348 {
349 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
350 uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
351 matchtype_t mt;
352 int err;
353
354 dsl_dir_snap_cmtime_update(ds->ds_dir);
355
356 if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
357 mt = MT_FIRST;
358 else
359 mt = MT_EXACT;
360
361 err = zap_remove_norm(mos, snapobj, name, mt, tx);
362 if (err == ENOTSUP && mt == MT_FIRST)
363 err = zap_remove(mos, snapobj, name, tx);
364
365 if (err == 0 && adj_cnt)
366 dsl_fs_ss_count_adjust(ds->ds_dir, -1,
367 DD_FIELD_SNAPSHOT_COUNT, tx);
368
369 return (err);
370 }
371
372 int
dsl_dataset_hold_obj(dsl_pool_t * dp,uint64_t dsobj,void * tag,dsl_dataset_t ** dsp)373 dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
374 dsl_dataset_t **dsp)
375 {
376 objset_t *mos = dp->dp_meta_objset;
377 dmu_buf_t *dbuf;
378 dsl_dataset_t *ds;
379 int err;
380 dmu_object_info_t doi;
381
382 ASSERT(dsl_pool_config_held(dp));
383
384 err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
385 if (err != 0)
386 return (err);
387
388 /* Make sure dsobj has the correct object type. */
389 dmu_object_info_from_db(dbuf, &doi);
390 if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
391 dmu_buf_rele(dbuf, tag);
392 return (SET_ERROR(EINVAL));
393 }
394
395 ds = dmu_buf_get_user(dbuf);
396 if (ds == NULL) {
397 dsl_dataset_t *winner = NULL;
398
399 ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
400 ds->ds_dbuf = dbuf;
401 ds->ds_object = dsobj;
402
403 mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
404 mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
405 mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
406 refcount_create(&ds->ds_longholds);
407
408 bplist_create(&ds->ds_pending_deadlist);
409 dsl_deadlist_open(&ds->ds_deadlist,
410 mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
411
412 list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
413 offsetof(dmu_sendarg_t, dsa_link));
414
415 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
416 int zaperr = zap_contains(mos, dsobj,
417 DS_FIELD_LARGE_BLOCKS);
418 if (zaperr != ENOENT) {
419 VERIFY0(zaperr);
420 ds->ds_large_blocks = B_TRUE;
421 }
422 }
423
424 if (err == 0) {
425 err = dsl_dir_hold_obj(dp,
426 dsl_dataset_phys(ds)->ds_dir_obj, NULL, ds,
427 &ds->ds_dir);
428 }
429 if (err != 0) {
430 mutex_destroy(&ds->ds_lock);
431 mutex_destroy(&ds->ds_opening_lock);
432 mutex_destroy(&ds->ds_sendstream_lock);
433 refcount_destroy(&ds->ds_longholds);
434 bplist_destroy(&ds->ds_pending_deadlist);
435 dsl_deadlist_close(&ds->ds_deadlist);
436 kmem_free(ds, sizeof (dsl_dataset_t));
437 dmu_buf_rele(dbuf, tag);
438 return (err);
439 }
440
441 if (!dsl_dataset_is_snapshot(ds)) {
442 ds->ds_snapname[0] = '\0';
443 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
444 err = dsl_dataset_hold_obj(dp,
445 dsl_dataset_phys(ds)->ds_prev_snap_obj,
446 ds, &ds->ds_prev);
447 }
448 if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
449 int zaperr = zap_lookup(mos, ds->ds_object,
450 DS_FIELD_BOOKMARK_NAMES,
451 sizeof (ds->ds_bookmarks), 1,
452 &ds->ds_bookmarks);
453 if (zaperr != ENOENT)
454 VERIFY0(zaperr);
455 }
456 } else {
457 if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
458 err = dsl_dataset_get_snapname(ds);
459 if (err == 0 &&
460 dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
461 err = zap_count(
462 ds->ds_dir->dd_pool->dp_meta_objset,
463 dsl_dataset_phys(ds)->ds_userrefs_obj,
464 &ds->ds_userrefs);
465 }
466 }
467
468 if (err == 0 && !dsl_dataset_is_snapshot(ds)) {
469 err = dsl_prop_get_int_ds(ds,
470 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
471 &ds->ds_reserved);
472 if (err == 0) {
473 err = dsl_prop_get_int_ds(ds,
474 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
475 &ds->ds_quota);
476 }
477 } else {
478 ds->ds_reserved = ds->ds_quota = 0;
479 }
480
481 if (err != 0 || (winner = dmu_buf_set_user_ie(dbuf, ds,
482 dsl_dataset_evict)) != NULL) {
483 bplist_destroy(&ds->ds_pending_deadlist);
484 dsl_deadlist_close(&ds->ds_deadlist);
485 if (ds->ds_prev)
486 dsl_dataset_rele(ds->ds_prev, ds);
487 dsl_dir_rele(ds->ds_dir, ds);
488 mutex_destroy(&ds->ds_lock);
489 mutex_destroy(&ds->ds_opening_lock);
490 mutex_destroy(&ds->ds_sendstream_lock);
491 refcount_destroy(&ds->ds_longholds);
492 kmem_free(ds, sizeof (dsl_dataset_t));
493 if (err != 0) {
494 dmu_buf_rele(dbuf, tag);
495 return (err);
496 }
497 ds = winner;
498 } else {
499 ds->ds_fsid_guid =
500 unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
501 }
502 }
503 ASSERT3P(ds->ds_dbuf, ==, dbuf);
504 ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
505 ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
506 spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
507 dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
508 *dsp = ds;
509 return (0);
510 }
511
512 int
dsl_dataset_hold(dsl_pool_t * dp,const char * name,void * tag,dsl_dataset_t ** dsp)513 dsl_dataset_hold(dsl_pool_t *dp, const char *name,
514 void *tag, dsl_dataset_t **dsp)
515 {
516 dsl_dir_t *dd;
517 const char *snapname;
518 uint64_t obj;
519 int err = 0;
520
521 err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
522 if (err != 0)
523 return (err);
524
525 ASSERT(dsl_pool_config_held(dp));
526 obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
527 if (obj != 0)
528 err = dsl_dataset_hold_obj(dp, obj, tag, dsp);
529 else
530 err = SET_ERROR(ENOENT);
531
532 /* we may be looking for a snapshot */
533 if (err == 0 && snapname != NULL) {
534 dsl_dataset_t *ds;
535
536 if (*snapname++ != '@') {
537 dsl_dataset_rele(*dsp, tag);
538 dsl_dir_rele(dd, FTAG);
539 return (SET_ERROR(ENOENT));
540 }
541
542 dprintf("looking for snapshot '%s'\n", snapname);
543 err = dsl_dataset_snap_lookup(*dsp, snapname, &obj);
544 if (err == 0)
545 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
546 dsl_dataset_rele(*dsp, tag);
547
548 if (err == 0) {
549 mutex_enter(&ds->ds_lock);
550 if (ds->ds_snapname[0] == 0)
551 (void) strlcpy(ds->ds_snapname, snapname,
552 sizeof (ds->ds_snapname));
553 mutex_exit(&ds->ds_lock);
554 *dsp = ds;
555 }
556 }
557
558 dsl_dir_rele(dd, FTAG);
559 return (err);
560 }
561
562 int
dsl_dataset_own_obj(dsl_pool_t * dp,uint64_t dsobj,void * tag,dsl_dataset_t ** dsp)563 dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj,
564 void *tag, dsl_dataset_t **dsp)
565 {
566 int err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
567 if (err != 0)
568 return (err);
569 if (!dsl_dataset_tryown(*dsp, tag)) {
570 dsl_dataset_rele(*dsp, tag);
571 *dsp = NULL;
572 return (SET_ERROR(EBUSY));
573 }
574 return (0);
575 }
576
577 int
dsl_dataset_own(dsl_pool_t * dp,const char * name,void * tag,dsl_dataset_t ** dsp)578 dsl_dataset_own(dsl_pool_t *dp, const char *name,
579 void *tag, dsl_dataset_t **dsp)
580 {
581 int err = dsl_dataset_hold(dp, name, tag, dsp);
582 if (err != 0)
583 return (err);
584 if (!dsl_dataset_tryown(*dsp, tag)) {
585 dsl_dataset_rele(*dsp, tag);
586 return (SET_ERROR(EBUSY));
587 }
588 return (0);
589 }
590
591 /*
592 * See the comment above dsl_pool_hold() for details. In summary, a long
593 * hold is used to prevent destruction of a dataset while the pool hold
594 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
595 *
596 * The dataset and pool must be held when this function is called. After it
597 * is called, the pool hold may be released while the dataset is still held
598 * and accessed.
599 */
600 void
dsl_dataset_long_hold(dsl_dataset_t * ds,void * tag)601 dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
602 {
603 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
604 (void) refcount_add(&ds->ds_longholds, tag);
605 }
606
607 void
dsl_dataset_long_rele(dsl_dataset_t * ds,void * tag)608 dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
609 {
610 (void) refcount_remove(&ds->ds_longholds, tag);
611 }
612
613 /* Return B_TRUE if there are any long holds on this dataset. */
614 boolean_t
dsl_dataset_long_held(dsl_dataset_t * ds)615 dsl_dataset_long_held(dsl_dataset_t *ds)
616 {
617 return (!refcount_is_zero(&ds->ds_longholds));
618 }
619
620 void
dsl_dataset_name(dsl_dataset_t * ds,char * name)621 dsl_dataset_name(dsl_dataset_t *ds, char *name)
622 {
623 if (ds == NULL) {
624 (void) strcpy(name, "mos");
625 } else {
626 dsl_dir_name(ds->ds_dir, name);
627 VERIFY0(dsl_dataset_get_snapname(ds));
628 if (ds->ds_snapname[0]) {
629 (void) strcat(name, "@");
630 /*
631 * We use a "recursive" mutex so that we
632 * can call dprintf_ds() with ds_lock held.
633 */
634 if (!MUTEX_HELD(&ds->ds_lock)) {
635 mutex_enter(&ds->ds_lock);
636 (void) strcat(name, ds->ds_snapname);
637 mutex_exit(&ds->ds_lock);
638 } else {
639 (void) strcat(name, ds->ds_snapname);
640 }
641 }
642 }
643 }
644
645 void
dsl_dataset_rele(dsl_dataset_t * ds,void * tag)646 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
647 {
648 dmu_buf_rele(ds->ds_dbuf, tag);
649 }
650
651 void
dsl_dataset_disown(dsl_dataset_t * ds,void * tag)652 dsl_dataset_disown(dsl_dataset_t *ds, void *tag)
653 {
654 ASSERT3P(ds->ds_owner, ==, tag);
655 ASSERT(ds->ds_dbuf != NULL);
656
657 mutex_enter(&ds->ds_lock);
658 ds->ds_owner = NULL;
659 mutex_exit(&ds->ds_lock);
660 dsl_dataset_long_rele(ds, tag);
661 dsl_dataset_rele(ds, tag);
662 }
663
664 boolean_t
dsl_dataset_tryown(dsl_dataset_t * ds,void * tag)665 dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
666 {
667 boolean_t gotit = FALSE;
668
669 mutex_enter(&ds->ds_lock);
670 if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
671 ds->ds_owner = tag;
672 dsl_dataset_long_hold(ds, tag);
673 gotit = TRUE;
674 }
675 mutex_exit(&ds->ds_lock);
676 return (gotit);
677 }
678
679 uint64_t
dsl_dataset_create_sync_dd(dsl_dir_t * dd,dsl_dataset_t * origin,uint64_t flags,dmu_tx_t * tx)680 dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
681 uint64_t flags, dmu_tx_t *tx)
682 {
683 dsl_pool_t *dp = dd->dd_pool;
684 dmu_buf_t *dbuf;
685 dsl_dataset_phys_t *dsphys;
686 uint64_t dsobj;
687 objset_t *mos = dp->dp_meta_objset;
688
689 if (origin == NULL)
690 origin = dp->dp_origin_snap;
691
692 ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
693 ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
694 ASSERT(dmu_tx_is_syncing(tx));
695 ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
696
697 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
698 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
699 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
700 dmu_buf_will_dirty(dbuf, tx);
701 dsphys = dbuf->db_data;
702 bzero(dsphys, sizeof (dsl_dataset_phys_t));
703 dsphys->ds_dir_obj = dd->dd_object;
704 dsphys->ds_flags = flags;
705 dsphys->ds_fsid_guid = unique_create();
706 do {
707 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
708 sizeof (dsphys->ds_guid));
709 } while (dsphys->ds_guid == 0);
710 dsphys->ds_snapnames_zapobj =
711 zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
712 DMU_OT_NONE, 0, tx);
713 dsphys->ds_creation_time = gethrestime_sec();
714 dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
715
716 if (origin == NULL) {
717 dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
718 } else {
719 dsl_dataset_t *ohds; /* head of the origin snapshot */
720
721 dsphys->ds_prev_snap_obj = origin->ds_object;
722 dsphys->ds_prev_snap_txg =
723 dsl_dataset_phys(origin)->ds_creation_txg;
724 dsphys->ds_referenced_bytes =
725 dsl_dataset_phys(origin)->ds_referenced_bytes;
726 dsphys->ds_compressed_bytes =
727 dsl_dataset_phys(origin)->ds_compressed_bytes;
728 dsphys->ds_uncompressed_bytes =
729 dsl_dataset_phys(origin)->ds_uncompressed_bytes;
730 dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
731
732 /*
733 * Inherit flags that describe the dataset's contents
734 * (INCONSISTENT) or properties (Case Insensitive).
735 */
736 dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
737 (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
738
739 if (origin->ds_large_blocks)
740 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
741
742 dmu_buf_will_dirty(origin->ds_dbuf, tx);
743 dsl_dataset_phys(origin)->ds_num_children++;
744
745 VERIFY0(dsl_dataset_hold_obj(dp,
746 dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
747 FTAG, &ohds));
748 dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
749 dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
750 dsl_dataset_rele(ohds, FTAG);
751
752 if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
753 if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
754 dsl_dataset_phys(origin)->ds_next_clones_obj =
755 zap_create(mos,
756 DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
757 }
758 VERIFY0(zap_add_int(mos,
759 dsl_dataset_phys(origin)->ds_next_clones_obj,
760 dsobj, tx));
761 }
762
763 dmu_buf_will_dirty(dd->dd_dbuf, tx);
764 dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
765 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
766 if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
767 dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
768 dsl_dir_phys(origin->ds_dir)->dd_clones =
769 zap_create(mos,
770 DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
771 }
772 VERIFY0(zap_add_int(mos,
773 dsl_dir_phys(origin->ds_dir)->dd_clones,
774 dsobj, tx));
775 }
776 }
777
778 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
779 dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
780
781 dmu_buf_rele(dbuf, FTAG);
782
783 dmu_buf_will_dirty(dd->dd_dbuf, tx);
784 dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
785
786 return (dsobj);
787 }
788
789 static void
dsl_dataset_zero_zil(dsl_dataset_t * ds,dmu_tx_t * tx)790 dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
791 {
792 objset_t *os;
793
794 VERIFY0(dmu_objset_from_ds(ds, &os));
795 bzero(&os->os_zil_header, sizeof (os->os_zil_header));
796 dsl_dataset_dirty(ds, tx);
797 }
798
799 uint64_t
dsl_dataset_create_sync(dsl_dir_t * pdd,const char * lastname,dsl_dataset_t * origin,uint64_t flags,cred_t * cr,dmu_tx_t * tx)800 dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
801 dsl_dataset_t *origin, uint64_t flags, cred_t *cr, dmu_tx_t *tx)
802 {
803 dsl_pool_t *dp = pdd->dd_pool;
804 uint64_t dsobj, ddobj;
805 dsl_dir_t *dd;
806
807 ASSERT(dmu_tx_is_syncing(tx));
808 ASSERT(lastname[0] != '@');
809
810 ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
811 VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
812
813 dsobj = dsl_dataset_create_sync_dd(dd, origin,
814 flags & ~DS_CREATE_FLAG_NODIRTY, tx);
815
816 dsl_deleg_set_create_perms(dd, tx, cr);
817
818 /*
819 * Since we're creating a new node we know it's a leaf, so we can
820 * initialize the counts if the limit feature is active.
821 */
822 if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
823 uint64_t cnt = 0;
824 objset_t *os = dd->dd_pool->dp_meta_objset;
825
826 dsl_dir_zapify(dd, tx);
827 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
828 sizeof (cnt), 1, &cnt, tx));
829 VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
830 sizeof (cnt), 1, &cnt, tx));
831 }
832
833 dsl_dir_rele(dd, FTAG);
834
835 /*
836 * If we are creating a clone, make sure we zero out any stale
837 * data from the origin snapshots zil header.
838 */
839 if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
840 dsl_dataset_t *ds;
841
842 VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
843 dsl_dataset_zero_zil(ds, tx);
844 dsl_dataset_rele(ds, FTAG);
845 }
846
847 return (dsobj);
848 }
849
850 #ifdef __FreeBSD__
851 /* FreeBSD ioctl compat begin */
852 struct destroyarg {
853 nvlist_t *nvl;
854 const char *snapname;
855 };
856
857 static int
dsl_check_snap_cb(const char * name,void * arg)858 dsl_check_snap_cb(const char *name, void *arg)
859 {
860 struct destroyarg *da = arg;
861 dsl_dataset_t *ds;
862 char *dsname;
863
864 dsname = kmem_asprintf("%s@%s", name, da->snapname);
865 fnvlist_add_boolean(da->nvl, dsname);
866 kmem_free(dsname, strlen(dsname) + 1);
867
868 return (0);
869 }
870
871 int
dmu_get_recursive_snaps_nvl(char * fsname,const char * snapname,nvlist_t * snaps)872 dmu_get_recursive_snaps_nvl(char *fsname, const char *snapname,
873 nvlist_t *snaps)
874 {
875 struct destroyarg *da;
876 int err;
877
878 da = kmem_zalloc(sizeof (struct destroyarg), KM_SLEEP);
879 da->nvl = snaps;
880 da->snapname = snapname;
881 err = dmu_objset_find(fsname, dsl_check_snap_cb, da,
882 DS_FIND_CHILDREN);
883 kmem_free(da, sizeof (struct destroyarg));
884
885 return (err);
886 }
887 /* FreeBSD ioctl compat end */
888 #endif /* __FreeBSD__ */
889
890 /*
891 * The unique space in the head dataset can be calculated by subtracting
892 * the space used in the most recent snapshot, that is still being used
893 * in this file system, from the space currently in use. To figure out
894 * the space in the most recent snapshot still in use, we need to take
895 * the total space used in the snapshot and subtract out the space that
896 * has been freed up since the snapshot was taken.
897 */
898 void
dsl_dataset_recalc_head_uniq(dsl_dataset_t * ds)899 dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
900 {
901 uint64_t mrs_used;
902 uint64_t dlused, dlcomp, dluncomp;
903
904 ASSERT(!dsl_dataset_is_snapshot(ds));
905
906 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
907 mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
908 else
909 mrs_used = 0;
910
911 dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
912
913 ASSERT3U(dlused, <=, mrs_used);
914 dsl_dataset_phys(ds)->ds_unique_bytes =
915 dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
916
917 if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
918 SPA_VERSION_UNIQUE_ACCURATE)
919 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
920 }
921
922 void
dsl_dataset_remove_from_next_clones(dsl_dataset_t * ds,uint64_t obj,dmu_tx_t * tx)923 dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
924 dmu_tx_t *tx)
925 {
926 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
927 uint64_t count;
928 int err;
929
930 ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
931 err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
932 obj, tx);
933 /*
934 * The err should not be ENOENT, but a bug in a previous version
935 * of the code could cause upgrade_clones_cb() to not set
936 * ds_next_snap_obj when it should, leading to a missing entry.
937 * If we knew that the pool was created after
938 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
939 * ENOENT. However, at least we can check that we don't have
940 * too many entries in the next_clones_obj even after failing to
941 * remove this one.
942 */
943 if (err != ENOENT)
944 VERIFY0(err);
945 ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
946 &count));
947 ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
948 }
949
950
951 blkptr_t *
dsl_dataset_get_blkptr(dsl_dataset_t * ds)952 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
953 {
954 return (&dsl_dataset_phys(ds)->ds_bp);
955 }
956
957 void
dsl_dataset_set_blkptr(dsl_dataset_t * ds,blkptr_t * bp,dmu_tx_t * tx)958 dsl_dataset_set_blkptr(dsl_dataset_t *ds, blkptr_t *bp, dmu_tx_t *tx)
959 {
960 ASSERT(dmu_tx_is_syncing(tx));
961 /* If it's the meta-objset, set dp_meta_rootbp */
962 if (ds == NULL) {
963 tx->tx_pool->dp_meta_rootbp = *bp;
964 } else {
965 dmu_buf_will_dirty(ds->ds_dbuf, tx);
966 dsl_dataset_phys(ds)->ds_bp = *bp;
967 }
968 }
969
970 spa_t *
dsl_dataset_get_spa(dsl_dataset_t * ds)971 dsl_dataset_get_spa(dsl_dataset_t *ds)
972 {
973 return (ds->ds_dir->dd_pool->dp_spa);
974 }
975
976 void
dsl_dataset_dirty(dsl_dataset_t * ds,dmu_tx_t * tx)977 dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
978 {
979 dsl_pool_t *dp;
980
981 if (ds == NULL) /* this is the meta-objset */
982 return;
983
984 ASSERT(ds->ds_objset != NULL);
985
986 if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
987 panic("dirtying snapshot!");
988
989 dp = ds->ds_dir->dd_pool;
990
991 if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
992 /* up the hold count until we can be written out */
993 dmu_buf_add_ref(ds->ds_dbuf, ds);
994 }
995 }
996
997 boolean_t
dsl_dataset_is_dirty(dsl_dataset_t * ds)998 dsl_dataset_is_dirty(dsl_dataset_t *ds)
999 {
1000 for (int t = 0; t < TXG_SIZE; t++) {
1001 if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1002 ds, t))
1003 return (B_TRUE);
1004 }
1005 return (B_FALSE);
1006 }
1007
1008 static int
dsl_dataset_snapshot_reserve_space(dsl_dataset_t * ds,dmu_tx_t * tx)1009 dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1010 {
1011 uint64_t asize;
1012
1013 if (!dmu_tx_is_syncing(tx))
1014 return (0);
1015
1016 /*
1017 * If there's an fs-only reservation, any blocks that might become
1018 * owned by the snapshot dataset must be accommodated by space
1019 * outside of the reservation.
1020 */
1021 ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1022 asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1023 if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1024 return (SET_ERROR(ENOSPC));
1025
1026 /*
1027 * Propagate any reserved space for this snapshot to other
1028 * snapshot checks in this sync group.
1029 */
1030 if (asize > 0)
1031 dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1032
1033 return (0);
1034 }
1035
1036 typedef struct dsl_dataset_snapshot_arg {
1037 nvlist_t *ddsa_snaps;
1038 nvlist_t *ddsa_props;
1039 nvlist_t *ddsa_errors;
1040 cred_t *ddsa_cr;
1041 } dsl_dataset_snapshot_arg_t;
1042
1043 int
dsl_dataset_snapshot_check_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx,boolean_t recv,uint64_t cnt,cred_t * cr)1044 dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1045 dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1046 {
1047 int error;
1048 uint64_t value;
1049
1050 ds->ds_trysnap_txg = tx->tx_txg;
1051
1052 if (!dmu_tx_is_syncing(tx))
1053 return (0);
1054
1055 /*
1056 * We don't allow multiple snapshots of the same txg. If there
1057 * is already one, try again.
1058 */
1059 if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1060 return (SET_ERROR(EAGAIN));
1061
1062 /*
1063 * Check for conflicting snapshot name.
1064 */
1065 error = dsl_dataset_snap_lookup(ds, snapname, &value);
1066 if (error == 0)
1067 return (SET_ERROR(EEXIST));
1068 if (error != ENOENT)
1069 return (error);
1070
1071 /*
1072 * We don't allow taking snapshots of inconsistent datasets, such as
1073 * those into which we are currently receiving. However, if we are
1074 * creating this snapshot as part of a receive, this check will be
1075 * executed atomically with respect to the completion of the receive
1076 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1077 * case we ignore this, knowing it will be fixed up for us shortly in
1078 * dmu_recv_end_sync().
1079 */
1080 if (!recv && DS_IS_INCONSISTENT(ds))
1081 return (SET_ERROR(EBUSY));
1082
1083 /*
1084 * Skip the check for temporary snapshots or if we have already checked
1085 * the counts in dsl_dataset_snapshot_check. This means we really only
1086 * check the count here when we're receiving a stream.
1087 */
1088 if (cnt != 0 && cr != NULL) {
1089 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1090 ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1091 if (error != 0)
1092 return (error);
1093 }
1094
1095 error = dsl_dataset_snapshot_reserve_space(ds, tx);
1096 if (error != 0)
1097 return (error);
1098
1099 return (0);
1100 }
1101
1102 static int
dsl_dataset_snapshot_check(void * arg,dmu_tx_t * tx)1103 dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1104 {
1105 dsl_dataset_snapshot_arg_t *ddsa = arg;
1106 dsl_pool_t *dp = dmu_tx_pool(tx);
1107 nvpair_t *pair;
1108 int rv = 0;
1109
1110 /*
1111 * Pre-compute how many total new snapshots will be created for each
1112 * level in the tree and below. This is needed for validating the
1113 * snapshot limit when either taking a recursive snapshot or when
1114 * taking multiple snapshots.
1115 *
1116 * The problem is that the counts are not actually adjusted when
1117 * we are checking, only when we finally sync. For a single snapshot,
1118 * this is easy, the count will increase by 1 at each node up the tree,
1119 * but its more complicated for the recursive/multiple snapshot case.
1120 *
1121 * The dsl_fs_ss_limit_check function does recursively check the count
1122 * at each level up the tree but since it is validating each snapshot
1123 * independently we need to be sure that we are validating the complete
1124 * count for the entire set of snapshots. We do this by rolling up the
1125 * counts for each component of the name into an nvlist and then
1126 * checking each of those cases with the aggregated count.
1127 *
1128 * This approach properly handles not only the recursive snapshot
1129 * case (where we get all of those on the ddsa_snaps list) but also
1130 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1131 * validate the limit on 'a' using a count of 2).
1132 *
1133 * We validate the snapshot names in the third loop and only report
1134 * name errors once.
1135 */
1136 if (dmu_tx_is_syncing(tx)) {
1137 nvlist_t *cnt_track = NULL;
1138 cnt_track = fnvlist_alloc();
1139
1140 /* Rollup aggregated counts into the cnt_track list */
1141 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1142 pair != NULL;
1143 pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1144 char *pdelim;
1145 uint64_t val;
1146 char nm[MAXPATHLEN];
1147
1148 (void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1149 pdelim = strchr(nm, '@');
1150 if (pdelim == NULL)
1151 continue;
1152 *pdelim = '\0';
1153
1154 do {
1155 if (nvlist_lookup_uint64(cnt_track, nm,
1156 &val) == 0) {
1157 /* update existing entry */
1158 fnvlist_add_uint64(cnt_track, nm,
1159 val + 1);
1160 } else {
1161 /* add to list */
1162 fnvlist_add_uint64(cnt_track, nm, 1);
1163 }
1164
1165 pdelim = strrchr(nm, '/');
1166 if (pdelim != NULL)
1167 *pdelim = '\0';
1168 } while (pdelim != NULL);
1169 }
1170
1171 /* Check aggregated counts at each level */
1172 for (pair = nvlist_next_nvpair(cnt_track, NULL);
1173 pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1174 int error = 0;
1175 char *name;
1176 uint64_t cnt = 0;
1177 dsl_dataset_t *ds;
1178
1179 name = nvpair_name(pair);
1180 cnt = fnvpair_value_uint64(pair);
1181 ASSERT(cnt > 0);
1182
1183 error = dsl_dataset_hold(dp, name, FTAG, &ds);
1184 if (error == 0) {
1185 error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1186 ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1187 ddsa->ddsa_cr);
1188 dsl_dataset_rele(ds, FTAG);
1189 }
1190
1191 if (error != 0) {
1192 if (ddsa->ddsa_errors != NULL)
1193 fnvlist_add_int32(ddsa->ddsa_errors,
1194 name, error);
1195 rv = error;
1196 /* only report one error for this check */
1197 break;
1198 }
1199 }
1200 nvlist_free(cnt_track);
1201 }
1202
1203 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1204 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1205 int error = 0;
1206 dsl_dataset_t *ds;
1207 char *name, *atp;
1208 char dsname[MAXNAMELEN];
1209
1210 name = nvpair_name(pair);
1211 if (strlen(name) >= MAXNAMELEN)
1212 error = SET_ERROR(ENAMETOOLONG);
1213 if (error == 0) {
1214 atp = strchr(name, '@');
1215 if (atp == NULL)
1216 error = SET_ERROR(EINVAL);
1217 if (error == 0)
1218 (void) strlcpy(dsname, name, atp - name + 1);
1219 }
1220 if (error == 0)
1221 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1222 if (error == 0) {
1223 /* passing 0/NULL skips dsl_fs_ss_limit_check */
1224 error = dsl_dataset_snapshot_check_impl(ds,
1225 atp + 1, tx, B_FALSE, 0, NULL);
1226 dsl_dataset_rele(ds, FTAG);
1227 }
1228
1229 if (error != 0) {
1230 if (ddsa->ddsa_errors != NULL) {
1231 fnvlist_add_int32(ddsa->ddsa_errors,
1232 name, error);
1233 }
1234 rv = error;
1235 }
1236 }
1237
1238 return (rv);
1239 }
1240
1241 void
dsl_dataset_snapshot_sync_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx)1242 dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1243 dmu_tx_t *tx)
1244 {
1245 static zil_header_t zero_zil;
1246
1247 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1248 dmu_buf_t *dbuf;
1249 dsl_dataset_phys_t *dsphys;
1250 uint64_t dsobj, crtxg;
1251 objset_t *mos = dp->dp_meta_objset;
1252 objset_t *os;
1253
1254 ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1255
1256 /*
1257 * If we are on an old pool, the zil must not be active, in which
1258 * case it will be zeroed. Usually zil_suspend() accomplishes this.
1259 */
1260 ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1261 dmu_objset_from_ds(ds, &os) != 0 ||
1262 bcmp(&os->os_phys->os_zil_header, &zero_zil,
1263 sizeof (zero_zil)) == 0);
1264
1265 dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1266
1267 /*
1268 * The origin's ds_creation_txg has to be < TXG_INITIAL
1269 */
1270 if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1271 crtxg = 1;
1272 else
1273 crtxg = tx->tx_txg;
1274
1275 dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1276 DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1277 VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1278 dmu_buf_will_dirty(dbuf, tx);
1279 dsphys = dbuf->db_data;
1280 bzero(dsphys, sizeof (dsl_dataset_phys_t));
1281 dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1282 dsphys->ds_fsid_guid = unique_create();
1283 do {
1284 (void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1285 sizeof (dsphys->ds_guid));
1286 } while (dsphys->ds_guid == 0);
1287 dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1288 dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1289 dsphys->ds_next_snap_obj = ds->ds_object;
1290 dsphys->ds_num_children = 1;
1291 dsphys->ds_creation_time = gethrestime_sec();
1292 dsphys->ds_creation_txg = crtxg;
1293 dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1294 dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1295 dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1296 dsphys->ds_uncompressed_bytes =
1297 dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1298 dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1299 dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1300 dmu_buf_rele(dbuf, FTAG);
1301
1302 if (ds->ds_large_blocks)
1303 dsl_dataset_activate_large_blocks_sync_impl(dsobj, tx);
1304
1305 ASSERT3U(ds->ds_prev != 0, ==,
1306 dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1307 if (ds->ds_prev) {
1308 uint64_t next_clones_obj =
1309 dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1310 ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1311 ds->ds_object ||
1312 dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1313 if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1314 ds->ds_object) {
1315 dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1316 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1317 dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1318 dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1319 } else if (next_clones_obj != 0) {
1320 dsl_dataset_remove_from_next_clones(ds->ds_prev,
1321 dsphys->ds_next_snap_obj, tx);
1322 VERIFY0(zap_add_int(mos,
1323 next_clones_obj, dsobj, tx));
1324 }
1325 }
1326
1327 /*
1328 * If we have a reference-reservation on this dataset, we will
1329 * need to increase the amount of refreservation being charged
1330 * since our unique space is going to zero.
1331 */
1332 if (ds->ds_reserved) {
1333 int64_t delta;
1334 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1335 delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1336 ds->ds_reserved);
1337 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1338 delta, 0, 0, tx);
1339 }
1340
1341 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1342 dsl_dataset_phys(ds)->ds_deadlist_obj =
1343 dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1344 dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1345 dsl_deadlist_close(&ds->ds_deadlist);
1346 dsl_deadlist_open(&ds->ds_deadlist, mos,
1347 dsl_dataset_phys(ds)->ds_deadlist_obj);
1348 dsl_deadlist_add_key(&ds->ds_deadlist,
1349 dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1350
1351 ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1352 dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1353 dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1354 dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1355 if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1356 dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1357
1358 VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1359 snapname, 8, 1, &dsobj, tx));
1360
1361 if (ds->ds_prev)
1362 dsl_dataset_rele(ds->ds_prev, ds);
1363 VERIFY0(dsl_dataset_hold_obj(dp,
1364 dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1365
1366 dsl_scan_ds_snapshotted(ds, tx);
1367
1368 dsl_dir_snap_cmtime_update(ds->ds_dir);
1369
1370 spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1371 }
1372
1373 static void
dsl_dataset_snapshot_sync(void * arg,dmu_tx_t * tx)1374 dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1375 {
1376 dsl_dataset_snapshot_arg_t *ddsa = arg;
1377 dsl_pool_t *dp = dmu_tx_pool(tx);
1378 nvpair_t *pair;
1379
1380 for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1381 pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1382 dsl_dataset_t *ds;
1383 char *name, *atp;
1384 char dsname[MAXNAMELEN];
1385
1386 name = nvpair_name(pair);
1387 atp = strchr(name, '@');
1388 (void) strlcpy(dsname, name, atp - name + 1);
1389 VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1390
1391 dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1392 if (ddsa->ddsa_props != NULL) {
1393 dsl_props_set_sync_impl(ds->ds_prev,
1394 ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1395 }
1396 dsl_dataset_rele(ds, FTAG);
1397 }
1398 }
1399
1400 /*
1401 * The snapshots must all be in the same pool.
1402 * All-or-nothing: if there are any failures, nothing will be modified.
1403 */
1404 int
dsl_dataset_snapshot(nvlist_t * snaps,nvlist_t * props,nvlist_t * errors)1405 dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1406 {
1407 dsl_dataset_snapshot_arg_t ddsa;
1408 nvpair_t *pair;
1409 boolean_t needsuspend;
1410 int error;
1411 spa_t *spa;
1412 char *firstname;
1413 nvlist_t *suspended = NULL;
1414
1415 pair = nvlist_next_nvpair(snaps, NULL);
1416 if (pair == NULL)
1417 return (0);
1418 firstname = nvpair_name(pair);
1419
1420 error = spa_open(firstname, &spa, FTAG);
1421 if (error != 0)
1422 return (error);
1423 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1424 spa_close(spa, FTAG);
1425
1426 if (needsuspend) {
1427 suspended = fnvlist_alloc();
1428 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1429 pair = nvlist_next_nvpair(snaps, pair)) {
1430 char fsname[MAXNAMELEN];
1431 char *snapname = nvpair_name(pair);
1432 char *atp;
1433 void *cookie;
1434
1435 atp = strchr(snapname, '@');
1436 if (atp == NULL) {
1437 error = SET_ERROR(EINVAL);
1438 break;
1439 }
1440 (void) strlcpy(fsname, snapname, atp - snapname + 1);
1441
1442 error = zil_suspend(fsname, &cookie);
1443 if (error != 0)
1444 break;
1445 fnvlist_add_uint64(suspended, fsname,
1446 (uintptr_t)cookie);
1447 }
1448 }
1449
1450 ddsa.ddsa_snaps = snaps;
1451 ddsa.ddsa_props = props;
1452 ddsa.ddsa_errors = errors;
1453 ddsa.ddsa_cr = CRED();
1454
1455 if (error == 0) {
1456 error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1457 dsl_dataset_snapshot_sync, &ddsa,
1458 fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1459 }
1460
1461 if (suspended != NULL) {
1462 for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1463 pair = nvlist_next_nvpair(suspended, pair)) {
1464 zil_resume((void *)(uintptr_t)
1465 fnvpair_value_uint64(pair));
1466 }
1467 fnvlist_free(suspended);
1468 }
1469
1470 #ifdef __FreeBSD__
1471 #ifdef _KERNEL
1472 if (error == 0) {
1473 for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1474 pair = nvlist_next_nvpair(snaps, pair)) {
1475 char *snapname = nvpair_name(pair);
1476 zvol_create_minors(snapname);
1477 }
1478 }
1479 #endif
1480 #endif
1481 return (error);
1482 }
1483
1484 typedef struct dsl_dataset_snapshot_tmp_arg {
1485 const char *ddsta_fsname;
1486 const char *ddsta_snapname;
1487 minor_t ddsta_cleanup_minor;
1488 const char *ddsta_htag;
1489 } dsl_dataset_snapshot_tmp_arg_t;
1490
1491 static int
dsl_dataset_snapshot_tmp_check(void * arg,dmu_tx_t * tx)1492 dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1493 {
1494 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1495 dsl_pool_t *dp = dmu_tx_pool(tx);
1496 dsl_dataset_t *ds;
1497 int error;
1498
1499 error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1500 if (error != 0)
1501 return (error);
1502
1503 /* NULL cred means no limit check for tmp snapshot */
1504 error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1505 tx, B_FALSE, 0, NULL);
1506 if (error != 0) {
1507 dsl_dataset_rele(ds, FTAG);
1508 return (error);
1509 }
1510
1511 if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1512 dsl_dataset_rele(ds, FTAG);
1513 return (SET_ERROR(ENOTSUP));
1514 }
1515 error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1516 B_TRUE, tx);
1517 if (error != 0) {
1518 dsl_dataset_rele(ds, FTAG);
1519 return (error);
1520 }
1521
1522 dsl_dataset_rele(ds, FTAG);
1523 return (0);
1524 }
1525
1526 static void
dsl_dataset_snapshot_tmp_sync(void * arg,dmu_tx_t * tx)1527 dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1528 {
1529 dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1530 dsl_pool_t *dp = dmu_tx_pool(tx);
1531 dsl_dataset_t *ds;
1532
1533 VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1534
1535 dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1536 dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1537 ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1538 dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1539
1540 dsl_dataset_rele(ds, FTAG);
1541 }
1542
1543 int
dsl_dataset_snapshot_tmp(const char * fsname,const char * snapname,minor_t cleanup_minor,const char * htag)1544 dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1545 minor_t cleanup_minor, const char *htag)
1546 {
1547 dsl_dataset_snapshot_tmp_arg_t ddsta;
1548 int error;
1549 spa_t *spa;
1550 boolean_t needsuspend;
1551 void *cookie;
1552
1553 ddsta.ddsta_fsname = fsname;
1554 ddsta.ddsta_snapname = snapname;
1555 ddsta.ddsta_cleanup_minor = cleanup_minor;
1556 ddsta.ddsta_htag = htag;
1557
1558 error = spa_open(fsname, &spa, FTAG);
1559 if (error != 0)
1560 return (error);
1561 needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1562 spa_close(spa, FTAG);
1563
1564 if (needsuspend) {
1565 error = zil_suspend(fsname, &cookie);
1566 if (error != 0)
1567 return (error);
1568 }
1569
1570 error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1571 dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1572
1573 if (needsuspend)
1574 zil_resume(cookie);
1575 return (error);
1576 }
1577
1578
1579 void
dsl_dataset_sync(dsl_dataset_t * ds,zio_t * zio,dmu_tx_t * tx)1580 dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1581 {
1582 ASSERT(dmu_tx_is_syncing(tx));
1583 ASSERT(ds->ds_objset != NULL);
1584 ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1585
1586 /*
1587 * in case we had to change ds_fsid_guid when we opened it,
1588 * sync it out now.
1589 */
1590 dmu_buf_will_dirty(ds->ds_dbuf, tx);
1591 dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1592
1593 dmu_objset_sync(ds->ds_objset, zio, tx);
1594
1595 if (ds->ds_need_large_blocks && !ds->ds_large_blocks) {
1596 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
1597 ds->ds_large_blocks = B_TRUE;
1598 }
1599 }
1600
1601 static void
get_clones_stat(dsl_dataset_t * ds,nvlist_t * nv)1602 get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1603 {
1604 uint64_t count = 0;
1605 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1606 zap_cursor_t zc;
1607 zap_attribute_t za;
1608 nvlist_t *propval = fnvlist_alloc();
1609 nvlist_t *val = fnvlist_alloc();
1610
1611 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1612
1613 /*
1614 * There may be missing entries in ds_next_clones_obj
1615 * due to a bug in a previous version of the code.
1616 * Only trust it if it has the right number of entries.
1617 */
1618 if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1619 VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1620 &count));
1621 }
1622 if (count != dsl_dataset_phys(ds)->ds_num_children - 1)
1623 goto fail;
1624 for (zap_cursor_init(&zc, mos,
1625 dsl_dataset_phys(ds)->ds_next_clones_obj);
1626 zap_cursor_retrieve(&zc, &za) == 0;
1627 zap_cursor_advance(&zc)) {
1628 dsl_dataset_t *clone;
1629 char buf[ZFS_MAXNAMELEN];
1630 VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1631 za.za_first_integer, FTAG, &clone));
1632 dsl_dir_name(clone->ds_dir, buf);
1633 fnvlist_add_boolean(val, buf);
1634 dsl_dataset_rele(clone, FTAG);
1635 }
1636 zap_cursor_fini(&zc);
1637 fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1638 fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES), propval);
1639 fail:
1640 nvlist_free(val);
1641 nvlist_free(propval);
1642 }
1643
1644 void
dsl_dataset_stats(dsl_dataset_t * ds,nvlist_t * nv)1645 dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
1646 {
1647 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1648 uint64_t refd, avail, uobjs, aobjs, ratio;
1649
1650 ASSERT(dsl_pool_config_held(dp));
1651
1652 ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
1653 (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
1654 dsl_dataset_phys(ds)->ds_compressed_bytes);
1655
1656 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO, ratio);
1657 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
1658 dsl_dataset_phys(ds)->ds_uncompressed_bytes);
1659
1660 if (dsl_dataset_is_snapshot(ds)) {
1661 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO, ratio);
1662 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
1663 dsl_dataset_phys(ds)->ds_unique_bytes);
1664 get_clones_stat(ds, nv);
1665 } else {
1666 if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
1667 char buf[MAXNAMELEN];
1668 dsl_dataset_name(ds->ds_prev, buf);
1669 dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP, buf);
1670 }
1671
1672 dsl_dir_stats(ds->ds_dir, nv);
1673 }
1674
1675 dsl_dataset_space(ds, &refd, &avail, &uobjs, &aobjs);
1676 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE, avail);
1677 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED, refd);
1678
1679 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
1680 dsl_dataset_phys(ds)->ds_creation_time);
1681 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
1682 dsl_dataset_phys(ds)->ds_creation_txg);
1683 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
1684 ds->ds_quota);
1685 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
1686 ds->ds_reserved);
1687 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
1688 dsl_dataset_phys(ds)->ds_guid);
1689 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
1690 dsl_dataset_phys(ds)->ds_unique_bytes);
1691 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
1692 ds->ds_object);
1693 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
1694 ds->ds_userrefs);
1695 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
1696 DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
1697
1698 if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
1699 uint64_t written, comp, uncomp;
1700 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1701 dsl_dataset_t *prev;
1702
1703 int err = dsl_dataset_hold_obj(dp,
1704 dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
1705 if (err == 0) {
1706 err = dsl_dataset_space_written(prev, ds, &written,
1707 &comp, &uncomp);
1708 dsl_dataset_rele(prev, FTAG);
1709 if (err == 0) {
1710 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
1711 written);
1712 }
1713 }
1714 }
1715 }
1716
1717 void
dsl_dataset_fast_stat(dsl_dataset_t * ds,dmu_objset_stats_t * stat)1718 dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
1719 {
1720 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1721 ASSERT(dsl_pool_config_held(dp));
1722
1723 stat->dds_creation_txg = dsl_dataset_phys(ds)->ds_creation_txg;
1724 stat->dds_inconsistent =
1725 dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT;
1726 stat->dds_guid = dsl_dataset_phys(ds)->ds_guid;
1727 stat->dds_origin[0] = '\0';
1728 if (dsl_dataset_is_snapshot(ds)) {
1729 stat->dds_is_snapshot = B_TRUE;
1730 stat->dds_num_clones =
1731 dsl_dataset_phys(ds)->ds_num_children - 1;
1732 } else {
1733 stat->dds_is_snapshot = B_FALSE;
1734 stat->dds_num_clones = 0;
1735
1736 if (dsl_dir_is_clone(ds->ds_dir)) {
1737 dsl_dataset_t *ods;
1738
1739 VERIFY0(dsl_dataset_hold_obj(dp,
1740 dsl_dir_phys(ds->ds_dir)->dd_origin_obj,
1741 FTAG, &ods));
1742 dsl_dataset_name(ods, stat->dds_origin);
1743 dsl_dataset_rele(ods, FTAG);
1744 }
1745 }
1746 }
1747
1748 uint64_t
dsl_dataset_fsid_guid(dsl_dataset_t * ds)1749 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
1750 {
1751 return (ds->ds_fsid_guid);
1752 }
1753
1754 void
dsl_dataset_space(dsl_dataset_t * ds,uint64_t * refdbytesp,uint64_t * availbytesp,uint64_t * usedobjsp,uint64_t * availobjsp)1755 dsl_dataset_space(dsl_dataset_t *ds,
1756 uint64_t *refdbytesp, uint64_t *availbytesp,
1757 uint64_t *usedobjsp, uint64_t *availobjsp)
1758 {
1759 *refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
1760 *availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
1761 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
1762 *availbytesp +=
1763 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
1764 if (ds->ds_quota != 0) {
1765 /*
1766 * Adjust available bytes according to refquota
1767 */
1768 if (*refdbytesp < ds->ds_quota)
1769 *availbytesp = MIN(*availbytesp,
1770 ds->ds_quota - *refdbytesp);
1771 else
1772 *availbytesp = 0;
1773 }
1774 *usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
1775 *availobjsp = DN_MAX_OBJECT - *usedobjsp;
1776 }
1777
1778 boolean_t
dsl_dataset_modified_since_snap(dsl_dataset_t * ds,dsl_dataset_t * snap)1779 dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
1780 {
1781 dsl_pool_t *dp = ds->ds_dir->dd_pool;
1782
1783 ASSERT(dsl_pool_config_held(dp));
1784 if (snap == NULL)
1785 return (B_FALSE);
1786 if (dsl_dataset_phys(ds)->ds_bp.blk_birth >
1787 dsl_dataset_phys(snap)->ds_creation_txg) {
1788 objset_t *os, *os_snap;
1789 /*
1790 * It may be that only the ZIL differs, because it was
1791 * reset in the head. Don't count that as being
1792 * modified.
1793 */
1794 if (dmu_objset_from_ds(ds, &os) != 0)
1795 return (B_TRUE);
1796 if (dmu_objset_from_ds(snap, &os_snap) != 0)
1797 return (B_TRUE);
1798 return (bcmp(&os->os_phys->os_meta_dnode,
1799 &os_snap->os_phys->os_meta_dnode,
1800 sizeof (os->os_phys->os_meta_dnode)) != 0);
1801 }
1802 return (B_FALSE);
1803 }
1804
1805 typedef struct dsl_dataset_rename_snapshot_arg {
1806 const char *ddrsa_fsname;
1807 const char *ddrsa_oldsnapname;
1808 const char *ddrsa_newsnapname;
1809 boolean_t ddrsa_recursive;
1810 dmu_tx_t *ddrsa_tx;
1811 } dsl_dataset_rename_snapshot_arg_t;
1812
1813 /* ARGSUSED */
1814 static int
dsl_dataset_rename_snapshot_check_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)1815 dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
1816 dsl_dataset_t *hds, void *arg)
1817 {
1818 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1819 int error;
1820 uint64_t val;
1821
1822 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1823 if (error != 0) {
1824 /* ignore nonexistent snapshots */
1825 return (error == ENOENT ? 0 : error);
1826 }
1827
1828 /* new name should not exist */
1829 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
1830 if (error == 0)
1831 error = SET_ERROR(EEXIST);
1832 else if (error == ENOENT)
1833 error = 0;
1834
1835 /* dataset name + 1 for the "@" + the new snapshot name must fit */
1836 if (dsl_dir_namelen(hds->ds_dir) + 1 +
1837 strlen(ddrsa->ddrsa_newsnapname) >= MAXNAMELEN)
1838 error = SET_ERROR(ENAMETOOLONG);
1839
1840 return (error);
1841 }
1842
1843 static int
dsl_dataset_rename_snapshot_check(void * arg,dmu_tx_t * tx)1844 dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
1845 {
1846 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1847 dsl_pool_t *dp = dmu_tx_pool(tx);
1848 dsl_dataset_t *hds;
1849 int error;
1850
1851 error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
1852 if (error != 0)
1853 return (error);
1854
1855 if (ddrsa->ddrsa_recursive) {
1856 error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1857 dsl_dataset_rename_snapshot_check_impl, ddrsa,
1858 DS_FIND_CHILDREN);
1859 } else {
1860 error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
1861 }
1862 dsl_dataset_rele(hds, FTAG);
1863 return (error);
1864 }
1865
1866 static int
dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)1867 dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
1868 dsl_dataset_t *hds, void *arg)
1869 {
1870 #ifdef __FreeBSD__
1871 #ifdef _KERNEL
1872 char *oldname, *newname;
1873 #endif
1874 #endif
1875 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1876 dsl_dataset_t *ds;
1877 uint64_t val;
1878 dmu_tx_t *tx = ddrsa->ddrsa_tx;
1879 int error;
1880
1881 error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
1882 ASSERT(error == 0 || error == ENOENT);
1883 if (error == ENOENT) {
1884 /* ignore nonexistent snapshots */
1885 return (0);
1886 }
1887
1888 VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
1889
1890 /* log before we change the name */
1891 spa_history_log_internal_ds(ds, "rename", tx,
1892 "-> @%s", ddrsa->ddrsa_newsnapname);
1893
1894 VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
1895 B_FALSE));
1896 mutex_enter(&ds->ds_lock);
1897 (void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
1898 mutex_exit(&ds->ds_lock);
1899 VERIFY0(zap_add(dp->dp_meta_objset,
1900 dsl_dataset_phys(hds)->ds_snapnames_zapobj,
1901 ds->ds_snapname, 8, 1, &ds->ds_object, tx));
1902
1903 #ifdef __FreeBSD__
1904 #ifdef _KERNEL
1905 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1906 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1907 snprintf(oldname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1908 ddrsa->ddrsa_oldsnapname);
1909 snprintf(newname, MAXPATHLEN, "%s@%s", ddrsa->ddrsa_fsname,
1910 ddrsa->ddrsa_newsnapname);
1911 zfsvfs_update_fromname(oldname, newname);
1912 zvol_rename_minors(oldname, newname);
1913 dsl_event_notify_rename(dsl_dataset_get_spa(ds), oldname, newname, ESC_ZFS_DATASET_RENAME);
1914 kmem_free(newname, MAXPATHLEN);
1915 kmem_free(oldname, MAXPATHLEN);
1916 #endif
1917 #endif
1918 dsl_dataset_rele(ds, FTAG);
1919
1920 return (0);
1921 }
1922
1923 static void
dsl_dataset_rename_snapshot_sync(void * arg,dmu_tx_t * tx)1924 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
1925 {
1926 dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
1927 dsl_pool_t *dp = dmu_tx_pool(tx);
1928 dsl_dataset_t *hds;
1929
1930 VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
1931 ddrsa->ddrsa_tx = tx;
1932 if (ddrsa->ddrsa_recursive) {
1933 VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
1934 dsl_dataset_rename_snapshot_sync_impl, ddrsa,
1935 DS_FIND_CHILDREN));
1936 } else {
1937 VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
1938 }
1939 dsl_dataset_rele(hds, FTAG);
1940 }
1941
1942 int
dsl_dataset_rename_snapshot(const char * fsname,const char * oldsnapname,const char * newsnapname,boolean_t recursive)1943 dsl_dataset_rename_snapshot(const char *fsname,
1944 const char *oldsnapname, const char *newsnapname, boolean_t recursive)
1945 {
1946 dsl_dataset_rename_snapshot_arg_t ddrsa;
1947
1948 ddrsa.ddrsa_fsname = fsname;
1949 ddrsa.ddrsa_oldsnapname = oldsnapname;
1950 ddrsa.ddrsa_newsnapname = newsnapname;
1951 ddrsa.ddrsa_recursive = recursive;
1952
1953 return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
1954 dsl_dataset_rename_snapshot_sync, &ddrsa,
1955 1, ZFS_SPACE_CHECK_RESERVED));
1956 }
1957
1958 /*
1959 * If we're doing an ownership handoff, we need to make sure that there is
1960 * only one long hold on the dataset. We're not allowed to change anything here
1961 * so we don't permanently release the long hold or regular hold here. We want
1962 * to do this only when syncing to avoid the dataset unexpectedly going away
1963 * when we release the long hold.
1964 */
1965 static int
dsl_dataset_handoff_check(dsl_dataset_t * ds,void * owner,dmu_tx_t * tx)1966 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
1967 {
1968 boolean_t held;
1969
1970 if (!dmu_tx_is_syncing(tx))
1971 return (0);
1972
1973 if (owner != NULL) {
1974 VERIFY3P(ds->ds_owner, ==, owner);
1975 dsl_dataset_long_rele(ds, owner);
1976 }
1977
1978 held = dsl_dataset_long_held(ds);
1979
1980 if (owner != NULL)
1981 dsl_dataset_long_hold(ds, owner);
1982
1983 if (held)
1984 return (SET_ERROR(EBUSY));
1985
1986 return (0);
1987 }
1988
1989 typedef struct dsl_dataset_rollback_arg {
1990 const char *ddra_fsname;
1991 void *ddra_owner;
1992 nvlist_t *ddra_result;
1993 } dsl_dataset_rollback_arg_t;
1994
1995 static int
dsl_dataset_rollback_check(void * arg,dmu_tx_t * tx)1996 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
1997 {
1998 dsl_dataset_rollback_arg_t *ddra = arg;
1999 dsl_pool_t *dp = dmu_tx_pool(tx);
2000 dsl_dataset_t *ds;
2001 int64_t unused_refres_delta;
2002 int error;
2003
2004 error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2005 if (error != 0)
2006 return (error);
2007
2008 /* must not be a snapshot */
2009 if (dsl_dataset_is_snapshot(ds)) {
2010 dsl_dataset_rele(ds, FTAG);
2011 return (SET_ERROR(EINVAL));
2012 }
2013
2014 /* must have a most recent snapshot */
2015 if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2016 dsl_dataset_rele(ds, FTAG);
2017 return (SET_ERROR(EINVAL));
2018 }
2019
2020 /* must not have any bookmarks after the most recent snapshot */
2021 nvlist_t *proprequest = fnvlist_alloc();
2022 fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2023 nvlist_t *bookmarks = fnvlist_alloc();
2024 error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2025 fnvlist_free(proprequest);
2026 if (error != 0)
2027 return (error);
2028 for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2029 pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2030 nvlist_t *valuenv =
2031 fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2032 zfs_prop_to_name(ZFS_PROP_CREATETXG));
2033 uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2034 if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2035 fnvlist_free(bookmarks);
2036 dsl_dataset_rele(ds, FTAG);
2037 return (SET_ERROR(EEXIST));
2038 }
2039 }
2040 fnvlist_free(bookmarks);
2041
2042 error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2043 if (error != 0) {
2044 dsl_dataset_rele(ds, FTAG);
2045 return (error);
2046 }
2047
2048 /*
2049 * Check if the snap we are rolling back to uses more than
2050 * the refquota.
2051 */
2052 if (ds->ds_quota != 0 &&
2053 dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2054 dsl_dataset_rele(ds, FTAG);
2055 return (SET_ERROR(EDQUOT));
2056 }
2057
2058 /*
2059 * When we do the clone swap, we will temporarily use more space
2060 * due to the refreservation (the head will no longer have any
2061 * unique space, so the entire amount of the refreservation will need
2062 * to be free). We will immediately destroy the clone, freeing
2063 * this space, but the freeing happens over many txg's.
2064 */
2065 unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2066 dsl_dataset_phys(ds)->ds_unique_bytes);
2067
2068 if (unused_refres_delta > 0 &&
2069 unused_refres_delta >
2070 dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2071 dsl_dataset_rele(ds, FTAG);
2072 return (SET_ERROR(ENOSPC));
2073 }
2074
2075 dsl_dataset_rele(ds, FTAG);
2076 return (0);
2077 }
2078
2079 static void
dsl_dataset_rollback_sync(void * arg,dmu_tx_t * tx)2080 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2081 {
2082 dsl_dataset_rollback_arg_t *ddra = arg;
2083 dsl_pool_t *dp = dmu_tx_pool(tx);
2084 dsl_dataset_t *ds, *clone;
2085 uint64_t cloneobj;
2086 char namebuf[ZFS_MAXNAMELEN];
2087
2088 VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2089
2090 dsl_dataset_name(ds->ds_prev, namebuf);
2091 fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2092
2093 cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2094 ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, tx);
2095
2096 VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2097
2098 dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2099 dsl_dataset_zero_zil(ds, tx);
2100
2101 dsl_destroy_head_sync_impl(clone, tx);
2102
2103 dsl_dataset_rele(clone, FTAG);
2104 dsl_dataset_rele(ds, FTAG);
2105 }
2106
2107 /*
2108 * Rolls back the given filesystem or volume to the most recent snapshot.
2109 * The name of the most recent snapshot will be returned under key "target"
2110 * in the result nvlist.
2111 *
2112 * If owner != NULL:
2113 * - The existing dataset MUST be owned by the specified owner at entry
2114 * - Upon return, dataset will still be held by the same owner, whether we
2115 * succeed or not.
2116 *
2117 * This mode is required any time the existing filesystem is mounted. See
2118 * notes above zfs_suspend_fs() for further details.
2119 */
2120 int
dsl_dataset_rollback(const char * fsname,void * owner,nvlist_t * result)2121 dsl_dataset_rollback(const char *fsname, void *owner, nvlist_t *result)
2122 {
2123 dsl_dataset_rollback_arg_t ddra;
2124
2125 ddra.ddra_fsname = fsname;
2126 ddra.ddra_owner = owner;
2127 ddra.ddra_result = result;
2128
2129 return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2130 dsl_dataset_rollback_sync, &ddra,
2131 1, ZFS_SPACE_CHECK_RESERVED));
2132 }
2133
2134 struct promotenode {
2135 list_node_t link;
2136 dsl_dataset_t *ds;
2137 };
2138
2139 typedef struct dsl_dataset_promote_arg {
2140 const char *ddpa_clonename;
2141 dsl_dataset_t *ddpa_clone;
2142 list_t shared_snaps, origin_snaps, clone_snaps;
2143 dsl_dataset_t *origin_origin; /* origin of the origin */
2144 uint64_t used, comp, uncomp, unique, cloneusedsnap, originusedsnap;
2145 char *err_ds;
2146 cred_t *cr;
2147 } dsl_dataset_promote_arg_t;
2148
2149 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2150 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2151 void *tag);
2152 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2153
2154 static int
dsl_dataset_promote_check(void * arg,dmu_tx_t * tx)2155 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2156 {
2157 dsl_dataset_promote_arg_t *ddpa = arg;
2158 dsl_pool_t *dp = dmu_tx_pool(tx);
2159 dsl_dataset_t *hds;
2160 struct promotenode *snap;
2161 dsl_dataset_t *origin_ds;
2162 int err;
2163 uint64_t unused;
2164 uint64_t ss_mv_cnt;
2165 size_t max_snap_len;
2166
2167 err = promote_hold(ddpa, dp, FTAG);
2168 if (err != 0)
2169 return (err);
2170
2171 hds = ddpa->ddpa_clone;
2172 max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2173
2174 if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2175 promote_rele(ddpa, FTAG);
2176 return (SET_ERROR(EXDEV));
2177 }
2178
2179 /*
2180 * Compute and check the amount of space to transfer. Since this is
2181 * so expensive, don't do the preliminary check.
2182 */
2183 if (!dmu_tx_is_syncing(tx)) {
2184 promote_rele(ddpa, FTAG);
2185 return (0);
2186 }
2187
2188 snap = list_head(&ddpa->shared_snaps);
2189 origin_ds = snap->ds;
2190
2191 /* compute origin's new unique space */
2192 snap = list_tail(&ddpa->clone_snaps);
2193 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2194 origin_ds->ds_object);
2195 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2196 dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2197 &ddpa->unique, &unused, &unused);
2198
2199 /*
2200 * Walk the snapshots that we are moving
2201 *
2202 * Compute space to transfer. Consider the incremental changes
2203 * to used by each snapshot:
2204 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2205 * So each snapshot gave birth to:
2206 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2207 * So a sequence would look like:
2208 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2209 * Which simplifies to:
2210 * uN + kN + kN-1 + ... + k1 + k0
2211 * Note however, if we stop before we reach the ORIGIN we get:
2212 * uN + kN + kN-1 + ... + kM - uM-1
2213 */
2214 ss_mv_cnt = 0;
2215 ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2216 ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2217 ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2218 for (snap = list_head(&ddpa->shared_snaps); snap;
2219 snap = list_next(&ddpa->shared_snaps, snap)) {
2220 uint64_t val, dlused, dlcomp, dluncomp;
2221 dsl_dataset_t *ds = snap->ds;
2222
2223 ss_mv_cnt++;
2224
2225 /*
2226 * If there are long holds, we won't be able to evict
2227 * the objset.
2228 */
2229 if (dsl_dataset_long_held(ds)) {
2230 err = SET_ERROR(EBUSY);
2231 goto out;
2232 }
2233
2234 /* Check that the snapshot name does not conflict */
2235 VERIFY0(dsl_dataset_get_snapname(ds));
2236 if (strlen(ds->ds_snapname) >= max_snap_len) {
2237 err = SET_ERROR(ENAMETOOLONG);
2238 goto out;
2239 }
2240 err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2241 if (err == 0) {
2242 (void) strcpy(ddpa->err_ds, snap->ds->ds_snapname);
2243 err = SET_ERROR(EEXIST);
2244 goto out;
2245 }
2246 if (err != ENOENT)
2247 goto out;
2248
2249 /* The very first snapshot does not have a deadlist */
2250 if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2251 continue;
2252
2253 dsl_deadlist_space(&ds->ds_deadlist,
2254 &dlused, &dlcomp, &dluncomp);
2255 ddpa->used += dlused;
2256 ddpa->comp += dlcomp;
2257 ddpa->uncomp += dluncomp;
2258 }
2259
2260 /*
2261 * If we are a clone of a clone then we never reached ORIGIN,
2262 * so we need to subtract out the clone origin's used space.
2263 */
2264 if (ddpa->origin_origin) {
2265 ddpa->used -=
2266 dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
2267 ddpa->comp -=
2268 dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
2269 ddpa->uncomp -=
2270 dsl_dataset_phys(ddpa->origin_origin)->
2271 ds_uncompressed_bytes;
2272 }
2273
2274 /* Check that there is enough space and limit headroom here */
2275 err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
2276 0, ss_mv_cnt, ddpa->used, ddpa->cr);
2277 if (err != 0)
2278 goto out;
2279
2280 /*
2281 * Compute the amounts of space that will be used by snapshots
2282 * after the promotion (for both origin and clone). For each,
2283 * it is the amount of space that will be on all of their
2284 * deadlists (that was not born before their new origin).
2285 */
2286 if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
2287 uint64_t space;
2288
2289 /*
2290 * Note, typically this will not be a clone of a clone,
2291 * so dd_origin_txg will be < TXG_INITIAL, so
2292 * these snaplist_space() -> dsl_deadlist_space_range()
2293 * calls will be fast because they do not have to
2294 * iterate over all bps.
2295 */
2296 snap = list_head(&ddpa->origin_snaps);
2297 err = snaplist_space(&ddpa->shared_snaps,
2298 snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
2299 if (err != 0)
2300 goto out;
2301
2302 err = snaplist_space(&ddpa->clone_snaps,
2303 snap->ds->ds_dir->dd_origin_txg, &space);
2304 if (err != 0)
2305 goto out;
2306 ddpa->cloneusedsnap += space;
2307 }
2308 if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
2309 DD_FLAG_USED_BREAKDOWN) {
2310 err = snaplist_space(&ddpa->origin_snaps,
2311 dsl_dataset_phys(origin_ds)->ds_creation_txg,
2312 &ddpa->originusedsnap);
2313 if (err != 0)
2314 goto out;
2315 }
2316
2317 out:
2318 promote_rele(ddpa, FTAG);
2319 return (err);
2320 }
2321
2322 static void
dsl_dataset_promote_sync(void * arg,dmu_tx_t * tx)2323 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
2324 {
2325 dsl_dataset_promote_arg_t *ddpa = arg;
2326 dsl_pool_t *dp = dmu_tx_pool(tx);
2327 dsl_dataset_t *hds;
2328 struct promotenode *snap;
2329 dsl_dataset_t *origin_ds;
2330 dsl_dataset_t *origin_head;
2331 dsl_dir_t *dd;
2332 dsl_dir_t *odd = NULL;
2333 uint64_t oldnext_obj;
2334 int64_t delta;
2335 #if defined(__FreeBSD__) && defined(_KERNEL)
2336 char *oldname, *newname;
2337 #endif
2338
2339 VERIFY0(promote_hold(ddpa, dp, FTAG));
2340 hds = ddpa->ddpa_clone;
2341
2342 ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
2343
2344 snap = list_head(&ddpa->shared_snaps);
2345 origin_ds = snap->ds;
2346 dd = hds->ds_dir;
2347
2348 snap = list_head(&ddpa->origin_snaps);
2349 origin_head = snap->ds;
2350
2351 /*
2352 * We need to explicitly open odd, since origin_ds's dd will be
2353 * changing.
2354 */
2355 VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
2356 NULL, FTAG, &odd));
2357
2358 /* change origin's next snap */
2359 dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
2360 oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
2361 snap = list_tail(&ddpa->clone_snaps);
2362 ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2363 origin_ds->ds_object);
2364 dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
2365
2366 /* change the origin's next clone */
2367 if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
2368 dsl_dataset_remove_from_next_clones(origin_ds,
2369 snap->ds->ds_object, tx);
2370 VERIFY0(zap_add_int(dp->dp_meta_objset,
2371 dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
2372 oldnext_obj, tx));
2373 }
2374
2375 /* change origin */
2376 dmu_buf_will_dirty(dd->dd_dbuf, tx);
2377 ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
2378 dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
2379 dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
2380 dmu_buf_will_dirty(odd->dd_dbuf, tx);
2381 dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
2382 origin_head->ds_dir->dd_origin_txg =
2383 dsl_dataset_phys(origin_ds)->ds_creation_txg;
2384
2385 /* change dd_clone entries */
2386 if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2387 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2388 dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
2389 VERIFY0(zap_add_int(dp->dp_meta_objset,
2390 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2391 hds->ds_object, tx));
2392
2393 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2394 dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
2395 origin_head->ds_object, tx));
2396 if (dsl_dir_phys(dd)->dd_clones == 0) {
2397 dsl_dir_phys(dd)->dd_clones =
2398 zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
2399 DMU_OT_NONE, 0, tx);
2400 }
2401 VERIFY0(zap_add_int(dp->dp_meta_objset,
2402 dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
2403 }
2404
2405 #if defined(__FreeBSD__) && defined(_KERNEL)
2406 /* Take the spa_namespace_lock early so zvol renames don't deadlock. */
2407 mutex_enter(&spa_namespace_lock);
2408
2409 oldname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2410 newname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
2411 #endif
2412
2413 /* move snapshots to this dir */
2414 for (snap = list_head(&ddpa->shared_snaps); snap;
2415 snap = list_next(&ddpa->shared_snaps, snap)) {
2416 dsl_dataset_t *ds = snap->ds;
2417
2418 /*
2419 * Property callbacks are registered to a particular
2420 * dsl_dir. Since ours is changing, evict the objset
2421 * so that they will be unregistered from the old dsl_dir.
2422 */
2423 if (ds->ds_objset) {
2424 dmu_objset_evict(ds->ds_objset);
2425 ds->ds_objset = NULL;
2426 }
2427
2428 /* move snap name entry */
2429 VERIFY0(dsl_dataset_get_snapname(ds));
2430 VERIFY0(dsl_dataset_snap_remove(origin_head,
2431 ds->ds_snapname, tx, B_TRUE));
2432 VERIFY0(zap_add(dp->dp_meta_objset,
2433 dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
2434 8, 1, &ds->ds_object, tx));
2435 dsl_fs_ss_count_adjust(hds->ds_dir, 1,
2436 DD_FIELD_SNAPSHOT_COUNT, tx);
2437
2438 /* change containing dsl_dir */
2439 dmu_buf_will_dirty(ds->ds_dbuf, tx);
2440 ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
2441 dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
2442 ASSERT3P(ds->ds_dir, ==, odd);
2443 dsl_dir_rele(ds->ds_dir, ds);
2444 VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
2445 NULL, ds, &ds->ds_dir));
2446
2447 #if defined(__FreeBSD__) && defined(_KERNEL)
2448 dsl_dataset_name(ds, newname);
2449 zfsvfs_update_fromname(oldname, newname);
2450 zvol_rename_minors(oldname, newname);
2451 #endif
2452
2453 /* move any clone references */
2454 if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
2455 spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
2456 zap_cursor_t zc;
2457 zap_attribute_t za;
2458
2459 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2460 dsl_dataset_phys(ds)->ds_next_clones_obj);
2461 zap_cursor_retrieve(&zc, &za) == 0;
2462 zap_cursor_advance(&zc)) {
2463 dsl_dataset_t *cnds;
2464 uint64_t o;
2465
2466 if (za.za_first_integer == oldnext_obj) {
2467 /*
2468 * We've already moved the
2469 * origin's reference.
2470 */
2471 continue;
2472 }
2473
2474 VERIFY0(dsl_dataset_hold_obj(dp,
2475 za.za_first_integer, FTAG, &cnds));
2476 o = dsl_dir_phys(cnds->ds_dir)->
2477 dd_head_dataset_obj;
2478
2479 VERIFY0(zap_remove_int(dp->dp_meta_objset,
2480 dsl_dir_phys(odd)->dd_clones, o, tx));
2481 VERIFY0(zap_add_int(dp->dp_meta_objset,
2482 dsl_dir_phys(dd)->dd_clones, o, tx));
2483 dsl_dataset_rele(cnds, FTAG);
2484 }
2485 zap_cursor_fini(&zc);
2486 }
2487
2488 ASSERT(!dsl_prop_hascb(ds));
2489 }
2490
2491 #if defined(__FreeBSD__) && defined(_KERNEL)
2492 mutex_exit(&spa_namespace_lock);
2493
2494 kmem_free(newname, MAXPATHLEN);
2495 kmem_free(oldname, MAXPATHLEN);
2496 #endif
2497 /*
2498 * Change space accounting.
2499 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
2500 * both be valid, or both be 0 (resulting in delta == 0). This
2501 * is true for each of {clone,origin} independently.
2502 */
2503
2504 delta = ddpa->cloneusedsnap -
2505 dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
2506 ASSERT3S(delta, >=, 0);
2507 ASSERT3U(ddpa->used, >=, delta);
2508 dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
2509 dsl_dir_diduse_space(dd, DD_USED_HEAD,
2510 ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
2511
2512 delta = ddpa->originusedsnap -
2513 dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
2514 ASSERT3S(delta, <=, 0);
2515 ASSERT3U(ddpa->used, >=, -delta);
2516 dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
2517 dsl_dir_diduse_space(odd, DD_USED_HEAD,
2518 -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
2519
2520 dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
2521
2522 /* log history record */
2523 spa_history_log_internal_ds(hds, "promote", tx, "");
2524
2525 dsl_dir_rele(odd, FTAG);
2526 promote_rele(ddpa, FTAG);
2527 }
2528
2529 /*
2530 * Make a list of dsl_dataset_t's for the snapshots between first_obj
2531 * (exclusive) and last_obj (inclusive). The list will be in reverse
2532 * order (last_obj will be the list_head()). If first_obj == 0, do all
2533 * snapshots back to this dataset's origin.
2534 */
2535 static int
snaplist_make(dsl_pool_t * dp,uint64_t first_obj,uint64_t last_obj,list_t * l,void * tag)2536 snaplist_make(dsl_pool_t *dp,
2537 uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
2538 {
2539 uint64_t obj = last_obj;
2540
2541 list_create(l, sizeof (struct promotenode),
2542 offsetof(struct promotenode, link));
2543
2544 while (obj != first_obj) {
2545 dsl_dataset_t *ds;
2546 struct promotenode *snap;
2547 int err;
2548
2549 err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
2550 ASSERT(err != ENOENT);
2551 if (err != 0)
2552 return (err);
2553
2554 if (first_obj == 0)
2555 first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
2556
2557 snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
2558 snap->ds = ds;
2559 list_insert_tail(l, snap);
2560 obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
2561 }
2562
2563 return (0);
2564 }
2565
2566 static int
snaplist_space(list_t * l,uint64_t mintxg,uint64_t * spacep)2567 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
2568 {
2569 struct promotenode *snap;
2570
2571 *spacep = 0;
2572 for (snap = list_head(l); snap; snap = list_next(l, snap)) {
2573 uint64_t used, comp, uncomp;
2574 dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2575 mintxg, UINT64_MAX, &used, &comp, &uncomp);
2576 *spacep += used;
2577 }
2578 return (0);
2579 }
2580
2581 static void
snaplist_destroy(list_t * l,void * tag)2582 snaplist_destroy(list_t *l, void *tag)
2583 {
2584 struct promotenode *snap;
2585
2586 if (l == NULL || !list_link_active(&l->list_head))
2587 return;
2588
2589 while ((snap = list_tail(l)) != NULL) {
2590 list_remove(l, snap);
2591 dsl_dataset_rele(snap->ds, tag);
2592 kmem_free(snap, sizeof (*snap));
2593 }
2594 list_destroy(l);
2595 }
2596
2597 static int
promote_hold(dsl_dataset_promote_arg_t * ddpa,dsl_pool_t * dp,void * tag)2598 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
2599 {
2600 int error;
2601 dsl_dir_t *dd;
2602 struct promotenode *snap;
2603
2604 error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
2605 &ddpa->ddpa_clone);
2606 if (error != 0)
2607 return (error);
2608 dd = ddpa->ddpa_clone->ds_dir;
2609
2610 if (dsl_dataset_is_snapshot(ddpa->ddpa_clone) ||
2611 !dsl_dir_is_clone(dd)) {
2612 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2613 return (SET_ERROR(EINVAL));
2614 }
2615
2616 error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
2617 &ddpa->shared_snaps, tag);
2618 if (error != 0)
2619 goto out;
2620
2621 error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
2622 &ddpa->clone_snaps, tag);
2623 if (error != 0)
2624 goto out;
2625
2626 snap = list_head(&ddpa->shared_snaps);
2627 ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
2628 error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
2629 dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
2630 &ddpa->origin_snaps, tag);
2631 if (error != 0)
2632 goto out;
2633
2634 if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
2635 error = dsl_dataset_hold_obj(dp,
2636 dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
2637 tag, &ddpa->origin_origin);
2638 if (error != 0)
2639 goto out;
2640 }
2641 out:
2642 if (error != 0)
2643 promote_rele(ddpa, tag);
2644 return (error);
2645 }
2646
2647 static void
promote_rele(dsl_dataset_promote_arg_t * ddpa,void * tag)2648 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
2649 {
2650 snaplist_destroy(&ddpa->shared_snaps, tag);
2651 snaplist_destroy(&ddpa->clone_snaps, tag);
2652 snaplist_destroy(&ddpa->origin_snaps, tag);
2653 if (ddpa->origin_origin != NULL)
2654 dsl_dataset_rele(ddpa->origin_origin, tag);
2655 dsl_dataset_rele(ddpa->ddpa_clone, tag);
2656 }
2657
2658 /*
2659 * Promote a clone.
2660 *
2661 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
2662 * in with the name. (It must be at least MAXNAMELEN bytes long.)
2663 */
2664 int
dsl_dataset_promote(const char * name,char * conflsnap)2665 dsl_dataset_promote(const char *name, char *conflsnap)
2666 {
2667 dsl_dataset_promote_arg_t ddpa = { 0 };
2668 uint64_t numsnaps;
2669 int error;
2670 objset_t *os;
2671
2672 /*
2673 * We will modify space proportional to the number of
2674 * snapshots. Compute numsnaps.
2675 */
2676 error = dmu_objset_hold(name, FTAG, &os);
2677 if (error != 0)
2678 return (error);
2679 error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
2680 dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
2681 &numsnaps);
2682 dmu_objset_rele(os, FTAG);
2683 if (error != 0)
2684 return (error);
2685
2686 ddpa.ddpa_clonename = name;
2687 ddpa.err_ds = conflsnap;
2688 ddpa.cr = CRED();
2689
2690 return (dsl_sync_task(name, dsl_dataset_promote_check,
2691 dsl_dataset_promote_sync, &ddpa,
2692 2 + numsnaps, ZFS_SPACE_CHECK_RESERVED));
2693 }
2694
2695 int
dsl_dataset_clone_swap_check_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,boolean_t force,void * owner,dmu_tx_t * tx)2696 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
2697 dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
2698 {
2699 int64_t unused_refres_delta;
2700
2701 /* they should both be heads */
2702 if (dsl_dataset_is_snapshot(clone) ||
2703 dsl_dataset_is_snapshot(origin_head))
2704 return (SET_ERROR(EINVAL));
2705
2706 /* if we are not forcing, the branch point should be just before them */
2707 if (!force && clone->ds_prev != origin_head->ds_prev)
2708 return (SET_ERROR(EINVAL));
2709
2710 /* clone should be the clone (unless they are unrelated) */
2711 if (clone->ds_prev != NULL &&
2712 clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
2713 origin_head->ds_dir != clone->ds_prev->ds_dir)
2714 return (SET_ERROR(EINVAL));
2715
2716 /* the clone should be a child of the origin */
2717 if (clone->ds_dir->dd_parent != origin_head->ds_dir)
2718 return (SET_ERROR(EINVAL));
2719
2720 /* origin_head shouldn't be modified unless 'force' */
2721 if (!force &&
2722 dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
2723 return (SET_ERROR(ETXTBSY));
2724
2725 /* origin_head should have no long holds (e.g. is not mounted) */
2726 if (dsl_dataset_handoff_check(origin_head, owner, tx))
2727 return (SET_ERROR(EBUSY));
2728
2729 /* check amount of any unconsumed refreservation */
2730 unused_refres_delta =
2731 (int64_t)MIN(origin_head->ds_reserved,
2732 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2733 (int64_t)MIN(origin_head->ds_reserved,
2734 dsl_dataset_phys(clone)->ds_unique_bytes);
2735
2736 if (unused_refres_delta > 0 &&
2737 unused_refres_delta >
2738 dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
2739 return (SET_ERROR(ENOSPC));
2740
2741 /* clone can't be over the head's refquota */
2742 if (origin_head->ds_quota != 0 &&
2743 dsl_dataset_phys(clone)->ds_referenced_bytes >
2744 origin_head->ds_quota)
2745 return (SET_ERROR(EDQUOT));
2746
2747 return (0);
2748 }
2749
2750 void
dsl_dataset_clone_swap_sync_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,dmu_tx_t * tx)2751 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
2752 dsl_dataset_t *origin_head, dmu_tx_t *tx)
2753 {
2754 dsl_pool_t *dp = dmu_tx_pool(tx);
2755 int64_t unused_refres_delta;
2756
2757 ASSERT(clone->ds_reserved == 0);
2758 ASSERT(origin_head->ds_quota == 0 ||
2759 dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota);
2760 ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
2761
2762 dmu_buf_will_dirty(clone->ds_dbuf, tx);
2763 dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
2764
2765 if (clone->ds_objset != NULL) {
2766 dmu_objset_evict(clone->ds_objset);
2767 clone->ds_objset = NULL;
2768 }
2769
2770 if (origin_head->ds_objset != NULL) {
2771 dmu_objset_evict(origin_head->ds_objset);
2772 origin_head->ds_objset = NULL;
2773 }
2774
2775 unused_refres_delta =
2776 (int64_t)MIN(origin_head->ds_reserved,
2777 dsl_dataset_phys(origin_head)->ds_unique_bytes) -
2778 (int64_t)MIN(origin_head->ds_reserved,
2779 dsl_dataset_phys(clone)->ds_unique_bytes);
2780
2781 /*
2782 * Reset origin's unique bytes, if it exists.
2783 */
2784 if (clone->ds_prev) {
2785 dsl_dataset_t *origin = clone->ds_prev;
2786 uint64_t comp, uncomp;
2787
2788 dmu_buf_will_dirty(origin->ds_dbuf, tx);
2789 dsl_deadlist_space_range(&clone->ds_deadlist,
2790 dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
2791 &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
2792 }
2793
2794 /* swap blkptrs */
2795 {
2796 blkptr_t tmp;
2797 tmp = dsl_dataset_phys(origin_head)->ds_bp;
2798 dsl_dataset_phys(origin_head)->ds_bp =
2799 dsl_dataset_phys(clone)->ds_bp;
2800 dsl_dataset_phys(clone)->ds_bp = tmp;
2801 }
2802
2803 /* set dd_*_bytes */
2804 {
2805 int64_t dused, dcomp, duncomp;
2806 uint64_t cdl_used, cdl_comp, cdl_uncomp;
2807 uint64_t odl_used, odl_comp, odl_uncomp;
2808
2809 ASSERT3U(dsl_dir_phys(clone->ds_dir)->
2810 dd_used_breakdown[DD_USED_SNAP], ==, 0);
2811
2812 dsl_deadlist_space(&clone->ds_deadlist,
2813 &cdl_used, &cdl_comp, &cdl_uncomp);
2814 dsl_deadlist_space(&origin_head->ds_deadlist,
2815 &odl_used, &odl_comp, &odl_uncomp);
2816
2817 dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
2818 cdl_used -
2819 (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
2820 odl_used);
2821 dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
2822 cdl_comp -
2823 (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
2824 odl_comp);
2825 duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
2826 cdl_uncomp -
2827 (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
2828 odl_uncomp);
2829
2830 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
2831 dused, dcomp, duncomp, tx);
2832 dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
2833 -dused, -dcomp, -duncomp, tx);
2834
2835 /*
2836 * The difference in the space used by snapshots is the
2837 * difference in snapshot space due to the head's
2838 * deadlist (since that's the only thing that's
2839 * changing that affects the snapused).
2840 */
2841 dsl_deadlist_space_range(&clone->ds_deadlist,
2842 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2843 &cdl_used, &cdl_comp, &cdl_uncomp);
2844 dsl_deadlist_space_range(&origin_head->ds_deadlist,
2845 origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
2846 &odl_used, &odl_comp, &odl_uncomp);
2847 dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
2848 DD_USED_HEAD, DD_USED_SNAP, NULL);
2849 }
2850
2851 /* swap ds_*_bytes */
2852 SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
2853 dsl_dataset_phys(clone)->ds_referenced_bytes);
2854 SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
2855 dsl_dataset_phys(clone)->ds_compressed_bytes);
2856 SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
2857 dsl_dataset_phys(clone)->ds_uncompressed_bytes);
2858 SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
2859 dsl_dataset_phys(clone)->ds_unique_bytes);
2860
2861 /* apply any parent delta for change in unconsumed refreservation */
2862 dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
2863 unused_refres_delta, 0, 0, tx);
2864
2865 /*
2866 * Swap deadlists.
2867 */
2868 dsl_deadlist_close(&clone->ds_deadlist);
2869 dsl_deadlist_close(&origin_head->ds_deadlist);
2870 SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
2871 dsl_dataset_phys(clone)->ds_deadlist_obj);
2872 dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
2873 dsl_dataset_phys(clone)->ds_deadlist_obj);
2874 dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
2875 dsl_dataset_phys(origin_head)->ds_deadlist_obj);
2876
2877 dsl_scan_ds_clone_swapped(origin_head, clone, tx);
2878
2879 spa_history_log_internal_ds(clone, "clone swap", tx,
2880 "parent=%s", origin_head->ds_dir->dd_myname);
2881 }
2882
2883 /*
2884 * Given a pool name and a dataset object number in that pool,
2885 * return the name of that dataset.
2886 */
2887 int
dsl_dsobj_to_dsname(char * pname,uint64_t obj,char * buf)2888 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
2889 {
2890 dsl_pool_t *dp;
2891 dsl_dataset_t *ds;
2892 int error;
2893
2894 error = dsl_pool_hold(pname, FTAG, &dp);
2895 if (error != 0)
2896 return (error);
2897
2898 error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
2899 if (error == 0) {
2900 dsl_dataset_name(ds, buf);
2901 dsl_dataset_rele(ds, FTAG);
2902 }
2903 dsl_pool_rele(dp, FTAG);
2904
2905 return (error);
2906 }
2907
2908 int
dsl_dataset_check_quota(dsl_dataset_t * ds,boolean_t check_quota,uint64_t asize,uint64_t inflight,uint64_t * used,uint64_t * ref_rsrv)2909 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
2910 uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
2911 {
2912 int error = 0;
2913
2914 ASSERT3S(asize, >, 0);
2915
2916 /*
2917 * *ref_rsrv is the portion of asize that will come from any
2918 * unconsumed refreservation space.
2919 */
2920 *ref_rsrv = 0;
2921
2922 mutex_enter(&ds->ds_lock);
2923 /*
2924 * Make a space adjustment for reserved bytes.
2925 */
2926 if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2927 ASSERT3U(*used, >=,
2928 ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2929 *used -=
2930 (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
2931 *ref_rsrv =
2932 asize - MIN(asize, parent_delta(ds, asize + inflight));
2933 }
2934
2935 if (!check_quota || ds->ds_quota == 0) {
2936 mutex_exit(&ds->ds_lock);
2937 return (0);
2938 }
2939 /*
2940 * If they are requesting more space, and our current estimate
2941 * is over quota, they get to try again unless the actual
2942 * on-disk is over quota and there are no pending changes (which
2943 * may free up space for us).
2944 */
2945 if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
2946 ds->ds_quota) {
2947 if (inflight > 0 ||
2948 dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
2949 error = SET_ERROR(ERESTART);
2950 else
2951 error = SET_ERROR(EDQUOT);
2952 }
2953 mutex_exit(&ds->ds_lock);
2954
2955 return (error);
2956 }
2957
2958 typedef struct dsl_dataset_set_qr_arg {
2959 const char *ddsqra_name;
2960 zprop_source_t ddsqra_source;
2961 uint64_t ddsqra_value;
2962 } dsl_dataset_set_qr_arg_t;
2963
2964
2965 /* ARGSUSED */
2966 static int
dsl_dataset_set_refquota_check(void * arg,dmu_tx_t * tx)2967 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
2968 {
2969 dsl_dataset_set_qr_arg_t *ddsqra = arg;
2970 dsl_pool_t *dp = dmu_tx_pool(tx);
2971 dsl_dataset_t *ds;
2972 int error;
2973 uint64_t newval;
2974
2975 if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
2976 return (SET_ERROR(ENOTSUP));
2977
2978 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
2979 if (error != 0)
2980 return (error);
2981
2982 if (dsl_dataset_is_snapshot(ds)) {
2983 dsl_dataset_rele(ds, FTAG);
2984 return (SET_ERROR(EINVAL));
2985 }
2986
2987 error = dsl_prop_predict(ds->ds_dir,
2988 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
2989 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
2990 if (error != 0) {
2991 dsl_dataset_rele(ds, FTAG);
2992 return (error);
2993 }
2994
2995 if (newval == 0) {
2996 dsl_dataset_rele(ds, FTAG);
2997 return (0);
2998 }
2999
3000 if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3001 newval < ds->ds_reserved) {
3002 dsl_dataset_rele(ds, FTAG);
3003 return (SET_ERROR(ENOSPC));
3004 }
3005
3006 dsl_dataset_rele(ds, FTAG);
3007 return (0);
3008 }
3009
3010 static void
dsl_dataset_set_refquota_sync(void * arg,dmu_tx_t * tx)3011 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3012 {
3013 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3014 dsl_pool_t *dp = dmu_tx_pool(tx);
3015 dsl_dataset_t *ds;
3016 uint64_t newval;
3017
3018 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3019
3020 dsl_prop_set_sync_impl(ds,
3021 zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3022 ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3023 &ddsqra->ddsqra_value, tx);
3024
3025 VERIFY0(dsl_prop_get_int_ds(ds,
3026 zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3027
3028 if (ds->ds_quota != newval) {
3029 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3030 ds->ds_quota = newval;
3031 }
3032 dsl_dataset_rele(ds, FTAG);
3033 }
3034
3035 int
dsl_dataset_set_refquota(const char * dsname,zprop_source_t source,uint64_t refquota)3036 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3037 uint64_t refquota)
3038 {
3039 dsl_dataset_set_qr_arg_t ddsqra;
3040
3041 ddsqra.ddsqra_name = dsname;
3042 ddsqra.ddsqra_source = source;
3043 ddsqra.ddsqra_value = refquota;
3044
3045 return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3046 dsl_dataset_set_refquota_sync, &ddsqra, 0, ZFS_SPACE_CHECK_NONE));
3047 }
3048
3049 static int
dsl_dataset_set_refreservation_check(void * arg,dmu_tx_t * tx)3050 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3051 {
3052 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3053 dsl_pool_t *dp = dmu_tx_pool(tx);
3054 dsl_dataset_t *ds;
3055 int error;
3056 uint64_t newval, unique;
3057
3058 if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3059 return (SET_ERROR(ENOTSUP));
3060
3061 error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3062 if (error != 0)
3063 return (error);
3064
3065 if (dsl_dataset_is_snapshot(ds)) {
3066 dsl_dataset_rele(ds, FTAG);
3067 return (SET_ERROR(EINVAL));
3068 }
3069
3070 error = dsl_prop_predict(ds->ds_dir,
3071 zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3072 ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3073 if (error != 0) {
3074 dsl_dataset_rele(ds, FTAG);
3075 return (error);
3076 }
3077
3078 /*
3079 * If we are doing the preliminary check in open context, the
3080 * space estimates may be inaccurate.
3081 */
3082 if (!dmu_tx_is_syncing(tx)) {
3083 dsl_dataset_rele(ds, FTAG);
3084 return (0);
3085 }
3086
3087 mutex_enter(&ds->ds_lock);
3088 if (!DS_UNIQUE_IS_ACCURATE(ds))
3089 dsl_dataset_recalc_head_uniq(ds);
3090 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3091 mutex_exit(&ds->ds_lock);
3092
3093 if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3094 uint64_t delta = MAX(unique, newval) -
3095 MAX(unique, ds->ds_reserved);
3096
3097 if (delta >
3098 dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3099 (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3100 dsl_dataset_rele(ds, FTAG);
3101 return (SET_ERROR(ENOSPC));
3102 }
3103 }
3104
3105 dsl_dataset_rele(ds, FTAG);
3106 return (0);
3107 }
3108
3109 void
dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t * ds,zprop_source_t source,uint64_t value,dmu_tx_t * tx)3110 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3111 zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3112 {
3113 uint64_t newval;
3114 uint64_t unique;
3115 int64_t delta;
3116
3117 dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3118 source, sizeof (value), 1, &value, tx);
3119
3120 VERIFY0(dsl_prop_get_int_ds(ds,
3121 zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3122
3123 dmu_buf_will_dirty(ds->ds_dbuf, tx);
3124 mutex_enter(&ds->ds_dir->dd_lock);
3125 mutex_enter(&ds->ds_lock);
3126 ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3127 unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3128 delta = MAX(0, (int64_t)(newval - unique)) -
3129 MAX(0, (int64_t)(ds->ds_reserved - unique));
3130 ds->ds_reserved = newval;
3131 mutex_exit(&ds->ds_lock);
3132
3133 dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3134 mutex_exit(&ds->ds_dir->dd_lock);
3135 }
3136
3137 static void
dsl_dataset_set_refreservation_sync(void * arg,dmu_tx_t * tx)3138 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3139 {
3140 dsl_dataset_set_qr_arg_t *ddsqra = arg;
3141 dsl_pool_t *dp = dmu_tx_pool(tx);
3142 dsl_dataset_t *ds;
3143
3144 VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3145 dsl_dataset_set_refreservation_sync_impl(ds,
3146 ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3147 dsl_dataset_rele(ds, FTAG);
3148 }
3149
3150 int
dsl_dataset_set_refreservation(const char * dsname,zprop_source_t source,uint64_t refreservation)3151 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3152 uint64_t refreservation)
3153 {
3154 dsl_dataset_set_qr_arg_t ddsqra;
3155
3156 ddsqra.ddsqra_name = dsname;
3157 ddsqra.ddsqra_source = source;
3158 ddsqra.ddsqra_value = refreservation;
3159
3160 return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3161 dsl_dataset_set_refreservation_sync, &ddsqra,
3162 0, ZFS_SPACE_CHECK_NONE));
3163 }
3164
3165 /*
3166 * Return (in *usedp) the amount of space written in new that is not
3167 * present in oldsnap. New may be a snapshot or the head. Old must be
3168 * a snapshot before new, in new's filesystem (or its origin). If not then
3169 * fail and return EINVAL.
3170 *
3171 * The written space is calculated by considering two components: First, we
3172 * ignore any freed space, and calculate the written as new's used space
3173 * minus old's used space. Next, we add in the amount of space that was freed
3174 * between the two snapshots, thus reducing new's used space relative to old's.
3175 * Specifically, this is the space that was born before old->ds_creation_txg,
3176 * and freed before new (ie. on new's deadlist or a previous deadlist).
3177 *
3178 * space freed [---------------------]
3179 * snapshots ---O-------O--------O-------O------
3180 * oldsnap new
3181 */
3182 int
dsl_dataset_space_written(dsl_dataset_t * oldsnap,dsl_dataset_t * new,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)3183 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
3184 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3185 {
3186 int err = 0;
3187 uint64_t snapobj;
3188 dsl_pool_t *dp = new->ds_dir->dd_pool;
3189
3190 ASSERT(dsl_pool_config_held(dp));
3191
3192 *usedp = 0;
3193 *usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
3194 *usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
3195
3196 *compp = 0;
3197 *compp += dsl_dataset_phys(new)->ds_compressed_bytes;
3198 *compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
3199
3200 *uncompp = 0;
3201 *uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
3202 *uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
3203
3204 snapobj = new->ds_object;
3205 while (snapobj != oldsnap->ds_object) {
3206 dsl_dataset_t *snap;
3207 uint64_t used, comp, uncomp;
3208
3209 if (snapobj == new->ds_object) {
3210 snap = new;
3211 } else {
3212 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
3213 if (err != 0)
3214 break;
3215 }
3216
3217 if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
3218 dsl_dataset_phys(oldsnap)->ds_creation_txg) {
3219 /*
3220 * The blocks in the deadlist can not be born after
3221 * ds_prev_snap_txg, so get the whole deadlist space,
3222 * which is more efficient (especially for old-format
3223 * deadlists). Unfortunately the deadlist code
3224 * doesn't have enough information to make this
3225 * optimization itself.
3226 */
3227 dsl_deadlist_space(&snap->ds_deadlist,
3228 &used, &comp, &uncomp);
3229 } else {
3230 dsl_deadlist_space_range(&snap->ds_deadlist,
3231 0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
3232 &used, &comp, &uncomp);
3233 }
3234 *usedp += used;
3235 *compp += comp;
3236 *uncompp += uncomp;
3237
3238 /*
3239 * If we get to the beginning of the chain of snapshots
3240 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
3241 * was not a snapshot of/before new.
3242 */
3243 snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
3244 if (snap != new)
3245 dsl_dataset_rele(snap, FTAG);
3246 if (snapobj == 0) {
3247 err = SET_ERROR(EINVAL);
3248 break;
3249 }
3250
3251 }
3252 return (err);
3253 }
3254
3255 /*
3256 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
3257 * lastsnap, and all snapshots in between are deleted.
3258 *
3259 * blocks that would be freed [---------------------------]
3260 * snapshots ---O-------O--------O-------O--------O
3261 * firstsnap lastsnap
3262 *
3263 * This is the set of blocks that were born after the snap before firstsnap,
3264 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
3265 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
3266 * We calculate this by iterating over the relevant deadlists (from the snap
3267 * after lastsnap, backward to the snap after firstsnap), summing up the
3268 * space on the deadlist that was born after the snap before firstsnap.
3269 */
3270 int
dsl_dataset_space_wouldfree(dsl_dataset_t * firstsnap,dsl_dataset_t * lastsnap,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)3271 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
3272 dsl_dataset_t *lastsnap,
3273 uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
3274 {
3275 int err = 0;
3276 uint64_t snapobj;
3277 dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
3278
3279 ASSERT(dsl_dataset_is_snapshot(firstsnap));
3280 ASSERT(dsl_dataset_is_snapshot(lastsnap));
3281
3282 /*
3283 * Check that the snapshots are in the same dsl_dir, and firstsnap
3284 * is before lastsnap.
3285 */
3286 if (firstsnap->ds_dir != lastsnap->ds_dir ||
3287 dsl_dataset_phys(firstsnap)->ds_creation_txg >
3288 dsl_dataset_phys(lastsnap)->ds_creation_txg)
3289 return (SET_ERROR(EINVAL));
3290
3291 *usedp = *compp = *uncompp = 0;
3292
3293 snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
3294 while (snapobj != firstsnap->ds_object) {
3295 dsl_dataset_t *ds;
3296 uint64_t used, comp, uncomp;
3297
3298 err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
3299 if (err != 0)
3300 break;
3301
3302 dsl_deadlist_space_range(&ds->ds_deadlist,
3303 dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
3304 &used, &comp, &uncomp);
3305 *usedp += used;
3306 *compp += comp;
3307 *uncompp += uncomp;
3308
3309 snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3310 ASSERT3U(snapobj, !=, 0);
3311 dsl_dataset_rele(ds, FTAG);
3312 }
3313 return (err);
3314 }
3315
3316 static int
dsl_dataset_activate_large_blocks_check(void * arg,dmu_tx_t * tx)3317 dsl_dataset_activate_large_blocks_check(void *arg, dmu_tx_t *tx)
3318 {
3319 const char *dsname = arg;
3320 dsl_dataset_t *ds;
3321 dsl_pool_t *dp = dmu_tx_pool(tx);
3322 int error = 0;
3323
3324 if (!spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_LARGE_BLOCKS))
3325 return (SET_ERROR(ENOTSUP));
3326
3327 ASSERT(spa_feature_is_enabled(dp->dp_spa,
3328 SPA_FEATURE_EXTENSIBLE_DATASET));
3329
3330 error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
3331 if (error != 0)
3332 return (error);
3333
3334 if (ds->ds_large_blocks)
3335 error = EALREADY;
3336 dsl_dataset_rele(ds, FTAG);
3337
3338 return (error);
3339 }
3340
3341 void
dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj,dmu_tx_t * tx)3342 dsl_dataset_activate_large_blocks_sync_impl(uint64_t dsobj, dmu_tx_t *tx)
3343 {
3344 spa_t *spa = dmu_tx_pool(tx)->dp_spa;
3345 objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
3346 uint64_t zero = 0;
3347
3348 spa_feature_incr(spa, SPA_FEATURE_LARGE_BLOCKS, tx);
3349 dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
3350
3351 VERIFY0(zap_add(mos, dsobj, DS_FIELD_LARGE_BLOCKS,
3352 sizeof (zero), 1, &zero, tx));
3353 }
3354
3355 static void
dsl_dataset_activate_large_blocks_sync(void * arg,dmu_tx_t * tx)3356 dsl_dataset_activate_large_blocks_sync(void *arg, dmu_tx_t *tx)
3357 {
3358 const char *dsname = arg;
3359 dsl_dataset_t *ds;
3360
3361 VERIFY0(dsl_dataset_hold(dmu_tx_pool(tx), dsname, FTAG, &ds));
3362
3363 dsl_dataset_activate_large_blocks_sync_impl(ds->ds_object, tx);
3364 ASSERT(!ds->ds_large_blocks);
3365 ds->ds_large_blocks = B_TRUE;
3366 dsl_dataset_rele(ds, FTAG);
3367 }
3368
3369 int
dsl_dataset_activate_large_blocks(const char * dsname)3370 dsl_dataset_activate_large_blocks(const char *dsname)
3371 {
3372 int error;
3373
3374 error = dsl_sync_task(dsname,
3375 dsl_dataset_activate_large_blocks_check,
3376 dsl_dataset_activate_large_blocks_sync, (void *)dsname,
3377 1, ZFS_SPACE_CHECK_RESERVED);
3378
3379 /*
3380 * EALREADY indicates that this dataset already supports large blocks.
3381 */
3382 if (error == EALREADY)
3383 error = 0;
3384 return (error);
3385 }
3386
3387 /*
3388 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
3389 * For example, they could both be snapshots of the same filesystem, and
3390 * 'earlier' is before 'later'. Or 'earlier' could be the origin of
3391 * 'later's filesystem. Or 'earlier' could be an older snapshot in the origin's
3392 * filesystem. Or 'earlier' could be the origin's origin.
3393 *
3394 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
3395 */
3396 boolean_t
dsl_dataset_is_before(dsl_dataset_t * later,dsl_dataset_t * earlier,uint64_t earlier_txg)3397 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
3398 uint64_t earlier_txg)
3399 {
3400 dsl_pool_t *dp = later->ds_dir->dd_pool;
3401 int error;
3402 boolean_t ret;
3403
3404 ASSERT(dsl_pool_config_held(dp));
3405 ASSERT(dsl_dataset_is_snapshot(earlier) || earlier_txg != 0);
3406
3407 if (earlier_txg == 0)
3408 earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
3409
3410 if (dsl_dataset_is_snapshot(later) &&
3411 earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
3412 return (B_FALSE);
3413
3414 if (later->ds_dir == earlier->ds_dir)
3415 return (B_TRUE);
3416 if (!dsl_dir_is_clone(later->ds_dir))
3417 return (B_FALSE);
3418
3419 if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
3420 return (B_TRUE);
3421 dsl_dataset_t *origin;
3422 error = dsl_dataset_hold_obj(dp,
3423 dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
3424 if (error != 0)
3425 return (B_FALSE);
3426 ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
3427 dsl_dataset_rele(origin, FTAG);
3428 return (ret);
3429 }
3430
3431
3432 void
dsl_dataset_zapify(dsl_dataset_t * ds,dmu_tx_t * tx)3433 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
3434 {
3435 objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
3436 dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
3437 }
3438
3439 void
dsl_event_notify(dsl_dataset_t * ds,const char * name)3440 dsl_event_notify(dsl_dataset_t *ds, const char *name)
3441 {
3442 #ifdef _KERNEL
3443 spa_t *spa;
3444 sysevent_t *ev;
3445 sysevent_attr_list_t *attr = NULL;
3446 sysevent_value_t value;
3447 sysevent_id_t eid;
3448 char dsname[MAXNAMELEN];
3449
3450 spa = dsl_dataset_get_spa(ds);
3451 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
3452 SE_SLEEP);
3453
3454 value.value_type = SE_DATA_TYPE_STRING;
3455 value.value.sv_string = spa_name(spa);
3456 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
3457 goto done;
3458
3459 value.value_type = SE_DATA_TYPE_UINT64;
3460 value.value.sv_uint64 = spa_guid(spa);
3461 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
3462 goto done;
3463
3464 if (ds) {
3465 dsl_dataset_name(ds, dsname);
3466 value.value_type = SE_DATA_TYPE_STRING;
3467 value.value.sv_string = dsname;
3468 if (sysevent_add_attr(&attr, ZFS_EV_DSL_NAME, &value,
3469 SE_SLEEP) != 0)
3470 goto done;
3471 }
3472
3473 if (sysevent_attach_attributes(ev, attr) != 0)
3474 goto done;
3475 attr = NULL;
3476
3477 (void) log_sysevent(ev, SE_SLEEP, &eid);
3478
3479 done:
3480 if (attr)
3481 sysevent_free_attr(attr);
3482 sysevent_free(ev);
3483 #endif
3484 }
3485
3486 void
dsl_event_notify_rename(spa_t * spa,const char * oldname,const char * newname,const char * name)3487 dsl_event_notify_rename(spa_t *spa, const char *oldname, const char *newname,
3488 const char *name)
3489 {
3490 #ifdef _KERNEL
3491 sysevent_t *ev;
3492 sysevent_attr_list_t *attr = NULL;
3493 sysevent_value_t value;
3494 sysevent_id_t eid;
3495 char dsname[MAXNAMELEN];
3496
3497 ev = sysevent_alloc(EC_ZFS, (char *)name, SUNW_KERN_PUB "zfs",
3498 SE_SLEEP);
3499
3500 value.value_type = SE_DATA_TYPE_STRING;
3501 value.value.sv_string = spa_name(spa);
3502 if (sysevent_add_attr(&attr, ZFS_EV_POOL_NAME, &value, SE_SLEEP) != 0)
3503 goto done;
3504
3505 value.value_type = SE_DATA_TYPE_UINT64;
3506 value.value.sv_uint64 = spa_guid(spa);
3507 if (sysevent_add_attr(&attr, ZFS_EV_POOL_GUID, &value, SE_SLEEP) != 0)
3508 goto done;
3509
3510 value.value_type = SE_DATA_TYPE_STRING;
3511 value.value.sv_string = (char *)oldname;
3512 if (sysevent_add_attr(&attr, ZFS_EV_DSL_NAME, &value,
3513 SE_SLEEP) != 0)
3514 goto done;
3515
3516 value.value_type = SE_DATA_TYPE_STRING;
3517 value.value.sv_string = (char *)newname;
3518 if (sysevent_add_attr(&attr, ZFS_EV_DSL_NEW_NAME, &value,
3519 SE_SLEEP) != 0)
3520 goto done;
3521
3522 if (sysevent_attach_attributes(ev, attr) != 0)
3523 goto done;
3524 attr = NULL;
3525
3526 (void) log_sysevent(ev, SE_SLEEP, &eid);
3527
3528 done:
3529 if (attr)
3530 sysevent_free_attr(attr);
3531 sysevent_free(ev);
3532 #endif
3533 }
3534
3535